2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
16 #include <linux/kvm_host.h>
20 #include "kvm_cache_regs.h"
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/vmalloc.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/ftrace_event.h>
29 #include <linux/slab.h>
33 #include <asm/virtext.h>
36 #define __ex(x) __kvm_handle_fault_on_reboot(x)
38 MODULE_AUTHOR("Qumranet");
39 MODULE_LICENSE("GPL");
41 #define IOPM_ALLOC_ORDER 2
42 #define MSRPM_ALLOC_ORDER 1
44 #define SEG_TYPE_LDT 2
45 #define SEG_TYPE_BUSY_TSS16 3
47 #define SVM_FEATURE_NPT (1 << 0)
48 #define SVM_FEATURE_LBRV (1 << 1)
49 #define SVM_FEATURE_SVML (1 << 2)
50 #define SVM_FEATURE_NRIP (1 << 3)
51 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
53 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
54 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
55 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
57 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
59 static const u32 host_save_user_msrs[] = {
61 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
64 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
67 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
77 /* These are the merged vectors */
80 /* gpa pointers to the real vectors */
84 /* A VMEXIT is required but not yet emulated */
87 /* cache for intercepts of the guest */
88 u16 intercept_cr_read;
89 u16 intercept_cr_write;
90 u16 intercept_dr_read;
91 u16 intercept_dr_write;
92 u32 intercept_exceptions;
97 #define MSRPM_OFFSETS 16
98 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
101 struct kvm_vcpu vcpu;
103 unsigned long vmcb_pa;
104 struct svm_cpu_data *svm_data;
105 uint64_t asid_generation;
106 uint64_t sysenter_esp;
107 uint64_t sysenter_eip;
111 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
116 struct nested_state nested;
120 unsigned int3_injected;
121 unsigned long int3_rip;
124 #define MSR_INVALID 0xffffffffU
126 static struct svm_direct_access_msrs {
127 u32 index; /* Index of the MSR */
128 bool always; /* True if intercept is always on */
129 } direct_access_msrs[] = {
130 { .index = MSR_K6_STAR, .always = true },
131 { .index = MSR_IA32_SYSENTER_CS, .always = true },
133 { .index = MSR_GS_BASE, .always = true },
134 { .index = MSR_FS_BASE, .always = true },
135 { .index = MSR_KERNEL_GS_BASE, .always = true },
136 { .index = MSR_LSTAR, .always = true },
137 { .index = MSR_CSTAR, .always = true },
138 { .index = MSR_SYSCALL_MASK, .always = true },
140 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
141 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
142 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
143 { .index = MSR_IA32_LASTINTTOIP, .always = false },
144 { .index = MSR_INVALID, .always = false },
147 /* enable NPT for AMD64 and X86 with PAE */
148 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
149 static bool npt_enabled = true;
151 static bool npt_enabled;
155 module_param(npt, int, S_IRUGO);
157 static int nested = 1;
158 module_param(nested, int, S_IRUGO);
160 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
161 static void svm_complete_interrupts(struct vcpu_svm *svm);
163 static int nested_svm_exit_handled(struct vcpu_svm *svm);
164 static int nested_svm_intercept(struct vcpu_svm *svm);
165 static int nested_svm_vmexit(struct vcpu_svm *svm);
166 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
167 bool has_error_code, u32 error_code);
169 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
171 return container_of(vcpu, struct vcpu_svm, vcpu);
174 static inline bool is_nested(struct vcpu_svm *svm)
176 return svm->nested.vmcb;
179 static inline void enable_gif(struct vcpu_svm *svm)
181 svm->vcpu.arch.hflags |= HF_GIF_MASK;
184 static inline void disable_gif(struct vcpu_svm *svm)
186 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
189 static inline bool gif_set(struct vcpu_svm *svm)
191 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
194 static unsigned long iopm_base;
196 struct kvm_ldttss_desc {
199 unsigned base1:8, type:5, dpl:2, p:1;
200 unsigned limit1:4, zero0:3, g:1, base2:8;
203 } __attribute__((packed));
205 struct svm_cpu_data {
211 struct kvm_ldttss_desc *tss_desc;
213 struct page *save_area;
216 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
217 static uint32_t svm_features;
219 struct svm_init_data {
224 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
226 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
227 #define MSRS_RANGE_SIZE 2048
228 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
230 static u32 svm_msrpm_offset(u32 msr)
235 for (i = 0; i < NUM_MSR_MAPS; i++) {
236 if (msr < msrpm_ranges[i] ||
237 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
240 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
241 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
243 /* Now we have the u8 offset - but need the u32 offset */
247 /* MSR not in any range */
251 #define MAX_INST_SIZE 15
253 static inline u32 svm_has(u32 feat)
255 return svm_features & feat;
258 static inline void clgi(void)
260 asm volatile (__ex(SVM_CLGI));
263 static inline void stgi(void)
265 asm volatile (__ex(SVM_STGI));
268 static inline void invlpga(unsigned long addr, u32 asid)
270 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
273 static inline void force_new_asid(struct kvm_vcpu *vcpu)
275 to_svm(vcpu)->asid_generation--;
278 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
280 force_new_asid(vcpu);
283 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
285 if (!npt_enabled && !(efer & EFER_LMA))
288 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
289 vcpu->arch.efer = efer;
292 static int is_external_interrupt(u32 info)
294 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
295 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
298 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
300 struct vcpu_svm *svm = to_svm(vcpu);
303 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
304 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
308 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
310 struct vcpu_svm *svm = to_svm(vcpu);
313 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
315 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
319 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
321 struct vcpu_svm *svm = to_svm(vcpu);
323 if (svm->vmcb->control.next_rip != 0)
324 svm->next_rip = svm->vmcb->control.next_rip;
326 if (!svm->next_rip) {
327 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
329 printk(KERN_DEBUG "%s: NOP\n", __func__);
332 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
333 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
334 __func__, kvm_rip_read(vcpu), svm->next_rip);
336 kvm_rip_write(vcpu, svm->next_rip);
337 svm_set_interrupt_shadow(vcpu, 0);
340 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
341 bool has_error_code, u32 error_code,
344 struct vcpu_svm *svm = to_svm(vcpu);
347 * If we are within a nested VM we'd better #VMEXIT and let the guest
348 * handle the exception
351 nested_svm_check_exception(svm, nr, has_error_code, error_code))
354 if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) {
355 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
358 * For guest debugging where we have to reinject #BP if some
359 * INT3 is guest-owned:
360 * Emulate nRIP by moving RIP forward. Will fail if injection
361 * raises a fault that is not intercepted. Still better than
362 * failing in all cases.
364 skip_emulated_instruction(&svm->vcpu);
365 rip = kvm_rip_read(&svm->vcpu);
366 svm->int3_rip = rip + svm->vmcb->save.cs.base;
367 svm->int3_injected = rip - old_rip;
370 svm->vmcb->control.event_inj = nr
372 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
373 | SVM_EVTINJ_TYPE_EXEPT;
374 svm->vmcb->control.event_inj_err = error_code;
377 static int has_svm(void)
381 if (!cpu_has_svm(&msg)) {
382 printk(KERN_INFO "has_svm: %s\n", msg);
389 static void svm_hardware_disable(void *garbage)
394 static int svm_hardware_enable(void *garbage)
397 struct svm_cpu_data *sd;
399 struct desc_ptr gdt_descr;
400 struct desc_struct *gdt;
401 int me = raw_smp_processor_id();
403 rdmsrl(MSR_EFER, efer);
404 if (efer & EFER_SVME)
408 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
412 sd = per_cpu(svm_data, me);
415 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
420 sd->asid_generation = 1;
421 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
422 sd->next_asid = sd->max_asid + 1;
424 native_store_gdt(&gdt_descr);
425 gdt = (struct desc_struct *)gdt_descr.address;
426 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
428 wrmsrl(MSR_EFER, efer | EFER_SVME);
430 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
435 static void svm_cpu_uninit(int cpu)
437 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
442 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
443 __free_page(sd->save_area);
447 static int svm_cpu_init(int cpu)
449 struct svm_cpu_data *sd;
452 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
456 sd->save_area = alloc_page(GFP_KERNEL);
461 per_cpu(svm_data, cpu) = sd;
471 static bool valid_msr_intercept(u32 index)
475 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
476 if (direct_access_msrs[i].index == index)
482 static void set_msr_interception(u32 *msrpm, unsigned msr,
485 u8 bit_read, bit_write;
490 * If this warning triggers extend the direct_access_msrs list at the
491 * beginning of the file
493 WARN_ON(!valid_msr_intercept(msr));
495 offset = svm_msrpm_offset(msr);
496 bit_read = 2 * (msr & 0x0f);
497 bit_write = 2 * (msr & 0x0f) + 1;
500 BUG_ON(offset == MSR_INVALID);
502 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
503 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
508 static void svm_vcpu_init_msrpm(u32 *msrpm)
512 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
514 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
515 if (!direct_access_msrs[i].always)
518 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
522 static void add_msr_offset(u32 offset)
526 for (i = 0; i < MSRPM_OFFSETS; ++i) {
528 /* Offset already in list? */
529 if (msrpm_offsets[i] == offset)
532 /* Slot used by another offset? */
533 if (msrpm_offsets[i] != MSR_INVALID)
536 /* Add offset to list */
537 msrpm_offsets[i] = offset;
543 * If this BUG triggers the msrpm_offsets table has an overflow. Just
544 * increase MSRPM_OFFSETS in this case.
549 static void init_msrpm_offsets(void)
553 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
555 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
558 offset = svm_msrpm_offset(direct_access_msrs[i].index);
559 BUG_ON(offset == MSR_INVALID);
561 add_msr_offset(offset);
565 static void svm_enable_lbrv(struct vcpu_svm *svm)
567 u32 *msrpm = svm->msrpm;
569 svm->vmcb->control.lbr_ctl = 1;
570 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
571 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
572 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
573 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
576 static void svm_disable_lbrv(struct vcpu_svm *svm)
578 u32 *msrpm = svm->msrpm;
580 svm->vmcb->control.lbr_ctl = 0;
581 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
582 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
583 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
584 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
587 static __init int svm_hardware_setup(void)
590 struct page *iopm_pages;
594 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
599 iopm_va = page_address(iopm_pages);
600 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
601 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
603 init_msrpm_offsets();
605 if (boot_cpu_has(X86_FEATURE_NX))
606 kvm_enable_efer_bits(EFER_NX);
608 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
609 kvm_enable_efer_bits(EFER_FFXSR);
612 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
613 kvm_enable_efer_bits(EFER_SVME);
616 for_each_possible_cpu(cpu) {
617 r = svm_cpu_init(cpu);
622 svm_features = cpuid_edx(SVM_CPUID_FUNC);
624 if (!svm_has(SVM_FEATURE_NPT))
627 if (npt_enabled && !npt) {
628 printk(KERN_INFO "kvm: Nested Paging disabled\n");
633 printk(KERN_INFO "kvm: Nested Paging enabled\n");
641 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
646 static __exit void svm_hardware_unsetup(void)
650 for_each_possible_cpu(cpu)
653 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
657 static void init_seg(struct vmcb_seg *seg)
660 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
661 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
666 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
669 seg->attrib = SVM_SELECTOR_P_MASK | type;
674 static void init_vmcb(struct vcpu_svm *svm)
676 struct vmcb_control_area *control = &svm->vmcb->control;
677 struct vmcb_save_area *save = &svm->vmcb->save;
679 svm->vcpu.fpu_active = 1;
681 control->intercept_cr_read = INTERCEPT_CR0_MASK |
685 control->intercept_cr_write = INTERCEPT_CR0_MASK |
690 control->intercept_dr_read = INTERCEPT_DR0_MASK |
699 control->intercept_dr_write = INTERCEPT_DR0_MASK |
708 control->intercept_exceptions = (1 << PF_VECTOR) |
713 control->intercept = (1ULL << INTERCEPT_INTR) |
714 (1ULL << INTERCEPT_NMI) |
715 (1ULL << INTERCEPT_SMI) |
716 (1ULL << INTERCEPT_SELECTIVE_CR0) |
717 (1ULL << INTERCEPT_CPUID) |
718 (1ULL << INTERCEPT_INVD) |
719 (1ULL << INTERCEPT_HLT) |
720 (1ULL << INTERCEPT_INVLPG) |
721 (1ULL << INTERCEPT_INVLPGA) |
722 (1ULL << INTERCEPT_IOIO_PROT) |
723 (1ULL << INTERCEPT_MSR_PROT) |
724 (1ULL << INTERCEPT_TASK_SWITCH) |
725 (1ULL << INTERCEPT_SHUTDOWN) |
726 (1ULL << INTERCEPT_VMRUN) |
727 (1ULL << INTERCEPT_VMMCALL) |
728 (1ULL << INTERCEPT_VMLOAD) |
729 (1ULL << INTERCEPT_VMSAVE) |
730 (1ULL << INTERCEPT_STGI) |
731 (1ULL << INTERCEPT_CLGI) |
732 (1ULL << INTERCEPT_SKINIT) |
733 (1ULL << INTERCEPT_WBINVD) |
734 (1ULL << INTERCEPT_MONITOR) |
735 (1ULL << INTERCEPT_MWAIT);
737 control->iopm_base_pa = iopm_base;
738 control->msrpm_base_pa = __pa(svm->msrpm);
739 control->tsc_offset = 0;
740 control->int_ctl = V_INTR_MASKING_MASK;
748 save->cs.selector = 0xf000;
749 /* Executable/Readable Code Segment */
750 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
751 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
752 save->cs.limit = 0xffff;
754 * cs.base should really be 0xffff0000, but vmx can't handle that, so
755 * be consistent with it.
757 * Replace when we have real mode working for vmx.
759 save->cs.base = 0xf0000;
761 save->gdtr.limit = 0xffff;
762 save->idtr.limit = 0xffff;
764 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
765 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
767 save->efer = EFER_SVME;
768 save->dr6 = 0xffff0ff0;
771 save->rip = 0x0000fff0;
772 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
775 * This is the guest-visible cr0 value.
776 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
778 svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
779 kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0);
781 save->cr4 = X86_CR4_PAE;
785 /* Setup VMCB for Nested Paging */
786 control->nested_ctl = 1;
787 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
788 (1ULL << INTERCEPT_INVLPG));
789 control->intercept_exceptions &= ~(1 << PF_VECTOR);
790 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
791 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
792 save->g_pat = 0x0007040600070406ULL;
796 force_new_asid(&svm->vcpu);
798 svm->nested.vmcb = 0;
799 svm->vcpu.arch.hflags = 0;
801 if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
802 control->pause_filter_count = 3000;
803 control->intercept |= (1ULL << INTERCEPT_PAUSE);
809 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
811 struct vcpu_svm *svm = to_svm(vcpu);
815 if (!kvm_vcpu_is_bsp(vcpu)) {
816 kvm_rip_write(vcpu, 0);
817 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
818 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
820 vcpu->arch.regs_avail = ~0;
821 vcpu->arch.regs_dirty = ~0;
826 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
828 struct vcpu_svm *svm;
830 struct page *msrpm_pages;
831 struct page *hsave_page;
832 struct page *nested_msrpm_pages;
835 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
841 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
846 page = alloc_page(GFP_KERNEL);
850 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
854 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
855 if (!nested_msrpm_pages)
858 hsave_page = alloc_page(GFP_KERNEL);
862 svm->nested.hsave = page_address(hsave_page);
864 svm->msrpm = page_address(msrpm_pages);
865 svm_vcpu_init_msrpm(svm->msrpm);
867 svm->nested.msrpm = page_address(nested_msrpm_pages);
868 svm_vcpu_init_msrpm(svm->nested.msrpm);
870 svm->vmcb = page_address(page);
871 clear_page(svm->vmcb);
872 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
873 svm->asid_generation = 0;
877 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
878 if (kvm_vcpu_is_bsp(&svm->vcpu))
879 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
884 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
886 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
890 kvm_vcpu_uninit(&svm->vcpu);
892 kmem_cache_free(kvm_vcpu_cache, svm);
897 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
899 struct vcpu_svm *svm = to_svm(vcpu);
901 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
902 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
903 __free_page(virt_to_page(svm->nested.hsave));
904 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
905 kvm_vcpu_uninit(vcpu);
906 kmem_cache_free(kvm_vcpu_cache, svm);
909 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
911 struct vcpu_svm *svm = to_svm(vcpu);
914 if (unlikely(cpu != vcpu->cpu)) {
917 if (check_tsc_unstable()) {
919 * Make sure that the guest sees a monotonically
922 delta = vcpu->arch.host_tsc - native_read_tsc();
923 svm->vmcb->control.tsc_offset += delta;
925 svm->nested.hsave->control.tsc_offset += delta;
928 kvm_migrate_timers(vcpu);
929 svm->asid_generation = 0;
932 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
933 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
936 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
938 struct vcpu_svm *svm = to_svm(vcpu);
941 ++vcpu->stat.host_state_reload;
942 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
943 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
945 vcpu->arch.host_tsc = native_read_tsc();
948 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
950 return to_svm(vcpu)->vmcb->save.rflags;
953 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
955 to_svm(vcpu)->vmcb->save.rflags = rflags;
958 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
961 case VCPU_EXREG_PDPTR:
962 BUG_ON(!npt_enabled);
963 load_pdptrs(vcpu, vcpu->arch.cr3);
970 static void svm_set_vintr(struct vcpu_svm *svm)
972 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
975 static void svm_clear_vintr(struct vcpu_svm *svm)
977 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
980 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
982 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
985 case VCPU_SREG_CS: return &save->cs;
986 case VCPU_SREG_DS: return &save->ds;
987 case VCPU_SREG_ES: return &save->es;
988 case VCPU_SREG_FS: return &save->fs;
989 case VCPU_SREG_GS: return &save->gs;
990 case VCPU_SREG_SS: return &save->ss;
991 case VCPU_SREG_TR: return &save->tr;
992 case VCPU_SREG_LDTR: return &save->ldtr;
998 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1000 struct vmcb_seg *s = svm_seg(vcpu, seg);
1005 static void svm_get_segment(struct kvm_vcpu *vcpu,
1006 struct kvm_segment *var, int seg)
1008 struct vmcb_seg *s = svm_seg(vcpu, seg);
1010 var->base = s->base;
1011 var->limit = s->limit;
1012 var->selector = s->selector;
1013 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1014 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1015 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1016 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1017 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1018 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1019 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1020 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1023 * AMD's VMCB does not have an explicit unusable field, so emulate it
1024 * for cross vendor migration purposes by "not present"
1026 var->unusable = !var->present || (var->type == 0);
1031 * SVM always stores 0 for the 'G' bit in the CS selector in
1032 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1033 * Intel's VMENTRY has a check on the 'G' bit.
1035 var->g = s->limit > 0xfffff;
1039 * Work around a bug where the busy flag in the tr selector
1049 * The accessed bit must always be set in the segment
1050 * descriptor cache, although it can be cleared in the
1051 * descriptor, the cached bit always remains at 1. Since
1052 * Intel has a check on this, set it here to support
1053 * cross-vendor migration.
1060 * On AMD CPUs sometimes the DB bit in the segment
1061 * descriptor is left as 1, although the whole segment has
1062 * been made unusable. Clear it here to pass an Intel VMX
1063 * entry check when cross vendor migrating.
1071 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1073 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1078 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1080 struct vcpu_svm *svm = to_svm(vcpu);
1082 dt->size = svm->vmcb->save.idtr.limit;
1083 dt->address = svm->vmcb->save.idtr.base;
1086 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1088 struct vcpu_svm *svm = to_svm(vcpu);
1090 svm->vmcb->save.idtr.limit = dt->size;
1091 svm->vmcb->save.idtr.base = dt->address ;
1094 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1096 struct vcpu_svm *svm = to_svm(vcpu);
1098 dt->size = svm->vmcb->save.gdtr.limit;
1099 dt->address = svm->vmcb->save.gdtr.base;
1102 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1104 struct vcpu_svm *svm = to_svm(vcpu);
1106 svm->vmcb->save.gdtr.limit = dt->size;
1107 svm->vmcb->save.gdtr.base = dt->address ;
1110 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1114 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1118 static void update_cr0_intercept(struct vcpu_svm *svm)
1120 struct vmcb *vmcb = svm->vmcb;
1121 ulong gcr0 = svm->vcpu.arch.cr0;
1122 u64 *hcr0 = &svm->vmcb->save.cr0;
1124 if (!svm->vcpu.fpu_active)
1125 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1127 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1128 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1131 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1132 vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1133 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1134 if (is_nested(svm)) {
1135 struct vmcb *hsave = svm->nested.hsave;
1137 hsave->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1138 hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1139 vmcb->control.intercept_cr_read |= svm->nested.intercept_cr_read;
1140 vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write;
1143 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1144 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1145 if (is_nested(svm)) {
1146 struct vmcb *hsave = svm->nested.hsave;
1148 hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1149 hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1154 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1156 struct vcpu_svm *svm = to_svm(vcpu);
1158 if (is_nested(svm)) {
1160 * We are here because we run in nested mode, the host kvm
1161 * intercepts cr0 writes but the l1 hypervisor does not.
1162 * But the L1 hypervisor may intercept selective cr0 writes.
1163 * This needs to be checked here.
1165 unsigned long old, new;
1167 /* Remove bits that would trigger a real cr0 write intercept */
1168 old = vcpu->arch.cr0 & SVM_CR0_SELECTIVE_MASK;
1169 new = cr0 & SVM_CR0_SELECTIVE_MASK;
1172 /* cr0 write with ts and mp unchanged */
1173 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
1174 if (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE)
1179 #ifdef CONFIG_X86_64
1180 if (vcpu->arch.efer & EFER_LME) {
1181 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1182 vcpu->arch.efer |= EFER_LMA;
1183 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1186 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1187 vcpu->arch.efer &= ~EFER_LMA;
1188 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1192 vcpu->arch.cr0 = cr0;
1195 cr0 |= X86_CR0_PG | X86_CR0_WP;
1197 if (!vcpu->fpu_active)
1200 * re-enable caching here because the QEMU bios
1201 * does not do it - this results in some delay at
1204 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1205 svm->vmcb->save.cr0 = cr0;
1206 update_cr0_intercept(svm);
1209 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1211 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1212 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1214 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1215 force_new_asid(vcpu);
1217 vcpu->arch.cr4 = cr4;
1220 cr4 |= host_cr4_mce;
1221 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1224 static void svm_set_segment(struct kvm_vcpu *vcpu,
1225 struct kvm_segment *var, int seg)
1227 struct vcpu_svm *svm = to_svm(vcpu);
1228 struct vmcb_seg *s = svm_seg(vcpu, seg);
1230 s->base = var->base;
1231 s->limit = var->limit;
1232 s->selector = var->selector;
1236 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1237 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1238 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1239 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1240 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1241 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1242 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1243 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1245 if (seg == VCPU_SREG_CS)
1247 = (svm->vmcb->save.cs.attrib
1248 >> SVM_SELECTOR_DPL_SHIFT) & 3;
1252 static void update_db_intercept(struct kvm_vcpu *vcpu)
1254 struct vcpu_svm *svm = to_svm(vcpu);
1256 svm->vmcb->control.intercept_exceptions &=
1257 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1259 if (svm->nmi_singlestep)
1260 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1262 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1263 if (vcpu->guest_debug &
1264 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1265 svm->vmcb->control.intercept_exceptions |=
1267 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1268 svm->vmcb->control.intercept_exceptions |=
1271 vcpu->guest_debug = 0;
1274 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1276 struct vcpu_svm *svm = to_svm(vcpu);
1278 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1279 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1281 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1283 update_db_intercept(vcpu);
1286 static void load_host_msrs(struct kvm_vcpu *vcpu)
1288 #ifdef CONFIG_X86_64
1289 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1293 static void save_host_msrs(struct kvm_vcpu *vcpu)
1295 #ifdef CONFIG_X86_64
1296 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1300 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1302 if (sd->next_asid > sd->max_asid) {
1303 ++sd->asid_generation;
1305 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1308 svm->asid_generation = sd->asid_generation;
1309 svm->vmcb->control.asid = sd->next_asid++;
1312 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1314 struct vcpu_svm *svm = to_svm(vcpu);
1316 svm->vmcb->save.dr7 = value;
1319 static int pf_interception(struct vcpu_svm *svm)
1324 fault_address = svm->vmcb->control.exit_info_2;
1325 error_code = svm->vmcb->control.exit_info_1;
1327 trace_kvm_page_fault(fault_address, error_code);
1328 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1329 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1330 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1333 static int db_interception(struct vcpu_svm *svm)
1335 struct kvm_run *kvm_run = svm->vcpu.run;
1337 if (!(svm->vcpu.guest_debug &
1338 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1339 !svm->nmi_singlestep) {
1340 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1344 if (svm->nmi_singlestep) {
1345 svm->nmi_singlestep = false;
1346 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1347 svm->vmcb->save.rflags &=
1348 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1349 update_db_intercept(&svm->vcpu);
1352 if (svm->vcpu.guest_debug &
1353 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1354 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1355 kvm_run->debug.arch.pc =
1356 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1357 kvm_run->debug.arch.exception = DB_VECTOR;
1364 static int bp_interception(struct vcpu_svm *svm)
1366 struct kvm_run *kvm_run = svm->vcpu.run;
1368 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1369 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1370 kvm_run->debug.arch.exception = BP_VECTOR;
1374 static int ud_interception(struct vcpu_svm *svm)
1378 er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1379 if (er != EMULATE_DONE)
1380 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1384 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1386 struct vcpu_svm *svm = to_svm(vcpu);
1389 if (is_nested(svm)) {
1392 h_excp = svm->nested.hsave->control.intercept_exceptions;
1393 n_excp = svm->nested.intercept_exceptions;
1394 h_excp &= ~(1 << NM_VECTOR);
1395 excp = h_excp | n_excp;
1397 excp = svm->vmcb->control.intercept_exceptions;
1398 excp &= ~(1 << NM_VECTOR);
1401 svm->vmcb->control.intercept_exceptions = excp;
1403 svm->vcpu.fpu_active = 1;
1404 update_cr0_intercept(svm);
1407 static int nm_interception(struct vcpu_svm *svm)
1409 svm_fpu_activate(&svm->vcpu);
1413 static int mc_interception(struct vcpu_svm *svm)
1416 * On an #MC intercept the MCE handler is not called automatically in
1417 * the host. So do it by hand here.
1421 /* not sure if we ever come back to this point */
1426 static int shutdown_interception(struct vcpu_svm *svm)
1428 struct kvm_run *kvm_run = svm->vcpu.run;
1431 * VMCB is undefined after a SHUTDOWN intercept
1432 * so reinitialize it.
1434 clear_page(svm->vmcb);
1437 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1441 static int io_interception(struct vcpu_svm *svm)
1443 struct kvm_vcpu *vcpu = &svm->vcpu;
1444 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1445 int size, in, string;
1448 ++svm->vcpu.stat.io_exits;
1449 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1450 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1452 return !(emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO);
1454 port = io_info >> 16;
1455 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1456 svm->next_rip = svm->vmcb->control.exit_info_2;
1457 skip_emulated_instruction(&svm->vcpu);
1459 return kvm_fast_pio_out(vcpu, size, port);
1462 static int nmi_interception(struct vcpu_svm *svm)
1467 static int intr_interception(struct vcpu_svm *svm)
1469 ++svm->vcpu.stat.irq_exits;
1473 static int nop_on_interception(struct vcpu_svm *svm)
1478 static int halt_interception(struct vcpu_svm *svm)
1480 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1481 skip_emulated_instruction(&svm->vcpu);
1482 return kvm_emulate_halt(&svm->vcpu);
1485 static int vmmcall_interception(struct vcpu_svm *svm)
1487 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1488 skip_emulated_instruction(&svm->vcpu);
1489 kvm_emulate_hypercall(&svm->vcpu);
1493 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1495 if (!(svm->vcpu.arch.efer & EFER_SVME)
1496 || !is_paging(&svm->vcpu)) {
1497 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1501 if (svm->vmcb->save.cpl) {
1502 kvm_inject_gp(&svm->vcpu, 0);
1509 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1510 bool has_error_code, u32 error_code)
1514 if (!is_nested(svm))
1517 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1518 svm->vmcb->control.exit_code_hi = 0;
1519 svm->vmcb->control.exit_info_1 = error_code;
1520 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1522 vmexit = nested_svm_intercept(svm);
1523 if (vmexit == NESTED_EXIT_DONE)
1524 svm->nested.exit_required = true;
1529 /* This function returns true if it is save to enable the irq window */
1530 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1532 if (!is_nested(svm))
1535 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1538 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1541 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1542 svm->vmcb->control.exit_info_1 = 0;
1543 svm->vmcb->control.exit_info_2 = 0;
1545 if (svm->nested.intercept & 1ULL) {
1547 * The #vmexit can't be emulated here directly because this
1548 * code path runs with irqs and preemtion disabled. A
1549 * #vmexit emulation might sleep. Only signal request for
1552 svm->nested.exit_required = true;
1553 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1560 /* This function returns true if it is save to enable the nmi window */
1561 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1563 if (!is_nested(svm))
1566 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1569 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1570 svm->nested.exit_required = true;
1575 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1581 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1582 if (is_error_page(page))
1590 kvm_release_page_clean(page);
1591 kvm_inject_gp(&svm->vcpu, 0);
1596 static void nested_svm_unmap(struct page *page)
1599 kvm_release_page_dirty(page);
1602 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1608 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
1609 return NESTED_EXIT_HOST;
1611 port = svm->vmcb->control.exit_info_1 >> 16;
1612 gpa = svm->nested.vmcb_iopm + (port / 8);
1616 if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
1619 return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1622 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1624 u32 offset, msr, value;
1627 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1628 return NESTED_EXIT_HOST;
1630 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1631 offset = svm_msrpm_offset(msr);
1632 write = svm->vmcb->control.exit_info_1 & 1;
1633 mask = 1 << ((2 * (msr & 0xf)) + write);
1635 if (offset == MSR_INVALID)
1636 return NESTED_EXIT_DONE;
1638 /* Offset is in 32 bit units but need in 8 bit units */
1641 if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
1642 return NESTED_EXIT_DONE;
1644 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1647 static int nested_svm_exit_special(struct vcpu_svm *svm)
1649 u32 exit_code = svm->vmcb->control.exit_code;
1651 switch (exit_code) {
1654 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
1655 return NESTED_EXIT_HOST;
1657 /* For now we are always handling NPFs when using them */
1659 return NESTED_EXIT_HOST;
1661 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1662 /* When we're shadowing, trap PFs */
1664 return NESTED_EXIT_HOST;
1666 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
1667 nm_interception(svm);
1673 return NESTED_EXIT_CONTINUE;
1677 * If this function returns true, this #vmexit was already handled
1679 static int nested_svm_intercept(struct vcpu_svm *svm)
1681 u32 exit_code = svm->vmcb->control.exit_code;
1682 int vmexit = NESTED_EXIT_HOST;
1684 switch (exit_code) {
1686 vmexit = nested_svm_exit_handled_msr(svm);
1689 vmexit = nested_svm_intercept_ioio(svm);
1691 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1692 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1693 if (svm->nested.intercept_cr_read & cr_bits)
1694 vmexit = NESTED_EXIT_DONE;
1697 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1698 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1699 if (svm->nested.intercept_cr_write & cr_bits)
1700 vmexit = NESTED_EXIT_DONE;
1703 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1704 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1705 if (svm->nested.intercept_dr_read & dr_bits)
1706 vmexit = NESTED_EXIT_DONE;
1709 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1710 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1711 if (svm->nested.intercept_dr_write & dr_bits)
1712 vmexit = NESTED_EXIT_DONE;
1715 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1716 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1717 if (svm->nested.intercept_exceptions & excp_bits)
1718 vmexit = NESTED_EXIT_DONE;
1721 case SVM_EXIT_ERR: {
1722 vmexit = NESTED_EXIT_DONE;
1726 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1727 if (svm->nested.intercept & exit_bits)
1728 vmexit = NESTED_EXIT_DONE;
1735 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1739 vmexit = nested_svm_intercept(svm);
1741 if (vmexit == NESTED_EXIT_DONE)
1742 nested_svm_vmexit(svm);
1747 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1749 struct vmcb_control_area *dst = &dst_vmcb->control;
1750 struct vmcb_control_area *from = &from_vmcb->control;
1752 dst->intercept_cr_read = from->intercept_cr_read;
1753 dst->intercept_cr_write = from->intercept_cr_write;
1754 dst->intercept_dr_read = from->intercept_dr_read;
1755 dst->intercept_dr_write = from->intercept_dr_write;
1756 dst->intercept_exceptions = from->intercept_exceptions;
1757 dst->intercept = from->intercept;
1758 dst->iopm_base_pa = from->iopm_base_pa;
1759 dst->msrpm_base_pa = from->msrpm_base_pa;
1760 dst->tsc_offset = from->tsc_offset;
1761 dst->asid = from->asid;
1762 dst->tlb_ctl = from->tlb_ctl;
1763 dst->int_ctl = from->int_ctl;
1764 dst->int_vector = from->int_vector;
1765 dst->int_state = from->int_state;
1766 dst->exit_code = from->exit_code;
1767 dst->exit_code_hi = from->exit_code_hi;
1768 dst->exit_info_1 = from->exit_info_1;
1769 dst->exit_info_2 = from->exit_info_2;
1770 dst->exit_int_info = from->exit_int_info;
1771 dst->exit_int_info_err = from->exit_int_info_err;
1772 dst->nested_ctl = from->nested_ctl;
1773 dst->event_inj = from->event_inj;
1774 dst->event_inj_err = from->event_inj_err;
1775 dst->nested_cr3 = from->nested_cr3;
1776 dst->lbr_ctl = from->lbr_ctl;
1779 static int nested_svm_vmexit(struct vcpu_svm *svm)
1781 struct vmcb *nested_vmcb;
1782 struct vmcb *hsave = svm->nested.hsave;
1783 struct vmcb *vmcb = svm->vmcb;
1786 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1787 vmcb->control.exit_info_1,
1788 vmcb->control.exit_info_2,
1789 vmcb->control.exit_int_info,
1790 vmcb->control.exit_int_info_err);
1792 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
1796 /* Exit nested SVM mode */
1797 svm->nested.vmcb = 0;
1799 /* Give the current vmcb to the guest */
1802 nested_vmcb->save.es = vmcb->save.es;
1803 nested_vmcb->save.cs = vmcb->save.cs;
1804 nested_vmcb->save.ss = vmcb->save.ss;
1805 nested_vmcb->save.ds = vmcb->save.ds;
1806 nested_vmcb->save.gdtr = vmcb->save.gdtr;
1807 nested_vmcb->save.idtr = vmcb->save.idtr;
1808 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
1809 nested_vmcb->save.cr3 = svm->vcpu.arch.cr3;
1810 nested_vmcb->save.cr2 = vmcb->save.cr2;
1811 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
1812 nested_vmcb->save.rflags = vmcb->save.rflags;
1813 nested_vmcb->save.rip = vmcb->save.rip;
1814 nested_vmcb->save.rsp = vmcb->save.rsp;
1815 nested_vmcb->save.rax = vmcb->save.rax;
1816 nested_vmcb->save.dr7 = vmcb->save.dr7;
1817 nested_vmcb->save.dr6 = vmcb->save.dr6;
1818 nested_vmcb->save.cpl = vmcb->save.cpl;
1820 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
1821 nested_vmcb->control.int_vector = vmcb->control.int_vector;
1822 nested_vmcb->control.int_state = vmcb->control.int_state;
1823 nested_vmcb->control.exit_code = vmcb->control.exit_code;
1824 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
1825 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
1826 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
1827 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
1828 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1831 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1832 * to make sure that we do not lose injected events. So check event_inj
1833 * here and copy it to exit_int_info if it is valid.
1834 * Exit_int_info and event_inj can't be both valid because the case
1835 * below only happens on a VMRUN instruction intercept which has
1836 * no valid exit_int_info set.
1838 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
1839 struct vmcb_control_area *nc = &nested_vmcb->control;
1841 nc->exit_int_info = vmcb->control.event_inj;
1842 nc->exit_int_info_err = vmcb->control.event_inj_err;
1845 nested_vmcb->control.tlb_ctl = 0;
1846 nested_vmcb->control.event_inj = 0;
1847 nested_vmcb->control.event_inj_err = 0;
1849 /* We always set V_INTR_MASKING and remember the old value in hflags */
1850 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1851 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1853 /* Restore the original control entries */
1854 copy_vmcb_control_area(vmcb, hsave);
1856 kvm_clear_exception_queue(&svm->vcpu);
1857 kvm_clear_interrupt_queue(&svm->vcpu);
1859 /* Restore selected save entries */
1860 svm->vmcb->save.es = hsave->save.es;
1861 svm->vmcb->save.cs = hsave->save.cs;
1862 svm->vmcb->save.ss = hsave->save.ss;
1863 svm->vmcb->save.ds = hsave->save.ds;
1864 svm->vmcb->save.gdtr = hsave->save.gdtr;
1865 svm->vmcb->save.idtr = hsave->save.idtr;
1866 svm->vmcb->save.rflags = hsave->save.rflags;
1867 svm_set_efer(&svm->vcpu, hsave->save.efer);
1868 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1869 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1871 svm->vmcb->save.cr3 = hsave->save.cr3;
1872 svm->vcpu.arch.cr3 = hsave->save.cr3;
1874 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1876 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1877 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1878 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1879 svm->vmcb->save.dr7 = 0;
1880 svm->vmcb->save.cpl = 0;
1881 svm->vmcb->control.exit_int_info = 0;
1883 nested_svm_unmap(page);
1885 kvm_mmu_reset_context(&svm->vcpu);
1886 kvm_mmu_load(&svm->vcpu);
1891 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
1894 * This function merges the msr permission bitmaps of kvm and the
1895 * nested vmcb. It is omptimized in that it only merges the parts where
1896 * the kvm msr permission bitmap may contain zero bits
1900 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1903 for (i = 0; i < MSRPM_OFFSETS; i++) {
1907 if (msrpm_offsets[i] == 0xffffffff)
1910 p = msrpm_offsets[i];
1911 offset = svm->nested.vmcb_msrpm + (p * 4);
1913 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
1916 svm->nested.msrpm[p] = svm->msrpm[p] | value;
1919 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1924 static bool nested_svm_vmrun(struct vcpu_svm *svm)
1926 struct vmcb *nested_vmcb;
1927 struct vmcb *hsave = svm->nested.hsave;
1928 struct vmcb *vmcb = svm->vmcb;
1932 vmcb_gpa = svm->vmcb->save.rax;
1934 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
1938 trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, vmcb_gpa,
1939 nested_vmcb->save.rip,
1940 nested_vmcb->control.int_ctl,
1941 nested_vmcb->control.event_inj,
1942 nested_vmcb->control.nested_ctl);
1944 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr_read,
1945 nested_vmcb->control.intercept_cr_write,
1946 nested_vmcb->control.intercept_exceptions,
1947 nested_vmcb->control.intercept);
1949 /* Clear internal status */
1950 kvm_clear_exception_queue(&svm->vcpu);
1951 kvm_clear_interrupt_queue(&svm->vcpu);
1954 * Save the old vmcb, so we don't need to pick what we save, but can
1955 * restore everything when a VMEXIT occurs
1957 hsave->save.es = vmcb->save.es;
1958 hsave->save.cs = vmcb->save.cs;
1959 hsave->save.ss = vmcb->save.ss;
1960 hsave->save.ds = vmcb->save.ds;
1961 hsave->save.gdtr = vmcb->save.gdtr;
1962 hsave->save.idtr = vmcb->save.idtr;
1963 hsave->save.efer = svm->vcpu.arch.efer;
1964 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
1965 hsave->save.cr4 = svm->vcpu.arch.cr4;
1966 hsave->save.rflags = vmcb->save.rflags;
1967 hsave->save.rip = svm->next_rip;
1968 hsave->save.rsp = vmcb->save.rsp;
1969 hsave->save.rax = vmcb->save.rax;
1971 hsave->save.cr3 = vmcb->save.cr3;
1973 hsave->save.cr3 = svm->vcpu.arch.cr3;
1975 copy_vmcb_control_area(hsave, vmcb);
1977 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1978 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1980 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1982 /* Load the nested guest state */
1983 svm->vmcb->save.es = nested_vmcb->save.es;
1984 svm->vmcb->save.cs = nested_vmcb->save.cs;
1985 svm->vmcb->save.ss = nested_vmcb->save.ss;
1986 svm->vmcb->save.ds = nested_vmcb->save.ds;
1987 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1988 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1989 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1990 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1991 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1992 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1994 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1995 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1997 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1999 /* Guest paging mode is active - reset mmu */
2000 kvm_mmu_reset_context(&svm->vcpu);
2002 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2003 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2004 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2005 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2007 /* In case we don't even reach vcpu_run, the fields are not updated */
2008 svm->vmcb->save.rax = nested_vmcb->save.rax;
2009 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2010 svm->vmcb->save.rip = nested_vmcb->save.rip;
2011 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2012 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2013 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2015 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2016 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2018 /* cache intercepts */
2019 svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
2020 svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
2021 svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
2022 svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
2023 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2024 svm->nested.intercept = nested_vmcb->control.intercept;
2026 force_new_asid(&svm->vcpu);
2027 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2028 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2029 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2031 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2033 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2034 /* We only want the cr8 intercept bits of the guest */
2035 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK;
2036 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2039 /* We don't want to see VMMCALLs from a nested guest */
2040 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMMCALL);
2043 * We don't want a nested guest to be more powerful than the guest, so
2044 * all intercepts are ORed
2046 svm->vmcb->control.intercept_cr_read |=
2047 nested_vmcb->control.intercept_cr_read;
2048 svm->vmcb->control.intercept_cr_write |=
2049 nested_vmcb->control.intercept_cr_write;
2050 svm->vmcb->control.intercept_dr_read |=
2051 nested_vmcb->control.intercept_dr_read;
2052 svm->vmcb->control.intercept_dr_write |=
2053 nested_vmcb->control.intercept_dr_write;
2054 svm->vmcb->control.intercept_exceptions |=
2055 nested_vmcb->control.intercept_exceptions;
2057 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
2059 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2060 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2061 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2062 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2063 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2064 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2066 nested_svm_unmap(page);
2068 /* nested_vmcb is our indicator if nested SVM is activated */
2069 svm->nested.vmcb = vmcb_gpa;
2076 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2078 to_vmcb->save.fs = from_vmcb->save.fs;
2079 to_vmcb->save.gs = from_vmcb->save.gs;
2080 to_vmcb->save.tr = from_vmcb->save.tr;
2081 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2082 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2083 to_vmcb->save.star = from_vmcb->save.star;
2084 to_vmcb->save.lstar = from_vmcb->save.lstar;
2085 to_vmcb->save.cstar = from_vmcb->save.cstar;
2086 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2087 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2088 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2089 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2092 static int vmload_interception(struct vcpu_svm *svm)
2094 struct vmcb *nested_vmcb;
2097 if (nested_svm_check_permissions(svm))
2100 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2101 skip_emulated_instruction(&svm->vcpu);
2103 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2107 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2108 nested_svm_unmap(page);
2113 static int vmsave_interception(struct vcpu_svm *svm)
2115 struct vmcb *nested_vmcb;
2118 if (nested_svm_check_permissions(svm))
2121 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2122 skip_emulated_instruction(&svm->vcpu);
2124 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2128 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2129 nested_svm_unmap(page);
2134 static int vmrun_interception(struct vcpu_svm *svm)
2136 if (nested_svm_check_permissions(svm))
2139 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2140 skip_emulated_instruction(&svm->vcpu);
2142 if (!nested_svm_vmrun(svm))
2145 if (!nested_svm_vmrun_msrpm(svm))
2152 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2153 svm->vmcb->control.exit_code_hi = 0;
2154 svm->vmcb->control.exit_info_1 = 0;
2155 svm->vmcb->control.exit_info_2 = 0;
2157 nested_svm_vmexit(svm);
2162 static int stgi_interception(struct vcpu_svm *svm)
2164 if (nested_svm_check_permissions(svm))
2167 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2168 skip_emulated_instruction(&svm->vcpu);
2175 static int clgi_interception(struct vcpu_svm *svm)
2177 if (nested_svm_check_permissions(svm))
2180 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2181 skip_emulated_instruction(&svm->vcpu);
2185 /* After a CLGI no interrupts should come */
2186 svm_clear_vintr(svm);
2187 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2192 static int invlpga_interception(struct vcpu_svm *svm)
2194 struct kvm_vcpu *vcpu = &svm->vcpu;
2196 trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2197 vcpu->arch.regs[VCPU_REGS_RAX]);
2199 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2200 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2202 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2203 skip_emulated_instruction(&svm->vcpu);
2207 static int skinit_interception(struct vcpu_svm *svm)
2209 trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2211 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2215 static int invalid_op_interception(struct vcpu_svm *svm)
2217 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2221 static int task_switch_interception(struct vcpu_svm *svm)
2225 int int_type = svm->vmcb->control.exit_int_info &
2226 SVM_EXITINTINFO_TYPE_MASK;
2227 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2229 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2231 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2232 bool has_error_code = false;
2235 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2237 if (svm->vmcb->control.exit_info_2 &
2238 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2239 reason = TASK_SWITCH_IRET;
2240 else if (svm->vmcb->control.exit_info_2 &
2241 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2242 reason = TASK_SWITCH_JMP;
2244 reason = TASK_SWITCH_GATE;
2246 reason = TASK_SWITCH_CALL;
2248 if (reason == TASK_SWITCH_GATE) {
2250 case SVM_EXITINTINFO_TYPE_NMI:
2251 svm->vcpu.arch.nmi_injected = false;
2253 case SVM_EXITINTINFO_TYPE_EXEPT:
2254 if (svm->vmcb->control.exit_info_2 &
2255 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2256 has_error_code = true;
2258 (u32)svm->vmcb->control.exit_info_2;
2260 kvm_clear_exception_queue(&svm->vcpu);
2262 case SVM_EXITINTINFO_TYPE_INTR:
2263 kvm_clear_interrupt_queue(&svm->vcpu);
2270 if (reason != TASK_SWITCH_GATE ||
2271 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2272 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2273 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2274 skip_emulated_instruction(&svm->vcpu);
2276 if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
2277 has_error_code, error_code) == EMULATE_FAIL) {
2278 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2279 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2280 svm->vcpu.run->internal.ndata = 0;
2286 static int cpuid_interception(struct vcpu_svm *svm)
2288 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2289 kvm_emulate_cpuid(&svm->vcpu);
2293 static int iret_interception(struct vcpu_svm *svm)
2295 ++svm->vcpu.stat.nmi_window_exits;
2296 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
2297 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2301 static int invlpg_interception(struct vcpu_svm *svm)
2303 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2304 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2308 static int emulate_on_interception(struct vcpu_svm *svm)
2310 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2311 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2315 static int cr8_write_interception(struct vcpu_svm *svm)
2317 struct kvm_run *kvm_run = svm->vcpu.run;
2319 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2320 /* instruction emulation calls kvm_set_cr8() */
2321 emulate_instruction(&svm->vcpu, 0, 0, 0);
2322 if (irqchip_in_kernel(svm->vcpu.kvm)) {
2323 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2326 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2328 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2332 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2334 struct vcpu_svm *svm = to_svm(vcpu);
2337 case MSR_IA32_TSC: {
2341 tsc_offset = svm->nested.hsave->control.tsc_offset;
2343 tsc_offset = svm->vmcb->control.tsc_offset;
2345 *data = tsc_offset + native_read_tsc();
2349 *data = svm->vmcb->save.star;
2351 #ifdef CONFIG_X86_64
2353 *data = svm->vmcb->save.lstar;
2356 *data = svm->vmcb->save.cstar;
2358 case MSR_KERNEL_GS_BASE:
2359 *data = svm->vmcb->save.kernel_gs_base;
2361 case MSR_SYSCALL_MASK:
2362 *data = svm->vmcb->save.sfmask;
2365 case MSR_IA32_SYSENTER_CS:
2366 *data = svm->vmcb->save.sysenter_cs;
2368 case MSR_IA32_SYSENTER_EIP:
2369 *data = svm->sysenter_eip;
2371 case MSR_IA32_SYSENTER_ESP:
2372 *data = svm->sysenter_esp;
2375 * Nobody will change the following 5 values in the VMCB so we can
2376 * safely return them on rdmsr. They will always be 0 until LBRV is
2379 case MSR_IA32_DEBUGCTLMSR:
2380 *data = svm->vmcb->save.dbgctl;
2382 case MSR_IA32_LASTBRANCHFROMIP:
2383 *data = svm->vmcb->save.br_from;
2385 case MSR_IA32_LASTBRANCHTOIP:
2386 *data = svm->vmcb->save.br_to;
2388 case MSR_IA32_LASTINTFROMIP:
2389 *data = svm->vmcb->save.last_excp_from;
2391 case MSR_IA32_LASTINTTOIP:
2392 *data = svm->vmcb->save.last_excp_to;
2394 case MSR_VM_HSAVE_PA:
2395 *data = svm->nested.hsave_msr;
2398 *data = svm->nested.vm_cr_msr;
2400 case MSR_IA32_UCODE_REV:
2404 return kvm_get_msr_common(vcpu, ecx, data);
2409 static int rdmsr_interception(struct vcpu_svm *svm)
2411 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2414 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2415 trace_kvm_msr_read_ex(ecx);
2416 kvm_inject_gp(&svm->vcpu, 0);
2418 trace_kvm_msr_read(ecx, data);
2420 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2421 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2422 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2423 skip_emulated_instruction(&svm->vcpu);
2428 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2430 struct vcpu_svm *svm = to_svm(vcpu);
2431 int svm_dis, chg_mask;
2433 if (data & ~SVM_VM_CR_VALID_MASK)
2436 chg_mask = SVM_VM_CR_VALID_MASK;
2438 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2439 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2441 svm->nested.vm_cr_msr &= ~chg_mask;
2442 svm->nested.vm_cr_msr |= (data & chg_mask);
2444 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2446 /* check for svm_disable while efer.svme is set */
2447 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2453 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2455 struct vcpu_svm *svm = to_svm(vcpu);
2458 case MSR_IA32_TSC: {
2459 u64 tsc_offset = data - native_read_tsc();
2460 u64 g_tsc_offset = 0;
2462 if (is_nested(svm)) {
2463 g_tsc_offset = svm->vmcb->control.tsc_offset -
2464 svm->nested.hsave->control.tsc_offset;
2465 svm->nested.hsave->control.tsc_offset = tsc_offset;
2468 svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset;
2473 svm->vmcb->save.star = data;
2475 #ifdef CONFIG_X86_64
2477 svm->vmcb->save.lstar = data;
2480 svm->vmcb->save.cstar = data;
2482 case MSR_KERNEL_GS_BASE:
2483 svm->vmcb->save.kernel_gs_base = data;
2485 case MSR_SYSCALL_MASK:
2486 svm->vmcb->save.sfmask = data;
2489 case MSR_IA32_SYSENTER_CS:
2490 svm->vmcb->save.sysenter_cs = data;
2492 case MSR_IA32_SYSENTER_EIP:
2493 svm->sysenter_eip = data;
2494 svm->vmcb->save.sysenter_eip = data;
2496 case MSR_IA32_SYSENTER_ESP:
2497 svm->sysenter_esp = data;
2498 svm->vmcb->save.sysenter_esp = data;
2500 case MSR_IA32_DEBUGCTLMSR:
2501 if (!svm_has(SVM_FEATURE_LBRV)) {
2502 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2506 if (data & DEBUGCTL_RESERVED_BITS)
2509 svm->vmcb->save.dbgctl = data;
2510 if (data & (1ULL<<0))
2511 svm_enable_lbrv(svm);
2513 svm_disable_lbrv(svm);
2515 case MSR_VM_HSAVE_PA:
2516 svm->nested.hsave_msr = data;
2519 return svm_set_vm_cr(vcpu, data);
2521 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2524 return kvm_set_msr_common(vcpu, ecx, data);
2529 static int wrmsr_interception(struct vcpu_svm *svm)
2531 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2532 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2533 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2536 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2537 if (svm_set_msr(&svm->vcpu, ecx, data)) {
2538 trace_kvm_msr_write_ex(ecx, data);
2539 kvm_inject_gp(&svm->vcpu, 0);
2541 trace_kvm_msr_write(ecx, data);
2542 skip_emulated_instruction(&svm->vcpu);
2547 static int msr_interception(struct vcpu_svm *svm)
2549 if (svm->vmcb->control.exit_info_1)
2550 return wrmsr_interception(svm);
2552 return rdmsr_interception(svm);
2555 static int interrupt_window_interception(struct vcpu_svm *svm)
2557 struct kvm_run *kvm_run = svm->vcpu.run;
2559 svm_clear_vintr(svm);
2560 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2562 * If the user space waits to inject interrupts, exit as soon as
2565 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2566 kvm_run->request_interrupt_window &&
2567 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2568 ++svm->vcpu.stat.irq_window_exits;
2569 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2576 static int pause_interception(struct vcpu_svm *svm)
2578 kvm_vcpu_on_spin(&(svm->vcpu));
2582 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2583 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2584 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2585 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2586 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2587 [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
2588 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
2589 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2590 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2591 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2592 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2593 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2594 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2595 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2596 [SVM_EXIT_READ_DR4] = emulate_on_interception,
2597 [SVM_EXIT_READ_DR5] = emulate_on_interception,
2598 [SVM_EXIT_READ_DR6] = emulate_on_interception,
2599 [SVM_EXIT_READ_DR7] = emulate_on_interception,
2600 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2601 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2602 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2603 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2604 [SVM_EXIT_WRITE_DR4] = emulate_on_interception,
2605 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2606 [SVM_EXIT_WRITE_DR6] = emulate_on_interception,
2607 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2608 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2609 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2610 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2611 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2612 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2613 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2614 [SVM_EXIT_INTR] = intr_interception,
2615 [SVM_EXIT_NMI] = nmi_interception,
2616 [SVM_EXIT_SMI] = nop_on_interception,
2617 [SVM_EXIT_INIT] = nop_on_interception,
2618 [SVM_EXIT_VINTR] = interrupt_window_interception,
2619 [SVM_EXIT_CPUID] = cpuid_interception,
2620 [SVM_EXIT_IRET] = iret_interception,
2621 [SVM_EXIT_INVD] = emulate_on_interception,
2622 [SVM_EXIT_PAUSE] = pause_interception,
2623 [SVM_EXIT_HLT] = halt_interception,
2624 [SVM_EXIT_INVLPG] = invlpg_interception,
2625 [SVM_EXIT_INVLPGA] = invlpga_interception,
2626 [SVM_EXIT_IOIO] = io_interception,
2627 [SVM_EXIT_MSR] = msr_interception,
2628 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2629 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2630 [SVM_EXIT_VMRUN] = vmrun_interception,
2631 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2632 [SVM_EXIT_VMLOAD] = vmload_interception,
2633 [SVM_EXIT_VMSAVE] = vmsave_interception,
2634 [SVM_EXIT_STGI] = stgi_interception,
2635 [SVM_EXIT_CLGI] = clgi_interception,
2636 [SVM_EXIT_SKINIT] = skinit_interception,
2637 [SVM_EXIT_WBINVD] = emulate_on_interception,
2638 [SVM_EXIT_MONITOR] = invalid_op_interception,
2639 [SVM_EXIT_MWAIT] = invalid_op_interception,
2640 [SVM_EXIT_NPF] = pf_interception,
2643 static int handle_exit(struct kvm_vcpu *vcpu)
2645 struct vcpu_svm *svm = to_svm(vcpu);
2646 struct kvm_run *kvm_run = vcpu->run;
2647 u32 exit_code = svm->vmcb->control.exit_code;
2649 trace_kvm_exit(exit_code, vcpu);
2651 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2652 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2654 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2656 if (unlikely(svm->nested.exit_required)) {
2657 nested_svm_vmexit(svm);
2658 svm->nested.exit_required = false;
2663 if (is_nested(svm)) {
2666 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2667 svm->vmcb->control.exit_info_1,
2668 svm->vmcb->control.exit_info_2,
2669 svm->vmcb->control.exit_int_info,
2670 svm->vmcb->control.exit_int_info_err);
2672 vmexit = nested_svm_exit_special(svm);
2674 if (vmexit == NESTED_EXIT_CONTINUE)
2675 vmexit = nested_svm_exit_handled(svm);
2677 if (vmexit == NESTED_EXIT_DONE)
2681 svm_complete_interrupts(svm);
2683 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2684 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2685 kvm_run->fail_entry.hardware_entry_failure_reason
2686 = svm->vmcb->control.exit_code;
2690 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2691 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2692 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2693 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2695 __func__, svm->vmcb->control.exit_int_info,
2698 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2699 || !svm_exit_handlers[exit_code]) {
2700 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2701 kvm_run->hw.hardware_exit_reason = exit_code;
2705 return svm_exit_handlers[exit_code](svm);
2708 static void reload_tss(struct kvm_vcpu *vcpu)
2710 int cpu = raw_smp_processor_id();
2712 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2713 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
2717 static void pre_svm_run(struct vcpu_svm *svm)
2719 int cpu = raw_smp_processor_id();
2721 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2723 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2724 /* FIXME: handle wraparound of asid_generation */
2725 if (svm->asid_generation != sd->asid_generation)
2729 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2731 struct vcpu_svm *svm = to_svm(vcpu);
2733 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2734 vcpu->arch.hflags |= HF_NMI_MASK;
2735 svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
2736 ++vcpu->stat.nmi_injections;
2739 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2741 struct vmcb_control_area *control;
2743 trace_kvm_inj_virq(irq);
2745 ++svm->vcpu.stat.irq_injections;
2746 control = &svm->vmcb->control;
2747 control->int_vector = irq;
2748 control->int_ctl &= ~V_INTR_PRIO_MASK;
2749 control->int_ctl |= V_IRQ_MASK |
2750 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2753 static void svm_set_irq(struct kvm_vcpu *vcpu)
2755 struct vcpu_svm *svm = to_svm(vcpu);
2757 BUG_ON(!(gif_set(svm)));
2759 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2760 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2763 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2765 struct vcpu_svm *svm = to_svm(vcpu);
2767 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2774 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2777 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2779 struct vcpu_svm *svm = to_svm(vcpu);
2780 struct vmcb *vmcb = svm->vmcb;
2782 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2783 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2784 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
2789 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
2791 struct vcpu_svm *svm = to_svm(vcpu);
2793 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
2796 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2798 struct vcpu_svm *svm = to_svm(vcpu);
2801 svm->vcpu.arch.hflags |= HF_NMI_MASK;
2802 svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
2804 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
2805 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
2809 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2811 struct vcpu_svm *svm = to_svm(vcpu);
2812 struct vmcb *vmcb = svm->vmcb;
2815 if (!gif_set(svm) ||
2816 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
2819 ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
2822 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
2827 static void enable_irq_window(struct kvm_vcpu *vcpu)
2829 struct vcpu_svm *svm = to_svm(vcpu);
2832 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
2833 * 1, because that's a separate STGI/VMRUN intercept. The next time we
2834 * get that intercept, this function will be called again though and
2835 * we'll get the vintr intercept.
2837 if (gif_set(svm) && nested_svm_intr(svm)) {
2839 svm_inject_irq(svm, 0x0);
2843 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2845 struct vcpu_svm *svm = to_svm(vcpu);
2847 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2849 return; /* IRET will cause a vm exit */
2852 * Something prevents NMI from been injected. Single step over possible
2853 * problem (IRET or exception injection or interrupt shadow)
2855 svm->nmi_singlestep = true;
2856 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2857 update_db_intercept(vcpu);
2860 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2865 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2867 force_new_asid(vcpu);
2870 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2874 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2876 struct vcpu_svm *svm = to_svm(vcpu);
2878 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2881 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2882 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2883 kvm_set_cr8(vcpu, cr8);
2887 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2889 struct vcpu_svm *svm = to_svm(vcpu);
2892 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2895 cr8 = kvm_get_cr8(vcpu);
2896 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2897 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2900 static void svm_complete_interrupts(struct vcpu_svm *svm)
2904 u32 exitintinfo = svm->vmcb->control.exit_int_info;
2905 unsigned int3_injected = svm->int3_injected;
2907 svm->int3_injected = 0;
2909 if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2910 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2912 svm->vcpu.arch.nmi_injected = false;
2913 kvm_clear_exception_queue(&svm->vcpu);
2914 kvm_clear_interrupt_queue(&svm->vcpu);
2916 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2919 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2920 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2923 case SVM_EXITINTINFO_TYPE_NMI:
2924 svm->vcpu.arch.nmi_injected = true;
2926 case SVM_EXITINTINFO_TYPE_EXEPT:
2928 * In case of software exceptions, do not reinject the vector,
2929 * but re-execute the instruction instead. Rewind RIP first
2930 * if we emulated INT3 before.
2932 if (kvm_exception_is_soft(vector)) {
2933 if (vector == BP_VECTOR && int3_injected &&
2934 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
2935 kvm_rip_write(&svm->vcpu,
2936 kvm_rip_read(&svm->vcpu) -
2940 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2941 u32 err = svm->vmcb->control.exit_int_info_err;
2942 kvm_requeue_exception_e(&svm->vcpu, vector, err);
2945 kvm_requeue_exception(&svm->vcpu, vector);
2947 case SVM_EXITINTINFO_TYPE_INTR:
2948 kvm_queue_interrupt(&svm->vcpu, vector, false);
2955 #ifdef CONFIG_X86_64
2961 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
2963 struct vcpu_svm *svm = to_svm(vcpu);
2968 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2969 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2970 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2973 * A vmexit emulation is required before the vcpu can be executed
2976 if (unlikely(svm->nested.exit_required))
2981 sync_lapic_to_cr8(vcpu);
2983 save_host_msrs(vcpu);
2984 fs_selector = kvm_read_fs();
2985 gs_selector = kvm_read_gs();
2986 ldt_selector = kvm_read_ldt();
2987 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2988 /* required for live migration with NPT */
2990 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2997 "push %%"R"bp; \n\t"
2998 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2999 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
3000 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
3001 "mov %c[rsi](%[svm]), %%"R"si \n\t"
3002 "mov %c[rdi](%[svm]), %%"R"di \n\t"
3003 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
3004 #ifdef CONFIG_X86_64
3005 "mov %c[r8](%[svm]), %%r8 \n\t"
3006 "mov %c[r9](%[svm]), %%r9 \n\t"
3007 "mov %c[r10](%[svm]), %%r10 \n\t"
3008 "mov %c[r11](%[svm]), %%r11 \n\t"
3009 "mov %c[r12](%[svm]), %%r12 \n\t"
3010 "mov %c[r13](%[svm]), %%r13 \n\t"
3011 "mov %c[r14](%[svm]), %%r14 \n\t"
3012 "mov %c[r15](%[svm]), %%r15 \n\t"
3015 /* Enter guest mode */
3017 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
3018 __ex(SVM_VMLOAD) "\n\t"
3019 __ex(SVM_VMRUN) "\n\t"
3020 __ex(SVM_VMSAVE) "\n\t"
3023 /* Save guest registers, load host registers */
3024 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
3025 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
3026 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
3027 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
3028 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
3029 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
3030 #ifdef CONFIG_X86_64
3031 "mov %%r8, %c[r8](%[svm]) \n\t"
3032 "mov %%r9, %c[r9](%[svm]) \n\t"
3033 "mov %%r10, %c[r10](%[svm]) \n\t"
3034 "mov %%r11, %c[r11](%[svm]) \n\t"
3035 "mov %%r12, %c[r12](%[svm]) \n\t"
3036 "mov %%r13, %c[r13](%[svm]) \n\t"
3037 "mov %%r14, %c[r14](%[svm]) \n\t"
3038 "mov %%r15, %c[r15](%[svm]) \n\t"
3043 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3044 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3045 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3046 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3047 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3048 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3049 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3050 #ifdef CONFIG_X86_64
3051 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3052 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3053 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3054 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3055 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3056 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3057 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3058 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3061 , R"bx", R"cx", R"dx", R"si", R"di"
3062 #ifdef CONFIG_X86_64
3063 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3067 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3068 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3069 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3070 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3072 kvm_load_fs(fs_selector);
3073 kvm_load_gs(gs_selector);
3074 kvm_load_ldt(ldt_selector);
3075 load_host_msrs(vcpu);
3079 local_irq_disable();
3083 sync_cr8_to_lapic(vcpu);
3088 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3089 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3095 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3097 struct vcpu_svm *svm = to_svm(vcpu);
3100 svm->vmcb->control.nested_cr3 = root;
3101 force_new_asid(vcpu);
3105 svm->vmcb->save.cr3 = root;
3106 force_new_asid(vcpu);
3109 static int is_disabled(void)
3113 rdmsrl(MSR_VM_CR, vm_cr);
3114 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3121 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3124 * Patch in the VMMCALL instruction:
3126 hypercall[0] = 0x0f;
3127 hypercall[1] = 0x01;
3128 hypercall[2] = 0xd9;
3131 static void svm_check_processor_compat(void *rtn)
3136 static bool svm_cpu_has_accelerated_tpr(void)
3141 static int get_npt_level(void)
3143 #ifdef CONFIG_X86_64
3144 return PT64_ROOT_LEVEL;
3146 return PT32E_ROOT_LEVEL;
3150 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3155 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3159 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
3163 entry->eax = 1; /* SVM revision 1 */
3164 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
3165 ASID emulation to nested SVM */
3166 entry->ecx = 0; /* Reserved */
3167 entry->edx = 0; /* Do not support any additional features */
3173 static const struct trace_print_flags svm_exit_reasons_str[] = {
3174 { SVM_EXIT_READ_CR0, "read_cr0" },
3175 { SVM_EXIT_READ_CR3, "read_cr3" },
3176 { SVM_EXIT_READ_CR4, "read_cr4" },
3177 { SVM_EXIT_READ_CR8, "read_cr8" },
3178 { SVM_EXIT_WRITE_CR0, "write_cr0" },
3179 { SVM_EXIT_WRITE_CR3, "write_cr3" },
3180 { SVM_EXIT_WRITE_CR4, "write_cr4" },
3181 { SVM_EXIT_WRITE_CR8, "write_cr8" },
3182 { SVM_EXIT_READ_DR0, "read_dr0" },
3183 { SVM_EXIT_READ_DR1, "read_dr1" },
3184 { SVM_EXIT_READ_DR2, "read_dr2" },
3185 { SVM_EXIT_READ_DR3, "read_dr3" },
3186 { SVM_EXIT_WRITE_DR0, "write_dr0" },
3187 { SVM_EXIT_WRITE_DR1, "write_dr1" },
3188 { SVM_EXIT_WRITE_DR2, "write_dr2" },
3189 { SVM_EXIT_WRITE_DR3, "write_dr3" },
3190 { SVM_EXIT_WRITE_DR5, "write_dr5" },
3191 { SVM_EXIT_WRITE_DR7, "write_dr7" },
3192 { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
3193 { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
3194 { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
3195 { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
3196 { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
3197 { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
3198 { SVM_EXIT_INTR, "interrupt" },
3199 { SVM_EXIT_NMI, "nmi" },
3200 { SVM_EXIT_SMI, "smi" },
3201 { SVM_EXIT_INIT, "init" },
3202 { SVM_EXIT_VINTR, "vintr" },
3203 { SVM_EXIT_CPUID, "cpuid" },
3204 { SVM_EXIT_INVD, "invd" },
3205 { SVM_EXIT_HLT, "hlt" },
3206 { SVM_EXIT_INVLPG, "invlpg" },
3207 { SVM_EXIT_INVLPGA, "invlpga" },
3208 { SVM_EXIT_IOIO, "io" },
3209 { SVM_EXIT_MSR, "msr" },
3210 { SVM_EXIT_TASK_SWITCH, "task_switch" },
3211 { SVM_EXIT_SHUTDOWN, "shutdown" },
3212 { SVM_EXIT_VMRUN, "vmrun" },
3213 { SVM_EXIT_VMMCALL, "hypercall" },
3214 { SVM_EXIT_VMLOAD, "vmload" },
3215 { SVM_EXIT_VMSAVE, "vmsave" },
3216 { SVM_EXIT_STGI, "stgi" },
3217 { SVM_EXIT_CLGI, "clgi" },
3218 { SVM_EXIT_SKINIT, "skinit" },
3219 { SVM_EXIT_WBINVD, "wbinvd" },
3220 { SVM_EXIT_MONITOR, "monitor" },
3221 { SVM_EXIT_MWAIT, "mwait" },
3222 { SVM_EXIT_NPF, "npf" },
3226 static int svm_get_lpage_level(void)
3228 return PT_PDPE_LEVEL;
3231 static bool svm_rdtscp_supported(void)
3236 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3238 struct vcpu_svm *svm = to_svm(vcpu);
3240 svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
3242 svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR;
3243 update_cr0_intercept(svm);
3246 static struct kvm_x86_ops svm_x86_ops = {
3247 .cpu_has_kvm_support = has_svm,
3248 .disabled_by_bios = is_disabled,
3249 .hardware_setup = svm_hardware_setup,
3250 .hardware_unsetup = svm_hardware_unsetup,
3251 .check_processor_compatibility = svm_check_processor_compat,
3252 .hardware_enable = svm_hardware_enable,
3253 .hardware_disable = svm_hardware_disable,
3254 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
3256 .vcpu_create = svm_create_vcpu,
3257 .vcpu_free = svm_free_vcpu,
3258 .vcpu_reset = svm_vcpu_reset,
3260 .prepare_guest_switch = svm_prepare_guest_switch,
3261 .vcpu_load = svm_vcpu_load,
3262 .vcpu_put = svm_vcpu_put,
3264 .set_guest_debug = svm_guest_debug,
3265 .get_msr = svm_get_msr,
3266 .set_msr = svm_set_msr,
3267 .get_segment_base = svm_get_segment_base,
3268 .get_segment = svm_get_segment,
3269 .set_segment = svm_set_segment,
3270 .get_cpl = svm_get_cpl,
3271 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
3272 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
3273 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
3274 .set_cr0 = svm_set_cr0,
3275 .set_cr3 = svm_set_cr3,
3276 .set_cr4 = svm_set_cr4,
3277 .set_efer = svm_set_efer,
3278 .get_idt = svm_get_idt,
3279 .set_idt = svm_set_idt,
3280 .get_gdt = svm_get_gdt,
3281 .set_gdt = svm_set_gdt,
3282 .set_dr7 = svm_set_dr7,
3283 .cache_reg = svm_cache_reg,
3284 .get_rflags = svm_get_rflags,
3285 .set_rflags = svm_set_rflags,
3286 .fpu_activate = svm_fpu_activate,
3287 .fpu_deactivate = svm_fpu_deactivate,
3289 .tlb_flush = svm_flush_tlb,
3291 .run = svm_vcpu_run,
3292 .handle_exit = handle_exit,
3293 .skip_emulated_instruction = skip_emulated_instruction,
3294 .set_interrupt_shadow = svm_set_interrupt_shadow,
3295 .get_interrupt_shadow = svm_get_interrupt_shadow,
3296 .patch_hypercall = svm_patch_hypercall,
3297 .set_irq = svm_set_irq,
3298 .set_nmi = svm_inject_nmi,
3299 .queue_exception = svm_queue_exception,
3300 .interrupt_allowed = svm_interrupt_allowed,
3301 .nmi_allowed = svm_nmi_allowed,
3302 .get_nmi_mask = svm_get_nmi_mask,
3303 .set_nmi_mask = svm_set_nmi_mask,
3304 .enable_nmi_window = enable_nmi_window,
3305 .enable_irq_window = enable_irq_window,
3306 .update_cr8_intercept = update_cr8_intercept,
3308 .set_tss_addr = svm_set_tss_addr,
3309 .get_tdp_level = get_npt_level,
3310 .get_mt_mask = svm_get_mt_mask,
3312 .exit_reasons_str = svm_exit_reasons_str,
3313 .get_lpage_level = svm_get_lpage_level,
3315 .cpuid_update = svm_cpuid_update,
3317 .rdtscp_supported = svm_rdtscp_supported,
3319 .set_supported_cpuid = svm_set_supported_cpuid,
3322 static int __init svm_init(void)
3324 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3325 __alignof__(struct vcpu_svm), THIS_MODULE);
3328 static void __exit svm_exit(void)
3333 module_init(svm_init)
3334 module_exit(svm_exit)