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.
16 #include <linux/kvm_host.h>
20 #include "kvm_cache_regs.h"
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/vmalloc.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/ftrace_event.h>
32 #include <asm/virtext.h>
35 #define __ex(x) __kvm_handle_fault_on_reboot(x)
37 MODULE_AUTHOR("Qumranet");
38 MODULE_LICENSE("GPL");
40 #define IOPM_ALLOC_ORDER 2
41 #define MSRPM_ALLOC_ORDER 1
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
46 #define SVM_FEATURE_NPT (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_FEATURE_SVML (1 << 2)
50 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
51 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
52 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
54 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
56 /* Turn on to get debugging output*/
57 /* #define NESTED_DEBUG */
60 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
62 #define nsvm_printk(fmt, args...) do {} while(0)
65 static const u32 host_save_user_msrs[] = {
67 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
70 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
73 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
82 /* These are the merged vectors */
85 /* gpa pointers to the real vectors */
88 /* cache for intercepts of the guest */
89 u16 intercept_cr_read;
90 u16 intercept_cr_write;
91 u16 intercept_dr_read;
92 u16 intercept_dr_write;
93 u32 intercept_exceptions;
101 unsigned long vmcb_pa;
102 struct svm_cpu_data *svm_data;
103 uint64_t asid_generation;
104 uint64_t sysenter_esp;
105 uint64_t sysenter_eip;
109 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
114 struct nested_state nested;
117 /* enable NPT for AMD64 and X86 with PAE */
118 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
119 static bool npt_enabled = true;
121 static bool npt_enabled = false;
125 module_param(npt, int, S_IRUGO);
127 static int nested = 1;
128 module_param(nested, int, S_IRUGO);
130 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
131 static void svm_complete_interrupts(struct vcpu_svm *svm);
133 static int nested_svm_exit_handled(struct vcpu_svm *svm);
134 static int nested_svm_vmexit(struct vcpu_svm *svm);
135 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
136 bool has_error_code, u32 error_code);
138 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
140 return container_of(vcpu, struct vcpu_svm, vcpu);
143 static inline bool is_nested(struct vcpu_svm *svm)
145 return svm->nested.vmcb;
148 static inline void enable_gif(struct vcpu_svm *svm)
150 svm->vcpu.arch.hflags |= HF_GIF_MASK;
153 static inline void disable_gif(struct vcpu_svm *svm)
155 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
158 static inline bool gif_set(struct vcpu_svm *svm)
160 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
163 static unsigned long iopm_base;
165 struct kvm_ldttss_desc {
168 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
169 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
172 } __attribute__((packed));
174 struct svm_cpu_data {
180 struct kvm_ldttss_desc *tss_desc;
182 struct page *save_area;
185 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
186 static uint32_t svm_features;
188 struct svm_init_data {
193 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
195 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
196 #define MSRS_RANGE_SIZE 2048
197 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
199 #define MAX_INST_SIZE 15
201 static inline u32 svm_has(u32 feat)
203 return svm_features & feat;
206 static inline void clgi(void)
208 asm volatile (__ex(SVM_CLGI));
211 static inline void stgi(void)
213 asm volatile (__ex(SVM_STGI));
216 static inline void invlpga(unsigned long addr, u32 asid)
218 asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
221 static inline void force_new_asid(struct kvm_vcpu *vcpu)
223 to_svm(vcpu)->asid_generation--;
226 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
228 force_new_asid(vcpu);
231 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
233 if (!npt_enabled && !(efer & EFER_LMA))
236 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
237 vcpu->arch.shadow_efer = efer;
240 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
241 bool has_error_code, u32 error_code)
243 struct vcpu_svm *svm = to_svm(vcpu);
245 /* If we are within a nested VM we'd better #VMEXIT and let the
246 guest handle the exception */
247 if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
250 svm->vmcb->control.event_inj = nr
252 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
253 | SVM_EVTINJ_TYPE_EXEPT;
254 svm->vmcb->control.event_inj_err = error_code;
257 static int is_external_interrupt(u32 info)
259 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
260 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
263 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
265 struct vcpu_svm *svm = to_svm(vcpu);
268 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
269 ret |= X86_SHADOW_INT_STI | X86_SHADOW_INT_MOV_SS;
273 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
275 struct vcpu_svm *svm = to_svm(vcpu);
278 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
280 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
284 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
286 struct vcpu_svm *svm = to_svm(vcpu);
288 if (!svm->next_rip) {
289 if (emulate_instruction(vcpu, vcpu->run, 0, 0, EMULTYPE_SKIP) !=
291 printk(KERN_DEBUG "%s: NOP\n", __func__);
294 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
295 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
296 __func__, kvm_rip_read(vcpu), svm->next_rip);
298 kvm_rip_write(vcpu, svm->next_rip);
299 svm_set_interrupt_shadow(vcpu, 0);
302 static int has_svm(void)
306 if (!cpu_has_svm(&msg)) {
307 printk(KERN_INFO "has_svm: %s\n", msg);
314 static void svm_hardware_disable(void *garbage)
319 static void svm_hardware_enable(void *garbage)
322 struct svm_cpu_data *sd;
324 struct descriptor_table gdt_descr;
325 struct desc_struct *gdt;
326 int me = raw_smp_processor_id();
329 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
332 sd = per_cpu(svm_data, me);
335 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
340 sd->asid_generation = 1;
341 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
342 sd->next_asid = sd->max_asid + 1;
344 kvm_get_gdt(&gdt_descr);
345 gdt = (struct desc_struct *)gdt_descr.base;
346 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
348 rdmsrl(MSR_EFER, efer);
349 wrmsrl(MSR_EFER, efer | EFER_SVME);
351 wrmsrl(MSR_VM_HSAVE_PA,
352 page_to_pfn(sd->save_area) << PAGE_SHIFT);
355 static void svm_cpu_uninit(int cpu)
357 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
362 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
363 __free_page(sd->save_area);
367 static int svm_cpu_init(int cpu)
369 struct svm_cpu_data *sd;
372 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
376 sd->save_area = alloc_page(GFP_KERNEL);
381 per_cpu(svm_data, cpu) = sd;
391 static void set_msr_interception(u32 *msrpm, unsigned msr,
396 for (i = 0; i < NUM_MSR_MAPS; i++) {
397 if (msr >= msrpm_ranges[i] &&
398 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
399 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
400 msrpm_ranges[i]) * 2;
402 u32 *base = msrpm + (msr_offset / 32);
403 u32 msr_shift = msr_offset % 32;
404 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
405 *base = (*base & ~(0x3 << msr_shift)) |
413 static void svm_vcpu_init_msrpm(u32 *msrpm)
415 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
418 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
419 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
420 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
421 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
422 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
423 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
425 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
426 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
429 static void svm_enable_lbrv(struct vcpu_svm *svm)
431 u32 *msrpm = svm->msrpm;
433 svm->vmcb->control.lbr_ctl = 1;
434 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
435 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
436 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
437 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
440 static void svm_disable_lbrv(struct vcpu_svm *svm)
442 u32 *msrpm = svm->msrpm;
444 svm->vmcb->control.lbr_ctl = 0;
445 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
446 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
447 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
448 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
451 static __init int svm_hardware_setup(void)
454 struct page *iopm_pages;
458 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
463 iopm_va = page_address(iopm_pages);
464 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
465 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
467 if (boot_cpu_has(X86_FEATURE_NX))
468 kvm_enable_efer_bits(EFER_NX);
470 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
471 kvm_enable_efer_bits(EFER_FFXSR);
474 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
475 kvm_enable_efer_bits(EFER_SVME);
478 for_each_online_cpu(cpu) {
479 r = svm_cpu_init(cpu);
484 svm_features = cpuid_edx(SVM_CPUID_FUNC);
486 if (!svm_has(SVM_FEATURE_NPT))
489 if (npt_enabled && !npt) {
490 printk(KERN_INFO "kvm: Nested Paging disabled\n");
495 printk(KERN_INFO "kvm: Nested Paging enabled\n");
503 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
508 static __exit void svm_hardware_unsetup(void)
512 for_each_online_cpu(cpu)
515 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
519 static void init_seg(struct vmcb_seg *seg)
522 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
523 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
528 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
531 seg->attrib = SVM_SELECTOR_P_MASK | type;
536 static void init_vmcb(struct vcpu_svm *svm)
538 struct vmcb_control_area *control = &svm->vmcb->control;
539 struct vmcb_save_area *save = &svm->vmcb->save;
541 control->intercept_cr_read = INTERCEPT_CR0_MASK |
545 control->intercept_cr_write = INTERCEPT_CR0_MASK |
550 control->intercept_dr_read = INTERCEPT_DR0_MASK |
555 control->intercept_dr_write = INTERCEPT_DR0_MASK |
562 control->intercept_exceptions = (1 << PF_VECTOR) |
567 control->intercept = (1ULL << INTERCEPT_INTR) |
568 (1ULL << INTERCEPT_NMI) |
569 (1ULL << INTERCEPT_SMI) |
570 (1ULL << INTERCEPT_CPUID) |
571 (1ULL << INTERCEPT_INVD) |
572 (1ULL << INTERCEPT_HLT) |
573 (1ULL << INTERCEPT_INVLPG) |
574 (1ULL << INTERCEPT_INVLPGA) |
575 (1ULL << INTERCEPT_IOIO_PROT) |
576 (1ULL << INTERCEPT_MSR_PROT) |
577 (1ULL << INTERCEPT_TASK_SWITCH) |
578 (1ULL << INTERCEPT_SHUTDOWN) |
579 (1ULL << INTERCEPT_VMRUN) |
580 (1ULL << INTERCEPT_VMMCALL) |
581 (1ULL << INTERCEPT_VMLOAD) |
582 (1ULL << INTERCEPT_VMSAVE) |
583 (1ULL << INTERCEPT_STGI) |
584 (1ULL << INTERCEPT_CLGI) |
585 (1ULL << INTERCEPT_SKINIT) |
586 (1ULL << INTERCEPT_WBINVD) |
587 (1ULL << INTERCEPT_MONITOR) |
588 (1ULL << INTERCEPT_MWAIT);
590 control->iopm_base_pa = iopm_base;
591 control->msrpm_base_pa = __pa(svm->msrpm);
592 control->tsc_offset = 0;
593 control->int_ctl = V_INTR_MASKING_MASK;
601 save->cs.selector = 0xf000;
602 /* Executable/Readable Code Segment */
603 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
604 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
605 save->cs.limit = 0xffff;
607 * cs.base should really be 0xffff0000, but vmx can't handle that, so
608 * be consistent with it.
610 * Replace when we have real mode working for vmx.
612 save->cs.base = 0xf0000;
614 save->gdtr.limit = 0xffff;
615 save->idtr.limit = 0xffff;
617 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
618 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
620 save->efer = EFER_SVME;
621 save->dr6 = 0xffff0ff0;
624 save->rip = 0x0000fff0;
625 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
628 * cr0 val on cpu init should be 0x60000010, we enable cpu
629 * cache by default. the orderly way is to enable cache in bios.
631 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
632 save->cr4 = X86_CR4_PAE;
636 /* Setup VMCB for Nested Paging */
637 control->nested_ctl = 1;
638 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
639 (1ULL << INTERCEPT_INVLPG));
640 control->intercept_exceptions &= ~(1 << PF_VECTOR);
641 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
643 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
645 save->g_pat = 0x0007040600070406ULL;
646 /* enable caching because the QEMU Bios doesn't enable it */
647 save->cr0 = X86_CR0_ET;
651 force_new_asid(&svm->vcpu);
653 svm->nested.vmcb = 0;
654 svm->vcpu.arch.hflags = 0;
659 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
661 struct vcpu_svm *svm = to_svm(vcpu);
665 if (!kvm_vcpu_is_bsp(vcpu)) {
666 kvm_rip_write(vcpu, 0);
667 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
668 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
670 vcpu->arch.regs_avail = ~0;
671 vcpu->arch.regs_dirty = ~0;
676 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
678 struct vcpu_svm *svm;
680 struct page *msrpm_pages;
681 struct page *hsave_page;
682 struct page *nested_msrpm_pages;
685 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
691 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
695 page = alloc_page(GFP_KERNEL);
702 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
706 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
707 if (!nested_msrpm_pages)
710 svm->msrpm = page_address(msrpm_pages);
711 svm_vcpu_init_msrpm(svm->msrpm);
713 hsave_page = alloc_page(GFP_KERNEL);
716 svm->nested.hsave = page_address(hsave_page);
718 svm->nested.msrpm = page_address(nested_msrpm_pages);
720 svm->vmcb = page_address(page);
721 clear_page(svm->vmcb);
722 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
723 svm->asid_generation = 0;
727 svm->vcpu.fpu_active = 1;
728 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
729 if (kvm_vcpu_is_bsp(&svm->vcpu))
730 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
735 kvm_vcpu_uninit(&svm->vcpu);
737 kmem_cache_free(kvm_vcpu_cache, svm);
742 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
744 struct vcpu_svm *svm = to_svm(vcpu);
746 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
747 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
748 __free_page(virt_to_page(svm->nested.hsave));
749 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
750 kvm_vcpu_uninit(vcpu);
751 kmem_cache_free(kvm_vcpu_cache, svm);
754 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
756 struct vcpu_svm *svm = to_svm(vcpu);
759 if (unlikely(cpu != vcpu->cpu)) {
763 * Make sure that the guest sees a monotonically
767 delta = vcpu->arch.host_tsc - tsc_this;
768 svm->vmcb->control.tsc_offset += delta;
770 kvm_migrate_timers(vcpu);
771 svm->asid_generation = 0;
774 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
775 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
778 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
780 struct vcpu_svm *svm = to_svm(vcpu);
783 ++vcpu->stat.host_state_reload;
784 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
785 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
787 rdtscll(vcpu->arch.host_tsc);
790 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
792 return to_svm(vcpu)->vmcb->save.rflags;
795 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
797 to_svm(vcpu)->vmcb->save.rflags = rflags;
800 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
803 case VCPU_EXREG_PDPTR:
804 BUG_ON(!npt_enabled);
805 load_pdptrs(vcpu, vcpu->arch.cr3);
812 static void svm_set_vintr(struct vcpu_svm *svm)
814 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
817 static void svm_clear_vintr(struct vcpu_svm *svm)
819 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
822 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
824 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
827 case VCPU_SREG_CS: return &save->cs;
828 case VCPU_SREG_DS: return &save->ds;
829 case VCPU_SREG_ES: return &save->es;
830 case VCPU_SREG_FS: return &save->fs;
831 case VCPU_SREG_GS: return &save->gs;
832 case VCPU_SREG_SS: return &save->ss;
833 case VCPU_SREG_TR: return &save->tr;
834 case VCPU_SREG_LDTR: return &save->ldtr;
840 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
842 struct vmcb_seg *s = svm_seg(vcpu, seg);
847 static void svm_get_segment(struct kvm_vcpu *vcpu,
848 struct kvm_segment *var, int seg)
850 struct vmcb_seg *s = svm_seg(vcpu, seg);
853 var->limit = s->limit;
854 var->selector = s->selector;
855 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
856 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
857 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
858 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
859 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
860 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
861 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
862 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
864 /* AMD's VMCB does not have an explicit unusable field, so emulate it
865 * for cross vendor migration purposes by "not present"
867 var->unusable = !var->present || (var->type == 0);
872 * SVM always stores 0 for the 'G' bit in the CS selector in
873 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
874 * Intel's VMENTRY has a check on the 'G' bit.
876 var->g = s->limit > 0xfffff;
880 * Work around a bug where the busy flag in the tr selector
890 * The accessed bit must always be set in the segment
891 * descriptor cache, although it can be cleared in the
892 * descriptor, the cached bit always remains at 1. Since
893 * Intel has a check on this, set it here to support
894 * cross-vendor migration.
900 /* On AMD CPUs sometimes the DB bit in the segment
901 * descriptor is left as 1, although the whole segment has
902 * been made unusable. Clear it here to pass an Intel VMX
903 * entry check when cross vendor migrating.
911 static int svm_get_cpl(struct kvm_vcpu *vcpu)
913 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
918 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
920 struct vcpu_svm *svm = to_svm(vcpu);
922 dt->limit = svm->vmcb->save.idtr.limit;
923 dt->base = svm->vmcb->save.idtr.base;
926 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
928 struct vcpu_svm *svm = to_svm(vcpu);
930 svm->vmcb->save.idtr.limit = dt->limit;
931 svm->vmcb->save.idtr.base = dt->base ;
934 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
936 struct vcpu_svm *svm = to_svm(vcpu);
938 dt->limit = svm->vmcb->save.gdtr.limit;
939 dt->base = svm->vmcb->save.gdtr.base;
942 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
944 struct vcpu_svm *svm = to_svm(vcpu);
946 svm->vmcb->save.gdtr.limit = dt->limit;
947 svm->vmcb->save.gdtr.base = dt->base ;
950 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
954 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
956 struct vcpu_svm *svm = to_svm(vcpu);
959 if (vcpu->arch.shadow_efer & EFER_LME) {
960 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
961 vcpu->arch.shadow_efer |= EFER_LMA;
962 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
965 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
966 vcpu->arch.shadow_efer &= ~EFER_LMA;
967 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
974 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
975 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
976 vcpu->fpu_active = 1;
979 vcpu->arch.cr0 = cr0;
980 cr0 |= X86_CR0_PG | X86_CR0_WP;
981 if (!vcpu->fpu_active) {
982 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
987 * re-enable caching here because the QEMU bios
988 * does not do it - this results in some delay at
991 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
992 svm->vmcb->save.cr0 = cr0;
995 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
997 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
998 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1000 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1001 force_new_asid(vcpu);
1003 vcpu->arch.cr4 = cr4;
1006 cr4 |= host_cr4_mce;
1007 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1010 static void svm_set_segment(struct kvm_vcpu *vcpu,
1011 struct kvm_segment *var, int seg)
1013 struct vcpu_svm *svm = to_svm(vcpu);
1014 struct vmcb_seg *s = svm_seg(vcpu, seg);
1016 s->base = var->base;
1017 s->limit = var->limit;
1018 s->selector = var->selector;
1022 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1023 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1024 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1025 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1026 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1027 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1028 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1029 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1031 if (seg == VCPU_SREG_CS)
1033 = (svm->vmcb->save.cs.attrib
1034 >> SVM_SELECTOR_DPL_SHIFT) & 3;
1038 static void update_db_intercept(struct kvm_vcpu *vcpu)
1040 struct vcpu_svm *svm = to_svm(vcpu);
1042 svm->vmcb->control.intercept_exceptions &=
1043 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1045 if (vcpu->arch.singlestep)
1046 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1048 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1049 if (vcpu->guest_debug &
1050 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1051 svm->vmcb->control.intercept_exceptions |=
1053 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1054 svm->vmcb->control.intercept_exceptions |=
1057 vcpu->guest_debug = 0;
1060 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1062 int old_debug = vcpu->guest_debug;
1063 struct vcpu_svm *svm = to_svm(vcpu);
1065 vcpu->guest_debug = dbg->control;
1067 update_db_intercept(vcpu);
1069 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1070 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1072 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1074 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1075 svm->vmcb->save.rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1076 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1077 svm->vmcb->save.rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1082 static void load_host_msrs(struct kvm_vcpu *vcpu)
1084 #ifdef CONFIG_X86_64
1085 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1089 static void save_host_msrs(struct kvm_vcpu *vcpu)
1091 #ifdef CONFIG_X86_64
1092 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1096 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1098 if (sd->next_asid > sd->max_asid) {
1099 ++sd->asid_generation;
1101 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1104 svm->asid_generation = sd->asid_generation;
1105 svm->vmcb->control.asid = sd->next_asid++;
1108 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
1110 struct vcpu_svm *svm = to_svm(vcpu);
1115 val = vcpu->arch.db[dr];
1118 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1119 val = vcpu->arch.dr6;
1121 val = svm->vmcb->save.dr6;
1124 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1125 val = vcpu->arch.dr7;
1127 val = svm->vmcb->save.dr7;
1136 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
1139 struct vcpu_svm *svm = to_svm(vcpu);
1145 vcpu->arch.db[dr] = value;
1146 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1147 vcpu->arch.eff_db[dr] = value;
1150 if (vcpu->arch.cr4 & X86_CR4_DE)
1151 *exception = UD_VECTOR;
1154 if (value & 0xffffffff00000000ULL) {
1155 *exception = GP_VECTOR;
1158 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1161 if (value & 0xffffffff00000000ULL) {
1162 *exception = GP_VECTOR;
1165 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1166 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1167 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1168 vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1172 /* FIXME: Possible case? */
1173 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1175 *exception = UD_VECTOR;
1180 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1185 fault_address = svm->vmcb->control.exit_info_2;
1186 error_code = svm->vmcb->control.exit_info_1;
1188 trace_kvm_page_fault(fault_address, error_code);
1189 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1190 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1191 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1194 static int db_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1196 if (!(svm->vcpu.guest_debug &
1197 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1198 !svm->vcpu.arch.singlestep) {
1199 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1203 if (svm->vcpu.arch.singlestep) {
1204 svm->vcpu.arch.singlestep = false;
1205 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1206 svm->vmcb->save.rflags &=
1207 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1208 update_db_intercept(&svm->vcpu);
1211 if (svm->vcpu.guest_debug &
1212 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)){
1213 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1214 kvm_run->debug.arch.pc =
1215 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1216 kvm_run->debug.arch.exception = DB_VECTOR;
1223 static int bp_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1225 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1226 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1227 kvm_run->debug.arch.exception = BP_VECTOR;
1231 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1235 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1236 if (er != EMULATE_DONE)
1237 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1241 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1243 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1244 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1245 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1246 svm->vcpu.fpu_active = 1;
1251 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1254 * On an #MC intercept the MCE handler is not called automatically in
1255 * the host. So do it by hand here.
1259 /* not sure if we ever come back to this point */
1264 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1267 * VMCB is undefined after a SHUTDOWN intercept
1268 * so reinitialize it.
1270 clear_page(svm->vmcb);
1273 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1277 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1279 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1280 int size, in, string;
1283 ++svm->vcpu.stat.io_exits;
1285 svm->next_rip = svm->vmcb->control.exit_info_2;
1287 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1290 if (emulate_instruction(&svm->vcpu,
1291 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1296 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1297 port = io_info >> 16;
1298 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1300 skip_emulated_instruction(&svm->vcpu);
1301 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1304 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1309 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1311 ++svm->vcpu.stat.irq_exits;
1315 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1320 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1322 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1323 skip_emulated_instruction(&svm->vcpu);
1324 return kvm_emulate_halt(&svm->vcpu);
1327 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1329 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1330 skip_emulated_instruction(&svm->vcpu);
1331 kvm_emulate_hypercall(&svm->vcpu);
1335 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1337 if (!(svm->vcpu.arch.shadow_efer & EFER_SVME)
1338 || !is_paging(&svm->vcpu)) {
1339 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1343 if (svm->vmcb->save.cpl) {
1344 kvm_inject_gp(&svm->vcpu, 0);
1351 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1352 bool has_error_code, u32 error_code)
1354 if (!is_nested(svm))
1357 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1358 svm->vmcb->control.exit_code_hi = 0;
1359 svm->vmcb->control.exit_info_1 = error_code;
1360 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1362 return nested_svm_exit_handled(svm);
1365 static inline int nested_svm_intr(struct vcpu_svm *svm)
1367 if (!is_nested(svm))
1370 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1373 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1376 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1378 if (nested_svm_exit_handled(svm)) {
1379 nsvm_printk("VMexit -> INTR\n");
1386 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, enum km_type idx)
1390 down_read(¤t->mm->mmap_sem);
1391 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1392 up_read(¤t->mm->mmap_sem);
1394 if (is_error_page(page))
1397 return kmap_atomic(page, idx);
1400 kvm_release_page_clean(page);
1401 kvm_inject_gp(&svm->vcpu, 0);
1406 static void nested_svm_unmap(void *addr, enum km_type idx)
1413 page = kmap_atomic_to_page(addr);
1415 kunmap_atomic(addr, idx);
1416 kvm_release_page_dirty(page);
1419 static bool nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1421 u32 param = svm->vmcb->control.exit_info_1 & 1;
1422 u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1427 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1430 msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, KM_USER0);
1440 case 0xc0000000 ... 0xc0001fff:
1441 t0 = (8192 + msr - 0xc0000000) * 2;
1445 case 0xc0010000 ... 0xc0011fff:
1446 t0 = (16384 + msr - 0xc0010000) * 2;
1455 ret = msrpm[t1] & ((1 << param) << t0);
1458 nested_svm_unmap(msrpm, KM_USER0);
1463 static int nested_svm_exit_special(struct vcpu_svm *svm)
1465 u32 exit_code = svm->vmcb->control.exit_code;
1467 switch (exit_code) {
1470 return NESTED_EXIT_HOST;
1471 /* For now we are always handling NPFs when using them */
1474 return NESTED_EXIT_HOST;
1476 /* When we're shadowing, trap PFs */
1477 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1479 return NESTED_EXIT_HOST;
1485 return NESTED_EXIT_CONTINUE;
1489 * If this function returns true, this #vmexit was already handled
1491 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1493 u32 exit_code = svm->vmcb->control.exit_code;
1494 int vmexit = NESTED_EXIT_HOST;
1496 switch (exit_code) {
1498 vmexit = nested_svm_exit_handled_msr(svm);
1500 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1501 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1502 if (svm->nested.intercept_cr_read & cr_bits)
1503 vmexit = NESTED_EXIT_DONE;
1506 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1507 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1508 if (svm->nested.intercept_cr_write & cr_bits)
1509 vmexit = NESTED_EXIT_DONE;
1512 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1513 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1514 if (svm->nested.intercept_dr_read & dr_bits)
1515 vmexit = NESTED_EXIT_DONE;
1518 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1519 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1520 if (svm->nested.intercept_dr_write & dr_bits)
1521 vmexit = NESTED_EXIT_DONE;
1524 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1525 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1526 if (svm->nested.intercept_exceptions & excp_bits)
1527 vmexit = NESTED_EXIT_DONE;
1531 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1532 nsvm_printk("exit code: 0x%x\n", exit_code);
1533 if (svm->nested.intercept & exit_bits)
1534 vmexit = NESTED_EXIT_DONE;
1538 if (vmexit == NESTED_EXIT_DONE) {
1539 nsvm_printk("#VMEXIT reason=%04x\n", exit_code);
1540 nested_svm_vmexit(svm);
1546 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1548 struct vmcb_control_area *dst = &dst_vmcb->control;
1549 struct vmcb_control_area *from = &from_vmcb->control;
1551 dst->intercept_cr_read = from->intercept_cr_read;
1552 dst->intercept_cr_write = from->intercept_cr_write;
1553 dst->intercept_dr_read = from->intercept_dr_read;
1554 dst->intercept_dr_write = from->intercept_dr_write;
1555 dst->intercept_exceptions = from->intercept_exceptions;
1556 dst->intercept = from->intercept;
1557 dst->iopm_base_pa = from->iopm_base_pa;
1558 dst->msrpm_base_pa = from->msrpm_base_pa;
1559 dst->tsc_offset = from->tsc_offset;
1560 dst->asid = from->asid;
1561 dst->tlb_ctl = from->tlb_ctl;
1562 dst->int_ctl = from->int_ctl;
1563 dst->int_vector = from->int_vector;
1564 dst->int_state = from->int_state;
1565 dst->exit_code = from->exit_code;
1566 dst->exit_code_hi = from->exit_code_hi;
1567 dst->exit_info_1 = from->exit_info_1;
1568 dst->exit_info_2 = from->exit_info_2;
1569 dst->exit_int_info = from->exit_int_info;
1570 dst->exit_int_info_err = from->exit_int_info_err;
1571 dst->nested_ctl = from->nested_ctl;
1572 dst->event_inj = from->event_inj;
1573 dst->event_inj_err = from->event_inj_err;
1574 dst->nested_cr3 = from->nested_cr3;
1575 dst->lbr_ctl = from->lbr_ctl;
1578 static int nested_svm_vmexit(struct vcpu_svm *svm)
1580 struct vmcb *nested_vmcb;
1581 struct vmcb *hsave = svm->nested.hsave;
1582 struct vmcb *vmcb = svm->vmcb;
1584 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, KM_USER0);
1588 /* Give the current vmcb to the guest */
1591 nested_vmcb->save.es = vmcb->save.es;
1592 nested_vmcb->save.cs = vmcb->save.cs;
1593 nested_vmcb->save.ss = vmcb->save.ss;
1594 nested_vmcb->save.ds = vmcb->save.ds;
1595 nested_vmcb->save.gdtr = vmcb->save.gdtr;
1596 nested_vmcb->save.idtr = vmcb->save.idtr;
1598 nested_vmcb->save.cr3 = vmcb->save.cr3;
1599 nested_vmcb->save.cr2 = vmcb->save.cr2;
1600 nested_vmcb->save.rflags = vmcb->save.rflags;
1601 nested_vmcb->save.rip = vmcb->save.rip;
1602 nested_vmcb->save.rsp = vmcb->save.rsp;
1603 nested_vmcb->save.rax = vmcb->save.rax;
1604 nested_vmcb->save.dr7 = vmcb->save.dr7;
1605 nested_vmcb->save.dr6 = vmcb->save.dr6;
1606 nested_vmcb->save.cpl = vmcb->save.cpl;
1608 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
1609 nested_vmcb->control.int_vector = vmcb->control.int_vector;
1610 nested_vmcb->control.int_state = vmcb->control.int_state;
1611 nested_vmcb->control.exit_code = vmcb->control.exit_code;
1612 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
1613 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
1614 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
1615 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
1616 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1617 nested_vmcb->control.tlb_ctl = 0;
1618 nested_vmcb->control.event_inj = 0;
1619 nested_vmcb->control.event_inj_err = 0;
1621 /* We always set V_INTR_MASKING and remember the old value in hflags */
1622 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1623 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1625 /* Restore the original control entries */
1626 copy_vmcb_control_area(vmcb, hsave);
1628 /* Kill any pending exceptions */
1629 if (svm->vcpu.arch.exception.pending == true)
1630 nsvm_printk("WARNING: Pending Exception\n");
1632 kvm_clear_exception_queue(&svm->vcpu);
1633 kvm_clear_interrupt_queue(&svm->vcpu);
1635 /* Restore selected save entries */
1636 svm->vmcb->save.es = hsave->save.es;
1637 svm->vmcb->save.cs = hsave->save.cs;
1638 svm->vmcb->save.ss = hsave->save.ss;
1639 svm->vmcb->save.ds = hsave->save.ds;
1640 svm->vmcb->save.gdtr = hsave->save.gdtr;
1641 svm->vmcb->save.idtr = hsave->save.idtr;
1642 svm->vmcb->save.rflags = hsave->save.rflags;
1643 svm_set_efer(&svm->vcpu, hsave->save.efer);
1644 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1645 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1647 svm->vmcb->save.cr3 = hsave->save.cr3;
1648 svm->vcpu.arch.cr3 = hsave->save.cr3;
1650 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1652 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1653 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1654 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1655 svm->vmcb->save.dr7 = 0;
1656 svm->vmcb->save.cpl = 0;
1657 svm->vmcb->control.exit_int_info = 0;
1659 /* Exit nested SVM mode */
1660 svm->nested.vmcb = 0;
1662 nested_svm_unmap(nested_vmcb, KM_USER0);
1664 kvm_mmu_reset_context(&svm->vcpu);
1665 kvm_mmu_load(&svm->vcpu);
1670 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
1675 nested_msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, KM_USER0);
1679 for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1680 svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1682 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1684 nested_svm_unmap(nested_msrpm, KM_USER0);
1689 static bool nested_svm_vmrun(struct vcpu_svm *svm)
1691 struct vmcb *nested_vmcb;
1692 struct vmcb *hsave = svm->nested.hsave;
1693 struct vmcb *vmcb = svm->vmcb;
1695 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1699 /* nested_vmcb is our indicator if nested SVM is activated */
1700 svm->nested.vmcb = svm->vmcb->save.rax;
1702 /* Clear internal status */
1703 kvm_clear_exception_queue(&svm->vcpu);
1704 kvm_clear_interrupt_queue(&svm->vcpu);
1706 /* Save the old vmcb, so we don't need to pick what we save, but
1707 can restore everything when a VMEXIT occurs */
1708 hsave->save.es = vmcb->save.es;
1709 hsave->save.cs = vmcb->save.cs;
1710 hsave->save.ss = vmcb->save.ss;
1711 hsave->save.ds = vmcb->save.ds;
1712 hsave->save.gdtr = vmcb->save.gdtr;
1713 hsave->save.idtr = vmcb->save.idtr;
1714 hsave->save.efer = svm->vcpu.arch.shadow_efer;
1715 hsave->save.cr0 = svm->vcpu.arch.cr0;
1716 hsave->save.cr4 = svm->vcpu.arch.cr4;
1717 hsave->save.rflags = vmcb->save.rflags;
1718 hsave->save.rip = svm->next_rip;
1719 hsave->save.rsp = vmcb->save.rsp;
1720 hsave->save.rax = vmcb->save.rax;
1722 hsave->save.cr3 = vmcb->save.cr3;
1724 hsave->save.cr3 = svm->vcpu.arch.cr3;
1726 copy_vmcb_control_area(hsave, vmcb);
1728 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1729 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1731 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1733 /* Load the nested guest state */
1734 svm->vmcb->save.es = nested_vmcb->save.es;
1735 svm->vmcb->save.cs = nested_vmcb->save.cs;
1736 svm->vmcb->save.ss = nested_vmcb->save.ss;
1737 svm->vmcb->save.ds = nested_vmcb->save.ds;
1738 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1739 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1740 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1741 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1742 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1743 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1745 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1746 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1748 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1749 kvm_mmu_reset_context(&svm->vcpu);
1751 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
1752 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1753 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1754 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1755 /* In case we don't even reach vcpu_run, the fields are not updated */
1756 svm->vmcb->save.rax = nested_vmcb->save.rax;
1757 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1758 svm->vmcb->save.rip = nested_vmcb->save.rip;
1759 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1760 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1761 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1763 /* We don't want a nested guest to be more powerful than the guest,
1764 so all intercepts are ORed */
1765 svm->vmcb->control.intercept_cr_read |=
1766 nested_vmcb->control.intercept_cr_read;
1767 svm->vmcb->control.intercept_cr_write |=
1768 nested_vmcb->control.intercept_cr_write;
1769 svm->vmcb->control.intercept_dr_read |=
1770 nested_vmcb->control.intercept_dr_read;
1771 svm->vmcb->control.intercept_dr_write |=
1772 nested_vmcb->control.intercept_dr_write;
1773 svm->vmcb->control.intercept_exceptions |=
1774 nested_vmcb->control.intercept_exceptions;
1776 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1778 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1780 /* cache intercepts */
1781 svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
1782 svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
1783 svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
1784 svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
1785 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
1786 svm->nested.intercept = nested_vmcb->control.intercept;
1788 force_new_asid(&svm->vcpu);
1789 svm->vmcb->control.exit_int_info = nested_vmcb->control.exit_int_info;
1790 svm->vmcb->control.exit_int_info_err = nested_vmcb->control.exit_int_info_err;
1791 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1792 if (nested_vmcb->control.int_ctl & V_IRQ_MASK) {
1793 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1794 nested_vmcb->control.int_ctl);
1796 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1797 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1799 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1801 nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1802 nested_vmcb->control.exit_int_info,
1803 nested_vmcb->control.int_state);
1805 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1806 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1807 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1808 if (nested_vmcb->control.event_inj & SVM_EVTINJ_VALID)
1809 nsvm_printk("Injecting Event: 0x%x\n",
1810 nested_vmcb->control.event_inj);
1811 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1812 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1814 nested_svm_unmap(nested_vmcb, KM_USER0);
1821 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1823 to_vmcb->save.fs = from_vmcb->save.fs;
1824 to_vmcb->save.gs = from_vmcb->save.gs;
1825 to_vmcb->save.tr = from_vmcb->save.tr;
1826 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1827 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1828 to_vmcb->save.star = from_vmcb->save.star;
1829 to_vmcb->save.lstar = from_vmcb->save.lstar;
1830 to_vmcb->save.cstar = from_vmcb->save.cstar;
1831 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1832 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1833 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1834 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1837 static int vmload_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1839 struct vmcb *nested_vmcb;
1841 if (nested_svm_check_permissions(svm))
1844 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1845 skip_emulated_instruction(&svm->vcpu);
1847 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1851 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
1852 nested_svm_unmap(nested_vmcb, KM_USER0);
1857 static int vmsave_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1859 struct vmcb *nested_vmcb;
1861 if (nested_svm_check_permissions(svm))
1864 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1865 skip_emulated_instruction(&svm->vcpu);
1867 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1871 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
1872 nested_svm_unmap(nested_vmcb, KM_USER0);
1877 static int vmrun_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1879 nsvm_printk("VMrun\n");
1881 if (nested_svm_check_permissions(svm))
1884 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1885 skip_emulated_instruction(&svm->vcpu);
1887 if (!nested_svm_vmrun(svm))
1890 if (!nested_svm_vmrun_msrpm(svm))
1897 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
1898 svm->vmcb->control.exit_code_hi = 0;
1899 svm->vmcb->control.exit_info_1 = 0;
1900 svm->vmcb->control.exit_info_2 = 0;
1902 nested_svm_vmexit(svm);
1907 static int stgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1909 if (nested_svm_check_permissions(svm))
1912 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1913 skip_emulated_instruction(&svm->vcpu);
1920 static int clgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1922 if (nested_svm_check_permissions(svm))
1925 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1926 skip_emulated_instruction(&svm->vcpu);
1930 /* After a CLGI no interrupts should come */
1931 svm_clear_vintr(svm);
1932 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1937 static int invlpga_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1939 struct kvm_vcpu *vcpu = &svm->vcpu;
1940 nsvm_printk("INVLPGA\n");
1942 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
1943 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
1945 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1946 skip_emulated_instruction(&svm->vcpu);
1950 static int invalid_op_interception(struct vcpu_svm *svm,
1951 struct kvm_run *kvm_run)
1953 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1957 static int task_switch_interception(struct vcpu_svm *svm,
1958 struct kvm_run *kvm_run)
1962 int int_type = svm->vmcb->control.exit_int_info &
1963 SVM_EXITINTINFO_TYPE_MASK;
1964 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
1966 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
1968 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
1970 tss_selector = (u16)svm->vmcb->control.exit_info_1;
1972 if (svm->vmcb->control.exit_info_2 &
1973 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1974 reason = TASK_SWITCH_IRET;
1975 else if (svm->vmcb->control.exit_info_2 &
1976 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1977 reason = TASK_SWITCH_JMP;
1979 reason = TASK_SWITCH_GATE;
1981 reason = TASK_SWITCH_CALL;
1983 if (reason == TASK_SWITCH_GATE) {
1985 case SVM_EXITINTINFO_TYPE_NMI:
1986 svm->vcpu.arch.nmi_injected = false;
1988 case SVM_EXITINTINFO_TYPE_EXEPT:
1989 kvm_clear_exception_queue(&svm->vcpu);
1991 case SVM_EXITINTINFO_TYPE_INTR:
1992 kvm_clear_interrupt_queue(&svm->vcpu);
1999 if (reason != TASK_SWITCH_GATE ||
2000 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2001 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2002 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2003 skip_emulated_instruction(&svm->vcpu);
2005 return kvm_task_switch(&svm->vcpu, tss_selector, reason);
2008 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2010 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2011 kvm_emulate_cpuid(&svm->vcpu);
2015 static int iret_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2017 ++svm->vcpu.stat.nmi_window_exits;
2018 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2019 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2023 static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2025 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
2026 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2030 static int emulate_on_interception(struct vcpu_svm *svm,
2031 struct kvm_run *kvm_run)
2033 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
2034 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2038 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2040 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2041 /* instruction emulation calls kvm_set_cr8() */
2042 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
2043 if (irqchip_in_kernel(svm->vcpu.kvm)) {
2044 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2047 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2049 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2053 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2055 struct vcpu_svm *svm = to_svm(vcpu);
2058 case MSR_IA32_TSC: {
2062 *data = svm->vmcb->control.tsc_offset + tsc;
2066 *data = svm->vmcb->save.star;
2068 #ifdef CONFIG_X86_64
2070 *data = svm->vmcb->save.lstar;
2073 *data = svm->vmcb->save.cstar;
2075 case MSR_KERNEL_GS_BASE:
2076 *data = svm->vmcb->save.kernel_gs_base;
2078 case MSR_SYSCALL_MASK:
2079 *data = svm->vmcb->save.sfmask;
2082 case MSR_IA32_SYSENTER_CS:
2083 *data = svm->vmcb->save.sysenter_cs;
2085 case MSR_IA32_SYSENTER_EIP:
2086 *data = svm->sysenter_eip;
2088 case MSR_IA32_SYSENTER_ESP:
2089 *data = svm->sysenter_esp;
2091 /* Nobody will change the following 5 values in the VMCB so
2092 we can safely return them on rdmsr. They will always be 0
2093 until LBRV is implemented. */
2094 case MSR_IA32_DEBUGCTLMSR:
2095 *data = svm->vmcb->save.dbgctl;
2097 case MSR_IA32_LASTBRANCHFROMIP:
2098 *data = svm->vmcb->save.br_from;
2100 case MSR_IA32_LASTBRANCHTOIP:
2101 *data = svm->vmcb->save.br_to;
2103 case MSR_IA32_LASTINTFROMIP:
2104 *data = svm->vmcb->save.last_excp_from;
2106 case MSR_IA32_LASTINTTOIP:
2107 *data = svm->vmcb->save.last_excp_to;
2109 case MSR_VM_HSAVE_PA:
2110 *data = svm->nested.hsave_msr;
2115 case MSR_IA32_UCODE_REV:
2119 return kvm_get_msr_common(vcpu, ecx, data);
2124 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2126 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2129 if (svm_get_msr(&svm->vcpu, ecx, &data))
2130 kvm_inject_gp(&svm->vcpu, 0);
2132 trace_kvm_msr_read(ecx, data);
2134 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2135 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2136 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2137 skip_emulated_instruction(&svm->vcpu);
2142 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2144 struct vcpu_svm *svm = to_svm(vcpu);
2147 case MSR_IA32_TSC: {
2151 svm->vmcb->control.tsc_offset = data - tsc;
2155 svm->vmcb->save.star = data;
2157 #ifdef CONFIG_X86_64
2159 svm->vmcb->save.lstar = data;
2162 svm->vmcb->save.cstar = data;
2164 case MSR_KERNEL_GS_BASE:
2165 svm->vmcb->save.kernel_gs_base = data;
2167 case MSR_SYSCALL_MASK:
2168 svm->vmcb->save.sfmask = data;
2171 case MSR_IA32_SYSENTER_CS:
2172 svm->vmcb->save.sysenter_cs = data;
2174 case MSR_IA32_SYSENTER_EIP:
2175 svm->sysenter_eip = data;
2176 svm->vmcb->save.sysenter_eip = data;
2178 case MSR_IA32_SYSENTER_ESP:
2179 svm->sysenter_esp = data;
2180 svm->vmcb->save.sysenter_esp = data;
2182 case MSR_IA32_DEBUGCTLMSR:
2183 if (!svm_has(SVM_FEATURE_LBRV)) {
2184 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2188 if (data & DEBUGCTL_RESERVED_BITS)
2191 svm->vmcb->save.dbgctl = data;
2192 if (data & (1ULL<<0))
2193 svm_enable_lbrv(svm);
2195 svm_disable_lbrv(svm);
2197 case MSR_VM_HSAVE_PA:
2198 svm->nested.hsave_msr = data;
2202 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2205 return kvm_set_msr_common(vcpu, ecx, data);
2210 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2212 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2213 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2214 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2216 trace_kvm_msr_write(ecx, data);
2218 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2219 if (svm_set_msr(&svm->vcpu, ecx, data))
2220 kvm_inject_gp(&svm->vcpu, 0);
2222 skip_emulated_instruction(&svm->vcpu);
2226 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2228 if (svm->vmcb->control.exit_info_1)
2229 return wrmsr_interception(svm, kvm_run);
2231 return rdmsr_interception(svm, kvm_run);
2234 static int interrupt_window_interception(struct vcpu_svm *svm,
2235 struct kvm_run *kvm_run)
2237 svm_clear_vintr(svm);
2238 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2240 * If the user space waits to inject interrupts, exit as soon as
2243 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2244 kvm_run->request_interrupt_window &&
2245 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2246 ++svm->vcpu.stat.irq_window_exits;
2247 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2254 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
2255 struct kvm_run *kvm_run) = {
2256 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2257 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2258 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2259 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2261 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
2262 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2263 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2264 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2265 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2266 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2267 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2268 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2269 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2270 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2271 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2272 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2273 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2274 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2275 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2276 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2277 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2278 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2279 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2280 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2281 [SVM_EXIT_INTR] = intr_interception,
2282 [SVM_EXIT_NMI] = nmi_interception,
2283 [SVM_EXIT_SMI] = nop_on_interception,
2284 [SVM_EXIT_INIT] = nop_on_interception,
2285 [SVM_EXIT_VINTR] = interrupt_window_interception,
2286 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2287 [SVM_EXIT_CPUID] = cpuid_interception,
2288 [SVM_EXIT_IRET] = iret_interception,
2289 [SVM_EXIT_INVD] = emulate_on_interception,
2290 [SVM_EXIT_HLT] = halt_interception,
2291 [SVM_EXIT_INVLPG] = invlpg_interception,
2292 [SVM_EXIT_INVLPGA] = invlpga_interception,
2293 [SVM_EXIT_IOIO] = io_interception,
2294 [SVM_EXIT_MSR] = msr_interception,
2295 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2296 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2297 [SVM_EXIT_VMRUN] = vmrun_interception,
2298 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2299 [SVM_EXIT_VMLOAD] = vmload_interception,
2300 [SVM_EXIT_VMSAVE] = vmsave_interception,
2301 [SVM_EXIT_STGI] = stgi_interception,
2302 [SVM_EXIT_CLGI] = clgi_interception,
2303 [SVM_EXIT_SKINIT] = invalid_op_interception,
2304 [SVM_EXIT_WBINVD] = emulate_on_interception,
2305 [SVM_EXIT_MONITOR] = invalid_op_interception,
2306 [SVM_EXIT_MWAIT] = invalid_op_interception,
2307 [SVM_EXIT_NPF] = pf_interception,
2310 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2312 struct vcpu_svm *svm = to_svm(vcpu);
2313 u32 exit_code = svm->vmcb->control.exit_code;
2315 trace_kvm_exit(exit_code, svm->vmcb->save.rip);
2317 if (is_nested(svm)) {
2320 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2321 exit_code, svm->vmcb->control.exit_info_1,
2322 svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
2324 vmexit = nested_svm_exit_special(svm);
2326 if (vmexit == NESTED_EXIT_CONTINUE)
2327 vmexit = nested_svm_exit_handled(svm);
2329 if (vmexit == NESTED_EXIT_DONE)
2333 svm_complete_interrupts(svm);
2337 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
2338 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
2341 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2342 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2344 kvm_mmu_reset_context(vcpu);
2350 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2351 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2352 kvm_run->fail_entry.hardware_entry_failure_reason
2353 = svm->vmcb->control.exit_code;
2357 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2358 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2359 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2360 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2362 __func__, svm->vmcb->control.exit_int_info,
2365 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2366 || !svm_exit_handlers[exit_code]) {
2367 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2368 kvm_run->hw.hardware_exit_reason = exit_code;
2372 return svm_exit_handlers[exit_code](svm, kvm_run);
2375 static void reload_tss(struct kvm_vcpu *vcpu)
2377 int cpu = raw_smp_processor_id();
2379 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2380 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
2384 static void pre_svm_run(struct vcpu_svm *svm)
2386 int cpu = raw_smp_processor_id();
2388 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2390 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2391 /* FIXME: handle wraparound of asid_generation */
2392 if (svm->asid_generation != sd->asid_generation)
2396 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2398 struct vcpu_svm *svm = to_svm(vcpu);
2400 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2401 vcpu->arch.hflags |= HF_NMI_MASK;
2402 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2403 ++vcpu->stat.nmi_injections;
2406 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2408 struct vmcb_control_area *control;
2410 trace_kvm_inj_virq(irq);
2412 ++svm->vcpu.stat.irq_injections;
2413 control = &svm->vmcb->control;
2414 control->int_vector = irq;
2415 control->int_ctl &= ~V_INTR_PRIO_MASK;
2416 control->int_ctl |= V_IRQ_MASK |
2417 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2420 static void svm_set_irq(struct kvm_vcpu *vcpu)
2422 struct vcpu_svm *svm = to_svm(vcpu);
2424 BUG_ON(!(gif_set(svm)));
2426 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2427 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2430 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2432 struct vcpu_svm *svm = to_svm(vcpu);
2438 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2441 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2443 struct vcpu_svm *svm = to_svm(vcpu);
2444 struct vmcb *vmcb = svm->vmcb;
2445 return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2446 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2449 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2451 struct vcpu_svm *svm = to_svm(vcpu);
2452 struct vmcb *vmcb = svm->vmcb;
2453 return (vmcb->save.rflags & X86_EFLAGS_IF) &&
2454 !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2456 !(is_nested(svm) && (svm->vcpu.arch.hflags & HF_VINTR_MASK));
2459 static void enable_irq_window(struct kvm_vcpu *vcpu)
2461 struct vcpu_svm *svm = to_svm(vcpu);
2462 nsvm_printk("Trying to open IRQ window\n");
2464 nested_svm_intr(svm);
2466 /* In case GIF=0 we can't rely on the CPU to tell us when
2467 * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
2468 * The next time we get that intercept, this function will be
2469 * called again though and we'll get the vintr intercept. */
2472 svm_inject_irq(svm, 0x0);
2476 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2478 struct vcpu_svm *svm = to_svm(vcpu);
2480 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2482 return; /* IRET will cause a vm exit */
2484 /* Something prevents NMI from been injected. Single step over
2485 possible problem (IRET or exception injection or interrupt
2487 vcpu->arch.singlestep = true;
2488 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2489 update_db_intercept(vcpu);
2492 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2497 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2499 force_new_asid(vcpu);
2502 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2506 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2508 struct vcpu_svm *svm = to_svm(vcpu);
2510 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2511 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2512 kvm_set_cr8(vcpu, cr8);
2516 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2518 struct vcpu_svm *svm = to_svm(vcpu);
2521 cr8 = kvm_get_cr8(vcpu);
2522 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2523 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2526 static void svm_complete_interrupts(struct vcpu_svm *svm)
2530 u32 exitintinfo = svm->vmcb->control.exit_int_info;
2532 if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2533 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2535 svm->vcpu.arch.nmi_injected = false;
2536 kvm_clear_exception_queue(&svm->vcpu);
2537 kvm_clear_interrupt_queue(&svm->vcpu);
2539 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2542 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2543 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2546 case SVM_EXITINTINFO_TYPE_NMI:
2547 svm->vcpu.arch.nmi_injected = true;
2549 case SVM_EXITINTINFO_TYPE_EXEPT:
2550 /* In case of software exception do not reinject an exception
2551 vector, but re-execute and instruction instead */
2554 if (kvm_exception_is_soft(vector))
2556 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2557 u32 err = svm->vmcb->control.exit_int_info_err;
2558 kvm_queue_exception_e(&svm->vcpu, vector, err);
2561 kvm_queue_exception(&svm->vcpu, vector);
2563 case SVM_EXITINTINFO_TYPE_INTR:
2564 kvm_queue_interrupt(&svm->vcpu, vector, false);
2571 #ifdef CONFIG_X86_64
2577 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2579 struct vcpu_svm *svm = to_svm(vcpu);
2584 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2585 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2586 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2590 sync_lapic_to_cr8(vcpu);
2592 save_host_msrs(vcpu);
2593 fs_selector = kvm_read_fs();
2594 gs_selector = kvm_read_gs();
2595 ldt_selector = kvm_read_ldt();
2596 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2597 /* required for live migration with NPT */
2599 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2606 "push %%"R"bp; \n\t"
2607 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2608 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2609 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2610 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2611 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2612 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2613 #ifdef CONFIG_X86_64
2614 "mov %c[r8](%[svm]), %%r8 \n\t"
2615 "mov %c[r9](%[svm]), %%r9 \n\t"
2616 "mov %c[r10](%[svm]), %%r10 \n\t"
2617 "mov %c[r11](%[svm]), %%r11 \n\t"
2618 "mov %c[r12](%[svm]), %%r12 \n\t"
2619 "mov %c[r13](%[svm]), %%r13 \n\t"
2620 "mov %c[r14](%[svm]), %%r14 \n\t"
2621 "mov %c[r15](%[svm]), %%r15 \n\t"
2624 /* Enter guest mode */
2626 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2627 __ex(SVM_VMLOAD) "\n\t"
2628 __ex(SVM_VMRUN) "\n\t"
2629 __ex(SVM_VMSAVE) "\n\t"
2632 /* Save guest registers, load host registers */
2633 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2634 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2635 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2636 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2637 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2638 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2639 #ifdef CONFIG_X86_64
2640 "mov %%r8, %c[r8](%[svm]) \n\t"
2641 "mov %%r9, %c[r9](%[svm]) \n\t"
2642 "mov %%r10, %c[r10](%[svm]) \n\t"
2643 "mov %%r11, %c[r11](%[svm]) \n\t"
2644 "mov %%r12, %c[r12](%[svm]) \n\t"
2645 "mov %%r13, %c[r13](%[svm]) \n\t"
2646 "mov %%r14, %c[r14](%[svm]) \n\t"
2647 "mov %%r15, %c[r15](%[svm]) \n\t"
2652 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2653 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2654 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2655 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2656 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2657 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2658 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2659 #ifdef CONFIG_X86_64
2660 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2661 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2662 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2663 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2664 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2665 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2666 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2667 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2670 , R"bx", R"cx", R"dx", R"si", R"di"
2671 #ifdef CONFIG_X86_64
2672 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2676 vcpu->arch.cr2 = svm->vmcb->save.cr2;
2677 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2678 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2679 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2681 kvm_load_fs(fs_selector);
2682 kvm_load_gs(gs_selector);
2683 kvm_load_ldt(ldt_selector);
2684 load_host_msrs(vcpu);
2688 local_irq_disable();
2692 sync_cr8_to_lapic(vcpu);
2697 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
2698 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
2704 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2706 struct vcpu_svm *svm = to_svm(vcpu);
2709 svm->vmcb->control.nested_cr3 = root;
2710 force_new_asid(vcpu);
2714 svm->vmcb->save.cr3 = root;
2715 force_new_asid(vcpu);
2717 if (vcpu->fpu_active) {
2718 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
2719 svm->vmcb->save.cr0 |= X86_CR0_TS;
2720 vcpu->fpu_active = 0;
2724 static int is_disabled(void)
2728 rdmsrl(MSR_VM_CR, vm_cr);
2729 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2736 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2739 * Patch in the VMMCALL instruction:
2741 hypercall[0] = 0x0f;
2742 hypercall[1] = 0x01;
2743 hypercall[2] = 0xd9;
2746 static void svm_check_processor_compat(void *rtn)
2751 static bool svm_cpu_has_accelerated_tpr(void)
2756 static int get_npt_level(void)
2758 #ifdef CONFIG_X86_64
2759 return PT64_ROOT_LEVEL;
2761 return PT32E_ROOT_LEVEL;
2765 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
2770 static const struct trace_print_flags svm_exit_reasons_str[] = {
2771 { SVM_EXIT_READ_CR0, "read_cr0" },
2772 { SVM_EXIT_READ_CR3, "read_cr3" },
2773 { SVM_EXIT_READ_CR4, "read_cr4" },
2774 { SVM_EXIT_READ_CR8, "read_cr8" },
2775 { SVM_EXIT_WRITE_CR0, "write_cr0" },
2776 { SVM_EXIT_WRITE_CR3, "write_cr3" },
2777 { SVM_EXIT_WRITE_CR4, "write_cr4" },
2778 { SVM_EXIT_WRITE_CR8, "write_cr8" },
2779 { SVM_EXIT_READ_DR0, "read_dr0" },
2780 { SVM_EXIT_READ_DR1, "read_dr1" },
2781 { SVM_EXIT_READ_DR2, "read_dr2" },
2782 { SVM_EXIT_READ_DR3, "read_dr3" },
2783 { SVM_EXIT_WRITE_DR0, "write_dr0" },
2784 { SVM_EXIT_WRITE_DR1, "write_dr1" },
2785 { SVM_EXIT_WRITE_DR2, "write_dr2" },
2786 { SVM_EXIT_WRITE_DR3, "write_dr3" },
2787 { SVM_EXIT_WRITE_DR5, "write_dr5" },
2788 { SVM_EXIT_WRITE_DR7, "write_dr7" },
2789 { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
2790 { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
2791 { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
2792 { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
2793 { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
2794 { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
2795 { SVM_EXIT_INTR, "interrupt" },
2796 { SVM_EXIT_NMI, "nmi" },
2797 { SVM_EXIT_SMI, "smi" },
2798 { SVM_EXIT_INIT, "init" },
2799 { SVM_EXIT_VINTR, "vintr" },
2800 { SVM_EXIT_CPUID, "cpuid" },
2801 { SVM_EXIT_INVD, "invd" },
2802 { SVM_EXIT_HLT, "hlt" },
2803 { SVM_EXIT_INVLPG, "invlpg" },
2804 { SVM_EXIT_INVLPGA, "invlpga" },
2805 { SVM_EXIT_IOIO, "io" },
2806 { SVM_EXIT_MSR, "msr" },
2807 { SVM_EXIT_TASK_SWITCH, "task_switch" },
2808 { SVM_EXIT_SHUTDOWN, "shutdown" },
2809 { SVM_EXIT_VMRUN, "vmrun" },
2810 { SVM_EXIT_VMMCALL, "hypercall" },
2811 { SVM_EXIT_VMLOAD, "vmload" },
2812 { SVM_EXIT_VMSAVE, "vmsave" },
2813 { SVM_EXIT_STGI, "stgi" },
2814 { SVM_EXIT_CLGI, "clgi" },
2815 { SVM_EXIT_SKINIT, "skinit" },
2816 { SVM_EXIT_WBINVD, "wbinvd" },
2817 { SVM_EXIT_MONITOR, "monitor" },
2818 { SVM_EXIT_MWAIT, "mwait" },
2819 { SVM_EXIT_NPF, "npf" },
2823 static bool svm_gb_page_enable(void)
2828 static struct kvm_x86_ops svm_x86_ops = {
2829 .cpu_has_kvm_support = has_svm,
2830 .disabled_by_bios = is_disabled,
2831 .hardware_setup = svm_hardware_setup,
2832 .hardware_unsetup = svm_hardware_unsetup,
2833 .check_processor_compatibility = svm_check_processor_compat,
2834 .hardware_enable = svm_hardware_enable,
2835 .hardware_disable = svm_hardware_disable,
2836 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2838 .vcpu_create = svm_create_vcpu,
2839 .vcpu_free = svm_free_vcpu,
2840 .vcpu_reset = svm_vcpu_reset,
2842 .prepare_guest_switch = svm_prepare_guest_switch,
2843 .vcpu_load = svm_vcpu_load,
2844 .vcpu_put = svm_vcpu_put,
2846 .set_guest_debug = svm_guest_debug,
2847 .get_msr = svm_get_msr,
2848 .set_msr = svm_set_msr,
2849 .get_segment_base = svm_get_segment_base,
2850 .get_segment = svm_get_segment,
2851 .set_segment = svm_set_segment,
2852 .get_cpl = svm_get_cpl,
2853 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2854 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2855 .set_cr0 = svm_set_cr0,
2856 .set_cr3 = svm_set_cr3,
2857 .set_cr4 = svm_set_cr4,
2858 .set_efer = svm_set_efer,
2859 .get_idt = svm_get_idt,
2860 .set_idt = svm_set_idt,
2861 .get_gdt = svm_get_gdt,
2862 .set_gdt = svm_set_gdt,
2863 .get_dr = svm_get_dr,
2864 .set_dr = svm_set_dr,
2865 .cache_reg = svm_cache_reg,
2866 .get_rflags = svm_get_rflags,
2867 .set_rflags = svm_set_rflags,
2869 .tlb_flush = svm_flush_tlb,
2871 .run = svm_vcpu_run,
2872 .handle_exit = handle_exit,
2873 .skip_emulated_instruction = skip_emulated_instruction,
2874 .set_interrupt_shadow = svm_set_interrupt_shadow,
2875 .get_interrupt_shadow = svm_get_interrupt_shadow,
2876 .patch_hypercall = svm_patch_hypercall,
2877 .set_irq = svm_set_irq,
2878 .set_nmi = svm_inject_nmi,
2879 .queue_exception = svm_queue_exception,
2880 .interrupt_allowed = svm_interrupt_allowed,
2881 .nmi_allowed = svm_nmi_allowed,
2882 .enable_nmi_window = enable_nmi_window,
2883 .enable_irq_window = enable_irq_window,
2884 .update_cr8_intercept = update_cr8_intercept,
2886 .set_tss_addr = svm_set_tss_addr,
2887 .get_tdp_level = get_npt_level,
2888 .get_mt_mask = svm_get_mt_mask,
2890 .exit_reasons_str = svm_exit_reasons_str,
2891 .gb_page_enable = svm_gb_page_enable,
2894 static int __init svm_init(void)
2896 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
2900 static void __exit svm_exit(void)
2905 module_init(svm_init)
2906 module_exit(svm_exit)