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)
273 struct svm_cpu_data *svm_data
274 = per_cpu(svm_data, raw_smp_processor_id());
279 wrmsrl(MSR_VM_HSAVE_PA, 0);
280 rdmsrl(MSR_EFER, efer);
281 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
282 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
283 __free_page(svm_data->save_area);
288 static void svm_hardware_enable(void *garbage)
291 struct svm_cpu_data *svm_data;
293 struct desc_ptr gdt_descr;
294 struct desc_struct *gdt;
295 int me = raw_smp_processor_id();
298 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
301 svm_data = per_cpu(svm_data, me);
304 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
309 svm_data->asid_generation = 1;
310 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
311 svm_data->next_asid = svm_data->max_asid + 1;
313 asm volatile ("sgdt %0" : "=m"(gdt_descr));
314 gdt = (struct desc_struct *)gdt_descr.address;
315 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
317 rdmsrl(MSR_EFER, efer);
318 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
320 wrmsrl(MSR_VM_HSAVE_PA,
321 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
324 static int svm_cpu_init(int cpu)
326 struct svm_cpu_data *svm_data;
329 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
333 svm_data->save_area = alloc_page(GFP_KERNEL);
335 if (!svm_data->save_area)
338 per_cpu(svm_data, cpu) = svm_data;
348 static void set_msr_interception(u32 *msrpm, unsigned msr,
353 for (i = 0; i < NUM_MSR_MAPS; i++) {
354 if (msr >= msrpm_ranges[i] &&
355 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
356 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
357 msrpm_ranges[i]) * 2;
359 u32 *base = msrpm + (msr_offset / 32);
360 u32 msr_shift = msr_offset % 32;
361 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
362 *base = (*base & ~(0x3 << msr_shift)) |
370 static void svm_vcpu_init_msrpm(u32 *msrpm)
372 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
375 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
376 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
377 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
378 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
379 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
380 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
382 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
383 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
384 set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
385 set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
388 static void svm_enable_lbrv(struct vcpu_svm *svm)
390 u32 *msrpm = svm->msrpm;
392 svm->vmcb->control.lbr_ctl = 1;
393 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
394 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
395 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
396 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
399 static void svm_disable_lbrv(struct vcpu_svm *svm)
401 u32 *msrpm = svm->msrpm;
403 svm->vmcb->control.lbr_ctl = 0;
404 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
405 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
406 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
407 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
410 static __init int svm_hardware_setup(void)
413 struct page *iopm_pages;
417 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
422 iopm_va = page_address(iopm_pages);
423 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
424 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
425 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
427 if (boot_cpu_has(X86_FEATURE_NX))
428 kvm_enable_efer_bits(EFER_NX);
430 for_each_online_cpu(cpu) {
431 r = svm_cpu_init(cpu);
436 svm_features = cpuid_edx(SVM_CPUID_FUNC);
438 if (!svm_has(SVM_FEATURE_NPT))
441 if (npt_enabled && !npt) {
442 printk(KERN_INFO "kvm: Nested Paging disabled\n");
447 printk(KERN_INFO "kvm: Nested Paging enabled\n");
454 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
459 static __exit void svm_hardware_unsetup(void)
461 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
465 static void init_seg(struct vmcb_seg *seg)
468 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
469 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
474 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
477 seg->attrib = SVM_SELECTOR_P_MASK | type;
482 static void init_vmcb(struct vcpu_svm *svm)
484 struct vmcb_control_area *control = &svm->vmcb->control;
485 struct vmcb_save_area *save = &svm->vmcb->save;
487 control->intercept_cr_read = INTERCEPT_CR0_MASK |
491 control->intercept_cr_write = INTERCEPT_CR0_MASK |
496 control->intercept_dr_read = INTERCEPT_DR0_MASK |
501 control->intercept_dr_write = INTERCEPT_DR0_MASK |
508 control->intercept_exceptions = (1 << PF_VECTOR) |
513 control->intercept = (1ULL << INTERCEPT_INTR) |
514 (1ULL << INTERCEPT_NMI) |
515 (1ULL << INTERCEPT_SMI) |
516 (1ULL << INTERCEPT_CPUID) |
517 (1ULL << INTERCEPT_INVD) |
518 (1ULL << INTERCEPT_HLT) |
519 (1ULL << INTERCEPT_INVLPGA) |
520 (1ULL << INTERCEPT_IOIO_PROT) |
521 (1ULL << INTERCEPT_MSR_PROT) |
522 (1ULL << INTERCEPT_TASK_SWITCH) |
523 (1ULL << INTERCEPT_SHUTDOWN) |
524 (1ULL << INTERCEPT_VMRUN) |
525 (1ULL << INTERCEPT_VMMCALL) |
526 (1ULL << INTERCEPT_VMLOAD) |
527 (1ULL << INTERCEPT_VMSAVE) |
528 (1ULL << INTERCEPT_STGI) |
529 (1ULL << INTERCEPT_CLGI) |
530 (1ULL << INTERCEPT_SKINIT) |
531 (1ULL << INTERCEPT_WBINVD) |
532 (1ULL << INTERCEPT_MONITOR) |
533 (1ULL << INTERCEPT_MWAIT);
535 control->iopm_base_pa = iopm_base;
536 control->msrpm_base_pa = __pa(svm->msrpm);
537 control->tsc_offset = 0;
538 control->int_ctl = V_INTR_MASKING_MASK;
546 save->cs.selector = 0xf000;
547 /* Executable/Readable Code Segment */
548 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
549 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
550 save->cs.limit = 0xffff;
552 * cs.base should really be 0xffff0000, but vmx can't handle that, so
553 * be consistent with it.
555 * Replace when we have real mode working for vmx.
557 save->cs.base = 0xf0000;
559 save->gdtr.limit = 0xffff;
560 save->idtr.limit = 0xffff;
562 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
563 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
565 save->efer = MSR_EFER_SVME_MASK;
566 save->dr6 = 0xffff0ff0;
569 save->rip = 0x0000fff0;
572 * cr0 val on cpu init should be 0x60000010, we enable cpu
573 * cache by default. the orderly way is to enable cache in bios.
575 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
576 save->cr4 = X86_CR4_PAE;
580 /* Setup VMCB for Nested Paging */
581 control->nested_ctl = 1;
582 control->intercept &= ~(1ULL << INTERCEPT_TASK_SWITCH);
583 control->intercept_exceptions &= ~(1 << PF_VECTOR);
584 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
586 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
588 save->g_pat = 0x0007040600070406ULL;
589 /* enable caching because the QEMU Bios doesn't enable it */
590 save->cr0 = X86_CR0_ET;
594 force_new_asid(&svm->vcpu);
597 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
599 struct vcpu_svm *svm = to_svm(vcpu);
603 if (vcpu->vcpu_id != 0) {
604 svm->vmcb->save.rip = 0;
605 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
606 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
612 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
614 struct vcpu_svm *svm;
616 struct page *msrpm_pages;
619 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
625 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
629 page = alloc_page(GFP_KERNEL);
636 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
639 svm->msrpm = page_address(msrpm_pages);
640 svm_vcpu_init_msrpm(svm->msrpm);
642 svm->vmcb = page_address(page);
643 clear_page(svm->vmcb);
644 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
645 svm->asid_generation = 0;
646 memset(svm->db_regs, 0, sizeof(svm->db_regs));
650 svm->vcpu.fpu_active = 1;
651 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
652 if (svm->vcpu.vcpu_id == 0)
653 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
658 kvm_vcpu_uninit(&svm->vcpu);
660 kmem_cache_free(kvm_vcpu_cache, svm);
665 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
667 struct vcpu_svm *svm = to_svm(vcpu);
669 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
670 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
671 kvm_vcpu_uninit(vcpu);
672 kmem_cache_free(kvm_vcpu_cache, svm);
675 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
677 struct vcpu_svm *svm = to_svm(vcpu);
680 if (unlikely(cpu != vcpu->cpu)) {
684 * Make sure that the guest sees a monotonically
688 delta = vcpu->arch.host_tsc - tsc_this;
689 svm->vmcb->control.tsc_offset += delta;
691 kvm_migrate_timers(vcpu);
694 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
695 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
698 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
700 struct vcpu_svm *svm = to_svm(vcpu);
703 ++vcpu->stat.host_state_reload;
704 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
705 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
707 rdtscll(vcpu->arch.host_tsc);
710 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
714 static void svm_cache_regs(struct kvm_vcpu *vcpu)
716 struct vcpu_svm *svm = to_svm(vcpu);
718 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
719 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
720 vcpu->arch.rip = svm->vmcb->save.rip;
723 static void svm_decache_regs(struct kvm_vcpu *vcpu)
725 struct vcpu_svm *svm = to_svm(vcpu);
726 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
727 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
728 svm->vmcb->save.rip = vcpu->arch.rip;
731 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
733 return to_svm(vcpu)->vmcb->save.rflags;
736 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
738 to_svm(vcpu)->vmcb->save.rflags = rflags;
741 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
743 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
746 case VCPU_SREG_CS: return &save->cs;
747 case VCPU_SREG_DS: return &save->ds;
748 case VCPU_SREG_ES: return &save->es;
749 case VCPU_SREG_FS: return &save->fs;
750 case VCPU_SREG_GS: return &save->gs;
751 case VCPU_SREG_SS: return &save->ss;
752 case VCPU_SREG_TR: return &save->tr;
753 case VCPU_SREG_LDTR: return &save->ldtr;
759 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
761 struct vmcb_seg *s = svm_seg(vcpu, seg);
766 static void svm_get_segment(struct kvm_vcpu *vcpu,
767 struct kvm_segment *var, int seg)
769 struct vmcb_seg *s = svm_seg(vcpu, seg);
772 var->limit = s->limit;
773 var->selector = s->selector;
774 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
775 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
776 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
777 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
778 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
779 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
780 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
781 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
782 var->unusable = !var->present;
785 static int svm_get_cpl(struct kvm_vcpu *vcpu)
787 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
792 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
794 struct vcpu_svm *svm = to_svm(vcpu);
796 dt->limit = svm->vmcb->save.idtr.limit;
797 dt->base = svm->vmcb->save.idtr.base;
800 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
802 struct vcpu_svm *svm = to_svm(vcpu);
804 svm->vmcb->save.idtr.limit = dt->limit;
805 svm->vmcb->save.idtr.base = dt->base ;
808 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
810 struct vcpu_svm *svm = to_svm(vcpu);
812 dt->limit = svm->vmcb->save.gdtr.limit;
813 dt->base = svm->vmcb->save.gdtr.base;
816 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
818 struct vcpu_svm *svm = to_svm(vcpu);
820 svm->vmcb->save.gdtr.limit = dt->limit;
821 svm->vmcb->save.gdtr.base = dt->base ;
824 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
828 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
830 struct vcpu_svm *svm = to_svm(vcpu);
833 if (vcpu->arch.shadow_efer & EFER_LME) {
834 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
835 vcpu->arch.shadow_efer |= EFER_LMA;
836 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
839 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
840 vcpu->arch.shadow_efer &= ~EFER_LMA;
841 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
848 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
849 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
850 vcpu->fpu_active = 1;
853 vcpu->arch.cr0 = cr0;
854 cr0 |= X86_CR0_PG | X86_CR0_WP;
855 if (!vcpu->fpu_active) {
856 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
861 * re-enable caching here because the QEMU bios
862 * does not do it - this results in some delay at
865 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
866 svm->vmcb->save.cr0 = cr0;
869 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
871 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
873 vcpu->arch.cr4 = cr4;
877 to_svm(vcpu)->vmcb->save.cr4 = cr4;
880 static void svm_set_segment(struct kvm_vcpu *vcpu,
881 struct kvm_segment *var, int seg)
883 struct vcpu_svm *svm = to_svm(vcpu);
884 struct vmcb_seg *s = svm_seg(vcpu, seg);
887 s->limit = var->limit;
888 s->selector = var->selector;
892 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
893 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
894 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
895 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
896 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
897 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
898 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
899 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
901 if (seg == VCPU_SREG_CS)
903 = (svm->vmcb->save.cs.attrib
904 >> SVM_SELECTOR_DPL_SHIFT) & 3;
908 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
913 static int svm_get_irq(struct kvm_vcpu *vcpu)
915 struct vcpu_svm *svm = to_svm(vcpu);
916 u32 exit_int_info = svm->vmcb->control.exit_int_info;
918 if (is_external_interrupt(exit_int_info))
919 return exit_int_info & SVM_EVTINJ_VEC_MASK;
923 static void load_host_msrs(struct kvm_vcpu *vcpu)
926 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
930 static void save_host_msrs(struct kvm_vcpu *vcpu)
933 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
937 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
939 if (svm_data->next_asid > svm_data->max_asid) {
940 ++svm_data->asid_generation;
941 svm_data->next_asid = 1;
942 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
945 svm->vcpu.cpu = svm_data->cpu;
946 svm->asid_generation = svm_data->asid_generation;
947 svm->vmcb->control.asid = svm_data->next_asid++;
950 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
952 unsigned long val = to_svm(vcpu)->db_regs[dr];
953 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
957 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
960 struct vcpu_svm *svm = to_svm(vcpu);
964 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
965 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
966 svm->vmcb->save.dr6 |= DR6_BD_MASK;
967 *exception = DB_VECTOR;
973 svm->db_regs[dr] = value;
976 if (vcpu->arch.cr4 & X86_CR4_DE) {
977 *exception = UD_VECTOR;
981 if (value & ~((1ULL << 32) - 1)) {
982 *exception = GP_VECTOR;
985 svm->vmcb->save.dr7 = value;
989 printk(KERN_DEBUG "%s: unexpected dr %u\n",
991 *exception = UD_VECTOR;
996 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
998 u32 exit_int_info = svm->vmcb->control.exit_int_info;
999 struct kvm *kvm = svm->vcpu.kvm;
1003 if (!irqchip_in_kernel(kvm) &&
1004 is_external_interrupt(exit_int_info))
1005 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1007 fault_address = svm->vmcb->control.exit_info_2;
1008 error_code = svm->vmcb->control.exit_info_1;
1011 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1012 (u32)fault_address, (u32)(fault_address >> 32),
1015 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1018 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1022 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1023 if (er != EMULATE_DONE)
1024 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1028 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1030 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1031 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1032 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1033 svm->vcpu.fpu_active = 1;
1038 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1041 * On an #MC intercept the MCE handler is not called automatically in
1042 * the host. So do it by hand here.
1046 /* not sure if we ever come back to this point */
1051 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1054 * VMCB is undefined after a SHUTDOWN intercept
1055 * so reinitialize it.
1057 clear_page(svm->vmcb);
1060 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1064 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1066 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1067 int size, down, in, string, rep;
1070 ++svm->vcpu.stat.io_exits;
1072 svm->next_rip = svm->vmcb->control.exit_info_2;
1074 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1077 if (emulate_instruction(&svm->vcpu,
1078 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1083 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1084 port = io_info >> 16;
1085 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1086 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1087 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1089 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1092 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1094 KVMTRACE_0D(NMI, &svm->vcpu, handler);
1098 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1100 ++svm->vcpu.stat.irq_exits;
1101 KVMTRACE_0D(INTR, &svm->vcpu, handler);
1105 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1110 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1112 svm->next_rip = svm->vmcb->save.rip + 1;
1113 skip_emulated_instruction(&svm->vcpu);
1114 return kvm_emulate_halt(&svm->vcpu);
1117 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1119 svm->next_rip = svm->vmcb->save.rip + 3;
1120 skip_emulated_instruction(&svm->vcpu);
1121 kvm_emulate_hypercall(&svm->vcpu);
1125 static int invalid_op_interception(struct vcpu_svm *svm,
1126 struct kvm_run *kvm_run)
1128 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1132 static int task_switch_interception(struct vcpu_svm *svm,
1133 struct kvm_run *kvm_run)
1137 tss_selector = (u16)svm->vmcb->control.exit_info_1;
1138 if (svm->vmcb->control.exit_info_2 &
1139 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1140 return kvm_task_switch(&svm->vcpu, tss_selector,
1142 if (svm->vmcb->control.exit_info_2 &
1143 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1144 return kvm_task_switch(&svm->vcpu, tss_selector,
1146 return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1149 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1151 svm->next_rip = svm->vmcb->save.rip + 2;
1152 kvm_emulate_cpuid(&svm->vcpu);
1156 static int emulate_on_interception(struct vcpu_svm *svm,
1157 struct kvm_run *kvm_run)
1159 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1160 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1164 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1166 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1167 if (irqchip_in_kernel(svm->vcpu.kvm))
1169 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1173 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1175 struct vcpu_svm *svm = to_svm(vcpu);
1178 case MSR_IA32_TIME_STAMP_COUNTER: {
1182 *data = svm->vmcb->control.tsc_offset + tsc;
1186 *data = svm->vmcb->save.star;
1188 #ifdef CONFIG_X86_64
1190 *data = svm->vmcb->save.lstar;
1193 *data = svm->vmcb->save.cstar;
1195 case MSR_KERNEL_GS_BASE:
1196 *data = svm->vmcb->save.kernel_gs_base;
1198 case MSR_SYSCALL_MASK:
1199 *data = svm->vmcb->save.sfmask;
1202 case MSR_IA32_SYSENTER_CS:
1203 *data = svm->vmcb->save.sysenter_cs;
1205 case MSR_IA32_SYSENTER_EIP:
1206 *data = svm->vmcb->save.sysenter_eip;
1208 case MSR_IA32_SYSENTER_ESP:
1209 *data = svm->vmcb->save.sysenter_esp;
1211 /* Nobody will change the following 5 values in the VMCB so
1212 we can safely return them on rdmsr. They will always be 0
1213 until LBRV is implemented. */
1214 case MSR_IA32_DEBUGCTLMSR:
1215 *data = svm->vmcb->save.dbgctl;
1217 case MSR_IA32_LASTBRANCHFROMIP:
1218 *data = svm->vmcb->save.br_from;
1220 case MSR_IA32_LASTBRANCHTOIP:
1221 *data = svm->vmcb->save.br_to;
1223 case MSR_IA32_LASTINTFROMIP:
1224 *data = svm->vmcb->save.last_excp_from;
1226 case MSR_IA32_LASTINTTOIP:
1227 *data = svm->vmcb->save.last_excp_to;
1230 return kvm_get_msr_common(vcpu, ecx, data);
1235 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1237 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1240 if (svm_get_msr(&svm->vcpu, ecx, &data))
1241 kvm_inject_gp(&svm->vcpu, 0);
1243 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1244 (u32)(data >> 32), handler);
1246 svm->vmcb->save.rax = data & 0xffffffff;
1247 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1248 svm->next_rip = svm->vmcb->save.rip + 2;
1249 skip_emulated_instruction(&svm->vcpu);
1254 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1256 struct vcpu_svm *svm = to_svm(vcpu);
1259 case MSR_IA32_TIME_STAMP_COUNTER: {
1263 svm->vmcb->control.tsc_offset = data - tsc;
1267 svm->vmcb->save.star = data;
1269 #ifdef CONFIG_X86_64
1271 svm->vmcb->save.lstar = data;
1274 svm->vmcb->save.cstar = data;
1276 case MSR_KERNEL_GS_BASE:
1277 svm->vmcb->save.kernel_gs_base = data;
1279 case MSR_SYSCALL_MASK:
1280 svm->vmcb->save.sfmask = data;
1283 case MSR_IA32_SYSENTER_CS:
1284 svm->vmcb->save.sysenter_cs = data;
1286 case MSR_IA32_SYSENTER_EIP:
1287 svm->vmcb->save.sysenter_eip = data;
1289 case MSR_IA32_SYSENTER_ESP:
1290 svm->vmcb->save.sysenter_esp = data;
1292 case MSR_IA32_DEBUGCTLMSR:
1293 if (!svm_has(SVM_FEATURE_LBRV)) {
1294 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
1298 if (data & DEBUGCTL_RESERVED_BITS)
1301 svm->vmcb->save.dbgctl = data;
1302 if (data & (1ULL<<0))
1303 svm_enable_lbrv(svm);
1305 svm_disable_lbrv(svm);
1307 case MSR_K7_EVNTSEL0:
1308 case MSR_K7_EVNTSEL1:
1309 case MSR_K7_EVNTSEL2:
1310 case MSR_K7_EVNTSEL3:
1312 * only support writing 0 to the performance counters for now
1313 * to make Windows happy. Should be replaced by a real
1314 * performance counter emulation later.
1321 return kvm_set_msr_common(vcpu, ecx, data);
1326 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1328 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1329 u64 data = (svm->vmcb->save.rax & -1u)
1330 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1332 KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
1335 svm->next_rip = svm->vmcb->save.rip + 2;
1336 if (svm_set_msr(&svm->vcpu, ecx, data))
1337 kvm_inject_gp(&svm->vcpu, 0);
1339 skip_emulated_instruction(&svm->vcpu);
1343 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1345 if (svm->vmcb->control.exit_info_1)
1346 return wrmsr_interception(svm, kvm_run);
1348 return rdmsr_interception(svm, kvm_run);
1351 static int interrupt_window_interception(struct vcpu_svm *svm,
1352 struct kvm_run *kvm_run)
1354 KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
1356 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1357 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1359 * If the user space waits to inject interrupts, exit as soon as
1362 if (kvm_run->request_interrupt_window &&
1363 !svm->vcpu.arch.irq_summary) {
1364 ++svm->vcpu.stat.irq_window_exits;
1365 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1372 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1373 struct kvm_run *kvm_run) = {
1374 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1375 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1376 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1377 [SVM_EXIT_READ_CR8] = emulate_on_interception,
1379 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1380 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1381 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1382 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
1383 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1384 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1385 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1386 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1387 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1388 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1389 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1390 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1391 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1392 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1393 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
1394 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1395 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1396 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
1397 [SVM_EXIT_INTR] = intr_interception,
1398 [SVM_EXIT_NMI] = nmi_interception,
1399 [SVM_EXIT_SMI] = nop_on_interception,
1400 [SVM_EXIT_INIT] = nop_on_interception,
1401 [SVM_EXIT_VINTR] = interrupt_window_interception,
1402 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1403 [SVM_EXIT_CPUID] = cpuid_interception,
1404 [SVM_EXIT_INVD] = emulate_on_interception,
1405 [SVM_EXIT_HLT] = halt_interception,
1406 [SVM_EXIT_INVLPG] = emulate_on_interception,
1407 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1408 [SVM_EXIT_IOIO] = io_interception,
1409 [SVM_EXIT_MSR] = msr_interception,
1410 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1411 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1412 [SVM_EXIT_VMRUN] = invalid_op_interception,
1413 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1414 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1415 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1416 [SVM_EXIT_STGI] = invalid_op_interception,
1417 [SVM_EXIT_CLGI] = invalid_op_interception,
1418 [SVM_EXIT_SKINIT] = invalid_op_interception,
1419 [SVM_EXIT_WBINVD] = emulate_on_interception,
1420 [SVM_EXIT_MONITOR] = invalid_op_interception,
1421 [SVM_EXIT_MWAIT] = invalid_op_interception,
1422 [SVM_EXIT_NPF] = pf_interception,
1425 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1427 struct vcpu_svm *svm = to_svm(vcpu);
1428 u32 exit_code = svm->vmcb->control.exit_code;
1430 KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
1431 (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
1435 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
1436 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
1439 vcpu->arch.cr0 = svm->vmcb->save.cr0;
1440 vcpu->arch.cr3 = svm->vmcb->save.cr3;
1441 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1442 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1443 kvm_inject_gp(vcpu, 0);
1448 kvm_mmu_reset_context(vcpu);
1455 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1456 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1457 kvm_run->fail_entry.hardware_entry_failure_reason
1458 = svm->vmcb->control.exit_code;
1462 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1463 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1464 exit_code != SVM_EXIT_NPF)
1465 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1467 __func__, svm->vmcb->control.exit_int_info,
1470 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1471 || !svm_exit_handlers[exit_code]) {
1472 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1473 kvm_run->hw.hardware_exit_reason = exit_code;
1477 return svm_exit_handlers[exit_code](svm, kvm_run);
1480 static void reload_tss(struct kvm_vcpu *vcpu)
1482 int cpu = raw_smp_processor_id();
1484 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1485 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1489 static void pre_svm_run(struct vcpu_svm *svm)
1491 int cpu = raw_smp_processor_id();
1493 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1495 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1496 if (svm->vcpu.cpu != cpu ||
1497 svm->asid_generation != svm_data->asid_generation)
1498 new_asid(svm, svm_data);
1502 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1504 struct vmcb_control_area *control;
1506 KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
1508 control = &svm->vmcb->control;
1509 control->int_vector = irq;
1510 control->int_ctl &= ~V_INTR_PRIO_MASK;
1511 control->int_ctl |= V_IRQ_MASK |
1512 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1515 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1517 struct vcpu_svm *svm = to_svm(vcpu);
1519 svm_inject_irq(svm, irq);
1522 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
1524 struct vcpu_svm *svm = to_svm(vcpu);
1525 struct vmcb *vmcb = svm->vmcb;
1528 if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
1531 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1533 max_irr = kvm_lapic_find_highest_irr(vcpu);
1537 tpr = kvm_lapic_get_cr8(vcpu) << 4;
1539 if (tpr >= (max_irr & 0xf0))
1540 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
1543 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1545 struct vcpu_svm *svm = to_svm(vcpu);
1546 struct vmcb *vmcb = svm->vmcb;
1547 int intr_vector = -1;
1549 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1550 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1551 intr_vector = vmcb->control.exit_int_info &
1552 SVM_EVTINJ_VEC_MASK;
1553 vmcb->control.exit_int_info = 0;
1554 svm_inject_irq(svm, intr_vector);
1558 if (vmcb->control.int_ctl & V_IRQ_MASK)
1561 if (!kvm_cpu_has_interrupt(vcpu))
1564 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1565 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1566 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1567 /* unable to deliver irq, set pending irq */
1568 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1569 svm_inject_irq(svm, 0x0);
1572 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1573 intr_vector = kvm_cpu_get_interrupt(vcpu);
1574 svm_inject_irq(svm, intr_vector);
1575 kvm_timer_intr_post(vcpu, intr_vector);
1577 update_cr8_intercept(vcpu);
1580 static void kvm_reput_irq(struct vcpu_svm *svm)
1582 struct vmcb_control_area *control = &svm->vmcb->control;
1584 if ((control->int_ctl & V_IRQ_MASK)
1585 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1586 control->int_ctl &= ~V_IRQ_MASK;
1587 push_irq(&svm->vcpu, control->int_vector);
1590 svm->vcpu.arch.interrupt_window_open =
1591 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1594 static void svm_do_inject_vector(struct vcpu_svm *svm)
1596 struct kvm_vcpu *vcpu = &svm->vcpu;
1597 int word_index = __ffs(vcpu->arch.irq_summary);
1598 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1599 int irq = word_index * BITS_PER_LONG + bit_index;
1601 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1602 if (!vcpu->arch.irq_pending[word_index])
1603 clear_bit(word_index, &vcpu->arch.irq_summary);
1604 svm_inject_irq(svm, irq);
1607 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1608 struct kvm_run *kvm_run)
1610 struct vcpu_svm *svm = to_svm(vcpu);
1611 struct vmcb_control_area *control = &svm->vmcb->control;
1613 svm->vcpu.arch.interrupt_window_open =
1614 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1615 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1617 if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1619 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1621 svm_do_inject_vector(svm);
1624 * Interrupts blocked. Wait for unblock.
1626 if (!svm->vcpu.arch.interrupt_window_open &&
1627 (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1628 control->intercept |= 1ULL << INTERCEPT_VINTR;
1630 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1633 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1638 static void save_db_regs(unsigned long *db_regs)
1640 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1641 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1642 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1643 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1646 static void load_db_regs(unsigned long *db_regs)
1648 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1649 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1650 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1651 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1654 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1656 force_new_asid(vcpu);
1659 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1663 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
1665 struct vcpu_svm *svm = to_svm(vcpu);
1667 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
1668 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
1669 kvm_lapic_set_tpr(vcpu, cr8);
1673 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
1675 struct vcpu_svm *svm = to_svm(vcpu);
1678 if (!irqchip_in_kernel(vcpu->kvm))
1681 cr8 = kvm_get_cr8(vcpu);
1682 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
1683 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
1686 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1688 struct vcpu_svm *svm = to_svm(vcpu);
1695 sync_lapic_to_cr8(vcpu);
1697 save_host_msrs(vcpu);
1698 fs_selector = read_fs();
1699 gs_selector = read_gs();
1700 ldt_selector = read_ldt();
1701 svm->host_cr2 = kvm_read_cr2();
1702 svm->host_dr6 = read_dr6();
1703 svm->host_dr7 = read_dr7();
1704 svm->vmcb->save.cr2 = vcpu->arch.cr2;
1705 /* required for live migration with NPT */
1707 svm->vmcb->save.cr3 = vcpu->arch.cr3;
1709 if (svm->vmcb->save.dr7 & 0xff) {
1711 save_db_regs(svm->host_db_regs);
1712 load_db_regs(svm->db_regs);
1720 #ifdef CONFIG_X86_64
1726 #ifdef CONFIG_X86_64
1727 "mov %c[rbx](%[svm]), %%rbx \n\t"
1728 "mov %c[rcx](%[svm]), %%rcx \n\t"
1729 "mov %c[rdx](%[svm]), %%rdx \n\t"
1730 "mov %c[rsi](%[svm]), %%rsi \n\t"
1731 "mov %c[rdi](%[svm]), %%rdi \n\t"
1732 "mov %c[rbp](%[svm]), %%rbp \n\t"
1733 "mov %c[r8](%[svm]), %%r8 \n\t"
1734 "mov %c[r9](%[svm]), %%r9 \n\t"
1735 "mov %c[r10](%[svm]), %%r10 \n\t"
1736 "mov %c[r11](%[svm]), %%r11 \n\t"
1737 "mov %c[r12](%[svm]), %%r12 \n\t"
1738 "mov %c[r13](%[svm]), %%r13 \n\t"
1739 "mov %c[r14](%[svm]), %%r14 \n\t"
1740 "mov %c[r15](%[svm]), %%r15 \n\t"
1742 "mov %c[rbx](%[svm]), %%ebx \n\t"
1743 "mov %c[rcx](%[svm]), %%ecx \n\t"
1744 "mov %c[rdx](%[svm]), %%edx \n\t"
1745 "mov %c[rsi](%[svm]), %%esi \n\t"
1746 "mov %c[rdi](%[svm]), %%edi \n\t"
1747 "mov %c[rbp](%[svm]), %%ebp \n\t"
1750 #ifdef CONFIG_X86_64
1751 /* Enter guest mode */
1753 "mov %c[vmcb](%[svm]), %%rax \n\t"
1759 /* Enter guest mode */
1761 "mov %c[vmcb](%[svm]), %%eax \n\t"
1768 /* Save guest registers, load host registers */
1769 #ifdef CONFIG_X86_64
1770 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1771 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1772 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1773 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1774 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1775 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1776 "mov %%r8, %c[r8](%[svm]) \n\t"
1777 "mov %%r9, %c[r9](%[svm]) \n\t"
1778 "mov %%r10, %c[r10](%[svm]) \n\t"
1779 "mov %%r11, %c[r11](%[svm]) \n\t"
1780 "mov %%r12, %c[r12](%[svm]) \n\t"
1781 "mov %%r13, %c[r13](%[svm]) \n\t"
1782 "mov %%r14, %c[r14](%[svm]) \n\t"
1783 "mov %%r15, %c[r15](%[svm]) \n\t"
1787 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1788 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1789 "mov %%edx, %c[rdx](%[svm]) \n\t"
1790 "mov %%esi, %c[rsi](%[svm]) \n\t"
1791 "mov %%edi, %c[rdi](%[svm]) \n\t"
1792 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1798 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1799 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1800 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1801 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1802 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1803 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1804 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1805 #ifdef CONFIG_X86_64
1806 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1807 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1808 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1809 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1810 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1811 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1812 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1813 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1816 #ifdef CONFIG_X86_64
1817 , "rbx", "rcx", "rdx", "rsi", "rdi"
1818 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1820 , "ebx", "ecx", "edx" , "esi", "edi"
1824 if ((svm->vmcb->save.dr7 & 0xff))
1825 load_db_regs(svm->host_db_regs);
1827 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1829 write_dr6(svm->host_dr6);
1830 write_dr7(svm->host_dr7);
1831 kvm_write_cr2(svm->host_cr2);
1833 load_fs(fs_selector);
1834 load_gs(gs_selector);
1835 load_ldt(ldt_selector);
1836 load_host_msrs(vcpu);
1840 local_irq_disable();
1844 sync_cr8_to_lapic(vcpu);
1849 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1851 struct vcpu_svm *svm = to_svm(vcpu);
1854 svm->vmcb->control.nested_cr3 = root;
1855 force_new_asid(vcpu);
1859 svm->vmcb->save.cr3 = root;
1860 force_new_asid(vcpu);
1862 if (vcpu->fpu_active) {
1863 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1864 svm->vmcb->save.cr0 |= X86_CR0_TS;
1865 vcpu->fpu_active = 0;
1869 static int is_disabled(void)
1873 rdmsrl(MSR_VM_CR, vm_cr);
1874 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1881 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1884 * Patch in the VMMCALL instruction:
1886 hypercall[0] = 0x0f;
1887 hypercall[1] = 0x01;
1888 hypercall[2] = 0xd9;
1891 static void svm_check_processor_compat(void *rtn)
1896 static bool svm_cpu_has_accelerated_tpr(void)
1901 static int get_npt_level(void)
1903 #ifdef CONFIG_X86_64
1904 return PT64_ROOT_LEVEL;
1906 return PT32E_ROOT_LEVEL;
1910 static struct kvm_x86_ops svm_x86_ops = {
1911 .cpu_has_kvm_support = has_svm,
1912 .disabled_by_bios = is_disabled,
1913 .hardware_setup = svm_hardware_setup,
1914 .hardware_unsetup = svm_hardware_unsetup,
1915 .check_processor_compatibility = svm_check_processor_compat,
1916 .hardware_enable = svm_hardware_enable,
1917 .hardware_disable = svm_hardware_disable,
1918 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1920 .vcpu_create = svm_create_vcpu,
1921 .vcpu_free = svm_free_vcpu,
1922 .vcpu_reset = svm_vcpu_reset,
1924 .prepare_guest_switch = svm_prepare_guest_switch,
1925 .vcpu_load = svm_vcpu_load,
1926 .vcpu_put = svm_vcpu_put,
1927 .vcpu_decache = svm_vcpu_decache,
1929 .set_guest_debug = svm_guest_debug,
1930 .get_msr = svm_get_msr,
1931 .set_msr = svm_set_msr,
1932 .get_segment_base = svm_get_segment_base,
1933 .get_segment = svm_get_segment,
1934 .set_segment = svm_set_segment,
1935 .get_cpl = svm_get_cpl,
1936 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1937 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1938 .set_cr0 = svm_set_cr0,
1939 .set_cr3 = svm_set_cr3,
1940 .set_cr4 = svm_set_cr4,
1941 .set_efer = svm_set_efer,
1942 .get_idt = svm_get_idt,
1943 .set_idt = svm_set_idt,
1944 .get_gdt = svm_get_gdt,
1945 .set_gdt = svm_set_gdt,
1946 .get_dr = svm_get_dr,
1947 .set_dr = svm_set_dr,
1948 .cache_regs = svm_cache_regs,
1949 .decache_regs = svm_decache_regs,
1950 .get_rflags = svm_get_rflags,
1951 .set_rflags = svm_set_rflags,
1953 .tlb_flush = svm_flush_tlb,
1955 .run = svm_vcpu_run,
1956 .handle_exit = handle_exit,
1957 .skip_emulated_instruction = skip_emulated_instruction,
1958 .patch_hypercall = svm_patch_hypercall,
1959 .get_irq = svm_get_irq,
1960 .set_irq = svm_set_irq,
1961 .queue_exception = svm_queue_exception,
1962 .exception_injected = svm_exception_injected,
1963 .inject_pending_irq = svm_intr_assist,
1964 .inject_pending_vectors = do_interrupt_requests,
1966 .set_tss_addr = svm_set_tss_addr,
1967 .get_tdp_level = get_npt_level,
1970 static int __init svm_init(void)
1972 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1976 static void __exit svm_exit(void)
1981 module_init(svm_init)
1982 module_exit(svm_exit)