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>
29 #include <linux/slab.h>
33 #include <asm/virtext.h>
36 #define __ex(x) __kvm_handle_fault_on_reboot(x)
38 MODULE_AUTHOR("Qumranet");
39 MODULE_LICENSE("GPL");
41 #define IOPM_ALLOC_ORDER 2
42 #define MSRPM_ALLOC_ORDER 1
44 #define SEG_TYPE_LDT 2
45 #define SEG_TYPE_BUSY_TSS16 3
47 #define SVM_FEATURE_NPT (1 << 0)
48 #define SVM_FEATURE_LBRV (1 << 1)
49 #define SVM_FEATURE_SVML (1 << 2)
50 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
52 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
53 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
54 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
56 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
58 static const u32 host_save_user_msrs[] = {
60 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
63 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
66 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
75 /* These are the merged vectors */
78 /* gpa pointers to the real vectors */
81 /* A VMEXIT is required but not yet emulated */
84 /* cache for intercepts of the guest */
85 u16 intercept_cr_read;
86 u16 intercept_cr_write;
87 u16 intercept_dr_read;
88 u16 intercept_dr_write;
89 u32 intercept_exceptions;
97 unsigned long vmcb_pa;
98 struct svm_cpu_data *svm_data;
99 uint64_t asid_generation;
100 uint64_t sysenter_esp;
101 uint64_t sysenter_eip;
105 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
110 struct nested_state nested;
115 /* enable NPT for AMD64 and X86 with PAE */
116 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
117 static bool npt_enabled = true;
119 static bool npt_enabled = false;
123 module_param(npt, int, S_IRUGO);
125 static int nested = 1;
126 module_param(nested, int, S_IRUGO);
128 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
129 static void svm_complete_interrupts(struct vcpu_svm *svm);
131 static int nested_svm_exit_handled(struct vcpu_svm *svm);
132 static int nested_svm_vmexit(struct vcpu_svm *svm);
133 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
134 bool has_error_code, u32 error_code);
136 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
138 return container_of(vcpu, struct vcpu_svm, vcpu);
141 static inline bool is_nested(struct vcpu_svm *svm)
143 return svm->nested.vmcb;
146 static inline void enable_gif(struct vcpu_svm *svm)
148 svm->vcpu.arch.hflags |= HF_GIF_MASK;
151 static inline void disable_gif(struct vcpu_svm *svm)
153 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
156 static inline bool gif_set(struct vcpu_svm *svm)
158 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
161 static unsigned long iopm_base;
163 struct kvm_ldttss_desc {
166 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
167 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
170 } __attribute__((packed));
172 struct svm_cpu_data {
178 struct kvm_ldttss_desc *tss_desc;
180 struct page *save_area;
183 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
184 static uint32_t svm_features;
186 struct svm_init_data {
191 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
193 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
194 #define MSRS_RANGE_SIZE 2048
195 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
197 #define MAX_INST_SIZE 15
199 static inline u32 svm_has(u32 feat)
201 return svm_features & feat;
204 static inline void clgi(void)
206 asm volatile (__ex(SVM_CLGI));
209 static inline void stgi(void)
211 asm volatile (__ex(SVM_STGI));
214 static inline void invlpga(unsigned long addr, u32 asid)
216 asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
219 static inline void force_new_asid(struct kvm_vcpu *vcpu)
221 to_svm(vcpu)->asid_generation--;
224 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
226 force_new_asid(vcpu);
229 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
231 if (!npt_enabled && !(efer & EFER_LMA))
234 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
235 vcpu->arch.efer = efer;
238 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
239 bool has_error_code, u32 error_code)
241 struct vcpu_svm *svm = to_svm(vcpu);
243 /* If we are within a nested VM we'd better #VMEXIT and let the
244 guest handle the exception */
245 if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
248 svm->vmcb->control.event_inj = nr
250 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
251 | SVM_EVTINJ_TYPE_EXEPT;
252 svm->vmcb->control.event_inj_err = error_code;
255 static int is_external_interrupt(u32 info)
257 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
258 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
261 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
263 struct vcpu_svm *svm = to_svm(vcpu);
266 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
267 ret |= X86_SHADOW_INT_STI | X86_SHADOW_INT_MOV_SS;
271 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
273 struct vcpu_svm *svm = to_svm(vcpu);
276 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
278 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
282 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
284 struct vcpu_svm *svm = to_svm(vcpu);
286 if (!svm->next_rip) {
287 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
289 printk(KERN_DEBUG "%s: NOP\n", __func__);
292 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
293 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
294 __func__, kvm_rip_read(vcpu), svm->next_rip);
296 kvm_rip_write(vcpu, svm->next_rip);
297 svm_set_interrupt_shadow(vcpu, 0);
300 static int has_svm(void)
304 if (!cpu_has_svm(&msg)) {
305 printk(KERN_INFO "has_svm: %s\n", msg);
312 static void svm_hardware_disable(void *garbage)
317 static int svm_hardware_enable(void *garbage)
320 struct svm_cpu_data *sd;
322 struct descriptor_table gdt_descr;
323 struct desc_struct *gdt;
324 int me = raw_smp_processor_id();
326 rdmsrl(MSR_EFER, efer);
327 if (efer & EFER_SVME)
331 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
335 sd = per_cpu(svm_data, me);
338 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
343 sd->asid_generation = 1;
344 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
345 sd->next_asid = sd->max_asid + 1;
347 kvm_get_gdt(&gdt_descr);
348 gdt = (struct desc_struct *)gdt_descr.base;
349 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
351 wrmsrl(MSR_EFER, efer | EFER_SVME);
353 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
358 static void svm_cpu_uninit(int cpu)
360 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
365 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
366 __free_page(sd->save_area);
370 static int svm_cpu_init(int cpu)
372 struct svm_cpu_data *sd;
375 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
379 sd->save_area = alloc_page(GFP_KERNEL);
384 per_cpu(svm_data, cpu) = sd;
394 static void set_msr_interception(u32 *msrpm, unsigned msr,
399 for (i = 0; i < NUM_MSR_MAPS; i++) {
400 if (msr >= msrpm_ranges[i] &&
401 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
402 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
403 msrpm_ranges[i]) * 2;
405 u32 *base = msrpm + (msr_offset / 32);
406 u32 msr_shift = msr_offset % 32;
407 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
408 *base = (*base & ~(0x3 << msr_shift)) |
416 static void svm_vcpu_init_msrpm(u32 *msrpm)
418 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
421 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
422 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
423 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
424 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
425 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
426 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
428 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
429 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
432 static void svm_enable_lbrv(struct vcpu_svm *svm)
434 u32 *msrpm = svm->msrpm;
436 svm->vmcb->control.lbr_ctl = 1;
437 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
438 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
439 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
440 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
443 static void svm_disable_lbrv(struct vcpu_svm *svm)
445 u32 *msrpm = svm->msrpm;
447 svm->vmcb->control.lbr_ctl = 0;
448 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
449 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
450 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
451 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
454 static __init int svm_hardware_setup(void)
457 struct page *iopm_pages;
461 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
466 iopm_va = page_address(iopm_pages);
467 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
468 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
470 if (boot_cpu_has(X86_FEATURE_NX))
471 kvm_enable_efer_bits(EFER_NX);
473 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
474 kvm_enable_efer_bits(EFER_FFXSR);
477 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
478 kvm_enable_efer_bits(EFER_SVME);
481 for_each_possible_cpu(cpu) {
482 r = svm_cpu_init(cpu);
487 svm_features = cpuid_edx(SVM_CPUID_FUNC);
489 if (!svm_has(SVM_FEATURE_NPT))
492 if (npt_enabled && !npt) {
493 printk(KERN_INFO "kvm: Nested Paging disabled\n");
498 printk(KERN_INFO "kvm: Nested Paging enabled\n");
506 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
511 static __exit void svm_hardware_unsetup(void)
515 for_each_possible_cpu(cpu)
518 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
522 static void init_seg(struct vmcb_seg *seg)
525 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
526 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
531 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
534 seg->attrib = SVM_SELECTOR_P_MASK | type;
539 static void init_vmcb(struct vcpu_svm *svm)
541 struct vmcb_control_area *control = &svm->vmcb->control;
542 struct vmcb_save_area *save = &svm->vmcb->save;
544 svm->vcpu.fpu_active = 1;
546 control->intercept_cr_read = INTERCEPT_CR0_MASK |
550 control->intercept_cr_write = INTERCEPT_CR0_MASK |
555 control->intercept_dr_read = INTERCEPT_DR0_MASK |
564 control->intercept_dr_write = INTERCEPT_DR0_MASK |
573 control->intercept_exceptions = (1 << PF_VECTOR) |
578 control->intercept = (1ULL << INTERCEPT_INTR) |
579 (1ULL << INTERCEPT_NMI) |
580 (1ULL << INTERCEPT_SMI) |
581 (1ULL << INTERCEPT_SELECTIVE_CR0) |
582 (1ULL << INTERCEPT_CPUID) |
583 (1ULL << INTERCEPT_INVD) |
584 (1ULL << INTERCEPT_HLT) |
585 (1ULL << INTERCEPT_INVLPG) |
586 (1ULL << INTERCEPT_INVLPGA) |
587 (1ULL << INTERCEPT_IOIO_PROT) |
588 (1ULL << INTERCEPT_MSR_PROT) |
589 (1ULL << INTERCEPT_TASK_SWITCH) |
590 (1ULL << INTERCEPT_SHUTDOWN) |
591 (1ULL << INTERCEPT_VMRUN) |
592 (1ULL << INTERCEPT_VMMCALL) |
593 (1ULL << INTERCEPT_VMLOAD) |
594 (1ULL << INTERCEPT_VMSAVE) |
595 (1ULL << INTERCEPT_STGI) |
596 (1ULL << INTERCEPT_CLGI) |
597 (1ULL << INTERCEPT_SKINIT) |
598 (1ULL << INTERCEPT_WBINVD) |
599 (1ULL << INTERCEPT_MONITOR) |
600 (1ULL << INTERCEPT_MWAIT);
602 control->iopm_base_pa = iopm_base;
603 control->msrpm_base_pa = __pa(svm->msrpm);
604 control->tsc_offset = 0;
605 control->int_ctl = V_INTR_MASKING_MASK;
613 save->cs.selector = 0xf000;
614 /* Executable/Readable Code Segment */
615 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
616 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
617 save->cs.limit = 0xffff;
619 * cs.base should really be 0xffff0000, but vmx can't handle that, so
620 * be consistent with it.
622 * Replace when we have real mode working for vmx.
624 save->cs.base = 0xf0000;
626 save->gdtr.limit = 0xffff;
627 save->idtr.limit = 0xffff;
629 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
630 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
632 save->efer = EFER_SVME;
633 save->dr6 = 0xffff0ff0;
636 save->rip = 0x0000fff0;
637 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
639 /* This is the guest-visible cr0 value.
640 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
642 svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
643 kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0);
645 save->cr4 = X86_CR4_PAE;
649 /* Setup VMCB for Nested Paging */
650 control->nested_ctl = 1;
651 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
652 (1ULL << INTERCEPT_INVLPG));
653 control->intercept_exceptions &= ~(1 << PF_VECTOR);
654 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
655 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
656 save->g_pat = 0x0007040600070406ULL;
660 force_new_asid(&svm->vcpu);
662 svm->nested.vmcb = 0;
663 svm->vcpu.arch.hflags = 0;
665 if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
666 control->pause_filter_count = 3000;
667 control->intercept |= (1ULL << INTERCEPT_PAUSE);
673 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
675 struct vcpu_svm *svm = to_svm(vcpu);
679 if (!kvm_vcpu_is_bsp(vcpu)) {
680 kvm_rip_write(vcpu, 0);
681 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
682 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
684 vcpu->arch.regs_avail = ~0;
685 vcpu->arch.regs_dirty = ~0;
690 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
692 struct vcpu_svm *svm;
694 struct page *msrpm_pages;
695 struct page *hsave_page;
696 struct page *nested_msrpm_pages;
699 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
705 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
709 page = alloc_page(GFP_KERNEL);
716 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
720 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
721 if (!nested_msrpm_pages)
724 svm->msrpm = page_address(msrpm_pages);
725 svm_vcpu_init_msrpm(svm->msrpm);
727 hsave_page = alloc_page(GFP_KERNEL);
730 svm->nested.hsave = page_address(hsave_page);
732 svm->nested.msrpm = page_address(nested_msrpm_pages);
734 svm->vmcb = page_address(page);
735 clear_page(svm->vmcb);
736 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
737 svm->asid_generation = 0;
741 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
742 if (kvm_vcpu_is_bsp(&svm->vcpu))
743 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
748 kvm_vcpu_uninit(&svm->vcpu);
750 kmem_cache_free(kvm_vcpu_cache, svm);
755 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
757 struct vcpu_svm *svm = to_svm(vcpu);
759 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
760 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
761 __free_page(virt_to_page(svm->nested.hsave));
762 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
763 kvm_vcpu_uninit(vcpu);
764 kmem_cache_free(kvm_vcpu_cache, svm);
767 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
769 struct vcpu_svm *svm = to_svm(vcpu);
772 if (unlikely(cpu != vcpu->cpu)) {
775 if (check_tsc_unstable()) {
777 * Make sure that the guest sees a monotonically
780 delta = vcpu->arch.host_tsc - native_read_tsc();
781 svm->vmcb->control.tsc_offset += delta;
783 svm->nested.hsave->control.tsc_offset += delta;
786 kvm_migrate_timers(vcpu);
787 svm->asid_generation = 0;
790 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
791 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
794 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
796 struct vcpu_svm *svm = to_svm(vcpu);
799 ++vcpu->stat.host_state_reload;
800 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
801 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
803 vcpu->arch.host_tsc = native_read_tsc();
806 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
808 return to_svm(vcpu)->vmcb->save.rflags;
811 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
813 to_svm(vcpu)->vmcb->save.rflags = rflags;
816 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
819 case VCPU_EXREG_PDPTR:
820 BUG_ON(!npt_enabled);
821 load_pdptrs(vcpu, vcpu->arch.cr3);
828 static void svm_set_vintr(struct vcpu_svm *svm)
830 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
833 static void svm_clear_vintr(struct vcpu_svm *svm)
835 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
838 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
840 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
843 case VCPU_SREG_CS: return &save->cs;
844 case VCPU_SREG_DS: return &save->ds;
845 case VCPU_SREG_ES: return &save->es;
846 case VCPU_SREG_FS: return &save->fs;
847 case VCPU_SREG_GS: return &save->gs;
848 case VCPU_SREG_SS: return &save->ss;
849 case VCPU_SREG_TR: return &save->tr;
850 case VCPU_SREG_LDTR: return &save->ldtr;
856 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
858 struct vmcb_seg *s = svm_seg(vcpu, seg);
863 static void svm_get_segment(struct kvm_vcpu *vcpu,
864 struct kvm_segment *var, int seg)
866 struct vmcb_seg *s = svm_seg(vcpu, seg);
869 var->limit = s->limit;
870 var->selector = s->selector;
871 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
872 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
873 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
874 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
875 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
876 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
877 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
878 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
880 /* AMD's VMCB does not have an explicit unusable field, so emulate it
881 * for cross vendor migration purposes by "not present"
883 var->unusable = !var->present || (var->type == 0);
888 * SVM always stores 0 for the 'G' bit in the CS selector in
889 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
890 * Intel's VMENTRY has a check on the 'G' bit.
892 var->g = s->limit > 0xfffff;
896 * Work around a bug where the busy flag in the tr selector
906 * The accessed bit must always be set in the segment
907 * descriptor cache, although it can be cleared in the
908 * descriptor, the cached bit always remains at 1. Since
909 * Intel has a check on this, set it here to support
910 * cross-vendor migration.
916 /* On AMD CPUs sometimes the DB bit in the segment
917 * descriptor is left as 1, although the whole segment has
918 * been made unusable. Clear it here to pass an Intel VMX
919 * entry check when cross vendor migrating.
927 static int svm_get_cpl(struct kvm_vcpu *vcpu)
929 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
934 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
936 struct vcpu_svm *svm = to_svm(vcpu);
938 dt->limit = svm->vmcb->save.idtr.limit;
939 dt->base = svm->vmcb->save.idtr.base;
942 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
944 struct vcpu_svm *svm = to_svm(vcpu);
946 svm->vmcb->save.idtr.limit = dt->limit;
947 svm->vmcb->save.idtr.base = dt->base ;
950 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
952 struct vcpu_svm *svm = to_svm(vcpu);
954 dt->limit = svm->vmcb->save.gdtr.limit;
955 dt->base = svm->vmcb->save.gdtr.base;
958 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
960 struct vcpu_svm *svm = to_svm(vcpu);
962 svm->vmcb->save.gdtr.limit = dt->limit;
963 svm->vmcb->save.gdtr.base = dt->base ;
966 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
970 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
974 static void update_cr0_intercept(struct vcpu_svm *svm)
976 ulong gcr0 = svm->vcpu.arch.cr0;
977 u64 *hcr0 = &svm->vmcb->save.cr0;
979 if (!svm->vcpu.fpu_active)
980 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
982 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
983 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
986 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
987 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
988 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
990 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
991 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
995 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
997 struct vcpu_svm *svm = to_svm(vcpu);
1000 if (vcpu->arch.efer & EFER_LME) {
1001 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1002 vcpu->arch.efer |= EFER_LMA;
1003 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1006 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1007 vcpu->arch.efer &= ~EFER_LMA;
1008 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1012 vcpu->arch.cr0 = cr0;
1015 cr0 |= X86_CR0_PG | X86_CR0_WP;
1017 if (!vcpu->fpu_active)
1020 * re-enable caching here because the QEMU bios
1021 * does not do it - this results in some delay at
1024 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1025 svm->vmcb->save.cr0 = cr0;
1026 update_cr0_intercept(svm);
1029 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1031 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1032 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1034 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1035 force_new_asid(vcpu);
1037 vcpu->arch.cr4 = cr4;
1040 cr4 |= host_cr4_mce;
1041 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1044 static void svm_set_segment(struct kvm_vcpu *vcpu,
1045 struct kvm_segment *var, int seg)
1047 struct vcpu_svm *svm = to_svm(vcpu);
1048 struct vmcb_seg *s = svm_seg(vcpu, seg);
1050 s->base = var->base;
1051 s->limit = var->limit;
1052 s->selector = var->selector;
1056 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1057 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1058 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1059 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1060 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1061 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1062 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1063 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1065 if (seg == VCPU_SREG_CS)
1067 = (svm->vmcb->save.cs.attrib
1068 >> SVM_SELECTOR_DPL_SHIFT) & 3;
1072 static void update_db_intercept(struct kvm_vcpu *vcpu)
1074 struct vcpu_svm *svm = to_svm(vcpu);
1076 svm->vmcb->control.intercept_exceptions &=
1077 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1079 if (svm->nmi_singlestep)
1080 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1082 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1083 if (vcpu->guest_debug &
1084 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1085 svm->vmcb->control.intercept_exceptions |=
1087 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1088 svm->vmcb->control.intercept_exceptions |=
1091 vcpu->guest_debug = 0;
1094 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1096 struct vcpu_svm *svm = to_svm(vcpu);
1098 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1099 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1101 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1103 update_db_intercept(vcpu);
1106 static void load_host_msrs(struct kvm_vcpu *vcpu)
1108 #ifdef CONFIG_X86_64
1109 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1113 static void save_host_msrs(struct kvm_vcpu *vcpu)
1115 #ifdef CONFIG_X86_64
1116 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1120 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1122 if (sd->next_asid > sd->max_asid) {
1123 ++sd->asid_generation;
1125 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1128 svm->asid_generation = sd->asid_generation;
1129 svm->vmcb->control.asid = sd->next_asid++;
1132 static int svm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *dest)
1134 struct vcpu_svm *svm = to_svm(vcpu);
1138 *dest = vcpu->arch.db[dr];
1141 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1142 return EMULATE_FAIL; /* will re-inject UD */
1145 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1146 *dest = vcpu->arch.dr6;
1148 *dest = svm->vmcb->save.dr6;
1151 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1152 return EMULATE_FAIL; /* will re-inject UD */
1155 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1156 *dest = vcpu->arch.dr7;
1158 *dest = svm->vmcb->save.dr7;
1162 return EMULATE_DONE;
1165 static int svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value)
1167 struct vcpu_svm *svm = to_svm(vcpu);
1171 vcpu->arch.db[dr] = value;
1172 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1173 vcpu->arch.eff_db[dr] = value;
1176 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1177 return EMULATE_FAIL; /* will re-inject UD */
1180 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1183 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1184 return EMULATE_FAIL; /* will re-inject UD */
1187 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1188 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1189 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1190 vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1195 return EMULATE_DONE;
1198 static int pf_interception(struct vcpu_svm *svm)
1203 fault_address = svm->vmcb->control.exit_info_2;
1204 error_code = svm->vmcb->control.exit_info_1;
1206 trace_kvm_page_fault(fault_address, error_code);
1207 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1208 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1209 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1212 static int db_interception(struct vcpu_svm *svm)
1214 struct kvm_run *kvm_run = svm->vcpu.run;
1216 if (!(svm->vcpu.guest_debug &
1217 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1218 !svm->nmi_singlestep) {
1219 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1223 if (svm->nmi_singlestep) {
1224 svm->nmi_singlestep = false;
1225 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1226 svm->vmcb->save.rflags &=
1227 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1228 update_db_intercept(&svm->vcpu);
1231 if (svm->vcpu.guest_debug &
1232 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)){
1233 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1234 kvm_run->debug.arch.pc =
1235 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1236 kvm_run->debug.arch.exception = DB_VECTOR;
1243 static int bp_interception(struct vcpu_svm *svm)
1245 struct kvm_run *kvm_run = svm->vcpu.run;
1247 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1248 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1249 kvm_run->debug.arch.exception = BP_VECTOR;
1253 static int ud_interception(struct vcpu_svm *svm)
1257 er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1258 if (er != EMULATE_DONE)
1259 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1263 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1265 struct vcpu_svm *svm = to_svm(vcpu);
1266 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1267 svm->vcpu.fpu_active = 1;
1268 update_cr0_intercept(svm);
1271 static int nm_interception(struct vcpu_svm *svm)
1273 svm_fpu_activate(&svm->vcpu);
1277 static int mc_interception(struct vcpu_svm *svm)
1280 * On an #MC intercept the MCE handler is not called automatically in
1281 * the host. So do it by hand here.
1285 /* not sure if we ever come back to this point */
1290 static int shutdown_interception(struct vcpu_svm *svm)
1292 struct kvm_run *kvm_run = svm->vcpu.run;
1295 * VMCB is undefined after a SHUTDOWN intercept
1296 * so reinitialize it.
1298 clear_page(svm->vmcb);
1301 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1305 static int io_interception(struct vcpu_svm *svm)
1307 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1308 int size, in, string;
1311 ++svm->vcpu.stat.io_exits;
1313 svm->next_rip = svm->vmcb->control.exit_info_2;
1315 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1318 if (emulate_instruction(&svm->vcpu,
1319 0, 0, 0) == EMULATE_DO_MMIO)
1324 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1325 port = io_info >> 16;
1326 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1328 skip_emulated_instruction(&svm->vcpu);
1329 return kvm_emulate_pio(&svm->vcpu, in, size, port);
1332 static int nmi_interception(struct vcpu_svm *svm)
1337 static int intr_interception(struct vcpu_svm *svm)
1339 ++svm->vcpu.stat.irq_exits;
1343 static int nop_on_interception(struct vcpu_svm *svm)
1348 static int halt_interception(struct vcpu_svm *svm)
1350 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1351 skip_emulated_instruction(&svm->vcpu);
1352 return kvm_emulate_halt(&svm->vcpu);
1355 static int vmmcall_interception(struct vcpu_svm *svm)
1357 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1358 skip_emulated_instruction(&svm->vcpu);
1359 kvm_emulate_hypercall(&svm->vcpu);
1363 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1365 if (!(svm->vcpu.arch.efer & EFER_SVME)
1366 || !is_paging(&svm->vcpu)) {
1367 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1371 if (svm->vmcb->save.cpl) {
1372 kvm_inject_gp(&svm->vcpu, 0);
1379 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1380 bool has_error_code, u32 error_code)
1382 if (!is_nested(svm))
1385 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1386 svm->vmcb->control.exit_code_hi = 0;
1387 svm->vmcb->control.exit_info_1 = error_code;
1388 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1390 return nested_svm_exit_handled(svm);
1393 static inline int nested_svm_intr(struct vcpu_svm *svm)
1395 if (!is_nested(svm))
1398 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1401 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1404 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1406 if (svm->nested.intercept & 1ULL) {
1408 * The #vmexit can't be emulated here directly because this
1409 * code path runs with irqs and preemtion disabled. A
1410 * #vmexit emulation might sleep. Only signal request for
1413 svm->nested.exit_required = true;
1414 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1421 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, enum km_type idx)
1425 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1426 if (is_error_page(page))
1429 return kmap_atomic(page, idx);
1432 kvm_release_page_clean(page);
1433 kvm_inject_gp(&svm->vcpu, 0);
1438 static void nested_svm_unmap(void *addr, enum km_type idx)
1445 page = kmap_atomic_to_page(addr);
1447 kunmap_atomic(addr, idx);
1448 kvm_release_page_dirty(page);
1451 static bool nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1453 u32 param = svm->vmcb->control.exit_info_1 & 1;
1454 u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1459 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1462 msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, KM_USER0);
1472 case 0xc0000000 ... 0xc0001fff:
1473 t0 = (8192 + msr - 0xc0000000) * 2;
1477 case 0xc0010000 ... 0xc0011fff:
1478 t0 = (16384 + msr - 0xc0010000) * 2;
1487 ret = msrpm[t1] & ((1 << param) << t0);
1490 nested_svm_unmap(msrpm, KM_USER0);
1495 static int nested_svm_exit_special(struct vcpu_svm *svm)
1497 u32 exit_code = svm->vmcb->control.exit_code;
1499 switch (exit_code) {
1502 return NESTED_EXIT_HOST;
1503 /* For now we are always handling NPFs when using them */
1506 return NESTED_EXIT_HOST;
1508 /* When we're shadowing, trap PFs */
1509 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1511 return NESTED_EXIT_HOST;
1517 return NESTED_EXIT_CONTINUE;
1521 * If this function returns true, this #vmexit was already handled
1523 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1525 u32 exit_code = svm->vmcb->control.exit_code;
1526 int vmexit = NESTED_EXIT_HOST;
1528 switch (exit_code) {
1530 vmexit = nested_svm_exit_handled_msr(svm);
1532 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1533 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1534 if (svm->nested.intercept_cr_read & cr_bits)
1535 vmexit = NESTED_EXIT_DONE;
1538 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1539 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1540 if (svm->nested.intercept_cr_write & cr_bits)
1541 vmexit = NESTED_EXIT_DONE;
1544 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1545 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1546 if (svm->nested.intercept_dr_read & dr_bits)
1547 vmexit = NESTED_EXIT_DONE;
1550 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1551 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1552 if (svm->nested.intercept_dr_write & dr_bits)
1553 vmexit = NESTED_EXIT_DONE;
1556 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1557 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1558 if (svm->nested.intercept_exceptions & excp_bits)
1559 vmexit = NESTED_EXIT_DONE;
1563 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1564 if (svm->nested.intercept & exit_bits)
1565 vmexit = NESTED_EXIT_DONE;
1569 if (vmexit == NESTED_EXIT_DONE) {
1570 nested_svm_vmexit(svm);
1576 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1578 struct vmcb_control_area *dst = &dst_vmcb->control;
1579 struct vmcb_control_area *from = &from_vmcb->control;
1581 dst->intercept_cr_read = from->intercept_cr_read;
1582 dst->intercept_cr_write = from->intercept_cr_write;
1583 dst->intercept_dr_read = from->intercept_dr_read;
1584 dst->intercept_dr_write = from->intercept_dr_write;
1585 dst->intercept_exceptions = from->intercept_exceptions;
1586 dst->intercept = from->intercept;
1587 dst->iopm_base_pa = from->iopm_base_pa;
1588 dst->msrpm_base_pa = from->msrpm_base_pa;
1589 dst->tsc_offset = from->tsc_offset;
1590 dst->asid = from->asid;
1591 dst->tlb_ctl = from->tlb_ctl;
1592 dst->int_ctl = from->int_ctl;
1593 dst->int_vector = from->int_vector;
1594 dst->int_state = from->int_state;
1595 dst->exit_code = from->exit_code;
1596 dst->exit_code_hi = from->exit_code_hi;
1597 dst->exit_info_1 = from->exit_info_1;
1598 dst->exit_info_2 = from->exit_info_2;
1599 dst->exit_int_info = from->exit_int_info;
1600 dst->exit_int_info_err = from->exit_int_info_err;
1601 dst->nested_ctl = from->nested_ctl;
1602 dst->event_inj = from->event_inj;
1603 dst->event_inj_err = from->event_inj_err;
1604 dst->nested_cr3 = from->nested_cr3;
1605 dst->lbr_ctl = from->lbr_ctl;
1608 static int nested_svm_vmexit(struct vcpu_svm *svm)
1610 struct vmcb *nested_vmcb;
1611 struct vmcb *hsave = svm->nested.hsave;
1612 struct vmcb *vmcb = svm->vmcb;
1614 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1615 vmcb->control.exit_info_1,
1616 vmcb->control.exit_info_2,
1617 vmcb->control.exit_int_info,
1618 vmcb->control.exit_int_info_err);
1620 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, KM_USER0);
1624 /* Give the current vmcb to the guest */
1627 nested_vmcb->save.es = vmcb->save.es;
1628 nested_vmcb->save.cs = vmcb->save.cs;
1629 nested_vmcb->save.ss = vmcb->save.ss;
1630 nested_vmcb->save.ds = vmcb->save.ds;
1631 nested_vmcb->save.gdtr = vmcb->save.gdtr;
1632 nested_vmcb->save.idtr = vmcb->save.idtr;
1634 nested_vmcb->save.cr3 = vmcb->save.cr3;
1635 nested_vmcb->save.cr2 = vmcb->save.cr2;
1636 nested_vmcb->save.rflags = vmcb->save.rflags;
1637 nested_vmcb->save.rip = vmcb->save.rip;
1638 nested_vmcb->save.rsp = vmcb->save.rsp;
1639 nested_vmcb->save.rax = vmcb->save.rax;
1640 nested_vmcb->save.dr7 = vmcb->save.dr7;
1641 nested_vmcb->save.dr6 = vmcb->save.dr6;
1642 nested_vmcb->save.cpl = vmcb->save.cpl;
1644 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
1645 nested_vmcb->control.int_vector = vmcb->control.int_vector;
1646 nested_vmcb->control.int_state = vmcb->control.int_state;
1647 nested_vmcb->control.exit_code = vmcb->control.exit_code;
1648 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
1649 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
1650 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
1651 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
1652 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1655 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1656 * to make sure that we do not lose injected events. So check event_inj
1657 * here and copy it to exit_int_info if it is valid.
1658 * Exit_int_info and event_inj can't be both valid because the case
1659 * below only happens on a VMRUN instruction intercept which has
1660 * no valid exit_int_info set.
1662 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
1663 struct vmcb_control_area *nc = &nested_vmcb->control;
1665 nc->exit_int_info = vmcb->control.event_inj;
1666 nc->exit_int_info_err = vmcb->control.event_inj_err;
1669 nested_vmcb->control.tlb_ctl = 0;
1670 nested_vmcb->control.event_inj = 0;
1671 nested_vmcb->control.event_inj_err = 0;
1673 /* We always set V_INTR_MASKING and remember the old value in hflags */
1674 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1675 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1677 /* Restore the original control entries */
1678 copy_vmcb_control_area(vmcb, hsave);
1680 kvm_clear_exception_queue(&svm->vcpu);
1681 kvm_clear_interrupt_queue(&svm->vcpu);
1683 /* Restore selected save entries */
1684 svm->vmcb->save.es = hsave->save.es;
1685 svm->vmcb->save.cs = hsave->save.cs;
1686 svm->vmcb->save.ss = hsave->save.ss;
1687 svm->vmcb->save.ds = hsave->save.ds;
1688 svm->vmcb->save.gdtr = hsave->save.gdtr;
1689 svm->vmcb->save.idtr = hsave->save.idtr;
1690 svm->vmcb->save.rflags = hsave->save.rflags;
1691 svm_set_efer(&svm->vcpu, hsave->save.efer);
1692 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1693 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1695 svm->vmcb->save.cr3 = hsave->save.cr3;
1696 svm->vcpu.arch.cr3 = hsave->save.cr3;
1698 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1700 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1701 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1702 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1703 svm->vmcb->save.dr7 = 0;
1704 svm->vmcb->save.cpl = 0;
1705 svm->vmcb->control.exit_int_info = 0;
1707 /* Exit nested SVM mode */
1708 svm->nested.vmcb = 0;
1710 nested_svm_unmap(nested_vmcb, KM_USER0);
1712 kvm_mmu_reset_context(&svm->vcpu);
1713 kvm_mmu_load(&svm->vcpu);
1718 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
1723 nested_msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, KM_USER0);
1727 for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1728 svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1730 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1732 nested_svm_unmap(nested_msrpm, KM_USER0);
1737 static bool nested_svm_vmrun(struct vcpu_svm *svm)
1739 struct vmcb *nested_vmcb;
1740 struct vmcb *hsave = svm->nested.hsave;
1741 struct vmcb *vmcb = svm->vmcb;
1743 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1747 /* nested_vmcb is our indicator if nested SVM is activated */
1748 svm->nested.vmcb = svm->vmcb->save.rax;
1750 trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, svm->nested.vmcb,
1751 nested_vmcb->save.rip,
1752 nested_vmcb->control.int_ctl,
1753 nested_vmcb->control.event_inj,
1754 nested_vmcb->control.nested_ctl);
1756 /* Clear internal status */
1757 kvm_clear_exception_queue(&svm->vcpu);
1758 kvm_clear_interrupt_queue(&svm->vcpu);
1760 /* Save the old vmcb, so we don't need to pick what we save, but
1761 can restore everything when a VMEXIT occurs */
1762 hsave->save.es = vmcb->save.es;
1763 hsave->save.cs = vmcb->save.cs;
1764 hsave->save.ss = vmcb->save.ss;
1765 hsave->save.ds = vmcb->save.ds;
1766 hsave->save.gdtr = vmcb->save.gdtr;
1767 hsave->save.idtr = vmcb->save.idtr;
1768 hsave->save.efer = svm->vcpu.arch.efer;
1769 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
1770 hsave->save.cr4 = svm->vcpu.arch.cr4;
1771 hsave->save.rflags = vmcb->save.rflags;
1772 hsave->save.rip = svm->next_rip;
1773 hsave->save.rsp = vmcb->save.rsp;
1774 hsave->save.rax = vmcb->save.rax;
1776 hsave->save.cr3 = vmcb->save.cr3;
1778 hsave->save.cr3 = svm->vcpu.arch.cr3;
1780 copy_vmcb_control_area(hsave, vmcb);
1782 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1783 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1785 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1787 /* Load the nested guest state */
1788 svm->vmcb->save.es = nested_vmcb->save.es;
1789 svm->vmcb->save.cs = nested_vmcb->save.cs;
1790 svm->vmcb->save.ss = nested_vmcb->save.ss;
1791 svm->vmcb->save.ds = nested_vmcb->save.ds;
1792 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1793 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1794 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1795 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1796 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1797 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1799 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1800 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1802 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1803 kvm_mmu_reset_context(&svm->vcpu);
1805 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
1806 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1807 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1808 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1809 /* In case we don't even reach vcpu_run, the fields are not updated */
1810 svm->vmcb->save.rax = nested_vmcb->save.rax;
1811 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1812 svm->vmcb->save.rip = nested_vmcb->save.rip;
1813 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1814 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1815 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1817 /* We don't want a nested guest to be more powerful than the guest,
1818 so all intercepts are ORed */
1819 svm->vmcb->control.intercept_cr_read |=
1820 nested_vmcb->control.intercept_cr_read;
1821 svm->vmcb->control.intercept_cr_write |=
1822 nested_vmcb->control.intercept_cr_write;
1823 svm->vmcb->control.intercept_dr_read |=
1824 nested_vmcb->control.intercept_dr_read;
1825 svm->vmcb->control.intercept_dr_write |=
1826 nested_vmcb->control.intercept_dr_write;
1827 svm->vmcb->control.intercept_exceptions |=
1828 nested_vmcb->control.intercept_exceptions;
1830 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1832 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1834 /* cache intercepts */
1835 svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
1836 svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
1837 svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
1838 svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
1839 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
1840 svm->nested.intercept = nested_vmcb->control.intercept;
1842 force_new_asid(&svm->vcpu);
1843 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1844 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1845 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1847 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1849 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1850 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1851 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1852 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1853 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1855 nested_svm_unmap(nested_vmcb, KM_USER0);
1862 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1864 to_vmcb->save.fs = from_vmcb->save.fs;
1865 to_vmcb->save.gs = from_vmcb->save.gs;
1866 to_vmcb->save.tr = from_vmcb->save.tr;
1867 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1868 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1869 to_vmcb->save.star = from_vmcb->save.star;
1870 to_vmcb->save.lstar = from_vmcb->save.lstar;
1871 to_vmcb->save.cstar = from_vmcb->save.cstar;
1872 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1873 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1874 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1875 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1878 static int vmload_interception(struct vcpu_svm *svm)
1880 struct vmcb *nested_vmcb;
1882 if (nested_svm_check_permissions(svm))
1885 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1886 skip_emulated_instruction(&svm->vcpu);
1888 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1892 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
1893 nested_svm_unmap(nested_vmcb, KM_USER0);
1898 static int vmsave_interception(struct vcpu_svm *svm)
1900 struct vmcb *nested_vmcb;
1902 if (nested_svm_check_permissions(svm))
1905 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1906 skip_emulated_instruction(&svm->vcpu);
1908 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, KM_USER0);
1912 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
1913 nested_svm_unmap(nested_vmcb, KM_USER0);
1918 static int vmrun_interception(struct vcpu_svm *svm)
1920 if (nested_svm_check_permissions(svm))
1923 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1924 skip_emulated_instruction(&svm->vcpu);
1926 if (!nested_svm_vmrun(svm))
1929 if (!nested_svm_vmrun_msrpm(svm))
1936 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
1937 svm->vmcb->control.exit_code_hi = 0;
1938 svm->vmcb->control.exit_info_1 = 0;
1939 svm->vmcb->control.exit_info_2 = 0;
1941 nested_svm_vmexit(svm);
1946 static int stgi_interception(struct vcpu_svm *svm)
1948 if (nested_svm_check_permissions(svm))
1951 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1952 skip_emulated_instruction(&svm->vcpu);
1959 static int clgi_interception(struct vcpu_svm *svm)
1961 if (nested_svm_check_permissions(svm))
1964 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1965 skip_emulated_instruction(&svm->vcpu);
1969 /* After a CLGI no interrupts should come */
1970 svm_clear_vintr(svm);
1971 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1976 static int invlpga_interception(struct vcpu_svm *svm)
1978 struct kvm_vcpu *vcpu = &svm->vcpu;
1980 trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
1981 vcpu->arch.regs[VCPU_REGS_RAX]);
1983 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
1984 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
1986 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1987 skip_emulated_instruction(&svm->vcpu);
1991 static int skinit_interception(struct vcpu_svm *svm)
1993 trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
1995 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1999 static int invalid_op_interception(struct vcpu_svm *svm)
2001 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2005 static int task_switch_interception(struct vcpu_svm *svm)
2009 int int_type = svm->vmcb->control.exit_int_info &
2010 SVM_EXITINTINFO_TYPE_MASK;
2011 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2013 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2015 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2017 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2019 if (svm->vmcb->control.exit_info_2 &
2020 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2021 reason = TASK_SWITCH_IRET;
2022 else if (svm->vmcb->control.exit_info_2 &
2023 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2024 reason = TASK_SWITCH_JMP;
2026 reason = TASK_SWITCH_GATE;
2028 reason = TASK_SWITCH_CALL;
2030 if (reason == TASK_SWITCH_GATE) {
2032 case SVM_EXITINTINFO_TYPE_NMI:
2033 svm->vcpu.arch.nmi_injected = false;
2035 case SVM_EXITINTINFO_TYPE_EXEPT:
2036 kvm_clear_exception_queue(&svm->vcpu);
2038 case SVM_EXITINTINFO_TYPE_INTR:
2039 kvm_clear_interrupt_queue(&svm->vcpu);
2046 if (reason != TASK_SWITCH_GATE ||
2047 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2048 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2049 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2050 skip_emulated_instruction(&svm->vcpu);
2052 return kvm_task_switch(&svm->vcpu, tss_selector, reason);
2055 static int cpuid_interception(struct vcpu_svm *svm)
2057 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2058 kvm_emulate_cpuid(&svm->vcpu);
2062 static int iret_interception(struct vcpu_svm *svm)
2064 ++svm->vcpu.stat.nmi_window_exits;
2065 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2066 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2070 static int invlpg_interception(struct vcpu_svm *svm)
2072 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2073 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2077 static int emulate_on_interception(struct vcpu_svm *svm)
2079 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2080 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2084 static int cr8_write_interception(struct vcpu_svm *svm)
2086 struct kvm_run *kvm_run = svm->vcpu.run;
2088 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2089 /* instruction emulation calls kvm_set_cr8() */
2090 emulate_instruction(&svm->vcpu, 0, 0, 0);
2091 if (irqchip_in_kernel(svm->vcpu.kvm)) {
2092 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2095 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2097 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2101 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2103 struct vcpu_svm *svm = to_svm(vcpu);
2106 case MSR_IA32_TSC: {
2110 tsc_offset = svm->nested.hsave->control.tsc_offset;
2112 tsc_offset = svm->vmcb->control.tsc_offset;
2114 *data = tsc_offset + native_read_tsc();
2118 *data = svm->vmcb->save.star;
2120 #ifdef CONFIG_X86_64
2122 *data = svm->vmcb->save.lstar;
2125 *data = svm->vmcb->save.cstar;
2127 case MSR_KERNEL_GS_BASE:
2128 *data = svm->vmcb->save.kernel_gs_base;
2130 case MSR_SYSCALL_MASK:
2131 *data = svm->vmcb->save.sfmask;
2134 case MSR_IA32_SYSENTER_CS:
2135 *data = svm->vmcb->save.sysenter_cs;
2137 case MSR_IA32_SYSENTER_EIP:
2138 *data = svm->sysenter_eip;
2140 case MSR_IA32_SYSENTER_ESP:
2141 *data = svm->sysenter_esp;
2143 /* Nobody will change the following 5 values in the VMCB so
2144 we can safely return them on rdmsr. They will always be 0
2145 until LBRV is implemented. */
2146 case MSR_IA32_DEBUGCTLMSR:
2147 *data = svm->vmcb->save.dbgctl;
2149 case MSR_IA32_LASTBRANCHFROMIP:
2150 *data = svm->vmcb->save.br_from;
2152 case MSR_IA32_LASTBRANCHTOIP:
2153 *data = svm->vmcb->save.br_to;
2155 case MSR_IA32_LASTINTFROMIP:
2156 *data = svm->vmcb->save.last_excp_from;
2158 case MSR_IA32_LASTINTTOIP:
2159 *data = svm->vmcb->save.last_excp_to;
2161 case MSR_VM_HSAVE_PA:
2162 *data = svm->nested.hsave_msr;
2167 case MSR_IA32_UCODE_REV:
2171 return kvm_get_msr_common(vcpu, ecx, data);
2176 static int rdmsr_interception(struct vcpu_svm *svm)
2178 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2181 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2182 trace_kvm_msr_read_ex(ecx);
2183 kvm_inject_gp(&svm->vcpu, 0);
2185 trace_kvm_msr_read(ecx, data);
2187 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2188 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2189 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2190 skip_emulated_instruction(&svm->vcpu);
2195 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2197 struct vcpu_svm *svm = to_svm(vcpu);
2200 case MSR_IA32_TSC: {
2201 u64 tsc_offset = data - native_read_tsc();
2202 u64 g_tsc_offset = 0;
2204 if (is_nested(svm)) {
2205 g_tsc_offset = svm->vmcb->control.tsc_offset -
2206 svm->nested.hsave->control.tsc_offset;
2207 svm->nested.hsave->control.tsc_offset = tsc_offset;
2210 svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset;
2215 svm->vmcb->save.star = data;
2217 #ifdef CONFIG_X86_64
2219 svm->vmcb->save.lstar = data;
2222 svm->vmcb->save.cstar = data;
2224 case MSR_KERNEL_GS_BASE:
2225 svm->vmcb->save.kernel_gs_base = data;
2227 case MSR_SYSCALL_MASK:
2228 svm->vmcb->save.sfmask = data;
2231 case MSR_IA32_SYSENTER_CS:
2232 svm->vmcb->save.sysenter_cs = data;
2234 case MSR_IA32_SYSENTER_EIP:
2235 svm->sysenter_eip = data;
2236 svm->vmcb->save.sysenter_eip = data;
2238 case MSR_IA32_SYSENTER_ESP:
2239 svm->sysenter_esp = data;
2240 svm->vmcb->save.sysenter_esp = data;
2242 case MSR_IA32_DEBUGCTLMSR:
2243 if (!svm_has(SVM_FEATURE_LBRV)) {
2244 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2248 if (data & DEBUGCTL_RESERVED_BITS)
2251 svm->vmcb->save.dbgctl = data;
2252 if (data & (1ULL<<0))
2253 svm_enable_lbrv(svm);
2255 svm_disable_lbrv(svm);
2257 case MSR_VM_HSAVE_PA:
2258 svm->nested.hsave_msr = data;
2262 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2265 return kvm_set_msr_common(vcpu, ecx, data);
2270 static int wrmsr_interception(struct vcpu_svm *svm)
2272 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2273 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2274 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2277 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2278 if (svm_set_msr(&svm->vcpu, ecx, data)) {
2279 trace_kvm_msr_write_ex(ecx, data);
2280 kvm_inject_gp(&svm->vcpu, 0);
2282 trace_kvm_msr_write(ecx, data);
2283 skip_emulated_instruction(&svm->vcpu);
2288 static int msr_interception(struct vcpu_svm *svm)
2290 if (svm->vmcb->control.exit_info_1)
2291 return wrmsr_interception(svm);
2293 return rdmsr_interception(svm);
2296 static int interrupt_window_interception(struct vcpu_svm *svm)
2298 struct kvm_run *kvm_run = svm->vcpu.run;
2300 svm_clear_vintr(svm);
2301 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2303 * If the user space waits to inject interrupts, exit as soon as
2306 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2307 kvm_run->request_interrupt_window &&
2308 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2309 ++svm->vcpu.stat.irq_window_exits;
2310 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2317 static int pause_interception(struct vcpu_svm *svm)
2319 kvm_vcpu_on_spin(&(svm->vcpu));
2323 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2324 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2325 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2326 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2327 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2328 [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
2329 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
2330 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2331 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2332 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2333 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2334 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2335 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2336 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2337 [SVM_EXIT_READ_DR4] = emulate_on_interception,
2338 [SVM_EXIT_READ_DR5] = emulate_on_interception,
2339 [SVM_EXIT_READ_DR6] = emulate_on_interception,
2340 [SVM_EXIT_READ_DR7] = emulate_on_interception,
2341 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2342 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2343 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2344 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2345 [SVM_EXIT_WRITE_DR4] = emulate_on_interception,
2346 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2347 [SVM_EXIT_WRITE_DR6] = emulate_on_interception,
2348 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2349 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2350 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2351 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2352 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2353 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2354 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2355 [SVM_EXIT_INTR] = intr_interception,
2356 [SVM_EXIT_NMI] = nmi_interception,
2357 [SVM_EXIT_SMI] = nop_on_interception,
2358 [SVM_EXIT_INIT] = nop_on_interception,
2359 [SVM_EXIT_VINTR] = interrupt_window_interception,
2360 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2361 [SVM_EXIT_CPUID] = cpuid_interception,
2362 [SVM_EXIT_IRET] = iret_interception,
2363 [SVM_EXIT_INVD] = emulate_on_interception,
2364 [SVM_EXIT_PAUSE] = pause_interception,
2365 [SVM_EXIT_HLT] = halt_interception,
2366 [SVM_EXIT_INVLPG] = invlpg_interception,
2367 [SVM_EXIT_INVLPGA] = invlpga_interception,
2368 [SVM_EXIT_IOIO] = io_interception,
2369 [SVM_EXIT_MSR] = msr_interception,
2370 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2371 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2372 [SVM_EXIT_VMRUN] = vmrun_interception,
2373 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2374 [SVM_EXIT_VMLOAD] = vmload_interception,
2375 [SVM_EXIT_VMSAVE] = vmsave_interception,
2376 [SVM_EXIT_STGI] = stgi_interception,
2377 [SVM_EXIT_CLGI] = clgi_interception,
2378 [SVM_EXIT_SKINIT] = skinit_interception,
2379 [SVM_EXIT_WBINVD] = emulate_on_interception,
2380 [SVM_EXIT_MONITOR] = invalid_op_interception,
2381 [SVM_EXIT_MWAIT] = invalid_op_interception,
2382 [SVM_EXIT_NPF] = pf_interception,
2385 static int handle_exit(struct kvm_vcpu *vcpu)
2387 struct vcpu_svm *svm = to_svm(vcpu);
2388 struct kvm_run *kvm_run = vcpu->run;
2389 u32 exit_code = svm->vmcb->control.exit_code;
2391 trace_kvm_exit(exit_code, svm->vmcb->save.rip);
2393 if (unlikely(svm->nested.exit_required)) {
2394 nested_svm_vmexit(svm);
2395 svm->nested.exit_required = false;
2400 if (is_nested(svm)) {
2403 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2404 svm->vmcb->control.exit_info_1,
2405 svm->vmcb->control.exit_info_2,
2406 svm->vmcb->control.exit_int_info,
2407 svm->vmcb->control.exit_int_info_err);
2409 vmexit = nested_svm_exit_special(svm);
2411 if (vmexit == NESTED_EXIT_CONTINUE)
2412 vmexit = nested_svm_exit_handled(svm);
2414 if (vmexit == NESTED_EXIT_DONE)
2418 svm_complete_interrupts(svm);
2420 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2421 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2423 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2425 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2426 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2427 kvm_run->fail_entry.hardware_entry_failure_reason
2428 = svm->vmcb->control.exit_code;
2432 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2433 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2434 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2435 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2437 __func__, svm->vmcb->control.exit_int_info,
2440 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2441 || !svm_exit_handlers[exit_code]) {
2442 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2443 kvm_run->hw.hardware_exit_reason = exit_code;
2447 return svm_exit_handlers[exit_code](svm);
2450 static void reload_tss(struct kvm_vcpu *vcpu)
2452 int cpu = raw_smp_processor_id();
2454 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2455 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
2459 static void pre_svm_run(struct vcpu_svm *svm)
2461 int cpu = raw_smp_processor_id();
2463 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2465 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2466 /* FIXME: handle wraparound of asid_generation */
2467 if (svm->asid_generation != sd->asid_generation)
2471 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2473 struct vcpu_svm *svm = to_svm(vcpu);
2475 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2476 vcpu->arch.hflags |= HF_NMI_MASK;
2477 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2478 ++vcpu->stat.nmi_injections;
2481 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2483 struct vmcb_control_area *control;
2485 trace_kvm_inj_virq(irq);
2487 ++svm->vcpu.stat.irq_injections;
2488 control = &svm->vmcb->control;
2489 control->int_vector = irq;
2490 control->int_ctl &= ~V_INTR_PRIO_MASK;
2491 control->int_ctl |= V_IRQ_MASK |
2492 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2495 static void svm_set_irq(struct kvm_vcpu *vcpu)
2497 struct vcpu_svm *svm = to_svm(vcpu);
2499 BUG_ON(!(gif_set(svm)));
2501 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2502 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2505 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2507 struct vcpu_svm *svm = to_svm(vcpu);
2513 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2516 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2518 struct vcpu_svm *svm = to_svm(vcpu);
2519 struct vmcb *vmcb = svm->vmcb;
2520 return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2521 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2524 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
2526 struct vcpu_svm *svm = to_svm(vcpu);
2528 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
2531 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2533 struct vcpu_svm *svm = to_svm(vcpu);
2536 svm->vcpu.arch.hflags |= HF_NMI_MASK;
2537 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2539 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
2540 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2544 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2546 struct vcpu_svm *svm = to_svm(vcpu);
2547 struct vmcb *vmcb = svm->vmcb;
2550 if (!gif_set(svm) ||
2551 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
2554 ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
2557 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
2562 static void enable_irq_window(struct kvm_vcpu *vcpu)
2564 struct vcpu_svm *svm = to_svm(vcpu);
2566 nested_svm_intr(svm);
2568 /* In case GIF=0 we can't rely on the CPU to tell us when
2569 * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
2570 * The next time we get that intercept, this function will be
2571 * called again though and we'll get the vintr intercept. */
2574 svm_inject_irq(svm, 0x0);
2578 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2580 struct vcpu_svm *svm = to_svm(vcpu);
2582 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2584 return; /* IRET will cause a vm exit */
2586 /* Something prevents NMI from been injected. Single step over
2587 possible problem (IRET or exception injection or interrupt
2589 svm->nmi_singlestep = true;
2590 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2591 update_db_intercept(vcpu);
2594 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2599 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2601 force_new_asid(vcpu);
2604 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2608 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2610 struct vcpu_svm *svm = to_svm(vcpu);
2612 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2613 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2614 kvm_set_cr8(vcpu, cr8);
2618 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2620 struct vcpu_svm *svm = to_svm(vcpu);
2623 cr8 = kvm_get_cr8(vcpu);
2624 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2625 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2628 static void svm_complete_interrupts(struct vcpu_svm *svm)
2632 u32 exitintinfo = svm->vmcb->control.exit_int_info;
2634 if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2635 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2637 svm->vcpu.arch.nmi_injected = false;
2638 kvm_clear_exception_queue(&svm->vcpu);
2639 kvm_clear_interrupt_queue(&svm->vcpu);
2641 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2644 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2645 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2648 case SVM_EXITINTINFO_TYPE_NMI:
2649 svm->vcpu.arch.nmi_injected = true;
2651 case SVM_EXITINTINFO_TYPE_EXEPT:
2652 /* In case of software exception do not reinject an exception
2653 vector, but re-execute and instruction instead */
2656 if (kvm_exception_is_soft(vector))
2658 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2659 u32 err = svm->vmcb->control.exit_int_info_err;
2660 kvm_queue_exception_e(&svm->vcpu, vector, err);
2663 kvm_queue_exception(&svm->vcpu, vector);
2665 case SVM_EXITINTINFO_TYPE_INTR:
2666 kvm_queue_interrupt(&svm->vcpu, vector, false);
2673 #ifdef CONFIG_X86_64
2679 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
2681 struct vcpu_svm *svm = to_svm(vcpu);
2687 * A vmexit emulation is required before the vcpu can be executed
2690 if (unlikely(svm->nested.exit_required))
2693 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2694 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2695 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2699 sync_lapic_to_cr8(vcpu);
2701 save_host_msrs(vcpu);
2702 fs_selector = kvm_read_fs();
2703 gs_selector = kvm_read_gs();
2704 ldt_selector = kvm_read_ldt();
2705 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2706 /* required for live migration with NPT */
2708 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2715 "push %%"R"bp; \n\t"
2716 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2717 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2718 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2719 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2720 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2721 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2722 #ifdef CONFIG_X86_64
2723 "mov %c[r8](%[svm]), %%r8 \n\t"
2724 "mov %c[r9](%[svm]), %%r9 \n\t"
2725 "mov %c[r10](%[svm]), %%r10 \n\t"
2726 "mov %c[r11](%[svm]), %%r11 \n\t"
2727 "mov %c[r12](%[svm]), %%r12 \n\t"
2728 "mov %c[r13](%[svm]), %%r13 \n\t"
2729 "mov %c[r14](%[svm]), %%r14 \n\t"
2730 "mov %c[r15](%[svm]), %%r15 \n\t"
2733 /* Enter guest mode */
2735 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2736 __ex(SVM_VMLOAD) "\n\t"
2737 __ex(SVM_VMRUN) "\n\t"
2738 __ex(SVM_VMSAVE) "\n\t"
2741 /* Save guest registers, load host registers */
2742 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2743 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2744 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2745 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2746 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2747 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2748 #ifdef CONFIG_X86_64
2749 "mov %%r8, %c[r8](%[svm]) \n\t"
2750 "mov %%r9, %c[r9](%[svm]) \n\t"
2751 "mov %%r10, %c[r10](%[svm]) \n\t"
2752 "mov %%r11, %c[r11](%[svm]) \n\t"
2753 "mov %%r12, %c[r12](%[svm]) \n\t"
2754 "mov %%r13, %c[r13](%[svm]) \n\t"
2755 "mov %%r14, %c[r14](%[svm]) \n\t"
2756 "mov %%r15, %c[r15](%[svm]) \n\t"
2761 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2762 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2763 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2764 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2765 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2766 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2767 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2768 #ifdef CONFIG_X86_64
2769 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2770 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2771 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2772 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2773 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2774 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2775 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2776 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2779 , R"bx", R"cx", R"dx", R"si", R"di"
2780 #ifdef CONFIG_X86_64
2781 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2785 vcpu->arch.cr2 = svm->vmcb->save.cr2;
2786 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2787 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2788 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2790 kvm_load_fs(fs_selector);
2791 kvm_load_gs(gs_selector);
2792 kvm_load_ldt(ldt_selector);
2793 load_host_msrs(vcpu);
2797 local_irq_disable();
2801 sync_cr8_to_lapic(vcpu);
2806 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
2807 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
2813 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2815 struct vcpu_svm *svm = to_svm(vcpu);
2818 svm->vmcb->control.nested_cr3 = root;
2819 force_new_asid(vcpu);
2823 svm->vmcb->save.cr3 = root;
2824 force_new_asid(vcpu);
2827 static int is_disabled(void)
2831 rdmsrl(MSR_VM_CR, vm_cr);
2832 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2839 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2842 * Patch in the VMMCALL instruction:
2844 hypercall[0] = 0x0f;
2845 hypercall[1] = 0x01;
2846 hypercall[2] = 0xd9;
2849 static void svm_check_processor_compat(void *rtn)
2854 static bool svm_cpu_has_accelerated_tpr(void)
2859 static int get_npt_level(void)
2861 #ifdef CONFIG_X86_64
2862 return PT64_ROOT_LEVEL;
2864 return PT32E_ROOT_LEVEL;
2868 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
2873 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
2877 static const struct trace_print_flags svm_exit_reasons_str[] = {
2878 { SVM_EXIT_READ_CR0, "read_cr0" },
2879 { SVM_EXIT_READ_CR3, "read_cr3" },
2880 { SVM_EXIT_READ_CR4, "read_cr4" },
2881 { SVM_EXIT_READ_CR8, "read_cr8" },
2882 { SVM_EXIT_WRITE_CR0, "write_cr0" },
2883 { SVM_EXIT_WRITE_CR3, "write_cr3" },
2884 { SVM_EXIT_WRITE_CR4, "write_cr4" },
2885 { SVM_EXIT_WRITE_CR8, "write_cr8" },
2886 { SVM_EXIT_READ_DR0, "read_dr0" },
2887 { SVM_EXIT_READ_DR1, "read_dr1" },
2888 { SVM_EXIT_READ_DR2, "read_dr2" },
2889 { SVM_EXIT_READ_DR3, "read_dr3" },
2890 { SVM_EXIT_WRITE_DR0, "write_dr0" },
2891 { SVM_EXIT_WRITE_DR1, "write_dr1" },
2892 { SVM_EXIT_WRITE_DR2, "write_dr2" },
2893 { SVM_EXIT_WRITE_DR3, "write_dr3" },
2894 { SVM_EXIT_WRITE_DR5, "write_dr5" },
2895 { SVM_EXIT_WRITE_DR7, "write_dr7" },
2896 { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
2897 { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
2898 { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
2899 { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
2900 { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
2901 { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
2902 { SVM_EXIT_INTR, "interrupt" },
2903 { SVM_EXIT_NMI, "nmi" },
2904 { SVM_EXIT_SMI, "smi" },
2905 { SVM_EXIT_INIT, "init" },
2906 { SVM_EXIT_VINTR, "vintr" },
2907 { SVM_EXIT_CPUID, "cpuid" },
2908 { SVM_EXIT_INVD, "invd" },
2909 { SVM_EXIT_HLT, "hlt" },
2910 { SVM_EXIT_INVLPG, "invlpg" },
2911 { SVM_EXIT_INVLPGA, "invlpga" },
2912 { SVM_EXIT_IOIO, "io" },
2913 { SVM_EXIT_MSR, "msr" },
2914 { SVM_EXIT_TASK_SWITCH, "task_switch" },
2915 { SVM_EXIT_SHUTDOWN, "shutdown" },
2916 { SVM_EXIT_VMRUN, "vmrun" },
2917 { SVM_EXIT_VMMCALL, "hypercall" },
2918 { SVM_EXIT_VMLOAD, "vmload" },
2919 { SVM_EXIT_VMSAVE, "vmsave" },
2920 { SVM_EXIT_STGI, "stgi" },
2921 { SVM_EXIT_CLGI, "clgi" },
2922 { SVM_EXIT_SKINIT, "skinit" },
2923 { SVM_EXIT_WBINVD, "wbinvd" },
2924 { SVM_EXIT_MONITOR, "monitor" },
2925 { SVM_EXIT_MWAIT, "mwait" },
2926 { SVM_EXIT_NPF, "npf" },
2930 static int svm_get_lpage_level(void)
2932 return PT_PDPE_LEVEL;
2935 static bool svm_rdtscp_supported(void)
2940 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
2942 struct vcpu_svm *svm = to_svm(vcpu);
2944 update_cr0_intercept(svm);
2945 svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
2948 static struct kvm_x86_ops svm_x86_ops = {
2949 .cpu_has_kvm_support = has_svm,
2950 .disabled_by_bios = is_disabled,
2951 .hardware_setup = svm_hardware_setup,
2952 .hardware_unsetup = svm_hardware_unsetup,
2953 .check_processor_compatibility = svm_check_processor_compat,
2954 .hardware_enable = svm_hardware_enable,
2955 .hardware_disable = svm_hardware_disable,
2956 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2958 .vcpu_create = svm_create_vcpu,
2959 .vcpu_free = svm_free_vcpu,
2960 .vcpu_reset = svm_vcpu_reset,
2962 .prepare_guest_switch = svm_prepare_guest_switch,
2963 .vcpu_load = svm_vcpu_load,
2964 .vcpu_put = svm_vcpu_put,
2966 .set_guest_debug = svm_guest_debug,
2967 .get_msr = svm_get_msr,
2968 .set_msr = svm_set_msr,
2969 .get_segment_base = svm_get_segment_base,
2970 .get_segment = svm_get_segment,
2971 .set_segment = svm_set_segment,
2972 .get_cpl = svm_get_cpl,
2973 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2974 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
2975 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2976 .set_cr0 = svm_set_cr0,
2977 .set_cr3 = svm_set_cr3,
2978 .set_cr4 = svm_set_cr4,
2979 .set_efer = svm_set_efer,
2980 .get_idt = svm_get_idt,
2981 .set_idt = svm_set_idt,
2982 .get_gdt = svm_get_gdt,
2983 .set_gdt = svm_set_gdt,
2984 .get_dr = svm_get_dr,
2985 .set_dr = svm_set_dr,
2986 .cache_reg = svm_cache_reg,
2987 .get_rflags = svm_get_rflags,
2988 .set_rflags = svm_set_rflags,
2989 .fpu_activate = svm_fpu_activate,
2990 .fpu_deactivate = svm_fpu_deactivate,
2992 .tlb_flush = svm_flush_tlb,
2994 .run = svm_vcpu_run,
2995 .handle_exit = handle_exit,
2996 .skip_emulated_instruction = skip_emulated_instruction,
2997 .set_interrupt_shadow = svm_set_interrupt_shadow,
2998 .get_interrupt_shadow = svm_get_interrupt_shadow,
2999 .patch_hypercall = svm_patch_hypercall,
3000 .set_irq = svm_set_irq,
3001 .set_nmi = svm_inject_nmi,
3002 .queue_exception = svm_queue_exception,
3003 .interrupt_allowed = svm_interrupt_allowed,
3004 .nmi_allowed = svm_nmi_allowed,
3005 .get_nmi_mask = svm_get_nmi_mask,
3006 .set_nmi_mask = svm_set_nmi_mask,
3007 .enable_nmi_window = enable_nmi_window,
3008 .enable_irq_window = enable_irq_window,
3009 .update_cr8_intercept = update_cr8_intercept,
3011 .set_tss_addr = svm_set_tss_addr,
3012 .get_tdp_level = get_npt_level,
3013 .get_mt_mask = svm_get_mt_mask,
3015 .exit_reasons_str = svm_exit_reasons_str,
3016 .get_lpage_level = svm_get_lpage_level,
3018 .cpuid_update = svm_cpuid_update,
3020 .rdtscp_supported = svm_rdtscp_supported,
3023 static int __init svm_init(void)
3025 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3029 static void __exit svm_exit(void)
3034 module_init(svm_init)
3035 module_exit(svm_exit)