Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[pandora-kernel.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         /*
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
547         save->dr6 = 0xffff0ff0;
548         save->dr7 = 0x400;
549         save->rflags = 2;
550         save->rip = 0x0000fff0;
551
552         /*
553          * cr0 val on cpu init should be 0x60000010, we enable cpu
554          * cache by default. the orderly way is to enable cache in bios.
555          */
556         save->cr0 = 0x00000010 | CR0_PG_MASK;
557         save->cr4 = CR4_PAE_MASK;
558         /* rdx = ?? */
559 }
560
561 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
562 {
563         struct page *page;
564         int r;
565
566         r = -ENOMEM;
567         vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
568         if (!vcpu->svm)
569                 goto out1;
570         page = alloc_page(GFP_KERNEL);
571         if (!page)
572                 goto out2;
573
574         vcpu->svm->vmcb = page_address(page);
575         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
576         vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
577         vcpu->svm->cr0 = 0x00000010;
578         vcpu->svm->asid_generation = 0;
579         memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
580         init_vmcb(vcpu->svm->vmcb);
581
582         fx_init(vcpu);
583
584         return 0;
585
586 out2:
587         kfree(vcpu->svm);
588 out1:
589         return r;
590 }
591
592 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
593 {
594         if (!vcpu->svm)
595                 return;
596         if (vcpu->svm->vmcb)
597                 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
598         kfree(vcpu->svm);
599 }
600
601 static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
602 {
603         get_cpu();
604         return vcpu;
605 }
606
607 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
608 {
609         put_cpu();
610 }
611
612 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
613 {
614 }
615
616 static void svm_cache_regs(struct kvm_vcpu *vcpu)
617 {
618         vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
619         vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
620         vcpu->rip = vcpu->svm->vmcb->save.rip;
621 }
622
623 static void svm_decache_regs(struct kvm_vcpu *vcpu)
624 {
625         vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
626         vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
627         vcpu->svm->vmcb->save.rip = vcpu->rip;
628 }
629
630 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
631 {
632         return vcpu->svm->vmcb->save.rflags;
633 }
634
635 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
636 {
637         vcpu->svm->vmcb->save.rflags = rflags;
638 }
639
640 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
641 {
642         struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
643
644         switch (seg) {
645         case VCPU_SREG_CS: return &save->cs;
646         case VCPU_SREG_DS: return &save->ds;
647         case VCPU_SREG_ES: return &save->es;
648         case VCPU_SREG_FS: return &save->fs;
649         case VCPU_SREG_GS: return &save->gs;
650         case VCPU_SREG_SS: return &save->ss;
651         case VCPU_SREG_TR: return &save->tr;
652         case VCPU_SREG_LDTR: return &save->ldtr;
653         }
654         BUG();
655         return NULL;
656 }
657
658 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
659 {
660         struct vmcb_seg *s = svm_seg(vcpu, seg);
661
662         return s->base;
663 }
664
665 static void svm_get_segment(struct kvm_vcpu *vcpu,
666                             struct kvm_segment *var, int seg)
667 {
668         struct vmcb_seg *s = svm_seg(vcpu, seg);
669
670         var->base = s->base;
671         var->limit = s->limit;
672         var->selector = s->selector;
673         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
674         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
675         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
676         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
677         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
678         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
679         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
680         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
681         var->unusable = !var->present;
682 }
683
684 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
685 {
686         struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
687
688         *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
689         *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
690 }
691
692 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
693 {
694         dt->limit = vcpu->svm->vmcb->save.idtr.limit;
695         dt->base = vcpu->svm->vmcb->save.idtr.base;
696 }
697
698 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
699 {
700         vcpu->svm->vmcb->save.idtr.limit = dt->limit;
701         vcpu->svm->vmcb->save.idtr.base = dt->base ;
702 }
703
704 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
705 {
706         dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
707         dt->base = vcpu->svm->vmcb->save.gdtr.base;
708 }
709
710 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
711 {
712         vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
713         vcpu->svm->vmcb->save.gdtr.base = dt->base ;
714 }
715
716 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
717 {
718 }
719
720 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
721 {
722 #ifdef CONFIG_X86_64
723         if (vcpu->shadow_efer & KVM_EFER_LME) {
724                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
725                         vcpu->shadow_efer |= KVM_EFER_LMA;
726                         vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
727                 }
728
729                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
730                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
731                         vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
732                 }
733         }
734 #endif
735         vcpu->svm->cr0 = cr0;
736         vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK | CR0_WP_MASK;
737         vcpu->cr0 = cr0;
738 }
739
740 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
741 {
742        vcpu->cr4 = cr4;
743        vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
744 }
745
746 static void svm_set_segment(struct kvm_vcpu *vcpu,
747                             struct kvm_segment *var, int seg)
748 {
749         struct vmcb_seg *s = svm_seg(vcpu, seg);
750
751         s->base = var->base;
752         s->limit = var->limit;
753         s->selector = var->selector;
754         if (var->unusable)
755                 s->attrib = 0;
756         else {
757                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
758                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
759                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
760                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
761                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
762                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
763                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
764                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
765         }
766         if (seg == VCPU_SREG_CS)
767                 vcpu->svm->vmcb->save.cpl
768                         = (vcpu->svm->vmcb->save.cs.attrib
769                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
770
771 }
772
773 /* FIXME:
774
775         vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
776         vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
777
778 */
779
780 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
781 {
782         return -EOPNOTSUPP;
783 }
784
785 static void load_host_msrs(struct kvm_vcpu *vcpu)
786 {
787         int i;
788
789         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
790                 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
791 }
792
793 static void save_host_msrs(struct kvm_vcpu *vcpu)
794 {
795         int i;
796
797         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
798                 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
799 }
800
801 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
802 {
803         if (svm_data->next_asid > svm_data->max_asid) {
804                 ++svm_data->asid_generation;
805                 svm_data->next_asid = 1;
806                 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
807         }
808
809         vcpu->cpu = svm_data->cpu;
810         vcpu->svm->asid_generation = svm_data->asid_generation;
811         vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
812 }
813
814 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
815 {
816         invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
817 }
818
819 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
820 {
821         return vcpu->svm->db_regs[dr];
822 }
823
824 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
825                        int *exception)
826 {
827         *exception = 0;
828
829         if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
830                 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
831                 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
832                 *exception = DB_VECTOR;
833                 return;
834         }
835
836         switch (dr) {
837         case 0 ... 3:
838                 vcpu->svm->db_regs[dr] = value;
839                 return;
840         case 4 ... 5:
841                 if (vcpu->cr4 & CR4_DE_MASK) {
842                         *exception = UD_VECTOR;
843                         return;
844                 }
845         case 7: {
846                 if (value & ~((1ULL << 32) - 1)) {
847                         *exception = GP_VECTOR;
848                         return;
849                 }
850                 vcpu->svm->vmcb->save.dr7 = value;
851                 return;
852         }
853         default:
854                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
855                        __FUNCTION__, dr);
856                 *exception = UD_VECTOR;
857                 return;
858         }
859 }
860
861 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
862 {
863         u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
864         u64 fault_address;
865         u32 error_code;
866         enum emulation_result er;
867         int r;
868
869         if (is_external_interrupt(exit_int_info))
870                 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
871
872         spin_lock(&vcpu->kvm->lock);
873
874         fault_address  = vcpu->svm->vmcb->control.exit_info_2;
875         error_code = vcpu->svm->vmcb->control.exit_info_1;
876         r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
877         if (r < 0) {
878                 spin_unlock(&vcpu->kvm->lock);
879                 return r;
880         }
881         if (!r) {
882                 spin_unlock(&vcpu->kvm->lock);
883                 return 1;
884         }
885         er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
886         spin_unlock(&vcpu->kvm->lock);
887
888         switch (er) {
889         case EMULATE_DONE:
890                 return 1;
891         case EMULATE_DO_MMIO:
892                 ++kvm_stat.mmio_exits;
893                 kvm_run->exit_reason = KVM_EXIT_MMIO;
894                 return 0;
895         case EMULATE_FAIL:
896                 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
897                 break;
898         default:
899                 BUG();
900         }
901
902         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
903         return 0;
904 }
905
906 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
907 {
908         /*
909          * VMCB is undefined after a SHUTDOWN intercept
910          * so reinitialize it.
911          */
912         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
913         init_vmcb(vcpu->svm->vmcb);
914
915         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
916         return 0;
917 }
918
919 static int io_get_override(struct kvm_vcpu *vcpu,
920                           struct vmcb_seg **seg,
921                           int *addr_override)
922 {
923         u8 inst[MAX_INST_SIZE];
924         unsigned ins_length;
925         gva_t rip;
926         int i;
927
928         rip =  vcpu->svm->vmcb->save.rip;
929         ins_length = vcpu->svm->next_rip - rip;
930         rip += vcpu->svm->vmcb->save.cs.base;
931
932         if (ins_length > MAX_INST_SIZE)
933                 printk(KERN_DEBUG
934                        "%s: inst length err, cs base 0x%llx rip 0x%llx "
935                        "next rip 0x%llx ins_length %u\n",
936                        __FUNCTION__,
937                        vcpu->svm->vmcb->save.cs.base,
938                        vcpu->svm->vmcb->save.rip,
939                        vcpu->svm->vmcb->control.exit_info_2,
940                        ins_length);
941
942         if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
943                 /* #PF */
944                 return 0;
945
946         *addr_override = 0;
947         *seg = NULL;
948         for (i = 0; i < ins_length; i++)
949                 switch (inst[i]) {
950                 case 0xf0:
951                 case 0xf2:
952                 case 0xf3:
953                 case 0x66:
954                         continue;
955                 case 0x67:
956                         *addr_override = 1;
957                         continue;
958                 case 0x2e:
959                         *seg = &vcpu->svm->vmcb->save.cs;
960                         continue;
961                 case 0x36:
962                         *seg = &vcpu->svm->vmcb->save.ss;
963                         continue;
964                 case 0x3e:
965                         *seg = &vcpu->svm->vmcb->save.ds;
966                         continue;
967                 case 0x26:
968                         *seg = &vcpu->svm->vmcb->save.es;
969                         continue;
970                 case 0x64:
971                         *seg = &vcpu->svm->vmcb->save.fs;
972                         continue;
973                 case 0x65:
974                         *seg = &vcpu->svm->vmcb->save.gs;
975                         continue;
976                 default:
977                         return 1;
978                 }
979         printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
980         return 0;
981 }
982
983 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
984 {
985         unsigned long addr_mask;
986         unsigned long *reg;
987         struct vmcb_seg *seg;
988         int addr_override;
989         struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
990         u16 cs_attrib = save_area->cs.attrib;
991         unsigned addr_size = get_addr_size(vcpu);
992
993         if (!io_get_override(vcpu, &seg, &addr_override))
994                 return 0;
995
996         if (addr_override)
997                 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
998
999         if (ins) {
1000                 reg = &vcpu->regs[VCPU_REGS_RDI];
1001                 seg = &vcpu->svm->vmcb->save.es;
1002         } else {
1003                 reg = &vcpu->regs[VCPU_REGS_RSI];
1004                 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
1005         }
1006
1007         addr_mask = ~0ULL >> (64 - (addr_size * 8));
1008
1009         if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1010             !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1011                 *address = (*reg & addr_mask);
1012                 return addr_mask;
1013         }
1014
1015         if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1016                 svm_inject_gp(vcpu, 0);
1017                 return 0;
1018         }
1019
1020         *address = (*reg & addr_mask) + seg->base;
1021         return addr_mask;
1022 }
1023
1024 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1025 {
1026         u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1027         int _in = io_info & SVM_IOIO_TYPE_MASK;
1028
1029         ++kvm_stat.io_exits;
1030
1031         vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1032
1033         kvm_run->exit_reason = KVM_EXIT_IO;
1034         kvm_run->io.port = io_info >> 16;
1035         kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1036         kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1037         kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1038         kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1039
1040         if (kvm_run->io.string) {
1041                 unsigned addr_mask;
1042
1043                 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1044                 if (!addr_mask) {
1045                         printk(KERN_DEBUG "%s: get io address failed\n", __FUNCTION__);
1046                         return 1;
1047                 }
1048
1049                 if (kvm_run->io.rep) {
1050                         kvm_run->io.count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1051                         kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1052                                                    & X86_EFLAGS_DF) != 0;
1053                 }
1054         } else {
1055                 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1056         }
1057         return 0;
1058 }
1059
1060
1061 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1062 {
1063         return 1;
1064 }
1065
1066 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1067 {
1068         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1069         skip_emulated_instruction(vcpu);
1070         if (vcpu->irq_summary)
1071                 return 1;
1072
1073         kvm_run->exit_reason = KVM_EXIT_HLT;
1074         ++kvm_stat.halt_exits;
1075         return 0;
1076 }
1077
1078 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1079 {
1080         inject_ud(vcpu);
1081         return 1;
1082 }
1083
1084 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1085 {
1086         printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1087         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1088         return 0;
1089 }
1090
1091 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1092 {
1093         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1094         kvm_run->exit_reason = KVM_EXIT_CPUID;
1095         return 0;
1096 }
1097
1098 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1099 {
1100         if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1101                 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1102         return 1;
1103 }
1104
1105 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1106 {
1107         switch (ecx) {
1108         case MSR_IA32_TIME_STAMP_COUNTER: {
1109                 u64 tsc;
1110
1111                 rdtscll(tsc);
1112                 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1113                 break;
1114         }
1115         case MSR_K6_STAR:
1116                 *data = vcpu->svm->vmcb->save.star;
1117                 break;
1118 #ifdef CONFIG_X86_64
1119         case MSR_LSTAR:
1120                 *data = vcpu->svm->vmcb->save.lstar;
1121                 break;
1122         case MSR_CSTAR:
1123                 *data = vcpu->svm->vmcb->save.cstar;
1124                 break;
1125         case MSR_KERNEL_GS_BASE:
1126                 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1127                 break;
1128         case MSR_SYSCALL_MASK:
1129                 *data = vcpu->svm->vmcb->save.sfmask;
1130                 break;
1131 #endif
1132         case MSR_IA32_SYSENTER_CS:
1133                 *data = vcpu->svm->vmcb->save.sysenter_cs;
1134                 break;
1135         case MSR_IA32_SYSENTER_EIP:
1136                 *data = vcpu->svm->vmcb->save.sysenter_eip;
1137                 break;
1138         case MSR_IA32_SYSENTER_ESP:
1139                 *data = vcpu->svm->vmcb->save.sysenter_esp;
1140                 break;
1141         default:
1142                 return kvm_get_msr_common(vcpu, ecx, data);
1143         }
1144         return 0;
1145 }
1146
1147 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1148 {
1149         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1150         u64 data;
1151
1152         if (svm_get_msr(vcpu, ecx, &data))
1153                 svm_inject_gp(vcpu, 0);
1154         else {
1155                 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1156                 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1157                 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1158                 skip_emulated_instruction(vcpu);
1159         }
1160         return 1;
1161 }
1162
1163 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1164 {
1165         switch (ecx) {
1166         case MSR_IA32_TIME_STAMP_COUNTER: {
1167                 u64 tsc;
1168
1169                 rdtscll(tsc);
1170                 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1171                 break;
1172         }
1173         case MSR_K6_STAR:
1174                 vcpu->svm->vmcb->save.star = data;
1175                 break;
1176 #ifdef CONFIG_X86_64
1177         case MSR_LSTAR:
1178                 vcpu->svm->vmcb->save.lstar = data;
1179                 break;
1180         case MSR_CSTAR:
1181                 vcpu->svm->vmcb->save.cstar = data;
1182                 break;
1183         case MSR_KERNEL_GS_BASE:
1184                 vcpu->svm->vmcb->save.kernel_gs_base = data;
1185                 break;
1186         case MSR_SYSCALL_MASK:
1187                 vcpu->svm->vmcb->save.sfmask = data;
1188                 break;
1189 #endif
1190         case MSR_IA32_SYSENTER_CS:
1191                 vcpu->svm->vmcb->save.sysenter_cs = data;
1192                 break;
1193         case MSR_IA32_SYSENTER_EIP:
1194                 vcpu->svm->vmcb->save.sysenter_eip = data;
1195                 break;
1196         case MSR_IA32_SYSENTER_ESP:
1197                 vcpu->svm->vmcb->save.sysenter_esp = data;
1198                 break;
1199         default:
1200                 return kvm_set_msr_common(vcpu, ecx, data);
1201         }
1202         return 0;
1203 }
1204
1205 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1206 {
1207         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1208         u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1209                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1210         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1211         if (svm_set_msr(vcpu, ecx, data))
1212                 svm_inject_gp(vcpu, 0);
1213         else
1214                 skip_emulated_instruction(vcpu);
1215         return 1;
1216 }
1217
1218 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1219 {
1220         if (vcpu->svm->vmcb->control.exit_info_1)
1221                 return wrmsr_interception(vcpu, kvm_run);
1222         else
1223                 return rdmsr_interception(vcpu, kvm_run);
1224 }
1225
1226 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1227                                    struct kvm_run *kvm_run)
1228 {
1229         /*
1230          * If the user space waits to inject interrupts, exit as soon as
1231          * possible
1232          */
1233         if (kvm_run->request_interrupt_window &&
1234             !vcpu->irq_summary) {
1235                 ++kvm_stat.irq_window_exits;
1236                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1237                 return 0;
1238         }
1239
1240         return 1;
1241 }
1242
1243 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1244                                       struct kvm_run *kvm_run) = {
1245         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1246         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1247         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1248         /* for now: */
1249         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1250         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1251         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1252         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1253         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1254         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1255         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1256         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1257         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1258         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1259         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1260         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1261         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1262         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1263         [SVM_EXIT_INTR]                         = nop_on_interception,
1264         [SVM_EXIT_NMI]                          = nop_on_interception,
1265         [SVM_EXIT_SMI]                          = nop_on_interception,
1266         [SVM_EXIT_INIT]                         = nop_on_interception,
1267         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1268         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1269         [SVM_EXIT_CPUID]                        = cpuid_interception,
1270         [SVM_EXIT_HLT]                          = halt_interception,
1271         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1272         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1273         [SVM_EXIT_IOIO]                         = io_interception,
1274         [SVM_EXIT_MSR]                          = msr_interception,
1275         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1276         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1277         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1278         [SVM_EXIT_VMMCALL]                      = invalid_op_interception,
1279         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1280         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1281         [SVM_EXIT_STGI]                         = invalid_op_interception,
1282         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1283         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1284 };
1285
1286
1287 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1288 {
1289         u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1290
1291         kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1292
1293         if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1294             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1295                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1296                        "exit_code 0x%x\n",
1297                        __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1298                        exit_code);
1299
1300         if (exit_code >= sizeof(svm_exit_handlers) / sizeof(*svm_exit_handlers)
1301             || svm_exit_handlers[exit_code] == 0) {
1302                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1303                 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1304                        __FUNCTION__,
1305                        exit_code,
1306                        vcpu->svm->vmcb->save.rip,
1307                        vcpu->cr0,
1308                        vcpu->svm->vmcb->save.rflags);
1309                 return 0;
1310         }
1311
1312         return svm_exit_handlers[exit_code](vcpu, kvm_run);
1313 }
1314
1315 static void reload_tss(struct kvm_vcpu *vcpu)
1316 {
1317         int cpu = raw_smp_processor_id();
1318
1319         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1320         svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1321         load_TR_desc();
1322 }
1323
1324 static void pre_svm_run(struct kvm_vcpu *vcpu)
1325 {
1326         int cpu = raw_smp_processor_id();
1327
1328         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1329
1330         vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1331         if (vcpu->cpu != cpu ||
1332             vcpu->svm->asid_generation != svm_data->asid_generation)
1333                 new_asid(vcpu, svm_data);
1334 }
1335
1336
1337 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1338 {
1339         struct vmcb_control_area *control;
1340
1341         control = &vcpu->svm->vmcb->control;
1342         control->int_vector = pop_irq(vcpu);
1343         control->int_ctl &= ~V_INTR_PRIO_MASK;
1344         control->int_ctl |= V_IRQ_MASK |
1345                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1346 }
1347
1348 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1349 {
1350         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1351
1352         if (control->int_ctl & V_IRQ_MASK) {
1353                 control->int_ctl &= ~V_IRQ_MASK;
1354                 push_irq(vcpu, control->int_vector);
1355         }
1356
1357         vcpu->interrupt_window_open =
1358                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1359 }
1360
1361 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1362                                        struct kvm_run *kvm_run)
1363 {
1364         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1365
1366         vcpu->interrupt_window_open =
1367                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1368                  (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1369
1370         if (vcpu->interrupt_window_open && vcpu->irq_summary)
1371                 /*
1372                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1373                  */
1374                 kvm_do_inject_irq(vcpu);
1375
1376         /*
1377          * Interrupts blocked.  Wait for unblock.
1378          */
1379         if (!vcpu->interrupt_window_open &&
1380             (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1381                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1382         } else
1383                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1384 }
1385
1386 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1387                               struct kvm_run *kvm_run)
1388 {
1389         kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1390                                                   vcpu->irq_summary == 0);
1391         kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1392         kvm_run->cr8 = vcpu->cr8;
1393         kvm_run->apic_base = vcpu->apic_base;
1394 }
1395
1396 /*
1397  * Check if userspace requested an interrupt window, and that the
1398  * interrupt window is open.
1399  *
1400  * No need to exit to userspace if we already have an interrupt queued.
1401  */
1402 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1403                                           struct kvm_run *kvm_run)
1404 {
1405         return (!vcpu->irq_summary &&
1406                 kvm_run->request_interrupt_window &&
1407                 vcpu->interrupt_window_open &&
1408                 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1409 }
1410
1411 static void save_db_regs(unsigned long *db_regs)
1412 {
1413         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1414         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1415         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1416         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1417 }
1418
1419 static void load_db_regs(unsigned long *db_regs)
1420 {
1421         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1422         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1423         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1424         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1425 }
1426
1427 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1428 {
1429         u16 fs_selector;
1430         u16 gs_selector;
1431         u16 ldt_selector;
1432         int r;
1433
1434 again:
1435         if (!vcpu->mmio_read_completed)
1436                 do_interrupt_requests(vcpu, kvm_run);
1437
1438         clgi();
1439
1440         pre_svm_run(vcpu);
1441
1442         save_host_msrs(vcpu);
1443         fs_selector = read_fs();
1444         gs_selector = read_gs();
1445         ldt_selector = read_ldt();
1446         vcpu->svm->host_cr2 = kvm_read_cr2();
1447         vcpu->svm->host_dr6 = read_dr6();
1448         vcpu->svm->host_dr7 = read_dr7();
1449         vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1450
1451         if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1452                 write_dr7(0);
1453                 save_db_regs(vcpu->svm->host_db_regs);
1454                 load_db_regs(vcpu->svm->db_regs);
1455         }
1456
1457         fx_save(vcpu->host_fx_image);
1458         fx_restore(vcpu->guest_fx_image);
1459
1460         asm volatile (
1461 #ifdef CONFIG_X86_64
1462                 "push %%rbx; push %%rcx; push %%rdx;"
1463                 "push %%rsi; push %%rdi; push %%rbp;"
1464                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1465                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1466 #else
1467                 "push %%ebx; push %%ecx; push %%edx;"
1468                 "push %%esi; push %%edi; push %%ebp;"
1469 #endif
1470
1471 #ifdef CONFIG_X86_64
1472                 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1473                 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1474                 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1475                 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1476                 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1477                 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1478                 "mov %c[r8](%[vcpu]),  %%r8  \n\t"
1479                 "mov %c[r9](%[vcpu]),  %%r9  \n\t"
1480                 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1481                 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1482                 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1483                 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1484                 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1485                 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1486 #else
1487                 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1488                 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1489                 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1490                 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1491                 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1492                 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1493 #endif
1494
1495 #ifdef CONFIG_X86_64
1496                 /* Enter guest mode */
1497                 "push %%rax \n\t"
1498                 "mov %c[svm](%[vcpu]), %%rax \n\t"
1499                 "mov %c[vmcb](%%rax), %%rax \n\t"
1500                 SVM_VMLOAD "\n\t"
1501                 SVM_VMRUN "\n\t"
1502                 SVM_VMSAVE "\n\t"
1503                 "pop %%rax \n\t"
1504 #else
1505                 /* Enter guest mode */
1506                 "push %%eax \n\t"
1507                 "mov %c[svm](%[vcpu]), %%eax \n\t"
1508                 "mov %c[vmcb](%%eax), %%eax \n\t"
1509                 SVM_VMLOAD "\n\t"
1510                 SVM_VMRUN "\n\t"
1511                 SVM_VMSAVE "\n\t"
1512                 "pop %%eax \n\t"
1513 #endif
1514
1515                 /* Save guest registers, load host registers */
1516 #ifdef CONFIG_X86_64
1517                 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1518                 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1519                 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1520                 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1521                 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1522                 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1523                 "mov %%r8,  %c[r8](%[vcpu]) \n\t"
1524                 "mov %%r9,  %c[r9](%[vcpu]) \n\t"
1525                 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1526                 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1527                 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1528                 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1529                 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1530                 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1531
1532                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1533                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1534                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1535                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1536 #else
1537                 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1538                 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1539                 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1540                 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1541                 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1542                 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1543
1544                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1545                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1546 #endif
1547                 :
1548                 : [vcpu]"a"(vcpu),
1549                   [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1550                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1551                   [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1552                   [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1553                   [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1554                   [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1555                   [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1556                   [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1557 #ifdef CONFIG_X86_64
1558                   ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1559                   [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1560                   [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1561                   [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1562                   [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1563                   [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1564                   [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1565                   [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1566 #endif
1567                 : "cc", "memory" );
1568
1569         fx_save(vcpu->guest_fx_image);
1570         fx_restore(vcpu->host_fx_image);
1571
1572         if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1573                 load_db_regs(vcpu->svm->host_db_regs);
1574
1575         vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1576
1577         write_dr6(vcpu->svm->host_dr6);
1578         write_dr7(vcpu->svm->host_dr7);
1579         kvm_write_cr2(vcpu->svm->host_cr2);
1580
1581         load_fs(fs_selector);
1582         load_gs(gs_selector);
1583         load_ldt(ldt_selector);
1584         load_host_msrs(vcpu);
1585
1586         reload_tss(vcpu);
1587
1588         /*
1589          * Profile KVM exit RIPs:
1590          */
1591         if (unlikely(prof_on == KVM_PROFILING))
1592                 profile_hit(KVM_PROFILING,
1593                         (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1594
1595         stgi();
1596
1597         kvm_reput_irq(vcpu);
1598
1599         vcpu->svm->next_rip = 0;
1600
1601         if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1602                 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1603                 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
1604                 post_kvm_run_save(vcpu, kvm_run);
1605                 return 0;
1606         }
1607
1608         r = handle_exit(vcpu, kvm_run);
1609         if (r > 0) {
1610                 if (signal_pending(current)) {
1611                         ++kvm_stat.signal_exits;
1612                         post_kvm_run_save(vcpu, kvm_run);
1613                         return -EINTR;
1614                 }
1615
1616                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1617                         ++kvm_stat.request_irq_exits;
1618                         post_kvm_run_save(vcpu, kvm_run);
1619                         return -EINTR;
1620                 }
1621                 kvm_resched(vcpu);
1622                 goto again;
1623         }
1624         post_kvm_run_save(vcpu, kvm_run);
1625         return r;
1626 }
1627
1628 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1629 {
1630         force_new_asid(vcpu);
1631 }
1632
1633 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1634 {
1635         vcpu->svm->vmcb->save.cr3 = root;
1636         force_new_asid(vcpu);
1637 }
1638
1639 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1640                                   unsigned long  addr,
1641                                   uint32_t err_code)
1642 {
1643         uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1644
1645         ++kvm_stat.pf_guest;
1646
1647         if (is_page_fault(exit_int_info)) {
1648
1649                 vcpu->svm->vmcb->control.event_inj_err = 0;
1650                 vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1651                                                         SVM_EVTINJ_VALID_ERR |
1652                                                         SVM_EVTINJ_TYPE_EXEPT |
1653                                                         DF_VECTOR;
1654                 return;
1655         }
1656         vcpu->cr2 = addr;
1657         vcpu->svm->vmcb->save.cr2 = addr;
1658         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1659                                                 SVM_EVTINJ_VALID_ERR |
1660                                                 SVM_EVTINJ_TYPE_EXEPT |
1661                                                 PF_VECTOR;
1662         vcpu->svm->vmcb->control.event_inj_err = err_code;
1663 }
1664
1665
1666 static int is_disabled(void)
1667 {
1668         return 0;
1669 }
1670
1671 static struct kvm_arch_ops svm_arch_ops = {
1672         .cpu_has_kvm_support = has_svm,
1673         .disabled_by_bios = is_disabled,
1674         .hardware_setup = svm_hardware_setup,
1675         .hardware_unsetup = svm_hardware_unsetup,
1676         .hardware_enable = svm_hardware_enable,
1677         .hardware_disable = svm_hardware_disable,
1678
1679         .vcpu_create = svm_create_vcpu,
1680         .vcpu_free = svm_free_vcpu,
1681
1682         .vcpu_load = svm_vcpu_load,
1683         .vcpu_put = svm_vcpu_put,
1684         .vcpu_decache = svm_vcpu_decache,
1685
1686         .set_guest_debug = svm_guest_debug,
1687         .get_msr = svm_get_msr,
1688         .set_msr = svm_set_msr,
1689         .get_segment_base = svm_get_segment_base,
1690         .get_segment = svm_get_segment,
1691         .set_segment = svm_set_segment,
1692         .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1693         .decache_cr0_cr4_guest_bits = svm_decache_cr0_cr4_guest_bits,
1694         .set_cr0 = svm_set_cr0,
1695         .set_cr0_no_modeswitch = svm_set_cr0,
1696         .set_cr3 = svm_set_cr3,
1697         .set_cr4 = svm_set_cr4,
1698         .set_efer = svm_set_efer,
1699         .get_idt = svm_get_idt,
1700         .set_idt = svm_set_idt,
1701         .get_gdt = svm_get_gdt,
1702         .set_gdt = svm_set_gdt,
1703         .get_dr = svm_get_dr,
1704         .set_dr = svm_set_dr,
1705         .cache_regs = svm_cache_regs,
1706         .decache_regs = svm_decache_regs,
1707         .get_rflags = svm_get_rflags,
1708         .set_rflags = svm_set_rflags,
1709
1710         .invlpg = svm_invlpg,
1711         .tlb_flush = svm_flush_tlb,
1712         .inject_page_fault = svm_inject_page_fault,
1713
1714         .inject_gp = svm_inject_gp,
1715
1716         .run = svm_vcpu_run,
1717         .skip_emulated_instruction = skip_emulated_instruction,
1718         .vcpu_setup = svm_vcpu_setup,
1719 };
1720
1721 static int __init svm_init(void)
1722 {
1723         return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1724 }
1725
1726 static void __exit svm_exit(void)
1727 {
1728         kvm_exit_arch();
1729 }
1730
1731 module_init(svm_init)
1732 module_exit(svm_exit)