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>
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/vmalloc.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
30 #define __ex(x) __kvm_handle_fault_on_reboot(x)
32 MODULE_AUTHOR("Qumranet");
33 MODULE_LICENSE("GPL");
35 #define IOPM_ALLOC_ORDER 2
36 #define MSRPM_ALLOC_ORDER 1
42 #define DR7_GD_MASK (1 << 13)
43 #define DR6_BD_MASK (1 << 13)
45 #define SEG_TYPE_LDT 2
46 #define SEG_TYPE_BUSY_TSS16 3
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
52 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
54 /* enable NPT for AMD64 and X86 with PAE */
55 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
56 static bool npt_enabled = true;
58 static bool npt_enabled = false;
62 module_param(npt, int, S_IRUGO);
64 static void kvm_reput_irq(struct vcpu_svm *svm);
66 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
68 return container_of(vcpu, struct vcpu_svm, vcpu);
71 static unsigned long iopm_base;
73 struct kvm_ldttss_desc {
76 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
77 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
80 } __attribute__((packed));
88 struct kvm_ldttss_desc *tss_desc;
90 struct page *save_area;
93 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
94 static uint32_t svm_features;
96 struct svm_init_data {
101 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
103 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
104 #define MSRS_RANGE_SIZE 2048
105 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
107 #define MAX_INST_SIZE 15
109 static inline u32 svm_has(u32 feat)
111 return svm_features & feat;
114 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
116 int word_index = __ffs(vcpu->arch.irq_summary);
117 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
118 int irq = word_index * BITS_PER_LONG + bit_index;
120 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
121 if (!vcpu->arch.irq_pending[word_index])
122 clear_bit(word_index, &vcpu->arch.irq_summary);
126 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
128 set_bit(irq, vcpu->arch.irq_pending);
129 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
132 static inline void clgi(void)
134 asm volatile (__ex(SVM_CLGI));
137 static inline void stgi(void)
139 asm volatile (__ex(SVM_STGI));
142 static inline void invlpga(unsigned long addr, u32 asid)
144 asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
147 static inline unsigned long kvm_read_cr2(void)
151 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
155 static inline void kvm_write_cr2(unsigned long val)
157 asm volatile ("mov %0, %%cr2" :: "r" (val));
160 static inline unsigned long read_dr6(void)
164 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
168 static inline void write_dr6(unsigned long val)
170 asm volatile ("mov %0, %%dr6" :: "r" (val));
173 static inline unsigned long read_dr7(void)
177 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
181 static inline void write_dr7(unsigned long val)
183 asm volatile ("mov %0, %%dr7" :: "r" (val));
186 static inline void force_new_asid(struct kvm_vcpu *vcpu)
188 to_svm(vcpu)->asid_generation--;
191 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
193 force_new_asid(vcpu);
196 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
198 if (!npt_enabled && !(efer & EFER_LMA))
201 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
202 vcpu->arch.shadow_efer = efer;
205 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
206 bool has_error_code, u32 error_code)
208 struct vcpu_svm *svm = to_svm(vcpu);
210 svm->vmcb->control.event_inj = nr
212 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
213 | SVM_EVTINJ_TYPE_EXEPT;
214 svm->vmcb->control.event_inj_err = error_code;
217 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
219 struct vcpu_svm *svm = to_svm(vcpu);
221 return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
224 static int is_external_interrupt(u32 info)
226 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
227 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
230 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
232 struct vcpu_svm *svm = to_svm(vcpu);
234 if (!svm->next_rip) {
235 printk(KERN_DEBUG "%s: NOP\n", __func__);
238 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
239 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
244 vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
245 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
247 vcpu->arch.interrupt_window_open = 1;
250 static int has_svm(void)
252 uint32_t eax, ebx, ecx, edx;
254 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
255 printk(KERN_INFO "has_svm: not amd\n");
259 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
260 if (eax < SVM_CPUID_FUNC) {
261 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
265 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
266 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
267 printk(KERN_DEBUG "has_svm: svm not available\n");
273 static void svm_hardware_disable(void *garbage)
277 wrmsrl(MSR_VM_HSAVE_PA, 0);
278 rdmsrl(MSR_EFER, efer);
279 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
282 static void svm_hardware_enable(void *garbage)
285 struct svm_cpu_data *svm_data;
287 struct desc_ptr gdt_descr;
288 struct desc_struct *gdt;
289 int me = raw_smp_processor_id();
292 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
295 svm_data = per_cpu(svm_data, me);
298 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
303 svm_data->asid_generation = 1;
304 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
305 svm_data->next_asid = svm_data->max_asid + 1;
307 asm volatile ("sgdt %0" : "=m"(gdt_descr));
308 gdt = (struct desc_struct *)gdt_descr.address;
309 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
311 rdmsrl(MSR_EFER, efer);
312 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
314 wrmsrl(MSR_VM_HSAVE_PA,
315 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
318 static void svm_cpu_uninit(int cpu)
320 struct svm_cpu_data *svm_data
321 = per_cpu(svm_data, raw_smp_processor_id());
326 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
327 __free_page(svm_data->save_area);
331 static int svm_cpu_init(int cpu)
333 struct svm_cpu_data *svm_data;
336 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
340 svm_data->save_area = alloc_page(GFP_KERNEL);
342 if (!svm_data->save_area)
345 per_cpu(svm_data, cpu) = svm_data;
355 static void set_msr_interception(u32 *msrpm, unsigned msr,
360 for (i = 0; i < NUM_MSR_MAPS; i++) {
361 if (msr >= msrpm_ranges[i] &&
362 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
363 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
364 msrpm_ranges[i]) * 2;
366 u32 *base = msrpm + (msr_offset / 32);
367 u32 msr_shift = msr_offset % 32;
368 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
369 *base = (*base & ~(0x3 << msr_shift)) |
377 static void svm_vcpu_init_msrpm(u32 *msrpm)
379 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
382 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
383 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
384 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
385 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
386 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
387 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
389 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
390 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
391 set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
392 set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
395 static void svm_enable_lbrv(struct vcpu_svm *svm)
397 u32 *msrpm = svm->msrpm;
399 svm->vmcb->control.lbr_ctl = 1;
400 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
401 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
402 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
403 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
406 static void svm_disable_lbrv(struct vcpu_svm *svm)
408 u32 *msrpm = svm->msrpm;
410 svm->vmcb->control.lbr_ctl = 0;
411 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
412 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
413 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
414 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
417 static __init int svm_hardware_setup(void)
420 struct page *iopm_pages;
424 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
429 iopm_va = page_address(iopm_pages);
430 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
431 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
432 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
434 if (boot_cpu_has(X86_FEATURE_NX))
435 kvm_enable_efer_bits(EFER_NX);
437 for_each_online_cpu(cpu) {
438 r = svm_cpu_init(cpu);
443 svm_features = cpuid_edx(SVM_CPUID_FUNC);
445 if (!svm_has(SVM_FEATURE_NPT))
448 if (npt_enabled && !npt) {
449 printk(KERN_INFO "kvm: Nested Paging disabled\n");
454 printk(KERN_INFO "kvm: Nested Paging enabled\n");
462 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
467 static __exit void svm_hardware_unsetup(void)
471 for_each_online_cpu(cpu)
474 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
478 static void init_seg(struct vmcb_seg *seg)
481 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
482 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
487 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
490 seg->attrib = SVM_SELECTOR_P_MASK | type;
495 static void init_vmcb(struct vcpu_svm *svm)
497 struct vmcb_control_area *control = &svm->vmcb->control;
498 struct vmcb_save_area *save = &svm->vmcb->save;
500 control->intercept_cr_read = INTERCEPT_CR0_MASK |
504 control->intercept_cr_write = INTERCEPT_CR0_MASK |
509 control->intercept_dr_read = INTERCEPT_DR0_MASK |
514 control->intercept_dr_write = INTERCEPT_DR0_MASK |
521 control->intercept_exceptions = (1 << PF_VECTOR) |
526 control->intercept = (1ULL << INTERCEPT_INTR) |
527 (1ULL << INTERCEPT_NMI) |
528 (1ULL << INTERCEPT_SMI) |
529 (1ULL << INTERCEPT_CPUID) |
530 (1ULL << INTERCEPT_INVD) |
531 (1ULL << INTERCEPT_HLT) |
532 (1ULL << INTERCEPT_INVLPGA) |
533 (1ULL << INTERCEPT_IOIO_PROT) |
534 (1ULL << INTERCEPT_MSR_PROT) |
535 (1ULL << INTERCEPT_TASK_SWITCH) |
536 (1ULL << INTERCEPT_SHUTDOWN) |
537 (1ULL << INTERCEPT_VMRUN) |
538 (1ULL << INTERCEPT_VMMCALL) |
539 (1ULL << INTERCEPT_VMLOAD) |
540 (1ULL << INTERCEPT_VMSAVE) |
541 (1ULL << INTERCEPT_STGI) |
542 (1ULL << INTERCEPT_CLGI) |
543 (1ULL << INTERCEPT_SKINIT) |
544 (1ULL << INTERCEPT_WBINVD) |
545 (1ULL << INTERCEPT_MONITOR) |
546 (1ULL << INTERCEPT_MWAIT);
548 control->iopm_base_pa = iopm_base;
549 control->msrpm_base_pa = __pa(svm->msrpm);
550 control->tsc_offset = 0;
551 control->int_ctl = V_INTR_MASKING_MASK;
559 save->cs.selector = 0xf000;
560 /* Executable/Readable Code Segment */
561 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
562 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
563 save->cs.limit = 0xffff;
565 * cs.base should really be 0xffff0000, but vmx can't handle that, so
566 * be consistent with it.
568 * Replace when we have real mode working for vmx.
570 save->cs.base = 0xf0000;
572 save->gdtr.limit = 0xffff;
573 save->idtr.limit = 0xffff;
575 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
576 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
578 save->efer = MSR_EFER_SVME_MASK;
579 save->dr6 = 0xffff0ff0;
582 save->rip = 0x0000fff0;
585 * cr0 val on cpu init should be 0x60000010, we enable cpu
586 * cache by default. the orderly way is to enable cache in bios.
588 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
589 save->cr4 = X86_CR4_PAE;
593 /* Setup VMCB for Nested Paging */
594 control->nested_ctl = 1;
595 control->intercept &= ~(1ULL << INTERCEPT_TASK_SWITCH);
596 control->intercept_exceptions &= ~(1 << PF_VECTOR);
597 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
599 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
601 save->g_pat = 0x0007040600070406ULL;
602 /* enable caching because the QEMU Bios doesn't enable it */
603 save->cr0 = X86_CR0_ET;
607 force_new_asid(&svm->vcpu);
610 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
612 struct vcpu_svm *svm = to_svm(vcpu);
616 if (vcpu->vcpu_id != 0) {
617 svm->vmcb->save.rip = 0;
618 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
619 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
625 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
627 struct vcpu_svm *svm;
629 struct page *msrpm_pages;
632 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
638 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
642 page = alloc_page(GFP_KERNEL);
649 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
652 svm->msrpm = page_address(msrpm_pages);
653 svm_vcpu_init_msrpm(svm->msrpm);
655 svm->vmcb = page_address(page);
656 clear_page(svm->vmcb);
657 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
658 svm->asid_generation = 0;
659 memset(svm->db_regs, 0, sizeof(svm->db_regs));
663 svm->vcpu.fpu_active = 1;
664 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
665 if (svm->vcpu.vcpu_id == 0)
666 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
671 kvm_vcpu_uninit(&svm->vcpu);
673 kmem_cache_free(kvm_vcpu_cache, svm);
678 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
680 struct vcpu_svm *svm = to_svm(vcpu);
682 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
683 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
684 kvm_vcpu_uninit(vcpu);
685 kmem_cache_free(kvm_vcpu_cache, svm);
688 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
690 struct vcpu_svm *svm = to_svm(vcpu);
693 if (unlikely(cpu != vcpu->cpu)) {
697 * Make sure that the guest sees a monotonically
701 delta = vcpu->arch.host_tsc - tsc_this;
702 svm->vmcb->control.tsc_offset += delta;
704 kvm_migrate_timers(vcpu);
707 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
708 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
711 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
713 struct vcpu_svm *svm = to_svm(vcpu);
716 ++vcpu->stat.host_state_reload;
717 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
718 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
720 rdtscll(vcpu->arch.host_tsc);
723 static void svm_cache_regs(struct kvm_vcpu *vcpu)
725 struct vcpu_svm *svm = to_svm(vcpu);
727 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
728 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
729 vcpu->arch.rip = svm->vmcb->save.rip;
732 static void svm_decache_regs(struct kvm_vcpu *vcpu)
734 struct vcpu_svm *svm = to_svm(vcpu);
735 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
736 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
737 svm->vmcb->save.rip = vcpu->arch.rip;
740 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
742 return to_svm(vcpu)->vmcb->save.rflags;
745 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
747 to_svm(vcpu)->vmcb->save.rflags = rflags;
750 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
752 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
755 case VCPU_SREG_CS: return &save->cs;
756 case VCPU_SREG_DS: return &save->ds;
757 case VCPU_SREG_ES: return &save->es;
758 case VCPU_SREG_FS: return &save->fs;
759 case VCPU_SREG_GS: return &save->gs;
760 case VCPU_SREG_SS: return &save->ss;
761 case VCPU_SREG_TR: return &save->tr;
762 case VCPU_SREG_LDTR: return &save->ldtr;
768 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
770 struct vmcb_seg *s = svm_seg(vcpu, seg);
775 static void svm_get_segment(struct kvm_vcpu *vcpu,
776 struct kvm_segment *var, int seg)
778 struct vmcb_seg *s = svm_seg(vcpu, seg);
781 var->limit = s->limit;
782 var->selector = s->selector;
783 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
784 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
785 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
786 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
787 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
788 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
789 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
790 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
791 var->unusable = !var->present;
794 static int svm_get_cpl(struct kvm_vcpu *vcpu)
796 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
801 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
803 struct vcpu_svm *svm = to_svm(vcpu);
805 dt->limit = svm->vmcb->save.idtr.limit;
806 dt->base = svm->vmcb->save.idtr.base;
809 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
811 struct vcpu_svm *svm = to_svm(vcpu);
813 svm->vmcb->save.idtr.limit = dt->limit;
814 svm->vmcb->save.idtr.base = dt->base ;
817 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
819 struct vcpu_svm *svm = to_svm(vcpu);
821 dt->limit = svm->vmcb->save.gdtr.limit;
822 dt->base = svm->vmcb->save.gdtr.base;
825 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
827 struct vcpu_svm *svm = to_svm(vcpu);
829 svm->vmcb->save.gdtr.limit = dt->limit;
830 svm->vmcb->save.gdtr.base = dt->base ;
833 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
837 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
839 struct vcpu_svm *svm = to_svm(vcpu);
842 if (vcpu->arch.shadow_efer & EFER_LME) {
843 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
844 vcpu->arch.shadow_efer |= EFER_LMA;
845 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
848 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
849 vcpu->arch.shadow_efer &= ~EFER_LMA;
850 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
857 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
858 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
859 vcpu->fpu_active = 1;
862 vcpu->arch.cr0 = cr0;
863 cr0 |= X86_CR0_PG | X86_CR0_WP;
864 if (!vcpu->fpu_active) {
865 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
870 * re-enable caching here because the QEMU bios
871 * does not do it - this results in some delay at
874 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
875 svm->vmcb->save.cr0 = cr0;
878 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
880 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
882 vcpu->arch.cr4 = cr4;
886 to_svm(vcpu)->vmcb->save.cr4 = cr4;
889 static void svm_set_segment(struct kvm_vcpu *vcpu,
890 struct kvm_segment *var, int seg)
892 struct vcpu_svm *svm = to_svm(vcpu);
893 struct vmcb_seg *s = svm_seg(vcpu, seg);
896 s->limit = var->limit;
897 s->selector = var->selector;
901 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
902 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
903 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
904 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
905 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
906 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
907 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
908 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
910 if (seg == VCPU_SREG_CS)
912 = (svm->vmcb->save.cs.attrib
913 >> SVM_SELECTOR_DPL_SHIFT) & 3;
917 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
922 static int svm_get_irq(struct kvm_vcpu *vcpu)
924 struct vcpu_svm *svm = to_svm(vcpu);
925 u32 exit_int_info = svm->vmcb->control.exit_int_info;
927 if (is_external_interrupt(exit_int_info))
928 return exit_int_info & SVM_EVTINJ_VEC_MASK;
932 static void load_host_msrs(struct kvm_vcpu *vcpu)
935 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
939 static void save_host_msrs(struct kvm_vcpu *vcpu)
942 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
946 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
948 if (svm_data->next_asid > svm_data->max_asid) {
949 ++svm_data->asid_generation;
950 svm_data->next_asid = 1;
951 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
954 svm->vcpu.cpu = svm_data->cpu;
955 svm->asid_generation = svm_data->asid_generation;
956 svm->vmcb->control.asid = svm_data->next_asid++;
959 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
961 unsigned long val = to_svm(vcpu)->db_regs[dr];
962 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
966 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
969 struct vcpu_svm *svm = to_svm(vcpu);
973 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
974 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
975 svm->vmcb->save.dr6 |= DR6_BD_MASK;
976 *exception = DB_VECTOR;
982 svm->db_regs[dr] = value;
985 if (vcpu->arch.cr4 & X86_CR4_DE) {
986 *exception = UD_VECTOR;
990 if (value & ~((1ULL << 32) - 1)) {
991 *exception = GP_VECTOR;
994 svm->vmcb->save.dr7 = value;
998 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1000 *exception = UD_VECTOR;
1005 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1007 u32 exit_int_info = svm->vmcb->control.exit_int_info;
1008 struct kvm *kvm = svm->vcpu.kvm;
1012 if (!irqchip_in_kernel(kvm) &&
1013 is_external_interrupt(exit_int_info))
1014 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1016 fault_address = svm->vmcb->control.exit_info_2;
1017 error_code = svm->vmcb->control.exit_info_1;
1020 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1021 (u32)fault_address, (u32)(fault_address >> 32),
1024 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1025 (u32)fault_address, (u32)(fault_address >> 32),
1028 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1031 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1035 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1036 if (er != EMULATE_DONE)
1037 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1041 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1043 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1044 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1045 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1046 svm->vcpu.fpu_active = 1;
1051 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1054 * On an #MC intercept the MCE handler is not called automatically in
1055 * the host. So do it by hand here.
1059 /* not sure if we ever come back to this point */
1064 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1067 * VMCB is undefined after a SHUTDOWN intercept
1068 * so reinitialize it.
1070 clear_page(svm->vmcb);
1073 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1077 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1079 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1080 int size, down, in, string, rep;
1083 ++svm->vcpu.stat.io_exits;
1085 svm->next_rip = svm->vmcb->control.exit_info_2;
1087 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1090 if (emulate_instruction(&svm->vcpu,
1091 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1096 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1097 port = io_info >> 16;
1098 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1099 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1100 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1102 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1105 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1107 KVMTRACE_0D(NMI, &svm->vcpu, handler);
1111 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1113 ++svm->vcpu.stat.irq_exits;
1114 KVMTRACE_0D(INTR, &svm->vcpu, handler);
1118 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1123 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1125 svm->next_rip = svm->vmcb->save.rip + 1;
1126 skip_emulated_instruction(&svm->vcpu);
1127 return kvm_emulate_halt(&svm->vcpu);
1130 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1132 svm->next_rip = svm->vmcb->save.rip + 3;
1133 skip_emulated_instruction(&svm->vcpu);
1134 kvm_emulate_hypercall(&svm->vcpu);
1138 static int invalid_op_interception(struct vcpu_svm *svm,
1139 struct kvm_run *kvm_run)
1141 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1145 static int task_switch_interception(struct vcpu_svm *svm,
1146 struct kvm_run *kvm_run)
1150 tss_selector = (u16)svm->vmcb->control.exit_info_1;
1151 if (svm->vmcb->control.exit_info_2 &
1152 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1153 return kvm_task_switch(&svm->vcpu, tss_selector,
1155 if (svm->vmcb->control.exit_info_2 &
1156 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1157 return kvm_task_switch(&svm->vcpu, tss_selector,
1159 return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1162 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1164 svm->next_rip = svm->vmcb->save.rip + 2;
1165 kvm_emulate_cpuid(&svm->vcpu);
1169 static int emulate_on_interception(struct vcpu_svm *svm,
1170 struct kvm_run *kvm_run)
1172 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1173 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1177 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1179 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1180 if (irqchip_in_kernel(svm->vcpu.kvm))
1182 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1186 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1188 struct vcpu_svm *svm = to_svm(vcpu);
1191 case MSR_IA32_TIME_STAMP_COUNTER: {
1195 *data = svm->vmcb->control.tsc_offset + tsc;
1199 *data = svm->vmcb->save.star;
1201 #ifdef CONFIG_X86_64
1203 *data = svm->vmcb->save.lstar;
1206 *data = svm->vmcb->save.cstar;
1208 case MSR_KERNEL_GS_BASE:
1209 *data = svm->vmcb->save.kernel_gs_base;
1211 case MSR_SYSCALL_MASK:
1212 *data = svm->vmcb->save.sfmask;
1215 case MSR_IA32_SYSENTER_CS:
1216 *data = svm->vmcb->save.sysenter_cs;
1218 case MSR_IA32_SYSENTER_EIP:
1219 *data = svm->vmcb->save.sysenter_eip;
1221 case MSR_IA32_SYSENTER_ESP:
1222 *data = svm->vmcb->save.sysenter_esp;
1224 /* Nobody will change the following 5 values in the VMCB so
1225 we can safely return them on rdmsr. They will always be 0
1226 until LBRV is implemented. */
1227 case MSR_IA32_DEBUGCTLMSR:
1228 *data = svm->vmcb->save.dbgctl;
1230 case MSR_IA32_LASTBRANCHFROMIP:
1231 *data = svm->vmcb->save.br_from;
1233 case MSR_IA32_LASTBRANCHTOIP:
1234 *data = svm->vmcb->save.br_to;
1236 case MSR_IA32_LASTINTFROMIP:
1237 *data = svm->vmcb->save.last_excp_from;
1239 case MSR_IA32_LASTINTTOIP:
1240 *data = svm->vmcb->save.last_excp_to;
1243 return kvm_get_msr_common(vcpu, ecx, data);
1248 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1250 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1253 if (svm_get_msr(&svm->vcpu, ecx, &data))
1254 kvm_inject_gp(&svm->vcpu, 0);
1256 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1257 (u32)(data >> 32), handler);
1259 svm->vmcb->save.rax = data & 0xffffffff;
1260 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1261 svm->next_rip = svm->vmcb->save.rip + 2;
1262 skip_emulated_instruction(&svm->vcpu);
1267 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1269 struct vcpu_svm *svm = to_svm(vcpu);
1272 case MSR_IA32_TIME_STAMP_COUNTER: {
1276 svm->vmcb->control.tsc_offset = data - tsc;
1280 svm->vmcb->save.star = data;
1282 #ifdef CONFIG_X86_64
1284 svm->vmcb->save.lstar = data;
1287 svm->vmcb->save.cstar = data;
1289 case MSR_KERNEL_GS_BASE:
1290 svm->vmcb->save.kernel_gs_base = data;
1292 case MSR_SYSCALL_MASK:
1293 svm->vmcb->save.sfmask = data;
1296 case MSR_IA32_SYSENTER_CS:
1297 svm->vmcb->save.sysenter_cs = data;
1299 case MSR_IA32_SYSENTER_EIP:
1300 svm->vmcb->save.sysenter_eip = data;
1302 case MSR_IA32_SYSENTER_ESP:
1303 svm->vmcb->save.sysenter_esp = data;
1305 case MSR_IA32_DEBUGCTLMSR:
1306 if (!svm_has(SVM_FEATURE_LBRV)) {
1307 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
1311 if (data & DEBUGCTL_RESERVED_BITS)
1314 svm->vmcb->save.dbgctl = data;
1315 if (data & (1ULL<<0))
1316 svm_enable_lbrv(svm);
1318 svm_disable_lbrv(svm);
1320 case MSR_K7_EVNTSEL0:
1321 case MSR_K7_EVNTSEL1:
1322 case MSR_K7_EVNTSEL2:
1323 case MSR_K7_EVNTSEL3:
1324 case MSR_K7_PERFCTR0:
1325 case MSR_K7_PERFCTR1:
1326 case MSR_K7_PERFCTR2:
1327 case MSR_K7_PERFCTR3:
1329 * Just discard all writes to the performance counters; this
1330 * should keep both older linux and windows 64-bit guests
1333 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx, data);
1337 return kvm_set_msr_common(vcpu, ecx, data);
1342 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1344 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1345 u64 data = (svm->vmcb->save.rax & -1u)
1346 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1348 KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
1351 svm->next_rip = svm->vmcb->save.rip + 2;
1352 if (svm_set_msr(&svm->vcpu, ecx, data))
1353 kvm_inject_gp(&svm->vcpu, 0);
1355 skip_emulated_instruction(&svm->vcpu);
1359 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1361 if (svm->vmcb->control.exit_info_1)
1362 return wrmsr_interception(svm, kvm_run);
1364 return rdmsr_interception(svm, kvm_run);
1367 static int interrupt_window_interception(struct vcpu_svm *svm,
1368 struct kvm_run *kvm_run)
1370 KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
1372 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1373 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1375 * If the user space waits to inject interrupts, exit as soon as
1378 if (kvm_run->request_interrupt_window &&
1379 !svm->vcpu.arch.irq_summary) {
1380 ++svm->vcpu.stat.irq_window_exits;
1381 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1388 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1389 struct kvm_run *kvm_run) = {
1390 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1391 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1392 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1393 [SVM_EXIT_READ_CR8] = emulate_on_interception,
1395 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1396 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1397 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1398 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
1399 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1400 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1401 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1402 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1403 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1404 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1405 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1406 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1407 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1408 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1409 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
1410 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1411 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1412 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
1413 [SVM_EXIT_INTR] = intr_interception,
1414 [SVM_EXIT_NMI] = nmi_interception,
1415 [SVM_EXIT_SMI] = nop_on_interception,
1416 [SVM_EXIT_INIT] = nop_on_interception,
1417 [SVM_EXIT_VINTR] = interrupt_window_interception,
1418 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1419 [SVM_EXIT_CPUID] = cpuid_interception,
1420 [SVM_EXIT_INVD] = emulate_on_interception,
1421 [SVM_EXIT_HLT] = halt_interception,
1422 [SVM_EXIT_INVLPG] = emulate_on_interception,
1423 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1424 [SVM_EXIT_IOIO] = io_interception,
1425 [SVM_EXIT_MSR] = msr_interception,
1426 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1427 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1428 [SVM_EXIT_VMRUN] = invalid_op_interception,
1429 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1430 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1431 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1432 [SVM_EXIT_STGI] = invalid_op_interception,
1433 [SVM_EXIT_CLGI] = invalid_op_interception,
1434 [SVM_EXIT_SKINIT] = invalid_op_interception,
1435 [SVM_EXIT_WBINVD] = emulate_on_interception,
1436 [SVM_EXIT_MONITOR] = invalid_op_interception,
1437 [SVM_EXIT_MWAIT] = invalid_op_interception,
1438 [SVM_EXIT_NPF] = pf_interception,
1441 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1443 struct vcpu_svm *svm = to_svm(vcpu);
1444 u32 exit_code = svm->vmcb->control.exit_code;
1446 KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
1447 (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
1451 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
1452 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
1455 vcpu->arch.cr0 = svm->vmcb->save.cr0;
1456 vcpu->arch.cr3 = svm->vmcb->save.cr3;
1457 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1458 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1459 kvm_inject_gp(vcpu, 0);
1464 kvm_mmu_reset_context(vcpu);
1471 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1472 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1473 kvm_run->fail_entry.hardware_entry_failure_reason
1474 = svm->vmcb->control.exit_code;
1478 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1479 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1480 exit_code != SVM_EXIT_NPF)
1481 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1483 __func__, svm->vmcb->control.exit_int_info,
1486 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1487 || !svm_exit_handlers[exit_code]) {
1488 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1489 kvm_run->hw.hardware_exit_reason = exit_code;
1493 return svm_exit_handlers[exit_code](svm, kvm_run);
1496 static void reload_tss(struct kvm_vcpu *vcpu)
1498 int cpu = raw_smp_processor_id();
1500 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1501 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1505 static void pre_svm_run(struct vcpu_svm *svm)
1507 int cpu = raw_smp_processor_id();
1509 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1511 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1512 if (svm->vcpu.cpu != cpu ||
1513 svm->asid_generation != svm_data->asid_generation)
1514 new_asid(svm, svm_data);
1518 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1520 struct vmcb_control_area *control;
1522 KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
1524 control = &svm->vmcb->control;
1525 control->int_vector = irq;
1526 control->int_ctl &= ~V_INTR_PRIO_MASK;
1527 control->int_ctl |= V_IRQ_MASK |
1528 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1531 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1533 struct vcpu_svm *svm = to_svm(vcpu);
1535 svm_inject_irq(svm, irq);
1538 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
1540 struct vcpu_svm *svm = to_svm(vcpu);
1541 struct vmcb *vmcb = svm->vmcb;
1544 if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
1547 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1549 max_irr = kvm_lapic_find_highest_irr(vcpu);
1553 tpr = kvm_lapic_get_cr8(vcpu) << 4;
1555 if (tpr >= (max_irr & 0xf0))
1556 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
1559 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1561 struct vcpu_svm *svm = to_svm(vcpu);
1562 struct vmcb *vmcb = svm->vmcb;
1563 int intr_vector = -1;
1565 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1566 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1567 intr_vector = vmcb->control.exit_int_info &
1568 SVM_EVTINJ_VEC_MASK;
1569 vmcb->control.exit_int_info = 0;
1570 svm_inject_irq(svm, intr_vector);
1574 if (vmcb->control.int_ctl & V_IRQ_MASK)
1577 if (!kvm_cpu_has_interrupt(vcpu))
1580 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1581 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1582 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1583 /* unable to deliver irq, set pending irq */
1584 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1585 svm_inject_irq(svm, 0x0);
1588 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1589 intr_vector = kvm_cpu_get_interrupt(vcpu);
1590 svm_inject_irq(svm, intr_vector);
1591 kvm_timer_intr_post(vcpu, intr_vector);
1593 update_cr8_intercept(vcpu);
1596 static void kvm_reput_irq(struct vcpu_svm *svm)
1598 struct vmcb_control_area *control = &svm->vmcb->control;
1600 if ((control->int_ctl & V_IRQ_MASK)
1601 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1602 control->int_ctl &= ~V_IRQ_MASK;
1603 push_irq(&svm->vcpu, control->int_vector);
1606 svm->vcpu.arch.interrupt_window_open =
1607 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1610 static void svm_do_inject_vector(struct vcpu_svm *svm)
1612 struct kvm_vcpu *vcpu = &svm->vcpu;
1613 int word_index = __ffs(vcpu->arch.irq_summary);
1614 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1615 int irq = word_index * BITS_PER_LONG + bit_index;
1617 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1618 if (!vcpu->arch.irq_pending[word_index])
1619 clear_bit(word_index, &vcpu->arch.irq_summary);
1620 svm_inject_irq(svm, irq);
1623 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1624 struct kvm_run *kvm_run)
1626 struct vcpu_svm *svm = to_svm(vcpu);
1627 struct vmcb_control_area *control = &svm->vmcb->control;
1629 svm->vcpu.arch.interrupt_window_open =
1630 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1631 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1633 if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1635 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1637 svm_do_inject_vector(svm);
1640 * Interrupts blocked. Wait for unblock.
1642 if (!svm->vcpu.arch.interrupt_window_open &&
1643 (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1644 control->intercept |= 1ULL << INTERCEPT_VINTR;
1646 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1649 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1654 static void save_db_regs(unsigned long *db_regs)
1656 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1657 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1658 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1659 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1662 static void load_db_regs(unsigned long *db_regs)
1664 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1665 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1666 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1667 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1670 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1672 force_new_asid(vcpu);
1675 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1679 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
1681 struct vcpu_svm *svm = to_svm(vcpu);
1683 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
1684 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
1685 kvm_lapic_set_tpr(vcpu, cr8);
1689 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
1691 struct vcpu_svm *svm = to_svm(vcpu);
1694 if (!irqchip_in_kernel(vcpu->kvm))
1697 cr8 = kvm_get_cr8(vcpu);
1698 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
1699 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
1702 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1704 struct vcpu_svm *svm = to_svm(vcpu);
1711 sync_lapic_to_cr8(vcpu);
1713 save_host_msrs(vcpu);
1714 fs_selector = kvm_read_fs();
1715 gs_selector = kvm_read_gs();
1716 ldt_selector = kvm_read_ldt();
1717 svm->host_cr2 = kvm_read_cr2();
1718 svm->host_dr6 = read_dr6();
1719 svm->host_dr7 = read_dr7();
1720 svm->vmcb->save.cr2 = vcpu->arch.cr2;
1721 /* required for live migration with NPT */
1723 svm->vmcb->save.cr3 = vcpu->arch.cr3;
1725 if (svm->vmcb->save.dr7 & 0xff) {
1727 save_db_regs(svm->host_db_regs);
1728 load_db_regs(svm->db_regs);
1736 #ifdef CONFIG_X86_64
1742 #ifdef CONFIG_X86_64
1743 "mov %c[rbx](%[svm]), %%rbx \n\t"
1744 "mov %c[rcx](%[svm]), %%rcx \n\t"
1745 "mov %c[rdx](%[svm]), %%rdx \n\t"
1746 "mov %c[rsi](%[svm]), %%rsi \n\t"
1747 "mov %c[rdi](%[svm]), %%rdi \n\t"
1748 "mov %c[rbp](%[svm]), %%rbp \n\t"
1749 "mov %c[r8](%[svm]), %%r8 \n\t"
1750 "mov %c[r9](%[svm]), %%r9 \n\t"
1751 "mov %c[r10](%[svm]), %%r10 \n\t"
1752 "mov %c[r11](%[svm]), %%r11 \n\t"
1753 "mov %c[r12](%[svm]), %%r12 \n\t"
1754 "mov %c[r13](%[svm]), %%r13 \n\t"
1755 "mov %c[r14](%[svm]), %%r14 \n\t"
1756 "mov %c[r15](%[svm]), %%r15 \n\t"
1758 "mov %c[rbx](%[svm]), %%ebx \n\t"
1759 "mov %c[rcx](%[svm]), %%ecx \n\t"
1760 "mov %c[rdx](%[svm]), %%edx \n\t"
1761 "mov %c[rsi](%[svm]), %%esi \n\t"
1762 "mov %c[rdi](%[svm]), %%edi \n\t"
1763 "mov %c[rbp](%[svm]), %%ebp \n\t"
1766 #ifdef CONFIG_X86_64
1767 /* Enter guest mode */
1769 "mov %c[vmcb](%[svm]), %%rax \n\t"
1770 __ex(SVM_VMLOAD) "\n\t"
1771 __ex(SVM_VMRUN) "\n\t"
1772 __ex(SVM_VMSAVE) "\n\t"
1775 /* Enter guest mode */
1777 "mov %c[vmcb](%[svm]), %%eax \n\t"
1778 __ex(SVM_VMLOAD) "\n\t"
1779 __ex(SVM_VMRUN) "\n\t"
1780 __ex(SVM_VMSAVE) "\n\t"
1784 /* Save guest registers, load host registers */
1785 #ifdef CONFIG_X86_64
1786 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1787 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1788 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1789 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1790 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1791 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1792 "mov %%r8, %c[r8](%[svm]) \n\t"
1793 "mov %%r9, %c[r9](%[svm]) \n\t"
1794 "mov %%r10, %c[r10](%[svm]) \n\t"
1795 "mov %%r11, %c[r11](%[svm]) \n\t"
1796 "mov %%r12, %c[r12](%[svm]) \n\t"
1797 "mov %%r13, %c[r13](%[svm]) \n\t"
1798 "mov %%r14, %c[r14](%[svm]) \n\t"
1799 "mov %%r15, %c[r15](%[svm]) \n\t"
1803 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1804 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1805 "mov %%edx, %c[rdx](%[svm]) \n\t"
1806 "mov %%esi, %c[rsi](%[svm]) \n\t"
1807 "mov %%edi, %c[rdi](%[svm]) \n\t"
1808 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1814 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1815 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1816 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1817 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1818 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1819 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1820 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1821 #ifdef CONFIG_X86_64
1822 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1823 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1824 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1825 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1826 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1827 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1828 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1829 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1832 #ifdef CONFIG_X86_64
1833 , "rbx", "rcx", "rdx", "rsi", "rdi"
1834 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1836 , "ebx", "ecx", "edx" , "esi", "edi"
1840 if ((svm->vmcb->save.dr7 & 0xff))
1841 load_db_regs(svm->host_db_regs);
1843 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1845 write_dr6(svm->host_dr6);
1846 write_dr7(svm->host_dr7);
1847 kvm_write_cr2(svm->host_cr2);
1849 kvm_load_fs(fs_selector);
1850 kvm_load_gs(gs_selector);
1851 kvm_load_ldt(ldt_selector);
1852 load_host_msrs(vcpu);
1856 local_irq_disable();
1860 sync_cr8_to_lapic(vcpu);
1865 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1867 struct vcpu_svm *svm = to_svm(vcpu);
1870 svm->vmcb->control.nested_cr3 = root;
1871 force_new_asid(vcpu);
1875 svm->vmcb->save.cr3 = root;
1876 force_new_asid(vcpu);
1878 if (vcpu->fpu_active) {
1879 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1880 svm->vmcb->save.cr0 |= X86_CR0_TS;
1881 vcpu->fpu_active = 0;
1885 static int is_disabled(void)
1889 rdmsrl(MSR_VM_CR, vm_cr);
1890 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1897 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1900 * Patch in the VMMCALL instruction:
1902 hypercall[0] = 0x0f;
1903 hypercall[1] = 0x01;
1904 hypercall[2] = 0xd9;
1907 static void svm_check_processor_compat(void *rtn)
1912 static bool svm_cpu_has_accelerated_tpr(void)
1917 static int get_npt_level(void)
1919 #ifdef CONFIG_X86_64
1920 return PT64_ROOT_LEVEL;
1922 return PT32E_ROOT_LEVEL;
1926 static struct kvm_x86_ops svm_x86_ops = {
1927 .cpu_has_kvm_support = has_svm,
1928 .disabled_by_bios = is_disabled,
1929 .hardware_setup = svm_hardware_setup,
1930 .hardware_unsetup = svm_hardware_unsetup,
1931 .check_processor_compatibility = svm_check_processor_compat,
1932 .hardware_enable = svm_hardware_enable,
1933 .hardware_disable = svm_hardware_disable,
1934 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1936 .vcpu_create = svm_create_vcpu,
1937 .vcpu_free = svm_free_vcpu,
1938 .vcpu_reset = svm_vcpu_reset,
1940 .prepare_guest_switch = svm_prepare_guest_switch,
1941 .vcpu_load = svm_vcpu_load,
1942 .vcpu_put = svm_vcpu_put,
1944 .set_guest_debug = svm_guest_debug,
1945 .get_msr = svm_get_msr,
1946 .set_msr = svm_set_msr,
1947 .get_segment_base = svm_get_segment_base,
1948 .get_segment = svm_get_segment,
1949 .set_segment = svm_set_segment,
1950 .get_cpl = svm_get_cpl,
1951 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1952 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1953 .set_cr0 = svm_set_cr0,
1954 .set_cr3 = svm_set_cr3,
1955 .set_cr4 = svm_set_cr4,
1956 .set_efer = svm_set_efer,
1957 .get_idt = svm_get_idt,
1958 .set_idt = svm_set_idt,
1959 .get_gdt = svm_get_gdt,
1960 .set_gdt = svm_set_gdt,
1961 .get_dr = svm_get_dr,
1962 .set_dr = svm_set_dr,
1963 .cache_regs = svm_cache_regs,
1964 .decache_regs = svm_decache_regs,
1965 .get_rflags = svm_get_rflags,
1966 .set_rflags = svm_set_rflags,
1968 .tlb_flush = svm_flush_tlb,
1970 .run = svm_vcpu_run,
1971 .handle_exit = handle_exit,
1972 .skip_emulated_instruction = skip_emulated_instruction,
1973 .patch_hypercall = svm_patch_hypercall,
1974 .get_irq = svm_get_irq,
1975 .set_irq = svm_set_irq,
1976 .queue_exception = svm_queue_exception,
1977 .exception_injected = svm_exception_injected,
1978 .inject_pending_irq = svm_intr_assist,
1979 .inject_pending_vectors = do_interrupt_requests,
1981 .set_tss_addr = svm_set_tss_addr,
1982 .get_tdp_level = get_npt_level,
1985 static int __init svm_init(void)
1987 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1991 static void __exit svm_exit(void)
1996 module_init(svm_init)
1997 module_exit(svm_exit)