2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
18 #define pr_fmt(fmt) "SVM: " fmt
20 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
29 #include <linux/module.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/kernel.h>
32 #include <linux/vmalloc.h>
33 #include <linux/highmem.h>
34 #include <linux/sched.h>
35 #include <linux/trace_events.h>
36 #include <linux/slab.h>
37 #include <linux/amd-iommu.h>
38 #include <linux/hashtable.h>
41 #include <asm/perf_event.h>
42 #include <asm/tlbflush.h>
44 #include <asm/debugreg.h>
45 #include <asm/kvm_para.h>
46 #include <asm/irq_remapping.h>
48 #include <asm/virtext.h>
51 #define __ex(x) __kvm_handle_fault_on_reboot(x)
53 MODULE_AUTHOR("Qumranet");
54 MODULE_LICENSE("GPL");
56 static const struct x86_cpu_id svm_cpu_id[] = {
57 X86_FEATURE_MATCH(X86_FEATURE_SVM),
60 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
62 #define IOPM_ALLOC_ORDER 2
63 #define MSRPM_ALLOC_ORDER 1
65 #define SEG_TYPE_LDT 2
66 #define SEG_TYPE_BUSY_TSS16 3
68 #define SVM_FEATURE_NPT (1 << 0)
69 #define SVM_FEATURE_LBRV (1 << 1)
70 #define SVM_FEATURE_SVML (1 << 2)
71 #define SVM_FEATURE_NRIP (1 << 3)
72 #define SVM_FEATURE_TSC_RATE (1 << 4)
73 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
74 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
75 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
76 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
78 #define SVM_AVIC_DOORBELL 0xc001011b
80 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
81 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
82 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
84 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
86 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
87 #define TSC_RATIO_MIN 0x0000000000000001ULL
88 #define TSC_RATIO_MAX 0x000000ffffffffffULL
90 #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF)
93 * 0xff is broadcast, so the max index allowed for physical APIC ID
94 * table is 0xfe. APIC IDs above 0xff are reserved.
96 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
98 #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1
99 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0
100 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF
102 /* AVIC GATAG is encoded using VM and VCPU IDs */
103 #define AVIC_VCPU_ID_BITS 8
104 #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1)
106 #define AVIC_VM_ID_BITS 24
107 #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS)
108 #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1)
110 #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
111 (y & AVIC_VCPU_ID_MASK))
112 #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
113 #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK)
115 static bool erratum_383_found __read_mostly;
117 static const u32 host_save_user_msrs[] = {
119 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
122 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
126 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
130 struct nested_state {
136 /* These are the merged vectors */
139 /* gpa pointers to the real vectors */
143 /* A VMEXIT is required but not yet emulated */
146 /* cache for intercepts of the guest */
149 u32 intercept_exceptions;
152 /* Nested Paging related state */
156 #define MSRPM_OFFSETS 16
157 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
160 * Set osvw_len to higher value when updated Revision Guides
161 * are published and we know what the new status bits are
163 static uint64_t osvw_len = 4, osvw_status;
166 struct kvm_vcpu vcpu;
168 unsigned long vmcb_pa;
169 struct svm_cpu_data *svm_data;
170 uint64_t asid_generation;
171 uint64_t sysenter_esp;
172 uint64_t sysenter_eip;
177 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
189 struct nested_state nested;
193 unsigned int3_injected;
194 unsigned long int3_rip;
197 /* cached guest cpuid flags for faster access */
198 bool nrips_enabled : 1;
201 struct page *avic_backing_page;
202 u64 *avic_physical_id_cache;
203 bool avic_is_running;
206 * Per-vcpu list of struct amd_svm_iommu_ir:
207 * This is used mainly to store interrupt remapping information used
208 * when update the vcpu affinity. This avoids the need to scan for
209 * IRTE and try to match ga_tag in the IOMMU driver.
211 struct list_head ir_list;
212 spinlock_t ir_list_lock;
216 * This is a wrapper of struct amd_iommu_ir_data.
218 struct amd_svm_iommu_ir {
219 struct list_head node; /* Used by SVM for per-vcpu ir_list */
220 void *data; /* Storing pointer to struct amd_ir_data */
223 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
224 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
226 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
227 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
228 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
229 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
231 static DEFINE_PER_CPU(u64, current_tsc_ratio);
232 #define TSC_RATIO_DEFAULT 0x0100000000ULL
234 #define MSR_INVALID 0xffffffffU
236 static const struct svm_direct_access_msrs {
237 u32 index; /* Index of the MSR */
238 bool always; /* True if intercept is always on */
239 } direct_access_msrs[] = {
240 { .index = MSR_STAR, .always = true },
241 { .index = MSR_IA32_SYSENTER_CS, .always = true },
243 { .index = MSR_GS_BASE, .always = true },
244 { .index = MSR_FS_BASE, .always = true },
245 { .index = MSR_KERNEL_GS_BASE, .always = true },
246 { .index = MSR_LSTAR, .always = true },
247 { .index = MSR_CSTAR, .always = true },
248 { .index = MSR_SYSCALL_MASK, .always = true },
250 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
251 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
252 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
253 { .index = MSR_IA32_LASTINTTOIP, .always = false },
254 { .index = MSR_INVALID, .always = false },
257 /* enable NPT for AMD64 and X86 with PAE */
258 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
259 static bool npt_enabled = true;
261 static bool npt_enabled;
264 /* allow nested paging (virtualized MMU) for all guests */
265 static int npt = true;
266 module_param(npt, int, S_IRUGO);
268 /* allow nested virtualization in KVM/SVM */
269 static int nested = true;
270 module_param(nested, int, S_IRUGO);
272 /* enable / disable AVIC */
274 #ifdef CONFIG_X86_LOCAL_APIC
275 module_param(avic, int, S_IRUGO);
278 /* AVIC VM ID bit masks and lock */
279 static DECLARE_BITMAP(avic_vm_id_bitmap, AVIC_VM_ID_NR);
280 static DEFINE_SPINLOCK(avic_vm_id_lock);
282 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
283 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
284 static void svm_complete_interrupts(struct vcpu_svm *svm);
286 static int nested_svm_exit_handled(struct vcpu_svm *svm);
287 static int nested_svm_intercept(struct vcpu_svm *svm);
288 static int nested_svm_vmexit(struct vcpu_svm *svm);
289 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
290 bool has_error_code, u32 error_code);
293 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
294 pause filter count */
295 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
296 VMCB_ASID, /* ASID */
297 VMCB_INTR, /* int_ctl, int_vector */
298 VMCB_NPT, /* npt_en, nCR3, gPAT */
299 VMCB_CR, /* CR0, CR3, CR4, EFER */
300 VMCB_DR, /* DR6, DR7 */
301 VMCB_DT, /* GDT, IDT */
302 VMCB_SEG, /* CS, DS, SS, ES, CPL */
303 VMCB_CR2, /* CR2 only */
304 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
305 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
306 * AVIC PHYSICAL_TABLE pointer,
307 * AVIC LOGICAL_TABLE pointer
312 /* TPR and CR2 are always written before VMRUN */
313 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
315 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
317 static inline void mark_all_dirty(struct vmcb *vmcb)
319 vmcb->control.clean = 0;
322 static inline void mark_all_clean(struct vmcb *vmcb)
324 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
325 & ~VMCB_ALWAYS_DIRTY_MASK;
328 static inline void mark_dirty(struct vmcb *vmcb, int bit)
330 vmcb->control.clean &= ~(1 << bit);
333 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
335 return container_of(vcpu, struct vcpu_svm, vcpu);
338 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
340 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
341 mark_dirty(svm->vmcb, VMCB_AVIC);
344 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
346 struct vcpu_svm *svm = to_svm(vcpu);
347 u64 *entry = svm->avic_physical_id_cache;
352 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
355 static void recalc_intercepts(struct vcpu_svm *svm)
357 struct vmcb_control_area *c, *h;
358 struct nested_state *g;
360 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
362 if (!is_guest_mode(&svm->vcpu))
365 c = &svm->vmcb->control;
366 h = &svm->nested.hsave->control;
369 c->intercept_cr = h->intercept_cr | g->intercept_cr;
370 c->intercept_dr = h->intercept_dr | g->intercept_dr;
371 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
372 c->intercept = h->intercept | g->intercept;
375 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
377 if (is_guest_mode(&svm->vcpu))
378 return svm->nested.hsave;
383 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
385 struct vmcb *vmcb = get_host_vmcb(svm);
387 vmcb->control.intercept_cr |= (1U << bit);
389 recalc_intercepts(svm);
392 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
394 struct vmcb *vmcb = get_host_vmcb(svm);
396 vmcb->control.intercept_cr &= ~(1U << bit);
398 recalc_intercepts(svm);
401 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
403 struct vmcb *vmcb = get_host_vmcb(svm);
405 return vmcb->control.intercept_cr & (1U << bit);
408 static inline void set_dr_intercepts(struct vcpu_svm *svm)
410 struct vmcb *vmcb = get_host_vmcb(svm);
412 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
413 | (1 << INTERCEPT_DR1_READ)
414 | (1 << INTERCEPT_DR2_READ)
415 | (1 << INTERCEPT_DR3_READ)
416 | (1 << INTERCEPT_DR4_READ)
417 | (1 << INTERCEPT_DR5_READ)
418 | (1 << INTERCEPT_DR6_READ)
419 | (1 << INTERCEPT_DR7_READ)
420 | (1 << INTERCEPT_DR0_WRITE)
421 | (1 << INTERCEPT_DR1_WRITE)
422 | (1 << INTERCEPT_DR2_WRITE)
423 | (1 << INTERCEPT_DR3_WRITE)
424 | (1 << INTERCEPT_DR4_WRITE)
425 | (1 << INTERCEPT_DR5_WRITE)
426 | (1 << INTERCEPT_DR6_WRITE)
427 | (1 << INTERCEPT_DR7_WRITE);
429 recalc_intercepts(svm);
432 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
434 struct vmcb *vmcb = get_host_vmcb(svm);
436 vmcb->control.intercept_dr = 0;
438 recalc_intercepts(svm);
441 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
443 struct vmcb *vmcb = get_host_vmcb(svm);
445 vmcb->control.intercept_exceptions |= (1U << bit);
447 recalc_intercepts(svm);
450 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
452 struct vmcb *vmcb = get_host_vmcb(svm);
454 vmcb->control.intercept_exceptions &= ~(1U << bit);
456 recalc_intercepts(svm);
459 static inline void set_intercept(struct vcpu_svm *svm, int bit)
461 struct vmcb *vmcb = get_host_vmcb(svm);
463 vmcb->control.intercept |= (1ULL << bit);
465 recalc_intercepts(svm);
468 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
470 struct vmcb *vmcb = get_host_vmcb(svm);
472 vmcb->control.intercept &= ~(1ULL << bit);
474 recalc_intercepts(svm);
477 static inline void enable_gif(struct vcpu_svm *svm)
479 svm->vcpu.arch.hflags |= HF_GIF_MASK;
482 static inline void disable_gif(struct vcpu_svm *svm)
484 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
487 static inline bool gif_set(struct vcpu_svm *svm)
489 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
492 static unsigned long iopm_base;
494 struct kvm_ldttss_desc {
497 unsigned base1:8, type:5, dpl:2, p:1;
498 unsigned limit1:4, zero0:3, g:1, base2:8;
501 } __attribute__((packed));
503 struct svm_cpu_data {
509 struct kvm_ldttss_desc *tss_desc;
511 struct page *save_area;
514 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
516 struct svm_init_data {
521 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
523 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
524 #define MSRS_RANGE_SIZE 2048
525 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
527 static u32 svm_msrpm_offset(u32 msr)
532 for (i = 0; i < NUM_MSR_MAPS; i++) {
533 if (msr < msrpm_ranges[i] ||
534 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
537 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
538 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
540 /* Now we have the u8 offset - but need the u32 offset */
544 /* MSR not in any range */
548 #define MAX_INST_SIZE 15
550 static inline void clgi(void)
552 asm volatile (__ex(SVM_CLGI));
555 static inline void stgi(void)
557 asm volatile (__ex(SVM_STGI));
560 static inline void invlpga(unsigned long addr, u32 asid)
562 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
565 static int get_npt_level(void)
568 return PT64_ROOT_LEVEL;
570 return PT32E_ROOT_LEVEL;
574 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
576 vcpu->arch.efer = efer;
577 if (!npt_enabled && !(efer & EFER_LMA))
580 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
581 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
584 static int is_external_interrupt(u32 info)
586 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
587 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
590 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
592 struct vcpu_svm *svm = to_svm(vcpu);
595 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
596 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
600 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
602 struct vcpu_svm *svm = to_svm(vcpu);
605 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
607 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
611 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
613 struct vcpu_svm *svm = to_svm(vcpu);
615 if (svm->vmcb->control.next_rip != 0) {
616 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
617 svm->next_rip = svm->vmcb->control.next_rip;
620 if (!svm->next_rip) {
621 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
623 printk(KERN_DEBUG "%s: NOP\n", __func__);
626 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
627 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
628 __func__, kvm_rip_read(vcpu), svm->next_rip);
630 kvm_rip_write(vcpu, svm->next_rip);
631 svm_set_interrupt_shadow(vcpu, 0);
634 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
635 bool has_error_code, u32 error_code,
638 struct vcpu_svm *svm = to_svm(vcpu);
641 * If we are within a nested VM we'd better #VMEXIT and let the guest
642 * handle the exception
645 nested_svm_check_exception(svm, nr, has_error_code, error_code))
648 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
649 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
652 * For guest debugging where we have to reinject #BP if some
653 * INT3 is guest-owned:
654 * Emulate nRIP by moving RIP forward. Will fail if injection
655 * raises a fault that is not intercepted. Still better than
656 * failing in all cases.
658 skip_emulated_instruction(&svm->vcpu);
659 rip = kvm_rip_read(&svm->vcpu);
660 svm->int3_rip = rip + svm->vmcb->save.cs.base;
661 svm->int3_injected = rip - old_rip;
664 svm->vmcb->control.event_inj = nr
666 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
667 | SVM_EVTINJ_TYPE_EXEPT;
668 svm->vmcb->control.event_inj_err = error_code;
671 static void svm_init_erratum_383(void)
677 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
680 /* Use _safe variants to not break nested virtualization */
681 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
687 low = lower_32_bits(val);
688 high = upper_32_bits(val);
690 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
692 erratum_383_found = true;
695 static void svm_init_osvw(struct kvm_vcpu *vcpu)
698 * Guests should see errata 400 and 415 as fixed (assuming that
699 * HLT and IO instructions are intercepted).
701 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
702 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
705 * By increasing VCPU's osvw.length to 3 we are telling the guest that
706 * all osvw.status bits inside that length, including bit 0 (which is
707 * reserved for erratum 298), are valid. However, if host processor's
708 * osvw_len is 0 then osvw_status[0] carries no information. We need to
709 * be conservative here and therefore we tell the guest that erratum 298
710 * is present (because we really don't know).
712 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
713 vcpu->arch.osvw.status |= 1;
716 static int has_svm(void)
720 if (!cpu_has_svm(&msg)) {
721 printk(KERN_INFO "has_svm: %s\n", msg);
728 static void svm_hardware_disable(void)
730 /* Make sure we clean up behind us */
731 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
732 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
736 amd_pmu_disable_virt();
739 static int svm_hardware_enable(void)
742 struct svm_cpu_data *sd;
744 struct desc_ptr gdt_descr;
745 struct desc_struct *gdt;
746 int me = raw_smp_processor_id();
748 rdmsrl(MSR_EFER, efer);
749 if (efer & EFER_SVME)
753 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
756 sd = per_cpu(svm_data, me);
758 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
762 sd->asid_generation = 1;
763 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
764 sd->next_asid = sd->max_asid + 1;
766 native_store_gdt(&gdt_descr);
767 gdt = (struct desc_struct *)gdt_descr.address;
768 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
770 wrmsrl(MSR_EFER, efer | EFER_SVME);
772 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
774 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
775 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
776 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
783 * Note that it is possible to have a system with mixed processor
784 * revisions and therefore different OSVW bits. If bits are not the same
785 * on different processors then choose the worst case (i.e. if erratum
786 * is present on one processor and not on another then assume that the
787 * erratum is present everywhere).
789 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
790 uint64_t len, status = 0;
793 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
795 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
799 osvw_status = osvw_len = 0;
803 osvw_status |= status;
804 osvw_status &= (1ULL << osvw_len) - 1;
807 osvw_status = osvw_len = 0;
809 svm_init_erratum_383();
811 amd_pmu_enable_virt();
816 static void svm_cpu_uninit(int cpu)
818 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
823 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
824 __free_page(sd->save_area);
828 static int svm_cpu_init(int cpu)
830 struct svm_cpu_data *sd;
833 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
837 sd->save_area = alloc_page(GFP_KERNEL);
842 per_cpu(svm_data, cpu) = sd;
852 static bool valid_msr_intercept(u32 index)
856 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
857 if (direct_access_msrs[i].index == index)
863 static void set_msr_interception(u32 *msrpm, unsigned msr,
866 u8 bit_read, bit_write;
871 * If this warning triggers extend the direct_access_msrs list at the
872 * beginning of the file
874 WARN_ON(!valid_msr_intercept(msr));
876 offset = svm_msrpm_offset(msr);
877 bit_read = 2 * (msr & 0x0f);
878 bit_write = 2 * (msr & 0x0f) + 1;
881 BUG_ON(offset == MSR_INVALID);
883 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
884 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
889 static void svm_vcpu_init_msrpm(u32 *msrpm)
893 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
895 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
896 if (!direct_access_msrs[i].always)
899 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
903 static void add_msr_offset(u32 offset)
907 for (i = 0; i < MSRPM_OFFSETS; ++i) {
909 /* Offset already in list? */
910 if (msrpm_offsets[i] == offset)
913 /* Slot used by another offset? */
914 if (msrpm_offsets[i] != MSR_INVALID)
917 /* Add offset to list */
918 msrpm_offsets[i] = offset;
924 * If this BUG triggers the msrpm_offsets table has an overflow. Just
925 * increase MSRPM_OFFSETS in this case.
930 static void init_msrpm_offsets(void)
934 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
936 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
939 offset = svm_msrpm_offset(direct_access_msrs[i].index);
940 BUG_ON(offset == MSR_INVALID);
942 add_msr_offset(offset);
946 static void svm_enable_lbrv(struct vcpu_svm *svm)
948 u32 *msrpm = svm->msrpm;
950 svm->vmcb->control.lbr_ctl = 1;
951 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
952 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
953 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
954 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
957 static void svm_disable_lbrv(struct vcpu_svm *svm)
959 u32 *msrpm = svm->msrpm;
961 svm->vmcb->control.lbr_ctl = 0;
962 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
963 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
964 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
965 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
969 * This hash table is used to map VM_ID to a struct kvm_arch,
970 * when handling AMD IOMMU GALOG notification to schedule in
973 #define SVM_VM_DATA_HASH_BITS 8
974 DECLARE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
975 static spinlock_t svm_vm_data_hash_lock;
978 * This function is called from IOMMU driver to notify
979 * SVM to schedule in a particular vCPU of a particular VM.
981 static int avic_ga_log_notifier(u32 ga_tag)
984 struct kvm_arch *ka = NULL;
985 struct kvm_vcpu *vcpu = NULL;
986 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
987 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
989 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
991 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
992 hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) {
993 struct kvm *kvm = container_of(ka, struct kvm, arch);
994 struct kvm_arch *vm_data = &kvm->arch;
996 if (vm_data->avic_vm_id != vm_id)
998 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1001 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1007 * At this point, the IOMMU should have already set the pending
1008 * bit in the vAPIC backing page. So, we just need to schedule
1011 if (vcpu->mode == OUTSIDE_GUEST_MODE)
1012 kvm_vcpu_wake_up(vcpu);
1017 static __init int svm_hardware_setup(void)
1020 struct page *iopm_pages;
1024 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1029 iopm_va = page_address(iopm_pages);
1030 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1031 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1033 init_msrpm_offsets();
1035 if (boot_cpu_has(X86_FEATURE_NX))
1036 kvm_enable_efer_bits(EFER_NX);
1038 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1039 kvm_enable_efer_bits(EFER_FFXSR);
1041 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1042 kvm_has_tsc_control = true;
1043 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1044 kvm_tsc_scaling_ratio_frac_bits = 32;
1048 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1049 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1052 for_each_possible_cpu(cpu) {
1053 r = svm_cpu_init(cpu);
1058 if (!boot_cpu_has(X86_FEATURE_NPT))
1059 npt_enabled = false;
1061 if (npt_enabled && !npt) {
1062 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1063 npt_enabled = false;
1067 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1074 !boot_cpu_has(X86_FEATURE_AVIC) ||
1075 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1078 pr_info("AVIC enabled\n");
1080 hash_init(svm_vm_data_hash);
1081 spin_lock_init(&svm_vm_data_hash_lock);
1082 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1089 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1094 static __exit void svm_hardware_unsetup(void)
1098 for_each_possible_cpu(cpu)
1099 svm_cpu_uninit(cpu);
1101 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1105 static void init_seg(struct vmcb_seg *seg)
1108 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1109 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1110 seg->limit = 0xffff;
1114 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1117 seg->attrib = SVM_SELECTOR_P_MASK | type;
1118 seg->limit = 0xffff;
1122 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1124 struct vcpu_svm *svm = to_svm(vcpu);
1126 return svm->vmcb->control.tsc_offset;
1129 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1131 struct vcpu_svm *svm = to_svm(vcpu);
1132 u64 g_tsc_offset = 0;
1134 if (is_guest_mode(vcpu)) {
1135 g_tsc_offset = svm->vmcb->control.tsc_offset -
1136 svm->nested.hsave->control.tsc_offset;
1137 svm->nested.hsave->control.tsc_offset = offset;
1139 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1140 svm->vmcb->control.tsc_offset,
1143 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1145 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1148 static void svm_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
1150 struct vcpu_svm *svm = to_svm(vcpu);
1152 svm->vmcb->control.tsc_offset += adjustment;
1153 if (is_guest_mode(vcpu))
1154 svm->nested.hsave->control.tsc_offset += adjustment;
1156 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1157 svm->vmcb->control.tsc_offset - adjustment,
1158 svm->vmcb->control.tsc_offset);
1160 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1163 static void avic_init_vmcb(struct vcpu_svm *svm)
1165 struct vmcb *vmcb = svm->vmcb;
1166 struct kvm_arch *vm_data = &svm->vcpu.kvm->arch;
1167 phys_addr_t bpa = page_to_phys(svm->avic_backing_page);
1168 phys_addr_t lpa = page_to_phys(vm_data->avic_logical_id_table_page);
1169 phys_addr_t ppa = page_to_phys(vm_data->avic_physical_id_table_page);
1171 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1172 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1173 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1174 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1175 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1176 svm->vcpu.arch.apicv_active = true;
1179 static void init_vmcb(struct vcpu_svm *svm)
1181 struct vmcb_control_area *control = &svm->vmcb->control;
1182 struct vmcb_save_area *save = &svm->vmcb->save;
1184 svm->vcpu.fpu_active = 1;
1185 svm->vcpu.arch.hflags = 0;
1187 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1188 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1189 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1190 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1191 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1192 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1193 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1194 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1196 set_dr_intercepts(svm);
1198 set_exception_intercept(svm, PF_VECTOR);
1199 set_exception_intercept(svm, UD_VECTOR);
1200 set_exception_intercept(svm, MC_VECTOR);
1201 set_exception_intercept(svm, AC_VECTOR);
1202 set_exception_intercept(svm, DB_VECTOR);
1204 set_intercept(svm, INTERCEPT_INTR);
1205 set_intercept(svm, INTERCEPT_NMI);
1206 set_intercept(svm, INTERCEPT_SMI);
1207 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1208 set_intercept(svm, INTERCEPT_RDPMC);
1209 set_intercept(svm, INTERCEPT_CPUID);
1210 set_intercept(svm, INTERCEPT_INVD);
1211 set_intercept(svm, INTERCEPT_HLT);
1212 set_intercept(svm, INTERCEPT_INVLPG);
1213 set_intercept(svm, INTERCEPT_INVLPGA);
1214 set_intercept(svm, INTERCEPT_IOIO_PROT);
1215 set_intercept(svm, INTERCEPT_MSR_PROT);
1216 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1217 set_intercept(svm, INTERCEPT_SHUTDOWN);
1218 set_intercept(svm, INTERCEPT_VMRUN);
1219 set_intercept(svm, INTERCEPT_VMMCALL);
1220 set_intercept(svm, INTERCEPT_VMLOAD);
1221 set_intercept(svm, INTERCEPT_VMSAVE);
1222 set_intercept(svm, INTERCEPT_STGI);
1223 set_intercept(svm, INTERCEPT_CLGI);
1224 set_intercept(svm, INTERCEPT_SKINIT);
1225 set_intercept(svm, INTERCEPT_WBINVD);
1226 set_intercept(svm, INTERCEPT_MONITOR);
1227 set_intercept(svm, INTERCEPT_MWAIT);
1228 set_intercept(svm, INTERCEPT_XSETBV);
1230 control->iopm_base_pa = iopm_base;
1231 control->msrpm_base_pa = __pa(svm->msrpm);
1232 control->int_ctl = V_INTR_MASKING_MASK;
1234 init_seg(&save->es);
1235 init_seg(&save->ss);
1236 init_seg(&save->ds);
1237 init_seg(&save->fs);
1238 init_seg(&save->gs);
1240 save->cs.selector = 0xf000;
1241 save->cs.base = 0xffff0000;
1242 /* Executable/Readable Code Segment */
1243 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1244 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1245 save->cs.limit = 0xffff;
1247 save->gdtr.limit = 0xffff;
1248 save->idtr.limit = 0xffff;
1250 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1251 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1253 svm_set_efer(&svm->vcpu, 0);
1254 save->dr6 = 0xffff0ff0;
1255 kvm_set_rflags(&svm->vcpu, 2);
1256 save->rip = 0x0000fff0;
1257 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1260 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1261 * It also updates the guest-visible cr0 value.
1263 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1264 kvm_mmu_reset_context(&svm->vcpu);
1266 save->cr4 = X86_CR4_PAE;
1270 /* Setup VMCB for Nested Paging */
1271 control->nested_ctl = 1;
1272 clr_intercept(svm, INTERCEPT_INVLPG);
1273 clr_exception_intercept(svm, PF_VECTOR);
1274 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1275 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1276 save->g_pat = svm->vcpu.arch.pat;
1280 svm->asid_generation = 0;
1282 svm->nested.vmcb = 0;
1283 svm->vcpu.arch.hflags = 0;
1285 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1286 control->pause_filter_count = 3000;
1287 set_intercept(svm, INTERCEPT_PAUSE);
1291 avic_init_vmcb(svm);
1293 mark_all_dirty(svm->vmcb);
1299 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu, int index)
1301 u64 *avic_physical_id_table;
1302 struct kvm_arch *vm_data = &vcpu->kvm->arch;
1304 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1307 avic_physical_id_table = page_address(vm_data->avic_physical_id_table_page);
1309 return &avic_physical_id_table[index];
1314 * AVIC hardware walks the nested page table to check permissions,
1315 * but does not use the SPA address specified in the leaf page
1316 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1317 * field of the VMCB. Therefore, we set up the
1318 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1320 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1322 struct kvm *kvm = vcpu->kvm;
1325 if (kvm->arch.apic_access_page_done)
1328 ret = x86_set_memory_region(kvm,
1329 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1330 APIC_DEFAULT_PHYS_BASE,
1335 kvm->arch.apic_access_page_done = true;
1339 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1342 u64 *entry, new_entry;
1343 int id = vcpu->vcpu_id;
1344 struct vcpu_svm *svm = to_svm(vcpu);
1346 ret = avic_init_access_page(vcpu);
1350 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1353 if (!svm->vcpu.arch.apic->regs)
1356 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1358 /* Setting AVIC backing page address in the phy APIC ID table */
1359 entry = avic_get_physical_id_entry(vcpu, id);
1363 new_entry = READ_ONCE(*entry);
1364 new_entry = (page_to_phys(svm->avic_backing_page) &
1365 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1366 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
1367 WRITE_ONCE(*entry, new_entry);
1369 svm->avic_physical_id_cache = entry;
1374 static inline int avic_get_next_vm_id(void)
1378 spin_lock(&avic_vm_id_lock);
1380 /* AVIC VM ID is one-based. */
1381 id = find_next_zero_bit(avic_vm_id_bitmap, AVIC_VM_ID_NR, 1);
1382 if (id <= AVIC_VM_ID_MASK)
1383 __set_bit(id, avic_vm_id_bitmap);
1387 spin_unlock(&avic_vm_id_lock);
1391 static inline int avic_free_vm_id(int id)
1393 if (id <= 0 || id > AVIC_VM_ID_MASK)
1396 spin_lock(&avic_vm_id_lock);
1397 __clear_bit(id, avic_vm_id_bitmap);
1398 spin_unlock(&avic_vm_id_lock);
1402 static void avic_vm_destroy(struct kvm *kvm)
1404 unsigned long flags;
1405 struct kvm_arch *vm_data = &kvm->arch;
1407 avic_free_vm_id(vm_data->avic_vm_id);
1409 if (vm_data->avic_logical_id_table_page)
1410 __free_page(vm_data->avic_logical_id_table_page);
1411 if (vm_data->avic_physical_id_table_page)
1412 __free_page(vm_data->avic_physical_id_table_page);
1414 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1415 hash_del(&vm_data->hnode);
1416 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1419 static int avic_vm_init(struct kvm *kvm)
1421 unsigned long flags;
1423 struct kvm_arch *vm_data = &kvm->arch;
1424 struct page *p_page;
1425 struct page *l_page;
1430 vm_data->avic_vm_id = avic_get_next_vm_id();
1431 if (vm_data->avic_vm_id < 0)
1432 return vm_data->avic_vm_id;
1434 /* Allocating physical APIC ID table (4KB) */
1435 p_page = alloc_page(GFP_KERNEL);
1439 vm_data->avic_physical_id_table_page = p_page;
1440 clear_page(page_address(p_page));
1442 /* Allocating logical APIC ID table (4KB) */
1443 l_page = alloc_page(GFP_KERNEL);
1447 vm_data->avic_logical_id_table_page = l_page;
1448 clear_page(page_address(l_page));
1450 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1451 hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id);
1452 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1457 avic_vm_destroy(kvm);
1462 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1465 unsigned long flags;
1466 struct amd_svm_iommu_ir *ir;
1467 struct vcpu_svm *svm = to_svm(vcpu);
1469 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1473 * Here, we go through the per-vcpu ir_list to update all existing
1474 * interrupt remapping table entry targeting this vcpu.
1476 spin_lock_irqsave(&svm->ir_list_lock, flags);
1478 if (list_empty(&svm->ir_list))
1481 list_for_each_entry(ir, &svm->ir_list, node) {
1482 ret = amd_iommu_update_ga(cpu, r, ir->data);
1487 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
1491 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1494 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1495 int h_physical_id = kvm_cpu_get_apicid(cpu);
1496 struct vcpu_svm *svm = to_svm(vcpu);
1498 if (!kvm_vcpu_apicv_active(vcpu))
1501 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
1504 entry = READ_ONCE(*(svm->avic_physical_id_cache));
1505 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
1507 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
1508 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
1510 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1511 if (svm->avic_is_running)
1512 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1514 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1515 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
1516 svm->avic_is_running);
1519 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
1522 struct vcpu_svm *svm = to_svm(vcpu);
1524 if (!kvm_vcpu_apicv_active(vcpu))
1527 entry = READ_ONCE(*(svm->avic_physical_id_cache));
1528 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
1529 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
1531 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1532 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1536 * This function is called during VCPU halt/unhalt.
1538 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
1540 struct vcpu_svm *svm = to_svm(vcpu);
1542 svm->avic_is_running = is_run;
1544 avic_vcpu_load(vcpu, vcpu->cpu);
1546 avic_vcpu_put(vcpu);
1549 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1551 struct vcpu_svm *svm = to_svm(vcpu);
1556 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1557 MSR_IA32_APICBASE_ENABLE;
1558 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1559 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1563 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1564 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1566 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1567 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1570 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1572 struct vcpu_svm *svm;
1574 struct page *msrpm_pages;
1575 struct page *hsave_page;
1576 struct page *nested_msrpm_pages;
1579 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1585 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1590 page = alloc_page(GFP_KERNEL);
1594 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1598 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1599 if (!nested_msrpm_pages)
1602 hsave_page = alloc_page(GFP_KERNEL);
1607 err = avic_init_backing_page(&svm->vcpu);
1611 INIT_LIST_HEAD(&svm->ir_list);
1612 spin_lock_init(&svm->ir_list_lock);
1615 /* We initialize this flag to true to make sure that the is_running
1616 * bit would be set the first time the vcpu is loaded.
1618 svm->avic_is_running = true;
1620 svm->nested.hsave = page_address(hsave_page);
1622 svm->msrpm = page_address(msrpm_pages);
1623 svm_vcpu_init_msrpm(svm->msrpm);
1625 svm->nested.msrpm = page_address(nested_msrpm_pages);
1626 svm_vcpu_init_msrpm(svm->nested.msrpm);
1628 svm->vmcb = page_address(page);
1629 clear_page(svm->vmcb);
1630 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1631 svm->asid_generation = 0;
1634 svm_init_osvw(&svm->vcpu);
1639 __free_page(hsave_page);
1641 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1643 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1647 kvm_vcpu_uninit(&svm->vcpu);
1649 kmem_cache_free(kvm_vcpu_cache, svm);
1651 return ERR_PTR(err);
1654 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1656 struct vcpu_svm *svm = to_svm(vcpu);
1658 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1659 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1660 __free_page(virt_to_page(svm->nested.hsave));
1661 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1662 kvm_vcpu_uninit(vcpu);
1663 kmem_cache_free(kvm_vcpu_cache, svm);
1666 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1668 struct vcpu_svm *svm = to_svm(vcpu);
1671 if (unlikely(cpu != vcpu->cpu)) {
1672 svm->asid_generation = 0;
1673 mark_all_dirty(svm->vmcb);
1676 #ifdef CONFIG_X86_64
1677 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1679 savesegment(fs, svm->host.fs);
1680 savesegment(gs, svm->host.gs);
1681 svm->host.ldt = kvm_read_ldt();
1683 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1684 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1686 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1687 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1688 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1689 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1690 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1693 /* This assumes that the kernel never uses MSR_TSC_AUX */
1694 if (static_cpu_has(X86_FEATURE_RDTSCP))
1695 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1697 avic_vcpu_load(vcpu, cpu);
1700 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1702 struct vcpu_svm *svm = to_svm(vcpu);
1705 avic_vcpu_put(vcpu);
1707 ++vcpu->stat.host_state_reload;
1708 kvm_load_ldt(svm->host.ldt);
1709 #ifdef CONFIG_X86_64
1710 loadsegment(fs, svm->host.fs);
1711 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
1712 load_gs_index(svm->host.gs);
1714 #ifdef CONFIG_X86_32_LAZY_GS
1715 loadsegment(gs, svm->host.gs);
1718 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1719 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1722 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
1724 avic_set_running(vcpu, false);
1727 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
1729 avic_set_running(vcpu, true);
1732 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1734 return to_svm(vcpu)->vmcb->save.rflags;
1737 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1740 * Any change of EFLAGS.VM is accompanied by a reload of SS
1741 * (caused by either a task switch or an inter-privilege IRET),
1742 * so we do not need to update the CPL here.
1744 to_svm(vcpu)->vmcb->save.rflags = rflags;
1747 static u32 svm_get_pkru(struct kvm_vcpu *vcpu)
1752 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1755 case VCPU_EXREG_PDPTR:
1756 BUG_ON(!npt_enabled);
1757 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1764 static void svm_set_vintr(struct vcpu_svm *svm)
1766 set_intercept(svm, INTERCEPT_VINTR);
1769 static void svm_clear_vintr(struct vcpu_svm *svm)
1771 clr_intercept(svm, INTERCEPT_VINTR);
1774 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1776 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1779 case VCPU_SREG_CS: return &save->cs;
1780 case VCPU_SREG_DS: return &save->ds;
1781 case VCPU_SREG_ES: return &save->es;
1782 case VCPU_SREG_FS: return &save->fs;
1783 case VCPU_SREG_GS: return &save->gs;
1784 case VCPU_SREG_SS: return &save->ss;
1785 case VCPU_SREG_TR: return &save->tr;
1786 case VCPU_SREG_LDTR: return &save->ldtr;
1792 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1794 struct vmcb_seg *s = svm_seg(vcpu, seg);
1799 static void svm_get_segment(struct kvm_vcpu *vcpu,
1800 struct kvm_segment *var, int seg)
1802 struct vmcb_seg *s = svm_seg(vcpu, seg);
1804 var->base = s->base;
1805 var->limit = s->limit;
1806 var->selector = s->selector;
1807 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1808 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1809 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1810 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1811 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1812 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1813 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1816 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1817 * However, the SVM spec states that the G bit is not observed by the
1818 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1819 * So let's synthesize a legal G bit for all segments, this helps
1820 * running KVM nested. It also helps cross-vendor migration, because
1821 * Intel's vmentry has a check on the 'G' bit.
1823 var->g = s->limit > 0xfffff;
1826 * AMD's VMCB does not have an explicit unusable field, so emulate it
1827 * for cross vendor migration purposes by "not present"
1829 var->unusable = !var->present || (var->type == 0);
1834 * Work around a bug where the busy flag in the tr selector
1844 * The accessed bit must always be set in the segment
1845 * descriptor cache, although it can be cleared in the
1846 * descriptor, the cached bit always remains at 1. Since
1847 * Intel has a check on this, set it here to support
1848 * cross-vendor migration.
1855 * On AMD CPUs sometimes the DB bit in the segment
1856 * descriptor is left as 1, although the whole segment has
1857 * been made unusable. Clear it here to pass an Intel VMX
1858 * entry check when cross vendor migrating.
1862 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1867 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1869 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1874 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1876 struct vcpu_svm *svm = to_svm(vcpu);
1878 dt->size = svm->vmcb->save.idtr.limit;
1879 dt->address = svm->vmcb->save.idtr.base;
1882 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1884 struct vcpu_svm *svm = to_svm(vcpu);
1886 svm->vmcb->save.idtr.limit = dt->size;
1887 svm->vmcb->save.idtr.base = dt->address ;
1888 mark_dirty(svm->vmcb, VMCB_DT);
1891 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1893 struct vcpu_svm *svm = to_svm(vcpu);
1895 dt->size = svm->vmcb->save.gdtr.limit;
1896 dt->address = svm->vmcb->save.gdtr.base;
1899 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1901 struct vcpu_svm *svm = to_svm(vcpu);
1903 svm->vmcb->save.gdtr.limit = dt->size;
1904 svm->vmcb->save.gdtr.base = dt->address ;
1905 mark_dirty(svm->vmcb, VMCB_DT);
1908 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1912 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1916 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1920 static void update_cr0_intercept(struct vcpu_svm *svm)
1922 ulong gcr0 = svm->vcpu.arch.cr0;
1923 u64 *hcr0 = &svm->vmcb->save.cr0;
1925 if (!svm->vcpu.fpu_active)
1926 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1928 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1929 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1931 mark_dirty(svm->vmcb, VMCB_CR);
1933 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1934 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1935 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1937 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1938 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1942 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1944 struct vcpu_svm *svm = to_svm(vcpu);
1946 #ifdef CONFIG_X86_64
1947 if (vcpu->arch.efer & EFER_LME) {
1948 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1949 vcpu->arch.efer |= EFER_LMA;
1950 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1953 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1954 vcpu->arch.efer &= ~EFER_LMA;
1955 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1959 vcpu->arch.cr0 = cr0;
1962 cr0 |= X86_CR0_PG | X86_CR0_WP;
1964 if (!vcpu->fpu_active)
1967 * re-enable caching here because the QEMU bios
1968 * does not do it - this results in some delay at
1971 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1972 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1973 svm->vmcb->save.cr0 = cr0;
1974 mark_dirty(svm->vmcb, VMCB_CR);
1975 update_cr0_intercept(svm);
1978 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1980 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1981 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1983 if (cr4 & X86_CR4_VMXE)
1986 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1987 svm_flush_tlb(vcpu);
1989 vcpu->arch.cr4 = cr4;
1992 cr4 |= host_cr4_mce;
1993 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1994 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1998 static void svm_set_segment(struct kvm_vcpu *vcpu,
1999 struct kvm_segment *var, int seg)
2001 struct vcpu_svm *svm = to_svm(vcpu);
2002 struct vmcb_seg *s = svm_seg(vcpu, seg);
2004 s->base = var->base;
2005 s->limit = var->limit;
2006 s->selector = var->selector;
2010 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2011 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2012 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2013 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
2014 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2015 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2016 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2017 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2021 * This is always accurate, except if SYSRET returned to a segment
2022 * with SS.DPL != 3. Intel does not have this quirk, and always
2023 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2024 * would entail passing the CPL to userspace and back.
2026 if (seg == VCPU_SREG_SS)
2027 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2029 mark_dirty(svm->vmcb, VMCB_SEG);
2032 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2034 struct vcpu_svm *svm = to_svm(vcpu);
2036 clr_exception_intercept(svm, BP_VECTOR);
2038 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2039 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2040 set_exception_intercept(svm, BP_VECTOR);
2042 vcpu->guest_debug = 0;
2045 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2047 if (sd->next_asid > sd->max_asid) {
2048 ++sd->asid_generation;
2050 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2053 svm->asid_generation = sd->asid_generation;
2054 svm->vmcb->control.asid = sd->next_asid++;
2056 mark_dirty(svm->vmcb, VMCB_ASID);
2059 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2061 return to_svm(vcpu)->vmcb->save.dr6;
2064 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2066 struct vcpu_svm *svm = to_svm(vcpu);
2068 svm->vmcb->save.dr6 = value;
2069 mark_dirty(svm->vmcb, VMCB_DR);
2072 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2074 struct vcpu_svm *svm = to_svm(vcpu);
2076 get_debugreg(vcpu->arch.db[0], 0);
2077 get_debugreg(vcpu->arch.db[1], 1);
2078 get_debugreg(vcpu->arch.db[2], 2);
2079 get_debugreg(vcpu->arch.db[3], 3);
2080 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2081 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2083 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2084 set_dr_intercepts(svm);
2087 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2089 struct vcpu_svm *svm = to_svm(vcpu);
2091 svm->vmcb->save.dr7 = value;
2092 mark_dirty(svm->vmcb, VMCB_DR);
2095 static int pf_interception(struct vcpu_svm *svm)
2097 u64 fault_address = svm->vmcb->control.exit_info_2;
2101 switch (svm->apf_reason) {
2103 error_code = svm->vmcb->control.exit_info_1;
2105 trace_kvm_page_fault(fault_address, error_code);
2106 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
2107 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
2108 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2109 svm->vmcb->control.insn_bytes,
2110 svm->vmcb->control.insn_len);
2112 case KVM_PV_REASON_PAGE_NOT_PRESENT:
2113 svm->apf_reason = 0;
2114 local_irq_disable();
2115 kvm_async_pf_task_wait(fault_address);
2118 case KVM_PV_REASON_PAGE_READY:
2119 svm->apf_reason = 0;
2120 local_irq_disable();
2121 kvm_async_pf_task_wake(fault_address);
2128 static int db_interception(struct vcpu_svm *svm)
2130 struct kvm_run *kvm_run = svm->vcpu.run;
2132 if (!(svm->vcpu.guest_debug &
2133 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2134 !svm->nmi_singlestep) {
2135 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2139 if (svm->nmi_singlestep) {
2140 svm->nmi_singlestep = false;
2141 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
2142 svm->vmcb->save.rflags &=
2143 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2146 if (svm->vcpu.guest_debug &
2147 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2148 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2149 kvm_run->debug.arch.pc =
2150 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2151 kvm_run->debug.arch.exception = DB_VECTOR;
2158 static int bp_interception(struct vcpu_svm *svm)
2160 struct kvm_run *kvm_run = svm->vcpu.run;
2162 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2163 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2164 kvm_run->debug.arch.exception = BP_VECTOR;
2168 static int ud_interception(struct vcpu_svm *svm)
2172 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
2173 if (er != EMULATE_DONE)
2174 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2178 static int ac_interception(struct vcpu_svm *svm)
2180 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2184 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
2186 struct vcpu_svm *svm = to_svm(vcpu);
2188 clr_exception_intercept(svm, NM_VECTOR);
2190 svm->vcpu.fpu_active = 1;
2191 update_cr0_intercept(svm);
2194 static int nm_interception(struct vcpu_svm *svm)
2196 svm_fpu_activate(&svm->vcpu);
2200 static bool is_erratum_383(void)
2205 if (!erratum_383_found)
2208 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2212 /* Bit 62 may or may not be set for this mce */
2213 value &= ~(1ULL << 62);
2215 if (value != 0xb600000000010015ULL)
2218 /* Clear MCi_STATUS registers */
2219 for (i = 0; i < 6; ++i)
2220 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2222 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2226 value &= ~(1ULL << 2);
2227 low = lower_32_bits(value);
2228 high = upper_32_bits(value);
2230 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2233 /* Flush tlb to evict multi-match entries */
2239 static void svm_handle_mce(struct vcpu_svm *svm)
2241 if (is_erratum_383()) {
2243 * Erratum 383 triggered. Guest state is corrupt so kill the
2246 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2248 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2254 * On an #MC intercept the MCE handler is not called automatically in
2255 * the host. So do it by hand here.
2259 /* not sure if we ever come back to this point */
2264 static int mc_interception(struct vcpu_svm *svm)
2269 static int shutdown_interception(struct vcpu_svm *svm)
2271 struct kvm_run *kvm_run = svm->vcpu.run;
2274 * VMCB is undefined after a SHUTDOWN intercept
2275 * so reinitialize it.
2277 clear_page(svm->vmcb);
2280 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2284 static int io_interception(struct vcpu_svm *svm)
2286 struct kvm_vcpu *vcpu = &svm->vcpu;
2287 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2288 int size, in, string;
2291 ++svm->vcpu.stat.io_exits;
2292 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2293 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2295 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2297 port = io_info >> 16;
2298 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2299 svm->next_rip = svm->vmcb->control.exit_info_2;
2300 skip_emulated_instruction(&svm->vcpu);
2302 return kvm_fast_pio_out(vcpu, size, port);
2305 static int nmi_interception(struct vcpu_svm *svm)
2310 static int intr_interception(struct vcpu_svm *svm)
2312 ++svm->vcpu.stat.irq_exits;
2316 static int nop_on_interception(struct vcpu_svm *svm)
2321 static int halt_interception(struct vcpu_svm *svm)
2323 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2324 return kvm_emulate_halt(&svm->vcpu);
2327 static int vmmcall_interception(struct vcpu_svm *svm)
2329 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2330 return kvm_emulate_hypercall(&svm->vcpu);
2333 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2335 struct vcpu_svm *svm = to_svm(vcpu);
2337 return svm->nested.nested_cr3;
2340 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2342 struct vcpu_svm *svm = to_svm(vcpu);
2343 u64 cr3 = svm->nested.nested_cr3;
2347 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2348 offset_in_page(cr3) + index * 8, 8);
2354 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2357 struct vcpu_svm *svm = to_svm(vcpu);
2359 svm->vmcb->control.nested_cr3 = root;
2360 mark_dirty(svm->vmcb, VMCB_NPT);
2361 svm_flush_tlb(vcpu);
2364 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2365 struct x86_exception *fault)
2367 struct vcpu_svm *svm = to_svm(vcpu);
2369 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2371 * TODO: track the cause of the nested page fault, and
2372 * correctly fill in the high bits of exit_info_1.
2374 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2375 svm->vmcb->control.exit_code_hi = 0;
2376 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2377 svm->vmcb->control.exit_info_2 = fault->address;
2380 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2381 svm->vmcb->control.exit_info_1 |= fault->error_code;
2384 * The present bit is always zero for page structure faults on real
2387 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2388 svm->vmcb->control.exit_info_1 &= ~1;
2390 nested_svm_vmexit(svm);
2393 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2395 WARN_ON(mmu_is_nested(vcpu));
2396 kvm_init_shadow_mmu(vcpu);
2397 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2398 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2399 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2400 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2401 vcpu->arch.mmu.shadow_root_level = get_npt_level();
2402 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2403 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2406 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2408 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2411 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2413 if (!(svm->vcpu.arch.efer & EFER_SVME)
2414 || !is_paging(&svm->vcpu)) {
2415 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2419 if (svm->vmcb->save.cpl) {
2420 kvm_inject_gp(&svm->vcpu, 0);
2427 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2428 bool has_error_code, u32 error_code)
2432 if (!is_guest_mode(&svm->vcpu))
2435 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2436 svm->vmcb->control.exit_code_hi = 0;
2437 svm->vmcb->control.exit_info_1 = error_code;
2438 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2440 vmexit = nested_svm_intercept(svm);
2441 if (vmexit == NESTED_EXIT_DONE)
2442 svm->nested.exit_required = true;
2447 /* This function returns true if it is save to enable the irq window */
2448 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2450 if (!is_guest_mode(&svm->vcpu))
2453 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2456 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2460 * if vmexit was already requested (by intercepted exception
2461 * for instance) do not overwrite it with "external interrupt"
2464 if (svm->nested.exit_required)
2467 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2468 svm->vmcb->control.exit_info_1 = 0;
2469 svm->vmcb->control.exit_info_2 = 0;
2471 if (svm->nested.intercept & 1ULL) {
2473 * The #vmexit can't be emulated here directly because this
2474 * code path runs with irqs and preemption disabled. A
2475 * #vmexit emulation might sleep. Only signal request for
2478 svm->nested.exit_required = true;
2479 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2486 /* This function returns true if it is save to enable the nmi window */
2487 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2489 if (!is_guest_mode(&svm->vcpu))
2492 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2495 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2496 svm->nested.exit_required = true;
2501 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2507 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2508 if (is_error_page(page))
2516 kvm_inject_gp(&svm->vcpu, 0);
2521 static void nested_svm_unmap(struct page *page)
2524 kvm_release_page_dirty(page);
2527 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2529 unsigned port, size, iopm_len;
2534 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2535 return NESTED_EXIT_HOST;
2537 port = svm->vmcb->control.exit_info_1 >> 16;
2538 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2539 SVM_IOIO_SIZE_SHIFT;
2540 gpa = svm->nested.vmcb_iopm + (port / 8);
2541 start_bit = port % 8;
2542 iopm_len = (start_bit + size > 8) ? 2 : 1;
2543 mask = (0xf >> (4 - size)) << start_bit;
2546 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2547 return NESTED_EXIT_DONE;
2549 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2552 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2554 u32 offset, msr, value;
2557 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2558 return NESTED_EXIT_HOST;
2560 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2561 offset = svm_msrpm_offset(msr);
2562 write = svm->vmcb->control.exit_info_1 & 1;
2563 mask = 1 << ((2 * (msr & 0xf)) + write);
2565 if (offset == MSR_INVALID)
2566 return NESTED_EXIT_DONE;
2568 /* Offset is in 32 bit units but need in 8 bit units */
2571 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2572 return NESTED_EXIT_DONE;
2574 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2577 static int nested_svm_exit_special(struct vcpu_svm *svm)
2579 u32 exit_code = svm->vmcb->control.exit_code;
2581 switch (exit_code) {
2584 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2585 return NESTED_EXIT_HOST;
2587 /* For now we are always handling NPFs when using them */
2589 return NESTED_EXIT_HOST;
2591 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2592 /* When we're shadowing, trap PFs, but not async PF */
2593 if (!npt_enabled && svm->apf_reason == 0)
2594 return NESTED_EXIT_HOST;
2596 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2597 nm_interception(svm);
2603 return NESTED_EXIT_CONTINUE;
2607 * If this function returns true, this #vmexit was already handled
2609 static int nested_svm_intercept(struct vcpu_svm *svm)
2611 u32 exit_code = svm->vmcb->control.exit_code;
2612 int vmexit = NESTED_EXIT_HOST;
2614 switch (exit_code) {
2616 vmexit = nested_svm_exit_handled_msr(svm);
2619 vmexit = nested_svm_intercept_ioio(svm);
2621 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2622 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2623 if (svm->nested.intercept_cr & bit)
2624 vmexit = NESTED_EXIT_DONE;
2627 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2628 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2629 if (svm->nested.intercept_dr & bit)
2630 vmexit = NESTED_EXIT_DONE;
2633 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2634 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2635 if (svm->nested.intercept_exceptions & excp_bits)
2636 vmexit = NESTED_EXIT_DONE;
2637 /* async page fault always cause vmexit */
2638 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2639 svm->apf_reason != 0)
2640 vmexit = NESTED_EXIT_DONE;
2643 case SVM_EXIT_ERR: {
2644 vmexit = NESTED_EXIT_DONE;
2648 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2649 if (svm->nested.intercept & exit_bits)
2650 vmexit = NESTED_EXIT_DONE;
2657 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2661 vmexit = nested_svm_intercept(svm);
2663 if (vmexit == NESTED_EXIT_DONE)
2664 nested_svm_vmexit(svm);
2669 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2671 struct vmcb_control_area *dst = &dst_vmcb->control;
2672 struct vmcb_control_area *from = &from_vmcb->control;
2674 dst->intercept_cr = from->intercept_cr;
2675 dst->intercept_dr = from->intercept_dr;
2676 dst->intercept_exceptions = from->intercept_exceptions;
2677 dst->intercept = from->intercept;
2678 dst->iopm_base_pa = from->iopm_base_pa;
2679 dst->msrpm_base_pa = from->msrpm_base_pa;
2680 dst->tsc_offset = from->tsc_offset;
2681 dst->asid = from->asid;
2682 dst->tlb_ctl = from->tlb_ctl;
2683 dst->int_ctl = from->int_ctl;
2684 dst->int_vector = from->int_vector;
2685 dst->int_state = from->int_state;
2686 dst->exit_code = from->exit_code;
2687 dst->exit_code_hi = from->exit_code_hi;
2688 dst->exit_info_1 = from->exit_info_1;
2689 dst->exit_info_2 = from->exit_info_2;
2690 dst->exit_int_info = from->exit_int_info;
2691 dst->exit_int_info_err = from->exit_int_info_err;
2692 dst->nested_ctl = from->nested_ctl;
2693 dst->event_inj = from->event_inj;
2694 dst->event_inj_err = from->event_inj_err;
2695 dst->nested_cr3 = from->nested_cr3;
2696 dst->lbr_ctl = from->lbr_ctl;
2699 static int nested_svm_vmexit(struct vcpu_svm *svm)
2701 struct vmcb *nested_vmcb;
2702 struct vmcb *hsave = svm->nested.hsave;
2703 struct vmcb *vmcb = svm->vmcb;
2706 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2707 vmcb->control.exit_info_1,
2708 vmcb->control.exit_info_2,
2709 vmcb->control.exit_int_info,
2710 vmcb->control.exit_int_info_err,
2713 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2717 /* Exit Guest-Mode */
2718 leave_guest_mode(&svm->vcpu);
2719 svm->nested.vmcb = 0;
2721 /* Give the current vmcb to the guest */
2724 nested_vmcb->save.es = vmcb->save.es;
2725 nested_vmcb->save.cs = vmcb->save.cs;
2726 nested_vmcb->save.ss = vmcb->save.ss;
2727 nested_vmcb->save.ds = vmcb->save.ds;
2728 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2729 nested_vmcb->save.idtr = vmcb->save.idtr;
2730 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2731 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2732 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2733 nested_vmcb->save.cr2 = vmcb->save.cr2;
2734 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2735 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2736 nested_vmcb->save.rip = vmcb->save.rip;
2737 nested_vmcb->save.rsp = vmcb->save.rsp;
2738 nested_vmcb->save.rax = vmcb->save.rax;
2739 nested_vmcb->save.dr7 = vmcb->save.dr7;
2740 nested_vmcb->save.dr6 = vmcb->save.dr6;
2741 nested_vmcb->save.cpl = vmcb->save.cpl;
2743 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2744 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2745 nested_vmcb->control.int_state = vmcb->control.int_state;
2746 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2747 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2748 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2749 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2750 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2751 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2753 if (svm->nrips_enabled)
2754 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2757 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2758 * to make sure that we do not lose injected events. So check event_inj
2759 * here and copy it to exit_int_info if it is valid.
2760 * Exit_int_info and event_inj can't be both valid because the case
2761 * below only happens on a VMRUN instruction intercept which has
2762 * no valid exit_int_info set.
2764 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2765 struct vmcb_control_area *nc = &nested_vmcb->control;
2767 nc->exit_int_info = vmcb->control.event_inj;
2768 nc->exit_int_info_err = vmcb->control.event_inj_err;
2771 nested_vmcb->control.tlb_ctl = 0;
2772 nested_vmcb->control.event_inj = 0;
2773 nested_vmcb->control.event_inj_err = 0;
2775 /* We always set V_INTR_MASKING and remember the old value in hflags */
2776 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2777 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2779 /* Restore the original control entries */
2780 copy_vmcb_control_area(vmcb, hsave);
2782 kvm_clear_exception_queue(&svm->vcpu);
2783 kvm_clear_interrupt_queue(&svm->vcpu);
2785 svm->nested.nested_cr3 = 0;
2787 /* Restore selected save entries */
2788 svm->vmcb->save.es = hsave->save.es;
2789 svm->vmcb->save.cs = hsave->save.cs;
2790 svm->vmcb->save.ss = hsave->save.ss;
2791 svm->vmcb->save.ds = hsave->save.ds;
2792 svm->vmcb->save.gdtr = hsave->save.gdtr;
2793 svm->vmcb->save.idtr = hsave->save.idtr;
2794 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2795 svm_set_efer(&svm->vcpu, hsave->save.efer);
2796 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2797 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2799 svm->vmcb->save.cr3 = hsave->save.cr3;
2800 svm->vcpu.arch.cr3 = hsave->save.cr3;
2802 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2804 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2805 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2806 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2807 svm->vmcb->save.dr7 = 0;
2808 svm->vmcb->save.cpl = 0;
2809 svm->vmcb->control.exit_int_info = 0;
2811 mark_all_dirty(svm->vmcb);
2813 nested_svm_unmap(page);
2815 nested_svm_uninit_mmu_context(&svm->vcpu);
2816 kvm_mmu_reset_context(&svm->vcpu);
2817 kvm_mmu_load(&svm->vcpu);
2822 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2825 * This function merges the msr permission bitmaps of kvm and the
2826 * nested vmcb. It is optimized in that it only merges the parts where
2827 * the kvm msr permission bitmap may contain zero bits
2831 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2834 for (i = 0; i < MSRPM_OFFSETS; i++) {
2838 if (msrpm_offsets[i] == 0xffffffff)
2841 p = msrpm_offsets[i];
2842 offset = svm->nested.vmcb_msrpm + (p * 4);
2844 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2847 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2850 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2855 static bool nested_vmcb_checks(struct vmcb *vmcb)
2857 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2860 if (vmcb->control.asid == 0)
2863 if (vmcb->control.nested_ctl && !npt_enabled)
2869 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2871 struct vmcb *nested_vmcb;
2872 struct vmcb *hsave = svm->nested.hsave;
2873 struct vmcb *vmcb = svm->vmcb;
2877 vmcb_gpa = svm->vmcb->save.rax;
2879 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2883 if (!nested_vmcb_checks(nested_vmcb)) {
2884 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2885 nested_vmcb->control.exit_code_hi = 0;
2886 nested_vmcb->control.exit_info_1 = 0;
2887 nested_vmcb->control.exit_info_2 = 0;
2889 nested_svm_unmap(page);
2894 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2895 nested_vmcb->save.rip,
2896 nested_vmcb->control.int_ctl,
2897 nested_vmcb->control.event_inj,
2898 nested_vmcb->control.nested_ctl);
2900 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2901 nested_vmcb->control.intercept_cr >> 16,
2902 nested_vmcb->control.intercept_exceptions,
2903 nested_vmcb->control.intercept);
2905 /* Clear internal status */
2906 kvm_clear_exception_queue(&svm->vcpu);
2907 kvm_clear_interrupt_queue(&svm->vcpu);
2910 * Save the old vmcb, so we don't need to pick what we save, but can
2911 * restore everything when a VMEXIT occurs
2913 hsave->save.es = vmcb->save.es;
2914 hsave->save.cs = vmcb->save.cs;
2915 hsave->save.ss = vmcb->save.ss;
2916 hsave->save.ds = vmcb->save.ds;
2917 hsave->save.gdtr = vmcb->save.gdtr;
2918 hsave->save.idtr = vmcb->save.idtr;
2919 hsave->save.efer = svm->vcpu.arch.efer;
2920 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2921 hsave->save.cr4 = svm->vcpu.arch.cr4;
2922 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2923 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2924 hsave->save.rsp = vmcb->save.rsp;
2925 hsave->save.rax = vmcb->save.rax;
2927 hsave->save.cr3 = vmcb->save.cr3;
2929 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2931 copy_vmcb_control_area(hsave, vmcb);
2933 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2934 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2936 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2938 if (nested_vmcb->control.nested_ctl) {
2939 kvm_mmu_unload(&svm->vcpu);
2940 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2941 nested_svm_init_mmu_context(&svm->vcpu);
2944 /* Load the nested guest state */
2945 svm->vmcb->save.es = nested_vmcb->save.es;
2946 svm->vmcb->save.cs = nested_vmcb->save.cs;
2947 svm->vmcb->save.ss = nested_vmcb->save.ss;
2948 svm->vmcb->save.ds = nested_vmcb->save.ds;
2949 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2950 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2951 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2952 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2953 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2954 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2956 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2957 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2959 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2961 /* Guest paging mode is active - reset mmu */
2962 kvm_mmu_reset_context(&svm->vcpu);
2964 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2965 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2966 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2967 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2969 /* In case we don't even reach vcpu_run, the fields are not updated */
2970 svm->vmcb->save.rax = nested_vmcb->save.rax;
2971 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2972 svm->vmcb->save.rip = nested_vmcb->save.rip;
2973 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2974 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2975 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2977 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2978 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2980 /* cache intercepts */
2981 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2982 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2983 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2984 svm->nested.intercept = nested_vmcb->control.intercept;
2986 svm_flush_tlb(&svm->vcpu);
2987 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2988 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2989 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2991 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2993 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2994 /* We only want the cr8 intercept bits of the guest */
2995 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2996 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2999 /* We don't want to see VMMCALLs from a nested guest */
3000 clr_intercept(svm, INTERCEPT_VMMCALL);
3002 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
3003 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3004 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3005 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
3006 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3007 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3009 nested_svm_unmap(page);
3011 /* Enter Guest-Mode */
3012 enter_guest_mode(&svm->vcpu);
3015 * Merge guest and host intercepts - must be called with vcpu in
3016 * guest-mode to take affect here
3018 recalc_intercepts(svm);
3020 svm->nested.vmcb = vmcb_gpa;
3024 mark_all_dirty(svm->vmcb);
3029 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3031 to_vmcb->save.fs = from_vmcb->save.fs;
3032 to_vmcb->save.gs = from_vmcb->save.gs;
3033 to_vmcb->save.tr = from_vmcb->save.tr;
3034 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3035 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3036 to_vmcb->save.star = from_vmcb->save.star;
3037 to_vmcb->save.lstar = from_vmcb->save.lstar;
3038 to_vmcb->save.cstar = from_vmcb->save.cstar;
3039 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3040 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3041 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3042 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3045 static int vmload_interception(struct vcpu_svm *svm)
3047 struct vmcb *nested_vmcb;
3050 if (nested_svm_check_permissions(svm))
3053 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3057 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3058 skip_emulated_instruction(&svm->vcpu);
3060 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3061 nested_svm_unmap(page);
3066 static int vmsave_interception(struct vcpu_svm *svm)
3068 struct vmcb *nested_vmcb;
3071 if (nested_svm_check_permissions(svm))
3074 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3078 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3079 skip_emulated_instruction(&svm->vcpu);
3081 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3082 nested_svm_unmap(page);
3087 static int vmrun_interception(struct vcpu_svm *svm)
3089 if (nested_svm_check_permissions(svm))
3092 /* Save rip after vmrun instruction */
3093 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3095 if (!nested_svm_vmrun(svm))
3098 if (!nested_svm_vmrun_msrpm(svm))
3105 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3106 svm->vmcb->control.exit_code_hi = 0;
3107 svm->vmcb->control.exit_info_1 = 0;
3108 svm->vmcb->control.exit_info_2 = 0;
3110 nested_svm_vmexit(svm);
3115 static int stgi_interception(struct vcpu_svm *svm)
3117 if (nested_svm_check_permissions(svm))
3120 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3121 skip_emulated_instruction(&svm->vcpu);
3122 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3129 static int clgi_interception(struct vcpu_svm *svm)
3131 if (nested_svm_check_permissions(svm))
3134 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3135 skip_emulated_instruction(&svm->vcpu);
3139 /* After a CLGI no interrupts should come */
3140 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3141 svm_clear_vintr(svm);
3142 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3143 mark_dirty(svm->vmcb, VMCB_INTR);
3149 static int invlpga_interception(struct vcpu_svm *svm)
3151 struct kvm_vcpu *vcpu = &svm->vcpu;
3153 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3154 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3156 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3157 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3159 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3160 skip_emulated_instruction(&svm->vcpu);
3164 static int skinit_interception(struct vcpu_svm *svm)
3166 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3168 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3172 static int wbinvd_interception(struct vcpu_svm *svm)
3174 kvm_emulate_wbinvd(&svm->vcpu);
3178 static int xsetbv_interception(struct vcpu_svm *svm)
3180 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3181 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3183 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3184 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3185 skip_emulated_instruction(&svm->vcpu);
3191 static int task_switch_interception(struct vcpu_svm *svm)
3195 int int_type = svm->vmcb->control.exit_int_info &
3196 SVM_EXITINTINFO_TYPE_MASK;
3197 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3199 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3201 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3202 bool has_error_code = false;
3205 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3207 if (svm->vmcb->control.exit_info_2 &
3208 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3209 reason = TASK_SWITCH_IRET;
3210 else if (svm->vmcb->control.exit_info_2 &
3211 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3212 reason = TASK_SWITCH_JMP;
3214 reason = TASK_SWITCH_GATE;
3216 reason = TASK_SWITCH_CALL;
3218 if (reason == TASK_SWITCH_GATE) {
3220 case SVM_EXITINTINFO_TYPE_NMI:
3221 svm->vcpu.arch.nmi_injected = false;
3223 case SVM_EXITINTINFO_TYPE_EXEPT:
3224 if (svm->vmcb->control.exit_info_2 &
3225 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3226 has_error_code = true;
3228 (u32)svm->vmcb->control.exit_info_2;
3230 kvm_clear_exception_queue(&svm->vcpu);
3232 case SVM_EXITINTINFO_TYPE_INTR:
3233 kvm_clear_interrupt_queue(&svm->vcpu);
3240 if (reason != TASK_SWITCH_GATE ||
3241 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3242 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3243 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3244 skip_emulated_instruction(&svm->vcpu);
3246 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3249 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3250 has_error_code, error_code) == EMULATE_FAIL) {
3251 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3252 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3253 svm->vcpu.run->internal.ndata = 0;
3259 static int cpuid_interception(struct vcpu_svm *svm)
3261 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3262 kvm_emulate_cpuid(&svm->vcpu);
3266 static int iret_interception(struct vcpu_svm *svm)
3268 ++svm->vcpu.stat.nmi_window_exits;
3269 clr_intercept(svm, INTERCEPT_IRET);
3270 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3271 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3272 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3276 static int invlpg_interception(struct vcpu_svm *svm)
3278 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3279 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3281 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3282 skip_emulated_instruction(&svm->vcpu);
3286 static int emulate_on_interception(struct vcpu_svm *svm)
3288 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3291 static int rdpmc_interception(struct vcpu_svm *svm)
3295 if (!static_cpu_has(X86_FEATURE_NRIPS))
3296 return emulate_on_interception(svm);
3298 err = kvm_rdpmc(&svm->vcpu);
3299 kvm_complete_insn_gp(&svm->vcpu, err);
3304 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3307 unsigned long cr0 = svm->vcpu.arch.cr0;
3311 intercept = svm->nested.intercept;
3313 if (!is_guest_mode(&svm->vcpu) ||
3314 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3317 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3318 val &= ~SVM_CR0_SELECTIVE_MASK;
3321 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3322 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3328 #define CR_VALID (1ULL << 63)
3330 static int cr_interception(struct vcpu_svm *svm)
3336 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3337 return emulate_on_interception(svm);
3339 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3340 return emulate_on_interception(svm);
3342 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3343 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3344 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3346 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3349 if (cr >= 16) { /* mov to cr */
3351 val = kvm_register_read(&svm->vcpu, reg);
3354 if (!check_selective_cr0_intercepted(svm, val))
3355 err = kvm_set_cr0(&svm->vcpu, val);
3361 err = kvm_set_cr3(&svm->vcpu, val);
3364 err = kvm_set_cr4(&svm->vcpu, val);
3367 err = kvm_set_cr8(&svm->vcpu, val);
3370 WARN(1, "unhandled write to CR%d", cr);
3371 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3374 } else { /* mov from cr */
3377 val = kvm_read_cr0(&svm->vcpu);
3380 val = svm->vcpu.arch.cr2;
3383 val = kvm_read_cr3(&svm->vcpu);
3386 val = kvm_read_cr4(&svm->vcpu);
3389 val = kvm_get_cr8(&svm->vcpu);
3392 WARN(1, "unhandled read from CR%d", cr);
3393 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3396 kvm_register_write(&svm->vcpu, reg, val);
3398 kvm_complete_insn_gp(&svm->vcpu, err);
3403 static int dr_interception(struct vcpu_svm *svm)
3408 if (svm->vcpu.guest_debug == 0) {
3410 * No more DR vmexits; force a reload of the debug registers
3411 * and reenter on this instruction. The next vmexit will
3412 * retrieve the full state of the debug registers.
3414 clr_dr_intercepts(svm);
3415 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3419 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3420 return emulate_on_interception(svm);
3422 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3423 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3425 if (dr >= 16) { /* mov to DRn */
3426 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3428 val = kvm_register_read(&svm->vcpu, reg);
3429 kvm_set_dr(&svm->vcpu, dr - 16, val);
3431 if (!kvm_require_dr(&svm->vcpu, dr))
3433 kvm_get_dr(&svm->vcpu, dr, &val);
3434 kvm_register_write(&svm->vcpu, reg, val);
3437 skip_emulated_instruction(&svm->vcpu);
3442 static int cr8_write_interception(struct vcpu_svm *svm)
3444 struct kvm_run *kvm_run = svm->vcpu.run;
3447 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3448 /* instruction emulation calls kvm_set_cr8() */
3449 r = cr_interception(svm);
3450 if (lapic_in_kernel(&svm->vcpu))
3452 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3454 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3458 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3460 struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3461 return vmcb->control.tsc_offset + host_tsc;
3464 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3466 struct vcpu_svm *svm = to_svm(vcpu);
3468 switch (msr_info->index) {
3469 case MSR_IA32_TSC: {
3470 msr_info->data = svm->vmcb->control.tsc_offset +
3471 kvm_scale_tsc(vcpu, rdtsc());
3476 msr_info->data = svm->vmcb->save.star;
3478 #ifdef CONFIG_X86_64
3480 msr_info->data = svm->vmcb->save.lstar;
3483 msr_info->data = svm->vmcb->save.cstar;
3485 case MSR_KERNEL_GS_BASE:
3486 msr_info->data = svm->vmcb->save.kernel_gs_base;
3488 case MSR_SYSCALL_MASK:
3489 msr_info->data = svm->vmcb->save.sfmask;
3492 case MSR_IA32_SYSENTER_CS:
3493 msr_info->data = svm->vmcb->save.sysenter_cs;
3495 case MSR_IA32_SYSENTER_EIP:
3496 msr_info->data = svm->sysenter_eip;
3498 case MSR_IA32_SYSENTER_ESP:
3499 msr_info->data = svm->sysenter_esp;
3502 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3504 msr_info->data = svm->tsc_aux;
3507 * Nobody will change the following 5 values in the VMCB so we can
3508 * safely return them on rdmsr. They will always be 0 until LBRV is
3511 case MSR_IA32_DEBUGCTLMSR:
3512 msr_info->data = svm->vmcb->save.dbgctl;
3514 case MSR_IA32_LASTBRANCHFROMIP:
3515 msr_info->data = svm->vmcb->save.br_from;
3517 case MSR_IA32_LASTBRANCHTOIP:
3518 msr_info->data = svm->vmcb->save.br_to;
3520 case MSR_IA32_LASTINTFROMIP:
3521 msr_info->data = svm->vmcb->save.last_excp_from;
3523 case MSR_IA32_LASTINTTOIP:
3524 msr_info->data = svm->vmcb->save.last_excp_to;
3526 case MSR_VM_HSAVE_PA:
3527 msr_info->data = svm->nested.hsave_msr;
3530 msr_info->data = svm->nested.vm_cr_msr;
3532 case MSR_IA32_UCODE_REV:
3533 msr_info->data = 0x01000065;
3535 case MSR_F15H_IC_CFG: {
3539 family = guest_cpuid_family(vcpu);
3540 model = guest_cpuid_model(vcpu);
3542 if (family < 0 || model < 0)
3543 return kvm_get_msr_common(vcpu, msr_info);
3547 if (family == 0x15 &&
3548 (model >= 0x2 && model < 0x20))
3549 msr_info->data = 0x1E;
3553 return kvm_get_msr_common(vcpu, msr_info);
3558 static int rdmsr_interception(struct vcpu_svm *svm)
3560 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3561 struct msr_data msr_info;
3563 msr_info.index = ecx;
3564 msr_info.host_initiated = false;
3565 if (svm_get_msr(&svm->vcpu, &msr_info)) {
3566 trace_kvm_msr_read_ex(ecx);
3567 kvm_inject_gp(&svm->vcpu, 0);
3569 trace_kvm_msr_read(ecx, msr_info.data);
3571 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3572 msr_info.data & 0xffffffff);
3573 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3574 msr_info.data >> 32);
3575 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3576 skip_emulated_instruction(&svm->vcpu);
3581 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3583 struct vcpu_svm *svm = to_svm(vcpu);
3584 int svm_dis, chg_mask;
3586 if (data & ~SVM_VM_CR_VALID_MASK)
3589 chg_mask = SVM_VM_CR_VALID_MASK;
3591 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3592 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3594 svm->nested.vm_cr_msr &= ~chg_mask;
3595 svm->nested.vm_cr_msr |= (data & chg_mask);
3597 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3599 /* check for svm_disable while efer.svme is set */
3600 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3606 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3608 struct vcpu_svm *svm = to_svm(vcpu);
3610 u32 ecx = msr->index;
3611 u64 data = msr->data;
3614 kvm_write_tsc(vcpu, msr);
3617 svm->vmcb->save.star = data;
3619 #ifdef CONFIG_X86_64
3621 svm->vmcb->save.lstar = data;
3624 svm->vmcb->save.cstar = data;
3626 case MSR_KERNEL_GS_BASE:
3627 svm->vmcb->save.kernel_gs_base = data;
3629 case MSR_SYSCALL_MASK:
3630 svm->vmcb->save.sfmask = data;
3633 case MSR_IA32_SYSENTER_CS:
3634 svm->vmcb->save.sysenter_cs = data;
3636 case MSR_IA32_SYSENTER_EIP:
3637 svm->sysenter_eip = data;
3638 svm->vmcb->save.sysenter_eip = data;
3640 case MSR_IA32_SYSENTER_ESP:
3641 svm->sysenter_esp = data;
3642 svm->vmcb->save.sysenter_esp = data;
3645 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3649 * This is rare, so we update the MSR here instead of using
3650 * direct_access_msrs. Doing that would require a rdmsr in
3653 svm->tsc_aux = data;
3654 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
3656 case MSR_IA32_DEBUGCTLMSR:
3657 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3658 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3662 if (data & DEBUGCTL_RESERVED_BITS)
3665 svm->vmcb->save.dbgctl = data;
3666 mark_dirty(svm->vmcb, VMCB_LBR);
3667 if (data & (1ULL<<0))
3668 svm_enable_lbrv(svm);
3670 svm_disable_lbrv(svm);
3672 case MSR_VM_HSAVE_PA:
3673 svm->nested.hsave_msr = data;
3676 return svm_set_vm_cr(vcpu, data);
3678 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3680 case MSR_IA32_APICBASE:
3681 if (kvm_vcpu_apicv_active(vcpu))
3682 avic_update_vapic_bar(to_svm(vcpu), data);
3683 /* Follow through */
3685 return kvm_set_msr_common(vcpu, msr);
3690 static int wrmsr_interception(struct vcpu_svm *svm)
3692 struct msr_data msr;
3693 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3694 u64 data = kvm_read_edx_eax(&svm->vcpu);
3698 msr.host_initiated = false;
3700 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3701 if (kvm_set_msr(&svm->vcpu, &msr)) {
3702 trace_kvm_msr_write_ex(ecx, data);
3703 kvm_inject_gp(&svm->vcpu, 0);
3705 trace_kvm_msr_write(ecx, data);
3706 skip_emulated_instruction(&svm->vcpu);
3711 static int msr_interception(struct vcpu_svm *svm)
3713 if (svm->vmcb->control.exit_info_1)
3714 return wrmsr_interception(svm);
3716 return rdmsr_interception(svm);
3719 static int interrupt_window_interception(struct vcpu_svm *svm)
3721 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3722 svm_clear_vintr(svm);
3723 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3724 mark_dirty(svm->vmcb, VMCB_INTR);
3725 ++svm->vcpu.stat.irq_window_exits;
3729 static int pause_interception(struct vcpu_svm *svm)
3731 kvm_vcpu_on_spin(&(svm->vcpu));
3735 static int nop_interception(struct vcpu_svm *svm)
3737 skip_emulated_instruction(&(svm->vcpu));
3741 static int monitor_interception(struct vcpu_svm *svm)
3743 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3744 return nop_interception(svm);
3747 static int mwait_interception(struct vcpu_svm *svm)
3749 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3750 return nop_interception(svm);
3753 enum avic_ipi_failure_cause {
3754 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
3755 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
3756 AVIC_IPI_FAILURE_INVALID_TARGET,
3757 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
3760 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
3762 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
3763 u32 icrl = svm->vmcb->control.exit_info_1;
3764 u32 id = svm->vmcb->control.exit_info_2 >> 32;
3765 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
3766 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3768 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
3771 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
3773 * AVIC hardware handles the generation of
3774 * IPIs when the specified Message Type is Fixed
3775 * (also known as fixed delivery mode) and
3776 * the Trigger Mode is edge-triggered. The hardware
3777 * also supports self and broadcast delivery modes
3778 * specified via the Destination Shorthand(DSH)
3779 * field of the ICRL. Logical and physical APIC ID
3780 * formats are supported. All other IPI types cause
3781 * a #VMEXIT, which needs to emulated.
3783 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
3784 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
3786 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
3788 struct kvm_vcpu *vcpu;
3789 struct kvm *kvm = svm->vcpu.kvm;
3790 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3793 * At this point, we expect that the AVIC HW has already
3794 * set the appropriate IRR bits on the valid target
3795 * vcpus. So, we just need to kick the appropriate vcpu.
3797 kvm_for_each_vcpu(i, vcpu, kvm) {
3798 bool m = kvm_apic_match_dest(vcpu, apic,
3799 icrl & KVM_APIC_SHORT_MASK,
3800 GET_APIC_DEST_FIELD(icrh),
3801 icrl & KVM_APIC_DEST_MASK);
3803 if (m && !avic_vcpu_is_running(vcpu))
3804 kvm_vcpu_wake_up(vcpu);
3808 case AVIC_IPI_FAILURE_INVALID_TARGET:
3810 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
3811 WARN_ONCE(1, "Invalid backing page\n");
3814 pr_err("Unknown IPI interception\n");
3820 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
3822 struct kvm_arch *vm_data = &vcpu->kvm->arch;
3824 u32 *logical_apic_id_table;
3825 int dlid = GET_APIC_LOGICAL_ID(ldr);
3830 if (flat) { /* flat */
3831 index = ffs(dlid) - 1;
3834 } else { /* cluster */
3835 int cluster = (dlid & 0xf0) >> 4;
3836 int apic = ffs(dlid & 0x0f) - 1;
3838 if ((apic < 0) || (apic > 7) ||
3841 index = (cluster << 2) + apic;
3844 logical_apic_id_table = (u32 *) page_address(vm_data->avic_logical_id_table_page);
3846 return &logical_apic_id_table[index];
3849 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
3853 u32 *entry, new_entry;
3855 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
3856 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
3860 new_entry = READ_ONCE(*entry);
3861 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
3862 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
3864 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3866 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3867 WRITE_ONCE(*entry, new_entry);
3872 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
3875 struct vcpu_svm *svm = to_svm(vcpu);
3876 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
3881 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
3882 if (ret && svm->ldr_reg) {
3883 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
3891 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
3894 struct vcpu_svm *svm = to_svm(vcpu);
3895 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
3896 u32 id = (apic_id_reg >> 24) & 0xff;
3898 if (vcpu->vcpu_id == id)
3901 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
3902 new = avic_get_physical_id_entry(vcpu, id);
3906 /* We need to move physical_id_entry to new offset */
3909 to_svm(vcpu)->avic_physical_id_cache = new;
3912 * Also update the guest physical APIC ID in the logical
3913 * APIC ID table entry if already setup the LDR.
3916 avic_handle_ldr_update(vcpu);
3921 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
3923 struct vcpu_svm *svm = to_svm(vcpu);
3924 struct kvm_arch *vm_data = &vcpu->kvm->arch;
3925 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
3926 u32 mod = (dfr >> 28) & 0xf;
3929 * We assume that all local APICs are using the same type.
3930 * If this changes, we need to flush the AVIC logical
3933 if (vm_data->ldr_mode == mod)
3936 clear_page(page_address(vm_data->avic_logical_id_table_page));
3937 vm_data->ldr_mode = mod;
3940 avic_handle_ldr_update(vcpu);
3944 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
3946 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3947 u32 offset = svm->vmcb->control.exit_info_1 &
3948 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
3952 if (avic_handle_apic_id_update(&svm->vcpu))
3956 if (avic_handle_ldr_update(&svm->vcpu))
3960 avic_handle_dfr_update(&svm->vcpu);
3966 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
3971 static bool is_avic_unaccelerated_access_trap(u32 offset)
4000 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4003 u32 offset = svm->vmcb->control.exit_info_1 &
4004 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4005 u32 vector = svm->vmcb->control.exit_info_2 &
4006 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4007 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4008 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4009 bool trap = is_avic_unaccelerated_access_trap(offset);
4011 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4012 trap, write, vector);
4015 WARN_ONCE(!write, "svm: Handling trap read.\n");
4016 ret = avic_unaccel_trap_write(svm);
4018 /* Handling Fault */
4019 ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4025 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4026 [SVM_EXIT_READ_CR0] = cr_interception,
4027 [SVM_EXIT_READ_CR3] = cr_interception,
4028 [SVM_EXIT_READ_CR4] = cr_interception,
4029 [SVM_EXIT_READ_CR8] = cr_interception,
4030 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4031 [SVM_EXIT_WRITE_CR0] = cr_interception,
4032 [SVM_EXIT_WRITE_CR3] = cr_interception,
4033 [SVM_EXIT_WRITE_CR4] = cr_interception,
4034 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4035 [SVM_EXIT_READ_DR0] = dr_interception,
4036 [SVM_EXIT_READ_DR1] = dr_interception,
4037 [SVM_EXIT_READ_DR2] = dr_interception,
4038 [SVM_EXIT_READ_DR3] = dr_interception,
4039 [SVM_EXIT_READ_DR4] = dr_interception,
4040 [SVM_EXIT_READ_DR5] = dr_interception,
4041 [SVM_EXIT_READ_DR6] = dr_interception,
4042 [SVM_EXIT_READ_DR7] = dr_interception,
4043 [SVM_EXIT_WRITE_DR0] = dr_interception,
4044 [SVM_EXIT_WRITE_DR1] = dr_interception,
4045 [SVM_EXIT_WRITE_DR2] = dr_interception,
4046 [SVM_EXIT_WRITE_DR3] = dr_interception,
4047 [SVM_EXIT_WRITE_DR4] = dr_interception,
4048 [SVM_EXIT_WRITE_DR5] = dr_interception,
4049 [SVM_EXIT_WRITE_DR6] = dr_interception,
4050 [SVM_EXIT_WRITE_DR7] = dr_interception,
4051 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4052 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4053 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4054 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4055 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
4056 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4057 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4058 [SVM_EXIT_INTR] = intr_interception,
4059 [SVM_EXIT_NMI] = nmi_interception,
4060 [SVM_EXIT_SMI] = nop_on_interception,
4061 [SVM_EXIT_INIT] = nop_on_interception,
4062 [SVM_EXIT_VINTR] = interrupt_window_interception,
4063 [SVM_EXIT_RDPMC] = rdpmc_interception,
4064 [SVM_EXIT_CPUID] = cpuid_interception,
4065 [SVM_EXIT_IRET] = iret_interception,
4066 [SVM_EXIT_INVD] = emulate_on_interception,
4067 [SVM_EXIT_PAUSE] = pause_interception,
4068 [SVM_EXIT_HLT] = halt_interception,
4069 [SVM_EXIT_INVLPG] = invlpg_interception,
4070 [SVM_EXIT_INVLPGA] = invlpga_interception,
4071 [SVM_EXIT_IOIO] = io_interception,
4072 [SVM_EXIT_MSR] = msr_interception,
4073 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4074 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4075 [SVM_EXIT_VMRUN] = vmrun_interception,
4076 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4077 [SVM_EXIT_VMLOAD] = vmload_interception,
4078 [SVM_EXIT_VMSAVE] = vmsave_interception,
4079 [SVM_EXIT_STGI] = stgi_interception,
4080 [SVM_EXIT_CLGI] = clgi_interception,
4081 [SVM_EXIT_SKINIT] = skinit_interception,
4082 [SVM_EXIT_WBINVD] = wbinvd_interception,
4083 [SVM_EXIT_MONITOR] = monitor_interception,
4084 [SVM_EXIT_MWAIT] = mwait_interception,
4085 [SVM_EXIT_XSETBV] = xsetbv_interception,
4086 [SVM_EXIT_NPF] = pf_interception,
4087 [SVM_EXIT_RSM] = emulate_on_interception,
4088 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4089 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4092 static void dump_vmcb(struct kvm_vcpu *vcpu)
4094 struct vcpu_svm *svm = to_svm(vcpu);
4095 struct vmcb_control_area *control = &svm->vmcb->control;
4096 struct vmcb_save_area *save = &svm->vmcb->save;
4098 pr_err("VMCB Control Area:\n");
4099 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4100 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4101 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4102 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4103 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4104 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4105 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4106 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4107 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4108 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4109 pr_err("%-20s%d\n", "asid:", control->asid);
4110 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4111 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4112 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4113 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4114 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4115 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4116 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4117 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4118 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4119 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4120 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4121 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4122 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4123 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4124 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
4125 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4126 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4127 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4128 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4129 pr_err("VMCB State Save Area:\n");
4130 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4132 save->es.selector, save->es.attrib,
4133 save->es.limit, save->es.base);
4134 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4136 save->cs.selector, save->cs.attrib,
4137 save->cs.limit, save->cs.base);
4138 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4140 save->ss.selector, save->ss.attrib,
4141 save->ss.limit, save->ss.base);
4142 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4144 save->ds.selector, save->ds.attrib,
4145 save->ds.limit, save->ds.base);
4146 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4148 save->fs.selector, save->fs.attrib,
4149 save->fs.limit, save->fs.base);
4150 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4152 save->gs.selector, save->gs.attrib,
4153 save->gs.limit, save->gs.base);
4154 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4156 save->gdtr.selector, save->gdtr.attrib,
4157 save->gdtr.limit, save->gdtr.base);
4158 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4160 save->ldtr.selector, save->ldtr.attrib,
4161 save->ldtr.limit, save->ldtr.base);
4162 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4164 save->idtr.selector, save->idtr.attrib,
4165 save->idtr.limit, save->idtr.base);
4166 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4168 save->tr.selector, save->tr.attrib,
4169 save->tr.limit, save->tr.base);
4170 pr_err("cpl: %d efer: %016llx\n",
4171 save->cpl, save->efer);
4172 pr_err("%-15s %016llx %-13s %016llx\n",
4173 "cr0:", save->cr0, "cr2:", save->cr2);
4174 pr_err("%-15s %016llx %-13s %016llx\n",
4175 "cr3:", save->cr3, "cr4:", save->cr4);
4176 pr_err("%-15s %016llx %-13s %016llx\n",
4177 "dr6:", save->dr6, "dr7:", save->dr7);
4178 pr_err("%-15s %016llx %-13s %016llx\n",
4179 "rip:", save->rip, "rflags:", save->rflags);
4180 pr_err("%-15s %016llx %-13s %016llx\n",
4181 "rsp:", save->rsp, "rax:", save->rax);
4182 pr_err("%-15s %016llx %-13s %016llx\n",
4183 "star:", save->star, "lstar:", save->lstar);
4184 pr_err("%-15s %016llx %-13s %016llx\n",
4185 "cstar:", save->cstar, "sfmask:", save->sfmask);
4186 pr_err("%-15s %016llx %-13s %016llx\n",
4187 "kernel_gs_base:", save->kernel_gs_base,
4188 "sysenter_cs:", save->sysenter_cs);
4189 pr_err("%-15s %016llx %-13s %016llx\n",
4190 "sysenter_esp:", save->sysenter_esp,
4191 "sysenter_eip:", save->sysenter_eip);
4192 pr_err("%-15s %016llx %-13s %016llx\n",
4193 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4194 pr_err("%-15s %016llx %-13s %016llx\n",
4195 "br_from:", save->br_from, "br_to:", save->br_to);
4196 pr_err("%-15s %016llx %-13s %016llx\n",
4197 "excp_from:", save->last_excp_from,
4198 "excp_to:", save->last_excp_to);
4201 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4203 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4205 *info1 = control->exit_info_1;
4206 *info2 = control->exit_info_2;
4209 static int handle_exit(struct kvm_vcpu *vcpu)
4211 struct vcpu_svm *svm = to_svm(vcpu);
4212 struct kvm_run *kvm_run = vcpu->run;
4213 u32 exit_code = svm->vmcb->control.exit_code;
4215 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4217 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4218 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4220 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4222 if (unlikely(svm->nested.exit_required)) {
4223 nested_svm_vmexit(svm);
4224 svm->nested.exit_required = false;
4229 if (is_guest_mode(vcpu)) {
4232 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4233 svm->vmcb->control.exit_info_1,
4234 svm->vmcb->control.exit_info_2,
4235 svm->vmcb->control.exit_int_info,
4236 svm->vmcb->control.exit_int_info_err,
4239 vmexit = nested_svm_exit_special(svm);
4241 if (vmexit == NESTED_EXIT_CONTINUE)
4242 vmexit = nested_svm_exit_handled(svm);
4244 if (vmexit == NESTED_EXIT_DONE)
4248 svm_complete_interrupts(svm);
4250 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4251 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4252 kvm_run->fail_entry.hardware_entry_failure_reason
4253 = svm->vmcb->control.exit_code;
4254 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4259 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4260 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4261 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4262 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4263 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4265 __func__, svm->vmcb->control.exit_int_info,
4268 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4269 || !svm_exit_handlers[exit_code]) {
4270 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4271 kvm_queue_exception(vcpu, UD_VECTOR);
4275 return svm_exit_handlers[exit_code](svm);
4278 static void reload_tss(struct kvm_vcpu *vcpu)
4280 int cpu = raw_smp_processor_id();
4282 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4283 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4287 static void pre_svm_run(struct vcpu_svm *svm)
4289 int cpu = raw_smp_processor_id();
4291 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4293 /* FIXME: handle wraparound of asid_generation */
4294 if (svm->asid_generation != sd->asid_generation)
4298 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
4300 struct vcpu_svm *svm = to_svm(vcpu);
4302 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
4303 vcpu->arch.hflags |= HF_NMI_MASK;
4304 set_intercept(svm, INTERCEPT_IRET);
4305 ++vcpu->stat.nmi_injections;
4308 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
4310 struct vmcb_control_area *control;
4312 /* The following fields are ignored when AVIC is enabled */
4313 control = &svm->vmcb->control;
4314 control->int_vector = irq;
4315 control->int_ctl &= ~V_INTR_PRIO_MASK;
4316 control->int_ctl |= V_IRQ_MASK |
4317 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
4318 mark_dirty(svm->vmcb, VMCB_INTR);
4321 static void svm_set_irq(struct kvm_vcpu *vcpu)
4323 struct vcpu_svm *svm = to_svm(vcpu);
4325 BUG_ON(!(gif_set(svm)));
4327 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
4328 ++vcpu->stat.irq_injections;
4330 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
4331 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
4334 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
4336 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
4339 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
4341 struct vcpu_svm *svm = to_svm(vcpu);
4343 if (svm_nested_virtualize_tpr(vcpu) ||
4344 kvm_vcpu_apicv_active(vcpu))
4347 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4353 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4356 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
4361 static bool svm_get_enable_apicv(void)
4366 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
4370 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
4374 /* Note: Currently only used by Hyper-V. */
4375 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4377 struct vcpu_svm *svm = to_svm(vcpu);
4378 struct vmcb *vmcb = svm->vmcb;
4383 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
4384 mark_dirty(vmcb, VMCB_INTR);
4387 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
4392 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4397 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
4399 kvm_lapic_set_irr(vec, vcpu->arch.apic);
4400 smp_mb__after_atomic();
4402 if (avic_vcpu_is_running(vcpu))
4403 wrmsrl(SVM_AVIC_DOORBELL,
4404 kvm_cpu_get_apicid(vcpu->cpu));
4406 kvm_vcpu_wake_up(vcpu);
4409 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4411 unsigned long flags;
4412 struct amd_svm_iommu_ir *cur;
4414 spin_lock_irqsave(&svm->ir_list_lock, flags);
4415 list_for_each_entry(cur, &svm->ir_list, node) {
4416 if (cur->data != pi->ir_data)
4418 list_del(&cur->node);
4422 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4425 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4428 unsigned long flags;
4429 struct amd_svm_iommu_ir *ir;
4432 * In some cases, the existing irte is updaed and re-set,
4433 * so we need to check here if it's already been * added
4436 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
4437 struct kvm *kvm = svm->vcpu.kvm;
4438 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
4439 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
4440 struct vcpu_svm *prev_svm;
4447 prev_svm = to_svm(prev_vcpu);
4448 svm_ir_list_del(prev_svm, pi);
4452 * Allocating new amd_iommu_pi_data, which will get
4453 * add to the per-vcpu ir_list.
4455 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
4460 ir->data = pi->ir_data;
4462 spin_lock_irqsave(&svm->ir_list_lock, flags);
4463 list_add(&ir->node, &svm->ir_list);
4464 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4471 * The HW cannot support posting multicast/broadcast
4472 * interrupts to a vCPU. So, we still use legacy interrupt
4473 * remapping for these kind of interrupts.
4475 * For lowest-priority interrupts, we only support
4476 * those with single CPU as the destination, e.g. user
4477 * configures the interrupts via /proc/irq or uses
4478 * irqbalance to make the interrupts single-CPU.
4481 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
4482 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
4484 struct kvm_lapic_irq irq;
4485 struct kvm_vcpu *vcpu = NULL;
4487 kvm_set_msi_irq(kvm, e, &irq);
4489 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
4490 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4491 __func__, irq.vector);
4495 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
4497 *svm = to_svm(vcpu);
4498 vcpu_info->pi_desc_addr = page_to_phys((*svm)->avic_backing_page);
4499 vcpu_info->vector = irq.vector;
4505 * svm_update_pi_irte - set IRTE for Posted-Interrupts
4508 * @host_irq: host irq of the interrupt
4509 * @guest_irq: gsi of the interrupt
4510 * @set: set or unset PI
4511 * returns 0 on success, < 0 on failure
4513 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
4514 uint32_t guest_irq, bool set)
4516 struct kvm_kernel_irq_routing_entry *e;
4517 struct kvm_irq_routing_table *irq_rt;
4518 int idx, ret = -EINVAL;
4520 if (!kvm_arch_has_assigned_device(kvm) ||
4521 !irq_remapping_cap(IRQ_POSTING_CAP))
4524 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4525 __func__, host_irq, guest_irq, set);
4527 idx = srcu_read_lock(&kvm->irq_srcu);
4528 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
4529 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
4531 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
4532 struct vcpu_data vcpu_info;
4533 struct vcpu_svm *svm = NULL;
4535 if (e->type != KVM_IRQ_ROUTING_MSI)
4539 * Here, we setup with legacy mode in the following cases:
4540 * 1. When cannot target interrupt to a specific vcpu.
4541 * 2. Unsetting posted interrupt.
4542 * 3. APIC virtialization is disabled for the vcpu.
4544 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
4545 kvm_vcpu_apicv_active(&svm->vcpu)) {
4546 struct amd_iommu_pi_data pi;
4548 /* Try to enable guest_mode in IRTE */
4549 pi.base = page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK;
4550 pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
4552 pi.is_guest_mode = true;
4553 pi.vcpu_data = &vcpu_info;
4554 ret = irq_set_vcpu_affinity(host_irq, &pi);
4557 * Here, we successfully setting up vcpu affinity in
4558 * IOMMU guest mode. Now, we need to store the posted
4559 * interrupt information in a per-vcpu ir_list so that
4560 * we can reference to them directly when we update vcpu
4561 * scheduling information in IOMMU irte.
4563 if (!ret && pi.is_guest_mode)
4564 svm_ir_list_add(svm, &pi);
4566 /* Use legacy mode in IRTE */
4567 struct amd_iommu_pi_data pi;
4570 * Here, pi is used to:
4571 * - Tell IOMMU to use legacy mode for this interrupt.
4572 * - Retrieve ga_tag of prior interrupt remapping data.
4574 pi.is_guest_mode = false;
4575 ret = irq_set_vcpu_affinity(host_irq, &pi);
4578 * Check if the posted interrupt was previously
4579 * setup with the guest_mode by checking if the ga_tag
4580 * was cached. If so, we need to clean up the per-vcpu
4583 if (!ret && pi.prev_ga_tag) {
4584 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
4585 struct kvm_vcpu *vcpu;
4587 vcpu = kvm_get_vcpu_by_id(kvm, id);
4589 svm_ir_list_del(to_svm(vcpu), &pi);
4594 trace_kvm_pi_irte_update(svm->vcpu.vcpu_id,
4597 vcpu_info.pi_desc_addr, set);
4601 pr_err("%s: failed to update PI IRTE\n", __func__);
4608 srcu_read_unlock(&kvm->irq_srcu, idx);
4612 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
4614 struct vcpu_svm *svm = to_svm(vcpu);
4615 struct vmcb *vmcb = svm->vmcb;
4617 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
4618 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
4619 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
4624 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
4626 struct vcpu_svm *svm = to_svm(vcpu);
4628 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
4631 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4633 struct vcpu_svm *svm = to_svm(vcpu);
4636 svm->vcpu.arch.hflags |= HF_NMI_MASK;
4637 set_intercept(svm, INTERCEPT_IRET);
4639 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
4640 clr_intercept(svm, INTERCEPT_IRET);
4644 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
4646 struct vcpu_svm *svm = to_svm(vcpu);
4647 struct vmcb *vmcb = svm->vmcb;
4650 if (!gif_set(svm) ||
4651 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
4654 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
4656 if (is_guest_mode(vcpu))
4657 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
4662 static void enable_irq_window(struct kvm_vcpu *vcpu)
4664 struct vcpu_svm *svm = to_svm(vcpu);
4666 if (kvm_vcpu_apicv_active(vcpu))
4670 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4671 * 1, because that's a separate STGI/VMRUN intercept. The next time we
4672 * get that intercept, this function will be called again though and
4673 * we'll get the vintr intercept.
4675 if (gif_set(svm) && nested_svm_intr(svm)) {
4677 svm_inject_irq(svm, 0x0);
4681 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4683 struct vcpu_svm *svm = to_svm(vcpu);
4685 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
4687 return; /* IRET will cause a vm exit */
4690 * Something prevents NMI from been injected. Single step over possible
4691 * problem (IRET or exception injection or interrupt shadow)
4693 svm->nmi_singlestep = true;
4694 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
4697 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
4702 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
4704 struct vcpu_svm *svm = to_svm(vcpu);
4706 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
4707 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
4709 svm->asid_generation--;
4712 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
4716 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
4718 struct vcpu_svm *svm = to_svm(vcpu);
4720 if (svm_nested_virtualize_tpr(vcpu))
4723 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
4724 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
4725 kvm_set_cr8(vcpu, cr8);
4729 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
4731 struct vcpu_svm *svm = to_svm(vcpu);
4734 if (svm_nested_virtualize_tpr(vcpu) ||
4735 kvm_vcpu_apicv_active(vcpu))
4738 cr8 = kvm_get_cr8(vcpu);
4739 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
4740 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
4743 static void svm_complete_interrupts(struct vcpu_svm *svm)
4747 u32 exitintinfo = svm->vmcb->control.exit_int_info;
4748 unsigned int3_injected = svm->int3_injected;
4750 svm->int3_injected = 0;
4753 * If we've made progress since setting HF_IRET_MASK, we've
4754 * executed an IRET and can allow NMI injection.
4756 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
4757 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
4758 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
4759 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4762 svm->vcpu.arch.nmi_injected = false;
4763 kvm_clear_exception_queue(&svm->vcpu);
4764 kvm_clear_interrupt_queue(&svm->vcpu);
4766 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
4769 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4771 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
4772 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
4775 case SVM_EXITINTINFO_TYPE_NMI:
4776 svm->vcpu.arch.nmi_injected = true;
4778 case SVM_EXITINTINFO_TYPE_EXEPT:
4780 * In case of software exceptions, do not reinject the vector,
4781 * but re-execute the instruction instead. Rewind RIP first
4782 * if we emulated INT3 before.
4784 if (kvm_exception_is_soft(vector)) {
4785 if (vector == BP_VECTOR && int3_injected &&
4786 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
4787 kvm_rip_write(&svm->vcpu,
4788 kvm_rip_read(&svm->vcpu) -
4792 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
4793 u32 err = svm->vmcb->control.exit_int_info_err;
4794 kvm_requeue_exception_e(&svm->vcpu, vector, err);
4797 kvm_requeue_exception(&svm->vcpu, vector);
4799 case SVM_EXITINTINFO_TYPE_INTR:
4800 kvm_queue_interrupt(&svm->vcpu, vector, false);
4807 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
4809 struct vcpu_svm *svm = to_svm(vcpu);
4810 struct vmcb_control_area *control = &svm->vmcb->control;
4812 control->exit_int_info = control->event_inj;
4813 control->exit_int_info_err = control->event_inj_err;
4814 control->event_inj = 0;
4815 svm_complete_interrupts(svm);
4818 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
4820 struct vcpu_svm *svm = to_svm(vcpu);
4822 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4823 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4824 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4827 * A vmexit emulation is required before the vcpu can be executed
4830 if (unlikely(svm->nested.exit_required))
4835 sync_lapic_to_cr8(vcpu);
4837 svm->vmcb->save.cr2 = vcpu->arch.cr2;
4844 "push %%" _ASM_BP "; \n\t"
4845 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4846 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4847 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4848 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4849 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4850 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4851 #ifdef CONFIG_X86_64
4852 "mov %c[r8](%[svm]), %%r8 \n\t"
4853 "mov %c[r9](%[svm]), %%r9 \n\t"
4854 "mov %c[r10](%[svm]), %%r10 \n\t"
4855 "mov %c[r11](%[svm]), %%r11 \n\t"
4856 "mov %c[r12](%[svm]), %%r12 \n\t"
4857 "mov %c[r13](%[svm]), %%r13 \n\t"
4858 "mov %c[r14](%[svm]), %%r14 \n\t"
4859 "mov %c[r15](%[svm]), %%r15 \n\t"
4862 /* Enter guest mode */
4863 "push %%" _ASM_AX " \n\t"
4864 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4865 __ex(SVM_VMLOAD) "\n\t"
4866 __ex(SVM_VMRUN) "\n\t"
4867 __ex(SVM_VMSAVE) "\n\t"
4868 "pop %%" _ASM_AX " \n\t"
4870 /* Save guest registers, load host registers */
4871 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4872 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4873 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4874 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4875 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4876 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4877 #ifdef CONFIG_X86_64
4878 "mov %%r8, %c[r8](%[svm]) \n\t"
4879 "mov %%r9, %c[r9](%[svm]) \n\t"
4880 "mov %%r10, %c[r10](%[svm]) \n\t"
4881 "mov %%r11, %c[r11](%[svm]) \n\t"
4882 "mov %%r12, %c[r12](%[svm]) \n\t"
4883 "mov %%r13, %c[r13](%[svm]) \n\t"
4884 "mov %%r14, %c[r14](%[svm]) \n\t"
4885 "mov %%r15, %c[r15](%[svm]) \n\t"
4890 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4891 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4892 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4893 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4894 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4895 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4896 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4897 #ifdef CONFIG_X86_64
4898 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4899 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4900 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4901 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4902 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4903 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4904 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4905 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4908 #ifdef CONFIG_X86_64
4909 , "rbx", "rcx", "rdx", "rsi", "rdi"
4910 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4912 , "ebx", "ecx", "edx", "esi", "edi"
4916 #ifdef CONFIG_X86_64
4917 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4919 loadsegment(fs, svm->host.fs);
4920 #ifndef CONFIG_X86_32_LAZY_GS
4921 loadsegment(gs, svm->host.gs);
4927 local_irq_disable();
4929 vcpu->arch.cr2 = svm->vmcb->save.cr2;
4930 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4931 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4932 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4934 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4935 kvm_before_handle_nmi(&svm->vcpu);
4939 /* Any pending NMI will happen here */
4941 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4942 kvm_after_handle_nmi(&svm->vcpu);
4944 sync_cr8_to_lapic(vcpu);
4948 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4950 /* if exit due to PF check for async PF */
4951 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4952 svm->apf_reason = kvm_read_and_reset_pf_reason();
4955 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4956 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4960 * We need to handle MC intercepts here before the vcpu has a chance to
4961 * change the physical cpu
4963 if (unlikely(svm->vmcb->control.exit_code ==
4964 SVM_EXIT_EXCP_BASE + MC_VECTOR))
4965 svm_handle_mce(svm);
4967 mark_all_clean(svm->vmcb);
4970 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4972 struct vcpu_svm *svm = to_svm(vcpu);
4974 svm->vmcb->save.cr3 = root;
4975 mark_dirty(svm->vmcb, VMCB_CR);
4976 svm_flush_tlb(vcpu);
4979 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4981 struct vcpu_svm *svm = to_svm(vcpu);
4983 svm->vmcb->control.nested_cr3 = root;
4984 mark_dirty(svm->vmcb, VMCB_NPT);
4986 /* Also sync guest cr3 here in case we live migrate */
4987 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4988 mark_dirty(svm->vmcb, VMCB_CR);
4990 svm_flush_tlb(vcpu);
4993 static int is_disabled(void)
4997 rdmsrl(MSR_VM_CR, vm_cr);
4998 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5005 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5008 * Patch in the VMMCALL instruction:
5010 hypercall[0] = 0x0f;
5011 hypercall[1] = 0x01;
5012 hypercall[2] = 0xd9;
5015 static void svm_check_processor_compat(void *rtn)
5020 static bool svm_cpu_has_accelerated_tpr(void)
5025 static bool svm_has_high_real_mode_segbase(void)
5030 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5035 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5037 struct vcpu_svm *svm = to_svm(vcpu);
5038 struct kvm_cpuid_entry2 *entry;
5040 /* Update nrips enabled cache */
5041 svm->nrips_enabled = !!guest_cpuid_has_nrips(&svm->vcpu);
5043 if (!kvm_vcpu_apicv_active(vcpu))
5046 entry = kvm_find_cpuid_entry(vcpu, 1, 0);
5048 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5051 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5056 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5060 entry->ecx |= (1 << 2); /* Set SVM bit */
5063 entry->eax = 1; /* SVM revision 1 */
5064 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5065 ASID emulation to nested SVM */
5066 entry->ecx = 0; /* Reserved */
5067 entry->edx = 0; /* Per default do not support any
5068 additional features */
5070 /* Support next_rip if host supports it */
5071 if (boot_cpu_has(X86_FEATURE_NRIPS))
5072 entry->edx |= SVM_FEATURE_NRIP;
5074 /* Support NPT for the guest if enabled */
5076 entry->edx |= SVM_FEATURE_NPT;
5082 static int svm_get_lpage_level(void)
5084 return PT_PDPE_LEVEL;
5087 static bool svm_rdtscp_supported(void)
5089 return boot_cpu_has(X86_FEATURE_RDTSCP);
5092 static bool svm_invpcid_supported(void)
5097 static bool svm_mpx_supported(void)
5102 static bool svm_xsaves_supported(void)
5107 static bool svm_has_wbinvd_exit(void)
5112 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
5114 struct vcpu_svm *svm = to_svm(vcpu);
5116 set_exception_intercept(svm, NM_VECTOR);
5117 update_cr0_intercept(svm);
5120 #define PRE_EX(exit) { .exit_code = (exit), \
5121 .stage = X86_ICPT_PRE_EXCEPT, }
5122 #define POST_EX(exit) { .exit_code = (exit), \
5123 .stage = X86_ICPT_POST_EXCEPT, }
5124 #define POST_MEM(exit) { .exit_code = (exit), \
5125 .stage = X86_ICPT_POST_MEMACCESS, }
5127 static const struct __x86_intercept {
5129 enum x86_intercept_stage stage;
5130 } x86_intercept_map[] = {
5131 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5132 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5133 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5134 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5135 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5136 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5137 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5138 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5139 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5140 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5141 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5142 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5143 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5144 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5145 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5146 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5147 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5148 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5149 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5150 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5151 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5152 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5153 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5154 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5155 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5156 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5157 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5158 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
5159 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
5160 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
5161 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
5162 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
5163 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
5164 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
5165 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
5166 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
5167 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
5168 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
5169 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
5170 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
5171 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
5172 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
5173 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
5174 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
5175 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
5176 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
5183 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5184 struct x86_instruction_info *info,
5185 enum x86_intercept_stage stage)
5187 struct vcpu_svm *svm = to_svm(vcpu);
5188 int vmexit, ret = X86EMUL_CONTINUE;
5189 struct __x86_intercept icpt_info;
5190 struct vmcb *vmcb = svm->vmcb;
5192 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5195 icpt_info = x86_intercept_map[info->intercept];
5197 if (stage != icpt_info.stage)
5200 switch (icpt_info.exit_code) {
5201 case SVM_EXIT_READ_CR0:
5202 if (info->intercept == x86_intercept_cr_read)
5203 icpt_info.exit_code += info->modrm_reg;
5205 case SVM_EXIT_WRITE_CR0: {
5206 unsigned long cr0, val;
5209 if (info->intercept == x86_intercept_cr_write)
5210 icpt_info.exit_code += info->modrm_reg;
5212 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
5213 info->intercept == x86_intercept_clts)
5216 intercept = svm->nested.intercept;
5218 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
5221 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
5222 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
5224 if (info->intercept == x86_intercept_lmsw) {
5227 /* lmsw can't clear PE - catch this here */
5228 if (cr0 & X86_CR0_PE)
5233 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
5237 case SVM_EXIT_READ_DR0:
5238 case SVM_EXIT_WRITE_DR0:
5239 icpt_info.exit_code += info->modrm_reg;
5242 if (info->intercept == x86_intercept_wrmsr)
5243 vmcb->control.exit_info_1 = 1;
5245 vmcb->control.exit_info_1 = 0;
5247 case SVM_EXIT_PAUSE:
5249 * We get this for NOP only, but pause
5250 * is rep not, check this here
5252 if (info->rep_prefix != REPE_PREFIX)
5254 case SVM_EXIT_IOIO: {
5258 if (info->intercept == x86_intercept_in ||
5259 info->intercept == x86_intercept_ins) {
5260 exit_info = ((info->src_val & 0xffff) << 16) |
5262 bytes = info->dst_bytes;
5264 exit_info = (info->dst_val & 0xffff) << 16;
5265 bytes = info->src_bytes;
5268 if (info->intercept == x86_intercept_outs ||
5269 info->intercept == x86_intercept_ins)
5270 exit_info |= SVM_IOIO_STR_MASK;
5272 if (info->rep_prefix)
5273 exit_info |= SVM_IOIO_REP_MASK;
5275 bytes = min(bytes, 4u);
5277 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
5279 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
5281 vmcb->control.exit_info_1 = exit_info;
5282 vmcb->control.exit_info_2 = info->next_rip;
5290 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5291 if (static_cpu_has(X86_FEATURE_NRIPS))
5292 vmcb->control.next_rip = info->next_rip;
5293 vmcb->control.exit_code = icpt_info.exit_code;
5294 vmexit = nested_svm_exit_handled(svm);
5296 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
5303 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
5307 * We must have an instruction with interrupts enabled, so
5308 * the timer interrupt isn't delayed by the interrupt shadow.
5311 local_irq_disable();
5314 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
5318 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
5320 if (avic_handle_apic_id_update(vcpu) != 0)
5322 if (avic_handle_dfr_update(vcpu) != 0)
5324 avic_handle_ldr_update(vcpu);
5327 static struct kvm_x86_ops svm_x86_ops = {
5328 .cpu_has_kvm_support = has_svm,
5329 .disabled_by_bios = is_disabled,
5330 .hardware_setup = svm_hardware_setup,
5331 .hardware_unsetup = svm_hardware_unsetup,
5332 .check_processor_compatibility = svm_check_processor_compat,
5333 .hardware_enable = svm_hardware_enable,
5334 .hardware_disable = svm_hardware_disable,
5335 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
5336 .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
5338 .vcpu_create = svm_create_vcpu,
5339 .vcpu_free = svm_free_vcpu,
5340 .vcpu_reset = svm_vcpu_reset,
5342 .vm_init = avic_vm_init,
5343 .vm_destroy = avic_vm_destroy,
5345 .prepare_guest_switch = svm_prepare_guest_switch,
5346 .vcpu_load = svm_vcpu_load,
5347 .vcpu_put = svm_vcpu_put,
5348 .vcpu_blocking = svm_vcpu_blocking,
5349 .vcpu_unblocking = svm_vcpu_unblocking,
5351 .update_bp_intercept = update_bp_intercept,
5352 .get_msr = svm_get_msr,
5353 .set_msr = svm_set_msr,
5354 .get_segment_base = svm_get_segment_base,
5355 .get_segment = svm_get_segment,
5356 .set_segment = svm_set_segment,
5357 .get_cpl = svm_get_cpl,
5358 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
5359 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
5360 .decache_cr3 = svm_decache_cr3,
5361 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
5362 .set_cr0 = svm_set_cr0,
5363 .set_cr3 = svm_set_cr3,
5364 .set_cr4 = svm_set_cr4,
5365 .set_efer = svm_set_efer,
5366 .get_idt = svm_get_idt,
5367 .set_idt = svm_set_idt,
5368 .get_gdt = svm_get_gdt,
5369 .set_gdt = svm_set_gdt,
5370 .get_dr6 = svm_get_dr6,
5371 .set_dr6 = svm_set_dr6,
5372 .set_dr7 = svm_set_dr7,
5373 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
5374 .cache_reg = svm_cache_reg,
5375 .get_rflags = svm_get_rflags,
5376 .set_rflags = svm_set_rflags,
5378 .get_pkru = svm_get_pkru,
5380 .fpu_activate = svm_fpu_activate,
5381 .fpu_deactivate = svm_fpu_deactivate,
5383 .tlb_flush = svm_flush_tlb,
5385 .run = svm_vcpu_run,
5386 .handle_exit = handle_exit,
5387 .skip_emulated_instruction = skip_emulated_instruction,
5388 .set_interrupt_shadow = svm_set_interrupt_shadow,
5389 .get_interrupt_shadow = svm_get_interrupt_shadow,
5390 .patch_hypercall = svm_patch_hypercall,
5391 .set_irq = svm_set_irq,
5392 .set_nmi = svm_inject_nmi,
5393 .queue_exception = svm_queue_exception,
5394 .cancel_injection = svm_cancel_injection,
5395 .interrupt_allowed = svm_interrupt_allowed,
5396 .nmi_allowed = svm_nmi_allowed,
5397 .get_nmi_mask = svm_get_nmi_mask,
5398 .set_nmi_mask = svm_set_nmi_mask,
5399 .enable_nmi_window = enable_nmi_window,
5400 .enable_irq_window = enable_irq_window,
5401 .update_cr8_intercept = update_cr8_intercept,
5402 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
5403 .get_enable_apicv = svm_get_enable_apicv,
5404 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
5405 .load_eoi_exitmap = svm_load_eoi_exitmap,
5406 .sync_pir_to_irr = svm_sync_pir_to_irr,
5407 .hwapic_irr_update = svm_hwapic_irr_update,
5408 .hwapic_isr_update = svm_hwapic_isr_update,
5409 .apicv_post_state_restore = avic_post_state_restore,
5411 .set_tss_addr = svm_set_tss_addr,
5412 .get_tdp_level = get_npt_level,
5413 .get_mt_mask = svm_get_mt_mask,
5415 .get_exit_info = svm_get_exit_info,
5417 .get_lpage_level = svm_get_lpage_level,
5419 .cpuid_update = svm_cpuid_update,
5421 .rdtscp_supported = svm_rdtscp_supported,
5422 .invpcid_supported = svm_invpcid_supported,
5423 .mpx_supported = svm_mpx_supported,
5424 .xsaves_supported = svm_xsaves_supported,
5426 .set_supported_cpuid = svm_set_supported_cpuid,
5428 .has_wbinvd_exit = svm_has_wbinvd_exit,
5430 .read_tsc_offset = svm_read_tsc_offset,
5431 .write_tsc_offset = svm_write_tsc_offset,
5432 .adjust_tsc_offset_guest = svm_adjust_tsc_offset_guest,
5433 .read_l1_tsc = svm_read_l1_tsc,
5435 .set_tdp_cr3 = set_tdp_cr3,
5437 .check_intercept = svm_check_intercept,
5438 .handle_external_intr = svm_handle_external_intr,
5440 .sched_in = svm_sched_in,
5442 .pmu_ops = &amd_pmu_ops,
5443 .deliver_posted_interrupt = svm_deliver_avic_intr,
5444 .update_pi_irte = svm_update_pi_irte,
5447 static int __init svm_init(void)
5449 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
5450 __alignof__(struct vcpu_svm), THIS_MODULE);
5453 static void __exit svm_exit(void)
5458 module_init(svm_init)
5459 module_exit(svm_exit)