2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
18 #include "x86_emulate.h"
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/vmalloc.h>
23 #include <linux/highmem.h>
24 #include <linux/profile.h>
25 #include <linux/sched.h>
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
52 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
54 return container_of(vcpu, struct vcpu_svm, vcpu);
57 unsigned long iopm_base;
58 unsigned long msrpm_base;
60 struct kvm_ldttss_desc {
63 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
64 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
67 } __attribute__((packed));
75 struct kvm_ldttss_desc *tss_desc;
77 struct page *save_area;
80 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
81 static uint32_t svm_features;
83 struct svm_init_data {
88 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
90 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
91 #define MSRS_RANGE_SIZE 2048
92 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
94 #define MAX_INST_SIZE 15
96 static inline u32 svm_has(u32 feat)
98 return svm_features & feat;
101 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
103 struct vmcb_save_area *sa = &to_svm(vcpu)->vmcb->save;
106 if (!(sa->cr0 & X86_CR0_PE) || (sa->rflags & X86_EFLAGS_VM))
109 cs_attrib = sa->cs.attrib;
111 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
112 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
115 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
117 int word_index = __ffs(vcpu->irq_summary);
118 int bit_index = __ffs(vcpu->irq_pending[word_index]);
119 int irq = word_index * BITS_PER_LONG + bit_index;
121 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
122 if (!vcpu->irq_pending[word_index])
123 clear_bit(word_index, &vcpu->irq_summary);
127 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
129 set_bit(irq, vcpu->irq_pending);
130 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
133 static inline void clgi(void)
135 asm volatile (SVM_CLGI);
138 static inline void stgi(void)
140 asm volatile (SVM_STGI);
143 static inline void invlpga(unsigned long addr, u32 asid)
145 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
148 static inline unsigned long kvm_read_cr2(void)
152 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
156 static inline void kvm_write_cr2(unsigned long val)
158 asm volatile ("mov %0, %%cr2" :: "r" (val));
161 static inline unsigned long read_dr6(void)
165 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
169 static inline void write_dr6(unsigned long val)
171 asm volatile ("mov %0, %%dr6" :: "r" (val));
174 static inline unsigned long read_dr7(void)
178 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
182 static inline void write_dr7(unsigned long val)
184 asm volatile ("mov %0, %%dr7" :: "r" (val));
187 static inline void force_new_asid(struct kvm_vcpu *vcpu)
189 to_svm(vcpu)->asid_generation--;
192 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
194 force_new_asid(vcpu);
197 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
199 if (!(efer & KVM_EFER_LMA))
200 efer &= ~KVM_EFER_LME;
202 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
203 vcpu->shadow_efer = efer;
206 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
208 struct vcpu_svm *svm = to_svm(vcpu);
210 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
211 SVM_EVTINJ_VALID_ERR |
212 SVM_EVTINJ_TYPE_EXEPT |
214 svm->vmcb->control.event_inj_err = error_code;
217 static void inject_ud(struct kvm_vcpu *vcpu)
219 to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
220 SVM_EVTINJ_TYPE_EXEPT |
224 static int is_page_fault(uint32_t info)
226 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
227 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
230 static int is_external_interrupt(u32 info)
232 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
233 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
236 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
238 struct vcpu_svm *svm = to_svm(vcpu);
240 if (!svm->next_rip) {
241 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
244 if (svm->next_rip - svm->vmcb->save.rip > 15) {
245 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
251 vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
252 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
254 vcpu->interrupt_window_open = 1;
257 static int has_svm(void)
259 uint32_t eax, ebx, ecx, edx;
261 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
262 printk(KERN_INFO "has_svm: not amd\n");
266 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
267 if (eax < SVM_CPUID_FUNC) {
268 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
272 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
273 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
274 printk(KERN_DEBUG "has_svm: svm not available\n");
280 static void svm_hardware_disable(void *garbage)
282 struct svm_cpu_data *svm_data
283 = per_cpu(svm_data, raw_smp_processor_id());
288 wrmsrl(MSR_VM_HSAVE_PA, 0);
289 rdmsrl(MSR_EFER, efer);
290 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
291 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
292 __free_page(svm_data->save_area);
297 static void svm_hardware_enable(void *garbage)
300 struct svm_cpu_data *svm_data;
303 struct desc_ptr gdt_descr;
305 struct Xgt_desc_struct gdt_descr;
307 struct desc_struct *gdt;
308 int me = raw_smp_processor_id();
311 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
314 svm_data = per_cpu(svm_data, me);
317 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
322 svm_data->asid_generation = 1;
323 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
324 svm_data->next_asid = svm_data->max_asid + 1;
325 svm_features = cpuid_edx(SVM_CPUID_FUNC);
327 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
328 gdt = (struct desc_struct *)gdt_descr.address;
329 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
331 rdmsrl(MSR_EFER, efer);
332 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
334 wrmsrl(MSR_VM_HSAVE_PA,
335 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
338 static int svm_cpu_init(int cpu)
340 struct svm_cpu_data *svm_data;
343 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
347 svm_data->save_area = alloc_page(GFP_KERNEL);
349 if (!svm_data->save_area)
352 per_cpu(svm_data, cpu) = svm_data;
362 static int set_msr_interception(u32 *msrpm, unsigned msr,
367 for (i = 0; i < NUM_MSR_MAPS; i++) {
368 if (msr >= msrpm_ranges[i] &&
369 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
370 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
371 msrpm_ranges[i]) * 2;
373 u32 *base = msrpm + (msr_offset / 32);
374 u32 msr_shift = msr_offset % 32;
375 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
376 *base = (*base & ~(0x3 << msr_shift)) |
381 printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
385 static __init int svm_hardware_setup(void)
388 struct page *iopm_pages;
389 struct page *msrpm_pages;
390 void *iopm_va, *msrpm_va;
393 kvm_emulator_want_group7_invlpg();
395 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
400 iopm_va = page_address(iopm_pages);
401 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
402 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
403 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
406 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
412 msrpm_va = page_address(msrpm_pages);
413 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
414 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
417 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
418 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
419 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
420 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
421 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
422 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
424 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
425 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
426 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
427 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
429 for_each_online_cpu(cpu) {
430 r = svm_cpu_init(cpu);
437 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
440 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
445 static __exit void svm_hardware_unsetup(void)
447 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
448 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
449 iopm_base = msrpm_base = 0;
452 static void init_seg(struct vmcb_seg *seg)
455 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
456 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
461 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
464 seg->attrib = SVM_SELECTOR_P_MASK | type;
469 static void init_vmcb(struct vmcb *vmcb)
471 struct vmcb_control_area *control = &vmcb->control;
472 struct vmcb_save_area *save = &vmcb->save;
474 control->intercept_cr_read = INTERCEPT_CR0_MASK |
478 control->intercept_cr_write = INTERCEPT_CR0_MASK |
482 control->intercept_dr_read = INTERCEPT_DR0_MASK |
487 control->intercept_dr_write = INTERCEPT_DR0_MASK |
494 control->intercept_exceptions = 1 << PF_VECTOR;
497 control->intercept = (1ULL << INTERCEPT_INTR) |
498 (1ULL << INTERCEPT_NMI) |
499 (1ULL << INTERCEPT_SMI) |
501 * selective cr0 intercept bug?
502 * 0: 0f 22 d8 mov %eax,%cr3
503 * 3: 0f 20 c0 mov %cr0,%eax
504 * 6: 0d 00 00 00 80 or $0x80000000,%eax
505 * b: 0f 22 c0 mov %eax,%cr0
506 * set cr3 ->interception
507 * get cr0 ->interception
508 * set cr0 -> no interception
510 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
511 (1ULL << INTERCEPT_CPUID) |
512 (1ULL << INTERCEPT_HLT) |
513 (1ULL << INTERCEPT_INVLPGA) |
514 (1ULL << INTERCEPT_IOIO_PROT) |
515 (1ULL << INTERCEPT_MSR_PROT) |
516 (1ULL << INTERCEPT_TASK_SWITCH) |
517 (1ULL << INTERCEPT_SHUTDOWN) |
518 (1ULL << INTERCEPT_VMRUN) |
519 (1ULL << INTERCEPT_VMMCALL) |
520 (1ULL << INTERCEPT_VMLOAD) |
521 (1ULL << INTERCEPT_VMSAVE) |
522 (1ULL << INTERCEPT_STGI) |
523 (1ULL << INTERCEPT_CLGI) |
524 (1ULL << INTERCEPT_SKINIT) |
525 (1ULL << INTERCEPT_MONITOR) |
526 (1ULL << INTERCEPT_MWAIT);
528 control->iopm_base_pa = iopm_base;
529 control->msrpm_base_pa = msrpm_base;
530 control->tsc_offset = 0;
531 control->int_ctl = V_INTR_MASKING_MASK;
539 save->cs.selector = 0xf000;
540 /* Executable/Readable Code Segment */
541 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
542 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
543 save->cs.limit = 0xffff;
545 * cs.base should really be 0xffff0000, but vmx can't handle that, so
546 * be consistent with it.
548 * Replace when we have real mode working for vmx.
550 save->cs.base = 0xf0000;
552 save->gdtr.limit = 0xffff;
553 save->idtr.limit = 0xffff;
555 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
556 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
558 save->efer = MSR_EFER_SVME_MASK;
560 save->dr6 = 0xffff0ff0;
563 save->rip = 0x0000fff0;
566 * cr0 val on cpu init should be 0x60000010, we enable cpu
567 * cache by default. the orderly way is to enable cache in bios.
569 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
570 save->cr4 = X86_CR4_PAE;
574 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
576 struct vcpu_svm *svm;
580 svm = kzalloc(sizeof *svm, GFP_KERNEL);
586 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
590 page = alloc_page(GFP_KERNEL);
596 svm->vmcb = page_address(page);
597 clear_page(svm->vmcb);
598 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
599 svm->asid_generation = 0;
600 memset(svm->db_regs, 0, sizeof(svm->db_regs));
601 init_vmcb(svm->vmcb);
604 svm->vcpu.fpu_active = 1;
605 svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
606 if (svm->vcpu.vcpu_id == 0)
607 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
612 kvm_vcpu_uninit(&svm->vcpu);
619 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
621 struct vcpu_svm *svm = to_svm(vcpu);
623 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
624 kvm_vcpu_uninit(vcpu);
628 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
630 struct vcpu_svm *svm = to_svm(vcpu);
633 if (unlikely(cpu != vcpu->cpu)) {
637 * Make sure that the guest sees a monotonically
641 delta = vcpu->host_tsc - tsc_this;
642 svm->vmcb->control.tsc_offset += delta;
646 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
647 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
650 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
652 struct vcpu_svm *svm = to_svm(vcpu);
655 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
656 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
658 rdtscll(vcpu->host_tsc);
661 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
665 static void svm_cache_regs(struct kvm_vcpu *vcpu)
667 struct vcpu_svm *svm = to_svm(vcpu);
669 vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
670 vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
671 vcpu->rip = svm->vmcb->save.rip;
674 static void svm_decache_regs(struct kvm_vcpu *vcpu)
676 struct vcpu_svm *svm = to_svm(vcpu);
677 svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
678 svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
679 svm->vmcb->save.rip = vcpu->rip;
682 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
684 return to_svm(vcpu)->vmcb->save.rflags;
687 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
689 to_svm(vcpu)->vmcb->save.rflags = rflags;
692 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
694 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
697 case VCPU_SREG_CS: return &save->cs;
698 case VCPU_SREG_DS: return &save->ds;
699 case VCPU_SREG_ES: return &save->es;
700 case VCPU_SREG_FS: return &save->fs;
701 case VCPU_SREG_GS: return &save->gs;
702 case VCPU_SREG_SS: return &save->ss;
703 case VCPU_SREG_TR: return &save->tr;
704 case VCPU_SREG_LDTR: return &save->ldtr;
710 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
712 struct vmcb_seg *s = svm_seg(vcpu, seg);
717 static void svm_get_segment(struct kvm_vcpu *vcpu,
718 struct kvm_segment *var, int seg)
720 struct vmcb_seg *s = svm_seg(vcpu, seg);
723 var->limit = s->limit;
724 var->selector = s->selector;
725 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
726 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
727 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
728 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
729 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
730 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
731 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
732 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
733 var->unusable = !var->present;
736 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
738 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
740 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
741 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
744 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
746 struct vcpu_svm *svm = to_svm(vcpu);
748 dt->limit = svm->vmcb->save.idtr.limit;
749 dt->base = svm->vmcb->save.idtr.base;
752 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
754 struct vcpu_svm *svm = to_svm(vcpu);
756 svm->vmcb->save.idtr.limit = dt->limit;
757 svm->vmcb->save.idtr.base = dt->base ;
760 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
762 struct vcpu_svm *svm = to_svm(vcpu);
764 dt->limit = svm->vmcb->save.gdtr.limit;
765 dt->base = svm->vmcb->save.gdtr.base;
768 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
770 struct vcpu_svm *svm = to_svm(vcpu);
772 svm->vmcb->save.gdtr.limit = dt->limit;
773 svm->vmcb->save.gdtr.base = dt->base ;
776 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
780 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
782 struct vcpu_svm *svm = to_svm(vcpu);
785 if (vcpu->shadow_efer & KVM_EFER_LME) {
786 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
787 vcpu->shadow_efer |= KVM_EFER_LMA;
788 svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
791 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG) ) {
792 vcpu->shadow_efer &= ~KVM_EFER_LMA;
793 svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
797 if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
798 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
799 vcpu->fpu_active = 1;
803 cr0 |= X86_CR0_PG | X86_CR0_WP;
804 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
805 svm->vmcb->save.cr0 = cr0;
808 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
811 to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
814 static void svm_set_segment(struct kvm_vcpu *vcpu,
815 struct kvm_segment *var, int seg)
817 struct vcpu_svm *svm = to_svm(vcpu);
818 struct vmcb_seg *s = svm_seg(vcpu, seg);
821 s->limit = var->limit;
822 s->selector = var->selector;
826 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
827 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
828 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
829 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
830 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
831 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
832 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
833 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
835 if (seg == VCPU_SREG_CS)
837 = (svm->vmcb->save.cs.attrib
838 >> SVM_SELECTOR_DPL_SHIFT) & 3;
844 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
845 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
849 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
854 static void load_host_msrs(struct kvm_vcpu *vcpu)
857 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
861 static void save_host_msrs(struct kvm_vcpu *vcpu)
864 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
868 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
870 struct vcpu_svm *svm = to_svm(vcpu);
872 if (svm_data->next_asid > svm_data->max_asid) {
873 ++svm_data->asid_generation;
874 svm_data->next_asid = 1;
875 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
878 vcpu->cpu = svm_data->cpu;
879 svm->asid_generation = svm_data->asid_generation;
880 svm->vmcb->control.asid = svm_data->next_asid++;
883 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
885 invlpga(address, to_svm(vcpu)->vmcb->control.asid); // is needed?
888 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
890 return to_svm(vcpu)->db_regs[dr];
893 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
896 struct vcpu_svm *svm = to_svm(vcpu);
900 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
901 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
902 svm->vmcb->save.dr6 |= DR6_BD_MASK;
903 *exception = DB_VECTOR;
909 svm->db_regs[dr] = value;
912 if (vcpu->cr4 & X86_CR4_DE) {
913 *exception = UD_VECTOR;
917 if (value & ~((1ULL << 32) - 1)) {
918 *exception = GP_VECTOR;
921 svm->vmcb->save.dr7 = value;
925 printk(KERN_DEBUG "%s: unexpected dr %u\n",
927 *exception = UD_VECTOR;
932 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
934 struct vcpu_svm *svm = to_svm(vcpu);
935 u32 exit_int_info = svm->vmcb->control.exit_int_info;
938 enum emulation_result er;
941 if (is_external_interrupt(exit_int_info))
942 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
944 spin_lock(&vcpu->kvm->lock);
946 fault_address = svm->vmcb->control.exit_info_2;
947 error_code = svm->vmcb->control.exit_info_1;
948 r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
950 spin_unlock(&vcpu->kvm->lock);
954 spin_unlock(&vcpu->kvm->lock);
957 er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
958 spin_unlock(&vcpu->kvm->lock);
963 case EMULATE_DO_MMIO:
964 ++vcpu->stat.mmio_exits;
967 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
973 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
977 static int nm_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
979 struct vcpu_svm *svm = to_svm(vcpu);
981 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
982 if (!(vcpu->cr0 & X86_CR0_TS))
983 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
984 vcpu->fpu_active = 1;
989 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
991 struct vcpu_svm *svm = to_svm(vcpu);
993 * VMCB is undefined after a SHUTDOWN intercept
994 * so reinitialize it.
996 clear_page(svm->vmcb);
997 init_vmcb(svm->vmcb);
999 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1003 static int io_get_override(struct kvm_vcpu *vcpu,
1004 struct vmcb_seg **seg,
1007 struct vcpu_svm *svm = to_svm(vcpu);
1008 u8 inst[MAX_INST_SIZE];
1009 unsigned ins_length;
1013 rip = svm->vmcb->save.rip;
1014 ins_length = svm->next_rip - rip;
1015 rip += svm->vmcb->save.cs.base;
1017 if (ins_length > MAX_INST_SIZE)
1019 "%s: inst length err, cs base 0x%llx rip 0x%llx "
1020 "next rip 0x%llx ins_length %u\n",
1022 svm->vmcb->save.cs.base,
1023 svm->vmcb->save.rip,
1024 svm->vmcb->control.exit_info_2,
1027 if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
1033 for (i = 0; i < ins_length; i++)
1044 *seg = &svm->vmcb->save.cs;
1047 *seg = &svm->vmcb->save.ss;
1050 *seg = &svm->vmcb->save.ds;
1053 *seg = &svm->vmcb->save.es;
1056 *seg = &svm->vmcb->save.fs;
1059 *seg = &svm->vmcb->save.gs;
1064 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
1068 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, gva_t *address)
1070 unsigned long addr_mask;
1072 struct vmcb_seg *seg;
1074 struct vcpu_svm *svm = to_svm(vcpu);
1075 struct vmcb_save_area *save_area = &svm->vmcb->save;
1076 u16 cs_attrib = save_area->cs.attrib;
1077 unsigned addr_size = get_addr_size(vcpu);
1079 if (!io_get_override(vcpu, &seg, &addr_override))
1083 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
1086 reg = &vcpu->regs[VCPU_REGS_RDI];
1087 seg = &svm->vmcb->save.es;
1089 reg = &vcpu->regs[VCPU_REGS_RSI];
1090 seg = (seg) ? seg : &svm->vmcb->save.ds;
1093 addr_mask = ~0ULL >> (64 - (addr_size * 8));
1095 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1096 !(svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1097 *address = (*reg & addr_mask);
1101 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1102 svm_inject_gp(vcpu, 0);
1106 *address = (*reg & addr_mask) + seg->base;
1110 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1112 struct vcpu_svm *svm = to_svm(vcpu);
1113 u32 io_info = svm->vmcb->control.exit_info_1; //address size bug?
1114 int size, down, in, string, rep;
1116 unsigned long count;
1119 ++vcpu->stat.io_exits;
1121 svm->next_rip = svm->vmcb->control.exit_info_2;
1123 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1124 port = io_info >> 16;
1125 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1126 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1127 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1129 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1134 addr_mask = io_adress(vcpu, in, &address);
1136 printk(KERN_DEBUG "%s: get io address failed\n",
1142 count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1144 return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1145 address, rep, port);
1148 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1153 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1155 struct vcpu_svm *svm = to_svm(vcpu);
1157 svm->next_rip = svm->vmcb->save.rip + 1;
1158 skip_emulated_instruction(vcpu);
1159 return kvm_emulate_halt(vcpu);
1162 static int vmmcall_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1164 struct vcpu_svm *svm = to_svm(vcpu);
1166 svm->next_rip = svm->vmcb->save.rip + 3;
1167 skip_emulated_instruction(vcpu);
1168 return kvm_hypercall(vcpu, kvm_run);
1171 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1177 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1179 printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1180 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1184 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1186 struct vcpu_svm *svm = to_svm(vcpu);
1188 svm->next_rip = svm->vmcb->save.rip + 2;
1189 kvm_emulate_cpuid(vcpu);
1193 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1195 if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1196 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1200 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1202 struct vcpu_svm *svm = to_svm(vcpu);
1205 case MSR_IA32_TIME_STAMP_COUNTER: {
1209 *data = svm->vmcb->control.tsc_offset + tsc;
1213 *data = svm->vmcb->save.star;
1215 #ifdef CONFIG_X86_64
1217 *data = svm->vmcb->save.lstar;
1220 *data = svm->vmcb->save.cstar;
1222 case MSR_KERNEL_GS_BASE:
1223 *data = svm->vmcb->save.kernel_gs_base;
1225 case MSR_SYSCALL_MASK:
1226 *data = svm->vmcb->save.sfmask;
1229 case MSR_IA32_SYSENTER_CS:
1230 *data = svm->vmcb->save.sysenter_cs;
1232 case MSR_IA32_SYSENTER_EIP:
1233 *data = svm->vmcb->save.sysenter_eip;
1235 case MSR_IA32_SYSENTER_ESP:
1236 *data = svm->vmcb->save.sysenter_esp;
1239 return kvm_get_msr_common(vcpu, ecx, data);
1244 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1246 struct vcpu_svm *svm = to_svm(vcpu);
1247 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1250 if (svm_get_msr(vcpu, ecx, &data))
1251 svm_inject_gp(vcpu, 0);
1253 svm->vmcb->save.rax = data & 0xffffffff;
1254 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1255 svm->next_rip = svm->vmcb->save.rip + 2;
1256 skip_emulated_instruction(vcpu);
1261 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1263 struct vcpu_svm *svm = to_svm(vcpu);
1266 case MSR_IA32_TIME_STAMP_COUNTER: {
1270 svm->vmcb->control.tsc_offset = data - tsc;
1274 svm->vmcb->save.star = data;
1276 #ifdef CONFIG_X86_64
1278 svm->vmcb->save.lstar = data;
1281 svm->vmcb->save.cstar = data;
1283 case MSR_KERNEL_GS_BASE:
1284 svm->vmcb->save.kernel_gs_base = data;
1286 case MSR_SYSCALL_MASK:
1287 svm->vmcb->save.sfmask = data;
1290 case MSR_IA32_SYSENTER_CS:
1291 svm->vmcb->save.sysenter_cs = data;
1293 case MSR_IA32_SYSENTER_EIP:
1294 svm->vmcb->save.sysenter_eip = data;
1296 case MSR_IA32_SYSENTER_ESP:
1297 svm->vmcb->save.sysenter_esp = data;
1300 return kvm_set_msr_common(vcpu, ecx, data);
1305 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1307 struct vcpu_svm *svm = to_svm(vcpu);
1308 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1309 u64 data = (svm->vmcb->save.rax & -1u)
1310 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1311 svm->next_rip = svm->vmcb->save.rip + 2;
1312 if (svm_set_msr(vcpu, ecx, data))
1313 svm_inject_gp(vcpu, 0);
1315 skip_emulated_instruction(vcpu);
1319 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1321 if (to_svm(vcpu)->vmcb->control.exit_info_1)
1322 return wrmsr_interception(vcpu, kvm_run);
1324 return rdmsr_interception(vcpu, kvm_run);
1327 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1328 struct kvm_run *kvm_run)
1331 * If the user space waits to inject interrupts, exit as soon as
1334 if (kvm_run->request_interrupt_window &&
1335 !vcpu->irq_summary) {
1336 ++vcpu->stat.irq_window_exits;
1337 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1344 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1345 struct kvm_run *kvm_run) = {
1346 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1347 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1348 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1350 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1351 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1352 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1353 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1354 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1355 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1356 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1357 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1358 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1359 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1360 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1361 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1362 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1363 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1364 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1365 [SVM_EXIT_INTR] = nop_on_interception,
1366 [SVM_EXIT_NMI] = nop_on_interception,
1367 [SVM_EXIT_SMI] = nop_on_interception,
1368 [SVM_EXIT_INIT] = nop_on_interception,
1369 [SVM_EXIT_VINTR] = interrupt_window_interception,
1370 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1371 [SVM_EXIT_CPUID] = cpuid_interception,
1372 [SVM_EXIT_HLT] = halt_interception,
1373 [SVM_EXIT_INVLPG] = emulate_on_interception,
1374 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1375 [SVM_EXIT_IOIO] = io_interception,
1376 [SVM_EXIT_MSR] = msr_interception,
1377 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1378 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1379 [SVM_EXIT_VMRUN] = invalid_op_interception,
1380 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1381 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1382 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1383 [SVM_EXIT_STGI] = invalid_op_interception,
1384 [SVM_EXIT_CLGI] = invalid_op_interception,
1385 [SVM_EXIT_SKINIT] = invalid_op_interception,
1386 [SVM_EXIT_MONITOR] = invalid_op_interception,
1387 [SVM_EXIT_MWAIT] = invalid_op_interception,
1391 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1393 struct vcpu_svm *svm = to_svm(vcpu);
1394 u32 exit_code = svm->vmcb->control.exit_code;
1396 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1397 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1398 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1400 __FUNCTION__, svm->vmcb->control.exit_int_info,
1403 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1404 || svm_exit_handlers[exit_code] == 0) {
1405 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1406 kvm_run->hw.hardware_exit_reason = exit_code;
1410 return svm_exit_handlers[exit_code](vcpu, kvm_run);
1413 static void reload_tss(struct kvm_vcpu *vcpu)
1415 int cpu = raw_smp_processor_id();
1417 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1418 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1422 static void pre_svm_run(struct kvm_vcpu *vcpu)
1424 struct vcpu_svm *svm = to_svm(vcpu);
1425 int cpu = raw_smp_processor_id();
1427 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1429 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1430 if (vcpu->cpu != cpu ||
1431 svm->asid_generation != svm_data->asid_generation)
1432 new_asid(vcpu, svm_data);
1436 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1438 struct vmcb_control_area *control;
1440 control = &to_svm(vcpu)->vmcb->control;
1441 control->int_vector = pop_irq(vcpu);
1442 control->int_ctl &= ~V_INTR_PRIO_MASK;
1443 control->int_ctl |= V_IRQ_MASK |
1444 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1447 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1449 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
1451 if (control->int_ctl & V_IRQ_MASK) {
1452 control->int_ctl &= ~V_IRQ_MASK;
1453 push_irq(vcpu, control->int_vector);
1456 vcpu->interrupt_window_open =
1457 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1460 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1461 struct kvm_run *kvm_run)
1463 struct vcpu_svm *svm = to_svm(vcpu);
1464 struct vmcb_control_area *control = &svm->vmcb->control;
1466 vcpu->interrupt_window_open =
1467 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1468 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1470 if (vcpu->interrupt_window_open && vcpu->irq_summary)
1472 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1474 kvm_do_inject_irq(vcpu);
1477 * Interrupts blocked. Wait for unblock.
1479 if (!vcpu->interrupt_window_open &&
1480 (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1481 control->intercept |= 1ULL << INTERCEPT_VINTR;
1483 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1486 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1487 struct kvm_run *kvm_run)
1489 struct vcpu_svm *svm = to_svm(vcpu);
1491 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1492 vcpu->irq_summary == 0);
1493 kvm_run->if_flag = (svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1494 kvm_run->cr8 = vcpu->cr8;
1495 kvm_run->apic_base = vcpu->apic_base;
1499 * Check if userspace requested an interrupt window, and that the
1500 * interrupt window is open.
1502 * No need to exit to userspace if we already have an interrupt queued.
1504 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1505 struct kvm_run *kvm_run)
1507 return (!vcpu->irq_summary &&
1508 kvm_run->request_interrupt_window &&
1509 vcpu->interrupt_window_open &&
1510 (to_svm(vcpu)->vmcb->save.rflags & X86_EFLAGS_IF));
1513 static void save_db_regs(unsigned long *db_regs)
1515 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1516 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1517 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1518 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1521 static void load_db_regs(unsigned long *db_regs)
1523 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1524 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1525 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1526 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1529 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1531 force_new_asid(vcpu);
1534 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1536 struct vcpu_svm *svm = to_svm(vcpu);
1543 r = kvm_mmu_reload(vcpu);
1547 if (!vcpu->mmio_read_completed)
1548 do_interrupt_requests(vcpu, kvm_run);
1552 vcpu->guest_mode = 1;
1554 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
1555 svm_flush_tlb(vcpu);
1559 save_host_msrs(vcpu);
1560 fs_selector = read_fs();
1561 gs_selector = read_gs();
1562 ldt_selector = read_ldt();
1563 svm->host_cr2 = kvm_read_cr2();
1564 svm->host_dr6 = read_dr6();
1565 svm->host_dr7 = read_dr7();
1566 svm->vmcb->save.cr2 = vcpu->cr2;
1568 if (svm->vmcb->save.dr7 & 0xff) {
1570 save_db_regs(svm->host_db_regs);
1571 load_db_regs(svm->db_regs);
1574 if (vcpu->fpu_active) {
1575 fx_save(vcpu->host_fx_image);
1576 fx_restore(vcpu->guest_fx_image);
1580 #ifdef CONFIG_X86_64
1581 "push %%rbx; push %%rcx; push %%rdx;"
1582 "push %%rsi; push %%rdi; push %%rbp;"
1583 "push %%r8; push %%r9; push %%r10; push %%r11;"
1584 "push %%r12; push %%r13; push %%r14; push %%r15;"
1586 "push %%ebx; push %%ecx; push %%edx;"
1587 "push %%esi; push %%edi; push %%ebp;"
1590 #ifdef CONFIG_X86_64
1591 "mov %c[rbx](%[svm]), %%rbx \n\t"
1592 "mov %c[rcx](%[svm]), %%rcx \n\t"
1593 "mov %c[rdx](%[svm]), %%rdx \n\t"
1594 "mov %c[rsi](%[svm]), %%rsi \n\t"
1595 "mov %c[rdi](%[svm]), %%rdi \n\t"
1596 "mov %c[rbp](%[svm]), %%rbp \n\t"
1597 "mov %c[r8](%[svm]), %%r8 \n\t"
1598 "mov %c[r9](%[svm]), %%r9 \n\t"
1599 "mov %c[r10](%[svm]), %%r10 \n\t"
1600 "mov %c[r11](%[svm]), %%r11 \n\t"
1601 "mov %c[r12](%[svm]), %%r12 \n\t"
1602 "mov %c[r13](%[svm]), %%r13 \n\t"
1603 "mov %c[r14](%[svm]), %%r14 \n\t"
1604 "mov %c[r15](%[svm]), %%r15 \n\t"
1606 "mov %c[rbx](%[svm]), %%ebx \n\t"
1607 "mov %c[rcx](%[svm]), %%ecx \n\t"
1608 "mov %c[rdx](%[svm]), %%edx \n\t"
1609 "mov %c[rsi](%[svm]), %%esi \n\t"
1610 "mov %c[rdi](%[svm]), %%edi \n\t"
1611 "mov %c[rbp](%[svm]), %%ebp \n\t"
1614 #ifdef CONFIG_X86_64
1615 /* Enter guest mode */
1617 "mov %c[vmcb](%[svm]), %%rax \n\t"
1623 /* Enter guest mode */
1625 "mov %c[vmcb](%[svm]), %%eax \n\t"
1632 /* Save guest registers, load host registers */
1633 #ifdef CONFIG_X86_64
1634 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1635 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1636 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1637 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1638 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1639 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1640 "mov %%r8, %c[r8](%[svm]) \n\t"
1641 "mov %%r9, %c[r9](%[svm]) \n\t"
1642 "mov %%r10, %c[r10](%[svm]) \n\t"
1643 "mov %%r11, %c[r11](%[svm]) \n\t"
1644 "mov %%r12, %c[r12](%[svm]) \n\t"
1645 "mov %%r13, %c[r13](%[svm]) \n\t"
1646 "mov %%r14, %c[r14](%[svm]) \n\t"
1647 "mov %%r15, %c[r15](%[svm]) \n\t"
1649 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1650 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1651 "pop %%rbp; pop %%rdi; pop %%rsi;"
1652 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1654 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1655 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1656 "mov %%edx, %c[rdx](%[svm]) \n\t"
1657 "mov %%esi, %c[rsi](%[svm]) \n\t"
1658 "mov %%edi, %c[rdi](%[svm]) \n\t"
1659 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1661 "pop %%ebp; pop %%edi; pop %%esi;"
1662 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1666 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1667 [rbx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBX])),
1668 [rcx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RCX])),
1669 [rdx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDX])),
1670 [rsi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RSI])),
1671 [rdi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDI])),
1672 [rbp]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBP]))
1673 #ifdef CONFIG_X86_64
1674 ,[r8 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R8])),
1675 [r9 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R9 ])),
1676 [r10]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R10])),
1677 [r11]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R11])),
1678 [r12]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R12])),
1679 [r13]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R13])),
1680 [r14]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R14])),
1681 [r15]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R15]))
1685 vcpu->guest_mode = 0;
1687 if (vcpu->fpu_active) {
1688 fx_save(vcpu->guest_fx_image);
1689 fx_restore(vcpu->host_fx_image);
1692 if ((svm->vmcb->save.dr7 & 0xff))
1693 load_db_regs(svm->host_db_regs);
1695 vcpu->cr2 = svm->vmcb->save.cr2;
1697 write_dr6(svm->host_dr6);
1698 write_dr7(svm->host_dr7);
1699 kvm_write_cr2(svm->host_cr2);
1701 load_fs(fs_selector);
1702 load_gs(gs_selector);
1703 load_ldt(ldt_selector);
1704 load_host_msrs(vcpu);
1709 * Profile KVM exit RIPs:
1711 if (unlikely(prof_on == KVM_PROFILING))
1712 profile_hit(KVM_PROFILING,
1713 (void *)(unsigned long)svm->vmcb->save.rip);
1717 kvm_reput_irq(vcpu);
1721 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1722 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1723 kvm_run->fail_entry.hardware_entry_failure_reason
1724 = svm->vmcb->control.exit_code;
1725 post_kvm_run_save(vcpu, kvm_run);
1729 r = handle_exit(vcpu, kvm_run);
1731 if (signal_pending(current)) {
1732 ++vcpu->stat.signal_exits;
1733 post_kvm_run_save(vcpu, kvm_run);
1734 kvm_run->exit_reason = KVM_EXIT_INTR;
1738 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1739 ++vcpu->stat.request_irq_exits;
1740 post_kvm_run_save(vcpu, kvm_run);
1741 kvm_run->exit_reason = KVM_EXIT_INTR;
1747 post_kvm_run_save(vcpu, kvm_run);
1751 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1753 struct vcpu_svm *svm = to_svm(vcpu);
1755 svm->vmcb->save.cr3 = root;
1756 force_new_asid(vcpu);
1758 if (vcpu->fpu_active) {
1759 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1760 svm->vmcb->save.cr0 |= X86_CR0_TS;
1761 vcpu->fpu_active = 0;
1765 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1769 struct vcpu_svm *svm = to_svm(vcpu);
1770 uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1772 ++vcpu->stat.pf_guest;
1774 if (is_page_fault(exit_int_info)) {
1776 svm->vmcb->control.event_inj_err = 0;
1777 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1778 SVM_EVTINJ_VALID_ERR |
1779 SVM_EVTINJ_TYPE_EXEPT |
1784 svm->vmcb->save.cr2 = addr;
1785 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1786 SVM_EVTINJ_VALID_ERR |
1787 SVM_EVTINJ_TYPE_EXEPT |
1789 svm->vmcb->control.event_inj_err = err_code;
1793 static int is_disabled(void)
1797 rdmsrl(MSR_VM_CR, vm_cr);
1798 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1805 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1808 * Patch in the VMMCALL instruction:
1810 hypercall[0] = 0x0f;
1811 hypercall[1] = 0x01;
1812 hypercall[2] = 0xd9;
1813 hypercall[3] = 0xc3;
1816 static struct kvm_arch_ops svm_arch_ops = {
1817 .cpu_has_kvm_support = has_svm,
1818 .disabled_by_bios = is_disabled,
1819 .hardware_setup = svm_hardware_setup,
1820 .hardware_unsetup = svm_hardware_unsetup,
1821 .hardware_enable = svm_hardware_enable,
1822 .hardware_disable = svm_hardware_disable,
1824 .vcpu_create = svm_create_vcpu,
1825 .vcpu_free = svm_free_vcpu,
1827 .vcpu_load = svm_vcpu_load,
1828 .vcpu_put = svm_vcpu_put,
1829 .vcpu_decache = svm_vcpu_decache,
1831 .set_guest_debug = svm_guest_debug,
1832 .get_msr = svm_get_msr,
1833 .set_msr = svm_set_msr,
1834 .get_segment_base = svm_get_segment_base,
1835 .get_segment = svm_get_segment,
1836 .set_segment = svm_set_segment,
1837 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1838 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1839 .set_cr0 = svm_set_cr0,
1840 .set_cr3 = svm_set_cr3,
1841 .set_cr4 = svm_set_cr4,
1842 .set_efer = svm_set_efer,
1843 .get_idt = svm_get_idt,
1844 .set_idt = svm_set_idt,
1845 .get_gdt = svm_get_gdt,
1846 .set_gdt = svm_set_gdt,
1847 .get_dr = svm_get_dr,
1848 .set_dr = svm_set_dr,
1849 .cache_regs = svm_cache_regs,
1850 .decache_regs = svm_decache_regs,
1851 .get_rflags = svm_get_rflags,
1852 .set_rflags = svm_set_rflags,
1854 .invlpg = svm_invlpg,
1855 .tlb_flush = svm_flush_tlb,
1856 .inject_page_fault = svm_inject_page_fault,
1858 .inject_gp = svm_inject_gp,
1860 .run = svm_vcpu_run,
1861 .skip_emulated_instruction = skip_emulated_instruction,
1862 .patch_hypercall = svm_patch_hypercall,
1865 static int __init svm_init(void)
1867 return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1870 static void __exit svm_exit(void)
1875 module_init(svm_init)
1876 module_exit(svm_exit)