KVM: SVM: Fix SMP with kernel apic
[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 "kvm_svm.h"
18 #include "x86_emulate.h"
19 #include "irq.h"
20
21 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/vmalloc.h>
24 #include <linux/highmem.h>
25 #include <linux/sched.h>
26
27 #include <asm/desc.h>
28
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
31
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
34
35 #define DB_VECTOR 1
36 #define UD_VECTOR 6
37 #define GP_VECTOR 13
38
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
41
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
44
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
47
48 #define SVM_FEATURE_NPT  (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
51
52 static void kvm_reput_irq(struct vcpu_svm *svm);
53
54 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
55 {
56         return container_of(vcpu, struct vcpu_svm, vcpu);
57 }
58
59 unsigned long iopm_base;
60 unsigned long msrpm_base;
61
62 struct kvm_ldttss_desc {
63         u16 limit0;
64         u16 base0;
65         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
66         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
67         u32 base3;
68         u32 zero1;
69 } __attribute__((packed));
70
71 struct svm_cpu_data {
72         int cpu;
73
74         u64 asid_generation;
75         u32 max_asid;
76         u32 next_asid;
77         struct kvm_ldttss_desc *tss_desc;
78
79         struct page *save_area;
80 };
81
82 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
83 static uint32_t svm_features;
84
85 struct svm_init_data {
86         int cpu;
87         int r;
88 };
89
90 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
91
92 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
93 #define MSRS_RANGE_SIZE 2048
94 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
95
96 #define MAX_INST_SIZE 15
97
98 static inline u32 svm_has(u32 feat)
99 {
100         return svm_features & feat;
101 }
102
103 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
104 {
105         int word_index = __ffs(vcpu->irq_summary);
106         int bit_index = __ffs(vcpu->irq_pending[word_index]);
107         int irq = word_index * BITS_PER_LONG + bit_index;
108
109         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
110         if (!vcpu->irq_pending[word_index])
111                 clear_bit(word_index, &vcpu->irq_summary);
112         return irq;
113 }
114
115 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
116 {
117         set_bit(irq, vcpu->irq_pending);
118         set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
119 }
120
121 static inline void clgi(void)
122 {
123         asm volatile (SVM_CLGI);
124 }
125
126 static inline void stgi(void)
127 {
128         asm volatile (SVM_STGI);
129 }
130
131 static inline void invlpga(unsigned long addr, u32 asid)
132 {
133         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
134 }
135
136 static inline unsigned long kvm_read_cr2(void)
137 {
138         unsigned long cr2;
139
140         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
141         return cr2;
142 }
143
144 static inline void kvm_write_cr2(unsigned long val)
145 {
146         asm volatile ("mov %0, %%cr2" :: "r" (val));
147 }
148
149 static inline unsigned long read_dr6(void)
150 {
151         unsigned long dr6;
152
153         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
154         return dr6;
155 }
156
157 static inline void write_dr6(unsigned long val)
158 {
159         asm volatile ("mov %0, %%dr6" :: "r" (val));
160 }
161
162 static inline unsigned long read_dr7(void)
163 {
164         unsigned long dr7;
165
166         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
167         return dr7;
168 }
169
170 static inline void write_dr7(unsigned long val)
171 {
172         asm volatile ("mov %0, %%dr7" :: "r" (val));
173 }
174
175 static inline void force_new_asid(struct kvm_vcpu *vcpu)
176 {
177         to_svm(vcpu)->asid_generation--;
178 }
179
180 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
181 {
182         force_new_asid(vcpu);
183 }
184
185 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
186 {
187         if (!(efer & KVM_EFER_LMA))
188                 efer &= ~KVM_EFER_LME;
189
190         to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
191         vcpu->shadow_efer = efer;
192 }
193
194 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
195 {
196         struct vcpu_svm *svm = to_svm(vcpu);
197
198         svm->vmcb->control.event_inj =          SVM_EVTINJ_VALID |
199                                                 SVM_EVTINJ_VALID_ERR |
200                                                 SVM_EVTINJ_TYPE_EXEPT |
201                                                 GP_VECTOR;
202         svm->vmcb->control.event_inj_err = error_code;
203 }
204
205 static void inject_ud(struct kvm_vcpu *vcpu)
206 {
207         to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
208                                                 SVM_EVTINJ_TYPE_EXEPT |
209                                                 UD_VECTOR;
210 }
211
212 static int is_page_fault(uint32_t info)
213 {
214         info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
215         return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
216 }
217
218 static int is_external_interrupt(u32 info)
219 {
220         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
221         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
222 }
223
224 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
225 {
226         struct vcpu_svm *svm = to_svm(vcpu);
227
228         if (!svm->next_rip) {
229                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
230                 return;
231         }
232         if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE) {
233                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
234                        __FUNCTION__,
235                        svm->vmcb->save.rip,
236                        svm->next_rip);
237         }
238
239         vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
240         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
241
242         vcpu->interrupt_window_open = 1;
243 }
244
245 static int has_svm(void)
246 {
247         uint32_t eax, ebx, ecx, edx;
248
249         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
250                 printk(KERN_INFO "has_svm: not amd\n");
251                 return 0;
252         }
253
254         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
255         if (eax < SVM_CPUID_FUNC) {
256                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
257                 return 0;
258         }
259
260         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
261         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
262                 printk(KERN_DEBUG "has_svm: svm not available\n");
263                 return 0;
264         }
265         return 1;
266 }
267
268 static void svm_hardware_disable(void *garbage)
269 {
270         struct svm_cpu_data *svm_data
271                 = per_cpu(svm_data, raw_smp_processor_id());
272
273         if (svm_data) {
274                 uint64_t efer;
275
276                 wrmsrl(MSR_VM_HSAVE_PA, 0);
277                 rdmsrl(MSR_EFER, efer);
278                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
279                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
280                 __free_page(svm_data->save_area);
281                 kfree(svm_data);
282         }
283 }
284
285 static void svm_hardware_enable(void *garbage)
286 {
287
288         struct svm_cpu_data *svm_data;
289         uint64_t efer;
290 #ifdef CONFIG_X86_64
291         struct desc_ptr gdt_descr;
292 #else
293         struct Xgt_desc_struct gdt_descr;
294 #endif
295         struct desc_struct *gdt;
296         int me = raw_smp_processor_id();
297
298         if (!has_svm()) {
299                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
300                 return;
301         }
302         svm_data = per_cpu(svm_data, me);
303
304         if (!svm_data) {
305                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
306                        me);
307                 return;
308         }
309
310         svm_data->asid_generation = 1;
311         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
312         svm_data->next_asid = svm_data->max_asid + 1;
313         svm_features = cpuid_edx(SVM_CPUID_FUNC);
314
315         asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
316         gdt = (struct desc_struct *)gdt_descr.address;
317         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
318
319         rdmsrl(MSR_EFER, efer);
320         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
321
322         wrmsrl(MSR_VM_HSAVE_PA,
323                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
324 }
325
326 static int svm_cpu_init(int cpu)
327 {
328         struct svm_cpu_data *svm_data;
329         int r;
330
331         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
332         if (!svm_data)
333                 return -ENOMEM;
334         svm_data->cpu = cpu;
335         svm_data->save_area = alloc_page(GFP_KERNEL);
336         r = -ENOMEM;
337         if (!svm_data->save_area)
338                 goto err_1;
339
340         per_cpu(svm_data, cpu) = svm_data;
341
342         return 0;
343
344 err_1:
345         kfree(svm_data);
346         return r;
347
348 }
349
350 static void set_msr_interception(u32 *msrpm, unsigned msr,
351                                  int read, int write)
352 {
353         int i;
354
355         for (i = 0; i < NUM_MSR_MAPS; i++) {
356                 if (msr >= msrpm_ranges[i] &&
357                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
358                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
359                                           msrpm_ranges[i]) * 2;
360
361                         u32 *base = msrpm + (msr_offset / 32);
362                         u32 msr_shift = msr_offset % 32;
363                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
364                         *base = (*base & ~(0x3 << msr_shift)) |
365                                 (mask << msr_shift);
366                         return;
367                 }
368         }
369         BUG();
370 }
371
372 static __init int svm_hardware_setup(void)
373 {
374         int cpu;
375         struct page *iopm_pages;
376         struct page *msrpm_pages;
377         void *iopm_va, *msrpm_va;
378         int r;
379
380         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
381
382         if (!iopm_pages)
383                 return -ENOMEM;
384
385         iopm_va = page_address(iopm_pages);
386         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
387         clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
388         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
389
390
391         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
392
393         r = -ENOMEM;
394         if (!msrpm_pages)
395                 goto err_1;
396
397         msrpm_va = page_address(msrpm_pages);
398         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
399         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
400
401 #ifdef CONFIG_X86_64
402         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
403         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
404         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
405         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
406         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
407         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
408 #endif
409         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
410         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
411         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
412         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
413
414         for_each_online_cpu(cpu) {
415                 r = svm_cpu_init(cpu);
416                 if (r)
417                         goto err_2;
418         }
419         return 0;
420
421 err_2:
422         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
423         msrpm_base = 0;
424 err_1:
425         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
426         iopm_base = 0;
427         return r;
428 }
429
430 static __exit void svm_hardware_unsetup(void)
431 {
432         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
433         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
434         iopm_base = msrpm_base = 0;
435 }
436
437 static void init_seg(struct vmcb_seg *seg)
438 {
439         seg->selector = 0;
440         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
441                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
442         seg->limit = 0xffff;
443         seg->base = 0;
444 }
445
446 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
447 {
448         seg->selector = 0;
449         seg->attrib = SVM_SELECTOR_P_MASK | type;
450         seg->limit = 0xffff;
451         seg->base = 0;
452 }
453
454 static void init_vmcb(struct vmcb *vmcb)
455 {
456         struct vmcb_control_area *control = &vmcb->control;
457         struct vmcb_save_area *save = &vmcb->save;
458
459         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
460                                         INTERCEPT_CR3_MASK |
461                                         INTERCEPT_CR4_MASK;
462
463         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
464                                         INTERCEPT_CR3_MASK |
465                                         INTERCEPT_CR4_MASK;
466
467         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
468                                         INTERCEPT_DR1_MASK |
469                                         INTERCEPT_DR2_MASK |
470                                         INTERCEPT_DR3_MASK;
471
472         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
473                                         INTERCEPT_DR1_MASK |
474                                         INTERCEPT_DR2_MASK |
475                                         INTERCEPT_DR3_MASK |
476                                         INTERCEPT_DR5_MASK |
477                                         INTERCEPT_DR7_MASK;
478
479         control->intercept_exceptions = 1 << PF_VECTOR;
480
481
482         control->intercept =    (1ULL << INTERCEPT_INTR) |
483                                 (1ULL << INTERCEPT_NMI) |
484                                 (1ULL << INTERCEPT_SMI) |
485                 /*
486                  * selective cr0 intercept bug?
487                  *      0:   0f 22 d8                mov    %eax,%cr3
488                  *      3:   0f 20 c0                mov    %cr0,%eax
489                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
490                  *      b:   0f 22 c0                mov    %eax,%cr0
491                  * set cr3 ->interception
492                  * get cr0 ->interception
493                  * set cr0 -> no interception
494                  */
495                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
496                                 (1ULL << INTERCEPT_CPUID) |
497                                 (1ULL << INTERCEPT_HLT) |
498                                 (1ULL << INTERCEPT_INVLPGA) |
499                                 (1ULL << INTERCEPT_IOIO_PROT) |
500                                 (1ULL << INTERCEPT_MSR_PROT) |
501                                 (1ULL << INTERCEPT_TASK_SWITCH) |
502                                 (1ULL << INTERCEPT_SHUTDOWN) |
503                                 (1ULL << INTERCEPT_VMRUN) |
504                                 (1ULL << INTERCEPT_VMMCALL) |
505                                 (1ULL << INTERCEPT_VMLOAD) |
506                                 (1ULL << INTERCEPT_VMSAVE) |
507                                 (1ULL << INTERCEPT_STGI) |
508                                 (1ULL << INTERCEPT_CLGI) |
509                                 (1ULL << INTERCEPT_SKINIT) |
510                                 (1ULL << INTERCEPT_MONITOR) |
511                                 (1ULL << INTERCEPT_MWAIT);
512
513         control->iopm_base_pa = iopm_base;
514         control->msrpm_base_pa = msrpm_base;
515         control->tsc_offset = 0;
516         control->int_ctl = V_INTR_MASKING_MASK;
517
518         init_seg(&save->es);
519         init_seg(&save->ss);
520         init_seg(&save->ds);
521         init_seg(&save->fs);
522         init_seg(&save->gs);
523
524         save->cs.selector = 0xf000;
525         /* Executable/Readable Code Segment */
526         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
527                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
528         save->cs.limit = 0xffff;
529         /*
530          * cs.base should really be 0xffff0000, but vmx can't handle that, so
531          * be consistent with it.
532          *
533          * Replace when we have real mode working for vmx.
534          */
535         save->cs.base = 0xf0000;
536
537         save->gdtr.limit = 0xffff;
538         save->idtr.limit = 0xffff;
539
540         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
541         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
542
543         save->efer = MSR_EFER_SVME_MASK;
544
545         save->dr6 = 0xffff0ff0;
546         save->dr7 = 0x400;
547         save->rflags = 2;
548         save->rip = 0x0000fff0;
549
550         /*
551          * cr0 val on cpu init should be 0x60000010, we enable cpu
552          * cache by default. the orderly way is to enable cache in bios.
553          */
554         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
555         save->cr4 = X86_CR4_PAE;
556         /* rdx = ?? */
557 }
558
559 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
560 {
561         struct vcpu_svm *svm = to_svm(vcpu);
562
563         init_vmcb(svm->vmcb);
564
565         if (vcpu->vcpu_id != 0) {
566                 svm->vmcb->save.rip = 0;
567                 svm->vmcb->save.cs.base = svm->vcpu.sipi_vector << 12;
568                 svm->vmcb->save.cs.selector = svm->vcpu.sipi_vector << 8;
569         }
570 }
571
572 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
573 {
574         struct vcpu_svm *svm;
575         struct page *page;
576         int err;
577
578         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
579         if (!svm) {
580                 err = -ENOMEM;
581                 goto out;
582         }
583
584         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
585         if (err)
586                 goto free_svm;
587
588         if (irqchip_in_kernel(kvm)) {
589                 err = kvm_create_lapic(&svm->vcpu);
590                 if (err < 0)
591                         goto free_svm;
592         }
593
594         page = alloc_page(GFP_KERNEL);
595         if (!page) {
596                 err = -ENOMEM;
597                 goto uninit;
598         }
599
600         svm->vmcb = page_address(page);
601         clear_page(svm->vmcb);
602         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
603         svm->asid_generation = 0;
604         memset(svm->db_regs, 0, sizeof(svm->db_regs));
605         init_vmcb(svm->vmcb);
606
607         fx_init(&svm->vcpu);
608         svm->vcpu.fpu_active = 1;
609         svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
610         if (svm->vcpu.vcpu_id == 0)
611                 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
612
613         return &svm->vcpu;
614
615 uninit:
616         kvm_vcpu_uninit(&svm->vcpu);
617 free_svm:
618         kmem_cache_free(kvm_vcpu_cache, svm);
619 out:
620         return ERR_PTR(err);
621 }
622
623 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
624 {
625         struct vcpu_svm *svm = to_svm(vcpu);
626
627         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
628         kvm_vcpu_uninit(vcpu);
629         kmem_cache_free(kvm_vcpu_cache, svm);
630 }
631
632 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
633 {
634         struct vcpu_svm *svm = to_svm(vcpu);
635         int i;
636
637         if (unlikely(cpu != vcpu->cpu)) {
638                 u64 tsc_this, delta;
639
640                 /*
641                  * Make sure that the guest sees a monotonically
642                  * increasing TSC.
643                  */
644                 rdtscll(tsc_this);
645                 delta = vcpu->host_tsc - tsc_this;
646                 svm->vmcb->control.tsc_offset += delta;
647                 vcpu->cpu = cpu;
648                 kvm_migrate_apic_timer(vcpu);
649         }
650
651         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
652                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
653 }
654
655 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
656 {
657         struct vcpu_svm *svm = to_svm(vcpu);
658         int i;
659
660         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
661                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
662
663         rdtscll(vcpu->host_tsc);
664 }
665
666 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
667 {
668 }
669
670 static void svm_cache_regs(struct kvm_vcpu *vcpu)
671 {
672         struct vcpu_svm *svm = to_svm(vcpu);
673
674         vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
675         vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
676         vcpu->rip = svm->vmcb->save.rip;
677 }
678
679 static void svm_decache_regs(struct kvm_vcpu *vcpu)
680 {
681         struct vcpu_svm *svm = to_svm(vcpu);
682         svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
683         svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
684         svm->vmcb->save.rip = vcpu->rip;
685 }
686
687 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
688 {
689         return to_svm(vcpu)->vmcb->save.rflags;
690 }
691
692 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
693 {
694         to_svm(vcpu)->vmcb->save.rflags = rflags;
695 }
696
697 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
698 {
699         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
700
701         switch (seg) {
702         case VCPU_SREG_CS: return &save->cs;
703         case VCPU_SREG_DS: return &save->ds;
704         case VCPU_SREG_ES: return &save->es;
705         case VCPU_SREG_FS: return &save->fs;
706         case VCPU_SREG_GS: return &save->gs;
707         case VCPU_SREG_SS: return &save->ss;
708         case VCPU_SREG_TR: return &save->tr;
709         case VCPU_SREG_LDTR: return &save->ldtr;
710         }
711         BUG();
712         return NULL;
713 }
714
715 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
716 {
717         struct vmcb_seg *s = svm_seg(vcpu, seg);
718
719         return s->base;
720 }
721
722 static void svm_get_segment(struct kvm_vcpu *vcpu,
723                             struct kvm_segment *var, int seg)
724 {
725         struct vmcb_seg *s = svm_seg(vcpu, seg);
726
727         var->base = s->base;
728         var->limit = s->limit;
729         var->selector = s->selector;
730         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
731         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
732         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
733         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
734         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
735         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
736         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
737         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
738         var->unusable = !var->present;
739 }
740
741 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
742 {
743         struct vcpu_svm *svm = to_svm(vcpu);
744
745         dt->limit = svm->vmcb->save.idtr.limit;
746         dt->base = svm->vmcb->save.idtr.base;
747 }
748
749 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
750 {
751         struct vcpu_svm *svm = to_svm(vcpu);
752
753         svm->vmcb->save.idtr.limit = dt->limit;
754         svm->vmcb->save.idtr.base = dt->base ;
755 }
756
757 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
758 {
759         struct vcpu_svm *svm = to_svm(vcpu);
760
761         dt->limit = svm->vmcb->save.gdtr.limit;
762         dt->base = svm->vmcb->save.gdtr.base;
763 }
764
765 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
766 {
767         struct vcpu_svm *svm = to_svm(vcpu);
768
769         svm->vmcb->save.gdtr.limit = dt->limit;
770         svm->vmcb->save.gdtr.base = dt->base ;
771 }
772
773 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
774 {
775 }
776
777 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
778 {
779         struct vcpu_svm *svm = to_svm(vcpu);
780
781 #ifdef CONFIG_X86_64
782         if (vcpu->shadow_efer & KVM_EFER_LME) {
783                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
784                         vcpu->shadow_efer |= KVM_EFER_LMA;
785                         svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
786                 }
787
788                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG) ) {
789                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
790                         svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
791                 }
792         }
793 #endif
794         if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
795                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
796                 vcpu->fpu_active = 1;
797         }
798
799         vcpu->cr0 = cr0;
800         cr0 |= X86_CR0_PG | X86_CR0_WP;
801         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
802         svm->vmcb->save.cr0 = cr0;
803 }
804
805 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
806 {
807        vcpu->cr4 = cr4;
808        to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
809 }
810
811 static void svm_set_segment(struct kvm_vcpu *vcpu,
812                             struct kvm_segment *var, int seg)
813 {
814         struct vcpu_svm *svm = to_svm(vcpu);
815         struct vmcb_seg *s = svm_seg(vcpu, seg);
816
817         s->base = var->base;
818         s->limit = var->limit;
819         s->selector = var->selector;
820         if (var->unusable)
821                 s->attrib = 0;
822         else {
823                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
824                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
825                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
826                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
827                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
828                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
829                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
830                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
831         }
832         if (seg == VCPU_SREG_CS)
833                 svm->vmcb->save.cpl
834                         = (svm->vmcb->save.cs.attrib
835                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
836
837 }
838
839 /* FIXME:
840
841         svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
842         svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
843
844 */
845
846 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
847 {
848         return -EOPNOTSUPP;
849 }
850
851 static int svm_get_irq(struct kvm_vcpu *vcpu)
852 {
853         struct vcpu_svm *svm = to_svm(vcpu);
854         u32 exit_int_info = svm->vmcb->control.exit_int_info;
855
856         if (is_external_interrupt(exit_int_info))
857                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
858         return -1;
859 }
860
861 static void load_host_msrs(struct kvm_vcpu *vcpu)
862 {
863 #ifdef CONFIG_X86_64
864         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
865 #endif
866 }
867
868 static void save_host_msrs(struct kvm_vcpu *vcpu)
869 {
870 #ifdef CONFIG_X86_64
871         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
872 #endif
873 }
874
875 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
876 {
877         if (svm_data->next_asid > svm_data->max_asid) {
878                 ++svm_data->asid_generation;
879                 svm_data->next_asid = 1;
880                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
881         }
882
883         svm->vcpu.cpu = svm_data->cpu;
884         svm->asid_generation = svm_data->asid_generation;
885         svm->vmcb->control.asid = svm_data->next_asid++;
886 }
887
888 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
889 {
890         return to_svm(vcpu)->db_regs[dr];
891 }
892
893 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
894                        int *exception)
895 {
896         struct vcpu_svm *svm = to_svm(vcpu);
897
898         *exception = 0;
899
900         if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
901                 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
902                 svm->vmcb->save.dr6 |= DR6_BD_MASK;
903                 *exception = DB_VECTOR;
904                 return;
905         }
906
907         switch (dr) {
908         case 0 ... 3:
909                 svm->db_regs[dr] = value;
910                 return;
911         case 4 ... 5:
912                 if (vcpu->cr4 & X86_CR4_DE) {
913                         *exception = UD_VECTOR;
914                         return;
915                 }
916         case 7: {
917                 if (value & ~((1ULL << 32) - 1)) {
918                         *exception = GP_VECTOR;
919                         return;
920                 }
921                 svm->vmcb->save.dr7 = value;
922                 return;
923         }
924         default:
925                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
926                        __FUNCTION__, dr);
927                 *exception = UD_VECTOR;
928                 return;
929         }
930 }
931
932 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
933 {
934         u32 exit_int_info = svm->vmcb->control.exit_int_info;
935         struct kvm *kvm = svm->vcpu.kvm;
936         u64 fault_address;
937         u32 error_code;
938         enum emulation_result er;
939         int r;
940
941         if (!irqchip_in_kernel(kvm) &&
942                 is_external_interrupt(exit_int_info))
943                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
944
945         mutex_lock(&kvm->lock);
946
947         fault_address  = svm->vmcb->control.exit_info_2;
948         error_code = svm->vmcb->control.exit_info_1;
949         r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
950         if (r < 0) {
951                 mutex_unlock(&kvm->lock);
952                 return r;
953         }
954         if (!r) {
955                 mutex_unlock(&kvm->lock);
956                 return 1;
957         }
958         er = emulate_instruction(&svm->vcpu, kvm_run, fault_address,
959                                  error_code);
960         mutex_unlock(&kvm->lock);
961
962         switch (er) {
963         case EMULATE_DONE:
964                 return 1;
965         case EMULATE_DO_MMIO:
966                 ++svm->vcpu.stat.mmio_exits;
967                 return 0;
968         case EMULATE_FAIL:
969                 kvm_report_emulation_failure(&svm->vcpu, "pagetable");
970                 break;
971         default:
972                 BUG();
973         }
974
975         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
976         return 0;
977 }
978
979 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
980 {
981         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
982         if (!(svm->vcpu.cr0 & X86_CR0_TS))
983                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
984         svm->vcpu.fpu_active = 1;
985
986         return 1;
987 }
988
989 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
990 {
991         /*
992          * VMCB is undefined after a SHUTDOWN intercept
993          * so reinitialize it.
994          */
995         clear_page(svm->vmcb);
996         init_vmcb(svm->vmcb);
997
998         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
999         return 0;
1000 }
1001
1002 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1003 {
1004         u32 io_info = svm->vmcb->control.exit_info_1; //address size bug?
1005         int size, down, in, string, rep;
1006         unsigned port;
1007
1008         ++svm->vcpu.stat.io_exits;
1009
1010         svm->next_rip = svm->vmcb->control.exit_info_2;
1011
1012         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1013
1014         if (string) {
1015                 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0) == EMULATE_DO_MMIO)
1016                         return 0;
1017                 return 1;
1018         }
1019
1020         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1021         port = io_info >> 16;
1022         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1023         rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1024         down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1025
1026         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1027 }
1028
1029 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1030 {
1031         return 1;
1032 }
1033
1034 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1035 {
1036         svm->next_rip = svm->vmcb->save.rip + 1;
1037         skip_emulated_instruction(&svm->vcpu);
1038         return kvm_emulate_halt(&svm->vcpu);
1039 }
1040
1041 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1042 {
1043         svm->next_rip = svm->vmcb->save.rip + 3;
1044         skip_emulated_instruction(&svm->vcpu);
1045         return kvm_hypercall(&svm->vcpu, kvm_run);
1046 }
1047
1048 static int invalid_op_interception(struct vcpu_svm *svm,
1049                                    struct kvm_run *kvm_run)
1050 {
1051         inject_ud(&svm->vcpu);
1052         return 1;
1053 }
1054
1055 static int task_switch_interception(struct vcpu_svm *svm,
1056                                     struct kvm_run *kvm_run)
1057 {
1058         pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1059         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1060         return 0;
1061 }
1062
1063 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1064 {
1065         svm->next_rip = svm->vmcb->save.rip + 2;
1066         kvm_emulate_cpuid(&svm->vcpu);
1067         return 1;
1068 }
1069
1070 static int emulate_on_interception(struct vcpu_svm *svm,
1071                                    struct kvm_run *kvm_run)
1072 {
1073         if (emulate_instruction(&svm->vcpu, NULL, 0, 0) != EMULATE_DONE)
1074                 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1075         return 1;
1076 }
1077
1078 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1079 {
1080         struct vcpu_svm *svm = to_svm(vcpu);
1081
1082         switch (ecx) {
1083         case MSR_IA32_TIME_STAMP_COUNTER: {
1084                 u64 tsc;
1085
1086                 rdtscll(tsc);
1087                 *data = svm->vmcb->control.tsc_offset + tsc;
1088                 break;
1089         }
1090         case MSR_K6_STAR:
1091                 *data = svm->vmcb->save.star;
1092                 break;
1093 #ifdef CONFIG_X86_64
1094         case MSR_LSTAR:
1095                 *data = svm->vmcb->save.lstar;
1096                 break;
1097         case MSR_CSTAR:
1098                 *data = svm->vmcb->save.cstar;
1099                 break;
1100         case MSR_KERNEL_GS_BASE:
1101                 *data = svm->vmcb->save.kernel_gs_base;
1102                 break;
1103         case MSR_SYSCALL_MASK:
1104                 *data = svm->vmcb->save.sfmask;
1105                 break;
1106 #endif
1107         case MSR_IA32_SYSENTER_CS:
1108                 *data = svm->vmcb->save.sysenter_cs;
1109                 break;
1110         case MSR_IA32_SYSENTER_EIP:
1111                 *data = svm->vmcb->save.sysenter_eip;
1112                 break;
1113         case MSR_IA32_SYSENTER_ESP:
1114                 *data = svm->vmcb->save.sysenter_esp;
1115                 break;
1116         default:
1117                 return kvm_get_msr_common(vcpu, ecx, data);
1118         }
1119         return 0;
1120 }
1121
1122 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1123 {
1124         u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1125         u64 data;
1126
1127         if (svm_get_msr(&svm->vcpu, ecx, &data))
1128                 svm_inject_gp(&svm->vcpu, 0);
1129         else {
1130                 svm->vmcb->save.rax = data & 0xffffffff;
1131                 svm->vcpu.regs[VCPU_REGS_RDX] = data >> 32;
1132                 svm->next_rip = svm->vmcb->save.rip + 2;
1133                 skip_emulated_instruction(&svm->vcpu);
1134         }
1135         return 1;
1136 }
1137
1138 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1139 {
1140         struct vcpu_svm *svm = to_svm(vcpu);
1141
1142         switch (ecx) {
1143         case MSR_IA32_TIME_STAMP_COUNTER: {
1144                 u64 tsc;
1145
1146                 rdtscll(tsc);
1147                 svm->vmcb->control.tsc_offset = data - tsc;
1148                 break;
1149         }
1150         case MSR_K6_STAR:
1151                 svm->vmcb->save.star = data;
1152                 break;
1153 #ifdef CONFIG_X86_64
1154         case MSR_LSTAR:
1155                 svm->vmcb->save.lstar = data;
1156                 break;
1157         case MSR_CSTAR:
1158                 svm->vmcb->save.cstar = data;
1159                 break;
1160         case MSR_KERNEL_GS_BASE:
1161                 svm->vmcb->save.kernel_gs_base = data;
1162                 break;
1163         case MSR_SYSCALL_MASK:
1164                 svm->vmcb->save.sfmask = data;
1165                 break;
1166 #endif
1167         case MSR_IA32_SYSENTER_CS:
1168                 svm->vmcb->save.sysenter_cs = data;
1169                 break;
1170         case MSR_IA32_SYSENTER_EIP:
1171                 svm->vmcb->save.sysenter_eip = data;
1172                 break;
1173         case MSR_IA32_SYSENTER_ESP:
1174                 svm->vmcb->save.sysenter_esp = data;
1175                 break;
1176         default:
1177                 return kvm_set_msr_common(vcpu, ecx, data);
1178         }
1179         return 0;
1180 }
1181
1182 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1183 {
1184         u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1185         u64 data = (svm->vmcb->save.rax & -1u)
1186                 | ((u64)(svm->vcpu.regs[VCPU_REGS_RDX] & -1u) << 32);
1187         svm->next_rip = svm->vmcb->save.rip + 2;
1188         if (svm_set_msr(&svm->vcpu, ecx, data))
1189                 svm_inject_gp(&svm->vcpu, 0);
1190         else
1191                 skip_emulated_instruction(&svm->vcpu);
1192         return 1;
1193 }
1194
1195 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1196 {
1197         if (svm->vmcb->control.exit_info_1)
1198                 return wrmsr_interception(svm, kvm_run);
1199         else
1200                 return rdmsr_interception(svm, kvm_run);
1201 }
1202
1203 static int interrupt_window_interception(struct vcpu_svm *svm,
1204                                    struct kvm_run *kvm_run)
1205 {
1206         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1207         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1208         /*
1209          * If the user space waits to inject interrupts, exit as soon as
1210          * possible
1211          */
1212         if (kvm_run->request_interrupt_window &&
1213             !svm->vcpu.irq_summary) {
1214                 ++svm->vcpu.stat.irq_window_exits;
1215                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1216                 return 0;
1217         }
1218
1219         return 1;
1220 }
1221
1222 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1223                                       struct kvm_run *kvm_run) = {
1224         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1225         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1226         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1227         /* for now: */
1228         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1229         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1230         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1231         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1232         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1233         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1234         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1235         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1236         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1237         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1238         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1239         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1240         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1241         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1242         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
1243         [SVM_EXIT_INTR]                         = nop_on_interception,
1244         [SVM_EXIT_NMI]                          = nop_on_interception,
1245         [SVM_EXIT_SMI]                          = nop_on_interception,
1246         [SVM_EXIT_INIT]                         = nop_on_interception,
1247         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1248         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1249         [SVM_EXIT_CPUID]                        = cpuid_interception,
1250         [SVM_EXIT_HLT]                          = halt_interception,
1251         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1252         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1253         [SVM_EXIT_IOIO]                         = io_interception,
1254         [SVM_EXIT_MSR]                          = msr_interception,
1255         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1256         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1257         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1258         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
1259         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1260         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1261         [SVM_EXIT_STGI]                         = invalid_op_interception,
1262         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1263         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1264         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
1265         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
1266 };
1267
1268
1269 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1270 {
1271         struct vcpu_svm *svm = to_svm(vcpu);
1272         u32 exit_code = svm->vmcb->control.exit_code;
1273
1274         kvm_reput_irq(svm);
1275
1276         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1277                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1278                 kvm_run->fail_entry.hardware_entry_failure_reason
1279                         = svm->vmcb->control.exit_code;
1280                 return 0;
1281         }
1282
1283         if (is_external_interrupt(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__, svm->vmcb->control.exit_int_info,
1288                        exit_code);
1289
1290         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1291             || svm_exit_handlers[exit_code] == 0) {
1292                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1293                 kvm_run->hw.hardware_exit_reason = exit_code;
1294                 return 0;
1295         }
1296
1297         return svm_exit_handlers[exit_code](svm, kvm_run);
1298 }
1299
1300 static void reload_tss(struct kvm_vcpu *vcpu)
1301 {
1302         int cpu = raw_smp_processor_id();
1303
1304         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1305         svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1306         load_TR_desc();
1307 }
1308
1309 static void pre_svm_run(struct vcpu_svm *svm)
1310 {
1311         int cpu = raw_smp_processor_id();
1312
1313         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1314
1315         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1316         if (svm->vcpu.cpu != cpu ||
1317             svm->asid_generation != svm_data->asid_generation)
1318                 new_asid(svm, svm_data);
1319 }
1320
1321
1322 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1323 {
1324         struct vmcb_control_area *control;
1325
1326         control = &svm->vmcb->control;
1327         control->int_vector = irq;
1328         control->int_ctl &= ~V_INTR_PRIO_MASK;
1329         control->int_ctl |= V_IRQ_MASK |
1330                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1331 }
1332
1333 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1334 {
1335         struct vcpu_svm *svm = to_svm(vcpu);
1336
1337         svm_inject_irq(svm, irq);
1338 }
1339
1340 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1341 {
1342         struct vcpu_svm *svm = to_svm(vcpu);
1343         struct vmcb *vmcb = svm->vmcb;
1344         int intr_vector = -1;
1345
1346         kvm_inject_pending_timer_irqs(vcpu);
1347         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1348             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1349                 intr_vector = vmcb->control.exit_int_info &
1350                               SVM_EVTINJ_VEC_MASK;
1351                 vmcb->control.exit_int_info = 0;
1352                 svm_inject_irq(svm, intr_vector);
1353                 return;
1354         }
1355
1356         if (vmcb->control.int_ctl & V_IRQ_MASK)
1357                 return;
1358
1359         if (!kvm_cpu_has_interrupt(vcpu))
1360                 return;
1361
1362         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1363             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1364             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1365                 /* unable to deliver irq, set pending irq */
1366                 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1367                 svm_inject_irq(svm, 0x0);
1368                 return;
1369         }
1370         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1371         intr_vector = kvm_cpu_get_interrupt(vcpu);
1372         svm_inject_irq(svm, intr_vector);
1373         kvm_timer_intr_post(vcpu, intr_vector);
1374 }
1375
1376 static void kvm_reput_irq(struct vcpu_svm *svm)
1377 {
1378         struct vmcb_control_area *control = &svm->vmcb->control;
1379
1380         if ((control->int_ctl & V_IRQ_MASK)
1381             && !irqchip_in_kernel(svm->vcpu.kvm)) {
1382                 control->int_ctl &= ~V_IRQ_MASK;
1383                 push_irq(&svm->vcpu, control->int_vector);
1384         }
1385
1386         svm->vcpu.interrupt_window_open =
1387                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1388 }
1389
1390 static void svm_do_inject_vector(struct vcpu_svm *svm)
1391 {
1392         struct kvm_vcpu *vcpu = &svm->vcpu;
1393         int word_index = __ffs(vcpu->irq_summary);
1394         int bit_index = __ffs(vcpu->irq_pending[word_index]);
1395         int irq = word_index * BITS_PER_LONG + bit_index;
1396
1397         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1398         if (!vcpu->irq_pending[word_index])
1399                 clear_bit(word_index, &vcpu->irq_summary);
1400         svm_inject_irq(svm, irq);
1401 }
1402
1403 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1404                                        struct kvm_run *kvm_run)
1405 {
1406         struct vcpu_svm *svm = to_svm(vcpu);
1407         struct vmcb_control_area *control = &svm->vmcb->control;
1408
1409         svm->vcpu.interrupt_window_open =
1410                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1411                  (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1412
1413         if (svm->vcpu.interrupt_window_open && svm->vcpu.irq_summary)
1414                 /*
1415                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1416                  */
1417                 svm_do_inject_vector(svm);
1418
1419         /*
1420          * Interrupts blocked.  Wait for unblock.
1421          */
1422         if (!svm->vcpu.interrupt_window_open &&
1423             (svm->vcpu.irq_summary || kvm_run->request_interrupt_window)) {
1424                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1425         } else
1426                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1427 }
1428
1429 static void save_db_regs(unsigned long *db_regs)
1430 {
1431         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1432         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1433         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1434         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1435 }
1436
1437 static void load_db_regs(unsigned long *db_regs)
1438 {
1439         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1440         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1441         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1442         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1443 }
1444
1445 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1446 {
1447         force_new_asid(vcpu);
1448 }
1449
1450 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1451 {
1452 }
1453
1454 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1455 {
1456         struct vcpu_svm *svm = to_svm(vcpu);
1457         u16 fs_selector;
1458         u16 gs_selector;
1459         u16 ldt_selector;
1460
1461         pre_svm_run(svm);
1462
1463         save_host_msrs(vcpu);
1464         fs_selector = read_fs();
1465         gs_selector = read_gs();
1466         ldt_selector = read_ldt();
1467         svm->host_cr2 = kvm_read_cr2();
1468         svm->host_dr6 = read_dr6();
1469         svm->host_dr7 = read_dr7();
1470         svm->vmcb->save.cr2 = vcpu->cr2;
1471
1472         if (svm->vmcb->save.dr7 & 0xff) {
1473                 write_dr7(0);
1474                 save_db_regs(svm->host_db_regs);
1475                 load_db_regs(svm->db_regs);
1476         }
1477
1478         clgi();
1479
1480         local_irq_enable();
1481
1482         asm volatile (
1483 #ifdef CONFIG_X86_64
1484                 "push %%rbx; push %%rcx; push %%rdx;"
1485                 "push %%rsi; push %%rdi; push %%rbp;"
1486                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1487                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1488 #else
1489                 "push %%ebx; push %%ecx; push %%edx;"
1490                 "push %%esi; push %%edi; push %%ebp;"
1491 #endif
1492
1493 #ifdef CONFIG_X86_64
1494                 "mov %c[rbx](%[svm]), %%rbx \n\t"
1495                 "mov %c[rcx](%[svm]), %%rcx \n\t"
1496                 "mov %c[rdx](%[svm]), %%rdx \n\t"
1497                 "mov %c[rsi](%[svm]), %%rsi \n\t"
1498                 "mov %c[rdi](%[svm]), %%rdi \n\t"
1499                 "mov %c[rbp](%[svm]), %%rbp \n\t"
1500                 "mov %c[r8](%[svm]),  %%r8  \n\t"
1501                 "mov %c[r9](%[svm]),  %%r9  \n\t"
1502                 "mov %c[r10](%[svm]), %%r10 \n\t"
1503                 "mov %c[r11](%[svm]), %%r11 \n\t"
1504                 "mov %c[r12](%[svm]), %%r12 \n\t"
1505                 "mov %c[r13](%[svm]), %%r13 \n\t"
1506                 "mov %c[r14](%[svm]), %%r14 \n\t"
1507                 "mov %c[r15](%[svm]), %%r15 \n\t"
1508 #else
1509                 "mov %c[rbx](%[svm]), %%ebx \n\t"
1510                 "mov %c[rcx](%[svm]), %%ecx \n\t"
1511                 "mov %c[rdx](%[svm]), %%edx \n\t"
1512                 "mov %c[rsi](%[svm]), %%esi \n\t"
1513                 "mov %c[rdi](%[svm]), %%edi \n\t"
1514                 "mov %c[rbp](%[svm]), %%ebp \n\t"
1515 #endif
1516
1517 #ifdef CONFIG_X86_64
1518                 /* Enter guest mode */
1519                 "push %%rax \n\t"
1520                 "mov %c[vmcb](%[svm]), %%rax \n\t"
1521                 SVM_VMLOAD "\n\t"
1522                 SVM_VMRUN "\n\t"
1523                 SVM_VMSAVE "\n\t"
1524                 "pop %%rax \n\t"
1525 #else
1526                 /* Enter guest mode */
1527                 "push %%eax \n\t"
1528                 "mov %c[vmcb](%[svm]), %%eax \n\t"
1529                 SVM_VMLOAD "\n\t"
1530                 SVM_VMRUN "\n\t"
1531                 SVM_VMSAVE "\n\t"
1532                 "pop %%eax \n\t"
1533 #endif
1534
1535                 /* Save guest registers, load host registers */
1536 #ifdef CONFIG_X86_64
1537                 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1538                 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1539                 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1540                 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1541                 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1542                 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1543                 "mov %%r8,  %c[r8](%[svm]) \n\t"
1544                 "mov %%r9,  %c[r9](%[svm]) \n\t"
1545                 "mov %%r10, %c[r10](%[svm]) \n\t"
1546                 "mov %%r11, %c[r11](%[svm]) \n\t"
1547                 "mov %%r12, %c[r12](%[svm]) \n\t"
1548                 "mov %%r13, %c[r13](%[svm]) \n\t"
1549                 "mov %%r14, %c[r14](%[svm]) \n\t"
1550                 "mov %%r15, %c[r15](%[svm]) \n\t"
1551
1552                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1553                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1554                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1555                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1556 #else
1557                 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1558                 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1559                 "mov %%edx, %c[rdx](%[svm]) \n\t"
1560                 "mov %%esi, %c[rsi](%[svm]) \n\t"
1561                 "mov %%edi, %c[rdi](%[svm]) \n\t"
1562                 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1563
1564                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1565                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1566 #endif
1567                 :
1568                 : [svm]"a"(svm),
1569                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1570                   [rbx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBX])),
1571                   [rcx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RCX])),
1572                   [rdx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDX])),
1573                   [rsi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RSI])),
1574                   [rdi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDI])),
1575                   [rbp]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBP]))
1576 #ifdef CONFIG_X86_64
1577                   ,[r8 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R8])),
1578                   [r9 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R9 ])),
1579                   [r10]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R10])),
1580                   [r11]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R11])),
1581                   [r12]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R12])),
1582                   [r13]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R13])),
1583                   [r14]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R14])),
1584                   [r15]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R15]))
1585 #endif
1586                 : "cc", "memory" );
1587
1588         local_irq_disable();
1589
1590         stgi();
1591
1592         if ((svm->vmcb->save.dr7 & 0xff))
1593                 load_db_regs(svm->host_db_regs);
1594
1595         vcpu->cr2 = svm->vmcb->save.cr2;
1596
1597         write_dr6(svm->host_dr6);
1598         write_dr7(svm->host_dr7);
1599         kvm_write_cr2(svm->host_cr2);
1600
1601         load_fs(fs_selector);
1602         load_gs(gs_selector);
1603         load_ldt(ldt_selector);
1604         load_host_msrs(vcpu);
1605
1606         reload_tss(vcpu);
1607
1608         svm->next_rip = 0;
1609 }
1610
1611 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1612 {
1613         struct vcpu_svm *svm = to_svm(vcpu);
1614
1615         svm->vmcb->save.cr3 = root;
1616         force_new_asid(vcpu);
1617
1618         if (vcpu->fpu_active) {
1619                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1620                 svm->vmcb->save.cr0 |= X86_CR0_TS;
1621                 vcpu->fpu_active = 0;
1622         }
1623 }
1624
1625 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1626                                   unsigned long  addr,
1627                                   uint32_t err_code)
1628 {
1629         struct vcpu_svm *svm = to_svm(vcpu);
1630         uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1631
1632         ++vcpu->stat.pf_guest;
1633
1634         if (is_page_fault(exit_int_info)) {
1635
1636                 svm->vmcb->control.event_inj_err = 0;
1637                 svm->vmcb->control.event_inj =  SVM_EVTINJ_VALID |
1638                                                 SVM_EVTINJ_VALID_ERR |
1639                                                 SVM_EVTINJ_TYPE_EXEPT |
1640                                                 DF_VECTOR;
1641                 return;
1642         }
1643         vcpu->cr2 = addr;
1644         svm->vmcb->save.cr2 = addr;
1645         svm->vmcb->control.event_inj =  SVM_EVTINJ_VALID |
1646                                         SVM_EVTINJ_VALID_ERR |
1647                                         SVM_EVTINJ_TYPE_EXEPT |
1648                                         PF_VECTOR;
1649         svm->vmcb->control.event_inj_err = err_code;
1650 }
1651
1652
1653 static int is_disabled(void)
1654 {
1655         u64 vm_cr;
1656
1657         rdmsrl(MSR_VM_CR, vm_cr);
1658         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1659                 return 1;
1660
1661         return 0;
1662 }
1663
1664 static void
1665 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1666 {
1667         /*
1668          * Patch in the VMMCALL instruction:
1669          */
1670         hypercall[0] = 0x0f;
1671         hypercall[1] = 0x01;
1672         hypercall[2] = 0xd9;
1673         hypercall[3] = 0xc3;
1674 }
1675
1676 static void svm_check_processor_compat(void *rtn)
1677 {
1678         *(int *)rtn = 0;
1679 }
1680
1681 static struct kvm_x86_ops svm_x86_ops = {
1682         .cpu_has_kvm_support = has_svm,
1683         .disabled_by_bios = is_disabled,
1684         .hardware_setup = svm_hardware_setup,
1685         .hardware_unsetup = svm_hardware_unsetup,
1686         .check_processor_compatibility = svm_check_processor_compat,
1687         .hardware_enable = svm_hardware_enable,
1688         .hardware_disable = svm_hardware_disable,
1689
1690         .vcpu_create = svm_create_vcpu,
1691         .vcpu_free = svm_free_vcpu,
1692         .vcpu_reset = svm_vcpu_reset,
1693
1694         .prepare_guest_switch = svm_prepare_guest_switch,
1695         .vcpu_load = svm_vcpu_load,
1696         .vcpu_put = svm_vcpu_put,
1697         .vcpu_decache = svm_vcpu_decache,
1698
1699         .set_guest_debug = svm_guest_debug,
1700         .get_msr = svm_get_msr,
1701         .set_msr = svm_set_msr,
1702         .get_segment_base = svm_get_segment_base,
1703         .get_segment = svm_get_segment,
1704         .set_segment = svm_set_segment,
1705         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1706         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1707         .set_cr0 = svm_set_cr0,
1708         .set_cr3 = svm_set_cr3,
1709         .set_cr4 = svm_set_cr4,
1710         .set_efer = svm_set_efer,
1711         .get_idt = svm_get_idt,
1712         .set_idt = svm_set_idt,
1713         .get_gdt = svm_get_gdt,
1714         .set_gdt = svm_set_gdt,
1715         .get_dr = svm_get_dr,
1716         .set_dr = svm_set_dr,
1717         .cache_regs = svm_cache_regs,
1718         .decache_regs = svm_decache_regs,
1719         .get_rflags = svm_get_rflags,
1720         .set_rflags = svm_set_rflags,
1721
1722         .tlb_flush = svm_flush_tlb,
1723         .inject_page_fault = svm_inject_page_fault,
1724
1725         .inject_gp = svm_inject_gp,
1726
1727         .run = svm_vcpu_run,
1728         .handle_exit = handle_exit,
1729         .skip_emulated_instruction = skip_emulated_instruction,
1730         .patch_hypercall = svm_patch_hypercall,
1731         .get_irq = svm_get_irq,
1732         .set_irq = svm_set_irq,
1733         .inject_pending_irq = svm_intr_assist,
1734         .inject_pending_vectors = do_interrupt_requests,
1735 };
1736
1737 static int __init svm_init(void)
1738 {
1739         return kvm_init_x86(&svm_x86_ops, sizeof(struct vcpu_svm),
1740                               THIS_MODULE);
1741 }
1742
1743 static void __exit svm_exit(void)
1744 {
1745         kvm_exit_x86();
1746 }
1747
1748 module_init(svm_init)
1749 module_exit(svm_exit)