Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-apm
[sfrench/cifs-2.6.git] / 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 <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/highmem.h>
20 #include <linux/profile.h>
21 #include <asm/desc.h>
22
23 #include "kvm_svm.h"
24 #include "x86_emulate.h"
25
26 MODULE_AUTHOR("Qumranet");
27 MODULE_LICENSE("GPL");
28
29 #define IOPM_ALLOC_ORDER 2
30 #define MSRPM_ALLOC_ORDER 1
31
32 #define DB_VECTOR 1
33 #define UD_VECTOR 6
34 #define GP_VECTOR 13
35
36 #define DR7_GD_MASK (1 << 13)
37 #define DR6_BD_MASK (1 << 13)
38 #define CR4_DE_MASK (1UL << 3)
39
40 #define SEG_TYPE_LDT 2
41 #define SEG_TYPE_BUSY_TSS16 3
42
43 #define KVM_EFER_LMA (1 << 10)
44 #define KVM_EFER_LME (1 << 8)
45
46 unsigned long iopm_base;
47 unsigned long msrpm_base;
48
49 struct kvm_ldttss_desc {
50         u16 limit0;
51         u16 base0;
52         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
53         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
54         u32 base3;
55         u32 zero1;
56 } __attribute__((packed));
57
58 struct svm_cpu_data {
59         int cpu;
60
61         uint64_t asid_generation;
62         uint32_t max_asid;
63         uint32_t next_asid;
64         struct kvm_ldttss_desc *tss_desc;
65
66         struct page *save_area;
67 };
68
69 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
70
71 struct svm_init_data {
72         int cpu;
73         int r;
74 };
75
76 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
77
78 #define NUM_MSR_MAPS (sizeof(msrpm_ranges) / sizeof(*msrpm_ranges))
79 #define MSRS_RANGE_SIZE 2048
80 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
81
82 #define MAX_INST_SIZE 15
83
84 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
85 {
86         struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
87         u16 cs_attrib;
88
89         if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
90                 return 2;
91
92         cs_attrib = sa->cs.attrib;
93
94         return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
95                                 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
96 }
97
98 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
99 {
100         int word_index = __ffs(vcpu->irq_summary);
101         int bit_index = __ffs(vcpu->irq_pending[word_index]);
102         int irq = word_index * BITS_PER_LONG + bit_index;
103
104         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
105         if (!vcpu->irq_pending[word_index])
106                 clear_bit(word_index, &vcpu->irq_summary);
107         return irq;
108 }
109
110 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
111 {
112         set_bit(irq, vcpu->irq_pending);
113         set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
114 }
115
116 static inline void clgi(void)
117 {
118         asm volatile (SVM_CLGI);
119 }
120
121 static inline void stgi(void)
122 {
123         asm volatile (SVM_STGI);
124 }
125
126 static inline void invlpga(unsigned long addr, u32 asid)
127 {
128         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
129 }
130
131 static inline unsigned long kvm_read_cr2(void)
132 {
133         unsigned long cr2;
134
135         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
136         return cr2;
137 }
138
139 static inline void kvm_write_cr2(unsigned long val)
140 {
141         asm volatile ("mov %0, %%cr2" :: "r" (val));
142 }
143
144 static inline unsigned long read_dr6(void)
145 {
146         unsigned long dr6;
147
148         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
149         return dr6;
150 }
151
152 static inline void write_dr6(unsigned long val)
153 {
154         asm volatile ("mov %0, %%dr6" :: "r" (val));
155 }
156
157 static inline unsigned long read_dr7(void)
158 {
159         unsigned long dr7;
160
161         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
162         return dr7;
163 }
164
165 static inline void write_dr7(unsigned long val)
166 {
167         asm volatile ("mov %0, %%dr7" :: "r" (val));
168 }
169
170 static inline void force_new_asid(struct kvm_vcpu *vcpu)
171 {
172         vcpu->svm->asid_generation--;
173 }
174
175 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
176 {
177         force_new_asid(vcpu);
178 }
179
180 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
181 {
182         if (!(efer & KVM_EFER_LMA))
183                 efer &= ~KVM_EFER_LME;
184
185         vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
186         vcpu->shadow_efer = efer;
187 }
188
189 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
190 {
191         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
192                                                 SVM_EVTINJ_VALID_ERR |
193                                                 SVM_EVTINJ_TYPE_EXEPT |
194                                                 GP_VECTOR;
195         vcpu->svm->vmcb->control.event_inj_err = error_code;
196 }
197
198 static void inject_ud(struct kvm_vcpu *vcpu)
199 {
200         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
201                                                 SVM_EVTINJ_TYPE_EXEPT |
202                                                 UD_VECTOR;
203 }
204
205 static void inject_db(struct kvm_vcpu *vcpu)
206 {
207         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
208                                                 SVM_EVTINJ_TYPE_EXEPT |
209                                                 DB_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         if (!vcpu->svm->next_rip) {
227                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
228                 return;
229         }
230         if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
231                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
232                        __FUNCTION__,
233                        vcpu->svm->vmcb->save.rip,
234                        vcpu->svm->next_rip);
235         }
236
237         vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
238         vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
239
240         vcpu->interrupt_window_open = 1;
241 }
242
243 static int has_svm(void)
244 {
245         uint32_t eax, ebx, ecx, edx;
246
247         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
248                 printk(KERN_INFO "has_svm: not amd\n");
249                 return 0;
250         }
251
252         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
253         if (eax < SVM_CPUID_FUNC) {
254                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
255                 return 0;
256         }
257
258         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
259         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
260                 printk(KERN_DEBUG "has_svm: svm not available\n");
261                 return 0;
262         }
263         return 1;
264 }
265
266 static void svm_hardware_disable(void *garbage)
267 {
268         struct svm_cpu_data *svm_data
269                 = per_cpu(svm_data, raw_smp_processor_id());
270
271         if (svm_data) {
272                 uint64_t efer;
273
274                 wrmsrl(MSR_VM_HSAVE_PA, 0);
275                 rdmsrl(MSR_EFER, efer);
276                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
277                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
278                 __free_page(svm_data->save_area);
279                 kfree(svm_data);
280         }
281 }
282
283 static void svm_hardware_enable(void *garbage)
284 {
285
286         struct svm_cpu_data *svm_data;
287         uint64_t efer;
288 #ifdef CONFIG_X86_64
289         struct desc_ptr gdt_descr;
290 #else
291         struct Xgt_desc_struct gdt_descr;
292 #endif
293         struct desc_struct *gdt;
294         int me = raw_smp_processor_id();
295
296         if (!has_svm()) {
297                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
298                 return;
299         }
300         svm_data = per_cpu(svm_data, me);
301
302         if (!svm_data) {
303                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
304                        me);
305                 return;
306         }
307
308         svm_data->asid_generation = 1;
309         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
310         svm_data->next_asid = svm_data->max_asid + 1;
311
312         asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
313         gdt = (struct desc_struct *)gdt_descr.address;
314         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
315
316         rdmsrl(MSR_EFER, efer);
317         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
318
319         wrmsrl(MSR_VM_HSAVE_PA,
320                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
321 }
322
323 static int svm_cpu_init(int cpu)
324 {
325         struct svm_cpu_data *svm_data;
326         int r;
327
328         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
329         if (!svm_data)
330                 return -ENOMEM;
331         svm_data->cpu = cpu;
332         svm_data->save_area = alloc_page(GFP_KERNEL);
333         r = -ENOMEM;
334         if (!svm_data->save_area)
335                 goto err_1;
336
337         per_cpu(svm_data, cpu) = svm_data;
338
339         return 0;
340
341 err_1:
342         kfree(svm_data);
343         return r;
344
345 }
346
347 static int set_msr_interception(u32 *msrpm, unsigned msr,
348                                 int read, int write)
349 {
350         int i;
351
352         for (i = 0; i < NUM_MSR_MAPS; i++) {
353                 if (msr >= msrpm_ranges[i] &&
354                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
355                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
356                                           msrpm_ranges[i]) * 2;
357
358                         u32 *base = msrpm + (msr_offset / 32);
359                         u32 msr_shift = msr_offset % 32;
360                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
361                         *base = (*base & ~(0x3 << msr_shift)) |
362                                 (mask << msr_shift);
363                         return 1;
364                 }
365         }
366         printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
367         return 0;
368 }
369
370 static __init int svm_hardware_setup(void)
371 {
372         int cpu;
373         struct page *iopm_pages;
374         struct page *msrpm_pages;
375         void *msrpm_va;
376         int r;
377
378         kvm_emulator_want_group7_invlpg();
379
380         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
381
382         if (!iopm_pages)
383                 return -ENOMEM;
384         memset(page_address(iopm_pages), 0xff,
385                                         PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
386         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
387
388
389         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
390
391         r = -ENOMEM;
392         if (!msrpm_pages)
393                 goto err_1;
394
395         msrpm_va = page_address(msrpm_pages);
396         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
397         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
398
399 #ifdef CONFIG_X86_64
400         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
401         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
402         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
403         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
404         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
405         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
406 #endif
407         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
408         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
409         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
410         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
411
412         for_each_online_cpu(cpu) {
413                 r = svm_cpu_init(cpu);
414                 if (r)
415                         goto err_2;
416         }
417         return 0;
418
419 err_2:
420         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
421         msrpm_base = 0;
422 err_1:
423         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
424         iopm_base = 0;
425         return r;
426 }
427
428 static __exit void svm_hardware_unsetup(void)
429 {
430         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
431         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
432         iopm_base = msrpm_base = 0;
433 }
434
435 static void init_seg(struct vmcb_seg *seg)
436 {
437         seg->selector = 0;
438         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
439                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
440         seg->limit = 0xffff;
441         seg->base = 0;
442 }
443
444 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
445 {
446         seg->selector = 0;
447         seg->attrib = SVM_SELECTOR_P_MASK | type;
448         seg->limit = 0xffff;
449         seg->base = 0;
450 }
451
452 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
453 {
454         return 0;
455 }
456
457 static void init_vmcb(struct vmcb *vmcb)
458 {
459         struct vmcb_control_area *control = &vmcb->control;
460         struct vmcb_save_area *save = &vmcb->save;
461         u64 tsc;
462
463         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
464                                         INTERCEPT_CR3_MASK |
465                                         INTERCEPT_CR4_MASK;
466
467         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
468                                         INTERCEPT_CR3_MASK |
469                                         INTERCEPT_CR4_MASK;
470
471         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
472                                         INTERCEPT_DR1_MASK |
473                                         INTERCEPT_DR2_MASK |
474                                         INTERCEPT_DR3_MASK;
475
476         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
477                                         INTERCEPT_DR1_MASK |
478                                         INTERCEPT_DR2_MASK |
479                                         INTERCEPT_DR3_MASK |
480                                         INTERCEPT_DR5_MASK |
481                                         INTERCEPT_DR7_MASK;
482
483         control->intercept_exceptions = 1 << PF_VECTOR;
484
485
486         control->intercept =    (1ULL << INTERCEPT_INTR) |
487                                 (1ULL << INTERCEPT_NMI) |
488                 /*
489                  * selective cr0 intercept bug?
490                  *      0:   0f 22 d8                mov    %eax,%cr3
491                  *      3:   0f 20 c0                mov    %cr0,%eax
492                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
493                  *      b:   0f 22 c0                mov    %eax,%cr0
494                  * set cr3 ->interception
495                  * get cr0 ->interception
496                  * set cr0 -> no interception
497                  */
498                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
499                                 (1ULL << INTERCEPT_CPUID) |
500                                 (1ULL << INTERCEPT_HLT) |
501                                 (1ULL << INTERCEPT_INVLPGA) |
502                                 (1ULL << INTERCEPT_IOIO_PROT) |
503                                 (1ULL << INTERCEPT_MSR_PROT) |
504                                 (1ULL << INTERCEPT_TASK_SWITCH) |
505                                 (1ULL << INTERCEPT_SHUTDOWN) |
506                                 (1ULL << INTERCEPT_VMRUN) |
507                                 (1ULL << INTERCEPT_VMMCALL) |
508                                 (1ULL << INTERCEPT_VMLOAD) |
509                                 (1ULL << INTERCEPT_VMSAVE) |
510                                 (1ULL << INTERCEPT_STGI) |
511                                 (1ULL << INTERCEPT_CLGI) |
512                                 (1ULL << INTERCEPT_SKINIT);
513
514         control->iopm_base_pa = iopm_base;
515         control->msrpm_base_pa = msrpm_base;
516         rdtscll(tsc);
517         control->tsc_offset = -tsc;
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         save->cs.base = 0xffff0000;
532
533         save->gdtr.limit = 0xffff;
534         save->idtr.limit = 0xffff;
535
536         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
537         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
538
539         save->efer = MSR_EFER_SVME_MASK;
540
541         save->dr6 = 0xffff0ff0;
542         save->dr7 = 0x400;
543         save->rflags = 2;
544         save->rip = 0x0000fff0;
545
546         /*
547          * cr0 val on cpu init should be 0x60000010, we enable cpu
548          * cache by default. the orderly way is to enable cache in bios.
549          */
550         save->cr0 = 0x00000010 | CR0_PG_MASK;
551         save->cr4 = CR4_PAE_MASK;
552         /* rdx = ?? */
553 }
554
555 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
556 {
557         struct page *page;
558         int r;
559
560         r = -ENOMEM;
561         vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
562         if (!vcpu->svm)
563                 goto out1;
564         page = alloc_page(GFP_KERNEL);
565         if (!page)
566                 goto out2;
567
568         vcpu->svm->vmcb = page_address(page);
569         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
570         vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
571         vcpu->svm->cr0 = 0x00000010;
572         vcpu->svm->asid_generation = 0;
573         memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
574         init_vmcb(vcpu->svm->vmcb);
575
576         fx_init(vcpu);
577
578         return 0;
579
580 out2:
581         kfree(vcpu->svm);
582 out1:
583         return r;
584 }
585
586 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
587 {
588         if (!vcpu->svm)
589                 return;
590         if (vcpu->svm->vmcb)
591                 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
592         kfree(vcpu->svm);
593 }
594
595 static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
596 {
597         get_cpu();
598         return vcpu;
599 }
600
601 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
602 {
603         put_cpu();
604 }
605
606 static void svm_cache_regs(struct kvm_vcpu *vcpu)
607 {
608         vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
609         vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
610         vcpu->rip = vcpu->svm->vmcb->save.rip;
611 }
612
613 static void svm_decache_regs(struct kvm_vcpu *vcpu)
614 {
615         vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
616         vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
617         vcpu->svm->vmcb->save.rip = vcpu->rip;
618 }
619
620 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
621 {
622         return vcpu->svm->vmcb->save.rflags;
623 }
624
625 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
626 {
627         vcpu->svm->vmcb->save.rflags = rflags;
628 }
629
630 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
631 {
632         struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
633
634         switch (seg) {
635         case VCPU_SREG_CS: return &save->cs;
636         case VCPU_SREG_DS: return &save->ds;
637         case VCPU_SREG_ES: return &save->es;
638         case VCPU_SREG_FS: return &save->fs;
639         case VCPU_SREG_GS: return &save->gs;
640         case VCPU_SREG_SS: return &save->ss;
641         case VCPU_SREG_TR: return &save->tr;
642         case VCPU_SREG_LDTR: return &save->ldtr;
643         }
644         BUG();
645         return NULL;
646 }
647
648 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
649 {
650         struct vmcb_seg *s = svm_seg(vcpu, seg);
651
652         return s->base;
653 }
654
655 static void svm_get_segment(struct kvm_vcpu *vcpu,
656                             struct kvm_segment *var, int seg)
657 {
658         struct vmcb_seg *s = svm_seg(vcpu, seg);
659
660         var->base = s->base;
661         var->limit = s->limit;
662         var->selector = s->selector;
663         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
664         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
665         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
666         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
667         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
668         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
669         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
670         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
671         var->unusable = !var->present;
672 }
673
674 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
675 {
676         struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
677
678         *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
679         *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
680 }
681
682 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
683 {
684         dt->limit = vcpu->svm->vmcb->save.idtr.limit;
685         dt->base = vcpu->svm->vmcb->save.idtr.base;
686 }
687
688 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
689 {
690         vcpu->svm->vmcb->save.idtr.limit = dt->limit;
691         vcpu->svm->vmcb->save.idtr.base = dt->base ;
692 }
693
694 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
695 {
696         dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
697         dt->base = vcpu->svm->vmcb->save.gdtr.base;
698 }
699
700 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
701 {
702         vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
703         vcpu->svm->vmcb->save.gdtr.base = dt->base ;
704 }
705
706 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
707 {
708 }
709
710 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
711 {
712 #ifdef CONFIG_X86_64
713         if (vcpu->shadow_efer & KVM_EFER_LME) {
714                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
715                         vcpu->shadow_efer |= KVM_EFER_LMA;
716                         vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
717                 }
718
719                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
720                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
721                         vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
722                 }
723         }
724 #endif
725         vcpu->svm->cr0 = cr0;
726         vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK;
727         vcpu->cr0 = cr0;
728 }
729
730 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
731 {
732        vcpu->cr4 = cr4;
733        vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
734 }
735
736 static void svm_set_segment(struct kvm_vcpu *vcpu,
737                             struct kvm_segment *var, int seg)
738 {
739         struct vmcb_seg *s = svm_seg(vcpu, seg);
740
741         s->base = var->base;
742         s->limit = var->limit;
743         s->selector = var->selector;
744         if (var->unusable)
745                 s->attrib = 0;
746         else {
747                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
748                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
749                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
750                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
751                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
752                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
753                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
754                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
755         }
756         if (seg == VCPU_SREG_CS)
757                 vcpu->svm->vmcb->save.cpl
758                         = (vcpu->svm->vmcb->save.cs.attrib
759                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
760
761 }
762
763 /* FIXME:
764
765         vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
766         vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
767
768 */
769
770 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
771 {
772         return -EOPNOTSUPP;
773 }
774
775 static void load_host_msrs(struct kvm_vcpu *vcpu)
776 {
777         int i;
778
779         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
780                 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
781 }
782
783 static void save_host_msrs(struct kvm_vcpu *vcpu)
784 {
785         int i;
786
787         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
788                 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
789 }
790
791 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
792 {
793         if (svm_data->next_asid > svm_data->max_asid) {
794                 ++svm_data->asid_generation;
795                 svm_data->next_asid = 1;
796                 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
797         }
798
799         vcpu->cpu = svm_data->cpu;
800         vcpu->svm->asid_generation = svm_data->asid_generation;
801         vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
802 }
803
804 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
805 {
806         invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
807 }
808
809 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
810 {
811         return vcpu->svm->db_regs[dr];
812 }
813
814 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
815                        int *exception)
816 {
817         *exception = 0;
818
819         if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
820                 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
821                 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
822                 *exception = DB_VECTOR;
823                 return;
824         }
825
826         switch (dr) {
827         case 0 ... 3:
828                 vcpu->svm->db_regs[dr] = value;
829                 return;
830         case 4 ... 5:
831                 if (vcpu->cr4 & CR4_DE_MASK) {
832                         *exception = UD_VECTOR;
833                         return;
834                 }
835         case 7: {
836                 if (value & ~((1ULL << 32) - 1)) {
837                         *exception = GP_VECTOR;
838                         return;
839                 }
840                 vcpu->svm->vmcb->save.dr7 = value;
841                 return;
842         }
843         default:
844                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
845                        __FUNCTION__, dr);
846                 *exception = UD_VECTOR;
847                 return;
848         }
849 }
850
851 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
852 {
853         u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
854         u64 fault_address;
855         u32 error_code;
856         enum emulation_result er;
857         int r;
858
859         if (is_external_interrupt(exit_int_info))
860                 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
861
862         spin_lock(&vcpu->kvm->lock);
863
864         fault_address  = vcpu->svm->vmcb->control.exit_info_2;
865         error_code = vcpu->svm->vmcb->control.exit_info_1;
866         r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
867         if (r < 0) {
868                 spin_unlock(&vcpu->kvm->lock);
869                 return r;
870         }
871         if (!r) {
872                 spin_unlock(&vcpu->kvm->lock);
873                 return 1;
874         }
875         er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
876         spin_unlock(&vcpu->kvm->lock);
877
878         switch (er) {
879         case EMULATE_DONE:
880                 return 1;
881         case EMULATE_DO_MMIO:
882                 ++kvm_stat.mmio_exits;
883                 kvm_run->exit_reason = KVM_EXIT_MMIO;
884                 return 0;
885         case EMULATE_FAIL:
886                 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
887                 break;
888         default:
889                 BUG();
890         }
891
892         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
893         return 0;
894 }
895
896 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
897 {
898         /*
899          * VMCB is undefined after a SHUTDOWN intercept
900          * so reinitialize it.
901          */
902         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
903         init_vmcb(vcpu->svm->vmcb);
904
905         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
906         return 0;
907 }
908
909 static int io_get_override(struct kvm_vcpu *vcpu,
910                           struct vmcb_seg **seg,
911                           int *addr_override)
912 {
913         u8 inst[MAX_INST_SIZE];
914         unsigned ins_length;
915         gva_t rip;
916         int i;
917
918         rip =  vcpu->svm->vmcb->save.rip;
919         ins_length = vcpu->svm->next_rip - rip;
920         rip += vcpu->svm->vmcb->save.cs.base;
921
922         if (ins_length > MAX_INST_SIZE)
923                 printk(KERN_DEBUG
924                        "%s: inst length err, cs base 0x%llx rip 0x%llx "
925                        "next rip 0x%llx ins_length %u\n",
926                        __FUNCTION__,
927                        vcpu->svm->vmcb->save.cs.base,
928                        vcpu->svm->vmcb->save.rip,
929                        vcpu->svm->vmcb->control.exit_info_2,
930                        ins_length);
931
932         if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
933                 /* #PF */
934                 return 0;
935
936         *addr_override = 0;
937         *seg = NULL;
938         for (i = 0; i < ins_length; i++)
939                 switch (inst[i]) {
940                 case 0xf0:
941                 case 0xf2:
942                 case 0xf3:
943                 case 0x66:
944                         continue;
945                 case 0x67:
946                         *addr_override = 1;
947                         continue;
948                 case 0x2e:
949                         *seg = &vcpu->svm->vmcb->save.cs;
950                         continue;
951                 case 0x36:
952                         *seg = &vcpu->svm->vmcb->save.ss;
953                         continue;
954                 case 0x3e:
955                         *seg = &vcpu->svm->vmcb->save.ds;
956                         continue;
957                 case 0x26:
958                         *seg = &vcpu->svm->vmcb->save.es;
959                         continue;
960                 case 0x64:
961                         *seg = &vcpu->svm->vmcb->save.fs;
962                         continue;
963                 case 0x65:
964                         *seg = &vcpu->svm->vmcb->save.gs;
965                         continue;
966                 default:
967                         return 1;
968                 }
969         printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
970         return 0;
971 }
972
973 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
974 {
975         unsigned long addr_mask;
976         unsigned long *reg;
977         struct vmcb_seg *seg;
978         int addr_override;
979         struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
980         u16 cs_attrib = save_area->cs.attrib;
981         unsigned addr_size = get_addr_size(vcpu);
982
983         if (!io_get_override(vcpu, &seg, &addr_override))
984                 return 0;
985
986         if (addr_override)
987                 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
988
989         if (ins) {
990                 reg = &vcpu->regs[VCPU_REGS_RDI];
991                 seg = &vcpu->svm->vmcb->save.es;
992         } else {
993                 reg = &vcpu->regs[VCPU_REGS_RSI];
994                 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
995         }
996
997         addr_mask = ~0ULL >> (64 - (addr_size * 8));
998
999         if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1000             !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1001                 *address = (*reg & addr_mask);
1002                 return addr_mask;
1003         }
1004
1005         if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1006                 svm_inject_gp(vcpu, 0);
1007                 return 0;
1008         }
1009
1010         *address = (*reg & addr_mask) + seg->base;
1011         return addr_mask;
1012 }
1013
1014 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1015 {
1016         u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1017         int _in = io_info & SVM_IOIO_TYPE_MASK;
1018
1019         ++kvm_stat.io_exits;
1020
1021         vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1022
1023         kvm_run->exit_reason = KVM_EXIT_IO;
1024         kvm_run->io.port = io_info >> 16;
1025         kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1026         kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1027         kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1028         kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1029
1030         if (kvm_run->io.string) {
1031                 unsigned addr_mask;
1032
1033                 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1034                 if (!addr_mask) {
1035                         printk(KERN_DEBUG "%s: get io address failed\n", __FUNCTION__);
1036                         return 1;
1037                 }
1038
1039                 if (kvm_run->io.rep) {
1040                         kvm_run->io.count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1041                         kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1042                                                    & X86_EFLAGS_DF) != 0;
1043                 }
1044         } else {
1045                 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1046         }
1047         return 0;
1048 }
1049
1050
1051 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1052 {
1053         return 1;
1054 }
1055
1056 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1057 {
1058         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1059         skip_emulated_instruction(vcpu);
1060         if (vcpu->irq_summary)
1061                 return 1;
1062
1063         kvm_run->exit_reason = KVM_EXIT_HLT;
1064         ++kvm_stat.halt_exits;
1065         return 0;
1066 }
1067
1068 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1069 {
1070         inject_ud(vcpu);
1071         return 1;
1072 }
1073
1074 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1075 {
1076         printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1077         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1078         return 0;
1079 }
1080
1081 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1082 {
1083         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1084         kvm_run->exit_reason = KVM_EXIT_CPUID;
1085         return 0;
1086 }
1087
1088 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1089 {
1090         if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1091                 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1092         return 1;
1093 }
1094
1095 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1096 {
1097         switch (ecx) {
1098         case MSR_IA32_TIME_STAMP_COUNTER: {
1099                 u64 tsc;
1100
1101                 rdtscll(tsc);
1102                 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1103                 break;
1104         }
1105         case MSR_K6_STAR:
1106                 *data = vcpu->svm->vmcb->save.star;
1107                 break;
1108 #ifdef CONFIG_X86_64
1109         case MSR_LSTAR:
1110                 *data = vcpu->svm->vmcb->save.lstar;
1111                 break;
1112         case MSR_CSTAR:
1113                 *data = vcpu->svm->vmcb->save.cstar;
1114                 break;
1115         case MSR_KERNEL_GS_BASE:
1116                 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1117                 break;
1118         case MSR_SYSCALL_MASK:
1119                 *data = vcpu->svm->vmcb->save.sfmask;
1120                 break;
1121 #endif
1122         case MSR_IA32_SYSENTER_CS:
1123                 *data = vcpu->svm->vmcb->save.sysenter_cs;
1124                 break;
1125         case MSR_IA32_SYSENTER_EIP:
1126                 *data = vcpu->svm->vmcb->save.sysenter_eip;
1127                 break;
1128         case MSR_IA32_SYSENTER_ESP:
1129                 *data = vcpu->svm->vmcb->save.sysenter_esp;
1130                 break;
1131         default:
1132                 return kvm_get_msr_common(vcpu, ecx, data);
1133         }
1134         return 0;
1135 }
1136
1137 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1138 {
1139         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1140         u64 data;
1141
1142         if (svm_get_msr(vcpu, ecx, &data))
1143                 svm_inject_gp(vcpu, 0);
1144         else {
1145                 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1146                 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1147                 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1148                 skip_emulated_instruction(vcpu);
1149         }
1150         return 1;
1151 }
1152
1153 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1154 {
1155         switch (ecx) {
1156         case MSR_IA32_TIME_STAMP_COUNTER: {
1157                 u64 tsc;
1158
1159                 rdtscll(tsc);
1160                 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1161                 break;
1162         }
1163         case MSR_K6_STAR:
1164                 vcpu->svm->vmcb->save.star = data;
1165                 break;
1166 #ifdef CONFIG_X86_64
1167         case MSR_LSTAR:
1168                 vcpu->svm->vmcb->save.lstar = data;
1169                 break;
1170         case MSR_CSTAR:
1171                 vcpu->svm->vmcb->save.cstar = data;
1172                 break;
1173         case MSR_KERNEL_GS_BASE:
1174                 vcpu->svm->vmcb->save.kernel_gs_base = data;
1175                 break;
1176         case MSR_SYSCALL_MASK:
1177                 vcpu->svm->vmcb->save.sfmask = data;
1178                 break;
1179 #endif
1180         case MSR_IA32_SYSENTER_CS:
1181                 vcpu->svm->vmcb->save.sysenter_cs = data;
1182                 break;
1183         case MSR_IA32_SYSENTER_EIP:
1184                 vcpu->svm->vmcb->save.sysenter_eip = data;
1185                 break;
1186         case MSR_IA32_SYSENTER_ESP:
1187                 vcpu->svm->vmcb->save.sysenter_esp = data;
1188                 break;
1189         default:
1190                 return kvm_set_msr_common(vcpu, ecx, data);
1191         }
1192         return 0;
1193 }
1194
1195 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1196 {
1197         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1198         u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1199                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1200         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1201         if (svm_set_msr(vcpu, ecx, data))
1202                 svm_inject_gp(vcpu, 0);
1203         else
1204                 skip_emulated_instruction(vcpu);
1205         return 1;
1206 }
1207
1208 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1209 {
1210         if (vcpu->svm->vmcb->control.exit_info_1)
1211                 return wrmsr_interception(vcpu, kvm_run);
1212         else
1213                 return rdmsr_interception(vcpu, kvm_run);
1214 }
1215
1216 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1217                                    struct kvm_run *kvm_run)
1218 {
1219         /*
1220          * If the user space waits to inject interrupts, exit as soon as
1221          * possible
1222          */
1223         if (kvm_run->request_interrupt_window &&
1224             !vcpu->irq_summary) {
1225                 ++kvm_stat.irq_window_exits;
1226                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1227                 return 0;
1228         }
1229
1230         return 1;
1231 }
1232
1233 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1234                                       struct kvm_run *kvm_run) = {
1235         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1236         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1237         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1238         /* for now: */
1239         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1240         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1241         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1242         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1243         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1244         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1245         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1246         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1247         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1248         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1249         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1250         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1251         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1252         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_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_HLT]                          = halt_interception,
1261         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1262         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1263         [SVM_EXIT_IOIO]                         = io_interception,
1264         [SVM_EXIT_MSR]                          = msr_interception,
1265         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1266         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1267         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1268         [SVM_EXIT_VMMCALL]                      = invalid_op_interception,
1269         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1270         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1271         [SVM_EXIT_STGI]                         = invalid_op_interception,
1272         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1273         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1274 };
1275
1276
1277 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1278 {
1279         u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1280
1281         kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1282
1283         if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1284             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1285                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1286                        "exit_code 0x%x\n",
1287                        __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1288                        exit_code);
1289
1290         if (exit_code >= sizeof(svm_exit_handlers) / sizeof(*svm_exit_handlers)
1291             || svm_exit_handlers[exit_code] == 0) {
1292                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1293                 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1294                        __FUNCTION__,
1295                        exit_code,
1296                        vcpu->svm->vmcb->save.rip,
1297                        vcpu->cr0,
1298                        vcpu->svm->vmcb->save.rflags);
1299                 return 0;
1300         }
1301
1302         return svm_exit_handlers[exit_code](vcpu, kvm_run);
1303 }
1304
1305 static void reload_tss(struct kvm_vcpu *vcpu)
1306 {
1307         int cpu = raw_smp_processor_id();
1308
1309         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1310         svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1311         load_TR_desc();
1312 }
1313
1314 static void pre_svm_run(struct kvm_vcpu *vcpu)
1315 {
1316         int cpu = raw_smp_processor_id();
1317
1318         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1319
1320         vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1321         if (vcpu->cpu != cpu ||
1322             vcpu->svm->asid_generation != svm_data->asid_generation)
1323                 new_asid(vcpu, svm_data);
1324 }
1325
1326
1327 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1328 {
1329         struct vmcb_control_area *control;
1330
1331         control = &vcpu->svm->vmcb->control;
1332         control->int_vector = pop_irq(vcpu);
1333         control->int_ctl &= ~V_INTR_PRIO_MASK;
1334         control->int_ctl |= V_IRQ_MASK |
1335                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1336 }
1337
1338 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1339 {
1340         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1341
1342         if (control->int_ctl & V_IRQ_MASK) {
1343                 control->int_ctl &= ~V_IRQ_MASK;
1344                 push_irq(vcpu, control->int_vector);
1345         }
1346
1347         vcpu->interrupt_window_open =
1348                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1349 }
1350
1351 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1352                                        struct kvm_run *kvm_run)
1353 {
1354         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1355
1356         vcpu->interrupt_window_open =
1357                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1358                  (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1359
1360         if (vcpu->interrupt_window_open && vcpu->irq_summary)
1361                 /*
1362                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1363                  */
1364                 kvm_do_inject_irq(vcpu);
1365
1366         /*
1367          * Interrupts blocked.  Wait for unblock.
1368          */
1369         if (!vcpu->interrupt_window_open &&
1370             (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1371                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1372         } else
1373                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1374 }
1375
1376 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1377                               struct kvm_run *kvm_run)
1378 {
1379         kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1380                                                   vcpu->irq_summary == 0);
1381         kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1382         kvm_run->cr8 = vcpu->cr8;
1383         kvm_run->apic_base = vcpu->apic_base;
1384 }
1385
1386 /*
1387  * Check if userspace requested an interrupt window, and that the
1388  * interrupt window is open.
1389  *
1390  * No need to exit to userspace if we already have an interrupt queued.
1391  */
1392 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1393                                           struct kvm_run *kvm_run)
1394 {
1395         return (!vcpu->irq_summary &&
1396                 kvm_run->request_interrupt_window &&
1397                 vcpu->interrupt_window_open &&
1398                 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1399 }
1400
1401 static void save_db_regs(unsigned long *db_regs)
1402 {
1403         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1404         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1405         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1406         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1407 }
1408
1409 static void load_db_regs(unsigned long *db_regs)
1410 {
1411         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1412         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1413         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1414         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1415 }
1416
1417 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1418 {
1419         u16 fs_selector;
1420         u16 gs_selector;
1421         u16 ldt_selector;
1422         int r;
1423
1424 again:
1425         if (!vcpu->mmio_read_completed)
1426                 do_interrupt_requests(vcpu, kvm_run);
1427
1428         clgi();
1429
1430         pre_svm_run(vcpu);
1431
1432         save_host_msrs(vcpu);
1433         fs_selector = read_fs();
1434         gs_selector = read_gs();
1435         ldt_selector = read_ldt();
1436         vcpu->svm->host_cr2 = kvm_read_cr2();
1437         vcpu->svm->host_dr6 = read_dr6();
1438         vcpu->svm->host_dr7 = read_dr7();
1439         vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1440
1441         if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1442                 write_dr7(0);
1443                 save_db_regs(vcpu->svm->host_db_regs);
1444                 load_db_regs(vcpu->svm->db_regs);
1445         }
1446
1447         fx_save(vcpu->host_fx_image);
1448         fx_restore(vcpu->guest_fx_image);
1449
1450         asm volatile (
1451 #ifdef CONFIG_X86_64
1452                 "push %%rbx; push %%rcx; push %%rdx;"
1453                 "push %%rsi; push %%rdi; push %%rbp;"
1454                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1455                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1456 #else
1457                 "push %%ebx; push %%ecx; push %%edx;"
1458                 "push %%esi; push %%edi; push %%ebp;"
1459 #endif
1460
1461 #ifdef CONFIG_X86_64
1462                 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1463                 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1464                 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1465                 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1466                 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1467                 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1468                 "mov %c[r8](%[vcpu]),  %%r8  \n\t"
1469                 "mov %c[r9](%[vcpu]),  %%r9  \n\t"
1470                 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1471                 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1472                 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1473                 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1474                 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1475                 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1476 #else
1477                 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1478                 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1479                 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1480                 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1481                 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1482                 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1483 #endif
1484
1485 #ifdef CONFIG_X86_64
1486                 /* Enter guest mode */
1487                 "push %%rax \n\t"
1488                 "mov %c[svm](%[vcpu]), %%rax \n\t"
1489                 "mov %c[vmcb](%%rax), %%rax \n\t"
1490                 SVM_VMLOAD "\n\t"
1491                 SVM_VMRUN "\n\t"
1492                 SVM_VMSAVE "\n\t"
1493                 "pop %%rax \n\t"
1494 #else
1495                 /* Enter guest mode */
1496                 "push %%eax \n\t"
1497                 "mov %c[svm](%[vcpu]), %%eax \n\t"
1498                 "mov %c[vmcb](%%eax), %%eax \n\t"
1499                 SVM_VMLOAD "\n\t"
1500                 SVM_VMRUN "\n\t"
1501                 SVM_VMSAVE "\n\t"
1502                 "pop %%eax \n\t"
1503 #endif
1504
1505                 /* Save guest registers, load host registers */
1506 #ifdef CONFIG_X86_64
1507                 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1508                 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1509                 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1510                 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1511                 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1512                 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1513                 "mov %%r8,  %c[r8](%[vcpu]) \n\t"
1514                 "mov %%r9,  %c[r9](%[vcpu]) \n\t"
1515                 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1516                 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1517                 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1518                 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1519                 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1520                 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1521
1522                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1523                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1524                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1525                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1526 #else
1527                 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1528                 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1529                 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1530                 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1531                 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1532                 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1533
1534                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1535                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1536 #endif
1537                 :
1538                 : [vcpu]"a"(vcpu),
1539                   [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1540                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1541                   [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1542                   [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1543                   [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1544                   [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1545                   [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1546                   [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1547 #ifdef CONFIG_X86_64
1548                   ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1549                   [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1550                   [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1551                   [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1552                   [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1553                   [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1554                   [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1555                   [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1556 #endif
1557                 : "cc", "memory" );
1558
1559         fx_save(vcpu->guest_fx_image);
1560         fx_restore(vcpu->host_fx_image);
1561
1562         if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1563                 load_db_regs(vcpu->svm->host_db_regs);
1564
1565         vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1566
1567         write_dr6(vcpu->svm->host_dr6);
1568         write_dr7(vcpu->svm->host_dr7);
1569         kvm_write_cr2(vcpu->svm->host_cr2);
1570
1571         load_fs(fs_selector);
1572         load_gs(gs_selector);
1573         load_ldt(ldt_selector);
1574         load_host_msrs(vcpu);
1575
1576         reload_tss(vcpu);
1577
1578         /*
1579          * Profile KVM exit RIPs:
1580          */
1581         if (unlikely(prof_on == KVM_PROFILING))
1582                 profile_hit(KVM_PROFILING,
1583                         (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1584
1585         stgi();
1586
1587         kvm_reput_irq(vcpu);
1588
1589         vcpu->svm->next_rip = 0;
1590
1591         if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1592                 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1593                 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
1594                 post_kvm_run_save(vcpu, kvm_run);
1595                 return 0;
1596         }
1597
1598         r = handle_exit(vcpu, kvm_run);
1599         if (r > 0) {
1600                 if (signal_pending(current)) {
1601                         ++kvm_stat.signal_exits;
1602                         post_kvm_run_save(vcpu, kvm_run);
1603                         return -EINTR;
1604                 }
1605
1606                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1607                         ++kvm_stat.request_irq_exits;
1608                         post_kvm_run_save(vcpu, kvm_run);
1609                         return -EINTR;
1610                 }
1611                 kvm_resched(vcpu);
1612                 goto again;
1613         }
1614         post_kvm_run_save(vcpu, kvm_run);
1615         return r;
1616 }
1617
1618 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1619 {
1620         force_new_asid(vcpu);
1621 }
1622
1623 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1624 {
1625         vcpu->svm->vmcb->save.cr3 = root;
1626         force_new_asid(vcpu);
1627 }
1628
1629 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1630                                   unsigned long  addr,
1631                                   uint32_t err_code)
1632 {
1633         uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1634
1635         ++kvm_stat.pf_guest;
1636
1637         if (is_page_fault(exit_int_info)) {
1638
1639                 vcpu->svm->vmcb->control.event_inj_err = 0;
1640                 vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1641                                                         SVM_EVTINJ_VALID_ERR |
1642                                                         SVM_EVTINJ_TYPE_EXEPT |
1643                                                         DF_VECTOR;
1644                 return;
1645         }
1646         vcpu->cr2 = addr;
1647         vcpu->svm->vmcb->save.cr2 = addr;
1648         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1649                                                 SVM_EVTINJ_VALID_ERR |
1650                                                 SVM_EVTINJ_TYPE_EXEPT |
1651                                                 PF_VECTOR;
1652         vcpu->svm->vmcb->control.event_inj_err = err_code;
1653 }
1654
1655
1656 static int is_disabled(void)
1657 {
1658         return 0;
1659 }
1660
1661 static struct kvm_arch_ops svm_arch_ops = {
1662         .cpu_has_kvm_support = has_svm,
1663         .disabled_by_bios = is_disabled,
1664         .hardware_setup = svm_hardware_setup,
1665         .hardware_unsetup = svm_hardware_unsetup,
1666         .hardware_enable = svm_hardware_enable,
1667         .hardware_disable = svm_hardware_disable,
1668
1669         .vcpu_create = svm_create_vcpu,
1670         .vcpu_free = svm_free_vcpu,
1671
1672         .vcpu_load = svm_vcpu_load,
1673         .vcpu_put = svm_vcpu_put,
1674
1675         .set_guest_debug = svm_guest_debug,
1676         .get_msr = svm_get_msr,
1677         .set_msr = svm_set_msr,
1678         .get_segment_base = svm_get_segment_base,
1679         .get_segment = svm_get_segment,
1680         .set_segment = svm_set_segment,
1681         .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1682         .decache_cr0_cr4_guest_bits = svm_decache_cr0_cr4_guest_bits,
1683         .set_cr0 = svm_set_cr0,
1684         .set_cr0_no_modeswitch = svm_set_cr0,
1685         .set_cr3 = svm_set_cr3,
1686         .set_cr4 = svm_set_cr4,
1687         .set_efer = svm_set_efer,
1688         .get_idt = svm_get_idt,
1689         .set_idt = svm_set_idt,
1690         .get_gdt = svm_get_gdt,
1691         .set_gdt = svm_set_gdt,
1692         .get_dr = svm_get_dr,
1693         .set_dr = svm_set_dr,
1694         .cache_regs = svm_cache_regs,
1695         .decache_regs = svm_decache_regs,
1696         .get_rflags = svm_get_rflags,
1697         .set_rflags = svm_set_rflags,
1698
1699         .invlpg = svm_invlpg,
1700         .tlb_flush = svm_flush_tlb,
1701         .inject_page_fault = svm_inject_page_fault,
1702
1703         .inject_gp = svm_inject_gp,
1704
1705         .run = svm_vcpu_run,
1706         .skip_emulated_instruction = skip_emulated_instruction,
1707         .vcpu_setup = svm_vcpu_setup,
1708 };
1709
1710 static int __init svm_init(void)
1711 {
1712         return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1713 }
1714
1715 static void __exit svm_exit(void)
1716 {
1717         kvm_exit_arch();
1718 }
1719
1720 module_init(svm_init)
1721 module_exit(svm_exit)