Merge branch 'linus' into sched/new-API-sched_setscheduler
[linux-2.6] / arch / x86 / kvm / vmx.c
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
18 #include "irq.h"
19 #include "vmx.h"
20 #include "mmu.h"
21
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29
30 #include <asm/io.h>
31 #include <asm/desc.h>
32
33 MODULE_AUTHOR("Qumranet");
34 MODULE_LICENSE("GPL");
35
36 static int bypass_guest_pf = 1;
37 module_param(bypass_guest_pf, bool, 0);
38
39 static int enable_vpid = 1;
40 module_param(enable_vpid, bool, 0);
41
42 static int flexpriority_enabled = 1;
43 module_param(flexpriority_enabled, bool, 0);
44
45 static int enable_ept = 1;
46 module_param(enable_ept, bool, 0);
47
48 struct vmcs {
49         u32 revision_id;
50         u32 abort;
51         char data[0];
52 };
53
54 struct vcpu_vmx {
55         struct kvm_vcpu       vcpu;
56         int                   launched;
57         u8                    fail;
58         u32                   idt_vectoring_info;
59         struct kvm_msr_entry *guest_msrs;
60         struct kvm_msr_entry *host_msrs;
61         int                   nmsrs;
62         int                   save_nmsrs;
63         int                   msr_offset_efer;
64 #ifdef CONFIG_X86_64
65         int                   msr_offset_kernel_gs_base;
66 #endif
67         struct vmcs          *vmcs;
68         struct {
69                 int           loaded;
70                 u16           fs_sel, gs_sel, ldt_sel;
71                 int           gs_ldt_reload_needed;
72                 int           fs_reload_needed;
73                 int           guest_efer_loaded;
74         } host_state;
75         struct {
76                 struct {
77                         bool pending;
78                         u8 vector;
79                         unsigned rip;
80                 } irq;
81         } rmode;
82         int vpid;
83 };
84
85 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
86 {
87         return container_of(vcpu, struct vcpu_vmx, vcpu);
88 }
89
90 static int init_rmode(struct kvm *kvm);
91
92 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
93 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
94
95 static struct page *vmx_io_bitmap_a;
96 static struct page *vmx_io_bitmap_b;
97 static struct page *vmx_msr_bitmap;
98
99 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
100 static DEFINE_SPINLOCK(vmx_vpid_lock);
101
102 static struct vmcs_config {
103         int size;
104         int order;
105         u32 revision_id;
106         u32 pin_based_exec_ctrl;
107         u32 cpu_based_exec_ctrl;
108         u32 cpu_based_2nd_exec_ctrl;
109         u32 vmexit_ctrl;
110         u32 vmentry_ctrl;
111 } vmcs_config;
112
113 struct vmx_capability {
114         u32 ept;
115         u32 vpid;
116 } vmx_capability;
117
118 #define VMX_SEGMENT_FIELD(seg)                                  \
119         [VCPU_SREG_##seg] = {                                   \
120                 .selector = GUEST_##seg##_SELECTOR,             \
121                 .base = GUEST_##seg##_BASE,                     \
122                 .limit = GUEST_##seg##_LIMIT,                   \
123                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
124         }
125
126 static struct kvm_vmx_segment_field {
127         unsigned selector;
128         unsigned base;
129         unsigned limit;
130         unsigned ar_bytes;
131 } kvm_vmx_segment_fields[] = {
132         VMX_SEGMENT_FIELD(CS),
133         VMX_SEGMENT_FIELD(DS),
134         VMX_SEGMENT_FIELD(ES),
135         VMX_SEGMENT_FIELD(FS),
136         VMX_SEGMENT_FIELD(GS),
137         VMX_SEGMENT_FIELD(SS),
138         VMX_SEGMENT_FIELD(TR),
139         VMX_SEGMENT_FIELD(LDTR),
140 };
141
142 /*
143  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
144  * away by decrementing the array size.
145  */
146 static const u32 vmx_msr_index[] = {
147 #ifdef CONFIG_X86_64
148         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
149 #endif
150         MSR_EFER, MSR_K6_STAR,
151 };
152 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
153
154 static void load_msrs(struct kvm_msr_entry *e, int n)
155 {
156         int i;
157
158         for (i = 0; i < n; ++i)
159                 wrmsrl(e[i].index, e[i].data);
160 }
161
162 static void save_msrs(struct kvm_msr_entry *e, int n)
163 {
164         int i;
165
166         for (i = 0; i < n; ++i)
167                 rdmsrl(e[i].index, e[i].data);
168 }
169
170 static inline int is_page_fault(u32 intr_info)
171 {
172         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
173                              INTR_INFO_VALID_MASK)) ==
174                 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
175 }
176
177 static inline int is_no_device(u32 intr_info)
178 {
179         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
180                              INTR_INFO_VALID_MASK)) ==
181                 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
182 }
183
184 static inline int is_invalid_opcode(u32 intr_info)
185 {
186         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
187                              INTR_INFO_VALID_MASK)) ==
188                 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
189 }
190
191 static inline int is_external_interrupt(u32 intr_info)
192 {
193         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
194                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
195 }
196
197 static inline int cpu_has_vmx_msr_bitmap(void)
198 {
199         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS);
200 }
201
202 static inline int cpu_has_vmx_tpr_shadow(void)
203 {
204         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
205 }
206
207 static inline int vm_need_tpr_shadow(struct kvm *kvm)
208 {
209         return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
210 }
211
212 static inline int cpu_has_secondary_exec_ctrls(void)
213 {
214         return (vmcs_config.cpu_based_exec_ctrl &
215                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
216 }
217
218 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
219 {
220         return flexpriority_enabled
221                 && (vmcs_config.cpu_based_2nd_exec_ctrl &
222                     SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
223 }
224
225 static inline int cpu_has_vmx_invept_individual_addr(void)
226 {
227         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT));
228 }
229
230 static inline int cpu_has_vmx_invept_context(void)
231 {
232         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT));
233 }
234
235 static inline int cpu_has_vmx_invept_global(void)
236 {
237         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT));
238 }
239
240 static inline int cpu_has_vmx_ept(void)
241 {
242         return (vmcs_config.cpu_based_2nd_exec_ctrl &
243                 SECONDARY_EXEC_ENABLE_EPT);
244 }
245
246 static inline int vm_need_ept(void)
247 {
248         return (cpu_has_vmx_ept() && enable_ept);
249 }
250
251 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
252 {
253         return ((cpu_has_vmx_virtualize_apic_accesses()) &&
254                 (irqchip_in_kernel(kvm)));
255 }
256
257 static inline int cpu_has_vmx_vpid(void)
258 {
259         return (vmcs_config.cpu_based_2nd_exec_ctrl &
260                 SECONDARY_EXEC_ENABLE_VPID);
261 }
262
263 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
264 {
265         int i;
266
267         for (i = 0; i < vmx->nmsrs; ++i)
268                 if (vmx->guest_msrs[i].index == msr)
269                         return i;
270         return -1;
271 }
272
273 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
274 {
275     struct {
276         u64 vpid : 16;
277         u64 rsvd : 48;
278         u64 gva;
279     } operand = { vpid, 0, gva };
280
281     asm volatile (ASM_VMX_INVVPID
282                   /* CF==1 or ZF==1 --> rc = -1 */
283                   "; ja 1f ; ud2 ; 1:"
284                   : : "a"(&operand), "c"(ext) : "cc", "memory");
285 }
286
287 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
288 {
289         struct {
290                 u64 eptp, gpa;
291         } operand = {eptp, gpa};
292
293         asm volatile (ASM_VMX_INVEPT
294                         /* CF==1 or ZF==1 --> rc = -1 */
295                         "; ja 1f ; ud2 ; 1:\n"
296                         : : "a" (&operand), "c" (ext) : "cc", "memory");
297 }
298
299 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
300 {
301         int i;
302
303         i = __find_msr_index(vmx, msr);
304         if (i >= 0)
305                 return &vmx->guest_msrs[i];
306         return NULL;
307 }
308
309 static void vmcs_clear(struct vmcs *vmcs)
310 {
311         u64 phys_addr = __pa(vmcs);
312         u8 error;
313
314         asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
315                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
316                       : "cc", "memory");
317         if (error)
318                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
319                        vmcs, phys_addr);
320 }
321
322 static void __vcpu_clear(void *arg)
323 {
324         struct vcpu_vmx *vmx = arg;
325         int cpu = raw_smp_processor_id();
326
327         if (vmx->vcpu.cpu == cpu)
328                 vmcs_clear(vmx->vmcs);
329         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
330                 per_cpu(current_vmcs, cpu) = NULL;
331         rdtscll(vmx->vcpu.arch.host_tsc);
332 }
333
334 static void vcpu_clear(struct vcpu_vmx *vmx)
335 {
336         if (vmx->vcpu.cpu == -1)
337                 return;
338         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
339         vmx->launched = 0;
340 }
341
342 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
343 {
344         if (vmx->vpid == 0)
345                 return;
346
347         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
348 }
349
350 static inline void ept_sync_global(void)
351 {
352         if (cpu_has_vmx_invept_global())
353                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
354 }
355
356 static inline void ept_sync_context(u64 eptp)
357 {
358         if (vm_need_ept()) {
359                 if (cpu_has_vmx_invept_context())
360                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
361                 else
362                         ept_sync_global();
363         }
364 }
365
366 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
367 {
368         if (vm_need_ept()) {
369                 if (cpu_has_vmx_invept_individual_addr())
370                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
371                                         eptp, gpa);
372                 else
373                         ept_sync_context(eptp);
374         }
375 }
376
377 static unsigned long vmcs_readl(unsigned long field)
378 {
379         unsigned long value;
380
381         asm volatile (ASM_VMX_VMREAD_RDX_RAX
382                       : "=a"(value) : "d"(field) : "cc");
383         return value;
384 }
385
386 static u16 vmcs_read16(unsigned long field)
387 {
388         return vmcs_readl(field);
389 }
390
391 static u32 vmcs_read32(unsigned long field)
392 {
393         return vmcs_readl(field);
394 }
395
396 static u64 vmcs_read64(unsigned long field)
397 {
398 #ifdef CONFIG_X86_64
399         return vmcs_readl(field);
400 #else
401         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
402 #endif
403 }
404
405 static noinline void vmwrite_error(unsigned long field, unsigned long value)
406 {
407         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
408                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
409         dump_stack();
410 }
411
412 static void vmcs_writel(unsigned long field, unsigned long value)
413 {
414         u8 error;
415
416         asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
417                        : "=q"(error) : "a"(value), "d"(field) : "cc");
418         if (unlikely(error))
419                 vmwrite_error(field, value);
420 }
421
422 static void vmcs_write16(unsigned long field, u16 value)
423 {
424         vmcs_writel(field, value);
425 }
426
427 static void vmcs_write32(unsigned long field, u32 value)
428 {
429         vmcs_writel(field, value);
430 }
431
432 static void vmcs_write64(unsigned long field, u64 value)
433 {
434 #ifdef CONFIG_X86_64
435         vmcs_writel(field, value);
436 #else
437         vmcs_writel(field, value);
438         asm volatile ("");
439         vmcs_writel(field+1, value >> 32);
440 #endif
441 }
442
443 static void vmcs_clear_bits(unsigned long field, u32 mask)
444 {
445         vmcs_writel(field, vmcs_readl(field) & ~mask);
446 }
447
448 static void vmcs_set_bits(unsigned long field, u32 mask)
449 {
450         vmcs_writel(field, vmcs_readl(field) | mask);
451 }
452
453 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
454 {
455         u32 eb;
456
457         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
458         if (!vcpu->fpu_active)
459                 eb |= 1u << NM_VECTOR;
460         if (vcpu->guest_debug.enabled)
461                 eb |= 1u << 1;
462         if (vcpu->arch.rmode.active)
463                 eb = ~0;
464         if (vm_need_ept())
465                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
466         vmcs_write32(EXCEPTION_BITMAP, eb);
467 }
468
469 static void reload_tss(void)
470 {
471         /*
472          * VT restores TR but not its size.  Useless.
473          */
474         struct descriptor_table gdt;
475         struct desc_struct *descs;
476
477         get_gdt(&gdt);
478         descs = (void *)gdt.base;
479         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
480         load_TR_desc();
481 }
482
483 static void load_transition_efer(struct vcpu_vmx *vmx)
484 {
485         int efer_offset = vmx->msr_offset_efer;
486         u64 host_efer = vmx->host_msrs[efer_offset].data;
487         u64 guest_efer = vmx->guest_msrs[efer_offset].data;
488         u64 ignore_bits;
489
490         if (efer_offset < 0)
491                 return;
492         /*
493          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
494          * outside long mode
495          */
496         ignore_bits = EFER_NX | EFER_SCE;
497 #ifdef CONFIG_X86_64
498         ignore_bits |= EFER_LMA | EFER_LME;
499         /* SCE is meaningful only in long mode on Intel */
500         if (guest_efer & EFER_LMA)
501                 ignore_bits &= ~(u64)EFER_SCE;
502 #endif
503         if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
504                 return;
505
506         vmx->host_state.guest_efer_loaded = 1;
507         guest_efer &= ~ignore_bits;
508         guest_efer |= host_efer & ignore_bits;
509         wrmsrl(MSR_EFER, guest_efer);
510         vmx->vcpu.stat.efer_reload++;
511 }
512
513 static void reload_host_efer(struct vcpu_vmx *vmx)
514 {
515         if (vmx->host_state.guest_efer_loaded) {
516                 vmx->host_state.guest_efer_loaded = 0;
517                 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
518         }
519 }
520
521 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
522 {
523         struct vcpu_vmx *vmx = to_vmx(vcpu);
524
525         if (vmx->host_state.loaded)
526                 return;
527
528         vmx->host_state.loaded = 1;
529         /*
530          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
531          * allow segment selectors with cpl > 0 or ti == 1.
532          */
533         vmx->host_state.ldt_sel = read_ldt();
534         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
535         vmx->host_state.fs_sel = read_fs();
536         if (!(vmx->host_state.fs_sel & 7)) {
537                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
538                 vmx->host_state.fs_reload_needed = 0;
539         } else {
540                 vmcs_write16(HOST_FS_SELECTOR, 0);
541                 vmx->host_state.fs_reload_needed = 1;
542         }
543         vmx->host_state.gs_sel = read_gs();
544         if (!(vmx->host_state.gs_sel & 7))
545                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
546         else {
547                 vmcs_write16(HOST_GS_SELECTOR, 0);
548                 vmx->host_state.gs_ldt_reload_needed = 1;
549         }
550
551 #ifdef CONFIG_X86_64
552         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
553         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
554 #else
555         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
556         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
557 #endif
558
559 #ifdef CONFIG_X86_64
560         if (is_long_mode(&vmx->vcpu))
561                 save_msrs(vmx->host_msrs +
562                           vmx->msr_offset_kernel_gs_base, 1);
563
564 #endif
565         load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
566         load_transition_efer(vmx);
567 }
568
569 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
570 {
571         unsigned long flags;
572
573         if (!vmx->host_state.loaded)
574                 return;
575
576         ++vmx->vcpu.stat.host_state_reload;
577         vmx->host_state.loaded = 0;
578         if (vmx->host_state.fs_reload_needed)
579                 load_fs(vmx->host_state.fs_sel);
580         if (vmx->host_state.gs_ldt_reload_needed) {
581                 load_ldt(vmx->host_state.ldt_sel);
582                 /*
583                  * If we have to reload gs, we must take care to
584                  * preserve our gs base.
585                  */
586                 local_irq_save(flags);
587                 load_gs(vmx->host_state.gs_sel);
588 #ifdef CONFIG_X86_64
589                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
590 #endif
591                 local_irq_restore(flags);
592         }
593         reload_tss();
594         save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
595         load_msrs(vmx->host_msrs, vmx->save_nmsrs);
596         reload_host_efer(vmx);
597 }
598
599 static void vmx_load_host_state(struct vcpu_vmx *vmx)
600 {
601         preempt_disable();
602         __vmx_load_host_state(vmx);
603         preempt_enable();
604 }
605
606 /*
607  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
608  * vcpu mutex is already taken.
609  */
610 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
611 {
612         struct vcpu_vmx *vmx = to_vmx(vcpu);
613         u64 phys_addr = __pa(vmx->vmcs);
614         u64 tsc_this, delta, new_offset;
615
616         if (vcpu->cpu != cpu) {
617                 vcpu_clear(vmx);
618                 kvm_migrate_timers(vcpu);
619                 vpid_sync_vcpu_all(vmx);
620         }
621
622         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
623                 u8 error;
624
625                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
626                 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
627                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
628                               : "cc");
629                 if (error)
630                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
631                                vmx->vmcs, phys_addr);
632         }
633
634         if (vcpu->cpu != cpu) {
635                 struct descriptor_table dt;
636                 unsigned long sysenter_esp;
637
638                 vcpu->cpu = cpu;
639                 /*
640                  * Linux uses per-cpu TSS and GDT, so set these when switching
641                  * processors.
642                  */
643                 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
644                 get_gdt(&dt);
645                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
646
647                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
648                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
649
650                 /*
651                  * Make sure the time stamp counter is monotonous.
652                  */
653                 rdtscll(tsc_this);
654                 if (tsc_this < vcpu->arch.host_tsc) {
655                         delta = vcpu->arch.host_tsc - tsc_this;
656                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
657                         vmcs_write64(TSC_OFFSET, new_offset);
658                 }
659         }
660 }
661
662 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
663 {
664         __vmx_load_host_state(to_vmx(vcpu));
665 }
666
667 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
668 {
669         if (vcpu->fpu_active)
670                 return;
671         vcpu->fpu_active = 1;
672         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
673         if (vcpu->arch.cr0 & X86_CR0_TS)
674                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
675         update_exception_bitmap(vcpu);
676 }
677
678 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
679 {
680         if (!vcpu->fpu_active)
681                 return;
682         vcpu->fpu_active = 0;
683         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
684         update_exception_bitmap(vcpu);
685 }
686
687 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
688 {
689         vcpu_clear(to_vmx(vcpu));
690 }
691
692 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
693 {
694         return vmcs_readl(GUEST_RFLAGS);
695 }
696
697 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
698 {
699         if (vcpu->arch.rmode.active)
700                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
701         vmcs_writel(GUEST_RFLAGS, rflags);
702 }
703
704 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
705 {
706         unsigned long rip;
707         u32 interruptibility;
708
709         rip = vmcs_readl(GUEST_RIP);
710         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
711         vmcs_writel(GUEST_RIP, rip);
712
713         /*
714          * We emulated an instruction, so temporary interrupt blocking
715          * should be removed, if set.
716          */
717         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
718         if (interruptibility & 3)
719                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
720                              interruptibility & ~3);
721         vcpu->arch.interrupt_window_open = 1;
722 }
723
724 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
725                                 bool has_error_code, u32 error_code)
726 {
727         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
728                      nr | INTR_TYPE_EXCEPTION
729                      | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
730                      | INTR_INFO_VALID_MASK);
731         if (has_error_code)
732                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
733 }
734
735 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
736 {
737         struct vcpu_vmx *vmx = to_vmx(vcpu);
738
739         return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
740 }
741
742 /*
743  * Swap MSR entry in host/guest MSR entry array.
744  */
745 #ifdef CONFIG_X86_64
746 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
747 {
748         struct kvm_msr_entry tmp;
749
750         tmp = vmx->guest_msrs[to];
751         vmx->guest_msrs[to] = vmx->guest_msrs[from];
752         vmx->guest_msrs[from] = tmp;
753         tmp = vmx->host_msrs[to];
754         vmx->host_msrs[to] = vmx->host_msrs[from];
755         vmx->host_msrs[from] = tmp;
756 }
757 #endif
758
759 /*
760  * Set up the vmcs to automatically save and restore system
761  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
762  * mode, as fiddling with msrs is very expensive.
763  */
764 static void setup_msrs(struct vcpu_vmx *vmx)
765 {
766         int save_nmsrs;
767
768         vmx_load_host_state(vmx);
769         save_nmsrs = 0;
770 #ifdef CONFIG_X86_64
771         if (is_long_mode(&vmx->vcpu)) {
772                 int index;
773
774                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
775                 if (index >= 0)
776                         move_msr_up(vmx, index, save_nmsrs++);
777                 index = __find_msr_index(vmx, MSR_LSTAR);
778                 if (index >= 0)
779                         move_msr_up(vmx, index, save_nmsrs++);
780                 index = __find_msr_index(vmx, MSR_CSTAR);
781                 if (index >= 0)
782                         move_msr_up(vmx, index, save_nmsrs++);
783                 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
784                 if (index >= 0)
785                         move_msr_up(vmx, index, save_nmsrs++);
786                 /*
787                  * MSR_K6_STAR is only needed on long mode guests, and only
788                  * if efer.sce is enabled.
789                  */
790                 index = __find_msr_index(vmx, MSR_K6_STAR);
791                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
792                         move_msr_up(vmx, index, save_nmsrs++);
793         }
794 #endif
795         vmx->save_nmsrs = save_nmsrs;
796
797 #ifdef CONFIG_X86_64
798         vmx->msr_offset_kernel_gs_base =
799                 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
800 #endif
801         vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
802 }
803
804 /*
805  * reads and returns guest's timestamp counter "register"
806  * guest_tsc = host_tsc + tsc_offset    -- 21.3
807  */
808 static u64 guest_read_tsc(void)
809 {
810         u64 host_tsc, tsc_offset;
811
812         rdtscll(host_tsc);
813         tsc_offset = vmcs_read64(TSC_OFFSET);
814         return host_tsc + tsc_offset;
815 }
816
817 /*
818  * writes 'guest_tsc' into guest's timestamp counter "register"
819  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
820  */
821 static void guest_write_tsc(u64 guest_tsc)
822 {
823         u64 host_tsc;
824
825         rdtscll(host_tsc);
826         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
827 }
828
829 /*
830  * Reads an msr value (of 'msr_index') into 'pdata'.
831  * Returns 0 on success, non-0 otherwise.
832  * Assumes vcpu_load() was already called.
833  */
834 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
835 {
836         u64 data;
837         struct kvm_msr_entry *msr;
838
839         if (!pdata) {
840                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
841                 return -EINVAL;
842         }
843
844         switch (msr_index) {
845 #ifdef CONFIG_X86_64
846         case MSR_FS_BASE:
847                 data = vmcs_readl(GUEST_FS_BASE);
848                 break;
849         case MSR_GS_BASE:
850                 data = vmcs_readl(GUEST_GS_BASE);
851                 break;
852         case MSR_EFER:
853                 return kvm_get_msr_common(vcpu, msr_index, pdata);
854 #endif
855         case MSR_IA32_TIME_STAMP_COUNTER:
856                 data = guest_read_tsc();
857                 break;
858         case MSR_IA32_SYSENTER_CS:
859                 data = vmcs_read32(GUEST_SYSENTER_CS);
860                 break;
861         case MSR_IA32_SYSENTER_EIP:
862                 data = vmcs_readl(GUEST_SYSENTER_EIP);
863                 break;
864         case MSR_IA32_SYSENTER_ESP:
865                 data = vmcs_readl(GUEST_SYSENTER_ESP);
866                 break;
867         default:
868                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
869                 if (msr) {
870                         data = msr->data;
871                         break;
872                 }
873                 return kvm_get_msr_common(vcpu, msr_index, pdata);
874         }
875
876         *pdata = data;
877         return 0;
878 }
879
880 /*
881  * Writes msr value into into the appropriate "register".
882  * Returns 0 on success, non-0 otherwise.
883  * Assumes vcpu_load() was already called.
884  */
885 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
886 {
887         struct vcpu_vmx *vmx = to_vmx(vcpu);
888         struct kvm_msr_entry *msr;
889         int ret = 0;
890
891         switch (msr_index) {
892 #ifdef CONFIG_X86_64
893         case MSR_EFER:
894                 vmx_load_host_state(vmx);
895                 ret = kvm_set_msr_common(vcpu, msr_index, data);
896                 break;
897         case MSR_FS_BASE:
898                 vmcs_writel(GUEST_FS_BASE, data);
899                 break;
900         case MSR_GS_BASE:
901                 vmcs_writel(GUEST_GS_BASE, data);
902                 break;
903 #endif
904         case MSR_IA32_SYSENTER_CS:
905                 vmcs_write32(GUEST_SYSENTER_CS, data);
906                 break;
907         case MSR_IA32_SYSENTER_EIP:
908                 vmcs_writel(GUEST_SYSENTER_EIP, data);
909                 break;
910         case MSR_IA32_SYSENTER_ESP:
911                 vmcs_writel(GUEST_SYSENTER_ESP, data);
912                 break;
913         case MSR_IA32_TIME_STAMP_COUNTER:
914                 guest_write_tsc(data);
915                 break;
916         default:
917                 vmx_load_host_state(vmx);
918                 msr = find_msr_entry(vmx, msr_index);
919                 if (msr) {
920                         msr->data = data;
921                         break;
922                 }
923                 ret = kvm_set_msr_common(vcpu, msr_index, data);
924         }
925
926         return ret;
927 }
928
929 /*
930  * Sync the rsp and rip registers into the vcpu structure.  This allows
931  * registers to be accessed by indexing vcpu->arch.regs.
932  */
933 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
934 {
935         vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
936         vcpu->arch.rip = vmcs_readl(GUEST_RIP);
937 }
938
939 /*
940  * Syncs rsp and rip back into the vmcs.  Should be called after possible
941  * modification.
942  */
943 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
944 {
945         vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
946         vmcs_writel(GUEST_RIP, vcpu->arch.rip);
947 }
948
949 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
950 {
951         unsigned long dr7 = 0x400;
952         int old_singlestep;
953
954         old_singlestep = vcpu->guest_debug.singlestep;
955
956         vcpu->guest_debug.enabled = dbg->enabled;
957         if (vcpu->guest_debug.enabled) {
958                 int i;
959
960                 dr7 |= 0x200;  /* exact */
961                 for (i = 0; i < 4; ++i) {
962                         if (!dbg->breakpoints[i].enabled)
963                                 continue;
964                         vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
965                         dr7 |= 2 << (i*2);    /* global enable */
966                         dr7 |= 0 << (i*4+16); /* execution breakpoint */
967                 }
968
969                 vcpu->guest_debug.singlestep = dbg->singlestep;
970         } else
971                 vcpu->guest_debug.singlestep = 0;
972
973         if (old_singlestep && !vcpu->guest_debug.singlestep) {
974                 unsigned long flags;
975
976                 flags = vmcs_readl(GUEST_RFLAGS);
977                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
978                 vmcs_writel(GUEST_RFLAGS, flags);
979         }
980
981         update_exception_bitmap(vcpu);
982         vmcs_writel(GUEST_DR7, dr7);
983
984         return 0;
985 }
986
987 static int vmx_get_irq(struct kvm_vcpu *vcpu)
988 {
989         struct vcpu_vmx *vmx = to_vmx(vcpu);
990         u32 idtv_info_field;
991
992         idtv_info_field = vmx->idt_vectoring_info;
993         if (idtv_info_field & INTR_INFO_VALID_MASK) {
994                 if (is_external_interrupt(idtv_info_field))
995                         return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
996                 else
997                         printk(KERN_DEBUG "pending exception: not handled yet\n");
998         }
999         return -1;
1000 }
1001
1002 static __init int cpu_has_kvm_support(void)
1003 {
1004         unsigned long ecx = cpuid_ecx(1);
1005         return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
1006 }
1007
1008 static __init int vmx_disabled_by_bios(void)
1009 {
1010         u64 msr;
1011
1012         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1013         return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
1014                        MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
1015             == MSR_IA32_FEATURE_CONTROL_LOCKED;
1016         /* locked but not enabled */
1017 }
1018
1019 static void hardware_enable(void *garbage)
1020 {
1021         int cpu = raw_smp_processor_id();
1022         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1023         u64 old;
1024
1025         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1026         if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
1027                     MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
1028             != (MSR_IA32_FEATURE_CONTROL_LOCKED |
1029                 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
1030                 /* enable and lock */
1031                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1032                        MSR_IA32_FEATURE_CONTROL_LOCKED |
1033                        MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
1034         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1035         asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
1036                       : "memory", "cc");
1037 }
1038
1039 static void hardware_disable(void *garbage)
1040 {
1041         asm volatile (ASM_VMX_VMXOFF : : : "cc");
1042         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1043 }
1044
1045 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1046                                       u32 msr, u32 *result)
1047 {
1048         u32 vmx_msr_low, vmx_msr_high;
1049         u32 ctl = ctl_min | ctl_opt;
1050
1051         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1052
1053         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1054         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1055
1056         /* Ensure minimum (required) set of control bits are supported. */
1057         if (ctl_min & ~ctl)
1058                 return -EIO;
1059
1060         *result = ctl;
1061         return 0;
1062 }
1063
1064 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1065 {
1066         u32 vmx_msr_low, vmx_msr_high;
1067         u32 min, opt, min2, opt2;
1068         u32 _pin_based_exec_control = 0;
1069         u32 _cpu_based_exec_control = 0;
1070         u32 _cpu_based_2nd_exec_control = 0;
1071         u32 _vmexit_control = 0;
1072         u32 _vmentry_control = 0;
1073
1074         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1075         opt = 0;
1076         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1077                                 &_pin_based_exec_control) < 0)
1078                 return -EIO;
1079
1080         min = CPU_BASED_HLT_EXITING |
1081 #ifdef CONFIG_X86_64
1082               CPU_BASED_CR8_LOAD_EXITING |
1083               CPU_BASED_CR8_STORE_EXITING |
1084 #endif
1085               CPU_BASED_CR3_LOAD_EXITING |
1086               CPU_BASED_CR3_STORE_EXITING |
1087               CPU_BASED_USE_IO_BITMAPS |
1088               CPU_BASED_MOV_DR_EXITING |
1089               CPU_BASED_USE_TSC_OFFSETING;
1090         opt = CPU_BASED_TPR_SHADOW |
1091               CPU_BASED_USE_MSR_BITMAPS |
1092               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1093         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1094                                 &_cpu_based_exec_control) < 0)
1095                 return -EIO;
1096 #ifdef CONFIG_X86_64
1097         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1098                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1099                                            ~CPU_BASED_CR8_STORE_EXITING;
1100 #endif
1101         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1102                 min2 = 0;
1103                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1104                         SECONDARY_EXEC_WBINVD_EXITING |
1105                         SECONDARY_EXEC_ENABLE_VPID |
1106                         SECONDARY_EXEC_ENABLE_EPT;
1107                 if (adjust_vmx_controls(min2, opt2,
1108                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1109                                         &_cpu_based_2nd_exec_control) < 0)
1110                         return -EIO;
1111         }
1112 #ifndef CONFIG_X86_64
1113         if (!(_cpu_based_2nd_exec_control &
1114                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1115                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1116 #endif
1117         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1118                 /* CR3 accesses don't need to cause VM Exits when EPT enabled */
1119                 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1120                          CPU_BASED_CR3_STORE_EXITING);
1121                 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1122                                         &_cpu_based_exec_control) < 0)
1123                         return -EIO;
1124                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1125                       vmx_capability.ept, vmx_capability.vpid);
1126         }
1127
1128         min = 0;
1129 #ifdef CONFIG_X86_64
1130         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1131 #endif
1132         opt = 0;
1133         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1134                                 &_vmexit_control) < 0)
1135                 return -EIO;
1136
1137         min = opt = 0;
1138         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1139                                 &_vmentry_control) < 0)
1140                 return -EIO;
1141
1142         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1143
1144         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1145         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1146                 return -EIO;
1147
1148 #ifdef CONFIG_X86_64
1149         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1150         if (vmx_msr_high & (1u<<16))
1151                 return -EIO;
1152 #endif
1153
1154         /* Require Write-Back (WB) memory type for VMCS accesses. */
1155         if (((vmx_msr_high >> 18) & 15) != 6)
1156                 return -EIO;
1157
1158         vmcs_conf->size = vmx_msr_high & 0x1fff;
1159         vmcs_conf->order = get_order(vmcs_config.size);
1160         vmcs_conf->revision_id = vmx_msr_low;
1161
1162         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1163         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1164         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1165         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1166         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1167
1168         return 0;
1169 }
1170
1171 static struct vmcs *alloc_vmcs_cpu(int cpu)
1172 {
1173         int node = cpu_to_node(cpu);
1174         struct page *pages;
1175         struct vmcs *vmcs;
1176
1177         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1178         if (!pages)
1179                 return NULL;
1180         vmcs = page_address(pages);
1181         memset(vmcs, 0, vmcs_config.size);
1182         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1183         return vmcs;
1184 }
1185
1186 static struct vmcs *alloc_vmcs(void)
1187 {
1188         return alloc_vmcs_cpu(raw_smp_processor_id());
1189 }
1190
1191 static void free_vmcs(struct vmcs *vmcs)
1192 {
1193         free_pages((unsigned long)vmcs, vmcs_config.order);
1194 }
1195
1196 static void free_kvm_area(void)
1197 {
1198         int cpu;
1199
1200         for_each_online_cpu(cpu)
1201                 free_vmcs(per_cpu(vmxarea, cpu));
1202 }
1203
1204 static __init int alloc_kvm_area(void)
1205 {
1206         int cpu;
1207
1208         for_each_online_cpu(cpu) {
1209                 struct vmcs *vmcs;
1210
1211                 vmcs = alloc_vmcs_cpu(cpu);
1212                 if (!vmcs) {
1213                         free_kvm_area();
1214                         return -ENOMEM;
1215                 }
1216
1217                 per_cpu(vmxarea, cpu) = vmcs;
1218         }
1219         return 0;
1220 }
1221
1222 static __init int hardware_setup(void)
1223 {
1224         if (setup_vmcs_config(&vmcs_config) < 0)
1225                 return -EIO;
1226
1227         if (boot_cpu_has(X86_FEATURE_NX))
1228                 kvm_enable_efer_bits(EFER_NX);
1229
1230         return alloc_kvm_area();
1231 }
1232
1233 static __exit void hardware_unsetup(void)
1234 {
1235         free_kvm_area();
1236 }
1237
1238 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1239 {
1240         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1241
1242         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1243                 vmcs_write16(sf->selector, save->selector);
1244                 vmcs_writel(sf->base, save->base);
1245                 vmcs_write32(sf->limit, save->limit);
1246                 vmcs_write32(sf->ar_bytes, save->ar);
1247         } else {
1248                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1249                         << AR_DPL_SHIFT;
1250                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1251         }
1252 }
1253
1254 static void enter_pmode(struct kvm_vcpu *vcpu)
1255 {
1256         unsigned long flags;
1257
1258         vcpu->arch.rmode.active = 0;
1259
1260         vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1261         vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1262         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1263
1264         flags = vmcs_readl(GUEST_RFLAGS);
1265         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1266         flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1267         vmcs_writel(GUEST_RFLAGS, flags);
1268
1269         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1270                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1271
1272         update_exception_bitmap(vcpu);
1273
1274         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1275         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1276         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1277         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1278
1279         vmcs_write16(GUEST_SS_SELECTOR, 0);
1280         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1281
1282         vmcs_write16(GUEST_CS_SELECTOR,
1283                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1284         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1285 }
1286
1287 static gva_t rmode_tss_base(struct kvm *kvm)
1288 {
1289         if (!kvm->arch.tss_addr) {
1290                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1291                                  kvm->memslots[0].npages - 3;
1292                 return base_gfn << PAGE_SHIFT;
1293         }
1294         return kvm->arch.tss_addr;
1295 }
1296
1297 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1298 {
1299         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1300
1301         save->selector = vmcs_read16(sf->selector);
1302         save->base = vmcs_readl(sf->base);
1303         save->limit = vmcs_read32(sf->limit);
1304         save->ar = vmcs_read32(sf->ar_bytes);
1305         vmcs_write16(sf->selector, save->base >> 4);
1306         vmcs_write32(sf->base, save->base & 0xfffff);
1307         vmcs_write32(sf->limit, 0xffff);
1308         vmcs_write32(sf->ar_bytes, 0xf3);
1309 }
1310
1311 static void enter_rmode(struct kvm_vcpu *vcpu)
1312 {
1313         unsigned long flags;
1314
1315         vcpu->arch.rmode.active = 1;
1316
1317         vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1318         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1319
1320         vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1321         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1322
1323         vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1324         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1325
1326         flags = vmcs_readl(GUEST_RFLAGS);
1327         vcpu->arch.rmode.save_iopl
1328                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1329
1330         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1331
1332         vmcs_writel(GUEST_RFLAGS, flags);
1333         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1334         update_exception_bitmap(vcpu);
1335
1336         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1337         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1338         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1339
1340         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1341         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1342         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1343                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1344         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1345
1346         fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1347         fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1348         fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1349         fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1350
1351         kvm_mmu_reset_context(vcpu);
1352         init_rmode(vcpu->kvm);
1353 }
1354
1355 #ifdef CONFIG_X86_64
1356
1357 static void enter_lmode(struct kvm_vcpu *vcpu)
1358 {
1359         u32 guest_tr_ar;
1360
1361         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1362         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1363                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1364                        __func__);
1365                 vmcs_write32(GUEST_TR_AR_BYTES,
1366                              (guest_tr_ar & ~AR_TYPE_MASK)
1367                              | AR_TYPE_BUSY_64_TSS);
1368         }
1369
1370         vcpu->arch.shadow_efer |= EFER_LMA;
1371
1372         find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1373         vmcs_write32(VM_ENTRY_CONTROLS,
1374                      vmcs_read32(VM_ENTRY_CONTROLS)
1375                      | VM_ENTRY_IA32E_MODE);
1376 }
1377
1378 static void exit_lmode(struct kvm_vcpu *vcpu)
1379 {
1380         vcpu->arch.shadow_efer &= ~EFER_LMA;
1381
1382         vmcs_write32(VM_ENTRY_CONTROLS,
1383                      vmcs_read32(VM_ENTRY_CONTROLS)
1384                      & ~VM_ENTRY_IA32E_MODE);
1385 }
1386
1387 #endif
1388
1389 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1390 {
1391         vpid_sync_vcpu_all(to_vmx(vcpu));
1392 }
1393
1394 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1395 {
1396         vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1397         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1398 }
1399
1400 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1401 {
1402         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1403                 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1404                         printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1405                         return;
1406                 }
1407                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1408                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1409                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1410                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1411         }
1412 }
1413
1414 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1415
1416 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1417                                         unsigned long cr0,
1418                                         struct kvm_vcpu *vcpu)
1419 {
1420         if (!(cr0 & X86_CR0_PG)) {
1421                 /* From paging/starting to nonpaging */
1422                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1423                              vmcs_config.cpu_based_exec_ctrl |
1424                              (CPU_BASED_CR3_LOAD_EXITING |
1425                               CPU_BASED_CR3_STORE_EXITING));
1426                 vcpu->arch.cr0 = cr0;
1427                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1428                 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1429                 *hw_cr0 &= ~X86_CR0_WP;
1430         } else if (!is_paging(vcpu)) {
1431                 /* From nonpaging to paging */
1432                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1433                              vmcs_config.cpu_based_exec_ctrl &
1434                              ~(CPU_BASED_CR3_LOAD_EXITING |
1435                                CPU_BASED_CR3_STORE_EXITING));
1436                 vcpu->arch.cr0 = cr0;
1437                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1438                 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1439                         *hw_cr0 &= ~X86_CR0_WP;
1440         }
1441 }
1442
1443 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1444                                         struct kvm_vcpu *vcpu)
1445 {
1446         if (!is_paging(vcpu)) {
1447                 *hw_cr4 &= ~X86_CR4_PAE;
1448                 *hw_cr4 |= X86_CR4_PSE;
1449         } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1450                 *hw_cr4 &= ~X86_CR4_PAE;
1451 }
1452
1453 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1454 {
1455         unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1456                                 KVM_VM_CR0_ALWAYS_ON;
1457
1458         vmx_fpu_deactivate(vcpu);
1459
1460         if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1461                 enter_pmode(vcpu);
1462
1463         if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1464                 enter_rmode(vcpu);
1465
1466 #ifdef CONFIG_X86_64
1467         if (vcpu->arch.shadow_efer & EFER_LME) {
1468                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1469                         enter_lmode(vcpu);
1470                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1471                         exit_lmode(vcpu);
1472         }
1473 #endif
1474
1475         if (vm_need_ept())
1476                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1477
1478         vmcs_writel(CR0_READ_SHADOW, cr0);
1479         vmcs_writel(GUEST_CR0, hw_cr0);
1480         vcpu->arch.cr0 = cr0;
1481
1482         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1483                 vmx_fpu_activate(vcpu);
1484 }
1485
1486 static u64 construct_eptp(unsigned long root_hpa)
1487 {
1488         u64 eptp;
1489
1490         /* TODO write the value reading from MSR */
1491         eptp = VMX_EPT_DEFAULT_MT |
1492                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1493         eptp |= (root_hpa & PAGE_MASK);
1494
1495         return eptp;
1496 }
1497
1498 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1499 {
1500         unsigned long guest_cr3;
1501         u64 eptp;
1502
1503         guest_cr3 = cr3;
1504         if (vm_need_ept()) {
1505                 eptp = construct_eptp(cr3);
1506                 vmcs_write64(EPT_POINTER, eptp);
1507                 ept_sync_context(eptp);
1508                 ept_load_pdptrs(vcpu);
1509                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1510                         VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1511         }
1512
1513         vmx_flush_tlb(vcpu);
1514         vmcs_writel(GUEST_CR3, guest_cr3);
1515         if (vcpu->arch.cr0 & X86_CR0_PE)
1516                 vmx_fpu_deactivate(vcpu);
1517 }
1518
1519 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1520 {
1521         unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.active ?
1522                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1523
1524         vcpu->arch.cr4 = cr4;
1525         if (vm_need_ept())
1526                 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1527
1528         vmcs_writel(CR4_READ_SHADOW, cr4);
1529         vmcs_writel(GUEST_CR4, hw_cr4);
1530 }
1531
1532 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1533 {
1534         struct vcpu_vmx *vmx = to_vmx(vcpu);
1535         struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1536
1537         vcpu->arch.shadow_efer = efer;
1538         if (!msr)
1539                 return;
1540         if (efer & EFER_LMA) {
1541                 vmcs_write32(VM_ENTRY_CONTROLS,
1542                                      vmcs_read32(VM_ENTRY_CONTROLS) |
1543                                      VM_ENTRY_IA32E_MODE);
1544                 msr->data = efer;
1545
1546         } else {
1547                 vmcs_write32(VM_ENTRY_CONTROLS,
1548                                      vmcs_read32(VM_ENTRY_CONTROLS) &
1549                                      ~VM_ENTRY_IA32E_MODE);
1550
1551                 msr->data = efer & ~EFER_LME;
1552         }
1553         setup_msrs(vmx);
1554 }
1555
1556 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1557 {
1558         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1559
1560         return vmcs_readl(sf->base);
1561 }
1562
1563 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1564                             struct kvm_segment *var, int seg)
1565 {
1566         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1567         u32 ar;
1568
1569         var->base = vmcs_readl(sf->base);
1570         var->limit = vmcs_read32(sf->limit);
1571         var->selector = vmcs_read16(sf->selector);
1572         ar = vmcs_read32(sf->ar_bytes);
1573         if (ar & AR_UNUSABLE_MASK)
1574                 ar = 0;
1575         var->type = ar & 15;
1576         var->s = (ar >> 4) & 1;
1577         var->dpl = (ar >> 5) & 3;
1578         var->present = (ar >> 7) & 1;
1579         var->avl = (ar >> 12) & 1;
1580         var->l = (ar >> 13) & 1;
1581         var->db = (ar >> 14) & 1;
1582         var->g = (ar >> 15) & 1;
1583         var->unusable = (ar >> 16) & 1;
1584 }
1585
1586 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1587 {
1588         struct kvm_segment kvm_seg;
1589
1590         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1591                 return 0;
1592
1593         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1594                 return 3;
1595
1596         vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1597         return kvm_seg.selector & 3;
1598 }
1599
1600 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1601 {
1602         u32 ar;
1603
1604         if (var->unusable)
1605                 ar = 1 << 16;
1606         else {
1607                 ar = var->type & 15;
1608                 ar |= (var->s & 1) << 4;
1609                 ar |= (var->dpl & 3) << 5;
1610                 ar |= (var->present & 1) << 7;
1611                 ar |= (var->avl & 1) << 12;
1612                 ar |= (var->l & 1) << 13;
1613                 ar |= (var->db & 1) << 14;
1614                 ar |= (var->g & 1) << 15;
1615         }
1616         if (ar == 0) /* a 0 value means unusable */
1617                 ar = AR_UNUSABLE_MASK;
1618
1619         return ar;
1620 }
1621
1622 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1623                             struct kvm_segment *var, int seg)
1624 {
1625         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1626         u32 ar;
1627
1628         if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1629                 vcpu->arch.rmode.tr.selector = var->selector;
1630                 vcpu->arch.rmode.tr.base = var->base;
1631                 vcpu->arch.rmode.tr.limit = var->limit;
1632                 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1633                 return;
1634         }
1635         vmcs_writel(sf->base, var->base);
1636         vmcs_write32(sf->limit, var->limit);
1637         vmcs_write16(sf->selector, var->selector);
1638         if (vcpu->arch.rmode.active && var->s) {
1639                 /*
1640                  * Hack real-mode segments into vm86 compatibility.
1641                  */
1642                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1643                         vmcs_writel(sf->base, 0xf0000);
1644                 ar = 0xf3;
1645         } else
1646                 ar = vmx_segment_access_rights(var);
1647         vmcs_write32(sf->ar_bytes, ar);
1648 }
1649
1650 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1651 {
1652         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1653
1654         *db = (ar >> 14) & 1;
1655         *l = (ar >> 13) & 1;
1656 }
1657
1658 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1659 {
1660         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1661         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1662 }
1663
1664 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1665 {
1666         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1667         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1668 }
1669
1670 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1671 {
1672         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1673         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1674 }
1675
1676 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1677 {
1678         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1679         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1680 }
1681
1682 static int init_rmode_tss(struct kvm *kvm)
1683 {
1684         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1685         u16 data = 0;
1686         int ret = 0;
1687         int r;
1688
1689         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1690         if (r < 0)
1691                 goto out;
1692         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1693         r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1694         if (r < 0)
1695                 goto out;
1696         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1697         if (r < 0)
1698                 goto out;
1699         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1700         if (r < 0)
1701                 goto out;
1702         data = ~0;
1703         r = kvm_write_guest_page(kvm, fn, &data,
1704                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1705                                  sizeof(u8));
1706         if (r < 0)
1707                 goto out;
1708
1709         ret = 1;
1710 out:
1711         return ret;
1712 }
1713
1714 static int init_rmode_identity_map(struct kvm *kvm)
1715 {
1716         int i, r, ret;
1717         pfn_t identity_map_pfn;
1718         u32 tmp;
1719
1720         if (!vm_need_ept())
1721                 return 1;
1722         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
1723                 printk(KERN_ERR "EPT: identity-mapping pagetable "
1724                         "haven't been allocated!\n");
1725                 return 0;
1726         }
1727         if (likely(kvm->arch.ept_identity_pagetable_done))
1728                 return 1;
1729         ret = 0;
1730         identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
1731         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
1732         if (r < 0)
1733                 goto out;
1734         /* Set up identity-mapping pagetable for EPT in real mode */
1735         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
1736                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
1737                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
1738                 r = kvm_write_guest_page(kvm, identity_map_pfn,
1739                                 &tmp, i * sizeof(tmp), sizeof(tmp));
1740                 if (r < 0)
1741                         goto out;
1742         }
1743         kvm->arch.ept_identity_pagetable_done = true;
1744         ret = 1;
1745 out:
1746         return ret;
1747 }
1748
1749 static void seg_setup(int seg)
1750 {
1751         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1752
1753         vmcs_write16(sf->selector, 0);
1754         vmcs_writel(sf->base, 0);
1755         vmcs_write32(sf->limit, 0xffff);
1756         vmcs_write32(sf->ar_bytes, 0x93);
1757 }
1758
1759 static int alloc_apic_access_page(struct kvm *kvm)
1760 {
1761         struct kvm_userspace_memory_region kvm_userspace_mem;
1762         int r = 0;
1763
1764         down_write(&kvm->slots_lock);
1765         if (kvm->arch.apic_access_page)
1766                 goto out;
1767         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
1768         kvm_userspace_mem.flags = 0;
1769         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
1770         kvm_userspace_mem.memory_size = PAGE_SIZE;
1771         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1772         if (r)
1773                 goto out;
1774
1775         down_read(&current->mm->mmap_sem);
1776         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
1777         up_read(&current->mm->mmap_sem);
1778 out:
1779         up_write(&kvm->slots_lock);
1780         return r;
1781 }
1782
1783 static int alloc_identity_pagetable(struct kvm *kvm)
1784 {
1785         struct kvm_userspace_memory_region kvm_userspace_mem;
1786         int r = 0;
1787
1788         down_write(&kvm->slots_lock);
1789         if (kvm->arch.ept_identity_pagetable)
1790                 goto out;
1791         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
1792         kvm_userspace_mem.flags = 0;
1793         kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1794         kvm_userspace_mem.memory_size = PAGE_SIZE;
1795         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1796         if (r)
1797                 goto out;
1798
1799         down_read(&current->mm->mmap_sem);
1800         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
1801                         VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
1802         up_read(&current->mm->mmap_sem);
1803 out:
1804         up_write(&kvm->slots_lock);
1805         return r;
1806 }
1807
1808 static void allocate_vpid(struct vcpu_vmx *vmx)
1809 {
1810         int vpid;
1811
1812         vmx->vpid = 0;
1813         if (!enable_vpid || !cpu_has_vmx_vpid())
1814                 return;
1815         spin_lock(&vmx_vpid_lock);
1816         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
1817         if (vpid < VMX_NR_VPIDS) {
1818                 vmx->vpid = vpid;
1819                 __set_bit(vpid, vmx_vpid_bitmap);
1820         }
1821         spin_unlock(&vmx_vpid_lock);
1822 }
1823
1824 void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
1825 {
1826         void *va;
1827
1828         if (!cpu_has_vmx_msr_bitmap())
1829                 return;
1830
1831         /*
1832          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
1833          * have the write-low and read-high bitmap offsets the wrong way round.
1834          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
1835          */
1836         va = kmap(msr_bitmap);
1837         if (msr <= 0x1fff) {
1838                 __clear_bit(msr, va + 0x000); /* read-low */
1839                 __clear_bit(msr, va + 0x800); /* write-low */
1840         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1841                 msr &= 0x1fff;
1842                 __clear_bit(msr, va + 0x400); /* read-high */
1843                 __clear_bit(msr, va + 0xc00); /* write-high */
1844         }
1845         kunmap(msr_bitmap);
1846 }
1847
1848 /*
1849  * Sets up the vmcs for emulated real mode.
1850  */
1851 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1852 {
1853         u32 host_sysenter_cs;
1854         u32 junk;
1855         unsigned long a;
1856         struct descriptor_table dt;
1857         int i;
1858         unsigned long kvm_vmx_return;
1859         u32 exec_control;
1860
1861         /* I/O */
1862         vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1863         vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1864
1865         if (cpu_has_vmx_msr_bitmap())
1866                 vmcs_write64(MSR_BITMAP, page_to_phys(vmx_msr_bitmap));
1867
1868         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1869
1870         /* Control */
1871         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1872                 vmcs_config.pin_based_exec_ctrl);
1873
1874         exec_control = vmcs_config.cpu_based_exec_ctrl;
1875         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1876                 exec_control &= ~CPU_BASED_TPR_SHADOW;
1877 #ifdef CONFIG_X86_64
1878                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1879                                 CPU_BASED_CR8_LOAD_EXITING;
1880 #endif
1881         }
1882         if (!vm_need_ept())
1883                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
1884                                 CPU_BASED_CR3_LOAD_EXITING;
1885         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1886
1887         if (cpu_has_secondary_exec_ctrls()) {
1888                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
1889                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1890                         exec_control &=
1891                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1892                 if (vmx->vpid == 0)
1893                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
1894                 if (!vm_need_ept())
1895                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
1896                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
1897         }
1898
1899         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1900         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
1901         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
1902
1903         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
1904         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
1905         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
1906
1907         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
1908         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1909         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1910         vmcs_write16(HOST_FS_SELECTOR, read_fs());    /* 22.2.4 */
1911         vmcs_write16(HOST_GS_SELECTOR, read_gs());    /* 22.2.4 */
1912         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1913 #ifdef CONFIG_X86_64
1914         rdmsrl(MSR_FS_BASE, a);
1915         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1916         rdmsrl(MSR_GS_BASE, a);
1917         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1918 #else
1919         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1920         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1921 #endif
1922
1923         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
1924
1925         get_idt(&dt);
1926         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
1927
1928         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1929         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1930         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1931         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1932         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1933
1934         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1935         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1936         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1937         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
1938         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1939         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
1940
1941         for (i = 0; i < NR_VMX_MSR; ++i) {
1942                 u32 index = vmx_msr_index[i];
1943                 u32 data_low, data_high;
1944                 u64 data;
1945                 int j = vmx->nmsrs;
1946
1947                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1948                         continue;
1949                 if (wrmsr_safe(index, data_low, data_high) < 0)
1950                         continue;
1951                 data = data_low | ((u64)data_high << 32);
1952                 vmx->host_msrs[j].index = index;
1953                 vmx->host_msrs[j].reserved = 0;
1954                 vmx->host_msrs[j].data = data;
1955                 vmx->guest_msrs[j] = vmx->host_msrs[j];
1956                 ++vmx->nmsrs;
1957         }
1958
1959         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
1960
1961         /* 22.2.1, 20.8.1 */
1962         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1963
1964         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1965         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1966
1967
1968         return 0;
1969 }
1970
1971 static int init_rmode(struct kvm *kvm)
1972 {
1973         if (!init_rmode_tss(kvm))
1974                 return 0;
1975         if (!init_rmode_identity_map(kvm))
1976                 return 0;
1977         return 1;
1978 }
1979
1980 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1981 {
1982         struct vcpu_vmx *vmx = to_vmx(vcpu);
1983         u64 msr;
1984         int ret;
1985
1986         down_read(&vcpu->kvm->slots_lock);
1987         if (!init_rmode(vmx->vcpu.kvm)) {
1988                 ret = -ENOMEM;
1989                 goto out;
1990         }
1991
1992         vmx->vcpu.arch.rmode.active = 0;
1993
1994         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
1995         kvm_set_cr8(&vmx->vcpu, 0);
1996         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1997         if (vmx->vcpu.vcpu_id == 0)
1998                 msr |= MSR_IA32_APICBASE_BSP;
1999         kvm_set_apic_base(&vmx->vcpu, msr);
2000
2001         fx_init(&vmx->vcpu);
2002
2003         /*
2004          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2005          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2006          */
2007         if (vmx->vcpu.vcpu_id == 0) {
2008                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2009                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2010         } else {
2011                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2012                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2013         }
2014         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2015         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2016
2017         seg_setup(VCPU_SREG_DS);
2018         seg_setup(VCPU_SREG_ES);
2019         seg_setup(VCPU_SREG_FS);
2020         seg_setup(VCPU_SREG_GS);
2021         seg_setup(VCPU_SREG_SS);
2022
2023         vmcs_write16(GUEST_TR_SELECTOR, 0);
2024         vmcs_writel(GUEST_TR_BASE, 0);
2025         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2026         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2027
2028         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2029         vmcs_writel(GUEST_LDTR_BASE, 0);
2030         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2031         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2032
2033         vmcs_write32(GUEST_SYSENTER_CS, 0);
2034         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2035         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2036
2037         vmcs_writel(GUEST_RFLAGS, 0x02);
2038         if (vmx->vcpu.vcpu_id == 0)
2039                 vmcs_writel(GUEST_RIP, 0xfff0);
2040         else
2041                 vmcs_writel(GUEST_RIP, 0);
2042         vmcs_writel(GUEST_RSP, 0);
2043
2044         /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
2045         vmcs_writel(GUEST_DR7, 0x400);
2046
2047         vmcs_writel(GUEST_GDTR_BASE, 0);
2048         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2049
2050         vmcs_writel(GUEST_IDTR_BASE, 0);
2051         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2052
2053         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2054         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2055         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2056
2057         guest_write_tsc(0);
2058
2059         /* Special registers */
2060         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2061
2062         setup_msrs(vmx);
2063
2064         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2065
2066         if (cpu_has_vmx_tpr_shadow()) {
2067                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2068                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2069                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2070                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2071                 vmcs_write32(TPR_THRESHOLD, 0);
2072         }
2073
2074         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2075                 vmcs_write64(APIC_ACCESS_ADDR,
2076                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2077
2078         if (vmx->vpid != 0)
2079                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2080
2081         vmx->vcpu.arch.cr0 = 0x60000010;
2082         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2083         vmx_set_cr4(&vmx->vcpu, 0);
2084         vmx_set_efer(&vmx->vcpu, 0);
2085         vmx_fpu_activate(&vmx->vcpu);
2086         update_exception_bitmap(&vmx->vcpu);
2087
2088         vpid_sync_vcpu_all(vmx);
2089
2090         ret = 0;
2091
2092 out:
2093         up_read(&vcpu->kvm->slots_lock);
2094         return ret;
2095 }
2096
2097 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2098 {
2099         struct vcpu_vmx *vmx = to_vmx(vcpu);
2100
2101         KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2102
2103         if (vcpu->arch.rmode.active) {
2104                 vmx->rmode.irq.pending = true;
2105                 vmx->rmode.irq.vector = irq;
2106                 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP);
2107                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2108                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2109                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2110                 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1);
2111                 return;
2112         }
2113         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2114                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
2115 }
2116
2117 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2118 {
2119         int word_index = __ffs(vcpu->arch.irq_summary);
2120         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2121         int irq = word_index * BITS_PER_LONG + bit_index;
2122
2123         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2124         if (!vcpu->arch.irq_pending[word_index])
2125                 clear_bit(word_index, &vcpu->arch.irq_summary);
2126         vmx_inject_irq(vcpu, irq);
2127 }
2128
2129
2130 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2131                                        struct kvm_run *kvm_run)
2132 {
2133         u32 cpu_based_vm_exec_control;
2134
2135         vcpu->arch.interrupt_window_open =
2136                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2137                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2138
2139         if (vcpu->arch.interrupt_window_open &&
2140             vcpu->arch.irq_summary &&
2141             !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
2142                 /*
2143                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
2144                  */
2145                 kvm_do_inject_irq(vcpu);
2146
2147         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2148         if (!vcpu->arch.interrupt_window_open &&
2149             (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
2150                 /*
2151                  * Interrupts blocked.  Wait for unblock.
2152                  */
2153                 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2154         else
2155                 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2156         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2157 }
2158
2159 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2160 {
2161         int ret;
2162         struct kvm_userspace_memory_region tss_mem = {
2163                 .slot = 8,
2164                 .guest_phys_addr = addr,
2165                 .memory_size = PAGE_SIZE * 3,
2166                 .flags = 0,
2167         };
2168
2169         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2170         if (ret)
2171                 return ret;
2172         kvm->arch.tss_addr = addr;
2173         return 0;
2174 }
2175
2176 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
2177 {
2178         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
2179
2180         set_debugreg(dbg->bp[0], 0);
2181         set_debugreg(dbg->bp[1], 1);
2182         set_debugreg(dbg->bp[2], 2);
2183         set_debugreg(dbg->bp[3], 3);
2184
2185         if (dbg->singlestep) {
2186                 unsigned long flags;
2187
2188                 flags = vmcs_readl(GUEST_RFLAGS);
2189                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
2190                 vmcs_writel(GUEST_RFLAGS, flags);
2191         }
2192 }
2193
2194 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2195                                   int vec, u32 err_code)
2196 {
2197         if (!vcpu->arch.rmode.active)
2198                 return 0;
2199
2200         /*
2201          * Instruction with address size override prefix opcode 0x67
2202          * Cause the #SS fault with 0 error code in VM86 mode.
2203          */
2204         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2205                 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2206                         return 1;
2207         return 0;
2208 }
2209
2210 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2211 {
2212         struct vcpu_vmx *vmx = to_vmx(vcpu);
2213         u32 intr_info, error_code;
2214         unsigned long cr2, rip;
2215         u32 vect_info;
2216         enum emulation_result er;
2217
2218         vect_info = vmx->idt_vectoring_info;
2219         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2220
2221         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2222                                                 !is_page_fault(intr_info))
2223                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2224                        "intr info 0x%x\n", __func__, vect_info, intr_info);
2225
2226         if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
2227                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
2228                 set_bit(irq, vcpu->arch.irq_pending);
2229                 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
2230         }
2231
2232         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2233                 return 1;  /* already handled by vmx_vcpu_run() */
2234
2235         if (is_no_device(intr_info)) {
2236                 vmx_fpu_activate(vcpu);
2237                 return 1;
2238         }
2239
2240         if (is_invalid_opcode(intr_info)) {
2241                 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2242                 if (er != EMULATE_DONE)
2243                         kvm_queue_exception(vcpu, UD_VECTOR);
2244                 return 1;
2245         }
2246
2247         error_code = 0;
2248         rip = vmcs_readl(GUEST_RIP);
2249         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2250                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2251         if (is_page_fault(intr_info)) {
2252                 /* EPT won't cause page fault directly */
2253                 if (vm_need_ept())
2254                         BUG();
2255                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2256                 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2257                             (u32)((u64)cr2 >> 32), handler);
2258                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2259         }
2260
2261         if (vcpu->arch.rmode.active &&
2262             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2263                                                                 error_code)) {
2264                 if (vcpu->arch.halt_request) {
2265                         vcpu->arch.halt_request = 0;
2266                         return kvm_emulate_halt(vcpu);
2267                 }
2268                 return 1;
2269         }
2270
2271         if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
2272             (INTR_TYPE_EXCEPTION | 1)) {
2273                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2274                 return 0;
2275         }
2276         kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2277         kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
2278         kvm_run->ex.error_code = error_code;
2279         return 0;
2280 }
2281
2282 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2283                                      struct kvm_run *kvm_run)
2284 {
2285         ++vcpu->stat.irq_exits;
2286         KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
2287         return 1;
2288 }
2289
2290 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2291 {
2292         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2293         return 0;
2294 }
2295
2296 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2297 {
2298         unsigned long exit_qualification;
2299         int size, down, in, string, rep;
2300         unsigned port;
2301
2302         ++vcpu->stat.io_exits;
2303         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2304         string = (exit_qualification & 16) != 0;
2305
2306         if (string) {
2307                 if (emulate_instruction(vcpu,
2308                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2309                         return 0;
2310                 return 1;
2311         }
2312
2313         size = (exit_qualification & 7) + 1;
2314         in = (exit_qualification & 8) != 0;
2315         down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
2316         rep = (exit_qualification & 32) != 0;
2317         port = exit_qualification >> 16;
2318
2319         return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2320 }
2321
2322 static void
2323 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2324 {
2325         /*
2326          * Patch in the VMCALL instruction:
2327          */
2328         hypercall[0] = 0x0f;
2329         hypercall[1] = 0x01;
2330         hypercall[2] = 0xc1;
2331 }
2332
2333 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2334 {
2335         unsigned long exit_qualification;
2336         int cr;
2337         int reg;
2338
2339         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2340         cr = exit_qualification & 15;
2341         reg = (exit_qualification >> 8) & 15;
2342         switch ((exit_qualification >> 4) & 3) {
2343         case 0: /* mov to cr */
2344                 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)vcpu->arch.regs[reg],
2345                             (u32)((u64)vcpu->arch.regs[reg] >> 32), handler);
2346                 switch (cr) {
2347                 case 0:
2348                         vcpu_load_rsp_rip(vcpu);
2349                         kvm_set_cr0(vcpu, vcpu->arch.regs[reg]);
2350                         skip_emulated_instruction(vcpu);
2351                         return 1;
2352                 case 3:
2353                         vcpu_load_rsp_rip(vcpu);
2354                         kvm_set_cr3(vcpu, vcpu->arch.regs[reg]);
2355                         skip_emulated_instruction(vcpu);
2356                         return 1;
2357                 case 4:
2358                         vcpu_load_rsp_rip(vcpu);
2359                         kvm_set_cr4(vcpu, vcpu->arch.regs[reg]);
2360                         skip_emulated_instruction(vcpu);
2361                         return 1;
2362                 case 8:
2363                         vcpu_load_rsp_rip(vcpu);
2364                         kvm_set_cr8(vcpu, vcpu->arch.regs[reg]);
2365                         skip_emulated_instruction(vcpu);
2366                         if (irqchip_in_kernel(vcpu->kvm))
2367                                 return 1;
2368                         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2369                         return 0;
2370                 };
2371                 break;
2372         case 2: /* clts */
2373                 vcpu_load_rsp_rip(vcpu);
2374                 vmx_fpu_deactivate(vcpu);
2375                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2376                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2377                 vmx_fpu_activate(vcpu);
2378                 KVMTRACE_0D(CLTS, vcpu, handler);
2379                 skip_emulated_instruction(vcpu);
2380                 return 1;
2381         case 1: /*mov from cr*/
2382                 switch (cr) {
2383                 case 3:
2384                         vcpu_load_rsp_rip(vcpu);
2385                         vcpu->arch.regs[reg] = vcpu->arch.cr3;
2386                         vcpu_put_rsp_rip(vcpu);
2387                         KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2388                                     (u32)vcpu->arch.regs[reg],
2389                                     (u32)((u64)vcpu->arch.regs[reg] >> 32),
2390                                     handler);
2391                         skip_emulated_instruction(vcpu);
2392                         return 1;
2393                 case 8:
2394                         vcpu_load_rsp_rip(vcpu);
2395                         vcpu->arch.regs[reg] = kvm_get_cr8(vcpu);
2396                         vcpu_put_rsp_rip(vcpu);
2397                         KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2398                                     (u32)vcpu->arch.regs[reg], handler);
2399                         skip_emulated_instruction(vcpu);
2400                         return 1;
2401                 }
2402                 break;
2403         case 3: /* lmsw */
2404                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2405
2406                 skip_emulated_instruction(vcpu);
2407                 return 1;
2408         default:
2409                 break;
2410         }
2411         kvm_run->exit_reason = 0;
2412         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2413                (int)(exit_qualification >> 4) & 3, cr);
2414         return 0;
2415 }
2416
2417 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2418 {
2419         unsigned long exit_qualification;
2420         unsigned long val;
2421         int dr, reg;
2422
2423         /*
2424          * FIXME: this code assumes the host is debugging the guest.
2425          *        need to deal with guest debugging itself too.
2426          */
2427         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2428         dr = exit_qualification & 7;
2429         reg = (exit_qualification >> 8) & 15;
2430         vcpu_load_rsp_rip(vcpu);
2431         if (exit_qualification & 16) {
2432                 /* mov from dr */
2433                 switch (dr) {
2434                 case 6:
2435                         val = 0xffff0ff0;
2436                         break;
2437                 case 7:
2438                         val = 0x400;
2439                         break;
2440                 default:
2441                         val = 0;
2442                 }
2443                 vcpu->arch.regs[reg] = val;
2444                 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2445         } else {
2446                 /* mov to dr */
2447         }
2448         vcpu_put_rsp_rip(vcpu);
2449         skip_emulated_instruction(vcpu);
2450         return 1;
2451 }
2452
2453 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2454 {
2455         kvm_emulate_cpuid(vcpu);
2456         return 1;
2457 }
2458
2459 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2460 {
2461         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2462         u64 data;
2463
2464         if (vmx_get_msr(vcpu, ecx, &data)) {
2465                 kvm_inject_gp(vcpu, 0);
2466                 return 1;
2467         }
2468
2469         KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2470                     handler);
2471
2472         /* FIXME: handling of bits 32:63 of rax, rdx */
2473         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2474         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2475         skip_emulated_instruction(vcpu);
2476         return 1;
2477 }
2478
2479 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2480 {
2481         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2482         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2483                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2484
2485         KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2486                     handler);
2487
2488         if (vmx_set_msr(vcpu, ecx, data) != 0) {
2489                 kvm_inject_gp(vcpu, 0);
2490                 return 1;
2491         }
2492
2493         skip_emulated_instruction(vcpu);
2494         return 1;
2495 }
2496
2497 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2498                                       struct kvm_run *kvm_run)
2499 {
2500         return 1;
2501 }
2502
2503 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2504                                    struct kvm_run *kvm_run)
2505 {
2506         u32 cpu_based_vm_exec_control;
2507
2508         /* clear pending irq */
2509         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2510         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2511         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2512
2513         KVMTRACE_0D(PEND_INTR, vcpu, handler);
2514
2515         /*
2516          * If the user space waits to inject interrupts, exit as soon as
2517          * possible
2518          */
2519         if (kvm_run->request_interrupt_window &&
2520             !vcpu->arch.irq_summary) {
2521                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2522                 ++vcpu->stat.irq_window_exits;
2523                 return 0;
2524         }
2525         return 1;
2526 }
2527
2528 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2529 {
2530         skip_emulated_instruction(vcpu);
2531         return kvm_emulate_halt(vcpu);
2532 }
2533
2534 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2535 {
2536         skip_emulated_instruction(vcpu);
2537         kvm_emulate_hypercall(vcpu);
2538         return 1;
2539 }
2540
2541 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2542 {
2543         skip_emulated_instruction(vcpu);
2544         /* TODO: Add support for VT-d/pass-through device */
2545         return 1;
2546 }
2547
2548 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2549 {
2550         u64 exit_qualification;
2551         enum emulation_result er;
2552         unsigned long offset;
2553
2554         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2555         offset = exit_qualification & 0xffful;
2556
2557         KVMTRACE_1D(APIC_ACCESS, vcpu, (u32)offset, handler);
2558
2559         er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2560
2561         if (er !=  EMULATE_DONE) {
2562                 printk(KERN_ERR
2563                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2564                        offset);
2565                 return -ENOTSUPP;
2566         }
2567         return 1;
2568 }
2569
2570 static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2571 {
2572         unsigned long exit_qualification;
2573         u16 tss_selector;
2574         int reason;
2575
2576         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2577
2578         reason = (u32)exit_qualification >> 30;
2579         tss_selector = exit_qualification;
2580
2581         return kvm_task_switch(vcpu, tss_selector, reason);
2582 }
2583
2584 static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2585 {
2586         u64 exit_qualification;
2587         enum emulation_result er;
2588         gpa_t gpa;
2589         unsigned long hva;
2590         int gla_validity;
2591         int r;
2592
2593         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2594
2595         if (exit_qualification & (1 << 6)) {
2596                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
2597                 return -ENOTSUPP;
2598         }
2599
2600         gla_validity = (exit_qualification >> 7) & 0x3;
2601         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
2602                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
2603                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2604                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2605                         (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2606                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2607                         (long unsigned int)exit_qualification);
2608                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2609                 kvm_run->hw.hardware_exit_reason = 0;
2610                 return -ENOTSUPP;
2611         }
2612
2613         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
2614         hva = gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT);
2615         if (!kvm_is_error_hva(hva)) {
2616                 r = kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
2617                 if (r < 0) {
2618                         printk(KERN_ERR "EPT: Not enough memory!\n");
2619                         return -ENOMEM;
2620                 }
2621                 return 1;
2622         } else {
2623                 /* must be MMIO */
2624                 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2625
2626                 if (er == EMULATE_FAIL) {
2627                         printk(KERN_ERR
2628                          "EPT: Fail to handle EPT violation vmexit!er is %d\n",
2629                          er);
2630                         printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
2631                          (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
2632                          (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
2633                         printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
2634                                 (long unsigned int)exit_qualification);
2635                         return -ENOTSUPP;
2636                 } else if (er == EMULATE_DO_MMIO)
2637                         return 0;
2638         }
2639         return 1;
2640 }
2641
2642 /*
2643  * The exit handlers return 1 if the exit was handled fully and guest execution
2644  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
2645  * to be done to userspace and return 0.
2646  */
2647 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2648                                       struct kvm_run *kvm_run) = {
2649         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
2650         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
2651         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
2652         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
2653         [EXIT_REASON_CR_ACCESS]               = handle_cr,
2654         [EXIT_REASON_DR_ACCESS]               = handle_dr,
2655         [EXIT_REASON_CPUID]                   = handle_cpuid,
2656         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
2657         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
2658         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
2659         [EXIT_REASON_HLT]                     = handle_halt,
2660         [EXIT_REASON_VMCALL]                  = handle_vmcall,
2661         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
2662         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
2663         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
2664         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
2665         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
2666 };
2667
2668 static const int kvm_vmx_max_exit_handlers =
2669         ARRAY_SIZE(kvm_vmx_exit_handlers);
2670
2671 /*
2672  * The guest has exited.  See if we can fix it or if we need userspace
2673  * assistance.
2674  */
2675 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2676 {
2677         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2678         struct vcpu_vmx *vmx = to_vmx(vcpu);
2679         u32 vectoring_info = vmx->idt_vectoring_info;
2680
2681         KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)vmcs_readl(GUEST_RIP),
2682                     (u32)((u64)vmcs_readl(GUEST_RIP) >> 32), entryexit);
2683
2684         /* Access CR3 don't cause VMExit in paging mode, so we need
2685          * to sync with guest real CR3. */
2686         if (vm_need_ept() && is_paging(vcpu)) {
2687                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2688                 ept_load_pdptrs(vcpu);
2689         }
2690
2691         if (unlikely(vmx->fail)) {
2692                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2693                 kvm_run->fail_entry.hardware_entry_failure_reason
2694                         = vmcs_read32(VM_INSTRUCTION_ERROR);
2695                 return 0;
2696         }
2697
2698         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2699                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
2700                         exit_reason != EXIT_REASON_EPT_VIOLATION))
2701                 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
2702                        "exit reason is 0x%x\n", __func__, exit_reason);
2703         if (exit_reason < kvm_vmx_max_exit_handlers
2704             && kvm_vmx_exit_handlers[exit_reason])
2705                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2706         else {
2707                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2708                 kvm_run->hw.hardware_exit_reason = exit_reason;
2709         }
2710         return 0;
2711 }
2712
2713 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2714 {
2715         int max_irr, tpr;
2716
2717         if (!vm_need_tpr_shadow(vcpu->kvm))
2718                 return;
2719
2720         if (!kvm_lapic_enabled(vcpu) ||
2721             ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2722                 vmcs_write32(TPR_THRESHOLD, 0);
2723                 return;
2724         }
2725
2726         tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2727         vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2728 }
2729
2730 static void enable_irq_window(struct kvm_vcpu *vcpu)
2731 {
2732         u32 cpu_based_vm_exec_control;
2733
2734         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2735         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2736         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2737 }
2738
2739 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2740 {
2741         struct vcpu_vmx *vmx = to_vmx(vcpu);
2742         u32 idtv_info_field, intr_info_field;
2743         int has_ext_irq, interrupt_window_open;
2744         int vector;
2745
2746         update_tpr_threshold(vcpu);
2747
2748         has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2749         intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
2750         idtv_info_field = vmx->idt_vectoring_info;
2751         if (intr_info_field & INTR_INFO_VALID_MASK) {
2752                 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2753                         /* TODO: fault when IDT_Vectoring */
2754                         if (printk_ratelimit())
2755                                 printk(KERN_ERR "Fault when IDT_Vectoring\n");
2756                 }
2757                 if (has_ext_irq)
2758                         enable_irq_window(vcpu);
2759                 return;
2760         }
2761         if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2762                 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2763                     == INTR_TYPE_EXT_INTR
2764                     && vcpu->arch.rmode.active) {
2765                         u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2766
2767                         vmx_inject_irq(vcpu, vect);
2768                         if (unlikely(has_ext_irq))
2769                                 enable_irq_window(vcpu);
2770                         return;
2771                 }
2772
2773                 KVMTRACE_1D(REDELIVER_EVT, vcpu, idtv_info_field, handler);
2774
2775                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2776                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2777                                 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2778
2779                 if (unlikely(idtv_info_field & INTR_INFO_DELIVER_CODE_MASK))
2780                         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2781                                 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2782                 if (unlikely(has_ext_irq))
2783                         enable_irq_window(vcpu);
2784                 return;
2785         }
2786         if (!has_ext_irq)
2787                 return;
2788         interrupt_window_open =
2789                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2790                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2791         if (interrupt_window_open) {
2792                 vector = kvm_cpu_get_interrupt(vcpu);
2793                 vmx_inject_irq(vcpu, vector);
2794                 kvm_timer_intr_post(vcpu, vector);
2795         } else
2796                 enable_irq_window(vcpu);
2797 }
2798
2799 /*
2800  * Failure to inject an interrupt should give us the information
2801  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
2802  * when fetching the interrupt redirection bitmap in the real-mode
2803  * tss, this doesn't happen.  So we do it ourselves.
2804  */
2805 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2806 {
2807         vmx->rmode.irq.pending = 0;
2808         if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip)
2809                 return;
2810         vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip);
2811         if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2812                 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2813                 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
2814                 return;
2815         }
2816         vmx->idt_vectoring_info =
2817                 VECTORING_INFO_VALID_MASK
2818                 | INTR_TYPE_EXT_INTR
2819                 | vmx->rmode.irq.vector;
2820 }
2821
2822 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2823 {
2824         struct vcpu_vmx *vmx = to_vmx(vcpu);
2825         u32 intr_info;
2826
2827         /*
2828          * Loading guest fpu may have cleared host cr0.ts
2829          */
2830         vmcs_writel(HOST_CR0, read_cr0());
2831
2832         asm(
2833                 /* Store host registers */
2834 #ifdef CONFIG_X86_64
2835                 "push %%rdx; push %%rbp;"
2836                 "push %%rcx \n\t"
2837 #else
2838                 "push %%edx; push %%ebp;"
2839                 "push %%ecx \n\t"
2840 #endif
2841                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2842                 /* Check if vmlaunch of vmresume is needed */
2843                 "cmpl $0, %c[launched](%0) \n\t"
2844                 /* Load guest registers.  Don't clobber flags. */
2845 #ifdef CONFIG_X86_64
2846                 "mov %c[cr2](%0), %%rax \n\t"
2847                 "mov %%rax, %%cr2 \n\t"
2848                 "mov %c[rax](%0), %%rax \n\t"
2849                 "mov %c[rbx](%0), %%rbx \n\t"
2850                 "mov %c[rdx](%0), %%rdx \n\t"
2851                 "mov %c[rsi](%0), %%rsi \n\t"
2852                 "mov %c[rdi](%0), %%rdi \n\t"
2853                 "mov %c[rbp](%0), %%rbp \n\t"
2854                 "mov %c[r8](%0),  %%r8  \n\t"
2855                 "mov %c[r9](%0),  %%r9  \n\t"
2856                 "mov %c[r10](%0), %%r10 \n\t"
2857                 "mov %c[r11](%0), %%r11 \n\t"
2858                 "mov %c[r12](%0), %%r12 \n\t"
2859                 "mov %c[r13](%0), %%r13 \n\t"
2860                 "mov %c[r14](%0), %%r14 \n\t"
2861                 "mov %c[r15](%0), %%r15 \n\t"
2862                 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
2863 #else
2864                 "mov %c[cr2](%0), %%eax \n\t"
2865                 "mov %%eax,   %%cr2 \n\t"
2866                 "mov %c[rax](%0), %%eax \n\t"
2867                 "mov %c[rbx](%0), %%ebx \n\t"
2868                 "mov %c[rdx](%0), %%edx \n\t"
2869                 "mov %c[rsi](%0), %%esi \n\t"
2870                 "mov %c[rdi](%0), %%edi \n\t"
2871                 "mov %c[rbp](%0), %%ebp \n\t"
2872                 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
2873 #endif
2874                 /* Enter guest mode */
2875                 "jne .Llaunched \n\t"
2876                 ASM_VMX_VMLAUNCH "\n\t"
2877                 "jmp .Lkvm_vmx_return \n\t"
2878                 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2879                 ".Lkvm_vmx_return: "
2880                 /* Save guest registers, load host registers, keep flags */
2881 #ifdef CONFIG_X86_64
2882                 "xchg %0,     (%%rsp) \n\t"
2883                 "mov %%rax, %c[rax](%0) \n\t"
2884                 "mov %%rbx, %c[rbx](%0) \n\t"
2885                 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
2886                 "mov %%rdx, %c[rdx](%0) \n\t"
2887                 "mov %%rsi, %c[rsi](%0) \n\t"
2888                 "mov %%rdi, %c[rdi](%0) \n\t"
2889                 "mov %%rbp, %c[rbp](%0) \n\t"
2890                 "mov %%r8,  %c[r8](%0) \n\t"
2891                 "mov %%r9,  %c[r9](%0) \n\t"
2892                 "mov %%r10, %c[r10](%0) \n\t"
2893                 "mov %%r11, %c[r11](%0) \n\t"
2894                 "mov %%r12, %c[r12](%0) \n\t"
2895                 "mov %%r13, %c[r13](%0) \n\t"
2896                 "mov %%r14, %c[r14](%0) \n\t"
2897                 "mov %%r15, %c[r15](%0) \n\t"
2898                 "mov %%cr2, %%rax   \n\t"
2899                 "mov %%rax, %c[cr2](%0) \n\t"
2900
2901                 "pop  %%rbp; pop  %%rbp; pop  %%rdx \n\t"
2902 #else
2903                 "xchg %0, (%%esp) \n\t"
2904                 "mov %%eax, %c[rax](%0) \n\t"
2905                 "mov %%ebx, %c[rbx](%0) \n\t"
2906                 "pushl (%%esp); popl %c[rcx](%0) \n\t"
2907                 "mov %%edx, %c[rdx](%0) \n\t"
2908                 "mov %%esi, %c[rsi](%0) \n\t"
2909                 "mov %%edi, %c[rdi](%0) \n\t"
2910                 "mov %%ebp, %c[rbp](%0) \n\t"
2911                 "mov %%cr2, %%eax  \n\t"
2912                 "mov %%eax, %c[cr2](%0) \n\t"
2913
2914                 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
2915 #endif
2916                 "setbe %c[fail](%0) \n\t"
2917               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
2918                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
2919                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
2920                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
2921                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
2922                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
2923                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
2924                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
2925                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
2926                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
2927 #ifdef CONFIG_X86_64
2928                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
2929                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
2930                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
2931                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
2932                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
2933                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
2934                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
2935                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
2936 #endif
2937                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
2938               : "cc", "memory"
2939 #ifdef CONFIG_X86_64
2940                 , "rbx", "rdi", "rsi"
2941                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
2942 #else
2943                 , "ebx", "edi", "rsi"
2944 #endif
2945               );
2946
2947         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2948         if (vmx->rmode.irq.pending)
2949                 fixup_rmode_irq(vmx);
2950
2951         vcpu->arch.interrupt_window_open =
2952                 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2953
2954         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2955         vmx->launched = 1;
2956
2957         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2958
2959         /* We need to handle NMIs before interrupts are enabled */
2960         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
2961                 KVMTRACE_0D(NMI, vcpu, handler);
2962                 asm("int $2");
2963         }
2964 }
2965
2966 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2967 {
2968         struct vcpu_vmx *vmx = to_vmx(vcpu);
2969
2970         if (vmx->vmcs) {
2971                 on_each_cpu(__vcpu_clear, vmx, 0, 1);
2972                 free_vmcs(vmx->vmcs);
2973                 vmx->vmcs = NULL;
2974         }
2975 }
2976
2977 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2978 {
2979         struct vcpu_vmx *vmx = to_vmx(vcpu);
2980
2981         spin_lock(&vmx_vpid_lock);
2982         if (vmx->vpid != 0)
2983                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2984         spin_unlock(&vmx_vpid_lock);
2985         vmx_free_vmcs(vcpu);
2986         kfree(vmx->host_msrs);
2987         kfree(vmx->guest_msrs);
2988         kvm_vcpu_uninit(vcpu);
2989         kmem_cache_free(kvm_vcpu_cache, vmx);
2990 }
2991
2992 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2993 {
2994         int err;
2995         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2996         int cpu;
2997
2998         if (!vmx)
2999                 return ERR_PTR(-ENOMEM);
3000
3001         allocate_vpid(vmx);
3002         if (id == 0 && vm_need_ept()) {
3003                 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3004                         VMX_EPT_WRITABLE_MASK |
3005                         VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT);
3006                 kvm_mmu_set_mask_ptes(0ull, VMX_EPT_FAKE_ACCESSED_MASK,
3007                                 VMX_EPT_FAKE_DIRTY_MASK, 0ull,
3008                                 VMX_EPT_EXECUTABLE_MASK);
3009                 kvm_enable_tdp();
3010         }
3011
3012         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3013         if (err)
3014                 goto free_vcpu;
3015
3016         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3017         if (!vmx->guest_msrs) {
3018                 err = -ENOMEM;
3019                 goto uninit_vcpu;
3020         }
3021
3022         vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3023         if (!vmx->host_msrs)
3024                 goto free_guest_msrs;
3025
3026         vmx->vmcs = alloc_vmcs();
3027         if (!vmx->vmcs)
3028                 goto free_msrs;
3029
3030         vmcs_clear(vmx->vmcs);
3031
3032         cpu = get_cpu();
3033         vmx_vcpu_load(&vmx->vcpu, cpu);
3034         err = vmx_vcpu_setup(vmx);
3035         vmx_vcpu_put(&vmx->vcpu);
3036         put_cpu();
3037         if (err)
3038                 goto free_vmcs;
3039         if (vm_need_virtualize_apic_accesses(kvm))
3040                 if (alloc_apic_access_page(kvm) != 0)
3041                         goto free_vmcs;
3042
3043         if (vm_need_ept())
3044                 if (alloc_identity_pagetable(kvm) != 0)
3045                         goto free_vmcs;
3046
3047         return &vmx->vcpu;
3048
3049 free_vmcs:
3050         free_vmcs(vmx->vmcs);
3051 free_msrs:
3052         kfree(vmx->host_msrs);
3053 free_guest_msrs:
3054         kfree(vmx->guest_msrs);
3055 uninit_vcpu:
3056         kvm_vcpu_uninit(&vmx->vcpu);
3057 free_vcpu:
3058         kmem_cache_free(kvm_vcpu_cache, vmx);
3059         return ERR_PTR(err);
3060 }
3061
3062 static void __init vmx_check_processor_compat(void *rtn)
3063 {
3064         struct vmcs_config vmcs_conf;
3065
3066         *(int *)rtn = 0;
3067         if (setup_vmcs_config(&vmcs_conf) < 0)
3068                 *(int *)rtn = -EIO;
3069         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3070                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3071                                 smp_processor_id());
3072                 *(int *)rtn = -EIO;
3073         }
3074 }
3075
3076 static int get_ept_level(void)
3077 {
3078         return VMX_EPT_DEFAULT_GAW + 1;
3079 }
3080
3081 static struct kvm_x86_ops vmx_x86_ops = {
3082         .cpu_has_kvm_support = cpu_has_kvm_support,
3083         .disabled_by_bios = vmx_disabled_by_bios,
3084         .hardware_setup = hardware_setup,
3085         .hardware_unsetup = hardware_unsetup,
3086         .check_processor_compatibility = vmx_check_processor_compat,
3087         .hardware_enable = hardware_enable,
3088         .hardware_disable = hardware_disable,
3089         .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
3090
3091         .vcpu_create = vmx_create_vcpu,
3092         .vcpu_free = vmx_free_vcpu,
3093         .vcpu_reset = vmx_vcpu_reset,
3094
3095         .prepare_guest_switch = vmx_save_host_state,
3096         .vcpu_load = vmx_vcpu_load,
3097         .vcpu_put = vmx_vcpu_put,
3098         .vcpu_decache = vmx_vcpu_decache,
3099
3100         .set_guest_debug = set_guest_debug,
3101         .guest_debug_pre = kvm_guest_debug_pre,
3102         .get_msr = vmx_get_msr,
3103         .set_msr = vmx_set_msr,
3104         .get_segment_base = vmx_get_segment_base,
3105         .get_segment = vmx_get_segment,
3106         .set_segment = vmx_set_segment,
3107         .get_cpl = vmx_get_cpl,
3108         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3109         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3110         .set_cr0 = vmx_set_cr0,
3111         .set_cr3 = vmx_set_cr3,
3112         .set_cr4 = vmx_set_cr4,
3113         .set_efer = vmx_set_efer,
3114         .get_idt = vmx_get_idt,
3115         .set_idt = vmx_set_idt,
3116         .get_gdt = vmx_get_gdt,
3117         .set_gdt = vmx_set_gdt,
3118         .cache_regs = vcpu_load_rsp_rip,
3119         .decache_regs = vcpu_put_rsp_rip,
3120         .get_rflags = vmx_get_rflags,
3121         .set_rflags = vmx_set_rflags,
3122
3123         .tlb_flush = vmx_flush_tlb,
3124
3125         .run = vmx_vcpu_run,
3126         .handle_exit = kvm_handle_exit,
3127         .skip_emulated_instruction = skip_emulated_instruction,
3128         .patch_hypercall = vmx_patch_hypercall,
3129         .get_irq = vmx_get_irq,
3130         .set_irq = vmx_inject_irq,
3131         .queue_exception = vmx_queue_exception,
3132         .exception_injected = vmx_exception_injected,
3133         .inject_pending_irq = vmx_intr_assist,
3134         .inject_pending_vectors = do_interrupt_requests,
3135
3136         .set_tss_addr = vmx_set_tss_addr,
3137         .get_tdp_level = get_ept_level,
3138 };
3139
3140 static int __init vmx_init(void)
3141 {
3142         void *va;
3143         int r;
3144
3145         vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3146         if (!vmx_io_bitmap_a)
3147                 return -ENOMEM;
3148
3149         vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3150         if (!vmx_io_bitmap_b) {
3151                 r = -ENOMEM;
3152                 goto out;
3153         }
3154
3155         vmx_msr_bitmap = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
3156         if (!vmx_msr_bitmap) {
3157                 r = -ENOMEM;
3158                 goto out1;
3159         }
3160
3161         /*
3162          * Allow direct access to the PC debug port (it is often used for I/O
3163          * delays, but the vmexits simply slow things down).
3164          */
3165         va = kmap(vmx_io_bitmap_a);
3166         memset(va, 0xff, PAGE_SIZE);
3167         clear_bit(0x80, va);
3168         kunmap(vmx_io_bitmap_a);
3169
3170         va = kmap(vmx_io_bitmap_b);
3171         memset(va, 0xff, PAGE_SIZE);
3172         kunmap(vmx_io_bitmap_b);
3173
3174         va = kmap(vmx_msr_bitmap);
3175         memset(va, 0xff, PAGE_SIZE);
3176         kunmap(vmx_msr_bitmap);
3177
3178         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3179
3180         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
3181         if (r)
3182                 goto out2;
3183
3184         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_FS_BASE);
3185         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_GS_BASE);
3186         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_CS);
3187         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_ESP);
3188         vmx_disable_intercept_for_msr(vmx_msr_bitmap, MSR_IA32_SYSENTER_EIP);
3189
3190         if (cpu_has_vmx_ept())
3191                 bypass_guest_pf = 0;
3192
3193         if (bypass_guest_pf)
3194                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3195
3196         ept_sync_global();
3197
3198         return 0;
3199
3200 out2:
3201         __free_page(vmx_msr_bitmap);
3202 out1:
3203         __free_page(vmx_io_bitmap_b);
3204 out:
3205         __free_page(vmx_io_bitmap_a);
3206         return r;
3207 }
3208
3209 static void __exit vmx_exit(void)
3210 {
3211         __free_page(vmx_msr_bitmap);
3212         __free_page(vmx_io_bitmap_b);
3213         __free_page(vmx_io_bitmap_a);
3214
3215         kvm_exit();
3216 }
3217
3218 module_init(vmx_init)
3219 module_exit(vmx_exit)