1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Yaniv Kamay <yaniv@qumranet.com>
12 * Avi Kivity <avi@qumranet.com>
18 #include <linux/kvm_types.h>
19 #include <linux/kvm_host.h>
20 #include <linux/bits.h>
23 #include <asm/sev-common.h>
25 #include "kvm_cache_regs.h"
27 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
29 #define IOPM_SIZE PAGE_SIZE * 3
30 #define MSRPM_SIZE PAGE_SIZE * 2
32 #define MAX_DIRECT_ACCESS_MSRS 21
33 #define MSRPM_OFFSETS 16
34 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
35 extern bool npt_enabled;
37 extern bool intercept_smi;
41 * VMCB_ALL_CLEAN_MASK might also need to
42 * be updated if this enum is modified.
45 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
47 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
49 VMCB_INTR, /* int_ctl, int_vector */
50 VMCB_NPT, /* npt_en, nCR3, gPAT */
51 VMCB_CR, /* CR0, CR3, CR4, EFER */
52 VMCB_DR, /* DR6, DR7 */
53 VMCB_DT, /* GDT, IDT */
54 VMCB_SEG, /* CS, DS, SS, ES, CPL */
55 VMCB_CR2, /* CR2 only */
56 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
57 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
58 * AVIC PHYSICAL_TABLE pointer,
59 * AVIC LOGICAL_TABLE pointer
61 VMCB_SW = 31, /* Reserved for hypervisor/software use */
64 #define VMCB_ALL_CLEAN_MASK ( \
65 (1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) | \
66 (1U << VMCB_ASID) | (1U << VMCB_INTR) | \
67 (1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) | \
68 (1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) | \
69 (1U << VMCB_LBR) | (1U << VMCB_AVIC) | \
72 /* TPR and CR2 are always written before VMRUN */
73 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
76 bool active; /* SEV enabled guest */
77 bool es_active; /* SEV-ES enabled guest */
78 unsigned int asid; /* ASID used for this guest */
79 unsigned int handle; /* SEV firmware handle */
80 int fd; /* SEV device fd */
81 unsigned long pages_locked; /* Number of pages locked */
82 struct list_head regions_list; /* List of registered regions */
83 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
84 struct kvm *enc_context_owner; /* Owner of copied encryption context */
85 struct list_head mirror_vms; /* List of VMs mirroring */
86 struct list_head mirror_entry; /* Use as a list entry of mirrors */
87 struct misc_cg *misc_cg; /* For misc cgroup accounting */
88 atomic_t migration_in_progress;
94 /* Struct members for AVIC */
96 struct page *avic_logical_id_table_page;
97 struct page *avic_physical_id_table_page;
98 struct hlist_node hnode;
100 struct kvm_sev_info sev_info;
105 struct kvm_vmcb_info {
109 uint64_t asid_generation;
112 struct vmcb_save_area_cached {
121 struct vmcb_ctrl_area_cached {
122 u32 intercepts[MAX_INTERCEPT];
123 u16 pause_filter_thresh;
124 u16 pause_filter_count;
138 u32 exit_int_info_err;
148 struct svm_nested_state {
149 struct kvm_vmcb_info vmcb02;
155 /* These are the merged vectors */
158 /* A VMRUN has started but has not yet been performed, so
159 * we cannot inject a nested vmexit yet. */
160 bool nested_run_pending;
162 /* cache for control fields of the guest */
163 struct vmcb_ctrl_area_cached ctl;
166 * Note: this struct is not kept up-to-date while L2 runs; it is only
167 * valid within nested_svm_vmrun.
169 struct vmcb_save_area_cached save;
174 * Indicates whether MSR bitmap for L2 needs to be rebuilt due to
175 * changes in MSR bitmap for L1 or switching to a different L2. Note,
176 * this flag can only be used reliably in conjunction with a paravirt L1
177 * which informs L0 whether any changes to MSR bitmap for L2 were done
180 bool force_msr_bitmap_recalc;
183 struct vcpu_sev_es_state {
185 struct sev_es_save_area *vmsa;
187 struct kvm_host_map ghcb_map;
188 bool received_first_sipi;
190 /* SEV-ES scratch area support */
198 struct kvm_vcpu vcpu;
199 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
201 struct kvm_vmcb_info vmcb01;
202 struct kvm_vmcb_info *current_vmcb;
203 struct svm_cpu_data *svm_data;
217 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
218 * translated into the appropriate L2_CFG bits on the host to
219 * perform speculative control.
227 struct svm_nested_state nested;
230 u64 nmi_singlestep_guest_rflags;
232 unsigned int3_injected;
233 unsigned long int3_rip;
235 /* optional nested SVM features that are enabled for this guest */
236 bool nrips_enabled : 1;
237 bool tsc_scaling_enabled : 1;
238 bool v_vmload_vmsave_enabled : 1;
239 bool lbrv_enabled : 1;
240 bool pause_filter_enabled : 1;
241 bool pause_threshold_enabled : 1;
242 bool vgif_enabled : 1;
246 struct page *avic_backing_page;
247 u64 *avic_physical_id_cache;
250 * Per-vcpu list of struct amd_svm_iommu_ir:
251 * This is used mainly to store interrupt remapping information used
252 * when update the vcpu affinity. This avoids the need to scan for
253 * IRTE and try to match ga_tag in the IOMMU driver.
255 struct list_head ir_list;
256 spinlock_t ir_list_lock;
258 /* Save desired MSR intercept (read: pass-through) state */
260 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
261 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
262 } shadow_msr_intercept;
264 struct vcpu_sev_es_state sev_es;
266 bool guest_state_loaded;
269 struct svm_cpu_data {
276 struct kvm_ldttss_desc *tss_desc;
278 struct page *save_area;
279 struct vmcb *current_vmcb;
281 /* index = sev_asid, value = vmcb pointer */
282 struct vmcb **sev_vmcbs;
285 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
287 void recalc_intercepts(struct vcpu_svm *svm);
289 static __always_inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
291 return container_of(kvm, struct kvm_svm, kvm);
294 static __always_inline bool sev_guest(struct kvm *kvm)
296 #ifdef CONFIG_KVM_AMD_SEV
297 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
305 static __always_inline bool sev_es_guest(struct kvm *kvm)
307 #ifdef CONFIG_KVM_AMD_SEV
308 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
310 return sev->es_active && !WARN_ON_ONCE(!sev->active);
316 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
318 vmcb->control.clean = 0;
321 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
323 vmcb->control.clean = VMCB_ALL_CLEAN_MASK
324 & ~VMCB_ALWAYS_DIRTY_MASK;
327 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
329 vmcb->control.clean &= ~(1 << bit);
332 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
334 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
337 static __always_inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
339 return container_of(vcpu, struct vcpu_svm, vcpu);
343 * Only the PDPTRs are loaded on demand into the shadow MMU. All other
344 * fields are synchronized on VM-Exit, because accessing the VMCB is cheap.
346 * CR3 might be out of date in the VMCB but it is not marked dirty; instead,
347 * KVM_REQ_LOAD_MMU_PGD is always requested when the cached vcpu->arch.cr3
348 * is changed. svm_load_mmu_pgd() then syncs the new CR3 value into the VMCB.
350 #define SVM_REGS_LAZY_LOAD_SET (1 << VCPU_EXREG_PDPTR)
352 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
354 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
355 __set_bit(bit, (unsigned long *)&control->intercepts);
358 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
360 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
361 __clear_bit(bit, (unsigned long *)&control->intercepts);
364 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
366 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
367 return test_bit(bit, (unsigned long *)&control->intercepts);
370 static inline bool vmcb12_is_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
372 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
373 return test_bit(bit, (unsigned long *)&control->intercepts);
376 static inline void set_dr_intercepts(struct vcpu_svm *svm)
378 struct vmcb *vmcb = svm->vmcb01.ptr;
380 if (!sev_es_guest(svm->vcpu.kvm)) {
381 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
382 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
383 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
384 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
385 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
386 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
387 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
388 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
389 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
390 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
391 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
392 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
393 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
394 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
397 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
398 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
400 recalc_intercepts(svm);
403 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
405 struct vmcb *vmcb = svm->vmcb01.ptr;
407 vmcb->control.intercepts[INTERCEPT_DR] = 0;
409 /* DR7 access must remain intercepted for an SEV-ES guest */
410 if (sev_es_guest(svm->vcpu.kvm)) {
411 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
412 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
415 recalc_intercepts(svm);
418 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
420 struct vmcb *vmcb = svm->vmcb01.ptr;
422 WARN_ON_ONCE(bit >= 32);
423 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
425 recalc_intercepts(svm);
428 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
430 struct vmcb *vmcb = svm->vmcb01.ptr;
432 WARN_ON_ONCE(bit >= 32);
433 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
435 recalc_intercepts(svm);
438 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
440 struct vmcb *vmcb = svm->vmcb01.ptr;
442 vmcb_set_intercept(&vmcb->control, bit);
444 recalc_intercepts(svm);
447 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
449 struct vmcb *vmcb = svm->vmcb01.ptr;
451 vmcb_clr_intercept(&vmcb->control, bit);
453 recalc_intercepts(svm);
456 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
458 return vmcb_is_intercept(&svm->vmcb->control, bit);
461 static inline bool nested_vgif_enabled(struct vcpu_svm *svm)
463 return svm->vgif_enabled && (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK);
466 static inline struct vmcb *get_vgif_vmcb(struct vcpu_svm *svm)
471 if (is_guest_mode(&svm->vcpu) && !nested_vgif_enabled(svm))
472 return svm->nested.vmcb02.ptr;
474 return svm->vmcb01.ptr;
477 static inline void enable_gif(struct vcpu_svm *svm)
479 struct vmcb *vmcb = get_vgif_vmcb(svm);
482 vmcb->control.int_ctl |= V_GIF_MASK;
484 svm->vcpu.arch.hflags |= HF_GIF_MASK;
487 static inline void disable_gif(struct vcpu_svm *svm)
489 struct vmcb *vmcb = get_vgif_vmcb(svm);
492 vmcb->control.int_ctl &= ~V_GIF_MASK;
494 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
497 static inline bool gif_set(struct vcpu_svm *svm)
499 struct vmcb *vmcb = get_vgif_vmcb(svm);
502 return !!(vmcb->control.int_ctl & V_GIF_MASK);
504 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
507 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
509 return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
513 #define MSR_INVALID 0xffffffffU
515 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
517 extern bool dump_invalid_vmcb;
519 u32 svm_msrpm_offset(u32 msr);
520 u32 *svm_vcpu_alloc_msrpm(void);
521 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
522 void svm_vcpu_free_msrpm(u32 *msrpm);
523 void svm_copy_lbrs(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
524 void svm_update_lbrv(struct kvm_vcpu *vcpu);
526 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
527 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
528 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
529 void disable_nmi_singlestep(struct vcpu_svm *svm);
530 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
531 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
532 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
533 void svm_set_gif(struct vcpu_svm *svm, bool value);
534 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
535 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
536 int read, int write);
537 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
538 int trig_mode, int vec);
542 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
543 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
544 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
546 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
548 struct vcpu_svm *svm = to_svm(vcpu);
550 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
553 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
555 return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
558 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
560 return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
563 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
565 return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
568 int enter_svm_guest_mode(struct kvm_vcpu *vcpu,
569 u64 vmcb_gpa, struct vmcb *vmcb12, bool from_vmrun);
570 void svm_leave_nested(struct kvm_vcpu *vcpu);
571 void svm_free_nested(struct vcpu_svm *svm);
572 int svm_allocate_nested(struct vcpu_svm *svm);
573 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
574 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
575 struct vmcb_save_area *from_save);
576 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
577 int nested_svm_vmexit(struct vcpu_svm *svm);
579 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
581 svm->vmcb->control.exit_code = exit_code;
582 svm->vmcb->control.exit_info_1 = 0;
583 svm->vmcb->control.exit_info_2 = 0;
584 return nested_svm_vmexit(svm);
587 int nested_svm_exit_handled(struct vcpu_svm *svm);
588 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
589 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
590 bool has_error_code, u32 error_code);
591 int nested_svm_exit_special(struct vcpu_svm *svm);
592 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
593 void __svm_write_tsc_multiplier(u64 multiplier);
594 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
595 struct vmcb_control_area *control);
596 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
597 struct vmcb_save_area *save);
598 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
599 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
600 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
602 extern struct kvm_x86_nested_ops svm_nested_ops;
606 int avic_ga_log_notifier(u32 ga_tag);
607 void avic_vm_destroy(struct kvm *kvm);
608 int avic_vm_init(struct kvm *kvm);
609 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb);
610 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
611 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
612 int avic_init_vcpu(struct vcpu_svm *svm);
613 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
614 void avic_vcpu_put(struct kvm_vcpu *vcpu);
615 void avic_apicv_post_state_restore(struct kvm_vcpu *vcpu);
616 void avic_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
617 void avic_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
618 bool avic_check_apicv_inhibit_reasons(enum kvm_apicv_inhibit reason);
619 void avic_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
620 void avic_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
621 bool avic_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
622 int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
623 uint32_t guest_irq, bool set);
624 void avic_vcpu_blocking(struct kvm_vcpu *vcpu);
625 void avic_vcpu_unblocking(struct kvm_vcpu *vcpu);
626 void avic_ring_doorbell(struct kvm_vcpu *vcpu);
627 unsigned long avic_vcpu_get_apicv_inhibit_reasons(struct kvm_vcpu *vcpu);
631 #define GHCB_VERSION_MAX 1ULL
632 #define GHCB_VERSION_MIN 1ULL
635 extern unsigned int max_sev_asid;
637 void sev_vm_destroy(struct kvm *kvm);
638 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp);
639 int sev_mem_enc_register_region(struct kvm *kvm,
640 struct kvm_enc_region *range);
641 int sev_mem_enc_unregister_region(struct kvm *kvm,
642 struct kvm_enc_region *range);
643 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd);
644 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd);
645 void sev_guest_memory_reclaimed(struct kvm *kvm);
647 void pre_sev_run(struct vcpu_svm *svm, int cpu);
648 void __init sev_set_cpu_caps(void);
649 void __init sev_hardware_setup(void);
650 void sev_hardware_unsetup(void);
651 int sev_cpu_init(struct svm_cpu_data *sd);
652 void sev_free_vcpu(struct kvm_vcpu *vcpu);
653 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
654 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
655 void sev_es_init_vmcb(struct vcpu_svm *svm);
656 void sev_es_vcpu_reset(struct vcpu_svm *svm);
657 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
658 void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa);
659 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
663 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
664 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);