KVM: VMX: When emulating on invalid vmx state, don't return to userspace unnecessarily
[sfrench/cifs-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
36 #define __ex(x) __kvm_handle_fault_on_reboot(x)
37
38 MODULE_AUTHOR("Qumranet");
39 MODULE_LICENSE("GPL");
40
41 static int bypass_guest_pf = 1;
42 module_param(bypass_guest_pf, bool, 0);
43
44 static int enable_vpid = 1;
45 module_param(enable_vpid, bool, 0);
46
47 static int flexpriority_enabled = 1;
48 module_param(flexpriority_enabled, bool, 0);
49
50 static int enable_ept = 1;
51 module_param(enable_ept, bool, 0);
52
53 static int emulate_invalid_guest_state = 0;
54 module_param(emulate_invalid_guest_state, bool, 0);
55
56 struct vmcs {
57         u32 revision_id;
58         u32 abort;
59         char data[0];
60 };
61
62 struct vcpu_vmx {
63         struct kvm_vcpu       vcpu;
64         struct list_head      local_vcpus_link;
65         unsigned long         host_rsp;
66         int                   launched;
67         u8                    fail;
68         u32                   idt_vectoring_info;
69         struct kvm_msr_entry *guest_msrs;
70         struct kvm_msr_entry *host_msrs;
71         int                   nmsrs;
72         int                   save_nmsrs;
73         int                   msr_offset_efer;
74 #ifdef CONFIG_X86_64
75         int                   msr_offset_kernel_gs_base;
76 #endif
77         struct vmcs          *vmcs;
78         struct {
79                 int           loaded;
80                 u16           fs_sel, gs_sel, ldt_sel;
81                 int           gs_ldt_reload_needed;
82                 int           fs_reload_needed;
83                 int           guest_efer_loaded;
84         } host_state;
85         struct {
86                 struct {
87                         bool pending;
88                         u8 vector;
89                         unsigned rip;
90                 } irq;
91         } rmode;
92         int vpid;
93         bool emulation_required;
94         enum emulation_result invalid_state_emulation_result;
95
96         /* Support for vnmi-less CPUs */
97         int soft_vnmi_blocked;
98         ktime_t entry_time;
99         s64 vnmi_blocked_time;
100 };
101
102 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
103 {
104         return container_of(vcpu, struct vcpu_vmx, vcpu);
105 }
106
107 static int init_rmode(struct kvm *kvm);
108 static u64 construct_eptp(unsigned long root_hpa);
109
110 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
111 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
112 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
113
114 static struct page *vmx_io_bitmap_a;
115 static struct page *vmx_io_bitmap_b;
116 static struct page *vmx_msr_bitmap;
117
118 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
119 static DEFINE_SPINLOCK(vmx_vpid_lock);
120
121 static struct vmcs_config {
122         int size;
123         int order;
124         u32 revision_id;
125         u32 pin_based_exec_ctrl;
126         u32 cpu_based_exec_ctrl;
127         u32 cpu_based_2nd_exec_ctrl;
128         u32 vmexit_ctrl;
129         u32 vmentry_ctrl;
130 } vmcs_config;
131
132 static struct vmx_capability {
133         u32 ept;
134         u32 vpid;
135 } vmx_capability;
136
137 #define VMX_SEGMENT_FIELD(seg)                                  \
138         [VCPU_SREG_##seg] = {                                   \
139                 .selector = GUEST_##seg##_SELECTOR,             \
140                 .base = GUEST_##seg##_BASE,                     \
141                 .limit = GUEST_##seg##_LIMIT,                   \
142                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
143         }
144
145 static struct kvm_vmx_segment_field {
146         unsigned selector;
147         unsigned base;
148         unsigned limit;
149         unsigned ar_bytes;
150 } kvm_vmx_segment_fields[] = {
151         VMX_SEGMENT_FIELD(CS),
152         VMX_SEGMENT_FIELD(DS),
153         VMX_SEGMENT_FIELD(ES),
154         VMX_SEGMENT_FIELD(FS),
155         VMX_SEGMENT_FIELD(GS),
156         VMX_SEGMENT_FIELD(SS),
157         VMX_SEGMENT_FIELD(TR),
158         VMX_SEGMENT_FIELD(LDTR),
159 };
160
161 /*
162  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
163  * away by decrementing the array size.
164  */
165 static const u32 vmx_msr_index[] = {
166 #ifdef CONFIG_X86_64
167         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
168 #endif
169         MSR_EFER, MSR_K6_STAR,
170 };
171 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
172
173 static void load_msrs(struct kvm_msr_entry *e, int n)
174 {
175         int i;
176
177         for (i = 0; i < n; ++i)
178                 wrmsrl(e[i].index, e[i].data);
179 }
180
181 static void save_msrs(struct kvm_msr_entry *e, int n)
182 {
183         int i;
184
185         for (i = 0; i < n; ++i)
186                 rdmsrl(e[i].index, e[i].data);
187 }
188
189 static inline int is_page_fault(u32 intr_info)
190 {
191         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
192                              INTR_INFO_VALID_MASK)) ==
193                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
194 }
195
196 static inline int is_no_device(u32 intr_info)
197 {
198         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
199                              INTR_INFO_VALID_MASK)) ==
200                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
201 }
202
203 static inline int is_invalid_opcode(u32 intr_info)
204 {
205         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
206                              INTR_INFO_VALID_MASK)) ==
207                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
208 }
209
210 static inline int is_external_interrupt(u32 intr_info)
211 {
212         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
213                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
214 }
215
216 static inline int cpu_has_vmx_msr_bitmap(void)
217 {
218         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS);
219 }
220
221 static inline int cpu_has_vmx_tpr_shadow(void)
222 {
223         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
224 }
225
226 static inline int vm_need_tpr_shadow(struct kvm *kvm)
227 {
228         return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
229 }
230
231 static inline int cpu_has_secondary_exec_ctrls(void)
232 {
233         return (vmcs_config.cpu_based_exec_ctrl &
234                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
235 }
236
237 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
238 {
239         return flexpriority_enabled
240                 && (vmcs_config.cpu_based_2nd_exec_ctrl &
241                     SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
242 }
243
244 static inline int cpu_has_vmx_invept_individual_addr(void)
245 {
246         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT));
247 }
248
249 static inline int cpu_has_vmx_invept_context(void)
250 {
251         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT));
252 }
253
254 static inline int cpu_has_vmx_invept_global(void)
255 {
256         return (!!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT));
257 }
258
259 static inline int cpu_has_vmx_ept(void)
260 {
261         return (vmcs_config.cpu_based_2nd_exec_ctrl &
262                 SECONDARY_EXEC_ENABLE_EPT);
263 }
264
265 static inline int vm_need_ept(void)
266 {
267         return (cpu_has_vmx_ept() && enable_ept);
268 }
269
270 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
271 {
272         return ((cpu_has_vmx_virtualize_apic_accesses()) &&
273                 (irqchip_in_kernel(kvm)));
274 }
275
276 static inline int cpu_has_vmx_vpid(void)
277 {
278         return (vmcs_config.cpu_based_2nd_exec_ctrl &
279                 SECONDARY_EXEC_ENABLE_VPID);
280 }
281
282 static inline int cpu_has_virtual_nmis(void)
283 {
284         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
285 }
286
287 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
288 {
289         int i;
290
291         for (i = 0; i < vmx->nmsrs; ++i)
292                 if (vmx->guest_msrs[i].index == msr)
293                         return i;
294         return -1;
295 }
296
297 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
298 {
299     struct {
300         u64 vpid : 16;
301         u64 rsvd : 48;
302         u64 gva;
303     } operand = { vpid, 0, gva };
304
305     asm volatile (__ex(ASM_VMX_INVVPID)
306                   /* CF==1 or ZF==1 --> rc = -1 */
307                   "; ja 1f ; ud2 ; 1:"
308                   : : "a"(&operand), "c"(ext) : "cc", "memory");
309 }
310
311 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
312 {
313         struct {
314                 u64 eptp, gpa;
315         } operand = {eptp, gpa};
316
317         asm volatile (__ex(ASM_VMX_INVEPT)
318                         /* CF==1 or ZF==1 --> rc = -1 */
319                         "; ja 1f ; ud2 ; 1:\n"
320                         : : "a" (&operand), "c" (ext) : "cc", "memory");
321 }
322
323 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
324 {
325         int i;
326
327         i = __find_msr_index(vmx, msr);
328         if (i >= 0)
329                 return &vmx->guest_msrs[i];
330         return NULL;
331 }
332
333 static void vmcs_clear(struct vmcs *vmcs)
334 {
335         u64 phys_addr = __pa(vmcs);
336         u8 error;
337
338         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
339                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
340                       : "cc", "memory");
341         if (error)
342                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
343                        vmcs, phys_addr);
344 }
345
346 static void __vcpu_clear(void *arg)
347 {
348         struct vcpu_vmx *vmx = arg;
349         int cpu = raw_smp_processor_id();
350
351         if (vmx->vcpu.cpu == cpu)
352                 vmcs_clear(vmx->vmcs);
353         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
354                 per_cpu(current_vmcs, cpu) = NULL;
355         rdtscll(vmx->vcpu.arch.host_tsc);
356         list_del(&vmx->local_vcpus_link);
357         vmx->vcpu.cpu = -1;
358         vmx->launched = 0;
359 }
360
361 static void vcpu_clear(struct vcpu_vmx *vmx)
362 {
363         if (vmx->vcpu.cpu == -1)
364                 return;
365         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
366 }
367
368 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
369 {
370         if (vmx->vpid == 0)
371                 return;
372
373         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
374 }
375
376 static inline void ept_sync_global(void)
377 {
378         if (cpu_has_vmx_invept_global())
379                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
380 }
381
382 static inline void ept_sync_context(u64 eptp)
383 {
384         if (vm_need_ept()) {
385                 if (cpu_has_vmx_invept_context())
386                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
387                 else
388                         ept_sync_global();
389         }
390 }
391
392 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
393 {
394         if (vm_need_ept()) {
395                 if (cpu_has_vmx_invept_individual_addr())
396                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
397                                         eptp, gpa);
398                 else
399                         ept_sync_context(eptp);
400         }
401 }
402
403 static unsigned long vmcs_readl(unsigned long field)
404 {
405         unsigned long value;
406
407         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
408                       : "=a"(value) : "d"(field) : "cc");
409         return value;
410 }
411
412 static u16 vmcs_read16(unsigned long field)
413 {
414         return vmcs_readl(field);
415 }
416
417 static u32 vmcs_read32(unsigned long field)
418 {
419         return vmcs_readl(field);
420 }
421
422 static u64 vmcs_read64(unsigned long field)
423 {
424 #ifdef CONFIG_X86_64
425         return vmcs_readl(field);
426 #else
427         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
428 #endif
429 }
430
431 static noinline void vmwrite_error(unsigned long field, unsigned long value)
432 {
433         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
434                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
435         dump_stack();
436 }
437
438 static void vmcs_writel(unsigned long field, unsigned long value)
439 {
440         u8 error;
441
442         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
443                        : "=q"(error) : "a"(value), "d"(field) : "cc");
444         if (unlikely(error))
445                 vmwrite_error(field, value);
446 }
447
448 static void vmcs_write16(unsigned long field, u16 value)
449 {
450         vmcs_writel(field, value);
451 }
452
453 static void vmcs_write32(unsigned long field, u32 value)
454 {
455         vmcs_writel(field, value);
456 }
457
458 static void vmcs_write64(unsigned long field, u64 value)
459 {
460         vmcs_writel(field, value);
461 #ifndef CONFIG_X86_64
462         asm volatile ("");
463         vmcs_writel(field+1, value >> 32);
464 #endif
465 }
466
467 static void vmcs_clear_bits(unsigned long field, u32 mask)
468 {
469         vmcs_writel(field, vmcs_readl(field) & ~mask);
470 }
471
472 static void vmcs_set_bits(unsigned long field, u32 mask)
473 {
474         vmcs_writel(field, vmcs_readl(field) | mask);
475 }
476
477 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
478 {
479         u32 eb;
480
481         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
482         if (!vcpu->fpu_active)
483                 eb |= 1u << NM_VECTOR;
484         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
485                 if (vcpu->guest_debug &
486                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
487                         eb |= 1u << DB_VECTOR;
488                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
489                         eb |= 1u << BP_VECTOR;
490         }
491         if (vcpu->arch.rmode.active)
492                 eb = ~0;
493         if (vm_need_ept())
494                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
495         vmcs_write32(EXCEPTION_BITMAP, eb);
496 }
497
498 static void reload_tss(void)
499 {
500         /*
501          * VT restores TR but not its size.  Useless.
502          */
503         struct descriptor_table gdt;
504         struct desc_struct *descs;
505
506         kvm_get_gdt(&gdt);
507         descs = (void *)gdt.base;
508         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
509         load_TR_desc();
510 }
511
512 static void load_transition_efer(struct vcpu_vmx *vmx)
513 {
514         int efer_offset = vmx->msr_offset_efer;
515         u64 host_efer = vmx->host_msrs[efer_offset].data;
516         u64 guest_efer = vmx->guest_msrs[efer_offset].data;
517         u64 ignore_bits;
518
519         if (efer_offset < 0)
520                 return;
521         /*
522          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
523          * outside long mode
524          */
525         ignore_bits = EFER_NX | EFER_SCE;
526 #ifdef CONFIG_X86_64
527         ignore_bits |= EFER_LMA | EFER_LME;
528         /* SCE is meaningful only in long mode on Intel */
529         if (guest_efer & EFER_LMA)
530                 ignore_bits &= ~(u64)EFER_SCE;
531 #endif
532         if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
533                 return;
534
535         vmx->host_state.guest_efer_loaded = 1;
536         guest_efer &= ~ignore_bits;
537         guest_efer |= host_efer & ignore_bits;
538         wrmsrl(MSR_EFER, guest_efer);
539         vmx->vcpu.stat.efer_reload++;
540 }
541
542 static void reload_host_efer(struct vcpu_vmx *vmx)
543 {
544         if (vmx->host_state.guest_efer_loaded) {
545                 vmx->host_state.guest_efer_loaded = 0;
546                 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
547         }
548 }
549
550 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
551 {
552         struct vcpu_vmx *vmx = to_vmx(vcpu);
553
554         if (vmx->host_state.loaded)
555                 return;
556
557         vmx->host_state.loaded = 1;
558         /*
559          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
560          * allow segment selectors with cpl > 0 or ti == 1.
561          */
562         vmx->host_state.ldt_sel = kvm_read_ldt();
563         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
564         vmx->host_state.fs_sel = kvm_read_fs();
565         if (!(vmx->host_state.fs_sel & 7)) {
566                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
567                 vmx->host_state.fs_reload_needed = 0;
568         } else {
569                 vmcs_write16(HOST_FS_SELECTOR, 0);
570                 vmx->host_state.fs_reload_needed = 1;
571         }
572         vmx->host_state.gs_sel = kvm_read_gs();
573         if (!(vmx->host_state.gs_sel & 7))
574                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
575         else {
576                 vmcs_write16(HOST_GS_SELECTOR, 0);
577                 vmx->host_state.gs_ldt_reload_needed = 1;
578         }
579
580 #ifdef CONFIG_X86_64
581         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
582         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
583 #else
584         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
585         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
586 #endif
587
588 #ifdef CONFIG_X86_64
589         if (is_long_mode(&vmx->vcpu))
590                 save_msrs(vmx->host_msrs +
591                           vmx->msr_offset_kernel_gs_base, 1);
592
593 #endif
594         load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
595         load_transition_efer(vmx);
596 }
597
598 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
599 {
600         unsigned long flags;
601
602         if (!vmx->host_state.loaded)
603                 return;
604
605         ++vmx->vcpu.stat.host_state_reload;
606         vmx->host_state.loaded = 0;
607         if (vmx->host_state.fs_reload_needed)
608                 kvm_load_fs(vmx->host_state.fs_sel);
609         if (vmx->host_state.gs_ldt_reload_needed) {
610                 kvm_load_ldt(vmx->host_state.ldt_sel);
611                 /*
612                  * If we have to reload gs, we must take care to
613                  * preserve our gs base.
614                  */
615                 local_irq_save(flags);
616                 kvm_load_gs(vmx->host_state.gs_sel);
617 #ifdef CONFIG_X86_64
618                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
619 #endif
620                 local_irq_restore(flags);
621         }
622         reload_tss();
623         save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
624         load_msrs(vmx->host_msrs, vmx->save_nmsrs);
625         reload_host_efer(vmx);
626 }
627
628 static void vmx_load_host_state(struct vcpu_vmx *vmx)
629 {
630         preempt_disable();
631         __vmx_load_host_state(vmx);
632         preempt_enable();
633 }
634
635 /*
636  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
637  * vcpu mutex is already taken.
638  */
639 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
640 {
641         struct vcpu_vmx *vmx = to_vmx(vcpu);
642         u64 phys_addr = __pa(vmx->vmcs);
643         u64 tsc_this, delta, new_offset;
644
645         if (vcpu->cpu != cpu) {
646                 vcpu_clear(vmx);
647                 kvm_migrate_timers(vcpu);
648                 vpid_sync_vcpu_all(vmx);
649                 local_irq_disable();
650                 list_add(&vmx->local_vcpus_link,
651                          &per_cpu(vcpus_on_cpu, cpu));
652                 local_irq_enable();
653         }
654
655         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
656                 u8 error;
657
658                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
659                 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
660                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
661                               : "cc");
662                 if (error)
663                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
664                                vmx->vmcs, phys_addr);
665         }
666
667         if (vcpu->cpu != cpu) {
668                 struct descriptor_table dt;
669                 unsigned long sysenter_esp;
670
671                 vcpu->cpu = cpu;
672                 /*
673                  * Linux uses per-cpu TSS and GDT, so set these when switching
674                  * processors.
675                  */
676                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
677                 kvm_get_gdt(&dt);
678                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
679
680                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
681                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
682
683                 /*
684                  * Make sure the time stamp counter is monotonous.
685                  */
686                 rdtscll(tsc_this);
687                 if (tsc_this < vcpu->arch.host_tsc) {
688                         delta = vcpu->arch.host_tsc - tsc_this;
689                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
690                         vmcs_write64(TSC_OFFSET, new_offset);
691                 }
692         }
693 }
694
695 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
696 {
697         __vmx_load_host_state(to_vmx(vcpu));
698 }
699
700 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
701 {
702         if (vcpu->fpu_active)
703                 return;
704         vcpu->fpu_active = 1;
705         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
706         if (vcpu->arch.cr0 & X86_CR0_TS)
707                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
708         update_exception_bitmap(vcpu);
709 }
710
711 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
712 {
713         if (!vcpu->fpu_active)
714                 return;
715         vcpu->fpu_active = 0;
716         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
717         update_exception_bitmap(vcpu);
718 }
719
720 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
721 {
722         return vmcs_readl(GUEST_RFLAGS);
723 }
724
725 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
726 {
727         if (vcpu->arch.rmode.active)
728                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
729         vmcs_writel(GUEST_RFLAGS, rflags);
730 }
731
732 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
733 {
734         unsigned long rip;
735         u32 interruptibility;
736
737         rip = kvm_rip_read(vcpu);
738         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
739         kvm_rip_write(vcpu, rip);
740
741         /*
742          * We emulated an instruction, so temporary interrupt blocking
743          * should be removed, if set.
744          */
745         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
746         if (interruptibility & 3)
747                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
748                              interruptibility & ~3);
749         vcpu->arch.interrupt_window_open = 1;
750 }
751
752 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
753                                 bool has_error_code, u32 error_code)
754 {
755         struct vcpu_vmx *vmx = to_vmx(vcpu);
756         u32 intr_info = nr | INTR_INFO_VALID_MASK;
757
758         if (has_error_code) {
759                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
760                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
761         }
762
763         if (vcpu->arch.rmode.active) {
764                 vmx->rmode.irq.pending = true;
765                 vmx->rmode.irq.vector = nr;
766                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
767                 if (nr == BP_VECTOR || nr == OF_VECTOR)
768                         vmx->rmode.irq.rip++;
769                 intr_info |= INTR_TYPE_SOFT_INTR;
770                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
771                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
772                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
773                 return;
774         }
775
776         if (nr == BP_VECTOR || nr == OF_VECTOR) {
777                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
778                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
779         } else
780                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
781
782         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
783 }
784
785 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
786 {
787         return false;
788 }
789
790 /*
791  * Swap MSR entry in host/guest MSR entry array.
792  */
793 #ifdef CONFIG_X86_64
794 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
795 {
796         struct kvm_msr_entry tmp;
797
798         tmp = vmx->guest_msrs[to];
799         vmx->guest_msrs[to] = vmx->guest_msrs[from];
800         vmx->guest_msrs[from] = tmp;
801         tmp = vmx->host_msrs[to];
802         vmx->host_msrs[to] = vmx->host_msrs[from];
803         vmx->host_msrs[from] = tmp;
804 }
805 #endif
806
807 /*
808  * Set up the vmcs to automatically save and restore system
809  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
810  * mode, as fiddling with msrs is very expensive.
811  */
812 static void setup_msrs(struct vcpu_vmx *vmx)
813 {
814         int save_nmsrs;
815
816         vmx_load_host_state(vmx);
817         save_nmsrs = 0;
818 #ifdef CONFIG_X86_64
819         if (is_long_mode(&vmx->vcpu)) {
820                 int index;
821
822                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
823                 if (index >= 0)
824                         move_msr_up(vmx, index, save_nmsrs++);
825                 index = __find_msr_index(vmx, MSR_LSTAR);
826                 if (index >= 0)
827                         move_msr_up(vmx, index, save_nmsrs++);
828                 index = __find_msr_index(vmx, MSR_CSTAR);
829                 if (index >= 0)
830                         move_msr_up(vmx, index, save_nmsrs++);
831                 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
832                 if (index >= 0)
833                         move_msr_up(vmx, index, save_nmsrs++);
834                 /*
835                  * MSR_K6_STAR is only needed on long mode guests, and only
836                  * if efer.sce is enabled.
837                  */
838                 index = __find_msr_index(vmx, MSR_K6_STAR);
839                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
840                         move_msr_up(vmx, index, save_nmsrs++);
841         }
842 #endif
843         vmx->save_nmsrs = save_nmsrs;
844
845 #ifdef CONFIG_X86_64
846         vmx->msr_offset_kernel_gs_base =
847                 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
848 #endif
849         vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
850 }
851
852 /*
853  * reads and returns guest's timestamp counter "register"
854  * guest_tsc = host_tsc + tsc_offset    -- 21.3
855  */
856 static u64 guest_read_tsc(void)
857 {
858         u64 host_tsc, tsc_offset;
859
860         rdtscll(host_tsc);
861         tsc_offset = vmcs_read64(TSC_OFFSET);
862         return host_tsc + tsc_offset;
863 }
864
865 /*
866  * writes 'guest_tsc' into guest's timestamp counter "register"
867  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
868  */
869 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
870 {
871         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
872 }
873
874 /*
875  * Reads an msr value (of 'msr_index') into 'pdata'.
876  * Returns 0 on success, non-0 otherwise.
877  * Assumes vcpu_load() was already called.
878  */
879 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
880 {
881         u64 data;
882         struct kvm_msr_entry *msr;
883
884         if (!pdata) {
885                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
886                 return -EINVAL;
887         }
888
889         switch (msr_index) {
890 #ifdef CONFIG_X86_64
891         case MSR_FS_BASE:
892                 data = vmcs_readl(GUEST_FS_BASE);
893                 break;
894         case MSR_GS_BASE:
895                 data = vmcs_readl(GUEST_GS_BASE);
896                 break;
897         case MSR_EFER:
898                 return kvm_get_msr_common(vcpu, msr_index, pdata);
899 #endif
900         case MSR_IA32_TIME_STAMP_COUNTER:
901                 data = guest_read_tsc();
902                 break;
903         case MSR_IA32_SYSENTER_CS:
904                 data = vmcs_read32(GUEST_SYSENTER_CS);
905                 break;
906         case MSR_IA32_SYSENTER_EIP:
907                 data = vmcs_readl(GUEST_SYSENTER_EIP);
908                 break;
909         case MSR_IA32_SYSENTER_ESP:
910                 data = vmcs_readl(GUEST_SYSENTER_ESP);
911                 break;
912         default:
913                 vmx_load_host_state(to_vmx(vcpu));
914                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
915                 if (msr) {
916                         data = msr->data;
917                         break;
918                 }
919                 return kvm_get_msr_common(vcpu, msr_index, pdata);
920         }
921
922         *pdata = data;
923         return 0;
924 }
925
926 /*
927  * Writes msr value into into the appropriate "register".
928  * Returns 0 on success, non-0 otherwise.
929  * Assumes vcpu_load() was already called.
930  */
931 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
932 {
933         struct vcpu_vmx *vmx = to_vmx(vcpu);
934         struct kvm_msr_entry *msr;
935         u64 host_tsc;
936         int ret = 0;
937
938         switch (msr_index) {
939 #ifdef CONFIG_X86_64
940         case MSR_EFER:
941                 vmx_load_host_state(vmx);
942                 ret = kvm_set_msr_common(vcpu, msr_index, data);
943                 break;
944         case MSR_FS_BASE:
945                 vmcs_writel(GUEST_FS_BASE, data);
946                 break;
947         case MSR_GS_BASE:
948                 vmcs_writel(GUEST_GS_BASE, data);
949                 break;
950 #endif
951         case MSR_IA32_SYSENTER_CS:
952                 vmcs_write32(GUEST_SYSENTER_CS, data);
953                 break;
954         case MSR_IA32_SYSENTER_EIP:
955                 vmcs_writel(GUEST_SYSENTER_EIP, data);
956                 break;
957         case MSR_IA32_SYSENTER_ESP:
958                 vmcs_writel(GUEST_SYSENTER_ESP, data);
959                 break;
960         case MSR_IA32_TIME_STAMP_COUNTER:
961                 rdtscll(host_tsc);
962                 guest_write_tsc(data, host_tsc);
963                 break;
964         case MSR_P6_PERFCTR0:
965         case MSR_P6_PERFCTR1:
966         case MSR_P6_EVNTSEL0:
967         case MSR_P6_EVNTSEL1:
968                 /*
969                  * Just discard all writes to the performance counters; this
970                  * should keep both older linux and windows 64-bit guests
971                  * happy
972                  */
973                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", msr_index, data);
974
975                 break;
976         case MSR_IA32_CR_PAT:
977                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
978                         vmcs_write64(GUEST_IA32_PAT, data);
979                         vcpu->arch.pat = data;
980                         break;
981                 }
982                 /* Otherwise falls through to kvm_set_msr_common */
983         default:
984                 vmx_load_host_state(vmx);
985                 msr = find_msr_entry(vmx, msr_index);
986                 if (msr) {
987                         msr->data = data;
988                         break;
989                 }
990                 ret = kvm_set_msr_common(vcpu, msr_index, data);
991         }
992
993         return ret;
994 }
995
996 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
997 {
998         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
999         switch (reg) {
1000         case VCPU_REGS_RSP:
1001                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1002                 break;
1003         case VCPU_REGS_RIP:
1004                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1005                 break;
1006         default:
1007                 break;
1008         }
1009 }
1010
1011 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1012 {
1013         int old_debug = vcpu->guest_debug;
1014         unsigned long flags;
1015
1016         vcpu->guest_debug = dbg->control;
1017         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1018                 vcpu->guest_debug = 0;
1019
1020         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1021                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1022         else
1023                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1024
1025         flags = vmcs_readl(GUEST_RFLAGS);
1026         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1027                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1028         else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1029                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1030         vmcs_writel(GUEST_RFLAGS, flags);
1031
1032         update_exception_bitmap(vcpu);
1033
1034         return 0;
1035 }
1036
1037 static int vmx_get_irq(struct kvm_vcpu *vcpu)
1038 {
1039         if (!vcpu->arch.interrupt.pending)
1040                 return -1;
1041         return vcpu->arch.interrupt.nr;
1042 }
1043
1044 static __init int cpu_has_kvm_support(void)
1045 {
1046         return cpu_has_vmx();
1047 }
1048
1049 static __init int vmx_disabled_by_bios(void)
1050 {
1051         u64 msr;
1052
1053         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1054         return (msr & (FEATURE_CONTROL_LOCKED |
1055                        FEATURE_CONTROL_VMXON_ENABLED))
1056             == FEATURE_CONTROL_LOCKED;
1057         /* locked but not enabled */
1058 }
1059
1060 static void hardware_enable(void *garbage)
1061 {
1062         int cpu = raw_smp_processor_id();
1063         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1064         u64 old;
1065
1066         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1067         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1068         if ((old & (FEATURE_CONTROL_LOCKED |
1069                     FEATURE_CONTROL_VMXON_ENABLED))
1070             != (FEATURE_CONTROL_LOCKED |
1071                 FEATURE_CONTROL_VMXON_ENABLED))
1072                 /* enable and lock */
1073                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1074                        FEATURE_CONTROL_LOCKED |
1075                        FEATURE_CONTROL_VMXON_ENABLED);
1076         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1077         asm volatile (ASM_VMX_VMXON_RAX
1078                       : : "a"(&phys_addr), "m"(phys_addr)
1079                       : "memory", "cc");
1080 }
1081
1082 static void vmclear_local_vcpus(void)
1083 {
1084         int cpu = raw_smp_processor_id();
1085         struct vcpu_vmx *vmx, *n;
1086
1087         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1088                                  local_vcpus_link)
1089                 __vcpu_clear(vmx);
1090 }
1091
1092
1093 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1094  * tricks.
1095  */
1096 static void kvm_cpu_vmxoff(void)
1097 {
1098         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1099         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1100 }
1101
1102 static void hardware_disable(void *garbage)
1103 {
1104         vmclear_local_vcpus();
1105         kvm_cpu_vmxoff();
1106 }
1107
1108 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1109                                       u32 msr, u32 *result)
1110 {
1111         u32 vmx_msr_low, vmx_msr_high;
1112         u32 ctl = ctl_min | ctl_opt;
1113
1114         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1115
1116         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1117         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1118
1119         /* Ensure minimum (required) set of control bits are supported. */
1120         if (ctl_min & ~ctl)
1121                 return -EIO;
1122
1123         *result = ctl;
1124         return 0;
1125 }
1126
1127 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1128 {
1129         u32 vmx_msr_low, vmx_msr_high;
1130         u32 min, opt, min2, opt2;
1131         u32 _pin_based_exec_control = 0;
1132         u32 _cpu_based_exec_control = 0;
1133         u32 _cpu_based_2nd_exec_control = 0;
1134         u32 _vmexit_control = 0;
1135         u32 _vmentry_control = 0;
1136
1137         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1138         opt = PIN_BASED_VIRTUAL_NMIS;
1139         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1140                                 &_pin_based_exec_control) < 0)
1141                 return -EIO;
1142
1143         min = CPU_BASED_HLT_EXITING |
1144 #ifdef CONFIG_X86_64
1145               CPU_BASED_CR8_LOAD_EXITING |
1146               CPU_BASED_CR8_STORE_EXITING |
1147 #endif
1148               CPU_BASED_CR3_LOAD_EXITING |
1149               CPU_BASED_CR3_STORE_EXITING |
1150               CPU_BASED_USE_IO_BITMAPS |
1151               CPU_BASED_MOV_DR_EXITING |
1152               CPU_BASED_USE_TSC_OFFSETING |
1153               CPU_BASED_INVLPG_EXITING;
1154         opt = CPU_BASED_TPR_SHADOW |
1155               CPU_BASED_USE_MSR_BITMAPS |
1156               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1157         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1158                                 &_cpu_based_exec_control) < 0)
1159                 return -EIO;
1160 #ifdef CONFIG_X86_64
1161         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1162                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1163                                            ~CPU_BASED_CR8_STORE_EXITING;
1164 #endif
1165         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1166                 min2 = 0;
1167                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1168                         SECONDARY_EXEC_WBINVD_EXITING |
1169                         SECONDARY_EXEC_ENABLE_VPID |
1170                         SECONDARY_EXEC_ENABLE_EPT;
1171                 if (adjust_vmx_controls(min2, opt2,
1172                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1173                                         &_cpu_based_2nd_exec_control) < 0)
1174                         return -EIO;
1175         }
1176 #ifndef CONFIG_X86_64
1177         if (!(_cpu_based_2nd_exec_control &
1178                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1179                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1180 #endif
1181         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1182                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1183                    enabled */
1184                 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1185                          CPU_BASED_CR3_STORE_EXITING |
1186                          CPU_BASED_INVLPG_EXITING);
1187                 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1188                                         &_cpu_based_exec_control) < 0)
1189                         return -EIO;
1190                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1191                       vmx_capability.ept, vmx_capability.vpid);
1192         }
1193
1194         min = 0;
1195 #ifdef CONFIG_X86_64
1196         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1197 #endif
1198         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1199         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1200                                 &_vmexit_control) < 0)
1201                 return -EIO;
1202
1203         min = 0;
1204         opt = VM_ENTRY_LOAD_IA32_PAT;
1205         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1206                                 &_vmentry_control) < 0)
1207                 return -EIO;
1208
1209         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1210
1211         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1212         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1213                 return -EIO;
1214
1215 #ifdef CONFIG_X86_64
1216         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1217         if (vmx_msr_high & (1u<<16))
1218                 return -EIO;
1219 #endif
1220
1221         /* Require Write-Back (WB) memory type for VMCS accesses. */
1222         if (((vmx_msr_high >> 18) & 15) != 6)
1223                 return -EIO;
1224
1225         vmcs_conf->size = vmx_msr_high & 0x1fff;
1226         vmcs_conf->order = get_order(vmcs_config.size);
1227         vmcs_conf->revision_id = vmx_msr_low;
1228
1229         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1230         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1231         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1232         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1233         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1234
1235         return 0;
1236 }
1237
1238 static struct vmcs *alloc_vmcs_cpu(int cpu)
1239 {
1240         int node = cpu_to_node(cpu);
1241         struct page *pages;
1242         struct vmcs *vmcs;
1243
1244         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1245         if (!pages)
1246                 return NULL;
1247         vmcs = page_address(pages);
1248         memset(vmcs, 0, vmcs_config.size);
1249         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1250         return vmcs;
1251 }
1252
1253 static struct vmcs *alloc_vmcs(void)
1254 {
1255         return alloc_vmcs_cpu(raw_smp_processor_id());
1256 }
1257
1258 static void free_vmcs(struct vmcs *vmcs)
1259 {
1260         free_pages((unsigned long)vmcs, vmcs_config.order);
1261 }
1262
1263 static void free_kvm_area(void)
1264 {
1265         int cpu;
1266
1267         for_each_online_cpu(cpu)
1268                 free_vmcs(per_cpu(vmxarea, cpu));
1269 }
1270
1271 static __init int alloc_kvm_area(void)
1272 {
1273         int cpu;
1274
1275         for_each_online_cpu(cpu) {
1276                 struct vmcs *vmcs;
1277
1278                 vmcs = alloc_vmcs_cpu(cpu);
1279                 if (!vmcs) {
1280                         free_kvm_area();
1281                         return -ENOMEM;
1282                 }
1283
1284                 per_cpu(vmxarea, cpu) = vmcs;
1285         }
1286         return 0;
1287 }
1288
1289 static __init int hardware_setup(void)
1290 {
1291         if (setup_vmcs_config(&vmcs_config) < 0)
1292                 return -EIO;
1293
1294         if (boot_cpu_has(X86_FEATURE_NX))
1295                 kvm_enable_efer_bits(EFER_NX);
1296
1297         return alloc_kvm_area();
1298 }
1299
1300 static __exit void hardware_unsetup(void)
1301 {
1302         free_kvm_area();
1303 }
1304
1305 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1306 {
1307         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1308
1309         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1310                 vmcs_write16(sf->selector, save->selector);
1311                 vmcs_writel(sf->base, save->base);
1312                 vmcs_write32(sf->limit, save->limit);
1313                 vmcs_write32(sf->ar_bytes, save->ar);
1314         } else {
1315                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1316                         << AR_DPL_SHIFT;
1317                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1318         }
1319 }
1320
1321 static void enter_pmode(struct kvm_vcpu *vcpu)
1322 {
1323         unsigned long flags;
1324         struct vcpu_vmx *vmx = to_vmx(vcpu);
1325
1326         vmx->emulation_required = 1;
1327         vcpu->arch.rmode.active = 0;
1328
1329         vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1330         vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1331         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1332
1333         flags = vmcs_readl(GUEST_RFLAGS);
1334         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1335         flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1336         vmcs_writel(GUEST_RFLAGS, flags);
1337
1338         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1339                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1340
1341         update_exception_bitmap(vcpu);
1342
1343         if (emulate_invalid_guest_state)
1344                 return;
1345
1346         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1347         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1348         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1349         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1350
1351         vmcs_write16(GUEST_SS_SELECTOR, 0);
1352         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1353
1354         vmcs_write16(GUEST_CS_SELECTOR,
1355                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1356         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1357 }
1358
1359 static gva_t rmode_tss_base(struct kvm *kvm)
1360 {
1361         if (!kvm->arch.tss_addr) {
1362                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1363                                  kvm->memslots[0].npages - 3;
1364                 return base_gfn << PAGE_SHIFT;
1365         }
1366         return kvm->arch.tss_addr;
1367 }
1368
1369 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1370 {
1371         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1372
1373         save->selector = vmcs_read16(sf->selector);
1374         save->base = vmcs_readl(sf->base);
1375         save->limit = vmcs_read32(sf->limit);
1376         save->ar = vmcs_read32(sf->ar_bytes);
1377         vmcs_write16(sf->selector, save->base >> 4);
1378         vmcs_write32(sf->base, save->base & 0xfffff);
1379         vmcs_write32(sf->limit, 0xffff);
1380         vmcs_write32(sf->ar_bytes, 0xf3);
1381 }
1382
1383 static void enter_rmode(struct kvm_vcpu *vcpu)
1384 {
1385         unsigned long flags;
1386         struct vcpu_vmx *vmx = to_vmx(vcpu);
1387
1388         vmx->emulation_required = 1;
1389         vcpu->arch.rmode.active = 1;
1390
1391         vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1392         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1393
1394         vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1395         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1396
1397         vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1398         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1399
1400         flags = vmcs_readl(GUEST_RFLAGS);
1401         vcpu->arch.rmode.save_iopl
1402                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1403
1404         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1405
1406         vmcs_writel(GUEST_RFLAGS, flags);
1407         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1408         update_exception_bitmap(vcpu);
1409
1410         if (emulate_invalid_guest_state)
1411                 goto continue_rmode;
1412
1413         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1414         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1415         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1416
1417         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1418         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1419         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1420                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1421         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1422
1423         fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1424         fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1425         fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1426         fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1427
1428 continue_rmode:
1429         kvm_mmu_reset_context(vcpu);
1430         init_rmode(vcpu->kvm);
1431 }
1432
1433 #ifdef CONFIG_X86_64
1434
1435 static void enter_lmode(struct kvm_vcpu *vcpu)
1436 {
1437         u32 guest_tr_ar;
1438
1439         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1440         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1441                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1442                        __func__);
1443                 vmcs_write32(GUEST_TR_AR_BYTES,
1444                              (guest_tr_ar & ~AR_TYPE_MASK)
1445                              | AR_TYPE_BUSY_64_TSS);
1446         }
1447
1448         vcpu->arch.shadow_efer |= EFER_LMA;
1449
1450         find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1451         vmcs_write32(VM_ENTRY_CONTROLS,
1452                      vmcs_read32(VM_ENTRY_CONTROLS)
1453                      | VM_ENTRY_IA32E_MODE);
1454 }
1455
1456 static void exit_lmode(struct kvm_vcpu *vcpu)
1457 {
1458         vcpu->arch.shadow_efer &= ~EFER_LMA;
1459
1460         vmcs_write32(VM_ENTRY_CONTROLS,
1461                      vmcs_read32(VM_ENTRY_CONTROLS)
1462                      & ~VM_ENTRY_IA32E_MODE);
1463 }
1464
1465 #endif
1466
1467 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1468 {
1469         vpid_sync_vcpu_all(to_vmx(vcpu));
1470         if (vm_need_ept())
1471                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1472 }
1473
1474 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1475 {
1476         vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1477         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1478 }
1479
1480 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1481 {
1482         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1483                 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1484                         printk(KERN_ERR "EPT: Fail to load pdptrs!\n");
1485                         return;
1486                 }
1487                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1488                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1489                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1490                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1491         }
1492 }
1493
1494 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1495
1496 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1497                                         unsigned long cr0,
1498                                         struct kvm_vcpu *vcpu)
1499 {
1500         if (!(cr0 & X86_CR0_PG)) {
1501                 /* From paging/starting to nonpaging */
1502                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1503                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1504                              (CPU_BASED_CR3_LOAD_EXITING |
1505                               CPU_BASED_CR3_STORE_EXITING));
1506                 vcpu->arch.cr0 = cr0;
1507                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1508                 *hw_cr0 |= X86_CR0_PE | X86_CR0_PG;
1509                 *hw_cr0 &= ~X86_CR0_WP;
1510         } else if (!is_paging(vcpu)) {
1511                 /* From nonpaging to paging */
1512                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1513                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1514                              ~(CPU_BASED_CR3_LOAD_EXITING |
1515                                CPU_BASED_CR3_STORE_EXITING));
1516                 vcpu->arch.cr0 = cr0;
1517                 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1518                 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1519                         *hw_cr0 &= ~X86_CR0_WP;
1520         }
1521 }
1522
1523 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1524                                         struct kvm_vcpu *vcpu)
1525 {
1526         if (!is_paging(vcpu)) {
1527                 *hw_cr4 &= ~X86_CR4_PAE;
1528                 *hw_cr4 |= X86_CR4_PSE;
1529         } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1530                 *hw_cr4 &= ~X86_CR4_PAE;
1531 }
1532
1533 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1534 {
1535         unsigned long hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) |
1536                                 KVM_VM_CR0_ALWAYS_ON;
1537
1538         vmx_fpu_deactivate(vcpu);
1539
1540         if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1541                 enter_pmode(vcpu);
1542
1543         if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1544                 enter_rmode(vcpu);
1545
1546 #ifdef CONFIG_X86_64
1547         if (vcpu->arch.shadow_efer & EFER_LME) {
1548                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1549                         enter_lmode(vcpu);
1550                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1551                         exit_lmode(vcpu);
1552         }
1553 #endif
1554
1555         if (vm_need_ept())
1556                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1557
1558         vmcs_writel(CR0_READ_SHADOW, cr0);
1559         vmcs_writel(GUEST_CR0, hw_cr0);
1560         vcpu->arch.cr0 = cr0;
1561
1562         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1563                 vmx_fpu_activate(vcpu);
1564 }
1565
1566 static u64 construct_eptp(unsigned long root_hpa)
1567 {
1568         u64 eptp;
1569
1570         /* TODO write the value reading from MSR */
1571         eptp = VMX_EPT_DEFAULT_MT |
1572                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1573         eptp |= (root_hpa & PAGE_MASK);
1574
1575         return eptp;
1576 }
1577
1578 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1579 {
1580         unsigned long guest_cr3;
1581         u64 eptp;
1582
1583         guest_cr3 = cr3;
1584         if (vm_need_ept()) {
1585                 eptp = construct_eptp(cr3);
1586                 vmcs_write64(EPT_POINTER, eptp);
1587                 ept_sync_context(eptp);
1588                 ept_load_pdptrs(vcpu);
1589                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1590                         VMX_EPT_IDENTITY_PAGETABLE_ADDR;
1591         }
1592
1593         vmx_flush_tlb(vcpu);
1594         vmcs_writel(GUEST_CR3, guest_cr3);
1595         if (vcpu->arch.cr0 & X86_CR0_PE)
1596                 vmx_fpu_deactivate(vcpu);
1597 }
1598
1599 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1600 {
1601         unsigned long hw_cr4 = cr4 | (vcpu->arch.rmode.active ?
1602                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1603
1604         vcpu->arch.cr4 = cr4;
1605         if (vm_need_ept())
1606                 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1607
1608         vmcs_writel(CR4_READ_SHADOW, cr4);
1609         vmcs_writel(GUEST_CR4, hw_cr4);
1610 }
1611
1612 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1613 {
1614         struct vcpu_vmx *vmx = to_vmx(vcpu);
1615         struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1616
1617         vcpu->arch.shadow_efer = efer;
1618         if (!msr)
1619                 return;
1620         if (efer & EFER_LMA) {
1621                 vmcs_write32(VM_ENTRY_CONTROLS,
1622                                      vmcs_read32(VM_ENTRY_CONTROLS) |
1623                                      VM_ENTRY_IA32E_MODE);
1624                 msr->data = efer;
1625
1626         } else {
1627                 vmcs_write32(VM_ENTRY_CONTROLS,
1628                                      vmcs_read32(VM_ENTRY_CONTROLS) &
1629                                      ~VM_ENTRY_IA32E_MODE);
1630
1631                 msr->data = efer & ~EFER_LME;
1632         }
1633         setup_msrs(vmx);
1634 }
1635
1636 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1637 {
1638         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1639
1640         return vmcs_readl(sf->base);
1641 }
1642
1643 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1644                             struct kvm_segment *var, int seg)
1645 {
1646         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1647         u32 ar;
1648
1649         var->base = vmcs_readl(sf->base);
1650         var->limit = vmcs_read32(sf->limit);
1651         var->selector = vmcs_read16(sf->selector);
1652         ar = vmcs_read32(sf->ar_bytes);
1653         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1654                 ar = 0;
1655         var->type = ar & 15;
1656         var->s = (ar >> 4) & 1;
1657         var->dpl = (ar >> 5) & 3;
1658         var->present = (ar >> 7) & 1;
1659         var->avl = (ar >> 12) & 1;
1660         var->l = (ar >> 13) & 1;
1661         var->db = (ar >> 14) & 1;
1662         var->g = (ar >> 15) & 1;
1663         var->unusable = (ar >> 16) & 1;
1664 }
1665
1666 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1667 {
1668         struct kvm_segment kvm_seg;
1669
1670         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1671                 return 0;
1672
1673         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1674                 return 3;
1675
1676         vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1677         return kvm_seg.selector & 3;
1678 }
1679
1680 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1681 {
1682         u32 ar;
1683
1684         if (var->unusable)
1685                 ar = 1 << 16;
1686         else {
1687                 ar = var->type & 15;
1688                 ar |= (var->s & 1) << 4;
1689                 ar |= (var->dpl & 3) << 5;
1690                 ar |= (var->present & 1) << 7;
1691                 ar |= (var->avl & 1) << 12;
1692                 ar |= (var->l & 1) << 13;
1693                 ar |= (var->db & 1) << 14;
1694                 ar |= (var->g & 1) << 15;
1695         }
1696         if (ar == 0) /* a 0 value means unusable */
1697                 ar = AR_UNUSABLE_MASK;
1698
1699         return ar;
1700 }
1701
1702 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1703                             struct kvm_segment *var, int seg)
1704 {
1705         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1706         u32 ar;
1707
1708         if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1709                 vcpu->arch.rmode.tr.selector = var->selector;
1710                 vcpu->arch.rmode.tr.base = var->base;
1711                 vcpu->arch.rmode.tr.limit = var->limit;
1712                 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1713                 return;
1714         }
1715         vmcs_writel(sf->base, var->base);
1716         vmcs_write32(sf->limit, var->limit);
1717         vmcs_write16(sf->selector, var->selector);
1718         if (vcpu->arch.rmode.active && var->s) {
1719                 /*
1720                  * Hack real-mode segments into vm86 compatibility.
1721                  */
1722                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1723                         vmcs_writel(sf->base, 0xf0000);
1724                 ar = 0xf3;
1725         } else
1726                 ar = vmx_segment_access_rights(var);
1727         vmcs_write32(sf->ar_bytes, ar);
1728 }
1729
1730 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1731 {
1732         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1733
1734         *db = (ar >> 14) & 1;
1735         *l = (ar >> 13) & 1;
1736 }
1737
1738 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1739 {
1740         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1741         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1742 }
1743
1744 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1745 {
1746         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1747         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1748 }
1749
1750 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1751 {
1752         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1753         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1754 }
1755
1756 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1757 {
1758         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1759         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1760 }
1761
1762 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1763 {
1764         struct kvm_segment var;
1765         u32 ar;
1766
1767         vmx_get_segment(vcpu, &var, seg);
1768         ar = vmx_segment_access_rights(&var);
1769
1770         if (var.base != (var.selector << 4))
1771                 return false;
1772         if (var.limit != 0xffff)
1773                 return false;
1774         if (ar != 0xf3)
1775                 return false;
1776
1777         return true;
1778 }
1779
1780 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1781 {
1782         struct kvm_segment cs;
1783         unsigned int cs_rpl;
1784
1785         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1786         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1787
1788         if (cs.unusable)
1789                 return false;
1790         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1791                 return false;
1792         if (!cs.s)
1793                 return false;
1794         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1795                 if (cs.dpl > cs_rpl)
1796                         return false;
1797         } else {
1798                 if (cs.dpl != cs_rpl)
1799                         return false;
1800         }
1801         if (!cs.present)
1802                 return false;
1803
1804         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1805         return true;
1806 }
1807
1808 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1809 {
1810         struct kvm_segment ss;
1811         unsigned int ss_rpl;
1812
1813         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1814         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1815
1816         if (ss.unusable)
1817                 return true;
1818         if (ss.type != 3 && ss.type != 7)
1819                 return false;
1820         if (!ss.s)
1821                 return false;
1822         if (ss.dpl != ss_rpl) /* DPL != RPL */
1823                 return false;
1824         if (!ss.present)
1825                 return false;
1826
1827         return true;
1828 }
1829
1830 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1831 {
1832         struct kvm_segment var;
1833         unsigned int rpl;
1834
1835         vmx_get_segment(vcpu, &var, seg);
1836         rpl = var.selector & SELECTOR_RPL_MASK;
1837
1838         if (var.unusable)
1839                 return true;
1840         if (!var.s)
1841                 return false;
1842         if (!var.present)
1843                 return false;
1844         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1845                 if (var.dpl < rpl) /* DPL < RPL */
1846                         return false;
1847         }
1848
1849         /* TODO: Add other members to kvm_segment_field to allow checking for other access
1850          * rights flags
1851          */
1852         return true;
1853 }
1854
1855 static bool tr_valid(struct kvm_vcpu *vcpu)
1856 {
1857         struct kvm_segment tr;
1858
1859         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1860
1861         if (tr.unusable)
1862                 return false;
1863         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
1864                 return false;
1865         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
1866                 return false;
1867         if (!tr.present)
1868                 return false;
1869
1870         return true;
1871 }
1872
1873 static bool ldtr_valid(struct kvm_vcpu *vcpu)
1874 {
1875         struct kvm_segment ldtr;
1876
1877         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1878
1879         if (ldtr.unusable)
1880                 return true;
1881         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
1882                 return false;
1883         if (ldtr.type != 2)
1884                 return false;
1885         if (!ldtr.present)
1886                 return false;
1887
1888         return true;
1889 }
1890
1891 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1892 {
1893         struct kvm_segment cs, ss;
1894
1895         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1896         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1897
1898         return ((cs.selector & SELECTOR_RPL_MASK) ==
1899                  (ss.selector & SELECTOR_RPL_MASK));
1900 }
1901
1902 /*
1903  * Check if guest state is valid. Returns true if valid, false if
1904  * not.
1905  * We assume that registers are always usable
1906  */
1907 static bool guest_state_valid(struct kvm_vcpu *vcpu)
1908 {
1909         /* real mode guest state checks */
1910         if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1911                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1912                         return false;
1913                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1914                         return false;
1915                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1916                         return false;
1917                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1918                         return false;
1919                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1920                         return false;
1921                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1922                         return false;
1923         } else {
1924         /* protected mode guest state checks */
1925                 if (!cs_ss_rpl_check(vcpu))
1926                         return false;
1927                 if (!code_segment_valid(vcpu))
1928                         return false;
1929                 if (!stack_segment_valid(vcpu))
1930                         return false;
1931                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1932                         return false;
1933                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1934                         return false;
1935                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1936                         return false;
1937                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1938                         return false;
1939                 if (!tr_valid(vcpu))
1940                         return false;
1941                 if (!ldtr_valid(vcpu))
1942                         return false;
1943         }
1944         /* TODO:
1945          * - Add checks on RIP
1946          * - Add checks on RFLAGS
1947          */
1948
1949         return true;
1950 }
1951
1952 static int init_rmode_tss(struct kvm *kvm)
1953 {
1954         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1955         u16 data = 0;
1956         int ret = 0;
1957         int r;
1958
1959         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1960         if (r < 0)
1961                 goto out;
1962         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1963         r = kvm_write_guest_page(kvm, fn++, &data,
1964                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
1965         if (r < 0)
1966                 goto out;
1967         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1968         if (r < 0)
1969                 goto out;
1970         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1971         if (r < 0)
1972                 goto out;
1973         data = ~0;
1974         r = kvm_write_guest_page(kvm, fn, &data,
1975                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1976                                  sizeof(u8));
1977         if (r < 0)
1978                 goto out;
1979
1980         ret = 1;
1981 out:
1982         return ret;
1983 }
1984
1985 static int init_rmode_identity_map(struct kvm *kvm)
1986 {
1987         int i, r, ret;
1988         pfn_t identity_map_pfn;
1989         u32 tmp;
1990
1991         if (!vm_need_ept())
1992                 return 1;
1993         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
1994                 printk(KERN_ERR "EPT: identity-mapping pagetable "
1995                         "haven't been allocated!\n");
1996                 return 0;
1997         }
1998         if (likely(kvm->arch.ept_identity_pagetable_done))
1999                 return 1;
2000         ret = 0;
2001         identity_map_pfn = VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT;
2002         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2003         if (r < 0)
2004                 goto out;
2005         /* Set up identity-mapping pagetable for EPT in real mode */
2006         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2007                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2008                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2009                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2010                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2011                 if (r < 0)
2012                         goto out;
2013         }
2014         kvm->arch.ept_identity_pagetable_done = true;
2015         ret = 1;
2016 out:
2017         return ret;
2018 }
2019
2020 static void seg_setup(int seg)
2021 {
2022         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2023
2024         vmcs_write16(sf->selector, 0);
2025         vmcs_writel(sf->base, 0);
2026         vmcs_write32(sf->limit, 0xffff);
2027         vmcs_write32(sf->ar_bytes, 0xf3);
2028 }
2029
2030 static int alloc_apic_access_page(struct kvm *kvm)
2031 {
2032         struct kvm_userspace_memory_region kvm_userspace_mem;
2033         int r = 0;
2034
2035         down_write(&kvm->slots_lock);
2036         if (kvm->arch.apic_access_page)
2037                 goto out;
2038         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2039         kvm_userspace_mem.flags = 0;
2040         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2041         kvm_userspace_mem.memory_size = PAGE_SIZE;
2042         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2043         if (r)
2044                 goto out;
2045
2046         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2047 out:
2048         up_write(&kvm->slots_lock);
2049         return r;
2050 }
2051
2052 static int alloc_identity_pagetable(struct kvm *kvm)
2053 {
2054         struct kvm_userspace_memory_region kvm_userspace_mem;
2055         int r = 0;
2056
2057         down_write(&kvm->slots_lock);
2058         if (kvm->arch.ept_identity_pagetable)
2059                 goto out;
2060         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2061         kvm_userspace_mem.flags = 0;
2062         kvm_userspace_mem.guest_phys_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
2063         kvm_userspace_mem.memory_size = PAGE_SIZE;
2064         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2065         if (r)
2066                 goto out;
2067
2068         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2069                         VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
2070 out:
2071         up_write(&kvm->slots_lock);
2072         return r;
2073 }
2074
2075 static void allocate_vpid(struct vcpu_vmx *vmx)
2076 {
2077         int vpid;
2078
2079         vmx->vpid = 0;
2080         if (!enable_vpid || !cpu_has_vmx_vpid())
2081                 return;
2082         spin_lock(&vmx_vpid_lock);
2083         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2084         if (vpid < VMX_NR_VPIDS) {
2085                 vmx->vpid = vpid;
2086                 __set_bit(vpid, vmx_vpid_bitmap);
2087         }
2088         spin_unlock(&vmx_vpid_lock);
2089 }
2090
2091 static void vmx_disable_intercept_for_msr(struct page *msr_bitmap, u32 msr)
2092 {
2093         void *va;
2094
2095         if (!cpu_has_vmx_msr_bitmap())
2096                 return;
2097
2098         /*
2099          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2100          * have the write-low and read-high bitmap offsets the wrong way round.
2101          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2102          */
2103         va = kmap(msr_bitmap);
2104         if (msr <= 0x1fff) {
2105                 __clear_bit(msr, va + 0x000); /* read-low */
2106                 __clear_bit(msr, va + 0x800); /* write-low */
2107         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2108                 msr &= 0x1fff;
2109                 __clear_bit(msr, va + 0x400); /* read-high */
2110                 __clear_bit(msr, va + 0xc00); /* write-high */
2111         }
2112         kunmap(msr_bitmap);
2113 }
2114
2115 /*
2116  * Sets up the vmcs for emulated real mode.
2117  */
2118 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2119 {
2120         u32 host_sysenter_cs, msr_low, msr_high;
2121         u32 junk;
2122         u64 host_pat, tsc_this, tsc_base;
2123         unsigned long a;
2124         struct descriptor_table dt;
2125         int i;
2126         unsigned long kvm_vmx_return;
2127         u32 exec_control;
2128
2129         /* I/O */
2130         vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
2131         vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
2132
2133         if (cpu_has_vmx_msr_bitmap())
2134                 vmcs_write64(MSR_BITMAP, page_to_phys(vmx_msr_bitmap));
2135
2136         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2137
2138         /* Control */
2139         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2140                 vmcs_config.pin_based_exec_ctrl);
2141
2142         exec_control = vmcs_config.cpu_based_exec_ctrl;
2143         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2144                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2145 #ifdef CONFIG_X86_64
2146                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2147                                 CPU_BASED_CR8_LOAD_EXITING;
2148 #endif
2149         }
2150         if (!vm_need_ept())
2151                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2152                                 CPU_BASED_CR3_LOAD_EXITING  |
2153                                 CPU_BASED_INVLPG_EXITING;
2154         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2155
2156         if (cpu_has_secondary_exec_ctrls()) {
2157                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2158                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2159                         exec_control &=
2160                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2161                 if (vmx->vpid == 0)
2162                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2163                 if (!vm_need_ept())
2164                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2165                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2166         }
2167
2168         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2169         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2170         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2171
2172         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
2173         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2174         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2175
2176         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2177         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2178         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2179         vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs());    /* 22.2.4 */
2180         vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs());    /* 22.2.4 */
2181         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2182 #ifdef CONFIG_X86_64
2183         rdmsrl(MSR_FS_BASE, a);
2184         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2185         rdmsrl(MSR_GS_BASE, a);
2186         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2187 #else
2188         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2189         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2190 #endif
2191
2192         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2193
2194         kvm_get_idt(&dt);
2195         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
2196
2197         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2198         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2199         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2200         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2201         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2202
2203         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2204         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2205         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2206         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2207         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2208         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2209
2210         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2211                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2212                 host_pat = msr_low | ((u64) msr_high << 32);
2213                 vmcs_write64(HOST_IA32_PAT, host_pat);
2214         }
2215         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2216                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2217                 host_pat = msr_low | ((u64) msr_high << 32);
2218                 /* Write the default value follow host pat */
2219                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2220                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2221                 vmx->vcpu.arch.pat = host_pat;
2222         }
2223
2224         for (i = 0; i < NR_VMX_MSR; ++i) {
2225                 u32 index = vmx_msr_index[i];
2226                 u32 data_low, data_high;
2227                 u64 data;
2228                 int j = vmx->nmsrs;
2229
2230                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2231                         continue;
2232                 if (wrmsr_safe(index, data_low, data_high) < 0)
2233                         continue;
2234                 data = data_low | ((u64)data_high << 32);
2235                 vmx->host_msrs[j].index = index;
2236                 vmx->host_msrs[j].reserved = 0;
2237                 vmx->host_msrs[j].data = data;
2238                 vmx->guest_msrs[j] = vmx->host_msrs[j];
2239                 ++vmx->nmsrs;
2240         }
2241
2242         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2243
2244         /* 22.2.1, 20.8.1 */
2245         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2246
2247         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2248         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2249
2250         tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2251         rdtscll(tsc_this);
2252         if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2253                 tsc_base = tsc_this;
2254
2255         guest_write_tsc(0, tsc_base);
2256
2257         return 0;
2258 }
2259
2260 static int init_rmode(struct kvm *kvm)
2261 {
2262         if (!init_rmode_tss(kvm))
2263                 return 0;
2264         if (!init_rmode_identity_map(kvm))
2265                 return 0;
2266         return 1;
2267 }
2268
2269 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2270 {
2271         struct vcpu_vmx *vmx = to_vmx(vcpu);
2272         u64 msr;
2273         int ret;
2274
2275         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2276         down_read(&vcpu->kvm->slots_lock);
2277         if (!init_rmode(vmx->vcpu.kvm)) {
2278                 ret = -ENOMEM;
2279                 goto out;
2280         }
2281
2282         vmx->vcpu.arch.rmode.active = 0;
2283
2284         vmx->soft_vnmi_blocked = 0;
2285
2286         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2287         kvm_set_cr8(&vmx->vcpu, 0);
2288         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2289         if (vmx->vcpu.vcpu_id == 0)
2290                 msr |= MSR_IA32_APICBASE_BSP;
2291         kvm_set_apic_base(&vmx->vcpu, msr);
2292
2293         fx_init(&vmx->vcpu);
2294
2295         seg_setup(VCPU_SREG_CS);
2296         /*
2297          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2298          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2299          */
2300         if (vmx->vcpu.vcpu_id == 0) {
2301                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2302                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2303         } else {
2304                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2305                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2306         }
2307
2308         seg_setup(VCPU_SREG_DS);
2309         seg_setup(VCPU_SREG_ES);
2310         seg_setup(VCPU_SREG_FS);
2311         seg_setup(VCPU_SREG_GS);
2312         seg_setup(VCPU_SREG_SS);
2313
2314         vmcs_write16(GUEST_TR_SELECTOR, 0);
2315         vmcs_writel(GUEST_TR_BASE, 0);
2316         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2317         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2318
2319         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2320         vmcs_writel(GUEST_LDTR_BASE, 0);
2321         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2322         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2323
2324         vmcs_write32(GUEST_SYSENTER_CS, 0);
2325         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2326         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2327
2328         vmcs_writel(GUEST_RFLAGS, 0x02);
2329         if (vmx->vcpu.vcpu_id == 0)
2330                 kvm_rip_write(vcpu, 0xfff0);
2331         else
2332                 kvm_rip_write(vcpu, 0);
2333         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2334
2335         vmcs_writel(GUEST_DR7, 0x400);
2336
2337         vmcs_writel(GUEST_GDTR_BASE, 0);
2338         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2339
2340         vmcs_writel(GUEST_IDTR_BASE, 0);
2341         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2342
2343         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2344         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2345         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2346
2347         /* Special registers */
2348         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2349
2350         setup_msrs(vmx);
2351
2352         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2353
2354         if (cpu_has_vmx_tpr_shadow()) {
2355                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2356                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2357                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2358                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2359                 vmcs_write32(TPR_THRESHOLD, 0);
2360         }
2361
2362         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2363                 vmcs_write64(APIC_ACCESS_ADDR,
2364                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2365
2366         if (vmx->vpid != 0)
2367                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2368
2369         vmx->vcpu.arch.cr0 = 0x60000010;
2370         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2371         vmx_set_cr4(&vmx->vcpu, 0);
2372         vmx_set_efer(&vmx->vcpu, 0);
2373         vmx_fpu_activate(&vmx->vcpu);
2374         update_exception_bitmap(&vmx->vcpu);
2375
2376         vpid_sync_vcpu_all(vmx);
2377
2378         ret = 0;
2379
2380         /* HACK: Don't enable emulation on guest boot/reset */
2381         vmx->emulation_required = 0;
2382
2383 out:
2384         up_read(&vcpu->kvm->slots_lock);
2385         return ret;
2386 }
2387
2388 static void enable_irq_window(struct kvm_vcpu *vcpu)
2389 {
2390         u32 cpu_based_vm_exec_control;
2391
2392         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2393         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2394         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2395 }
2396
2397 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2398 {
2399         u32 cpu_based_vm_exec_control;
2400
2401         if (!cpu_has_virtual_nmis()) {
2402                 enable_irq_window(vcpu);
2403                 return;
2404         }
2405
2406         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2407         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2408         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2409 }
2410
2411 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2412 {
2413         struct vcpu_vmx *vmx = to_vmx(vcpu);
2414
2415         KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2416
2417         ++vcpu->stat.irq_injections;
2418         if (vcpu->arch.rmode.active) {
2419                 vmx->rmode.irq.pending = true;
2420                 vmx->rmode.irq.vector = irq;
2421                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2422                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2423                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2424                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2425                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2426                 return;
2427         }
2428         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2429                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
2430 }
2431
2432 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2433 {
2434         struct vcpu_vmx *vmx = to_vmx(vcpu);
2435
2436         if (!cpu_has_virtual_nmis()) {
2437                 /*
2438                  * Tracking the NMI-blocked state in software is built upon
2439                  * finding the next open IRQ window. This, in turn, depends on
2440                  * well-behaving guests: They have to keep IRQs disabled at
2441                  * least as long as the NMI handler runs. Otherwise we may
2442                  * cause NMI nesting, maybe breaking the guest. But as this is
2443                  * highly unlikely, we can live with the residual risk.
2444                  */
2445                 vmx->soft_vnmi_blocked = 1;
2446                 vmx->vnmi_blocked_time = 0;
2447         }
2448
2449         ++vcpu->stat.nmi_injections;
2450         if (vcpu->arch.rmode.active) {
2451                 vmx->rmode.irq.pending = true;
2452                 vmx->rmode.irq.vector = NMI_VECTOR;
2453                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2454                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2455                              NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2456                              INTR_INFO_VALID_MASK);
2457                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2458                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2459                 return;
2460         }
2461         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2462                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2463 }
2464
2465 static void vmx_update_window_states(struct kvm_vcpu *vcpu)
2466 {
2467         u32 guest_intr = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2468
2469         vcpu->arch.nmi_window_open =
2470                 !(guest_intr & (GUEST_INTR_STATE_STI |
2471                                 GUEST_INTR_STATE_MOV_SS |
2472                                 GUEST_INTR_STATE_NMI));
2473         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2474                 vcpu->arch.nmi_window_open = 0;
2475
2476         vcpu->arch.interrupt_window_open =
2477                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2478                  !(guest_intr & (GUEST_INTR_STATE_STI |
2479                                  GUEST_INTR_STATE_MOV_SS)));
2480 }
2481
2482 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2483 {
2484         int word_index = __ffs(vcpu->arch.irq_summary);
2485         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2486         int irq = word_index * BITS_PER_LONG + bit_index;
2487
2488         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2489         if (!vcpu->arch.irq_pending[word_index])
2490                 clear_bit(word_index, &vcpu->arch.irq_summary);
2491         kvm_queue_interrupt(vcpu, irq);
2492 }
2493
2494 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2495                                        struct kvm_run *kvm_run)
2496 {
2497         vmx_update_window_states(vcpu);
2498
2499         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
2500                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2501                                 GUEST_INTR_STATE_STI |
2502                                 GUEST_INTR_STATE_MOV_SS);
2503
2504         if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
2505                 if (vcpu->arch.interrupt.pending) {
2506                         enable_nmi_window(vcpu);
2507                 } else if (vcpu->arch.nmi_window_open) {
2508                         vcpu->arch.nmi_pending = false;
2509                         vcpu->arch.nmi_injected = true;
2510                 } else {
2511                         enable_nmi_window(vcpu);
2512                         return;
2513                 }
2514         }
2515         if (vcpu->arch.nmi_injected) {
2516                 vmx_inject_nmi(vcpu);
2517                 if (vcpu->arch.nmi_pending)
2518                         enable_nmi_window(vcpu);
2519                 else if (vcpu->arch.irq_summary
2520                          || kvm_run->request_interrupt_window)
2521                         enable_irq_window(vcpu);
2522                 return;
2523         }
2524
2525         if (vcpu->arch.interrupt_window_open) {
2526                 if (vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2527                         kvm_do_inject_irq(vcpu);
2528
2529                 if (vcpu->arch.interrupt.pending)
2530                         vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2531         }
2532         if (!vcpu->arch.interrupt_window_open &&
2533             (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
2534                 enable_irq_window(vcpu);
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 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2595 {
2596         struct vcpu_vmx *vmx = to_vmx(vcpu);
2597         u32 intr_info, ex_no, error_code;
2598         unsigned long cr2, rip, dr6;
2599         u32 vect_info;
2600         enum emulation_result er;
2601
2602         vect_info = vmx->idt_vectoring_info;
2603         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2604
2605         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2606                                                 !is_page_fault(intr_info))
2607                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2608                        "intr info 0x%x\n", __func__, vect_info, intr_info);
2609
2610         if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
2611                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
2612                 set_bit(irq, vcpu->arch.irq_pending);
2613                 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
2614         }
2615
2616         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2617                 return 1;  /* already handled by vmx_vcpu_run() */
2618
2619         if (is_no_device(intr_info)) {
2620                 vmx_fpu_activate(vcpu);
2621                 return 1;
2622         }
2623
2624         if (is_invalid_opcode(intr_info)) {
2625                 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2626                 if (er != EMULATE_DONE)
2627                         kvm_queue_exception(vcpu, UD_VECTOR);
2628                 return 1;
2629         }
2630
2631         error_code = 0;
2632         rip = kvm_rip_read(vcpu);
2633         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2634                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2635         if (is_page_fault(intr_info)) {
2636                 /* EPT won't cause page fault directly */
2637                 if (vm_need_ept())
2638                         BUG();
2639                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2640                 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2641                             (u32)((u64)cr2 >> 32), handler);
2642                 if (vcpu->arch.interrupt.pending || vcpu->arch.exception.pending)
2643                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
2644                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2645         }
2646
2647         if (vcpu->arch.rmode.active &&
2648             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2649                                                                 error_code)) {
2650                 if (vcpu->arch.halt_request) {
2651                         vcpu->arch.halt_request = 0;
2652                         return kvm_emulate_halt(vcpu);
2653                 }
2654                 return 1;
2655         }
2656
2657         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2658         switch (ex_no) {
2659         case DB_VECTOR:
2660                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2661                 if (!(vcpu->guest_debug &
2662                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2663                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2664                         kvm_queue_exception(vcpu, DB_VECTOR);
2665                         return 1;
2666                 }
2667                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2668                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2669                 /* fall through */
2670         case BP_VECTOR:
2671                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2672                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2673                 kvm_run->debug.arch.exception = ex_no;
2674                 break;
2675         default:
2676                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2677                 kvm_run->ex.exception = ex_no;
2678                 kvm_run->ex.error_code = error_code;
2679                 break;
2680         }
2681         return 0;
2682 }
2683
2684 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2685                                      struct kvm_run *kvm_run)
2686 {
2687         ++vcpu->stat.irq_exits;
2688         KVMTRACE_1D(INTR, vcpu, vmcs_read32(VM_EXIT_INTR_INFO), handler);
2689         return 1;
2690 }
2691
2692 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2693 {
2694         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2695         return 0;
2696 }
2697
2698 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2699 {
2700         unsigned long exit_qualification;
2701         int size, down, in, string, rep;
2702         unsigned port;
2703
2704         ++vcpu->stat.io_exits;
2705         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2706         string = (exit_qualification & 16) != 0;
2707
2708         if (string) {
2709                 if (emulate_instruction(vcpu,
2710                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2711                         return 0;
2712                 return 1;
2713         }
2714
2715         size = (exit_qualification & 7) + 1;
2716         in = (exit_qualification & 8) != 0;
2717         down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
2718         rep = (exit_qualification & 32) != 0;
2719         port = exit_qualification >> 16;
2720
2721         skip_emulated_instruction(vcpu);
2722         return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2723 }
2724
2725 static void
2726 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2727 {
2728         /*
2729          * Patch in the VMCALL instruction:
2730          */
2731         hypercall[0] = 0x0f;
2732         hypercall[1] = 0x01;
2733         hypercall[2] = 0xc1;
2734 }
2735
2736 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2737 {
2738         unsigned long exit_qualification;
2739         int cr;
2740         int reg;
2741
2742         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2743         cr = exit_qualification & 15;
2744         reg = (exit_qualification >> 8) & 15;
2745         switch ((exit_qualification >> 4) & 3) {
2746         case 0: /* mov to cr */
2747                 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2748                             (u32)kvm_register_read(vcpu, reg),
2749                             (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2750                             handler);
2751                 switch (cr) {
2752                 case 0:
2753                         kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
2754                         skip_emulated_instruction(vcpu);
2755                         return 1;
2756                 case 3:
2757                         kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
2758                         skip_emulated_instruction(vcpu);
2759                         return 1;
2760                 case 4:
2761                         kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
2762                         skip_emulated_instruction(vcpu);
2763                         return 1;
2764                 case 8:
2765                         kvm_set_cr8(vcpu, kvm_register_read(vcpu, reg));
2766                         skip_emulated_instruction(vcpu);
2767                         if (irqchip_in_kernel(vcpu->kvm))
2768                                 return 1;
2769                         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2770                         return 0;
2771                 };
2772                 break;
2773         case 2: /* clts */
2774                 vmx_fpu_deactivate(vcpu);
2775                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2776                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2777                 vmx_fpu_activate(vcpu);
2778                 KVMTRACE_0D(CLTS, vcpu, handler);
2779                 skip_emulated_instruction(vcpu);
2780                 return 1;
2781         case 1: /*mov from cr*/
2782                 switch (cr) {
2783                 case 3:
2784                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2785                         KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2786                                     (u32)kvm_register_read(vcpu, reg),
2787                                     (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2788                                     handler);
2789                         skip_emulated_instruction(vcpu);
2790                         return 1;
2791                 case 8:
2792                         kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2793                         KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2794                                     (u32)kvm_register_read(vcpu, reg), handler);
2795                         skip_emulated_instruction(vcpu);
2796                         return 1;
2797                 }
2798                 break;
2799         case 3: /* lmsw */
2800                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2801
2802                 skip_emulated_instruction(vcpu);
2803                 return 1;
2804         default:
2805                 break;
2806         }
2807         kvm_run->exit_reason = 0;
2808         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2809                (int)(exit_qualification >> 4) & 3, cr);
2810         return 0;
2811 }
2812
2813 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2814 {
2815         unsigned long exit_qualification;
2816         unsigned long val;
2817         int dr, reg;
2818
2819         dr = vmcs_readl(GUEST_DR7);
2820         if (dr & DR7_GD) {
2821                 /*
2822                  * As the vm-exit takes precedence over the debug trap, we
2823                  * need to emulate the latter, either for the host or the
2824                  * guest debugging itself.
2825                  */
2826                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2827                         kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2828                         kvm_run->debug.arch.dr7 = dr;
2829                         kvm_run->debug.arch.pc =
2830                                 vmcs_readl(GUEST_CS_BASE) +
2831                                 vmcs_readl(GUEST_RIP);
2832                         kvm_run->debug.arch.exception = DB_VECTOR;
2833                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
2834                         return 0;
2835                 } else {
2836                         vcpu->arch.dr7 &= ~DR7_GD;
2837                         vcpu->arch.dr6 |= DR6_BD;
2838                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2839                         kvm_queue_exception(vcpu, DB_VECTOR);
2840                         return 1;
2841                 }
2842         }
2843
2844         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2845         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2846         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2847         if (exit_qualification & TYPE_MOV_FROM_DR) {
2848                 switch (dr) {
2849                 case 0 ... 3:
2850                         val = vcpu->arch.db[dr];
2851                         break;
2852                 case 6:
2853                         val = vcpu->arch.dr6;
2854                         break;
2855                 case 7:
2856                         val = vcpu->arch.dr7;
2857                         break;
2858                 default:
2859                         val = 0;
2860                 }
2861                 kvm_register_write(vcpu, reg, val);
2862                 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2863         } else {
2864                 val = vcpu->arch.regs[reg];
2865                 switch (dr) {
2866                 case 0 ... 3:
2867                         vcpu->arch.db[dr] = val;
2868                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2869                                 vcpu->arch.eff_db[dr] = val;
2870                         break;
2871                 case 4 ... 5:
2872                         if (vcpu->arch.cr4 & X86_CR4_DE)
2873                                 kvm_queue_exception(vcpu, UD_VECTOR);
2874                         break;
2875                 case 6:
2876                         if (val & 0xffffffff00000000ULL) {
2877                                 kvm_queue_exception(vcpu, GP_VECTOR);
2878                                 break;
2879                         }
2880                         vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2881                         break;
2882                 case 7:
2883                         if (val & 0xffffffff00000000ULL) {
2884                                 kvm_queue_exception(vcpu, GP_VECTOR);
2885                                 break;
2886                         }
2887                         vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2888                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2889                                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2890                                 vcpu->arch.switch_db_regs =
2891                                         (val & DR7_BP_EN_MASK);
2892                         }
2893                         break;
2894                 }
2895                 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)val, handler);
2896         }
2897         skip_emulated_instruction(vcpu);
2898         return 1;
2899 }
2900
2901 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2902 {
2903         kvm_emulate_cpuid(vcpu);
2904         return 1;
2905 }
2906
2907 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2908 {
2909         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2910         u64 data;
2911
2912         if (vmx_get_msr(vcpu, ecx, &data)) {
2913                 kvm_inject_gp(vcpu, 0);
2914                 return 1;
2915         }
2916
2917         KVMTRACE_3D(MSR_READ, vcpu, ecx, (u32)data, (u32)(data >> 32),
2918                     handler);
2919
2920         /* FIXME: handling of bits 32:63 of rax, rdx */
2921         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2922         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2923         skip_emulated_instruction(vcpu);
2924         return 1;
2925 }
2926
2927 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2928 {
2929         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2930         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2931                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2932
2933         KVMTRACE_3D(MSR_WRITE, vcpu, ecx, (u32)data, (u32)(data >> 32),
2934                     handler);
2935
2936         if (vmx_set_msr(vcpu, ecx, data) != 0) {
2937                 kvm_inject_gp(vcpu, 0);
2938                 return 1;
2939         }
2940
2941         skip_emulated_instruction(vcpu);
2942         return 1;
2943 }
2944
2945 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2946                                       struct kvm_run *kvm_run)
2947 {
2948         return 1;
2949 }
2950
2951 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2952                                    struct kvm_run *kvm_run)
2953 {
2954         u32 cpu_based_vm_exec_control;
2955
2956         /* clear pending irq */
2957         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2958         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2959         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2960
2961         KVMTRACE_0D(PEND_INTR, vcpu, handler);
2962         ++vcpu->stat.irq_window_exits;
2963
2964         /*
2965          * If the user space waits to inject interrupts, exit as soon as
2966          * possible
2967          */
2968         if (kvm_run->request_interrupt_window &&
2969             !vcpu->arch.irq_summary) {
2970                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2971                 return 0;
2972         }
2973         return 1;
2974 }
2975
2976 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2977 {
2978         skip_emulated_instruction(vcpu);
2979         return kvm_emulate_halt(vcpu);
2980 }
2981
2982 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2983 {
2984         skip_emulated_instruction(vcpu);
2985         kvm_emulate_hypercall(vcpu);
2986         return 1;
2987 }
2988
2989 static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2990 {
2991         u64 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2992
2993         kvm_mmu_invlpg(vcpu, exit_qualification);
2994         skip_emulated_instruction(vcpu);
2995         return 1;
2996 }
2997
2998 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2999 {
3000         skip_emulated_instruction(vcpu);
3001         /* TODO: Add support for VT-d/pass-through device */
3002         return 1;
3003 }
3004
3005 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3006 {
3007         u64 exit_qualification;
3008         enum emulation_result er;
3009         unsigned long offset;
3010
3011         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
3012         offset = exit_qualification & 0xffful;
3013
3014         er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3015
3016         if (er !=  EMULATE_DONE) {
3017                 printk(KERN_ERR
3018                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3019                        offset);
3020                 return -ENOTSUPP;
3021         }
3022         return 1;
3023 }
3024
3025 static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3026 {
3027         struct vcpu_vmx *vmx = to_vmx(vcpu);
3028         unsigned long exit_qualification;
3029         u16 tss_selector;
3030         int reason;
3031
3032         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3033
3034         reason = (u32)exit_qualification >> 30;
3035         if (reason == TASK_SWITCH_GATE && vmx->vcpu.arch.nmi_injected &&
3036             (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
3037             (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK)
3038             == INTR_TYPE_NMI_INTR) {
3039                 vcpu->arch.nmi_injected = false;
3040                 if (cpu_has_virtual_nmis())
3041                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3042                                       GUEST_INTR_STATE_NMI);
3043         }
3044         tss_selector = exit_qualification;
3045
3046         if (!kvm_task_switch(vcpu, tss_selector, reason))
3047                 return 0;
3048
3049         /* clear all local breakpoint enable flags */
3050         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3051
3052         /*
3053          * TODO: What about debug traps on tss switch?
3054          *       Are we supposed to inject them and update dr6?
3055          */
3056
3057         return 1;
3058 }
3059
3060 static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3061 {
3062         u64 exit_qualification;
3063         enum emulation_result er;
3064         gpa_t gpa;
3065         unsigned long hva;
3066         int gla_validity;
3067         int r;
3068
3069         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
3070
3071         if (exit_qualification & (1 << 6)) {
3072                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3073                 return -ENOTSUPP;
3074         }
3075
3076         gla_validity = (exit_qualification >> 7) & 0x3;
3077         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3078                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3079                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3080                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3081                         (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
3082                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3083                         (long unsigned int)exit_qualification);
3084                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3085                 kvm_run->hw.hardware_exit_reason = 0;
3086                 return -ENOTSUPP;
3087         }
3088
3089         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3090         hva = gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT);
3091         if (!kvm_is_error_hva(hva)) {
3092                 r = kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3093                 if (r < 0) {
3094                         printk(KERN_ERR "EPT: Not enough memory!\n");
3095                         return -ENOMEM;
3096                 }
3097                 return 1;
3098         } else {
3099                 /* must be MMIO */
3100                 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3101
3102                 if (er == EMULATE_FAIL) {
3103                         printk(KERN_ERR
3104                          "EPT: Fail to handle EPT violation vmexit!er is %d\n",
3105                          er);
3106                         printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3107                          (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3108                          (long unsigned int)vmcs_read64(GUEST_LINEAR_ADDRESS));
3109                         printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3110                                 (long unsigned int)exit_qualification);
3111                         return -ENOTSUPP;
3112                 } else if (er == EMULATE_DO_MMIO)
3113                         return 0;
3114         }
3115         return 1;
3116 }
3117
3118 static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3119 {
3120         u32 cpu_based_vm_exec_control;
3121
3122         /* clear pending NMI */
3123         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3124         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3125         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3126         ++vcpu->stat.nmi_window_exits;
3127
3128         return 1;
3129 }
3130
3131 static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3132                                 struct kvm_run *kvm_run)
3133 {
3134         struct vcpu_vmx *vmx = to_vmx(vcpu);
3135         enum emulation_result err = EMULATE_DONE;
3136
3137         preempt_enable();
3138         local_irq_enable();
3139
3140         while (!guest_state_valid(vcpu)) {
3141                 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3142
3143                 if (err == EMULATE_DO_MMIO)
3144                         break;
3145
3146                 if (err != EMULATE_DONE) {
3147                         kvm_report_emulation_failure(vcpu, "emulation failure");
3148                         return;
3149                 }
3150
3151                 if (signal_pending(current))
3152                         break;
3153                 if (need_resched())
3154                         schedule();
3155         }
3156
3157         local_irq_disable();
3158         preempt_disable();
3159
3160         vmx->invalid_state_emulation_result = err;
3161 }
3162
3163 /*
3164  * The exit handlers return 1 if the exit was handled fully and guest execution
3165  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3166  * to be done to userspace and return 0.
3167  */
3168 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3169                                       struct kvm_run *kvm_run) = {
3170         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3171         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3172         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3173         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3174         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3175         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3176         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3177         [EXIT_REASON_CPUID]                   = handle_cpuid,
3178         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3179         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3180         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3181         [EXIT_REASON_HLT]                     = handle_halt,
3182         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3183         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3184         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3185         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3186         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3187         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3188         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3189 };
3190
3191 static const int kvm_vmx_max_exit_handlers =
3192         ARRAY_SIZE(kvm_vmx_exit_handlers);
3193
3194 /*
3195  * The guest has exited.  See if we can fix it or if we need userspace
3196  * assistance.
3197  */
3198 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3199 {
3200         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
3201         struct vcpu_vmx *vmx = to_vmx(vcpu);
3202         u32 vectoring_info = vmx->idt_vectoring_info;
3203
3204         KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
3205                     (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
3206
3207         /* If we need to emulate an MMIO from handle_invalid_guest_state
3208          * we just return 0 */
3209         if (vmx->emulation_required && emulate_invalid_guest_state) {
3210                 if (guest_state_valid(vcpu))
3211                         vmx->emulation_required = 0;
3212                 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
3213         }
3214
3215         /* Access CR3 don't cause VMExit in paging mode, so we need
3216          * to sync with guest real CR3. */
3217         if (vm_need_ept() && is_paging(vcpu)) {
3218                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3219                 ept_load_pdptrs(vcpu);
3220         }
3221
3222         if (unlikely(vmx->fail)) {
3223                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3224                 kvm_run->fail_entry.hardware_entry_failure_reason
3225                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3226                 return 0;
3227         }
3228
3229         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3230                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3231                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3232                         exit_reason != EXIT_REASON_TASK_SWITCH))
3233                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3234                        "(0x%x) and exit reason is 0x%x\n",
3235                        __func__, vectoring_info, exit_reason);
3236
3237         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3238                 if (vcpu->arch.interrupt_window_open) {
3239                         vmx->soft_vnmi_blocked = 0;
3240                         vcpu->arch.nmi_window_open = 1;
3241                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3242                            vcpu->arch.nmi_pending) {
3243                         /*
3244                          * This CPU don't support us in finding the end of an
3245                          * NMI-blocked window if the guest runs with IRQs
3246                          * disabled. So we pull the trigger after 1 s of
3247                          * futile waiting, but inform the user about this.
3248                          */
3249                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3250                                "state on VCPU %d after 1 s timeout\n",
3251                                __func__, vcpu->vcpu_id);
3252                         vmx->soft_vnmi_blocked = 0;
3253                         vmx->vcpu.arch.nmi_window_open = 1;
3254                 }
3255         }
3256
3257         if (exit_reason < kvm_vmx_max_exit_handlers
3258             && kvm_vmx_exit_handlers[exit_reason])
3259                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3260         else {
3261                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3262                 kvm_run->hw.hardware_exit_reason = exit_reason;
3263         }
3264         return 0;
3265 }
3266
3267 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
3268 {
3269         int max_irr, tpr;
3270
3271         if (!vm_need_tpr_shadow(vcpu->kvm))
3272                 return;
3273
3274         if (!kvm_lapic_enabled(vcpu) ||
3275             ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
3276                 vmcs_write32(TPR_THRESHOLD, 0);
3277                 return;
3278         }
3279
3280         tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
3281         vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
3282 }
3283
3284 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3285 {
3286         u32 exit_intr_info;
3287         u32 idt_vectoring_info;
3288         bool unblock_nmi;
3289         u8 vector;
3290         int type;
3291         bool idtv_info_valid;
3292         u32 error;
3293
3294         exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3295         if (cpu_has_virtual_nmis()) {
3296                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3297                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3298                 /*
3299                  * SDM 3: 25.7.1.2
3300                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3301                  * a guest IRET fault.
3302                  */
3303                 if (unblock_nmi && vector != DF_VECTOR)
3304                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3305                                       GUEST_INTR_STATE_NMI);
3306         } else if (unlikely(vmx->soft_vnmi_blocked))
3307                 vmx->vnmi_blocked_time +=
3308                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3309
3310         idt_vectoring_info = vmx->idt_vectoring_info;
3311         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3312         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3313         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3314         if (vmx->vcpu.arch.nmi_injected) {
3315                 /*
3316                  * SDM 3: 25.7.1.2
3317                  * Clear bit "block by NMI" before VM entry if a NMI delivery
3318                  * faulted.
3319                  */
3320                 if (idtv_info_valid && type == INTR_TYPE_NMI_INTR)
3321                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3322                                         GUEST_INTR_STATE_NMI);
3323                 else
3324                         vmx->vcpu.arch.nmi_injected = false;
3325         }
3326         kvm_clear_exception_queue(&vmx->vcpu);
3327         if (idtv_info_valid && (type == INTR_TYPE_HARD_EXCEPTION ||
3328                                 type == INTR_TYPE_SOFT_EXCEPTION)) {
3329                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3330                         error = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3331                         kvm_queue_exception_e(&vmx->vcpu, vector, error);
3332                 } else
3333                         kvm_queue_exception(&vmx->vcpu, vector);
3334                 vmx->idt_vectoring_info = 0;
3335         }
3336         kvm_clear_interrupt_queue(&vmx->vcpu);
3337         if (idtv_info_valid && type == INTR_TYPE_EXT_INTR) {
3338                 kvm_queue_interrupt(&vmx->vcpu, vector);
3339                 vmx->idt_vectoring_info = 0;
3340         }
3341 }
3342
3343 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3344 {
3345         update_tpr_threshold(vcpu);
3346
3347         vmx_update_window_states(vcpu);
3348
3349         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3350                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3351                                 GUEST_INTR_STATE_STI |
3352                                 GUEST_INTR_STATE_MOV_SS);
3353