32d6ae8fb60e339eba6b85968676776d44293c8b
[linux-2.6.git] / 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 "mmu.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/mm.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
27 #include <linux/moduleparam.h>
28 #include "kvm_cache_regs.h"
29 #include "x86.h"
30
31 #include <asm/io.h>
32 #include <asm/desc.h>
33 #include <asm/vmx.h>
34 #include <asm/virtext.h>
35 #include <asm/mce.h>
36
37 #define __ex(x) __kvm_handle_fault_on_reboot(x)
38
39 MODULE_AUTHOR("Qumranet");
40 MODULE_LICENSE("GPL");
41
42 static int __read_mostly bypass_guest_pf = 1;
43 module_param(bypass_guest_pf, bool, S_IRUGO);
44
45 static int __read_mostly enable_vpid = 1;
46 module_param_named(vpid, enable_vpid, bool, 0444);
47
48 static int __read_mostly flexpriority_enabled = 1;
49 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
50
51 static int __read_mostly enable_ept = 1;
52 module_param_named(ept, enable_ept, bool, S_IRUGO);
53
54 static int __read_mostly emulate_invalid_guest_state = 0;
55 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
56
57 struct vmcs {
58         u32 revision_id;
59         u32 abort;
60         char data[0];
61 };
62
63 struct vcpu_vmx {
64         struct kvm_vcpu       vcpu;
65         struct list_head      local_vcpus_link;
66         unsigned long         host_rsp;
67         int                   launched;
68         u8                    fail;
69         u32                   idt_vectoring_info;
70         struct kvm_msr_entry *guest_msrs;
71         struct kvm_msr_entry *host_msrs;
72         int                   nmsrs;
73         int                   save_nmsrs;
74         int                   msr_offset_efer;
75 #ifdef CONFIG_X86_64
76         int                   msr_offset_kernel_gs_base;
77 #endif
78         struct vmcs          *vmcs;
79         struct {
80                 int           loaded;
81                 u16           fs_sel, gs_sel, ldt_sel;
82                 int           gs_ldt_reload_needed;
83                 int           fs_reload_needed;
84                 int           guest_efer_loaded;
85         } host_state;
86         struct {
87                 struct {
88                         bool pending;
89                         u8 vector;
90                         unsigned rip;
91                 } irq;
92         } rmode;
93         int vpid;
94         bool emulation_required;
95         enum emulation_result invalid_state_emulation_result;
96
97         /* Support for vnmi-less CPUs */
98         int soft_vnmi_blocked;
99         ktime_t entry_time;
100         s64 vnmi_blocked_time;
101         u32 exit_reason;
102 };
103
104 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
105 {
106         return container_of(vcpu, struct vcpu_vmx, vcpu);
107 }
108
109 static int init_rmode(struct kvm *kvm);
110 static u64 construct_eptp(unsigned long root_hpa);
111
112 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
113 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
114 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
115
116 static unsigned long *vmx_io_bitmap_a;
117 static unsigned long *vmx_io_bitmap_b;
118 static unsigned long *vmx_msr_bitmap_legacy;
119 static unsigned long *vmx_msr_bitmap_longmode;
120
121 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
122 static DEFINE_SPINLOCK(vmx_vpid_lock);
123
124 static struct vmcs_config {
125         int size;
126         int order;
127         u32 revision_id;
128         u32 pin_based_exec_ctrl;
129         u32 cpu_based_exec_ctrl;
130         u32 cpu_based_2nd_exec_ctrl;
131         u32 vmexit_ctrl;
132         u32 vmentry_ctrl;
133 } vmcs_config;
134
135 static struct vmx_capability {
136         u32 ept;
137         u32 vpid;
138 } vmx_capability;
139
140 #define VMX_SEGMENT_FIELD(seg)                                  \
141         [VCPU_SREG_##seg] = {                                   \
142                 .selector = GUEST_##seg##_SELECTOR,             \
143                 .base = GUEST_##seg##_BASE,                     \
144                 .limit = GUEST_##seg##_LIMIT,                   \
145                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
146         }
147
148 static struct kvm_vmx_segment_field {
149         unsigned selector;
150         unsigned base;
151         unsigned limit;
152         unsigned ar_bytes;
153 } kvm_vmx_segment_fields[] = {
154         VMX_SEGMENT_FIELD(CS),
155         VMX_SEGMENT_FIELD(DS),
156         VMX_SEGMENT_FIELD(ES),
157         VMX_SEGMENT_FIELD(FS),
158         VMX_SEGMENT_FIELD(GS),
159         VMX_SEGMENT_FIELD(SS),
160         VMX_SEGMENT_FIELD(TR),
161         VMX_SEGMENT_FIELD(LDTR),
162 };
163
164 /*
165  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
166  * away by decrementing the array size.
167  */
168 static const u32 vmx_msr_index[] = {
169 #ifdef CONFIG_X86_64
170         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
171 #endif
172         MSR_EFER, MSR_K6_STAR,
173 };
174 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
175
176 static void load_msrs(struct kvm_msr_entry *e, int n)
177 {
178         int i;
179
180         for (i = 0; i < n; ++i)
181                 wrmsrl(e[i].index, e[i].data);
182 }
183
184 static void save_msrs(struct kvm_msr_entry *e, int n)
185 {
186         int i;
187
188         for (i = 0; i < n; ++i)
189                 rdmsrl(e[i].index, e[i].data);
190 }
191
192 static inline int is_page_fault(u32 intr_info)
193 {
194         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
195                              INTR_INFO_VALID_MASK)) ==
196                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
197 }
198
199 static inline int is_no_device(u32 intr_info)
200 {
201         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
202                              INTR_INFO_VALID_MASK)) ==
203                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
204 }
205
206 static inline int is_invalid_opcode(u32 intr_info)
207 {
208         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
209                              INTR_INFO_VALID_MASK)) ==
210                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
211 }
212
213 static inline int is_external_interrupt(u32 intr_info)
214 {
215         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
216                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
217 }
218
219 static inline int is_machine_check(u32 intr_info)
220 {
221         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
222                              INTR_INFO_VALID_MASK)) ==
223                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
224 }
225
226 static inline int cpu_has_vmx_msr_bitmap(void)
227 {
228         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
229 }
230
231 static inline int cpu_has_vmx_tpr_shadow(void)
232 {
233         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
234 }
235
236 static inline int vm_need_tpr_shadow(struct kvm *kvm)
237 {
238         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
239 }
240
241 static inline int cpu_has_secondary_exec_ctrls(void)
242 {
243         return vmcs_config.cpu_based_exec_ctrl &
244                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
245 }
246
247 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
248 {
249         return vmcs_config.cpu_based_2nd_exec_ctrl &
250                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
251 }
252
253 static inline bool cpu_has_vmx_flexpriority(void)
254 {
255         return cpu_has_vmx_tpr_shadow() &&
256                 cpu_has_vmx_virtualize_apic_accesses();
257 }
258
259 static inline int cpu_has_vmx_invept_individual_addr(void)
260 {
261         return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
262 }
263
264 static inline int cpu_has_vmx_invept_context(void)
265 {
266         return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
267 }
268
269 static inline int cpu_has_vmx_invept_global(void)
270 {
271         return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
272 }
273
274 static inline int cpu_has_vmx_ept(void)
275 {
276         return vmcs_config.cpu_based_2nd_exec_ctrl &
277                 SECONDARY_EXEC_ENABLE_EPT;
278 }
279
280 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
281 {
282         return flexpriority_enabled &&
283                 (cpu_has_vmx_virtualize_apic_accesses()) &&
284                 (irqchip_in_kernel(kvm));
285 }
286
287 static inline int cpu_has_vmx_vpid(void)
288 {
289         return vmcs_config.cpu_based_2nd_exec_ctrl &
290                 SECONDARY_EXEC_ENABLE_VPID;
291 }
292
293 static inline int cpu_has_virtual_nmis(void)
294 {
295         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
296 }
297
298 static inline bool report_flexpriority(void)
299 {
300         return flexpriority_enabled;
301 }
302
303 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
304 {
305         int i;
306
307         for (i = 0; i < vmx->nmsrs; ++i)
308                 if (vmx->guest_msrs[i].index == msr)
309                         return i;
310         return -1;
311 }
312
313 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
314 {
315     struct {
316         u64 vpid : 16;
317         u64 rsvd : 48;
318         u64 gva;
319     } operand = { vpid, 0, gva };
320
321     asm volatile (__ex(ASM_VMX_INVVPID)
322                   /* CF==1 or ZF==1 --> rc = -1 */
323                   "; ja 1f ; ud2 ; 1:"
324                   : : "a"(&operand), "c"(ext) : "cc", "memory");
325 }
326
327 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
328 {
329         struct {
330                 u64 eptp, gpa;
331         } operand = {eptp, gpa};
332
333         asm volatile (__ex(ASM_VMX_INVEPT)
334                         /* CF==1 or ZF==1 --> rc = -1 */
335                         "; ja 1f ; ud2 ; 1:\n"
336                         : : "a" (&operand), "c" (ext) : "cc", "memory");
337 }
338
339 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
340 {
341         int i;
342
343         i = __find_msr_index(vmx, msr);
344         if (i >= 0)
345                 return &vmx->guest_msrs[i];
346         return NULL;
347 }
348
349 static void vmcs_clear(struct vmcs *vmcs)
350 {
351         u64 phys_addr = __pa(vmcs);
352         u8 error;
353
354         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
355                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
356                       : "cc", "memory");
357         if (error)
358                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
359                        vmcs, phys_addr);
360 }
361
362 static void __vcpu_clear(void *arg)
363 {
364         struct vcpu_vmx *vmx = arg;
365         int cpu = raw_smp_processor_id();
366
367         if (vmx->vcpu.cpu == cpu)
368                 vmcs_clear(vmx->vmcs);
369         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
370                 per_cpu(current_vmcs, cpu) = NULL;
371         rdtscll(vmx->vcpu.arch.host_tsc);
372         list_del(&vmx->local_vcpus_link);
373         vmx->vcpu.cpu = -1;
374         vmx->launched = 0;
375 }
376
377 static void vcpu_clear(struct vcpu_vmx *vmx)
378 {
379         if (vmx->vcpu.cpu == -1)
380                 return;
381         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
382 }
383
384 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
385 {
386         if (vmx->vpid == 0)
387                 return;
388
389         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
390 }
391
392 static inline void ept_sync_global(void)
393 {
394         if (cpu_has_vmx_invept_global())
395                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
396 }
397
398 static inline void ept_sync_context(u64 eptp)
399 {
400         if (enable_ept) {
401                 if (cpu_has_vmx_invept_context())
402                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
403                 else
404                         ept_sync_global();
405         }
406 }
407
408 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
409 {
410         if (enable_ept) {
411                 if (cpu_has_vmx_invept_individual_addr())
412                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
413                                         eptp, gpa);
414                 else
415                         ept_sync_context(eptp);
416         }
417 }
418
419 static unsigned long vmcs_readl(unsigned long field)
420 {
421         unsigned long value;
422
423         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
424                       : "=a"(value) : "d"(field) : "cc");
425         return value;
426 }
427
428 static u16 vmcs_read16(unsigned long field)
429 {
430         return vmcs_readl(field);
431 }
432
433 static u32 vmcs_read32(unsigned long field)
434 {
435         return vmcs_readl(field);
436 }
437
438 static u64 vmcs_read64(unsigned long field)
439 {
440 #ifdef CONFIG_X86_64
441         return vmcs_readl(field);
442 #else
443         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
444 #endif
445 }
446
447 static noinline void vmwrite_error(unsigned long field, unsigned long value)
448 {
449         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
450                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
451         dump_stack();
452 }
453
454 static void vmcs_writel(unsigned long field, unsigned long value)
455 {
456         u8 error;
457
458         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
459                        : "=q"(error) : "a"(value), "d"(field) : "cc");
460         if (unlikely(error))
461                 vmwrite_error(field, value);
462 }
463
464 static void vmcs_write16(unsigned long field, u16 value)
465 {
466         vmcs_writel(field, value);
467 }
468
469 static void vmcs_write32(unsigned long field, u32 value)
470 {
471         vmcs_writel(field, value);
472 }
473
474 static void vmcs_write64(unsigned long field, u64 value)
475 {
476         vmcs_writel(field, value);
477 #ifndef CONFIG_X86_64
478         asm volatile ("");
479         vmcs_writel(field+1, value >> 32);
480 #endif
481 }
482
483 static void vmcs_clear_bits(unsigned long field, u32 mask)
484 {
485         vmcs_writel(field, vmcs_readl(field) & ~mask);
486 }
487
488 static void vmcs_set_bits(unsigned long field, u32 mask)
489 {
490         vmcs_writel(field, vmcs_readl(field) | mask);
491 }
492
493 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
494 {
495         u32 eb;
496
497         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
498         if (!vcpu->fpu_active)
499                 eb |= 1u << NM_VECTOR;
500         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
501                 if (vcpu->guest_debug &
502                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
503                         eb |= 1u << DB_VECTOR;
504                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
505                         eb |= 1u << BP_VECTOR;
506         }
507         if (vcpu->arch.rmode.vm86_active)
508                 eb = ~0;
509         if (enable_ept)
510                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
511         vmcs_write32(EXCEPTION_BITMAP, eb);
512 }
513
514 static void reload_tss(void)
515 {
516         /*
517          * VT restores TR but not its size.  Useless.
518          */
519         struct descriptor_table gdt;
520         struct desc_struct *descs;
521
522         kvm_get_gdt(&gdt);
523         descs = (void *)gdt.base;
524         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
525         load_TR_desc();
526 }
527
528 static void load_transition_efer(struct vcpu_vmx *vmx)
529 {
530         int efer_offset = vmx->msr_offset_efer;
531         u64 host_efer = vmx->host_msrs[efer_offset].data;
532         u64 guest_efer = vmx->guest_msrs[efer_offset].data;
533         u64 ignore_bits;
534
535         if (efer_offset < 0)
536                 return;
537         /*
538          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
539          * outside long mode
540          */
541         ignore_bits = EFER_NX | EFER_SCE;
542 #ifdef CONFIG_X86_64
543         ignore_bits |= EFER_LMA | EFER_LME;
544         /* SCE is meaningful only in long mode on Intel */
545         if (guest_efer & EFER_LMA)
546                 ignore_bits &= ~(u64)EFER_SCE;
547 #endif
548         if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
549                 return;
550
551         vmx->host_state.guest_efer_loaded = 1;
552         guest_efer &= ~ignore_bits;
553         guest_efer |= host_efer & ignore_bits;
554         wrmsrl(MSR_EFER, guest_efer);
555         vmx->vcpu.stat.efer_reload++;
556 }
557
558 static void reload_host_efer(struct vcpu_vmx *vmx)
559 {
560         if (vmx->host_state.guest_efer_loaded) {
561                 vmx->host_state.guest_efer_loaded = 0;
562                 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
563         }
564 }
565
566 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
567 {
568         struct vcpu_vmx *vmx = to_vmx(vcpu);
569
570         if (vmx->host_state.loaded)
571                 return;
572
573         vmx->host_state.loaded = 1;
574         /*
575          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
576          * allow segment selectors with cpl > 0 or ti == 1.
577          */
578         vmx->host_state.ldt_sel = kvm_read_ldt();
579         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
580         vmx->host_state.fs_sel = kvm_read_fs();
581         if (!(vmx->host_state.fs_sel & 7)) {
582                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
583                 vmx->host_state.fs_reload_needed = 0;
584         } else {
585                 vmcs_write16(HOST_FS_SELECTOR, 0);
586                 vmx->host_state.fs_reload_needed = 1;
587         }
588         vmx->host_state.gs_sel = kvm_read_gs();
589         if (!(vmx->host_state.gs_sel & 7))
590                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
591         else {
592                 vmcs_write16(HOST_GS_SELECTOR, 0);
593                 vmx->host_state.gs_ldt_reload_needed = 1;
594         }
595
596 #ifdef CONFIG_X86_64
597         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
598         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
599 #else
600         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
601         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
602 #endif
603
604 #ifdef CONFIG_X86_64
605         if (is_long_mode(&vmx->vcpu))
606                 save_msrs(vmx->host_msrs +
607                           vmx->msr_offset_kernel_gs_base, 1);
608
609 #endif
610         load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
611         load_transition_efer(vmx);
612 }
613
614 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
615 {
616         unsigned long flags;
617
618         if (!vmx->host_state.loaded)
619                 return;
620
621         ++vmx->vcpu.stat.host_state_reload;
622         vmx->host_state.loaded = 0;
623         if (vmx->host_state.fs_reload_needed)
624                 kvm_load_fs(vmx->host_state.fs_sel);
625         if (vmx->host_state.gs_ldt_reload_needed) {
626                 kvm_load_ldt(vmx->host_state.ldt_sel);
627                 /*
628                  * If we have to reload gs, we must take care to
629                  * preserve our gs base.
630                  */
631                 local_irq_save(flags);
632                 kvm_load_gs(vmx->host_state.gs_sel);
633 #ifdef CONFIG_X86_64
634                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
635 #endif
636                 local_irq_restore(flags);
637         }
638         reload_tss();
639         save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
640         load_msrs(vmx->host_msrs, vmx->save_nmsrs);
641         reload_host_efer(vmx);
642 }
643
644 static void vmx_load_host_state(struct vcpu_vmx *vmx)
645 {
646         preempt_disable();
647         __vmx_load_host_state(vmx);
648         preempt_enable();
649 }
650
651 /*
652  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
653  * vcpu mutex is already taken.
654  */
655 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
656 {
657         struct vcpu_vmx *vmx = to_vmx(vcpu);
658         u64 phys_addr = __pa(vmx->vmcs);
659         u64 tsc_this, delta, new_offset;
660
661         if (vcpu->cpu != cpu) {
662                 vcpu_clear(vmx);
663                 kvm_migrate_timers(vcpu);
664                 vpid_sync_vcpu_all(vmx);
665                 local_irq_disable();
666                 list_add(&vmx->local_vcpus_link,
667                          &per_cpu(vcpus_on_cpu, cpu));
668                 local_irq_enable();
669         }
670
671         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
672                 u8 error;
673
674                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
675                 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
676                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
677                               : "cc");
678                 if (error)
679                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
680                                vmx->vmcs, phys_addr);
681         }
682
683         if (vcpu->cpu != cpu) {
684                 struct descriptor_table dt;
685                 unsigned long sysenter_esp;
686
687                 vcpu->cpu = cpu;
688                 /*
689                  * Linux uses per-cpu TSS and GDT, so set these when switching
690                  * processors.
691                  */
692                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
693                 kvm_get_gdt(&dt);
694                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
695
696                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
697                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
698
699                 /*
700                  * Make sure the time stamp counter is monotonous.
701                  */
702                 rdtscll(tsc_this);
703                 if (tsc_this < vcpu->arch.host_tsc) {
704                         delta = vcpu->arch.host_tsc - tsc_this;
705                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
706                         vmcs_write64(TSC_OFFSET, new_offset);
707                 }
708         }
709 }
710
711 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
712 {
713         __vmx_load_host_state(to_vmx(vcpu));
714 }
715
716 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
717 {
718         if (vcpu->fpu_active)
719                 return;
720         vcpu->fpu_active = 1;
721         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
722         if (vcpu->arch.cr0 & X86_CR0_TS)
723                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
724         update_exception_bitmap(vcpu);
725 }
726
727 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
728 {
729         if (!vcpu->fpu_active)
730                 return;
731         vcpu->fpu_active = 0;
732         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
733         update_exception_bitmap(vcpu);
734 }
735
736 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
737 {
738         return vmcs_readl(GUEST_RFLAGS);
739 }
740
741 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
742 {
743         if (vcpu->arch.rmode.vm86_active)
744                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
745         vmcs_writel(GUEST_RFLAGS, rflags);
746 }
747
748 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
749 {
750         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
751         int ret = 0;
752
753         if (interruptibility & GUEST_INTR_STATE_STI)
754                 ret |= X86_SHADOW_INT_STI;
755         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
756                 ret |= X86_SHADOW_INT_MOV_SS;
757
758         return ret & mask;
759 }
760
761 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
762 {
763         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
764         u32 interruptibility = interruptibility_old;
765
766         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
767
768         if (mask & X86_SHADOW_INT_MOV_SS)
769                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
770         if (mask & X86_SHADOW_INT_STI)
771                 interruptibility |= GUEST_INTR_STATE_STI;
772
773         if ((interruptibility != interruptibility_old))
774                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
775 }
776
777 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
778 {
779         unsigned long rip;
780
781         rip = kvm_rip_read(vcpu);
782         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
783         kvm_rip_write(vcpu, rip);
784
785         /* skipping an emulated instruction also counts */
786         vmx_set_interrupt_shadow(vcpu, 0);
787 }
788
789 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
790                                 bool has_error_code, u32 error_code)
791 {
792         struct vcpu_vmx *vmx = to_vmx(vcpu);
793         u32 intr_info = nr | INTR_INFO_VALID_MASK;
794
795         if (has_error_code) {
796                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
797                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
798         }
799
800         if (vcpu->arch.rmode.vm86_active) {
801                 vmx->rmode.irq.pending = true;
802                 vmx->rmode.irq.vector = nr;
803                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
804                 if (nr == BP_VECTOR || nr == OF_VECTOR)
805                         vmx->rmode.irq.rip++;
806                 intr_info |= INTR_TYPE_SOFT_INTR;
807                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
808                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
809                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
810                 return;
811         }
812
813         if (kvm_exception_is_soft(nr)) {
814                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
815                              vmx->vcpu.arch.event_exit_inst_len);
816                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
817         } else
818                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
819
820         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
821 }
822
823 /*
824  * Swap MSR entry in host/guest MSR entry array.
825  */
826 #ifdef CONFIG_X86_64
827 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
828 {
829         struct kvm_msr_entry tmp;
830
831         tmp = vmx->guest_msrs[to];
832         vmx->guest_msrs[to] = vmx->guest_msrs[from];
833         vmx->guest_msrs[from] = tmp;
834         tmp = vmx->host_msrs[to];
835         vmx->host_msrs[to] = vmx->host_msrs[from];
836         vmx->host_msrs[from] = tmp;
837 }
838 #endif
839
840 /*
841  * Set up the vmcs to automatically save and restore system
842  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
843  * mode, as fiddling with msrs is very expensive.
844  */
845 static void setup_msrs(struct vcpu_vmx *vmx)
846 {
847         int save_nmsrs;
848         unsigned long *msr_bitmap;
849
850         vmx_load_host_state(vmx);
851         save_nmsrs = 0;
852 #ifdef CONFIG_X86_64
853         if (is_long_mode(&vmx->vcpu)) {
854                 int index;
855
856                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
857                 if (index >= 0)
858                         move_msr_up(vmx, index, save_nmsrs++);
859                 index = __find_msr_index(vmx, MSR_LSTAR);
860                 if (index >= 0)
861                         move_msr_up(vmx, index, save_nmsrs++);
862                 index = __find_msr_index(vmx, MSR_CSTAR);
863                 if (index >= 0)
864                         move_msr_up(vmx, index, save_nmsrs++);
865                 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
866                 if (index >= 0)
867                         move_msr_up(vmx, index, save_nmsrs++);
868                 /*
869                  * MSR_K6_STAR is only needed on long mode guests, and only
870                  * if efer.sce is enabled.
871                  */
872                 index = __find_msr_index(vmx, MSR_K6_STAR);
873                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
874                         move_msr_up(vmx, index, save_nmsrs++);
875         }
876 #endif
877         vmx->save_nmsrs = save_nmsrs;
878
879 #ifdef CONFIG_X86_64
880         vmx->msr_offset_kernel_gs_base =
881                 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
882 #endif
883         vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
884
885         if (cpu_has_vmx_msr_bitmap()) {
886                 if (is_long_mode(&vmx->vcpu))
887                         msr_bitmap = vmx_msr_bitmap_longmode;
888                 else
889                         msr_bitmap = vmx_msr_bitmap_legacy;
890
891                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
892         }
893 }
894
895 /*
896  * reads and returns guest's timestamp counter "register"
897  * guest_tsc = host_tsc + tsc_offset    -- 21.3
898  */
899 static u64 guest_read_tsc(void)
900 {
901         u64 host_tsc, tsc_offset;
902
903         rdtscll(host_tsc);
904         tsc_offset = vmcs_read64(TSC_OFFSET);
905         return host_tsc + tsc_offset;
906 }
907
908 /*
909  * writes 'guest_tsc' into guest's timestamp counter "register"
910  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
911  */
912 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
913 {
914         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
915 }
916
917 /*
918  * Reads an msr value (of 'msr_index') into 'pdata'.
919  * Returns 0 on success, non-0 otherwise.
920  * Assumes vcpu_load() was already called.
921  */
922 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
923 {
924         u64 data;
925         struct kvm_msr_entry *msr;
926
927         if (!pdata) {
928                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
929                 return -EINVAL;
930         }
931
932         switch (msr_index) {
933 #ifdef CONFIG_X86_64
934         case MSR_FS_BASE:
935                 data = vmcs_readl(GUEST_FS_BASE);
936                 break;
937         case MSR_GS_BASE:
938                 data = vmcs_readl(GUEST_GS_BASE);
939                 break;
940         case MSR_EFER:
941                 return kvm_get_msr_common(vcpu, msr_index, pdata);
942 #endif
943         case MSR_IA32_TIME_STAMP_COUNTER:
944                 data = guest_read_tsc();
945                 break;
946         case MSR_IA32_SYSENTER_CS:
947                 data = vmcs_read32(GUEST_SYSENTER_CS);
948                 break;
949         case MSR_IA32_SYSENTER_EIP:
950                 data = vmcs_readl(GUEST_SYSENTER_EIP);
951                 break;
952         case MSR_IA32_SYSENTER_ESP:
953                 data = vmcs_readl(GUEST_SYSENTER_ESP);
954                 break;
955         default:
956                 vmx_load_host_state(to_vmx(vcpu));
957                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
958                 if (msr) {
959                         data = msr->data;
960                         break;
961                 }
962                 return kvm_get_msr_common(vcpu, msr_index, pdata);
963         }
964
965         *pdata = data;
966         return 0;
967 }
968
969 /*
970  * Writes msr value into into the appropriate "register".
971  * Returns 0 on success, non-0 otherwise.
972  * Assumes vcpu_load() was already called.
973  */
974 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
975 {
976         struct vcpu_vmx *vmx = to_vmx(vcpu);
977         struct kvm_msr_entry *msr;
978         u64 host_tsc;
979         int ret = 0;
980
981         switch (msr_index) {
982         case MSR_EFER:
983                 vmx_load_host_state(vmx);
984                 ret = kvm_set_msr_common(vcpu, msr_index, data);
985                 break;
986 #ifdef CONFIG_X86_64
987         case MSR_FS_BASE:
988                 vmcs_writel(GUEST_FS_BASE, data);
989                 break;
990         case MSR_GS_BASE:
991                 vmcs_writel(GUEST_GS_BASE, data);
992                 break;
993 #endif
994         case MSR_IA32_SYSENTER_CS:
995                 vmcs_write32(GUEST_SYSENTER_CS, data);
996                 break;
997         case MSR_IA32_SYSENTER_EIP:
998                 vmcs_writel(GUEST_SYSENTER_EIP, data);
999                 break;
1000         case MSR_IA32_SYSENTER_ESP:
1001                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1002                 break;
1003         case MSR_IA32_TIME_STAMP_COUNTER:
1004                 rdtscll(host_tsc);
1005                 guest_write_tsc(data, host_tsc);
1006                 break;
1007         case MSR_P6_PERFCTR0:
1008         case MSR_P6_PERFCTR1:
1009         case MSR_P6_EVNTSEL0:
1010         case MSR_P6_EVNTSEL1:
1011                 /*
1012                  * Just discard all writes to the performance counters; this
1013                  * should keep both older linux and windows 64-bit guests
1014                  * happy
1015                  */
1016                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
1017
1018                 break;
1019         case MSR_IA32_CR_PAT:
1020                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1021                         vmcs_write64(GUEST_IA32_PAT, data);
1022                         vcpu->arch.pat = data;
1023                         break;
1024                 }
1025                 /* Otherwise falls through to kvm_set_msr_common */
1026         default:
1027                 vmx_load_host_state(vmx);
1028                 msr = find_msr_entry(vmx, msr_index);
1029                 if (msr) {
1030                         msr->data = data;
1031                         break;
1032                 }
1033                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1034         }
1035
1036         return ret;
1037 }
1038
1039 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1040 {
1041         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1042         switch (reg) {
1043         case VCPU_REGS_RSP:
1044                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1045                 break;
1046         case VCPU_REGS_RIP:
1047                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1048                 break;
1049         default:
1050                 break;
1051         }
1052 }
1053
1054 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1055 {
1056         int old_debug = vcpu->guest_debug;
1057         unsigned long flags;
1058
1059         vcpu->guest_debug = dbg->control;
1060         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1061                 vcpu->guest_debug = 0;
1062
1063         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1064                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1065         else
1066                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1067
1068         flags = vmcs_readl(GUEST_RFLAGS);
1069         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1070                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1071         else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1072                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1073         vmcs_writel(GUEST_RFLAGS, flags);
1074
1075         update_exception_bitmap(vcpu);
1076
1077         return 0;
1078 }
1079
1080 static __init int cpu_has_kvm_support(void)
1081 {
1082         return cpu_has_vmx();
1083 }
1084
1085 static __init int vmx_disabled_by_bios(void)
1086 {
1087         u64 msr;
1088
1089         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1090         return (msr & (FEATURE_CONTROL_LOCKED |
1091                        FEATURE_CONTROL_VMXON_ENABLED))
1092             == FEATURE_CONTROL_LOCKED;
1093         /* locked but not enabled */
1094 }
1095
1096 static void hardware_enable(void *garbage)
1097 {
1098         int cpu = raw_smp_processor_id();
1099         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1100         u64 old;
1101
1102         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1103         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1104         if ((old & (FEATURE_CONTROL_LOCKED |
1105                     FEATURE_CONTROL_VMXON_ENABLED))
1106             != (FEATURE_CONTROL_LOCKED |
1107                 FEATURE_CONTROL_VMXON_ENABLED))
1108                 /* enable and lock */
1109                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1110                        FEATURE_CONTROL_LOCKED |
1111                        FEATURE_CONTROL_VMXON_ENABLED);
1112         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1113         asm volatile (ASM_VMX_VMXON_RAX
1114                       : : "a"(&phys_addr), "m"(phys_addr)
1115                       : "memory", "cc");
1116 }
1117
1118 static void vmclear_local_vcpus(void)
1119 {
1120         int cpu = raw_smp_processor_id();
1121         struct vcpu_vmx *vmx, *n;
1122
1123         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1124                                  local_vcpus_link)
1125                 __vcpu_clear(vmx);
1126 }
1127
1128
1129 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1130  * tricks.
1131  */
1132 static void kvm_cpu_vmxoff(void)
1133 {
1134         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1135         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1136 }
1137
1138 static void hardware_disable(void *garbage)
1139 {
1140         vmclear_local_vcpus();
1141         kvm_cpu_vmxoff();
1142 }
1143
1144 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1145                                       u32 msr, u32 *result)
1146 {
1147         u32 vmx_msr_low, vmx_msr_high;
1148         u32 ctl = ctl_min | ctl_opt;
1149
1150         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1151
1152         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1153         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1154
1155         /* Ensure minimum (required) set of control bits are supported. */
1156         if (ctl_min & ~ctl)
1157                 return -EIO;
1158
1159         *result = ctl;
1160         return 0;
1161 }
1162
1163 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1164 {
1165         u32 vmx_msr_low, vmx_msr_high;
1166         u32 min, opt, min2, opt2;
1167         u32 _pin_based_exec_control = 0;
1168         u32 _cpu_based_exec_control = 0;
1169         u32 _cpu_based_2nd_exec_control = 0;
1170         u32 _vmexit_control = 0;
1171         u32 _vmentry_control = 0;
1172
1173         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1174         opt = PIN_BASED_VIRTUAL_NMIS;
1175         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1176                                 &_pin_based_exec_control) < 0)
1177                 return -EIO;
1178
1179         min = CPU_BASED_HLT_EXITING |
1180 #ifdef CONFIG_X86_64
1181               CPU_BASED_CR8_LOAD_EXITING |
1182               CPU_BASED_CR8_STORE_EXITING |
1183 #endif
1184               CPU_BASED_CR3_LOAD_EXITING |
1185               CPU_BASED_CR3_STORE_EXITING |
1186               CPU_BASED_USE_IO_BITMAPS |
1187               CPU_BASED_MOV_DR_EXITING |
1188               CPU_BASED_USE_TSC_OFFSETING |
1189               CPU_BASED_INVLPG_EXITING;
1190         opt = CPU_BASED_TPR_SHADOW |
1191               CPU_BASED_USE_MSR_BITMAPS |
1192               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1193         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1194                                 &_cpu_based_exec_control) < 0)
1195                 return -EIO;
1196 #ifdef CONFIG_X86_64
1197         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1198                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1199                                            ~CPU_BASED_CR8_STORE_EXITING;
1200 #endif
1201         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1202                 min2 = 0;
1203                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1204                         SECONDARY_EXEC_WBINVD_EXITING |
1205                         SECONDARY_EXEC_ENABLE_VPID |
1206                         SECONDARY_EXEC_ENABLE_EPT;
1207                 if (adjust_vmx_controls(min2, opt2,
1208                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1209                                         &_cpu_based_2nd_exec_control) < 0)
1210                         return -EIO;
1211         }
1212 #ifndef CONFIG_X86_64
1213         if (!(_cpu_based_2nd_exec_control &
1214                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1215                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1216 #endif
1217         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1218                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1219                    enabled */
1220                 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1221                          CPU_BASED_CR3_STORE_EXITING |
1222                          CPU_BASED_INVLPG_EXITING);
1223                 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1224                                         &_cpu_based_exec_control) < 0)
1225                         return -EIO;
1226                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1227                       vmx_capability.ept, vmx_capability.vpid);
1228         }
1229
1230         min = 0;
1231 #ifdef CONFIG_X86_64
1232         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1233 #endif
1234         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1235         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1236                                 &_vmexit_control) < 0)
1237                 return -EIO;
1238
1239         min = 0;
1240         opt = VM_ENTRY_LOAD_IA32_PAT;
1241         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1242                                 &_vmentry_control) < 0)
1243                 return -EIO;
1244
1245         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1246
1247         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1248         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1249                 return -EIO;
1250
1251 #ifdef CONFIG_X86_64
1252         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1253         if (vmx_msr_high & (1u<<16))
1254                 return -EIO;
1255 #endif
1256
1257         /* Require Write-Back (WB) memory type for VMCS accesses. */
1258         if (((vmx_msr_high >> 18) & 15) != 6)
1259                 return -EIO;
1260
1261         vmcs_conf->size = vmx_msr_high & 0x1fff;
1262         vmcs_conf->order = get_order(vmcs_config.size);
1263         vmcs_conf->revision_id = vmx_msr_low;
1264
1265         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1266         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1267         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1268         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1269         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1270
1271         return 0;
1272 }
1273
1274 static struct vmcs *alloc_vmcs_cpu(int cpu)
1275 {
1276         int node = cpu_to_node(cpu);
1277         struct page *pages;
1278         struct vmcs *vmcs;
1279
1280         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1281         if (!pages)
1282                 return NULL;
1283         vmcs = page_address(pages);
1284         memset(vmcs, 0, vmcs_config.size);
1285         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1286         return vmcs;
1287 }
1288
1289 static struct vmcs *alloc_vmcs(void)
1290 {
1291         return alloc_vmcs_cpu(raw_smp_processor_id());
1292 }
1293
1294 static void free_vmcs(struct vmcs *vmcs)
1295 {
1296         free_pages((unsigned long)vmcs, vmcs_config.order);
1297 }
1298
1299 static void free_kvm_area(void)
1300 {
1301         int cpu;
1302
1303         for_each_online_cpu(cpu)
1304                 free_vmcs(per_cpu(vmxarea, cpu));
1305 }
1306
1307 static __init int alloc_kvm_area(void)
1308 {
1309         int cpu;
1310
1311         for_each_online_cpu(cpu) {
1312                 struct vmcs *vmcs;
1313
1314                 vmcs = alloc_vmcs_cpu(cpu);
1315                 if (!vmcs) {
1316                         free_kvm_area();
1317                         return -ENOMEM;
1318                 }
1319
1320                 per_cpu(vmxarea, cpu) = vmcs;
1321         }
1322         return 0;
1323 }
1324
1325 static __init int hardware_setup(void)
1326 {
1327         if (setup_vmcs_config(&vmcs_config) < 0)
1328                 return -EIO;
1329
1330         if (boot_cpu_has(X86_FEATURE_NX))
1331                 kvm_enable_efer_bits(EFER_NX);
1332
1333         if (!cpu_has_vmx_vpid())
1334                 enable_vpid = 0;
1335
1336         if (!cpu_has_vmx_ept())
1337                 enable_ept = 0;
1338
1339         if (!cpu_has_vmx_flexpriority())
1340                 flexpriority_enabled = 0;
1341
1342         if (!cpu_has_vmx_tpr_shadow())
1343                 kvm_x86_ops->update_cr8_intercept = NULL;
1344
1345         return alloc_kvm_area();
1346 }
1347
1348 static __exit void hardware_unsetup(void)
1349 {
1350         free_kvm_area();
1351 }
1352
1353 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1354 {
1355         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1356
1357         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1358                 vmcs_write16(sf->selector, save->selector);
1359                 vmcs_writel(sf->base, save->base);
1360                 vmcs_write32(sf->limit, save->limit);
1361                 vmcs_write32(sf->ar_bytes, save->ar);
1362         } else {
1363                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1364                         << AR_DPL_SHIFT;
1365                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1366         }
1367 }
1368
1369 static void enter_pmode(struct kvm_vcpu *vcpu)
1370 {
1371         unsigned long flags;
1372         struct vcpu_vmx *vmx = to_vmx(vcpu);
1373
1374         vmx->emulation_required = 1;
1375         vcpu->arch.rmode.vm86_active = 0;
1376
1377         vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1378         vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1379         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1380
1381         flags = vmcs_readl(GUEST_RFLAGS);
1382         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1383         flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1384         vmcs_writel(GUEST_RFLAGS, flags);
1385
1386         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1387                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1388
1389         update_exception_bitmap(vcpu);
1390
1391         if (emulate_invalid_guest_state)
1392                 return;
1393
1394         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1395         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1396         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1397         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1398
1399         vmcs_write16(GUEST_SS_SELECTOR, 0);
1400         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1401
1402         vmcs_write16(GUEST_CS_SELECTOR,
1403                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1404         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1405 }
1406
1407 static gva_t rmode_tss_base(struct kvm *kvm)
1408 {
1409         if (!kvm->arch.tss_addr) {
1410                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1411                                  kvm->memslots[0].npages - 3;
1412                 return base_gfn << PAGE_SHIFT;
1413         }
1414         return kvm->arch.tss_addr;
1415 }
1416
1417 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1418 {
1419         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1420
1421         save->selector = vmcs_read16(sf->selector);
1422         save->base = vmcs_readl(sf->base);
1423         save->limit = vmcs_read32(sf->limit);
1424         save->ar = vmcs_read32(sf->ar_bytes);
1425         vmcs_write16(sf->selector, save->base >> 4);
1426         vmcs_write32(sf->base, save->base & 0xfffff);
1427         vmcs_write32(sf->limit, 0xffff);
1428         vmcs_write32(sf->ar_bytes, 0xf3);
1429 }
1430
1431 static void enter_rmode(struct kvm_vcpu *vcpu)
1432 {
1433         unsigned long flags;
1434         struct vcpu_vmx *vmx = to_vmx(vcpu);
1435
1436         vmx->emulation_required = 1;
1437         vcpu->arch.rmode.vm86_active = 1;
1438
1439         vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1440         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1441
1442         vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1443         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1444
1445         vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1446         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1447
1448         flags = vmcs_readl(GUEST_RFLAGS);
1449         vcpu->arch.rmode.save_iopl
1450                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1451
1452         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1453
1454         vmcs_writel(GUEST_RFLAGS, flags);
1455         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1456         update_exception_bitmap(vcpu);
1457
1458         if (emulate_invalid_guest_state)
1459                 goto continue_rmode;
1460
1461         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1462         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1463         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1464
1465         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1466         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1467         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1468                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1469         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1470
1471         fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1472         fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1473         fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1474         fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1475
1476 continue_rmode:
1477         kvm_mmu_reset_context(vcpu);
1478         init_rmode(vcpu->kvm);
1479 }
1480
1481 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1482 {
1483         struct vcpu_vmx *vmx = to_vmx(vcpu);
1484         struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1485
1486         vcpu->arch.shadow_efer = efer;
1487         if (!msr)
1488                 return;
1489         if (efer & EFER_LMA) {
1490                 vmcs_write32(VM_ENTRY_CONTROLS,
1491                              vmcs_read32(VM_ENTRY_CONTROLS) |
1492                              VM_ENTRY_IA32E_MODE);
1493                 msr->data = efer;
1494         } else {
1495                 vmcs_write32(VM_ENTRY_CONTROLS,
1496                              vmcs_read32(VM_ENTRY_CONTROLS) &
1497                              ~VM_ENTRY_IA32E_MODE);
1498
1499                 msr->data = efer & ~EFER_LME;
1500         }
1501         setup_msrs(vmx);
1502 }
1503
1504 #ifdef CONFIG_X86_64
1505
1506 static void enter_lmode(struct kvm_vcpu *vcpu)
1507 {
1508         u32 guest_tr_ar;
1509
1510         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1511         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1512                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1513                        __func__);
1514                 vmcs_write32(GUEST_TR_AR_BYTES,
1515                              (guest_tr_ar & ~AR_TYPE_MASK)
1516                              | AR_TYPE_BUSY_64_TSS);
1517         }
1518         vcpu->arch.shadow_efer |= EFER_LMA;
1519         vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1520 }
1521
1522 static void exit_lmode(struct kvm_vcpu *vcpu)
1523 {
1524         vcpu->arch.shadow_efer &= ~EFER_LMA;
1525
1526         vmcs_write32(VM_ENTRY_CONTROLS,
1527                      vmcs_read32(VM_ENTRY_CONTROLS)
1528                      & ~VM_ENTRY_IA32E_MODE);
1529 }
1530
1531 #endif
1532
1533 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1534 {
1535         vpid_sync_vcpu_all(to_vmx(vcpu));
1536         if (enable_ept)
1537                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1538 }
1539
1540 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1541 {
1542         vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1543         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1544 }
1545
1546 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1547 {
1548         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1549                 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1550                         printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1551                         return;
1552                 }
1553                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1554                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1555                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1556                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1557         }
1558 }
1559
1560 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1561
1562 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1563                                         unsigned long cr0,
1564                                         struct kvm_vcpu *vcpu)
1565 {
1566         if (!(cr0 & X86_CR0_PG)) {
1567                 /* From paging/starting to nonpaging */
1568                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1569                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1570                              (CPU_BASED_CR3_LOAD_EXITING |
1571                               CPU_BASED_CR3_STORE_EXITING));
1572                 vcpu->arch.cr0 = cr0;
1573                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1574                 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1575                 *hw_cr0 &= ~X86_CR0_WP;
1576         } else if (!is_paging(vcpu)) {
1577                 /* From nonpaging to paging */
1578                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1579                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1580                              ~(CPU_BASED_CR3_LOAD_EXITING |
1581                                CPU_BASED_CR3_STORE_EXITING));
1582                 vcpu->arch.cr0 = cr0;
1583                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1584                 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1585                         *hw_cr0 &= ~X86_CR0_WP;
1586         }
1587 }
1588
1589 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1590                                         struct kvm_vcpu *vcpu)
1591 {
1592         if (!is_paging(vcpu)) {
1593                 *hw_cr4 &= ~X86_CR4_PAE;
1594                 *hw_cr4 |= X86_CR4_PSE;
1595         } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1596                 *hw_cr4 &= ~X86_CR4_PAE;
1597 }
1598
1599 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1600 {
1601         unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1602                                 KVM_VM_CR0_ALWAYS_ON;
1603
1604         vmx_fpu_deactivate(vcpu);
1605
1606         if (vcpu->arch.rmode.vm86_active && (cr0 & X86_CR0_PE))
1607                 enter_pmode(vcpu);
1608
1609         if (!vcpu->arch.rmode.vm86_active && !(cr0 & X86_CR0_PE))
1610                 enter_rmode(vcpu);
1611
1612 #ifdef CONFIG_X86_64
1613         if (vcpu->arch.shadow_efer & EFER_LME) {
1614                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1615                         enter_lmode(vcpu);
1616                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1617                         exit_lmode(vcpu);
1618         }
1619 #endif
1620
1621         if (enable_ept)
1622                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1623
1624         vmcs_writel(CR0_READ_SHADOW, cr0);
1625         vmcs_writel(GUEST_CR0, hw_cr0);
1626         vcpu->arch.cr0 = cr0;
1627
1628         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1629                 vmx_fpu_activate(vcpu);
1630 }
1631
1632 static u64 construct_eptp(unsigned long root_hpa)
1633 {
1634         u64 eptp;
1635
1636         /* TODO write the value reading from MSR */
1637         eptp = VMX_EPT_DEFAULT_MT |
1638                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1639         eptp |= (root_hpa & PAGE_MASK);
1640
1641         return eptp;
1642 }
1643
1644 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1645 {
1646         unsigned long guest_cr3;
1647         u64 eptp;
1648
1649         guest_cr3 = cr3;
1650         if (enable_ept) {
1651                 eptp = construct_eptp(cr3);
1652                 vmcs_write64(EPT_POINTER, eptp);
1653                 ept_sync_context(eptp);
1654                 ept_load_pdptrs(vcpu);
1655                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1656                         VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1657         }
1658
1659         vmx_flush_tlb(vcpu);
1660         vmcs_writel(GUEST_CR3, guest_cr3);
1661         if (vcpu->arch.cr0 & X86_CR0_PE)
1662                 vmx_fpu_deactivate(vcpu);
1663 }
1664
1665 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1666 {
1667         unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.vm86_active ?
1668                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1669
1670         vcpu->arch.cr4 = cr4;
1671         if (enable_ept)
1672                 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1673
1674         vmcs_writel(CR4_READ_SHADOW, cr4);
1675         vmcs_writel(GUEST_CR4, hw_cr4);
1676 }
1677
1678 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1679 {
1680         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1681
1682         return vmcs_readl(sf->base);
1683 }
1684
1685 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1686                             struct kvm_segment *var, int seg)
1687 {
1688         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1689         u32 ar;
1690
1691         var->base = vmcs_readl(sf->base);
1692         var->limit = vmcs_read32(sf->limit);
1693         var->selector = vmcs_read16(sf->selector);
1694         ar = vmcs_read32(sf->ar_bytes);
1695         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1696                 ar = 0;
1697         var->type = ar & 15;
1698         var->s = (ar >> 4) & 1;
1699         var->dpl = (ar >> 5) & 3;
1700         var->present = (ar >> 7) & 1;
1701         var->avl = (ar >> 12) & 1;
1702         var->l = (ar >> 13) & 1;
1703         var->db = (ar >> 14) & 1;
1704         var->g = (ar >> 15) & 1;
1705         var->unusable = (ar >> 16) & 1;
1706 }
1707
1708 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1709 {
1710         struct kvm_segment kvm_seg;
1711
1712         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1713                 return 0;
1714
1715         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1716                 return 3;
1717
1718         vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1719         return kvm_seg.selector & 3;
1720 }
1721
1722 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1723 {
1724         u32 ar;
1725
1726         if (var->unusable)
1727                 ar = 1 << 16;
1728         else {
1729                 ar = var->type & 15;
1730                 ar |= (var->s & 1) << 4;
1731                 ar |= (var->dpl & 3) << 5;
1732                 ar |= (var->present & 1) << 7;
1733                 ar |= (var->avl & 1) << 12;
1734                 ar |= (var->l & 1) << 13;
1735                 ar |= (var->db & 1) << 14;
1736                 ar |= (var->g & 1) << 15;
1737         }
1738         if (ar == 0) /* a 0 value means unusable */
1739                 ar = AR_UNUSABLE_MASK;
1740
1741         return ar;
1742 }
1743
1744 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1745                             struct kvm_segment *var, int seg)
1746 {
1747         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1748         u32 ar;
1749
1750         if (vcpu->arch.rmode.vm86_active && seg == VCPU_SREG_TR) {
1751                 vcpu->arch.rmode.tr.selector = var->selector;
1752                 vcpu->arch.rmode.tr.base = var->base;
1753                 vcpu->arch.rmode.tr.limit = var->limit;
1754                 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1755                 return;
1756         }
1757         vmcs_writel(sf->base, var->base);
1758         vmcs_write32(sf->limit, var->limit);
1759         vmcs_write16(sf->selector, var->selector);
1760         if (vcpu->arch.rmode.vm86_active && var->s) {
1761                 /*
1762                  * Hack real-mode segments into vm86 compatibility.
1763                  */
1764                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1765                         vmcs_writel(sf->base, 0xf0000);
1766                 ar = 0xf3;
1767         } else
1768                 ar = vmx_segment_access_rights(var);
1769         vmcs_write32(sf->ar_bytes, ar);
1770 }
1771
1772 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1773 {
1774         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1775
1776         *db = (ar >> 14) & 1;
1777         *l = (ar >> 13) & 1;
1778 }
1779
1780 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1781 {
1782         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1783         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1784 }
1785
1786 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1787 {
1788         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1789         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1790 }
1791
1792 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1793 {
1794         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1795         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1796 }
1797
1798 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1799 {
1800         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1801         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1802 }
1803
1804 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1805 {
1806         struct kvm_segment var;
1807         u32 ar;
1808
1809         vmx_get_segment(vcpu, &var, seg);
1810         ar = vmx_segment_access_rights(&var);
1811
1812         if (var.base != (var.selector << 4))
1813                 return false;
1814         if (var.limit != 0xffff)
1815                 return false;
1816         if (ar != 0xf3)
1817                 return false;
1818
1819         return true;
1820 }
1821
1822 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1823 {
1824         struct kvm_segment cs;
1825         unsigned int cs_rpl;
1826
1827         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1828         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1829
1830         if (cs.unusable)
1831                 return false;
1832         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1833                 return false;
1834         if (!cs.s)
1835                 return false;
1836         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1837                 if (cs.dpl > cs_rpl)
1838                         return false;
1839         } else {
1840                 if (cs.dpl != cs_rpl)
1841                         return false;
1842         }
1843         if (!cs.present)
1844                 return false;
1845
1846         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1847         return true;
1848 }
1849
1850 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1851 {
1852         struct kvm_segment ss;
1853         unsigned int ss_rpl;
1854
1855         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1856         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1857
1858         if (ss.unusable)
1859                 return true;
1860         if (ss.type != 3 && ss.type != 7)
1861                 return false;
1862         if (!ss.s)
1863                 return false;
1864         if (ss.dpl != ss_rpl) /* DPL != RPL */
1865                 return false;
1866         if (!ss.present)
1867                 return false;
1868
1869         return true;
1870 }
1871
1872 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1873 {
1874         struct kvm_segment var;
1875         unsigned int rpl;
1876
1877         vmx_get_segment(vcpu, &var, seg);
1878         rpl = var.selector & SELECTOR_RPL_MASK;
1879
1880         if (var.unusable)
1881                 return true;
1882         if (!var.s)
1883                 return false;
1884         if (!var.present)
1885                 return false;
1886         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1887                 if (var.dpl < rpl) /* DPL < RPL */
1888                         return false;
1889         }
1890
1891         /* TODO: Add other members to kvm_segment_field to allow checking for other access
1892          * rights flags
1893          */
1894         return true;
1895 }
1896
1897 static bool tr_valid(struct kvm_vcpu *vcpu)
1898 {
1899         struct kvm_segment tr;
1900
1901         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1902
1903         if (tr.unusable)
1904                 return false;
1905         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
1906                 return false;
1907         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
1908                 return false;
1909         if (!tr.present)
1910                 return false;
1911
1912         return true;
1913 }
1914
1915 static bool ldtr_valid(struct kvm_vcpu *vcpu)
1916 {
1917         struct kvm_segment ldtr;
1918
1919         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1920
1921         if (ldtr.unusable)
1922                 return true;
1923         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
1924                 return false;
1925         if (ldtr.type != 2)
1926                 return false;
1927         if (!ldtr.present)
1928                 return false;
1929
1930         return true;
1931 }
1932
1933 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1934 {
1935         struct kvm_segment cs, ss;
1936
1937         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1938         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1939
1940         return ((cs.selector & SELECTOR_RPL_MASK) ==
1941                  (ss.selector & SELECTOR_RPL_MASK));
1942 }
1943
1944 /*
1945  * Check if guest state is valid. Returns true if valid, false if
1946  * not.
1947  * We assume that registers are always usable
1948  */
1949 static bool guest_state_valid(struct kvm_vcpu *vcpu)
1950 {
1951         /* real mode guest state checks */
1952         if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1953                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1954                         return false;
1955                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1956                         return false;
1957                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1958                         return false;
1959                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1960                         return false;
1961                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1962                         return false;
1963                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1964                         return false;
1965         } else {
1966         /* protected mode guest state checks */
1967                 if (!cs_ss_rpl_check(vcpu))
1968                         return false;
1969                 if (!code_segment_valid(vcpu))
1970                         return false;
1971                 if (!stack_segment_valid(vcpu))
1972                         return false;
1973                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1974                         return false;
1975                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1976                         return false;
1977                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1978                         return false;
1979                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1980                         return false;
1981                 if (!tr_valid(vcpu))
1982                         return false;
1983                 if (!ldtr_valid(vcpu))
1984                         return false;
1985         }
1986         /* TODO:
1987          * - Add checks on RIP
1988          * - Add checks on RFLAGS
1989          */
1990
1991         return true;
1992 }
1993
1994 static int init_rmode_tss(struct kvm *kvm)
1995 {
1996         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1997         u16 data = 0;
1998         int ret = 0;
1999         int r;
2000
2001         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2002         if (r < 0)
2003                 goto out;
2004         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2005         r = kvm_write_guest_page(kvm, fn++, &data,
2006                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2007         if (r < 0)
2008                 goto out;
2009         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2010         if (r < 0)
2011                 goto out;
2012         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2013         if (r < 0)
2014                 goto out;
2015         data = ~0;
2016         r = kvm_write_guest_page(kvm, fn, &data,
2017                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2018                                  sizeof(u8));
2019         if (r < 0)
2020                 goto out;
2021
2022         ret = 1;
2023 out:
2024         return ret;
2025 }
2026
2027 static int init_rmode_identity_map(struct kvm *kvm)
2028 {
2029         int i, r, ret;
2030         pfn_t identity_map_pfn;
2031         u32 tmp;
2032
2033         if (!enable_ept)
2034                 return 1;
2035         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2036                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2037                         "haven't been allocated!\n");
2038                 return 0;
2039         }
2040         if (likely(kvm->arch.ept_identity_pagetable_done))
2041                 return 1;
2042         ret = 0;
2043         identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
2044         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2045         if (r < 0)
2046                 goto out;
2047         /* Set up identity-mapping pagetable for EPT in real mode */
2048         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2049                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2050                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2051                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2052                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2053                 if (r < 0)
2054                         goto out;
2055         }
2056         kvm->arch.ept_identity_pagetable_done = true;
2057         ret = 1;
2058 out:
2059         return ret;
2060 }
2061
2062 static void seg_setup(int seg)
2063 {
2064         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2065
2066         vmcs_write16(sf->selector, 0);
2067         vmcs_writel(sf->base, 0);
2068         vmcs_write32(sf->limit, 0xffff);
2069         vmcs_write32(sf->ar_bytes, 0xf3);
2070 }
2071
2072 static int alloc_apic_access_page(struct kvm *kvm)
2073 {
2074         struct kvm_userspace_memory_region kvm_userspace_mem;
2075         int r = 0;
2076
2077         down_write(&kvm->slots_lock);
2078         if (kvm->arch.apic_access_page)
2079                 goto out;
2080         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2081         kvm_userspace_mem.flags = 0;
2082         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2083         kvm_userspace_mem.memory_size = PAGE_SIZE;
2084         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2085         if (r)
2086                 goto out;
2087
2088         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2089 out:
2090         up_write(&kvm->slots_lock);
2091         return r;
2092 }
2093
2094 static int alloc_identity_pagetable(struct kvm *kvm)
2095 {
2096         struct kvm_userspace_memory_region kvm_userspace_mem;
2097         int r = 0;
2098
2099         down_write(&kvm->slots_lock);
2100         if (kvm->arch.ept_identity_pagetable)
2101                 goto out;
2102         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2103         kvm_userspace_mem.flags = 0;
2104         kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2105         kvm_userspace_mem.memory_size = PAGE_SIZE;
2106         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2107         if (r)
2108                 goto out;
2109
2110         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2111                         VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
2112 out:
2113         up_write(&kvm->slots_lock);
2114         return r;
2115 }
2116
2117 static void allocate_vpid(struct vcpu_vmx *vmx)
2118 {
2119         int vpid;
2120
2121         vmx->vpid = 0;
2122         if (!enable_vpid)
2123                 return;
2124         spin_lock(&vmx_vpid_lock);
2125         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2126         if (vpid < VMX_NR_VPIDS) {
2127                 vmx->vpid = vpid;
2128                 __set_bit(vpid, vmx_vpid_bitmap);
2129         }
2130         spin_unlock(&vmx_vpid_lock);
2131 }
2132
2133 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2134 {
2135         int f = sizeof(unsigned long);
2136
2137         if (!cpu_has_vmx_msr_bitmap())
2138                 return;
2139
2140         /*
2141          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2142          * have the write-low and read-high bitmap offsets the wrong way round.
2143          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2144          */
2145         if (msr <= 0x1fff) {
2146                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2147                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2148         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2149                 msr &= 0x1fff;
2150                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2151                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2152         }
2153 }
2154
2155 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2156 {
2157         if (!longmode_only)
2158                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2159         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2160 }
2161
2162 /*
2163  * Sets up the vmcs for emulated real mode.
2164  */
2165 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2166 {
2167         u32 host_sysenter_cs, msr_low, msr_high;
2168         u32 junk;
2169         u64 host_pat, tsc_this, tsc_base;
2170         unsigned long a;
2171         struct descriptor_table dt;
2172         int i;
2173         unsigned long kvm_vmx_return;
2174         u32 exec_control;
2175
2176         /* I/O */
2177         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2178         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2179
2180         if (cpu_has_vmx_msr_bitmap())
2181                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2182
2183         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2184
2185         /* Control */
2186         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2187                 vmcs_config.pin_based_exec_ctrl);
2188
2189         exec_control = vmcs_config.cpu_based_exec_ctrl;
2190         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2191                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2192 #ifdef CONFIG_X86_64
2193                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2194                                 CPU_BASED_CR8_LOAD_EXITING;
2195 #endif
2196         }
2197         if (!enable_ept)
2198                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2199                                 CPU_BASED_CR3_LOAD_EXITING  |
2200                                 CPU_BASED_INVLPG_EXITING;
2201         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2202
2203         if (cpu_has_secondary_exec_ctrls()) {
2204                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2205                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2206                         exec_control &=
2207                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2208                 if (vmx->vpid == 0)
2209                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2210                 if (!enable_ept)
2211                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2212                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2213         }
2214
2215         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2216         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2217         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2218
2219         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
2220         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2221         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2222
2223         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2224         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2225         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2226         vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs());    /* 22.2.4 */
2227         vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs());    /* 22.2.4 */
2228         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2229 #ifdef CONFIG_X86_64
2230         rdmsrl(MSR_FS_BASE, a);
2231         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2232         rdmsrl(MSR_GS_BASE, a);
2233         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2234 #else
2235         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2236         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2237 #endif
2238
2239         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2240
2241         kvm_get_idt(&dt);
2242         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
2243
2244         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2245         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2246         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2247         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2248         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2249
2250         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2251         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2252         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2253         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2254         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2255         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2256
2257         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2258                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2259                 host_pat = msr_low | ((u64) msr_high << 32);
2260                 vmcs_write64(HOST_IA32_PAT, host_pat);
2261         }
2262         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2263                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2264                 host_pat = msr_low | ((u64) msr_high << 32);
2265                 /* Write the default value follow host pat */
2266                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2267                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2268                 vmx->vcpu.arch.pat = host_pat;
2269         }
2270
2271         for (i = 0; i < NR_VMX_MSR; ++i) {
2272                 u32 index = vmx_msr_index[i];
2273                 u32 data_low, data_high;
2274                 u64 data;
2275                 int j = vmx->nmsrs;
2276
2277                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2278                         continue;
2279                 if (wrmsr_safe(index, data_low, data_high) < 0)
2280                         continue;
2281                 data = data_low | ((u64)data_high << 32);
2282                 vmx->host_msrs[j].index = index;
2283                 vmx->host_msrs[j].reserved = 0;
2284                 vmx->host_msrs[j].data = data;
2285                 vmx->guest_msrs[j] = vmx->host_msrs[j];
2286                 ++vmx->nmsrs;
2287         }
2288
2289         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2290
2291         /* 22.2.1, 20.8.1 */
2292         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2293
2294         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2295         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2296
2297         tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2298         rdtscll(tsc_this);
2299         if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2300                 tsc_base = tsc_this;
2301
2302         guest_write_tsc(0, tsc_base);
2303
2304         return 0;
2305 }
2306
2307 static int init_rmode(struct kvm *kvm)
2308 {
2309         if (!init_rmode_tss(kvm))
2310                 return 0;
2311         if (!init_rmode_identity_map(kvm))
2312                 return 0;
2313         return 1;
2314 }
2315
2316 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2317 {
2318         struct vcpu_vmx *vmx = to_vmx(vcpu);
2319         u64 msr;
2320         int ret;
2321
2322         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2323         down_read(&vcpu->kvm->slots_lock);
2324         if (!init_rmode(vmx->vcpu.kvm)) {
2325                 ret = -ENOMEM;
2326                 goto out;
2327         }
2328
2329         vmx->vcpu.arch.rmode.vm86_active = 0;
2330
2331         vmx->soft_vnmi_blocked = 0;
2332
2333         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2334         kvm_set_cr8(&vmx->vcpu, 0);
2335         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2336         if (vmx->vcpu.vcpu_id == 0)
2337                 msr |= MSR_IA32_APICBASE_BSP;
2338         kvm_set_apic_base(&vmx->vcpu, msr);
2339
2340         fx_init(&vmx->vcpu);
2341
2342         seg_setup(VCPU_SREG_CS);
2343         /*
2344          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2345          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2346          */
2347         if (vmx->vcpu.vcpu_id == 0) {
2348                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2349                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2350         } else {
2351                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2352                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2353         }
2354
2355         seg_setup(VCPU_SREG_DS);
2356         seg_setup(VCPU_SREG_ES);
2357         seg_setup(VCPU_SREG_FS);
2358         seg_setup(VCPU_SREG_GS);
2359         seg_setup(VCPU_SREG_SS);
2360
2361         vmcs_write16(GUEST_TR_SELECTOR, 0);
2362         vmcs_writel(GUEST_TR_BASE, 0);
2363         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2364         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2365
2366         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2367         vmcs_writel(GUEST_LDTR_BASE, 0);
2368         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2369         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2370
2371         vmcs_write32(GUEST_SYSENTER_CS, 0);
2372         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2373         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2374
2375         vmcs_writel(GUEST_RFLAGS, 0x02);
2376         if (vmx->vcpu.vcpu_id == 0)
2377                 kvm_rip_write(vcpu, 0xfff0);
2378         else
2379                 kvm_rip_write(vcpu, 0);
2380         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2381
2382         vmcs_writel(GUEST_DR7, 0x400);
2383
2384         vmcs_writel(GUEST_GDTR_BASE, 0);
2385         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2386
2387         vmcs_writel(GUEST_IDTR_BASE, 0);
2388         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2389
2390         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2391         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2392         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2393
2394         /* Special registers */
2395         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2396
2397         setup_msrs(vmx);
2398
2399         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2400
2401         if (cpu_has_vmx_tpr_shadow()) {
2402                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2403                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2404                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2405                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2406                 vmcs_write32(TPR_THRESHOLD, 0);
2407         }
2408
2409         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2410                 vmcs_write64(APIC_ACCESS_ADDR,
2411                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2412
2413         if (vmx->vpid != 0)
2414                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2415
2416         vmx->vcpu.arch.cr0 = 0x60000010;
2417         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2418         vmx_set_cr4(&vmx->vcpu, 0);
2419         vmx_set_efer(&vmx->vcpu, 0);
2420         vmx_fpu_activate(&vmx->vcpu);
2421         update_exception_bitmap(&vmx->vcpu);
2422
2423         vpid_sync_vcpu_all(vmx);
2424
2425         ret = 0;
2426
2427         /* HACK: Don't enable emulation on guest boot/reset */
2428         vmx->emulation_required = 0;
2429
2430 out:
2431         up_read(&vcpu->kvm->slots_lock);
2432         return ret;
2433 }
2434
2435 static void enable_irq_window(struct kvm_vcpu *vcpu)
2436 {
2437         u32 cpu_based_vm_exec_control;
2438
2439         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2440         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2441         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2442 }
2443
2444 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2445 {
2446         u32 cpu_based_vm_exec_control;
2447
2448         if (!cpu_has_virtual_nmis()) {
2449                 enable_irq_window(vcpu);
2450                 return;
2451         }
2452
2453         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2454         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2455         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2456 }
2457
2458 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2459 {
2460         struct vcpu_vmx *vmx = to_vmx(vcpu);
2461         uint32_t intr;
2462         int irq = vcpu->arch.interrupt.nr;
2463
2464         KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2465
2466         ++vcpu->stat.irq_injections;
2467         if (vcpu->arch.rmode.vm86_active) {
2468                 vmx->rmode.irq.pending = true;
2469                 vmx->rmode.irq.vector = irq;
2470                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2471                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2472                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2473                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2474                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2475                 return;
2476         }
2477         intr = irq | INTR_INFO_VALID_MASK;
2478         if (vcpu->arch.interrupt.soft) {
2479                 intr |= INTR_TYPE_SOFT_INTR;
2480                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2481                              vmx->vcpu.arch.event_exit_inst_len);
2482         } else
2483                 intr |= INTR_TYPE_EXT_INTR;
2484         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2485 }
2486
2487 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2488 {
2489         struct vcpu_vmx *vmx = to_vmx(vcpu);
2490
2491         if (!cpu_has_virtual_nmis()) {
2492                 /*
2493                  * Tracking the NMI-blocked state in software is built upon
2494                  * finding the next open IRQ window. This, in turn, depends on
2495                  * well-behaving guests: They have to keep IRQs disabled at
2496                  * least as long as the NMI handler runs. Otherwise we may
2497                  * cause NMI nesting, maybe breaking the guest. But as this is
2498                  * highly unlikely, we can live with the residual risk.
2499                  */
2500                 vmx->soft_vnmi_blocked = 1;
2501                 vmx->vnmi_blocked_time = 0;
2502         }
2503
2504         ++vcpu->stat.nmi_injections;
2505         if (vcpu->arch.rmode.vm86_active) {
2506                 vmx->rmode.irq.pending = true;
2507                 vmx->rmode.irq.vector = NMI_VECTOR;
2508                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2509                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2510                              NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2511                              INTR_INFO_VALID_MASK);
2512                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2513                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2514                 return;
2515         }
2516         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2517                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2518 }
2519
2520 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2521 {
2522         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2523                 return 0;
2524
2525         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2526                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2527                                 GUEST_INTR_STATE_NMI));
2528 }
2529
2530 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2531 {
2532         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2533                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2534                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2535 }
2536
2537 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2538 {
2539         int ret;
2540         struct kvm_userspace_memory_region tss_mem = {
2541                 .slot = TSS_PRIVATE_MEMSLOT,
2542                 .guest_phys_addr = addr,
2543                 .memory_size = PAGE_SIZE * 3,
2544                 .flags = 0,
2545         };
2546
2547         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2548         if (ret)
2549                 return ret;
2550         kvm->arch.tss_addr = addr;
2551         return 0;
2552 }
2553
2554 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2555                                   int vec, u32 err_code)
2556 {
2557         /*
2558          * Instruction with address size override prefix opcode 0x67
2559          * Cause the #SS fault with 0 error code in VM86 mode.
2560          */
2561         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2562                 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2563                         return 1;
2564         /*
2565          * Forward all other exceptions that are valid in real mode.
2566          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2567          *        the required debugging infrastructure rework.
2568          */
2569         switch (vec) {
2570         case DB_VECTOR:
2571                 if (vcpu->guest_debug &
2572                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2573                         return 0;
2574                 kvm_queue_exception(vcpu, vec);
2575                 return 1;
2576         case BP_VECTOR:
2577                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2578                         return 0;
2579                 /* fall through */
2580         case DE_VECTOR:
2581         case OF_VECTOR:
2582         case BR_VECTOR:
2583         case UD_VECTOR:
2584         case DF_VECTOR:
2585         case SS_VECTOR:
2586         case GP_VECTOR:
2587         case MF_VECTOR:
2588                 kvm_queue_exception(vcpu, vec);
2589                 return 1;
2590         }
2591         return 0;
2592 }
2593
2594 /*
2595  * Trigger machine check on the host. We assume all the MSRs are already set up
2596  * by the CPU and that we still run on the same CPU as the MCE occurred on.
2597  * We pass a fake environment to the machine check handler because we want
2598  * the guest to be always treated like user space, no matter what context
2599  * it used internally.
2600  */
2601 static void kvm_machine_check(void)
2602 {
2603 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2604         struct pt_regs regs = {
2605                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2606                 .flags = X86_EFLAGS_IF,
2607         };
2608
2609         do_machine_check(&regs, 0);
2610 #endif
2611 }
2612
2613 static int handle_machine_check(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2614 {
2615         /* already handled by vcpu_run */
2616         return 1;
2617 }
2618
2619 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2620 {
2621         struct vcpu_vmx *vmx = to_vmx(vcpu);
2622         u32 intr_info, ex_no, error_code;
2623         unsigned long cr2, rip, dr6;
2624         u32 vect_info;
2625         enum emulation_result er;
2626
2627         vect_info = vmx->idt_vectoring_info;
2628         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2629
2630         if (is_machine_check(intr_info))
2631                 return handle_machine_check(vcpu, kvm_run);
2632
2633         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2634                                                 !is_page_fault(intr_info))
2635                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2636                        "intr info 0x%x\n", __func__, vect_info, intr_info);
2637
2638         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2639                 return 1;  /* already handled by vmx_vcpu_run() */
2640
2641         if (is_no_device(intr_info)) {
2642                 vmx_fpu_activate(vcpu);
2643                 return 1;
2644         }
2645
2646         if (is_invalid_opcode(intr_info)) {
2647                 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2648                 if (er != EMULATE_DONE)
2649                         kvm_queue_exception(vcpu, UD_VECTOR);
2650                 return 1;
2651         }
2652
2653         error_code = 0;
2654         rip = kvm_rip_read(vcpu);
2655         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2656                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2657         if (is_page_fault(intr_info)) {
2658                 /* EPT won't cause page fault directly */
2659                 if (enable_ept)
2660                         BUG();
2661                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2662                 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2663                             (u32)((u64)cr2 >> 32), handler);
2664                 if (kvm_event_needs_reinjection(vcpu))
2665                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
2666                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2667         }
2668
2669         if (vcpu->arch.rmode.vm86_active &&
2670             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2671                                                                 error_code)) {
2672                 if (vcpu->arch.halt_request) {
2673                         vcpu->arch.halt_request = 0;
2674                         return kvm_emulate_halt(vcpu);
2675                 }
2676                 return 1;
2677         }
2678
2679         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2680         switch (ex_no) {
2681         case DB_VECTOR:
2682                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2683                 if (!(vcpu->guest_debug &
2684                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2685                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2686                         kvm_queue_exception(vcpu, DB_VECTOR);
2687                         return 1;
2688                 }
2689                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2690                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2691                 /* fall through */
2692         case BP_VECTOR:
2693                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2694                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2695                 kvm_run->debug.arch.exception = ex_no;
2696                 break;
2697         default:
2698                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2699                 kvm_run->ex.exception = ex_no;
2700                 kvm_run->ex.error_code = error_code;
2701                 break;
2702         }
2703         return 0;
2704 }
2705
2706 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2707                                      struct kvm_run *kvm_run)
2708 {
2709         ++vcpu->stat.irq_exits;
2710         KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
2711         return 1;
2712 }
2713
2714 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2715 {
2716         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2717         return 0;
2718 }
2719
2720 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2721 {
2722         unsigned long exit_qualification;
2723         int size, in, string;
2724         unsigned port;
2725
2726         ++vcpu->stat.io_exits;
2727         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2728         string = (exit_qualification & 16) != 0;
2729
2730         if (string) {
2731                 if (emulate_instruction(vcpu,
2732                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2733                         return 0;
2734                 return 1;
2735         }
2736
2737         size = (exit_qualification & 7) + 1;
2738         in = (exit_qualification & 8) != 0;
2739         port = exit_qualification >> 16;
2740
2741         skip_emulated_instruction(vcpu);
2742         return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2743 }
2744
2745 static void
2746 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2747 {
2748         /*
2749          * Patch in the VMCALL instruction:
2750          */
2751         hypercall[0] = 0x0f;
2752         hypercall[1] = 0x01;
2753         hypercall[2] = 0xc1;
2754 }
2755
2756 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2757 {
2758         unsigned long exit_qualification;
2759         int cr;
2760         int reg;
2761
2762         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2763         cr = exit_qualification & 15;
2764         reg = (exit_qualification >> 8) & 15;
2765         switch ((exit_qualification >> 4) & 3) {
2766         case 0: /* mov to cr */
2767                 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2768                             (u32)kvm_register_read(vcpu, reg),
2769                             (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2770                             handler);
2771                 switch (cr) {
2772                 case 0:
2773                         kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
2774                         skip_emulated_instruction(vcpu);
2775                         return 1;
2776                 case 3:
2777                         kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
2778                         skip_emulated_instruction(vcpu);
2779                         return 1;
2780                 case 4:
2781                         kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
2782                         skip_emulated_instruction(vcpu);
2783                         return 1;
2784                 case 8: {
2785                                 u8 cr8_prev = kvm_get_cr8(vcpu);
2786                                 u8 cr8 = kvm_register_read(vcpu, reg);
2787                                 kvm_set_cr8(vcpu, cr8);
2788                                 skip_emulated_instruction(vcpu);
2789                                 if (irqchip_in_kernel(vcpu->kvm))
2790                                         return 1;
2791                                 if (cr8_prev <= cr8)
2792                                         return 1;
2793                                 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2794                                 return 0;
2795                         }
2796                 };
2797                 break;
2798         case 2: /* clts */
2799                 vmx_fpu_deactivate(vcpu);
2800                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2801                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2802                 vmx_fpu_activate(vcpu);
2803                 KVMTRACE_0D(CLTS, vcpu, handler);
2804                 skip_emulated_instruction(vcpu);
2805                 return 1;
2806         case 1: /*mov from cr*/
2807                 switch (cr) {
2808                 case 3:
2809                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2810                         KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2811                                     (u32)kvm_register_read(vcpu, reg),
2812                                     (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2813                                     handler);
2814                         skip_emulated_instruction(vcpu);
2815                         return 1;
2816                 case 8:
2817                         kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2818                         KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2819                                     (u32)kvm_register_read(vcpu, reg), handler);
2820                         skip_emulated_instruction(vcpu);
2821                         return 1;
2822                 }
2823                 break;
2824         case 3: /* lmsw */
2825                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2826
2827                 skip_emulated_instruction(vcpu);
2828                 return 1;
2829         default:
2830                 break;
2831         }
2832         kvm_run->exit_reason = 0;
2833         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2834                (int)(exit_qualification >> 4) & 3, cr);
2835         return 0;
2836 }
2837
2838 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2839 {
2840         unsigned long exit_qualification;
2841         unsigned long val;
2842         int dr, reg;
2843
2844         dr = vmcs_readl(GUEST_DR7);
2845         if (dr & DR7_GD) {
2846                 /*
2847                  * As the vm-exit takes precedence over the debug trap, we
2848                  * need to emulate the latter, either for the host or the
2849                  * guest debugging itself.
2850                  */
2851                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2852                         kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2853                         kvm_run->debug.arch.dr7 = dr;
2854                         kvm_run->debug.arch.pc =
2855                                 vmcs_readl(GUEST_CS_BASE) +
2856                                 vmcs_readl(GUEST_RIP);
2857                         kvm_run->debug.arch.exception = DB_VECTOR;
2858                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
2859                         return 0;
2860                 } else {
2861                         vcpu->arch.dr7 &= ~DR7_GD;
2862                         vcpu->arch.dr6 |= DR6_BD;
2863                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2864                         kvm_queue_exception(vcpu, DB_VECTOR);
2865                         return 1;
2866                 }
2867         }
2868
2869         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2870         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2871         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2872         if (exit_qualification & TYPE_MOV_FROM_DR) {
2873                 switch (dr) {
2874                 case 0 ... 3:
2875                         val = vcpu->arch.db[dr];
2876                         break;
2877                 case 6:
2878                         val = vcpu->arch.dr6;
2879                         break;
2880                 case 7:
2881                         val = vcpu->arch.dr7;
2882                         break;
2883                 default:
2884                         val = 0;
2885                 }
2886                 kvm_register_write(vcpu, reg, val);
2887                 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2888         } else {
2889                 val = vcpu->arch.regs[reg];
2890                 switch (dr) {
2891                 case 0 ... 3:
2892                         vcpu->arch.db[dr] = val;
2893                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2894                                 vcpu->arch.eff_db[dr] = val;
2895                         break;
2896                 case 4 ... 5:
2897                         if (vcpu->arch.cr4 & X86_CR4_DE)
2898                                 kvm_queue_exception(vcpu, UD_VECTOR);
2899                         break;
2900                 case 6:
2901                         if (val & 0xffffffff00000000ULL) {
2902                                 kvm_queue_exception(vcpu, GP_VECTOR);
2903                                 break;
2904                         }
2905                         vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2906                         break;
2907                 case 7:
2908                         if (val & 0xffffffff00000000ULL) {
2909                                 kvm_queue_exception(vcpu, GP_VECTOR);
2910                                 break;
2911                         }
2912                         vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2913                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2914                                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2915                                 vcpu->arch.switch_db_regs =
2916                                         (val & DR7_BP_EN_MASK);
2917                         }
2918                         break;
2919                 }
2920                 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)val, handler);
2921         }
2922         skip_emulated_instruction(vcpu);
2923         return 1;
2924 }
2925
2926 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2927 {
2928         kvm_emulate_cpuid(vcpu);
2929         return 1;
2930 }
2931
2932 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2933 {
2934         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2935         u64 data;
2936
2937         if (vmx_get_msr(vcpu, ecx, &data)) {
2938                 kvm_inject_gp(vcpu, 0);
2939                 return 1;
2940         }
2941
2942         KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2943                     handler);
2944
2945         /* FIXME: handling of bits 32:63 of rax, rdx */
2946         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2947         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2948         skip_emulated_instruction(vcpu);
2949         return 1;
2950 }
2951
2952 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2953 {
2954         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2955         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2956                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2957
2958         KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2959                     handler);
2960
2961         if (vmx_set_msr(vcpu, ecx, data) != 0) {
2962                 kvm_inject_gp(vcpu, 0);
2963                 return 1;
2964         }
2965
2966         skip_emulated_instruction(vcpu);
2967         return 1;
2968 }
2969
2970 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2971                                       struct kvm_run *kvm_run)
2972 {
2973         return 1;
2974 }
2975
2976 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2977                                    struct kvm_run *kvm_run)
2978 {
2979         u32 cpu_based_vm_exec_control;
2980
2981         /* clear pending irq */
2982         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2983         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2984         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2985
2986         KVMTRACE_0D(PEND_INTR, vcpu, handler);
2987         ++vcpu->stat.irq_window_exits;
2988
2989         /*
2990          * If the user space waits to inject interrupts, exit as soon as
2991          * possible
2992          */
2993         if (!irqchip_in_kernel(vcpu->kvm) &&
2994             kvm_run->request_interrupt_window &&
2995             !kvm_cpu_has_interrupt(vcpu)) {
2996                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2997                 return 0;
2998         }
2999         return 1;
3000 }
3001
3002 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3003 {
3004         skip_emulated_instruction(vcpu);
3005         return kvm_emulate_halt(vcpu);
3006 }
3007
3008 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3009 {
3010         skip_emulated_instruction(vcpu);
3011         kvm_emulate_hypercall(vcpu);
3012         return 1;
3013 }
3014
3015 static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3016 {
3017         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3018
3019         kvm_mmu_invlpg(vcpu, exit_qualification);
3020         skip_emulated_instruction(vcpu);
3021         return 1;
3022 }
3023
3024 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3025 {
3026         skip_emulated_instruction(vcpu);
3027         /* TODO: Add support for VT-d/pass-through device */
3028         return 1;
3029 }
3030
3031 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3032 {
3033         unsigned long exit_qualification;
3034         enum emulation_result er;
3035         unsigned long offset;
3036
3037         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3038         offset = exit_qualification & 0xffful;
3039
3040         er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3041
3042         if (er !=  EMULATE_DONE) {
3043                 printk(KERN_ERR
3044                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3045                        offset);
3046                 return -ENOTSUPP;
3047         }
3048         return 1;
3049 }
3050
3051 static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3052 {
3053         struct vcpu_vmx *vmx = to_vmx(vcpu);
3054         unsigned long exit_qualification;
3055         u16 tss_selector;
3056         int reason, type, idt_v;
3057
3058         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3059         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3060
3061         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3062
3063         reason = (u32)exit_qualification >> 30;
3064         if (reason == TASK_SWITCH_GATE && idt_v) {
3065                 switch (type) {
3066                 case INTR_TYPE_NMI_INTR:
3067                         vcpu->arch.nmi_injected = false;
3068                         if (cpu_has_virtual_nmis())
3069                                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3070                                               GUEST_INTR_STATE_NMI);
3071                         break;
3072                 case INTR_TYPE_EXT_INTR:
3073                 case INTR_TYPE_SOFT_INTR:
3074                         kvm_clear_interrupt_queue(vcpu);
3075                         break;
3076                 case INTR_TYPE_HARD_EXCEPTION:
3077                 case INTR_TYPE_SOFT_EXCEPTION:
3078                         kvm_clear_exception_queue(vcpu);
3079                         break;
3080                 default:
3081                         break;
3082                 }
3083         }
3084         tss_selector = exit_qualification;
3085
3086         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3087                        type != INTR_TYPE_EXT_INTR &&
3088                        type != INTR_TYPE_NMI_INTR))
3089                 skip_emulated_instruction(vcpu);
3090
3091         if (!kvm_task_switch(vcpu, tss_selector, reason))
3092                 return 0;
3093
3094         /* clear all local breakpoint enable flags */
3095         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3096
3097         /*
3098          * TODO: What about debug traps on tss switch?
3099          *       Are we supposed to inject them and update dr6?
3100          */
3101
3102         return 1;
3103 }
3104
3105 static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3106 {
3107         unsigned long exit_qualification;
3108         gpa_t gpa;
3109         int gla_validity;
3110
3111         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3112
3113         if (exit_qualification & (1 << 6)) {
3114                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3115                 return -ENOTSUPP;
3116         }
3117
3118         gla_validity = (exit_qualification >> 7) & 0x3;
3119         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3120                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3121                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3122                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3123                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3124                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3125                         (long unsigned int)exit_qualification);
3126                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3127                 kvm_run->hw.hardware_exit_reason = 0;
3128                 return -ENOTSUPP;
3129         }
3130
3131         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3132         return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3133 }
3134
3135 static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3136 {
3137         u32 cpu_based_vm_exec_control;
3138
3139         /* clear pending NMI */
3140         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3141         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3142         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3143         ++vcpu->stat.nmi_window_exits;
3144
3145         return 1;
3146 }
3147
3148 static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3149                                 struct kvm_run *kvm_run)
3150 {
3151         struct vcpu_vmx *vmx = to_vmx(vcpu);
3152         enum emulation_result err = EMULATE_DONE;
3153
3154         preempt_enable();
3155         local_irq_enable();
3156
3157         while (!guest_state_valid(vcpu)) {
3158                 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3159
3160                 if (err == EMULATE_DO_MMIO)
3161                         break;
3162
3163                 if (err != EMULATE_DONE) {
3164                         kvm_report_emulation_failure(vcpu, "emulation failure");
3165                         return;
3166                 }
3167
3168                 if (signal_pending(current))
3169                         break;
3170                 if (need_resched())
3171                         schedule();
3172         }
3173
3174         local_irq_disable();
3175         preempt_disable();
3176
3177         vmx->invalid_state_emulation_result = err;
3178 }
3179
3180 /*
3181  * The exit handlers return 1 if the exit was handled fully and guest execution
3182  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3183  * to be done to userspace and return 0.
3184  */
3185 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3186                                       struct kvm_run *kvm_run) = {
3187         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3188         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3189         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3190         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3191         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3192         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3193         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3194         [EXIT_REASON_CPUID]                   = handle_cpuid,
3195         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3196         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3197         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3198         [EXIT_REASON_HLT]                     = handle_halt,
3199         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3200         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3201         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3202         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3203         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3204         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3205         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3206         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3207 };
3208
3209 static const int kvm_vmx_max_exit_handlers =
3210         ARRAY_SIZE(kvm_vmx_exit_handlers);
3211
3212 /*
3213  * The guest has exited.  See if we can fix it or if we need userspace
3214  * assistance.
3215  */
3216 static int vmx_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3217 {
3218         struct vcpu_vmx *vmx = to_vmx(vcpu);
3219         u32 exit_reason = vmx->exit_reason;
3220         u32 vectoring_info = vmx->idt_vectoring_info;
3221
3222         KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3223                     (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
3224
3225         /* If we need to emulate an MMIO from handle_invalid_guest_state
3226          * we just return 0 */
3227         if (vmx->emulation_required && emulate_invalid_guest_state) {
3228                 if (guest_state_valid(vcpu))
3229                         vmx->emulation_required = 0;
3230                 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
3231         }
3232
3233         /* Access CR3 don't cause VMExit in paging mode, so we need
3234          * to sync with guest real CR3. */
3235         if (enable_ept && is_paging(vcpu)) {
3236                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3237                 ept_load_pdptrs(vcpu);
3238         }
3239
3240         if (unlikely(vmx->fail)) {
3241                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3242                 kvm_run->fail_entry.hardware_entry_failure_reason
3243                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3244                 return 0;
3245         }
3246
3247         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3248                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3249                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3250                         exit_reason != EXIT_REASON_TASK_SWITCH))
3251                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3252                        "(0x%x) and exit reason is 0x%x\n",
3253                        __func__, vectoring_info, exit_reason);
3254
3255         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3256                 if (vmx_interrupt_allowed(vcpu)) {
3257                         vmx->soft_vnmi_blocked = 0;
3258                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3259                            vcpu->arch.nmi_pending) {
3260                         /*
3261                          * This CPU don't support us in finding the end of an
3262                          * NMI-blocked window if the guest runs with IRQs
3263                          * disabled. So we pull the trigger after 1 s of
3264                          * futile waiting, but inform the user about this.
3265                          */
3266                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3267                                "state on VCPU %d after 1 s timeout\n",
3268                                __func__, vcpu->vcpu_id);
3269                         vmx->soft_vnmi_blocked = 0;
3270                 }
3271         }
3272
3273         if (exit_reason < kvm_vmx_max_exit_handlers
3274             && kvm_vmx_exit_handlers[exit_reason])
3275                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3276         else {
3277                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3278                 kvm_run->hw.hardware_exit_reason = exit_reason;
3279         }
3280         return 0;
3281 }
3282
3283 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3284 {
3285         if (irr == -1 || tpr < irr) {
3286                 vmcs_write32(TPR_THRESHOLD, 0);
3287                 return;
3288         }
3289
3290         vmcs_write32(TPR_THRESHOLD, irr);
3291 }
3292
3293 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3294 {
3295         u32 exit_intr_info;
3296         u32 idt_vectoring_info = vmx->idt_vectoring_info;
3297         bool unblock_nmi;
3298         u8 vector;
3299         int type;
3300         bool idtv_info_valid;
3301
3302         exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3303
3304         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3305
3306         /* Handle machine checks before interrupts are enabled */
3307         if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3308             || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3309                 && is_machine_check(exit_intr_info)))
3310                 kvm_machine_check();
3311
3312         /* We need to handle NMIs before interrupts are enabled */
3313         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3314             (exit_intr_info & INTR_INFO_VALID_MASK)) {
3315                 KVMTRACE_0D(NMI, &vmx->vcpu, handler);
3316                 asm("int $2");
3317         }
3318
3319         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3320
3321         if (cpu_has_virtual_nmis()) {
3322                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3323                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3324                 /*
3325                  * SDM 3: 27.7.1.2 (September 2008)
3326                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3327                  * a guest IRET fault.
3328                  * SDM 3: 23.2.2 (September 2008)
3329                  * Bit 12 is undefined in any of the following cases:
3330                  *  If the VM exit sets the valid bit in the IDT-vectoring
3331                  *   information field.
3332                  *  If the VM exit is due to a double fault.
3333                  */
3334                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3335                     vector != DF_VECTOR && !idtv_info_valid)
3336                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3337                                       GUEST_INTR_STATE_NMI);
3338         } else if (unlikely(vmx->soft_vnmi_blocked))
3339                 vmx->vnmi_blocked_time +=
3340                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3341
3342         vmx->vcpu.arch.nmi_injected = false;
3343         kvm_clear_exception_queue(&vmx->vcpu);
3344         kvm_clear_interrupt_queue(&vmx->vcpu);
3345
3346         if (!idtv_info_valid)
3347                 return;
3348
3349         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3350         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3351
3352         switch (type) {
3353         case INTR_TYPE_NMI_INTR:
3354                 vmx->vcpu.arch.nmi_injected = true;
3355                 /*
3356                  * SDM 3: 27.7.1.2 (September 2008)
3357                  * Clear bit "block by NMI" before VM entry if a NMI
3358                  * delivery faulted.
3359                  */
3360                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3361                                 GUEST_INTR_STATE_NMI);
3362                 break;
3363         case INTR_TYPE_SOFT_EXCEPTION:
3364                 vmx->vcpu.arch.event_exit_inst_len =
3365                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3366                 /* fall through */
3367         case INTR_TYPE_HARD_EXCEPTION:
3368                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3369                         u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3370                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
3371                 } else
3372                         kvm_queue_exception(&vmx->vcpu, vector);
3373                 break;
3374         case INTR_TYPE_SOFT_INTR:
3375                 vmx->vcpu.arch.event_exit_inst_len =
3376                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3377                 /* fall through */
3378         case INTR_TYPE_EXT_INTR:
3379                 kvm_queue_interrupt(&vmx->vcpu, vector,
3380                         type == INTR_TYPE_SOFT_INTR);
3381                 break;
3382         default:
3383                 break;
3384         }
3385 }
3386
3387 /*
3388  * Failure to inject an interrupt should give us the information
3389  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
3390  * when fetching the interrupt redirection bitmap in the real-mode
3391  * tss, this doesn't happen.  So we do it ourselves.
3392  */
3393 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3394 {
3395         vmx->rmode.irq.pending = 0;
3396         if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3397                 return;
3398         kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3399         if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3400                 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3401                 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3402                 return;
3403         }
3404         vmx->idt_vectoring_info =
3405                 VECTORING_INFO_VALID_MASK
3406                 | INTR_TYPE_EXT_INTR
3407                 | vmx->rmode.irq.vector;
3408 }
3409
3410 #ifdef CONFIG_X86_64
3411 #define R "r"
3412 #define Q "q"
3413 #else
3414 #define R "e"
3415 #define Q "l"
3416 #endif
3417
3418 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3419 {
3420         struct vcpu_vmx *vmx = to_vmx(vcpu);
3421
3422         /* Record the guest's net vcpu time for enforced NMI injections. */
3423         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3424                 vmx->entry_time = ktime_get();
3425
3426         /* Handle invalid guest state instead of entering VMX */
3427         if (vmx->emulation_required && emulate_invalid_guest_state) {
3428                 handle_invalid_guest_state(vcpu, kvm_run);
3429                 return;
3430         }
3431
3432         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3433                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3434         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3435                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3436
3437         /*
3438          * Loading guest fpu may have cleared host cr0.ts
3439          */
3440         vmcs_writel(HOST_CR0, read_cr0());
3441
3442         set_debugreg(vcpu->arch.dr6, 6);
3443
3444         asm(
3445                 /* Store host registers */
3446                 "push %%"R"dx; push %%"R"bp;"
3447                 "push %%"R"cx \n\t"
3448                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3449                 "je 1f \n\t"
3450                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3451                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3452                 "1: \n\t"
3453                 /* Check if vmlaunch of vmresume is needed */
3454                 "cmpl $0, %c[launched](%0) \n\t"
3455                 /* Load guest registers.  Don't clobber flags. */
3456                 "mov %c[cr2](%0), %%"R"ax \n\t"
3457                 "mov %%"R"ax, %%cr2 \n\t"
3458                 "mov %c[rax](%0), %%"R"ax \n\t"
3459                 "mov %c[rbx](%0), %%"R"bx \n\t"
3460                 "mov %c[rdx](%0), %%"R"dx \n\t"
3461                 "mov %c[rsi](%0), %%"R"si \n\t"
3462                 "mov %c[rdi](%0), %%"R"di \n\t"
3463                 "mov %c[rbp](%0), %%"R"bp \n\t"
3464 #ifdef CONFIG_X86_64
3465                 "mov %c[r8](%0),  %%r8  \n\t"
3466                 "mov %c[r9](%0),  %%r9  \n\t"
3467                 "mov %c[r10](%0), %%r10 \n\t"
3468                 "mov %c[r11](%0), %%r11 \n\t"
3469                 "mov %c[r12](%0), %%r12 \n\t"
3470                 "mov %c[r13](%0), %%r13 \n\t"
3471                 "mov %c[r14](%0), %%r14 \n\t"
3472                 "mov %c[r15](%0), %%r15 \n\t"
3473 #endif
3474                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3475
3476                 /* Enter guest mode */
3477                 "jne .Llaunched \n\t"
3478                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3479                 "jmp .Lkvm_vmx_return \n\t"
3480                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3481                 ".Lkvm_vmx_return: "
3482                 /* Save guest registers, load host registers, keep flags */
3483                 "xchg %0,     (%%"R"sp) \n\t"
3484                 "mov %%"R"ax, %c[rax](%0) \n\t"
3485                 "mov %%"R"bx, %c[rbx](%0) \n\t"
3486                 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3487                 "mov %%"R"dx, %c[rdx](%0) \n\t"
3488                 "mov %%"R"si, %c[rsi](%0) \n\t"
3489                 "mov %%"R"di, %c[rdi](%0) \n\t"
3490                 "mov %%"R"bp, %c[rbp](%0) \n\t"
3491 #ifdef CONFIG_X86_64
3492                 "mov %%r8,  %c[r8](%0) \n\t"
3493                 "mov %%r9,  %c[r9](%0) \n\t"
3494                 "mov %%r10, %c[r10](%0) \n\t"
3495                 "mov %%r11, %c[r11](%0) \n\t"
3496                 "mov %%r12, %c[r12](%0) \n\t"
3497                 "mov %%r13, %c[r13](%0) \n\t"
3498                 "mov %%r14, %c[r14](%0) \n\t"
3499                 "mov %%r15, %c[r15](%0) \n\t"
3500 #endif
3501                 "mov %%cr2, %%"R"ax   \n\t"
3502                 "mov %%"R"ax, %c[cr2](%0) \n\t"
3503
3504                 "pop  %%"R"bp; pop  %%"R"bp; pop  %%"R"dx \n\t"
3505                 "setbe %c[fail](%0) \n\t"
3506               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3507                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3508                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3509                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3510                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3511                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3512                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3513                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3514                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3515                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3516                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3517 #ifdef CONFIG_X86_64
3518                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3519                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3520                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3521                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3522                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3523                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3524                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3525                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3526 #endif
3527                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3528               : "cc", "memory"
3529                 , R"bx", R"di", R"si"
3530 #ifdef CONFIG_X86_64
3531                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3532 #endif
3533               );
3534
3535         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3536         vcpu->arch.regs_dirty = 0;
3537
3538         get_debugreg(vcpu->arch.dr6, 6);
3539
3540         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3541         if (vmx->rmode.irq.pending)
3542                 fixup_rmode_irq(vmx);
3543
3544         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3545         vmx->launched = 1;
3546
3547         vmx_complete_interrupts(vmx);
3548 }
3549
3550 #undef R
3551 #undef Q
3552
3553 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3554 {
3555         struct vcpu_vmx *vmx = to_vmx(vcpu);
3556
3557         if (vmx->vmcs) {
3558                 vcpu_clear(vmx);
3559                 free_vmcs(vmx->vmcs);
3560                 vmx->vmcs = NULL;
3561         }
3562 }
3563
3564 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3565 {
3566         struct vcpu_vmx *vmx = to_vmx(vcpu);
3567
3568         spin_lock(&vmx_vpid_lock);
3569         if (vmx->vpid != 0)
3570                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3571         spin_unlock(&vmx_vpid_lock);
3572         vmx_free_vmcs(vcpu);
3573         kfree(vmx->host_msrs);
3574         kfree(vmx->guest_msrs);
3575         kvm_vcpu_uninit(vcpu);
3576         kmem_cache_free(kvm_vcpu_cache, vmx);
3577 }
3578
3579 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3580 {
3581         int err;
3582         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3583         int cpu;
3584
3585         if (!vmx)
3586                 return ERR_PTR(-ENOMEM);
3587
3588         allocate_vpid(vmx);
3589
3590         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3591         if (err)
3592                 goto free_vcpu;
3593
3594         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3595         if (!vmx->guest_msrs) {
3596                 err = -ENOMEM;
3597                 goto uninit_vcpu;
3598         }
3599
3600         vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3601         if (!vmx->host_msrs)
3602                 goto free_guest_msrs;
3603
3604         vmx->vmcs = alloc_vmcs();
3605         if (!vmx->vmcs)
3606                 goto free_msrs;
3607
3608         vmcs_clear(vmx->vmcs);
3609
3610         cpu = get_cpu();
3611         vmx_vcpu_load(&vmx->vcpu, cpu);
3612         err = vmx_vcpu_setup(vmx);
3613         vmx_vcpu_put(&vmx->vcpu);
3614         put_cpu();
3615         if (err)
3616                 goto free_vmcs;
3617         if (vm_need_virtualize_apic_accesses(kvm))
3618                 if (alloc_apic_access_page(kvm) != 0)
3619                         goto free_vmcs;
3620
3621         if (enable_ept)
3622                 if (alloc_identity_pagetable(kvm) != 0)
3623                         goto free_vmcs;
3624
3625         return &vmx->vcpu;
3626
3627 free_vmcs:
3628         free_vmcs(vmx->vmcs);
3629 free_msrs:
3630         kfree(vmx->host_msrs);
3631 free_guest_msrs:
3632         kfree(vmx->guest_msrs);
3633 uninit_vcpu:
3634         kvm_vcpu_uninit(&vmx->vcpu);
3635 free_vcpu:
3636         kmem_cache_free(kvm_vcpu_cache, vmx);
3637         return ERR_PTR(err);
3638 }
3639
3640 static void __init vmx_check_processor_compat(void *rtn)
3641 {
3642         struct vmcs_config vmcs_conf;
3643
3644         *(int *)rtn = 0;
3645         if (setup_vmcs_config(&vmcs_conf) < 0)
3646                 *(int *)rtn = -EIO;
3647         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3648                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3649                                 smp_processor_id());
3650                 *(int *)rtn = -EIO;
3651         }
3652 }
3653
3654 static int get_ept_level(void)
3655 {
3656         return VMX_EPT_DEFAULT_GAW + 1;
3657 }
3658
3659 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3660 {
3661         u64 ret;
3662
3663         /* For VT-d and EPT combination
3664          * 1. MMIO: always map as UC
3665          * 2. EPT with VT-d:
3666          *   a. VT-d without snooping control feature: can't guarantee the
3667          *      result, try to trust guest.
3668          *   b. VT-d with snooping control feature: snooping control feature of
3669          *      VT-d engine can guarantee the cache correctness. Just set it
3670          *      to WB to keep consistent with host. So the same as item 3.
3671          * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3672          *    consistent with host MTRR
3673          */
3674         if (is_mmio)
3675                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
3676         else if (vcpu->kvm->arch.iommu_domain &&
3677                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3678                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3679                       VMX_EPT_MT_EPTE_SHIFT;
3680         else
3681                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3682                         | VMX_EPT_IGMT_BIT;
3683
3684         return ret;
3685 }
3686
3687 static struct kvm_x86_ops vmx_x86_ops = {
3688         .cpu_has_kvm_support = cpu_has_kvm_support,
3689         .disabled_by_bios = vmx_disabled_by_bios,
3690         .hardware_setup = hardware_setup,
3691         .hardware_unsetup = hardware_unsetup,
3692         .check_processor_compatibility = vmx_check_processor_compat,
3693         .hardware_enable = hardware_enable,
3694         .hardware_disable = hardware_disable,
3695         .cpu_has_accelerated_tpr = report_flexpriority,
3696
3697         .vcpu_create = vmx_create_vcpu,
3698         .vcpu_free = vmx_free_vcpu,
3699         .vcpu_reset = vmx_vcpu_reset,
3700
3701         .prepare_guest_switch = vmx_save_host_state,
3702         .vcpu_load = vmx_vcpu_load,
3703         .vcpu_put = vmx_vcpu_put,
3704
3705         .set_guest_debug = set_guest_debug,
3706         .get_msr = vmx_get_msr,
3707         .set_msr = vmx_set_msr,
3708         .get_segment_base = vmx_get_segment_base,
3709         .get_segment = vmx_get_segment,
3710         .set_segment = vmx_set_segment,
3711         .get_cpl = vmx_get_cpl,
3712         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3713         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3714         .set_cr0 = vmx_set_cr0,
3715         .set_cr3 = vmx_set_cr3,
3716         .set_cr4 = vmx_set_cr4,
3717         .set_efer = vmx_set_efer,
3718         .get_idt = vmx_get_idt,
3719         .set_idt = vmx_set_idt,
3720         .get_gdt = vmx_get_gdt,
3721         .set_gdt = vmx_set_gdt,
3722         .cache_reg = vmx_cache_reg,
3723         .get_rflags = vmx_get_rflags,
3724         .set_rflags = vmx_set_rflags,
3725
3726         .tlb_flush = vmx_flush_tlb,
3727
3728         .run = vmx_vcpu_run,
3729         .handle_exit = vmx_handle_exit,
3730         .skip_emulated_instruction = skip_emulated_instruction,
3731         .set_interrupt_shadow = vmx_set_interrupt_shadow,
3732         .get_interrupt_shadow = vmx_get_interrupt_shadow,
3733         .patch_hypercall = vmx_patch_hypercall,
3734         .set_irq = vmx_inject_irq,
3735         .set_nmi = vmx_inject_nmi,
3736         .queue_exception = vmx_queue_exception,
3737         .interrupt_allowed = vmx_interrupt_allowed,
3738         .nmi_allowed = vmx_nmi_allowed,
3739         .enable_nmi_window = enable_nmi_window,
3740         .enable_irq_window = enable_irq_window,
3741         .update_cr8_intercept = update_cr8_intercept,
3742
3743         .set_tss_addr = vmx_set_tss_addr,
3744         .get_tdp_level = get_ept_level,
3745         .get_mt_mask = vmx_get_mt_mask,
3746 };
3747
3748 static int __init vmx_init(void)
3749 {
3750         int r;
3751
3752         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
3753         if (!vmx_io_bitmap_a)
3754                 return -ENOMEM;
3755
3756         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
3757         if (!vmx_io_bitmap_b) {
3758                 r = -ENOMEM;
3759                 goto out;
3760         }
3761
3762         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
3763         if (!vmx_msr_bitmap_legacy) {
3764                 r = -ENOMEM;
3765                 goto out1;
3766         }
3767
3768         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
3769         if (!vmx_msr_bitmap_longmode) {
3770                 r = -ENOMEM;
3771                 goto out2;
3772         }
3773
3774         /*
3775          * Allow direct access to the PC debug port (it is often used for I/O
3776          * delays, but the vmexits simply slow things down).
3777          */
3778         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
3779         clear_bit(0x80, vmx_io_bitmap_a);
3780
3781         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
3782
3783         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
3784         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
3785
3786         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
3787
3788         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
3789         if (r)
3790                 goto out3;
3791
3792         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
3793         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
3794         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
3795         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
3796         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
3797         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
3798
3799         if (enable_ept) {
3800                 bypass_guest_pf = 0;
3801                 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3802                         VMX_EPT_WRITABLE_MASK);
3803                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
3804                                 VMX_EPT_EXECUTABLE_MASK);
3805                 kvm_enable_tdp();
3806         } else
3807                 kvm_disable_tdp();
3808
3809         if (bypass_guest_pf)
3810                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
3811
3812         ept_sync_global();
3813
3814         return 0;
3815
3816 out3:
3817         free_page((unsigned long)vmx_msr_bitmap_longmode);
3818 out2:
3819         free_page((unsigned long)vmx_msr_bitmap_legacy);
3820 out1:
3821         free_page((unsigned long)vmx_io_bitmap_b);
3822 out:
3823         free_page((unsigned long)vmx_io_bitmap_a);
3824         return r;
3825 }
3826
3827 static void __exit vmx_exit(void)
3828 {
3829         free_page((unsigned long)vmx_msr_bitmap_legacy);
3830         free_page((unsigned long)vmx_msr_bitmap_longmode);
3831         free_page((unsigned long)vmx_io_bitmap_b);
3832         free_page((unsigned long)vmx_io_bitmap_a);
3833
3834         kvm_exit();
3835 }
3836
3837 module_init(vmx_init)
3838 module_exit(vmx_exit)