2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
17 #include <linux/kvm_host.h>
21 #include "kvm_cache_regs.h"
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/vmalloc.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
32 #include <asm/tlbflush.h>
34 #include <asm/kvm_para.h>
36 #include <asm/virtext.h>
39 #define __ex(x) __kvm_handle_fault_on_reboot(x)
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
44 #define IOPM_ALLOC_ORDER 2
45 #define MSRPM_ALLOC_ORDER 1
47 #define SEG_TYPE_LDT 2
48 #define SEG_TYPE_BUSY_TSS16 3
50 #define SVM_FEATURE_NPT (1 << 0)
51 #define SVM_FEATURE_LBRV (1 << 1)
52 #define SVM_FEATURE_SVML (1 << 2)
53 #define SVM_FEATURE_NRIP (1 << 3)
54 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
56 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
57 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
58 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
60 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
62 static bool erratum_383_found __read_mostly;
64 static const u32 host_save_user_msrs[] = {
66 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
69 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
72 #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 */
89 /* A VMEXIT is required but not yet emulated */
93 * If we vmexit during an instruction emulation we need this to restore
94 * the l1 guest rip after the emulation
96 unsigned long vmexit_rip;
97 unsigned long vmexit_rsp;
98 unsigned long vmexit_rax;
100 /* cache for intercepts of the guest */
101 u16 intercept_cr_read;
102 u16 intercept_cr_write;
103 u16 intercept_dr_read;
104 u16 intercept_dr_write;
105 u32 intercept_exceptions;
108 /* Nested Paging related state */
112 #define MSRPM_OFFSETS 16
113 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
116 struct kvm_vcpu vcpu;
118 unsigned long vmcb_pa;
119 struct svm_cpu_data *svm_data;
120 uint64_t asid_generation;
121 uint64_t sysenter_esp;
122 uint64_t sysenter_eip;
126 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
136 struct nested_state nested;
140 unsigned int3_injected;
141 unsigned long int3_rip;
145 #define MSR_INVALID 0xffffffffU
147 static struct svm_direct_access_msrs {
148 u32 index; /* Index of the MSR */
149 bool always; /* True if intercept is always on */
150 } direct_access_msrs[] = {
151 { .index = MSR_STAR, .always = true },
152 { .index = MSR_IA32_SYSENTER_CS, .always = true },
154 { .index = MSR_GS_BASE, .always = true },
155 { .index = MSR_FS_BASE, .always = true },
156 { .index = MSR_KERNEL_GS_BASE, .always = true },
157 { .index = MSR_LSTAR, .always = true },
158 { .index = MSR_CSTAR, .always = true },
159 { .index = MSR_SYSCALL_MASK, .always = true },
161 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
162 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
163 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
164 { .index = MSR_IA32_LASTINTTOIP, .always = false },
165 { .index = MSR_INVALID, .always = false },
168 /* enable NPT for AMD64 and X86 with PAE */
169 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
170 static bool npt_enabled = true;
172 static bool npt_enabled;
176 module_param(npt, int, S_IRUGO);
178 static int nested = 1;
179 module_param(nested, int, S_IRUGO);
181 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
182 static void svm_complete_interrupts(struct vcpu_svm *svm);
184 static int nested_svm_exit_handled(struct vcpu_svm *svm);
185 static int nested_svm_intercept(struct vcpu_svm *svm);
186 static int nested_svm_vmexit(struct vcpu_svm *svm);
187 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
188 bool has_error_code, u32 error_code);
190 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
192 return container_of(vcpu, struct vcpu_svm, vcpu);
195 static inline bool is_nested(struct vcpu_svm *svm)
197 return svm->nested.vmcb;
200 static inline void enable_gif(struct vcpu_svm *svm)
202 svm->vcpu.arch.hflags |= HF_GIF_MASK;
205 static inline void disable_gif(struct vcpu_svm *svm)
207 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
210 static inline bool gif_set(struct vcpu_svm *svm)
212 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
215 static unsigned long iopm_base;
217 struct kvm_ldttss_desc {
220 unsigned base1:8, type:5, dpl:2, p:1;
221 unsigned limit1:4, zero0:3, g:1, base2:8;
224 } __attribute__((packed));
226 struct svm_cpu_data {
232 struct kvm_ldttss_desc *tss_desc;
234 struct page *save_area;
237 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
238 static uint32_t svm_features;
240 struct svm_init_data {
245 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
247 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
248 #define MSRS_RANGE_SIZE 2048
249 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
251 static u32 svm_msrpm_offset(u32 msr)
256 for (i = 0; i < NUM_MSR_MAPS; i++) {
257 if (msr < msrpm_ranges[i] ||
258 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
261 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
262 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
264 /* Now we have the u8 offset - but need the u32 offset */
268 /* MSR not in any range */
272 #define MAX_INST_SIZE 15
274 static inline void clgi(void)
276 asm volatile (__ex(SVM_CLGI));
279 static inline void stgi(void)
281 asm volatile (__ex(SVM_STGI));
284 static inline void invlpga(unsigned long addr, u32 asid)
286 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
289 static inline void force_new_asid(struct kvm_vcpu *vcpu)
291 to_svm(vcpu)->asid_generation--;
294 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
296 force_new_asid(vcpu);
299 static int get_npt_level(void)
302 return PT64_ROOT_LEVEL;
304 return PT32E_ROOT_LEVEL;
308 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
310 vcpu->arch.efer = efer;
311 if (!npt_enabled && !(efer & EFER_LMA))
314 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
317 static int is_external_interrupt(u32 info)
319 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
320 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
323 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
325 struct vcpu_svm *svm = to_svm(vcpu);
328 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
329 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
333 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
335 struct vcpu_svm *svm = to_svm(vcpu);
338 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
340 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
344 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
346 struct vcpu_svm *svm = to_svm(vcpu);
348 if (svm->vmcb->control.next_rip != 0)
349 svm->next_rip = svm->vmcb->control.next_rip;
351 if (!svm->next_rip) {
352 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
354 printk(KERN_DEBUG "%s: NOP\n", __func__);
357 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
358 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
359 __func__, kvm_rip_read(vcpu), svm->next_rip);
361 kvm_rip_write(vcpu, svm->next_rip);
362 svm_set_interrupt_shadow(vcpu, 0);
365 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
366 bool has_error_code, u32 error_code,
369 struct vcpu_svm *svm = to_svm(vcpu);
372 * If we are within a nested VM we'd better #VMEXIT and let the guest
373 * handle the exception
376 nested_svm_check_exception(svm, nr, has_error_code, error_code))
379 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
380 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
383 * For guest debugging where we have to reinject #BP if some
384 * INT3 is guest-owned:
385 * Emulate nRIP by moving RIP forward. Will fail if injection
386 * raises a fault that is not intercepted. Still better than
387 * failing in all cases.
389 skip_emulated_instruction(&svm->vcpu);
390 rip = kvm_rip_read(&svm->vcpu);
391 svm->int3_rip = rip + svm->vmcb->save.cs.base;
392 svm->int3_injected = rip - old_rip;
395 svm->vmcb->control.event_inj = nr
397 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
398 | SVM_EVTINJ_TYPE_EXEPT;
399 svm->vmcb->control.event_inj_err = error_code;
402 static void svm_init_erratum_383(void)
408 if (!cpu_has_amd_erratum(amd_erratum_383))
411 /* Use _safe variants to not break nested virtualization */
412 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
418 low = lower_32_bits(val);
419 high = upper_32_bits(val);
421 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
423 erratum_383_found = true;
426 static int has_svm(void)
430 if (!cpu_has_svm(&msg)) {
431 printk(KERN_INFO "has_svm: %s\n", msg);
438 static void svm_hardware_disable(void *garbage)
443 static int svm_hardware_enable(void *garbage)
446 struct svm_cpu_data *sd;
448 struct desc_ptr gdt_descr;
449 struct desc_struct *gdt;
450 int me = raw_smp_processor_id();
452 rdmsrl(MSR_EFER, efer);
453 if (efer & EFER_SVME)
457 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
461 sd = per_cpu(svm_data, me);
464 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
469 sd->asid_generation = 1;
470 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
471 sd->next_asid = sd->max_asid + 1;
473 native_store_gdt(&gdt_descr);
474 gdt = (struct desc_struct *)gdt_descr.address;
475 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
477 wrmsrl(MSR_EFER, efer | EFER_SVME);
479 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
481 svm_init_erratum_383();
486 static void svm_cpu_uninit(int cpu)
488 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
493 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
494 __free_page(sd->save_area);
498 static int svm_cpu_init(int cpu)
500 struct svm_cpu_data *sd;
503 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
507 sd->save_area = alloc_page(GFP_KERNEL);
512 per_cpu(svm_data, cpu) = sd;
522 static bool valid_msr_intercept(u32 index)
526 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
527 if (direct_access_msrs[i].index == index)
533 static void set_msr_interception(u32 *msrpm, unsigned msr,
536 u8 bit_read, bit_write;
541 * If this warning triggers extend the direct_access_msrs list at the
542 * beginning of the file
544 WARN_ON(!valid_msr_intercept(msr));
546 offset = svm_msrpm_offset(msr);
547 bit_read = 2 * (msr & 0x0f);
548 bit_write = 2 * (msr & 0x0f) + 1;
551 BUG_ON(offset == MSR_INVALID);
553 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
554 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
559 static void svm_vcpu_init_msrpm(u32 *msrpm)
563 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
565 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
566 if (!direct_access_msrs[i].always)
569 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
573 static void add_msr_offset(u32 offset)
577 for (i = 0; i < MSRPM_OFFSETS; ++i) {
579 /* Offset already in list? */
580 if (msrpm_offsets[i] == offset)
583 /* Slot used by another offset? */
584 if (msrpm_offsets[i] != MSR_INVALID)
587 /* Add offset to list */
588 msrpm_offsets[i] = offset;
594 * If this BUG triggers the msrpm_offsets table has an overflow. Just
595 * increase MSRPM_OFFSETS in this case.
600 static void init_msrpm_offsets(void)
604 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
606 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
609 offset = svm_msrpm_offset(direct_access_msrs[i].index);
610 BUG_ON(offset == MSR_INVALID);
612 add_msr_offset(offset);
616 static void svm_enable_lbrv(struct vcpu_svm *svm)
618 u32 *msrpm = svm->msrpm;
620 svm->vmcb->control.lbr_ctl = 1;
621 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
622 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
623 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
624 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
627 static void svm_disable_lbrv(struct vcpu_svm *svm)
629 u32 *msrpm = svm->msrpm;
631 svm->vmcb->control.lbr_ctl = 0;
632 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
633 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
634 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
635 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
638 static __init int svm_hardware_setup(void)
641 struct page *iopm_pages;
645 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
650 iopm_va = page_address(iopm_pages);
651 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
652 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
654 init_msrpm_offsets();
656 if (boot_cpu_has(X86_FEATURE_NX))
657 kvm_enable_efer_bits(EFER_NX);
659 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
660 kvm_enable_efer_bits(EFER_FFXSR);
663 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
664 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
667 for_each_possible_cpu(cpu) {
668 r = svm_cpu_init(cpu);
673 svm_features = cpuid_edx(SVM_CPUID_FUNC);
675 if (!boot_cpu_has(X86_FEATURE_NPT))
678 if (npt_enabled && !npt) {
679 printk(KERN_INFO "kvm: Nested Paging disabled\n");
684 printk(KERN_INFO "kvm: Nested Paging enabled\n");
692 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
697 static __exit void svm_hardware_unsetup(void)
701 for_each_possible_cpu(cpu)
704 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
708 static void init_seg(struct vmcb_seg *seg)
711 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
712 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
717 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
720 seg->attrib = SVM_SELECTOR_P_MASK | type;
725 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
727 struct vcpu_svm *svm = to_svm(vcpu);
728 u64 g_tsc_offset = 0;
730 if (is_nested(svm)) {
731 g_tsc_offset = svm->vmcb->control.tsc_offset -
732 svm->nested.hsave->control.tsc_offset;
733 svm->nested.hsave->control.tsc_offset = offset;
736 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
739 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
741 struct vcpu_svm *svm = to_svm(vcpu);
743 svm->vmcb->control.tsc_offset += adjustment;
745 svm->nested.hsave->control.tsc_offset += adjustment;
748 static void init_vmcb(struct vcpu_svm *svm)
750 struct vmcb_control_area *control = &svm->vmcb->control;
751 struct vmcb_save_area *save = &svm->vmcb->save;
753 svm->vcpu.fpu_active = 1;
755 control->intercept_cr_read = INTERCEPT_CR0_MASK |
759 control->intercept_cr_write = INTERCEPT_CR0_MASK |
764 control->intercept_dr_read = INTERCEPT_DR0_MASK |
773 control->intercept_dr_write = INTERCEPT_DR0_MASK |
782 control->intercept_exceptions = (1 << PF_VECTOR) |
787 control->intercept = (1ULL << INTERCEPT_INTR) |
788 (1ULL << INTERCEPT_NMI) |
789 (1ULL << INTERCEPT_SMI) |
790 (1ULL << INTERCEPT_SELECTIVE_CR0) |
791 (1ULL << INTERCEPT_CPUID) |
792 (1ULL << INTERCEPT_INVD) |
793 (1ULL << INTERCEPT_HLT) |
794 (1ULL << INTERCEPT_INVLPG) |
795 (1ULL << INTERCEPT_INVLPGA) |
796 (1ULL << INTERCEPT_IOIO_PROT) |
797 (1ULL << INTERCEPT_MSR_PROT) |
798 (1ULL << INTERCEPT_TASK_SWITCH) |
799 (1ULL << INTERCEPT_SHUTDOWN) |
800 (1ULL << INTERCEPT_VMRUN) |
801 (1ULL << INTERCEPT_VMMCALL) |
802 (1ULL << INTERCEPT_VMLOAD) |
803 (1ULL << INTERCEPT_VMSAVE) |
804 (1ULL << INTERCEPT_STGI) |
805 (1ULL << INTERCEPT_CLGI) |
806 (1ULL << INTERCEPT_SKINIT) |
807 (1ULL << INTERCEPT_WBINVD) |
808 (1ULL << INTERCEPT_MONITOR) |
809 (1ULL << INTERCEPT_MWAIT);
811 control->iopm_base_pa = iopm_base;
812 control->msrpm_base_pa = __pa(svm->msrpm);
813 control->int_ctl = V_INTR_MASKING_MASK;
821 save->cs.selector = 0xf000;
822 /* Executable/Readable Code Segment */
823 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
824 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
825 save->cs.limit = 0xffff;
827 * cs.base should really be 0xffff0000, but vmx can't handle that, so
828 * be consistent with it.
830 * Replace when we have real mode working for vmx.
832 save->cs.base = 0xf0000;
834 save->gdtr.limit = 0xffff;
835 save->idtr.limit = 0xffff;
837 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
838 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
840 svm_set_efer(&svm->vcpu, 0);
841 save->dr6 = 0xffff0ff0;
844 save->rip = 0x0000fff0;
845 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
848 * This is the guest-visible cr0 value.
849 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
851 svm->vcpu.arch.cr0 = 0;
852 (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
854 save->cr4 = X86_CR4_PAE;
858 /* Setup VMCB for Nested Paging */
859 control->nested_ctl = 1;
860 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
861 (1ULL << INTERCEPT_INVLPG));
862 control->intercept_exceptions &= ~(1 << PF_VECTOR);
863 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
864 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
865 save->g_pat = 0x0007040600070406ULL;
869 force_new_asid(&svm->vcpu);
871 svm->nested.vmcb = 0;
872 svm->vcpu.arch.hflags = 0;
874 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
875 control->pause_filter_count = 3000;
876 control->intercept |= (1ULL << INTERCEPT_PAUSE);
882 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
884 struct vcpu_svm *svm = to_svm(vcpu);
888 if (!kvm_vcpu_is_bsp(vcpu)) {
889 kvm_rip_write(vcpu, 0);
890 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
891 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
893 vcpu->arch.regs_avail = ~0;
894 vcpu->arch.regs_dirty = ~0;
899 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
901 struct vcpu_svm *svm;
903 struct page *msrpm_pages;
904 struct page *hsave_page;
905 struct page *nested_msrpm_pages;
908 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
914 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
919 page = alloc_page(GFP_KERNEL);
923 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
927 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
928 if (!nested_msrpm_pages)
931 hsave_page = alloc_page(GFP_KERNEL);
935 svm->nested.hsave = page_address(hsave_page);
937 svm->msrpm = page_address(msrpm_pages);
938 svm_vcpu_init_msrpm(svm->msrpm);
940 svm->nested.msrpm = page_address(nested_msrpm_pages);
941 svm_vcpu_init_msrpm(svm->nested.msrpm);
943 svm->vmcb = page_address(page);
944 clear_page(svm->vmcb);
945 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
946 svm->asid_generation = 0;
948 kvm_write_tsc(&svm->vcpu, 0);
950 err = fx_init(&svm->vcpu);
954 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
955 if (kvm_vcpu_is_bsp(&svm->vcpu))
956 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
961 __free_page(hsave_page);
963 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
965 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
969 kvm_vcpu_uninit(&svm->vcpu);
971 kmem_cache_free(kvm_vcpu_cache, svm);
976 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
978 struct vcpu_svm *svm = to_svm(vcpu);
980 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
981 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
982 __free_page(virt_to_page(svm->nested.hsave));
983 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
984 kvm_vcpu_uninit(vcpu);
985 kmem_cache_free(kvm_vcpu_cache, svm);
988 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
990 struct vcpu_svm *svm = to_svm(vcpu);
993 if (unlikely(cpu != vcpu->cpu)) {
994 svm->asid_generation = 0;
998 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1000 savesegment(fs, svm->host.fs);
1001 savesegment(gs, svm->host.gs);
1002 svm->host.ldt = kvm_read_ldt();
1004 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1005 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1008 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1010 struct vcpu_svm *svm = to_svm(vcpu);
1013 ++vcpu->stat.host_state_reload;
1014 kvm_load_ldt(svm->host.ldt);
1015 #ifdef CONFIG_X86_64
1016 loadsegment(fs, svm->host.fs);
1017 load_gs_index(svm->host.gs);
1018 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1020 loadsegment(gs, svm->host.gs);
1022 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1023 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1026 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1028 return to_svm(vcpu)->vmcb->save.rflags;
1031 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1033 to_svm(vcpu)->vmcb->save.rflags = rflags;
1036 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1039 case VCPU_EXREG_PDPTR:
1040 BUG_ON(!npt_enabled);
1041 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
1048 static void svm_set_vintr(struct vcpu_svm *svm)
1050 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
1053 static void svm_clear_vintr(struct vcpu_svm *svm)
1055 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1058 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1060 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1063 case VCPU_SREG_CS: return &save->cs;
1064 case VCPU_SREG_DS: return &save->ds;
1065 case VCPU_SREG_ES: return &save->es;
1066 case VCPU_SREG_FS: return &save->fs;
1067 case VCPU_SREG_GS: return &save->gs;
1068 case VCPU_SREG_SS: return &save->ss;
1069 case VCPU_SREG_TR: return &save->tr;
1070 case VCPU_SREG_LDTR: return &save->ldtr;
1076 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1078 struct vmcb_seg *s = svm_seg(vcpu, seg);
1083 static void svm_get_segment(struct kvm_vcpu *vcpu,
1084 struct kvm_segment *var, int seg)
1086 struct vmcb_seg *s = svm_seg(vcpu, seg);
1088 var->base = s->base;
1089 var->limit = s->limit;
1090 var->selector = s->selector;
1091 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1092 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1093 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1094 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1095 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1096 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1097 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1098 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1101 * AMD's VMCB does not have an explicit unusable field, so emulate it
1102 * for cross vendor migration purposes by "not present"
1104 var->unusable = !var->present || (var->type == 0);
1109 * SVM always stores 0 for the 'G' bit in the CS selector in
1110 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1111 * Intel's VMENTRY has a check on the 'G' bit.
1113 var->g = s->limit > 0xfffff;
1117 * Work around a bug where the busy flag in the tr selector
1127 * The accessed bit must always be set in the segment
1128 * descriptor cache, although it can be cleared in the
1129 * descriptor, the cached bit always remains at 1. Since
1130 * Intel has a check on this, set it here to support
1131 * cross-vendor migration.
1138 * On AMD CPUs sometimes the DB bit in the segment
1139 * descriptor is left as 1, although the whole segment has
1140 * been made unusable. Clear it here to pass an Intel VMX
1141 * entry check when cross vendor migrating.
1149 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1151 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1156 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1158 struct vcpu_svm *svm = to_svm(vcpu);
1160 dt->size = svm->vmcb->save.idtr.limit;
1161 dt->address = svm->vmcb->save.idtr.base;
1164 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1166 struct vcpu_svm *svm = to_svm(vcpu);
1168 svm->vmcb->save.idtr.limit = dt->size;
1169 svm->vmcb->save.idtr.base = dt->address ;
1172 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1174 struct vcpu_svm *svm = to_svm(vcpu);
1176 dt->size = svm->vmcb->save.gdtr.limit;
1177 dt->address = svm->vmcb->save.gdtr.base;
1180 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1182 struct vcpu_svm *svm = to_svm(vcpu);
1184 svm->vmcb->save.gdtr.limit = dt->size;
1185 svm->vmcb->save.gdtr.base = dt->address ;
1188 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1192 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1196 static void update_cr0_intercept(struct vcpu_svm *svm)
1198 struct vmcb *vmcb = svm->vmcb;
1199 ulong gcr0 = svm->vcpu.arch.cr0;
1200 u64 *hcr0 = &svm->vmcb->save.cr0;
1202 if (!svm->vcpu.fpu_active)
1203 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1205 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1206 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1209 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1210 vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1211 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1212 if (is_nested(svm)) {
1213 struct vmcb *hsave = svm->nested.hsave;
1215 hsave->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1216 hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1217 vmcb->control.intercept_cr_read |= svm->nested.intercept_cr_read;
1218 vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write;
1221 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1222 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1223 if (is_nested(svm)) {
1224 struct vmcb *hsave = svm->nested.hsave;
1226 hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1227 hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1232 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1234 struct vcpu_svm *svm = to_svm(vcpu);
1236 if (is_nested(svm)) {
1238 * We are here because we run in nested mode, the host kvm
1239 * intercepts cr0 writes but the l1 hypervisor does not.
1240 * But the L1 hypervisor may intercept selective cr0 writes.
1241 * This needs to be checked here.
1243 unsigned long old, new;
1245 /* Remove bits that would trigger a real cr0 write intercept */
1246 old = vcpu->arch.cr0 & SVM_CR0_SELECTIVE_MASK;
1247 new = cr0 & SVM_CR0_SELECTIVE_MASK;
1250 /* cr0 write with ts and mp unchanged */
1251 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
1252 if (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE) {
1253 svm->nested.vmexit_rip = kvm_rip_read(vcpu);
1254 svm->nested.vmexit_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
1255 svm->nested.vmexit_rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
1261 #ifdef CONFIG_X86_64
1262 if (vcpu->arch.efer & EFER_LME) {
1263 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1264 vcpu->arch.efer |= EFER_LMA;
1265 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1268 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1269 vcpu->arch.efer &= ~EFER_LMA;
1270 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1274 vcpu->arch.cr0 = cr0;
1277 cr0 |= X86_CR0_PG | X86_CR0_WP;
1279 if (!vcpu->fpu_active)
1282 * re-enable caching here because the QEMU bios
1283 * does not do it - this results in some delay at
1286 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1287 svm->vmcb->save.cr0 = cr0;
1288 update_cr0_intercept(svm);
1291 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1293 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1294 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1296 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1297 force_new_asid(vcpu);
1299 vcpu->arch.cr4 = cr4;
1302 cr4 |= host_cr4_mce;
1303 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1306 static void svm_set_segment(struct kvm_vcpu *vcpu,
1307 struct kvm_segment *var, int seg)
1309 struct vcpu_svm *svm = to_svm(vcpu);
1310 struct vmcb_seg *s = svm_seg(vcpu, seg);
1312 s->base = var->base;
1313 s->limit = var->limit;
1314 s->selector = var->selector;
1318 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1319 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1320 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1321 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1322 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1323 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1324 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1325 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1327 if (seg == VCPU_SREG_CS)
1329 = (svm->vmcb->save.cs.attrib
1330 >> SVM_SELECTOR_DPL_SHIFT) & 3;
1334 static void update_db_intercept(struct kvm_vcpu *vcpu)
1336 struct vcpu_svm *svm = to_svm(vcpu);
1338 svm->vmcb->control.intercept_exceptions &=
1339 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1341 if (svm->nmi_singlestep)
1342 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1344 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1345 if (vcpu->guest_debug &
1346 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1347 svm->vmcb->control.intercept_exceptions |=
1349 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1350 svm->vmcb->control.intercept_exceptions |=
1353 vcpu->guest_debug = 0;
1356 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1358 struct vcpu_svm *svm = to_svm(vcpu);
1360 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1361 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1363 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1365 update_db_intercept(vcpu);
1368 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1370 if (sd->next_asid > sd->max_asid) {
1371 ++sd->asid_generation;
1373 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1376 svm->asid_generation = sd->asid_generation;
1377 svm->vmcb->control.asid = sd->next_asid++;
1380 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1382 struct vcpu_svm *svm = to_svm(vcpu);
1384 svm->vmcb->save.dr7 = value;
1387 static int pf_interception(struct vcpu_svm *svm)
1389 u64 fault_address = svm->vmcb->control.exit_info_2;
1393 switch (svm->apf_reason) {
1395 error_code = svm->vmcb->control.exit_info_1;
1397 trace_kvm_page_fault(fault_address, error_code);
1398 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1399 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1400 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1402 case KVM_PV_REASON_PAGE_NOT_PRESENT:
1403 svm->apf_reason = 0;
1404 local_irq_disable();
1405 kvm_async_pf_task_wait(fault_address);
1408 case KVM_PV_REASON_PAGE_READY:
1409 svm->apf_reason = 0;
1410 local_irq_disable();
1411 kvm_async_pf_task_wake(fault_address);
1418 static int db_interception(struct vcpu_svm *svm)
1420 struct kvm_run *kvm_run = svm->vcpu.run;
1422 if (!(svm->vcpu.guest_debug &
1423 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1424 !svm->nmi_singlestep) {
1425 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1429 if (svm->nmi_singlestep) {
1430 svm->nmi_singlestep = false;
1431 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1432 svm->vmcb->save.rflags &=
1433 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1434 update_db_intercept(&svm->vcpu);
1437 if (svm->vcpu.guest_debug &
1438 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1439 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1440 kvm_run->debug.arch.pc =
1441 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1442 kvm_run->debug.arch.exception = DB_VECTOR;
1449 static int bp_interception(struct vcpu_svm *svm)
1451 struct kvm_run *kvm_run = svm->vcpu.run;
1453 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1454 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1455 kvm_run->debug.arch.exception = BP_VECTOR;
1459 static int ud_interception(struct vcpu_svm *svm)
1463 er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1464 if (er != EMULATE_DONE)
1465 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1469 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1471 struct vcpu_svm *svm = to_svm(vcpu);
1474 if (is_nested(svm)) {
1477 h_excp = svm->nested.hsave->control.intercept_exceptions;
1478 n_excp = svm->nested.intercept_exceptions;
1479 h_excp &= ~(1 << NM_VECTOR);
1480 excp = h_excp | n_excp;
1482 excp = svm->vmcb->control.intercept_exceptions;
1483 excp &= ~(1 << NM_VECTOR);
1486 svm->vmcb->control.intercept_exceptions = excp;
1488 svm->vcpu.fpu_active = 1;
1489 update_cr0_intercept(svm);
1492 static int nm_interception(struct vcpu_svm *svm)
1494 svm_fpu_activate(&svm->vcpu);
1498 static bool is_erratum_383(void)
1503 if (!erratum_383_found)
1506 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1510 /* Bit 62 may or may not be set for this mce */
1511 value &= ~(1ULL << 62);
1513 if (value != 0xb600000000010015ULL)
1516 /* Clear MCi_STATUS registers */
1517 for (i = 0; i < 6; ++i)
1518 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1520 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1524 value &= ~(1ULL << 2);
1525 low = lower_32_bits(value);
1526 high = upper_32_bits(value);
1528 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1531 /* Flush tlb to evict multi-match entries */
1537 static void svm_handle_mce(struct vcpu_svm *svm)
1539 if (is_erratum_383()) {
1541 * Erratum 383 triggered. Guest state is corrupt so kill the
1544 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1546 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1552 * On an #MC intercept the MCE handler is not called automatically in
1553 * the host. So do it by hand here.
1557 /* not sure if we ever come back to this point */
1562 static int mc_interception(struct vcpu_svm *svm)
1567 static int shutdown_interception(struct vcpu_svm *svm)
1569 struct kvm_run *kvm_run = svm->vcpu.run;
1572 * VMCB is undefined after a SHUTDOWN intercept
1573 * so reinitialize it.
1575 clear_page(svm->vmcb);
1578 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1582 static int io_interception(struct vcpu_svm *svm)
1584 struct kvm_vcpu *vcpu = &svm->vcpu;
1585 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1586 int size, in, string;
1589 ++svm->vcpu.stat.io_exits;
1590 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1591 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1593 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
1595 port = io_info >> 16;
1596 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1597 svm->next_rip = svm->vmcb->control.exit_info_2;
1598 skip_emulated_instruction(&svm->vcpu);
1600 return kvm_fast_pio_out(vcpu, size, port);
1603 static int nmi_interception(struct vcpu_svm *svm)
1608 static int intr_interception(struct vcpu_svm *svm)
1610 ++svm->vcpu.stat.irq_exits;
1614 static int nop_on_interception(struct vcpu_svm *svm)
1619 static int halt_interception(struct vcpu_svm *svm)
1621 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1622 skip_emulated_instruction(&svm->vcpu);
1623 return kvm_emulate_halt(&svm->vcpu);
1626 static int vmmcall_interception(struct vcpu_svm *svm)
1628 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1629 skip_emulated_instruction(&svm->vcpu);
1630 kvm_emulate_hypercall(&svm->vcpu);
1634 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1636 struct vcpu_svm *svm = to_svm(vcpu);
1638 return svm->nested.nested_cr3;
1641 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1644 struct vcpu_svm *svm = to_svm(vcpu);
1646 svm->vmcb->control.nested_cr3 = root;
1647 force_new_asid(vcpu);
1650 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu)
1652 struct vcpu_svm *svm = to_svm(vcpu);
1654 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1655 svm->vmcb->control.exit_code_hi = 0;
1656 svm->vmcb->control.exit_info_1 = vcpu->arch.fault.error_code;
1657 svm->vmcb->control.exit_info_2 = vcpu->arch.fault.address;
1659 nested_svm_vmexit(svm);
1662 static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1666 r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1668 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
1669 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
1670 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1671 vcpu->arch.mmu.shadow_root_level = get_npt_level();
1672 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
1677 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1679 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1682 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1684 if (!(svm->vcpu.arch.efer & EFER_SVME)
1685 || !is_paging(&svm->vcpu)) {
1686 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1690 if (svm->vmcb->save.cpl) {
1691 kvm_inject_gp(&svm->vcpu, 0);
1698 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1699 bool has_error_code, u32 error_code)
1703 if (!is_nested(svm))
1706 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1707 svm->vmcb->control.exit_code_hi = 0;
1708 svm->vmcb->control.exit_info_1 = error_code;
1709 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1711 vmexit = nested_svm_intercept(svm);
1712 if (vmexit == NESTED_EXIT_DONE)
1713 svm->nested.exit_required = true;
1718 /* This function returns true if it is save to enable the irq window */
1719 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1721 if (!is_nested(svm))
1724 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1727 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1731 * if vmexit was already requested (by intercepted exception
1732 * for instance) do not overwrite it with "external interrupt"
1735 if (svm->nested.exit_required)
1738 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1739 svm->vmcb->control.exit_info_1 = 0;
1740 svm->vmcb->control.exit_info_2 = 0;
1742 if (svm->nested.intercept & 1ULL) {
1744 * The #vmexit can't be emulated here directly because this
1745 * code path runs with irqs and preemtion disabled. A
1746 * #vmexit emulation might sleep. Only signal request for
1749 svm->nested.exit_required = true;
1750 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1757 /* This function returns true if it is save to enable the nmi window */
1758 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1760 if (!is_nested(svm))
1763 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1766 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1767 svm->nested.exit_required = true;
1772 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1778 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1779 if (is_error_page(page))
1787 kvm_release_page_clean(page);
1788 kvm_inject_gp(&svm->vcpu, 0);
1793 static void nested_svm_unmap(struct page *page)
1796 kvm_release_page_dirty(page);
1799 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1805 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
1806 return NESTED_EXIT_HOST;
1808 port = svm->vmcb->control.exit_info_1 >> 16;
1809 gpa = svm->nested.vmcb_iopm + (port / 8);
1813 if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
1816 return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1819 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1821 u32 offset, msr, value;
1824 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1825 return NESTED_EXIT_HOST;
1827 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1828 offset = svm_msrpm_offset(msr);
1829 write = svm->vmcb->control.exit_info_1 & 1;
1830 mask = 1 << ((2 * (msr & 0xf)) + write);
1832 if (offset == MSR_INVALID)
1833 return NESTED_EXIT_DONE;
1835 /* Offset is in 32 bit units but need in 8 bit units */
1838 if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
1839 return NESTED_EXIT_DONE;
1841 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1844 static int nested_svm_exit_special(struct vcpu_svm *svm)
1846 u32 exit_code = svm->vmcb->control.exit_code;
1848 switch (exit_code) {
1851 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
1852 return NESTED_EXIT_HOST;
1854 /* For now we are always handling NPFs when using them */
1856 return NESTED_EXIT_HOST;
1858 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1859 /* When we're shadowing, trap PFs, but not async PF */
1860 if (!npt_enabled && svm->apf_reason == 0)
1861 return NESTED_EXIT_HOST;
1863 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
1864 nm_interception(svm);
1870 return NESTED_EXIT_CONTINUE;
1874 * If this function returns true, this #vmexit was already handled
1876 static int nested_svm_intercept(struct vcpu_svm *svm)
1878 u32 exit_code = svm->vmcb->control.exit_code;
1879 int vmexit = NESTED_EXIT_HOST;
1881 switch (exit_code) {
1883 vmexit = nested_svm_exit_handled_msr(svm);
1886 vmexit = nested_svm_intercept_ioio(svm);
1888 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1889 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1890 if (svm->nested.intercept_cr_read & cr_bits)
1891 vmexit = NESTED_EXIT_DONE;
1894 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1895 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1896 if (svm->nested.intercept_cr_write & cr_bits)
1897 vmexit = NESTED_EXIT_DONE;
1900 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1901 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1902 if (svm->nested.intercept_dr_read & dr_bits)
1903 vmexit = NESTED_EXIT_DONE;
1906 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1907 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1908 if (svm->nested.intercept_dr_write & dr_bits)
1909 vmexit = NESTED_EXIT_DONE;
1912 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1913 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1914 if (svm->nested.intercept_exceptions & excp_bits)
1915 vmexit = NESTED_EXIT_DONE;
1916 /* async page fault always cause vmexit */
1917 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
1918 svm->apf_reason != 0)
1919 vmexit = NESTED_EXIT_DONE;
1922 case SVM_EXIT_ERR: {
1923 vmexit = NESTED_EXIT_DONE;
1927 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1928 if (svm->nested.intercept & exit_bits)
1929 vmexit = NESTED_EXIT_DONE;
1936 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1940 vmexit = nested_svm_intercept(svm);
1942 if (vmexit == NESTED_EXIT_DONE)
1943 nested_svm_vmexit(svm);
1948 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1950 struct vmcb_control_area *dst = &dst_vmcb->control;
1951 struct vmcb_control_area *from = &from_vmcb->control;
1953 dst->intercept_cr_read = from->intercept_cr_read;
1954 dst->intercept_cr_write = from->intercept_cr_write;
1955 dst->intercept_dr_read = from->intercept_dr_read;
1956 dst->intercept_dr_write = from->intercept_dr_write;
1957 dst->intercept_exceptions = from->intercept_exceptions;
1958 dst->intercept = from->intercept;
1959 dst->iopm_base_pa = from->iopm_base_pa;
1960 dst->msrpm_base_pa = from->msrpm_base_pa;
1961 dst->tsc_offset = from->tsc_offset;
1962 dst->asid = from->asid;
1963 dst->tlb_ctl = from->tlb_ctl;
1964 dst->int_ctl = from->int_ctl;
1965 dst->int_vector = from->int_vector;
1966 dst->int_state = from->int_state;
1967 dst->exit_code = from->exit_code;
1968 dst->exit_code_hi = from->exit_code_hi;
1969 dst->exit_info_1 = from->exit_info_1;
1970 dst->exit_info_2 = from->exit_info_2;
1971 dst->exit_int_info = from->exit_int_info;
1972 dst->exit_int_info_err = from->exit_int_info_err;
1973 dst->nested_ctl = from->nested_ctl;
1974 dst->event_inj = from->event_inj;
1975 dst->event_inj_err = from->event_inj_err;
1976 dst->nested_cr3 = from->nested_cr3;
1977 dst->lbr_ctl = from->lbr_ctl;
1980 static int nested_svm_vmexit(struct vcpu_svm *svm)
1982 struct vmcb *nested_vmcb;
1983 struct vmcb *hsave = svm->nested.hsave;
1984 struct vmcb *vmcb = svm->vmcb;
1987 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1988 vmcb->control.exit_info_1,
1989 vmcb->control.exit_info_2,
1990 vmcb->control.exit_int_info,
1991 vmcb->control.exit_int_info_err);
1993 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
1997 /* Exit nested SVM mode */
1998 svm->nested.vmcb = 0;
2000 /* Give the current vmcb to the guest */
2003 nested_vmcb->save.es = vmcb->save.es;
2004 nested_vmcb->save.cs = vmcb->save.cs;
2005 nested_vmcb->save.ss = vmcb->save.ss;
2006 nested_vmcb->save.ds = vmcb->save.ds;
2007 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2008 nested_vmcb->save.idtr = vmcb->save.idtr;
2009 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2010 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2011 nested_vmcb->save.cr3 = svm->vcpu.arch.cr3;
2012 nested_vmcb->save.cr2 = vmcb->save.cr2;
2013 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2014 nested_vmcb->save.rflags = vmcb->save.rflags;
2015 nested_vmcb->save.rip = vmcb->save.rip;
2016 nested_vmcb->save.rsp = vmcb->save.rsp;
2017 nested_vmcb->save.rax = vmcb->save.rax;
2018 nested_vmcb->save.dr7 = vmcb->save.dr7;
2019 nested_vmcb->save.dr6 = vmcb->save.dr6;
2020 nested_vmcb->save.cpl = vmcb->save.cpl;
2022 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2023 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2024 nested_vmcb->control.int_state = vmcb->control.int_state;
2025 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2026 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2027 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2028 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2029 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2030 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2031 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2034 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2035 * to make sure that we do not lose injected events. So check event_inj
2036 * here and copy it to exit_int_info if it is valid.
2037 * Exit_int_info and event_inj can't be both valid because the case
2038 * below only happens on a VMRUN instruction intercept which has
2039 * no valid exit_int_info set.
2041 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2042 struct vmcb_control_area *nc = &nested_vmcb->control;
2044 nc->exit_int_info = vmcb->control.event_inj;
2045 nc->exit_int_info_err = vmcb->control.event_inj_err;
2048 nested_vmcb->control.tlb_ctl = 0;
2049 nested_vmcb->control.event_inj = 0;
2050 nested_vmcb->control.event_inj_err = 0;
2052 /* We always set V_INTR_MASKING and remember the old value in hflags */
2053 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2054 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2056 /* Restore the original control entries */
2057 copy_vmcb_control_area(vmcb, hsave);
2059 kvm_clear_exception_queue(&svm->vcpu);
2060 kvm_clear_interrupt_queue(&svm->vcpu);
2062 svm->nested.nested_cr3 = 0;
2064 /* Restore selected save entries */
2065 svm->vmcb->save.es = hsave->save.es;
2066 svm->vmcb->save.cs = hsave->save.cs;
2067 svm->vmcb->save.ss = hsave->save.ss;
2068 svm->vmcb->save.ds = hsave->save.ds;
2069 svm->vmcb->save.gdtr = hsave->save.gdtr;
2070 svm->vmcb->save.idtr = hsave->save.idtr;
2071 svm->vmcb->save.rflags = hsave->save.rflags;
2072 svm_set_efer(&svm->vcpu, hsave->save.efer);
2073 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2074 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2076 svm->vmcb->save.cr3 = hsave->save.cr3;
2077 svm->vcpu.arch.cr3 = hsave->save.cr3;
2079 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2081 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2082 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2083 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2084 svm->vmcb->save.dr7 = 0;
2085 svm->vmcb->save.cpl = 0;
2086 svm->vmcb->control.exit_int_info = 0;
2088 nested_svm_unmap(page);
2090 nested_svm_uninit_mmu_context(&svm->vcpu);
2091 kvm_mmu_reset_context(&svm->vcpu);
2092 kvm_mmu_load(&svm->vcpu);
2097 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2100 * This function merges the msr permission bitmaps of kvm and the
2101 * nested vmcb. It is omptimized in that it only merges the parts where
2102 * the kvm msr permission bitmap may contain zero bits
2106 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2109 for (i = 0; i < MSRPM_OFFSETS; i++) {
2113 if (msrpm_offsets[i] == 0xffffffff)
2116 p = msrpm_offsets[i];
2117 offset = svm->nested.vmcb_msrpm + (p * 4);
2119 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2122 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2125 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2130 static bool nested_vmcb_checks(struct vmcb *vmcb)
2132 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2135 if (vmcb->control.asid == 0)
2138 if (vmcb->control.nested_ctl && !npt_enabled)
2144 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2146 struct vmcb *nested_vmcb;
2147 struct vmcb *hsave = svm->nested.hsave;
2148 struct vmcb *vmcb = svm->vmcb;
2152 vmcb_gpa = svm->vmcb->save.rax;
2154 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2158 if (!nested_vmcb_checks(nested_vmcb)) {
2159 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2160 nested_vmcb->control.exit_code_hi = 0;
2161 nested_vmcb->control.exit_info_1 = 0;
2162 nested_vmcb->control.exit_info_2 = 0;
2164 nested_svm_unmap(page);
2169 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2170 nested_vmcb->save.rip,
2171 nested_vmcb->control.int_ctl,
2172 nested_vmcb->control.event_inj,
2173 nested_vmcb->control.nested_ctl);
2175 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr_read,
2176 nested_vmcb->control.intercept_cr_write,
2177 nested_vmcb->control.intercept_exceptions,
2178 nested_vmcb->control.intercept);
2180 /* Clear internal status */
2181 kvm_clear_exception_queue(&svm->vcpu);
2182 kvm_clear_interrupt_queue(&svm->vcpu);
2185 * Save the old vmcb, so we don't need to pick what we save, but can
2186 * restore everything when a VMEXIT occurs
2188 hsave->save.es = vmcb->save.es;
2189 hsave->save.cs = vmcb->save.cs;
2190 hsave->save.ss = vmcb->save.ss;
2191 hsave->save.ds = vmcb->save.ds;
2192 hsave->save.gdtr = vmcb->save.gdtr;
2193 hsave->save.idtr = vmcb->save.idtr;
2194 hsave->save.efer = svm->vcpu.arch.efer;
2195 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2196 hsave->save.cr4 = svm->vcpu.arch.cr4;
2197 hsave->save.rflags = vmcb->save.rflags;
2198 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2199 hsave->save.rsp = vmcb->save.rsp;
2200 hsave->save.rax = vmcb->save.rax;
2202 hsave->save.cr3 = vmcb->save.cr3;
2204 hsave->save.cr3 = svm->vcpu.arch.cr3;
2206 copy_vmcb_control_area(hsave, vmcb);
2208 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
2209 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2211 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2213 if (nested_vmcb->control.nested_ctl) {
2214 kvm_mmu_unload(&svm->vcpu);
2215 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2216 nested_svm_init_mmu_context(&svm->vcpu);
2219 /* Load the nested guest state */
2220 svm->vmcb->save.es = nested_vmcb->save.es;
2221 svm->vmcb->save.cs = nested_vmcb->save.cs;
2222 svm->vmcb->save.ss = nested_vmcb->save.ss;
2223 svm->vmcb->save.ds = nested_vmcb->save.ds;
2224 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2225 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2226 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
2227 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2228 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2229 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2231 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2232 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2234 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2236 /* Guest paging mode is active - reset mmu */
2237 kvm_mmu_reset_context(&svm->vcpu);
2239 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2240 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2241 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2242 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2244 /* In case we don't even reach vcpu_run, the fields are not updated */
2245 svm->vmcb->save.rax = nested_vmcb->save.rax;
2246 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2247 svm->vmcb->save.rip = nested_vmcb->save.rip;
2248 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2249 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2250 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2252 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2253 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2255 /* cache intercepts */
2256 svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
2257 svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
2258 svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
2259 svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
2260 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2261 svm->nested.intercept = nested_vmcb->control.intercept;
2263 force_new_asid(&svm->vcpu);
2264 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2265 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2266 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2268 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2270 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2271 /* We only want the cr8 intercept bits of the guest */
2272 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK;
2273 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2276 /* We don't want to see VMMCALLs from a nested guest */
2277 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMMCALL);
2280 * We don't want a nested guest to be more powerful than the guest, so
2281 * all intercepts are ORed
2283 svm->vmcb->control.intercept_cr_read |=
2284 nested_vmcb->control.intercept_cr_read;
2285 svm->vmcb->control.intercept_cr_write |=
2286 nested_vmcb->control.intercept_cr_write;
2287 svm->vmcb->control.intercept_dr_read |=
2288 nested_vmcb->control.intercept_dr_read;
2289 svm->vmcb->control.intercept_dr_write |=
2290 nested_vmcb->control.intercept_dr_write;
2291 svm->vmcb->control.intercept_exceptions |=
2292 nested_vmcb->control.intercept_exceptions;
2294 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
2296 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2297 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2298 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2299 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2300 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2301 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2303 nested_svm_unmap(page);
2305 /* nested_vmcb is our indicator if nested SVM is activated */
2306 svm->nested.vmcb = vmcb_gpa;
2313 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2315 to_vmcb->save.fs = from_vmcb->save.fs;
2316 to_vmcb->save.gs = from_vmcb->save.gs;
2317 to_vmcb->save.tr = from_vmcb->save.tr;
2318 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2319 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2320 to_vmcb->save.star = from_vmcb->save.star;
2321 to_vmcb->save.lstar = from_vmcb->save.lstar;
2322 to_vmcb->save.cstar = from_vmcb->save.cstar;
2323 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2324 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2325 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2326 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2329 static int vmload_interception(struct vcpu_svm *svm)
2331 struct vmcb *nested_vmcb;
2334 if (nested_svm_check_permissions(svm))
2337 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2338 skip_emulated_instruction(&svm->vcpu);
2340 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2344 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2345 nested_svm_unmap(page);
2350 static int vmsave_interception(struct vcpu_svm *svm)
2352 struct vmcb *nested_vmcb;
2355 if (nested_svm_check_permissions(svm))
2358 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2359 skip_emulated_instruction(&svm->vcpu);
2361 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2365 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2366 nested_svm_unmap(page);
2371 static int vmrun_interception(struct vcpu_svm *svm)
2373 if (nested_svm_check_permissions(svm))
2376 /* Save rip after vmrun instruction */
2377 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2379 if (!nested_svm_vmrun(svm))
2382 if (!nested_svm_vmrun_msrpm(svm))
2389 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2390 svm->vmcb->control.exit_code_hi = 0;
2391 svm->vmcb->control.exit_info_1 = 0;
2392 svm->vmcb->control.exit_info_2 = 0;
2394 nested_svm_vmexit(svm);
2399 static int stgi_interception(struct vcpu_svm *svm)
2401 if (nested_svm_check_permissions(svm))
2404 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2405 skip_emulated_instruction(&svm->vcpu);
2406 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2413 static int clgi_interception(struct vcpu_svm *svm)
2415 if (nested_svm_check_permissions(svm))
2418 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2419 skip_emulated_instruction(&svm->vcpu);
2423 /* After a CLGI no interrupts should come */
2424 svm_clear_vintr(svm);
2425 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2430 static int invlpga_interception(struct vcpu_svm *svm)
2432 struct kvm_vcpu *vcpu = &svm->vcpu;
2434 trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2435 vcpu->arch.regs[VCPU_REGS_RAX]);
2437 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2438 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2440 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2441 skip_emulated_instruction(&svm->vcpu);
2445 static int skinit_interception(struct vcpu_svm *svm)
2447 trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2449 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2453 static int invalid_op_interception(struct vcpu_svm *svm)
2455 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2459 static int task_switch_interception(struct vcpu_svm *svm)
2463 int int_type = svm->vmcb->control.exit_int_info &
2464 SVM_EXITINTINFO_TYPE_MASK;
2465 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2467 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2469 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2470 bool has_error_code = false;
2473 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2475 if (svm->vmcb->control.exit_info_2 &
2476 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2477 reason = TASK_SWITCH_IRET;
2478 else if (svm->vmcb->control.exit_info_2 &
2479 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2480 reason = TASK_SWITCH_JMP;
2482 reason = TASK_SWITCH_GATE;
2484 reason = TASK_SWITCH_CALL;
2486 if (reason == TASK_SWITCH_GATE) {
2488 case SVM_EXITINTINFO_TYPE_NMI:
2489 svm->vcpu.arch.nmi_injected = false;
2491 case SVM_EXITINTINFO_TYPE_EXEPT:
2492 if (svm->vmcb->control.exit_info_2 &
2493 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2494 has_error_code = true;
2496 (u32)svm->vmcb->control.exit_info_2;
2498 kvm_clear_exception_queue(&svm->vcpu);
2500 case SVM_EXITINTINFO_TYPE_INTR:
2501 kvm_clear_interrupt_queue(&svm->vcpu);
2508 if (reason != TASK_SWITCH_GATE ||
2509 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2510 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2511 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2512 skip_emulated_instruction(&svm->vcpu);
2514 if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
2515 has_error_code, error_code) == EMULATE_FAIL) {
2516 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2517 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2518 svm->vcpu.run->internal.ndata = 0;
2524 static int cpuid_interception(struct vcpu_svm *svm)
2526 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2527 kvm_emulate_cpuid(&svm->vcpu);
2531 static int iret_interception(struct vcpu_svm *svm)
2533 ++svm->vcpu.stat.nmi_window_exits;
2534 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
2535 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2539 static int invlpg_interception(struct vcpu_svm *svm)
2541 return emulate_instruction(&svm->vcpu, 0, 0, 0) == EMULATE_DONE;
2544 static int emulate_on_interception(struct vcpu_svm *svm)
2546 return emulate_instruction(&svm->vcpu, 0, 0, 0) == EMULATE_DONE;
2549 static int cr0_write_interception(struct vcpu_svm *svm)
2551 struct kvm_vcpu *vcpu = &svm->vcpu;
2554 r = emulate_instruction(&svm->vcpu, 0, 0, 0);
2556 if (svm->nested.vmexit_rip) {
2557 kvm_register_write(vcpu, VCPU_REGS_RIP, svm->nested.vmexit_rip);
2558 kvm_register_write(vcpu, VCPU_REGS_RSP, svm->nested.vmexit_rsp);
2559 kvm_register_write(vcpu, VCPU_REGS_RAX, svm->nested.vmexit_rax);
2560 svm->nested.vmexit_rip = 0;
2563 return r == EMULATE_DONE;
2566 static int cr8_write_interception(struct vcpu_svm *svm)
2568 struct kvm_run *kvm_run = svm->vcpu.run;
2570 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2571 /* instruction emulation calls kvm_set_cr8() */
2572 emulate_instruction(&svm->vcpu, 0, 0, 0);
2573 if (irqchip_in_kernel(svm->vcpu.kvm)) {
2574 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2577 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2579 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2583 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2585 struct vcpu_svm *svm = to_svm(vcpu);
2588 case MSR_IA32_TSC: {
2592 tsc_offset = svm->nested.hsave->control.tsc_offset;
2594 tsc_offset = svm->vmcb->control.tsc_offset;
2596 *data = tsc_offset + native_read_tsc();
2600 *data = svm->vmcb->save.star;
2602 #ifdef CONFIG_X86_64
2604 *data = svm->vmcb->save.lstar;
2607 *data = svm->vmcb->save.cstar;
2609 case MSR_KERNEL_GS_BASE:
2610 *data = svm->vmcb->save.kernel_gs_base;
2612 case MSR_SYSCALL_MASK:
2613 *data = svm->vmcb->save.sfmask;
2616 case MSR_IA32_SYSENTER_CS:
2617 *data = svm->vmcb->save.sysenter_cs;
2619 case MSR_IA32_SYSENTER_EIP:
2620 *data = svm->sysenter_eip;
2622 case MSR_IA32_SYSENTER_ESP:
2623 *data = svm->sysenter_esp;
2626 * Nobody will change the following 5 values in the VMCB so we can
2627 * safely return them on rdmsr. They will always be 0 until LBRV is
2630 case MSR_IA32_DEBUGCTLMSR:
2631 *data = svm->vmcb->save.dbgctl;
2633 case MSR_IA32_LASTBRANCHFROMIP:
2634 *data = svm->vmcb->save.br_from;
2636 case MSR_IA32_LASTBRANCHTOIP:
2637 *data = svm->vmcb->save.br_to;
2639 case MSR_IA32_LASTINTFROMIP:
2640 *data = svm->vmcb->save.last_excp_from;
2642 case MSR_IA32_LASTINTTOIP:
2643 *data = svm->vmcb->save.last_excp_to;
2645 case MSR_VM_HSAVE_PA:
2646 *data = svm->nested.hsave_msr;
2649 *data = svm->nested.vm_cr_msr;
2651 case MSR_IA32_UCODE_REV:
2655 return kvm_get_msr_common(vcpu, ecx, data);
2660 static int rdmsr_interception(struct vcpu_svm *svm)
2662 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2665 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2666 trace_kvm_msr_read_ex(ecx);
2667 kvm_inject_gp(&svm->vcpu, 0);
2669 trace_kvm_msr_read(ecx, data);
2671 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2672 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2673 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2674 skip_emulated_instruction(&svm->vcpu);
2679 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2681 struct vcpu_svm *svm = to_svm(vcpu);
2682 int svm_dis, chg_mask;
2684 if (data & ~SVM_VM_CR_VALID_MASK)
2687 chg_mask = SVM_VM_CR_VALID_MASK;
2689 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2690 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2692 svm->nested.vm_cr_msr &= ~chg_mask;
2693 svm->nested.vm_cr_msr |= (data & chg_mask);
2695 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2697 /* check for svm_disable while efer.svme is set */
2698 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2704 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2706 struct vcpu_svm *svm = to_svm(vcpu);
2710 kvm_write_tsc(vcpu, data);
2713 svm->vmcb->save.star = data;
2715 #ifdef CONFIG_X86_64
2717 svm->vmcb->save.lstar = data;
2720 svm->vmcb->save.cstar = data;
2722 case MSR_KERNEL_GS_BASE:
2723 svm->vmcb->save.kernel_gs_base = data;
2725 case MSR_SYSCALL_MASK:
2726 svm->vmcb->save.sfmask = data;
2729 case MSR_IA32_SYSENTER_CS:
2730 svm->vmcb->save.sysenter_cs = data;
2732 case MSR_IA32_SYSENTER_EIP:
2733 svm->sysenter_eip = data;
2734 svm->vmcb->save.sysenter_eip = data;
2736 case MSR_IA32_SYSENTER_ESP:
2737 svm->sysenter_esp = data;
2738 svm->vmcb->save.sysenter_esp = data;
2740 case MSR_IA32_DEBUGCTLMSR:
2741 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
2742 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2746 if (data & DEBUGCTL_RESERVED_BITS)
2749 svm->vmcb->save.dbgctl = data;
2750 if (data & (1ULL<<0))
2751 svm_enable_lbrv(svm);
2753 svm_disable_lbrv(svm);
2755 case MSR_VM_HSAVE_PA:
2756 svm->nested.hsave_msr = data;
2759 return svm_set_vm_cr(vcpu, data);
2761 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2764 return kvm_set_msr_common(vcpu, ecx, data);
2769 static int wrmsr_interception(struct vcpu_svm *svm)
2771 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2772 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2773 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2776 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2777 if (svm_set_msr(&svm->vcpu, ecx, data)) {
2778 trace_kvm_msr_write_ex(ecx, data);
2779 kvm_inject_gp(&svm->vcpu, 0);
2781 trace_kvm_msr_write(ecx, data);
2782 skip_emulated_instruction(&svm->vcpu);
2787 static int msr_interception(struct vcpu_svm *svm)
2789 if (svm->vmcb->control.exit_info_1)
2790 return wrmsr_interception(svm);
2792 return rdmsr_interception(svm);
2795 static int interrupt_window_interception(struct vcpu_svm *svm)
2797 struct kvm_run *kvm_run = svm->vcpu.run;
2799 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2800 svm_clear_vintr(svm);
2801 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2803 * If the user space waits to inject interrupts, exit as soon as
2806 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2807 kvm_run->request_interrupt_window &&
2808 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2809 ++svm->vcpu.stat.irq_window_exits;
2810 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2817 static int pause_interception(struct vcpu_svm *svm)
2819 kvm_vcpu_on_spin(&(svm->vcpu));
2823 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2824 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2825 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2826 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2827 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2828 [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
2829 [SVM_EXIT_WRITE_CR0] = cr0_write_interception,
2830 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2831 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2832 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2833 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2834 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2835 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2836 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2837 [SVM_EXIT_READ_DR4] = emulate_on_interception,
2838 [SVM_EXIT_READ_DR5] = emulate_on_interception,
2839 [SVM_EXIT_READ_DR6] = emulate_on_interception,
2840 [SVM_EXIT_READ_DR7] = emulate_on_interception,
2841 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2842 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2843 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2844 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2845 [SVM_EXIT_WRITE_DR4] = emulate_on_interception,
2846 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2847 [SVM_EXIT_WRITE_DR6] = emulate_on_interception,
2848 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2849 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2850 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2851 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2852 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2853 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2854 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2855 [SVM_EXIT_INTR] = intr_interception,
2856 [SVM_EXIT_NMI] = nmi_interception,
2857 [SVM_EXIT_SMI] = nop_on_interception,
2858 [SVM_EXIT_INIT] = nop_on_interception,
2859 [SVM_EXIT_VINTR] = interrupt_window_interception,
2860 [SVM_EXIT_CPUID] = cpuid_interception,
2861 [SVM_EXIT_IRET] = iret_interception,
2862 [SVM_EXIT_INVD] = emulate_on_interception,
2863 [SVM_EXIT_PAUSE] = pause_interception,
2864 [SVM_EXIT_HLT] = halt_interception,
2865 [SVM_EXIT_INVLPG] = invlpg_interception,
2866 [SVM_EXIT_INVLPGA] = invlpga_interception,
2867 [SVM_EXIT_IOIO] = io_interception,
2868 [SVM_EXIT_MSR] = msr_interception,
2869 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2870 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2871 [SVM_EXIT_VMRUN] = vmrun_interception,
2872 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2873 [SVM_EXIT_VMLOAD] = vmload_interception,
2874 [SVM_EXIT_VMSAVE] = vmsave_interception,
2875 [SVM_EXIT_STGI] = stgi_interception,
2876 [SVM_EXIT_CLGI] = clgi_interception,
2877 [SVM_EXIT_SKINIT] = skinit_interception,
2878 [SVM_EXIT_WBINVD] = emulate_on_interception,
2879 [SVM_EXIT_MONITOR] = invalid_op_interception,
2880 [SVM_EXIT_MWAIT] = invalid_op_interception,
2881 [SVM_EXIT_NPF] = pf_interception,
2884 void dump_vmcb(struct kvm_vcpu *vcpu)
2886 struct vcpu_svm *svm = to_svm(vcpu);
2887 struct vmcb_control_area *control = &svm->vmcb->control;
2888 struct vmcb_save_area *save = &svm->vmcb->save;
2890 pr_err("VMCB Control Area:\n");
2891 pr_err("cr_read: %04x\n", control->intercept_cr_read);
2892 pr_err("cr_write: %04x\n", control->intercept_cr_write);
2893 pr_err("dr_read: %04x\n", control->intercept_dr_read);
2894 pr_err("dr_write: %04x\n", control->intercept_dr_write);
2895 pr_err("exceptions: %08x\n", control->intercept_exceptions);
2896 pr_err("intercepts: %016llx\n", control->intercept);
2897 pr_err("pause filter count: %d\n", control->pause_filter_count);
2898 pr_err("iopm_base_pa: %016llx\n", control->iopm_base_pa);
2899 pr_err("msrpm_base_pa: %016llx\n", control->msrpm_base_pa);
2900 pr_err("tsc_offset: %016llx\n", control->tsc_offset);
2901 pr_err("asid: %d\n", control->asid);
2902 pr_err("tlb_ctl: %d\n", control->tlb_ctl);
2903 pr_err("int_ctl: %08x\n", control->int_ctl);
2904 pr_err("int_vector: %08x\n", control->int_vector);
2905 pr_err("int_state: %08x\n", control->int_state);
2906 pr_err("exit_code: %08x\n", control->exit_code);
2907 pr_err("exit_info1: %016llx\n", control->exit_info_1);
2908 pr_err("exit_info2: %016llx\n", control->exit_info_2);
2909 pr_err("exit_int_info: %08x\n", control->exit_int_info);
2910 pr_err("exit_int_info_err: %08x\n", control->exit_int_info_err);
2911 pr_err("nested_ctl: %lld\n", control->nested_ctl);
2912 pr_err("nested_cr3: %016llx\n", control->nested_cr3);
2913 pr_err("event_inj: %08x\n", control->event_inj);
2914 pr_err("event_inj_err: %08x\n", control->event_inj_err);
2915 pr_err("lbr_ctl: %lld\n", control->lbr_ctl);
2916 pr_err("next_rip: %016llx\n", control->next_rip);
2917 pr_err("VMCB State Save Area:\n");
2918 pr_err("es: s: %04x a: %04x l: %08x b: %016llx\n",
2919 save->es.selector, save->es.attrib,
2920 save->es.limit, save->es.base);
2921 pr_err("cs: s: %04x a: %04x l: %08x b: %016llx\n",
2922 save->cs.selector, save->cs.attrib,
2923 save->cs.limit, save->cs.base);
2924 pr_err("ss: s: %04x a: %04x l: %08x b: %016llx\n",
2925 save->ss.selector, save->ss.attrib,
2926 save->ss.limit, save->ss.base);
2927 pr_err("ds: s: %04x a: %04x l: %08x b: %016llx\n",
2928 save->ds.selector, save->ds.attrib,
2929 save->ds.limit, save->ds.base);
2930 pr_err("fs: s: %04x a: %04x l: %08x b: %016llx\n",
2931 save->fs.selector, save->fs.attrib,
2932 save->fs.limit, save->fs.base);
2933 pr_err("gs: s: %04x a: %04x l: %08x b: %016llx\n",
2934 save->gs.selector, save->gs.attrib,
2935 save->gs.limit, save->gs.base);
2936 pr_err("gdtr: s: %04x a: %04x l: %08x b: %016llx\n",
2937 save->gdtr.selector, save->gdtr.attrib,
2938 save->gdtr.limit, save->gdtr.base);
2939 pr_err("ldtr: s: %04x a: %04x l: %08x b: %016llx\n",
2940 save->ldtr.selector, save->ldtr.attrib,
2941 save->ldtr.limit, save->ldtr.base);
2942 pr_err("idtr: s: %04x a: %04x l: %08x b: %016llx\n",
2943 save->idtr.selector, save->idtr.attrib,
2944 save->idtr.limit, save->idtr.base);
2945 pr_err("tr: s: %04x a: %04x l: %08x b: %016llx\n",
2946 save->tr.selector, save->tr.attrib,
2947 save->tr.limit, save->tr.base);
2948 pr_err("cpl: %d efer: %016llx\n",
2949 save->cpl, save->efer);
2950 pr_err("cr0: %016llx cr2: %016llx\n",
2951 save->cr0, save->cr2);
2952 pr_err("cr3: %016llx cr4: %016llx\n",
2953 save->cr3, save->cr4);
2954 pr_err("dr6: %016llx dr7: %016llx\n",
2955 save->dr6, save->dr7);
2956 pr_err("rip: %016llx rflags: %016llx\n",
2957 save->rip, save->rflags);
2958 pr_err("rsp: %016llx rax: %016llx\n",
2959 save->rsp, save->rax);
2960 pr_err("star: %016llx lstar: %016llx\n",
2961 save->star, save->lstar);
2962 pr_err("cstar: %016llx sfmask: %016llx\n",
2963 save->cstar, save->sfmask);
2964 pr_err("kernel_gs_base: %016llx sysenter_cs: %016llx\n",
2965 save->kernel_gs_base, save->sysenter_cs);
2966 pr_err("sysenter_esp: %016llx sysenter_eip: %016llx\n",
2967 save->sysenter_esp, save->sysenter_eip);
2968 pr_err("gpat: %016llx dbgctl: %016llx\n",
2969 save->g_pat, save->dbgctl);
2970 pr_err("br_from: %016llx br_to: %016llx\n",
2971 save->br_from, save->br_to);
2972 pr_err("excp_from: %016llx excp_to: %016llx\n",
2973 save->last_excp_from, save->last_excp_to);
2977 static int handle_exit(struct kvm_vcpu *vcpu)
2979 struct vcpu_svm *svm = to_svm(vcpu);
2980 struct kvm_run *kvm_run = vcpu->run;
2981 u32 exit_code = svm->vmcb->control.exit_code;
2983 trace_kvm_exit(exit_code, vcpu);
2985 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2986 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2988 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2990 if (unlikely(svm->nested.exit_required)) {
2991 nested_svm_vmexit(svm);
2992 svm->nested.exit_required = false;
2997 if (is_nested(svm)) {
3000 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3001 svm->vmcb->control.exit_info_1,
3002 svm->vmcb->control.exit_info_2,
3003 svm->vmcb->control.exit_int_info,
3004 svm->vmcb->control.exit_int_info_err);
3006 vmexit = nested_svm_exit_special(svm);
3008 if (vmexit == NESTED_EXIT_CONTINUE)
3009 vmexit = nested_svm_exit_handled(svm);
3011 if (vmexit == NESTED_EXIT_DONE)
3015 svm_complete_interrupts(svm);
3017 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3018 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3019 kvm_run->fail_entry.hardware_entry_failure_reason
3020 = svm->vmcb->control.exit_code;
3021 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3026 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3027 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3028 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3029 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3030 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
3032 __func__, svm->vmcb->control.exit_int_info,
3035 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3036 || !svm_exit_handlers[exit_code]) {
3037 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3038 kvm_run->hw.hardware_exit_reason = exit_code;
3042 return svm_exit_handlers[exit_code](svm);
3045 static void reload_tss(struct kvm_vcpu *vcpu)
3047 int cpu = raw_smp_processor_id();
3049 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3050 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3054 static void pre_svm_run(struct vcpu_svm *svm)
3056 int cpu = raw_smp_processor_id();
3058 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3060 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3061 /* FIXME: handle wraparound of asid_generation */
3062 if (svm->asid_generation != sd->asid_generation)
3066 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3068 struct vcpu_svm *svm = to_svm(vcpu);
3070 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3071 vcpu->arch.hflags |= HF_NMI_MASK;
3072 svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
3073 ++vcpu->stat.nmi_injections;
3076 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3078 struct vmcb_control_area *control;
3080 control = &svm->vmcb->control;
3081 control->int_vector = irq;
3082 control->int_ctl &= ~V_INTR_PRIO_MASK;
3083 control->int_ctl |= V_IRQ_MASK |
3084 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3087 static void svm_set_irq(struct kvm_vcpu *vcpu)
3089 struct vcpu_svm *svm = to_svm(vcpu);
3091 BUG_ON(!(gif_set(svm)));
3093 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3094 ++vcpu->stat.irq_injections;
3096 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3097 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3100 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3102 struct vcpu_svm *svm = to_svm(vcpu);
3104 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3111 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
3114 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3116 struct vcpu_svm *svm = to_svm(vcpu);
3117 struct vmcb *vmcb = svm->vmcb;
3119 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3120 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3121 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3126 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3128 struct vcpu_svm *svm = to_svm(vcpu);
3130 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3133 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3135 struct vcpu_svm *svm = to_svm(vcpu);
3138 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3139 svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
3141 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3142 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
3146 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3148 struct vcpu_svm *svm = to_svm(vcpu);
3149 struct vmcb *vmcb = svm->vmcb;
3152 if (!gif_set(svm) ||
3153 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3156 ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
3159 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3164 static void enable_irq_window(struct kvm_vcpu *vcpu)
3166 struct vcpu_svm *svm = to_svm(vcpu);
3169 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3170 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3171 * get that intercept, this function will be called again though and
3172 * we'll get the vintr intercept.
3174 if (gif_set(svm) && nested_svm_intr(svm)) {
3176 svm_inject_irq(svm, 0x0);
3180 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3182 struct vcpu_svm *svm = to_svm(vcpu);
3184 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3186 return; /* IRET will cause a vm exit */
3189 * Something prevents NMI from been injected. Single step over possible
3190 * problem (IRET or exception injection or interrupt shadow)
3192 svm->nmi_singlestep = true;
3193 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3194 update_db_intercept(vcpu);
3197 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3202 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3204 force_new_asid(vcpu);
3207 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3211 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3213 struct vcpu_svm *svm = to_svm(vcpu);
3215 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3218 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
3219 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3220 kvm_set_cr8(vcpu, cr8);
3224 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3226 struct vcpu_svm *svm = to_svm(vcpu);
3229 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3232 cr8 = kvm_get_cr8(vcpu);
3233 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3234 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3237 static void svm_complete_interrupts(struct vcpu_svm *svm)
3241 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3242 unsigned int3_injected = svm->int3_injected;
3244 svm->int3_injected = 0;
3246 if (svm->vcpu.arch.hflags & HF_IRET_MASK) {
3247 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3248 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3251 svm->vcpu.arch.nmi_injected = false;
3252 kvm_clear_exception_queue(&svm->vcpu);
3253 kvm_clear_interrupt_queue(&svm->vcpu);
3255 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3258 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3260 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3261 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3264 case SVM_EXITINTINFO_TYPE_NMI:
3265 svm->vcpu.arch.nmi_injected = true;
3267 case SVM_EXITINTINFO_TYPE_EXEPT:
3269 * In case of software exceptions, do not reinject the vector,
3270 * but re-execute the instruction instead. Rewind RIP first
3271 * if we emulated INT3 before.
3273 if (kvm_exception_is_soft(vector)) {
3274 if (vector == BP_VECTOR && int3_injected &&
3275 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3276 kvm_rip_write(&svm->vcpu,
3277 kvm_rip_read(&svm->vcpu) -
3281 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3282 u32 err = svm->vmcb->control.exit_int_info_err;
3283 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3286 kvm_requeue_exception(&svm->vcpu, vector);
3288 case SVM_EXITINTINFO_TYPE_INTR:
3289 kvm_queue_interrupt(&svm->vcpu, vector, false);
3296 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3298 struct vcpu_svm *svm = to_svm(vcpu);
3299 struct vmcb_control_area *control = &svm->vmcb->control;
3301 control->exit_int_info = control->event_inj;
3302 control->exit_int_info_err = control->event_inj_err;
3303 control->event_inj = 0;
3304 svm_complete_interrupts(svm);
3307 #ifdef CONFIG_X86_64
3313 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3315 struct vcpu_svm *svm = to_svm(vcpu);
3317 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3318 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3319 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3322 * A vmexit emulation is required before the vcpu can be executed
3325 if (unlikely(svm->nested.exit_required))
3330 sync_lapic_to_cr8(vcpu);
3332 svm->vmcb->save.cr2 = vcpu->arch.cr2;
3339 "push %%"R"bp; \n\t"
3340 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
3341 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
3342 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
3343 "mov %c[rsi](%[svm]), %%"R"si \n\t"
3344 "mov %c[rdi](%[svm]), %%"R"di \n\t"
3345 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
3346 #ifdef CONFIG_X86_64
3347 "mov %c[r8](%[svm]), %%r8 \n\t"
3348 "mov %c[r9](%[svm]), %%r9 \n\t"
3349 "mov %c[r10](%[svm]), %%r10 \n\t"
3350 "mov %c[r11](%[svm]), %%r11 \n\t"
3351 "mov %c[r12](%[svm]), %%r12 \n\t"
3352 "mov %c[r13](%[svm]), %%r13 \n\t"
3353 "mov %c[r14](%[svm]), %%r14 \n\t"
3354 "mov %c[r15](%[svm]), %%r15 \n\t"
3357 /* Enter guest mode */
3359 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
3360 __ex(SVM_VMLOAD) "\n\t"
3361 __ex(SVM_VMRUN) "\n\t"
3362 __ex(SVM_VMSAVE) "\n\t"
3365 /* Save guest registers, load host registers */
3366 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
3367 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
3368 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
3369 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
3370 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
3371 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
3372 #ifdef CONFIG_X86_64
3373 "mov %%r8, %c[r8](%[svm]) \n\t"
3374 "mov %%r9, %c[r9](%[svm]) \n\t"
3375 "mov %%r10, %c[r10](%[svm]) \n\t"
3376 "mov %%r11, %c[r11](%[svm]) \n\t"
3377 "mov %%r12, %c[r12](%[svm]) \n\t"
3378 "mov %%r13, %c[r13](%[svm]) \n\t"
3379 "mov %%r14, %c[r14](%[svm]) \n\t"
3380 "mov %%r15, %c[r15](%[svm]) \n\t"
3385 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3386 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3387 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3388 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3389 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3390 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3391 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3392 #ifdef CONFIG_X86_64
3393 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3394 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3395 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3396 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3397 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3398 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3399 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3400 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3403 , R"bx", R"cx", R"dx", R"si", R"di"
3404 #ifdef CONFIG_X86_64
3405 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3409 #ifdef CONFIG_X86_64
3410 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3412 loadsegment(fs, svm->host.fs);
3417 local_irq_disable();
3421 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3422 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3423 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3424 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3426 sync_cr8_to_lapic(vcpu);
3430 /* if exit due to PF check for async PF */
3431 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3432 svm->apf_reason = kvm_read_and_reset_pf_reason();
3435 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3436 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3440 * We need to handle MC intercepts here before the vcpu has a chance to
3441 * change the physical cpu
3443 if (unlikely(svm->vmcb->control.exit_code ==
3444 SVM_EXIT_EXCP_BASE + MC_VECTOR))
3445 svm_handle_mce(svm);
3450 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3452 struct vcpu_svm *svm = to_svm(vcpu);
3454 svm->vmcb->save.cr3 = root;
3455 force_new_asid(vcpu);
3458 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3460 struct vcpu_svm *svm = to_svm(vcpu);
3462 svm->vmcb->control.nested_cr3 = root;
3464 /* Also sync guest cr3 here in case we live migrate */
3465 svm->vmcb->save.cr3 = vcpu->arch.cr3;
3467 force_new_asid(vcpu);
3470 static int is_disabled(void)
3474 rdmsrl(MSR_VM_CR, vm_cr);
3475 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3482 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3485 * Patch in the VMMCALL instruction:
3487 hypercall[0] = 0x0f;
3488 hypercall[1] = 0x01;
3489 hypercall[2] = 0xd9;
3492 static void svm_check_processor_compat(void *rtn)
3497 static bool svm_cpu_has_accelerated_tpr(void)
3502 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3507 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3511 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
3515 /* Mask out xsave bit as long as it is not supported by SVM */
3516 entry->ecx &= ~(bit(X86_FEATURE_XSAVE));
3520 entry->ecx |= (1 << 2); /* Set SVM bit */
3523 entry->eax = 1; /* SVM revision 1 */
3524 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
3525 ASID emulation to nested SVM */
3526 entry->ecx = 0; /* Reserved */
3527 entry->edx = 0; /* Per default do not support any
3528 additional features */
3530 /* Support next_rip if host supports it */
3531 if (boot_cpu_has(X86_FEATURE_NRIPS))
3532 entry->edx |= SVM_FEATURE_NRIP;
3534 /* Support NPT for the guest if enabled */
3536 entry->edx |= SVM_FEATURE_NPT;
3542 static const struct trace_print_flags svm_exit_reasons_str[] = {
3543 { SVM_EXIT_READ_CR0, "read_cr0" },
3544 { SVM_EXIT_READ_CR3, "read_cr3" },
3545 { SVM_EXIT_READ_CR4, "read_cr4" },
3546 { SVM_EXIT_READ_CR8, "read_cr8" },
3547 { SVM_EXIT_WRITE_CR0, "write_cr0" },
3548 { SVM_EXIT_WRITE_CR3, "write_cr3" },
3549 { SVM_EXIT_WRITE_CR4, "write_cr4" },
3550 { SVM_EXIT_WRITE_CR8, "write_cr8" },
3551 { SVM_EXIT_READ_DR0, "read_dr0" },
3552 { SVM_EXIT_READ_DR1, "read_dr1" },
3553 { SVM_EXIT_READ_DR2, "read_dr2" },
3554 { SVM_EXIT_READ_DR3, "read_dr3" },
3555 { SVM_EXIT_WRITE_DR0, "write_dr0" },
3556 { SVM_EXIT_WRITE_DR1, "write_dr1" },
3557 { SVM_EXIT_WRITE_DR2, "write_dr2" },
3558 { SVM_EXIT_WRITE_DR3, "write_dr3" },
3559 { SVM_EXIT_WRITE_DR5, "write_dr5" },
3560 { SVM_EXIT_WRITE_DR7, "write_dr7" },
3561 { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
3562 { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
3563 { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
3564 { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
3565 { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
3566 { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
3567 { SVM_EXIT_INTR, "interrupt" },
3568 { SVM_EXIT_NMI, "nmi" },
3569 { SVM_EXIT_SMI, "smi" },
3570 { SVM_EXIT_INIT, "init" },
3571 { SVM_EXIT_VINTR, "vintr" },
3572 { SVM_EXIT_CPUID, "cpuid" },
3573 { SVM_EXIT_INVD, "invd" },
3574 { SVM_EXIT_HLT, "hlt" },
3575 { SVM_EXIT_INVLPG, "invlpg" },
3576 { SVM_EXIT_INVLPGA, "invlpga" },
3577 { SVM_EXIT_IOIO, "io" },
3578 { SVM_EXIT_MSR, "msr" },
3579 { SVM_EXIT_TASK_SWITCH, "task_switch" },
3580 { SVM_EXIT_SHUTDOWN, "shutdown" },
3581 { SVM_EXIT_VMRUN, "vmrun" },
3582 { SVM_EXIT_VMMCALL, "hypercall" },
3583 { SVM_EXIT_VMLOAD, "vmload" },
3584 { SVM_EXIT_VMSAVE, "vmsave" },
3585 { SVM_EXIT_STGI, "stgi" },
3586 { SVM_EXIT_CLGI, "clgi" },
3587 { SVM_EXIT_SKINIT, "skinit" },
3588 { SVM_EXIT_WBINVD, "wbinvd" },
3589 { SVM_EXIT_MONITOR, "monitor" },
3590 { SVM_EXIT_MWAIT, "mwait" },
3591 { SVM_EXIT_NPF, "npf" },
3595 static int svm_get_lpage_level(void)
3597 return PT_PDPE_LEVEL;
3600 static bool svm_rdtscp_supported(void)
3605 static bool svm_has_wbinvd_exit(void)
3610 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3612 struct vcpu_svm *svm = to_svm(vcpu);
3614 svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
3616 svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR;
3617 update_cr0_intercept(svm);
3620 static struct kvm_x86_ops svm_x86_ops = {
3621 .cpu_has_kvm_support = has_svm,
3622 .disabled_by_bios = is_disabled,
3623 .hardware_setup = svm_hardware_setup,
3624 .hardware_unsetup = svm_hardware_unsetup,
3625 .check_processor_compatibility = svm_check_processor_compat,
3626 .hardware_enable = svm_hardware_enable,
3627 .hardware_disable = svm_hardware_disable,
3628 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
3630 .vcpu_create = svm_create_vcpu,
3631 .vcpu_free = svm_free_vcpu,
3632 .vcpu_reset = svm_vcpu_reset,
3634 .prepare_guest_switch = svm_prepare_guest_switch,
3635 .vcpu_load = svm_vcpu_load,
3636 .vcpu_put = svm_vcpu_put,
3638 .set_guest_debug = svm_guest_debug,
3639 .get_msr = svm_get_msr,
3640 .set_msr = svm_set_msr,
3641 .get_segment_base = svm_get_segment_base,
3642 .get_segment = svm_get_segment,
3643 .set_segment = svm_set_segment,
3644 .get_cpl = svm_get_cpl,
3645 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
3646 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
3647 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
3648 .set_cr0 = svm_set_cr0,
3649 .set_cr3 = svm_set_cr3,
3650 .set_cr4 = svm_set_cr4,
3651 .set_efer = svm_set_efer,
3652 .get_idt = svm_get_idt,
3653 .set_idt = svm_set_idt,
3654 .get_gdt = svm_get_gdt,
3655 .set_gdt = svm_set_gdt,
3656 .set_dr7 = svm_set_dr7,
3657 .cache_reg = svm_cache_reg,
3658 .get_rflags = svm_get_rflags,
3659 .set_rflags = svm_set_rflags,
3660 .fpu_activate = svm_fpu_activate,
3661 .fpu_deactivate = svm_fpu_deactivate,
3663 .tlb_flush = svm_flush_tlb,
3665 .run = svm_vcpu_run,
3666 .handle_exit = handle_exit,
3667 .skip_emulated_instruction = skip_emulated_instruction,
3668 .set_interrupt_shadow = svm_set_interrupt_shadow,
3669 .get_interrupt_shadow = svm_get_interrupt_shadow,
3670 .patch_hypercall = svm_patch_hypercall,
3671 .set_irq = svm_set_irq,
3672 .set_nmi = svm_inject_nmi,
3673 .queue_exception = svm_queue_exception,
3674 .cancel_injection = svm_cancel_injection,
3675 .interrupt_allowed = svm_interrupt_allowed,
3676 .nmi_allowed = svm_nmi_allowed,
3677 .get_nmi_mask = svm_get_nmi_mask,
3678 .set_nmi_mask = svm_set_nmi_mask,
3679 .enable_nmi_window = enable_nmi_window,
3680 .enable_irq_window = enable_irq_window,
3681 .update_cr8_intercept = update_cr8_intercept,
3683 .set_tss_addr = svm_set_tss_addr,
3684 .get_tdp_level = get_npt_level,
3685 .get_mt_mask = svm_get_mt_mask,
3687 .exit_reasons_str = svm_exit_reasons_str,
3688 .get_lpage_level = svm_get_lpage_level,
3690 .cpuid_update = svm_cpuid_update,
3692 .rdtscp_supported = svm_rdtscp_supported,
3694 .set_supported_cpuid = svm_set_supported_cpuid,
3696 .has_wbinvd_exit = svm_has_wbinvd_exit,
3698 .write_tsc_offset = svm_write_tsc_offset,
3699 .adjust_tsc_offset = svm_adjust_tsc_offset,
3701 .set_tdp_cr3 = set_tdp_cr3,
3704 static int __init svm_init(void)
3706 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3707 __alignof__(struct vcpu_svm), THIS_MODULE);
3710 static void __exit svm_exit(void)
3715 module_init(svm_init)
3716 module_exit(svm_exit)