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 MODULE_AUTHOR("Qumranet");
31 MODULE_LICENSE("GPL");
33 #define IOPM_ALLOC_ORDER 2
34 #define MSRPM_ALLOC_ORDER 1
40 #define DR7_GD_MASK (1 << 13)
41 #define DR6_BD_MASK (1 << 13)
43 #define SEG_TYPE_LDT 2
44 #define SEG_TYPE_BUSY_TSS16 3
46 #define SVM_FEATURE_NPT (1 << 0)
47 #define SVM_FEATURE_LBRV (1 << 1)
48 #define SVM_DEATURE_SVML (1 << 2)
50 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
52 /* enable NPT for AMD64 and X86 with PAE */
53 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
54 static bool npt_enabled = true;
56 static bool npt_enabled = false;
60 module_param(npt, int, S_IRUGO);
62 static void kvm_reput_irq(struct vcpu_svm *svm);
64 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
66 return container_of(vcpu, struct vcpu_svm, vcpu);
69 static unsigned long iopm_base;
71 struct kvm_ldttss_desc {
74 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
75 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
78 } __attribute__((packed));
86 struct kvm_ldttss_desc *tss_desc;
88 struct page *save_area;
91 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
92 static uint32_t svm_features;
94 struct svm_init_data {
99 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
101 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
102 #define MSRS_RANGE_SIZE 2048
103 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
105 #define MAX_INST_SIZE 15
107 static inline u32 svm_has(u32 feat)
109 return svm_features & feat;
112 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
114 int word_index = __ffs(vcpu->arch.irq_summary);
115 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
116 int irq = word_index * BITS_PER_LONG + bit_index;
118 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
119 if (!vcpu->arch.irq_pending[word_index])
120 clear_bit(word_index, &vcpu->arch.irq_summary);
124 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
126 set_bit(irq, vcpu->arch.irq_pending);
127 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
130 static inline void clgi(void)
132 asm volatile (SVM_CLGI);
135 static inline void stgi(void)
137 asm volatile (SVM_STGI);
140 static inline void invlpga(unsigned long addr, u32 asid)
142 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
145 static inline unsigned long kvm_read_cr2(void)
149 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
153 static inline void kvm_write_cr2(unsigned long val)
155 asm volatile ("mov %0, %%cr2" :: "r" (val));
158 static inline unsigned long read_dr6(void)
162 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
166 static inline void write_dr6(unsigned long val)
168 asm volatile ("mov %0, %%dr6" :: "r" (val));
171 static inline unsigned long read_dr7(void)
175 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
179 static inline void write_dr7(unsigned long val)
181 asm volatile ("mov %0, %%dr7" :: "r" (val));
184 static inline void force_new_asid(struct kvm_vcpu *vcpu)
186 to_svm(vcpu)->asid_generation--;
189 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
191 force_new_asid(vcpu);
194 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
196 if (!npt_enabled && !(efer & EFER_LMA))
199 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
200 vcpu->arch.shadow_efer = efer;
203 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
204 bool has_error_code, u32 error_code)
206 struct vcpu_svm *svm = to_svm(vcpu);
208 svm->vmcb->control.event_inj = nr
210 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
211 | SVM_EVTINJ_TYPE_EXEPT;
212 svm->vmcb->control.event_inj_err = error_code;
215 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
217 struct vcpu_svm *svm = to_svm(vcpu);
219 return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
222 static int is_external_interrupt(u32 info)
224 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
225 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
228 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
230 struct vcpu_svm *svm = to_svm(vcpu);
232 if (!svm->next_rip) {
233 printk(KERN_DEBUG "%s: NOP\n", __func__);
236 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
237 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
242 vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
243 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
245 vcpu->arch.interrupt_window_open = 1;
248 static int has_svm(void)
250 uint32_t eax, ebx, ecx, edx;
252 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
253 printk(KERN_INFO "has_svm: not amd\n");
257 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
258 if (eax < SVM_CPUID_FUNC) {
259 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
263 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
264 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
265 printk(KERN_DEBUG "has_svm: svm not available\n");
271 static void svm_hardware_disable(void *garbage)
275 wrmsrl(MSR_VM_HSAVE_PA, 0);
276 rdmsrl(MSR_EFER, efer);
277 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
280 static void svm_hardware_enable(void *garbage)
283 struct svm_cpu_data *svm_data;
285 struct desc_ptr gdt_descr;
286 struct desc_struct *gdt;
287 int me = raw_smp_processor_id();
290 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
293 svm_data = per_cpu(svm_data, me);
296 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
301 svm_data->asid_generation = 1;
302 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
303 svm_data->next_asid = svm_data->max_asid + 1;
305 asm volatile ("sgdt %0" : "=m"(gdt_descr));
306 gdt = (struct desc_struct *)gdt_descr.address;
307 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
309 rdmsrl(MSR_EFER, efer);
310 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
312 wrmsrl(MSR_VM_HSAVE_PA,
313 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
316 static void svm_cpu_uninit(int cpu)
318 struct svm_cpu_data *svm_data
319 = per_cpu(svm_data, raw_smp_processor_id());
324 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
325 __free_page(svm_data->save_area);
329 static int svm_cpu_init(int cpu)
331 struct svm_cpu_data *svm_data;
334 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
338 svm_data->save_area = alloc_page(GFP_KERNEL);
340 if (!svm_data->save_area)
343 per_cpu(svm_data, cpu) = svm_data;
353 static void set_msr_interception(u32 *msrpm, unsigned msr,
358 for (i = 0; i < NUM_MSR_MAPS; i++) {
359 if (msr >= msrpm_ranges[i] &&
360 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
361 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
362 msrpm_ranges[i]) * 2;
364 u32 *base = msrpm + (msr_offset / 32);
365 u32 msr_shift = msr_offset % 32;
366 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
367 *base = (*base & ~(0x3 << msr_shift)) |
375 static void svm_vcpu_init_msrpm(u32 *msrpm)
377 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
380 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
381 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
382 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
383 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
384 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
385 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
387 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
388 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
389 set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
390 set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
393 static void svm_enable_lbrv(struct vcpu_svm *svm)
395 u32 *msrpm = svm->msrpm;
397 svm->vmcb->control.lbr_ctl = 1;
398 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
399 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
400 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
401 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
404 static void svm_disable_lbrv(struct vcpu_svm *svm)
406 u32 *msrpm = svm->msrpm;
408 svm->vmcb->control.lbr_ctl = 0;
409 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
410 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
411 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
412 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
415 static __init int svm_hardware_setup(void)
418 struct page *iopm_pages;
422 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
427 iopm_va = page_address(iopm_pages);
428 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
429 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
430 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
432 if (boot_cpu_has(X86_FEATURE_NX))
433 kvm_enable_efer_bits(EFER_NX);
435 for_each_online_cpu(cpu) {
436 r = svm_cpu_init(cpu);
441 svm_features = cpuid_edx(SVM_CPUID_FUNC);
443 if (!svm_has(SVM_FEATURE_NPT))
446 if (npt_enabled && !npt) {
447 printk(KERN_INFO "kvm: Nested Paging disabled\n");
452 printk(KERN_INFO "kvm: Nested Paging enabled\n");
459 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
464 static __exit void svm_hardware_unsetup(void)
468 for_each_online_cpu(cpu)
471 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
475 static void init_seg(struct vmcb_seg *seg)
478 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
479 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
484 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
487 seg->attrib = SVM_SELECTOR_P_MASK | type;
492 static void init_vmcb(struct vcpu_svm *svm)
494 struct vmcb_control_area *control = &svm->vmcb->control;
495 struct vmcb_save_area *save = &svm->vmcb->save;
497 control->intercept_cr_read = INTERCEPT_CR0_MASK |
501 control->intercept_cr_write = INTERCEPT_CR0_MASK |
506 control->intercept_dr_read = INTERCEPT_DR0_MASK |
511 control->intercept_dr_write = INTERCEPT_DR0_MASK |
518 control->intercept_exceptions = (1 << PF_VECTOR) |
523 control->intercept = (1ULL << INTERCEPT_INTR) |
524 (1ULL << INTERCEPT_NMI) |
525 (1ULL << INTERCEPT_SMI) |
526 (1ULL << INTERCEPT_CPUID) |
527 (1ULL << INTERCEPT_INVD) |
528 (1ULL << INTERCEPT_HLT) |
529 (1ULL << INTERCEPT_INVLPGA) |
530 (1ULL << INTERCEPT_IOIO_PROT) |
531 (1ULL << INTERCEPT_MSR_PROT) |
532 (1ULL << INTERCEPT_TASK_SWITCH) |
533 (1ULL << INTERCEPT_SHUTDOWN) |
534 (1ULL << INTERCEPT_VMRUN) |
535 (1ULL << INTERCEPT_VMMCALL) |
536 (1ULL << INTERCEPT_VMLOAD) |
537 (1ULL << INTERCEPT_VMSAVE) |
538 (1ULL << INTERCEPT_STGI) |
539 (1ULL << INTERCEPT_CLGI) |
540 (1ULL << INTERCEPT_SKINIT) |
541 (1ULL << INTERCEPT_WBINVD) |
542 (1ULL << INTERCEPT_MONITOR) |
543 (1ULL << INTERCEPT_MWAIT);
545 control->iopm_base_pa = iopm_base;
546 control->msrpm_base_pa = __pa(svm->msrpm);
547 control->tsc_offset = 0;
548 control->int_ctl = V_INTR_MASKING_MASK;
556 save->cs.selector = 0xf000;
557 /* Executable/Readable Code Segment */
558 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
559 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
560 save->cs.limit = 0xffff;
562 * cs.base should really be 0xffff0000, but vmx can't handle that, so
563 * be consistent with it.
565 * Replace when we have real mode working for vmx.
567 save->cs.base = 0xf0000;
569 save->gdtr.limit = 0xffff;
570 save->idtr.limit = 0xffff;
572 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
573 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
575 save->efer = MSR_EFER_SVME_MASK;
576 save->dr6 = 0xffff0ff0;
579 save->rip = 0x0000fff0;
582 * cr0 val on cpu init should be 0x60000010, we enable cpu
583 * cache by default. the orderly way is to enable cache in bios.
585 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
586 save->cr4 = X86_CR4_PAE;
590 /* Setup VMCB for Nested Paging */
591 control->nested_ctl = 1;
592 control->intercept &= ~(1ULL << INTERCEPT_TASK_SWITCH);
593 control->intercept_exceptions &= ~(1 << PF_VECTOR);
594 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
596 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
598 save->g_pat = 0x0007040600070406ULL;
599 /* enable caching because the QEMU Bios doesn't enable it */
600 save->cr0 = X86_CR0_ET;
604 force_new_asid(&svm->vcpu);
607 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
609 struct vcpu_svm *svm = to_svm(vcpu);
613 if (vcpu->vcpu_id != 0) {
614 svm->vmcb->save.rip = 0;
615 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
616 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
622 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
624 struct vcpu_svm *svm;
626 struct page *msrpm_pages;
629 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
635 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
639 page = alloc_page(GFP_KERNEL);
646 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
649 svm->msrpm = page_address(msrpm_pages);
650 svm_vcpu_init_msrpm(svm->msrpm);
652 svm->vmcb = page_address(page);
653 clear_page(svm->vmcb);
654 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
655 svm->asid_generation = 0;
656 memset(svm->db_regs, 0, sizeof(svm->db_regs));
660 svm->vcpu.fpu_active = 1;
661 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
662 if (svm->vcpu.vcpu_id == 0)
663 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
668 kvm_vcpu_uninit(&svm->vcpu);
670 kmem_cache_free(kvm_vcpu_cache, svm);
675 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
677 struct vcpu_svm *svm = to_svm(vcpu);
679 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
680 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
681 kvm_vcpu_uninit(vcpu);
682 kmem_cache_free(kvm_vcpu_cache, svm);
685 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
687 struct vcpu_svm *svm = to_svm(vcpu);
690 if (unlikely(cpu != vcpu->cpu)) {
694 * Make sure that the guest sees a monotonically
698 delta = vcpu->arch.host_tsc - tsc_this;
699 svm->vmcb->control.tsc_offset += delta;
701 kvm_migrate_timers(vcpu);
704 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
705 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
708 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
710 struct vcpu_svm *svm = to_svm(vcpu);
713 ++vcpu->stat.host_state_reload;
714 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
715 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
717 rdtscll(vcpu->arch.host_tsc);
720 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
724 static void svm_cache_regs(struct kvm_vcpu *vcpu)
726 struct vcpu_svm *svm = to_svm(vcpu);
728 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
729 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
730 vcpu->arch.rip = svm->vmcb->save.rip;
733 static void svm_decache_regs(struct kvm_vcpu *vcpu)
735 struct vcpu_svm *svm = to_svm(vcpu);
736 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
737 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
738 svm->vmcb->save.rip = vcpu->arch.rip;
741 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
743 return to_svm(vcpu)->vmcb->save.rflags;
746 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
748 to_svm(vcpu)->vmcb->save.rflags = rflags;
751 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
753 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
756 case VCPU_SREG_CS: return &save->cs;
757 case VCPU_SREG_DS: return &save->ds;
758 case VCPU_SREG_ES: return &save->es;
759 case VCPU_SREG_FS: return &save->fs;
760 case VCPU_SREG_GS: return &save->gs;
761 case VCPU_SREG_SS: return &save->ss;
762 case VCPU_SREG_TR: return &save->tr;
763 case VCPU_SREG_LDTR: return &save->ldtr;
769 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
771 struct vmcb_seg *s = svm_seg(vcpu, seg);
776 static void svm_get_segment(struct kvm_vcpu *vcpu,
777 struct kvm_segment *var, int seg)
779 struct vmcb_seg *s = svm_seg(vcpu, seg);
782 var->limit = s->limit;
783 var->selector = s->selector;
784 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
785 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
786 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
787 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
788 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
789 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
790 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
791 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
792 var->unusable = !var->present;
795 static int svm_get_cpl(struct kvm_vcpu *vcpu)
797 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
802 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
804 struct vcpu_svm *svm = to_svm(vcpu);
806 dt->limit = svm->vmcb->save.idtr.limit;
807 dt->base = svm->vmcb->save.idtr.base;
810 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
812 struct vcpu_svm *svm = to_svm(vcpu);
814 svm->vmcb->save.idtr.limit = dt->limit;
815 svm->vmcb->save.idtr.base = dt->base ;
818 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
820 struct vcpu_svm *svm = to_svm(vcpu);
822 dt->limit = svm->vmcb->save.gdtr.limit;
823 dt->base = svm->vmcb->save.gdtr.base;
826 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
828 struct vcpu_svm *svm = to_svm(vcpu);
830 svm->vmcb->save.gdtr.limit = dt->limit;
831 svm->vmcb->save.gdtr.base = dt->base ;
834 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
838 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
840 struct vcpu_svm *svm = to_svm(vcpu);
843 if (vcpu->arch.shadow_efer & EFER_LME) {
844 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
845 vcpu->arch.shadow_efer |= EFER_LMA;
846 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
849 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
850 vcpu->arch.shadow_efer &= ~EFER_LMA;
851 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
858 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
859 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
860 vcpu->fpu_active = 1;
863 vcpu->arch.cr0 = cr0;
864 cr0 |= X86_CR0_PG | X86_CR0_WP;
865 if (!vcpu->fpu_active) {
866 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
871 * re-enable caching here because the QEMU bios
872 * does not do it - this results in some delay at
875 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
876 svm->vmcb->save.cr0 = cr0;
879 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
881 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
883 vcpu->arch.cr4 = cr4;
887 to_svm(vcpu)->vmcb->save.cr4 = cr4;
890 static void svm_set_segment(struct kvm_vcpu *vcpu,
891 struct kvm_segment *var, int seg)
893 struct vcpu_svm *svm = to_svm(vcpu);
894 struct vmcb_seg *s = svm_seg(vcpu, seg);
897 s->limit = var->limit;
898 s->selector = var->selector;
902 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
903 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
904 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
905 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
906 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
907 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
908 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
909 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
911 if (seg == VCPU_SREG_CS)
913 = (svm->vmcb->save.cs.attrib
914 >> SVM_SELECTOR_DPL_SHIFT) & 3;
918 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
923 static int svm_get_irq(struct kvm_vcpu *vcpu)
925 struct vcpu_svm *svm = to_svm(vcpu);
926 u32 exit_int_info = svm->vmcb->control.exit_int_info;
928 if (is_external_interrupt(exit_int_info))
929 return exit_int_info & SVM_EVTINJ_VEC_MASK;
933 static void load_host_msrs(struct kvm_vcpu *vcpu)
936 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
940 static void save_host_msrs(struct kvm_vcpu *vcpu)
943 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
947 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
949 if (svm_data->next_asid > svm_data->max_asid) {
950 ++svm_data->asid_generation;
951 svm_data->next_asid = 1;
952 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
955 svm->vcpu.cpu = svm_data->cpu;
956 svm->asid_generation = svm_data->asid_generation;
957 svm->vmcb->control.asid = svm_data->next_asid++;
960 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
962 return to_svm(vcpu)->db_regs[dr];
965 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
968 struct vcpu_svm *svm = to_svm(vcpu);
972 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
973 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
974 svm->vmcb->save.dr6 |= DR6_BD_MASK;
975 *exception = DB_VECTOR;
981 svm->db_regs[dr] = value;
984 if (vcpu->arch.cr4 & X86_CR4_DE) {
985 *exception = UD_VECTOR;
989 if (value & ~((1ULL << 32) - 1)) {
990 *exception = GP_VECTOR;
993 svm->vmcb->save.dr7 = value;
997 printk(KERN_DEBUG "%s: unexpected dr %u\n",
999 *exception = UD_VECTOR;
1004 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1006 u32 exit_int_info = svm->vmcb->control.exit_int_info;
1007 struct kvm *kvm = svm->vcpu.kvm;
1011 if (!irqchip_in_kernel(kvm) &&
1012 is_external_interrupt(exit_int_info))
1013 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1015 fault_address = svm->vmcb->control.exit_info_2;
1016 error_code = svm->vmcb->control.exit_info_1;
1017 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1020 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1024 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1025 if (er != EMULATE_DONE)
1026 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1030 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1032 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1033 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1034 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1035 svm->vcpu.fpu_active = 1;
1040 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1043 * On an #MC intercept the MCE handler is not called automatically in
1044 * the host. So do it by hand here.
1048 /* not sure if we ever come back to this point */
1053 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1056 * VMCB is undefined after a SHUTDOWN intercept
1057 * so reinitialize it.
1059 clear_page(svm->vmcb);
1062 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1066 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1068 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1069 int size, down, in, string, rep;
1072 ++svm->vcpu.stat.io_exits;
1074 svm->next_rip = svm->vmcb->control.exit_info_2;
1076 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1079 if (emulate_instruction(&svm->vcpu,
1080 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1085 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1086 port = io_info >> 16;
1087 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1088 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1089 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1091 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1094 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1099 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1101 svm->next_rip = svm->vmcb->save.rip + 1;
1102 skip_emulated_instruction(&svm->vcpu);
1103 return kvm_emulate_halt(&svm->vcpu);
1106 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1108 svm->next_rip = svm->vmcb->save.rip + 3;
1109 skip_emulated_instruction(&svm->vcpu);
1110 kvm_emulate_hypercall(&svm->vcpu);
1114 static int invalid_op_interception(struct vcpu_svm *svm,
1115 struct kvm_run *kvm_run)
1117 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1121 static int task_switch_interception(struct vcpu_svm *svm,
1122 struct kvm_run *kvm_run)
1126 tss_selector = (u16)svm->vmcb->control.exit_info_1;
1127 if (svm->vmcb->control.exit_info_2 &
1128 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1129 return kvm_task_switch(&svm->vcpu, tss_selector,
1131 if (svm->vmcb->control.exit_info_2 &
1132 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1133 return kvm_task_switch(&svm->vcpu, tss_selector,
1135 return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1138 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1140 svm->next_rip = svm->vmcb->save.rip + 2;
1141 kvm_emulate_cpuid(&svm->vcpu);
1145 static int emulate_on_interception(struct vcpu_svm *svm,
1146 struct kvm_run *kvm_run)
1148 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1149 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1153 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1155 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1156 if (irqchip_in_kernel(svm->vcpu.kvm))
1158 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1162 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1164 struct vcpu_svm *svm = to_svm(vcpu);
1167 case MSR_IA32_TIME_STAMP_COUNTER: {
1171 *data = svm->vmcb->control.tsc_offset + tsc;
1175 *data = svm->vmcb->save.star;
1177 #ifdef CONFIG_X86_64
1179 *data = svm->vmcb->save.lstar;
1182 *data = svm->vmcb->save.cstar;
1184 case MSR_KERNEL_GS_BASE:
1185 *data = svm->vmcb->save.kernel_gs_base;
1187 case MSR_SYSCALL_MASK:
1188 *data = svm->vmcb->save.sfmask;
1191 case MSR_IA32_SYSENTER_CS:
1192 *data = svm->vmcb->save.sysenter_cs;
1194 case MSR_IA32_SYSENTER_EIP:
1195 *data = svm->vmcb->save.sysenter_eip;
1197 case MSR_IA32_SYSENTER_ESP:
1198 *data = svm->vmcb->save.sysenter_esp;
1200 /* Nobody will change the following 5 values in the VMCB so
1201 we can safely return them on rdmsr. They will always be 0
1202 until LBRV is implemented. */
1203 case MSR_IA32_DEBUGCTLMSR:
1204 *data = svm->vmcb->save.dbgctl;
1206 case MSR_IA32_LASTBRANCHFROMIP:
1207 *data = svm->vmcb->save.br_from;
1209 case MSR_IA32_LASTBRANCHTOIP:
1210 *data = svm->vmcb->save.br_to;
1212 case MSR_IA32_LASTINTFROMIP:
1213 *data = svm->vmcb->save.last_excp_from;
1215 case MSR_IA32_LASTINTTOIP:
1216 *data = svm->vmcb->save.last_excp_to;
1219 return kvm_get_msr_common(vcpu, ecx, data);
1224 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1226 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1229 if (svm_get_msr(&svm->vcpu, ecx, &data))
1230 kvm_inject_gp(&svm->vcpu, 0);
1232 svm->vmcb->save.rax = data & 0xffffffff;
1233 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1234 svm->next_rip = svm->vmcb->save.rip + 2;
1235 skip_emulated_instruction(&svm->vcpu);
1240 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1242 struct vcpu_svm *svm = to_svm(vcpu);
1245 case MSR_IA32_TIME_STAMP_COUNTER: {
1249 svm->vmcb->control.tsc_offset = data - tsc;
1253 svm->vmcb->save.star = data;
1255 #ifdef CONFIG_X86_64
1257 svm->vmcb->save.lstar = data;
1260 svm->vmcb->save.cstar = data;
1262 case MSR_KERNEL_GS_BASE:
1263 svm->vmcb->save.kernel_gs_base = data;
1265 case MSR_SYSCALL_MASK:
1266 svm->vmcb->save.sfmask = data;
1269 case MSR_IA32_SYSENTER_CS:
1270 svm->vmcb->save.sysenter_cs = data;
1272 case MSR_IA32_SYSENTER_EIP:
1273 svm->vmcb->save.sysenter_eip = data;
1275 case MSR_IA32_SYSENTER_ESP:
1276 svm->vmcb->save.sysenter_esp = data;
1278 case MSR_IA32_DEBUGCTLMSR:
1279 if (!svm_has(SVM_FEATURE_LBRV)) {
1280 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
1284 if (data & DEBUGCTL_RESERVED_BITS)
1287 svm->vmcb->save.dbgctl = data;
1288 if (data & (1ULL<<0))
1289 svm_enable_lbrv(svm);
1291 svm_disable_lbrv(svm);
1293 case MSR_K7_EVNTSEL0:
1294 case MSR_K7_EVNTSEL1:
1295 case MSR_K7_EVNTSEL2:
1296 case MSR_K7_EVNTSEL3:
1298 * only support writing 0 to the performance counters for now
1299 * to make Windows happy. Should be replaced by a real
1300 * performance counter emulation later.
1307 return kvm_set_msr_common(vcpu, ecx, data);
1312 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1314 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1315 u64 data = (svm->vmcb->save.rax & -1u)
1316 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1317 svm->next_rip = svm->vmcb->save.rip + 2;
1318 if (svm_set_msr(&svm->vcpu, ecx, data))
1319 kvm_inject_gp(&svm->vcpu, 0);
1321 skip_emulated_instruction(&svm->vcpu);
1325 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1327 if (svm->vmcb->control.exit_info_1)
1328 return wrmsr_interception(svm, kvm_run);
1330 return rdmsr_interception(svm, kvm_run);
1333 static int interrupt_window_interception(struct vcpu_svm *svm,
1334 struct kvm_run *kvm_run)
1336 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1337 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1339 * If the user space waits to inject interrupts, exit as soon as
1342 if (kvm_run->request_interrupt_window &&
1343 !svm->vcpu.arch.irq_summary) {
1344 ++svm->vcpu.stat.irq_window_exits;
1345 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1352 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1353 struct kvm_run *kvm_run) = {
1354 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1355 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1356 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1357 [SVM_EXIT_READ_CR8] = emulate_on_interception,
1359 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1360 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1361 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1362 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
1363 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1364 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1365 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1366 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1367 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1368 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1369 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1370 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1371 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1372 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1373 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
1374 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1375 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1376 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
1377 [SVM_EXIT_INTR] = nop_on_interception,
1378 [SVM_EXIT_NMI] = nop_on_interception,
1379 [SVM_EXIT_SMI] = nop_on_interception,
1380 [SVM_EXIT_INIT] = nop_on_interception,
1381 [SVM_EXIT_VINTR] = interrupt_window_interception,
1382 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1383 [SVM_EXIT_CPUID] = cpuid_interception,
1384 [SVM_EXIT_INVD] = emulate_on_interception,
1385 [SVM_EXIT_HLT] = halt_interception,
1386 [SVM_EXIT_INVLPG] = emulate_on_interception,
1387 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1388 [SVM_EXIT_IOIO] = io_interception,
1389 [SVM_EXIT_MSR] = msr_interception,
1390 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1391 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1392 [SVM_EXIT_VMRUN] = invalid_op_interception,
1393 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1394 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1395 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1396 [SVM_EXIT_STGI] = invalid_op_interception,
1397 [SVM_EXIT_CLGI] = invalid_op_interception,
1398 [SVM_EXIT_SKINIT] = invalid_op_interception,
1399 [SVM_EXIT_WBINVD] = emulate_on_interception,
1400 [SVM_EXIT_MONITOR] = invalid_op_interception,
1401 [SVM_EXIT_MWAIT] = invalid_op_interception,
1402 [SVM_EXIT_NPF] = pf_interception,
1405 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1407 struct vcpu_svm *svm = to_svm(vcpu);
1408 u32 exit_code = svm->vmcb->control.exit_code;
1412 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
1413 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
1416 vcpu->arch.cr0 = svm->vmcb->save.cr0;
1417 vcpu->arch.cr3 = svm->vmcb->save.cr3;
1418 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1419 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1420 kvm_inject_gp(vcpu, 0);
1425 kvm_mmu_reset_context(vcpu);
1432 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1433 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1434 kvm_run->fail_entry.hardware_entry_failure_reason
1435 = svm->vmcb->control.exit_code;
1439 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1440 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1441 exit_code != SVM_EXIT_NPF)
1442 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1444 __func__, svm->vmcb->control.exit_int_info,
1447 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1448 || !svm_exit_handlers[exit_code]) {
1449 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1450 kvm_run->hw.hardware_exit_reason = exit_code;
1454 return svm_exit_handlers[exit_code](svm, kvm_run);
1457 static void reload_tss(struct kvm_vcpu *vcpu)
1459 int cpu = raw_smp_processor_id();
1461 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1462 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1466 static void pre_svm_run(struct vcpu_svm *svm)
1468 int cpu = raw_smp_processor_id();
1470 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1472 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1473 if (svm->vcpu.cpu != cpu ||
1474 svm->asid_generation != svm_data->asid_generation)
1475 new_asid(svm, svm_data);
1479 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1481 struct vmcb_control_area *control;
1483 control = &svm->vmcb->control;
1484 control->int_vector = irq;
1485 control->int_ctl &= ~V_INTR_PRIO_MASK;
1486 control->int_ctl |= V_IRQ_MASK |
1487 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1490 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1492 struct vcpu_svm *svm = to_svm(vcpu);
1494 svm_inject_irq(svm, irq);
1497 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
1499 struct vcpu_svm *svm = to_svm(vcpu);
1500 struct vmcb *vmcb = svm->vmcb;
1503 if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
1506 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1508 max_irr = kvm_lapic_find_highest_irr(vcpu);
1512 tpr = kvm_lapic_get_cr8(vcpu) << 4;
1514 if (tpr >= (max_irr & 0xf0))
1515 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
1518 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1520 struct vcpu_svm *svm = to_svm(vcpu);
1521 struct vmcb *vmcb = svm->vmcb;
1522 int intr_vector = -1;
1524 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1525 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1526 intr_vector = vmcb->control.exit_int_info &
1527 SVM_EVTINJ_VEC_MASK;
1528 vmcb->control.exit_int_info = 0;
1529 svm_inject_irq(svm, intr_vector);
1533 if (vmcb->control.int_ctl & V_IRQ_MASK)
1536 if (!kvm_cpu_has_interrupt(vcpu))
1539 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1540 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1541 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1542 /* unable to deliver irq, set pending irq */
1543 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1544 svm_inject_irq(svm, 0x0);
1547 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1548 intr_vector = kvm_cpu_get_interrupt(vcpu);
1549 svm_inject_irq(svm, intr_vector);
1550 kvm_timer_intr_post(vcpu, intr_vector);
1552 update_cr8_intercept(vcpu);
1555 static void kvm_reput_irq(struct vcpu_svm *svm)
1557 struct vmcb_control_area *control = &svm->vmcb->control;
1559 if ((control->int_ctl & V_IRQ_MASK)
1560 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1561 control->int_ctl &= ~V_IRQ_MASK;
1562 push_irq(&svm->vcpu, control->int_vector);
1565 svm->vcpu.arch.interrupt_window_open =
1566 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1569 static void svm_do_inject_vector(struct vcpu_svm *svm)
1571 struct kvm_vcpu *vcpu = &svm->vcpu;
1572 int word_index = __ffs(vcpu->arch.irq_summary);
1573 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1574 int irq = word_index * BITS_PER_LONG + bit_index;
1576 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1577 if (!vcpu->arch.irq_pending[word_index])
1578 clear_bit(word_index, &vcpu->arch.irq_summary);
1579 svm_inject_irq(svm, irq);
1582 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1583 struct kvm_run *kvm_run)
1585 struct vcpu_svm *svm = to_svm(vcpu);
1586 struct vmcb_control_area *control = &svm->vmcb->control;
1588 svm->vcpu.arch.interrupt_window_open =
1589 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1590 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1592 if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1594 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1596 svm_do_inject_vector(svm);
1599 * Interrupts blocked. Wait for unblock.
1601 if (!svm->vcpu.arch.interrupt_window_open &&
1602 (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1603 control->intercept |= 1ULL << INTERCEPT_VINTR;
1605 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1608 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1613 static void save_db_regs(unsigned long *db_regs)
1615 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1616 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1617 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1618 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1621 static void load_db_regs(unsigned long *db_regs)
1623 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1624 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1625 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1626 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1629 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1631 force_new_asid(vcpu);
1634 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1638 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
1640 struct vcpu_svm *svm = to_svm(vcpu);
1642 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
1643 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
1644 kvm_lapic_set_tpr(vcpu, cr8);
1648 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
1650 struct vcpu_svm *svm = to_svm(vcpu);
1653 if (!irqchip_in_kernel(vcpu->kvm))
1656 cr8 = kvm_get_cr8(vcpu);
1657 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
1658 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
1661 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1663 struct vcpu_svm *svm = to_svm(vcpu);
1670 sync_lapic_to_cr8(vcpu);
1672 save_host_msrs(vcpu);
1673 fs_selector = read_fs();
1674 gs_selector = read_gs();
1675 ldt_selector = read_ldt();
1676 svm->host_cr2 = kvm_read_cr2();
1677 svm->host_dr6 = read_dr6();
1678 svm->host_dr7 = read_dr7();
1679 svm->vmcb->save.cr2 = vcpu->arch.cr2;
1680 /* required for live migration with NPT */
1682 svm->vmcb->save.cr3 = vcpu->arch.cr3;
1684 if (svm->vmcb->save.dr7 & 0xff) {
1686 save_db_regs(svm->host_db_regs);
1687 load_db_regs(svm->db_regs);
1695 #ifdef CONFIG_X86_64
1701 #ifdef CONFIG_X86_64
1702 "mov %c[rbx](%[svm]), %%rbx \n\t"
1703 "mov %c[rcx](%[svm]), %%rcx \n\t"
1704 "mov %c[rdx](%[svm]), %%rdx \n\t"
1705 "mov %c[rsi](%[svm]), %%rsi \n\t"
1706 "mov %c[rdi](%[svm]), %%rdi \n\t"
1707 "mov %c[rbp](%[svm]), %%rbp \n\t"
1708 "mov %c[r8](%[svm]), %%r8 \n\t"
1709 "mov %c[r9](%[svm]), %%r9 \n\t"
1710 "mov %c[r10](%[svm]), %%r10 \n\t"
1711 "mov %c[r11](%[svm]), %%r11 \n\t"
1712 "mov %c[r12](%[svm]), %%r12 \n\t"
1713 "mov %c[r13](%[svm]), %%r13 \n\t"
1714 "mov %c[r14](%[svm]), %%r14 \n\t"
1715 "mov %c[r15](%[svm]), %%r15 \n\t"
1717 "mov %c[rbx](%[svm]), %%ebx \n\t"
1718 "mov %c[rcx](%[svm]), %%ecx \n\t"
1719 "mov %c[rdx](%[svm]), %%edx \n\t"
1720 "mov %c[rsi](%[svm]), %%esi \n\t"
1721 "mov %c[rdi](%[svm]), %%edi \n\t"
1722 "mov %c[rbp](%[svm]), %%ebp \n\t"
1725 #ifdef CONFIG_X86_64
1726 /* Enter guest mode */
1728 "mov %c[vmcb](%[svm]), %%rax \n\t"
1734 /* Enter guest mode */
1736 "mov %c[vmcb](%[svm]), %%eax \n\t"
1743 /* Save guest registers, load host registers */
1744 #ifdef CONFIG_X86_64
1745 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1746 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1747 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1748 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1749 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1750 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1751 "mov %%r8, %c[r8](%[svm]) \n\t"
1752 "mov %%r9, %c[r9](%[svm]) \n\t"
1753 "mov %%r10, %c[r10](%[svm]) \n\t"
1754 "mov %%r11, %c[r11](%[svm]) \n\t"
1755 "mov %%r12, %c[r12](%[svm]) \n\t"
1756 "mov %%r13, %c[r13](%[svm]) \n\t"
1757 "mov %%r14, %c[r14](%[svm]) \n\t"
1758 "mov %%r15, %c[r15](%[svm]) \n\t"
1762 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1763 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1764 "mov %%edx, %c[rdx](%[svm]) \n\t"
1765 "mov %%esi, %c[rsi](%[svm]) \n\t"
1766 "mov %%edi, %c[rdi](%[svm]) \n\t"
1767 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1773 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1774 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1775 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1776 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1777 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1778 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1779 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1780 #ifdef CONFIG_X86_64
1781 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1782 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1783 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1784 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1785 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1786 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1787 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1788 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1791 #ifdef CONFIG_X86_64
1792 , "rbx", "rcx", "rdx", "rsi", "rdi"
1793 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1795 , "ebx", "ecx", "edx" , "esi", "edi"
1799 if ((svm->vmcb->save.dr7 & 0xff))
1800 load_db_regs(svm->host_db_regs);
1802 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1804 write_dr6(svm->host_dr6);
1805 write_dr7(svm->host_dr7);
1806 kvm_write_cr2(svm->host_cr2);
1808 load_fs(fs_selector);
1809 load_gs(gs_selector);
1810 load_ldt(ldt_selector);
1811 load_host_msrs(vcpu);
1815 local_irq_disable();
1819 sync_cr8_to_lapic(vcpu);
1824 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1826 struct vcpu_svm *svm = to_svm(vcpu);
1829 svm->vmcb->control.nested_cr3 = root;
1830 force_new_asid(vcpu);
1834 svm->vmcb->save.cr3 = root;
1835 force_new_asid(vcpu);
1837 if (vcpu->fpu_active) {
1838 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1839 svm->vmcb->save.cr0 |= X86_CR0_TS;
1840 vcpu->fpu_active = 0;
1844 static int is_disabled(void)
1848 rdmsrl(MSR_VM_CR, vm_cr);
1849 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1856 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1859 * Patch in the VMMCALL instruction:
1861 hypercall[0] = 0x0f;
1862 hypercall[1] = 0x01;
1863 hypercall[2] = 0xd9;
1866 static void svm_check_processor_compat(void *rtn)
1871 static bool svm_cpu_has_accelerated_tpr(void)
1876 static int get_npt_level(void)
1878 #ifdef CONFIG_X86_64
1879 return PT64_ROOT_LEVEL;
1881 return PT32E_ROOT_LEVEL;
1885 static struct kvm_x86_ops svm_x86_ops = {
1886 .cpu_has_kvm_support = has_svm,
1887 .disabled_by_bios = is_disabled,
1888 .hardware_setup = svm_hardware_setup,
1889 .hardware_unsetup = svm_hardware_unsetup,
1890 .check_processor_compatibility = svm_check_processor_compat,
1891 .hardware_enable = svm_hardware_enable,
1892 .hardware_disable = svm_hardware_disable,
1893 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1895 .vcpu_create = svm_create_vcpu,
1896 .vcpu_free = svm_free_vcpu,
1897 .vcpu_reset = svm_vcpu_reset,
1899 .prepare_guest_switch = svm_prepare_guest_switch,
1900 .vcpu_load = svm_vcpu_load,
1901 .vcpu_put = svm_vcpu_put,
1902 .vcpu_decache = svm_vcpu_decache,
1904 .set_guest_debug = svm_guest_debug,
1905 .get_msr = svm_get_msr,
1906 .set_msr = svm_set_msr,
1907 .get_segment_base = svm_get_segment_base,
1908 .get_segment = svm_get_segment,
1909 .set_segment = svm_set_segment,
1910 .get_cpl = svm_get_cpl,
1911 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1912 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1913 .set_cr0 = svm_set_cr0,
1914 .set_cr3 = svm_set_cr3,
1915 .set_cr4 = svm_set_cr4,
1916 .set_efer = svm_set_efer,
1917 .get_idt = svm_get_idt,
1918 .set_idt = svm_set_idt,
1919 .get_gdt = svm_get_gdt,
1920 .set_gdt = svm_set_gdt,
1921 .get_dr = svm_get_dr,
1922 .set_dr = svm_set_dr,
1923 .cache_regs = svm_cache_regs,
1924 .decache_regs = svm_decache_regs,
1925 .get_rflags = svm_get_rflags,
1926 .set_rflags = svm_set_rflags,
1928 .tlb_flush = svm_flush_tlb,
1930 .run = svm_vcpu_run,
1931 .handle_exit = handle_exit,
1932 .skip_emulated_instruction = skip_emulated_instruction,
1933 .patch_hypercall = svm_patch_hypercall,
1934 .get_irq = svm_get_irq,
1935 .set_irq = svm_set_irq,
1936 .queue_exception = svm_queue_exception,
1937 .exception_injected = svm_exception_injected,
1938 .inject_pending_irq = svm_intr_assist,
1939 .inject_pending_vectors = do_interrupt_requests,
1941 .set_tss_addr = svm_set_tss_addr,
1942 .get_tdp_level = get_npt_level,
1945 static int __init svm_init(void)
1947 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1951 static void __exit svm_exit(void)
1956 module_init(svm_init)
1957 module_exit(svm_exit)