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 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
27 #define IOPM_SIZE PAGE_SIZE * 3
28 #define MSRPM_SIZE PAGE_SIZE * 2
30 #define MAX_DIRECT_ACCESS_MSRS 20
31 #define MSRPM_OFFSETS 16
32 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
33 extern bool npt_enabled;
34 extern bool intercept_smi;
38 * VMCB_ALL_CLEAN_MASK might also need to
39 * be updated if this enum is modified.
42 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
44 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
46 VMCB_INTR, /* int_ctl, int_vector */
47 VMCB_NPT, /* npt_en, nCR3, gPAT */
48 VMCB_CR, /* CR0, CR3, CR4, EFER */
49 VMCB_DR, /* DR6, DR7 */
50 VMCB_DT, /* GDT, IDT */
51 VMCB_SEG, /* CS, DS, SS, ES, CPL */
52 VMCB_CR2, /* CR2 only */
53 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
54 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
55 * AVIC PHYSICAL_TABLE pointer,
56 * AVIC LOGICAL_TABLE pointer
58 VMCB_SW = 31, /* Reserved for hypervisor/software use */
61 #define VMCB_ALL_CLEAN_MASK ( \
62 (1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) | \
63 (1U << VMCB_ASID) | (1U << VMCB_INTR) | \
64 (1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) | \
65 (1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) | \
66 (1U << VMCB_LBR) | (1U << VMCB_AVIC) | \
69 /* TPR and CR2 are always written before VMRUN */
70 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
73 bool active; /* SEV enabled guest */
74 bool es_active; /* SEV-ES enabled guest */
75 unsigned int asid; /* ASID used for this guest */
76 unsigned int handle; /* SEV firmware handle */
77 int fd; /* SEV device fd */
78 unsigned long pages_locked; /* Number of pages locked */
79 struct list_head regions_list; /* List of registered regions */
80 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
81 struct kvm *enc_context_owner; /* Owner of copied encryption context */
82 struct misc_cg *misc_cg; /* For misc cgroup accounting */
88 /* Struct members for AVIC */
90 struct page *avic_logical_id_table_page;
91 struct page *avic_physical_id_table_page;
92 struct hlist_node hnode;
94 struct kvm_sev_info sev_info;
99 struct kvm_vmcb_info {
103 uint64_t asid_generation;
106 struct svm_nested_state {
107 struct kvm_vmcb_info vmcb02;
113 /* These are the merged vectors */
116 /* A VMRUN has started but has not yet been performed, so
117 * we cannot inject a nested vmexit yet. */
118 bool nested_run_pending;
120 /* cache for control fields of the guest */
121 struct vmcb_control_area ctl;
127 struct kvm_vcpu vcpu;
128 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
130 struct kvm_vmcb_info vmcb01;
131 struct kvm_vmcb_info *current_vmcb;
132 struct svm_cpu_data *svm_data;
144 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
145 * translated into the appropriate L2_CFG bits on the host to
146 * perform speculative control.
154 struct svm_nested_state nested;
157 u64 nmi_singlestep_guest_rflags;
159 unsigned int3_injected;
160 unsigned long int3_rip;
162 /* cached guest cpuid flags for faster access */
163 bool nrips_enabled : 1;
167 struct page *avic_backing_page;
168 u64 *avic_physical_id_cache;
169 bool avic_is_running;
172 * Per-vcpu list of struct amd_svm_iommu_ir:
173 * This is used mainly to store interrupt remapping information used
174 * when update the vcpu affinity. This avoids the need to scan for
175 * IRTE and try to match ga_tag in the IOMMU driver.
177 struct list_head ir_list;
178 spinlock_t ir_list_lock;
180 /* Save desired MSR intercept (read: pass-through) state */
182 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
183 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
184 } shadow_msr_intercept;
187 struct vmcb_save_area *vmsa;
189 struct kvm_host_map ghcb_map;
190 bool received_first_sipi;
192 /* SEV-ES scratch area support */
198 bool guest_state_loaded;
201 struct svm_cpu_data {
208 struct kvm_ldttss_desc *tss_desc;
210 struct page *save_area;
211 struct vmcb *current_vmcb;
213 /* index = sev_asid, value = vmcb pointer */
214 struct vmcb **sev_vmcbs;
217 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
219 void recalc_intercepts(struct vcpu_svm *svm);
221 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
223 return container_of(kvm, struct kvm_svm, kvm);
226 static inline bool sev_guest(struct kvm *kvm)
228 #ifdef CONFIG_KVM_AMD_SEV
229 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
237 static inline bool sev_es_guest(struct kvm *kvm)
239 #ifdef CONFIG_KVM_AMD_SEV
240 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
242 return sev_guest(kvm) && sev->es_active;
248 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
250 vmcb->control.clean = 0;
253 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
255 vmcb->control.clean = VMCB_ALL_CLEAN_MASK
256 & ~VMCB_ALWAYS_DIRTY_MASK;
259 static inline bool vmcb_is_clean(struct vmcb *vmcb, int bit)
261 return (vmcb->control.clean & (1 << bit));
264 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
266 vmcb->control.clean &= ~(1 << bit);
269 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
271 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
274 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
276 return container_of(vcpu, struct vcpu_svm, vcpu);
279 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
281 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
282 __set_bit(bit, (unsigned long *)&control->intercepts);
285 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
287 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
288 __clear_bit(bit, (unsigned long *)&control->intercepts);
291 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
293 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
294 return test_bit(bit, (unsigned long *)&control->intercepts);
297 static inline void set_dr_intercepts(struct vcpu_svm *svm)
299 struct vmcb *vmcb = svm->vmcb01.ptr;
301 if (!sev_es_guest(svm->vcpu.kvm)) {
302 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
303 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
304 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
305 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
306 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
307 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
308 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
309 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
310 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
311 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
312 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
313 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
314 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
315 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
318 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
319 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
321 recalc_intercepts(svm);
324 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
326 struct vmcb *vmcb = svm->vmcb01.ptr;
328 vmcb->control.intercepts[INTERCEPT_DR] = 0;
330 /* DR7 access must remain intercepted for an SEV-ES guest */
331 if (sev_es_guest(svm->vcpu.kvm)) {
332 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
333 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
336 recalc_intercepts(svm);
339 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
341 struct vmcb *vmcb = svm->vmcb01.ptr;
343 WARN_ON_ONCE(bit >= 32);
344 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
346 recalc_intercepts(svm);
349 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
351 struct vmcb *vmcb = svm->vmcb01.ptr;
353 WARN_ON_ONCE(bit >= 32);
354 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
356 recalc_intercepts(svm);
359 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
361 struct vmcb *vmcb = svm->vmcb01.ptr;
363 vmcb_set_intercept(&vmcb->control, bit);
365 recalc_intercepts(svm);
368 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
370 struct vmcb *vmcb = svm->vmcb01.ptr;
372 vmcb_clr_intercept(&vmcb->control, bit);
374 recalc_intercepts(svm);
377 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
379 return vmcb_is_intercept(&svm->vmcb->control, bit);
382 static inline bool vgif_enabled(struct vcpu_svm *svm)
384 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
387 static inline void enable_gif(struct vcpu_svm *svm)
389 if (vgif_enabled(svm))
390 svm->vmcb->control.int_ctl |= V_GIF_MASK;
392 svm->vcpu.arch.hflags |= HF_GIF_MASK;
395 static inline void disable_gif(struct vcpu_svm *svm)
397 if (vgif_enabled(svm))
398 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
400 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
403 static inline bool gif_set(struct vcpu_svm *svm)
405 if (vgif_enabled(svm))
406 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
408 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
412 #define MSR_INVALID 0xffffffffU
414 extern bool dump_invalid_vmcb;
416 u32 svm_msrpm_offset(u32 msr);
417 u32 *svm_vcpu_alloc_msrpm(void);
418 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
419 void svm_vcpu_free_msrpm(u32 *msrpm);
421 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
422 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
423 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
424 void svm_flush_tlb(struct kvm_vcpu *vcpu);
425 void disable_nmi_singlestep(struct vcpu_svm *svm);
426 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
427 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
428 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
429 void svm_set_gif(struct vcpu_svm *svm, bool value);
430 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
431 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
432 int read, int write);
436 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
437 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
438 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
440 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
442 struct vcpu_svm *svm = to_svm(vcpu);
444 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
447 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
449 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
452 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
454 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
457 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
459 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
462 int enter_svm_guest_mode(struct kvm_vcpu *vcpu,
463 u64 vmcb_gpa, struct vmcb *vmcb12, bool from_vmrun);
464 void svm_leave_nested(struct vcpu_svm *svm);
465 void svm_free_nested(struct vcpu_svm *svm);
466 int svm_allocate_nested(struct vcpu_svm *svm);
467 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
468 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
469 struct vmcb_save_area *from_save);
470 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
471 int nested_svm_vmexit(struct vcpu_svm *svm);
473 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
475 svm->vmcb->control.exit_code = exit_code;
476 svm->vmcb->control.exit_info_1 = 0;
477 svm->vmcb->control.exit_info_2 = 0;
478 return nested_svm_vmexit(svm);
481 int nested_svm_exit_handled(struct vcpu_svm *svm);
482 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
483 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
484 bool has_error_code, u32 error_code);
485 int nested_svm_exit_special(struct vcpu_svm *svm);
486 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
487 struct vmcb_control_area *control);
488 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
489 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
490 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
492 extern struct kvm_x86_nested_ops svm_nested_ops;
496 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
497 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
498 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
500 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
501 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
502 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
503 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
505 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
507 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
509 struct vcpu_svm *svm = to_svm(vcpu);
510 u64 *entry = svm->avic_physical_id_cache;
515 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
518 int avic_ga_log_notifier(u32 ga_tag);
519 void avic_vm_destroy(struct kvm *kvm);
520 int avic_vm_init(struct kvm *kvm);
521 void avic_init_vmcb(struct vcpu_svm *svm);
522 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
523 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
524 int avic_init_vcpu(struct vcpu_svm *svm);
525 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
526 void avic_vcpu_put(struct kvm_vcpu *vcpu);
527 void avic_post_state_restore(struct kvm_vcpu *vcpu);
528 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
529 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
530 bool svm_check_apicv_inhibit_reasons(ulong bit);
531 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
532 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
533 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
534 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
535 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
536 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
537 uint32_t guest_irq, bool set);
538 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
539 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
543 #define GHCB_VERSION_MAX 1ULL
544 #define GHCB_VERSION_MIN 1ULL
547 extern unsigned int max_sev_asid;
549 void sev_vm_destroy(struct kvm *kvm);
550 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
551 int svm_register_enc_region(struct kvm *kvm,
552 struct kvm_enc_region *range);
553 int svm_unregister_enc_region(struct kvm *kvm,
554 struct kvm_enc_region *range);
555 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
556 void pre_sev_run(struct vcpu_svm *svm, int cpu);
557 void __init sev_set_cpu_caps(void);
558 void __init sev_hardware_setup(void);
559 void sev_hardware_teardown(void);
560 int sev_cpu_init(struct svm_cpu_data *sd);
561 void sev_free_vcpu(struct kvm_vcpu *vcpu);
562 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
563 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
564 void sev_es_init_vmcb(struct vcpu_svm *svm);
565 void sev_es_create_vcpu(struct vcpu_svm *svm);
566 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
567 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
568 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
572 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
573 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);