]> bbs.cooldavid.org Git - net-next-2.6.git/blame - arch/x86/kvm/vmx.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: VMX: Add module option to disable flexpriority
[net-next-2.6.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>
e495606d 29
6aa8b732 30#include <asm/io.h>
3b3be0d1 31#include <asm/desc.h>
6aa8b732 32
6aa8b732
AK		 33MODULE_AUTHOR("Qumranet");
34MODULE_LICENSE("GPL");
35
c7addb90
AK		 36static int bypass_guest_pf = 1;
37module_param(bypass_guest_pf, bool, 0);
38
2384d2b3
SY		 39static int enable_vpid = 1;
40module_param(enable_vpid, bool, 0);
41
4c9fc8ef
AK		 42static int flexpriority_enabled = 1;
43module_param(flexpriority_enabled, bool, 0);
44
a2fa3e9f
GH		 45struct vmcs {
46 u32 revision_id;
47 u32 abort;
48 char data[0];
49};
50
51struct vcpu_vmx {
fb3f0f51 52 struct kvm_vcpu vcpu;
a2fa3e9f 53 int launched;
29bd8a78 54 u8 fail;
1155f76a 55 u32 idt_vectoring_info;
a2fa3e9f
GH		 56 struct kvm_msr_entry *guest_msrs;
57 struct kvm_msr_entry *host_msrs;
58 int nmsrs;
59 int save_nmsrs;
60 int msr_offset_efer;
61#ifdef CONFIG_X86_64
62 int msr_offset_kernel_gs_base;
63#endif
64 struct vmcs *vmcs;
65 struct {
66 int loaded;
67 u16 fs_sel, gs_sel, ldt_sel;
152d3f2f
LV		 68 int gs_ldt_reload_needed;
69 int fs_reload_needed;
51c6cf66 70 int guest_efer_loaded;
d77c26fc 71 } host_state;
9c8cba37
AK		 72 struct {
73 struct {
74 bool pending;
75 u8 vector;
76 unsigned rip;
77 } irq;
78 } rmode;
2384d2b3 79 int vpid;
a2fa3e9f
GH		 80};
81
82static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
83{
fb3f0f51 84 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 85}
86
75880a01
AK		 87static int init_rmode_tss(struct kvm *kvm);
88
6aa8b732
AK		 89static DEFINE_PER_CPU(struct vmcs *, vmxarea);
90static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
91
fdef3ad1
HQ		 92static struct page *vmx_io_bitmap_a;
93static struct page *vmx_io_bitmap_b;
94
2384d2b3
SY		 95static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
96static DEFINE_SPINLOCK(vmx_vpid_lock);
97
1c3d14fe 98static struct vmcs_config {
6aa8b732
AK		 99 int size;
100 int order;
101 u32 revision_id;
1c3d14fe
YS		 102 u32 pin_based_exec_ctrl;
103 u32 cpu_based_exec_ctrl;
f78e0e2e 104 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 105 u32 vmexit_ctrl;
106 u32 vmentry_ctrl;
107} vmcs_config;
6aa8b732
AK		 108
109#define VMX_SEGMENT_FIELD(seg) \
110 [VCPU_SREG_##seg] = { \
111 .selector = GUEST_##seg##_SELECTOR, \
112 .base = GUEST_##seg##_BASE, \
113 .limit = GUEST_##seg##_LIMIT, \
114 .ar_bytes = GUEST_##seg##_AR_BYTES, \
115 }
116
117static struct kvm_vmx_segment_field {
118 unsigned selector;
119 unsigned base;
120 unsigned limit;
121 unsigned ar_bytes;
122} kvm_vmx_segment_fields[] = {
123 VMX_SEGMENT_FIELD(CS),
124 VMX_SEGMENT_FIELD(DS),
125 VMX_SEGMENT_FIELD(ES),
126 VMX_SEGMENT_FIELD(FS),
127 VMX_SEGMENT_FIELD(GS),
128 VMX_SEGMENT_FIELD(SS),
129 VMX_SEGMENT_FIELD(TR),
130 VMX_SEGMENT_FIELD(LDTR),
131};
132
4d56c8a7
AK		 133/*
134 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
135 * away by decrementing the array size.
136 */
6aa8b732 137static const u32 vmx_msr_index[] = {
05b3e0c2 138#ifdef CONFIG_X86_64
6aa8b732
AK		 139 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
140#endif
141 MSR_EFER, MSR_K6_STAR,
142};
9d8f549d 143#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 144
a2fa3e9f
GH		 145static void load_msrs(struct kvm_msr_entry *e, int n)
146{
147 int i;
148
149 for (i = 0; i < n; ++i)
150 wrmsrl(e[i].index, e[i].data);
151}
152
153static void save_msrs(struct kvm_msr_entry *e, int n)
154{
155 int i;
156
157 for (i = 0; i < n; ++i)
158 rdmsrl(e[i].index, e[i].data);
159}
160
6aa8b732
AK		 161static inline int is_page_fault(u32 intr_info)
162{
163 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
164 INTR_INFO_VALID_MASK)) ==
165 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
166}
167
2ab455cc
AL		 168static inline int is_no_device(u32 intr_info)
169{
170 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
171 INTR_INFO_VALID_MASK)) ==
172 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
173}
174
7aa81cc0
AL		 175static inline int is_invalid_opcode(u32 intr_info)
176{
177 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
178 INTR_INFO_VALID_MASK)) ==
179 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
180}
181
6aa8b732
AK		 182static inline int is_external_interrupt(u32 intr_info)
183{
184 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
185 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
186}
187
6e5d865c
YS		 188static inline int cpu_has_vmx_tpr_shadow(void)
189{
190 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
191}
192
193static inline int vm_need_tpr_shadow(struct kvm *kvm)
194{
195 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
196}
197
f78e0e2e
SY		 198static inline int cpu_has_secondary_exec_ctrls(void)
199{
200 return (vmcs_config.cpu_based_exec_ctrl &
201 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
202}
203
774ead3a 204static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 205{
4c9fc8ef
AK		 206 return flexpriority_enabled
207 && (vmcs_config.cpu_based_2nd_exec_ctrl &
208 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
f78e0e2e
SY		 209}
210
211static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
212{
213 return ((cpu_has_vmx_virtualize_apic_accesses()) &&
214 (irqchip_in_kernel(kvm)));
215}
216
2384d2b3
SY		 217static inline int cpu_has_vmx_vpid(void)
218{
219 return (vmcs_config.cpu_based_2nd_exec_ctrl &
220 SECONDARY_EXEC_ENABLE_VPID);
221}
222
8b9cf98c 223static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 224{
225 int i;
226
a2fa3e9f
GH		 227 for (i = 0; i < vmx->nmsrs; ++i)
228 if (vmx->guest_msrs[i].index == msr)
a75beee6
ED		 229 return i;
230 return -1;
231}
232
2384d2b3
SY		 233static inline void __invvpid(int ext, u16 vpid, gva_t gva)
234{
235 struct {
236 u64 vpid : 16;
237 u64 rsvd : 48;
238 u64 gva;
239 } operand = { vpid, 0, gva };
240
241 asm volatile (ASM_VMX_INVVPID
242 /* CF==1 or ZF==1 --> rc = -1 */
243 "; ja 1f ; ud2 ; 1:"
244 : : "a"(&operand), "c"(ext) : "cc", "memory");
245}
246
8b9cf98c 247static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 248{
249 int i;
250
8b9cf98c 251 i = __find_msr_index(vmx, msr);
a75beee6 252 if (i >= 0)
a2fa3e9f 253 return &vmx->guest_msrs[i];
8b6d44c7 254 return NULL;
7725f0ba
AK		 255}
256
6aa8b732
AK		 257static void vmcs_clear(struct vmcs *vmcs)
258{
259 u64 phys_addr = __pa(vmcs);
260 u8 error;
261
262 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
263 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
264 : "cc", "memory");
265 if (error)
266 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
267 vmcs, phys_addr);
268}
269
270static void __vcpu_clear(void *arg)
271{
8b9cf98c 272 struct vcpu_vmx *vmx = arg;
d3b2c338 273 int cpu = raw_smp_processor_id();
6aa8b732 274
8b9cf98c 275 if (vmx->vcpu.cpu == cpu)
a2fa3e9f
GH		 276 vmcs_clear(vmx->vmcs);
277 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
6aa8b732 278 per_cpu(current_vmcs, cpu) = NULL;
ad312c7c 279 rdtscll(vmx->vcpu.arch.host_tsc);
6aa8b732
AK		 280}
281
8b9cf98c 282static void vcpu_clear(struct vcpu_vmx *vmx)
8d0be2b3 283{
eae5ecb5
AK		 284 if (vmx->vcpu.cpu == -1)
285 return;
f566e09f 286 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
8b9cf98c 287 vmx->launched = 0;
8d0be2b3
AK		 288}
289
2384d2b3
SY		 290static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
291{
292 if (vmx->vpid == 0)
293 return;
294
295 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
296}
297
6aa8b732
AK		 298static unsigned long vmcs_readl(unsigned long field)
299{
300 unsigned long value;
301
302 asm volatile (ASM_VMX_VMREAD_RDX_RAX
303 : "=a"(value) : "d"(field) : "cc");
304 return value;
305}
306
307static u16 vmcs_read16(unsigned long field)
308{
309 return vmcs_readl(field);
310}
311
312static u32 vmcs_read32(unsigned long field)
313{
314 return vmcs_readl(field);
315}
316
317static u64 vmcs_read64(unsigned long field)
318{
05b3e0c2 319#ifdef CONFIG_X86_64
6aa8b732
AK		 320 return vmcs_readl(field);
321#else
322 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
323#endif
324}
325
e52de1b8
AK		 326static noinline void vmwrite_error(unsigned long field, unsigned long value)
327{
328 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
329 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
330 dump_stack();
331}
332
6aa8b732
AK		 333static void vmcs_writel(unsigned long field, unsigned long value)
334{
335 u8 error;
336
337 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
d77c26fc 338 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 339 if (unlikely(error))
340 vmwrite_error(field, value);
6aa8b732
AK		 341}
342
343static void vmcs_write16(unsigned long field, u16 value)
344{
345 vmcs_writel(field, value);
346}
347
348static void vmcs_write32(unsigned long field, u32 value)
349{
350 vmcs_writel(field, value);
351}
352
353static void vmcs_write64(unsigned long field, u64 value)
354{
05b3e0c2 355#ifdef CONFIG_X86_64
6aa8b732
AK		 356 vmcs_writel(field, value);
357#else
358 vmcs_writel(field, value);
359 asm volatile ("");
360 vmcs_writel(field+1, value >> 32);
361#endif
362}
363
2ab455cc
AL		 364static void vmcs_clear_bits(unsigned long field, u32 mask)
365{
366 vmcs_writel(field, vmcs_readl(field) & ~mask);
367}
368
369static void vmcs_set_bits(unsigned long field, u32 mask)
370{
371 vmcs_writel(field, vmcs_readl(field) | mask);
372}
373
abd3f2d6
AK		 374static void update_exception_bitmap(struct kvm_vcpu *vcpu)
375{
376 u32 eb;
377
7aa81cc0 378 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
abd3f2d6
AK		 379 if (!vcpu->fpu_active)
380 eb |= 1u << NM_VECTOR;
381 if (vcpu->guest_debug.enabled)
382 eb |= 1u << 1;
ad312c7c 383 if (vcpu->arch.rmode.active)
abd3f2d6
AK		 384 eb = ~0;
385 vmcs_write32(EXCEPTION_BITMAP, eb);
386}
387
33ed6329
AK		 388static void reload_tss(void)
389{
33ed6329
AK		 390 /*
391 * VT restores TR but not its size. Useless.
392 */
393 struct descriptor_table gdt;
a5f61300 394 struct desc_struct *descs;
33ed6329
AK		 395
396 get_gdt(&gdt);
397 descs = (void *)gdt.base;
398 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
399 load_TR_desc();
33ed6329
AK		 400}
401
8b9cf98c 402static void load_transition_efer(struct vcpu_vmx *vmx)
2cc51560 403{
a2fa3e9f 404 int efer_offset = vmx->msr_offset_efer;
51c6cf66
AK		 405 u64 host_efer = vmx->host_msrs[efer_offset].data;
406 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
407 u64 ignore_bits;
408
409 if (efer_offset < 0)
410 return;
411 /*
412 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
413 * outside long mode
414 */
415 ignore_bits = EFER_NX | EFER_SCE;
416#ifdef CONFIG_X86_64
417 ignore_bits |= EFER_LMA | EFER_LME;
418 /* SCE is meaningful only in long mode on Intel */
419 if (guest_efer & EFER_LMA)
420 ignore_bits &= ~(u64)EFER_SCE;
421#endif
422 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
423 return;
2cc51560 424
51c6cf66
AK		 425 vmx->host_state.guest_efer_loaded = 1;
426 guest_efer &= ~ignore_bits;
427 guest_efer |= host_efer & ignore_bits;
428 wrmsrl(MSR_EFER, guest_efer);
8b9cf98c 429 vmx->vcpu.stat.efer_reload++;
2cc51560
ED		 430}
431
51c6cf66
AK		 432static void reload_host_efer(struct vcpu_vmx *vmx)
433{
434 if (vmx->host_state.guest_efer_loaded) {
435 vmx->host_state.guest_efer_loaded = 0;
436 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
437 }
438}
439
04d2cc77 440static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 441{
04d2cc77
AK		 442 struct vcpu_vmx *vmx = to_vmx(vcpu);
443
a2fa3e9f 444 if (vmx->host_state.loaded)
33ed6329
AK		 445 return;
446
a2fa3e9f 447 vmx->host_state.loaded = 1;
33ed6329
AK		 448 /*
449 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
450 * allow segment selectors with cpl > 0 or ti == 1.
451 */
a2fa3e9f 452 vmx->host_state.ldt_sel = read_ldt();
152d3f2f 453 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
a2fa3e9f 454 vmx->host_state.fs_sel = read_fs();
152d3f2f 455 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 456 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 457 vmx->host_state.fs_reload_needed = 0;
458 } else {
33ed6329 459 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 460 vmx->host_state.fs_reload_needed = 1;
33ed6329 461 }
a2fa3e9f
GH		 462 vmx->host_state.gs_sel = read_gs();
463 if (!(vmx->host_state.gs_sel & 7))
464 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 465 else {
466 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 467 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 468 }
469
470#ifdef CONFIG_X86_64
471 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
472 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
473#else
a2fa3e9f
GH		 474 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
475 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 476#endif
707c0874
AK		 477
478#ifdef CONFIG_X86_64
d77c26fc 479 if (is_long_mode(&vmx->vcpu))
a2fa3e9f
GH		 480 save_msrs(vmx->host_msrs +
481 vmx->msr_offset_kernel_gs_base, 1);
d77c26fc 482
707c0874 483#endif
a2fa3e9f 484 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
51c6cf66 485 load_transition_efer(vmx);
33ed6329
AK		 486}
487
8b9cf98c 488static void vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 489{
15ad7146 490 unsigned long flags;
33ed6329 491
a2fa3e9f 492 if (!vmx->host_state.loaded)
33ed6329
AK		 493 return;
494
e1beb1d3 495 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 496 vmx->host_state.loaded = 0;
152d3f2f 497 if (vmx->host_state.fs_reload_needed)
a2fa3e9f 498 load_fs(vmx->host_state.fs_sel);
152d3f2f
LV		 499 if (vmx->host_state.gs_ldt_reload_needed) {
500 load_ldt(vmx->host_state.ldt_sel);
33ed6329
AK		 501 /*
502 * If we have to reload gs, we must take care to
503 * preserve our gs base.
504 */
15ad7146 505 local_irq_save(flags);
a2fa3e9f 506 load_gs(vmx->host_state.gs_sel);
33ed6329
AK		 507#ifdef CONFIG_X86_64
508 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
509#endif
15ad7146 510 local_irq_restore(flags);
33ed6329 511 }
152d3f2f 512 reload_tss();
a2fa3e9f
GH		 513 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
514 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
51c6cf66 515 reload_host_efer(vmx);
33ed6329
AK		 516}
517
6aa8b732
AK		 518/*
519 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
520 * vcpu mutex is already taken.
521 */
15ad7146 522static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 523{
a2fa3e9f
GH		 524 struct vcpu_vmx *vmx = to_vmx(vcpu);
525 u64 phys_addr = __pa(vmx->vmcs);
019960ae 526 u64 tsc_this, delta, new_offset;
6aa8b732 527
a3d7f85f 528 if (vcpu->cpu != cpu) {
8b9cf98c 529 vcpu_clear(vmx);
a3d7f85f 530 kvm_migrate_apic_timer(vcpu);
2384d2b3 531 vpid_sync_vcpu_all(vmx);
a3d7f85f 532 }
6aa8b732 533
a2fa3e9f 534 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
6aa8b732
AK		 535 u8 error;
536
a2fa3e9f 537 per_cpu(current_vmcs, cpu) = vmx->vmcs;
6aa8b732
AK		 538 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
539 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
540 : "cc");
541 if (error)
542 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
a2fa3e9f 543 vmx->vmcs, phys_addr);
6aa8b732
AK		 544 }
545
546 if (vcpu->cpu != cpu) {
547 struct descriptor_table dt;
548 unsigned long sysenter_esp;
549
550 vcpu->cpu = cpu;
551 /*
552 * Linux uses per-cpu TSS and GDT, so set these when switching
553 * processors.
554 */
555 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
556 get_gdt(&dt);
557 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
558
559 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
560 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
7700270e
AK		 561
562 /*
563 * Make sure the time stamp counter is monotonous.
564 */
565 rdtscll(tsc_this);
019960ae
AK		 566 if (tsc_this < vcpu->arch.host_tsc) {
567 delta = vcpu->arch.host_tsc - tsc_this;
568 new_offset = vmcs_read64(TSC_OFFSET) + delta;
569 vmcs_write64(TSC_OFFSET, new_offset);
570 }
6aa8b732 571 }
6aa8b732
AK		 572}
573
574static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
575{
8b9cf98c 576 vmx_load_host_state(to_vmx(vcpu));
6aa8b732
AK		 577}
578
5fd86fcf
AK		 579static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
580{
581 if (vcpu->fpu_active)
582 return;
583 vcpu->fpu_active = 1;
707d92fa 584 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
ad312c7c 585 if (vcpu->arch.cr0 & X86_CR0_TS)
707d92fa 586 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 587 update_exception_bitmap(vcpu);
588}
589
590static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
591{
592 if (!vcpu->fpu_active)
593 return;
594 vcpu->fpu_active = 0;
707d92fa 595 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
5fd86fcf
AK		 596 update_exception_bitmap(vcpu);
597}
598
774c47f1
AK		 599static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
600{
8b9cf98c 601 vcpu_clear(to_vmx(vcpu));
774c47f1
AK		 602}
603
6aa8b732
AK		 604static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
605{
606 return vmcs_readl(GUEST_RFLAGS);
607}
608
609static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
610{
ad312c7c 611 if (vcpu->arch.rmode.active)
053de044 612 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 613 vmcs_writel(GUEST_RFLAGS, rflags);
614}
615
616static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
617{
618 unsigned long rip;
619 u32 interruptibility;
620
621 rip = vmcs_readl(GUEST_RIP);
622 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
623 vmcs_writel(GUEST_RIP, rip);
624
625 /*
626 * We emulated an instruction, so temporary interrupt blocking
627 * should be removed, if set.
628 */
629 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
630 if (interruptibility & 3)
631 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
632 interruptibility & ~3);
ad312c7c 633 vcpu->arch.interrupt_window_open = 1;
6aa8b732
AK		 634}
635
298101da
AK		 636static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
637 bool has_error_code, u32 error_code)
638{
639 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
640 nr | INTR_TYPE_EXCEPTION
2e11384c 641 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
298101da
AK		 642 | INTR_INFO_VALID_MASK);
643 if (has_error_code)
644 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
645}
646
647static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
648{
649 struct vcpu_vmx *vmx = to_vmx(vcpu);
650
651 return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
652}
653
a75beee6
ED		 654/*
655 * Swap MSR entry in host/guest MSR entry array.
656 */
54e11fa1 657#ifdef CONFIG_X86_64
8b9cf98c 658static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 659{
a2fa3e9f
GH		 660 struct kvm_msr_entry tmp;
661
662 tmp = vmx->guest_msrs[to];
663 vmx->guest_msrs[to] = vmx->guest_msrs[from];
664 vmx->guest_msrs[from] = tmp;
665 tmp = vmx->host_msrs[to];
666 vmx->host_msrs[to] = vmx->host_msrs[from];
667 vmx->host_msrs[from] = tmp;
a75beee6 668}
54e11fa1 669#endif
a75beee6 670
e38aea3e
AK		 671/*
672 * Set up the vmcs to automatically save and restore system
673 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
674 * mode, as fiddling with msrs is very expensive.
675 */
8b9cf98c 676static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 677{
2cc51560 678 int save_nmsrs;
e38aea3e 679
33f9c505 680 vmx_load_host_state(vmx);
a75beee6
ED		 681 save_nmsrs = 0;
682#ifdef CONFIG_X86_64
8b9cf98c 683 if (is_long_mode(&vmx->vcpu)) {
2cc51560
ED		 684 int index;
685
8b9cf98c 686 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 687 if (index >= 0)
8b9cf98c
RR		 688 move_msr_up(vmx, index, save_nmsrs++);
689 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 690 if (index >= 0)
8b9cf98c
RR		 691 move_msr_up(vmx, index, save_nmsrs++);
692 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 693 if (index >= 0)
8b9cf98c
RR		 694 move_msr_up(vmx, index, save_nmsrs++);
695 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
a75beee6 696 if (index >= 0)
8b9cf98c 697 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 698 /*
699 * MSR_K6_STAR is only needed on long mode guests, and only
700 * if efer.sce is enabled.
701 */
8b9cf98c 702 index = __find_msr_index(vmx, MSR_K6_STAR);
ad312c7c 703 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
8b9cf98c 704 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 705 }
706#endif
a2fa3e9f 707 vmx->save_nmsrs = save_nmsrs;
e38aea3e 708
4d56c8a7 709#ifdef CONFIG_X86_64
a2fa3e9f 710 vmx->msr_offset_kernel_gs_base =
8b9cf98c 711 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
4d56c8a7 712#endif
8b9cf98c 713 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
e38aea3e
AK		 714}
715
6aa8b732
AK		 716/*
717 * reads and returns guest's timestamp counter "register"
718 * guest_tsc = host_tsc + tsc_offset -- 21.3
719 */
720static u64 guest_read_tsc(void)
721{
722 u64 host_tsc, tsc_offset;
723
724 rdtscll(host_tsc);
725 tsc_offset = vmcs_read64(TSC_OFFSET);
726 return host_tsc + tsc_offset;
727}
728
729/*
730 * writes 'guest_tsc' into guest's timestamp counter "register"
731 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
732 */
733static void guest_write_tsc(u64 guest_tsc)
734{
735 u64 host_tsc;
736
737 rdtscll(host_tsc);
738 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
739}
740
6aa8b732
AK		 741/*
742 * Reads an msr value (of 'msr_index') into 'pdata'.
743 * Returns 0 on success, non-0 otherwise.
744 * Assumes vcpu_load() was already called.
745 */
746static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
747{
748 u64 data;
a2fa3e9f 749 struct kvm_msr_entry *msr;
6aa8b732
AK		 750
751 if (!pdata) {
752 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
753 return -EINVAL;
754 }
755
756 switch (msr_index) {
05b3e0c2 757#ifdef CONFIG_X86_64
6aa8b732
AK		 758 case MSR_FS_BASE:
759 data = vmcs_readl(GUEST_FS_BASE);
760 break;
761 case MSR_GS_BASE:
762 data = vmcs_readl(GUEST_GS_BASE);
763 break;
764 case MSR_EFER:
3bab1f5d 765 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 766#endif
767 case MSR_IA32_TIME_STAMP_COUNTER:
768 data = guest_read_tsc();
769 break;
770 case MSR_IA32_SYSENTER_CS:
771 data = vmcs_read32(GUEST_SYSENTER_CS);
772 break;
773 case MSR_IA32_SYSENTER_EIP:
f5b42c33 774 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 775 break;
776 case MSR_IA32_SYSENTER_ESP:
f5b42c33 777 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 778 break;
6aa8b732 779 default:
8b9cf98c 780 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d
AK		 781 if (msr) {
782 data = msr->data;
783 break;
6aa8b732 784 }
3bab1f5d 785 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 786 }
787
788 *pdata = data;
789 return 0;
790}
791
792/*
793 * Writes msr value into into the appropriate "register".
794 * Returns 0 on success, non-0 otherwise.
795 * Assumes vcpu_load() was already called.
796 */
797static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
798{
a2fa3e9f
GH		 799 struct vcpu_vmx *vmx = to_vmx(vcpu);
800 struct kvm_msr_entry *msr;
2cc51560
ED		 801 int ret = 0;
802
6aa8b732 803 switch (msr_index) {
05b3e0c2 804#ifdef CONFIG_X86_64
3bab1f5d 805 case MSR_EFER:
2cc51560 806 ret = kvm_set_msr_common(vcpu, msr_index, data);
51c6cf66
AK		 807 if (vmx->host_state.loaded) {
808 reload_host_efer(vmx);
8b9cf98c 809 load_transition_efer(vmx);
51c6cf66 810 }
2cc51560 811 break;
6aa8b732
AK		 812 case MSR_FS_BASE:
813 vmcs_writel(GUEST_FS_BASE, data);
814 break;
815 case MSR_GS_BASE:
816 vmcs_writel(GUEST_GS_BASE, data);
817 break;
818#endif
819 case MSR_IA32_SYSENTER_CS:
820 vmcs_write32(GUEST_SYSENTER_CS, data);
821 break;
822 case MSR_IA32_SYSENTER_EIP:
f5b42c33 823 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 824 break;
825 case MSR_IA32_SYSENTER_ESP:
f5b42c33 826 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 827 break;
d27d4aca 828 case MSR_IA32_TIME_STAMP_COUNTER:
6aa8b732
AK		 829 guest_write_tsc(data);
830 break;
6aa8b732 831 default:
8b9cf98c 832 msr = find_msr_entry(vmx, msr_index);
3bab1f5d
AK		 833 if (msr) {
834 msr->data = data;
a2fa3e9f
GH		 835 if (vmx->host_state.loaded)
836 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
3bab1f5d 837 break;
6aa8b732 838 }
2cc51560 839 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 840 }
841
2cc51560 842 return ret;
6aa8b732
AK		 843}
844
845/*
846 * Sync the rsp and rip registers into the vcpu structure. This allows
ad312c7c 847 * registers to be accessed by indexing vcpu->arch.regs.
6aa8b732
AK		 848 */
849static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
850{
ad312c7c
ZX		 851 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
852 vcpu->arch.rip = vmcs_readl(GUEST_RIP);
6aa8b732
AK		 853}
854
855/*
856 * Syncs rsp and rip back into the vmcs. Should be called after possible
857 * modification.
858 */
859static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
860{
ad312c7c
ZX		 861 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
862 vmcs_writel(GUEST_RIP, vcpu->arch.rip);
6aa8b732
AK		 863}
864
865static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
866{
867 unsigned long dr7 = 0x400;
6aa8b732
AK		 868 int old_singlestep;
869
6aa8b732
AK		 870 old_singlestep = vcpu->guest_debug.singlestep;
871
872 vcpu->guest_debug.enabled = dbg->enabled;
873 if (vcpu->guest_debug.enabled) {
874 int i;
875
876 dr7 |= 0x200; /* exact */
877 for (i = 0; i < 4; ++i) {
878 if (!dbg->breakpoints[i].enabled)
879 continue;
880 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
881 dr7 |= 2 << (i*2); /* global enable */
882 dr7 |= 0 << (i*4+16); /* execution breakpoint */
883 }
884
6aa8b732 885 vcpu->guest_debug.singlestep = dbg->singlestep;
abd3f2d6 886 } else
6aa8b732 887 vcpu->guest_debug.singlestep = 0;
6aa8b732
AK		 888
889 if (old_singlestep && !vcpu->guest_debug.singlestep) {
890 unsigned long flags;
891
892 flags = vmcs_readl(GUEST_RFLAGS);
893 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
894 vmcs_writel(GUEST_RFLAGS, flags);
895 }
896
abd3f2d6 897 update_exception_bitmap(vcpu);
6aa8b732
AK		 898 vmcs_writel(GUEST_DR7, dr7);
899
900 return 0;
901}
902
2a8067f1
ED		 903static int vmx_get_irq(struct kvm_vcpu *vcpu)
904{
1155f76a 905 struct vcpu_vmx *vmx = to_vmx(vcpu);
2a8067f1
ED		 906 u32 idtv_info_field;
907
1155f76a 908 idtv_info_field = vmx->idt_vectoring_info;
2a8067f1
ED		 909 if (idtv_info_field & INTR_INFO_VALID_MASK) {
910 if (is_external_interrupt(idtv_info_field))
911 return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
912 else
d77c26fc 913 printk(KERN_DEBUG "pending exception: not handled yet\n");
2a8067f1
ED		 914 }
915 return -1;
916}
917
6aa8b732
AK		 918static __init int cpu_has_kvm_support(void)
919{
920 unsigned long ecx = cpuid_ecx(1);
921 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
922}
923
924static __init int vmx_disabled_by_bios(void)
925{
926 u64 msr;
927
928 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
62b3ffb8
YS		 929 return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
930 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
931 == MSR_IA32_FEATURE_CONTROL_LOCKED;
932 /* locked but not enabled */
6aa8b732
AK		 933}
934
774c47f1 935static void hardware_enable(void *garbage)
6aa8b732
AK		 936{
937 int cpu = raw_smp_processor_id();
938 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
939 u64 old;
940
941 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
62b3ffb8
YS		 942 if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
943 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
944 != (MSR_IA32_FEATURE_CONTROL_LOCKED |
945 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
6aa8b732 946 /* enable and lock */
62b3ffb8
YS		 947 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
948 MSR_IA32_FEATURE_CONTROL_LOCKED |
949 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
66aee91a 950 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
6aa8b732
AK		 951 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
952 : "memory", "cc");
953}
954
955static void hardware_disable(void *garbage)
956{
957 asm volatile (ASM_VMX_VMXOFF : : : "cc");
958}
959
1c3d14fe 960static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 961 u32 msr, u32 *result)
1c3d14fe
YS		 962{
963 u32 vmx_msr_low, vmx_msr_high;
964 u32 ctl = ctl_min | ctl_opt;
965
966 rdmsr(msr, vmx_msr_low, vmx_msr_high);
967
968 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
969 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
970
971 /* Ensure minimum (required) set of control bits are supported. */
972 if (ctl_min & ~ctl)
002c7f7c 973 return -EIO;
1c3d14fe
YS		 974
975 *result = ctl;
976 return 0;
977}
978
002c7f7c 979static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 980{
981 u32 vmx_msr_low, vmx_msr_high;
1c3d14fe
YS		 982 u32 min, opt;
983 u32 _pin_based_exec_control = 0;
984 u32 _cpu_based_exec_control = 0;
f78e0e2e 985 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 986 u32 _vmexit_control = 0;
987 u32 _vmentry_control = 0;
988
989 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
990 opt = 0;
991 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
992 &_pin_based_exec_control) < 0)
002c7f7c 993 return -EIO;
1c3d14fe
YS		 994
995 min = CPU_BASED_HLT_EXITING |
996#ifdef CONFIG_X86_64
997 CPU_BASED_CR8_LOAD_EXITING |
998 CPU_BASED_CR8_STORE_EXITING |
999#endif
1000 CPU_BASED_USE_IO_BITMAPS |
1001 CPU_BASED_MOV_DR_EXITING |
1002 CPU_BASED_USE_TSC_OFFSETING;
f78e0e2e
SY		 1003 opt = CPU_BASED_TPR_SHADOW |
1004 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 1005 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1006 &_cpu_based_exec_control) < 0)
002c7f7c 1007 return -EIO;
6e5d865c
YS		 1008#ifdef CONFIG_X86_64
1009 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1010 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1011 ~CPU_BASED_CR8_STORE_EXITING;
1012#endif
f78e0e2e
SY		 1013 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1014 min = 0;
e5edaa01 1015 opt = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3
SY		 1016 SECONDARY_EXEC_WBINVD_EXITING |
1017 SECONDARY_EXEC_ENABLE_VPID;
f78e0e2e
SY		 1018 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS2,
1019 &_cpu_based_2nd_exec_control) < 0)
1020 return -EIO;
1021 }
1022#ifndef CONFIG_X86_64
1023 if (!(_cpu_based_2nd_exec_control &
1024 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1025 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1026#endif
1c3d14fe
YS		 1027
1028 min = 0;
1029#ifdef CONFIG_X86_64
1030 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1031#endif
1032 opt = 0;
1033 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1034 &_vmexit_control) < 0)
002c7f7c 1035 return -EIO;
1c3d14fe
YS		 1036
1037 min = opt = 0;
1038 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1039 &_vmentry_control) < 0)
002c7f7c 1040 return -EIO;
6aa8b732 1041
c68876fd 1042 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 1043
1044 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1045 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 1046 return -EIO;
1c3d14fe
YS		 1047
1048#ifdef CONFIG_X86_64
1049 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1050 if (vmx_msr_high & (1u<<16))
002c7f7c 1051 return -EIO;
1c3d14fe
YS		 1052#endif
1053
1054 /* Require Write-Back (WB) memory type for VMCS accesses. */
1055 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 1056 return -EIO;
1c3d14fe 1057
002c7f7c
YS		 1058 vmcs_conf->size = vmx_msr_high & 0x1fff;
1059 vmcs_conf->order = get_order(vmcs_config.size);
1060 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 1061
002c7f7c
YS		 1062 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1063 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 1064 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 1065 vmcs_conf->vmexit_ctrl = _vmexit_control;
1066 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe
YS		 1067
1068 return 0;
c68876fd 1069}
6aa8b732
AK		 1070
1071static struct vmcs *alloc_vmcs_cpu(int cpu)
1072{
1073 int node = cpu_to_node(cpu);
1074 struct page *pages;
1075 struct vmcs *vmcs;
1076
1c3d14fe 1077 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 1078 if (!pages)
1079 return NULL;
1080 vmcs = page_address(pages);
1c3d14fe
YS		 1081 memset(vmcs, 0, vmcs_config.size);
1082 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 1083 return vmcs;
1084}
1085
1086static struct vmcs *alloc_vmcs(void)
1087{
d3b2c338 1088 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 1089}
1090
1091static void free_vmcs(struct vmcs *vmcs)
1092{
1c3d14fe 1093 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 1094}
1095
39959588 1096static void free_kvm_area(void)
6aa8b732
AK		 1097{
1098 int cpu;
1099
1100 for_each_online_cpu(cpu)
1101 free_vmcs(per_cpu(vmxarea, cpu));
1102}
1103
6aa8b732
AK		 1104static __init int alloc_kvm_area(void)
1105{
1106 int cpu;
1107
1108 for_each_online_cpu(cpu) {
1109 struct vmcs *vmcs;
1110
1111 vmcs = alloc_vmcs_cpu(cpu);
1112 if (!vmcs) {
1113 free_kvm_area();
1114 return -ENOMEM;
1115 }
1116
1117 per_cpu(vmxarea, cpu) = vmcs;
1118 }
1119 return 0;
1120}
1121
1122static __init int hardware_setup(void)
1123{
002c7f7c
YS		 1124 if (setup_vmcs_config(&vmcs_config) < 0)
1125 return -EIO;
50a37eb4
JR		 1126
1127 if (boot_cpu_has(X86_FEATURE_NX))
1128 kvm_enable_efer_bits(EFER_NX);
1129
6aa8b732
AK		 1130 return alloc_kvm_area();
1131}
1132
1133static __exit void hardware_unsetup(void)
1134{
1135 free_kvm_area();
1136}
1137
6aa8b732
AK		 1138static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1139{
1140 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1141
6af11b9e 1142 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
6aa8b732
AK		 1143 vmcs_write16(sf->selector, save->selector);
1144 vmcs_writel(sf->base, save->base);
1145 vmcs_write32(sf->limit, save->limit);
1146 vmcs_write32(sf->ar_bytes, save->ar);
1147 } else {
1148 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1149 << AR_DPL_SHIFT;
1150 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1151 }
1152}
1153
1154static void enter_pmode(struct kvm_vcpu *vcpu)
1155{
1156 unsigned long flags;
1157
ad312c7c 1158 vcpu->arch.rmode.active = 0;
6aa8b732 1159
ad312c7c
ZX		 1160 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1161 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1162 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
6aa8b732
AK		 1163
1164 flags = vmcs_readl(GUEST_RFLAGS);
053de044 1165 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
ad312c7c 1166 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
6aa8b732
AK		 1167 vmcs_writel(GUEST_RFLAGS, flags);
1168
66aee91a
RR		 1169 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1170 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 1171
1172 update_exception_bitmap(vcpu);
1173
ad312c7c
ZX		 1174 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1175 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1176 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1177 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
6aa8b732
AK		 1178
1179 vmcs_write16(GUEST_SS_SELECTOR, 0);
1180 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1181
1182 vmcs_write16(GUEST_CS_SELECTOR,
1183 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1184 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1185}
1186
d77c26fc 1187static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 1188{
bfc6d222 1189 if (!kvm->arch.tss_addr) {
cbc94022
IE		 1190 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1191 kvm->memslots[0].npages - 3;
1192 return base_gfn << PAGE_SHIFT;
1193 }
bfc6d222 1194 return kvm->arch.tss_addr;
6aa8b732
AK		 1195}
1196
1197static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1198{
1199 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1200
1201 save->selector = vmcs_read16(sf->selector);
1202 save->base = vmcs_readl(sf->base);
1203 save->limit = vmcs_read32(sf->limit);
1204 save->ar = vmcs_read32(sf->ar_bytes);
15b00f32
JK		 1205 vmcs_write16(sf->selector, save->base >> 4);
1206 vmcs_write32(sf->base, save->base & 0xfffff);
6aa8b732
AK		 1207 vmcs_write32(sf->limit, 0xffff);
1208 vmcs_write32(sf->ar_bytes, 0xf3);
1209}
1210
1211static void enter_rmode(struct kvm_vcpu *vcpu)
1212{
1213 unsigned long flags;
1214
ad312c7c 1215 vcpu->arch.rmode.active = 1;
6aa8b732 1216
ad312c7c 1217 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
6aa8b732
AK		 1218 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1219
ad312c7c 1220 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
6aa8b732
AK		 1221 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1222
ad312c7c 1223 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
6aa8b732
AK		 1224 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1225
1226 flags = vmcs_readl(GUEST_RFLAGS);
ad312c7c
ZX		 1227 vcpu->arch.rmode.save_iopl
1228 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
6aa8b732 1229
053de044 1230 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 1231
1232 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 1233 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 1234 update_exception_bitmap(vcpu);
1235
1236 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1237 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1238 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1239
1240 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
abacf8df 1241 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
8cb5b033
AK		 1242 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1243 vmcs_writel(GUEST_CS_BASE, 0xf0000);
6aa8b732
AK		 1244 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1245
ad312c7c
ZX		 1246 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1247 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1248 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1249 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
75880a01 1250
8668a3c4 1251 kvm_mmu_reset_context(vcpu);
75880a01 1252 init_rmode_tss(vcpu->kvm);
6aa8b732
AK		 1253}
1254
05b3e0c2 1255#ifdef CONFIG_X86_64
6aa8b732
AK		 1256
1257static void enter_lmode(struct kvm_vcpu *vcpu)
1258{
1259 u32 guest_tr_ar;
1260
1261 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1262 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1263 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
b8688d51 1264 __func__);
6aa8b732
AK		 1265 vmcs_write32(GUEST_TR_AR_BYTES,
1266 (guest_tr_ar & ~AR_TYPE_MASK)
1267 | AR_TYPE_BUSY_64_TSS);
1268 }
1269
ad312c7c 1270 vcpu->arch.shadow_efer |= EFER_LMA;
6aa8b732 1271
8b9cf98c 1272 find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
6aa8b732
AK		 1273 vmcs_write32(VM_ENTRY_CONTROLS,
1274 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1275 | VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1276}
1277
1278static void exit_lmode(struct kvm_vcpu *vcpu)
1279{
ad312c7c 1280 vcpu->arch.shadow_efer &= ~EFER_LMA;
6aa8b732
AK		 1281
1282 vmcs_write32(VM_ENTRY_CONTROLS,
1283 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 1284 & ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1285}
1286
1287#endif
1288
2384d2b3
SY		 1289static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1290{
1291 vpid_sync_vcpu_all(to_vmx(vcpu));
1292}
1293
25c4c276 1294static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 1295{
ad312c7c
ZX		 1296 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1297 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
399badf3
AK		 1298}
1299
6aa8b732
AK		 1300static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1301{
5fd86fcf
AK		 1302 vmx_fpu_deactivate(vcpu);
1303
ad312c7c 1304 if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 1305 enter_pmode(vcpu);
1306
ad312c7c 1307 if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 1308 enter_rmode(vcpu);
1309
05b3e0c2 1310#ifdef CONFIG_X86_64
ad312c7c 1311 if (vcpu->arch.shadow_efer & EFER_LME) {
707d92fa 1312 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 1313 enter_lmode(vcpu);
707d92fa 1314 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 1315 exit_lmode(vcpu);
1316 }
1317#endif
1318
1319 vmcs_writel(CR0_READ_SHADOW, cr0);
1320 vmcs_writel(GUEST_CR0,
1321 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
ad312c7c 1322 vcpu->arch.cr0 = cr0;
5fd86fcf 1323
707d92fa 1324 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
5fd86fcf 1325 vmx_fpu_activate(vcpu);
6aa8b732
AK		 1326}
1327
6aa8b732
AK		 1328static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1329{
2384d2b3 1330 vmx_flush_tlb(vcpu);
6aa8b732 1331 vmcs_writel(GUEST_CR3, cr3);
ad312c7c 1332 if (vcpu->arch.cr0 & X86_CR0_PE)
5fd86fcf 1333 vmx_fpu_deactivate(vcpu);
6aa8b732
AK		 1334}
1335
1336static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1337{
1338 vmcs_writel(CR4_READ_SHADOW, cr4);
ad312c7c 1339 vmcs_writel(GUEST_CR4, cr4 | (vcpu->arch.rmode.active ?
6aa8b732 1340 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
ad312c7c 1341 vcpu->arch.cr4 = cr4;
6aa8b732
AK		 1342}
1343
6aa8b732
AK		 1344static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1345{
8b9cf98c
RR		 1346 struct vcpu_vmx *vmx = to_vmx(vcpu);
1347 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
6aa8b732 1348
ad312c7c 1349 vcpu->arch.shadow_efer = efer;
9f62e19a
JR		 1350 if (!msr)
1351 return;
6aa8b732
AK		 1352 if (efer & EFER_LMA) {
1353 vmcs_write32(VM_ENTRY_CONTROLS,
1354 vmcs_read32(VM_ENTRY_CONTROLS) |
1e4e6e00 1355 VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1356 msr->data = efer;
1357
1358 } else {
1359 vmcs_write32(VM_ENTRY_CONTROLS,
1360 vmcs_read32(VM_ENTRY_CONTROLS) &
1e4e6e00 1361 ~VM_ENTRY_IA32E_MODE);
6aa8b732
AK		 1362
1363 msr->data = efer & ~EFER_LME;
1364 }
8b9cf98c 1365 setup_msrs(vmx);
6aa8b732
AK		 1366}
1367
6aa8b732
AK		 1368static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1369{
1370 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1371
1372 return vmcs_readl(sf->base);
1373}
1374
1375static void vmx_get_segment(struct kvm_vcpu *vcpu,
1376 struct kvm_segment *var, int seg)
1377{
1378 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1379 u32 ar;
1380
1381 var->base = vmcs_readl(sf->base);
1382 var->limit = vmcs_read32(sf->limit);
1383 var->selector = vmcs_read16(sf->selector);
1384 ar = vmcs_read32(sf->ar_bytes);
1385 if (ar & AR_UNUSABLE_MASK)
1386 ar = 0;
1387 var->type = ar & 15;
1388 var->s = (ar >> 4) & 1;
1389 var->dpl = (ar >> 5) & 3;
1390 var->present = (ar >> 7) & 1;
1391 var->avl = (ar >> 12) & 1;
1392 var->l = (ar >> 13) & 1;
1393 var->db = (ar >> 14) & 1;
1394 var->g = (ar >> 15) & 1;
1395 var->unusable = (ar >> 16) & 1;
1396}
1397
653e3108 1398static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 1399{
6aa8b732
AK		 1400 u32 ar;
1401
653e3108 1402 if (var->unusable)
6aa8b732
AK		 1403 ar = 1 << 16;
1404 else {
1405 ar = var->type & 15;
1406 ar |= (var->s & 1) << 4;
1407 ar |= (var->dpl & 3) << 5;
1408 ar |= (var->present & 1) << 7;
1409 ar |= (var->avl & 1) << 12;
1410 ar |= (var->l & 1) << 13;
1411 ar |= (var->db & 1) << 14;
1412 ar |= (var->g & 1) << 15;
1413 }
f7fbf1fd
UL		 1414 if (ar == 0) /* a 0 value means unusable */
1415 ar = AR_UNUSABLE_MASK;
653e3108
AK		 1416
1417 return ar;
1418}
1419
1420static void vmx_set_segment(struct kvm_vcpu *vcpu,
1421 struct kvm_segment *var, int seg)
1422{
1423 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1424 u32 ar;
1425
ad312c7c
ZX		 1426 if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1427 vcpu->arch.rmode.tr.selector = var->selector;
1428 vcpu->arch.rmode.tr.base = var->base;
1429 vcpu->arch.rmode.tr.limit = var->limit;
1430 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
653e3108
AK		 1431 return;
1432 }
1433 vmcs_writel(sf->base, var->base);
1434 vmcs_write32(sf->limit, var->limit);
1435 vmcs_write16(sf->selector, var->selector);
ad312c7c 1436 if (vcpu->arch.rmode.active && var->s) {
653e3108
AK		 1437 /*
1438 * Hack real-mode segments into vm86 compatibility.
1439 */
1440 if (var->base == 0xffff0000 && var->selector == 0xf000)
1441 vmcs_writel(sf->base, 0xf0000);
1442 ar = 0xf3;
1443 } else
1444 ar = vmx_segment_access_rights(var);
6aa8b732
AK		 1445 vmcs_write32(sf->ar_bytes, ar);
1446}
1447
6aa8b732
AK		 1448static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1449{
1450 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1451
1452 *db = (ar >> 14) & 1;
1453 *l = (ar >> 13) & 1;
1454}
1455
1456static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1457{
1458 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1459 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1460}
1461
1462static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1463{
1464 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1465 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1466}
1467
1468static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1469{
1470 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1471 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1472}
1473
1474static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1475{
1476 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1477 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1478}
1479
d77c26fc 1480static int init_rmode_tss(struct kvm *kvm)
6aa8b732 1481{
6aa8b732 1482 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde 1483 u16 data = 0;
10589a46 1484 int ret = 0;
195aefde 1485 int r;
6aa8b732 1486
707a18a5 1487 down_read(&kvm->slots_lock);
195aefde
IE		 1488 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1489 if (r < 0)
10589a46 1490 goto out;
195aefde
IE		 1491 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1492 r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1493 if (r < 0)
10589a46 1494 goto out;
195aefde
IE		 1495 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1496 if (r < 0)
10589a46 1497 goto out;
195aefde
IE		 1498 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1499 if (r < 0)
10589a46 1500 goto out;
195aefde 1501 data = ~0;
10589a46
MT		 1502 r = kvm_write_guest_page(kvm, fn, &data,
1503 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1504 sizeof(u8));
195aefde 1505 if (r < 0)
10589a46
MT		 1506 goto out;
1507
1508 ret = 1;
1509out:
707a18a5 1510 up_read(&kvm->slots_lock);
10589a46 1511 return ret;
6aa8b732
AK		 1512}
1513
6aa8b732
AK		 1514static void seg_setup(int seg)
1515{
1516 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1517
1518 vmcs_write16(sf->selector, 0);
1519 vmcs_writel(sf->base, 0);
1520 vmcs_write32(sf->limit, 0xffff);
1521 vmcs_write32(sf->ar_bytes, 0x93);
1522}
1523
f78e0e2e
SY		 1524static int alloc_apic_access_page(struct kvm *kvm)
1525{
1526 struct kvm_userspace_memory_region kvm_userspace_mem;
1527 int r = 0;
1528
72dc67a6 1529 down_write(&kvm->slots_lock);
bfc6d222 1530 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 1531 goto out;
1532 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
1533 kvm_userspace_mem.flags = 0;
1534 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
1535 kvm_userspace_mem.memory_size = PAGE_SIZE;
1536 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1537 if (r)
1538 goto out;
72dc67a6
IE		 1539
1540 down_read(&current->mm->mmap_sem);
bfc6d222 1541 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
72dc67a6 1542 up_read(&current->mm->mmap_sem);
f78e0e2e 1543out:
72dc67a6 1544 up_write(&kvm->slots_lock);
f78e0e2e
SY		 1545 return r;
1546}
1547
2384d2b3
SY		 1548static void allocate_vpid(struct vcpu_vmx *vmx)
1549{
1550 int vpid;
1551
1552 vmx->vpid = 0;
1553 if (!enable_vpid || !cpu_has_vmx_vpid())
1554 return;
1555 spin_lock(&vmx_vpid_lock);
1556 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
1557 if (vpid < VMX_NR_VPIDS) {
1558 vmx->vpid = vpid;
1559 __set_bit(vpid, vmx_vpid_bitmap);
1560 }
1561 spin_unlock(&vmx_vpid_lock);
1562}
1563
6aa8b732
AK		 1564/*
1565 * Sets up the vmcs for emulated real mode.
1566 */
8b9cf98c 1567static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732
AK		 1568{
1569 u32 host_sysenter_cs;
1570 u32 junk;
1571 unsigned long a;
1572 struct descriptor_table dt;
1573 int i;
cd2276a7 1574 unsigned long kvm_vmx_return;
6e5d865c 1575 u32 exec_control;
6aa8b732 1576
6aa8b732 1577 /* I/O */
fdef3ad1
HQ		 1578 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1579 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
6aa8b732 1580
6aa8b732
AK		 1581 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1582
6aa8b732 1583 /* Control */
1c3d14fe
YS		 1584 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1585 vmcs_config.pin_based_exec_ctrl);
6e5d865c
YS		 1586
1587 exec_control = vmcs_config.cpu_based_exec_ctrl;
1588 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1589 exec_control &= ~CPU_BASED_TPR_SHADOW;
1590#ifdef CONFIG_X86_64
1591 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1592 CPU_BASED_CR8_LOAD_EXITING;
1593#endif
1594 }
1595 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6aa8b732 1596
83ff3b9d
SY		 1597 if (cpu_has_secondary_exec_ctrls()) {
1598 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
1599 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1600 exec_control &=
1601 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2384d2b3
SY		 1602 if (vmx->vpid == 0)
1603 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
83ff3b9d
SY		 1604 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
1605 }
f78e0e2e 1606
c7addb90
AK		 1607 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1608 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
6aa8b732
AK		 1609 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1610
1611 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1612 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1613 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1614
1615 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1616 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1617 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1618 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1619 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1620 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
05b3e0c2 1621#ifdef CONFIG_X86_64
6aa8b732
AK		 1622 rdmsrl(MSR_FS_BASE, a);
1623 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1624 rdmsrl(MSR_GS_BASE, a);
1625 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1626#else
1627 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1628 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1629#endif
1630
1631 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1632
1633 get_idt(&dt);
1634 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1635
d77c26fc 1636 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
cd2276a7 1637 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2cc51560
ED		 1638 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1639 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1640 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
6aa8b732
AK		 1641
1642 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1643 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1644 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1645 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1646 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1647 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1648
6aa8b732
AK		 1649 for (i = 0; i < NR_VMX_MSR; ++i) {
1650 u32 index = vmx_msr_index[i];
1651 u32 data_low, data_high;
1652 u64 data;
a2fa3e9f 1653 int j = vmx->nmsrs;
6aa8b732
AK		 1654
1655 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1656 continue;
432bd6cb
AK		 1657 if (wrmsr_safe(index, data_low, data_high) < 0)
1658 continue;
6aa8b732 1659 data = data_low | ((u64)data_high << 32);
a2fa3e9f
GH		 1660 vmx->host_msrs[j].index = index;
1661 vmx->host_msrs[j].reserved = 0;
1662 vmx->host_msrs[j].data = data;
1663 vmx->guest_msrs[j] = vmx->host_msrs[j];
1664 ++vmx->nmsrs;
6aa8b732 1665 }
6aa8b732 1666
1c3d14fe 1667 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 1668
1669 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 1670 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1671
e00c8cf2
AK		 1672 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1673 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1674
f78e0e2e 1675
e00c8cf2
AK		 1676 return 0;
1677}
1678
1679static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1680{
1681 struct vcpu_vmx *vmx = to_vmx(vcpu);
1682 u64 msr;
1683 int ret;
1684
1685 if (!init_rmode_tss(vmx->vcpu.kvm)) {
1686 ret = -ENOMEM;
1687 goto out;
1688 }
1689
ad312c7c 1690 vmx->vcpu.arch.rmode.active = 0;
e00c8cf2 1691
ad312c7c 1692 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 1693 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2
AK		 1694 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1695 if (vmx->vcpu.vcpu_id == 0)
1696 msr |= MSR_IA32_APICBASE_BSP;
1697 kvm_set_apic_base(&vmx->vcpu, msr);
1698
1699 fx_init(&vmx->vcpu);
1700
1701 /*
1702 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1703 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1704 */
1705 if (vmx->vcpu.vcpu_id == 0) {
1706 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1707 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1708 } else {
ad312c7c
ZX		 1709 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
1710 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2
AK		 1711 }
1712 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1713 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1714
1715 seg_setup(VCPU_SREG_DS);
1716 seg_setup(VCPU_SREG_ES);
1717 seg_setup(VCPU_SREG_FS);
1718 seg_setup(VCPU_SREG_GS);
1719 seg_setup(VCPU_SREG_SS);
1720
1721 vmcs_write16(GUEST_TR_SELECTOR, 0);
1722 vmcs_writel(GUEST_TR_BASE, 0);
1723 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1724 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1725
1726 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1727 vmcs_writel(GUEST_LDTR_BASE, 0);
1728 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1729 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1730
1731 vmcs_write32(GUEST_SYSENTER_CS, 0);
1732 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1733 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1734
1735 vmcs_writel(GUEST_RFLAGS, 0x02);
1736 if (vmx->vcpu.vcpu_id == 0)
1737 vmcs_writel(GUEST_RIP, 0xfff0);
1738 else
1739 vmcs_writel(GUEST_RIP, 0);
1740 vmcs_writel(GUEST_RSP, 0);
1741
1742 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
1743 vmcs_writel(GUEST_DR7, 0x400);
1744
1745 vmcs_writel(GUEST_GDTR_BASE, 0);
1746 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1747
1748 vmcs_writel(GUEST_IDTR_BASE, 0);
1749 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1750
1751 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1752 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1753 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1754
1755 guest_write_tsc(0);
1756
1757 /* Special registers */
1758 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1759
1760 setup_msrs(vmx);
1761
6aa8b732
AK		 1762 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1763
f78e0e2e
SY		 1764 if (cpu_has_vmx_tpr_shadow()) {
1765 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
1766 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
1767 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
ad312c7c 1768 page_to_phys(vmx->vcpu.arch.apic->regs_page));
f78e0e2e
SY		 1769 vmcs_write32(TPR_THRESHOLD, 0);
1770 }
1771
1772 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1773 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 1774 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 1775
2384d2b3
SY		 1776 if (vmx->vpid != 0)
1777 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
1778
ad312c7c
ZX		 1779 vmx->vcpu.arch.cr0 = 0x60000010;
1780 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
8b9cf98c 1781 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 1782 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 1783 vmx_fpu_activate(&vmx->vcpu);
1784 update_exception_bitmap(&vmx->vcpu);
6aa8b732 1785
2384d2b3
SY		 1786 vpid_sync_vcpu_all(vmx);
1787
6aa8b732
AK		 1788 return 0;
1789
6aa8b732
AK		 1790out:
1791 return ret;
1792}
1793
85f455f7
ED		 1794static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
1795{
9c8cba37
AK		 1796 struct vcpu_vmx *vmx = to_vmx(vcpu);
1797
ad312c7c 1798 if (vcpu->arch.rmode.active) {
9c8cba37
AK		 1799 vmx->rmode.irq.pending = true;
1800 vmx->rmode.irq.vector = irq;
1801 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP);
9c5623e3
AK		 1802 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1803 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
1804 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
9c8cba37 1805 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1);
85f455f7
ED		 1806 return;
1807 }
1808 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1809 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1810}
1811
6aa8b732
AK		 1812static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1813{
ad312c7c
ZX		 1814 int word_index = __ffs(vcpu->arch.irq_summary);
1815 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
6aa8b732
AK		 1816 int irq = word_index * BITS_PER_LONG + bit_index;
1817
ad312c7c
ZX		 1818 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1819 if (!vcpu->arch.irq_pending[word_index])
1820 clear_bit(word_index, &vcpu->arch.irq_summary);
85f455f7 1821 vmx_inject_irq(vcpu, irq);
6aa8b732
AK		 1822}
1823
c1150d8c
DL		 1824
1825static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1826 struct kvm_run *kvm_run)
6aa8b732 1827{
c1150d8c
DL		 1828 u32 cpu_based_vm_exec_control;
1829
ad312c7c 1830 vcpu->arch.interrupt_window_open =
c1150d8c
DL		 1831 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1832 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1833
ad312c7c
ZX		 1834 if (vcpu->arch.interrupt_window_open &&
1835 vcpu->arch.irq_summary &&
c1150d8c 1836 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
6aa8b732 1837 /*
c1150d8c 1838 * If interrupts enabled, and not blocked by sti or mov ss. Good.
6aa8b732
AK		 1839 */
1840 kvm_do_inject_irq(vcpu);
c1150d8c
DL		 1841
1842 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
ad312c7c
ZX		 1843 if (!vcpu->arch.interrupt_window_open &&
1844 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
6aa8b732
AK		 1845 /*
1846 * Interrupts blocked. Wait for unblock.
1847 */
c1150d8c
DL		 1848 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1849 else
1850 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1851 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6aa8b732
AK		 1852}
1853
cbc94022
IE		 1854static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
1855{
1856 int ret;
1857 struct kvm_userspace_memory_region tss_mem = {
1858 .slot = 8,
1859 .guest_phys_addr = addr,
1860 .memory_size = PAGE_SIZE * 3,
1861 .flags = 0,
1862 };
1863
1864 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
1865 if (ret)
1866 return ret;
bfc6d222 1867 kvm->arch.tss_addr = addr;
cbc94022
IE		 1868 return 0;
1869}
1870
6aa8b732
AK		 1871static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1872{
1873 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1874
1875 set_debugreg(dbg->bp[0], 0);
1876 set_debugreg(dbg->bp[1], 1);
1877 set_debugreg(dbg->bp[2], 2);
1878 set_debugreg(dbg->bp[3], 3);
1879
1880 if (dbg->singlestep) {
1881 unsigned long flags;
1882
1883 flags = vmcs_readl(GUEST_RFLAGS);
1884 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1885 vmcs_writel(GUEST_RFLAGS, flags);
1886 }
1887}
1888
1889static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1890 int vec, u32 err_code)
1891{
ad312c7c 1892 if (!vcpu->arch.rmode.active)
6aa8b732
AK		 1893 return 0;
1894
b3f37707
NK		 1895 /*
1896 * Instruction with address size override prefix opcode 0x67
1897 * Cause the #SS fault with 0 error code in VM86 mode.
1898 */
1899 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3427318f 1900 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
6aa8b732
AK		 1901 return 1;
1902 return 0;
1903}
1904
1905static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1906{
1155f76a 1907 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732
AK		 1908 u32 intr_info, error_code;
1909 unsigned long cr2, rip;
1910 u32 vect_info;
1911 enum emulation_result er;
1912
1155f76a 1913 vect_info = vmx->idt_vectoring_info;
6aa8b732
AK		 1914 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1915
1916 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
d77c26fc 1917 !is_page_fault(intr_info))
6aa8b732 1918 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
b8688d51 1919 "intr info 0x%x\n", __func__, vect_info, intr_info);
6aa8b732 1920
85f455f7 1921 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
6aa8b732 1922 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
ad312c7c
ZX		 1923 set_bit(irq, vcpu->arch.irq_pending);
1924 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
6aa8b732
AK		 1925 }
1926
1b6269db
AK		 1927 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
1928 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 1929
1930 if (is_no_device(intr_info)) {
5fd86fcf 1931 vmx_fpu_activate(vcpu);
2ab455cc
AL		 1932 return 1;
1933 }
1934
7aa81cc0 1935 if (is_invalid_opcode(intr_info)) {
571008da 1936 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
7aa81cc0 1937 if (er != EMULATE_DONE)
7ee5d940 1938 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 1939 return 1;
1940 }
1941
6aa8b732
AK		 1942 error_code = 0;
1943 rip = vmcs_readl(GUEST_RIP);
2e11384c 1944 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 1945 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1946 if (is_page_fault(intr_info)) {
1947 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3067714c 1948 return kvm_mmu_page_fault(vcpu, cr2, error_code);
6aa8b732
AK		 1949 }
1950
ad312c7c 1951 if (vcpu->arch.rmode.active &&
6aa8b732 1952 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 1953 error_code)) {
ad312c7c
ZX		 1954 if (vcpu->arch.halt_request) {
1955 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 1956 return kvm_emulate_halt(vcpu);
1957 }
6aa8b732 1958 return 1;
72d6e5a0 1959 }
6aa8b732 1960
d77c26fc
MD		 1961 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
1962 (INTR_TYPE_EXCEPTION | 1)) {
6aa8b732
AK		 1963 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1964 return 0;
1965 }
1966 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1967 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1968 kvm_run->ex.error_code = error_code;
1969 return 0;
1970}
1971
1972static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1973 struct kvm_run *kvm_run)
1974{
1165f5fe 1975 ++vcpu->stat.irq_exits;
6aa8b732
AK		 1976 return 1;
1977}
1978
988ad74f
AK		 1979static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1980{
1981 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1982 return 0;
1983}
6aa8b732 1984
6aa8b732
AK		 1985static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1986{
bfdaab09 1987 unsigned long exit_qualification;
039576c0
AK		 1988 int size, down, in, string, rep;
1989 unsigned port;
6aa8b732 1990
1165f5fe 1991 ++vcpu->stat.io_exits;
bfdaab09 1992 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 1993 string = (exit_qualification & 16) != 0;
e70669ab
LV		 1994
1995 if (string) {
3427318f
LV		 1996 if (emulate_instruction(vcpu,
1997 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
e70669ab
LV		 1998 return 0;
1999 return 1;
2000 }
2001
2002 size = (exit_qualification & 7) + 1;
2003 in = (exit_qualification & 8) != 0;
039576c0 2004 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
039576c0
AK		 2005 rep = (exit_qualification & 32) != 0;
2006 port = exit_qualification >> 16;
e70669ab 2007
3090dd73 2008 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
6aa8b732
AK		 2009}
2010
102d8325
IM		 2011static void
2012vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2013{
2014 /*
2015 * Patch in the VMCALL instruction:
2016 */
2017 hypercall[0] = 0x0f;
2018 hypercall[1] = 0x01;
2019 hypercall[2] = 0xc1;
102d8325
IM		 2020}
2021
6aa8b732
AK		 2022static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2023{
bfdaab09 2024 unsigned long exit_qualification;
6aa8b732
AK		 2025 int cr;
2026 int reg;
2027
bfdaab09 2028 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2029 cr = exit_qualification & 15;
2030 reg = (exit_qualification >> 8) & 15;
2031 switch ((exit_qualification >> 4) & 3) {
2032 case 0: /* mov to cr */
2033 switch (cr) {
2034 case 0:
2035 vcpu_load_rsp_rip(vcpu);
2d3ad1f4 2036 kvm_set_cr0(vcpu, vcpu->arch.regs[reg]);
6aa8b732
AK		 2037 skip_emulated_instruction(vcpu);
2038 return 1;
2039 case 3:
2040 vcpu_load_rsp_rip(vcpu);
2d3ad1f4 2041 kvm_set_cr3(vcpu, vcpu->arch.regs[reg]);
6aa8b732
AK		 2042 skip_emulated_instruction(vcpu);
2043 return 1;
2044 case 4:
2045 vcpu_load_rsp_rip(vcpu);
2d3ad1f4 2046 kvm_set_cr4(vcpu, vcpu->arch.regs[reg]);
6aa8b732
AK		 2047 skip_emulated_instruction(vcpu);
2048 return 1;
2049 case 8:
2050 vcpu_load_rsp_rip(vcpu);
2d3ad1f4 2051 kvm_set_cr8(vcpu, vcpu->arch.regs[reg]);
6aa8b732 2052 skip_emulated_instruction(vcpu);
e5314067
AK		 2053 if (irqchip_in_kernel(vcpu->kvm))
2054 return 1;
253abdee
YS		 2055 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2056 return 0;
6aa8b732
AK		 2057 };
2058 break;
25c4c276
AL		 2059 case 2: /* clts */
2060 vcpu_load_rsp_rip(vcpu);
5fd86fcf 2061 vmx_fpu_deactivate(vcpu);
ad312c7c
ZX		 2062 vcpu->arch.cr0 &= ~X86_CR0_TS;
2063 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf 2064 vmx_fpu_activate(vcpu);
25c4c276
AL		 2065 skip_emulated_instruction(vcpu);
2066 return 1;
6aa8b732
AK		 2067 case 1: /*mov from cr*/
2068 switch (cr) {
2069 case 3:
2070 vcpu_load_rsp_rip(vcpu);
ad312c7c 2071 vcpu->arch.regs[reg] = vcpu->arch.cr3;
6aa8b732
AK		 2072 vcpu_put_rsp_rip(vcpu);
2073 skip_emulated_instruction(vcpu);
2074 return 1;
2075 case 8:
6aa8b732 2076 vcpu_load_rsp_rip(vcpu);
2d3ad1f4 2077 vcpu->arch.regs[reg] = kvm_get_cr8(vcpu);
6aa8b732
AK		 2078 vcpu_put_rsp_rip(vcpu);
2079 skip_emulated_instruction(vcpu);
2080 return 1;
2081 }
2082 break;
2083 case 3: /* lmsw */
2d3ad1f4 2084 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
6aa8b732
AK		 2085
2086 skip_emulated_instruction(vcpu);
2087 return 1;
2088 default:
2089 break;
2090 }
2091 kvm_run->exit_reason = 0;
f0242478 2092 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 2093 (int)(exit_qualification >> 4) & 3, cr);
2094 return 0;
2095}
2096
2097static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2098{
bfdaab09 2099 unsigned long exit_qualification;
6aa8b732
AK		 2100 unsigned long val;
2101 int dr, reg;
2102
2103 /*
2104 * FIXME: this code assumes the host is debugging the guest.
2105 * need to deal with guest debugging itself too.
2106 */
bfdaab09 2107 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 2108 dr = exit_qualification & 7;
2109 reg = (exit_qualification >> 8) & 15;
2110 vcpu_load_rsp_rip(vcpu);
2111 if (exit_qualification & 16) {
2112 /* mov from dr */
2113 switch (dr) {
2114 case 6:
2115 val = 0xffff0ff0;
2116 break;
2117 case 7:
2118 val = 0x400;
2119 break;
2120 default:
2121 val = 0;
2122 }
ad312c7c 2123 vcpu->arch.regs[reg] = val;
6aa8b732
AK		 2124 } else {
2125 /* mov to dr */
2126 }
2127 vcpu_put_rsp_rip(vcpu);
2128 skip_emulated_instruction(vcpu);
2129 return 1;
2130}
2131
2132static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2133{
06465c5a
AK		 2134 kvm_emulate_cpuid(vcpu);
2135 return 1;
6aa8b732
AK		 2136}
2137
2138static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2139{
ad312c7c 2140 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 2141 u64 data;
2142
2143 if (vmx_get_msr(vcpu, ecx, &data)) {
c1a5d4f9 2144 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2145 return 1;
2146 }
2147
2148 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 2149 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2150 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 2151 skip_emulated_instruction(vcpu);
2152 return 1;
2153}
2154
2155static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2156{
ad312c7c
ZX		 2157 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2158 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2159 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732
AK		 2160
2161 if (vmx_set_msr(vcpu, ecx, data) != 0) {
c1a5d4f9 2162 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 2163 return 1;
2164 }
2165
2166 skip_emulated_instruction(vcpu);
2167 return 1;
2168}
2169
6e5d865c
YS		 2170static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2171 struct kvm_run *kvm_run)
2172{
2173 return 1;
2174}
2175
6aa8b732
AK		 2176static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2177 struct kvm_run *kvm_run)
2178{
85f455f7
ED		 2179 u32 cpu_based_vm_exec_control;
2180
2181 /* clear pending irq */
2182 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2183 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2184 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
c1150d8c
DL		 2185 /*
2186 * If the user space waits to inject interrupts, exit as soon as
2187 * possible
2188 */
2189 if (kvm_run->request_interrupt_window &&
ad312c7c 2190 !vcpu->arch.irq_summary) {
c1150d8c 2191 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1165f5fe 2192 ++vcpu->stat.irq_window_exits;
c1150d8c
DL		 2193 return 0;
2194 }
6aa8b732
AK		 2195 return 1;
2196}
2197
2198static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2199{
2200 skip_emulated_instruction(vcpu);
d3bef15f 2201 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 2202}
2203
c21415e8
IM		 2204static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2205{
510043da 2206 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 2207 kvm_emulate_hypercall(vcpu);
2208 return 1;
c21415e8
IM		 2209}
2210
e5edaa01
ED		 2211static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2212{
2213 skip_emulated_instruction(vcpu);
2214 /* TODO: Add support for VT-d/pass-through device */
2215 return 1;
2216}
2217
f78e0e2e
SY		 2218static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2219{
2220 u64 exit_qualification;
2221 enum emulation_result er;
2222 unsigned long offset;
2223
2224 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2225 offset = exit_qualification & 0xffful;
2226
2227 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2228
2229 if (er != EMULATE_DONE) {
2230 printk(KERN_ERR
2231 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2232 offset);
2233 return -ENOTSUPP;
2234 }
2235 return 1;
2236}
2237
6aa8b732
AK		 2238/*
2239 * The exit handlers return 1 if the exit was handled fully and guest execution
2240 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
2241 * to be done to userspace and return 0.
2242 */
2243static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2244 struct kvm_run *kvm_run) = {
2245 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
2246 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 2247 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6aa8b732 2248 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 2249 [EXIT_REASON_CR_ACCESS] = handle_cr,
2250 [EXIT_REASON_DR_ACCESS] = handle_dr,
2251 [EXIT_REASON_CPUID] = handle_cpuid,
2252 [EXIT_REASON_MSR_READ] = handle_rdmsr,
2253 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
2254 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
2255 [EXIT_REASON_HLT] = handle_halt,
c21415e8 2256 [EXIT_REASON_VMCALL] = handle_vmcall,
f78e0e2e
SY		 2257 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
2258 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 2259 [EXIT_REASON_WBINVD] = handle_wbinvd,
6aa8b732
AK		 2260};
2261
2262static const int kvm_vmx_max_exit_handlers =
50a3485c 2263 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732
AK		 2264
2265/*
2266 * The guest has exited. See if we can fix it or if we need userspace
2267 * assistance.
2268 */
2269static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2270{
6aa8b732 2271 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
29bd8a78 2272 struct vcpu_vmx *vmx = to_vmx(vcpu);
1155f76a 2273 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78
AK		 2274
2275 if (unlikely(vmx->fail)) {
2276 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2277 kvm_run->fail_entry.hardware_entry_failure_reason
2278 = vmcs_read32(VM_INSTRUCTION_ERROR);
2279 return 0;
2280 }
6aa8b732 2281
d77c26fc
MD		 2282 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2283 exit_reason != EXIT_REASON_EXCEPTION_NMI)
6aa8b732 2284 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
b8688d51 2285 "exit reason is 0x%x\n", __func__, exit_reason);
6aa8b732
AK		 2286 if (exit_reason < kvm_vmx_max_exit_handlers
2287 && kvm_vmx_exit_handlers[exit_reason])
2288 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2289 else {
2290 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2291 kvm_run->hw.hardware_exit_reason = exit_reason;
2292 }
2293 return 0;
2294}
2295
6e5d865c
YS		 2296static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2297{
2298 int max_irr, tpr;
2299
2300 if (!vm_need_tpr_shadow(vcpu->kvm))
2301 return;
2302
2303 if (!kvm_lapic_enabled(vcpu) ||
2304 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2305 vmcs_write32(TPR_THRESHOLD, 0);
2306 return;
2307 }
2308
2309 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2310 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2311}
2312
85f455f7
ED		 2313static void enable_irq_window(struct kvm_vcpu *vcpu)
2314{
2315 u32 cpu_based_vm_exec_control;
2316
2317 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2318 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2319 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2320}
2321
2322static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2323{
1155f76a 2324 struct vcpu_vmx *vmx = to_vmx(vcpu);
85f455f7
ED		 2325 u32 idtv_info_field, intr_info_field;
2326 int has_ext_irq, interrupt_window_open;
1b9778da 2327 int vector;
85f455f7 2328
6e5d865c
YS		 2329 update_tpr_threshold(vcpu);
2330
85f455f7
ED		 2331 has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2332 intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
1155f76a 2333 idtv_info_field = vmx->idt_vectoring_info;
85f455f7
ED		 2334 if (intr_info_field & INTR_INFO_VALID_MASK) {
2335 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2336 /* TODO: fault when IDT_Vectoring */
9584bf2c
RH		 2337 if (printk_ratelimit())
2338 printk(KERN_ERR "Fault when IDT_Vectoring\n");
85f455f7
ED		 2339 }
2340 if (has_ext_irq)
2341 enable_irq_window(vcpu);
2342 return;
2343 }
2344 if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
9c8cba37
AK		 2345 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2346 == INTR_TYPE_EXT_INTR
ad312c7c 2347 && vcpu->arch.rmode.active) {
9c8cba37
AK		 2348 u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2349
2350 vmx_inject_irq(vcpu, vect);
2351 if (unlikely(has_ext_irq))
2352 enable_irq_window(vcpu);
2353 return;
2354 }
2355
85f455f7
ED		 2356 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2357 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2358 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2359
2e11384c 2360 if (unlikely(idtv_info_field & INTR_INFO_DELIVER_CODE_MASK))
85f455f7
ED		 2361 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2362 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2363 if (unlikely(has_ext_irq))
2364 enable_irq_window(vcpu);
2365 return;
2366 }
2367 if (!has_ext_irq)
2368 return;
2369 interrupt_window_open =
2370 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2371 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1b9778da
ED		 2372 if (interrupt_window_open) {
2373 vector = kvm_cpu_get_interrupt(vcpu);
2374 vmx_inject_irq(vcpu, vector);
2375 kvm_timer_intr_post(vcpu, vector);
2376 } else
85f455f7
ED		 2377 enable_irq_window(vcpu);
2378}
2379
9c8cba37
AK		 2380/*
2381 * Failure to inject an interrupt should give us the information
2382 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
2383 * when fetching the interrupt redirection bitmap in the real-mode
2384 * tss, this doesn't happen. So we do it ourselves.
2385 */
2386static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2387{
2388 vmx->rmode.irq.pending = 0;
2389 if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip)
2390 return;
2391 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip);
2392 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2393 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2394 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
2395 return;
2396 }
2397 vmx->idt_vectoring_info =
2398 VECTORING_INFO_VALID_MASK
2399 | INTR_TYPE_EXT_INTR
2400 | vmx->rmode.irq.vector;
2401}
2402
04d2cc77 2403static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 2404{
a2fa3e9f 2405 struct vcpu_vmx *vmx = to_vmx(vcpu);
1b6269db 2406 u32 intr_info;
e6adf283
AK		 2407
2408 /*
2409 * Loading guest fpu may have cleared host cr0.ts
2410 */
2411 vmcs_writel(HOST_CR0, read_cr0());
2412
d77c26fc 2413 asm(
6aa8b732 2414 /* Store host registers */
05b3e0c2 2415#ifdef CONFIG_X86_64
c2036300 2416 "push %%rdx; push %%rbp;"
6aa8b732 2417 "push %%rcx \n\t"
6aa8b732 2418#else
ff593e5a
LV		 2419 "push %%edx; push %%ebp;"
2420 "push %%ecx \n\t"
6aa8b732 2421#endif
c2036300 2422 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
6aa8b732 2423 /* Check if vmlaunch of vmresume is needed */
e08aa78a 2424 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 2425 /* Load guest registers. Don't clobber flags. */
05b3e0c2 2426#ifdef CONFIG_X86_64
e08aa78a 2427 "mov %c[cr2](%0), %%rax \n\t"
6aa8b732 2428 "mov %%rax, %%cr2 \n\t"
e08aa78a
AK		 2429 "mov %c[rax](%0), %%rax \n\t"
2430 "mov %c[rbx](%0), %%rbx \n\t"
2431 "mov %c[rdx](%0), %%rdx \n\t"
2432 "mov %c[rsi](%0), %%rsi \n\t"
2433 "mov %c[rdi](%0), %%rdi \n\t"
2434 "mov %c[rbp](%0), %%rbp \n\t"
2435 "mov %c[r8](%0), %%r8 \n\t"
2436 "mov %c[r9](%0), %%r9 \n\t"
2437 "mov %c[r10](%0), %%r10 \n\t"
2438 "mov %c[r11](%0), %%r11 \n\t"
2439 "mov %c[r12](%0), %%r12 \n\t"
2440 "mov %c[r13](%0), %%r13 \n\t"
2441 "mov %c[r14](%0), %%r14 \n\t"
2442 "mov %c[r15](%0), %%r15 \n\t"
2443 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
6aa8b732 2444#else
e08aa78a 2445 "mov %c[cr2](%0), %%eax \n\t"
6aa8b732 2446 "mov %%eax, %%cr2 \n\t"
e08aa78a
AK		 2447 "mov %c[rax](%0), %%eax \n\t"
2448 "mov %c[rbx](%0), %%ebx \n\t"
2449 "mov %c[rdx](%0), %%edx \n\t"
2450 "mov %c[rsi](%0), %%esi \n\t"
2451 "mov %c[rdi](%0), %%edi \n\t"
2452 "mov %c[rbp](%0), %%ebp \n\t"
2453 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
6aa8b732
AK		 2454#endif
2455 /* Enter guest mode */
cd2276a7 2456 "jne .Llaunched \n\t"
6aa8b732 2457 ASM_VMX_VMLAUNCH "\n\t"
cd2276a7
AK		 2458 "jmp .Lkvm_vmx_return \n\t"
2459 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2460 ".Lkvm_vmx_return: "
6aa8b732 2461 /* Save guest registers, load host registers, keep flags */
05b3e0c2 2462#ifdef CONFIG_X86_64
e08aa78a
AK		 2463 "xchg %0, (%%rsp) \n\t"
2464 "mov %%rax, %c[rax](%0) \n\t"
2465 "mov %%rbx, %c[rbx](%0) \n\t"
2466 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
2467 "mov %%rdx, %c[rdx](%0) \n\t"
2468 "mov %%rsi, %c[rsi](%0) \n\t"
2469 "mov %%rdi, %c[rdi](%0) \n\t"
2470 "mov %%rbp, %c[rbp](%0) \n\t"
2471 "mov %%r8, %c[r8](%0) \n\t"
2472 "mov %%r9, %c[r9](%0) \n\t"
2473 "mov %%r10, %c[r10](%0) \n\t"
2474 "mov %%r11, %c[r11](%0) \n\t"
2475 "mov %%r12, %c[r12](%0) \n\t"
2476 "mov %%r13, %c[r13](%0) \n\t"
2477 "mov %%r14, %c[r14](%0) \n\t"
2478 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 2479 "mov %%cr2, %%rax \n\t"
e08aa78a 2480 "mov %%rax, %c[cr2](%0) \n\t"
6aa8b732 2481
e08aa78a 2482 "pop %%rbp; pop %%rbp; pop %%rdx \n\t"
6aa8b732 2483#else
e08aa78a
AK		 2484 "xchg %0, (%%esp) \n\t"
2485 "mov %%eax, %c[rax](%0) \n\t"
2486 "mov %%ebx, %c[rbx](%0) \n\t"
2487 "pushl (%%esp); popl %c[rcx](%0) \n\t"
2488 "mov %%edx, %c[rdx](%0) \n\t"
2489 "mov %%esi, %c[rsi](%0) \n\t"
2490 "mov %%edi, %c[rdi](%0) \n\t"
2491 "mov %%ebp, %c[rbp](%0) \n\t"
6aa8b732 2492 "mov %%cr2, %%eax \n\t"
e08aa78a 2493 "mov %%eax, %c[cr2](%0) \n\t"
6aa8b732 2494
e08aa78a 2495 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
6aa8b732 2496#endif
e08aa78a
AK		 2497 "setbe %c[fail](%0) \n\t"
2498 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
2499 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
2500 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
ad312c7c
ZX		 2501 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
2502 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
2503 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
2504 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
2505 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
2506 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
2507 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 2508#ifdef CONFIG_X86_64
ad312c7c
ZX		 2509 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
2510 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
2511 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
2512 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
2513 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
2514 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
2515 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
2516 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 2517#endif
ad312c7c 2518 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
c2036300
LV		 2519 : "cc", "memory"
2520#ifdef CONFIG_X86_64
2521 , "rbx", "rdi", "rsi"
2522 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
ff593e5a
LV		 2523#else
2524 , "ebx", "edi", "rsi"
c2036300
LV		 2525#endif
2526 );
6aa8b732 2527
1155f76a 2528 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9c8cba37
AK		 2529 if (vmx->rmode.irq.pending)
2530 fixup_rmode_irq(vmx);
1155f76a 2531
ad312c7c 2532 vcpu->arch.interrupt_window_open =
d77c26fc 2533 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
6aa8b732 2534
d77c26fc 2535 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
15ad7146 2536 vmx->launched = 1;
1b6269db
AK		 2537
2538 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2539
2540 /* We need to handle NMIs before interrupts are enabled */
2541 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2542 asm("int $2");
6aa8b732
AK		 2543}
2544
6aa8b732
AK		 2545static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2546{
a2fa3e9f
GH		 2547 struct vcpu_vmx *vmx = to_vmx(vcpu);
2548
2549 if (vmx->vmcs) {
8b9cf98c 2550 on_each_cpu(__vcpu_clear, vmx, 0, 1);
a2fa3e9f
GH		 2551 free_vmcs(vmx->vmcs);
2552 vmx->vmcs = NULL;
6aa8b732
AK		 2553 }
2554}
2555
2556static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2557{
fb3f0f51
RR		 2558 struct vcpu_vmx *vmx = to_vmx(vcpu);
2559
2384d2b3
SY		 2560 spin_lock(&vmx_vpid_lock);
2561 if (vmx->vpid != 0)
2562 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2563 spin_unlock(&vmx_vpid_lock);
6aa8b732 2564 vmx_free_vmcs(vcpu);
fb3f0f51
RR		 2565 kfree(vmx->host_msrs);
2566 kfree(vmx->guest_msrs);
2567 kvm_vcpu_uninit(vcpu);
a4770347 2568 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 2569}
2570
fb3f0f51 2571static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 2572{
fb3f0f51 2573 int err;
c16f862d 2574 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 2575 int cpu;
6aa8b732 2576
a2fa3e9f 2577 if (!vmx)
fb3f0f51
RR		 2578 return ERR_PTR(-ENOMEM);
2579
2384d2b3
SY		 2580 allocate_vpid(vmx);
2581
fb3f0f51
RR		 2582 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
2583 if (err)
2584 goto free_vcpu;
965b58a5 2585
a2fa3e9f 2586 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
fb3f0f51
RR		 2587 if (!vmx->guest_msrs) {
2588 err = -ENOMEM;
2589 goto uninit_vcpu;
2590 }
965b58a5 2591
a2fa3e9f
GH		 2592 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2593 if (!vmx->host_msrs)
fb3f0f51 2594 goto free_guest_msrs;
965b58a5 2595
a2fa3e9f
GH		 2596 vmx->vmcs = alloc_vmcs();
2597 if (!vmx->vmcs)
fb3f0f51 2598 goto free_msrs;
a2fa3e9f
GH		 2599
2600 vmcs_clear(vmx->vmcs);
2601
15ad7146
AK		 2602 cpu = get_cpu();
2603 vmx_vcpu_load(&vmx->vcpu, cpu);
8b9cf98c 2604 err = vmx_vcpu_setup(vmx);
fb3f0f51 2605 vmx_vcpu_put(&vmx->vcpu);
15ad7146 2606 put_cpu();
fb3f0f51
RR		 2607 if (err)
2608 goto free_vmcs;
5e4a0b3c
MT		 2609 if (vm_need_virtualize_apic_accesses(kvm))
2610 if (alloc_apic_access_page(kvm) != 0)
2611 goto free_vmcs;
fb3f0f51
RR		 2612
2613 return &vmx->vcpu;
2614
2615free_vmcs:
2616 free_vmcs(vmx->vmcs);
2617free_msrs:
2618 kfree(vmx->host_msrs);
2619free_guest_msrs:
2620 kfree(vmx->guest_msrs);
2621uninit_vcpu:
2622 kvm_vcpu_uninit(&vmx->vcpu);
2623free_vcpu:
a4770347 2624 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 2625 return ERR_PTR(err);
6aa8b732
AK		 2626}
2627
002c7f7c
YS		 2628static void __init vmx_check_processor_compat(void *rtn)
2629{
2630 struct vmcs_config vmcs_conf;
2631
2632 *(int *)rtn = 0;
2633 if (setup_vmcs_config(&vmcs_conf) < 0)
2634 *(int *)rtn = -EIO;
2635 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
2636 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
2637 smp_processor_id());
2638 *(int *)rtn = -EIO;
2639 }
2640}
2641
cbdd1bea 2642static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 2643 .cpu_has_kvm_support = cpu_has_kvm_support,
2644 .disabled_by_bios = vmx_disabled_by_bios,
2645 .hardware_setup = hardware_setup,
2646 .hardware_unsetup = hardware_unsetup,
002c7f7c 2647 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 2648 .hardware_enable = hardware_enable,
2649 .hardware_disable = hardware_disable,
774ead3a 2650 .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
6aa8b732
AK		 2651
2652 .vcpu_create = vmx_create_vcpu,
2653 .vcpu_free = vmx_free_vcpu,
04d2cc77 2654 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 2655
04d2cc77 2656 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 2657 .vcpu_load = vmx_vcpu_load,
2658 .vcpu_put = vmx_vcpu_put,
774c47f1 2659 .vcpu_decache = vmx_vcpu_decache,
6aa8b732
AK		 2660
2661 .set_guest_debug = set_guest_debug,
04d2cc77 2662 .guest_debug_pre = kvm_guest_debug_pre,
6aa8b732
AK		 2663 .get_msr = vmx_get_msr,
2664 .set_msr = vmx_set_msr,
2665 .get_segment_base = vmx_get_segment_base,
2666 .get_segment = vmx_get_segment,
2667 .set_segment = vmx_set_segment,
6aa8b732 2668 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
25c4c276 2669 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 2670 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 2671 .set_cr3 = vmx_set_cr3,
2672 .set_cr4 = vmx_set_cr4,
6aa8b732 2673 .set_efer = vmx_set_efer,
6aa8b732
AK		 2674 .get_idt = vmx_get_idt,
2675 .set_idt = vmx_set_idt,
2676 .get_gdt = vmx_get_gdt,
2677 .set_gdt = vmx_set_gdt,
2678 .cache_regs = vcpu_load_rsp_rip,
2679 .decache_regs = vcpu_put_rsp_rip,
2680 .get_rflags = vmx_get_rflags,
2681 .set_rflags = vmx_set_rflags,
2682
2683 .tlb_flush = vmx_flush_tlb,
6aa8b732 2684
6aa8b732 2685 .run = vmx_vcpu_run,
04d2cc77 2686 .handle_exit = kvm_handle_exit,
6aa8b732 2687 .skip_emulated_instruction = skip_emulated_instruction,
102d8325 2688 .patch_hypercall = vmx_patch_hypercall,
2a8067f1
ED		 2689 .get_irq = vmx_get_irq,
2690 .set_irq = vmx_inject_irq,
298101da
AK		 2691 .queue_exception = vmx_queue_exception,
2692 .exception_injected = vmx_exception_injected,
04d2cc77
AK		 2693 .inject_pending_irq = vmx_intr_assist,
2694 .inject_pending_vectors = do_interrupt_requests,
cbc94022
IE		 2695
2696 .set_tss_addr = vmx_set_tss_addr,
6aa8b732
AK		 2697};
2698
2699static int __init vmx_init(void)
2700{
fdef3ad1
HQ		 2701 void *iova;
2702 int r;
2703
2704 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2705 if (!vmx_io_bitmap_a)
2706 return -ENOMEM;
2707
2708 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2709 if (!vmx_io_bitmap_b) {
2710 r = -ENOMEM;
2711 goto out;
2712 }
2713
2714 /*
2715 * Allow direct access to the PC debug port (it is often used for I/O
2716 * delays, but the vmexits simply slow things down).
2717 */
2718 iova = kmap(vmx_io_bitmap_a);
2719 memset(iova, 0xff, PAGE_SIZE);
2720 clear_bit(0x80, iova);
cd0536d7 2721 kunmap(vmx_io_bitmap_a);
fdef3ad1
HQ		 2722
2723 iova = kmap(vmx_io_bitmap_b);
2724 memset(iova, 0xff, PAGE_SIZE);
cd0536d7 2725 kunmap(vmx_io_bitmap_b);
fdef3ad1 2726
2384d2b3
SY		 2727 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
2728
cb498ea2 2729 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
fdef3ad1
HQ		 2730 if (r)
2731 goto out1;
2732
c7addb90
AK		 2733 if (bypass_guest_pf)
2734 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
2735
fdef3ad1
HQ		 2736 return 0;
2737
2738out1:
2739 __free_page(vmx_io_bitmap_b);
2740out:
2741 __free_page(vmx_io_bitmap_a);
2742 return r;
6aa8b732
AK		 2743}
2744
2745static void __exit vmx_exit(void)
2746{
fdef3ad1
HQ		 2747 __free_page(vmx_io_bitmap_b);
2748 __free_page(vmx_io_bitmap_a);
2749
cb498ea2 2750 kvm_exit();
6aa8b732
AK		 2751}
2752
2753module_init(vmx_init)
2754module_exit(vmx_exit)
Clone of davem's net-next-2.6 tree