KVM: SVM: Add microcode patch level dummy
[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 #include "kvm_cache_regs.h"
22
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/vmalloc.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28
29 #include <asm/desc.h>
30
31 #include <asm/virtext.h>
32
33 #define __ex(x) __kvm_handle_fault_on_reboot(x)
34
35 MODULE_AUTHOR("Qumranet");
36 MODULE_LICENSE("GPL");
37
38 #define IOPM_ALLOC_ORDER 2
39 #define MSRPM_ALLOC_ORDER 1
40
41 #define SEG_TYPE_LDT 2
42 #define SEG_TYPE_BUSY_TSS16 3
43
44 #define SVM_FEATURE_NPT  (1 << 0)
45 #define SVM_FEATURE_LBRV (1 << 1)
46 #define SVM_FEATURE_SVML (1 << 2)
47
48 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
49
50 /* Turn on to get debugging output*/
51 /* #define NESTED_DEBUG */
52
53 #ifdef NESTED_DEBUG
54 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
55 #else
56 #define nsvm_printk(fmt, args...) do {} while(0)
57 #endif
58
59 /* enable NPT for AMD64 and X86 with PAE */
60 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
61 static bool npt_enabled = true;
62 #else
63 static bool npt_enabled = false;
64 #endif
65 static int npt = 1;
66
67 module_param(npt, int, S_IRUGO);
68
69 static int nested = 0;
70 module_param(nested, int, S_IRUGO);
71
72 static void kvm_reput_irq(struct vcpu_svm *svm);
73 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
74
75 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override);
76 static int nested_svm_vmexit(struct vcpu_svm *svm);
77 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
78                              void *arg2, void *opaque);
79 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
80                                       bool has_error_code, u32 error_code);
81
82 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
83 {
84         return container_of(vcpu, struct vcpu_svm, vcpu);
85 }
86
87 static inline bool is_nested(struct vcpu_svm *svm)
88 {
89         return svm->nested_vmcb;
90 }
91
92 static unsigned long iopm_base;
93
94 struct kvm_ldttss_desc {
95         u16 limit0;
96         u16 base0;
97         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
98         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
99         u32 base3;
100         u32 zero1;
101 } __attribute__((packed));
102
103 struct svm_cpu_data {
104         int cpu;
105
106         u64 asid_generation;
107         u32 max_asid;
108         u32 next_asid;
109         struct kvm_ldttss_desc *tss_desc;
110
111         struct page *save_area;
112 };
113
114 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
115 static uint32_t svm_features;
116
117 struct svm_init_data {
118         int cpu;
119         int r;
120 };
121
122 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
123
124 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
125 #define MSRS_RANGE_SIZE 2048
126 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
127
128 #define MAX_INST_SIZE 15
129
130 static inline u32 svm_has(u32 feat)
131 {
132         return svm_features & feat;
133 }
134
135 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
136 {
137         int word_index = __ffs(vcpu->arch.irq_summary);
138         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
139         int irq = word_index * BITS_PER_LONG + bit_index;
140
141         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
142         if (!vcpu->arch.irq_pending[word_index])
143                 clear_bit(word_index, &vcpu->arch.irq_summary);
144         return irq;
145 }
146
147 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
148 {
149         set_bit(irq, vcpu->arch.irq_pending);
150         set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
151 }
152
153 static inline void clgi(void)
154 {
155         asm volatile (__ex(SVM_CLGI));
156 }
157
158 static inline void stgi(void)
159 {
160         asm volatile (__ex(SVM_STGI));
161 }
162
163 static inline void invlpga(unsigned long addr, u32 asid)
164 {
165         asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
166 }
167
168 static inline unsigned long kvm_read_cr2(void)
169 {
170         unsigned long cr2;
171
172         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
173         return cr2;
174 }
175
176 static inline void kvm_write_cr2(unsigned long val)
177 {
178         asm volatile ("mov %0, %%cr2" :: "r" (val));
179 }
180
181 static inline void force_new_asid(struct kvm_vcpu *vcpu)
182 {
183         to_svm(vcpu)->asid_generation--;
184 }
185
186 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
187 {
188         force_new_asid(vcpu);
189 }
190
191 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
192 {
193         if (!npt_enabled && !(efer & EFER_LMA))
194                 efer &= ~EFER_LME;
195
196         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
197         vcpu->arch.shadow_efer = efer;
198 }
199
200 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
201                                 bool has_error_code, u32 error_code)
202 {
203         struct vcpu_svm *svm = to_svm(vcpu);
204
205         /* If we are within a nested VM we'd better #VMEXIT and let the
206            guest handle the exception */
207         if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
208                 return;
209
210         svm->vmcb->control.event_inj = nr
211                 | SVM_EVTINJ_VALID
212                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
213                 | SVM_EVTINJ_TYPE_EXEPT;
214         svm->vmcb->control.event_inj_err = error_code;
215 }
216
217 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
218 {
219         struct vcpu_svm *svm = to_svm(vcpu);
220
221         return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
222 }
223
224 static int is_external_interrupt(u32 info)
225 {
226         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
227         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
228 }
229
230 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
231 {
232         struct vcpu_svm *svm = to_svm(vcpu);
233
234         if (!svm->next_rip) {
235                 printk(KERN_DEBUG "%s: NOP\n", __func__);
236                 return;
237         }
238         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
239                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
240                        __func__, kvm_rip_read(vcpu), svm->next_rip);
241
242         kvm_rip_write(vcpu, svm->next_rip);
243         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
244
245         vcpu->arch.interrupt_window_open = (svm->vcpu.arch.hflags & HF_GIF_MASK);
246 }
247
248 static int has_svm(void)
249 {
250         const char *msg;
251
252         if (!cpu_has_svm(&msg)) {
253                 printk(KERN_INFO "has_svn: %s\n", msg);
254                 return 0;
255         }
256
257         return 1;
258 }
259
260 static void svm_hardware_disable(void *garbage)
261 {
262         cpu_svm_disable();
263 }
264
265 static void svm_hardware_enable(void *garbage)
266 {
267
268         struct svm_cpu_data *svm_data;
269         uint64_t efer;
270         struct desc_ptr gdt_descr;
271         struct desc_struct *gdt;
272         int me = raw_smp_processor_id();
273
274         if (!has_svm()) {
275                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
276                 return;
277         }
278         svm_data = per_cpu(svm_data, me);
279
280         if (!svm_data) {
281                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
282                        me);
283                 return;
284         }
285
286         svm_data->asid_generation = 1;
287         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
288         svm_data->next_asid = svm_data->max_asid + 1;
289
290         asm volatile ("sgdt %0" : "=m"(gdt_descr));
291         gdt = (struct desc_struct *)gdt_descr.address;
292         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
293
294         rdmsrl(MSR_EFER, efer);
295         wrmsrl(MSR_EFER, efer | EFER_SVME);
296
297         wrmsrl(MSR_VM_HSAVE_PA,
298                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
299 }
300
301 static void svm_cpu_uninit(int cpu)
302 {
303         struct svm_cpu_data *svm_data
304                 = per_cpu(svm_data, raw_smp_processor_id());
305
306         if (!svm_data)
307                 return;
308
309         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
310         __free_page(svm_data->save_area);
311         kfree(svm_data);
312 }
313
314 static int svm_cpu_init(int cpu)
315 {
316         struct svm_cpu_data *svm_data;
317         int r;
318
319         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
320         if (!svm_data)
321                 return -ENOMEM;
322         svm_data->cpu = cpu;
323         svm_data->save_area = alloc_page(GFP_KERNEL);
324         r = -ENOMEM;
325         if (!svm_data->save_area)
326                 goto err_1;
327
328         per_cpu(svm_data, cpu) = svm_data;
329
330         return 0;
331
332 err_1:
333         kfree(svm_data);
334         return r;
335
336 }
337
338 static void set_msr_interception(u32 *msrpm, unsigned msr,
339                                  int read, int write)
340 {
341         int i;
342
343         for (i = 0; i < NUM_MSR_MAPS; i++) {
344                 if (msr >= msrpm_ranges[i] &&
345                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
346                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
347                                           msrpm_ranges[i]) * 2;
348
349                         u32 *base = msrpm + (msr_offset / 32);
350                         u32 msr_shift = msr_offset % 32;
351                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
352                         *base = (*base & ~(0x3 << msr_shift)) |
353                                 (mask << msr_shift);
354                         return;
355                 }
356         }
357         BUG();
358 }
359
360 static void svm_vcpu_init_msrpm(u32 *msrpm)
361 {
362         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
363
364 #ifdef CONFIG_X86_64
365         set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
366         set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
367         set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
368         set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
369         set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
370         set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
371 #endif
372         set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
373         set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
374         set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
375         set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
376 }
377
378 static void svm_enable_lbrv(struct vcpu_svm *svm)
379 {
380         u32 *msrpm = svm->msrpm;
381
382         svm->vmcb->control.lbr_ctl = 1;
383         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
384         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
385         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
386         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
387 }
388
389 static void svm_disable_lbrv(struct vcpu_svm *svm)
390 {
391         u32 *msrpm = svm->msrpm;
392
393         svm->vmcb->control.lbr_ctl = 0;
394         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
395         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
396         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
397         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
398 }
399
400 static __init int svm_hardware_setup(void)
401 {
402         int cpu;
403         struct page *iopm_pages;
404         void *iopm_va;
405         int r;
406
407         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
408
409         if (!iopm_pages)
410                 return -ENOMEM;
411
412         iopm_va = page_address(iopm_pages);
413         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
414         clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
415         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
416
417         if (boot_cpu_has(X86_FEATURE_NX))
418                 kvm_enable_efer_bits(EFER_NX);
419
420         if (nested) {
421                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
422                 kvm_enable_efer_bits(EFER_SVME);
423         }
424
425         for_each_online_cpu(cpu) {
426                 r = svm_cpu_init(cpu);
427                 if (r)
428                         goto err;
429         }
430
431         svm_features = cpuid_edx(SVM_CPUID_FUNC);
432
433         if (!svm_has(SVM_FEATURE_NPT))
434                 npt_enabled = false;
435
436         if (npt_enabled && !npt) {
437                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
438                 npt_enabled = false;
439         }
440
441         if (npt_enabled) {
442                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
443                 kvm_enable_tdp();
444         } else
445                 kvm_disable_tdp();
446
447         return 0;
448
449 err:
450         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
451         iopm_base = 0;
452         return r;
453 }
454
455 static __exit void svm_hardware_unsetup(void)
456 {
457         int cpu;
458
459         for_each_online_cpu(cpu)
460                 svm_cpu_uninit(cpu);
461
462         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
463         iopm_base = 0;
464 }
465
466 static void init_seg(struct vmcb_seg *seg)
467 {
468         seg->selector = 0;
469         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
470                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
471         seg->limit = 0xffff;
472         seg->base = 0;
473 }
474
475 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
476 {
477         seg->selector = 0;
478         seg->attrib = SVM_SELECTOR_P_MASK | type;
479         seg->limit = 0xffff;
480         seg->base = 0;
481 }
482
483 static void init_vmcb(struct vcpu_svm *svm)
484 {
485         struct vmcb_control_area *control = &svm->vmcb->control;
486         struct vmcb_save_area *save = &svm->vmcb->save;
487
488         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
489                                         INTERCEPT_CR3_MASK |
490                                         INTERCEPT_CR4_MASK;
491
492         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
493                                         INTERCEPT_CR3_MASK |
494                                         INTERCEPT_CR4_MASK |
495                                         INTERCEPT_CR8_MASK;
496
497         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
498                                         INTERCEPT_DR1_MASK |
499                                         INTERCEPT_DR2_MASK |
500                                         INTERCEPT_DR3_MASK;
501
502         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
503                                         INTERCEPT_DR1_MASK |
504                                         INTERCEPT_DR2_MASK |
505                                         INTERCEPT_DR3_MASK |
506                                         INTERCEPT_DR5_MASK |
507                                         INTERCEPT_DR7_MASK;
508
509         control->intercept_exceptions = (1 << PF_VECTOR) |
510                                         (1 << UD_VECTOR) |
511                                         (1 << MC_VECTOR);
512
513
514         control->intercept =    (1ULL << INTERCEPT_INTR) |
515                                 (1ULL << INTERCEPT_NMI) |
516                                 (1ULL << INTERCEPT_SMI) |
517                                 (1ULL << INTERCEPT_CPUID) |
518                                 (1ULL << INTERCEPT_INVD) |
519                                 (1ULL << INTERCEPT_HLT) |
520                                 (1ULL << INTERCEPT_INVLPG) |
521                                 (1ULL << INTERCEPT_INVLPGA) |
522                                 (1ULL << INTERCEPT_IOIO_PROT) |
523                                 (1ULL << INTERCEPT_MSR_PROT) |
524                                 (1ULL << INTERCEPT_TASK_SWITCH) |
525                                 (1ULL << INTERCEPT_SHUTDOWN) |
526                                 (1ULL << INTERCEPT_VMRUN) |
527                                 (1ULL << INTERCEPT_VMMCALL) |
528                                 (1ULL << INTERCEPT_VMLOAD) |
529                                 (1ULL << INTERCEPT_VMSAVE) |
530                                 (1ULL << INTERCEPT_STGI) |
531                                 (1ULL << INTERCEPT_CLGI) |
532                                 (1ULL << INTERCEPT_SKINIT) |
533                                 (1ULL << INTERCEPT_WBINVD) |
534                                 (1ULL << INTERCEPT_MONITOR) |
535                                 (1ULL << INTERCEPT_MWAIT);
536
537         control->iopm_base_pa = iopm_base;
538         control->msrpm_base_pa = __pa(svm->msrpm);
539         control->tsc_offset = 0;
540         control->int_ctl = V_INTR_MASKING_MASK;
541
542         init_seg(&save->es);
543         init_seg(&save->ss);
544         init_seg(&save->ds);
545         init_seg(&save->fs);
546         init_seg(&save->gs);
547
548         save->cs.selector = 0xf000;
549         /* Executable/Readable Code Segment */
550         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
551                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
552         save->cs.limit = 0xffff;
553         /*
554          * cs.base should really be 0xffff0000, but vmx can't handle that, so
555          * be consistent with it.
556          *
557          * Replace when we have real mode working for vmx.
558          */
559         save->cs.base = 0xf0000;
560
561         save->gdtr.limit = 0xffff;
562         save->idtr.limit = 0xffff;
563
564         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
565         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
566
567         save->efer = EFER_SVME;
568         save->dr6 = 0xffff0ff0;
569         save->dr7 = 0x400;
570         save->rflags = 2;
571         save->rip = 0x0000fff0;
572         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
573
574         /*
575          * cr0 val on cpu init should be 0x60000010, we enable cpu
576          * cache by default. the orderly way is to enable cache in bios.
577          */
578         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
579         save->cr4 = X86_CR4_PAE;
580         /* rdx = ?? */
581
582         if (npt_enabled) {
583                 /* Setup VMCB for Nested Paging */
584                 control->nested_ctl = 1;
585                 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
586                                         (1ULL << INTERCEPT_INVLPG));
587                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
588                 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
589                                                 INTERCEPT_CR3_MASK);
590                 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
591                                                  INTERCEPT_CR3_MASK);
592                 save->g_pat = 0x0007040600070406ULL;
593                 /* enable caching because the QEMU Bios doesn't enable it */
594                 save->cr0 = X86_CR0_ET;
595                 save->cr3 = 0;
596                 save->cr4 = 0;
597         }
598         force_new_asid(&svm->vcpu);
599
600         svm->nested_vmcb = 0;
601         svm->vcpu.arch.hflags = HF_GIF_MASK;
602 }
603
604 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
605 {
606         struct vcpu_svm *svm = to_svm(vcpu);
607
608         init_vmcb(svm);
609
610         if (vcpu->vcpu_id != 0) {
611                 kvm_rip_write(vcpu, 0);
612                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
613                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
614         }
615         vcpu->arch.regs_avail = ~0;
616         vcpu->arch.regs_dirty = ~0;
617
618         return 0;
619 }
620
621 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
622 {
623         struct vcpu_svm *svm;
624         struct page *page;
625         struct page *msrpm_pages;
626         struct page *hsave_page;
627         struct page *nested_msrpm_pages;
628         int err;
629
630         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
631         if (!svm) {
632                 err = -ENOMEM;
633                 goto out;
634         }
635
636         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
637         if (err)
638                 goto free_svm;
639
640         page = alloc_page(GFP_KERNEL);
641         if (!page) {
642                 err = -ENOMEM;
643                 goto uninit;
644         }
645
646         err = -ENOMEM;
647         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
648         if (!msrpm_pages)
649                 goto uninit;
650
651         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
652         if (!nested_msrpm_pages)
653                 goto uninit;
654
655         svm->msrpm = page_address(msrpm_pages);
656         svm_vcpu_init_msrpm(svm->msrpm);
657
658         hsave_page = alloc_page(GFP_KERNEL);
659         if (!hsave_page)
660                 goto uninit;
661         svm->hsave = page_address(hsave_page);
662
663         svm->nested_msrpm = page_address(nested_msrpm_pages);
664
665         svm->vmcb = page_address(page);
666         clear_page(svm->vmcb);
667         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
668         svm->asid_generation = 0;
669         init_vmcb(svm);
670
671         fx_init(&svm->vcpu);
672         svm->vcpu.fpu_active = 1;
673         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
674         if (svm->vcpu.vcpu_id == 0)
675                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
676
677         return &svm->vcpu;
678
679 uninit:
680         kvm_vcpu_uninit(&svm->vcpu);
681 free_svm:
682         kmem_cache_free(kvm_vcpu_cache, svm);
683 out:
684         return ERR_PTR(err);
685 }
686
687 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
688 {
689         struct vcpu_svm *svm = to_svm(vcpu);
690
691         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
692         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
693         __free_page(virt_to_page(svm->hsave));
694         __free_pages(virt_to_page(svm->nested_msrpm), MSRPM_ALLOC_ORDER);
695         kvm_vcpu_uninit(vcpu);
696         kmem_cache_free(kvm_vcpu_cache, svm);
697 }
698
699 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
700 {
701         struct vcpu_svm *svm = to_svm(vcpu);
702         int i;
703
704         if (unlikely(cpu != vcpu->cpu)) {
705                 u64 tsc_this, delta;
706
707                 /*
708                  * Make sure that the guest sees a monotonically
709                  * increasing TSC.
710                  */
711                 rdtscll(tsc_this);
712                 delta = vcpu->arch.host_tsc - tsc_this;
713                 svm->vmcb->control.tsc_offset += delta;
714                 vcpu->cpu = cpu;
715                 kvm_migrate_timers(vcpu);
716         }
717
718         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
719                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
720 }
721
722 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
723 {
724         struct vcpu_svm *svm = to_svm(vcpu);
725         int i;
726
727         ++vcpu->stat.host_state_reload;
728         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
729                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
730
731         rdtscll(vcpu->arch.host_tsc);
732 }
733
734 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
735 {
736         return to_svm(vcpu)->vmcb->save.rflags;
737 }
738
739 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
740 {
741         to_svm(vcpu)->vmcb->save.rflags = rflags;
742 }
743
744 static void svm_set_vintr(struct vcpu_svm *svm)
745 {
746         svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
747 }
748
749 static void svm_clear_vintr(struct vcpu_svm *svm)
750 {
751         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
752 }
753
754 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
755 {
756         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
757
758         switch (seg) {
759         case VCPU_SREG_CS: return &save->cs;
760         case VCPU_SREG_DS: return &save->ds;
761         case VCPU_SREG_ES: return &save->es;
762         case VCPU_SREG_FS: return &save->fs;
763         case VCPU_SREG_GS: return &save->gs;
764         case VCPU_SREG_SS: return &save->ss;
765         case VCPU_SREG_TR: return &save->tr;
766         case VCPU_SREG_LDTR: return &save->ldtr;
767         }
768         BUG();
769         return NULL;
770 }
771
772 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
773 {
774         struct vmcb_seg *s = svm_seg(vcpu, seg);
775
776         return s->base;
777 }
778
779 static void svm_get_segment(struct kvm_vcpu *vcpu,
780                             struct kvm_segment *var, int seg)
781 {
782         struct vmcb_seg *s = svm_seg(vcpu, seg);
783
784         var->base = s->base;
785         var->limit = s->limit;
786         var->selector = s->selector;
787         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
788         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
789         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
790         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
791         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
792         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
793         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
794         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
795
796         /*
797          * SVM always stores 0 for the 'G' bit in the CS selector in
798          * the VMCB on a VMEXIT. This hurts cross-vendor migration:
799          * Intel's VMENTRY has a check on the 'G' bit.
800          */
801         if (seg == VCPU_SREG_CS)
802                 var->g = s->limit > 0xfffff;
803
804         /*
805          * Work around a bug where the busy flag in the tr selector
806          * isn't exposed
807          */
808         if (seg == VCPU_SREG_TR)
809                 var->type |= 0x2;
810
811         var->unusable = !var->present;
812 }
813
814 static int svm_get_cpl(struct kvm_vcpu *vcpu)
815 {
816         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
817
818         return save->cpl;
819 }
820
821 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
822 {
823         struct vcpu_svm *svm = to_svm(vcpu);
824
825         dt->limit = svm->vmcb->save.idtr.limit;
826         dt->base = svm->vmcb->save.idtr.base;
827 }
828
829 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
830 {
831         struct vcpu_svm *svm = to_svm(vcpu);
832
833         svm->vmcb->save.idtr.limit = dt->limit;
834         svm->vmcb->save.idtr.base = dt->base ;
835 }
836
837 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
838 {
839         struct vcpu_svm *svm = to_svm(vcpu);
840
841         dt->limit = svm->vmcb->save.gdtr.limit;
842         dt->base = svm->vmcb->save.gdtr.base;
843 }
844
845 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
846 {
847         struct vcpu_svm *svm = to_svm(vcpu);
848
849         svm->vmcb->save.gdtr.limit = dt->limit;
850         svm->vmcb->save.gdtr.base = dt->base ;
851 }
852
853 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
854 {
855 }
856
857 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
858 {
859         struct vcpu_svm *svm = to_svm(vcpu);
860
861 #ifdef CONFIG_X86_64
862         if (vcpu->arch.shadow_efer & EFER_LME) {
863                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
864                         vcpu->arch.shadow_efer |= EFER_LMA;
865                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
866                 }
867
868                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
869                         vcpu->arch.shadow_efer &= ~EFER_LMA;
870                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
871                 }
872         }
873 #endif
874         if (npt_enabled)
875                 goto set;
876
877         if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
878                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
879                 vcpu->fpu_active = 1;
880         }
881
882         vcpu->arch.cr0 = cr0;
883         cr0 |= X86_CR0_PG | X86_CR0_WP;
884         if (!vcpu->fpu_active) {
885                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
886                 cr0 |= X86_CR0_TS;
887         }
888 set:
889         /*
890          * re-enable caching here because the QEMU bios
891          * does not do it - this results in some delay at
892          * reboot
893          */
894         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
895         svm->vmcb->save.cr0 = cr0;
896 }
897
898 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
899 {
900         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
901         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
902
903         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
904                 force_new_asid(vcpu);
905
906         vcpu->arch.cr4 = cr4;
907         if (!npt_enabled)
908                 cr4 |= X86_CR4_PAE;
909         cr4 |= host_cr4_mce;
910         to_svm(vcpu)->vmcb->save.cr4 = cr4;
911 }
912
913 static void svm_set_segment(struct kvm_vcpu *vcpu,
914                             struct kvm_segment *var, int seg)
915 {
916         struct vcpu_svm *svm = to_svm(vcpu);
917         struct vmcb_seg *s = svm_seg(vcpu, seg);
918
919         s->base = var->base;
920         s->limit = var->limit;
921         s->selector = var->selector;
922         if (var->unusable)
923                 s->attrib = 0;
924         else {
925                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
926                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
927                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
928                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
929                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
930                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
931                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
932                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
933         }
934         if (seg == VCPU_SREG_CS)
935                 svm->vmcb->save.cpl
936                         = (svm->vmcb->save.cs.attrib
937                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
938
939 }
940
941 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
942 {
943         int old_debug = vcpu->guest_debug;
944         struct vcpu_svm *svm = to_svm(vcpu);
945
946         vcpu->guest_debug = dbg->control;
947
948         svm->vmcb->control.intercept_exceptions &=
949                 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
950         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
951                 if (vcpu->guest_debug &
952                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
953                         svm->vmcb->control.intercept_exceptions |=
954                                 1 << DB_VECTOR;
955                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
956                         svm->vmcb->control.intercept_exceptions |=
957                                 1 << BP_VECTOR;
958         } else
959                 vcpu->guest_debug = 0;
960
961         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
962                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
963         else
964                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
965
966         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
967                 svm->vmcb->save.rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
968         else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
969                 svm->vmcb->save.rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
970
971         return 0;
972 }
973
974 static int svm_get_irq(struct kvm_vcpu *vcpu)
975 {
976         struct vcpu_svm *svm = to_svm(vcpu);
977         u32 exit_int_info = svm->vmcb->control.exit_int_info;
978
979         if (is_external_interrupt(exit_int_info))
980                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
981         return -1;
982 }
983
984 static void load_host_msrs(struct kvm_vcpu *vcpu)
985 {
986 #ifdef CONFIG_X86_64
987         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
988 #endif
989 }
990
991 static void save_host_msrs(struct kvm_vcpu *vcpu)
992 {
993 #ifdef CONFIG_X86_64
994         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
995 #endif
996 }
997
998 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
999 {
1000         if (svm_data->next_asid > svm_data->max_asid) {
1001                 ++svm_data->asid_generation;
1002                 svm_data->next_asid = 1;
1003                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1004         }
1005
1006         svm->vcpu.cpu = svm_data->cpu;
1007         svm->asid_generation = svm_data->asid_generation;
1008         svm->vmcb->control.asid = svm_data->next_asid++;
1009 }
1010
1011 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
1012 {
1013         struct vcpu_svm *svm = to_svm(vcpu);
1014         unsigned long val;
1015
1016         switch (dr) {
1017         case 0 ... 3:
1018                 val = vcpu->arch.db[dr];
1019                 break;
1020         case 6:
1021                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1022                         val = vcpu->arch.dr6;
1023                 else
1024                         val = svm->vmcb->save.dr6;
1025                 break;
1026         case 7:
1027                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1028                         val = vcpu->arch.dr7;
1029                 else
1030                         val = svm->vmcb->save.dr7;
1031                 break;
1032         default:
1033                 val = 0;
1034         }
1035
1036         KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
1037         return val;
1038 }
1039
1040 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
1041                        int *exception)
1042 {
1043         struct vcpu_svm *svm = to_svm(vcpu);
1044
1045         KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)value, handler);
1046
1047         *exception = 0;
1048
1049         switch (dr) {
1050         case 0 ... 3:
1051                 vcpu->arch.db[dr] = value;
1052                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1053                         vcpu->arch.eff_db[dr] = value;
1054                 return;
1055         case 4 ... 5:
1056                 if (vcpu->arch.cr4 & X86_CR4_DE)
1057                         *exception = UD_VECTOR;
1058                 return;
1059         case 6:
1060                 if (value & 0xffffffff00000000ULL) {
1061                         *exception = GP_VECTOR;
1062                         return;
1063                 }
1064                 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1065                 return;
1066         case 7:
1067                 if (value & 0xffffffff00000000ULL) {
1068                         *exception = GP_VECTOR;
1069                         return;
1070                 }
1071                 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1072                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1073                         svm->vmcb->save.dr7 = vcpu->arch.dr7;
1074                         vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1075                 }
1076                 return;
1077         default:
1078                 /* FIXME: Possible case? */
1079                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1080                        __func__, dr);
1081                 *exception = UD_VECTOR;
1082                 return;
1083         }
1084 }
1085
1086 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1087 {
1088         u32 exit_int_info = svm->vmcb->control.exit_int_info;
1089         struct kvm *kvm = svm->vcpu.kvm;
1090         u64 fault_address;
1091         u32 error_code;
1092         bool event_injection = false;
1093
1094         if (!irqchip_in_kernel(kvm) &&
1095             is_external_interrupt(exit_int_info)) {
1096                 event_injection = true;
1097                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1098         }
1099
1100         fault_address  = svm->vmcb->control.exit_info_2;
1101         error_code = svm->vmcb->control.exit_info_1;
1102
1103         if (!npt_enabled)
1104                 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1105                             (u32)fault_address, (u32)(fault_address >> 32),
1106                             handler);
1107         else
1108                 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1109                             (u32)fault_address, (u32)(fault_address >> 32),
1110                             handler);
1111         /*
1112          * FIXME: Tis shouldn't be necessary here, but there is a flush
1113          * missing in the MMU code. Until we find this bug, flush the
1114          * complete TLB here on an NPF
1115          */
1116         if (npt_enabled)
1117                 svm_flush_tlb(&svm->vcpu);
1118
1119         if (!npt_enabled && event_injection)
1120                 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1121         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1122 }
1123
1124 static int db_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1125 {
1126         if (!(svm->vcpu.guest_debug &
1127               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
1128                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1129                 return 1;
1130         }
1131         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1132         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1133         kvm_run->debug.arch.exception = DB_VECTOR;
1134         return 0;
1135 }
1136
1137 static int bp_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1138 {
1139         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1140         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1141         kvm_run->debug.arch.exception = BP_VECTOR;
1142         return 0;
1143 }
1144
1145 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1146 {
1147         int er;
1148
1149         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1150         if (er != EMULATE_DONE)
1151                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1152         return 1;
1153 }
1154
1155 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1156 {
1157         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1158         if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1159                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1160         svm->vcpu.fpu_active = 1;
1161
1162         return 1;
1163 }
1164
1165 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1166 {
1167         /*
1168          * On an #MC intercept the MCE handler is not called automatically in
1169          * the host. So do it by hand here.
1170          */
1171         asm volatile (
1172                 "int $0x12\n");
1173         /* not sure if we ever come back to this point */
1174
1175         return 1;
1176 }
1177
1178 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1179 {
1180         /*
1181          * VMCB is undefined after a SHUTDOWN intercept
1182          * so reinitialize it.
1183          */
1184         clear_page(svm->vmcb);
1185         init_vmcb(svm);
1186
1187         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1188         return 0;
1189 }
1190
1191 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1192 {
1193         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1194         int size, down, in, string, rep;
1195         unsigned port;
1196
1197         ++svm->vcpu.stat.io_exits;
1198
1199         svm->next_rip = svm->vmcb->control.exit_info_2;
1200
1201         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1202
1203         if (string) {
1204                 if (emulate_instruction(&svm->vcpu,
1205                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1206                         return 0;
1207                 return 1;
1208         }
1209
1210         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1211         port = io_info >> 16;
1212         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1213         rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1214         down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1215
1216         skip_emulated_instruction(&svm->vcpu);
1217         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1218 }
1219
1220 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1221 {
1222         KVMTRACE_0D(NMI, &svm->vcpu, handler);
1223         return 1;
1224 }
1225
1226 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1227 {
1228         ++svm->vcpu.stat.irq_exits;
1229         KVMTRACE_0D(INTR, &svm->vcpu, handler);
1230         return 1;
1231 }
1232
1233 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1234 {
1235         return 1;
1236 }
1237
1238 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1239 {
1240         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1241         skip_emulated_instruction(&svm->vcpu);
1242         return kvm_emulate_halt(&svm->vcpu);
1243 }
1244
1245 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1246 {
1247         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1248         skip_emulated_instruction(&svm->vcpu);
1249         kvm_emulate_hypercall(&svm->vcpu);
1250         return 1;
1251 }
1252
1253 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1254 {
1255         if (!(svm->vcpu.arch.shadow_efer & EFER_SVME)
1256             || !is_paging(&svm->vcpu)) {
1257                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1258                 return 1;
1259         }
1260
1261         if (svm->vmcb->save.cpl) {
1262                 kvm_inject_gp(&svm->vcpu, 0);
1263                 return 1;
1264         }
1265
1266        return 0;
1267 }
1268
1269 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1270                                       bool has_error_code, u32 error_code)
1271 {
1272         if (is_nested(svm)) {
1273                 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1274                 svm->vmcb->control.exit_code_hi = 0;
1275                 svm->vmcb->control.exit_info_1 = error_code;
1276                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1277                 if (nested_svm_exit_handled(svm, false)) {
1278                         nsvm_printk("VMexit -> EXCP 0x%x\n", nr);
1279
1280                         nested_svm_vmexit(svm);
1281                         return 1;
1282                 }
1283         }
1284
1285         return 0;
1286 }
1287
1288 static inline int nested_svm_intr(struct vcpu_svm *svm)
1289 {
1290         if (is_nested(svm)) {
1291                 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1292                         return 0;
1293
1294                 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1295                         return 0;
1296
1297                 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1298
1299                 if (nested_svm_exit_handled(svm, false)) {
1300                         nsvm_printk("VMexit -> INTR\n");
1301                         nested_svm_vmexit(svm);
1302                         return 1;
1303                 }
1304         }
1305
1306         return 0;
1307 }
1308
1309 static struct page *nested_svm_get_page(struct vcpu_svm *svm, u64 gpa)
1310 {
1311         struct page *page;
1312
1313         down_read(&current->mm->mmap_sem);
1314         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1315         up_read(&current->mm->mmap_sem);
1316
1317         if (is_error_page(page)) {
1318                 printk(KERN_INFO "%s: could not find page at 0x%llx\n",
1319                        __func__, gpa);
1320                 kvm_release_page_clean(page);
1321                 kvm_inject_gp(&svm->vcpu, 0);
1322                 return NULL;
1323         }
1324         return page;
1325 }
1326
1327 static int nested_svm_do(struct vcpu_svm *svm,
1328                          u64 arg1_gpa, u64 arg2_gpa, void *opaque,
1329                          int (*handler)(struct vcpu_svm *svm,
1330                                         void *arg1,
1331                                         void *arg2,
1332                                         void *opaque))
1333 {
1334         struct page *arg1_page;
1335         struct page *arg2_page = NULL;
1336         void *arg1;
1337         void *arg2 = NULL;
1338         int retval;
1339
1340         arg1_page = nested_svm_get_page(svm, arg1_gpa);
1341         if(arg1_page == NULL)
1342                 return 1;
1343
1344         if (arg2_gpa) {
1345                 arg2_page = nested_svm_get_page(svm, arg2_gpa);
1346                 if(arg2_page == NULL) {
1347                         kvm_release_page_clean(arg1_page);
1348                         return 1;
1349                 }
1350         }
1351
1352         arg1 = kmap_atomic(arg1_page, KM_USER0);
1353         if (arg2_gpa)
1354                 arg2 = kmap_atomic(arg2_page, KM_USER1);
1355
1356         retval = handler(svm, arg1, arg2, opaque);
1357
1358         kunmap_atomic(arg1, KM_USER0);
1359         if (arg2_gpa)
1360                 kunmap_atomic(arg2, KM_USER1);
1361
1362         kvm_release_page_dirty(arg1_page);
1363         if (arg2_gpa)
1364                 kvm_release_page_dirty(arg2_page);
1365
1366         return retval;
1367 }
1368
1369 static int nested_svm_exit_handled_real(struct vcpu_svm *svm,
1370                                         void *arg1,
1371                                         void *arg2,
1372                                         void *opaque)
1373 {
1374         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1375         bool kvm_overrides = *(bool *)opaque;
1376         u32 exit_code = svm->vmcb->control.exit_code;
1377
1378         if (kvm_overrides) {
1379                 switch (exit_code) {
1380                 case SVM_EXIT_INTR:
1381                 case SVM_EXIT_NMI:
1382                         return 0;
1383                 /* For now we are always handling NPFs when using them */
1384                 case SVM_EXIT_NPF:
1385                         if (npt_enabled)
1386                                 return 0;
1387                         break;
1388                 /* When we're shadowing, trap PFs */
1389                 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1390                         if (!npt_enabled)
1391                                 return 0;
1392                         break;
1393                 default:
1394                         break;
1395                 }
1396         }
1397
1398         switch (exit_code) {
1399         case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1400                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1401                 if (nested_vmcb->control.intercept_cr_read & cr_bits)
1402                         return 1;
1403                 break;
1404         }
1405         case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1406                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1407                 if (nested_vmcb->control.intercept_cr_write & cr_bits)
1408                         return 1;
1409                 break;
1410         }
1411         case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1412                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1413                 if (nested_vmcb->control.intercept_dr_read & dr_bits)
1414                         return 1;
1415                 break;
1416         }
1417         case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1418                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1419                 if (nested_vmcb->control.intercept_dr_write & dr_bits)
1420                         return 1;
1421                 break;
1422         }
1423         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1424                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1425                 if (nested_vmcb->control.intercept_exceptions & excp_bits)
1426                         return 1;
1427                 break;
1428         }
1429         default: {
1430                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1431                 nsvm_printk("exit code: 0x%x\n", exit_code);
1432                 if (nested_vmcb->control.intercept & exit_bits)
1433                         return 1;
1434         }
1435         }
1436
1437         return 0;
1438 }
1439
1440 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm,
1441                                        void *arg1, void *arg2,
1442                                        void *opaque)
1443 {
1444         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1445         u8 *msrpm = (u8 *)arg2;
1446         u32 t0, t1;
1447         u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1448         u32 param = svm->vmcb->control.exit_info_1 & 1;
1449
1450         if (!(nested_vmcb->control.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1451                 return 0;
1452
1453         switch(msr) {
1454         case 0 ... 0x1fff:
1455                 t0 = (msr * 2) % 8;
1456                 t1 = msr / 8;
1457                 break;
1458         case 0xc0000000 ... 0xc0001fff:
1459                 t0 = (8192 + msr - 0xc0000000) * 2;
1460                 t1 = (t0 / 8);
1461                 t0 %= 8;
1462                 break;
1463         case 0xc0010000 ... 0xc0011fff:
1464                 t0 = (16384 + msr - 0xc0010000) * 2;
1465                 t1 = (t0 / 8);
1466                 t0 %= 8;
1467                 break;
1468         default:
1469                 return 1;
1470                 break;
1471         }
1472         if (msrpm[t1] & ((1 << param) << t0))
1473                 return 1;
1474
1475         return 0;
1476 }
1477
1478 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override)
1479 {
1480         bool k = kvm_override;
1481
1482         switch (svm->vmcb->control.exit_code) {
1483         case SVM_EXIT_MSR:
1484                 return nested_svm_do(svm, svm->nested_vmcb,
1485                                      svm->nested_vmcb_msrpm, NULL,
1486                                      nested_svm_exit_handled_msr);
1487         default: break;
1488         }
1489
1490         return nested_svm_do(svm, svm->nested_vmcb, 0, &k,
1491                              nested_svm_exit_handled_real);
1492 }
1493
1494 static int nested_svm_vmexit_real(struct vcpu_svm *svm, void *arg1,
1495                                   void *arg2, void *opaque)
1496 {
1497         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1498         struct vmcb *hsave = svm->hsave;
1499         u64 nested_save[] = { nested_vmcb->save.cr0,
1500                               nested_vmcb->save.cr3,
1501                               nested_vmcb->save.cr4,
1502                               nested_vmcb->save.efer,
1503                               nested_vmcb->control.intercept_cr_read,
1504                               nested_vmcb->control.intercept_cr_write,
1505                               nested_vmcb->control.intercept_dr_read,
1506                               nested_vmcb->control.intercept_dr_write,
1507                               nested_vmcb->control.intercept_exceptions,
1508                               nested_vmcb->control.intercept,
1509                               nested_vmcb->control.msrpm_base_pa,
1510                               nested_vmcb->control.iopm_base_pa,
1511                               nested_vmcb->control.tsc_offset };
1512
1513         /* Give the current vmcb to the guest */
1514         memcpy(nested_vmcb, svm->vmcb, sizeof(struct vmcb));
1515         nested_vmcb->save.cr0 = nested_save[0];
1516         if (!npt_enabled)
1517                 nested_vmcb->save.cr3 = nested_save[1];
1518         nested_vmcb->save.cr4 = nested_save[2];
1519         nested_vmcb->save.efer = nested_save[3];
1520         nested_vmcb->control.intercept_cr_read = nested_save[4];
1521         nested_vmcb->control.intercept_cr_write = nested_save[5];
1522         nested_vmcb->control.intercept_dr_read = nested_save[6];
1523         nested_vmcb->control.intercept_dr_write = nested_save[7];
1524         nested_vmcb->control.intercept_exceptions = nested_save[8];
1525         nested_vmcb->control.intercept = nested_save[9];
1526         nested_vmcb->control.msrpm_base_pa = nested_save[10];
1527         nested_vmcb->control.iopm_base_pa = nested_save[11];
1528         nested_vmcb->control.tsc_offset = nested_save[12];
1529
1530         /* We always set V_INTR_MASKING and remember the old value in hflags */
1531         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1532                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1533
1534         if ((nested_vmcb->control.int_ctl & V_IRQ_MASK) &&
1535             (nested_vmcb->control.int_vector)) {
1536                 nsvm_printk("WARNING: IRQ 0x%x still enabled on #VMEXIT\n",
1537                                 nested_vmcb->control.int_vector);
1538         }
1539
1540         /* Restore the original control entries */
1541         svm->vmcb->control = hsave->control;
1542
1543         /* Kill any pending exceptions */
1544         if (svm->vcpu.arch.exception.pending == true)
1545                 nsvm_printk("WARNING: Pending Exception\n");
1546         svm->vcpu.arch.exception.pending = false;
1547
1548         /* Restore selected save entries */
1549         svm->vmcb->save.es = hsave->save.es;
1550         svm->vmcb->save.cs = hsave->save.cs;
1551         svm->vmcb->save.ss = hsave->save.ss;
1552         svm->vmcb->save.ds = hsave->save.ds;
1553         svm->vmcb->save.gdtr = hsave->save.gdtr;
1554         svm->vmcb->save.idtr = hsave->save.idtr;
1555         svm->vmcb->save.rflags = hsave->save.rflags;
1556         svm_set_efer(&svm->vcpu, hsave->save.efer);
1557         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1558         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1559         if (npt_enabled) {
1560                 svm->vmcb->save.cr3 = hsave->save.cr3;
1561                 svm->vcpu.arch.cr3 = hsave->save.cr3;
1562         } else {
1563                 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1564         }
1565         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1566         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1567         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1568         svm->vmcb->save.dr7 = 0;
1569         svm->vmcb->save.cpl = 0;
1570         svm->vmcb->control.exit_int_info = 0;
1571
1572         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1573         /* Exit nested SVM mode */
1574         svm->nested_vmcb = 0;
1575
1576         return 0;
1577 }
1578
1579 static int nested_svm_vmexit(struct vcpu_svm *svm)
1580 {
1581         nsvm_printk("VMexit\n");
1582         if (nested_svm_do(svm, svm->nested_vmcb, 0,
1583                           NULL, nested_svm_vmexit_real))
1584                 return 1;
1585
1586         kvm_mmu_reset_context(&svm->vcpu);
1587         kvm_mmu_load(&svm->vcpu);
1588
1589         return 0;
1590 }
1591
1592 static int nested_svm_vmrun_msrpm(struct vcpu_svm *svm, void *arg1,
1593                                   void *arg2, void *opaque)
1594 {
1595         int i;
1596         u32 *nested_msrpm = (u32*)arg1;
1597         for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1598                 svm->nested_msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1599         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested_msrpm);
1600
1601         return 0;
1602 }
1603
1604 static int nested_svm_vmrun(struct vcpu_svm *svm, void *arg1,
1605                             void *arg2, void *opaque)
1606 {
1607         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1608         struct vmcb *hsave = svm->hsave;
1609
1610         /* nested_vmcb is our indicator if nested SVM is activated */
1611         svm->nested_vmcb = svm->vmcb->save.rax;
1612
1613         /* Clear internal status */
1614         svm->vcpu.arch.exception.pending = false;
1615
1616         /* Save the old vmcb, so we don't need to pick what we save, but
1617            can restore everything when a VMEXIT occurs */
1618         memcpy(hsave, svm->vmcb, sizeof(struct vmcb));
1619         /* We need to remember the original CR3 in the SPT case */
1620         if (!npt_enabled)
1621                 hsave->save.cr3 = svm->vcpu.arch.cr3;
1622         hsave->save.cr4 = svm->vcpu.arch.cr4;
1623         hsave->save.rip = svm->next_rip;
1624
1625         if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1626                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1627         else
1628                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1629
1630         /* Load the nested guest state */
1631         svm->vmcb->save.es = nested_vmcb->save.es;
1632         svm->vmcb->save.cs = nested_vmcb->save.cs;
1633         svm->vmcb->save.ss = nested_vmcb->save.ss;
1634         svm->vmcb->save.ds = nested_vmcb->save.ds;
1635         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1636         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1637         svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1638         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1639         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1640         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1641         if (npt_enabled) {
1642                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1643                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1644         } else {
1645                 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1646                 kvm_mmu_reset_context(&svm->vcpu);
1647         }
1648         svm->vmcb->save.cr2 = nested_vmcb->save.cr2;
1649         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1650         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1651         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1652         /* In case we don't even reach vcpu_run, the fields are not updated */
1653         svm->vmcb->save.rax = nested_vmcb->save.rax;
1654         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1655         svm->vmcb->save.rip = nested_vmcb->save.rip;
1656         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1657         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1658         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1659
1660         /* We don't want a nested guest to be more powerful than the guest,
1661            so all intercepts are ORed */
1662         svm->vmcb->control.intercept_cr_read |=
1663                 nested_vmcb->control.intercept_cr_read;
1664         svm->vmcb->control.intercept_cr_write |=
1665                 nested_vmcb->control.intercept_cr_write;
1666         svm->vmcb->control.intercept_dr_read |=
1667                 nested_vmcb->control.intercept_dr_read;
1668         svm->vmcb->control.intercept_dr_write |=
1669                 nested_vmcb->control.intercept_dr_write;
1670         svm->vmcb->control.intercept_exceptions |=
1671                 nested_vmcb->control.intercept_exceptions;
1672
1673         svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1674
1675         svm->nested_vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1676
1677         force_new_asid(&svm->vcpu);
1678         svm->vmcb->control.exit_int_info = nested_vmcb->control.exit_int_info;
1679         svm->vmcb->control.exit_int_info_err = nested_vmcb->control.exit_int_info_err;
1680         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1681         if (nested_vmcb->control.int_ctl & V_IRQ_MASK) {
1682                 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1683                                 nested_vmcb->control.int_ctl);
1684         }
1685         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1686                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1687         else
1688                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1689
1690         nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1691                         nested_vmcb->control.exit_int_info,
1692                         nested_vmcb->control.int_state);
1693
1694         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1695         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1696         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1697         if (nested_vmcb->control.event_inj & SVM_EVTINJ_VALID)
1698                 nsvm_printk("Injecting Event: 0x%x\n",
1699                                 nested_vmcb->control.event_inj);
1700         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1701         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1702
1703         svm->vcpu.arch.hflags |= HF_GIF_MASK;
1704
1705         return 0;
1706 }
1707
1708 static int nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1709 {
1710         to_vmcb->save.fs = from_vmcb->save.fs;
1711         to_vmcb->save.gs = from_vmcb->save.gs;
1712         to_vmcb->save.tr = from_vmcb->save.tr;
1713         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1714         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1715         to_vmcb->save.star = from_vmcb->save.star;
1716         to_vmcb->save.lstar = from_vmcb->save.lstar;
1717         to_vmcb->save.cstar = from_vmcb->save.cstar;
1718         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1719         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1720         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1721         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1722
1723         return 1;
1724 }
1725
1726 static int nested_svm_vmload(struct vcpu_svm *svm, void *nested_vmcb,
1727                              void *arg2, void *opaque)
1728 {
1729         return nested_svm_vmloadsave((struct vmcb *)nested_vmcb, svm->vmcb);
1730 }
1731
1732 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
1733                              void *arg2, void *opaque)
1734 {
1735         return nested_svm_vmloadsave(svm->vmcb, (struct vmcb *)nested_vmcb);
1736 }
1737
1738 static int vmload_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1739 {
1740         if (nested_svm_check_permissions(svm))
1741                 return 1;
1742
1743         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1744         skip_emulated_instruction(&svm->vcpu);
1745
1746         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmload);
1747
1748         return 1;
1749 }
1750
1751 static int vmsave_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1752 {
1753         if (nested_svm_check_permissions(svm))
1754                 return 1;
1755
1756         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1757         skip_emulated_instruction(&svm->vcpu);
1758
1759         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmsave);
1760
1761         return 1;
1762 }
1763
1764 static int vmrun_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1765 {
1766         nsvm_printk("VMrun\n");
1767         if (nested_svm_check_permissions(svm))
1768                 return 1;
1769
1770         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1771         skip_emulated_instruction(&svm->vcpu);
1772
1773         if (nested_svm_do(svm, svm->vmcb->save.rax, 0,
1774                           NULL, nested_svm_vmrun))
1775                 return 1;
1776
1777         if (nested_svm_do(svm, svm->nested_vmcb_msrpm, 0,
1778                       NULL, nested_svm_vmrun_msrpm))
1779                 return 1;
1780
1781         return 1;
1782 }
1783
1784 static int stgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1785 {
1786         if (nested_svm_check_permissions(svm))
1787                 return 1;
1788
1789         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1790         skip_emulated_instruction(&svm->vcpu);
1791
1792         svm->vcpu.arch.hflags |= HF_GIF_MASK;
1793
1794         return 1;
1795 }
1796
1797 static int clgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1798 {
1799         if (nested_svm_check_permissions(svm))
1800                 return 1;
1801
1802         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1803         skip_emulated_instruction(&svm->vcpu);
1804
1805         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1806
1807         /* After a CLGI no interrupts should come */
1808         svm_clear_vintr(svm);
1809         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1810
1811         return 1;
1812 }
1813
1814 static int invalid_op_interception(struct vcpu_svm *svm,
1815                                    struct kvm_run *kvm_run)
1816 {
1817         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1818         return 1;
1819 }
1820
1821 static int task_switch_interception(struct vcpu_svm *svm,
1822                                     struct kvm_run *kvm_run)
1823 {
1824         u16 tss_selector;
1825
1826         tss_selector = (u16)svm->vmcb->control.exit_info_1;
1827         if (svm->vmcb->control.exit_info_2 &
1828             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1829                 return kvm_task_switch(&svm->vcpu, tss_selector,
1830                                        TASK_SWITCH_IRET);
1831         if (svm->vmcb->control.exit_info_2 &
1832             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1833                 return kvm_task_switch(&svm->vcpu, tss_selector,
1834                                        TASK_SWITCH_JMP);
1835         return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1836 }
1837
1838 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1839 {
1840         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1841         kvm_emulate_cpuid(&svm->vcpu);
1842         return 1;
1843 }
1844
1845 static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1846 {
1847         if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
1848                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1849         return 1;
1850 }
1851
1852 static int emulate_on_interception(struct vcpu_svm *svm,
1853                                    struct kvm_run *kvm_run)
1854 {
1855         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1856                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1857         return 1;
1858 }
1859
1860 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1861 {
1862         emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1863         if (irqchip_in_kernel(svm->vcpu.kvm))
1864                 return 1;
1865         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1866         return 0;
1867 }
1868
1869 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1870 {
1871         struct vcpu_svm *svm = to_svm(vcpu);
1872
1873         switch (ecx) {
1874         case MSR_IA32_TIME_STAMP_COUNTER: {
1875                 u64 tsc;
1876
1877                 rdtscll(tsc);
1878                 *data = svm->vmcb->control.tsc_offset + tsc;
1879                 break;
1880         }
1881         case MSR_K6_STAR:
1882                 *data = svm->vmcb->save.star;
1883                 break;
1884 #ifdef CONFIG_X86_64
1885         case MSR_LSTAR:
1886                 *data = svm->vmcb->save.lstar;
1887                 break;
1888         case MSR_CSTAR:
1889                 *data = svm->vmcb->save.cstar;
1890                 break;
1891         case MSR_KERNEL_GS_BASE:
1892                 *data = svm->vmcb->save.kernel_gs_base;
1893                 break;
1894         case MSR_SYSCALL_MASK:
1895                 *data = svm->vmcb->save.sfmask;
1896                 break;
1897 #endif
1898         case MSR_IA32_SYSENTER_CS:
1899                 *data = svm->vmcb->save.sysenter_cs;
1900                 break;
1901         case MSR_IA32_SYSENTER_EIP:
1902                 *data = svm->vmcb->save.sysenter_eip;
1903                 break;
1904         case MSR_IA32_SYSENTER_ESP:
1905                 *data = svm->vmcb->save.sysenter_esp;
1906                 break;
1907         /* Nobody will change the following 5 values in the VMCB so
1908            we can safely return them on rdmsr. They will always be 0
1909            until LBRV is implemented. */
1910         case MSR_IA32_DEBUGCTLMSR:
1911                 *data = svm->vmcb->save.dbgctl;
1912                 break;
1913         case MSR_IA32_LASTBRANCHFROMIP:
1914                 *data = svm->vmcb->save.br_from;
1915                 break;
1916         case MSR_IA32_LASTBRANCHTOIP:
1917                 *data = svm->vmcb->save.br_to;
1918                 break;
1919         case MSR_IA32_LASTINTFROMIP:
1920                 *data = svm->vmcb->save.last_excp_from;
1921                 break;
1922         case MSR_IA32_LASTINTTOIP:
1923                 *data = svm->vmcb->save.last_excp_to;
1924                 break;
1925         case MSR_VM_HSAVE_PA:
1926                 *data = svm->hsave_msr;
1927                 break;
1928         case MSR_VM_CR:
1929                 *data = 0;
1930                 break;
1931         case MSR_IA32_UCODE_REV:
1932                 *data = 0x01000065;
1933                 break;
1934         default:
1935                 return kvm_get_msr_common(vcpu, ecx, data);
1936         }
1937         return 0;
1938 }
1939
1940 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1941 {
1942         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1943         u64 data;
1944
1945         if (svm_get_msr(&svm->vcpu, ecx, &data))
1946                 kvm_inject_gp(&svm->vcpu, 0);
1947         else {
1948                 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1949                             (u32)(data >> 32), handler);
1950
1951                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
1952                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1953                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1954                 skip_emulated_instruction(&svm->vcpu);
1955         }
1956         return 1;
1957 }
1958
1959 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1960 {
1961         struct vcpu_svm *svm = to_svm(vcpu);
1962
1963         switch (ecx) {
1964         case MSR_IA32_TIME_STAMP_COUNTER: {
1965                 u64 tsc;
1966
1967                 rdtscll(tsc);
1968                 svm->vmcb->control.tsc_offset = data - tsc;
1969                 break;
1970         }
1971         case MSR_K6_STAR:
1972                 svm->vmcb->save.star = data;
1973                 break;
1974 #ifdef CONFIG_X86_64
1975         case MSR_LSTAR:
1976                 svm->vmcb->save.lstar = data;
1977                 break;
1978         case MSR_CSTAR:
1979                 svm->vmcb->save.cstar = data;
1980                 break;
1981         case MSR_KERNEL_GS_BASE:
1982                 svm->vmcb->save.kernel_gs_base = data;
1983                 break;
1984         case MSR_SYSCALL_MASK:
1985                 svm->vmcb->save.sfmask = data;
1986                 break;
1987 #endif
1988         case MSR_IA32_SYSENTER_CS:
1989                 svm->vmcb->save.sysenter_cs = data;
1990                 break;
1991         case MSR_IA32_SYSENTER_EIP:
1992                 svm->vmcb->save.sysenter_eip = data;
1993                 break;
1994         case MSR_IA32_SYSENTER_ESP:
1995                 svm->vmcb->save.sysenter_esp = data;
1996                 break;
1997         case MSR_IA32_DEBUGCTLMSR:
1998                 if (!svm_has(SVM_FEATURE_LBRV)) {
1999                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2000                                         __func__, data);
2001                         break;
2002                 }
2003                 if (data & DEBUGCTL_RESERVED_BITS)
2004                         return 1;
2005
2006                 svm->vmcb->save.dbgctl = data;
2007                 if (data & (1ULL<<0))
2008                         svm_enable_lbrv(svm);
2009                 else
2010                         svm_disable_lbrv(svm);
2011                 break;
2012         case MSR_K7_EVNTSEL0:
2013         case MSR_K7_EVNTSEL1:
2014         case MSR_K7_EVNTSEL2:
2015         case MSR_K7_EVNTSEL3:
2016         case MSR_K7_PERFCTR0:
2017         case MSR_K7_PERFCTR1:
2018         case MSR_K7_PERFCTR2:
2019         case MSR_K7_PERFCTR3:
2020                 /*
2021                  * Just discard all writes to the performance counters; this
2022                  * should keep both older linux and windows 64-bit guests
2023                  * happy
2024                  */
2025                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx, data);
2026
2027                 break;
2028         case MSR_VM_HSAVE_PA:
2029                 svm->hsave_msr = data;
2030                 break;
2031         default:
2032                 return kvm_set_msr_common(vcpu, ecx, data);
2033         }
2034         return 0;
2035 }
2036
2037 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2038 {
2039         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2040         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2041                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2042
2043         KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
2044                     handler);
2045
2046         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2047         if (svm_set_msr(&svm->vcpu, ecx, data))
2048                 kvm_inject_gp(&svm->vcpu, 0);
2049         else
2050                 skip_emulated_instruction(&svm->vcpu);
2051         return 1;
2052 }
2053
2054 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2055 {
2056         if (svm->vmcb->control.exit_info_1)
2057                 return wrmsr_interception(svm, kvm_run);
2058         else
2059                 return rdmsr_interception(svm, kvm_run);
2060 }
2061
2062 static int interrupt_window_interception(struct vcpu_svm *svm,
2063                                    struct kvm_run *kvm_run)
2064 {
2065         KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
2066
2067         svm_clear_vintr(svm);
2068         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2069         /*
2070          * If the user space waits to inject interrupts, exit as soon as
2071          * possible
2072          */
2073         if (kvm_run->request_interrupt_window &&
2074             !svm->vcpu.arch.irq_summary) {
2075                 ++svm->vcpu.stat.irq_window_exits;
2076                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2077                 return 0;
2078         }
2079
2080         return 1;
2081 }
2082
2083 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
2084                                       struct kvm_run *kvm_run) = {
2085         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
2086         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
2087         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
2088         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
2089         /* for now: */
2090         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
2091         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
2092         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
2093         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
2094         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
2095         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
2096         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
2097         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
2098         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
2099         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
2100         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
2101         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
2102         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
2103         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
2104         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
2105         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
2106         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
2107         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
2108         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
2109         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
2110         [SVM_EXIT_INTR]                         = intr_interception,
2111         [SVM_EXIT_NMI]                          = nmi_interception,
2112         [SVM_EXIT_SMI]                          = nop_on_interception,
2113         [SVM_EXIT_INIT]                         = nop_on_interception,
2114         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
2115         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
2116         [SVM_EXIT_CPUID]                        = cpuid_interception,
2117         [SVM_EXIT_INVD]                         = emulate_on_interception,
2118         [SVM_EXIT_HLT]                          = halt_interception,
2119         [SVM_EXIT_INVLPG]                       = invlpg_interception,
2120         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
2121         [SVM_EXIT_IOIO]                         = io_interception,
2122         [SVM_EXIT_MSR]                          = msr_interception,
2123         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
2124         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
2125         [SVM_EXIT_VMRUN]                        = vmrun_interception,
2126         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
2127         [SVM_EXIT_VMLOAD]                       = vmload_interception,
2128         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
2129         [SVM_EXIT_STGI]                         = stgi_interception,
2130         [SVM_EXIT_CLGI]                         = clgi_interception,
2131         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
2132         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
2133         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
2134         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
2135         [SVM_EXIT_NPF]                          = pf_interception,
2136 };
2137
2138 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2139 {
2140         struct vcpu_svm *svm = to_svm(vcpu);
2141         u32 exit_code = svm->vmcb->control.exit_code;
2142
2143         KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
2144                     (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
2145
2146         if (is_nested(svm)) {
2147                 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2148                             exit_code, svm->vmcb->control.exit_info_1,
2149                             svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
2150                 if (nested_svm_exit_handled(svm, true)) {
2151                         nested_svm_vmexit(svm);
2152                         nsvm_printk("-> #VMEXIT\n");
2153                         return 1;
2154                 }
2155         }
2156
2157         if (npt_enabled) {
2158                 int mmu_reload = 0;
2159                 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
2160                         svm_set_cr0(vcpu, svm->vmcb->save.cr0);
2161                         mmu_reload = 1;
2162                 }
2163                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2164                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2165                 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2166                         if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
2167                                 kvm_inject_gp(vcpu, 0);
2168                                 return 1;
2169                         }
2170                 }
2171                 if (mmu_reload) {
2172                         kvm_mmu_reset_context(vcpu);
2173                         kvm_mmu_load(vcpu);
2174                 }
2175         }
2176
2177         kvm_reput_irq(svm);
2178
2179         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2180                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2181                 kvm_run->fail_entry.hardware_entry_failure_reason
2182                         = svm->vmcb->control.exit_code;
2183                 return 0;
2184         }
2185
2186         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2187             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2188             exit_code != SVM_EXIT_NPF)
2189                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2190                        "exit_code 0x%x\n",
2191                        __func__, svm->vmcb->control.exit_int_info,
2192                        exit_code);
2193
2194         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2195             || !svm_exit_handlers[exit_code]) {
2196                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2197                 kvm_run->hw.hardware_exit_reason = exit_code;
2198                 return 0;
2199         }
2200
2201         return svm_exit_handlers[exit_code](svm, kvm_run);
2202 }
2203
2204 static void reload_tss(struct kvm_vcpu *vcpu)
2205 {
2206         int cpu = raw_smp_processor_id();
2207
2208         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2209         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
2210         load_TR_desc();
2211 }
2212
2213 static void pre_svm_run(struct vcpu_svm *svm)
2214 {
2215         int cpu = raw_smp_processor_id();
2216
2217         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2218
2219         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2220         if (svm->vcpu.cpu != cpu ||
2221             svm->asid_generation != svm_data->asid_generation)
2222                 new_asid(svm, svm_data);
2223 }
2224
2225
2226 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2227 {
2228         struct vmcb_control_area *control;
2229
2230         KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
2231
2232         ++svm->vcpu.stat.irq_injections;
2233         control = &svm->vmcb->control;
2234         control->int_vector = irq;
2235         control->int_ctl &= ~V_INTR_PRIO_MASK;
2236         control->int_ctl |= V_IRQ_MASK |
2237                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2238 }
2239
2240 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
2241 {
2242         struct vcpu_svm *svm = to_svm(vcpu);
2243
2244         nested_svm_intr(svm);
2245
2246         svm_inject_irq(svm, irq);
2247 }
2248
2249 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
2250 {
2251         struct vcpu_svm *svm = to_svm(vcpu);
2252         struct vmcb *vmcb = svm->vmcb;
2253         int max_irr, tpr;
2254
2255         if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
2256                 return;
2257
2258         vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2259
2260         max_irr = kvm_lapic_find_highest_irr(vcpu);
2261         if (max_irr == -1)
2262                 return;
2263
2264         tpr = kvm_lapic_get_cr8(vcpu) << 4;
2265
2266         if (tpr >= (max_irr & 0xf0))
2267                 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2268 }
2269
2270 static void svm_intr_assist(struct kvm_vcpu *vcpu)
2271 {
2272         struct vcpu_svm *svm = to_svm(vcpu);
2273         struct vmcb *vmcb = svm->vmcb;
2274         int intr_vector = -1;
2275
2276         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
2277             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
2278                 intr_vector = vmcb->control.exit_int_info &
2279                               SVM_EVTINJ_VEC_MASK;
2280                 vmcb->control.exit_int_info = 0;
2281                 svm_inject_irq(svm, intr_vector);
2282                 goto out;
2283         }
2284
2285         if (vmcb->control.int_ctl & V_IRQ_MASK)
2286                 goto out;
2287
2288         if (!kvm_cpu_has_interrupt(vcpu))
2289                 goto out;
2290
2291         if (nested_svm_intr(svm))
2292                 goto out;
2293
2294         if (!(svm->vcpu.arch.hflags & HF_GIF_MASK))
2295                 goto out;
2296
2297         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
2298             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
2299             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
2300                 /* unable to deliver irq, set pending irq */
2301                 svm_set_vintr(svm);
2302                 svm_inject_irq(svm, 0x0);
2303                 goto out;
2304         }
2305         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
2306         intr_vector = kvm_cpu_get_interrupt(vcpu);
2307         svm_inject_irq(svm, intr_vector);
2308 out:
2309         update_cr8_intercept(vcpu);
2310 }
2311
2312 static void kvm_reput_irq(struct vcpu_svm *svm)
2313 {
2314         struct vmcb_control_area *control = &svm->vmcb->control;
2315
2316         if ((control->int_ctl & V_IRQ_MASK)
2317             && !irqchip_in_kernel(svm->vcpu.kvm)) {
2318                 control->int_ctl &= ~V_IRQ_MASK;
2319                 push_irq(&svm->vcpu, control->int_vector);
2320         }
2321
2322         svm->vcpu.arch.interrupt_window_open =
2323                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2324                  (svm->vcpu.arch.hflags & HF_GIF_MASK);
2325 }
2326
2327 static void svm_do_inject_vector(struct vcpu_svm *svm)
2328 {
2329         struct kvm_vcpu *vcpu = &svm->vcpu;
2330         int word_index = __ffs(vcpu->arch.irq_summary);
2331         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2332         int irq = word_index * BITS_PER_LONG + bit_index;
2333
2334         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2335         if (!vcpu->arch.irq_pending[word_index])
2336                 clear_bit(word_index, &vcpu->arch.irq_summary);
2337         svm_inject_irq(svm, irq);
2338 }
2339
2340 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2341                                        struct kvm_run *kvm_run)
2342 {
2343         struct vcpu_svm *svm = to_svm(vcpu);
2344         struct vmcb_control_area *control = &svm->vmcb->control;
2345
2346         if (nested_svm_intr(svm))
2347                 return;
2348
2349         svm->vcpu.arch.interrupt_window_open =
2350                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2351                  (svm->vmcb->save.rflags & X86_EFLAGS_IF) &&
2352                  (svm->vcpu.arch.hflags & HF_GIF_MASK));
2353
2354         if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
2355                 /*
2356                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
2357                  */
2358                 svm_do_inject_vector(svm);
2359
2360         /*
2361          * Interrupts blocked.  Wait for unblock.
2362          */
2363         if (!svm->vcpu.arch.interrupt_window_open &&
2364             (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
2365                 svm_set_vintr(svm);
2366         else
2367                 svm_clear_vintr(svm);
2368 }
2369
2370 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2371 {
2372         return 0;
2373 }
2374
2375 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2376 {
2377         force_new_asid(vcpu);
2378 }
2379
2380 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2381 {
2382 }
2383
2384 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2385 {
2386         struct vcpu_svm *svm = to_svm(vcpu);
2387
2388         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2389                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2390                 kvm_lapic_set_tpr(vcpu, cr8);
2391         }
2392 }
2393
2394 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2395 {
2396         struct vcpu_svm *svm = to_svm(vcpu);
2397         u64 cr8;
2398
2399         if (!irqchip_in_kernel(vcpu->kvm))
2400                 return;
2401
2402         cr8 = kvm_get_cr8(vcpu);
2403         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2404         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2405 }
2406
2407 #ifdef CONFIG_X86_64
2408 #define R "r"
2409 #else
2410 #define R "e"
2411 #endif
2412
2413 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2414 {
2415         struct vcpu_svm *svm = to_svm(vcpu);
2416         u16 fs_selector;
2417         u16 gs_selector;
2418         u16 ldt_selector;
2419
2420         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2421         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2422         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2423
2424         pre_svm_run(svm);
2425
2426         sync_lapic_to_cr8(vcpu);
2427
2428         save_host_msrs(vcpu);
2429         fs_selector = kvm_read_fs();
2430         gs_selector = kvm_read_gs();
2431         ldt_selector = kvm_read_ldt();
2432         svm->host_cr2 = kvm_read_cr2();
2433         if (!is_nested(svm))
2434                 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2435         /* required for live migration with NPT */
2436         if (npt_enabled)
2437                 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2438
2439         clgi();
2440
2441         local_irq_enable();
2442
2443         asm volatile (
2444                 "push %%"R"bp; \n\t"
2445                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2446                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2447                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2448                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2449                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2450                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2451 #ifdef CONFIG_X86_64
2452                 "mov %c[r8](%[svm]),  %%r8  \n\t"
2453                 "mov %c[r9](%[svm]),  %%r9  \n\t"
2454                 "mov %c[r10](%[svm]), %%r10 \n\t"
2455                 "mov %c[r11](%[svm]), %%r11 \n\t"
2456                 "mov %c[r12](%[svm]), %%r12 \n\t"
2457                 "mov %c[r13](%[svm]), %%r13 \n\t"
2458                 "mov %c[r14](%[svm]), %%r14 \n\t"
2459                 "mov %c[r15](%[svm]), %%r15 \n\t"
2460 #endif
2461
2462                 /* Enter guest mode */
2463                 "push %%"R"ax \n\t"
2464                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2465                 __ex(SVM_VMLOAD) "\n\t"
2466                 __ex(SVM_VMRUN) "\n\t"
2467                 __ex(SVM_VMSAVE) "\n\t"
2468                 "pop %%"R"ax \n\t"
2469
2470                 /* Save guest registers, load host registers */
2471                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2472                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2473                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2474                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2475                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2476                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2477 #ifdef CONFIG_X86_64
2478                 "mov %%r8,  %c[r8](%[svm]) \n\t"
2479                 "mov %%r9,  %c[r9](%[svm]) \n\t"
2480                 "mov %%r10, %c[r10](%[svm]) \n\t"
2481                 "mov %%r11, %c[r11](%[svm]) \n\t"
2482                 "mov %%r12, %c[r12](%[svm]) \n\t"
2483                 "mov %%r13, %c[r13](%[svm]) \n\t"
2484                 "mov %%r14, %c[r14](%[svm]) \n\t"
2485                 "mov %%r15, %c[r15](%[svm]) \n\t"
2486 #endif
2487                 "pop %%"R"bp"
2488                 :
2489                 : [svm]"a"(svm),
2490                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2491                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2492                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2493                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2494                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2495                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2496                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2497 #ifdef CONFIG_X86_64
2498                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2499                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2500                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2501                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2502                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2503                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2504                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2505                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2506 #endif
2507                 : "cc", "memory"
2508                 , R"bx", R"cx", R"dx", R"si", R"di"
2509 #ifdef CONFIG_X86_64
2510                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2511 #endif
2512                 );
2513
2514         vcpu->arch.cr2 = svm->vmcb->save.cr2;
2515         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2516         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2517         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2518
2519         kvm_write_cr2(svm->host_cr2);
2520
2521         kvm_load_fs(fs_selector);
2522         kvm_load_gs(gs_selector);
2523         kvm_load_ldt(ldt_selector);
2524         load_host_msrs(vcpu);
2525
2526         reload_tss(vcpu);
2527
2528         local_irq_disable();
2529
2530         stgi();
2531
2532         sync_cr8_to_lapic(vcpu);
2533
2534         svm->next_rip = 0;
2535 }
2536
2537 #undef R
2538
2539 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2540 {
2541         struct vcpu_svm *svm = to_svm(vcpu);
2542
2543         if (npt_enabled) {
2544                 svm->vmcb->control.nested_cr3 = root;
2545                 force_new_asid(vcpu);
2546                 return;
2547         }
2548
2549         svm->vmcb->save.cr3 = root;
2550         force_new_asid(vcpu);
2551
2552         if (vcpu->fpu_active) {
2553                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
2554                 svm->vmcb->save.cr0 |= X86_CR0_TS;
2555                 vcpu->fpu_active = 0;
2556         }
2557 }
2558
2559 static int is_disabled(void)
2560 {
2561         u64 vm_cr;
2562
2563         rdmsrl(MSR_VM_CR, vm_cr);
2564         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2565                 return 1;
2566
2567         return 0;
2568 }
2569
2570 static void
2571 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2572 {
2573         /*
2574          * Patch in the VMMCALL instruction:
2575          */
2576         hypercall[0] = 0x0f;
2577         hypercall[1] = 0x01;
2578         hypercall[2] = 0xd9;
2579 }
2580
2581 static void svm_check_processor_compat(void *rtn)
2582 {
2583         *(int *)rtn = 0;
2584 }
2585
2586 static bool svm_cpu_has_accelerated_tpr(void)
2587 {
2588         return false;
2589 }
2590
2591 static int get_npt_level(void)
2592 {
2593 #ifdef CONFIG_X86_64
2594         return PT64_ROOT_LEVEL;
2595 #else
2596         return PT32E_ROOT_LEVEL;
2597 #endif
2598 }
2599
2600 static int svm_get_mt_mask_shift(void)
2601 {
2602         return 0;
2603 }
2604
2605 static struct kvm_x86_ops svm_x86_ops = {
2606         .cpu_has_kvm_support = has_svm,
2607         .disabled_by_bios = is_disabled,
2608         .hardware_setup = svm_hardware_setup,
2609         .hardware_unsetup = svm_hardware_unsetup,
2610         .check_processor_compatibility = svm_check_processor_compat,
2611         .hardware_enable = svm_hardware_enable,
2612         .hardware_disable = svm_hardware_disable,
2613         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2614
2615         .vcpu_create = svm_create_vcpu,
2616         .vcpu_free = svm_free_vcpu,
2617         .vcpu_reset = svm_vcpu_reset,
2618
2619         .prepare_guest_switch = svm_prepare_guest_switch,
2620         .vcpu_load = svm_vcpu_load,
2621         .vcpu_put = svm_vcpu_put,
2622
2623         .set_guest_debug = svm_guest_debug,
2624         .get_msr = svm_get_msr,
2625         .set_msr = svm_set_msr,
2626         .get_segment_base = svm_get_segment_base,
2627         .get_segment = svm_get_segment,
2628         .set_segment = svm_set_segment,
2629         .get_cpl = svm_get_cpl,
2630         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2631         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2632         .set_cr0 = svm_set_cr0,
2633         .set_cr3 = svm_set_cr3,
2634         .set_cr4 = svm_set_cr4,
2635         .set_efer = svm_set_efer,
2636         .get_idt = svm_get_idt,
2637         .set_idt = svm_set_idt,
2638         .get_gdt = svm_get_gdt,
2639         .set_gdt = svm_set_gdt,
2640         .get_dr = svm_get_dr,
2641         .set_dr = svm_set_dr,
2642         .get_rflags = svm_get_rflags,
2643         .set_rflags = svm_set_rflags,
2644
2645         .tlb_flush = svm_flush_tlb,
2646
2647         .run = svm_vcpu_run,
2648         .handle_exit = handle_exit,
2649         .skip_emulated_instruction = skip_emulated_instruction,
2650         .patch_hypercall = svm_patch_hypercall,
2651         .get_irq = svm_get_irq,
2652         .set_irq = svm_set_irq,
2653         .queue_exception = svm_queue_exception,
2654         .exception_injected = svm_exception_injected,
2655         .inject_pending_irq = svm_intr_assist,
2656         .inject_pending_vectors = do_interrupt_requests,
2657
2658         .set_tss_addr = svm_set_tss_addr,
2659         .get_tdp_level = get_npt_level,
2660         .get_mt_mask_shift = svm_get_mt_mask_shift,
2661 };
2662
2663 static int __init svm_init(void)
2664 {
2665         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
2666                               THIS_MODULE);
2667 }
2668
2669 static void __exit svm_exit(void)
2670 {
2671         kvm_exit();
2672 }
2673
2674 module_init(svm_init)
2675 module_exit(svm_exit)