KVM: x86: add functions to get the cpl of vcpu
[linux-2.6] / arch / x86 / kvm / vmx.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  *
9  * Authors:
10  *   Avi Kivity   <avi@qumranet.com>
11  *   Yaniv Kamay  <yaniv@qumranet.com>
12  *
13  * This work is licensed under the terms of the GNU GPL, version 2.  See
14  * the COPYING file in the top-level directory.
15  *
16  */
17
18 #include "irq.h"
19 #include "vmx.h"
20 #include "mmu.h"
21
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29
30 #include <asm/io.h>
31 #include <asm/desc.h>
32
33 MODULE_AUTHOR("Qumranet");
34 MODULE_LICENSE("GPL");
35
36 static int bypass_guest_pf = 1;
37 module_param(bypass_guest_pf, bool, 0);
38
39 static int enable_vpid = 1;
40 module_param(enable_vpid, bool, 0);
41
42 static int flexpriority_enabled = 1;
43 module_param(flexpriority_enabled, bool, 0);
44
45 struct vmcs {
46         u32 revision_id;
47         u32 abort;
48         char data[0];
49 };
50
51 struct vcpu_vmx {
52         struct kvm_vcpu       vcpu;
53         int                   launched;
54         u8                    fail;
55         u32                   idt_vectoring_info;
56         struct kvm_msr_entry *guest_msrs;
57         struct kvm_msr_entry *host_msrs;
58         int                   nmsrs;
59         int                   save_nmsrs;
60         int                   msr_offset_efer;
61 #ifdef CONFIG_X86_64
62         int                   msr_offset_kernel_gs_base;
63 #endif
64         struct vmcs          *vmcs;
65         struct {
66                 int           loaded;
67                 u16           fs_sel, gs_sel, ldt_sel;
68                 int           gs_ldt_reload_needed;
69                 int           fs_reload_needed;
70                 int           guest_efer_loaded;
71         } host_state;
72         struct {
73                 struct {
74                         bool pending;
75                         u8 vector;
76                         unsigned rip;
77                 } irq;
78         } rmode;
79         int vpid;
80 };
81
82 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
83 {
84         return container_of(vcpu, struct vcpu_vmx, vcpu);
85 }
86
87 static int init_rmode_tss(struct kvm *kvm);
88
89 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
90 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
91
92 static struct page *vmx_io_bitmap_a;
93 static struct page *vmx_io_bitmap_b;
94
95 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
96 static DEFINE_SPINLOCK(vmx_vpid_lock);
97
98 static struct vmcs_config {
99         int size;
100         int order;
101         u32 revision_id;
102         u32 pin_based_exec_ctrl;
103         u32 cpu_based_exec_ctrl;
104         u32 cpu_based_2nd_exec_ctrl;
105         u32 vmexit_ctrl;
106         u32 vmentry_ctrl;
107 } vmcs_config;
108
109 #define VMX_SEGMENT_FIELD(seg)                                  \
110         [VCPU_SREG_##seg] = {                                   \
111                 .selector = GUEST_##seg##_SELECTOR,             \
112                 .base = GUEST_##seg##_BASE,                     \
113                 .limit = GUEST_##seg##_LIMIT,                   \
114                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
115         }
116
117 static struct kvm_vmx_segment_field {
118         unsigned selector;
119         unsigned base;
120         unsigned limit;
121         unsigned ar_bytes;
122 } kvm_vmx_segment_fields[] = {
123         VMX_SEGMENT_FIELD(CS),
124         VMX_SEGMENT_FIELD(DS),
125         VMX_SEGMENT_FIELD(ES),
126         VMX_SEGMENT_FIELD(FS),
127         VMX_SEGMENT_FIELD(GS),
128         VMX_SEGMENT_FIELD(SS),
129         VMX_SEGMENT_FIELD(TR),
130         VMX_SEGMENT_FIELD(LDTR),
131 };
132
133 /*
134  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
135  * away by decrementing the array size.
136  */
137 static const u32 vmx_msr_index[] = {
138 #ifdef CONFIG_X86_64
139         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
140 #endif
141         MSR_EFER, MSR_K6_STAR,
142 };
143 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
144
145 static void load_msrs(struct kvm_msr_entry *e, int n)
146 {
147         int i;
148
149         for (i = 0; i < n; ++i)
150                 wrmsrl(e[i].index, e[i].data);
151 }
152
153 static void save_msrs(struct kvm_msr_entry *e, int n)
154 {
155         int i;
156
157         for (i = 0; i < n; ++i)
158                 rdmsrl(e[i].index, e[i].data);
159 }
160
161 static inline int is_page_fault(u32 intr_info)
162 {
163         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
164                              INTR_INFO_VALID_MASK)) ==
165                 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
166 }
167
168 static inline int is_no_device(u32 intr_info)
169 {
170         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
171                              INTR_INFO_VALID_MASK)) ==
172                 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
173 }
174
175 static inline int is_invalid_opcode(u32 intr_info)
176 {
177         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
178                              INTR_INFO_VALID_MASK)) ==
179                 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
180 }
181
182 static inline int is_external_interrupt(u32 intr_info)
183 {
184         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
185                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
186 }
187
188 static inline int cpu_has_vmx_tpr_shadow(void)
189 {
190         return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
191 }
192
193 static inline int vm_need_tpr_shadow(struct kvm *kvm)
194 {
195         return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
196 }
197
198 static inline int cpu_has_secondary_exec_ctrls(void)
199 {
200         return (vmcs_config.cpu_based_exec_ctrl &
201                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
202 }
203
204 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
205 {
206         return flexpriority_enabled
207                 && (vmcs_config.cpu_based_2nd_exec_ctrl &
208                     SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
209 }
210
211 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
212 {
213         return ((cpu_has_vmx_virtualize_apic_accesses()) &&
214                 (irqchip_in_kernel(kvm)));
215 }
216
217 static inline int cpu_has_vmx_vpid(void)
218 {
219         return (vmcs_config.cpu_based_2nd_exec_ctrl &
220                 SECONDARY_EXEC_ENABLE_VPID);
221 }
222
223 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
224 {
225         int i;
226
227         for (i = 0; i < vmx->nmsrs; ++i)
228                 if (vmx->guest_msrs[i].index == msr)
229                         return i;
230         return -1;
231 }
232
233 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
234 {
235     struct {
236         u64 vpid : 16;
237         u64 rsvd : 48;
238         u64 gva;
239     } operand = { vpid, 0, gva };
240
241     asm volatile (ASM_VMX_INVVPID
242                   /* CF==1 or ZF==1 --> rc = -1 */
243                   "; ja 1f ; ud2 ; 1:"
244                   : : "a"(&operand), "c"(ext) : "cc", "memory");
245 }
246
247 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
248 {
249         int i;
250
251         i = __find_msr_index(vmx, msr);
252         if (i >= 0)
253                 return &vmx->guest_msrs[i];
254         return NULL;
255 }
256
257 static void vmcs_clear(struct vmcs *vmcs)
258 {
259         u64 phys_addr = __pa(vmcs);
260         u8 error;
261
262         asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
263                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
264                       : "cc", "memory");
265         if (error)
266                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
267                        vmcs, phys_addr);
268 }
269
270 static void __vcpu_clear(void *arg)
271 {
272         struct vcpu_vmx *vmx = arg;
273         int cpu = raw_smp_processor_id();
274
275         if (vmx->vcpu.cpu == cpu)
276                 vmcs_clear(vmx->vmcs);
277         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
278                 per_cpu(current_vmcs, cpu) = NULL;
279         rdtscll(vmx->vcpu.arch.host_tsc);
280 }
281
282 static void vcpu_clear(struct vcpu_vmx *vmx)
283 {
284         if (vmx->vcpu.cpu == -1)
285                 return;
286         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
287         vmx->launched = 0;
288 }
289
290 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
291 {
292         if (vmx->vpid == 0)
293                 return;
294
295         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
296 }
297
298 static unsigned long vmcs_readl(unsigned long field)
299 {
300         unsigned long value;
301
302         asm volatile (ASM_VMX_VMREAD_RDX_RAX
303                       : "=a"(value) : "d"(field) : "cc");
304         return value;
305 }
306
307 static u16 vmcs_read16(unsigned long field)
308 {
309         return vmcs_readl(field);
310 }
311
312 static u32 vmcs_read32(unsigned long field)
313 {
314         return vmcs_readl(field);
315 }
316
317 static u64 vmcs_read64(unsigned long field)
318 {
319 #ifdef CONFIG_X86_64
320         return vmcs_readl(field);
321 #else
322         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
323 #endif
324 }
325
326 static noinline void vmwrite_error(unsigned long field, unsigned long value)
327 {
328         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
329                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
330         dump_stack();
331 }
332
333 static void vmcs_writel(unsigned long field, unsigned long value)
334 {
335         u8 error;
336
337         asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
338                        : "=q"(error) : "a"(value), "d"(field) : "cc");
339         if (unlikely(error))
340                 vmwrite_error(field, value);
341 }
342
343 static void vmcs_write16(unsigned long field, u16 value)
344 {
345         vmcs_writel(field, value);
346 }
347
348 static void vmcs_write32(unsigned long field, u32 value)
349 {
350         vmcs_writel(field, value);
351 }
352
353 static void vmcs_write64(unsigned long field, u64 value)
354 {
355 #ifdef CONFIG_X86_64
356         vmcs_writel(field, value);
357 #else
358         vmcs_writel(field, value);
359         asm volatile ("");
360         vmcs_writel(field+1, value >> 32);
361 #endif
362 }
363
364 static void vmcs_clear_bits(unsigned long field, u32 mask)
365 {
366         vmcs_writel(field, vmcs_readl(field) & ~mask);
367 }
368
369 static void vmcs_set_bits(unsigned long field, u32 mask)
370 {
371         vmcs_writel(field, vmcs_readl(field) | mask);
372 }
373
374 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
375 {
376         u32 eb;
377
378         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
379         if (!vcpu->fpu_active)
380                 eb |= 1u << NM_VECTOR;
381         if (vcpu->guest_debug.enabled)
382                 eb |= 1u << 1;
383         if (vcpu->arch.rmode.active)
384                 eb = ~0;
385         vmcs_write32(EXCEPTION_BITMAP, eb);
386 }
387
388 static void reload_tss(void)
389 {
390         /*
391          * VT restores TR but not its size.  Useless.
392          */
393         struct descriptor_table gdt;
394         struct desc_struct *descs;
395
396         get_gdt(&gdt);
397         descs = (void *)gdt.base;
398         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
399         load_TR_desc();
400 }
401
402 static void load_transition_efer(struct vcpu_vmx *vmx)
403 {
404         int efer_offset = vmx->msr_offset_efer;
405         u64 host_efer = vmx->host_msrs[efer_offset].data;
406         u64 guest_efer = vmx->guest_msrs[efer_offset].data;
407         u64 ignore_bits;
408
409         if (efer_offset < 0)
410                 return;
411         /*
412          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
413          * outside long mode
414          */
415         ignore_bits = EFER_NX | EFER_SCE;
416 #ifdef CONFIG_X86_64
417         ignore_bits |= EFER_LMA | EFER_LME;
418         /* SCE is meaningful only in long mode on Intel */
419         if (guest_efer & EFER_LMA)
420                 ignore_bits &= ~(u64)EFER_SCE;
421 #endif
422         if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
423                 return;
424
425         vmx->host_state.guest_efer_loaded = 1;
426         guest_efer &= ~ignore_bits;
427         guest_efer |= host_efer & ignore_bits;
428         wrmsrl(MSR_EFER, guest_efer);
429         vmx->vcpu.stat.efer_reload++;
430 }
431
432 static void reload_host_efer(struct vcpu_vmx *vmx)
433 {
434         if (vmx->host_state.guest_efer_loaded) {
435                 vmx->host_state.guest_efer_loaded = 0;
436                 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
437         }
438 }
439
440 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
441 {
442         struct vcpu_vmx *vmx = to_vmx(vcpu);
443
444         if (vmx->host_state.loaded)
445                 return;
446
447         vmx->host_state.loaded = 1;
448         /*
449          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
450          * allow segment selectors with cpl > 0 or ti == 1.
451          */
452         vmx->host_state.ldt_sel = read_ldt();
453         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
454         vmx->host_state.fs_sel = read_fs();
455         if (!(vmx->host_state.fs_sel & 7)) {
456                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
457                 vmx->host_state.fs_reload_needed = 0;
458         } else {
459                 vmcs_write16(HOST_FS_SELECTOR, 0);
460                 vmx->host_state.fs_reload_needed = 1;
461         }
462         vmx->host_state.gs_sel = read_gs();
463         if (!(vmx->host_state.gs_sel & 7))
464                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
465         else {
466                 vmcs_write16(HOST_GS_SELECTOR, 0);
467                 vmx->host_state.gs_ldt_reload_needed = 1;
468         }
469
470 #ifdef CONFIG_X86_64
471         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
472         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
473 #else
474         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
475         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
476 #endif
477
478 #ifdef CONFIG_X86_64
479         if (is_long_mode(&vmx->vcpu))
480                 save_msrs(vmx->host_msrs +
481                           vmx->msr_offset_kernel_gs_base, 1);
482
483 #endif
484         load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
485         load_transition_efer(vmx);
486 }
487
488 static void vmx_load_host_state(struct vcpu_vmx *vmx)
489 {
490         unsigned long flags;
491
492         if (!vmx->host_state.loaded)
493                 return;
494
495         ++vmx->vcpu.stat.host_state_reload;
496         vmx->host_state.loaded = 0;
497         if (vmx->host_state.fs_reload_needed)
498                 load_fs(vmx->host_state.fs_sel);
499         if (vmx->host_state.gs_ldt_reload_needed) {
500                 load_ldt(vmx->host_state.ldt_sel);
501                 /*
502                  * If we have to reload gs, we must take care to
503                  * preserve our gs base.
504                  */
505                 local_irq_save(flags);
506                 load_gs(vmx->host_state.gs_sel);
507 #ifdef CONFIG_X86_64
508                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
509 #endif
510                 local_irq_restore(flags);
511         }
512         reload_tss();
513         save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
514         load_msrs(vmx->host_msrs, vmx->save_nmsrs);
515         reload_host_efer(vmx);
516 }
517
518 /*
519  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
520  * vcpu mutex is already taken.
521  */
522 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
523 {
524         struct vcpu_vmx *vmx = to_vmx(vcpu);
525         u64 phys_addr = __pa(vmx->vmcs);
526         u64 tsc_this, delta, new_offset;
527
528         if (vcpu->cpu != cpu) {
529                 vcpu_clear(vmx);
530                 kvm_migrate_apic_timer(vcpu);
531                 vpid_sync_vcpu_all(vmx);
532         }
533
534         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
535                 u8 error;
536
537                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
538                 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
539                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
540                               : "cc");
541                 if (error)
542                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
543                                vmx->vmcs, phys_addr);
544         }
545
546         if (vcpu->cpu != cpu) {
547                 struct descriptor_table dt;
548                 unsigned long sysenter_esp;
549
550                 vcpu->cpu = cpu;
551                 /*
552                  * Linux uses per-cpu TSS and GDT, so set these when switching
553                  * processors.
554                  */
555                 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
556                 get_gdt(&dt);
557                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
558
559                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
560                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
561
562                 /*
563                  * Make sure the time stamp counter is monotonous.
564                  */
565                 rdtscll(tsc_this);
566                 if (tsc_this < vcpu->arch.host_tsc) {
567                         delta = vcpu->arch.host_tsc - tsc_this;
568                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
569                         vmcs_write64(TSC_OFFSET, new_offset);
570                 }
571         }
572 }
573
574 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
575 {
576         vmx_load_host_state(to_vmx(vcpu));
577 }
578
579 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
580 {
581         if (vcpu->fpu_active)
582                 return;
583         vcpu->fpu_active = 1;
584         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
585         if (vcpu->arch.cr0 & X86_CR0_TS)
586                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
587         update_exception_bitmap(vcpu);
588 }
589
590 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
591 {
592         if (!vcpu->fpu_active)
593                 return;
594         vcpu->fpu_active = 0;
595         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
596         update_exception_bitmap(vcpu);
597 }
598
599 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
600 {
601         vcpu_clear(to_vmx(vcpu));
602 }
603
604 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
605 {
606         return vmcs_readl(GUEST_RFLAGS);
607 }
608
609 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
610 {
611         if (vcpu->arch.rmode.active)
612                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
613         vmcs_writel(GUEST_RFLAGS, rflags);
614 }
615
616 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
617 {
618         unsigned long rip;
619         u32 interruptibility;
620
621         rip = vmcs_readl(GUEST_RIP);
622         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
623         vmcs_writel(GUEST_RIP, rip);
624
625         /*
626          * We emulated an instruction, so temporary interrupt blocking
627          * should be removed, if set.
628          */
629         interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
630         if (interruptibility & 3)
631                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
632                              interruptibility & ~3);
633         vcpu->arch.interrupt_window_open = 1;
634 }
635
636 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
637                                 bool has_error_code, u32 error_code)
638 {
639         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
640                      nr | INTR_TYPE_EXCEPTION
641                      | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
642                      | INTR_INFO_VALID_MASK);
643         if (has_error_code)
644                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
645 }
646
647 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
648 {
649         struct vcpu_vmx *vmx = to_vmx(vcpu);
650
651         return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
652 }
653
654 /*
655  * Swap MSR entry in host/guest MSR entry array.
656  */
657 #ifdef CONFIG_X86_64
658 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
659 {
660         struct kvm_msr_entry tmp;
661
662         tmp = vmx->guest_msrs[to];
663         vmx->guest_msrs[to] = vmx->guest_msrs[from];
664         vmx->guest_msrs[from] = tmp;
665         tmp = vmx->host_msrs[to];
666         vmx->host_msrs[to] = vmx->host_msrs[from];
667         vmx->host_msrs[from] = tmp;
668 }
669 #endif
670
671 /*
672  * Set up the vmcs to automatically save and restore system
673  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
674  * mode, as fiddling with msrs is very expensive.
675  */
676 static void setup_msrs(struct vcpu_vmx *vmx)
677 {
678         int save_nmsrs;
679
680         vmx_load_host_state(vmx);
681         save_nmsrs = 0;
682 #ifdef CONFIG_X86_64
683         if (is_long_mode(&vmx->vcpu)) {
684                 int index;
685
686                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
687                 if (index >= 0)
688                         move_msr_up(vmx, index, save_nmsrs++);
689                 index = __find_msr_index(vmx, MSR_LSTAR);
690                 if (index >= 0)
691                         move_msr_up(vmx, index, save_nmsrs++);
692                 index = __find_msr_index(vmx, MSR_CSTAR);
693                 if (index >= 0)
694                         move_msr_up(vmx, index, save_nmsrs++);
695                 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
696                 if (index >= 0)
697                         move_msr_up(vmx, index, save_nmsrs++);
698                 /*
699                  * MSR_K6_STAR is only needed on long mode guests, and only
700                  * if efer.sce is enabled.
701                  */
702                 index = __find_msr_index(vmx, MSR_K6_STAR);
703                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
704                         move_msr_up(vmx, index, save_nmsrs++);
705         }
706 #endif
707         vmx->save_nmsrs = save_nmsrs;
708
709 #ifdef CONFIG_X86_64
710         vmx->msr_offset_kernel_gs_base =
711                 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
712 #endif
713         vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
714 }
715
716 /*
717  * reads and returns guest's timestamp counter "register"
718  * guest_tsc = host_tsc + tsc_offset    -- 21.3
719  */
720 static u64 guest_read_tsc(void)
721 {
722         u64 host_tsc, tsc_offset;
723
724         rdtscll(host_tsc);
725         tsc_offset = vmcs_read64(TSC_OFFSET);
726         return host_tsc + tsc_offset;
727 }
728
729 /*
730  * writes 'guest_tsc' into guest's timestamp counter "register"
731  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
732  */
733 static void guest_write_tsc(u64 guest_tsc)
734 {
735         u64 host_tsc;
736
737         rdtscll(host_tsc);
738         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
739 }
740
741 /*
742  * Reads an msr value (of 'msr_index') into 'pdata'.
743  * Returns 0 on success, non-0 otherwise.
744  * Assumes vcpu_load() was already called.
745  */
746 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
747 {
748         u64 data;
749         struct kvm_msr_entry *msr;
750
751         if (!pdata) {
752                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
753                 return -EINVAL;
754         }
755
756         switch (msr_index) {
757 #ifdef CONFIG_X86_64
758         case MSR_FS_BASE:
759                 data = vmcs_readl(GUEST_FS_BASE);
760                 break;
761         case MSR_GS_BASE:
762                 data = vmcs_readl(GUEST_GS_BASE);
763                 break;
764         case MSR_EFER:
765                 return kvm_get_msr_common(vcpu, msr_index, pdata);
766 #endif
767         case MSR_IA32_TIME_STAMP_COUNTER:
768                 data = guest_read_tsc();
769                 break;
770         case MSR_IA32_SYSENTER_CS:
771                 data = vmcs_read32(GUEST_SYSENTER_CS);
772                 break;
773         case MSR_IA32_SYSENTER_EIP:
774                 data = vmcs_readl(GUEST_SYSENTER_EIP);
775                 break;
776         case MSR_IA32_SYSENTER_ESP:
777                 data = vmcs_readl(GUEST_SYSENTER_ESP);
778                 break;
779         default:
780                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
781                 if (msr) {
782                         data = msr->data;
783                         break;
784                 }
785                 return kvm_get_msr_common(vcpu, msr_index, pdata);
786         }
787
788         *pdata = data;
789         return 0;
790 }
791
792 /*
793  * Writes msr value into into the appropriate "register".
794  * Returns 0 on success, non-0 otherwise.
795  * Assumes vcpu_load() was already called.
796  */
797 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
798 {
799         struct vcpu_vmx *vmx = to_vmx(vcpu);
800         struct kvm_msr_entry *msr;
801         int ret = 0;
802
803         switch (msr_index) {
804 #ifdef CONFIG_X86_64
805         case MSR_EFER:
806                 ret = kvm_set_msr_common(vcpu, msr_index, data);
807                 if (vmx->host_state.loaded) {
808                         reload_host_efer(vmx);
809                         load_transition_efer(vmx);
810                 }
811                 break;
812         case MSR_FS_BASE:
813                 vmcs_writel(GUEST_FS_BASE, data);
814                 break;
815         case MSR_GS_BASE:
816                 vmcs_writel(GUEST_GS_BASE, data);
817                 break;
818 #endif
819         case MSR_IA32_SYSENTER_CS:
820                 vmcs_write32(GUEST_SYSENTER_CS, data);
821                 break;
822         case MSR_IA32_SYSENTER_EIP:
823                 vmcs_writel(GUEST_SYSENTER_EIP, data);
824                 break;
825         case MSR_IA32_SYSENTER_ESP:
826                 vmcs_writel(GUEST_SYSENTER_ESP, data);
827                 break;
828         case MSR_IA32_TIME_STAMP_COUNTER:
829                 guest_write_tsc(data);
830                 break;
831         default:
832                 msr = find_msr_entry(vmx, msr_index);
833                 if (msr) {
834                         msr->data = data;
835                         if (vmx->host_state.loaded)
836                                 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
837                         break;
838                 }
839                 ret = kvm_set_msr_common(vcpu, msr_index, data);
840         }
841
842         return ret;
843 }
844
845 /*
846  * Sync the rsp and rip registers into the vcpu structure.  This allows
847  * registers to be accessed by indexing vcpu->arch.regs.
848  */
849 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
850 {
851         vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
852         vcpu->arch.rip = vmcs_readl(GUEST_RIP);
853 }
854
855 /*
856  * Syncs rsp and rip back into the vmcs.  Should be called after possible
857  * modification.
858  */
859 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
860 {
861         vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
862         vmcs_writel(GUEST_RIP, vcpu->arch.rip);
863 }
864
865 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
866 {
867         unsigned long dr7 = 0x400;
868         int old_singlestep;
869
870         old_singlestep = vcpu->guest_debug.singlestep;
871
872         vcpu->guest_debug.enabled = dbg->enabled;
873         if (vcpu->guest_debug.enabled) {
874                 int i;
875
876                 dr7 |= 0x200;  /* exact */
877                 for (i = 0; i < 4; ++i) {
878                         if (!dbg->breakpoints[i].enabled)
879                                 continue;
880                         vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
881                         dr7 |= 2 << (i*2);    /* global enable */
882                         dr7 |= 0 << (i*4+16); /* execution breakpoint */
883                 }
884
885                 vcpu->guest_debug.singlestep = dbg->singlestep;
886         } else
887                 vcpu->guest_debug.singlestep = 0;
888
889         if (old_singlestep && !vcpu->guest_debug.singlestep) {
890                 unsigned long flags;
891
892                 flags = vmcs_readl(GUEST_RFLAGS);
893                 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
894                 vmcs_writel(GUEST_RFLAGS, flags);
895         }
896
897         update_exception_bitmap(vcpu);
898         vmcs_writel(GUEST_DR7, dr7);
899
900         return 0;
901 }
902
903 static int vmx_get_irq(struct kvm_vcpu *vcpu)
904 {
905         struct vcpu_vmx *vmx = to_vmx(vcpu);
906         u32 idtv_info_field;
907
908         idtv_info_field = vmx->idt_vectoring_info;
909         if (idtv_info_field & INTR_INFO_VALID_MASK) {
910                 if (is_external_interrupt(idtv_info_field))
911                         return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
912                 else
913                         printk(KERN_DEBUG "pending exception: not handled yet\n");
914         }
915         return -1;
916 }
917
918 static __init int cpu_has_kvm_support(void)
919 {
920         unsigned long ecx = cpuid_ecx(1);
921         return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
922 }
923
924 static __init int vmx_disabled_by_bios(void)
925 {
926         u64 msr;
927
928         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
929         return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
930                        MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
931             == MSR_IA32_FEATURE_CONTROL_LOCKED;
932         /* locked but not enabled */
933 }
934
935 static void hardware_enable(void *garbage)
936 {
937         int cpu = raw_smp_processor_id();
938         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
939         u64 old;
940
941         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
942         if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
943                     MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
944             != (MSR_IA32_FEATURE_CONTROL_LOCKED |
945                 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
946                 /* enable and lock */
947                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
948                        MSR_IA32_FEATURE_CONTROL_LOCKED |
949                        MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
950         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
951         asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
952                       : "memory", "cc");
953 }
954
955 static void hardware_disable(void *garbage)
956 {
957         asm volatile (ASM_VMX_VMXOFF : : : "cc");
958 }
959
960 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
961                                       u32 msr, u32 *result)
962 {
963         u32 vmx_msr_low, vmx_msr_high;
964         u32 ctl = ctl_min | ctl_opt;
965
966         rdmsr(msr, vmx_msr_low, vmx_msr_high);
967
968         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
969         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
970
971         /* Ensure minimum (required) set of control bits are supported. */
972         if (ctl_min & ~ctl)
973                 return -EIO;
974
975         *result = ctl;
976         return 0;
977 }
978
979 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
980 {
981         u32 vmx_msr_low, vmx_msr_high;
982         u32 min, opt;
983         u32 _pin_based_exec_control = 0;
984         u32 _cpu_based_exec_control = 0;
985         u32 _cpu_based_2nd_exec_control = 0;
986         u32 _vmexit_control = 0;
987         u32 _vmentry_control = 0;
988
989         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
990         opt = 0;
991         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
992                                 &_pin_based_exec_control) < 0)
993                 return -EIO;
994
995         min = CPU_BASED_HLT_EXITING |
996 #ifdef CONFIG_X86_64
997               CPU_BASED_CR8_LOAD_EXITING |
998               CPU_BASED_CR8_STORE_EXITING |
999 #endif
1000               CPU_BASED_USE_IO_BITMAPS |
1001               CPU_BASED_MOV_DR_EXITING |
1002               CPU_BASED_USE_TSC_OFFSETING;
1003         opt = CPU_BASED_TPR_SHADOW |
1004               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1005         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1006                                 &_cpu_based_exec_control) < 0)
1007                 return -EIO;
1008 #ifdef CONFIG_X86_64
1009         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1010                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1011                                            ~CPU_BASED_CR8_STORE_EXITING;
1012 #endif
1013         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1014                 min = 0;
1015                 opt = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1016                         SECONDARY_EXEC_WBINVD_EXITING |
1017                         SECONDARY_EXEC_ENABLE_VPID;
1018                 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS2,
1019                                         &_cpu_based_2nd_exec_control) < 0)
1020                         return -EIO;
1021         }
1022 #ifndef CONFIG_X86_64
1023         if (!(_cpu_based_2nd_exec_control &
1024                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1025                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1026 #endif
1027
1028         min = 0;
1029 #ifdef CONFIG_X86_64
1030         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1031 #endif
1032         opt = 0;
1033         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1034                                 &_vmexit_control) < 0)
1035                 return -EIO;
1036
1037         min = opt = 0;
1038         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1039                                 &_vmentry_control) < 0)
1040                 return -EIO;
1041
1042         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1043
1044         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1045         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1046                 return -EIO;
1047
1048 #ifdef CONFIG_X86_64
1049         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1050         if (vmx_msr_high & (1u<<16))
1051                 return -EIO;
1052 #endif
1053
1054         /* Require Write-Back (WB) memory type for VMCS accesses. */
1055         if (((vmx_msr_high >> 18) & 15) != 6)
1056                 return -EIO;
1057
1058         vmcs_conf->size = vmx_msr_high & 0x1fff;
1059         vmcs_conf->order = get_order(vmcs_config.size);
1060         vmcs_conf->revision_id = vmx_msr_low;
1061
1062         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1063         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1064         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1065         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1066         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1067
1068         return 0;
1069 }
1070
1071 static struct vmcs *alloc_vmcs_cpu(int cpu)
1072 {
1073         int node = cpu_to_node(cpu);
1074         struct page *pages;
1075         struct vmcs *vmcs;
1076
1077         pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1078         if (!pages)
1079                 return NULL;
1080         vmcs = page_address(pages);
1081         memset(vmcs, 0, vmcs_config.size);
1082         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1083         return vmcs;
1084 }
1085
1086 static struct vmcs *alloc_vmcs(void)
1087 {
1088         return alloc_vmcs_cpu(raw_smp_processor_id());
1089 }
1090
1091 static void free_vmcs(struct vmcs *vmcs)
1092 {
1093         free_pages((unsigned long)vmcs, vmcs_config.order);
1094 }
1095
1096 static void free_kvm_area(void)
1097 {
1098         int cpu;
1099
1100         for_each_online_cpu(cpu)
1101                 free_vmcs(per_cpu(vmxarea, cpu));
1102 }
1103
1104 static __init int alloc_kvm_area(void)
1105 {
1106         int cpu;
1107
1108         for_each_online_cpu(cpu) {
1109                 struct vmcs *vmcs;
1110
1111                 vmcs = alloc_vmcs_cpu(cpu);
1112                 if (!vmcs) {
1113                         free_kvm_area();
1114                         return -ENOMEM;
1115                 }
1116
1117                 per_cpu(vmxarea, cpu) = vmcs;
1118         }
1119         return 0;
1120 }
1121
1122 static __init int hardware_setup(void)
1123 {
1124         if (setup_vmcs_config(&vmcs_config) < 0)
1125                 return -EIO;
1126
1127         if (boot_cpu_has(X86_FEATURE_NX))
1128                 kvm_enable_efer_bits(EFER_NX);
1129
1130         return alloc_kvm_area();
1131 }
1132
1133 static __exit void hardware_unsetup(void)
1134 {
1135         free_kvm_area();
1136 }
1137
1138 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1139 {
1140         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1141
1142         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1143                 vmcs_write16(sf->selector, save->selector);
1144                 vmcs_writel(sf->base, save->base);
1145                 vmcs_write32(sf->limit, save->limit);
1146                 vmcs_write32(sf->ar_bytes, save->ar);
1147         } else {
1148                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1149                         << AR_DPL_SHIFT;
1150                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1151         }
1152 }
1153
1154 static void enter_pmode(struct kvm_vcpu *vcpu)
1155 {
1156         unsigned long flags;
1157
1158         vcpu->arch.rmode.active = 0;
1159
1160         vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1161         vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1162         vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1163
1164         flags = vmcs_readl(GUEST_RFLAGS);
1165         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1166         flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1167         vmcs_writel(GUEST_RFLAGS, flags);
1168
1169         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1170                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1171
1172         update_exception_bitmap(vcpu);
1173
1174         fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1175         fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1176         fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1177         fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1178
1179         vmcs_write16(GUEST_SS_SELECTOR, 0);
1180         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1181
1182         vmcs_write16(GUEST_CS_SELECTOR,
1183                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1184         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1185 }
1186
1187 static gva_t rmode_tss_base(struct kvm *kvm)
1188 {
1189         if (!kvm->arch.tss_addr) {
1190                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1191                                  kvm->memslots[0].npages - 3;
1192                 return base_gfn << PAGE_SHIFT;
1193         }
1194         return kvm->arch.tss_addr;
1195 }
1196
1197 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1198 {
1199         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1200
1201         save->selector = vmcs_read16(sf->selector);
1202         save->base = vmcs_readl(sf->base);
1203         save->limit = vmcs_read32(sf->limit);
1204         save->ar = vmcs_read32(sf->ar_bytes);
1205         vmcs_write16(sf->selector, save->base >> 4);
1206         vmcs_write32(sf->base, save->base & 0xfffff);
1207         vmcs_write32(sf->limit, 0xffff);
1208         vmcs_write32(sf->ar_bytes, 0xf3);
1209 }
1210
1211 static void enter_rmode(struct kvm_vcpu *vcpu)
1212 {
1213         unsigned long flags;
1214
1215         vcpu->arch.rmode.active = 1;
1216
1217         vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1218         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1219
1220         vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1221         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1222
1223         vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1224         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1225
1226         flags = vmcs_readl(GUEST_RFLAGS);
1227         vcpu->arch.rmode.save_iopl
1228                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1229
1230         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1231
1232         vmcs_writel(GUEST_RFLAGS, flags);
1233         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1234         update_exception_bitmap(vcpu);
1235
1236         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1237         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1238         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1239
1240         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1241         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1242         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1243                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1244         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1245
1246         fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1247         fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1248         fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1249         fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1250
1251         kvm_mmu_reset_context(vcpu);
1252         init_rmode_tss(vcpu->kvm);
1253 }
1254
1255 #ifdef CONFIG_X86_64
1256
1257 static void enter_lmode(struct kvm_vcpu *vcpu)
1258 {
1259         u32 guest_tr_ar;
1260
1261         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1262         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1263                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1264                        __func__);
1265                 vmcs_write32(GUEST_TR_AR_BYTES,
1266                              (guest_tr_ar & ~AR_TYPE_MASK)
1267                              | AR_TYPE_BUSY_64_TSS);
1268         }
1269
1270         vcpu->arch.shadow_efer |= EFER_LMA;
1271
1272         find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1273         vmcs_write32(VM_ENTRY_CONTROLS,
1274                      vmcs_read32(VM_ENTRY_CONTROLS)
1275                      | VM_ENTRY_IA32E_MODE);
1276 }
1277
1278 static void exit_lmode(struct kvm_vcpu *vcpu)
1279 {
1280         vcpu->arch.shadow_efer &= ~EFER_LMA;
1281
1282         vmcs_write32(VM_ENTRY_CONTROLS,
1283                      vmcs_read32(VM_ENTRY_CONTROLS)
1284                      & ~VM_ENTRY_IA32E_MODE);
1285 }
1286
1287 #endif
1288
1289 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1290 {
1291         vpid_sync_vcpu_all(to_vmx(vcpu));
1292 }
1293
1294 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1295 {
1296         vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1297         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1298 }
1299
1300 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1301 {
1302         vmx_fpu_deactivate(vcpu);
1303
1304         if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1305                 enter_pmode(vcpu);
1306
1307         if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1308                 enter_rmode(vcpu);
1309
1310 #ifdef CONFIG_X86_64
1311         if (vcpu->arch.shadow_efer & EFER_LME) {
1312                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1313                         enter_lmode(vcpu);
1314                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1315                         exit_lmode(vcpu);
1316         }
1317 #endif
1318
1319         vmcs_writel(CR0_READ_SHADOW, cr0);
1320         vmcs_writel(GUEST_CR0,
1321                     (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
1322         vcpu->arch.cr0 = cr0;
1323
1324         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1325                 vmx_fpu_activate(vcpu);
1326 }
1327
1328 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1329 {
1330         vmx_flush_tlb(vcpu);
1331         vmcs_writel(GUEST_CR3, cr3);
1332         if (vcpu->arch.cr0 & X86_CR0_PE)
1333                 vmx_fpu_deactivate(vcpu);
1334 }
1335
1336 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1337 {
1338         vmcs_writel(CR4_READ_SHADOW, cr4);
1339         vmcs_writel(GUEST_CR4, cr4 | (vcpu->arch.rmode.active ?
1340                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
1341         vcpu->arch.cr4 = cr4;
1342 }
1343
1344 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1345 {
1346         struct vcpu_vmx *vmx = to_vmx(vcpu);
1347         struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1348
1349         vcpu->arch.shadow_efer = efer;
1350         if (!msr)
1351                 return;
1352         if (efer & EFER_LMA) {
1353                 vmcs_write32(VM_ENTRY_CONTROLS,
1354                                      vmcs_read32(VM_ENTRY_CONTROLS) |
1355                                      VM_ENTRY_IA32E_MODE);
1356                 msr->data = efer;
1357
1358         } else {
1359                 vmcs_write32(VM_ENTRY_CONTROLS,
1360                                      vmcs_read32(VM_ENTRY_CONTROLS) &
1361                                      ~VM_ENTRY_IA32E_MODE);
1362
1363                 msr->data = efer & ~EFER_LME;
1364         }
1365         setup_msrs(vmx);
1366 }
1367
1368 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1369 {
1370         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1371
1372         return vmcs_readl(sf->base);
1373 }
1374
1375 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1376                             struct kvm_segment *var, int seg)
1377 {
1378         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1379         u32 ar;
1380
1381         var->base = vmcs_readl(sf->base);
1382         var->limit = vmcs_read32(sf->limit);
1383         var->selector = vmcs_read16(sf->selector);
1384         ar = vmcs_read32(sf->ar_bytes);
1385         if (ar & AR_UNUSABLE_MASK)
1386                 ar = 0;
1387         var->type = ar & 15;
1388         var->s = (ar >> 4) & 1;
1389         var->dpl = (ar >> 5) & 3;
1390         var->present = (ar >> 7) & 1;
1391         var->avl = (ar >> 12) & 1;
1392         var->l = (ar >> 13) & 1;
1393         var->db = (ar >> 14) & 1;
1394         var->g = (ar >> 15) & 1;
1395         var->unusable = (ar >> 16) & 1;
1396 }
1397
1398 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1399 {
1400         struct kvm_segment kvm_seg;
1401
1402         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1403                 return 0;
1404
1405         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1406                 return 3;
1407
1408         vmx_get_segment(vcpu, &kvm_seg, VCPU_SREG_CS);
1409         return kvm_seg.selector & 3;
1410 }
1411
1412 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1413 {
1414         u32 ar;
1415
1416         if (var->unusable)
1417                 ar = 1 << 16;
1418         else {
1419                 ar = var->type & 15;
1420                 ar |= (var->s & 1) << 4;
1421                 ar |= (var->dpl & 3) << 5;
1422                 ar |= (var->present & 1) << 7;
1423                 ar |= (var->avl & 1) << 12;
1424                 ar |= (var->l & 1) << 13;
1425                 ar |= (var->db & 1) << 14;
1426                 ar |= (var->g & 1) << 15;
1427         }
1428         if (ar == 0) /* a 0 value means unusable */
1429                 ar = AR_UNUSABLE_MASK;
1430
1431         return ar;
1432 }
1433
1434 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1435                             struct kvm_segment *var, int seg)
1436 {
1437         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1438         u32 ar;
1439
1440         if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1441                 vcpu->arch.rmode.tr.selector = var->selector;
1442                 vcpu->arch.rmode.tr.base = var->base;
1443                 vcpu->arch.rmode.tr.limit = var->limit;
1444                 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1445                 return;
1446         }
1447         vmcs_writel(sf->base, var->base);
1448         vmcs_write32(sf->limit, var->limit);
1449         vmcs_write16(sf->selector, var->selector);
1450         if (vcpu->arch.rmode.active && var->s) {
1451                 /*
1452                  * Hack real-mode segments into vm86 compatibility.
1453                  */
1454                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1455                         vmcs_writel(sf->base, 0xf0000);
1456                 ar = 0xf3;
1457         } else
1458                 ar = vmx_segment_access_rights(var);
1459         vmcs_write32(sf->ar_bytes, ar);
1460 }
1461
1462 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1463 {
1464         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1465
1466         *db = (ar >> 14) & 1;
1467         *l = (ar >> 13) & 1;
1468 }
1469
1470 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1471 {
1472         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1473         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1474 }
1475
1476 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1477 {
1478         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1479         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1480 }
1481
1482 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1483 {
1484         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1485         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1486 }
1487
1488 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1489 {
1490         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1491         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1492 }
1493
1494 static int init_rmode_tss(struct kvm *kvm)
1495 {
1496         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1497         u16 data = 0;
1498         int ret = 0;
1499         int r;
1500
1501         down_read(&kvm->slots_lock);
1502         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1503         if (r < 0)
1504                 goto out;
1505         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1506         r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1507         if (r < 0)
1508                 goto out;
1509         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1510         if (r < 0)
1511                 goto out;
1512         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1513         if (r < 0)
1514                 goto out;
1515         data = ~0;
1516         r = kvm_write_guest_page(kvm, fn, &data,
1517                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1518                                  sizeof(u8));
1519         if (r < 0)
1520                 goto out;
1521
1522         ret = 1;
1523 out:
1524         up_read(&kvm->slots_lock);
1525         return ret;
1526 }
1527
1528 static void seg_setup(int seg)
1529 {
1530         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1531
1532         vmcs_write16(sf->selector, 0);
1533         vmcs_writel(sf->base, 0);
1534         vmcs_write32(sf->limit, 0xffff);
1535         vmcs_write32(sf->ar_bytes, 0x93);
1536 }
1537
1538 static int alloc_apic_access_page(struct kvm *kvm)
1539 {
1540         struct kvm_userspace_memory_region kvm_userspace_mem;
1541         int r = 0;
1542
1543         down_write(&kvm->slots_lock);
1544         if (kvm->arch.apic_access_page)
1545                 goto out;
1546         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
1547         kvm_userspace_mem.flags = 0;
1548         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
1549         kvm_userspace_mem.memory_size = PAGE_SIZE;
1550         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1551         if (r)
1552                 goto out;
1553
1554         down_read(&current->mm->mmap_sem);
1555         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
1556         up_read(&current->mm->mmap_sem);
1557 out:
1558         up_write(&kvm->slots_lock);
1559         return r;
1560 }
1561
1562 static void allocate_vpid(struct vcpu_vmx *vmx)
1563 {
1564         int vpid;
1565
1566         vmx->vpid = 0;
1567         if (!enable_vpid || !cpu_has_vmx_vpid())
1568                 return;
1569         spin_lock(&vmx_vpid_lock);
1570         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
1571         if (vpid < VMX_NR_VPIDS) {
1572                 vmx->vpid = vpid;
1573                 __set_bit(vpid, vmx_vpid_bitmap);
1574         }
1575         spin_unlock(&vmx_vpid_lock);
1576 }
1577
1578 /*
1579  * Sets up the vmcs for emulated real mode.
1580  */
1581 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1582 {
1583         u32 host_sysenter_cs;
1584         u32 junk;
1585         unsigned long a;
1586         struct descriptor_table dt;
1587         int i;
1588         unsigned long kvm_vmx_return;
1589         u32 exec_control;
1590
1591         /* I/O */
1592         vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1593         vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1594
1595         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1596
1597         /* Control */
1598         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1599                 vmcs_config.pin_based_exec_ctrl);
1600
1601         exec_control = vmcs_config.cpu_based_exec_ctrl;
1602         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1603                 exec_control &= ~CPU_BASED_TPR_SHADOW;
1604 #ifdef CONFIG_X86_64
1605                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1606                                 CPU_BASED_CR8_LOAD_EXITING;
1607 #endif
1608         }
1609         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1610
1611         if (cpu_has_secondary_exec_ctrls()) {
1612                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
1613                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1614                         exec_control &=
1615                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1616                 if (vmx->vpid == 0)
1617                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
1618                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
1619         }
1620
1621         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1622         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
1623         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
1624
1625         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
1626         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
1627         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
1628
1629         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
1630         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1631         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1632         vmcs_write16(HOST_FS_SELECTOR, read_fs());    /* 22.2.4 */
1633         vmcs_write16(HOST_GS_SELECTOR, read_gs());    /* 22.2.4 */
1634         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
1635 #ifdef CONFIG_X86_64
1636         rdmsrl(MSR_FS_BASE, a);
1637         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1638         rdmsrl(MSR_GS_BASE, a);
1639         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1640 #else
1641         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1642         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1643 #endif
1644
1645         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
1646
1647         get_idt(&dt);
1648         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
1649
1650         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1651         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1652         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1653         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1654         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1655
1656         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1657         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1658         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1659         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
1660         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1661         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
1662
1663         for (i = 0; i < NR_VMX_MSR; ++i) {
1664                 u32 index = vmx_msr_index[i];
1665                 u32 data_low, data_high;
1666                 u64 data;
1667                 int j = vmx->nmsrs;
1668
1669                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1670                         continue;
1671                 if (wrmsr_safe(index, data_low, data_high) < 0)
1672                         continue;
1673                 data = data_low | ((u64)data_high << 32);
1674                 vmx->host_msrs[j].index = index;
1675                 vmx->host_msrs[j].reserved = 0;
1676                 vmx->host_msrs[j].data = data;
1677                 vmx->guest_msrs[j] = vmx->host_msrs[j];
1678                 ++vmx->nmsrs;
1679         }
1680
1681         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
1682
1683         /* 22.2.1, 20.8.1 */
1684         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1685
1686         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1687         vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1688
1689
1690         return 0;
1691 }
1692
1693 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1694 {
1695         struct vcpu_vmx *vmx = to_vmx(vcpu);
1696         u64 msr;
1697         int ret;
1698
1699         if (!init_rmode_tss(vmx->vcpu.kvm)) {
1700                 ret = -ENOMEM;
1701                 goto out;
1702         }
1703
1704         vmx->vcpu.arch.rmode.active = 0;
1705
1706         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
1707         kvm_set_cr8(&vmx->vcpu, 0);
1708         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1709         if (vmx->vcpu.vcpu_id == 0)
1710                 msr |= MSR_IA32_APICBASE_BSP;
1711         kvm_set_apic_base(&vmx->vcpu, msr);
1712
1713         fx_init(&vmx->vcpu);
1714
1715         /*
1716          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1717          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
1718          */
1719         if (vmx->vcpu.vcpu_id == 0) {
1720                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1721                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1722         } else {
1723                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
1724                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
1725         }
1726         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1727         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1728
1729         seg_setup(VCPU_SREG_DS);
1730         seg_setup(VCPU_SREG_ES);
1731         seg_setup(VCPU_SREG_FS);
1732         seg_setup(VCPU_SREG_GS);
1733         seg_setup(VCPU_SREG_SS);
1734
1735         vmcs_write16(GUEST_TR_SELECTOR, 0);
1736         vmcs_writel(GUEST_TR_BASE, 0);
1737         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1738         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1739
1740         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1741         vmcs_writel(GUEST_LDTR_BASE, 0);
1742         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1743         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1744
1745         vmcs_write32(GUEST_SYSENTER_CS, 0);
1746         vmcs_writel(GUEST_SYSENTER_ESP, 0);
1747         vmcs_writel(GUEST_SYSENTER_EIP, 0);
1748
1749         vmcs_writel(GUEST_RFLAGS, 0x02);
1750         if (vmx->vcpu.vcpu_id == 0)
1751                 vmcs_writel(GUEST_RIP, 0xfff0);
1752         else
1753                 vmcs_writel(GUEST_RIP, 0);
1754         vmcs_writel(GUEST_RSP, 0);
1755
1756         /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
1757         vmcs_writel(GUEST_DR7, 0x400);
1758
1759         vmcs_writel(GUEST_GDTR_BASE, 0);
1760         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1761
1762         vmcs_writel(GUEST_IDTR_BASE, 0);
1763         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1764
1765         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1766         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1767         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1768
1769         guest_write_tsc(0);
1770
1771         /* Special registers */
1772         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1773
1774         setup_msrs(vmx);
1775
1776         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
1777
1778         if (cpu_has_vmx_tpr_shadow()) {
1779                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
1780                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
1781                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
1782                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
1783                 vmcs_write32(TPR_THRESHOLD, 0);
1784         }
1785
1786         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1787                 vmcs_write64(APIC_ACCESS_ADDR,
1788                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
1789
1790         if (vmx->vpid != 0)
1791                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
1792
1793         vmx->vcpu.arch.cr0 = 0x60000010;
1794         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
1795         vmx_set_cr4(&vmx->vcpu, 0);
1796         vmx_set_efer(&vmx->vcpu, 0);
1797         vmx_fpu_activate(&vmx->vcpu);
1798         update_exception_bitmap(&vmx->vcpu);
1799
1800         vpid_sync_vcpu_all(vmx);
1801
1802         return 0;
1803
1804 out:
1805         return ret;
1806 }
1807
1808 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
1809 {
1810         struct vcpu_vmx *vmx = to_vmx(vcpu);
1811
1812         if (vcpu->arch.rmode.active) {
1813                 vmx->rmode.irq.pending = true;
1814                 vmx->rmode.irq.vector = irq;
1815                 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP);
1816                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1817                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
1818                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
1819                 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1);
1820                 return;
1821         }
1822         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1823                         irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1824 }
1825
1826 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1827 {
1828         int word_index = __ffs(vcpu->arch.irq_summary);
1829         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1830         int irq = word_index * BITS_PER_LONG + bit_index;
1831
1832         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1833         if (!vcpu->arch.irq_pending[word_index])
1834                 clear_bit(word_index, &vcpu->arch.irq_summary);
1835         vmx_inject_irq(vcpu, irq);
1836 }
1837
1838
1839 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1840                                        struct kvm_run *kvm_run)
1841 {
1842         u32 cpu_based_vm_exec_control;
1843
1844         vcpu->arch.interrupt_window_open =
1845                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1846                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1847
1848         if (vcpu->arch.interrupt_window_open &&
1849             vcpu->arch.irq_summary &&
1850             !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1851                 /*
1852                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1853                  */
1854                 kvm_do_inject_irq(vcpu);
1855
1856         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1857         if (!vcpu->arch.interrupt_window_open &&
1858             (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
1859                 /*
1860                  * Interrupts blocked.  Wait for unblock.
1861                  */
1862                 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1863         else
1864                 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1865         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1866 }
1867
1868 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
1869 {
1870         int ret;
1871         struct kvm_userspace_memory_region tss_mem = {
1872                 .slot = 8,
1873                 .guest_phys_addr = addr,
1874                 .memory_size = PAGE_SIZE * 3,
1875                 .flags = 0,
1876         };
1877
1878         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
1879         if (ret)
1880                 return ret;
1881         kvm->arch.tss_addr = addr;
1882         return 0;
1883 }
1884
1885 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1886 {
1887         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1888
1889         set_debugreg(dbg->bp[0], 0);
1890         set_debugreg(dbg->bp[1], 1);
1891         set_debugreg(dbg->bp[2], 2);
1892         set_debugreg(dbg->bp[3], 3);
1893
1894         if (dbg->singlestep) {
1895                 unsigned long flags;
1896
1897                 flags = vmcs_readl(GUEST_RFLAGS);
1898                 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1899                 vmcs_writel(GUEST_RFLAGS, flags);
1900         }
1901 }
1902
1903 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1904                                   int vec, u32 err_code)
1905 {
1906         if (!vcpu->arch.rmode.active)
1907                 return 0;
1908
1909         /*
1910          * Instruction with address size override prefix opcode 0x67
1911          * Cause the #SS fault with 0 error code in VM86 mode.
1912          */
1913         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1914                 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
1915                         return 1;
1916         return 0;
1917 }
1918
1919 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1920 {
1921         struct vcpu_vmx *vmx = to_vmx(vcpu);
1922         u32 intr_info, error_code;
1923         unsigned long cr2, rip;
1924         u32 vect_info;
1925         enum emulation_result er;
1926
1927         vect_info = vmx->idt_vectoring_info;
1928         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1929
1930         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1931                                                 !is_page_fault(intr_info))
1932                 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1933                        "intr info 0x%x\n", __func__, vect_info, intr_info);
1934
1935         if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
1936                 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1937                 set_bit(irq, vcpu->arch.irq_pending);
1938                 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1939         }
1940
1941         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
1942                 return 1;  /* already handled by vmx_vcpu_run() */
1943
1944         if (is_no_device(intr_info)) {
1945                 vmx_fpu_activate(vcpu);
1946                 return 1;
1947         }
1948
1949         if (is_invalid_opcode(intr_info)) {
1950                 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1951                 if (er != EMULATE_DONE)
1952                         kvm_queue_exception(vcpu, UD_VECTOR);
1953                 return 1;
1954         }
1955
1956         error_code = 0;
1957         rip = vmcs_readl(GUEST_RIP);
1958         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
1959                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1960         if (is_page_fault(intr_info)) {
1961                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1962                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
1963         }
1964
1965         if (vcpu->arch.rmode.active &&
1966             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1967                                                                 error_code)) {
1968                 if (vcpu->arch.halt_request) {
1969                         vcpu->arch.halt_request = 0;
1970                         return kvm_emulate_halt(vcpu);
1971                 }
1972                 return 1;
1973         }
1974
1975         if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
1976             (INTR_TYPE_EXCEPTION | 1)) {
1977                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1978                 return 0;
1979         }
1980         kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1981         kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1982         kvm_run->ex.error_code = error_code;
1983         return 0;
1984 }
1985
1986 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1987                                      struct kvm_run *kvm_run)
1988 {
1989         ++vcpu->stat.irq_exits;
1990         return 1;
1991 }
1992
1993 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1994 {
1995         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1996         return 0;
1997 }
1998
1999 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2000 {
2001         unsigned long exit_qualification;
2002         int size, down, in, string, rep;
2003         unsigned port;
2004
2005         ++vcpu->stat.io_exits;
2006         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2007         string = (exit_qualification & 16) != 0;
2008
2009         if (string) {
2010                 if (emulate_instruction(vcpu,
2011                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2012                         return 0;
2013                 return 1;
2014         }
2015
2016         size = (exit_qualification & 7) + 1;
2017         in = (exit_qualification & 8) != 0;
2018         down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
2019         rep = (exit_qualification & 32) != 0;
2020         port = exit_qualification >> 16;
2021
2022         return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2023 }
2024
2025 static void
2026 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2027 {
2028         /*
2029          * Patch in the VMCALL instruction:
2030          */
2031         hypercall[0] = 0x0f;
2032         hypercall[1] = 0x01;
2033         hypercall[2] = 0xc1;
2034 }
2035
2036 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2037 {
2038         unsigned long exit_qualification;
2039         int cr;
2040         int reg;
2041
2042         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2043         cr = exit_qualification & 15;
2044         reg = (exit_qualification >> 8) & 15;
2045         switch ((exit_qualification >> 4) & 3) {
2046         case 0: /* mov to cr */
2047                 switch (cr) {
2048                 case 0:
2049                         vcpu_load_rsp_rip(vcpu);
2050                         kvm_set_cr0(vcpu, vcpu->arch.regs[reg]);
2051                         skip_emulated_instruction(vcpu);
2052                         return 1;
2053                 case 3:
2054                         vcpu_load_rsp_rip(vcpu);
2055                         kvm_set_cr3(vcpu, vcpu->arch.regs[reg]);
2056                         skip_emulated_instruction(vcpu);
2057                         return 1;
2058                 case 4:
2059                         vcpu_load_rsp_rip(vcpu);
2060                         kvm_set_cr4(vcpu, vcpu->arch.regs[reg]);
2061                         skip_emulated_instruction(vcpu);
2062                         return 1;
2063                 case 8:
2064                         vcpu_load_rsp_rip(vcpu);
2065                         kvm_set_cr8(vcpu, vcpu->arch.regs[reg]);
2066                         skip_emulated_instruction(vcpu);
2067                         if (irqchip_in_kernel(vcpu->kvm))
2068                                 return 1;
2069                         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2070                         return 0;
2071                 };
2072                 break;
2073         case 2: /* clts */
2074                 vcpu_load_rsp_rip(vcpu);
2075                 vmx_fpu_deactivate(vcpu);
2076                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2077                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2078                 vmx_fpu_activate(vcpu);
2079                 skip_emulated_instruction(vcpu);
2080                 return 1;
2081         case 1: /*mov from cr*/
2082                 switch (cr) {
2083                 case 3:
2084                         vcpu_load_rsp_rip(vcpu);
2085                         vcpu->arch.regs[reg] = vcpu->arch.cr3;
2086                         vcpu_put_rsp_rip(vcpu);
2087                         skip_emulated_instruction(vcpu);
2088                         return 1;
2089                 case 8:
2090                         vcpu_load_rsp_rip(vcpu);
2091                         vcpu->arch.regs[reg] = kvm_get_cr8(vcpu);
2092                         vcpu_put_rsp_rip(vcpu);
2093                         skip_emulated_instruction(vcpu);
2094                         return 1;
2095                 }
2096                 break;
2097         case 3: /* lmsw */
2098                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2099
2100                 skip_emulated_instruction(vcpu);
2101                 return 1;
2102         default:
2103                 break;
2104         }
2105         kvm_run->exit_reason = 0;
2106         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2107                (int)(exit_qualification >> 4) & 3, cr);
2108         return 0;
2109 }
2110
2111 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2112 {
2113         unsigned long exit_qualification;
2114         unsigned long val;
2115         int dr, reg;
2116
2117         /*
2118          * FIXME: this code assumes the host is debugging the guest.
2119          *        need to deal with guest debugging itself too.
2120          */
2121         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2122         dr = exit_qualification & 7;
2123         reg = (exit_qualification >> 8) & 15;
2124         vcpu_load_rsp_rip(vcpu);
2125         if (exit_qualification & 16) {
2126                 /* mov from dr */
2127                 switch (dr) {
2128                 case 6:
2129                         val = 0xffff0ff0;
2130                         break;
2131                 case 7:
2132                         val = 0x400;
2133                         break;
2134                 default:
2135                         val = 0;
2136                 }
2137                 vcpu->arch.regs[reg] = val;
2138         } else {
2139                 /* mov to dr */
2140         }
2141         vcpu_put_rsp_rip(vcpu);
2142         skip_emulated_instruction(vcpu);
2143         return 1;
2144 }
2145
2146 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2147 {
2148         kvm_emulate_cpuid(vcpu);
2149         return 1;
2150 }
2151
2152 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2153 {
2154         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2155         u64 data;
2156
2157         if (vmx_get_msr(vcpu, ecx, &data)) {
2158                 kvm_inject_gp(vcpu, 0);
2159                 return 1;
2160         }
2161
2162         /* FIXME: handling of bits 32:63 of rax, rdx */
2163         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2164         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2165         skip_emulated_instruction(vcpu);
2166         return 1;
2167 }
2168
2169 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2170 {
2171         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2172         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2173                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2174
2175         if (vmx_set_msr(vcpu, ecx, data) != 0) {
2176                 kvm_inject_gp(vcpu, 0);
2177                 return 1;
2178         }
2179
2180         skip_emulated_instruction(vcpu);
2181         return 1;
2182 }
2183
2184 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2185                                       struct kvm_run *kvm_run)
2186 {
2187         return 1;
2188 }
2189
2190 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2191                                    struct kvm_run *kvm_run)
2192 {
2193         u32 cpu_based_vm_exec_control;
2194
2195         /* clear pending irq */
2196         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2197         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2198         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2199         /*
2200          * If the user space waits to inject interrupts, exit as soon as
2201          * possible
2202          */
2203         if (kvm_run->request_interrupt_window &&
2204             !vcpu->arch.irq_summary) {
2205                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2206                 ++vcpu->stat.irq_window_exits;
2207                 return 0;
2208         }
2209         return 1;
2210 }
2211
2212 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2213 {
2214         skip_emulated_instruction(vcpu);
2215         return kvm_emulate_halt(vcpu);
2216 }
2217
2218 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2219 {
2220         skip_emulated_instruction(vcpu);
2221         kvm_emulate_hypercall(vcpu);
2222         return 1;
2223 }
2224
2225 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2226 {
2227         skip_emulated_instruction(vcpu);
2228         /* TODO: Add support for VT-d/pass-through device */
2229         return 1;
2230 }
2231
2232 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2233 {
2234         u64 exit_qualification;
2235         enum emulation_result er;
2236         unsigned long offset;
2237
2238         exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2239         offset = exit_qualification & 0xffful;
2240
2241         er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2242
2243         if (er !=  EMULATE_DONE) {
2244                 printk(KERN_ERR
2245                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2246                        offset);
2247                 return -ENOTSUPP;
2248         }
2249         return 1;
2250 }
2251
2252 /*
2253  * The exit handlers return 1 if the exit was handled fully and guest execution
2254  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
2255  * to be done to userspace and return 0.
2256  */
2257 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2258                                       struct kvm_run *kvm_run) = {
2259         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
2260         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
2261         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
2262         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
2263         [EXIT_REASON_CR_ACCESS]               = handle_cr,
2264         [EXIT_REASON_DR_ACCESS]               = handle_dr,
2265         [EXIT_REASON_CPUID]                   = handle_cpuid,
2266         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
2267         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
2268         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
2269         [EXIT_REASON_HLT]                     = handle_halt,
2270         [EXIT_REASON_VMCALL]                  = handle_vmcall,
2271         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
2272         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
2273         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
2274 };
2275
2276 static const int kvm_vmx_max_exit_handlers =
2277         ARRAY_SIZE(kvm_vmx_exit_handlers);
2278
2279 /*
2280  * The guest has exited.  See if we can fix it or if we need userspace
2281  * assistance.
2282  */
2283 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2284 {
2285         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2286         struct vcpu_vmx *vmx = to_vmx(vcpu);
2287         u32 vectoring_info = vmx->idt_vectoring_info;
2288
2289         if (unlikely(vmx->fail)) {
2290                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2291                 kvm_run->fail_entry.hardware_entry_failure_reason
2292                         = vmcs_read32(VM_INSTRUCTION_ERROR);
2293                 return 0;
2294         }
2295
2296         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2297                                 exit_reason != EXIT_REASON_EXCEPTION_NMI)
2298                 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
2299                        "exit reason is 0x%x\n", __func__, exit_reason);
2300         if (exit_reason < kvm_vmx_max_exit_handlers
2301             && kvm_vmx_exit_handlers[exit_reason])
2302                 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2303         else {
2304                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2305                 kvm_run->hw.hardware_exit_reason = exit_reason;
2306         }
2307         return 0;
2308 }
2309
2310 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2311 {
2312         int max_irr, tpr;
2313
2314         if (!vm_need_tpr_shadow(vcpu->kvm))
2315                 return;
2316
2317         if (!kvm_lapic_enabled(vcpu) ||
2318             ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2319                 vmcs_write32(TPR_THRESHOLD, 0);
2320                 return;
2321         }
2322
2323         tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2324         vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2325 }
2326
2327 static void enable_irq_window(struct kvm_vcpu *vcpu)
2328 {
2329         u32 cpu_based_vm_exec_control;
2330
2331         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2332         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2333         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2334 }
2335
2336 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2337 {
2338         struct vcpu_vmx *vmx = to_vmx(vcpu);
2339         u32 idtv_info_field, intr_info_field;
2340         int has_ext_irq, interrupt_window_open;
2341         int vector;
2342
2343         update_tpr_threshold(vcpu);
2344
2345         has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2346         intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
2347         idtv_info_field = vmx->idt_vectoring_info;
2348         if (intr_info_field & INTR_INFO_VALID_MASK) {
2349                 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2350                         /* TODO: fault when IDT_Vectoring */
2351                         if (printk_ratelimit())
2352                                 printk(KERN_ERR "Fault when IDT_Vectoring\n");
2353                 }
2354                 if (has_ext_irq)
2355                         enable_irq_window(vcpu);
2356                 return;
2357         }
2358         if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2359                 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2360                     == INTR_TYPE_EXT_INTR
2361                     && vcpu->arch.rmode.active) {
2362                         u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2363
2364                         vmx_inject_irq(vcpu, vect);
2365                         if (unlikely(has_ext_irq))
2366                                 enable_irq_window(vcpu);
2367                         return;
2368                 }
2369
2370                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2371                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2372                                 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2373
2374                 if (unlikely(idtv_info_field & INTR_INFO_DELIVER_CODE_MASK))
2375                         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2376                                 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2377                 if (unlikely(has_ext_irq))
2378                         enable_irq_window(vcpu);
2379                 return;
2380         }
2381         if (!has_ext_irq)
2382                 return;
2383         interrupt_window_open =
2384                 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2385                  (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2386         if (interrupt_window_open) {
2387                 vector = kvm_cpu_get_interrupt(vcpu);
2388                 vmx_inject_irq(vcpu, vector);
2389                 kvm_timer_intr_post(vcpu, vector);
2390         } else
2391                 enable_irq_window(vcpu);
2392 }
2393
2394 /*
2395  * Failure to inject an interrupt should give us the information
2396  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
2397  * when fetching the interrupt redirection bitmap in the real-mode
2398  * tss, this doesn't happen.  So we do it ourselves.
2399  */
2400 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2401 {
2402         vmx->rmode.irq.pending = 0;
2403         if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip)
2404                 return;
2405         vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip);
2406         if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2407                 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2408                 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
2409                 return;
2410         }
2411         vmx->idt_vectoring_info =
2412                 VECTORING_INFO_VALID_MASK
2413                 | INTR_TYPE_EXT_INTR
2414                 | vmx->rmode.irq.vector;
2415 }
2416
2417 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2418 {
2419         struct vcpu_vmx *vmx = to_vmx(vcpu);
2420         u32 intr_info;
2421
2422         /*
2423          * Loading guest fpu may have cleared host cr0.ts
2424          */
2425         vmcs_writel(HOST_CR0, read_cr0());
2426
2427         asm(
2428                 /* Store host registers */
2429 #ifdef CONFIG_X86_64
2430                 "push %%rdx; push %%rbp;"
2431                 "push %%rcx \n\t"
2432 #else
2433                 "push %%edx; push %%ebp;"
2434                 "push %%ecx \n\t"
2435 #endif
2436                 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2437                 /* Check if vmlaunch of vmresume is needed */
2438                 "cmpl $0, %c[launched](%0) \n\t"
2439                 /* Load guest registers.  Don't clobber flags. */
2440 #ifdef CONFIG_X86_64
2441                 "mov %c[cr2](%0), %%rax \n\t"
2442                 "mov %%rax, %%cr2 \n\t"
2443                 "mov %c[rax](%0), %%rax \n\t"
2444                 "mov %c[rbx](%0), %%rbx \n\t"
2445                 "mov %c[rdx](%0), %%rdx \n\t"
2446                 "mov %c[rsi](%0), %%rsi \n\t"
2447                 "mov %c[rdi](%0), %%rdi \n\t"
2448                 "mov %c[rbp](%0), %%rbp \n\t"
2449                 "mov %c[r8](%0),  %%r8  \n\t"
2450                 "mov %c[r9](%0),  %%r9  \n\t"
2451                 "mov %c[r10](%0), %%r10 \n\t"
2452                 "mov %c[r11](%0), %%r11 \n\t"
2453                 "mov %c[r12](%0), %%r12 \n\t"
2454                 "mov %c[r13](%0), %%r13 \n\t"
2455                 "mov %c[r14](%0), %%r14 \n\t"
2456                 "mov %c[r15](%0), %%r15 \n\t"
2457                 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
2458 #else
2459                 "mov %c[cr2](%0), %%eax \n\t"
2460                 "mov %%eax,   %%cr2 \n\t"
2461                 "mov %c[rax](%0), %%eax \n\t"
2462                 "mov %c[rbx](%0), %%ebx \n\t"
2463                 "mov %c[rdx](%0), %%edx \n\t"
2464                 "mov %c[rsi](%0), %%esi \n\t"
2465                 "mov %c[rdi](%0), %%edi \n\t"
2466                 "mov %c[rbp](%0), %%ebp \n\t"
2467                 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
2468 #endif
2469                 /* Enter guest mode */
2470                 "jne .Llaunched \n\t"
2471                 ASM_VMX_VMLAUNCH "\n\t"
2472                 "jmp .Lkvm_vmx_return \n\t"
2473                 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2474                 ".Lkvm_vmx_return: "
2475                 /* Save guest registers, load host registers, keep flags */
2476 #ifdef CONFIG_X86_64
2477                 "xchg %0,     (%%rsp) \n\t"
2478                 "mov %%rax, %c[rax](%0) \n\t"
2479                 "mov %%rbx, %c[rbx](%0) \n\t"
2480                 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
2481                 "mov %%rdx, %c[rdx](%0) \n\t"
2482                 "mov %%rsi, %c[rsi](%0) \n\t"
2483                 "mov %%rdi, %c[rdi](%0) \n\t"
2484                 "mov %%rbp, %c[rbp](%0) \n\t"
2485                 "mov %%r8,  %c[r8](%0) \n\t"
2486                 "mov %%r9,  %c[r9](%0) \n\t"
2487                 "mov %%r10, %c[r10](%0) \n\t"
2488                 "mov %%r11, %c[r11](%0) \n\t"
2489                 "mov %%r12, %c[r12](%0) \n\t"
2490                 "mov %%r13, %c[r13](%0) \n\t"
2491                 "mov %%r14, %c[r14](%0) \n\t"
2492                 "mov %%r15, %c[r15](%0) \n\t"
2493                 "mov %%cr2, %%rax   \n\t"
2494                 "mov %%rax, %c[cr2](%0) \n\t"
2495
2496                 "pop  %%rbp; pop  %%rbp; pop  %%rdx \n\t"
2497 #else
2498                 "xchg %0, (%%esp) \n\t"
2499                 "mov %%eax, %c[rax](%0) \n\t"
2500                 "mov %%ebx, %c[rbx](%0) \n\t"
2501                 "pushl (%%esp); popl %c[rcx](%0) \n\t"
2502                 "mov %%edx, %c[rdx](%0) \n\t"
2503                 "mov %%esi, %c[rsi](%0) \n\t"
2504                 "mov %%edi, %c[rdi](%0) \n\t"
2505                 "mov %%ebp, %c[rbp](%0) \n\t"
2506                 "mov %%cr2, %%eax  \n\t"
2507                 "mov %%eax, %c[cr2](%0) \n\t"
2508
2509                 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
2510 #endif
2511                 "setbe %c[fail](%0) \n\t"
2512               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
2513                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
2514                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
2515                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
2516                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
2517                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
2518                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
2519                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
2520                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
2521                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
2522 #ifdef CONFIG_X86_64
2523                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
2524                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
2525                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
2526                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
2527                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
2528                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
2529                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
2530                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
2531 #endif
2532                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
2533               : "cc", "memory"
2534 #ifdef CONFIG_X86_64
2535                 , "rbx", "rdi", "rsi"
2536                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
2537 #else
2538                 , "ebx", "edi", "rsi"
2539 #endif
2540               );
2541
2542         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2543         if (vmx->rmode.irq.pending)
2544                 fixup_rmode_irq(vmx);
2545
2546         vcpu->arch.interrupt_window_open =
2547                 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2548
2549         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2550         vmx->launched = 1;
2551
2552         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2553
2554         /* We need to handle NMIs before interrupts are enabled */
2555         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2556                 asm("int $2");
2557 }
2558
2559 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2560 {
2561         struct vcpu_vmx *vmx = to_vmx(vcpu);
2562
2563         if (vmx->vmcs) {
2564                 on_each_cpu(__vcpu_clear, vmx, 0, 1);
2565                 free_vmcs(vmx->vmcs);
2566                 vmx->vmcs = NULL;
2567         }
2568 }
2569
2570 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2571 {
2572         struct vcpu_vmx *vmx = to_vmx(vcpu);
2573
2574         spin_lock(&vmx_vpid_lock);
2575         if (vmx->vpid != 0)
2576                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2577         spin_unlock(&vmx_vpid_lock);
2578         vmx_free_vmcs(vcpu);
2579         kfree(vmx->host_msrs);
2580         kfree(vmx->guest_msrs);
2581         kvm_vcpu_uninit(vcpu);
2582         kmem_cache_free(kvm_vcpu_cache, vmx);
2583 }
2584
2585 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2586 {
2587         int err;
2588         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2589         int cpu;
2590
2591         if (!vmx)
2592                 return ERR_PTR(-ENOMEM);
2593
2594         allocate_vpid(vmx);
2595
2596         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
2597         if (err)
2598                 goto free_vcpu;
2599
2600         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2601         if (!vmx->guest_msrs) {
2602                 err = -ENOMEM;
2603                 goto uninit_vcpu;
2604         }
2605
2606         vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2607         if (!vmx->host_msrs)
2608                 goto free_guest_msrs;
2609
2610         vmx->vmcs = alloc_vmcs();
2611         if (!vmx->vmcs)
2612                 goto free_msrs;
2613
2614         vmcs_clear(vmx->vmcs);
2615
2616         cpu = get_cpu();
2617         vmx_vcpu_load(&vmx->vcpu, cpu);
2618         err = vmx_vcpu_setup(vmx);
2619         vmx_vcpu_put(&vmx->vcpu);
2620         put_cpu();
2621         if (err)
2622                 goto free_vmcs;
2623         if (vm_need_virtualize_apic_accesses(kvm))
2624                 if (alloc_apic_access_page(kvm) != 0)
2625                         goto free_vmcs;
2626
2627         return &vmx->vcpu;
2628
2629 free_vmcs:
2630         free_vmcs(vmx->vmcs);
2631 free_msrs:
2632         kfree(vmx->host_msrs);
2633 free_guest_msrs:
2634         kfree(vmx->guest_msrs);
2635 uninit_vcpu:
2636         kvm_vcpu_uninit(&vmx->vcpu);
2637 free_vcpu:
2638         kmem_cache_free(kvm_vcpu_cache, vmx);
2639         return ERR_PTR(err);
2640 }
2641
2642 static void __init vmx_check_processor_compat(void *rtn)
2643 {
2644         struct vmcs_config vmcs_conf;
2645
2646         *(int *)rtn = 0;
2647         if (setup_vmcs_config(&vmcs_conf) < 0)
2648                 *(int *)rtn = -EIO;
2649         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
2650                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
2651                                 smp_processor_id());
2652                 *(int *)rtn = -EIO;
2653         }
2654 }
2655
2656 static struct kvm_x86_ops vmx_x86_ops = {
2657         .cpu_has_kvm_support = cpu_has_kvm_support,
2658         .disabled_by_bios = vmx_disabled_by_bios,
2659         .hardware_setup = hardware_setup,
2660         .hardware_unsetup = hardware_unsetup,
2661         .check_processor_compatibility = vmx_check_processor_compat,
2662         .hardware_enable = hardware_enable,
2663         .hardware_disable = hardware_disable,
2664         .cpu_has_accelerated_tpr = cpu_has_vmx_virtualize_apic_accesses,
2665
2666         .vcpu_create = vmx_create_vcpu,
2667         .vcpu_free = vmx_free_vcpu,
2668         .vcpu_reset = vmx_vcpu_reset,
2669
2670         .prepare_guest_switch = vmx_save_host_state,
2671         .vcpu_load = vmx_vcpu_load,
2672         .vcpu_put = vmx_vcpu_put,
2673         .vcpu_decache = vmx_vcpu_decache,
2674
2675         .set_guest_debug = set_guest_debug,
2676         .guest_debug_pre = kvm_guest_debug_pre,
2677         .get_msr = vmx_get_msr,
2678         .set_msr = vmx_set_msr,
2679         .get_segment_base = vmx_get_segment_base,
2680         .get_segment = vmx_get_segment,
2681         .set_segment = vmx_set_segment,
2682         .get_cpl = vmx_get_cpl,
2683         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2684         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2685         .set_cr0 = vmx_set_cr0,
2686         .set_cr3 = vmx_set_cr3,
2687         .set_cr4 = vmx_set_cr4,
2688         .set_efer = vmx_set_efer,
2689         .get_idt = vmx_get_idt,
2690         .set_idt = vmx_set_idt,
2691         .get_gdt = vmx_get_gdt,
2692         .set_gdt = vmx_set_gdt,
2693         .cache_regs = vcpu_load_rsp_rip,
2694         .decache_regs = vcpu_put_rsp_rip,
2695         .get_rflags = vmx_get_rflags,
2696         .set_rflags = vmx_set_rflags,
2697
2698         .tlb_flush = vmx_flush_tlb,
2699
2700         .run = vmx_vcpu_run,
2701         .handle_exit = kvm_handle_exit,
2702         .skip_emulated_instruction = skip_emulated_instruction,
2703         .patch_hypercall = vmx_patch_hypercall,
2704         .get_irq = vmx_get_irq,
2705         .set_irq = vmx_inject_irq,
2706         .queue_exception = vmx_queue_exception,
2707         .exception_injected = vmx_exception_injected,
2708         .inject_pending_irq = vmx_intr_assist,
2709         .inject_pending_vectors = do_interrupt_requests,
2710
2711         .set_tss_addr = vmx_set_tss_addr,
2712 };
2713
2714 static int __init vmx_init(void)
2715 {
2716         void *iova;
2717         int r;
2718
2719         vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2720         if (!vmx_io_bitmap_a)
2721                 return -ENOMEM;
2722
2723         vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2724         if (!vmx_io_bitmap_b) {
2725                 r = -ENOMEM;
2726                 goto out;
2727         }
2728
2729         /*
2730          * Allow direct access to the PC debug port (it is often used for I/O
2731          * delays, but the vmexits simply slow things down).
2732          */
2733         iova = kmap(vmx_io_bitmap_a);
2734         memset(iova, 0xff, PAGE_SIZE);
2735         clear_bit(0x80, iova);
2736         kunmap(vmx_io_bitmap_a);
2737
2738         iova = kmap(vmx_io_bitmap_b);
2739         memset(iova, 0xff, PAGE_SIZE);
2740         kunmap(vmx_io_bitmap_b);
2741
2742         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
2743
2744         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
2745         if (r)
2746                 goto out1;
2747
2748         if (bypass_guest_pf)
2749                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
2750
2751         return 0;
2752
2753 out1:
2754         __free_page(vmx_io_bitmap_b);
2755 out:
2756         __free_page(vmx_io_bitmap_a);
2757         return r;
2758 }
2759
2760 static void __exit vmx_exit(void)
2761 {
2762         __free_page(vmx_io_bitmap_b);
2763         __free_page(vmx_io_bitmap_a);
2764
2765         kvm_exit();
2766 }
2767
2768 module_init(vmx_init)
2769 module_exit(vmx_exit)