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>
21 #include "kvm_cache_regs.h"
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/vmalloc.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
32 #include <asm/virtext.h>
34 #define __ex(x) __kvm_handle_fault_on_reboot(x)
36 MODULE_AUTHOR("Qumranet");
37 MODULE_LICENSE("GPL");
39 #define IOPM_ALLOC_ORDER 2
40 #define MSRPM_ALLOC_ORDER 1
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define SVM_FEATURE_NPT (1 << 0)
46 #define SVM_FEATURE_LBRV (1 << 1)
47 #define SVM_FEATURE_SVML (1 << 2)
49 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
51 /* Turn on to get debugging output*/
52 /* #define NESTED_DEBUG */
55 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
57 #define nsvm_printk(fmt, args...) do {} while(0)
60 /* enable NPT for AMD64 and X86 with PAE */
61 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
62 static bool npt_enabled = true;
64 static bool npt_enabled = false;
68 module_param(npt, int, S_IRUGO);
70 static int nested = 0;
71 module_param(nested, int, S_IRUGO);
73 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
75 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override);
76 static int nested_svm_vmexit(struct vcpu_svm *svm);
77 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
78 void *arg2, void *opaque);
79 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
80 bool has_error_code, u32 error_code);
82 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
84 return container_of(vcpu, struct vcpu_svm, vcpu);
87 static inline bool is_nested(struct vcpu_svm *svm)
89 return svm->nested_vmcb;
92 static unsigned long iopm_base;
94 struct kvm_ldttss_desc {
97 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
98 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
101 } __attribute__((packed));
103 struct svm_cpu_data {
109 struct kvm_ldttss_desc *tss_desc;
111 struct page *save_area;
114 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
115 static uint32_t svm_features;
117 struct svm_init_data {
122 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
124 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
125 #define MSRS_RANGE_SIZE 2048
126 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
128 #define MAX_INST_SIZE 15
130 static inline u32 svm_has(u32 feat)
132 return svm_features & feat;
135 static inline void clgi(void)
137 asm volatile (__ex(SVM_CLGI));
140 static inline void stgi(void)
142 asm volatile (__ex(SVM_STGI));
145 static inline void invlpga(unsigned long addr, u32 asid)
147 asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
150 static inline unsigned long kvm_read_cr2(void)
154 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
158 static inline void kvm_write_cr2(unsigned long val)
160 asm volatile ("mov %0, %%cr2" :: "r" (val));
163 static inline void force_new_asid(struct kvm_vcpu *vcpu)
165 to_svm(vcpu)->asid_generation--;
168 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
170 force_new_asid(vcpu);
173 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
175 if (!npt_enabled && !(efer & EFER_LMA))
178 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
179 vcpu->arch.shadow_efer = efer;
182 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
183 bool has_error_code, u32 error_code)
185 struct vcpu_svm *svm = to_svm(vcpu);
187 /* If we are within a nested VM we'd better #VMEXIT and let the
188 guest handle the exception */
189 if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
192 svm->vmcb->control.event_inj = nr
194 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
195 | SVM_EVTINJ_TYPE_EXEPT;
196 svm->vmcb->control.event_inj_err = error_code;
199 static int is_external_interrupt(u32 info)
201 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
202 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
205 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
207 struct vcpu_svm *svm = to_svm(vcpu);
210 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
211 ret |= X86_SHADOW_INT_STI | X86_SHADOW_INT_MOV_SS;
215 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
217 struct vcpu_svm *svm = to_svm(vcpu);
220 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
222 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
226 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
228 struct vcpu_svm *svm = to_svm(vcpu);
230 if (!svm->next_rip) {
231 if (emulate_instruction(vcpu, vcpu->run, 0, 0, EMULTYPE_SKIP) !=
233 printk(KERN_DEBUG "%s: NOP\n", __func__);
236 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
237 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
238 __func__, kvm_rip_read(vcpu), svm->next_rip);
240 kvm_rip_write(vcpu, svm->next_rip);
241 svm_set_interrupt_shadow(vcpu, 0);
244 static int has_svm(void)
248 if (!cpu_has_svm(&msg)) {
249 printk(KERN_INFO "has_svm: %s\n", msg);
256 static void svm_hardware_disable(void *garbage)
261 static void svm_hardware_enable(void *garbage)
264 struct svm_cpu_data *svm_data;
266 struct desc_ptr gdt_descr;
267 struct desc_struct *gdt;
268 int me = raw_smp_processor_id();
271 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
274 svm_data = per_cpu(svm_data, me);
277 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
282 svm_data->asid_generation = 1;
283 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
284 svm_data->next_asid = svm_data->max_asid + 1;
286 asm volatile ("sgdt %0" : "=m"(gdt_descr));
287 gdt = (struct desc_struct *)gdt_descr.address;
288 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
290 rdmsrl(MSR_EFER, efer);
291 wrmsrl(MSR_EFER, efer | EFER_SVME);
293 wrmsrl(MSR_VM_HSAVE_PA,
294 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
297 static void svm_cpu_uninit(int cpu)
299 struct svm_cpu_data *svm_data
300 = per_cpu(svm_data, raw_smp_processor_id());
305 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
306 __free_page(svm_data->save_area);
310 static int svm_cpu_init(int cpu)
312 struct svm_cpu_data *svm_data;
315 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
319 svm_data->save_area = alloc_page(GFP_KERNEL);
321 if (!svm_data->save_area)
324 per_cpu(svm_data, cpu) = svm_data;
334 static void set_msr_interception(u32 *msrpm, unsigned msr,
339 for (i = 0; i < NUM_MSR_MAPS; i++) {
340 if (msr >= msrpm_ranges[i] &&
341 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
342 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
343 msrpm_ranges[i]) * 2;
345 u32 *base = msrpm + (msr_offset / 32);
346 u32 msr_shift = msr_offset % 32;
347 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
348 *base = (*base & ~(0x3 << msr_shift)) |
356 static void svm_vcpu_init_msrpm(u32 *msrpm)
358 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
361 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
362 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
363 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
364 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
365 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
366 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
368 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
369 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
370 set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
371 set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
374 static void svm_enable_lbrv(struct vcpu_svm *svm)
376 u32 *msrpm = svm->msrpm;
378 svm->vmcb->control.lbr_ctl = 1;
379 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
380 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
381 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
382 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
385 static void svm_disable_lbrv(struct vcpu_svm *svm)
387 u32 *msrpm = svm->msrpm;
389 svm->vmcb->control.lbr_ctl = 0;
390 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
391 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
392 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
393 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
396 static __init int svm_hardware_setup(void)
399 struct page *iopm_pages;
403 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
408 iopm_va = page_address(iopm_pages);
409 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
410 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
412 if (boot_cpu_has(X86_FEATURE_NX))
413 kvm_enable_efer_bits(EFER_NX);
415 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
416 kvm_enable_efer_bits(EFER_FFXSR);
419 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
420 kvm_enable_efer_bits(EFER_SVME);
423 for_each_online_cpu(cpu) {
424 r = svm_cpu_init(cpu);
429 svm_features = cpuid_edx(SVM_CPUID_FUNC);
431 if (!svm_has(SVM_FEATURE_NPT))
434 if (npt_enabled && !npt) {
435 printk(KERN_INFO "kvm: Nested Paging disabled\n");
440 printk(KERN_INFO "kvm: Nested Paging enabled\n");
448 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
453 static __exit void svm_hardware_unsetup(void)
457 for_each_online_cpu(cpu)
460 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
464 static void init_seg(struct vmcb_seg *seg)
467 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
468 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
473 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
476 seg->attrib = SVM_SELECTOR_P_MASK | type;
481 static void init_vmcb(struct vcpu_svm *svm)
483 struct vmcb_control_area *control = &svm->vmcb->control;
484 struct vmcb_save_area *save = &svm->vmcb->save;
486 control->intercept_cr_read = INTERCEPT_CR0_MASK |
490 control->intercept_cr_write = INTERCEPT_CR0_MASK |
495 control->intercept_dr_read = INTERCEPT_DR0_MASK |
500 control->intercept_dr_write = INTERCEPT_DR0_MASK |
507 control->intercept_exceptions = (1 << PF_VECTOR) |
512 control->intercept = (1ULL << INTERCEPT_INTR) |
513 (1ULL << INTERCEPT_NMI) |
514 (1ULL << INTERCEPT_SMI) |
515 (1ULL << INTERCEPT_CPUID) |
516 (1ULL << INTERCEPT_INVD) |
517 (1ULL << INTERCEPT_HLT) |
518 (1ULL << INTERCEPT_INVLPG) |
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 = EFER_SVME;
566 save->dr6 = 0xffff0ff0;
569 save->rip = 0x0000fff0;
570 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
573 * cr0 val on cpu init should be 0x60000010, we enable cpu
574 * cache by default. the orderly way is to enable cache in bios.
576 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
577 save->cr4 = X86_CR4_PAE;
581 /* Setup VMCB for Nested Paging */
582 control->nested_ctl = 1;
583 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
584 (1ULL << INTERCEPT_INVLPG));
585 control->intercept_exceptions &= ~(1 << PF_VECTOR);
586 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
588 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
590 save->g_pat = 0x0007040600070406ULL;
591 /* enable caching because the QEMU Bios doesn't enable it */
592 save->cr0 = X86_CR0_ET;
596 force_new_asid(&svm->vcpu);
598 svm->nested_vmcb = 0;
599 svm->vcpu.arch.hflags = HF_GIF_MASK;
602 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
604 struct vcpu_svm *svm = to_svm(vcpu);
608 if (vcpu->vcpu_id != 0) {
609 kvm_rip_write(vcpu, 0);
610 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
611 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
613 vcpu->arch.regs_avail = ~0;
614 vcpu->arch.regs_dirty = ~0;
619 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
621 struct vcpu_svm *svm;
623 struct page *msrpm_pages;
624 struct page *hsave_page;
625 struct page *nested_msrpm_pages;
628 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
634 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
638 page = alloc_page(GFP_KERNEL);
645 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
649 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
650 if (!nested_msrpm_pages)
653 svm->msrpm = page_address(msrpm_pages);
654 svm_vcpu_init_msrpm(svm->msrpm);
656 hsave_page = alloc_page(GFP_KERNEL);
659 svm->hsave = page_address(hsave_page);
661 svm->nested_msrpm = page_address(nested_msrpm_pages);
663 svm->vmcb = page_address(page);
664 clear_page(svm->vmcb);
665 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
666 svm->asid_generation = 0;
670 svm->vcpu.fpu_active = 1;
671 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
672 if (svm->vcpu.vcpu_id == 0)
673 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
678 kvm_vcpu_uninit(&svm->vcpu);
680 kmem_cache_free(kvm_vcpu_cache, svm);
685 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
687 struct vcpu_svm *svm = to_svm(vcpu);
689 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
690 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
691 __free_page(virt_to_page(svm->hsave));
692 __free_pages(virt_to_page(svm->nested_msrpm), MSRPM_ALLOC_ORDER);
693 kvm_vcpu_uninit(vcpu);
694 kmem_cache_free(kvm_vcpu_cache, svm);
697 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
699 struct vcpu_svm *svm = to_svm(vcpu);
702 if (unlikely(cpu != vcpu->cpu)) {
706 * Make sure that the guest sees a monotonically
710 delta = vcpu->arch.host_tsc - tsc_this;
711 svm->vmcb->control.tsc_offset += delta;
713 kvm_migrate_timers(vcpu);
716 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
717 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
720 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
722 struct vcpu_svm *svm = to_svm(vcpu);
725 ++vcpu->stat.host_state_reload;
726 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
727 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
729 rdtscll(vcpu->arch.host_tsc);
732 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
734 return to_svm(vcpu)->vmcb->save.rflags;
737 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
739 to_svm(vcpu)->vmcb->save.rflags = rflags;
742 static void svm_set_vintr(struct vcpu_svm *svm)
744 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
747 static void svm_clear_vintr(struct vcpu_svm *svm)
749 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
752 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
754 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
757 case VCPU_SREG_CS: return &save->cs;
758 case VCPU_SREG_DS: return &save->ds;
759 case VCPU_SREG_ES: return &save->es;
760 case VCPU_SREG_FS: return &save->fs;
761 case VCPU_SREG_GS: return &save->gs;
762 case VCPU_SREG_SS: return &save->ss;
763 case VCPU_SREG_TR: return &save->tr;
764 case VCPU_SREG_LDTR: return &save->ldtr;
770 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
772 struct vmcb_seg *s = svm_seg(vcpu, seg);
777 static void svm_get_segment(struct kvm_vcpu *vcpu,
778 struct kvm_segment *var, int seg)
780 struct vmcb_seg *s = svm_seg(vcpu, seg);
783 var->limit = s->limit;
784 var->selector = s->selector;
785 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
786 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
787 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
788 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
789 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
790 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
791 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
792 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
794 /* AMD's VMCB does not have an explicit unusable field, so emulate it
795 * for cross vendor migration purposes by "not present"
797 var->unusable = !var->present || (var->type == 0);
802 * SVM always stores 0 for the 'G' bit in the CS selector in
803 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
804 * Intel's VMENTRY has a check on the 'G' bit.
806 var->g = s->limit > 0xfffff;
810 * Work around a bug where the busy flag in the tr selector
820 * The accessed bit must always be set in the segment
821 * descriptor cache, although it can be cleared in the
822 * descriptor, the cached bit always remains at 1. Since
823 * Intel has a check on this, set it here to support
824 * cross-vendor migration.
830 /* On AMD CPUs sometimes the DB bit in the segment
831 * descriptor is left as 1, although the whole segment has
832 * been made unusable. Clear it here to pass an Intel VMX
833 * entry check when cross vendor migrating.
841 static int svm_get_cpl(struct kvm_vcpu *vcpu)
843 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
848 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
850 struct vcpu_svm *svm = to_svm(vcpu);
852 dt->limit = svm->vmcb->save.idtr.limit;
853 dt->base = svm->vmcb->save.idtr.base;
856 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
858 struct vcpu_svm *svm = to_svm(vcpu);
860 svm->vmcb->save.idtr.limit = dt->limit;
861 svm->vmcb->save.idtr.base = dt->base ;
864 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
866 struct vcpu_svm *svm = to_svm(vcpu);
868 dt->limit = svm->vmcb->save.gdtr.limit;
869 dt->base = svm->vmcb->save.gdtr.base;
872 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
874 struct vcpu_svm *svm = to_svm(vcpu);
876 svm->vmcb->save.gdtr.limit = dt->limit;
877 svm->vmcb->save.gdtr.base = dt->base ;
880 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
884 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
886 struct vcpu_svm *svm = to_svm(vcpu);
889 if (vcpu->arch.shadow_efer & EFER_LME) {
890 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
891 vcpu->arch.shadow_efer |= EFER_LMA;
892 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
895 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
896 vcpu->arch.shadow_efer &= ~EFER_LMA;
897 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
904 if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
905 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
906 vcpu->fpu_active = 1;
909 vcpu->arch.cr0 = cr0;
910 cr0 |= X86_CR0_PG | X86_CR0_WP;
911 if (!vcpu->fpu_active) {
912 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
917 * re-enable caching here because the QEMU bios
918 * does not do it - this results in some delay at
921 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
922 svm->vmcb->save.cr0 = cr0;
925 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
927 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
928 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
930 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
931 force_new_asid(vcpu);
933 vcpu->arch.cr4 = cr4;
937 to_svm(vcpu)->vmcb->save.cr4 = cr4;
940 static void svm_set_segment(struct kvm_vcpu *vcpu,
941 struct kvm_segment *var, int seg)
943 struct vcpu_svm *svm = to_svm(vcpu);
944 struct vmcb_seg *s = svm_seg(vcpu, seg);
947 s->limit = var->limit;
948 s->selector = var->selector;
952 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
953 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
954 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
955 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
956 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
957 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
958 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
959 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
961 if (seg == VCPU_SREG_CS)
963 = (svm->vmcb->save.cs.attrib
964 >> SVM_SELECTOR_DPL_SHIFT) & 3;
968 static void update_db_intercept(struct kvm_vcpu *vcpu)
970 struct vcpu_svm *svm = to_svm(vcpu);
972 svm->vmcb->control.intercept_exceptions &=
973 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
975 if (vcpu->arch.singlestep)
976 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
978 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
979 if (vcpu->guest_debug &
980 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
981 svm->vmcb->control.intercept_exceptions |=
983 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
984 svm->vmcb->control.intercept_exceptions |=
987 vcpu->guest_debug = 0;
990 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
992 int old_debug = vcpu->guest_debug;
993 struct vcpu_svm *svm = to_svm(vcpu);
995 vcpu->guest_debug = dbg->control;
997 update_db_intercept(vcpu);
999 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1000 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1002 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1004 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1005 svm->vmcb->save.rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1006 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1007 svm->vmcb->save.rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1012 static void load_host_msrs(struct kvm_vcpu *vcpu)
1014 #ifdef CONFIG_X86_64
1015 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1019 static void save_host_msrs(struct kvm_vcpu *vcpu)
1021 #ifdef CONFIG_X86_64
1022 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1026 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
1028 if (svm_data->next_asid > svm_data->max_asid) {
1029 ++svm_data->asid_generation;
1030 svm_data->next_asid = 1;
1031 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1034 svm->vcpu.cpu = svm_data->cpu;
1035 svm->asid_generation = svm_data->asid_generation;
1036 svm->vmcb->control.asid = svm_data->next_asid++;
1039 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
1041 struct vcpu_svm *svm = to_svm(vcpu);
1046 val = vcpu->arch.db[dr];
1049 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1050 val = vcpu->arch.dr6;
1052 val = svm->vmcb->save.dr6;
1055 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1056 val = vcpu->arch.dr7;
1058 val = svm->vmcb->save.dr7;
1064 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
1068 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
1071 struct vcpu_svm *svm = to_svm(vcpu);
1073 KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)value, handler);
1079 vcpu->arch.db[dr] = value;
1080 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1081 vcpu->arch.eff_db[dr] = value;
1084 if (vcpu->arch.cr4 & X86_CR4_DE)
1085 *exception = UD_VECTOR;
1088 if (value & 0xffffffff00000000ULL) {
1089 *exception = GP_VECTOR;
1092 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1095 if (value & 0xffffffff00000000ULL) {
1096 *exception = GP_VECTOR;
1099 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1100 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1101 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1102 vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1106 /* FIXME: Possible case? */
1107 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1109 *exception = UD_VECTOR;
1114 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1119 fault_address = svm->vmcb->control.exit_info_2;
1120 error_code = svm->vmcb->control.exit_info_1;
1123 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1124 (u32)fault_address, (u32)(fault_address >> 32),
1127 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1128 (u32)fault_address, (u32)(fault_address >> 32),
1131 * FIXME: Tis shouldn't be necessary here, but there is a flush
1132 * missing in the MMU code. Until we find this bug, flush the
1133 * complete TLB here on an NPF
1136 svm_flush_tlb(&svm->vcpu);
1138 if (kvm_event_needs_reinjection(&svm->vcpu))
1139 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1141 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1144 static int db_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1146 if (!(svm->vcpu.guest_debug &
1147 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1148 !svm->vcpu.arch.singlestep) {
1149 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1153 if (svm->vcpu.arch.singlestep) {
1154 svm->vcpu.arch.singlestep = false;
1155 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1156 svm->vmcb->save.rflags &=
1157 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1158 update_db_intercept(&svm->vcpu);
1161 if (svm->vcpu.guest_debug &
1162 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)){
1163 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1164 kvm_run->debug.arch.pc =
1165 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1166 kvm_run->debug.arch.exception = DB_VECTOR;
1173 static int bp_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1175 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1176 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1177 kvm_run->debug.arch.exception = BP_VECTOR;
1181 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1185 er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1186 if (er != EMULATE_DONE)
1187 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1191 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1193 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1194 if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1195 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1196 svm->vcpu.fpu_active = 1;
1201 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1204 * On an #MC intercept the MCE handler is not called automatically in
1205 * the host. So do it by hand here.
1209 /* not sure if we ever come back to this point */
1214 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1217 * VMCB is undefined after a SHUTDOWN intercept
1218 * so reinitialize it.
1220 clear_page(svm->vmcb);
1223 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1227 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1229 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1230 int size, in, string;
1233 ++svm->vcpu.stat.io_exits;
1235 svm->next_rip = svm->vmcb->control.exit_info_2;
1237 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1240 if (emulate_instruction(&svm->vcpu,
1241 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1246 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1247 port = io_info >> 16;
1248 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1250 skip_emulated_instruction(&svm->vcpu);
1251 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1254 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1256 KVMTRACE_0D(NMI, &svm->vcpu, handler);
1260 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1262 ++svm->vcpu.stat.irq_exits;
1263 KVMTRACE_0D(INTR, &svm->vcpu, handler);
1267 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1272 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1274 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1275 skip_emulated_instruction(&svm->vcpu);
1276 return kvm_emulate_halt(&svm->vcpu);
1279 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1281 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1282 skip_emulated_instruction(&svm->vcpu);
1283 kvm_emulate_hypercall(&svm->vcpu);
1287 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1289 if (!(svm->vcpu.arch.shadow_efer & EFER_SVME)
1290 || !is_paging(&svm->vcpu)) {
1291 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1295 if (svm->vmcb->save.cpl) {
1296 kvm_inject_gp(&svm->vcpu, 0);
1303 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1304 bool has_error_code, u32 error_code)
1306 if (is_nested(svm)) {
1307 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1308 svm->vmcb->control.exit_code_hi = 0;
1309 svm->vmcb->control.exit_info_1 = error_code;
1310 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1311 if (nested_svm_exit_handled(svm, false)) {
1312 nsvm_printk("VMexit -> EXCP 0x%x\n", nr);
1314 nested_svm_vmexit(svm);
1322 static inline int nested_svm_intr(struct vcpu_svm *svm)
1324 if (is_nested(svm)) {
1325 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1328 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1331 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1333 if (nested_svm_exit_handled(svm, false)) {
1334 nsvm_printk("VMexit -> INTR\n");
1335 nested_svm_vmexit(svm);
1343 static struct page *nested_svm_get_page(struct vcpu_svm *svm, u64 gpa)
1347 down_read(¤t->mm->mmap_sem);
1348 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1349 up_read(¤t->mm->mmap_sem);
1351 if (is_error_page(page)) {
1352 printk(KERN_INFO "%s: could not find page at 0x%llx\n",
1354 kvm_release_page_clean(page);
1355 kvm_inject_gp(&svm->vcpu, 0);
1361 static int nested_svm_do(struct vcpu_svm *svm,
1362 u64 arg1_gpa, u64 arg2_gpa, void *opaque,
1363 int (*handler)(struct vcpu_svm *svm,
1368 struct page *arg1_page;
1369 struct page *arg2_page = NULL;
1374 arg1_page = nested_svm_get_page(svm, arg1_gpa);
1375 if(arg1_page == NULL)
1379 arg2_page = nested_svm_get_page(svm, arg2_gpa);
1380 if(arg2_page == NULL) {
1381 kvm_release_page_clean(arg1_page);
1386 arg1 = kmap_atomic(arg1_page, KM_USER0);
1388 arg2 = kmap_atomic(arg2_page, KM_USER1);
1390 retval = handler(svm, arg1, arg2, opaque);
1392 kunmap_atomic(arg1, KM_USER0);
1394 kunmap_atomic(arg2, KM_USER1);
1396 kvm_release_page_dirty(arg1_page);
1398 kvm_release_page_dirty(arg2_page);
1403 static int nested_svm_exit_handled_real(struct vcpu_svm *svm,
1408 struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1409 bool kvm_overrides = *(bool *)opaque;
1410 u32 exit_code = svm->vmcb->control.exit_code;
1412 if (kvm_overrides) {
1413 switch (exit_code) {
1417 /* For now we are always handling NPFs when using them */
1422 /* When we're shadowing, trap PFs */
1423 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1432 switch (exit_code) {
1433 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1434 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1435 if (nested_vmcb->control.intercept_cr_read & cr_bits)
1439 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1440 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1441 if (nested_vmcb->control.intercept_cr_write & cr_bits)
1445 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1446 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1447 if (nested_vmcb->control.intercept_dr_read & dr_bits)
1451 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1452 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1453 if (nested_vmcb->control.intercept_dr_write & dr_bits)
1457 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1458 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1459 if (nested_vmcb->control.intercept_exceptions & excp_bits)
1464 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1465 nsvm_printk("exit code: 0x%x\n", exit_code);
1466 if (nested_vmcb->control.intercept & exit_bits)
1474 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm,
1475 void *arg1, void *arg2,
1478 struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1479 u8 *msrpm = (u8 *)arg2;
1481 u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1482 u32 param = svm->vmcb->control.exit_info_1 & 1;
1484 if (!(nested_vmcb->control.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1492 case 0xc0000000 ... 0xc0001fff:
1493 t0 = (8192 + msr - 0xc0000000) * 2;
1497 case 0xc0010000 ... 0xc0011fff:
1498 t0 = (16384 + msr - 0xc0010000) * 2;
1506 if (msrpm[t1] & ((1 << param) << t0))
1512 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override)
1514 bool k = kvm_override;
1516 switch (svm->vmcb->control.exit_code) {
1518 return nested_svm_do(svm, svm->nested_vmcb,
1519 svm->nested_vmcb_msrpm, NULL,
1520 nested_svm_exit_handled_msr);
1524 return nested_svm_do(svm, svm->nested_vmcb, 0, &k,
1525 nested_svm_exit_handled_real);
1528 static int nested_svm_vmexit_real(struct vcpu_svm *svm, void *arg1,
1529 void *arg2, void *opaque)
1531 struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1532 struct vmcb *hsave = svm->hsave;
1533 u64 nested_save[] = { nested_vmcb->save.cr0,
1534 nested_vmcb->save.cr3,
1535 nested_vmcb->save.cr4,
1536 nested_vmcb->save.efer,
1537 nested_vmcb->control.intercept_cr_read,
1538 nested_vmcb->control.intercept_cr_write,
1539 nested_vmcb->control.intercept_dr_read,
1540 nested_vmcb->control.intercept_dr_write,
1541 nested_vmcb->control.intercept_exceptions,
1542 nested_vmcb->control.intercept,
1543 nested_vmcb->control.msrpm_base_pa,
1544 nested_vmcb->control.iopm_base_pa,
1545 nested_vmcb->control.tsc_offset };
1547 /* Give the current vmcb to the guest */
1548 memcpy(nested_vmcb, svm->vmcb, sizeof(struct vmcb));
1549 nested_vmcb->save.cr0 = nested_save[0];
1551 nested_vmcb->save.cr3 = nested_save[1];
1552 nested_vmcb->save.cr4 = nested_save[2];
1553 nested_vmcb->save.efer = nested_save[3];
1554 nested_vmcb->control.intercept_cr_read = nested_save[4];
1555 nested_vmcb->control.intercept_cr_write = nested_save[5];
1556 nested_vmcb->control.intercept_dr_read = nested_save[6];
1557 nested_vmcb->control.intercept_dr_write = nested_save[7];
1558 nested_vmcb->control.intercept_exceptions = nested_save[8];
1559 nested_vmcb->control.intercept = nested_save[9];
1560 nested_vmcb->control.msrpm_base_pa = nested_save[10];
1561 nested_vmcb->control.iopm_base_pa = nested_save[11];
1562 nested_vmcb->control.tsc_offset = nested_save[12];
1564 /* We always set V_INTR_MASKING and remember the old value in hflags */
1565 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1566 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1568 if ((nested_vmcb->control.int_ctl & V_IRQ_MASK) &&
1569 (nested_vmcb->control.int_vector)) {
1570 nsvm_printk("WARNING: IRQ 0x%x still enabled on #VMEXIT\n",
1571 nested_vmcb->control.int_vector);
1574 /* Restore the original control entries */
1575 svm->vmcb->control = hsave->control;
1577 /* Kill any pending exceptions */
1578 if (svm->vcpu.arch.exception.pending == true)
1579 nsvm_printk("WARNING: Pending Exception\n");
1580 svm->vcpu.arch.exception.pending = false;
1582 /* Restore selected save entries */
1583 svm->vmcb->save.es = hsave->save.es;
1584 svm->vmcb->save.cs = hsave->save.cs;
1585 svm->vmcb->save.ss = hsave->save.ss;
1586 svm->vmcb->save.ds = hsave->save.ds;
1587 svm->vmcb->save.gdtr = hsave->save.gdtr;
1588 svm->vmcb->save.idtr = hsave->save.idtr;
1589 svm->vmcb->save.rflags = hsave->save.rflags;
1590 svm_set_efer(&svm->vcpu, hsave->save.efer);
1591 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1592 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1594 svm->vmcb->save.cr3 = hsave->save.cr3;
1595 svm->vcpu.arch.cr3 = hsave->save.cr3;
1597 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1599 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1600 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1601 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1602 svm->vmcb->save.dr7 = 0;
1603 svm->vmcb->save.cpl = 0;
1604 svm->vmcb->control.exit_int_info = 0;
1606 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1607 /* Exit nested SVM mode */
1608 svm->nested_vmcb = 0;
1613 static int nested_svm_vmexit(struct vcpu_svm *svm)
1615 nsvm_printk("VMexit\n");
1616 if (nested_svm_do(svm, svm->nested_vmcb, 0,
1617 NULL, nested_svm_vmexit_real))
1620 kvm_mmu_reset_context(&svm->vcpu);
1621 kvm_mmu_load(&svm->vcpu);
1626 static int nested_svm_vmrun_msrpm(struct vcpu_svm *svm, void *arg1,
1627 void *arg2, void *opaque)
1630 u32 *nested_msrpm = (u32*)arg1;
1631 for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1632 svm->nested_msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1633 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested_msrpm);
1638 static int nested_svm_vmrun(struct vcpu_svm *svm, void *arg1,
1639 void *arg2, void *opaque)
1641 struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1642 struct vmcb *hsave = svm->hsave;
1644 /* nested_vmcb is our indicator if nested SVM is activated */
1645 svm->nested_vmcb = svm->vmcb->save.rax;
1647 /* Clear internal status */
1648 svm->vcpu.arch.exception.pending = false;
1650 /* Save the old vmcb, so we don't need to pick what we save, but
1651 can restore everything when a VMEXIT occurs */
1652 memcpy(hsave, svm->vmcb, sizeof(struct vmcb));
1653 /* We need to remember the original CR3 in the SPT case */
1655 hsave->save.cr3 = svm->vcpu.arch.cr3;
1656 hsave->save.cr4 = svm->vcpu.arch.cr4;
1657 hsave->save.rip = svm->next_rip;
1659 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1660 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1662 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1664 /* Load the nested guest state */
1665 svm->vmcb->save.es = nested_vmcb->save.es;
1666 svm->vmcb->save.cs = nested_vmcb->save.cs;
1667 svm->vmcb->save.ss = nested_vmcb->save.ss;
1668 svm->vmcb->save.ds = nested_vmcb->save.ds;
1669 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1670 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1671 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1672 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1673 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1674 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1676 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1677 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1679 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1680 kvm_mmu_reset_context(&svm->vcpu);
1682 svm->vmcb->save.cr2 = nested_vmcb->save.cr2;
1683 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1684 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1685 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1686 /* In case we don't even reach vcpu_run, the fields are not updated */
1687 svm->vmcb->save.rax = nested_vmcb->save.rax;
1688 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1689 svm->vmcb->save.rip = nested_vmcb->save.rip;
1690 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1691 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1692 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1694 /* We don't want a nested guest to be more powerful than the guest,
1695 so all intercepts are ORed */
1696 svm->vmcb->control.intercept_cr_read |=
1697 nested_vmcb->control.intercept_cr_read;
1698 svm->vmcb->control.intercept_cr_write |=
1699 nested_vmcb->control.intercept_cr_write;
1700 svm->vmcb->control.intercept_dr_read |=
1701 nested_vmcb->control.intercept_dr_read;
1702 svm->vmcb->control.intercept_dr_write |=
1703 nested_vmcb->control.intercept_dr_write;
1704 svm->vmcb->control.intercept_exceptions |=
1705 nested_vmcb->control.intercept_exceptions;
1707 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1709 svm->nested_vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1711 force_new_asid(&svm->vcpu);
1712 svm->vmcb->control.exit_int_info = nested_vmcb->control.exit_int_info;
1713 svm->vmcb->control.exit_int_info_err = nested_vmcb->control.exit_int_info_err;
1714 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1715 if (nested_vmcb->control.int_ctl & V_IRQ_MASK) {
1716 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1717 nested_vmcb->control.int_ctl);
1719 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1720 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1722 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1724 nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1725 nested_vmcb->control.exit_int_info,
1726 nested_vmcb->control.int_state);
1728 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1729 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1730 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1731 if (nested_vmcb->control.event_inj & SVM_EVTINJ_VALID)
1732 nsvm_printk("Injecting Event: 0x%x\n",
1733 nested_vmcb->control.event_inj);
1734 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1735 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1737 svm->vcpu.arch.hflags |= HF_GIF_MASK;
1742 static int nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1744 to_vmcb->save.fs = from_vmcb->save.fs;
1745 to_vmcb->save.gs = from_vmcb->save.gs;
1746 to_vmcb->save.tr = from_vmcb->save.tr;
1747 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1748 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1749 to_vmcb->save.star = from_vmcb->save.star;
1750 to_vmcb->save.lstar = from_vmcb->save.lstar;
1751 to_vmcb->save.cstar = from_vmcb->save.cstar;
1752 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1753 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1754 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1755 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1760 static int nested_svm_vmload(struct vcpu_svm *svm, void *nested_vmcb,
1761 void *arg2, void *opaque)
1763 return nested_svm_vmloadsave((struct vmcb *)nested_vmcb, svm->vmcb);
1766 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
1767 void *arg2, void *opaque)
1769 return nested_svm_vmloadsave(svm->vmcb, (struct vmcb *)nested_vmcb);
1772 static int vmload_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1774 if (nested_svm_check_permissions(svm))
1777 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1778 skip_emulated_instruction(&svm->vcpu);
1780 nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmload);
1785 static int vmsave_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1787 if (nested_svm_check_permissions(svm))
1790 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1791 skip_emulated_instruction(&svm->vcpu);
1793 nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmsave);
1798 static int vmrun_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1800 nsvm_printk("VMrun\n");
1801 if (nested_svm_check_permissions(svm))
1804 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1805 skip_emulated_instruction(&svm->vcpu);
1807 if (nested_svm_do(svm, svm->vmcb->save.rax, 0,
1808 NULL, nested_svm_vmrun))
1811 if (nested_svm_do(svm, svm->nested_vmcb_msrpm, 0,
1812 NULL, nested_svm_vmrun_msrpm))
1818 static int stgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1820 if (nested_svm_check_permissions(svm))
1823 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1824 skip_emulated_instruction(&svm->vcpu);
1826 svm->vcpu.arch.hflags |= HF_GIF_MASK;
1831 static int clgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1833 if (nested_svm_check_permissions(svm))
1836 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1837 skip_emulated_instruction(&svm->vcpu);
1839 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1841 /* After a CLGI no interrupts should come */
1842 svm_clear_vintr(svm);
1843 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1848 static int invalid_op_interception(struct vcpu_svm *svm,
1849 struct kvm_run *kvm_run)
1851 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1855 static int task_switch_interception(struct vcpu_svm *svm,
1856 struct kvm_run *kvm_run)
1860 int int_type = svm->vmcb->control.exit_int_info &
1861 SVM_EXITINTINFO_TYPE_MASK;
1862 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
1864 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
1866 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
1868 tss_selector = (u16)svm->vmcb->control.exit_info_1;
1870 if (svm->vmcb->control.exit_info_2 &
1871 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1872 reason = TASK_SWITCH_IRET;
1873 else if (svm->vmcb->control.exit_info_2 &
1874 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1875 reason = TASK_SWITCH_JMP;
1877 reason = TASK_SWITCH_GATE;
1879 reason = TASK_SWITCH_CALL;
1881 if (reason == TASK_SWITCH_GATE) {
1883 case SVM_EXITINTINFO_TYPE_NMI:
1884 svm->vcpu.arch.nmi_injected = false;
1886 case SVM_EXITINTINFO_TYPE_EXEPT:
1887 kvm_clear_exception_queue(&svm->vcpu);
1889 case SVM_EXITINTINFO_TYPE_INTR:
1890 kvm_clear_interrupt_queue(&svm->vcpu);
1897 if (reason != TASK_SWITCH_GATE ||
1898 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
1899 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
1900 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
1901 skip_emulated_instruction(&svm->vcpu);
1903 return kvm_task_switch(&svm->vcpu, tss_selector, reason);
1906 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1908 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1909 kvm_emulate_cpuid(&svm->vcpu);
1913 static int iret_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1915 ++svm->vcpu.stat.nmi_window_exits;
1916 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
1917 svm->vcpu.arch.hflags |= HF_IRET_MASK;
1921 static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1923 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
1924 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1928 static int emulate_on_interception(struct vcpu_svm *svm,
1929 struct kvm_run *kvm_run)
1931 if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1932 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1936 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1938 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
1939 /* instruction emulation calls kvm_set_cr8() */
1940 emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1941 if (irqchip_in_kernel(svm->vcpu.kvm)) {
1942 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1945 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
1947 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1951 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1953 struct vcpu_svm *svm = to_svm(vcpu);
1956 case MSR_IA32_TIME_STAMP_COUNTER: {
1960 *data = svm->vmcb->control.tsc_offset + tsc;
1964 *data = svm->vmcb->save.star;
1966 #ifdef CONFIG_X86_64
1968 *data = svm->vmcb->save.lstar;
1971 *data = svm->vmcb->save.cstar;
1973 case MSR_KERNEL_GS_BASE:
1974 *data = svm->vmcb->save.kernel_gs_base;
1976 case MSR_SYSCALL_MASK:
1977 *data = svm->vmcb->save.sfmask;
1980 case MSR_IA32_SYSENTER_CS:
1981 *data = svm->vmcb->save.sysenter_cs;
1983 case MSR_IA32_SYSENTER_EIP:
1984 *data = svm->vmcb->save.sysenter_eip;
1986 case MSR_IA32_SYSENTER_ESP:
1987 *data = svm->vmcb->save.sysenter_esp;
1989 /* Nobody will change the following 5 values in the VMCB so
1990 we can safely return them on rdmsr. They will always be 0
1991 until LBRV is implemented. */
1992 case MSR_IA32_DEBUGCTLMSR:
1993 *data = svm->vmcb->save.dbgctl;
1995 case MSR_IA32_LASTBRANCHFROMIP:
1996 *data = svm->vmcb->save.br_from;
1998 case MSR_IA32_LASTBRANCHTOIP:
1999 *data = svm->vmcb->save.br_to;
2001 case MSR_IA32_LASTINTFROMIP:
2002 *data = svm->vmcb->save.last_excp_from;
2004 case MSR_IA32_LASTINTTOIP:
2005 *data = svm->vmcb->save.last_excp_to;
2007 case MSR_VM_HSAVE_PA:
2008 *data = svm->hsave_msr;
2013 case MSR_IA32_UCODE_REV:
2017 return kvm_get_msr_common(vcpu, ecx, data);
2022 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2024 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2027 if (svm_get_msr(&svm->vcpu, ecx, &data))
2028 kvm_inject_gp(&svm->vcpu, 0);
2030 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
2031 (u32)(data >> 32), handler);
2033 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2034 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2035 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2036 skip_emulated_instruction(&svm->vcpu);
2041 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2043 struct vcpu_svm *svm = to_svm(vcpu);
2046 case MSR_IA32_TIME_STAMP_COUNTER: {
2050 svm->vmcb->control.tsc_offset = data - tsc;
2054 svm->vmcb->save.star = data;
2056 #ifdef CONFIG_X86_64
2058 svm->vmcb->save.lstar = data;
2061 svm->vmcb->save.cstar = data;
2063 case MSR_KERNEL_GS_BASE:
2064 svm->vmcb->save.kernel_gs_base = data;
2066 case MSR_SYSCALL_MASK:
2067 svm->vmcb->save.sfmask = data;
2070 case MSR_IA32_SYSENTER_CS:
2071 svm->vmcb->save.sysenter_cs = data;
2073 case MSR_IA32_SYSENTER_EIP:
2074 svm->vmcb->save.sysenter_eip = data;
2076 case MSR_IA32_SYSENTER_ESP:
2077 svm->vmcb->save.sysenter_esp = data;
2079 case MSR_IA32_DEBUGCTLMSR:
2080 if (!svm_has(SVM_FEATURE_LBRV)) {
2081 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2085 if (data & DEBUGCTL_RESERVED_BITS)
2088 svm->vmcb->save.dbgctl = data;
2089 if (data & (1ULL<<0))
2090 svm_enable_lbrv(svm);
2092 svm_disable_lbrv(svm);
2094 case MSR_K7_EVNTSEL0:
2095 case MSR_K7_EVNTSEL1:
2096 case MSR_K7_EVNTSEL2:
2097 case MSR_K7_EVNTSEL3:
2098 case MSR_K7_PERFCTR0:
2099 case MSR_K7_PERFCTR1:
2100 case MSR_K7_PERFCTR2:
2101 case MSR_K7_PERFCTR3:
2103 * Just discard all writes to the performance counters; this
2104 * should keep both older linux and windows 64-bit guests
2107 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx, data);
2110 case MSR_VM_HSAVE_PA:
2111 svm->hsave_msr = data;
2114 return kvm_set_msr_common(vcpu, ecx, data);
2119 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2121 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2122 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2123 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2125 KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
2128 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2129 if (svm_set_msr(&svm->vcpu, ecx, data))
2130 kvm_inject_gp(&svm->vcpu, 0);
2132 skip_emulated_instruction(&svm->vcpu);
2136 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2138 if (svm->vmcb->control.exit_info_1)
2139 return wrmsr_interception(svm, kvm_run);
2141 return rdmsr_interception(svm, kvm_run);
2144 static int interrupt_window_interception(struct vcpu_svm *svm,
2145 struct kvm_run *kvm_run)
2147 KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
2149 svm_clear_vintr(svm);
2150 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2152 * If the user space waits to inject interrupts, exit as soon as
2155 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2156 kvm_run->request_interrupt_window &&
2157 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2158 ++svm->vcpu.stat.irq_window_exits;
2159 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2166 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
2167 struct kvm_run *kvm_run) = {
2168 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2169 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2170 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2171 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2173 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
2174 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2175 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2176 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2177 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2178 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2179 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2180 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2181 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2182 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2183 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2184 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2185 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2186 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2187 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2188 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2189 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2190 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2191 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2192 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2193 [SVM_EXIT_INTR] = intr_interception,
2194 [SVM_EXIT_NMI] = nmi_interception,
2195 [SVM_EXIT_SMI] = nop_on_interception,
2196 [SVM_EXIT_INIT] = nop_on_interception,
2197 [SVM_EXIT_VINTR] = interrupt_window_interception,
2198 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
2199 [SVM_EXIT_CPUID] = cpuid_interception,
2200 [SVM_EXIT_IRET] = iret_interception,
2201 [SVM_EXIT_INVD] = emulate_on_interception,
2202 [SVM_EXIT_HLT] = halt_interception,
2203 [SVM_EXIT_INVLPG] = invlpg_interception,
2204 [SVM_EXIT_INVLPGA] = invalid_op_interception,
2205 [SVM_EXIT_IOIO] = io_interception,
2206 [SVM_EXIT_MSR] = msr_interception,
2207 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2208 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2209 [SVM_EXIT_VMRUN] = vmrun_interception,
2210 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2211 [SVM_EXIT_VMLOAD] = vmload_interception,
2212 [SVM_EXIT_VMSAVE] = vmsave_interception,
2213 [SVM_EXIT_STGI] = stgi_interception,
2214 [SVM_EXIT_CLGI] = clgi_interception,
2215 [SVM_EXIT_SKINIT] = invalid_op_interception,
2216 [SVM_EXIT_WBINVD] = emulate_on_interception,
2217 [SVM_EXIT_MONITOR] = invalid_op_interception,
2218 [SVM_EXIT_MWAIT] = invalid_op_interception,
2219 [SVM_EXIT_NPF] = pf_interception,
2222 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2224 struct vcpu_svm *svm = to_svm(vcpu);
2225 u32 exit_code = svm->vmcb->control.exit_code;
2227 KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
2228 (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
2230 if (is_nested(svm)) {
2231 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2232 exit_code, svm->vmcb->control.exit_info_1,
2233 svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
2234 if (nested_svm_exit_handled(svm, true)) {
2235 nested_svm_vmexit(svm);
2236 nsvm_printk("-> #VMEXIT\n");
2243 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
2244 svm_set_cr0(vcpu, svm->vmcb->save.cr0);
2247 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2248 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2249 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2250 if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
2251 kvm_inject_gp(vcpu, 0);
2256 kvm_mmu_reset_context(vcpu);
2262 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2263 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2264 kvm_run->fail_entry.hardware_entry_failure_reason
2265 = svm->vmcb->control.exit_code;
2269 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2270 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2271 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2272 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2274 __func__, svm->vmcb->control.exit_int_info,
2277 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2278 || !svm_exit_handlers[exit_code]) {
2279 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2280 kvm_run->hw.hardware_exit_reason = exit_code;
2284 return svm_exit_handlers[exit_code](svm, kvm_run);
2287 static void reload_tss(struct kvm_vcpu *vcpu)
2289 int cpu = raw_smp_processor_id();
2291 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2292 svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
2296 static void pre_svm_run(struct vcpu_svm *svm)
2298 int cpu = raw_smp_processor_id();
2300 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2302 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2303 if (svm->vcpu.cpu != cpu ||
2304 svm->asid_generation != svm_data->asid_generation)
2305 new_asid(svm, svm_data);
2308 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2310 struct vcpu_svm *svm = to_svm(vcpu);
2312 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2313 vcpu->arch.hflags |= HF_NMI_MASK;
2314 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2315 ++vcpu->stat.nmi_injections;
2318 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2320 struct vmcb_control_area *control;
2322 KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
2324 ++svm->vcpu.stat.irq_injections;
2325 control = &svm->vmcb->control;
2326 control->int_vector = irq;
2327 control->int_ctl &= ~V_INTR_PRIO_MASK;
2328 control->int_ctl |= V_IRQ_MASK |
2329 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2332 static void svm_queue_irq(struct kvm_vcpu *vcpu, unsigned nr)
2334 struct vcpu_svm *svm = to_svm(vcpu);
2336 svm->vmcb->control.event_inj = nr |
2337 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2340 static void svm_set_irq(struct kvm_vcpu *vcpu)
2342 struct vcpu_svm *svm = to_svm(vcpu);
2344 nested_svm_intr(svm);
2346 svm_queue_irq(vcpu, vcpu->arch.interrupt.nr);
2349 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2351 struct vcpu_svm *svm = to_svm(vcpu);
2357 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2360 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2362 struct vcpu_svm *svm = to_svm(vcpu);
2363 struct vmcb *vmcb = svm->vmcb;
2364 return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2365 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2368 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2370 struct vcpu_svm *svm = to_svm(vcpu);
2371 struct vmcb *vmcb = svm->vmcb;
2372 return (vmcb->save.rflags & X86_EFLAGS_IF) &&
2373 !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2374 (svm->vcpu.arch.hflags & HF_GIF_MASK);
2377 static void enable_irq_window(struct kvm_vcpu *vcpu)
2379 svm_set_vintr(to_svm(vcpu));
2380 svm_inject_irq(to_svm(vcpu), 0x0);
2383 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2385 struct vcpu_svm *svm = to_svm(vcpu);
2387 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2389 return; /* IRET will cause a vm exit */
2391 /* Something prevents NMI from been injected. Single step over
2392 possible problem (IRET or exception injection or interrupt
2394 vcpu->arch.singlestep = true;
2395 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2396 update_db_intercept(vcpu);
2399 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2404 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2406 force_new_asid(vcpu);
2409 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2413 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2415 struct vcpu_svm *svm = to_svm(vcpu);
2417 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2418 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2419 kvm_set_cr8(vcpu, cr8);
2423 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2425 struct vcpu_svm *svm = to_svm(vcpu);
2428 cr8 = kvm_get_cr8(vcpu);
2429 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2430 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2433 static void svm_complete_interrupts(struct vcpu_svm *svm)
2437 u32 exitintinfo = svm->vmcb->control.exit_int_info;
2439 if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2440 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2442 svm->vcpu.arch.nmi_injected = false;
2443 kvm_clear_exception_queue(&svm->vcpu);
2444 kvm_clear_interrupt_queue(&svm->vcpu);
2446 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2449 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2450 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2453 case SVM_EXITINTINFO_TYPE_NMI:
2454 svm->vcpu.arch.nmi_injected = true;
2456 case SVM_EXITINTINFO_TYPE_EXEPT:
2457 /* In case of software exception do not reinject an exception
2458 vector, but re-execute and instruction instead */
2459 if (kvm_exception_is_soft(vector))
2461 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2462 u32 err = svm->vmcb->control.exit_int_info_err;
2463 kvm_queue_exception_e(&svm->vcpu, vector, err);
2466 kvm_queue_exception(&svm->vcpu, vector);
2468 case SVM_EXITINTINFO_TYPE_INTR:
2469 kvm_queue_interrupt(&svm->vcpu, vector, false);
2476 #ifdef CONFIG_X86_64
2482 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2484 struct vcpu_svm *svm = to_svm(vcpu);
2489 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2490 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2491 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2495 sync_lapic_to_cr8(vcpu);
2497 save_host_msrs(vcpu);
2498 fs_selector = kvm_read_fs();
2499 gs_selector = kvm_read_gs();
2500 ldt_selector = kvm_read_ldt();
2501 svm->host_cr2 = kvm_read_cr2();
2502 if (!is_nested(svm))
2503 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2504 /* required for live migration with NPT */
2506 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2513 "push %%"R"bp; \n\t"
2514 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2515 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2516 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2517 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2518 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2519 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2520 #ifdef CONFIG_X86_64
2521 "mov %c[r8](%[svm]), %%r8 \n\t"
2522 "mov %c[r9](%[svm]), %%r9 \n\t"
2523 "mov %c[r10](%[svm]), %%r10 \n\t"
2524 "mov %c[r11](%[svm]), %%r11 \n\t"
2525 "mov %c[r12](%[svm]), %%r12 \n\t"
2526 "mov %c[r13](%[svm]), %%r13 \n\t"
2527 "mov %c[r14](%[svm]), %%r14 \n\t"
2528 "mov %c[r15](%[svm]), %%r15 \n\t"
2531 /* Enter guest mode */
2533 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2534 __ex(SVM_VMLOAD) "\n\t"
2535 __ex(SVM_VMRUN) "\n\t"
2536 __ex(SVM_VMSAVE) "\n\t"
2539 /* Save guest registers, load host registers */
2540 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2541 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2542 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2543 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2544 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2545 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2546 #ifdef CONFIG_X86_64
2547 "mov %%r8, %c[r8](%[svm]) \n\t"
2548 "mov %%r9, %c[r9](%[svm]) \n\t"
2549 "mov %%r10, %c[r10](%[svm]) \n\t"
2550 "mov %%r11, %c[r11](%[svm]) \n\t"
2551 "mov %%r12, %c[r12](%[svm]) \n\t"
2552 "mov %%r13, %c[r13](%[svm]) \n\t"
2553 "mov %%r14, %c[r14](%[svm]) \n\t"
2554 "mov %%r15, %c[r15](%[svm]) \n\t"
2559 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2560 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2561 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2562 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2563 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2564 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2565 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2566 #ifdef CONFIG_X86_64
2567 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2568 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2569 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2570 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2571 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2572 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2573 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2574 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2577 , R"bx", R"cx", R"dx", R"si", R"di"
2578 #ifdef CONFIG_X86_64
2579 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2583 vcpu->arch.cr2 = svm->vmcb->save.cr2;
2584 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2585 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2586 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2588 kvm_write_cr2(svm->host_cr2);
2590 kvm_load_fs(fs_selector);
2591 kvm_load_gs(gs_selector);
2592 kvm_load_ldt(ldt_selector);
2593 load_host_msrs(vcpu);
2597 local_irq_disable();
2601 sync_cr8_to_lapic(vcpu);
2605 svm_complete_interrupts(svm);
2610 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2612 struct vcpu_svm *svm = to_svm(vcpu);
2615 svm->vmcb->control.nested_cr3 = root;
2616 force_new_asid(vcpu);
2620 svm->vmcb->save.cr3 = root;
2621 force_new_asid(vcpu);
2623 if (vcpu->fpu_active) {
2624 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
2625 svm->vmcb->save.cr0 |= X86_CR0_TS;
2626 vcpu->fpu_active = 0;
2630 static int is_disabled(void)
2634 rdmsrl(MSR_VM_CR, vm_cr);
2635 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2642 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2645 * Patch in the VMMCALL instruction:
2647 hypercall[0] = 0x0f;
2648 hypercall[1] = 0x01;
2649 hypercall[2] = 0xd9;
2652 static void svm_check_processor_compat(void *rtn)
2657 static bool svm_cpu_has_accelerated_tpr(void)
2662 static int get_npt_level(void)
2664 #ifdef CONFIG_X86_64
2665 return PT64_ROOT_LEVEL;
2667 return PT32E_ROOT_LEVEL;
2671 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
2676 static struct kvm_x86_ops svm_x86_ops = {
2677 .cpu_has_kvm_support = has_svm,
2678 .disabled_by_bios = is_disabled,
2679 .hardware_setup = svm_hardware_setup,
2680 .hardware_unsetup = svm_hardware_unsetup,
2681 .check_processor_compatibility = svm_check_processor_compat,
2682 .hardware_enable = svm_hardware_enable,
2683 .hardware_disable = svm_hardware_disable,
2684 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2686 .vcpu_create = svm_create_vcpu,
2687 .vcpu_free = svm_free_vcpu,
2688 .vcpu_reset = svm_vcpu_reset,
2690 .prepare_guest_switch = svm_prepare_guest_switch,
2691 .vcpu_load = svm_vcpu_load,
2692 .vcpu_put = svm_vcpu_put,
2694 .set_guest_debug = svm_guest_debug,
2695 .get_msr = svm_get_msr,
2696 .set_msr = svm_set_msr,
2697 .get_segment_base = svm_get_segment_base,
2698 .get_segment = svm_get_segment,
2699 .set_segment = svm_set_segment,
2700 .get_cpl = svm_get_cpl,
2701 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2702 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2703 .set_cr0 = svm_set_cr0,
2704 .set_cr3 = svm_set_cr3,
2705 .set_cr4 = svm_set_cr4,
2706 .set_efer = svm_set_efer,
2707 .get_idt = svm_get_idt,
2708 .set_idt = svm_set_idt,
2709 .get_gdt = svm_get_gdt,
2710 .set_gdt = svm_set_gdt,
2711 .get_dr = svm_get_dr,
2712 .set_dr = svm_set_dr,
2713 .get_rflags = svm_get_rflags,
2714 .set_rflags = svm_set_rflags,
2716 .tlb_flush = svm_flush_tlb,
2718 .run = svm_vcpu_run,
2719 .handle_exit = handle_exit,
2720 .skip_emulated_instruction = skip_emulated_instruction,
2721 .set_interrupt_shadow = svm_set_interrupt_shadow,
2722 .get_interrupt_shadow = svm_get_interrupt_shadow,
2723 .patch_hypercall = svm_patch_hypercall,
2724 .set_irq = svm_set_irq,
2725 .set_nmi = svm_inject_nmi,
2726 .queue_exception = svm_queue_exception,
2727 .interrupt_allowed = svm_interrupt_allowed,
2728 .nmi_allowed = svm_nmi_allowed,
2729 .enable_nmi_window = enable_nmi_window,
2730 .enable_irq_window = enable_irq_window,
2731 .update_cr8_intercept = update_cr8_intercept,
2733 .set_tss_addr = svm_set_tss_addr,
2734 .get_tdp_level = get_npt_level,
2735 .get_mt_mask = svm_get_mt_mask,
2738 static int __init svm_init(void)
2740 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
2744 static void __exit svm_exit(void)
2749 module_init(svm_init)
2750 module_exit(svm_exit)