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.
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/highmem.h>
20 #include <linux/profile.h>
24 #include "x86_emulate.h"
26 MODULE_AUTHOR("Qumranet");
27 MODULE_LICENSE("GPL");
29 #define IOPM_ALLOC_ORDER 2
30 #define MSRPM_ALLOC_ORDER 1
36 #define DR7_GD_MASK (1 << 13)
37 #define DR6_BD_MASK (1 << 13)
38 #define CR4_DE_MASK (1UL << 3)
40 #define SEG_TYPE_LDT 2
41 #define SEG_TYPE_BUSY_TSS16 3
43 #define KVM_EFER_LMA (1 << 10)
44 #define KVM_EFER_LME (1 << 8)
46 unsigned long iopm_base;
47 unsigned long msrpm_base;
49 struct kvm_ldttss_desc {
52 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
53 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
56 } __attribute__((packed));
61 uint64_t asid_generation;
64 struct kvm_ldttss_desc *tss_desc;
66 struct page *save_area;
69 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
71 struct svm_init_data {
76 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
78 #define NUM_MSR_MAPS (sizeof(msrpm_ranges) / sizeof(*msrpm_ranges))
79 #define MSRS_RANGE_SIZE 2048
80 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
82 #define MAX_INST_SIZE 15
84 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
86 struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
89 if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
92 cs_attrib = sa->cs.attrib;
94 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
95 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
98 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
100 int word_index = __ffs(vcpu->irq_summary);
101 int bit_index = __ffs(vcpu->irq_pending[word_index]);
102 int irq = word_index * BITS_PER_LONG + bit_index;
104 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
105 if (!vcpu->irq_pending[word_index])
106 clear_bit(word_index, &vcpu->irq_summary);
110 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
112 set_bit(irq, vcpu->irq_pending);
113 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
116 static inline void clgi(void)
118 asm volatile (SVM_CLGI);
121 static inline void stgi(void)
123 asm volatile (SVM_STGI);
126 static inline void invlpga(unsigned long addr, u32 asid)
128 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
131 static inline unsigned long kvm_read_cr2(void)
135 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
139 static inline void kvm_write_cr2(unsigned long val)
141 asm volatile ("mov %0, %%cr2" :: "r" (val));
144 static inline unsigned long read_dr6(void)
148 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
152 static inline void write_dr6(unsigned long val)
154 asm volatile ("mov %0, %%dr6" :: "r" (val));
157 static inline unsigned long read_dr7(void)
161 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
165 static inline void write_dr7(unsigned long val)
167 asm volatile ("mov %0, %%dr7" :: "r" (val));
170 static inline void force_new_asid(struct kvm_vcpu *vcpu)
172 vcpu->svm->asid_generation--;
175 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
177 force_new_asid(vcpu);
180 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
182 if (!(efer & KVM_EFER_LMA))
183 efer &= ~KVM_EFER_LME;
185 vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
186 vcpu->shadow_efer = efer;
189 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
191 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
192 SVM_EVTINJ_VALID_ERR |
193 SVM_EVTINJ_TYPE_EXEPT |
195 vcpu->svm->vmcb->control.event_inj_err = error_code;
198 static void inject_ud(struct kvm_vcpu *vcpu)
200 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
201 SVM_EVTINJ_TYPE_EXEPT |
205 static void inject_db(struct kvm_vcpu *vcpu)
207 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
208 SVM_EVTINJ_TYPE_EXEPT |
212 static int is_page_fault(uint32_t info)
214 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
215 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
218 static int is_external_interrupt(u32 info)
220 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
221 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
224 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
226 if (!vcpu->svm->next_rip) {
227 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
230 if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
231 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
233 vcpu->svm->vmcb->save.rip,
234 vcpu->svm->next_rip);
237 vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
238 vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
240 vcpu->interrupt_window_open = 1;
243 static int has_svm(void)
245 uint32_t eax, ebx, ecx, edx;
247 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
248 printk(KERN_INFO "has_svm: not amd\n");
252 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
253 if (eax < SVM_CPUID_FUNC) {
254 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
258 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
259 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
260 printk(KERN_DEBUG "has_svm: svm not available\n");
266 static void svm_hardware_disable(void *garbage)
268 struct svm_cpu_data *svm_data
269 = per_cpu(svm_data, raw_smp_processor_id());
274 wrmsrl(MSR_VM_HSAVE_PA, 0);
275 rdmsrl(MSR_EFER, efer);
276 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
277 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
278 __free_page(svm_data->save_area);
283 static void svm_hardware_enable(void *garbage)
286 struct svm_cpu_data *svm_data;
289 struct desc_ptr gdt_descr;
291 struct Xgt_desc_struct gdt_descr;
293 struct desc_struct *gdt;
294 int me = raw_smp_processor_id();
297 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
300 svm_data = per_cpu(svm_data, me);
303 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
308 svm_data->asid_generation = 1;
309 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
310 svm_data->next_asid = svm_data->max_asid + 1;
312 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
313 gdt = (struct desc_struct *)gdt_descr.address;
314 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
316 rdmsrl(MSR_EFER, efer);
317 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
319 wrmsrl(MSR_VM_HSAVE_PA,
320 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
323 static int svm_cpu_init(int cpu)
325 struct svm_cpu_data *svm_data;
328 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
332 svm_data->save_area = alloc_page(GFP_KERNEL);
334 if (!svm_data->save_area)
337 per_cpu(svm_data, cpu) = svm_data;
347 static int set_msr_interception(u32 *msrpm, unsigned msr,
352 for (i = 0; i < NUM_MSR_MAPS; i++) {
353 if (msr >= msrpm_ranges[i] &&
354 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
355 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
356 msrpm_ranges[i]) * 2;
358 u32 *base = msrpm + (msr_offset / 32);
359 u32 msr_shift = msr_offset % 32;
360 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
361 *base = (*base & ~(0x3 << msr_shift)) |
366 printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
370 static __init int svm_hardware_setup(void)
373 struct page *iopm_pages;
374 struct page *msrpm_pages;
378 kvm_emulator_want_group7_invlpg();
380 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
384 memset(page_address(iopm_pages), 0xff,
385 PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
386 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
389 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
395 msrpm_va = page_address(msrpm_pages);
396 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
397 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
400 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
401 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
402 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
403 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
404 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
405 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
407 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
408 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
409 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
410 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
412 for_each_online_cpu(cpu) {
413 r = svm_cpu_init(cpu);
420 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
423 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
428 static __exit void svm_hardware_unsetup(void)
430 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
431 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
432 iopm_base = msrpm_base = 0;
435 static void init_seg(struct vmcb_seg *seg)
438 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
439 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
444 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
447 seg->attrib = SVM_SELECTOR_P_MASK | type;
452 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
457 static void init_vmcb(struct vmcb *vmcb)
459 struct vmcb_control_area *control = &vmcb->control;
460 struct vmcb_save_area *save = &vmcb->save;
463 control->intercept_cr_read = INTERCEPT_CR0_MASK |
467 control->intercept_cr_write = INTERCEPT_CR0_MASK |
471 control->intercept_dr_read = INTERCEPT_DR0_MASK |
476 control->intercept_dr_write = INTERCEPT_DR0_MASK |
483 control->intercept_exceptions = 1 << PF_VECTOR;
486 control->intercept = (1ULL << INTERCEPT_INTR) |
487 (1ULL << INTERCEPT_NMI) |
489 * selective cr0 intercept bug?
490 * 0: 0f 22 d8 mov %eax,%cr3
491 * 3: 0f 20 c0 mov %cr0,%eax
492 * 6: 0d 00 00 00 80 or $0x80000000,%eax
493 * b: 0f 22 c0 mov %eax,%cr0
494 * set cr3 ->interception
495 * get cr0 ->interception
496 * set cr0 -> no interception
498 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
499 (1ULL << INTERCEPT_CPUID) |
500 (1ULL << INTERCEPT_HLT) |
501 (1ULL << INTERCEPT_INVLPGA) |
502 (1ULL << INTERCEPT_IOIO_PROT) |
503 (1ULL << INTERCEPT_MSR_PROT) |
504 (1ULL << INTERCEPT_TASK_SWITCH) |
505 (1ULL << INTERCEPT_SHUTDOWN) |
506 (1ULL << INTERCEPT_VMRUN) |
507 (1ULL << INTERCEPT_VMMCALL) |
508 (1ULL << INTERCEPT_VMLOAD) |
509 (1ULL << INTERCEPT_VMSAVE) |
510 (1ULL << INTERCEPT_STGI) |
511 (1ULL << INTERCEPT_CLGI) |
512 (1ULL << INTERCEPT_SKINIT);
514 control->iopm_base_pa = iopm_base;
515 control->msrpm_base_pa = msrpm_base;
517 control->tsc_offset = -tsc;
518 control->int_ctl = V_INTR_MASKING_MASK;
526 save->cs.selector = 0xf000;
527 /* Executable/Readable Code Segment */
528 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
529 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
530 save->cs.limit = 0xffff;
532 * cs.base should really be 0xffff0000, but vmx can't handle that, so
533 * be consistent with it.
535 * Replace when we have real mode working for vmx.
537 save->cs.base = 0xf0000;
539 save->gdtr.limit = 0xffff;
540 save->idtr.limit = 0xffff;
542 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
543 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
545 save->efer = MSR_EFER_SVME_MASK;
547 save->dr6 = 0xffff0ff0;
550 save->rip = 0x0000fff0;
553 * cr0 val on cpu init should be 0x60000010, we enable cpu
554 * cache by default. the orderly way is to enable cache in bios.
556 save->cr0 = 0x00000010 | CR0_PG_MASK;
557 save->cr4 = CR4_PAE_MASK;
561 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
567 vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
570 page = alloc_page(GFP_KERNEL);
574 vcpu->svm->vmcb = page_address(page);
575 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
576 vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
577 vcpu->svm->cr0 = 0x00000010;
578 vcpu->svm->asid_generation = 0;
579 memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
580 init_vmcb(vcpu->svm->vmcb);
592 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
597 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
601 static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
607 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
612 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
616 static void svm_cache_regs(struct kvm_vcpu *vcpu)
618 vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
619 vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
620 vcpu->rip = vcpu->svm->vmcb->save.rip;
623 static void svm_decache_regs(struct kvm_vcpu *vcpu)
625 vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
626 vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
627 vcpu->svm->vmcb->save.rip = vcpu->rip;
630 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
632 return vcpu->svm->vmcb->save.rflags;
635 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
637 vcpu->svm->vmcb->save.rflags = rflags;
640 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
642 struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
645 case VCPU_SREG_CS: return &save->cs;
646 case VCPU_SREG_DS: return &save->ds;
647 case VCPU_SREG_ES: return &save->es;
648 case VCPU_SREG_FS: return &save->fs;
649 case VCPU_SREG_GS: return &save->gs;
650 case VCPU_SREG_SS: return &save->ss;
651 case VCPU_SREG_TR: return &save->tr;
652 case VCPU_SREG_LDTR: return &save->ldtr;
658 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
660 struct vmcb_seg *s = svm_seg(vcpu, seg);
665 static void svm_get_segment(struct kvm_vcpu *vcpu,
666 struct kvm_segment *var, int seg)
668 struct vmcb_seg *s = svm_seg(vcpu, seg);
671 var->limit = s->limit;
672 var->selector = s->selector;
673 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
674 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
675 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
676 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
677 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
678 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
679 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
680 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
681 var->unusable = !var->present;
684 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
686 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
688 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
689 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
692 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
694 dt->limit = vcpu->svm->vmcb->save.idtr.limit;
695 dt->base = vcpu->svm->vmcb->save.idtr.base;
698 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
700 vcpu->svm->vmcb->save.idtr.limit = dt->limit;
701 vcpu->svm->vmcb->save.idtr.base = dt->base ;
704 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
706 dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
707 dt->base = vcpu->svm->vmcb->save.gdtr.base;
710 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
712 vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
713 vcpu->svm->vmcb->save.gdtr.base = dt->base ;
716 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
720 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
723 if (vcpu->shadow_efer & KVM_EFER_LME) {
724 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
725 vcpu->shadow_efer |= KVM_EFER_LMA;
726 vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
729 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
730 vcpu->shadow_efer &= ~KVM_EFER_LMA;
731 vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
735 vcpu->svm->cr0 = cr0;
736 vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK | CR0_WP_MASK;
740 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
743 vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
746 static void svm_set_segment(struct kvm_vcpu *vcpu,
747 struct kvm_segment *var, int seg)
749 struct vmcb_seg *s = svm_seg(vcpu, seg);
752 s->limit = var->limit;
753 s->selector = var->selector;
757 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
758 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
759 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
760 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
761 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
762 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
763 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
764 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
766 if (seg == VCPU_SREG_CS)
767 vcpu->svm->vmcb->save.cpl
768 = (vcpu->svm->vmcb->save.cs.attrib
769 >> SVM_SELECTOR_DPL_SHIFT) & 3;
775 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
776 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
780 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
785 static void load_host_msrs(struct kvm_vcpu *vcpu)
789 for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
790 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
793 static void save_host_msrs(struct kvm_vcpu *vcpu)
797 for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
798 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
801 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
803 if (svm_data->next_asid > svm_data->max_asid) {
804 ++svm_data->asid_generation;
805 svm_data->next_asid = 1;
806 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
809 vcpu->cpu = svm_data->cpu;
810 vcpu->svm->asid_generation = svm_data->asid_generation;
811 vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
814 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
816 invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
819 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
821 return vcpu->svm->db_regs[dr];
824 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
829 if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
830 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
831 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
832 *exception = DB_VECTOR;
838 vcpu->svm->db_regs[dr] = value;
841 if (vcpu->cr4 & CR4_DE_MASK) {
842 *exception = UD_VECTOR;
846 if (value & ~((1ULL << 32) - 1)) {
847 *exception = GP_VECTOR;
850 vcpu->svm->vmcb->save.dr7 = value;
854 printk(KERN_DEBUG "%s: unexpected dr %u\n",
856 *exception = UD_VECTOR;
861 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
863 u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
866 enum emulation_result er;
869 if (is_external_interrupt(exit_int_info))
870 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
872 spin_lock(&vcpu->kvm->lock);
874 fault_address = vcpu->svm->vmcb->control.exit_info_2;
875 error_code = vcpu->svm->vmcb->control.exit_info_1;
876 r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
878 spin_unlock(&vcpu->kvm->lock);
882 spin_unlock(&vcpu->kvm->lock);
885 er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
886 spin_unlock(&vcpu->kvm->lock);
891 case EMULATE_DO_MMIO:
892 ++kvm_stat.mmio_exits;
893 kvm_run->exit_reason = KVM_EXIT_MMIO;
896 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
902 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
906 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
909 * VMCB is undefined after a SHUTDOWN intercept
910 * so reinitialize it.
912 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
913 init_vmcb(vcpu->svm->vmcb);
915 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
919 static int io_get_override(struct kvm_vcpu *vcpu,
920 struct vmcb_seg **seg,
923 u8 inst[MAX_INST_SIZE];
928 rip = vcpu->svm->vmcb->save.rip;
929 ins_length = vcpu->svm->next_rip - rip;
930 rip += vcpu->svm->vmcb->save.cs.base;
932 if (ins_length > MAX_INST_SIZE)
934 "%s: inst length err, cs base 0x%llx rip 0x%llx "
935 "next rip 0x%llx ins_length %u\n",
937 vcpu->svm->vmcb->save.cs.base,
938 vcpu->svm->vmcb->save.rip,
939 vcpu->svm->vmcb->control.exit_info_2,
942 if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
948 for (i = 0; i < ins_length; i++)
959 *seg = &vcpu->svm->vmcb->save.cs;
962 *seg = &vcpu->svm->vmcb->save.ss;
965 *seg = &vcpu->svm->vmcb->save.ds;
968 *seg = &vcpu->svm->vmcb->save.es;
971 *seg = &vcpu->svm->vmcb->save.fs;
974 *seg = &vcpu->svm->vmcb->save.gs;
979 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
983 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
985 unsigned long addr_mask;
987 struct vmcb_seg *seg;
989 struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
990 u16 cs_attrib = save_area->cs.attrib;
991 unsigned addr_size = get_addr_size(vcpu);
993 if (!io_get_override(vcpu, &seg, &addr_override))
997 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
1000 reg = &vcpu->regs[VCPU_REGS_RDI];
1001 seg = &vcpu->svm->vmcb->save.es;
1003 reg = &vcpu->regs[VCPU_REGS_RSI];
1004 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
1007 addr_mask = ~0ULL >> (64 - (addr_size * 8));
1009 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1010 !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1011 *address = (*reg & addr_mask);
1015 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1016 svm_inject_gp(vcpu, 0);
1020 *address = (*reg & addr_mask) + seg->base;
1024 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1026 u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1027 int _in = io_info & SVM_IOIO_TYPE_MASK;
1029 ++kvm_stat.io_exits;
1031 vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1033 kvm_run->exit_reason = KVM_EXIT_IO;
1034 kvm_run->io.port = io_info >> 16;
1035 kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1036 kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1037 kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1038 kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1040 if (kvm_run->io.string) {
1043 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1045 printk(KERN_DEBUG "%s: get io address failed\n", __FUNCTION__);
1049 if (kvm_run->io.rep) {
1050 kvm_run->io.count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1051 kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1052 & X86_EFLAGS_DF) != 0;
1055 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1061 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1066 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1068 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1069 skip_emulated_instruction(vcpu);
1070 if (vcpu->irq_summary)
1073 kvm_run->exit_reason = KVM_EXIT_HLT;
1074 ++kvm_stat.halt_exits;
1078 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1084 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1086 printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1087 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1091 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1093 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1094 kvm_run->exit_reason = KVM_EXIT_CPUID;
1098 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1100 if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1101 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1105 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1108 case MSR_IA32_TIME_STAMP_COUNTER: {
1112 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1116 *data = vcpu->svm->vmcb->save.star;
1118 #ifdef CONFIG_X86_64
1120 *data = vcpu->svm->vmcb->save.lstar;
1123 *data = vcpu->svm->vmcb->save.cstar;
1125 case MSR_KERNEL_GS_BASE:
1126 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1128 case MSR_SYSCALL_MASK:
1129 *data = vcpu->svm->vmcb->save.sfmask;
1132 case MSR_IA32_SYSENTER_CS:
1133 *data = vcpu->svm->vmcb->save.sysenter_cs;
1135 case MSR_IA32_SYSENTER_EIP:
1136 *data = vcpu->svm->vmcb->save.sysenter_eip;
1138 case MSR_IA32_SYSENTER_ESP:
1139 *data = vcpu->svm->vmcb->save.sysenter_esp;
1142 return kvm_get_msr_common(vcpu, ecx, data);
1147 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1149 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1152 if (svm_get_msr(vcpu, ecx, &data))
1153 svm_inject_gp(vcpu, 0);
1155 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1156 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1157 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1158 skip_emulated_instruction(vcpu);
1163 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1166 case MSR_IA32_TIME_STAMP_COUNTER: {
1170 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1174 vcpu->svm->vmcb->save.star = data;
1176 #ifdef CONFIG_X86_64
1178 vcpu->svm->vmcb->save.lstar = data;
1181 vcpu->svm->vmcb->save.cstar = data;
1183 case MSR_KERNEL_GS_BASE:
1184 vcpu->svm->vmcb->save.kernel_gs_base = data;
1186 case MSR_SYSCALL_MASK:
1187 vcpu->svm->vmcb->save.sfmask = data;
1190 case MSR_IA32_SYSENTER_CS:
1191 vcpu->svm->vmcb->save.sysenter_cs = data;
1193 case MSR_IA32_SYSENTER_EIP:
1194 vcpu->svm->vmcb->save.sysenter_eip = data;
1196 case MSR_IA32_SYSENTER_ESP:
1197 vcpu->svm->vmcb->save.sysenter_esp = data;
1200 return kvm_set_msr_common(vcpu, ecx, data);
1205 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1207 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1208 u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1209 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1210 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1211 if (svm_set_msr(vcpu, ecx, data))
1212 svm_inject_gp(vcpu, 0);
1214 skip_emulated_instruction(vcpu);
1218 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1220 if (vcpu->svm->vmcb->control.exit_info_1)
1221 return wrmsr_interception(vcpu, kvm_run);
1223 return rdmsr_interception(vcpu, kvm_run);
1226 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1227 struct kvm_run *kvm_run)
1230 * If the user space waits to inject interrupts, exit as soon as
1233 if (kvm_run->request_interrupt_window &&
1234 !vcpu->irq_summary) {
1235 ++kvm_stat.irq_window_exits;
1236 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1243 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1244 struct kvm_run *kvm_run) = {
1245 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1246 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1247 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1249 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1250 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1251 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1252 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1253 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1254 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1255 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1256 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1257 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1258 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1259 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1260 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1261 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1262 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1263 [SVM_EXIT_INTR] = nop_on_interception,
1264 [SVM_EXIT_NMI] = nop_on_interception,
1265 [SVM_EXIT_SMI] = nop_on_interception,
1266 [SVM_EXIT_INIT] = nop_on_interception,
1267 [SVM_EXIT_VINTR] = interrupt_window_interception,
1268 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1269 [SVM_EXIT_CPUID] = cpuid_interception,
1270 [SVM_EXIT_HLT] = halt_interception,
1271 [SVM_EXIT_INVLPG] = emulate_on_interception,
1272 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1273 [SVM_EXIT_IOIO] = io_interception,
1274 [SVM_EXIT_MSR] = msr_interception,
1275 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1276 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1277 [SVM_EXIT_VMRUN] = invalid_op_interception,
1278 [SVM_EXIT_VMMCALL] = invalid_op_interception,
1279 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1280 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1281 [SVM_EXIT_STGI] = invalid_op_interception,
1282 [SVM_EXIT_CLGI] = invalid_op_interception,
1283 [SVM_EXIT_SKINIT] = invalid_op_interception,
1287 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1289 u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1291 kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1293 if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1294 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1295 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1297 __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1300 if (exit_code >= sizeof(svm_exit_handlers) / sizeof(*svm_exit_handlers)
1301 || svm_exit_handlers[exit_code] == 0) {
1302 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1303 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1306 vcpu->svm->vmcb->save.rip,
1308 vcpu->svm->vmcb->save.rflags);
1312 return svm_exit_handlers[exit_code](vcpu, kvm_run);
1315 static void reload_tss(struct kvm_vcpu *vcpu)
1317 int cpu = raw_smp_processor_id();
1319 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1320 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1324 static void pre_svm_run(struct kvm_vcpu *vcpu)
1326 int cpu = raw_smp_processor_id();
1328 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1330 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1331 if (vcpu->cpu != cpu ||
1332 vcpu->svm->asid_generation != svm_data->asid_generation)
1333 new_asid(vcpu, svm_data);
1337 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1339 struct vmcb_control_area *control;
1341 control = &vcpu->svm->vmcb->control;
1342 control->int_vector = pop_irq(vcpu);
1343 control->int_ctl &= ~V_INTR_PRIO_MASK;
1344 control->int_ctl |= V_IRQ_MASK |
1345 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1348 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1350 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1352 if (control->int_ctl & V_IRQ_MASK) {
1353 control->int_ctl &= ~V_IRQ_MASK;
1354 push_irq(vcpu, control->int_vector);
1357 vcpu->interrupt_window_open =
1358 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1361 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1362 struct kvm_run *kvm_run)
1364 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1366 vcpu->interrupt_window_open =
1367 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1368 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1370 if (vcpu->interrupt_window_open && vcpu->irq_summary)
1372 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1374 kvm_do_inject_irq(vcpu);
1377 * Interrupts blocked. Wait for unblock.
1379 if (!vcpu->interrupt_window_open &&
1380 (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1381 control->intercept |= 1ULL << INTERCEPT_VINTR;
1383 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1386 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1387 struct kvm_run *kvm_run)
1389 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1390 vcpu->irq_summary == 0);
1391 kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1392 kvm_run->cr8 = vcpu->cr8;
1393 kvm_run->apic_base = vcpu->apic_base;
1397 * Check if userspace requested an interrupt window, and that the
1398 * interrupt window is open.
1400 * No need to exit to userspace if we already have an interrupt queued.
1402 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1403 struct kvm_run *kvm_run)
1405 return (!vcpu->irq_summary &&
1406 kvm_run->request_interrupt_window &&
1407 vcpu->interrupt_window_open &&
1408 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1411 static void save_db_regs(unsigned long *db_regs)
1413 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1414 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1415 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1416 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1419 static void load_db_regs(unsigned long *db_regs)
1421 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1422 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1423 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1424 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1427 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1435 if (!vcpu->mmio_read_completed)
1436 do_interrupt_requests(vcpu, kvm_run);
1442 save_host_msrs(vcpu);
1443 fs_selector = read_fs();
1444 gs_selector = read_gs();
1445 ldt_selector = read_ldt();
1446 vcpu->svm->host_cr2 = kvm_read_cr2();
1447 vcpu->svm->host_dr6 = read_dr6();
1448 vcpu->svm->host_dr7 = read_dr7();
1449 vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1451 if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1453 save_db_regs(vcpu->svm->host_db_regs);
1454 load_db_regs(vcpu->svm->db_regs);
1457 fx_save(vcpu->host_fx_image);
1458 fx_restore(vcpu->guest_fx_image);
1461 #ifdef CONFIG_X86_64
1462 "push %%rbx; push %%rcx; push %%rdx;"
1463 "push %%rsi; push %%rdi; push %%rbp;"
1464 "push %%r8; push %%r9; push %%r10; push %%r11;"
1465 "push %%r12; push %%r13; push %%r14; push %%r15;"
1467 "push %%ebx; push %%ecx; push %%edx;"
1468 "push %%esi; push %%edi; push %%ebp;"
1471 #ifdef CONFIG_X86_64
1472 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1473 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1474 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1475 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1476 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1477 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1478 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1479 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1480 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1481 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1482 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1483 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1484 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1485 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1487 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1488 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1489 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1490 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1491 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1492 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1495 #ifdef CONFIG_X86_64
1496 /* Enter guest mode */
1498 "mov %c[svm](%[vcpu]), %%rax \n\t"
1499 "mov %c[vmcb](%%rax), %%rax \n\t"
1505 /* Enter guest mode */
1507 "mov %c[svm](%[vcpu]), %%eax \n\t"
1508 "mov %c[vmcb](%%eax), %%eax \n\t"
1515 /* Save guest registers, load host registers */
1516 #ifdef CONFIG_X86_64
1517 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1518 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1519 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1520 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1521 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1522 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1523 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1524 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1525 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1526 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1527 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1528 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1529 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1530 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1532 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1533 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1534 "pop %%rbp; pop %%rdi; pop %%rsi;"
1535 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1537 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1538 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1539 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1540 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1541 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1542 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1544 "pop %%ebp; pop %%edi; pop %%esi;"
1545 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1549 [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1550 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1551 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1552 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1553 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1554 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1555 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1556 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1557 #ifdef CONFIG_X86_64
1558 ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1559 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1560 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1561 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1562 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1563 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1564 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1565 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1569 fx_save(vcpu->guest_fx_image);
1570 fx_restore(vcpu->host_fx_image);
1572 if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1573 load_db_regs(vcpu->svm->host_db_regs);
1575 vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1577 write_dr6(vcpu->svm->host_dr6);
1578 write_dr7(vcpu->svm->host_dr7);
1579 kvm_write_cr2(vcpu->svm->host_cr2);
1581 load_fs(fs_selector);
1582 load_gs(gs_selector);
1583 load_ldt(ldt_selector);
1584 load_host_msrs(vcpu);
1589 * Profile KVM exit RIPs:
1591 if (unlikely(prof_on == KVM_PROFILING))
1592 profile_hit(KVM_PROFILING,
1593 (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1597 kvm_reput_irq(vcpu);
1599 vcpu->svm->next_rip = 0;
1601 if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1602 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1603 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
1604 post_kvm_run_save(vcpu, kvm_run);
1608 r = handle_exit(vcpu, kvm_run);
1610 if (signal_pending(current)) {
1611 ++kvm_stat.signal_exits;
1612 post_kvm_run_save(vcpu, kvm_run);
1616 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1617 ++kvm_stat.request_irq_exits;
1618 post_kvm_run_save(vcpu, kvm_run);
1624 post_kvm_run_save(vcpu, kvm_run);
1628 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1630 force_new_asid(vcpu);
1633 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1635 vcpu->svm->vmcb->save.cr3 = root;
1636 force_new_asid(vcpu);
1639 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1643 uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1645 ++kvm_stat.pf_guest;
1647 if (is_page_fault(exit_int_info)) {
1649 vcpu->svm->vmcb->control.event_inj_err = 0;
1650 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1651 SVM_EVTINJ_VALID_ERR |
1652 SVM_EVTINJ_TYPE_EXEPT |
1657 vcpu->svm->vmcb->save.cr2 = addr;
1658 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1659 SVM_EVTINJ_VALID_ERR |
1660 SVM_EVTINJ_TYPE_EXEPT |
1662 vcpu->svm->vmcb->control.event_inj_err = err_code;
1666 static int is_disabled(void)
1671 static struct kvm_arch_ops svm_arch_ops = {
1672 .cpu_has_kvm_support = has_svm,
1673 .disabled_by_bios = is_disabled,
1674 .hardware_setup = svm_hardware_setup,
1675 .hardware_unsetup = svm_hardware_unsetup,
1676 .hardware_enable = svm_hardware_enable,
1677 .hardware_disable = svm_hardware_disable,
1679 .vcpu_create = svm_create_vcpu,
1680 .vcpu_free = svm_free_vcpu,
1682 .vcpu_load = svm_vcpu_load,
1683 .vcpu_put = svm_vcpu_put,
1684 .vcpu_decache = svm_vcpu_decache,
1686 .set_guest_debug = svm_guest_debug,
1687 .get_msr = svm_get_msr,
1688 .set_msr = svm_set_msr,
1689 .get_segment_base = svm_get_segment_base,
1690 .get_segment = svm_get_segment,
1691 .set_segment = svm_set_segment,
1692 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1693 .decache_cr0_cr4_guest_bits = svm_decache_cr0_cr4_guest_bits,
1694 .set_cr0 = svm_set_cr0,
1695 .set_cr0_no_modeswitch = svm_set_cr0,
1696 .set_cr3 = svm_set_cr3,
1697 .set_cr4 = svm_set_cr4,
1698 .set_efer = svm_set_efer,
1699 .get_idt = svm_get_idt,
1700 .set_idt = svm_set_idt,
1701 .get_gdt = svm_get_gdt,
1702 .set_gdt = svm_set_gdt,
1703 .get_dr = svm_get_dr,
1704 .set_dr = svm_set_dr,
1705 .cache_regs = svm_cache_regs,
1706 .decache_regs = svm_decache_regs,
1707 .get_rflags = svm_get_rflags,
1708 .set_rflags = svm_set_rflags,
1710 .invlpg = svm_invlpg,
1711 .tlb_flush = svm_flush_tlb,
1712 .inject_page_fault = svm_inject_page_fault,
1714 .inject_gp = svm_inject_gp,
1716 .run = svm_vcpu_run,
1717 .skip_emulated_instruction = skip_emulated_instruction,
1718 .vcpu_setup = svm_vcpu_setup,
1721 static int __init svm_init(void)
1723 return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1726 static void __exit svm_exit(void)
1731 module_init(svm_init)
1732 module_exit(svm_exit)