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"
21 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/vmalloc.h>
24 #include <linux/highmem.h>
25 #include <linux/profile.h>
26 #include <linux/sched.h>
30 MODULE_AUTHOR("Qumranet");
31 MODULE_LICENSE("GPL");
33 #define IOPM_ALLOC_ORDER 2
34 #define MSRPM_ALLOC_ORDER 1
40 #define DR7_GD_MASK (1 << 13)
41 #define DR6_BD_MASK (1 << 13)
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
46 #define KVM_EFER_LMA (1 << 10)
47 #define KVM_EFER_LME (1 << 8)
49 #define SVM_FEATURE_NPT (1 << 0)
50 #define SVM_FEATURE_LBRV (1 << 1)
51 #define SVM_DEATURE_SVML (1 << 2)
53 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
55 return container_of(vcpu, struct vcpu_svm, vcpu);
58 unsigned long iopm_base;
59 unsigned long msrpm_base;
61 struct kvm_ldttss_desc {
64 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
65 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
68 } __attribute__((packed));
76 struct kvm_ldttss_desc *tss_desc;
78 struct page *save_area;
81 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
82 static uint32_t svm_features;
84 struct svm_init_data {
89 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
91 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
92 #define MSRS_RANGE_SIZE 2048
93 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
95 #define MAX_INST_SIZE 15
97 static inline u32 svm_has(u32 feat)
99 return svm_features & feat;
102 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
104 int word_index = __ffs(vcpu->irq_summary);
105 int bit_index = __ffs(vcpu->irq_pending[word_index]);
106 int irq = word_index * BITS_PER_LONG + bit_index;
108 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
109 if (!vcpu->irq_pending[word_index])
110 clear_bit(word_index, &vcpu->irq_summary);
114 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
116 set_bit(irq, vcpu->irq_pending);
117 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
120 static inline void clgi(void)
122 asm volatile (SVM_CLGI);
125 static inline void stgi(void)
127 asm volatile (SVM_STGI);
130 static inline void invlpga(unsigned long addr, u32 asid)
132 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
135 static inline unsigned long kvm_read_cr2(void)
139 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
143 static inline void kvm_write_cr2(unsigned long val)
145 asm volatile ("mov %0, %%cr2" :: "r" (val));
148 static inline unsigned long read_dr6(void)
152 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
156 static inline void write_dr6(unsigned long val)
158 asm volatile ("mov %0, %%dr6" :: "r" (val));
161 static inline unsigned long read_dr7(void)
165 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
169 static inline void write_dr7(unsigned long val)
171 asm volatile ("mov %0, %%dr7" :: "r" (val));
174 static inline void force_new_asid(struct kvm_vcpu *vcpu)
176 to_svm(vcpu)->asid_generation--;
179 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
181 force_new_asid(vcpu);
184 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
186 if (!(efer & KVM_EFER_LMA))
187 efer &= ~KVM_EFER_LME;
189 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
190 vcpu->shadow_efer = efer;
193 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
195 struct vcpu_svm *svm = to_svm(vcpu);
197 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
198 SVM_EVTINJ_VALID_ERR |
199 SVM_EVTINJ_TYPE_EXEPT |
201 svm->vmcb->control.event_inj_err = error_code;
204 static void inject_ud(struct kvm_vcpu *vcpu)
206 to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
207 SVM_EVTINJ_TYPE_EXEPT |
211 static int is_page_fault(uint32_t info)
213 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
214 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
217 static int is_external_interrupt(u32 info)
219 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
220 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
223 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
225 struct vcpu_svm *svm = to_svm(vcpu);
227 if (!svm->next_rip) {
228 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
231 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE) {
232 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
238 vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
239 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
241 vcpu->interrupt_window_open = 1;
244 static int has_svm(void)
246 uint32_t eax, ebx, ecx, edx;
248 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
249 printk(KERN_INFO "has_svm: not amd\n");
253 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
254 if (eax < SVM_CPUID_FUNC) {
255 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
259 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
260 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
261 printk(KERN_DEBUG "has_svm: svm not available\n");
267 static void svm_hardware_disable(void *garbage)
269 struct svm_cpu_data *svm_data
270 = per_cpu(svm_data, raw_smp_processor_id());
275 wrmsrl(MSR_VM_HSAVE_PA, 0);
276 rdmsrl(MSR_EFER, efer);
277 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
278 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
279 __free_page(svm_data->save_area);
284 static void svm_hardware_enable(void *garbage)
287 struct svm_cpu_data *svm_data;
290 struct desc_ptr gdt_descr;
292 struct Xgt_desc_struct gdt_descr;
294 struct desc_struct *gdt;
295 int me = raw_smp_processor_id();
298 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
301 svm_data = per_cpu(svm_data, me);
304 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
309 svm_data->asid_generation = 1;
310 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
311 svm_data->next_asid = svm_data->max_asid + 1;
312 svm_features = cpuid_edx(SVM_CPUID_FUNC);
314 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
315 gdt = (struct desc_struct *)gdt_descr.address;
316 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
318 rdmsrl(MSR_EFER, efer);
319 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
321 wrmsrl(MSR_VM_HSAVE_PA,
322 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
325 static int svm_cpu_init(int cpu)
327 struct svm_cpu_data *svm_data;
330 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
334 svm_data->save_area = alloc_page(GFP_KERNEL);
336 if (!svm_data->save_area)
339 per_cpu(svm_data, cpu) = svm_data;
349 static void set_msr_interception(u32 *msrpm, unsigned msr,
354 for (i = 0; i < NUM_MSR_MAPS; i++) {
355 if (msr >= msrpm_ranges[i] &&
356 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
357 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
358 msrpm_ranges[i]) * 2;
360 u32 *base = msrpm + (msr_offset / 32);
361 u32 msr_shift = msr_offset % 32;
362 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
363 *base = (*base & ~(0x3 << msr_shift)) |
371 static __init int svm_hardware_setup(void)
374 struct page *iopm_pages;
375 struct page *msrpm_pages;
376 void *iopm_va, *msrpm_va;
379 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
384 iopm_va = page_address(iopm_pages);
385 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
386 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
387 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
390 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
396 msrpm_va = page_address(msrpm_pages);
397 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
398 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
401 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
402 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
403 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
404 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
405 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
406 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
408 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
409 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
410 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
411 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
413 for_each_online_cpu(cpu) {
414 r = svm_cpu_init(cpu);
421 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
424 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
429 static __exit void svm_hardware_unsetup(void)
431 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
432 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
433 iopm_base = msrpm_base = 0;
436 static void init_seg(struct vmcb_seg *seg)
439 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
440 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
445 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
448 seg->attrib = SVM_SELECTOR_P_MASK | type;
453 static void init_vmcb(struct vmcb *vmcb)
455 struct vmcb_control_area *control = &vmcb->control;
456 struct vmcb_save_area *save = &vmcb->save;
458 control->intercept_cr_read = INTERCEPT_CR0_MASK |
462 control->intercept_cr_write = INTERCEPT_CR0_MASK |
466 control->intercept_dr_read = INTERCEPT_DR0_MASK |
471 control->intercept_dr_write = INTERCEPT_DR0_MASK |
478 control->intercept_exceptions = 1 << PF_VECTOR;
481 control->intercept = (1ULL << INTERCEPT_INTR) |
482 (1ULL << INTERCEPT_NMI) |
483 (1ULL << INTERCEPT_SMI) |
485 * selective cr0 intercept bug?
486 * 0: 0f 22 d8 mov %eax,%cr3
487 * 3: 0f 20 c0 mov %cr0,%eax
488 * 6: 0d 00 00 00 80 or $0x80000000,%eax
489 * b: 0f 22 c0 mov %eax,%cr0
490 * set cr3 ->interception
491 * get cr0 ->interception
492 * set cr0 -> no interception
494 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
495 (1ULL << INTERCEPT_CPUID) |
496 (1ULL << INTERCEPT_HLT) |
497 (1ULL << INTERCEPT_INVLPGA) |
498 (1ULL << INTERCEPT_IOIO_PROT) |
499 (1ULL << INTERCEPT_MSR_PROT) |
500 (1ULL << INTERCEPT_TASK_SWITCH) |
501 (1ULL << INTERCEPT_SHUTDOWN) |
502 (1ULL << INTERCEPT_VMRUN) |
503 (1ULL << INTERCEPT_VMMCALL) |
504 (1ULL << INTERCEPT_VMLOAD) |
505 (1ULL << INTERCEPT_VMSAVE) |
506 (1ULL << INTERCEPT_STGI) |
507 (1ULL << INTERCEPT_CLGI) |
508 (1ULL << INTERCEPT_SKINIT) |
509 (1ULL << INTERCEPT_MONITOR) |
510 (1ULL << INTERCEPT_MWAIT);
512 control->iopm_base_pa = iopm_base;
513 control->msrpm_base_pa = msrpm_base;
514 control->tsc_offset = 0;
515 control->int_ctl = V_INTR_MASKING_MASK;
523 save->cs.selector = 0xf000;
524 /* Executable/Readable Code Segment */
525 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
526 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
527 save->cs.limit = 0xffff;
529 * cs.base should really be 0xffff0000, but vmx can't handle that, so
530 * be consistent with it.
532 * Replace when we have real mode working for vmx.
534 save->cs.base = 0xf0000;
536 save->gdtr.limit = 0xffff;
537 save->idtr.limit = 0xffff;
539 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
540 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
542 save->efer = MSR_EFER_SVME_MASK;
544 save->dr6 = 0xffff0ff0;
547 save->rip = 0x0000fff0;
550 * cr0 val on cpu init should be 0x60000010, we enable cpu
551 * cache by default. the orderly way is to enable cache in bios.
553 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
554 save->cr4 = X86_CR4_PAE;
558 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
560 struct vcpu_svm *svm;
564 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
570 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
574 if (irqchip_in_kernel(kvm)) {
575 err = kvm_create_lapic(&svm->vcpu);
580 page = alloc_page(GFP_KERNEL);
586 svm->vmcb = page_address(page);
587 clear_page(svm->vmcb);
588 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
589 svm->asid_generation = 0;
590 memset(svm->db_regs, 0, sizeof(svm->db_regs));
591 init_vmcb(svm->vmcb);
594 svm->vcpu.fpu_active = 1;
595 svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
596 if (svm->vcpu.vcpu_id == 0)
597 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
602 kvm_vcpu_uninit(&svm->vcpu);
604 kmem_cache_free(kvm_vcpu_cache, svm);
609 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
611 struct vcpu_svm *svm = to_svm(vcpu);
613 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
614 kvm_vcpu_uninit(vcpu);
615 kmem_cache_free(kvm_vcpu_cache, svm);
618 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
620 struct vcpu_svm *svm = to_svm(vcpu);
623 if (unlikely(cpu != vcpu->cpu)) {
627 * Make sure that the guest sees a monotonically
631 delta = vcpu->host_tsc - tsc_this;
632 svm->vmcb->control.tsc_offset += delta;
634 kvm_migrate_apic_timer(vcpu);
637 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
638 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
641 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
643 struct vcpu_svm *svm = to_svm(vcpu);
646 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
647 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
649 rdtscll(vcpu->host_tsc);
652 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
656 static void svm_cache_regs(struct kvm_vcpu *vcpu)
658 struct vcpu_svm *svm = to_svm(vcpu);
660 vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
661 vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
662 vcpu->rip = svm->vmcb->save.rip;
665 static void svm_decache_regs(struct kvm_vcpu *vcpu)
667 struct vcpu_svm *svm = to_svm(vcpu);
668 svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
669 svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
670 svm->vmcb->save.rip = vcpu->rip;
673 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
675 return to_svm(vcpu)->vmcb->save.rflags;
678 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
680 to_svm(vcpu)->vmcb->save.rflags = rflags;
683 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
685 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
688 case VCPU_SREG_CS: return &save->cs;
689 case VCPU_SREG_DS: return &save->ds;
690 case VCPU_SREG_ES: return &save->es;
691 case VCPU_SREG_FS: return &save->fs;
692 case VCPU_SREG_GS: return &save->gs;
693 case VCPU_SREG_SS: return &save->ss;
694 case VCPU_SREG_TR: return &save->tr;
695 case VCPU_SREG_LDTR: return &save->ldtr;
701 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
703 struct vmcb_seg *s = svm_seg(vcpu, seg);
708 static void svm_get_segment(struct kvm_vcpu *vcpu,
709 struct kvm_segment *var, int seg)
711 struct vmcb_seg *s = svm_seg(vcpu, seg);
714 var->limit = s->limit;
715 var->selector = s->selector;
716 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
717 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
718 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
719 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
720 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
721 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
722 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
723 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
724 var->unusable = !var->present;
727 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
729 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
731 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
732 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
735 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
737 struct vcpu_svm *svm = to_svm(vcpu);
739 dt->limit = svm->vmcb->save.idtr.limit;
740 dt->base = svm->vmcb->save.idtr.base;
743 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
745 struct vcpu_svm *svm = to_svm(vcpu);
747 svm->vmcb->save.idtr.limit = dt->limit;
748 svm->vmcb->save.idtr.base = dt->base ;
751 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
753 struct vcpu_svm *svm = to_svm(vcpu);
755 dt->limit = svm->vmcb->save.gdtr.limit;
756 dt->base = svm->vmcb->save.gdtr.base;
759 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
761 struct vcpu_svm *svm = to_svm(vcpu);
763 svm->vmcb->save.gdtr.limit = dt->limit;
764 svm->vmcb->save.gdtr.base = dt->base ;
767 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
771 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
773 struct vcpu_svm *svm = to_svm(vcpu);
776 if (vcpu->shadow_efer & KVM_EFER_LME) {
777 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
778 vcpu->shadow_efer |= KVM_EFER_LMA;
779 svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
782 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG) ) {
783 vcpu->shadow_efer &= ~KVM_EFER_LMA;
784 svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
788 if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
789 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
790 vcpu->fpu_active = 1;
794 cr0 |= X86_CR0_PG | X86_CR0_WP;
795 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
796 svm->vmcb->save.cr0 = cr0;
799 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
802 to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
805 static void svm_set_segment(struct kvm_vcpu *vcpu,
806 struct kvm_segment *var, int seg)
808 struct vcpu_svm *svm = to_svm(vcpu);
809 struct vmcb_seg *s = svm_seg(vcpu, seg);
812 s->limit = var->limit;
813 s->selector = var->selector;
817 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
818 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
819 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
820 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
821 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
822 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
823 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
824 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
826 if (seg == VCPU_SREG_CS)
828 = (svm->vmcb->save.cs.attrib
829 >> SVM_SELECTOR_DPL_SHIFT) & 3;
835 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
836 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
840 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
845 static int svm_get_irq(struct kvm_vcpu *vcpu)
847 struct vcpu_svm *svm = to_svm(vcpu);
848 u32 exit_int_info = svm->vmcb->control.exit_int_info;
850 if (is_external_interrupt(exit_int_info))
851 return exit_int_info & SVM_EVTINJ_VEC_MASK;
855 static void load_host_msrs(struct kvm_vcpu *vcpu)
858 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
862 static void save_host_msrs(struct kvm_vcpu *vcpu)
865 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
869 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
871 if (svm_data->next_asid > svm_data->max_asid) {
872 ++svm_data->asid_generation;
873 svm_data->next_asid = 1;
874 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
877 svm->vcpu.cpu = svm_data->cpu;
878 svm->asid_generation = svm_data->asid_generation;
879 svm->vmcb->control.asid = svm_data->next_asid++;
882 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
884 return to_svm(vcpu)->db_regs[dr];
887 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
890 struct vcpu_svm *svm = to_svm(vcpu);
894 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
895 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
896 svm->vmcb->save.dr6 |= DR6_BD_MASK;
897 *exception = DB_VECTOR;
903 svm->db_regs[dr] = value;
906 if (vcpu->cr4 & X86_CR4_DE) {
907 *exception = UD_VECTOR;
911 if (value & ~((1ULL << 32) - 1)) {
912 *exception = GP_VECTOR;
915 svm->vmcb->save.dr7 = value;
919 printk(KERN_DEBUG "%s: unexpected dr %u\n",
921 *exception = UD_VECTOR;
926 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
928 u32 exit_int_info = svm->vmcb->control.exit_int_info;
929 struct kvm *kvm = svm->vcpu.kvm;
932 enum emulation_result er;
935 if (!irqchip_in_kernel(kvm) &&
936 is_external_interrupt(exit_int_info))
937 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
939 mutex_lock(&kvm->lock);
941 fault_address = svm->vmcb->control.exit_info_2;
942 error_code = svm->vmcb->control.exit_info_1;
943 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
945 mutex_unlock(&kvm->lock);
949 mutex_unlock(&kvm->lock);
952 er = emulate_instruction(&svm->vcpu, kvm_run, fault_address,
954 mutex_unlock(&kvm->lock);
959 case EMULATE_DO_MMIO:
960 ++svm->vcpu.stat.mmio_exits;
963 vcpu_printf(&svm->vcpu, "%s: emulate fail\n", __FUNCTION__);
969 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
973 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
975 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
976 if (!(svm->vcpu.cr0 & X86_CR0_TS))
977 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
978 svm->vcpu.fpu_active = 1;
983 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
986 * VMCB is undefined after a SHUTDOWN intercept
987 * so reinitialize it.
989 clear_page(svm->vmcb);
990 init_vmcb(svm->vmcb);
992 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
996 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
998 u32 io_info = svm->vmcb->control.exit_info_1; //address size bug?
999 int size, down, in, string, rep;
1002 ++svm->vcpu.stat.io_exits;
1004 svm->next_rip = svm->vmcb->control.exit_info_2;
1006 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1009 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0) == EMULATE_DO_MMIO)
1014 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1015 port = io_info >> 16;
1016 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1017 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1018 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1020 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1023 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1028 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1030 svm->next_rip = svm->vmcb->save.rip + 1;
1031 skip_emulated_instruction(&svm->vcpu);
1032 return kvm_emulate_halt(&svm->vcpu);
1035 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1037 svm->next_rip = svm->vmcb->save.rip + 3;
1038 skip_emulated_instruction(&svm->vcpu);
1039 return kvm_hypercall(&svm->vcpu, kvm_run);
1042 static int invalid_op_interception(struct vcpu_svm *svm,
1043 struct kvm_run *kvm_run)
1045 inject_ud(&svm->vcpu);
1049 static int task_switch_interception(struct vcpu_svm *svm,
1050 struct kvm_run *kvm_run)
1052 pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1053 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1057 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1059 svm->next_rip = svm->vmcb->save.rip + 2;
1060 kvm_emulate_cpuid(&svm->vcpu);
1064 static int emulate_on_interception(struct vcpu_svm *svm,
1065 struct kvm_run *kvm_run)
1067 if (emulate_instruction(&svm->vcpu, NULL, 0, 0) != EMULATE_DONE)
1068 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1072 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1074 struct vcpu_svm *svm = to_svm(vcpu);
1077 case MSR_IA32_TIME_STAMP_COUNTER: {
1081 *data = svm->vmcb->control.tsc_offset + tsc;
1085 *data = svm->vmcb->save.star;
1087 #ifdef CONFIG_X86_64
1089 *data = svm->vmcb->save.lstar;
1092 *data = svm->vmcb->save.cstar;
1094 case MSR_KERNEL_GS_BASE:
1095 *data = svm->vmcb->save.kernel_gs_base;
1097 case MSR_SYSCALL_MASK:
1098 *data = svm->vmcb->save.sfmask;
1101 case MSR_IA32_SYSENTER_CS:
1102 *data = svm->vmcb->save.sysenter_cs;
1104 case MSR_IA32_SYSENTER_EIP:
1105 *data = svm->vmcb->save.sysenter_eip;
1107 case MSR_IA32_SYSENTER_ESP:
1108 *data = svm->vmcb->save.sysenter_esp;
1111 return kvm_get_msr_common(vcpu, ecx, data);
1116 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1118 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1121 if (svm_get_msr(&svm->vcpu, ecx, &data))
1122 svm_inject_gp(&svm->vcpu, 0);
1124 svm->vmcb->save.rax = data & 0xffffffff;
1125 svm->vcpu.regs[VCPU_REGS_RDX] = data >> 32;
1126 svm->next_rip = svm->vmcb->save.rip + 2;
1127 skip_emulated_instruction(&svm->vcpu);
1132 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1134 struct vcpu_svm *svm = to_svm(vcpu);
1137 case MSR_IA32_TIME_STAMP_COUNTER: {
1141 svm->vmcb->control.tsc_offset = data - tsc;
1145 svm->vmcb->save.star = data;
1147 #ifdef CONFIG_X86_64
1149 svm->vmcb->save.lstar = data;
1152 svm->vmcb->save.cstar = data;
1154 case MSR_KERNEL_GS_BASE:
1155 svm->vmcb->save.kernel_gs_base = data;
1157 case MSR_SYSCALL_MASK:
1158 svm->vmcb->save.sfmask = data;
1161 case MSR_IA32_SYSENTER_CS:
1162 svm->vmcb->save.sysenter_cs = data;
1164 case MSR_IA32_SYSENTER_EIP:
1165 svm->vmcb->save.sysenter_eip = data;
1167 case MSR_IA32_SYSENTER_ESP:
1168 svm->vmcb->save.sysenter_esp = data;
1171 return kvm_set_msr_common(vcpu, ecx, data);
1176 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1178 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1179 u64 data = (svm->vmcb->save.rax & -1u)
1180 | ((u64)(svm->vcpu.regs[VCPU_REGS_RDX] & -1u) << 32);
1181 svm->next_rip = svm->vmcb->save.rip + 2;
1182 if (svm_set_msr(&svm->vcpu, ecx, data))
1183 svm_inject_gp(&svm->vcpu, 0);
1185 skip_emulated_instruction(&svm->vcpu);
1189 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1191 if (svm->vmcb->control.exit_info_1)
1192 return wrmsr_interception(svm, kvm_run);
1194 return rdmsr_interception(svm, kvm_run);
1197 static int interrupt_window_interception(struct vcpu_svm *svm,
1198 struct kvm_run *kvm_run)
1200 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1201 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1203 * If the user space waits to inject interrupts, exit as soon as
1206 if (kvm_run->request_interrupt_window &&
1207 !svm->vcpu.irq_summary) {
1208 ++svm->vcpu.stat.irq_window_exits;
1209 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1216 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1217 struct kvm_run *kvm_run) = {
1218 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1219 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1220 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1222 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1223 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1224 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1225 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1226 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1227 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1228 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1229 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1230 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1231 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1232 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1233 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1234 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1235 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1236 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1237 [SVM_EXIT_INTR] = nop_on_interception,
1238 [SVM_EXIT_NMI] = nop_on_interception,
1239 [SVM_EXIT_SMI] = nop_on_interception,
1240 [SVM_EXIT_INIT] = nop_on_interception,
1241 [SVM_EXIT_VINTR] = interrupt_window_interception,
1242 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1243 [SVM_EXIT_CPUID] = cpuid_interception,
1244 [SVM_EXIT_HLT] = halt_interception,
1245 [SVM_EXIT_INVLPG] = emulate_on_interception,
1246 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1247 [SVM_EXIT_IOIO] = io_interception,
1248 [SVM_EXIT_MSR] = msr_interception,
1249 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1250 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1251 [SVM_EXIT_VMRUN] = invalid_op_interception,
1252 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1253 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1254 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1255 [SVM_EXIT_STGI] = invalid_op_interception,
1256 [SVM_EXIT_CLGI] = invalid_op_interception,
1257 [SVM_EXIT_SKINIT] = invalid_op_interception,
1258 [SVM_EXIT_MONITOR] = invalid_op_interception,
1259 [SVM_EXIT_MWAIT] = invalid_op_interception,
1263 static int handle_exit(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1265 u32 exit_code = svm->vmcb->control.exit_code;
1267 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1268 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1269 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1271 __FUNCTION__, svm->vmcb->control.exit_int_info,
1274 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1275 || svm_exit_handlers[exit_code] == 0) {
1276 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1277 kvm_run->hw.hardware_exit_reason = exit_code;
1281 return svm_exit_handlers[exit_code](svm, kvm_run);
1284 static void reload_tss(struct kvm_vcpu *vcpu)
1286 int cpu = raw_smp_processor_id();
1288 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1289 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1293 static void pre_svm_run(struct vcpu_svm *svm)
1295 int cpu = raw_smp_processor_id();
1297 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1299 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1300 if (svm->vcpu.cpu != cpu ||
1301 svm->asid_generation != svm_data->asid_generation)
1302 new_asid(svm, svm_data);
1306 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1308 struct vmcb_control_area *control;
1310 control = &svm->vmcb->control;
1311 control->int_vector = irq;
1312 control->int_ctl &= ~V_INTR_PRIO_MASK;
1313 control->int_ctl |= V_IRQ_MASK |
1314 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1317 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1319 struct vcpu_svm *svm = to_svm(vcpu);
1321 svm_inject_irq(svm, irq);
1324 static void svm_intr_assist(struct vcpu_svm *svm)
1326 struct vmcb *vmcb = svm->vmcb;
1327 int intr_vector = -1;
1328 struct kvm_vcpu *vcpu = &svm->vcpu;
1330 kvm_inject_pending_timer_irqs(vcpu);
1331 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1332 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1333 intr_vector = vmcb->control.exit_int_info &
1334 SVM_EVTINJ_VEC_MASK;
1335 vmcb->control.exit_int_info = 0;
1336 svm_inject_irq(svm, intr_vector);
1340 if (vmcb->control.int_ctl & V_IRQ_MASK)
1343 if (!kvm_cpu_has_interrupt(vcpu))
1346 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1347 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1348 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1349 /* unable to deliver irq, set pending irq */
1350 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1351 svm_inject_irq(svm, 0x0);
1354 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1355 intr_vector = kvm_cpu_get_interrupt(vcpu);
1356 svm_inject_irq(svm, intr_vector);
1357 kvm_timer_intr_post(vcpu, intr_vector);
1360 static void kvm_reput_irq(struct vcpu_svm *svm)
1362 struct vmcb_control_area *control = &svm->vmcb->control;
1364 if ((control->int_ctl & V_IRQ_MASK)
1365 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1366 control->int_ctl &= ~V_IRQ_MASK;
1367 push_irq(&svm->vcpu, control->int_vector);
1370 svm->vcpu.interrupt_window_open =
1371 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1374 static void svm_do_inject_vector(struct vcpu_svm *svm)
1376 struct kvm_vcpu *vcpu = &svm->vcpu;
1377 int word_index = __ffs(vcpu->irq_summary);
1378 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1379 int irq = word_index * BITS_PER_LONG + bit_index;
1381 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1382 if (!vcpu->irq_pending[word_index])
1383 clear_bit(word_index, &vcpu->irq_summary);
1384 svm_inject_irq(svm, irq);
1387 static void do_interrupt_requests(struct vcpu_svm *svm,
1388 struct kvm_run *kvm_run)
1390 struct vmcb_control_area *control = &svm->vmcb->control;
1392 svm->vcpu.interrupt_window_open =
1393 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1394 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1396 if (svm->vcpu.interrupt_window_open && svm->vcpu.irq_summary)
1398 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1400 svm_do_inject_vector(svm);
1403 * Interrupts blocked. Wait for unblock.
1405 if (!svm->vcpu.interrupt_window_open &&
1406 (svm->vcpu.irq_summary || kvm_run->request_interrupt_window)) {
1407 control->intercept |= 1ULL << INTERCEPT_VINTR;
1409 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1412 static void post_kvm_run_save(struct vcpu_svm *svm,
1413 struct kvm_run *kvm_run)
1415 if (irqchip_in_kernel(svm->vcpu.kvm))
1416 kvm_run->ready_for_interrupt_injection = 1;
1418 kvm_run->ready_for_interrupt_injection =
1419 (svm->vcpu.interrupt_window_open &&
1420 svm->vcpu.irq_summary == 0);
1421 kvm_run->if_flag = (svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1422 kvm_run->cr8 = get_cr8(&svm->vcpu);
1423 kvm_run->apic_base = kvm_get_apic_base(&svm->vcpu);
1427 * Check if userspace requested an interrupt window, and that the
1428 * interrupt window is open.
1430 * No need to exit to userspace if we already have an interrupt queued.
1432 static int dm_request_for_irq_injection(struct vcpu_svm *svm,
1433 struct kvm_run *kvm_run)
1435 return (!svm->vcpu.irq_summary &&
1436 kvm_run->request_interrupt_window &&
1437 svm->vcpu.interrupt_window_open &&
1438 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1441 static void save_db_regs(unsigned long *db_regs)
1443 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1444 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1445 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1446 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1449 static void load_db_regs(unsigned long *db_regs)
1451 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1452 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1453 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1454 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1457 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1459 force_new_asid(vcpu);
1462 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1464 struct vcpu_svm *svm = to_svm(vcpu);
1471 r = kvm_mmu_reload(vcpu);
1477 if (signal_pending(current)) {
1479 ++vcpu->stat.signal_exits;
1480 post_kvm_run_save(svm, kvm_run);
1481 kvm_run->exit_reason = KVM_EXIT_INTR;
1485 if (irqchip_in_kernel(vcpu->kvm))
1486 svm_intr_assist(svm);
1487 else if (!vcpu->mmio_read_completed)
1488 do_interrupt_requests(svm, kvm_run);
1490 vcpu->guest_mode = 1;
1492 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
1493 svm_flush_tlb(vcpu);
1497 save_host_msrs(vcpu);
1498 fs_selector = read_fs();
1499 gs_selector = read_gs();
1500 ldt_selector = read_ldt();
1501 svm->host_cr2 = kvm_read_cr2();
1502 svm->host_dr6 = read_dr6();
1503 svm->host_dr7 = read_dr7();
1504 svm->vmcb->save.cr2 = vcpu->cr2;
1506 if (svm->vmcb->save.dr7 & 0xff) {
1508 save_db_regs(svm->host_db_regs);
1509 load_db_regs(svm->db_regs);
1512 if (vcpu->fpu_active) {
1513 fx_save(&vcpu->host_fx_image);
1514 fx_restore(&vcpu->guest_fx_image);
1518 #ifdef CONFIG_X86_64
1519 "push %%rbx; push %%rcx; push %%rdx;"
1520 "push %%rsi; push %%rdi; push %%rbp;"
1521 "push %%r8; push %%r9; push %%r10; push %%r11;"
1522 "push %%r12; push %%r13; push %%r14; push %%r15;"
1524 "push %%ebx; push %%ecx; push %%edx;"
1525 "push %%esi; push %%edi; push %%ebp;"
1528 #ifdef CONFIG_X86_64
1529 "mov %c[rbx](%[svm]), %%rbx \n\t"
1530 "mov %c[rcx](%[svm]), %%rcx \n\t"
1531 "mov %c[rdx](%[svm]), %%rdx \n\t"
1532 "mov %c[rsi](%[svm]), %%rsi \n\t"
1533 "mov %c[rdi](%[svm]), %%rdi \n\t"
1534 "mov %c[rbp](%[svm]), %%rbp \n\t"
1535 "mov %c[r8](%[svm]), %%r8 \n\t"
1536 "mov %c[r9](%[svm]), %%r9 \n\t"
1537 "mov %c[r10](%[svm]), %%r10 \n\t"
1538 "mov %c[r11](%[svm]), %%r11 \n\t"
1539 "mov %c[r12](%[svm]), %%r12 \n\t"
1540 "mov %c[r13](%[svm]), %%r13 \n\t"
1541 "mov %c[r14](%[svm]), %%r14 \n\t"
1542 "mov %c[r15](%[svm]), %%r15 \n\t"
1544 "mov %c[rbx](%[svm]), %%ebx \n\t"
1545 "mov %c[rcx](%[svm]), %%ecx \n\t"
1546 "mov %c[rdx](%[svm]), %%edx \n\t"
1547 "mov %c[rsi](%[svm]), %%esi \n\t"
1548 "mov %c[rdi](%[svm]), %%edi \n\t"
1549 "mov %c[rbp](%[svm]), %%ebp \n\t"
1552 #ifdef CONFIG_X86_64
1553 /* Enter guest mode */
1555 "mov %c[vmcb](%[svm]), %%rax \n\t"
1561 /* Enter guest mode */
1563 "mov %c[vmcb](%[svm]), %%eax \n\t"
1570 /* Save guest registers, load host registers */
1571 #ifdef CONFIG_X86_64
1572 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1573 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1574 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1575 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1576 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1577 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1578 "mov %%r8, %c[r8](%[svm]) \n\t"
1579 "mov %%r9, %c[r9](%[svm]) \n\t"
1580 "mov %%r10, %c[r10](%[svm]) \n\t"
1581 "mov %%r11, %c[r11](%[svm]) \n\t"
1582 "mov %%r12, %c[r12](%[svm]) \n\t"
1583 "mov %%r13, %c[r13](%[svm]) \n\t"
1584 "mov %%r14, %c[r14](%[svm]) \n\t"
1585 "mov %%r15, %c[r15](%[svm]) \n\t"
1587 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1588 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1589 "pop %%rbp; pop %%rdi; pop %%rsi;"
1590 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1592 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1593 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1594 "mov %%edx, %c[rdx](%[svm]) \n\t"
1595 "mov %%esi, %c[rsi](%[svm]) \n\t"
1596 "mov %%edi, %c[rdi](%[svm]) \n\t"
1597 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1599 "pop %%ebp; pop %%edi; pop %%esi;"
1600 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1604 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1605 [rbx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBX])),
1606 [rcx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RCX])),
1607 [rdx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDX])),
1608 [rsi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RSI])),
1609 [rdi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDI])),
1610 [rbp]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBP]))
1611 #ifdef CONFIG_X86_64
1612 ,[r8 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R8])),
1613 [r9 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R9 ])),
1614 [r10]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R10])),
1615 [r11]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R11])),
1616 [r12]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R12])),
1617 [r13]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R13])),
1618 [r14]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R14])),
1619 [r15]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R15]))
1623 vcpu->guest_mode = 0;
1625 if (vcpu->fpu_active) {
1626 fx_save(&vcpu->guest_fx_image);
1627 fx_restore(&vcpu->host_fx_image);
1630 if ((svm->vmcb->save.dr7 & 0xff))
1631 load_db_regs(svm->host_db_regs);
1633 vcpu->cr2 = svm->vmcb->save.cr2;
1635 write_dr6(svm->host_dr6);
1636 write_dr7(svm->host_dr7);
1637 kvm_write_cr2(svm->host_cr2);
1639 load_fs(fs_selector);
1640 load_gs(gs_selector);
1641 load_ldt(ldt_selector);
1642 load_host_msrs(vcpu);
1647 * Profile KVM exit RIPs:
1649 if (unlikely(prof_on == KVM_PROFILING))
1650 profile_hit(KVM_PROFILING,
1651 (void *)(unsigned long)svm->vmcb->save.rip);
1659 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1660 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1661 kvm_run->fail_entry.hardware_entry_failure_reason
1662 = svm->vmcb->control.exit_code;
1663 post_kvm_run_save(svm, kvm_run);
1667 r = handle_exit(svm, kvm_run);
1669 if (dm_request_for_irq_injection(svm, kvm_run)) {
1670 ++vcpu->stat.request_irq_exits;
1671 post_kvm_run_save(svm, kvm_run);
1672 kvm_run->exit_reason = KVM_EXIT_INTR;
1678 post_kvm_run_save(svm, kvm_run);
1682 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1684 struct vcpu_svm *svm = to_svm(vcpu);
1686 svm->vmcb->save.cr3 = root;
1687 force_new_asid(vcpu);
1689 if (vcpu->fpu_active) {
1690 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1691 svm->vmcb->save.cr0 |= X86_CR0_TS;
1692 vcpu->fpu_active = 0;
1696 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1700 struct vcpu_svm *svm = to_svm(vcpu);
1701 uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1703 ++vcpu->stat.pf_guest;
1705 if (is_page_fault(exit_int_info)) {
1707 svm->vmcb->control.event_inj_err = 0;
1708 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1709 SVM_EVTINJ_VALID_ERR |
1710 SVM_EVTINJ_TYPE_EXEPT |
1715 svm->vmcb->save.cr2 = addr;
1716 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1717 SVM_EVTINJ_VALID_ERR |
1718 SVM_EVTINJ_TYPE_EXEPT |
1720 svm->vmcb->control.event_inj_err = err_code;
1724 static int is_disabled(void)
1728 rdmsrl(MSR_VM_CR, vm_cr);
1729 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1736 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1739 * Patch in the VMMCALL instruction:
1741 hypercall[0] = 0x0f;
1742 hypercall[1] = 0x01;
1743 hypercall[2] = 0xd9;
1744 hypercall[3] = 0xc3;
1747 static void svm_check_processor_compat(void *rtn)
1752 static struct kvm_arch_ops svm_arch_ops = {
1753 .cpu_has_kvm_support = has_svm,
1754 .disabled_by_bios = is_disabled,
1755 .hardware_setup = svm_hardware_setup,
1756 .hardware_unsetup = svm_hardware_unsetup,
1757 .check_processor_compatibility = svm_check_processor_compat,
1758 .hardware_enable = svm_hardware_enable,
1759 .hardware_disable = svm_hardware_disable,
1761 .vcpu_create = svm_create_vcpu,
1762 .vcpu_free = svm_free_vcpu,
1764 .vcpu_load = svm_vcpu_load,
1765 .vcpu_put = svm_vcpu_put,
1766 .vcpu_decache = svm_vcpu_decache,
1768 .set_guest_debug = svm_guest_debug,
1769 .get_msr = svm_get_msr,
1770 .set_msr = svm_set_msr,
1771 .get_segment_base = svm_get_segment_base,
1772 .get_segment = svm_get_segment,
1773 .set_segment = svm_set_segment,
1774 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1775 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1776 .set_cr0 = svm_set_cr0,
1777 .set_cr3 = svm_set_cr3,
1778 .set_cr4 = svm_set_cr4,
1779 .set_efer = svm_set_efer,
1780 .get_idt = svm_get_idt,
1781 .set_idt = svm_set_idt,
1782 .get_gdt = svm_get_gdt,
1783 .set_gdt = svm_set_gdt,
1784 .get_dr = svm_get_dr,
1785 .set_dr = svm_set_dr,
1786 .cache_regs = svm_cache_regs,
1787 .decache_regs = svm_decache_regs,
1788 .get_rflags = svm_get_rflags,
1789 .set_rflags = svm_set_rflags,
1791 .tlb_flush = svm_flush_tlb,
1792 .inject_page_fault = svm_inject_page_fault,
1794 .inject_gp = svm_inject_gp,
1796 .run = svm_vcpu_run,
1797 .skip_emulated_instruction = skip_emulated_instruction,
1798 .patch_hypercall = svm_patch_hypercall,
1799 .get_irq = svm_get_irq,
1800 .set_irq = svm_set_irq,
1803 static int __init svm_init(void)
1805 return kvm_init_arch(&svm_arch_ops, sizeof(struct vcpu_svm),
1809 static void __exit svm_exit(void)
1814 module_init(svm_init)
1815 module_exit(svm_exit)