]> err.no Git - linux-2.6/blob - drivers/kvm/svm.c
KVM: Move apic timer interrupt backlog processing to common code
[linux-2.6] / drivers / kvm / svm.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * AMD SVM support
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  *
8  * Authors:
9  *   Yaniv Kamay  <yaniv@qumranet.com>
10  *   Avi Kivity   <avi@qumranet.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2.  See
13  * the COPYING file in the top-level directory.
14  *
15  */
16
17 #include "kvm_svm.h"
18 #include "x86_emulate.h"
19 #include "irq.h"
20
21 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/vmalloc.h>
24 #include <linux/highmem.h>
25 #include <linux/sched.h>
26
27 #include <asm/desc.h>
28
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
31
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
34
35 #define DB_VECTOR 1
36 #define UD_VECTOR 6
37 #define GP_VECTOR 13
38
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
41
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
44
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
47
48 #define SVM_FEATURE_NPT  (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
51
52 static void kvm_reput_irq(struct vcpu_svm *svm);
53
54 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
55 {
56         return container_of(vcpu, struct vcpu_svm, vcpu);
57 }
58
59 unsigned long iopm_base;
60 unsigned long msrpm_base;
61
62 struct kvm_ldttss_desc {
63         u16 limit0;
64         u16 base0;
65         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
66         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
67         u32 base3;
68         u32 zero1;
69 } __attribute__((packed));
70
71 struct svm_cpu_data {
72         int cpu;
73
74         u64 asid_generation;
75         u32 max_asid;
76         u32 next_asid;
77         struct kvm_ldttss_desc *tss_desc;
78
79         struct page *save_area;
80 };
81
82 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
83 static uint32_t svm_features;
84
85 struct svm_init_data {
86         int cpu;
87         int r;
88 };
89
90 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
91
92 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
93 #define MSRS_RANGE_SIZE 2048
94 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
95
96 #define MAX_INST_SIZE 15
97
98 static inline u32 svm_has(u32 feat)
99 {
100         return svm_features & feat;
101 }
102
103 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
104 {
105         int word_index = __ffs(vcpu->irq_summary);
106         int bit_index = __ffs(vcpu->irq_pending[word_index]);
107         int irq = word_index * BITS_PER_LONG + bit_index;
108
109         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
110         if (!vcpu->irq_pending[word_index])
111                 clear_bit(word_index, &vcpu->irq_summary);
112         return irq;
113 }
114
115 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
116 {
117         set_bit(irq, vcpu->irq_pending);
118         set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
119 }
120
121 static inline void clgi(void)
122 {
123         asm volatile (SVM_CLGI);
124 }
125
126 static inline void stgi(void)
127 {
128         asm volatile (SVM_STGI);
129 }
130
131 static inline void invlpga(unsigned long addr, u32 asid)
132 {
133         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
134 }
135
136 static inline unsigned long kvm_read_cr2(void)
137 {
138         unsigned long cr2;
139
140         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
141         return cr2;
142 }
143
144 static inline void kvm_write_cr2(unsigned long val)
145 {
146         asm volatile ("mov %0, %%cr2" :: "r" (val));
147 }
148
149 static inline unsigned long read_dr6(void)
150 {
151         unsigned long dr6;
152
153         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
154         return dr6;
155 }
156
157 static inline void write_dr6(unsigned long val)
158 {
159         asm volatile ("mov %0, %%dr6" :: "r" (val));
160 }
161
162 static inline unsigned long read_dr7(void)
163 {
164         unsigned long dr7;
165
166         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
167         return dr7;
168 }
169
170 static inline void write_dr7(unsigned long val)
171 {
172         asm volatile ("mov %0, %%dr7" :: "r" (val));
173 }
174
175 static inline void force_new_asid(struct kvm_vcpu *vcpu)
176 {
177         to_svm(vcpu)->asid_generation--;
178 }
179
180 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
181 {
182         force_new_asid(vcpu);
183 }
184
185 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
186 {
187         if (!(efer & KVM_EFER_LMA))
188                 efer &= ~KVM_EFER_LME;
189
190         to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
191         vcpu->shadow_efer = efer;
192 }
193
194 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
195 {
196         struct vcpu_svm *svm = to_svm(vcpu);
197
198         svm->vmcb->control.event_inj =          SVM_EVTINJ_VALID |
199                                                 SVM_EVTINJ_VALID_ERR |
200                                                 SVM_EVTINJ_TYPE_EXEPT |
201                                                 GP_VECTOR;
202         svm->vmcb->control.event_inj_err = error_code;
203 }
204
205 static void inject_ud(struct kvm_vcpu *vcpu)
206 {
207         to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
208                                                 SVM_EVTINJ_TYPE_EXEPT |
209                                                 UD_VECTOR;
210 }
211
212 static int is_page_fault(uint32_t info)
213 {
214         info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
215         return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
216 }
217
218 static int is_external_interrupt(u32 info)
219 {
220         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
221         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
222 }
223
224 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
225 {
226         struct vcpu_svm *svm = to_svm(vcpu);
227
228         if (!svm->next_rip) {
229                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
230                 return;
231         }
232         if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
233                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
234                        __FUNCTION__,
235                        svm->vmcb->save.rip,
236                        svm->next_rip);
237
238         vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
239         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
240
241         vcpu->interrupt_window_open = 1;
242 }
243
244 static int has_svm(void)
245 {
246         uint32_t eax, ebx, ecx, edx;
247
248         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
249                 printk(KERN_INFO "has_svm: not amd\n");
250                 return 0;
251         }
252
253         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
254         if (eax < SVM_CPUID_FUNC) {
255                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
256                 return 0;
257         }
258
259         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
260         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
261                 printk(KERN_DEBUG "has_svm: svm not available\n");
262                 return 0;
263         }
264         return 1;
265 }
266
267 static void svm_hardware_disable(void *garbage)
268 {
269         struct svm_cpu_data *svm_data
270                 = per_cpu(svm_data, raw_smp_processor_id());
271
272         if (svm_data) {
273                 uint64_t efer;
274
275                 wrmsrl(MSR_VM_HSAVE_PA, 0);
276                 rdmsrl(MSR_EFER, efer);
277                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
278                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
279                 __free_page(svm_data->save_area);
280                 kfree(svm_data);
281         }
282 }
283
284 static void svm_hardware_enable(void *garbage)
285 {
286
287         struct svm_cpu_data *svm_data;
288         uint64_t efer;
289 #ifdef CONFIG_X86_64
290         struct desc_ptr gdt_descr;
291 #else
292         struct desc_ptr gdt_descr;
293 #endif
294         struct desc_struct *gdt;
295         int me = raw_smp_processor_id();
296
297         if (!has_svm()) {
298                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
299                 return;
300         }
301         svm_data = per_cpu(svm_data, me);
302
303         if (!svm_data) {
304                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
305                        me);
306                 return;
307         }
308
309         svm_data->asid_generation = 1;
310         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
311         svm_data->next_asid = svm_data->max_asid + 1;
312         svm_features = cpuid_edx(SVM_CPUID_FUNC);
313
314         asm volatile ("sgdt %0" : "=m"(gdt_descr));
315         gdt = (struct desc_struct *)gdt_descr.address;
316         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
317
318         rdmsrl(MSR_EFER, efer);
319         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
320
321         wrmsrl(MSR_VM_HSAVE_PA,
322                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
323 }
324
325 static int svm_cpu_init(int cpu)
326 {
327         struct svm_cpu_data *svm_data;
328         int r;
329
330         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
331         if (!svm_data)
332                 return -ENOMEM;
333         svm_data->cpu = cpu;
334         svm_data->save_area = alloc_page(GFP_KERNEL);
335         r = -ENOMEM;
336         if (!svm_data->save_area)
337                 goto err_1;
338
339         per_cpu(svm_data, cpu) = svm_data;
340
341         return 0;
342
343 err_1:
344         kfree(svm_data);
345         return r;
346
347 }
348
349 static void set_msr_interception(u32 *msrpm, unsigned msr,
350                                  int read, int write)
351 {
352         int i;
353
354         for (i = 0; i < NUM_MSR_MAPS; i++) {
355                 if (msr >= msrpm_ranges[i] &&
356                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
357                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
358                                           msrpm_ranges[i]) * 2;
359
360                         u32 *base = msrpm + (msr_offset / 32);
361                         u32 msr_shift = msr_offset % 32;
362                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
363                         *base = (*base & ~(0x3 << msr_shift)) |
364                                 (mask << msr_shift);
365                         return;
366                 }
367         }
368         BUG();
369 }
370
371 static __init int svm_hardware_setup(void)
372 {
373         int cpu;
374         struct page *iopm_pages;
375         struct page *msrpm_pages;
376         void *iopm_va, *msrpm_va;
377         int r;
378
379         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
380
381         if (!iopm_pages)
382                 return -ENOMEM;
383
384         iopm_va = page_address(iopm_pages);
385         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
386         clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
387         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
388
389
390         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
391
392         r = -ENOMEM;
393         if (!msrpm_pages)
394                 goto err_1;
395
396         msrpm_va = page_address(msrpm_pages);
397         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
398         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
399
400 #ifdef CONFIG_X86_64
401         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
402         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
403         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
404         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
405         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
406         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
407 #endif
408         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
409         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
410         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
411         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
412
413         for_each_online_cpu(cpu) {
414                 r = svm_cpu_init(cpu);
415                 if (r)
416                         goto err_2;
417         }
418         return 0;
419
420 err_2:
421         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
422         msrpm_base = 0;
423 err_1:
424         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
425         iopm_base = 0;
426         return r;
427 }
428
429 static __exit void svm_hardware_unsetup(void)
430 {
431         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
432         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
433         iopm_base = msrpm_base = 0;
434 }
435
436 static void init_seg(struct vmcb_seg *seg)
437 {
438         seg->selector = 0;
439         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
440                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
441         seg->limit = 0xffff;
442         seg->base = 0;
443 }
444
445 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
446 {
447         seg->selector = 0;
448         seg->attrib = SVM_SELECTOR_P_MASK | type;
449         seg->limit = 0xffff;
450         seg->base = 0;
451 }
452
453 static void init_vmcb(struct vmcb *vmcb)
454 {
455         struct vmcb_control_area *control = &vmcb->control;
456         struct vmcb_save_area *save = &vmcb->save;
457
458         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
459                                         INTERCEPT_CR3_MASK |
460                                         INTERCEPT_CR4_MASK;
461
462         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
463                                         INTERCEPT_CR3_MASK |
464                                         INTERCEPT_CR4_MASK;
465
466         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
467                                         INTERCEPT_DR1_MASK |
468                                         INTERCEPT_DR2_MASK |
469                                         INTERCEPT_DR3_MASK;
470
471         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
472                                         INTERCEPT_DR1_MASK |
473                                         INTERCEPT_DR2_MASK |
474                                         INTERCEPT_DR3_MASK |
475                                         INTERCEPT_DR5_MASK |
476                                         INTERCEPT_DR7_MASK;
477
478         control->intercept_exceptions = (1 << PF_VECTOR) |
479                                         (1 << UD_VECTOR);
480
481
482         control->intercept =    (1ULL << INTERCEPT_INTR) |
483                                 (1ULL << INTERCEPT_NMI) |
484                                 (1ULL << INTERCEPT_SMI) |
485                 /*
486                  * selective cr0 intercept bug?
487                  *      0:   0f 22 d8                mov    %eax,%cr3
488                  *      3:   0f 20 c0                mov    %cr0,%eax
489                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
490                  *      b:   0f 22 c0                mov    %eax,%cr0
491                  * set cr3 ->interception
492                  * get cr0 ->interception
493                  * set cr0 -> no interception
494                  */
495                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
496                                 (1ULL << INTERCEPT_CPUID) |
497                                 (1ULL << INTERCEPT_INVD) |
498                                 (1ULL << INTERCEPT_HLT) |
499                                 (1ULL << INTERCEPT_INVLPGA) |
500                                 (1ULL << INTERCEPT_IOIO_PROT) |
501                                 (1ULL << INTERCEPT_MSR_PROT) |
502                                 (1ULL << INTERCEPT_TASK_SWITCH) |
503                                 (1ULL << INTERCEPT_SHUTDOWN) |
504                                 (1ULL << INTERCEPT_VMRUN) |
505                                 (1ULL << INTERCEPT_VMMCALL) |
506                                 (1ULL << INTERCEPT_VMLOAD) |
507                                 (1ULL << INTERCEPT_VMSAVE) |
508                                 (1ULL << INTERCEPT_STGI) |
509                                 (1ULL << INTERCEPT_CLGI) |
510                                 (1ULL << INTERCEPT_SKINIT) |
511                                 (1ULL << INTERCEPT_WBINVD) |
512                                 (1ULL << INTERCEPT_MONITOR) |
513                                 (1ULL << INTERCEPT_MWAIT);
514
515         control->iopm_base_pa = iopm_base;
516         control->msrpm_base_pa = msrpm_base;
517         control->tsc_offset = 0;
518         control->int_ctl = V_INTR_MASKING_MASK;
519
520         init_seg(&save->es);
521         init_seg(&save->ss);
522         init_seg(&save->ds);
523         init_seg(&save->fs);
524         init_seg(&save->gs);
525
526         save->cs.selector = 0xf000;
527         /* Executable/Readable Code Segment */
528         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
529                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
530         save->cs.limit = 0xffff;
531         /*
532          * cs.base should really be 0xffff0000, but vmx can't handle that, so
533          * be consistent with it.
534          *
535          * Replace when we have real mode working for vmx.
536          */
537         save->cs.base = 0xf0000;
538
539         save->gdtr.limit = 0xffff;
540         save->idtr.limit = 0xffff;
541
542         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
543         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
544
545         save->efer = MSR_EFER_SVME_MASK;
546         save->dr6 = 0xffff0ff0;
547         save->dr7 = 0x400;
548         save->rflags = 2;
549         save->rip = 0x0000fff0;
550
551         /*
552          * cr0 val on cpu init should be 0x60000010, we enable cpu
553          * cache by default. the orderly way is to enable cache in bios.
554          */
555         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
556         save->cr4 = X86_CR4_PAE;
557         /* rdx = ?? */
558 }
559
560 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
561 {
562         struct vcpu_svm *svm = to_svm(vcpu);
563
564         init_vmcb(svm->vmcb);
565
566         if (vcpu->vcpu_id != 0) {
567                 svm->vmcb->save.rip = 0;
568                 svm->vmcb->save.cs.base = svm->vcpu.sipi_vector << 12;
569                 svm->vmcb->save.cs.selector = svm->vcpu.sipi_vector << 8;
570         }
571 }
572
573 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
574 {
575         struct vcpu_svm *svm;
576         struct page *page;
577         int err;
578
579         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
580         if (!svm) {
581                 err = -ENOMEM;
582                 goto out;
583         }
584
585         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
586         if (err)
587                 goto free_svm;
588
589         page = alloc_page(GFP_KERNEL);
590         if (!page) {
591                 err = -ENOMEM;
592                 goto uninit;
593         }
594
595         svm->vmcb = page_address(page);
596         clear_page(svm->vmcb);
597         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
598         svm->asid_generation = 0;
599         memset(svm->db_regs, 0, sizeof(svm->db_regs));
600         init_vmcb(svm->vmcb);
601
602         fx_init(&svm->vcpu);
603         svm->vcpu.fpu_active = 1;
604         svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
605         if (svm->vcpu.vcpu_id == 0)
606                 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
607
608         return &svm->vcpu;
609
610 uninit:
611         kvm_vcpu_uninit(&svm->vcpu);
612 free_svm:
613         kmem_cache_free(kvm_vcpu_cache, svm);
614 out:
615         return ERR_PTR(err);
616 }
617
618 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
619 {
620         struct vcpu_svm *svm = to_svm(vcpu);
621
622         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
623         kvm_vcpu_uninit(vcpu);
624         kmem_cache_free(kvm_vcpu_cache, svm);
625 }
626
627 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
628 {
629         struct vcpu_svm *svm = to_svm(vcpu);
630         int i;
631
632         if (unlikely(cpu != vcpu->cpu)) {
633                 u64 tsc_this, delta;
634
635                 /*
636                  * Make sure that the guest sees a monotonically
637                  * increasing TSC.
638                  */
639                 rdtscll(tsc_this);
640                 delta = vcpu->host_tsc - tsc_this;
641                 svm->vmcb->control.tsc_offset += delta;
642                 vcpu->cpu = cpu;
643                 kvm_migrate_apic_timer(vcpu);
644         }
645
646         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
647                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
648 }
649
650 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
651 {
652         struct vcpu_svm *svm = to_svm(vcpu);
653         int i;
654
655         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
656                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
657
658         rdtscll(vcpu->host_tsc);
659         kvm_put_guest_fpu(vcpu);
660 }
661
662 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
663 {
664 }
665
666 static void svm_cache_regs(struct kvm_vcpu *vcpu)
667 {
668         struct vcpu_svm *svm = to_svm(vcpu);
669
670         vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
671         vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
672         vcpu->rip = svm->vmcb->save.rip;
673 }
674
675 static void svm_decache_regs(struct kvm_vcpu *vcpu)
676 {
677         struct vcpu_svm *svm = to_svm(vcpu);
678         svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
679         svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
680         svm->vmcb->save.rip = vcpu->rip;
681 }
682
683 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
684 {
685         return to_svm(vcpu)->vmcb->save.rflags;
686 }
687
688 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
689 {
690         to_svm(vcpu)->vmcb->save.rflags = rflags;
691 }
692
693 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
694 {
695         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
696
697         switch (seg) {
698         case VCPU_SREG_CS: return &save->cs;
699         case VCPU_SREG_DS: return &save->ds;
700         case VCPU_SREG_ES: return &save->es;
701         case VCPU_SREG_FS: return &save->fs;
702         case VCPU_SREG_GS: return &save->gs;
703         case VCPU_SREG_SS: return &save->ss;
704         case VCPU_SREG_TR: return &save->tr;
705         case VCPU_SREG_LDTR: return &save->ldtr;
706         }
707         BUG();
708         return NULL;
709 }
710
711 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
712 {
713         struct vmcb_seg *s = svm_seg(vcpu, seg);
714
715         return s->base;
716 }
717
718 static void svm_get_segment(struct kvm_vcpu *vcpu,
719                             struct kvm_segment *var, int seg)
720 {
721         struct vmcb_seg *s = svm_seg(vcpu, seg);
722
723         var->base = s->base;
724         var->limit = s->limit;
725         var->selector = s->selector;
726         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
727         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
728         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
729         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
730         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
731         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
732         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
733         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
734         var->unusable = !var->present;
735 }
736
737 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
738 {
739         struct vcpu_svm *svm = to_svm(vcpu);
740
741         dt->limit = svm->vmcb->save.idtr.limit;
742         dt->base = svm->vmcb->save.idtr.base;
743 }
744
745 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
746 {
747         struct vcpu_svm *svm = to_svm(vcpu);
748
749         svm->vmcb->save.idtr.limit = dt->limit;
750         svm->vmcb->save.idtr.base = dt->base ;
751 }
752
753 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
754 {
755         struct vcpu_svm *svm = to_svm(vcpu);
756
757         dt->limit = svm->vmcb->save.gdtr.limit;
758         dt->base = svm->vmcb->save.gdtr.base;
759 }
760
761 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
762 {
763         struct vcpu_svm *svm = to_svm(vcpu);
764
765         svm->vmcb->save.gdtr.limit = dt->limit;
766         svm->vmcb->save.gdtr.base = dt->base ;
767 }
768
769 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
770 {
771 }
772
773 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
774 {
775         struct vcpu_svm *svm = to_svm(vcpu);
776
777 #ifdef CONFIG_X86_64
778         if (vcpu->shadow_efer & KVM_EFER_LME) {
779                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
780                         vcpu->shadow_efer |= KVM_EFER_LMA;
781                         svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
782                 }
783
784                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
785                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
786                         svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
787                 }
788         }
789 #endif
790         if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
791                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
792                 vcpu->fpu_active = 1;
793         }
794
795         vcpu->cr0 = cr0;
796         cr0 |= X86_CR0_PG | X86_CR0_WP;
797         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
798         svm->vmcb->save.cr0 = cr0;
799 }
800
801 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
802 {
803        vcpu->cr4 = cr4;
804        to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
805 }
806
807 static void svm_set_segment(struct kvm_vcpu *vcpu,
808                             struct kvm_segment *var, int seg)
809 {
810         struct vcpu_svm *svm = to_svm(vcpu);
811         struct vmcb_seg *s = svm_seg(vcpu, seg);
812
813         s->base = var->base;
814         s->limit = var->limit;
815         s->selector = var->selector;
816         if (var->unusable)
817                 s->attrib = 0;
818         else {
819                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
820                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
821                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
822                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
823                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
824                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
825                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
826                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
827         }
828         if (seg == VCPU_SREG_CS)
829                 svm->vmcb->save.cpl
830                         = (svm->vmcb->save.cs.attrib
831                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
832
833 }
834
835 /* FIXME:
836
837         svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
838         svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
839
840 */
841
842 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
843 {
844         return -EOPNOTSUPP;
845 }
846
847 static int svm_get_irq(struct kvm_vcpu *vcpu)
848 {
849         struct vcpu_svm *svm = to_svm(vcpu);
850         u32 exit_int_info = svm->vmcb->control.exit_int_info;
851
852         if (is_external_interrupt(exit_int_info))
853                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
854         return -1;
855 }
856
857 static void load_host_msrs(struct kvm_vcpu *vcpu)
858 {
859 #ifdef CONFIG_X86_64
860         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
861 #endif
862 }
863
864 static void save_host_msrs(struct kvm_vcpu *vcpu)
865 {
866 #ifdef CONFIG_X86_64
867         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
868 #endif
869 }
870
871 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
872 {
873         if (svm_data->next_asid > svm_data->max_asid) {
874                 ++svm_data->asid_generation;
875                 svm_data->next_asid = 1;
876                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
877         }
878
879         svm->vcpu.cpu = svm_data->cpu;
880         svm->asid_generation = svm_data->asid_generation;
881         svm->vmcb->control.asid = svm_data->next_asid++;
882 }
883
884 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
885 {
886         return to_svm(vcpu)->db_regs[dr];
887 }
888
889 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
890                        int *exception)
891 {
892         struct vcpu_svm *svm = to_svm(vcpu);
893
894         *exception = 0;
895
896         if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
897                 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
898                 svm->vmcb->save.dr6 |= DR6_BD_MASK;
899                 *exception = DB_VECTOR;
900                 return;
901         }
902
903         switch (dr) {
904         case 0 ... 3:
905                 svm->db_regs[dr] = value;
906                 return;
907         case 4 ... 5:
908                 if (vcpu->cr4 & X86_CR4_DE) {
909                         *exception = UD_VECTOR;
910                         return;
911                 }
912         case 7: {
913                 if (value & ~((1ULL << 32) - 1)) {
914                         *exception = GP_VECTOR;
915                         return;
916                 }
917                 svm->vmcb->save.dr7 = value;
918                 return;
919         }
920         default:
921                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
922                        __FUNCTION__, dr);
923                 *exception = UD_VECTOR;
924                 return;
925         }
926 }
927
928 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
929 {
930         u32 exit_int_info = svm->vmcb->control.exit_int_info;
931         struct kvm *kvm = svm->vcpu.kvm;
932         u64 fault_address;
933         u32 error_code;
934         enum emulation_result er;
935         int r;
936
937         if (!irqchip_in_kernel(kvm) &&
938                 is_external_interrupt(exit_int_info))
939                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
940
941         mutex_lock(&kvm->lock);
942
943         fault_address  = svm->vmcb->control.exit_info_2;
944         error_code = svm->vmcb->control.exit_info_1;
945         r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
946         if (r < 0) {
947                 mutex_unlock(&kvm->lock);
948                 return r;
949         }
950         if (!r) {
951                 mutex_unlock(&kvm->lock);
952                 return 1;
953         }
954         er = emulate_instruction(&svm->vcpu, kvm_run, fault_address,
955                                  error_code, 0);
956         mutex_unlock(&kvm->lock);
957
958         switch (er) {
959         case EMULATE_DONE:
960                 return 1;
961         case EMULATE_DO_MMIO:
962                 ++svm->vcpu.stat.mmio_exits;
963                 return 0;
964         case EMULATE_FAIL:
965                 kvm_report_emulation_failure(&svm->vcpu, "pagetable");
966                 break;
967         default:
968                 BUG();
969         }
970
971         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
972         return 0;
973 }
974
975 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
976 {
977         int er;
978
979         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0);
980         if (er != EMULATE_DONE)
981                 inject_ud(&svm->vcpu);
982
983         return 1;
984 }
985
986 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
987 {
988         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
989         if (!(svm->vcpu.cr0 & X86_CR0_TS))
990                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
991         svm->vcpu.fpu_active = 1;
992
993         return 1;
994 }
995
996 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
997 {
998         /*
999          * VMCB is undefined after a SHUTDOWN intercept
1000          * so reinitialize it.
1001          */
1002         clear_page(svm->vmcb);
1003         init_vmcb(svm->vmcb);
1004
1005         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1006         return 0;
1007 }
1008
1009 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1010 {
1011         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1012         int size, down, in, string, rep;
1013         unsigned port;
1014
1015         ++svm->vcpu.stat.io_exits;
1016
1017         svm->next_rip = svm->vmcb->control.exit_info_2;
1018
1019         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1020
1021         if (string) {
1022                 if (emulate_instruction(&svm->vcpu,
1023                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1024                         return 0;
1025                 return 1;
1026         }
1027
1028         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1029         port = io_info >> 16;
1030         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1031         rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1032         down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1033
1034         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1035 }
1036
1037 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1038 {
1039         return 1;
1040 }
1041
1042 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1043 {
1044         svm->next_rip = svm->vmcb->save.rip + 1;
1045         skip_emulated_instruction(&svm->vcpu);
1046         return kvm_emulate_halt(&svm->vcpu);
1047 }
1048
1049 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1050 {
1051         svm->next_rip = svm->vmcb->save.rip + 3;
1052         skip_emulated_instruction(&svm->vcpu);
1053         kvm_emulate_hypercall(&svm->vcpu);
1054         return 1;
1055 }
1056
1057 static int invalid_op_interception(struct vcpu_svm *svm,
1058                                    struct kvm_run *kvm_run)
1059 {
1060         inject_ud(&svm->vcpu);
1061         return 1;
1062 }
1063
1064 static int task_switch_interception(struct vcpu_svm *svm,
1065                                     struct kvm_run *kvm_run)
1066 {
1067         pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1068         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1069         return 0;
1070 }
1071
1072 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1073 {
1074         svm->next_rip = svm->vmcb->save.rip + 2;
1075         kvm_emulate_cpuid(&svm->vcpu);
1076         return 1;
1077 }
1078
1079 static int emulate_on_interception(struct vcpu_svm *svm,
1080                                    struct kvm_run *kvm_run)
1081 {
1082         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1083                 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1084         return 1;
1085 }
1086
1087 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1088 {
1089         struct vcpu_svm *svm = to_svm(vcpu);
1090
1091         switch (ecx) {
1092         case MSR_IA32_TIME_STAMP_COUNTER: {
1093                 u64 tsc;
1094
1095                 rdtscll(tsc);
1096                 *data = svm->vmcb->control.tsc_offset + tsc;
1097                 break;
1098         }
1099         case MSR_K6_STAR:
1100                 *data = svm->vmcb->save.star;
1101                 break;
1102 #ifdef CONFIG_X86_64
1103         case MSR_LSTAR:
1104                 *data = svm->vmcb->save.lstar;
1105                 break;
1106         case MSR_CSTAR:
1107                 *data = svm->vmcb->save.cstar;
1108                 break;
1109         case MSR_KERNEL_GS_BASE:
1110                 *data = svm->vmcb->save.kernel_gs_base;
1111                 break;
1112         case MSR_SYSCALL_MASK:
1113                 *data = svm->vmcb->save.sfmask;
1114                 break;
1115 #endif
1116         case MSR_IA32_SYSENTER_CS:
1117                 *data = svm->vmcb->save.sysenter_cs;
1118                 break;
1119         case MSR_IA32_SYSENTER_EIP:
1120                 *data = svm->vmcb->save.sysenter_eip;
1121                 break;
1122         case MSR_IA32_SYSENTER_ESP:
1123                 *data = svm->vmcb->save.sysenter_esp;
1124                 break;
1125         default:
1126                 return kvm_get_msr_common(vcpu, ecx, data);
1127         }
1128         return 0;
1129 }
1130
1131 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1132 {
1133         u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1134         u64 data;
1135
1136         if (svm_get_msr(&svm->vcpu, ecx, &data))
1137                 svm_inject_gp(&svm->vcpu, 0);
1138         else {
1139                 svm->vmcb->save.rax = data & 0xffffffff;
1140                 svm->vcpu.regs[VCPU_REGS_RDX] = data >> 32;
1141                 svm->next_rip = svm->vmcb->save.rip + 2;
1142                 skip_emulated_instruction(&svm->vcpu);
1143         }
1144         return 1;
1145 }
1146
1147 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1148 {
1149         struct vcpu_svm *svm = to_svm(vcpu);
1150
1151         switch (ecx) {
1152         case MSR_IA32_TIME_STAMP_COUNTER: {
1153                 u64 tsc;
1154
1155                 rdtscll(tsc);
1156                 svm->vmcb->control.tsc_offset = data - tsc;
1157                 break;
1158         }
1159         case MSR_K6_STAR:
1160                 svm->vmcb->save.star = data;
1161                 break;
1162 #ifdef CONFIG_X86_64
1163         case MSR_LSTAR:
1164                 svm->vmcb->save.lstar = data;
1165                 break;
1166         case MSR_CSTAR:
1167                 svm->vmcb->save.cstar = data;
1168                 break;
1169         case MSR_KERNEL_GS_BASE:
1170                 svm->vmcb->save.kernel_gs_base = data;
1171                 break;
1172         case MSR_SYSCALL_MASK:
1173                 svm->vmcb->save.sfmask = data;
1174                 break;
1175 #endif
1176         case MSR_IA32_SYSENTER_CS:
1177                 svm->vmcb->save.sysenter_cs = data;
1178                 break;
1179         case MSR_IA32_SYSENTER_EIP:
1180                 svm->vmcb->save.sysenter_eip = data;
1181                 break;
1182         case MSR_IA32_SYSENTER_ESP:
1183                 svm->vmcb->save.sysenter_esp = data;
1184                 break;
1185         default:
1186                 return kvm_set_msr_common(vcpu, ecx, data);
1187         }
1188         return 0;
1189 }
1190
1191 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1192 {
1193         u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1194         u64 data = (svm->vmcb->save.rax & -1u)
1195                 | ((u64)(svm->vcpu.regs[VCPU_REGS_RDX] & -1u) << 32);
1196         svm->next_rip = svm->vmcb->save.rip + 2;
1197         if (svm_set_msr(&svm->vcpu, ecx, data))
1198                 svm_inject_gp(&svm->vcpu, 0);
1199         else
1200                 skip_emulated_instruction(&svm->vcpu);
1201         return 1;
1202 }
1203
1204 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1205 {
1206         if (svm->vmcb->control.exit_info_1)
1207                 return wrmsr_interception(svm, kvm_run);
1208         else
1209                 return rdmsr_interception(svm, kvm_run);
1210 }
1211
1212 static int interrupt_window_interception(struct vcpu_svm *svm,
1213                                    struct kvm_run *kvm_run)
1214 {
1215         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1216         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1217         /*
1218          * If the user space waits to inject interrupts, exit as soon as
1219          * possible
1220          */
1221         if (kvm_run->request_interrupt_window &&
1222             !svm->vcpu.irq_summary) {
1223                 ++svm->vcpu.stat.irq_window_exits;
1224                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1225                 return 0;
1226         }
1227
1228         return 1;
1229 }
1230
1231 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1232                                       struct kvm_run *kvm_run) = {
1233         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1234         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1235         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1236         /* for now: */
1237         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1238         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1239         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1240         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1241         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1242         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1243         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1244         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1245         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1246         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1247         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1248         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1249         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1250         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
1251         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1252         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
1253         [SVM_EXIT_INTR]                         = nop_on_interception,
1254         [SVM_EXIT_NMI]                          = nop_on_interception,
1255         [SVM_EXIT_SMI]                          = nop_on_interception,
1256         [SVM_EXIT_INIT]                         = nop_on_interception,
1257         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1258         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1259         [SVM_EXIT_CPUID]                        = cpuid_interception,
1260         [SVM_EXIT_INVD]                         = emulate_on_interception,
1261         [SVM_EXIT_HLT]                          = halt_interception,
1262         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1263         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1264         [SVM_EXIT_IOIO]                         = io_interception,
1265         [SVM_EXIT_MSR]                          = msr_interception,
1266         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1267         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1268         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1269         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
1270         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1271         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1272         [SVM_EXIT_STGI]                         = invalid_op_interception,
1273         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1274         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1275         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
1276         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
1277         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
1278 };
1279
1280
1281 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1282 {
1283         struct vcpu_svm *svm = to_svm(vcpu);
1284         u32 exit_code = svm->vmcb->control.exit_code;
1285
1286         kvm_reput_irq(svm);
1287
1288         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1289                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1290                 kvm_run->fail_entry.hardware_entry_failure_reason
1291                         = svm->vmcb->control.exit_code;
1292                 return 0;
1293         }
1294
1295         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1296             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1297                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1298                        "exit_code 0x%x\n",
1299                        __FUNCTION__, svm->vmcb->control.exit_int_info,
1300                        exit_code);
1301
1302         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1303             || svm_exit_handlers[exit_code] == 0) {
1304                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1305                 kvm_run->hw.hardware_exit_reason = exit_code;
1306                 return 0;
1307         }
1308
1309         return svm_exit_handlers[exit_code](svm, kvm_run);
1310 }
1311
1312 static void reload_tss(struct kvm_vcpu *vcpu)
1313 {
1314         int cpu = raw_smp_processor_id();
1315
1316         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1317         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1318         load_TR_desc();
1319 }
1320
1321 static void pre_svm_run(struct vcpu_svm *svm)
1322 {
1323         int cpu = raw_smp_processor_id();
1324
1325         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1326
1327         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1328         if (svm->vcpu.cpu != cpu ||
1329             svm->asid_generation != svm_data->asid_generation)
1330                 new_asid(svm, svm_data);
1331 }
1332
1333
1334 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1335 {
1336         struct vmcb_control_area *control;
1337
1338         control = &svm->vmcb->control;
1339         control->int_vector = irq;
1340         control->int_ctl &= ~V_INTR_PRIO_MASK;
1341         control->int_ctl |= V_IRQ_MASK |
1342                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1343 }
1344
1345 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1346 {
1347         struct vcpu_svm *svm = to_svm(vcpu);
1348
1349         svm_inject_irq(svm, irq);
1350 }
1351
1352 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1353 {
1354         struct vcpu_svm *svm = to_svm(vcpu);
1355         struct vmcb *vmcb = svm->vmcb;
1356         int intr_vector = -1;
1357
1358         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1359             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1360                 intr_vector = vmcb->control.exit_int_info &
1361                               SVM_EVTINJ_VEC_MASK;
1362                 vmcb->control.exit_int_info = 0;
1363                 svm_inject_irq(svm, intr_vector);
1364                 return;
1365         }
1366
1367         if (vmcb->control.int_ctl & V_IRQ_MASK)
1368                 return;
1369
1370         if (!kvm_cpu_has_interrupt(vcpu))
1371                 return;
1372
1373         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1374             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1375             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1376                 /* unable to deliver irq, set pending irq */
1377                 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1378                 svm_inject_irq(svm, 0x0);
1379                 return;
1380         }
1381         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1382         intr_vector = kvm_cpu_get_interrupt(vcpu);
1383         svm_inject_irq(svm, intr_vector);
1384         kvm_timer_intr_post(vcpu, intr_vector);
1385 }
1386
1387 static void kvm_reput_irq(struct vcpu_svm *svm)
1388 {
1389         struct vmcb_control_area *control = &svm->vmcb->control;
1390
1391         if ((control->int_ctl & V_IRQ_MASK)
1392             && !irqchip_in_kernel(svm->vcpu.kvm)) {
1393                 control->int_ctl &= ~V_IRQ_MASK;
1394                 push_irq(&svm->vcpu, control->int_vector);
1395         }
1396
1397         svm->vcpu.interrupt_window_open =
1398                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1399 }
1400
1401 static void svm_do_inject_vector(struct vcpu_svm *svm)
1402 {
1403         struct kvm_vcpu *vcpu = &svm->vcpu;
1404         int word_index = __ffs(vcpu->irq_summary);
1405         int bit_index = __ffs(vcpu->irq_pending[word_index]);
1406         int irq = word_index * BITS_PER_LONG + bit_index;
1407
1408         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1409         if (!vcpu->irq_pending[word_index])
1410                 clear_bit(word_index, &vcpu->irq_summary);
1411         svm_inject_irq(svm, irq);
1412 }
1413
1414 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1415                                        struct kvm_run *kvm_run)
1416 {
1417         struct vcpu_svm *svm = to_svm(vcpu);
1418         struct vmcb_control_area *control = &svm->vmcb->control;
1419
1420         svm->vcpu.interrupt_window_open =
1421                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1422                  (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1423
1424         if (svm->vcpu.interrupt_window_open && svm->vcpu.irq_summary)
1425                 /*
1426                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1427                  */
1428                 svm_do_inject_vector(svm);
1429
1430         /*
1431          * Interrupts blocked.  Wait for unblock.
1432          */
1433         if (!svm->vcpu.interrupt_window_open &&
1434             (svm->vcpu.irq_summary || kvm_run->request_interrupt_window))
1435                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1436          else
1437                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1438 }
1439
1440 static void save_db_regs(unsigned long *db_regs)
1441 {
1442         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1443         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1444         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1445         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1446 }
1447
1448 static void load_db_regs(unsigned long *db_regs)
1449 {
1450         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1451         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1452         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1453         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1454 }
1455
1456 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1457 {
1458         force_new_asid(vcpu);
1459 }
1460
1461 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1462 {
1463 }
1464
1465 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1466 {
1467         struct vcpu_svm *svm = to_svm(vcpu);
1468         u16 fs_selector;
1469         u16 gs_selector;
1470         u16 ldt_selector;
1471
1472         pre_svm_run(svm);
1473
1474         save_host_msrs(vcpu);
1475         fs_selector = read_fs();
1476         gs_selector = read_gs();
1477         ldt_selector = read_ldt();
1478         svm->host_cr2 = kvm_read_cr2();
1479         svm->host_dr6 = read_dr6();
1480         svm->host_dr7 = read_dr7();
1481         svm->vmcb->save.cr2 = vcpu->cr2;
1482
1483         if (svm->vmcb->save.dr7 & 0xff) {
1484                 write_dr7(0);
1485                 save_db_regs(svm->host_db_regs);
1486                 load_db_regs(svm->db_regs);
1487         }
1488
1489         clgi();
1490
1491         local_irq_enable();
1492
1493         asm volatile (
1494 #ifdef CONFIG_X86_64
1495                 "push %%rbx; push %%rcx; push %%rdx;"
1496                 "push %%rsi; push %%rdi; push %%rbp;"
1497                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1498                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1499 #else
1500                 "push %%ebx; push %%ecx; push %%edx;"
1501                 "push %%esi; push %%edi; push %%ebp;"
1502 #endif
1503
1504 #ifdef CONFIG_X86_64
1505                 "mov %c[rbx](%[svm]), %%rbx \n\t"
1506                 "mov %c[rcx](%[svm]), %%rcx \n\t"
1507                 "mov %c[rdx](%[svm]), %%rdx \n\t"
1508                 "mov %c[rsi](%[svm]), %%rsi \n\t"
1509                 "mov %c[rdi](%[svm]), %%rdi \n\t"
1510                 "mov %c[rbp](%[svm]), %%rbp \n\t"
1511                 "mov %c[r8](%[svm]),  %%r8  \n\t"
1512                 "mov %c[r9](%[svm]),  %%r9  \n\t"
1513                 "mov %c[r10](%[svm]), %%r10 \n\t"
1514                 "mov %c[r11](%[svm]), %%r11 \n\t"
1515                 "mov %c[r12](%[svm]), %%r12 \n\t"
1516                 "mov %c[r13](%[svm]), %%r13 \n\t"
1517                 "mov %c[r14](%[svm]), %%r14 \n\t"
1518                 "mov %c[r15](%[svm]), %%r15 \n\t"
1519 #else
1520                 "mov %c[rbx](%[svm]), %%ebx \n\t"
1521                 "mov %c[rcx](%[svm]), %%ecx \n\t"
1522                 "mov %c[rdx](%[svm]), %%edx \n\t"
1523                 "mov %c[rsi](%[svm]), %%esi \n\t"
1524                 "mov %c[rdi](%[svm]), %%edi \n\t"
1525                 "mov %c[rbp](%[svm]), %%ebp \n\t"
1526 #endif
1527
1528 #ifdef CONFIG_X86_64
1529                 /* Enter guest mode */
1530                 "push %%rax \n\t"
1531                 "mov %c[vmcb](%[svm]), %%rax \n\t"
1532                 SVM_VMLOAD "\n\t"
1533                 SVM_VMRUN "\n\t"
1534                 SVM_VMSAVE "\n\t"
1535                 "pop %%rax \n\t"
1536 #else
1537                 /* Enter guest mode */
1538                 "push %%eax \n\t"
1539                 "mov %c[vmcb](%[svm]), %%eax \n\t"
1540                 SVM_VMLOAD "\n\t"
1541                 SVM_VMRUN "\n\t"
1542                 SVM_VMSAVE "\n\t"
1543                 "pop %%eax \n\t"
1544 #endif
1545
1546                 /* Save guest registers, load host registers */
1547 #ifdef CONFIG_X86_64
1548                 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1549                 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1550                 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1551                 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1552                 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1553                 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1554                 "mov %%r8,  %c[r8](%[svm]) \n\t"
1555                 "mov %%r9,  %c[r9](%[svm]) \n\t"
1556                 "mov %%r10, %c[r10](%[svm]) \n\t"
1557                 "mov %%r11, %c[r11](%[svm]) \n\t"
1558                 "mov %%r12, %c[r12](%[svm]) \n\t"
1559                 "mov %%r13, %c[r13](%[svm]) \n\t"
1560                 "mov %%r14, %c[r14](%[svm]) \n\t"
1561                 "mov %%r15, %c[r15](%[svm]) \n\t"
1562
1563                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1564                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1565                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1566                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1567 #else
1568                 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1569                 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1570                 "mov %%edx, %c[rdx](%[svm]) \n\t"
1571                 "mov %%esi, %c[rsi](%[svm]) \n\t"
1572                 "mov %%edi, %c[rdi](%[svm]) \n\t"
1573                 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1574
1575                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1576                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1577 #endif
1578                 :
1579                 : [svm]"a"(svm),
1580                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1581                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RBX])),
1582                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RCX])),
1583                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RDX])),
1584                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RSI])),
1585                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RDI])),
1586                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_RBP]))
1587 #ifdef CONFIG_X86_64
1588                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R8])),
1589                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R9])),
1590                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R10])),
1591                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R11])),
1592                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R12])),
1593                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R13])),
1594                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R14])),
1595                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.regs[VCPU_REGS_R15]))
1596 #endif
1597                 : "cc", "memory");
1598
1599         if ((svm->vmcb->save.dr7 & 0xff))
1600                 load_db_regs(svm->host_db_regs);
1601
1602         vcpu->cr2 = svm->vmcb->save.cr2;
1603
1604         write_dr6(svm->host_dr6);
1605         write_dr7(svm->host_dr7);
1606         kvm_write_cr2(svm->host_cr2);
1607
1608         load_fs(fs_selector);
1609         load_gs(gs_selector);
1610         load_ldt(ldt_selector);
1611         load_host_msrs(vcpu);
1612
1613         reload_tss(vcpu);
1614
1615         local_irq_disable();
1616
1617         stgi();
1618
1619         svm->next_rip = 0;
1620 }
1621
1622 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1623 {
1624         struct vcpu_svm *svm = to_svm(vcpu);
1625
1626         svm->vmcb->save.cr3 = root;
1627         force_new_asid(vcpu);
1628
1629         if (vcpu->fpu_active) {
1630                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1631                 svm->vmcb->save.cr0 |= X86_CR0_TS;
1632                 vcpu->fpu_active = 0;
1633         }
1634 }
1635
1636 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1637                                   unsigned long  addr,
1638                                   uint32_t err_code)
1639 {
1640         struct vcpu_svm *svm = to_svm(vcpu);
1641         uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1642
1643         ++vcpu->stat.pf_guest;
1644
1645         if (is_page_fault(exit_int_info)) {
1646
1647                 svm->vmcb->control.event_inj_err = 0;
1648                 svm->vmcb->control.event_inj =  SVM_EVTINJ_VALID |
1649                                                 SVM_EVTINJ_VALID_ERR |
1650                                                 SVM_EVTINJ_TYPE_EXEPT |
1651                                                 DF_VECTOR;
1652                 return;
1653         }
1654         vcpu->cr2 = addr;
1655         svm->vmcb->save.cr2 = addr;
1656         svm->vmcb->control.event_inj =  SVM_EVTINJ_VALID |
1657                                         SVM_EVTINJ_VALID_ERR |
1658                                         SVM_EVTINJ_TYPE_EXEPT |
1659                                         PF_VECTOR;
1660         svm->vmcb->control.event_inj_err = err_code;
1661 }
1662
1663
1664 static int is_disabled(void)
1665 {
1666         u64 vm_cr;
1667
1668         rdmsrl(MSR_VM_CR, vm_cr);
1669         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1670                 return 1;
1671
1672         return 0;
1673 }
1674
1675 static void
1676 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1677 {
1678         /*
1679          * Patch in the VMMCALL instruction:
1680          */
1681         hypercall[0] = 0x0f;
1682         hypercall[1] = 0x01;
1683         hypercall[2] = 0xd9;
1684 }
1685
1686 static void svm_check_processor_compat(void *rtn)
1687 {
1688         *(int *)rtn = 0;
1689 }
1690
1691 static struct kvm_x86_ops svm_x86_ops = {
1692         .cpu_has_kvm_support = has_svm,
1693         .disabled_by_bios = is_disabled,
1694         .hardware_setup = svm_hardware_setup,
1695         .hardware_unsetup = svm_hardware_unsetup,
1696         .check_processor_compatibility = svm_check_processor_compat,
1697         .hardware_enable = svm_hardware_enable,
1698         .hardware_disable = svm_hardware_disable,
1699
1700         .vcpu_create = svm_create_vcpu,
1701         .vcpu_free = svm_free_vcpu,
1702         .vcpu_reset = svm_vcpu_reset,
1703
1704         .prepare_guest_switch = svm_prepare_guest_switch,
1705         .vcpu_load = svm_vcpu_load,
1706         .vcpu_put = svm_vcpu_put,
1707         .vcpu_decache = svm_vcpu_decache,
1708
1709         .set_guest_debug = svm_guest_debug,
1710         .get_msr = svm_get_msr,
1711         .set_msr = svm_set_msr,
1712         .get_segment_base = svm_get_segment_base,
1713         .get_segment = svm_get_segment,
1714         .set_segment = svm_set_segment,
1715         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1716         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1717         .set_cr0 = svm_set_cr0,
1718         .set_cr3 = svm_set_cr3,
1719         .set_cr4 = svm_set_cr4,
1720         .set_efer = svm_set_efer,
1721         .get_idt = svm_get_idt,
1722         .set_idt = svm_set_idt,
1723         .get_gdt = svm_get_gdt,
1724         .set_gdt = svm_set_gdt,
1725         .get_dr = svm_get_dr,
1726         .set_dr = svm_set_dr,
1727         .cache_regs = svm_cache_regs,
1728         .decache_regs = svm_decache_regs,
1729         .get_rflags = svm_get_rflags,
1730         .set_rflags = svm_set_rflags,
1731
1732         .tlb_flush = svm_flush_tlb,
1733         .inject_page_fault = svm_inject_page_fault,
1734
1735         .inject_gp = svm_inject_gp,
1736
1737         .run = svm_vcpu_run,
1738         .handle_exit = handle_exit,
1739         .skip_emulated_instruction = skip_emulated_instruction,
1740         .patch_hypercall = svm_patch_hypercall,
1741         .get_irq = svm_get_irq,
1742         .set_irq = svm_set_irq,
1743         .inject_pending_irq = svm_intr_assist,
1744         .inject_pending_vectors = do_interrupt_requests,
1745 };
1746
1747 static int __init svm_init(void)
1748 {
1749         return kvm_init_x86(&svm_x86_ops, sizeof(struct vcpu_svm),
1750                               THIS_MODULE);
1751 }
1752
1753 static void __exit svm_exit(void)
1754 {
1755         kvm_exit_x86();
1756 }
1757
1758 module_init(svm_init)
1759 module_exit(svm_exit)