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>
39 #include <linux/frame.h>
42 #include <asm/perf_event.h>
43 #include <asm/tlbflush.h>
45 #include <asm/debugreg.h>
46 #include <asm/kvm_para.h>
47 #include <asm/irq_remapping.h>
49 #include <asm/virtext.h>
52 #define __ex(x) __kvm_handle_fault_on_reboot(x)
54 MODULE_AUTHOR("Qumranet");
55 MODULE_LICENSE("GPL");
57 static const struct x86_cpu_id svm_cpu_id[] = {
58 X86_FEATURE_MATCH(X86_FEATURE_SVM),
61 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
63 #define IOPM_ALLOC_ORDER 2
64 #define MSRPM_ALLOC_ORDER 1
66 #define SEG_TYPE_LDT 2
67 #define SEG_TYPE_BUSY_TSS16 3
69 #define SVM_FEATURE_NPT (1 << 0)
70 #define SVM_FEATURE_LBRV (1 << 1)
71 #define SVM_FEATURE_SVML (1 << 2)
72 #define SVM_FEATURE_NRIP (1 << 3)
73 #define SVM_FEATURE_TSC_RATE (1 << 4)
74 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
75 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
76 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
77 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
79 #define SVM_AVIC_DOORBELL 0xc001011b
81 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
82 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
83 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
85 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
87 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
88 #define TSC_RATIO_MIN 0x0000000000000001ULL
89 #define TSC_RATIO_MAX 0x000000ffffffffffULL
91 #define AVIC_HPA_MASK ~((0xFFFULL << 52) | 0xFFF)
94 * 0xff is broadcast, so the max index allowed for physical APIC ID
95 * table is 0xfe. APIC IDs above 0xff are reserved.
97 #define AVIC_MAX_PHYSICAL_ID_COUNT 255
99 #define AVIC_UNACCEL_ACCESS_WRITE_MASK 1
100 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK 0xFF0
101 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK 0xFFFFFFFF
103 /* AVIC GATAG is encoded using VM and VCPU IDs */
104 #define AVIC_VCPU_ID_BITS 8
105 #define AVIC_VCPU_ID_MASK ((1 << AVIC_VCPU_ID_BITS) - 1)
107 #define AVIC_VM_ID_BITS 24
108 #define AVIC_VM_ID_NR (1 << AVIC_VM_ID_BITS)
109 #define AVIC_VM_ID_MASK ((1 << AVIC_VM_ID_BITS) - 1)
111 #define AVIC_GATAG(x, y) (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
112 (y & AVIC_VCPU_ID_MASK))
113 #define AVIC_GATAG_TO_VMID(x) ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
114 #define AVIC_GATAG_TO_VCPUID(x) (x & AVIC_VCPU_ID_MASK)
116 static bool erratum_383_found __read_mostly;
118 static const u32 host_save_user_msrs[] = {
120 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
123 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
127 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
131 struct nested_state {
137 /* These are the merged vectors */
140 /* gpa pointers to the real vectors */
144 /* A VMEXIT is required but not yet emulated */
147 /* cache for intercepts of the guest */
150 u32 intercept_exceptions;
153 /* Nested Paging related state */
157 #define MSRPM_OFFSETS 16
158 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
161 * Set osvw_len to higher value when updated Revision Guides
162 * are published and we know what the new status bits are
164 static uint64_t osvw_len = 4, osvw_status;
167 struct kvm_vcpu vcpu;
169 unsigned long vmcb_pa;
170 struct svm_cpu_data *svm_data;
171 uint64_t asid_generation;
172 uint64_t sysenter_esp;
173 uint64_t sysenter_eip;
178 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
190 struct nested_state nested;
193 u64 nmi_singlestep_guest_rflags;
195 unsigned int3_injected;
196 unsigned long int3_rip;
199 /* cached guest cpuid flags for faster access */
200 bool nrips_enabled : 1;
203 struct page *avic_backing_page;
204 u64 *avic_physical_id_cache;
205 bool avic_is_running;
208 * Per-vcpu list of struct amd_svm_iommu_ir:
209 * This is used mainly to store interrupt remapping information used
210 * when update the vcpu affinity. This avoids the need to scan for
211 * IRTE and try to match ga_tag in the IOMMU driver.
213 struct list_head ir_list;
214 spinlock_t ir_list_lock;
218 * This is a wrapper of struct amd_iommu_ir_data.
220 struct amd_svm_iommu_ir {
221 struct list_head node; /* Used by SVM for per-vcpu ir_list */
222 void *data; /* Storing pointer to struct amd_ir_data */
225 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
226 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
228 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
229 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
230 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
231 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
233 static DEFINE_PER_CPU(u64, current_tsc_ratio);
234 #define TSC_RATIO_DEFAULT 0x0100000000ULL
236 #define MSR_INVALID 0xffffffffU
238 static const struct svm_direct_access_msrs {
239 u32 index; /* Index of the MSR */
240 bool always; /* True if intercept is always on */
241 } direct_access_msrs[] = {
242 { .index = MSR_STAR, .always = true },
243 { .index = MSR_IA32_SYSENTER_CS, .always = true },
245 { .index = MSR_GS_BASE, .always = true },
246 { .index = MSR_FS_BASE, .always = true },
247 { .index = MSR_KERNEL_GS_BASE, .always = true },
248 { .index = MSR_LSTAR, .always = true },
249 { .index = MSR_CSTAR, .always = true },
250 { .index = MSR_SYSCALL_MASK, .always = true },
252 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
253 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
254 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
255 { .index = MSR_IA32_LASTINTTOIP, .always = false },
256 { .index = MSR_INVALID, .always = false },
259 /* enable NPT for AMD64 and X86 with PAE */
260 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
261 static bool npt_enabled = true;
263 static bool npt_enabled;
266 /* allow nested paging (virtualized MMU) for all guests */
267 static int npt = true;
268 module_param(npt, int, S_IRUGO);
270 /* allow nested virtualization in KVM/SVM */
271 static int nested = true;
272 module_param(nested, int, S_IRUGO);
274 /* enable / disable AVIC */
276 #ifdef CONFIG_X86_LOCAL_APIC
277 module_param(avic, int, S_IRUGO);
280 /* AVIC VM ID bit masks and lock */
281 static DECLARE_BITMAP(avic_vm_id_bitmap, AVIC_VM_ID_NR);
282 static DEFINE_SPINLOCK(avic_vm_id_lock);
284 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
285 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
286 static void svm_complete_interrupts(struct vcpu_svm *svm);
288 static int nested_svm_exit_handled(struct vcpu_svm *svm);
289 static int nested_svm_intercept(struct vcpu_svm *svm);
290 static int nested_svm_vmexit(struct vcpu_svm *svm);
291 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
292 bool has_error_code, u32 error_code);
295 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
296 pause filter count */
297 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
298 VMCB_ASID, /* ASID */
299 VMCB_INTR, /* int_ctl, int_vector */
300 VMCB_NPT, /* npt_en, nCR3, gPAT */
301 VMCB_CR, /* CR0, CR3, CR4, EFER */
302 VMCB_DR, /* DR6, DR7 */
303 VMCB_DT, /* GDT, IDT */
304 VMCB_SEG, /* CS, DS, SS, ES, CPL */
305 VMCB_CR2, /* CR2 only */
306 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
307 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
308 * AVIC PHYSICAL_TABLE pointer,
309 * AVIC LOGICAL_TABLE pointer
314 /* TPR and CR2 are always written before VMRUN */
315 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
317 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
319 static inline void mark_all_dirty(struct vmcb *vmcb)
321 vmcb->control.clean = 0;
324 static inline void mark_all_clean(struct vmcb *vmcb)
326 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
327 & ~VMCB_ALWAYS_DIRTY_MASK;
330 static inline void mark_dirty(struct vmcb *vmcb, int bit)
332 vmcb->control.clean &= ~(1 << bit);
335 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
337 return container_of(vcpu, struct vcpu_svm, vcpu);
340 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
342 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
343 mark_dirty(svm->vmcb, VMCB_AVIC);
346 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
348 struct vcpu_svm *svm = to_svm(vcpu);
349 u64 *entry = svm->avic_physical_id_cache;
354 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
357 static void recalc_intercepts(struct vcpu_svm *svm)
359 struct vmcb_control_area *c, *h;
360 struct nested_state *g;
362 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
364 if (!is_guest_mode(&svm->vcpu))
367 c = &svm->vmcb->control;
368 h = &svm->nested.hsave->control;
371 c->intercept_cr = h->intercept_cr | g->intercept_cr;
372 c->intercept_dr = h->intercept_dr | g->intercept_dr;
373 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
374 c->intercept = h->intercept | g->intercept;
377 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
379 if (is_guest_mode(&svm->vcpu))
380 return svm->nested.hsave;
385 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
387 struct vmcb *vmcb = get_host_vmcb(svm);
389 vmcb->control.intercept_cr |= (1U << bit);
391 recalc_intercepts(svm);
394 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
396 struct vmcb *vmcb = get_host_vmcb(svm);
398 vmcb->control.intercept_cr &= ~(1U << bit);
400 recalc_intercepts(svm);
403 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
405 struct vmcb *vmcb = get_host_vmcb(svm);
407 return vmcb->control.intercept_cr & (1U << bit);
410 static inline void set_dr_intercepts(struct vcpu_svm *svm)
412 struct vmcb *vmcb = get_host_vmcb(svm);
414 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
415 | (1 << INTERCEPT_DR1_READ)
416 | (1 << INTERCEPT_DR2_READ)
417 | (1 << INTERCEPT_DR3_READ)
418 | (1 << INTERCEPT_DR4_READ)
419 | (1 << INTERCEPT_DR5_READ)
420 | (1 << INTERCEPT_DR6_READ)
421 | (1 << INTERCEPT_DR7_READ)
422 | (1 << INTERCEPT_DR0_WRITE)
423 | (1 << INTERCEPT_DR1_WRITE)
424 | (1 << INTERCEPT_DR2_WRITE)
425 | (1 << INTERCEPT_DR3_WRITE)
426 | (1 << INTERCEPT_DR4_WRITE)
427 | (1 << INTERCEPT_DR5_WRITE)
428 | (1 << INTERCEPT_DR6_WRITE)
429 | (1 << INTERCEPT_DR7_WRITE);
431 recalc_intercepts(svm);
434 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
436 struct vmcb *vmcb = get_host_vmcb(svm);
438 vmcb->control.intercept_dr = 0;
440 recalc_intercepts(svm);
443 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
445 struct vmcb *vmcb = get_host_vmcb(svm);
447 vmcb->control.intercept_exceptions |= (1U << bit);
449 recalc_intercepts(svm);
452 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
454 struct vmcb *vmcb = get_host_vmcb(svm);
456 vmcb->control.intercept_exceptions &= ~(1U << bit);
458 recalc_intercepts(svm);
461 static inline void set_intercept(struct vcpu_svm *svm, int bit)
463 struct vmcb *vmcb = get_host_vmcb(svm);
465 vmcb->control.intercept |= (1ULL << bit);
467 recalc_intercepts(svm);
470 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
472 struct vmcb *vmcb = get_host_vmcb(svm);
474 vmcb->control.intercept &= ~(1ULL << bit);
476 recalc_intercepts(svm);
479 static inline void enable_gif(struct vcpu_svm *svm)
481 svm->vcpu.arch.hflags |= HF_GIF_MASK;
484 static inline void disable_gif(struct vcpu_svm *svm)
486 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
489 static inline bool gif_set(struct vcpu_svm *svm)
491 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
494 static unsigned long iopm_base;
496 struct kvm_ldttss_desc {
499 unsigned base1:8, type:5, dpl:2, p:1;
500 unsigned limit1:4, zero0:3, g:1, base2:8;
503 } __attribute__((packed));
505 struct svm_cpu_data {
511 struct kvm_ldttss_desc *tss_desc;
513 struct page *save_area;
516 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
518 struct svm_init_data {
523 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
525 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
526 #define MSRS_RANGE_SIZE 2048
527 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
529 static u32 svm_msrpm_offset(u32 msr)
534 for (i = 0; i < NUM_MSR_MAPS; i++) {
535 if (msr < msrpm_ranges[i] ||
536 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
539 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
540 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
542 /* Now we have the u8 offset - but need the u32 offset */
546 /* MSR not in any range */
550 #define MAX_INST_SIZE 15
552 static inline void clgi(void)
554 asm volatile (__ex(SVM_CLGI));
557 static inline void stgi(void)
559 asm volatile (__ex(SVM_STGI));
562 static inline void invlpga(unsigned long addr, u32 asid)
564 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
567 static int get_npt_level(void)
570 return PT64_ROOT_LEVEL;
572 return PT32E_ROOT_LEVEL;
576 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
578 vcpu->arch.efer = efer;
579 if (!npt_enabled && !(efer & EFER_LMA))
582 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
583 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
586 static int is_external_interrupt(u32 info)
588 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
589 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
592 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
594 struct vcpu_svm *svm = to_svm(vcpu);
597 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
598 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
602 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
604 struct vcpu_svm *svm = to_svm(vcpu);
607 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
609 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
613 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
615 struct vcpu_svm *svm = to_svm(vcpu);
617 if (svm->vmcb->control.next_rip != 0) {
618 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
619 svm->next_rip = svm->vmcb->control.next_rip;
622 if (!svm->next_rip) {
623 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
625 printk(KERN_DEBUG "%s: NOP\n", __func__);
628 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
629 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
630 __func__, kvm_rip_read(vcpu), svm->next_rip);
632 kvm_rip_write(vcpu, svm->next_rip);
633 svm_set_interrupt_shadow(vcpu, 0);
636 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
637 bool has_error_code, u32 error_code,
640 struct vcpu_svm *svm = to_svm(vcpu);
643 * If we are within a nested VM we'd better #VMEXIT and let the guest
644 * handle the exception
647 nested_svm_check_exception(svm, nr, has_error_code, error_code))
650 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
651 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
654 * For guest debugging where we have to reinject #BP if some
655 * INT3 is guest-owned:
656 * Emulate nRIP by moving RIP forward. Will fail if injection
657 * raises a fault that is not intercepted. Still better than
658 * failing in all cases.
660 skip_emulated_instruction(&svm->vcpu);
661 rip = kvm_rip_read(&svm->vcpu);
662 svm->int3_rip = rip + svm->vmcb->save.cs.base;
663 svm->int3_injected = rip - old_rip;
666 svm->vmcb->control.event_inj = nr
668 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
669 | SVM_EVTINJ_TYPE_EXEPT;
670 svm->vmcb->control.event_inj_err = error_code;
673 static void svm_init_erratum_383(void)
679 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
682 /* Use _safe variants to not break nested virtualization */
683 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
689 low = lower_32_bits(val);
690 high = upper_32_bits(val);
692 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
694 erratum_383_found = true;
697 static void svm_init_osvw(struct kvm_vcpu *vcpu)
700 * Guests should see errata 400 and 415 as fixed (assuming that
701 * HLT and IO instructions are intercepted).
703 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
704 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
707 * By increasing VCPU's osvw.length to 3 we are telling the guest that
708 * all osvw.status bits inside that length, including bit 0 (which is
709 * reserved for erratum 298), are valid. However, if host processor's
710 * osvw_len is 0 then osvw_status[0] carries no information. We need to
711 * be conservative here and therefore we tell the guest that erratum 298
712 * is present (because we really don't know).
714 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
715 vcpu->arch.osvw.status |= 1;
718 static int has_svm(void)
722 if (!cpu_has_svm(&msg)) {
723 printk(KERN_INFO "has_svm: %s\n", msg);
730 static void svm_hardware_disable(void)
732 /* Make sure we clean up behind us */
733 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
734 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
738 amd_pmu_disable_virt();
741 static int svm_hardware_enable(void)
744 struct svm_cpu_data *sd;
746 struct desc_struct *gdt;
747 int me = raw_smp_processor_id();
749 rdmsrl(MSR_EFER, efer);
750 if (efer & EFER_SVME)
754 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
757 sd = per_cpu(svm_data, me);
759 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
763 sd->asid_generation = 1;
764 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
765 sd->next_asid = sd->max_asid + 1;
767 gdt = get_current_gdt_rw();
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);
968 static void disable_nmi_singlestep(struct vcpu_svm *svm)
970 svm->nmi_singlestep = false;
971 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
972 /* Clear our flags if they were not set by the guest */
973 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
974 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
975 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
976 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
981 * This hash table is used to map VM_ID to a struct kvm_arch,
982 * when handling AMD IOMMU GALOG notification to schedule in
985 #define SVM_VM_DATA_HASH_BITS 8
986 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
987 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
990 * This function is called from IOMMU driver to notify
991 * SVM to schedule in a particular vCPU of a particular VM.
993 static int avic_ga_log_notifier(u32 ga_tag)
996 struct kvm_arch *ka = NULL;
997 struct kvm_vcpu *vcpu = NULL;
998 u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
999 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1001 pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1003 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1004 hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) {
1005 struct kvm *kvm = container_of(ka, struct kvm, arch);
1006 struct kvm_arch *vm_data = &kvm->arch;
1008 if (vm_data->avic_vm_id != vm_id)
1010 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1013 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1019 * At this point, the IOMMU should have already set the pending
1020 * bit in the vAPIC backing page. So, we just need to schedule
1023 if (vcpu->mode == OUTSIDE_GUEST_MODE)
1024 kvm_vcpu_wake_up(vcpu);
1029 static __init int svm_hardware_setup(void)
1032 struct page *iopm_pages;
1036 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1041 iopm_va = page_address(iopm_pages);
1042 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1043 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1045 init_msrpm_offsets();
1047 if (boot_cpu_has(X86_FEATURE_NX))
1048 kvm_enable_efer_bits(EFER_NX);
1050 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1051 kvm_enable_efer_bits(EFER_FFXSR);
1053 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1054 kvm_has_tsc_control = true;
1055 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1056 kvm_tsc_scaling_ratio_frac_bits = 32;
1060 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1061 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1064 for_each_possible_cpu(cpu) {
1065 r = svm_cpu_init(cpu);
1070 if (!boot_cpu_has(X86_FEATURE_NPT))
1071 npt_enabled = false;
1073 if (npt_enabled && !npt) {
1074 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1075 npt_enabled = false;
1079 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1086 !boot_cpu_has(X86_FEATURE_AVIC) ||
1087 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1090 pr_info("AVIC enabled\n");
1092 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1099 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1104 static __exit void svm_hardware_unsetup(void)
1108 for_each_possible_cpu(cpu)
1109 svm_cpu_uninit(cpu);
1111 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1115 static void init_seg(struct vmcb_seg *seg)
1118 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1119 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1120 seg->limit = 0xffff;
1124 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1127 seg->attrib = SVM_SELECTOR_P_MASK | type;
1128 seg->limit = 0xffff;
1132 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1134 struct vcpu_svm *svm = to_svm(vcpu);
1135 u64 g_tsc_offset = 0;
1137 if (is_guest_mode(vcpu)) {
1138 g_tsc_offset = svm->vmcb->control.tsc_offset -
1139 svm->nested.hsave->control.tsc_offset;
1140 svm->nested.hsave->control.tsc_offset = offset;
1142 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1143 svm->vmcb->control.tsc_offset,
1146 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1148 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1151 static void avic_init_vmcb(struct vcpu_svm *svm)
1153 struct vmcb *vmcb = svm->vmcb;
1154 struct kvm_arch *vm_data = &svm->vcpu.kvm->arch;
1155 phys_addr_t bpa = page_to_phys(svm->avic_backing_page);
1156 phys_addr_t lpa = page_to_phys(vm_data->avic_logical_id_table_page);
1157 phys_addr_t ppa = page_to_phys(vm_data->avic_physical_id_table_page);
1159 vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1160 vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1161 vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1162 vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1163 vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1164 svm->vcpu.arch.apicv_active = true;
1167 static void init_vmcb(struct vcpu_svm *svm)
1169 struct vmcb_control_area *control = &svm->vmcb->control;
1170 struct vmcb_save_area *save = &svm->vmcb->save;
1172 svm->vcpu.arch.hflags = 0;
1174 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1175 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1176 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1177 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1178 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1179 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1180 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1181 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1183 set_dr_intercepts(svm);
1185 set_exception_intercept(svm, PF_VECTOR);
1186 set_exception_intercept(svm, UD_VECTOR);
1187 set_exception_intercept(svm, MC_VECTOR);
1188 set_exception_intercept(svm, AC_VECTOR);
1189 set_exception_intercept(svm, DB_VECTOR);
1191 set_intercept(svm, INTERCEPT_INTR);
1192 set_intercept(svm, INTERCEPT_NMI);
1193 set_intercept(svm, INTERCEPT_SMI);
1194 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1195 set_intercept(svm, INTERCEPT_RDPMC);
1196 set_intercept(svm, INTERCEPT_CPUID);
1197 set_intercept(svm, INTERCEPT_INVD);
1198 set_intercept(svm, INTERCEPT_HLT);
1199 set_intercept(svm, INTERCEPT_INVLPG);
1200 set_intercept(svm, INTERCEPT_INVLPGA);
1201 set_intercept(svm, INTERCEPT_IOIO_PROT);
1202 set_intercept(svm, INTERCEPT_MSR_PROT);
1203 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1204 set_intercept(svm, INTERCEPT_SHUTDOWN);
1205 set_intercept(svm, INTERCEPT_VMRUN);
1206 set_intercept(svm, INTERCEPT_VMMCALL);
1207 set_intercept(svm, INTERCEPT_VMLOAD);
1208 set_intercept(svm, INTERCEPT_VMSAVE);
1209 set_intercept(svm, INTERCEPT_STGI);
1210 set_intercept(svm, INTERCEPT_CLGI);
1211 set_intercept(svm, INTERCEPT_SKINIT);
1212 set_intercept(svm, INTERCEPT_WBINVD);
1213 set_intercept(svm, INTERCEPT_XSETBV);
1215 if (!kvm_mwait_in_guest()) {
1216 set_intercept(svm, INTERCEPT_MONITOR);
1217 set_intercept(svm, INTERCEPT_MWAIT);
1220 control->iopm_base_pa = iopm_base;
1221 control->msrpm_base_pa = __pa(svm->msrpm);
1222 control->int_ctl = V_INTR_MASKING_MASK;
1224 init_seg(&save->es);
1225 init_seg(&save->ss);
1226 init_seg(&save->ds);
1227 init_seg(&save->fs);
1228 init_seg(&save->gs);
1230 save->cs.selector = 0xf000;
1231 save->cs.base = 0xffff0000;
1232 /* Executable/Readable Code Segment */
1233 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1234 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1235 save->cs.limit = 0xffff;
1237 save->gdtr.limit = 0xffff;
1238 save->idtr.limit = 0xffff;
1240 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1241 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1243 svm_set_efer(&svm->vcpu, 0);
1244 save->dr6 = 0xffff0ff0;
1245 kvm_set_rflags(&svm->vcpu, 2);
1246 save->rip = 0x0000fff0;
1247 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1250 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1251 * It also updates the guest-visible cr0 value.
1253 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1254 kvm_mmu_reset_context(&svm->vcpu);
1256 save->cr4 = X86_CR4_PAE;
1260 /* Setup VMCB for Nested Paging */
1261 control->nested_ctl = 1;
1262 clr_intercept(svm, INTERCEPT_INVLPG);
1263 clr_exception_intercept(svm, PF_VECTOR);
1264 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1265 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1266 save->g_pat = svm->vcpu.arch.pat;
1270 svm->asid_generation = 0;
1272 svm->nested.vmcb = 0;
1273 svm->vcpu.arch.hflags = 0;
1275 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1276 control->pause_filter_count = 3000;
1277 set_intercept(svm, INTERCEPT_PAUSE);
1281 avic_init_vmcb(svm);
1283 mark_all_dirty(svm->vmcb);
1289 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1292 u64 *avic_physical_id_table;
1293 struct kvm_arch *vm_data = &vcpu->kvm->arch;
1295 if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1298 avic_physical_id_table = page_address(vm_data->avic_physical_id_table_page);
1300 return &avic_physical_id_table[index];
1305 * AVIC hardware walks the nested page table to check permissions,
1306 * but does not use the SPA address specified in the leaf page
1307 * table entry since it uses address in the AVIC_BACKING_PAGE pointer
1308 * field of the VMCB. Therefore, we set up the
1309 * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1311 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1313 struct kvm *kvm = vcpu->kvm;
1316 if (kvm->arch.apic_access_page_done)
1319 ret = x86_set_memory_region(kvm,
1320 APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1321 APIC_DEFAULT_PHYS_BASE,
1326 kvm->arch.apic_access_page_done = true;
1330 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1333 u64 *entry, new_entry;
1334 int id = vcpu->vcpu_id;
1335 struct vcpu_svm *svm = to_svm(vcpu);
1337 ret = avic_init_access_page(vcpu);
1341 if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1344 if (!svm->vcpu.arch.apic->regs)
1347 svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1349 /* Setting AVIC backing page address in the phy APIC ID table */
1350 entry = avic_get_physical_id_entry(vcpu, id);
1354 new_entry = READ_ONCE(*entry);
1355 new_entry = (page_to_phys(svm->avic_backing_page) &
1356 AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1357 AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
1358 WRITE_ONCE(*entry, new_entry);
1360 svm->avic_physical_id_cache = entry;
1365 static inline int avic_get_next_vm_id(void)
1369 spin_lock(&avic_vm_id_lock);
1371 /* AVIC VM ID is one-based. */
1372 id = find_next_zero_bit(avic_vm_id_bitmap, AVIC_VM_ID_NR, 1);
1373 if (id <= AVIC_VM_ID_MASK)
1374 __set_bit(id, avic_vm_id_bitmap);
1378 spin_unlock(&avic_vm_id_lock);
1382 static inline int avic_free_vm_id(int id)
1384 if (id <= 0 || id > AVIC_VM_ID_MASK)
1387 spin_lock(&avic_vm_id_lock);
1388 __clear_bit(id, avic_vm_id_bitmap);
1389 spin_unlock(&avic_vm_id_lock);
1393 static void avic_vm_destroy(struct kvm *kvm)
1395 unsigned long flags;
1396 struct kvm_arch *vm_data = &kvm->arch;
1401 avic_free_vm_id(vm_data->avic_vm_id);
1403 if (vm_data->avic_logical_id_table_page)
1404 __free_page(vm_data->avic_logical_id_table_page);
1405 if (vm_data->avic_physical_id_table_page)
1406 __free_page(vm_data->avic_physical_id_table_page);
1408 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1409 hash_del(&vm_data->hnode);
1410 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1413 static int avic_vm_init(struct kvm *kvm)
1415 unsigned long flags;
1416 int vm_id, err = -ENOMEM;
1417 struct kvm_arch *vm_data = &kvm->arch;
1418 struct page *p_page;
1419 struct page *l_page;
1424 vm_id = avic_get_next_vm_id();
1427 vm_data->avic_vm_id = (u32)vm_id;
1429 /* Allocating physical APIC ID table (4KB) */
1430 p_page = alloc_page(GFP_KERNEL);
1434 vm_data->avic_physical_id_table_page = p_page;
1435 clear_page(page_address(p_page));
1437 /* Allocating logical APIC ID table (4KB) */
1438 l_page = alloc_page(GFP_KERNEL);
1442 vm_data->avic_logical_id_table_page = l_page;
1443 clear_page(page_address(l_page));
1445 spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1446 hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id);
1447 spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1452 avic_vm_destroy(kvm);
1457 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1460 unsigned long flags;
1461 struct amd_svm_iommu_ir *ir;
1462 struct vcpu_svm *svm = to_svm(vcpu);
1464 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1468 * Here, we go through the per-vcpu ir_list to update all existing
1469 * interrupt remapping table entry targeting this vcpu.
1471 spin_lock_irqsave(&svm->ir_list_lock, flags);
1473 if (list_empty(&svm->ir_list))
1476 list_for_each_entry(ir, &svm->ir_list, node) {
1477 ret = amd_iommu_update_ga(cpu, r, ir->data);
1482 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
1486 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1489 /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1490 int h_physical_id = kvm_cpu_get_apicid(cpu);
1491 struct vcpu_svm *svm = to_svm(vcpu);
1493 if (!kvm_vcpu_apicv_active(vcpu))
1496 if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
1499 entry = READ_ONCE(*(svm->avic_physical_id_cache));
1500 WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
1502 entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
1503 entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
1505 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1506 if (svm->avic_is_running)
1507 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1509 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1510 avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
1511 svm->avic_is_running);
1514 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
1517 struct vcpu_svm *svm = to_svm(vcpu);
1519 if (!kvm_vcpu_apicv_active(vcpu))
1522 entry = READ_ONCE(*(svm->avic_physical_id_cache));
1523 if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
1524 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
1526 entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1527 WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1531 * This function is called during VCPU halt/unhalt.
1533 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
1535 struct vcpu_svm *svm = to_svm(vcpu);
1537 svm->avic_is_running = is_run;
1539 avic_vcpu_load(vcpu, vcpu->cpu);
1541 avic_vcpu_put(vcpu);
1544 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1546 struct vcpu_svm *svm = to_svm(vcpu);
1551 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1552 MSR_IA32_APICBASE_ENABLE;
1553 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1554 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1558 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1559 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1561 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1562 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1565 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1567 struct vcpu_svm *svm;
1569 struct page *msrpm_pages;
1570 struct page *hsave_page;
1571 struct page *nested_msrpm_pages;
1574 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1580 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1585 page = alloc_page(GFP_KERNEL);
1589 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1593 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1594 if (!nested_msrpm_pages)
1597 hsave_page = alloc_page(GFP_KERNEL);
1602 err = avic_init_backing_page(&svm->vcpu);
1606 INIT_LIST_HEAD(&svm->ir_list);
1607 spin_lock_init(&svm->ir_list_lock);
1610 /* We initialize this flag to true to make sure that the is_running
1611 * bit would be set the first time the vcpu is loaded.
1613 svm->avic_is_running = true;
1615 svm->nested.hsave = page_address(hsave_page);
1617 svm->msrpm = page_address(msrpm_pages);
1618 svm_vcpu_init_msrpm(svm->msrpm);
1620 svm->nested.msrpm = page_address(nested_msrpm_pages);
1621 svm_vcpu_init_msrpm(svm->nested.msrpm);
1623 svm->vmcb = page_address(page);
1624 clear_page(svm->vmcb);
1625 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1626 svm->asid_generation = 0;
1629 svm_init_osvw(&svm->vcpu);
1634 __free_page(hsave_page);
1636 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1638 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1642 kvm_vcpu_uninit(&svm->vcpu);
1644 kmem_cache_free(kvm_vcpu_cache, svm);
1646 return ERR_PTR(err);
1649 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1651 struct vcpu_svm *svm = to_svm(vcpu);
1653 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1654 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1655 __free_page(virt_to_page(svm->nested.hsave));
1656 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1657 kvm_vcpu_uninit(vcpu);
1658 kmem_cache_free(kvm_vcpu_cache, svm);
1661 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1663 struct vcpu_svm *svm = to_svm(vcpu);
1666 if (unlikely(cpu != vcpu->cpu)) {
1667 svm->asid_generation = 0;
1668 mark_all_dirty(svm->vmcb);
1671 #ifdef CONFIG_X86_64
1672 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1674 savesegment(fs, svm->host.fs);
1675 savesegment(gs, svm->host.gs);
1676 svm->host.ldt = kvm_read_ldt();
1678 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1679 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1681 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1682 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1683 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1684 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1685 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1688 /* This assumes that the kernel never uses MSR_TSC_AUX */
1689 if (static_cpu_has(X86_FEATURE_RDTSCP))
1690 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1692 avic_vcpu_load(vcpu, cpu);
1695 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1697 struct vcpu_svm *svm = to_svm(vcpu);
1700 avic_vcpu_put(vcpu);
1702 ++vcpu->stat.host_state_reload;
1703 kvm_load_ldt(svm->host.ldt);
1704 #ifdef CONFIG_X86_64
1705 loadsegment(fs, svm->host.fs);
1706 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
1707 load_gs_index(svm->host.gs);
1709 #ifdef CONFIG_X86_32_LAZY_GS
1710 loadsegment(gs, svm->host.gs);
1713 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1714 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1717 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
1719 avic_set_running(vcpu, false);
1722 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
1724 avic_set_running(vcpu, true);
1727 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1729 struct vcpu_svm *svm = to_svm(vcpu);
1730 unsigned long rflags = svm->vmcb->save.rflags;
1732 if (svm->nmi_singlestep) {
1733 /* Hide our flags if they were not set by the guest */
1734 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1735 rflags &= ~X86_EFLAGS_TF;
1736 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1737 rflags &= ~X86_EFLAGS_RF;
1742 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1744 if (to_svm(vcpu)->nmi_singlestep)
1745 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1748 * Any change of EFLAGS.VM is accompanied by a reload of SS
1749 * (caused by either a task switch or an inter-privilege IRET),
1750 * so we do not need to update the CPL here.
1752 to_svm(vcpu)->vmcb->save.rflags = rflags;
1755 static u32 svm_get_pkru(struct kvm_vcpu *vcpu)
1760 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1763 case VCPU_EXREG_PDPTR:
1764 BUG_ON(!npt_enabled);
1765 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1772 static void svm_set_vintr(struct vcpu_svm *svm)
1774 set_intercept(svm, INTERCEPT_VINTR);
1777 static void svm_clear_vintr(struct vcpu_svm *svm)
1779 clr_intercept(svm, INTERCEPT_VINTR);
1782 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1784 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1787 case VCPU_SREG_CS: return &save->cs;
1788 case VCPU_SREG_DS: return &save->ds;
1789 case VCPU_SREG_ES: return &save->es;
1790 case VCPU_SREG_FS: return &save->fs;
1791 case VCPU_SREG_GS: return &save->gs;
1792 case VCPU_SREG_SS: return &save->ss;
1793 case VCPU_SREG_TR: return &save->tr;
1794 case VCPU_SREG_LDTR: return &save->ldtr;
1800 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1802 struct vmcb_seg *s = svm_seg(vcpu, seg);
1807 static void svm_get_segment(struct kvm_vcpu *vcpu,
1808 struct kvm_segment *var, int seg)
1810 struct vmcb_seg *s = svm_seg(vcpu, seg);
1812 var->base = s->base;
1813 var->limit = s->limit;
1814 var->selector = s->selector;
1815 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1816 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1817 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1818 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1819 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1820 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1821 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1824 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1825 * However, the SVM spec states that the G bit is not observed by the
1826 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1827 * So let's synthesize a legal G bit for all segments, this helps
1828 * running KVM nested. It also helps cross-vendor migration, because
1829 * Intel's vmentry has a check on the 'G' bit.
1831 var->g = s->limit > 0xfffff;
1834 * AMD's VMCB does not have an explicit unusable field, so emulate it
1835 * for cross vendor migration purposes by "not present"
1837 var->unusable = !var->present;
1842 * Work around a bug where the busy flag in the tr selector
1852 * The accessed bit must always be set in the segment
1853 * descriptor cache, although it can be cleared in the
1854 * descriptor, the cached bit always remains at 1. Since
1855 * Intel has a check on this, set it here to support
1856 * cross-vendor migration.
1863 * On AMD CPUs sometimes the DB bit in the segment
1864 * descriptor is left as 1, although the whole segment has
1865 * been made unusable. Clear it here to pass an Intel VMX
1866 * entry check when cross vendor migrating.
1870 /* This is symmetric with svm_set_segment() */
1871 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1876 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1878 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1883 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1885 struct vcpu_svm *svm = to_svm(vcpu);
1887 dt->size = svm->vmcb->save.idtr.limit;
1888 dt->address = svm->vmcb->save.idtr.base;
1891 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1893 struct vcpu_svm *svm = to_svm(vcpu);
1895 svm->vmcb->save.idtr.limit = dt->size;
1896 svm->vmcb->save.idtr.base = dt->address ;
1897 mark_dirty(svm->vmcb, VMCB_DT);
1900 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1902 struct vcpu_svm *svm = to_svm(vcpu);
1904 dt->size = svm->vmcb->save.gdtr.limit;
1905 dt->address = svm->vmcb->save.gdtr.base;
1908 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1910 struct vcpu_svm *svm = to_svm(vcpu);
1912 svm->vmcb->save.gdtr.limit = dt->size;
1913 svm->vmcb->save.gdtr.base = dt->address ;
1914 mark_dirty(svm->vmcb, VMCB_DT);
1917 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1921 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1925 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1929 static void update_cr0_intercept(struct vcpu_svm *svm)
1931 ulong gcr0 = svm->vcpu.arch.cr0;
1932 u64 *hcr0 = &svm->vmcb->save.cr0;
1934 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1935 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1937 mark_dirty(svm->vmcb, VMCB_CR);
1939 if (gcr0 == *hcr0) {
1940 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1941 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1943 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1944 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1948 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1950 struct vcpu_svm *svm = to_svm(vcpu);
1952 #ifdef CONFIG_X86_64
1953 if (vcpu->arch.efer & EFER_LME) {
1954 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1955 vcpu->arch.efer |= EFER_LMA;
1956 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1959 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1960 vcpu->arch.efer &= ~EFER_LMA;
1961 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1965 vcpu->arch.cr0 = cr0;
1968 cr0 |= X86_CR0_PG | X86_CR0_WP;
1971 * re-enable caching here because the QEMU bios
1972 * does not do it - this results in some delay at
1975 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1976 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1977 svm->vmcb->save.cr0 = cr0;
1978 mark_dirty(svm->vmcb, VMCB_CR);
1979 update_cr0_intercept(svm);
1982 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1984 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1985 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1987 if (cr4 & X86_CR4_VMXE)
1990 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1991 svm_flush_tlb(vcpu);
1993 vcpu->arch.cr4 = cr4;
1996 cr4 |= host_cr4_mce;
1997 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1998 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2002 static void svm_set_segment(struct kvm_vcpu *vcpu,
2003 struct kvm_segment *var, int seg)
2005 struct vcpu_svm *svm = to_svm(vcpu);
2006 struct vmcb_seg *s = svm_seg(vcpu, seg);
2008 s->base = var->base;
2009 s->limit = var->limit;
2010 s->selector = var->selector;
2011 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2012 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2013 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2014 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2015 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2016 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2017 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2018 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 /* This is symmetric with svm_get_segment() */
2028 svm->vmcb->save.cpl = (var->dpl & 3);
2030 mark_dirty(svm->vmcb, VMCB_SEG);
2033 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2035 struct vcpu_svm *svm = to_svm(vcpu);
2037 clr_exception_intercept(svm, BP_VECTOR);
2039 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2040 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2041 set_exception_intercept(svm, BP_VECTOR);
2043 vcpu->guest_debug = 0;
2046 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2048 if (sd->next_asid > sd->max_asid) {
2049 ++sd->asid_generation;
2051 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2054 svm->asid_generation = sd->asid_generation;
2055 svm->vmcb->control.asid = sd->next_asid++;
2057 mark_dirty(svm->vmcb, VMCB_ASID);
2060 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2062 return to_svm(vcpu)->vmcb->save.dr6;
2065 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2067 struct vcpu_svm *svm = to_svm(vcpu);
2069 svm->vmcb->save.dr6 = value;
2070 mark_dirty(svm->vmcb, VMCB_DR);
2073 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2075 struct vcpu_svm *svm = to_svm(vcpu);
2077 get_debugreg(vcpu->arch.db[0], 0);
2078 get_debugreg(vcpu->arch.db[1], 1);
2079 get_debugreg(vcpu->arch.db[2], 2);
2080 get_debugreg(vcpu->arch.db[3], 3);
2081 vcpu->arch.dr6 = svm_get_dr6(vcpu);
2082 vcpu->arch.dr7 = svm->vmcb->save.dr7;
2084 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2085 set_dr_intercepts(svm);
2088 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2090 struct vcpu_svm *svm = to_svm(vcpu);
2092 svm->vmcb->save.dr7 = value;
2093 mark_dirty(svm->vmcb, VMCB_DR);
2096 static int pf_interception(struct vcpu_svm *svm)
2098 u64 fault_address = svm->vmcb->control.exit_info_2;
2102 switch (svm->apf_reason) {
2104 error_code = svm->vmcb->control.exit_info_1;
2106 trace_kvm_page_fault(fault_address, error_code);
2107 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
2108 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
2109 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2110 svm->vmcb->control.insn_bytes,
2111 svm->vmcb->control.insn_len);
2113 case KVM_PV_REASON_PAGE_NOT_PRESENT:
2114 svm->apf_reason = 0;
2115 local_irq_disable();
2116 kvm_async_pf_task_wait(fault_address);
2119 case KVM_PV_REASON_PAGE_READY:
2120 svm->apf_reason = 0;
2121 local_irq_disable();
2122 kvm_async_pf_task_wake(fault_address);
2129 static int db_interception(struct vcpu_svm *svm)
2131 struct kvm_run *kvm_run = svm->vcpu.run;
2133 if (!(svm->vcpu.guest_debug &
2134 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2135 !svm->nmi_singlestep) {
2136 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2140 if (svm->nmi_singlestep) {
2141 disable_nmi_singlestep(svm);
2144 if (svm->vcpu.guest_debug &
2145 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2146 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2147 kvm_run->debug.arch.pc =
2148 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2149 kvm_run->debug.arch.exception = DB_VECTOR;
2156 static int bp_interception(struct vcpu_svm *svm)
2158 struct kvm_run *kvm_run = svm->vcpu.run;
2160 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2161 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2162 kvm_run->debug.arch.exception = BP_VECTOR;
2166 static int ud_interception(struct vcpu_svm *svm)
2170 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
2171 if (er != EMULATE_DONE)
2172 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2176 static int ac_interception(struct vcpu_svm *svm)
2178 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2182 static bool is_erratum_383(void)
2187 if (!erratum_383_found)
2190 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2194 /* Bit 62 may or may not be set for this mce */
2195 value &= ~(1ULL << 62);
2197 if (value != 0xb600000000010015ULL)
2200 /* Clear MCi_STATUS registers */
2201 for (i = 0; i < 6; ++i)
2202 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2204 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2208 value &= ~(1ULL << 2);
2209 low = lower_32_bits(value);
2210 high = upper_32_bits(value);
2212 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2215 /* Flush tlb to evict multi-match entries */
2221 static void svm_handle_mce(struct vcpu_svm *svm)
2223 if (is_erratum_383()) {
2225 * Erratum 383 triggered. Guest state is corrupt so kill the
2228 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2230 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2236 * On an #MC intercept the MCE handler is not called automatically in
2237 * the host. So do it by hand here.
2241 /* not sure if we ever come back to this point */
2246 static int mc_interception(struct vcpu_svm *svm)
2251 static int shutdown_interception(struct vcpu_svm *svm)
2253 struct kvm_run *kvm_run = svm->vcpu.run;
2256 * VMCB is undefined after a SHUTDOWN intercept
2257 * so reinitialize it.
2259 clear_page(svm->vmcb);
2262 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2266 static int io_interception(struct vcpu_svm *svm)
2268 struct kvm_vcpu *vcpu = &svm->vcpu;
2269 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2270 int size, in, string;
2273 ++svm->vcpu.stat.io_exits;
2274 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2275 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2277 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2279 port = io_info >> 16;
2280 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2281 svm->next_rip = svm->vmcb->control.exit_info_2;
2282 skip_emulated_instruction(&svm->vcpu);
2284 return in ? kvm_fast_pio_in(vcpu, size, port)
2285 : kvm_fast_pio_out(vcpu, size, port);
2288 static int nmi_interception(struct vcpu_svm *svm)
2293 static int intr_interception(struct vcpu_svm *svm)
2295 ++svm->vcpu.stat.irq_exits;
2299 static int nop_on_interception(struct vcpu_svm *svm)
2304 static int halt_interception(struct vcpu_svm *svm)
2306 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2307 return kvm_emulate_halt(&svm->vcpu);
2310 static int vmmcall_interception(struct vcpu_svm *svm)
2312 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2313 return kvm_emulate_hypercall(&svm->vcpu);
2316 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2318 struct vcpu_svm *svm = to_svm(vcpu);
2320 return svm->nested.nested_cr3;
2323 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2325 struct vcpu_svm *svm = to_svm(vcpu);
2326 u64 cr3 = svm->nested.nested_cr3;
2330 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2331 offset_in_page(cr3) + index * 8, 8);
2337 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2340 struct vcpu_svm *svm = to_svm(vcpu);
2342 svm->vmcb->control.nested_cr3 = root;
2343 mark_dirty(svm->vmcb, VMCB_NPT);
2344 svm_flush_tlb(vcpu);
2347 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2348 struct x86_exception *fault)
2350 struct vcpu_svm *svm = to_svm(vcpu);
2352 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2354 * TODO: track the cause of the nested page fault, and
2355 * correctly fill in the high bits of exit_info_1.
2357 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2358 svm->vmcb->control.exit_code_hi = 0;
2359 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2360 svm->vmcb->control.exit_info_2 = fault->address;
2363 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2364 svm->vmcb->control.exit_info_1 |= fault->error_code;
2367 * The present bit is always zero for page structure faults on real
2370 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2371 svm->vmcb->control.exit_info_1 &= ~1;
2373 nested_svm_vmexit(svm);
2376 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2378 WARN_ON(mmu_is_nested(vcpu));
2379 kvm_init_shadow_mmu(vcpu);
2380 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2381 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2382 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2383 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2384 vcpu->arch.mmu.shadow_root_level = get_npt_level();
2385 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2386 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2389 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2391 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2394 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2396 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2397 !is_paging(&svm->vcpu)) {
2398 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2402 if (svm->vmcb->save.cpl) {
2403 kvm_inject_gp(&svm->vcpu, 0);
2410 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2411 bool has_error_code, u32 error_code)
2415 if (!is_guest_mode(&svm->vcpu))
2418 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2419 svm->vmcb->control.exit_code_hi = 0;
2420 svm->vmcb->control.exit_info_1 = error_code;
2421 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2423 vmexit = nested_svm_intercept(svm);
2424 if (vmexit == NESTED_EXIT_DONE)
2425 svm->nested.exit_required = true;
2430 /* This function returns true if it is save to enable the irq window */
2431 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2433 if (!is_guest_mode(&svm->vcpu))
2436 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2439 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2443 * if vmexit was already requested (by intercepted exception
2444 * for instance) do not overwrite it with "external interrupt"
2447 if (svm->nested.exit_required)
2450 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2451 svm->vmcb->control.exit_info_1 = 0;
2452 svm->vmcb->control.exit_info_2 = 0;
2454 if (svm->nested.intercept & 1ULL) {
2456 * The #vmexit can't be emulated here directly because this
2457 * code path runs with irqs and preemption disabled. A
2458 * #vmexit emulation might sleep. Only signal request for
2461 svm->nested.exit_required = true;
2462 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2469 /* This function returns true if it is save to enable the nmi window */
2470 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2472 if (!is_guest_mode(&svm->vcpu))
2475 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2478 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2479 svm->nested.exit_required = true;
2484 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2490 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2491 if (is_error_page(page))
2499 kvm_inject_gp(&svm->vcpu, 0);
2504 static void nested_svm_unmap(struct page *page)
2507 kvm_release_page_dirty(page);
2510 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2512 unsigned port, size, iopm_len;
2517 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2518 return NESTED_EXIT_HOST;
2520 port = svm->vmcb->control.exit_info_1 >> 16;
2521 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2522 SVM_IOIO_SIZE_SHIFT;
2523 gpa = svm->nested.vmcb_iopm + (port / 8);
2524 start_bit = port % 8;
2525 iopm_len = (start_bit + size > 8) ? 2 : 1;
2526 mask = (0xf >> (4 - size)) << start_bit;
2529 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2530 return NESTED_EXIT_DONE;
2532 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2535 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2537 u32 offset, msr, value;
2540 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2541 return NESTED_EXIT_HOST;
2543 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2544 offset = svm_msrpm_offset(msr);
2545 write = svm->vmcb->control.exit_info_1 & 1;
2546 mask = 1 << ((2 * (msr & 0xf)) + write);
2548 if (offset == MSR_INVALID)
2549 return NESTED_EXIT_DONE;
2551 /* Offset is in 32 bit units but need in 8 bit units */
2554 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2555 return NESTED_EXIT_DONE;
2557 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2560 /* DB exceptions for our internal use must not cause vmexit */
2561 static int nested_svm_intercept_db(struct vcpu_svm *svm)
2565 /* if we're not singlestepping, it's not ours */
2566 if (!svm->nmi_singlestep)
2567 return NESTED_EXIT_DONE;
2569 /* if it's not a singlestep exception, it's not ours */
2570 if (kvm_get_dr(&svm->vcpu, 6, &dr6))
2571 return NESTED_EXIT_DONE;
2572 if (!(dr6 & DR6_BS))
2573 return NESTED_EXIT_DONE;
2575 /* if the guest is singlestepping, it should get the vmexit */
2576 if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
2577 disable_nmi_singlestep(svm);
2578 return NESTED_EXIT_DONE;
2581 /* it's ours, the nested hypervisor must not see this one */
2582 return NESTED_EXIT_HOST;
2585 static int nested_svm_exit_special(struct vcpu_svm *svm)
2587 u32 exit_code = svm->vmcb->control.exit_code;
2589 switch (exit_code) {
2592 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2593 return NESTED_EXIT_HOST;
2595 /* For now we are always handling NPFs when using them */
2597 return NESTED_EXIT_HOST;
2599 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2600 /* When we're shadowing, trap PFs, but not async PF */
2601 if (!npt_enabled && svm->apf_reason == 0)
2602 return NESTED_EXIT_HOST;
2608 return NESTED_EXIT_CONTINUE;
2612 * If this function returns true, this #vmexit was already handled
2614 static int nested_svm_intercept(struct vcpu_svm *svm)
2616 u32 exit_code = svm->vmcb->control.exit_code;
2617 int vmexit = NESTED_EXIT_HOST;
2619 switch (exit_code) {
2621 vmexit = nested_svm_exit_handled_msr(svm);
2624 vmexit = nested_svm_intercept_ioio(svm);
2626 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2627 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2628 if (svm->nested.intercept_cr & bit)
2629 vmexit = NESTED_EXIT_DONE;
2632 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2633 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2634 if (svm->nested.intercept_dr & bit)
2635 vmexit = NESTED_EXIT_DONE;
2638 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2639 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2640 if (svm->nested.intercept_exceptions & excp_bits) {
2641 if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
2642 vmexit = nested_svm_intercept_db(svm);
2644 vmexit = NESTED_EXIT_DONE;
2646 /* async page fault always cause vmexit */
2647 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2648 svm->apf_reason != 0)
2649 vmexit = NESTED_EXIT_DONE;
2652 case SVM_EXIT_ERR: {
2653 vmexit = NESTED_EXIT_DONE;
2657 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2658 if (svm->nested.intercept & exit_bits)
2659 vmexit = NESTED_EXIT_DONE;
2666 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2670 vmexit = nested_svm_intercept(svm);
2672 if (vmexit == NESTED_EXIT_DONE)
2673 nested_svm_vmexit(svm);
2678 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2680 struct vmcb_control_area *dst = &dst_vmcb->control;
2681 struct vmcb_control_area *from = &from_vmcb->control;
2683 dst->intercept_cr = from->intercept_cr;
2684 dst->intercept_dr = from->intercept_dr;
2685 dst->intercept_exceptions = from->intercept_exceptions;
2686 dst->intercept = from->intercept;
2687 dst->iopm_base_pa = from->iopm_base_pa;
2688 dst->msrpm_base_pa = from->msrpm_base_pa;
2689 dst->tsc_offset = from->tsc_offset;
2690 dst->asid = from->asid;
2691 dst->tlb_ctl = from->tlb_ctl;
2692 dst->int_ctl = from->int_ctl;
2693 dst->int_vector = from->int_vector;
2694 dst->int_state = from->int_state;
2695 dst->exit_code = from->exit_code;
2696 dst->exit_code_hi = from->exit_code_hi;
2697 dst->exit_info_1 = from->exit_info_1;
2698 dst->exit_info_2 = from->exit_info_2;
2699 dst->exit_int_info = from->exit_int_info;
2700 dst->exit_int_info_err = from->exit_int_info_err;
2701 dst->nested_ctl = from->nested_ctl;
2702 dst->event_inj = from->event_inj;
2703 dst->event_inj_err = from->event_inj_err;
2704 dst->nested_cr3 = from->nested_cr3;
2705 dst->lbr_ctl = from->lbr_ctl;
2708 static int nested_svm_vmexit(struct vcpu_svm *svm)
2710 struct vmcb *nested_vmcb;
2711 struct vmcb *hsave = svm->nested.hsave;
2712 struct vmcb *vmcb = svm->vmcb;
2715 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2716 vmcb->control.exit_info_1,
2717 vmcb->control.exit_info_2,
2718 vmcb->control.exit_int_info,
2719 vmcb->control.exit_int_info_err,
2722 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2726 /* Exit Guest-Mode */
2727 leave_guest_mode(&svm->vcpu);
2728 svm->nested.vmcb = 0;
2730 /* Give the current vmcb to the guest */
2733 nested_vmcb->save.es = vmcb->save.es;
2734 nested_vmcb->save.cs = vmcb->save.cs;
2735 nested_vmcb->save.ss = vmcb->save.ss;
2736 nested_vmcb->save.ds = vmcb->save.ds;
2737 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2738 nested_vmcb->save.idtr = vmcb->save.idtr;
2739 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2740 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2741 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2742 nested_vmcb->save.cr2 = vmcb->save.cr2;
2743 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2744 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2745 nested_vmcb->save.rip = vmcb->save.rip;
2746 nested_vmcb->save.rsp = vmcb->save.rsp;
2747 nested_vmcb->save.rax = vmcb->save.rax;
2748 nested_vmcb->save.dr7 = vmcb->save.dr7;
2749 nested_vmcb->save.dr6 = vmcb->save.dr6;
2750 nested_vmcb->save.cpl = vmcb->save.cpl;
2752 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2753 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2754 nested_vmcb->control.int_state = vmcb->control.int_state;
2755 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2756 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2757 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2758 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2759 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2760 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2762 if (svm->nrips_enabled)
2763 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2766 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2767 * to make sure that we do not lose injected events. So check event_inj
2768 * here and copy it to exit_int_info if it is valid.
2769 * Exit_int_info and event_inj can't be both valid because the case
2770 * below only happens on a VMRUN instruction intercept which has
2771 * no valid exit_int_info set.
2773 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2774 struct vmcb_control_area *nc = &nested_vmcb->control;
2776 nc->exit_int_info = vmcb->control.event_inj;
2777 nc->exit_int_info_err = vmcb->control.event_inj_err;
2780 nested_vmcb->control.tlb_ctl = 0;
2781 nested_vmcb->control.event_inj = 0;
2782 nested_vmcb->control.event_inj_err = 0;
2784 /* We always set V_INTR_MASKING and remember the old value in hflags */
2785 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2786 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2788 /* Restore the original control entries */
2789 copy_vmcb_control_area(vmcb, hsave);
2791 kvm_clear_exception_queue(&svm->vcpu);
2792 kvm_clear_interrupt_queue(&svm->vcpu);
2794 svm->nested.nested_cr3 = 0;
2796 /* Restore selected save entries */
2797 svm->vmcb->save.es = hsave->save.es;
2798 svm->vmcb->save.cs = hsave->save.cs;
2799 svm->vmcb->save.ss = hsave->save.ss;
2800 svm->vmcb->save.ds = hsave->save.ds;
2801 svm->vmcb->save.gdtr = hsave->save.gdtr;
2802 svm->vmcb->save.idtr = hsave->save.idtr;
2803 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2804 svm_set_efer(&svm->vcpu, hsave->save.efer);
2805 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2806 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2808 svm->vmcb->save.cr3 = hsave->save.cr3;
2809 svm->vcpu.arch.cr3 = hsave->save.cr3;
2811 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2813 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2814 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2815 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2816 svm->vmcb->save.dr7 = 0;
2817 svm->vmcb->save.cpl = 0;
2818 svm->vmcb->control.exit_int_info = 0;
2820 mark_all_dirty(svm->vmcb);
2822 nested_svm_unmap(page);
2824 nested_svm_uninit_mmu_context(&svm->vcpu);
2825 kvm_mmu_reset_context(&svm->vcpu);
2826 kvm_mmu_load(&svm->vcpu);
2831 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2834 * This function merges the msr permission bitmaps of kvm and the
2835 * nested vmcb. It is optimized in that it only merges the parts where
2836 * the kvm msr permission bitmap may contain zero bits
2840 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2843 for (i = 0; i < MSRPM_OFFSETS; i++) {
2847 if (msrpm_offsets[i] == 0xffffffff)
2850 p = msrpm_offsets[i];
2851 offset = svm->nested.vmcb_msrpm + (p * 4);
2853 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2856 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2859 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2864 static bool nested_vmcb_checks(struct vmcb *vmcb)
2866 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2869 if (vmcb->control.asid == 0)
2872 if (vmcb->control.nested_ctl && !npt_enabled)
2878 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2880 struct vmcb *nested_vmcb;
2881 struct vmcb *hsave = svm->nested.hsave;
2882 struct vmcb *vmcb = svm->vmcb;
2886 vmcb_gpa = svm->vmcb->save.rax;
2888 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2892 if (!nested_vmcb_checks(nested_vmcb)) {
2893 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2894 nested_vmcb->control.exit_code_hi = 0;
2895 nested_vmcb->control.exit_info_1 = 0;
2896 nested_vmcb->control.exit_info_2 = 0;
2898 nested_svm_unmap(page);
2903 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2904 nested_vmcb->save.rip,
2905 nested_vmcb->control.int_ctl,
2906 nested_vmcb->control.event_inj,
2907 nested_vmcb->control.nested_ctl);
2909 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2910 nested_vmcb->control.intercept_cr >> 16,
2911 nested_vmcb->control.intercept_exceptions,
2912 nested_vmcb->control.intercept);
2914 /* Clear internal status */
2915 kvm_clear_exception_queue(&svm->vcpu);
2916 kvm_clear_interrupt_queue(&svm->vcpu);
2919 * Save the old vmcb, so we don't need to pick what we save, but can
2920 * restore everything when a VMEXIT occurs
2922 hsave->save.es = vmcb->save.es;
2923 hsave->save.cs = vmcb->save.cs;
2924 hsave->save.ss = vmcb->save.ss;
2925 hsave->save.ds = vmcb->save.ds;
2926 hsave->save.gdtr = vmcb->save.gdtr;
2927 hsave->save.idtr = vmcb->save.idtr;
2928 hsave->save.efer = svm->vcpu.arch.efer;
2929 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2930 hsave->save.cr4 = svm->vcpu.arch.cr4;
2931 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2932 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2933 hsave->save.rsp = vmcb->save.rsp;
2934 hsave->save.rax = vmcb->save.rax;
2936 hsave->save.cr3 = vmcb->save.cr3;
2938 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2940 copy_vmcb_control_area(hsave, vmcb);
2942 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2943 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2945 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2947 if (nested_vmcb->control.nested_ctl) {
2948 kvm_mmu_unload(&svm->vcpu);
2949 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2950 nested_svm_init_mmu_context(&svm->vcpu);
2953 /* Load the nested guest state */
2954 svm->vmcb->save.es = nested_vmcb->save.es;
2955 svm->vmcb->save.cs = nested_vmcb->save.cs;
2956 svm->vmcb->save.ss = nested_vmcb->save.ss;
2957 svm->vmcb->save.ds = nested_vmcb->save.ds;
2958 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2959 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2960 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2961 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2962 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2963 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2965 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2966 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2968 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2970 /* Guest paging mode is active - reset mmu */
2971 kvm_mmu_reset_context(&svm->vcpu);
2973 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2974 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2975 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2976 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2978 /* In case we don't even reach vcpu_run, the fields are not updated */
2979 svm->vmcb->save.rax = nested_vmcb->save.rax;
2980 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2981 svm->vmcb->save.rip = nested_vmcb->save.rip;
2982 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2983 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2984 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2986 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2987 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2989 /* cache intercepts */
2990 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2991 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2992 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2993 svm->nested.intercept = nested_vmcb->control.intercept;
2995 svm_flush_tlb(&svm->vcpu);
2996 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2997 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2998 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3000 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3002 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3003 /* We only want the cr8 intercept bits of the guest */
3004 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3005 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3008 /* We don't want to see VMMCALLs from a nested guest */
3009 clr_intercept(svm, INTERCEPT_VMMCALL);
3011 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
3012 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3013 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3014 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
3015 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3016 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3018 nested_svm_unmap(page);
3020 /* Enter Guest-Mode */
3021 enter_guest_mode(&svm->vcpu);
3024 * Merge guest and host intercepts - must be called with vcpu in
3025 * guest-mode to take affect here
3027 recalc_intercepts(svm);
3029 svm->nested.vmcb = vmcb_gpa;
3033 mark_all_dirty(svm->vmcb);
3038 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3040 to_vmcb->save.fs = from_vmcb->save.fs;
3041 to_vmcb->save.gs = from_vmcb->save.gs;
3042 to_vmcb->save.tr = from_vmcb->save.tr;
3043 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3044 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3045 to_vmcb->save.star = from_vmcb->save.star;
3046 to_vmcb->save.lstar = from_vmcb->save.lstar;
3047 to_vmcb->save.cstar = from_vmcb->save.cstar;
3048 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3049 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3050 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3051 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3054 static int vmload_interception(struct vcpu_svm *svm)
3056 struct vmcb *nested_vmcb;
3059 if (nested_svm_check_permissions(svm))
3062 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3066 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3067 skip_emulated_instruction(&svm->vcpu);
3069 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3070 nested_svm_unmap(page);
3075 static int vmsave_interception(struct vcpu_svm *svm)
3077 struct vmcb *nested_vmcb;
3080 if (nested_svm_check_permissions(svm))
3083 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3087 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3088 skip_emulated_instruction(&svm->vcpu);
3090 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3091 nested_svm_unmap(page);
3096 static int vmrun_interception(struct vcpu_svm *svm)
3098 if (nested_svm_check_permissions(svm))
3101 /* Save rip after vmrun instruction */
3102 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3104 if (!nested_svm_vmrun(svm))
3107 if (!nested_svm_vmrun_msrpm(svm))
3114 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
3115 svm->vmcb->control.exit_code_hi = 0;
3116 svm->vmcb->control.exit_info_1 = 0;
3117 svm->vmcb->control.exit_info_2 = 0;
3119 nested_svm_vmexit(svm);
3124 static int stgi_interception(struct vcpu_svm *svm)
3126 if (nested_svm_check_permissions(svm))
3129 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3130 skip_emulated_instruction(&svm->vcpu);
3131 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3138 static int clgi_interception(struct vcpu_svm *svm)
3140 if (nested_svm_check_permissions(svm))
3143 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3144 skip_emulated_instruction(&svm->vcpu);
3148 /* After a CLGI no interrupts should come */
3149 if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3150 svm_clear_vintr(svm);
3151 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3152 mark_dirty(svm->vmcb, VMCB_INTR);
3158 static int invlpga_interception(struct vcpu_svm *svm)
3160 struct kvm_vcpu *vcpu = &svm->vcpu;
3162 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3163 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3165 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3166 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3168 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3169 skip_emulated_instruction(&svm->vcpu);
3173 static int skinit_interception(struct vcpu_svm *svm)
3175 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3177 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3181 static int wbinvd_interception(struct vcpu_svm *svm)
3183 return kvm_emulate_wbinvd(&svm->vcpu);
3186 static int xsetbv_interception(struct vcpu_svm *svm)
3188 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3189 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3191 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3192 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3193 skip_emulated_instruction(&svm->vcpu);
3199 static int task_switch_interception(struct vcpu_svm *svm)
3203 int int_type = svm->vmcb->control.exit_int_info &
3204 SVM_EXITINTINFO_TYPE_MASK;
3205 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3207 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3209 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3210 bool has_error_code = false;
3213 tss_selector = (u16)svm->vmcb->control.exit_info_1;
3215 if (svm->vmcb->control.exit_info_2 &
3216 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3217 reason = TASK_SWITCH_IRET;
3218 else if (svm->vmcb->control.exit_info_2 &
3219 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3220 reason = TASK_SWITCH_JMP;
3222 reason = TASK_SWITCH_GATE;
3224 reason = TASK_SWITCH_CALL;
3226 if (reason == TASK_SWITCH_GATE) {
3228 case SVM_EXITINTINFO_TYPE_NMI:
3229 svm->vcpu.arch.nmi_injected = false;
3231 case SVM_EXITINTINFO_TYPE_EXEPT:
3232 if (svm->vmcb->control.exit_info_2 &
3233 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3234 has_error_code = true;
3236 (u32)svm->vmcb->control.exit_info_2;
3238 kvm_clear_exception_queue(&svm->vcpu);
3240 case SVM_EXITINTINFO_TYPE_INTR:
3241 kvm_clear_interrupt_queue(&svm->vcpu);
3248 if (reason != TASK_SWITCH_GATE ||
3249 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3250 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3251 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3252 skip_emulated_instruction(&svm->vcpu);
3254 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3257 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3258 has_error_code, error_code) == EMULATE_FAIL) {
3259 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3260 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3261 svm->vcpu.run->internal.ndata = 0;
3267 static int cpuid_interception(struct vcpu_svm *svm)
3269 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3270 return kvm_emulate_cpuid(&svm->vcpu);
3273 static int iret_interception(struct vcpu_svm *svm)
3275 ++svm->vcpu.stat.nmi_window_exits;
3276 clr_intercept(svm, INTERCEPT_IRET);
3277 svm->vcpu.arch.hflags |= HF_IRET_MASK;
3278 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3279 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3283 static int invlpg_interception(struct vcpu_svm *svm)
3285 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3286 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3288 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3289 skip_emulated_instruction(&svm->vcpu);
3293 static int emulate_on_interception(struct vcpu_svm *svm)
3295 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3298 static int rdpmc_interception(struct vcpu_svm *svm)
3302 if (!static_cpu_has(X86_FEATURE_NRIPS))
3303 return emulate_on_interception(svm);
3305 err = kvm_rdpmc(&svm->vcpu);
3306 return kvm_complete_insn_gp(&svm->vcpu, err);
3309 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3312 unsigned long cr0 = svm->vcpu.arch.cr0;
3316 intercept = svm->nested.intercept;
3318 if (!is_guest_mode(&svm->vcpu) ||
3319 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3322 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3323 val &= ~SVM_CR0_SELECTIVE_MASK;
3326 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3327 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3333 #define CR_VALID (1ULL << 63)
3335 static int cr_interception(struct vcpu_svm *svm)
3341 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3342 return emulate_on_interception(svm);
3344 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3345 return emulate_on_interception(svm);
3347 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3348 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3349 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3351 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3354 if (cr >= 16) { /* mov to cr */
3356 val = kvm_register_read(&svm->vcpu, reg);
3359 if (!check_selective_cr0_intercepted(svm, val))
3360 err = kvm_set_cr0(&svm->vcpu, val);
3366 err = kvm_set_cr3(&svm->vcpu, val);
3369 err = kvm_set_cr4(&svm->vcpu, val);
3372 err = kvm_set_cr8(&svm->vcpu, val);
3375 WARN(1, "unhandled write to CR%d", cr);
3376 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3379 } else { /* mov from cr */
3382 val = kvm_read_cr0(&svm->vcpu);
3385 val = svm->vcpu.arch.cr2;
3388 val = kvm_read_cr3(&svm->vcpu);
3391 val = kvm_read_cr4(&svm->vcpu);
3394 val = kvm_get_cr8(&svm->vcpu);
3397 WARN(1, "unhandled read from CR%d", cr);
3398 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3401 kvm_register_write(&svm->vcpu, reg, val);
3403 return kvm_complete_insn_gp(&svm->vcpu, err);
3406 static int dr_interception(struct vcpu_svm *svm)
3411 if (svm->vcpu.guest_debug == 0) {
3413 * No more DR vmexits; force a reload of the debug registers
3414 * and reenter on this instruction. The next vmexit will
3415 * retrieve the full state of the debug registers.
3417 clr_dr_intercepts(svm);
3418 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3422 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3423 return emulate_on_interception(svm);
3425 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3426 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3428 if (dr >= 16) { /* mov to DRn */
3429 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3431 val = kvm_register_read(&svm->vcpu, reg);
3432 kvm_set_dr(&svm->vcpu, dr - 16, val);
3434 if (!kvm_require_dr(&svm->vcpu, dr))
3436 kvm_get_dr(&svm->vcpu, dr, &val);
3437 kvm_register_write(&svm->vcpu, reg, val);
3440 skip_emulated_instruction(&svm->vcpu);
3445 static int cr8_write_interception(struct vcpu_svm *svm)
3447 struct kvm_run *kvm_run = svm->vcpu.run;
3450 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3451 /* instruction emulation calls kvm_set_cr8() */
3452 r = cr_interception(svm);
3453 if (lapic_in_kernel(&svm->vcpu))
3455 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3457 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3461 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3463 struct vcpu_svm *svm = to_svm(vcpu);
3465 switch (msr_info->index) {
3466 case MSR_IA32_TSC: {
3467 msr_info->data = svm->vmcb->control.tsc_offset +
3468 kvm_scale_tsc(vcpu, rdtsc());
3473 msr_info->data = svm->vmcb->save.star;
3475 #ifdef CONFIG_X86_64
3477 msr_info->data = svm->vmcb->save.lstar;
3480 msr_info->data = svm->vmcb->save.cstar;
3482 case MSR_KERNEL_GS_BASE:
3483 msr_info->data = svm->vmcb->save.kernel_gs_base;
3485 case MSR_SYSCALL_MASK:
3486 msr_info->data = svm->vmcb->save.sfmask;
3489 case MSR_IA32_SYSENTER_CS:
3490 msr_info->data = svm->vmcb->save.sysenter_cs;
3492 case MSR_IA32_SYSENTER_EIP:
3493 msr_info->data = svm->sysenter_eip;
3495 case MSR_IA32_SYSENTER_ESP:
3496 msr_info->data = svm->sysenter_esp;
3499 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3501 msr_info->data = svm->tsc_aux;
3504 * Nobody will change the following 5 values in the VMCB so we can
3505 * safely return them on rdmsr. They will always be 0 until LBRV is
3508 case MSR_IA32_DEBUGCTLMSR:
3509 msr_info->data = svm->vmcb->save.dbgctl;
3511 case MSR_IA32_LASTBRANCHFROMIP:
3512 msr_info->data = svm->vmcb->save.br_from;
3514 case MSR_IA32_LASTBRANCHTOIP:
3515 msr_info->data = svm->vmcb->save.br_to;
3517 case MSR_IA32_LASTINTFROMIP:
3518 msr_info->data = svm->vmcb->save.last_excp_from;
3520 case MSR_IA32_LASTINTTOIP:
3521 msr_info->data = svm->vmcb->save.last_excp_to;
3523 case MSR_VM_HSAVE_PA:
3524 msr_info->data = svm->nested.hsave_msr;
3527 msr_info->data = svm->nested.vm_cr_msr;
3529 case MSR_IA32_UCODE_REV:
3530 msr_info->data = 0x01000065;
3532 case MSR_F15H_IC_CFG: {
3536 family = guest_cpuid_family(vcpu);
3537 model = guest_cpuid_model(vcpu);
3539 if (family < 0 || model < 0)
3540 return kvm_get_msr_common(vcpu, msr_info);
3544 if (family == 0x15 &&
3545 (model >= 0x2 && model < 0x20))
3546 msr_info->data = 0x1E;
3550 return kvm_get_msr_common(vcpu, msr_info);
3555 static int rdmsr_interception(struct vcpu_svm *svm)
3557 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3558 struct msr_data msr_info;
3560 msr_info.index = ecx;
3561 msr_info.host_initiated = false;
3562 if (svm_get_msr(&svm->vcpu, &msr_info)) {
3563 trace_kvm_msr_read_ex(ecx);
3564 kvm_inject_gp(&svm->vcpu, 0);
3566 trace_kvm_msr_read(ecx, msr_info.data);
3568 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3569 msr_info.data & 0xffffffff);
3570 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3571 msr_info.data >> 32);
3572 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3573 skip_emulated_instruction(&svm->vcpu);
3578 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3580 struct vcpu_svm *svm = to_svm(vcpu);
3581 int svm_dis, chg_mask;
3583 if (data & ~SVM_VM_CR_VALID_MASK)
3586 chg_mask = SVM_VM_CR_VALID_MASK;
3588 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3589 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3591 svm->nested.vm_cr_msr &= ~chg_mask;
3592 svm->nested.vm_cr_msr |= (data & chg_mask);
3594 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3596 /* check for svm_disable while efer.svme is set */
3597 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3603 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3605 struct vcpu_svm *svm = to_svm(vcpu);
3607 u32 ecx = msr->index;
3608 u64 data = msr->data;
3611 kvm_write_tsc(vcpu, msr);
3614 svm->vmcb->save.star = data;
3616 #ifdef CONFIG_X86_64
3618 svm->vmcb->save.lstar = data;
3621 svm->vmcb->save.cstar = data;
3623 case MSR_KERNEL_GS_BASE:
3624 svm->vmcb->save.kernel_gs_base = data;
3626 case MSR_SYSCALL_MASK:
3627 svm->vmcb->save.sfmask = data;
3630 case MSR_IA32_SYSENTER_CS:
3631 svm->vmcb->save.sysenter_cs = data;
3633 case MSR_IA32_SYSENTER_EIP:
3634 svm->sysenter_eip = data;
3635 svm->vmcb->save.sysenter_eip = data;
3637 case MSR_IA32_SYSENTER_ESP:
3638 svm->sysenter_esp = data;
3639 svm->vmcb->save.sysenter_esp = data;
3642 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3646 * This is rare, so we update the MSR here instead of using
3647 * direct_access_msrs. Doing that would require a rdmsr in
3650 svm->tsc_aux = data;
3651 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
3653 case MSR_IA32_DEBUGCTLMSR:
3654 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3655 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3659 if (data & DEBUGCTL_RESERVED_BITS)
3662 svm->vmcb->save.dbgctl = data;
3663 mark_dirty(svm->vmcb, VMCB_LBR);
3664 if (data & (1ULL<<0))
3665 svm_enable_lbrv(svm);
3667 svm_disable_lbrv(svm);
3669 case MSR_VM_HSAVE_PA:
3670 svm->nested.hsave_msr = data;
3673 return svm_set_vm_cr(vcpu, data);
3675 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3677 case MSR_IA32_APICBASE:
3678 if (kvm_vcpu_apicv_active(vcpu))
3679 avic_update_vapic_bar(to_svm(vcpu), data);
3680 /* Follow through */
3682 return kvm_set_msr_common(vcpu, msr);
3687 static int wrmsr_interception(struct vcpu_svm *svm)
3689 struct msr_data msr;
3690 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3691 u64 data = kvm_read_edx_eax(&svm->vcpu);
3695 msr.host_initiated = false;
3697 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3698 if (kvm_set_msr(&svm->vcpu, &msr)) {
3699 trace_kvm_msr_write_ex(ecx, data);
3700 kvm_inject_gp(&svm->vcpu, 0);
3702 trace_kvm_msr_write(ecx, data);
3703 skip_emulated_instruction(&svm->vcpu);
3708 static int msr_interception(struct vcpu_svm *svm)
3710 if (svm->vmcb->control.exit_info_1)
3711 return wrmsr_interception(svm);
3713 return rdmsr_interception(svm);
3716 static int interrupt_window_interception(struct vcpu_svm *svm)
3718 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3719 svm_clear_vintr(svm);
3720 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3721 mark_dirty(svm->vmcb, VMCB_INTR);
3722 ++svm->vcpu.stat.irq_window_exits;
3726 static int pause_interception(struct vcpu_svm *svm)
3728 kvm_vcpu_on_spin(&(svm->vcpu));
3732 static int nop_interception(struct vcpu_svm *svm)
3734 skip_emulated_instruction(&(svm->vcpu));
3738 static int monitor_interception(struct vcpu_svm *svm)
3740 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3741 return nop_interception(svm);
3744 static int mwait_interception(struct vcpu_svm *svm)
3746 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3747 return nop_interception(svm);
3750 enum avic_ipi_failure_cause {
3751 AVIC_IPI_FAILURE_INVALID_INT_TYPE,
3752 AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
3753 AVIC_IPI_FAILURE_INVALID_TARGET,
3754 AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
3757 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
3759 u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
3760 u32 icrl = svm->vmcb->control.exit_info_1;
3761 u32 id = svm->vmcb->control.exit_info_2 >> 32;
3762 u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
3763 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3765 trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
3768 case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
3770 * AVIC hardware handles the generation of
3771 * IPIs when the specified Message Type is Fixed
3772 * (also known as fixed delivery mode) and
3773 * the Trigger Mode is edge-triggered. The hardware
3774 * also supports self and broadcast delivery modes
3775 * specified via the Destination Shorthand(DSH)
3776 * field of the ICRL. Logical and physical APIC ID
3777 * formats are supported. All other IPI types cause
3778 * a #VMEXIT, which needs to emulated.
3780 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
3781 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
3783 case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
3785 struct kvm_vcpu *vcpu;
3786 struct kvm *kvm = svm->vcpu.kvm;
3787 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3790 * At this point, we expect that the AVIC HW has already
3791 * set the appropriate IRR bits on the valid target
3792 * vcpus. So, we just need to kick the appropriate vcpu.
3794 kvm_for_each_vcpu(i, vcpu, kvm) {
3795 bool m = kvm_apic_match_dest(vcpu, apic,
3796 icrl & KVM_APIC_SHORT_MASK,
3797 GET_APIC_DEST_FIELD(icrh),
3798 icrl & KVM_APIC_DEST_MASK);
3800 if (m && !avic_vcpu_is_running(vcpu))
3801 kvm_vcpu_wake_up(vcpu);
3805 case AVIC_IPI_FAILURE_INVALID_TARGET:
3807 case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
3808 WARN_ONCE(1, "Invalid backing page\n");
3811 pr_err("Unknown IPI interception\n");
3817 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
3819 struct kvm_arch *vm_data = &vcpu->kvm->arch;
3821 u32 *logical_apic_id_table;
3822 int dlid = GET_APIC_LOGICAL_ID(ldr);
3827 if (flat) { /* flat */
3828 index = ffs(dlid) - 1;
3831 } else { /* cluster */
3832 int cluster = (dlid & 0xf0) >> 4;
3833 int apic = ffs(dlid & 0x0f) - 1;
3835 if ((apic < 0) || (apic > 7) ||
3838 index = (cluster << 2) + apic;
3841 logical_apic_id_table = (u32 *) page_address(vm_data->avic_logical_id_table_page);
3843 return &logical_apic_id_table[index];
3846 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
3850 u32 *entry, new_entry;
3852 flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
3853 entry = avic_get_logical_id_entry(vcpu, ldr, flat);
3857 new_entry = READ_ONCE(*entry);
3858 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
3859 new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
3861 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3863 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3864 WRITE_ONCE(*entry, new_entry);
3869 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
3872 struct vcpu_svm *svm = to_svm(vcpu);
3873 u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
3878 ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
3879 if (ret && svm->ldr_reg) {
3880 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
3888 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
3891 struct vcpu_svm *svm = to_svm(vcpu);
3892 u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
3893 u32 id = (apic_id_reg >> 24) & 0xff;
3895 if (vcpu->vcpu_id == id)
3898 old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
3899 new = avic_get_physical_id_entry(vcpu, id);
3903 /* We need to move physical_id_entry to new offset */
3906 to_svm(vcpu)->avic_physical_id_cache = new;
3909 * Also update the guest physical APIC ID in the logical
3910 * APIC ID table entry if already setup the LDR.
3913 avic_handle_ldr_update(vcpu);
3918 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
3920 struct vcpu_svm *svm = to_svm(vcpu);
3921 struct kvm_arch *vm_data = &vcpu->kvm->arch;
3922 u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
3923 u32 mod = (dfr >> 28) & 0xf;
3926 * We assume that all local APICs are using the same type.
3927 * If this changes, we need to flush the AVIC logical
3930 if (vm_data->ldr_mode == mod)
3933 clear_page(page_address(vm_data->avic_logical_id_table_page));
3934 vm_data->ldr_mode = mod;
3937 avic_handle_ldr_update(vcpu);
3941 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
3943 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3944 u32 offset = svm->vmcb->control.exit_info_1 &
3945 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
3949 if (avic_handle_apic_id_update(&svm->vcpu))
3953 if (avic_handle_ldr_update(&svm->vcpu))
3957 avic_handle_dfr_update(&svm->vcpu);
3963 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
3968 static bool is_avic_unaccelerated_access_trap(u32 offset)
3997 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4000 u32 offset = svm->vmcb->control.exit_info_1 &
4001 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4002 u32 vector = svm->vmcb->control.exit_info_2 &
4003 AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4004 bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4005 AVIC_UNACCEL_ACCESS_WRITE_MASK;
4006 bool trap = is_avic_unaccelerated_access_trap(offset);
4008 trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4009 trap, write, vector);
4012 WARN_ONCE(!write, "svm: Handling trap read.\n");
4013 ret = avic_unaccel_trap_write(svm);
4015 /* Handling Fault */
4016 ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4022 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4023 [SVM_EXIT_READ_CR0] = cr_interception,
4024 [SVM_EXIT_READ_CR3] = cr_interception,
4025 [SVM_EXIT_READ_CR4] = cr_interception,
4026 [SVM_EXIT_READ_CR8] = cr_interception,
4027 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
4028 [SVM_EXIT_WRITE_CR0] = cr_interception,
4029 [SVM_EXIT_WRITE_CR3] = cr_interception,
4030 [SVM_EXIT_WRITE_CR4] = cr_interception,
4031 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
4032 [SVM_EXIT_READ_DR0] = dr_interception,
4033 [SVM_EXIT_READ_DR1] = dr_interception,
4034 [SVM_EXIT_READ_DR2] = dr_interception,
4035 [SVM_EXIT_READ_DR3] = dr_interception,
4036 [SVM_EXIT_READ_DR4] = dr_interception,
4037 [SVM_EXIT_READ_DR5] = dr_interception,
4038 [SVM_EXIT_READ_DR6] = dr_interception,
4039 [SVM_EXIT_READ_DR7] = dr_interception,
4040 [SVM_EXIT_WRITE_DR0] = dr_interception,
4041 [SVM_EXIT_WRITE_DR1] = dr_interception,
4042 [SVM_EXIT_WRITE_DR2] = dr_interception,
4043 [SVM_EXIT_WRITE_DR3] = dr_interception,
4044 [SVM_EXIT_WRITE_DR4] = dr_interception,
4045 [SVM_EXIT_WRITE_DR5] = dr_interception,
4046 [SVM_EXIT_WRITE_DR6] = dr_interception,
4047 [SVM_EXIT_WRITE_DR7] = dr_interception,
4048 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
4049 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
4050 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
4051 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
4052 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
4053 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
4054 [SVM_EXIT_INTR] = intr_interception,
4055 [SVM_EXIT_NMI] = nmi_interception,
4056 [SVM_EXIT_SMI] = nop_on_interception,
4057 [SVM_EXIT_INIT] = nop_on_interception,
4058 [SVM_EXIT_VINTR] = interrupt_window_interception,
4059 [SVM_EXIT_RDPMC] = rdpmc_interception,
4060 [SVM_EXIT_CPUID] = cpuid_interception,
4061 [SVM_EXIT_IRET] = iret_interception,
4062 [SVM_EXIT_INVD] = emulate_on_interception,
4063 [SVM_EXIT_PAUSE] = pause_interception,
4064 [SVM_EXIT_HLT] = halt_interception,
4065 [SVM_EXIT_INVLPG] = invlpg_interception,
4066 [SVM_EXIT_INVLPGA] = invlpga_interception,
4067 [SVM_EXIT_IOIO] = io_interception,
4068 [SVM_EXIT_MSR] = msr_interception,
4069 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
4070 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
4071 [SVM_EXIT_VMRUN] = vmrun_interception,
4072 [SVM_EXIT_VMMCALL] = vmmcall_interception,
4073 [SVM_EXIT_VMLOAD] = vmload_interception,
4074 [SVM_EXIT_VMSAVE] = vmsave_interception,
4075 [SVM_EXIT_STGI] = stgi_interception,
4076 [SVM_EXIT_CLGI] = clgi_interception,
4077 [SVM_EXIT_SKINIT] = skinit_interception,
4078 [SVM_EXIT_WBINVD] = wbinvd_interception,
4079 [SVM_EXIT_MONITOR] = monitor_interception,
4080 [SVM_EXIT_MWAIT] = mwait_interception,
4081 [SVM_EXIT_XSETBV] = xsetbv_interception,
4082 [SVM_EXIT_NPF] = pf_interception,
4083 [SVM_EXIT_RSM] = emulate_on_interception,
4084 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
4085 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
4088 static void dump_vmcb(struct kvm_vcpu *vcpu)
4090 struct vcpu_svm *svm = to_svm(vcpu);
4091 struct vmcb_control_area *control = &svm->vmcb->control;
4092 struct vmcb_save_area *save = &svm->vmcb->save;
4094 pr_err("VMCB Control Area:\n");
4095 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4096 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4097 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4098 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4099 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4100 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4101 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4102 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4103 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4104 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4105 pr_err("%-20s%d\n", "asid:", control->asid);
4106 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4107 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4108 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4109 pr_err("%-20s%08x\n", "int_state:", control->int_state);
4110 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4111 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4112 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4113 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4114 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4115 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4116 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4117 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4118 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4119 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4120 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
4121 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4122 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4123 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4124 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4125 pr_err("VMCB State Save Area:\n");
4126 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4128 save->es.selector, save->es.attrib,
4129 save->es.limit, save->es.base);
4130 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4132 save->cs.selector, save->cs.attrib,
4133 save->cs.limit, save->cs.base);
4134 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4136 save->ss.selector, save->ss.attrib,
4137 save->ss.limit, save->ss.base);
4138 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4140 save->ds.selector, save->ds.attrib,
4141 save->ds.limit, save->ds.base);
4142 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4144 save->fs.selector, save->fs.attrib,
4145 save->fs.limit, save->fs.base);
4146 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4148 save->gs.selector, save->gs.attrib,
4149 save->gs.limit, save->gs.base);
4150 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4152 save->gdtr.selector, save->gdtr.attrib,
4153 save->gdtr.limit, save->gdtr.base);
4154 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4156 save->ldtr.selector, save->ldtr.attrib,
4157 save->ldtr.limit, save->ldtr.base);
4158 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4160 save->idtr.selector, save->idtr.attrib,
4161 save->idtr.limit, save->idtr.base);
4162 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4164 save->tr.selector, save->tr.attrib,
4165 save->tr.limit, save->tr.base);
4166 pr_err("cpl: %d efer: %016llx\n",
4167 save->cpl, save->efer);
4168 pr_err("%-15s %016llx %-13s %016llx\n",
4169 "cr0:", save->cr0, "cr2:", save->cr2);
4170 pr_err("%-15s %016llx %-13s %016llx\n",
4171 "cr3:", save->cr3, "cr4:", save->cr4);
4172 pr_err("%-15s %016llx %-13s %016llx\n",
4173 "dr6:", save->dr6, "dr7:", save->dr7);
4174 pr_err("%-15s %016llx %-13s %016llx\n",
4175 "rip:", save->rip, "rflags:", save->rflags);
4176 pr_err("%-15s %016llx %-13s %016llx\n",
4177 "rsp:", save->rsp, "rax:", save->rax);
4178 pr_err("%-15s %016llx %-13s %016llx\n",
4179 "star:", save->star, "lstar:", save->lstar);
4180 pr_err("%-15s %016llx %-13s %016llx\n",
4181 "cstar:", save->cstar, "sfmask:", save->sfmask);
4182 pr_err("%-15s %016llx %-13s %016llx\n",
4183 "kernel_gs_base:", save->kernel_gs_base,
4184 "sysenter_cs:", save->sysenter_cs);
4185 pr_err("%-15s %016llx %-13s %016llx\n",
4186 "sysenter_esp:", save->sysenter_esp,
4187 "sysenter_eip:", save->sysenter_eip);
4188 pr_err("%-15s %016llx %-13s %016llx\n",
4189 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4190 pr_err("%-15s %016llx %-13s %016llx\n",
4191 "br_from:", save->br_from, "br_to:", save->br_to);
4192 pr_err("%-15s %016llx %-13s %016llx\n",
4193 "excp_from:", save->last_excp_from,
4194 "excp_to:", save->last_excp_to);
4197 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4199 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4201 *info1 = control->exit_info_1;
4202 *info2 = control->exit_info_2;
4205 static int handle_exit(struct kvm_vcpu *vcpu)
4207 struct vcpu_svm *svm = to_svm(vcpu);
4208 struct kvm_run *kvm_run = vcpu->run;
4209 u32 exit_code = svm->vmcb->control.exit_code;
4211 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4213 vcpu->arch.gpa_available = (exit_code == SVM_EXIT_NPF);
4215 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4216 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4218 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4220 if (unlikely(svm->nested.exit_required)) {
4221 nested_svm_vmexit(svm);
4222 svm->nested.exit_required = false;
4227 if (is_guest_mode(vcpu)) {
4230 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4231 svm->vmcb->control.exit_info_1,
4232 svm->vmcb->control.exit_info_2,
4233 svm->vmcb->control.exit_int_info,
4234 svm->vmcb->control.exit_int_info_err,
4237 vmexit = nested_svm_exit_special(svm);
4239 if (vmexit == NESTED_EXIT_CONTINUE)
4240 vmexit = nested_svm_exit_handled(svm);
4242 if (vmexit == NESTED_EXIT_DONE)
4246 svm_complete_interrupts(svm);
4248 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4249 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4250 kvm_run->fail_entry.hardware_entry_failure_reason
4251 = svm->vmcb->control.exit_code;
4252 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4257 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4258 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4259 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4260 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4261 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4263 __func__, svm->vmcb->control.exit_int_info,
4266 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4267 || !svm_exit_handlers[exit_code]) {
4268 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4269 kvm_queue_exception(vcpu, UD_VECTOR);
4273 return svm_exit_handlers[exit_code](svm);
4276 static void reload_tss(struct kvm_vcpu *vcpu)
4278 int cpu = raw_smp_processor_id();
4280 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4281 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4285 static void pre_svm_run(struct vcpu_svm *svm)
4287 int cpu = raw_smp_processor_id();
4289 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4291 /* FIXME: handle wraparound of asid_generation */
4292 if (svm->asid_generation != sd->asid_generation)
4296 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
4298 struct vcpu_svm *svm = to_svm(vcpu);
4300 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
4301 vcpu->arch.hflags |= HF_NMI_MASK;
4302 set_intercept(svm, INTERCEPT_IRET);
4303 ++vcpu->stat.nmi_injections;
4306 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
4308 struct vmcb_control_area *control;
4310 /* The following fields are ignored when AVIC is enabled */
4311 control = &svm->vmcb->control;
4312 control->int_vector = irq;
4313 control->int_ctl &= ~V_INTR_PRIO_MASK;
4314 control->int_ctl |= V_IRQ_MASK |
4315 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
4316 mark_dirty(svm->vmcb, VMCB_INTR);
4319 static void svm_set_irq(struct kvm_vcpu *vcpu)
4321 struct vcpu_svm *svm = to_svm(vcpu);
4323 BUG_ON(!(gif_set(svm)));
4325 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
4326 ++vcpu->stat.irq_injections;
4328 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
4329 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
4332 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
4334 return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
4337 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
4339 struct vcpu_svm *svm = to_svm(vcpu);
4341 if (svm_nested_virtualize_tpr(vcpu) ||
4342 kvm_vcpu_apicv_active(vcpu))
4345 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4351 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4354 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
4359 static bool svm_get_enable_apicv(void)
4364 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
4368 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
4372 /* Note: Currently only used by Hyper-V. */
4373 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4375 struct vcpu_svm *svm = to_svm(vcpu);
4376 struct vmcb *vmcb = svm->vmcb;
4381 vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
4382 mark_dirty(vmcb, VMCB_INTR);
4385 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
4390 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
4392 kvm_lapic_set_irr(vec, vcpu->arch.apic);
4393 smp_mb__after_atomic();
4395 if (avic_vcpu_is_running(vcpu))
4396 wrmsrl(SVM_AVIC_DOORBELL,
4397 kvm_cpu_get_apicid(vcpu->cpu));
4399 kvm_vcpu_wake_up(vcpu);
4402 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4404 unsigned long flags;
4405 struct amd_svm_iommu_ir *cur;
4407 spin_lock_irqsave(&svm->ir_list_lock, flags);
4408 list_for_each_entry(cur, &svm->ir_list, node) {
4409 if (cur->data != pi->ir_data)
4411 list_del(&cur->node);
4415 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4418 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4421 unsigned long flags;
4422 struct amd_svm_iommu_ir *ir;
4425 * In some cases, the existing irte is updaed and re-set,
4426 * so we need to check here if it's already been * added
4429 if (pi->ir_data && (pi->prev_ga_tag != 0)) {
4430 struct kvm *kvm = svm->vcpu.kvm;
4431 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
4432 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
4433 struct vcpu_svm *prev_svm;
4440 prev_svm = to_svm(prev_vcpu);
4441 svm_ir_list_del(prev_svm, pi);
4445 * Allocating new amd_iommu_pi_data, which will get
4446 * add to the per-vcpu ir_list.
4448 ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
4453 ir->data = pi->ir_data;
4455 spin_lock_irqsave(&svm->ir_list_lock, flags);
4456 list_add(&ir->node, &svm->ir_list);
4457 spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4464 * The HW cannot support posting multicast/broadcast
4465 * interrupts to a vCPU. So, we still use legacy interrupt
4466 * remapping for these kind of interrupts.
4468 * For lowest-priority interrupts, we only support
4469 * those with single CPU as the destination, e.g. user
4470 * configures the interrupts via /proc/irq or uses
4471 * irqbalance to make the interrupts single-CPU.
4474 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
4475 struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
4477 struct kvm_lapic_irq irq;
4478 struct kvm_vcpu *vcpu = NULL;
4480 kvm_set_msi_irq(kvm, e, &irq);
4482 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
4483 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4484 __func__, irq.vector);
4488 pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
4490 *svm = to_svm(vcpu);
4491 vcpu_info->pi_desc_addr = page_to_phys((*svm)->avic_backing_page);
4492 vcpu_info->vector = irq.vector;
4498 * svm_update_pi_irte - set IRTE for Posted-Interrupts
4501 * @host_irq: host irq of the interrupt
4502 * @guest_irq: gsi of the interrupt
4503 * @set: set or unset PI
4504 * returns 0 on success, < 0 on failure
4506 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
4507 uint32_t guest_irq, bool set)
4509 struct kvm_kernel_irq_routing_entry *e;
4510 struct kvm_irq_routing_table *irq_rt;
4511 int idx, ret = -EINVAL;
4513 if (!kvm_arch_has_assigned_device(kvm) ||
4514 !irq_remapping_cap(IRQ_POSTING_CAP))
4517 pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4518 __func__, host_irq, guest_irq, set);
4520 idx = srcu_read_lock(&kvm->irq_srcu);
4521 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
4522 WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
4524 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
4525 struct vcpu_data vcpu_info;
4526 struct vcpu_svm *svm = NULL;
4528 if (e->type != KVM_IRQ_ROUTING_MSI)
4532 * Here, we setup with legacy mode in the following cases:
4533 * 1. When cannot target interrupt to a specific vcpu.
4534 * 2. Unsetting posted interrupt.
4535 * 3. APIC virtialization is disabled for the vcpu.
4537 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
4538 kvm_vcpu_apicv_active(&svm->vcpu)) {
4539 struct amd_iommu_pi_data pi;
4541 /* Try to enable guest_mode in IRTE */
4542 pi.base = page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK;
4543 pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
4545 pi.is_guest_mode = true;
4546 pi.vcpu_data = &vcpu_info;
4547 ret = irq_set_vcpu_affinity(host_irq, &pi);
4550 * Here, we successfully setting up vcpu affinity in
4551 * IOMMU guest mode. Now, we need to store the posted
4552 * interrupt information in a per-vcpu ir_list so that
4553 * we can reference to them directly when we update vcpu
4554 * scheduling information in IOMMU irte.
4556 if (!ret && pi.is_guest_mode)
4557 svm_ir_list_add(svm, &pi);
4559 /* Use legacy mode in IRTE */
4560 struct amd_iommu_pi_data pi;
4563 * Here, pi is used to:
4564 * - Tell IOMMU to use legacy mode for this interrupt.
4565 * - Retrieve ga_tag of prior interrupt remapping data.
4567 pi.is_guest_mode = false;
4568 ret = irq_set_vcpu_affinity(host_irq, &pi);
4571 * Check if the posted interrupt was previously
4572 * setup with the guest_mode by checking if the ga_tag
4573 * was cached. If so, we need to clean up the per-vcpu
4576 if (!ret && pi.prev_ga_tag) {
4577 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
4578 struct kvm_vcpu *vcpu;
4580 vcpu = kvm_get_vcpu_by_id(kvm, id);
4582 svm_ir_list_del(to_svm(vcpu), &pi);
4587 trace_kvm_pi_irte_update(svm->vcpu.vcpu_id,
4590 vcpu_info.pi_desc_addr, set);
4594 pr_err("%s: failed to update PI IRTE\n", __func__);
4601 srcu_read_unlock(&kvm->irq_srcu, idx);
4605 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
4607 struct vcpu_svm *svm = to_svm(vcpu);
4608 struct vmcb *vmcb = svm->vmcb;
4610 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
4611 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
4612 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
4617 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
4619 struct vcpu_svm *svm = to_svm(vcpu);
4621 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
4624 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4626 struct vcpu_svm *svm = to_svm(vcpu);
4629 svm->vcpu.arch.hflags |= HF_NMI_MASK;
4630 set_intercept(svm, INTERCEPT_IRET);
4632 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
4633 clr_intercept(svm, INTERCEPT_IRET);
4637 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
4639 struct vcpu_svm *svm = to_svm(vcpu);
4640 struct vmcb *vmcb = svm->vmcb;
4643 if (!gif_set(svm) ||
4644 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
4647 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
4649 if (is_guest_mode(vcpu))
4650 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
4655 static void enable_irq_window(struct kvm_vcpu *vcpu)
4657 struct vcpu_svm *svm = to_svm(vcpu);
4659 if (kvm_vcpu_apicv_active(vcpu))
4663 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4664 * 1, because that's a separate STGI/VMRUN intercept. The next time we
4665 * get that intercept, this function will be called again though and
4666 * we'll get the vintr intercept.
4668 if (gif_set(svm) && nested_svm_intr(svm)) {
4670 svm_inject_irq(svm, 0x0);
4674 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4676 struct vcpu_svm *svm = to_svm(vcpu);
4678 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
4680 return; /* IRET will cause a vm exit */
4682 if ((svm->vcpu.arch.hflags & HF_GIF_MASK) == 0)
4683 return; /* STGI will cause a vm exit */
4685 if (svm->nested.exit_required)
4686 return; /* we're not going to run the guest yet */
4689 * Something prevents NMI from been injected. Single step over possible
4690 * problem (IRET or exception injection or interrupt shadow)
4692 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
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))
4834 * Disable singlestep if we're injecting an interrupt/exception.
4835 * We don't want our modified rflags to be pushed on the stack where
4836 * we might not be able to easily reset them if we disabled NMI
4839 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
4841 * Event injection happens before external interrupts cause a
4842 * vmexit and interrupts are disabled here, so smp_send_reschedule
4843 * is enough to force an immediate vmexit.
4845 disable_nmi_singlestep(svm);
4846 smp_send_reschedule(vcpu->cpu);
4851 sync_lapic_to_cr8(vcpu);
4853 svm->vmcb->save.cr2 = vcpu->arch.cr2;
4860 "push %%" _ASM_BP "; \n\t"
4861 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4862 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4863 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4864 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4865 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4866 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4867 #ifdef CONFIG_X86_64
4868 "mov %c[r8](%[svm]), %%r8 \n\t"
4869 "mov %c[r9](%[svm]), %%r9 \n\t"
4870 "mov %c[r10](%[svm]), %%r10 \n\t"
4871 "mov %c[r11](%[svm]), %%r11 \n\t"
4872 "mov %c[r12](%[svm]), %%r12 \n\t"
4873 "mov %c[r13](%[svm]), %%r13 \n\t"
4874 "mov %c[r14](%[svm]), %%r14 \n\t"
4875 "mov %c[r15](%[svm]), %%r15 \n\t"
4878 /* Enter guest mode */
4879 "push %%" _ASM_AX " \n\t"
4880 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4881 __ex(SVM_VMLOAD) "\n\t"
4882 __ex(SVM_VMRUN) "\n\t"
4883 __ex(SVM_VMSAVE) "\n\t"
4884 "pop %%" _ASM_AX " \n\t"
4886 /* Save guest registers, load host registers */
4887 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4888 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4889 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4890 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4891 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4892 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4893 #ifdef CONFIG_X86_64
4894 "mov %%r8, %c[r8](%[svm]) \n\t"
4895 "mov %%r9, %c[r9](%[svm]) \n\t"
4896 "mov %%r10, %c[r10](%[svm]) \n\t"
4897 "mov %%r11, %c[r11](%[svm]) \n\t"
4898 "mov %%r12, %c[r12](%[svm]) \n\t"
4899 "mov %%r13, %c[r13](%[svm]) \n\t"
4900 "mov %%r14, %c[r14](%[svm]) \n\t"
4901 "mov %%r15, %c[r15](%[svm]) \n\t"
4906 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4907 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4908 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4909 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4910 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4911 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4912 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4913 #ifdef CONFIG_X86_64
4914 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4915 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4916 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4917 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4918 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4919 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4920 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4921 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4924 #ifdef CONFIG_X86_64
4925 , "rbx", "rcx", "rdx", "rsi", "rdi"
4926 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4928 , "ebx", "ecx", "edx", "esi", "edi"
4932 #ifdef CONFIG_X86_64
4933 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4935 loadsegment(fs, svm->host.fs);
4936 #ifndef CONFIG_X86_32_LAZY_GS
4937 loadsegment(gs, svm->host.gs);
4943 local_irq_disable();
4945 vcpu->arch.cr2 = svm->vmcb->save.cr2;
4946 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4947 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4948 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4950 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4951 kvm_before_handle_nmi(&svm->vcpu);
4955 /* Any pending NMI will happen here */
4957 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4958 kvm_after_handle_nmi(&svm->vcpu);
4960 sync_cr8_to_lapic(vcpu);
4964 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4966 /* if exit due to PF check for async PF */
4967 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4968 svm->apf_reason = kvm_read_and_reset_pf_reason();
4971 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4972 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4976 * We need to handle MC intercepts here before the vcpu has a chance to
4977 * change the physical cpu
4979 if (unlikely(svm->vmcb->control.exit_code ==
4980 SVM_EXIT_EXCP_BASE + MC_VECTOR))
4981 svm_handle_mce(svm);
4983 mark_all_clean(svm->vmcb);
4985 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
4987 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4989 struct vcpu_svm *svm = to_svm(vcpu);
4991 svm->vmcb->save.cr3 = root;
4992 mark_dirty(svm->vmcb, VMCB_CR);
4993 svm_flush_tlb(vcpu);
4996 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4998 struct vcpu_svm *svm = to_svm(vcpu);
5000 svm->vmcb->control.nested_cr3 = root;
5001 mark_dirty(svm->vmcb, VMCB_NPT);
5003 /* Also sync guest cr3 here in case we live migrate */
5004 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5005 mark_dirty(svm->vmcb, VMCB_CR);
5007 svm_flush_tlb(vcpu);
5010 static int is_disabled(void)
5014 rdmsrl(MSR_VM_CR, vm_cr);
5015 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5022 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5025 * Patch in the VMMCALL instruction:
5027 hypercall[0] = 0x0f;
5028 hypercall[1] = 0x01;
5029 hypercall[2] = 0xd9;
5032 static void svm_check_processor_compat(void *rtn)
5037 static bool svm_cpu_has_accelerated_tpr(void)
5042 static bool svm_has_high_real_mode_segbase(void)
5047 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5052 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5054 struct vcpu_svm *svm = to_svm(vcpu);
5055 struct kvm_cpuid_entry2 *entry;
5057 /* Update nrips enabled cache */
5058 svm->nrips_enabled = !!guest_cpuid_has_nrips(&svm->vcpu);
5060 if (!kvm_vcpu_apicv_active(vcpu))
5063 entry = kvm_find_cpuid_entry(vcpu, 1, 0);
5065 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5068 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5073 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5077 entry->ecx |= (1 << 2); /* Set SVM bit */
5080 entry->eax = 1; /* SVM revision 1 */
5081 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5082 ASID emulation to nested SVM */
5083 entry->ecx = 0; /* Reserved */
5084 entry->edx = 0; /* Per default do not support any
5085 additional features */
5087 /* Support next_rip if host supports it */
5088 if (boot_cpu_has(X86_FEATURE_NRIPS))
5089 entry->edx |= SVM_FEATURE_NRIP;
5091 /* Support NPT for the guest if enabled */
5093 entry->edx |= SVM_FEATURE_NPT;
5099 static int svm_get_lpage_level(void)
5101 return PT_PDPE_LEVEL;
5104 static bool svm_rdtscp_supported(void)
5106 return boot_cpu_has(X86_FEATURE_RDTSCP);
5109 static bool svm_invpcid_supported(void)
5114 static bool svm_mpx_supported(void)
5119 static bool svm_xsaves_supported(void)
5124 static bool svm_has_wbinvd_exit(void)
5129 #define PRE_EX(exit) { .exit_code = (exit), \
5130 .stage = X86_ICPT_PRE_EXCEPT, }
5131 #define POST_EX(exit) { .exit_code = (exit), \
5132 .stage = X86_ICPT_POST_EXCEPT, }
5133 #define POST_MEM(exit) { .exit_code = (exit), \
5134 .stage = X86_ICPT_POST_MEMACCESS, }
5136 static const struct __x86_intercept {
5138 enum x86_intercept_stage stage;
5139 } x86_intercept_map[] = {
5140 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
5141 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
5142 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
5143 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
5144 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
5145 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
5146 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
5147 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
5148 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
5149 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
5150 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
5151 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
5152 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
5153 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
5154 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
5155 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
5156 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
5157 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
5158 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
5159 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
5160 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
5161 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
5162 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
5163 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
5164 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
5165 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
5166 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
5167 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
5168 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
5169 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
5170 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
5171 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
5172 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
5173 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
5174 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
5175 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
5176 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
5177 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
5178 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
5179 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
5180 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
5181 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
5182 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
5183 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
5184 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
5185 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
5192 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5193 struct x86_instruction_info *info,
5194 enum x86_intercept_stage stage)
5196 struct vcpu_svm *svm = to_svm(vcpu);
5197 int vmexit, ret = X86EMUL_CONTINUE;
5198 struct __x86_intercept icpt_info;
5199 struct vmcb *vmcb = svm->vmcb;
5201 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5204 icpt_info = x86_intercept_map[info->intercept];
5206 if (stage != icpt_info.stage)
5209 switch (icpt_info.exit_code) {
5210 case SVM_EXIT_READ_CR0:
5211 if (info->intercept == x86_intercept_cr_read)
5212 icpt_info.exit_code += info->modrm_reg;
5214 case SVM_EXIT_WRITE_CR0: {
5215 unsigned long cr0, val;
5218 if (info->intercept == x86_intercept_cr_write)
5219 icpt_info.exit_code += info->modrm_reg;
5221 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
5222 info->intercept == x86_intercept_clts)
5225 intercept = svm->nested.intercept;
5227 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
5230 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
5231 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
5233 if (info->intercept == x86_intercept_lmsw) {
5236 /* lmsw can't clear PE - catch this here */
5237 if (cr0 & X86_CR0_PE)
5242 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
5246 case SVM_EXIT_READ_DR0:
5247 case SVM_EXIT_WRITE_DR0:
5248 icpt_info.exit_code += info->modrm_reg;
5251 if (info->intercept == x86_intercept_wrmsr)
5252 vmcb->control.exit_info_1 = 1;
5254 vmcb->control.exit_info_1 = 0;
5256 case SVM_EXIT_PAUSE:
5258 * We get this for NOP only, but pause
5259 * is rep not, check this here
5261 if (info->rep_prefix != REPE_PREFIX)
5263 case SVM_EXIT_IOIO: {
5267 if (info->intercept == x86_intercept_in ||
5268 info->intercept == x86_intercept_ins) {
5269 exit_info = ((info->src_val & 0xffff) << 16) |
5271 bytes = info->dst_bytes;
5273 exit_info = (info->dst_val & 0xffff) << 16;
5274 bytes = info->src_bytes;
5277 if (info->intercept == x86_intercept_outs ||
5278 info->intercept == x86_intercept_ins)
5279 exit_info |= SVM_IOIO_STR_MASK;
5281 if (info->rep_prefix)
5282 exit_info |= SVM_IOIO_REP_MASK;
5284 bytes = min(bytes, 4u);
5286 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
5288 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
5290 vmcb->control.exit_info_1 = exit_info;
5291 vmcb->control.exit_info_2 = info->next_rip;
5299 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5300 if (static_cpu_has(X86_FEATURE_NRIPS))
5301 vmcb->control.next_rip = info->next_rip;
5302 vmcb->control.exit_code = icpt_info.exit_code;
5303 vmexit = nested_svm_exit_handled(svm);
5305 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
5312 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
5316 * We must have an instruction with interrupts enabled, so
5317 * the timer interrupt isn't delayed by the interrupt shadow.
5320 local_irq_disable();
5323 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
5327 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
5329 if (avic_handle_apic_id_update(vcpu) != 0)
5331 if (avic_handle_dfr_update(vcpu) != 0)
5333 avic_handle_ldr_update(vcpu);
5336 static void svm_setup_mce(struct kvm_vcpu *vcpu)
5338 /* [63:9] are reserved. */
5339 vcpu->arch.mcg_cap &= 0x1ff;
5342 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
5343 .cpu_has_kvm_support = has_svm,
5344 .disabled_by_bios = is_disabled,
5345 .hardware_setup = svm_hardware_setup,
5346 .hardware_unsetup = svm_hardware_unsetup,
5347 .check_processor_compatibility = svm_check_processor_compat,
5348 .hardware_enable = svm_hardware_enable,
5349 .hardware_disable = svm_hardware_disable,
5350 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
5351 .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
5353 .vcpu_create = svm_create_vcpu,
5354 .vcpu_free = svm_free_vcpu,
5355 .vcpu_reset = svm_vcpu_reset,
5357 .vm_init = avic_vm_init,
5358 .vm_destroy = avic_vm_destroy,
5360 .prepare_guest_switch = svm_prepare_guest_switch,
5361 .vcpu_load = svm_vcpu_load,
5362 .vcpu_put = svm_vcpu_put,
5363 .vcpu_blocking = svm_vcpu_blocking,
5364 .vcpu_unblocking = svm_vcpu_unblocking,
5366 .update_bp_intercept = update_bp_intercept,
5367 .get_msr = svm_get_msr,
5368 .set_msr = svm_set_msr,
5369 .get_segment_base = svm_get_segment_base,
5370 .get_segment = svm_get_segment,
5371 .set_segment = svm_set_segment,
5372 .get_cpl = svm_get_cpl,
5373 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
5374 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
5375 .decache_cr3 = svm_decache_cr3,
5376 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
5377 .set_cr0 = svm_set_cr0,
5378 .set_cr3 = svm_set_cr3,
5379 .set_cr4 = svm_set_cr4,
5380 .set_efer = svm_set_efer,
5381 .get_idt = svm_get_idt,
5382 .set_idt = svm_set_idt,
5383 .get_gdt = svm_get_gdt,
5384 .set_gdt = svm_set_gdt,
5385 .get_dr6 = svm_get_dr6,
5386 .set_dr6 = svm_set_dr6,
5387 .set_dr7 = svm_set_dr7,
5388 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
5389 .cache_reg = svm_cache_reg,
5390 .get_rflags = svm_get_rflags,
5391 .set_rflags = svm_set_rflags,
5393 .get_pkru = svm_get_pkru,
5395 .tlb_flush = svm_flush_tlb,
5397 .run = svm_vcpu_run,
5398 .handle_exit = handle_exit,
5399 .skip_emulated_instruction = skip_emulated_instruction,
5400 .set_interrupt_shadow = svm_set_interrupt_shadow,
5401 .get_interrupt_shadow = svm_get_interrupt_shadow,
5402 .patch_hypercall = svm_patch_hypercall,
5403 .set_irq = svm_set_irq,
5404 .set_nmi = svm_inject_nmi,
5405 .queue_exception = svm_queue_exception,
5406 .cancel_injection = svm_cancel_injection,
5407 .interrupt_allowed = svm_interrupt_allowed,
5408 .nmi_allowed = svm_nmi_allowed,
5409 .get_nmi_mask = svm_get_nmi_mask,
5410 .set_nmi_mask = svm_set_nmi_mask,
5411 .enable_nmi_window = enable_nmi_window,
5412 .enable_irq_window = enable_irq_window,
5413 .update_cr8_intercept = update_cr8_intercept,
5414 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
5415 .get_enable_apicv = svm_get_enable_apicv,
5416 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
5417 .load_eoi_exitmap = svm_load_eoi_exitmap,
5418 .hwapic_irr_update = svm_hwapic_irr_update,
5419 .hwapic_isr_update = svm_hwapic_isr_update,
5420 .apicv_post_state_restore = avic_post_state_restore,
5422 .set_tss_addr = svm_set_tss_addr,
5423 .get_tdp_level = get_npt_level,
5424 .get_mt_mask = svm_get_mt_mask,
5426 .get_exit_info = svm_get_exit_info,
5428 .get_lpage_level = svm_get_lpage_level,
5430 .cpuid_update = svm_cpuid_update,
5432 .rdtscp_supported = svm_rdtscp_supported,
5433 .invpcid_supported = svm_invpcid_supported,
5434 .mpx_supported = svm_mpx_supported,
5435 .xsaves_supported = svm_xsaves_supported,
5437 .set_supported_cpuid = svm_set_supported_cpuid,
5439 .has_wbinvd_exit = svm_has_wbinvd_exit,
5441 .write_tsc_offset = svm_write_tsc_offset,
5443 .set_tdp_cr3 = set_tdp_cr3,
5445 .check_intercept = svm_check_intercept,
5446 .handle_external_intr = svm_handle_external_intr,
5448 .sched_in = svm_sched_in,
5450 .pmu_ops = &amd_pmu_ops,
5451 .deliver_posted_interrupt = svm_deliver_avic_intr,
5452 .update_pi_irte = svm_update_pi_irte,
5453 .setup_mce = svm_setup_mce,
5456 static int __init svm_init(void)
5458 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
5459 __alignof__(struct vcpu_svm), THIS_MODULE);
5462 static void __exit svm_exit(void)
5467 module_init(svm_init)
5468 module_exit(svm_exit)