]> err.no Git - linux-2.6/blob - arch/x86/kvm/svm.c
KVM: SVM: Fix lazy FPU switching
[linux-2.6] / arch / x86 / 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 #include <linux/kvm_host.h>
17
18 #include "kvm_svm.h"
19 #include "irq.h"
20 #include "mmu.h"
21
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/vmalloc.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
27
28 #include <asm/desc.h>
29
30 MODULE_AUTHOR("Qumranet");
31 MODULE_LICENSE("GPL");
32
33 #define IOPM_ALLOC_ORDER 2
34 #define MSRPM_ALLOC_ORDER 1
35
36 #define DB_VECTOR 1
37 #define UD_VECTOR 6
38 #define GP_VECTOR 13
39
40 #define DR7_GD_MASK (1 << 13)
41 #define DR6_BD_MASK (1 << 13)
42
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
45
46 #define SVM_FEATURE_NPT  (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_DEATURE_SVML (1 << 2)
49
50 static void kvm_reput_irq(struct vcpu_svm *svm);
51
52 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
53 {
54         return container_of(vcpu, struct vcpu_svm, vcpu);
55 }
56
57 unsigned long iopm_base;
58 unsigned long msrpm_base;
59
60 struct kvm_ldttss_desc {
61         u16 limit0;
62         u16 base0;
63         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
64         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
65         u32 base3;
66         u32 zero1;
67 } __attribute__((packed));
68
69 struct svm_cpu_data {
70         int cpu;
71
72         u64 asid_generation;
73         u32 max_asid;
74         u32 next_asid;
75         struct kvm_ldttss_desc *tss_desc;
76
77         struct page *save_area;
78 };
79
80 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
81 static uint32_t svm_features;
82
83 struct svm_init_data {
84         int cpu;
85         int r;
86 };
87
88 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
89
90 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
91 #define MSRS_RANGE_SIZE 2048
92 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
93
94 #define MAX_INST_SIZE 15
95
96 static inline u32 svm_has(u32 feat)
97 {
98         return svm_features & feat;
99 }
100
101 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
102 {
103         int word_index = __ffs(vcpu->arch.irq_summary);
104         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
105         int irq = word_index * BITS_PER_LONG + bit_index;
106
107         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
108         if (!vcpu->arch.irq_pending[word_index])
109                 clear_bit(word_index, &vcpu->arch.irq_summary);
110         return irq;
111 }
112
113 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
114 {
115         set_bit(irq, vcpu->arch.irq_pending);
116         set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
117 }
118
119 static inline void clgi(void)
120 {
121         asm volatile (SVM_CLGI);
122 }
123
124 static inline void stgi(void)
125 {
126         asm volatile (SVM_STGI);
127 }
128
129 static inline void invlpga(unsigned long addr, u32 asid)
130 {
131         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
132 }
133
134 static inline unsigned long kvm_read_cr2(void)
135 {
136         unsigned long cr2;
137
138         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
139         return cr2;
140 }
141
142 static inline void kvm_write_cr2(unsigned long val)
143 {
144         asm volatile ("mov %0, %%cr2" :: "r" (val));
145 }
146
147 static inline unsigned long read_dr6(void)
148 {
149         unsigned long dr6;
150
151         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
152         return dr6;
153 }
154
155 static inline void write_dr6(unsigned long val)
156 {
157         asm volatile ("mov %0, %%dr6" :: "r" (val));
158 }
159
160 static inline unsigned long read_dr7(void)
161 {
162         unsigned long dr7;
163
164         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
165         return dr7;
166 }
167
168 static inline void write_dr7(unsigned long val)
169 {
170         asm volatile ("mov %0, %%dr7" :: "r" (val));
171 }
172
173 static inline void force_new_asid(struct kvm_vcpu *vcpu)
174 {
175         to_svm(vcpu)->asid_generation--;
176 }
177
178 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
179 {
180         force_new_asid(vcpu);
181 }
182
183 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
184 {
185         if (!(efer & EFER_LMA))
186                 efer &= ~EFER_LME;
187
188         to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
189         vcpu->arch.shadow_efer = efer;
190 }
191
192 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
193                                 bool has_error_code, u32 error_code)
194 {
195         struct vcpu_svm *svm = to_svm(vcpu);
196
197         svm->vmcb->control.event_inj = nr
198                 | SVM_EVTINJ_VALID
199                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
200                 | SVM_EVTINJ_TYPE_EXEPT;
201         svm->vmcb->control.event_inj_err = error_code;
202 }
203
204 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
205 {
206         struct vcpu_svm *svm = to_svm(vcpu);
207
208         return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
209 }
210
211 static int is_external_interrupt(u32 info)
212 {
213         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
214         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
215 }
216
217 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
218 {
219         struct vcpu_svm *svm = to_svm(vcpu);
220
221         if (!svm->next_rip) {
222                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
223                 return;
224         }
225         if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
226                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
227                        __FUNCTION__,
228                        svm->vmcb->save.rip,
229                        svm->next_rip);
230
231         vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
232         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
233
234         vcpu->arch.interrupt_window_open = 1;
235 }
236
237 static int has_svm(void)
238 {
239         uint32_t eax, ebx, ecx, edx;
240
241         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
242                 printk(KERN_INFO "has_svm: not amd\n");
243                 return 0;
244         }
245
246         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
247         if (eax < SVM_CPUID_FUNC) {
248                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
249                 return 0;
250         }
251
252         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
253         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
254                 printk(KERN_DEBUG "has_svm: svm not available\n");
255                 return 0;
256         }
257         return 1;
258 }
259
260 static void svm_hardware_disable(void *garbage)
261 {
262         struct svm_cpu_data *svm_data
263                 = per_cpu(svm_data, raw_smp_processor_id());
264
265         if (svm_data) {
266                 uint64_t efer;
267
268                 wrmsrl(MSR_VM_HSAVE_PA, 0);
269                 rdmsrl(MSR_EFER, efer);
270                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
271                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
272                 __free_page(svm_data->save_area);
273                 kfree(svm_data);
274         }
275 }
276
277 static void svm_hardware_enable(void *garbage)
278 {
279
280         struct svm_cpu_data *svm_data;
281         uint64_t efer;
282 #ifdef CONFIG_X86_64
283         struct desc_ptr gdt_descr;
284 #else
285         struct desc_ptr gdt_descr;
286 #endif
287         struct desc_struct *gdt;
288         int me = raw_smp_processor_id();
289
290         if (!has_svm()) {
291                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
292                 return;
293         }
294         svm_data = per_cpu(svm_data, me);
295
296         if (!svm_data) {
297                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
298                        me);
299                 return;
300         }
301
302         svm_data->asid_generation = 1;
303         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
304         svm_data->next_asid = svm_data->max_asid + 1;
305         svm_features = cpuid_edx(SVM_CPUID_FUNC);
306
307         asm volatile ("sgdt %0" : "=m"(gdt_descr));
308         gdt = (struct desc_struct *)gdt_descr.address;
309         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
310
311         rdmsrl(MSR_EFER, efer);
312         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
313
314         wrmsrl(MSR_VM_HSAVE_PA,
315                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
316 }
317
318 static int svm_cpu_init(int cpu)
319 {
320         struct svm_cpu_data *svm_data;
321         int r;
322
323         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
324         if (!svm_data)
325                 return -ENOMEM;
326         svm_data->cpu = cpu;
327         svm_data->save_area = alloc_page(GFP_KERNEL);
328         r = -ENOMEM;
329         if (!svm_data->save_area)
330                 goto err_1;
331
332         per_cpu(svm_data, cpu) = svm_data;
333
334         return 0;
335
336 err_1:
337         kfree(svm_data);
338         return r;
339
340 }
341
342 static void set_msr_interception(u32 *msrpm, unsigned msr,
343                                  int read, int write)
344 {
345         int i;
346
347         for (i = 0; i < NUM_MSR_MAPS; i++) {
348                 if (msr >= msrpm_ranges[i] &&
349                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
350                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
351                                           msrpm_ranges[i]) * 2;
352
353                         u32 *base = msrpm + (msr_offset / 32);
354                         u32 msr_shift = msr_offset % 32;
355                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
356                         *base = (*base & ~(0x3 << msr_shift)) |
357                                 (mask << msr_shift);
358                         return;
359                 }
360         }
361         BUG();
362 }
363
364 static __init int svm_hardware_setup(void)
365 {
366         int cpu;
367         struct page *iopm_pages;
368         struct page *msrpm_pages;
369         void *iopm_va, *msrpm_va;
370         int r;
371
372         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
373
374         if (!iopm_pages)
375                 return -ENOMEM;
376
377         iopm_va = page_address(iopm_pages);
378         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
379         clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
380         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
381
382
383         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
384
385         r = -ENOMEM;
386         if (!msrpm_pages)
387                 goto err_1;
388
389         msrpm_va = page_address(msrpm_pages);
390         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
391         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
392
393 #ifdef CONFIG_X86_64
394         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
395         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
396         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
397         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
398         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
399         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
400 #endif
401         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
402         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
403         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
404         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
405
406         for_each_online_cpu(cpu) {
407                 r = svm_cpu_init(cpu);
408                 if (r)
409                         goto err_2;
410         }
411         return 0;
412
413 err_2:
414         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
415         msrpm_base = 0;
416 err_1:
417         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
418         iopm_base = 0;
419         return r;
420 }
421
422 static __exit void svm_hardware_unsetup(void)
423 {
424         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
425         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
426         iopm_base = msrpm_base = 0;
427 }
428
429 static void init_seg(struct vmcb_seg *seg)
430 {
431         seg->selector = 0;
432         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
433                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
434         seg->limit = 0xffff;
435         seg->base = 0;
436 }
437
438 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
439 {
440         seg->selector = 0;
441         seg->attrib = SVM_SELECTOR_P_MASK | type;
442         seg->limit = 0xffff;
443         seg->base = 0;
444 }
445
446 static void init_vmcb(struct vmcb *vmcb)
447 {
448         struct vmcb_control_area *control = &vmcb->control;
449         struct vmcb_save_area *save = &vmcb->save;
450
451         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
452                                         INTERCEPT_CR3_MASK |
453                                         INTERCEPT_CR4_MASK |
454                                         INTERCEPT_CR8_MASK;
455
456         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
457                                         INTERCEPT_CR3_MASK |
458                                         INTERCEPT_CR4_MASK |
459                                         INTERCEPT_CR8_MASK;
460
461         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
462                                         INTERCEPT_DR1_MASK |
463                                         INTERCEPT_DR2_MASK |
464                                         INTERCEPT_DR3_MASK;
465
466         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
467                                         INTERCEPT_DR1_MASK |
468                                         INTERCEPT_DR2_MASK |
469                                         INTERCEPT_DR3_MASK |
470                                         INTERCEPT_DR5_MASK |
471                                         INTERCEPT_DR7_MASK;
472
473         control->intercept_exceptions = (1 << PF_VECTOR) |
474                                         (1 << UD_VECTOR);
475
476
477         control->intercept =    (1ULL << INTERCEPT_INTR) |
478                                 (1ULL << INTERCEPT_NMI) |
479                                 (1ULL << INTERCEPT_SMI) |
480                 /*
481                  * selective cr0 intercept bug?
482                  *      0:   0f 22 d8                mov    %eax,%cr3
483                  *      3:   0f 20 c0                mov    %cr0,%eax
484                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
485                  *      b:   0f 22 c0                mov    %eax,%cr0
486                  * set cr3 ->interception
487                  * get cr0 ->interception
488                  * set cr0 -> no interception
489                  */
490                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
491                                 (1ULL << INTERCEPT_CPUID) |
492                                 (1ULL << INTERCEPT_INVD) |
493                                 (1ULL << INTERCEPT_HLT) |
494                                 (1ULL << INTERCEPT_INVLPGA) |
495                                 (1ULL << INTERCEPT_IOIO_PROT) |
496                                 (1ULL << INTERCEPT_MSR_PROT) |
497                                 (1ULL << INTERCEPT_TASK_SWITCH) |
498                                 (1ULL << INTERCEPT_SHUTDOWN) |
499                                 (1ULL << INTERCEPT_VMRUN) |
500                                 (1ULL << INTERCEPT_VMMCALL) |
501                                 (1ULL << INTERCEPT_VMLOAD) |
502                                 (1ULL << INTERCEPT_VMSAVE) |
503                                 (1ULL << INTERCEPT_STGI) |
504                                 (1ULL << INTERCEPT_CLGI) |
505                                 (1ULL << INTERCEPT_SKINIT) |
506                                 (1ULL << INTERCEPT_WBINVD) |
507                                 (1ULL << INTERCEPT_MONITOR) |
508                                 (1ULL << INTERCEPT_MWAIT);
509
510         control->iopm_base_pa = iopm_base;
511         control->msrpm_base_pa = msrpm_base;
512         control->tsc_offset = 0;
513         control->int_ctl = V_INTR_MASKING_MASK;
514
515         init_seg(&save->es);
516         init_seg(&save->ss);
517         init_seg(&save->ds);
518         init_seg(&save->fs);
519         init_seg(&save->gs);
520
521         save->cs.selector = 0xf000;
522         /* Executable/Readable Code Segment */
523         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
524                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
525         save->cs.limit = 0xffff;
526         /*
527          * cs.base should really be 0xffff0000, but vmx can't handle that, so
528          * be consistent with it.
529          *
530          * Replace when we have real mode working for vmx.
531          */
532         save->cs.base = 0xf0000;
533
534         save->gdtr.limit = 0xffff;
535         save->idtr.limit = 0xffff;
536
537         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
538         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
539
540         save->efer = MSR_EFER_SVME_MASK;
541         save->dr6 = 0xffff0ff0;
542         save->dr7 = 0x400;
543         save->rflags = 2;
544         save->rip = 0x0000fff0;
545
546         /*
547          * cr0 val on cpu init should be 0x60000010, we enable cpu
548          * cache by default. the orderly way is to enable cache in bios.
549          */
550         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
551         save->cr4 = X86_CR4_PAE;
552         /* rdx = ?? */
553 }
554
555 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
556 {
557         struct vcpu_svm *svm = to_svm(vcpu);
558
559         init_vmcb(svm->vmcb);
560
561         if (vcpu->vcpu_id != 0) {
562                 svm->vmcb->save.rip = 0;
563                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
564                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
565         }
566
567         return 0;
568 }
569
570 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
571 {
572         struct vcpu_svm *svm;
573         struct page *page;
574         int err;
575
576         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
577         if (!svm) {
578                 err = -ENOMEM;
579                 goto out;
580         }
581
582         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
583         if (err)
584                 goto free_svm;
585
586         page = alloc_page(GFP_KERNEL);
587         if (!page) {
588                 err = -ENOMEM;
589                 goto uninit;
590         }
591
592         svm->vmcb = page_address(page);
593         clear_page(svm->vmcb);
594         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
595         svm->asid_generation = 0;
596         memset(svm->db_regs, 0, sizeof(svm->db_regs));
597         init_vmcb(svm->vmcb);
598
599         fx_init(&svm->vcpu);
600         svm->vcpu.fpu_active = 1;
601         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
602         if (svm->vcpu.vcpu_id == 0)
603                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
604
605         return &svm->vcpu;
606
607 uninit:
608         kvm_vcpu_uninit(&svm->vcpu);
609 free_svm:
610         kmem_cache_free(kvm_vcpu_cache, svm);
611 out:
612         return ERR_PTR(err);
613 }
614
615 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
616 {
617         struct vcpu_svm *svm = to_svm(vcpu);
618
619         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
620         kvm_vcpu_uninit(vcpu);
621         kmem_cache_free(kvm_vcpu_cache, svm);
622 }
623
624 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
625 {
626         struct vcpu_svm *svm = to_svm(vcpu);
627         int i;
628
629         if (unlikely(cpu != vcpu->cpu)) {
630                 u64 tsc_this, delta;
631
632                 /*
633                  * Make sure that the guest sees a monotonically
634                  * increasing TSC.
635                  */
636                 rdtscll(tsc_this);
637                 delta = vcpu->arch.host_tsc - tsc_this;
638                 svm->vmcb->control.tsc_offset += delta;
639                 vcpu->cpu = cpu;
640                 kvm_migrate_apic_timer(vcpu);
641         }
642
643         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
644                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
645 }
646
647 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
648 {
649         struct vcpu_svm *svm = to_svm(vcpu);
650         int i;
651
652         ++vcpu->stat.host_state_reload;
653         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
654                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
655
656         rdtscll(vcpu->arch.host_tsc);
657 }
658
659 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
660 {
661 }
662
663 static void svm_cache_regs(struct kvm_vcpu *vcpu)
664 {
665         struct vcpu_svm *svm = to_svm(vcpu);
666
667         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
668         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
669         vcpu->arch.rip = svm->vmcb->save.rip;
670 }
671
672 static void svm_decache_regs(struct kvm_vcpu *vcpu)
673 {
674         struct vcpu_svm *svm = to_svm(vcpu);
675         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
676         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
677         svm->vmcb->save.rip = vcpu->arch.rip;
678 }
679
680 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
681 {
682         return to_svm(vcpu)->vmcb->save.rflags;
683 }
684
685 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
686 {
687         to_svm(vcpu)->vmcb->save.rflags = rflags;
688 }
689
690 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
691 {
692         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
693
694         switch (seg) {
695         case VCPU_SREG_CS: return &save->cs;
696         case VCPU_SREG_DS: return &save->ds;
697         case VCPU_SREG_ES: return &save->es;
698         case VCPU_SREG_FS: return &save->fs;
699         case VCPU_SREG_GS: return &save->gs;
700         case VCPU_SREG_SS: return &save->ss;
701         case VCPU_SREG_TR: return &save->tr;
702         case VCPU_SREG_LDTR: return &save->ldtr;
703         }
704         BUG();
705         return NULL;
706 }
707
708 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
709 {
710         struct vmcb_seg *s = svm_seg(vcpu, seg);
711
712         return s->base;
713 }
714
715 static void svm_get_segment(struct kvm_vcpu *vcpu,
716                             struct kvm_segment *var, int seg)
717 {
718         struct vmcb_seg *s = svm_seg(vcpu, seg);
719
720         var->base = s->base;
721         var->limit = s->limit;
722         var->selector = s->selector;
723         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
724         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
725         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
726         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
727         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
728         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
729         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
730         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
731         var->unusable = !var->present;
732 }
733
734 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
735 {
736         struct vcpu_svm *svm = to_svm(vcpu);
737
738         dt->limit = svm->vmcb->save.idtr.limit;
739         dt->base = svm->vmcb->save.idtr.base;
740 }
741
742 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
743 {
744         struct vcpu_svm *svm = to_svm(vcpu);
745
746         svm->vmcb->save.idtr.limit = dt->limit;
747         svm->vmcb->save.idtr.base = dt->base ;
748 }
749
750 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
751 {
752         struct vcpu_svm *svm = to_svm(vcpu);
753
754         dt->limit = svm->vmcb->save.gdtr.limit;
755         dt->base = svm->vmcb->save.gdtr.base;
756 }
757
758 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
759 {
760         struct vcpu_svm *svm = to_svm(vcpu);
761
762         svm->vmcb->save.gdtr.limit = dt->limit;
763         svm->vmcb->save.gdtr.base = dt->base ;
764 }
765
766 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
767 {
768 }
769
770 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
771 {
772         struct vcpu_svm *svm = to_svm(vcpu);
773
774 #ifdef CONFIG_X86_64
775         if (vcpu->arch.shadow_efer & EFER_LME) {
776                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
777                         vcpu->arch.shadow_efer |= EFER_LMA;
778                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
779                 }
780
781                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
782                         vcpu->arch.shadow_efer &= ~EFER_LMA;
783                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
784                 }
785         }
786 #endif
787         if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
788                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
789                 vcpu->fpu_active = 1;
790         }
791
792         vcpu->arch.cr0 = cr0;
793         cr0 |= X86_CR0_PG | X86_CR0_WP;
794         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
795         if (!vcpu->fpu_active)
796                 cr0 |= X86_CR0_TS;
797         svm->vmcb->save.cr0 = cr0;
798 }
799
800 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
801 {
802        vcpu->arch.cr4 = cr4;
803        to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
804 }
805
806 static void svm_set_segment(struct kvm_vcpu *vcpu,
807                             struct kvm_segment *var, int seg)
808 {
809         struct vcpu_svm *svm = to_svm(vcpu);
810         struct vmcb_seg *s = svm_seg(vcpu, seg);
811
812         s->base = var->base;
813         s->limit = var->limit;
814         s->selector = var->selector;
815         if (var->unusable)
816                 s->attrib = 0;
817         else {
818                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
819                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
820                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
821                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
822                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
823                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
824                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
825                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
826         }
827         if (seg == VCPU_SREG_CS)
828                 svm->vmcb->save.cpl
829                         = (svm->vmcb->save.cs.attrib
830                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
831
832 }
833
834 /* FIXME:
835
836         svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
837         svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
838
839 */
840
841 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
842 {
843         return -EOPNOTSUPP;
844 }
845
846 static int svm_get_irq(struct kvm_vcpu *vcpu)
847 {
848         struct vcpu_svm *svm = to_svm(vcpu);
849         u32 exit_int_info = svm->vmcb->control.exit_int_info;
850
851         if (is_external_interrupt(exit_int_info))
852                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
853         return -1;
854 }
855
856 static void load_host_msrs(struct kvm_vcpu *vcpu)
857 {
858 #ifdef CONFIG_X86_64
859         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
860 #endif
861 }
862
863 static void save_host_msrs(struct kvm_vcpu *vcpu)
864 {
865 #ifdef CONFIG_X86_64
866         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
867 #endif
868 }
869
870 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
871 {
872         if (svm_data->next_asid > svm_data->max_asid) {
873                 ++svm_data->asid_generation;
874                 svm_data->next_asid = 1;
875                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
876         }
877
878         svm->vcpu.cpu = svm_data->cpu;
879         svm->asid_generation = svm_data->asid_generation;
880         svm->vmcb->control.asid = svm_data->next_asid++;
881 }
882
883 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
884 {
885         return to_svm(vcpu)->db_regs[dr];
886 }
887
888 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
889                        int *exception)
890 {
891         struct vcpu_svm *svm = to_svm(vcpu);
892
893         *exception = 0;
894
895         if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
896                 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
897                 svm->vmcb->save.dr6 |= DR6_BD_MASK;
898                 *exception = DB_VECTOR;
899                 return;
900         }
901
902         switch (dr) {
903         case 0 ... 3:
904                 svm->db_regs[dr] = value;
905                 return;
906         case 4 ... 5:
907                 if (vcpu->arch.cr4 & X86_CR4_DE) {
908                         *exception = UD_VECTOR;
909                         return;
910                 }
911         case 7: {
912                 if (value & ~((1ULL << 32) - 1)) {
913                         *exception = GP_VECTOR;
914                         return;
915                 }
916                 svm->vmcb->save.dr7 = value;
917                 return;
918         }
919         default:
920                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
921                        __FUNCTION__, dr);
922                 *exception = UD_VECTOR;
923                 return;
924         }
925 }
926
927 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
928 {
929         u32 exit_int_info = svm->vmcb->control.exit_int_info;
930         struct kvm *kvm = svm->vcpu.kvm;
931         u64 fault_address;
932         u32 error_code;
933
934         if (!irqchip_in_kernel(kvm) &&
935                 is_external_interrupt(exit_int_info))
936                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
937
938         fault_address  = svm->vmcb->control.exit_info_2;
939         error_code = svm->vmcb->control.exit_info_1;
940         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
941 }
942
943 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
944 {
945         int er;
946
947         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
948         if (er != EMULATE_DONE)
949                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
950         return 1;
951 }
952
953 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
954 {
955         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
956         if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
957                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
958         svm->vcpu.fpu_active = 1;
959
960         return 1;
961 }
962
963 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
964 {
965         /*
966          * VMCB is undefined after a SHUTDOWN intercept
967          * so reinitialize it.
968          */
969         clear_page(svm->vmcb);
970         init_vmcb(svm->vmcb);
971
972         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
973         return 0;
974 }
975
976 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
977 {
978         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
979         int size, down, in, string, rep;
980         unsigned port;
981
982         ++svm->vcpu.stat.io_exits;
983
984         svm->next_rip = svm->vmcb->control.exit_info_2;
985
986         string = (io_info & SVM_IOIO_STR_MASK) != 0;
987
988         if (string) {
989                 if (emulate_instruction(&svm->vcpu,
990                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
991                         return 0;
992                 return 1;
993         }
994
995         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
996         port = io_info >> 16;
997         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
998         rep = (io_info & SVM_IOIO_REP_MASK) != 0;
999         down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1000
1001         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1002 }
1003
1004 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1005 {
1006         return 1;
1007 }
1008
1009 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1010 {
1011         svm->next_rip = svm->vmcb->save.rip + 1;
1012         skip_emulated_instruction(&svm->vcpu);
1013         return kvm_emulate_halt(&svm->vcpu);
1014 }
1015
1016 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1017 {
1018         svm->next_rip = svm->vmcb->save.rip + 3;
1019         skip_emulated_instruction(&svm->vcpu);
1020         kvm_emulate_hypercall(&svm->vcpu);
1021         return 1;
1022 }
1023
1024 static int invalid_op_interception(struct vcpu_svm *svm,
1025                                    struct kvm_run *kvm_run)
1026 {
1027         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1028         return 1;
1029 }
1030
1031 static int task_switch_interception(struct vcpu_svm *svm,
1032                                     struct kvm_run *kvm_run)
1033 {
1034         pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1035         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1036         return 0;
1037 }
1038
1039 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1040 {
1041         svm->next_rip = svm->vmcb->save.rip + 2;
1042         kvm_emulate_cpuid(&svm->vcpu);
1043         return 1;
1044 }
1045
1046 static int emulate_on_interception(struct vcpu_svm *svm,
1047                                    struct kvm_run *kvm_run)
1048 {
1049         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1050                 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1051         return 1;
1052 }
1053
1054 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1055 {
1056         emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1057         if (irqchip_in_kernel(svm->vcpu.kvm))
1058                 return 1;
1059         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1060         return 0;
1061 }
1062
1063 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1064 {
1065         struct vcpu_svm *svm = to_svm(vcpu);
1066
1067         switch (ecx) {
1068         case MSR_IA32_TIME_STAMP_COUNTER: {
1069                 u64 tsc;
1070
1071                 rdtscll(tsc);
1072                 *data = svm->vmcb->control.tsc_offset + tsc;
1073                 break;
1074         }
1075         case MSR_K6_STAR:
1076                 *data = svm->vmcb->save.star;
1077                 break;
1078 #ifdef CONFIG_X86_64
1079         case MSR_LSTAR:
1080                 *data = svm->vmcb->save.lstar;
1081                 break;
1082         case MSR_CSTAR:
1083                 *data = svm->vmcb->save.cstar;
1084                 break;
1085         case MSR_KERNEL_GS_BASE:
1086                 *data = svm->vmcb->save.kernel_gs_base;
1087                 break;
1088         case MSR_SYSCALL_MASK:
1089                 *data = svm->vmcb->save.sfmask;
1090                 break;
1091 #endif
1092         case MSR_IA32_SYSENTER_CS:
1093                 *data = svm->vmcb->save.sysenter_cs;
1094                 break;
1095         case MSR_IA32_SYSENTER_EIP:
1096                 *data = svm->vmcb->save.sysenter_eip;
1097                 break;
1098         case MSR_IA32_SYSENTER_ESP:
1099                 *data = svm->vmcb->save.sysenter_esp;
1100                 break;
1101         default:
1102                 return kvm_get_msr_common(vcpu, ecx, data);
1103         }
1104         return 0;
1105 }
1106
1107 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1108 {
1109         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1110         u64 data;
1111
1112         if (svm_get_msr(&svm->vcpu, ecx, &data))
1113                 kvm_inject_gp(&svm->vcpu, 0);
1114         else {
1115                 svm->vmcb->save.rax = data & 0xffffffff;
1116                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1117                 svm->next_rip = svm->vmcb->save.rip + 2;
1118                 skip_emulated_instruction(&svm->vcpu);
1119         }
1120         return 1;
1121 }
1122
1123 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1124 {
1125         struct vcpu_svm *svm = to_svm(vcpu);
1126
1127         switch (ecx) {
1128         case MSR_IA32_TIME_STAMP_COUNTER: {
1129                 u64 tsc;
1130
1131                 rdtscll(tsc);
1132                 svm->vmcb->control.tsc_offset = data - tsc;
1133                 break;
1134         }
1135         case MSR_K6_STAR:
1136                 svm->vmcb->save.star = data;
1137                 break;
1138 #ifdef CONFIG_X86_64
1139         case MSR_LSTAR:
1140                 svm->vmcb->save.lstar = data;
1141                 break;
1142         case MSR_CSTAR:
1143                 svm->vmcb->save.cstar = data;
1144                 break;
1145         case MSR_KERNEL_GS_BASE:
1146                 svm->vmcb->save.kernel_gs_base = data;
1147                 break;
1148         case MSR_SYSCALL_MASK:
1149                 svm->vmcb->save.sfmask = data;
1150                 break;
1151 #endif
1152         case MSR_IA32_SYSENTER_CS:
1153                 svm->vmcb->save.sysenter_cs = data;
1154                 break;
1155         case MSR_IA32_SYSENTER_EIP:
1156                 svm->vmcb->save.sysenter_eip = data;
1157                 break;
1158         case MSR_IA32_SYSENTER_ESP:
1159                 svm->vmcb->save.sysenter_esp = data;
1160                 break;
1161         case MSR_K7_EVNTSEL0:
1162         case MSR_K7_EVNTSEL1:
1163         case MSR_K7_EVNTSEL2:
1164         case MSR_K7_EVNTSEL3:
1165                 /*
1166                  * only support writing 0 to the performance counters for now
1167                  * to make Windows happy. Should be replaced by a real
1168                  * performance counter emulation later.
1169                  */
1170                 if (data != 0)
1171                         goto unhandled;
1172                 break;
1173         default:
1174         unhandled:
1175                 return kvm_set_msr_common(vcpu, ecx, data);
1176         }
1177         return 0;
1178 }
1179
1180 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1181 {
1182         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1183         u64 data = (svm->vmcb->save.rax & -1u)
1184                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1185         svm->next_rip = svm->vmcb->save.rip + 2;
1186         if (svm_set_msr(&svm->vcpu, ecx, data))
1187                 kvm_inject_gp(&svm->vcpu, 0);
1188         else
1189                 skip_emulated_instruction(&svm->vcpu);
1190         return 1;
1191 }
1192
1193 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1194 {
1195         if (svm->vmcb->control.exit_info_1)
1196                 return wrmsr_interception(svm, kvm_run);
1197         else
1198                 return rdmsr_interception(svm, kvm_run);
1199 }
1200
1201 static int interrupt_window_interception(struct vcpu_svm *svm,
1202                                    struct kvm_run *kvm_run)
1203 {
1204         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1205         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1206         /*
1207          * If the user space waits to inject interrupts, exit as soon as
1208          * possible
1209          */
1210         if (kvm_run->request_interrupt_window &&
1211             !svm->vcpu.arch.irq_summary) {
1212                 ++svm->vcpu.stat.irq_window_exits;
1213                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1214                 return 0;
1215         }
1216
1217         return 1;
1218 }
1219
1220 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1221                                       struct kvm_run *kvm_run) = {
1222         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1223         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1224         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1225         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
1226         /* for now: */
1227         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1228         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1229         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1230         [SVM_EXIT_WRITE_CR8]                    = cr8_write_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 + UD_VECTOR]        = ud_interception,
1242         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1243         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
1244         [SVM_EXIT_INTR]                         = nop_on_interception,
1245         [SVM_EXIT_NMI]                          = nop_on_interception,
1246         [SVM_EXIT_SMI]                          = nop_on_interception,
1247         [SVM_EXIT_INIT]                         = nop_on_interception,
1248         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1249         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1250         [SVM_EXIT_CPUID]                        = cpuid_interception,
1251         [SVM_EXIT_INVD]                         = emulate_on_interception,
1252         [SVM_EXIT_HLT]                          = halt_interception,
1253         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1254         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1255         [SVM_EXIT_IOIO]                         = io_interception,
1256         [SVM_EXIT_MSR]                          = msr_interception,
1257         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1258         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1259         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1260         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
1261         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1262         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1263         [SVM_EXIT_STGI]                         = invalid_op_interception,
1264         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1265         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1266         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
1267         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
1268         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
1269 };
1270
1271
1272 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1273 {
1274         struct vcpu_svm *svm = to_svm(vcpu);
1275         u32 exit_code = svm->vmcb->control.exit_code;
1276
1277         kvm_reput_irq(svm);
1278
1279         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1280                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1281                 kvm_run->fail_entry.hardware_entry_failure_reason
1282                         = svm->vmcb->control.exit_code;
1283                 return 0;
1284         }
1285
1286         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1287             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1288                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1289                        "exit_code 0x%x\n",
1290                        __FUNCTION__, svm->vmcb->control.exit_int_info,
1291                        exit_code);
1292
1293         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1294             || !svm_exit_handlers[exit_code]) {
1295                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1296                 kvm_run->hw.hardware_exit_reason = exit_code;
1297                 return 0;
1298         }
1299
1300         return svm_exit_handlers[exit_code](svm, kvm_run);
1301 }
1302
1303 static void reload_tss(struct kvm_vcpu *vcpu)
1304 {
1305         int cpu = raw_smp_processor_id();
1306
1307         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1308         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1309         load_TR_desc();
1310 }
1311
1312 static void pre_svm_run(struct vcpu_svm *svm)
1313 {
1314         int cpu = raw_smp_processor_id();
1315
1316         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1317
1318         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1319         if (svm->vcpu.cpu != cpu ||
1320             svm->asid_generation != svm_data->asid_generation)
1321                 new_asid(svm, svm_data);
1322 }
1323
1324
1325 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1326 {
1327         struct vmcb_control_area *control;
1328
1329         control = &svm->vmcb->control;
1330         control->int_vector = irq;
1331         control->int_ctl &= ~V_INTR_PRIO_MASK;
1332         control->int_ctl |= V_IRQ_MASK |
1333                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1334 }
1335
1336 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1337 {
1338         struct vcpu_svm *svm = to_svm(vcpu);
1339
1340         svm_inject_irq(svm, irq);
1341 }
1342
1343 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1344 {
1345         struct vcpu_svm *svm = to_svm(vcpu);
1346         struct vmcb *vmcb = svm->vmcb;
1347         int intr_vector = -1;
1348
1349         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1350             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1351                 intr_vector = vmcb->control.exit_int_info &
1352                               SVM_EVTINJ_VEC_MASK;
1353                 vmcb->control.exit_int_info = 0;
1354                 svm_inject_irq(svm, intr_vector);
1355                 return;
1356         }
1357
1358         if (vmcb->control.int_ctl & V_IRQ_MASK)
1359                 return;
1360
1361         if (!kvm_cpu_has_interrupt(vcpu))
1362                 return;
1363
1364         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1365             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1366             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1367                 /* unable to deliver irq, set pending irq */
1368                 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1369                 svm_inject_irq(svm, 0x0);
1370                 return;
1371         }
1372         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1373         intr_vector = kvm_cpu_get_interrupt(vcpu);
1374         svm_inject_irq(svm, intr_vector);
1375         kvm_timer_intr_post(vcpu, intr_vector);
1376 }
1377
1378 static void kvm_reput_irq(struct vcpu_svm *svm)
1379 {
1380         struct vmcb_control_area *control = &svm->vmcb->control;
1381
1382         if ((control->int_ctl & V_IRQ_MASK)
1383             && !irqchip_in_kernel(svm->vcpu.kvm)) {
1384                 control->int_ctl &= ~V_IRQ_MASK;
1385                 push_irq(&svm->vcpu, control->int_vector);
1386         }
1387
1388         svm->vcpu.arch.interrupt_window_open =
1389                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1390 }
1391
1392 static void svm_do_inject_vector(struct vcpu_svm *svm)
1393 {
1394         struct kvm_vcpu *vcpu = &svm->vcpu;
1395         int word_index = __ffs(vcpu->arch.irq_summary);
1396         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1397         int irq = word_index * BITS_PER_LONG + bit_index;
1398
1399         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1400         if (!vcpu->arch.irq_pending[word_index])
1401                 clear_bit(word_index, &vcpu->arch.irq_summary);
1402         svm_inject_irq(svm, irq);
1403 }
1404
1405 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1406                                        struct kvm_run *kvm_run)
1407 {
1408         struct vcpu_svm *svm = to_svm(vcpu);
1409         struct vmcb_control_area *control = &svm->vmcb->control;
1410
1411         svm->vcpu.arch.interrupt_window_open =
1412                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1413                  (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1414
1415         if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1416                 /*
1417                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1418                  */
1419                 svm_do_inject_vector(svm);
1420
1421         /*
1422          * Interrupts blocked.  Wait for unblock.
1423          */
1424         if (!svm->vcpu.arch.interrupt_window_open &&
1425             (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1426                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1427          else
1428                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1429 }
1430
1431 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1432 {
1433         return 0;
1434 }
1435
1436 static void save_db_regs(unsigned long *db_regs)
1437 {
1438         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1439         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1440         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1441         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1442 }
1443
1444 static void load_db_regs(unsigned long *db_regs)
1445 {
1446         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1447         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1448         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1449         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1450 }
1451
1452 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1453 {
1454         force_new_asid(vcpu);
1455 }
1456
1457 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1458 {
1459 }
1460
1461 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1462 {
1463         struct vcpu_svm *svm = to_svm(vcpu);
1464         u16 fs_selector;
1465         u16 gs_selector;
1466         u16 ldt_selector;
1467
1468         pre_svm_run(svm);
1469
1470         save_host_msrs(vcpu);
1471         fs_selector = read_fs();
1472         gs_selector = read_gs();
1473         ldt_selector = read_ldt();
1474         svm->host_cr2 = kvm_read_cr2();
1475         svm->host_dr6 = read_dr6();
1476         svm->host_dr7 = read_dr7();
1477         svm->vmcb->save.cr2 = vcpu->arch.cr2;
1478
1479         if (svm->vmcb->save.dr7 & 0xff) {
1480                 write_dr7(0);
1481                 save_db_regs(svm->host_db_regs);
1482                 load_db_regs(svm->db_regs);
1483         }
1484
1485         clgi();
1486
1487         local_irq_enable();
1488
1489         asm volatile (
1490 #ifdef CONFIG_X86_64
1491                 "push %%rbp; \n\t"
1492 #else
1493                 "push %%ebp; \n\t"
1494 #endif
1495
1496 #ifdef CONFIG_X86_64
1497                 "mov %c[rbx](%[svm]), %%rbx \n\t"
1498                 "mov %c[rcx](%[svm]), %%rcx \n\t"
1499                 "mov %c[rdx](%[svm]), %%rdx \n\t"
1500                 "mov %c[rsi](%[svm]), %%rsi \n\t"
1501                 "mov %c[rdi](%[svm]), %%rdi \n\t"
1502                 "mov %c[rbp](%[svm]), %%rbp \n\t"
1503                 "mov %c[r8](%[svm]),  %%r8  \n\t"
1504                 "mov %c[r9](%[svm]),  %%r9  \n\t"
1505                 "mov %c[r10](%[svm]), %%r10 \n\t"
1506                 "mov %c[r11](%[svm]), %%r11 \n\t"
1507                 "mov %c[r12](%[svm]), %%r12 \n\t"
1508                 "mov %c[r13](%[svm]), %%r13 \n\t"
1509                 "mov %c[r14](%[svm]), %%r14 \n\t"
1510                 "mov %c[r15](%[svm]), %%r15 \n\t"
1511 #else
1512                 "mov %c[rbx](%[svm]), %%ebx \n\t"
1513                 "mov %c[rcx](%[svm]), %%ecx \n\t"
1514                 "mov %c[rdx](%[svm]), %%edx \n\t"
1515                 "mov %c[rsi](%[svm]), %%esi \n\t"
1516                 "mov %c[rdi](%[svm]), %%edi \n\t"
1517                 "mov %c[rbp](%[svm]), %%ebp \n\t"
1518 #endif
1519
1520 #ifdef CONFIG_X86_64
1521                 /* Enter guest mode */
1522                 "push %%rax \n\t"
1523                 "mov %c[vmcb](%[svm]), %%rax \n\t"
1524                 SVM_VMLOAD "\n\t"
1525                 SVM_VMRUN "\n\t"
1526                 SVM_VMSAVE "\n\t"
1527                 "pop %%rax \n\t"
1528 #else
1529                 /* Enter guest mode */
1530                 "push %%eax \n\t"
1531                 "mov %c[vmcb](%[svm]), %%eax \n\t"
1532                 SVM_VMLOAD "\n\t"
1533                 SVM_VMRUN "\n\t"
1534                 SVM_VMSAVE "\n\t"
1535                 "pop %%eax \n\t"
1536 #endif
1537
1538                 /* Save guest registers, load host registers */
1539 #ifdef CONFIG_X86_64
1540                 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1541                 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1542                 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1543                 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1544                 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1545                 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1546                 "mov %%r8,  %c[r8](%[svm]) \n\t"
1547                 "mov %%r9,  %c[r9](%[svm]) \n\t"
1548                 "mov %%r10, %c[r10](%[svm]) \n\t"
1549                 "mov %%r11, %c[r11](%[svm]) \n\t"
1550                 "mov %%r12, %c[r12](%[svm]) \n\t"
1551                 "mov %%r13, %c[r13](%[svm]) \n\t"
1552                 "mov %%r14, %c[r14](%[svm]) \n\t"
1553                 "mov %%r15, %c[r15](%[svm]) \n\t"
1554
1555                 "pop  %%rbp; \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; \n\t"
1565 #endif
1566                 :
1567                 : [svm]"a"(svm),
1568                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1569                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1570                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1571                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1572                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1573                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1574                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1575 #ifdef CONFIG_X86_64
1576                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1577                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1578                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1579                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1580                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1581                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1582                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1583                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1584 #endif
1585                 : "cc", "memory"
1586 #ifdef CONFIG_X86_64
1587                 , "rbx", "rcx", "rdx", "rsi", "rdi"
1588                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1589 #else
1590                 , "ebx", "ecx", "edx" , "esi", "edi"
1591 #endif
1592                 );
1593
1594         if ((svm->vmcb->save.dr7 & 0xff))
1595                 load_db_regs(svm->host_db_regs);
1596
1597         vcpu->arch.cr2 = svm->vmcb->save.cr2;
1598
1599         write_dr6(svm->host_dr6);
1600         write_dr7(svm->host_dr7);
1601         kvm_write_cr2(svm->host_cr2);
1602
1603         load_fs(fs_selector);
1604         load_gs(gs_selector);
1605         load_ldt(ldt_selector);
1606         load_host_msrs(vcpu);
1607
1608         reload_tss(vcpu);
1609
1610         local_irq_disable();
1611
1612         stgi();
1613
1614         svm->next_rip = 0;
1615 }
1616
1617 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1618 {
1619         struct vcpu_svm *svm = to_svm(vcpu);
1620
1621         svm->vmcb->save.cr3 = root;
1622         force_new_asid(vcpu);
1623
1624         if (vcpu->fpu_active) {
1625                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1626                 svm->vmcb->save.cr0 |= X86_CR0_TS;
1627                 vcpu->fpu_active = 0;
1628         }
1629 }
1630
1631 static int is_disabled(void)
1632 {
1633         u64 vm_cr;
1634
1635         rdmsrl(MSR_VM_CR, vm_cr);
1636         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1637                 return 1;
1638
1639         return 0;
1640 }
1641
1642 static void
1643 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1644 {
1645         /*
1646          * Patch in the VMMCALL instruction:
1647          */
1648         hypercall[0] = 0x0f;
1649         hypercall[1] = 0x01;
1650         hypercall[2] = 0xd9;
1651 }
1652
1653 static void svm_check_processor_compat(void *rtn)
1654 {
1655         *(int *)rtn = 0;
1656 }
1657
1658 static bool svm_cpu_has_accelerated_tpr(void)
1659 {
1660         return false;
1661 }
1662
1663 static struct kvm_x86_ops svm_x86_ops = {
1664         .cpu_has_kvm_support = has_svm,
1665         .disabled_by_bios = is_disabled,
1666         .hardware_setup = svm_hardware_setup,
1667         .hardware_unsetup = svm_hardware_unsetup,
1668         .check_processor_compatibility = svm_check_processor_compat,
1669         .hardware_enable = svm_hardware_enable,
1670         .hardware_disable = svm_hardware_disable,
1671         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1672
1673         .vcpu_create = svm_create_vcpu,
1674         .vcpu_free = svm_free_vcpu,
1675         .vcpu_reset = svm_vcpu_reset,
1676
1677         .prepare_guest_switch = svm_prepare_guest_switch,
1678         .vcpu_load = svm_vcpu_load,
1679         .vcpu_put = svm_vcpu_put,
1680         .vcpu_decache = svm_vcpu_decache,
1681
1682         .set_guest_debug = svm_guest_debug,
1683         .get_msr = svm_get_msr,
1684         .set_msr = svm_set_msr,
1685         .get_segment_base = svm_get_segment_base,
1686         .get_segment = svm_get_segment,
1687         .set_segment = svm_set_segment,
1688         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1689         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1690         .set_cr0 = svm_set_cr0,
1691         .set_cr3 = svm_set_cr3,
1692         .set_cr4 = svm_set_cr4,
1693         .set_efer = svm_set_efer,
1694         .get_idt = svm_get_idt,
1695         .set_idt = svm_set_idt,
1696         .get_gdt = svm_get_gdt,
1697         .set_gdt = svm_set_gdt,
1698         .get_dr = svm_get_dr,
1699         .set_dr = svm_set_dr,
1700         .cache_regs = svm_cache_regs,
1701         .decache_regs = svm_decache_regs,
1702         .get_rflags = svm_get_rflags,
1703         .set_rflags = svm_set_rflags,
1704
1705         .tlb_flush = svm_flush_tlb,
1706
1707         .run = svm_vcpu_run,
1708         .handle_exit = handle_exit,
1709         .skip_emulated_instruction = skip_emulated_instruction,
1710         .patch_hypercall = svm_patch_hypercall,
1711         .get_irq = svm_get_irq,
1712         .set_irq = svm_set_irq,
1713         .queue_exception = svm_queue_exception,
1714         .exception_injected = svm_exception_injected,
1715         .inject_pending_irq = svm_intr_assist,
1716         .inject_pending_vectors = do_interrupt_requests,
1717
1718         .set_tss_addr = svm_set_tss_addr,
1719 };
1720
1721 static int __init svm_init(void)
1722 {
1723         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1724                               THIS_MODULE);
1725 }
1726
1727 static void __exit svm_exit(void)
1728 {
1729         kvm_exit();
1730 }
1731
1732 module_init(svm_init)
1733 module_exit(svm_exit)