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 */
83 atomic_t migration_in_progress;
89 /* Struct members for AVIC */
91 struct page *avic_logical_id_table_page;
92 struct page *avic_physical_id_table_page;
93 struct hlist_node hnode;
95 struct kvm_sev_info sev_info;
100 struct kvm_vmcb_info {
104 uint64_t asid_generation;
107 struct svm_nested_state {
108 struct kvm_vmcb_info vmcb02;
114 /* These are the merged vectors */
117 /* A VMRUN has started but has not yet been performed, so
118 * we cannot inject a nested vmexit yet. */
119 bool nested_run_pending;
121 /* cache for control fields of the guest */
122 struct vmcb_control_area ctl;
127 struct vcpu_sev_es_state {
129 struct vmcb_save_area *vmsa;
131 struct kvm_host_map ghcb_map;
132 bool received_first_sipi;
134 /* SEV-ES scratch area support */
142 struct kvm_vcpu vcpu;
143 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
145 struct kvm_vmcb_info vmcb01;
146 struct kvm_vmcb_info *current_vmcb;
147 struct svm_cpu_data *svm_data;
161 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
162 * translated into the appropriate L2_CFG bits on the host to
163 * perform speculative control.
171 struct svm_nested_state nested;
174 u64 nmi_singlestep_guest_rflags;
176 unsigned int3_injected;
177 unsigned long int3_rip;
179 /* cached guest cpuid flags for faster access */
180 bool nrips_enabled : 1;
181 bool tsc_scaling_enabled : 1;
185 struct page *avic_backing_page;
186 u64 *avic_physical_id_cache;
187 bool avic_is_running;
190 * Per-vcpu list of struct amd_svm_iommu_ir:
191 * This is used mainly to store interrupt remapping information used
192 * when update the vcpu affinity. This avoids the need to scan for
193 * IRTE and try to match ga_tag in the IOMMU driver.
195 struct list_head ir_list;
196 spinlock_t ir_list_lock;
198 /* Save desired MSR intercept (read: pass-through) state */
200 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
201 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
202 } shadow_msr_intercept;
204 struct vcpu_sev_es_state sev_es;
206 bool guest_state_loaded;
209 struct svm_cpu_data {
216 struct kvm_ldttss_desc *tss_desc;
218 struct page *save_area;
219 struct vmcb *current_vmcb;
221 /* index = sev_asid, value = vmcb pointer */
222 struct vmcb **sev_vmcbs;
225 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
227 void recalc_intercepts(struct vcpu_svm *svm);
229 static __always_inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
231 return container_of(kvm, struct kvm_svm, kvm);
234 static __always_inline bool sev_guest(struct kvm *kvm)
236 #ifdef CONFIG_KVM_AMD_SEV
237 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
245 static __always_inline bool sev_es_guest(struct kvm *kvm)
247 #ifdef CONFIG_KVM_AMD_SEV
248 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
250 return sev->es_active && !WARN_ON_ONCE(!sev->active);
256 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
258 vmcb->control.clean = 0;
261 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
263 vmcb->control.clean = VMCB_ALL_CLEAN_MASK
264 & ~VMCB_ALWAYS_DIRTY_MASK;
267 static inline bool vmcb_is_clean(struct vmcb *vmcb, int bit)
269 return (vmcb->control.clean & (1 << bit));
272 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
274 vmcb->control.clean &= ~(1 << bit);
277 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
279 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
282 static __always_inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
284 return container_of(vcpu, struct vcpu_svm, vcpu);
287 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
289 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
290 __set_bit(bit, (unsigned long *)&control->intercepts);
293 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
295 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
296 __clear_bit(bit, (unsigned long *)&control->intercepts);
299 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
301 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
302 return test_bit(bit, (unsigned long *)&control->intercepts);
305 static inline void set_dr_intercepts(struct vcpu_svm *svm)
307 struct vmcb *vmcb = svm->vmcb01.ptr;
309 if (!sev_es_guest(svm->vcpu.kvm)) {
310 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
311 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
312 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
313 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
314 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
315 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
316 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
317 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
318 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
319 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
320 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
321 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
322 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
323 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
326 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
327 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
329 recalc_intercepts(svm);
332 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
334 struct vmcb *vmcb = svm->vmcb01.ptr;
336 vmcb->control.intercepts[INTERCEPT_DR] = 0;
338 /* DR7 access must remain intercepted for an SEV-ES guest */
339 if (sev_es_guest(svm->vcpu.kvm)) {
340 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
341 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
344 recalc_intercepts(svm);
347 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
349 struct vmcb *vmcb = svm->vmcb01.ptr;
351 WARN_ON_ONCE(bit >= 32);
352 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
354 recalc_intercepts(svm);
357 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
359 struct vmcb *vmcb = svm->vmcb01.ptr;
361 WARN_ON_ONCE(bit >= 32);
362 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
364 recalc_intercepts(svm);
367 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
369 struct vmcb *vmcb = svm->vmcb01.ptr;
371 vmcb_set_intercept(&vmcb->control, bit);
373 recalc_intercepts(svm);
376 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
378 struct vmcb *vmcb = svm->vmcb01.ptr;
380 vmcb_clr_intercept(&vmcb->control, bit);
382 recalc_intercepts(svm);
385 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
387 return vmcb_is_intercept(&svm->vmcb->control, bit);
390 static inline bool vgif_enabled(struct vcpu_svm *svm)
392 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
395 static inline void enable_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 void disable_gif(struct vcpu_svm *svm)
405 if (vgif_enabled(svm))
406 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
408 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
411 static inline bool gif_set(struct vcpu_svm *svm)
413 if (vgif_enabled(svm))
414 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
416 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
420 #define MSR_INVALID 0xffffffffU
422 extern bool dump_invalid_vmcb;
424 u32 svm_msrpm_offset(u32 msr);
425 u32 *svm_vcpu_alloc_msrpm(void);
426 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
427 void svm_vcpu_free_msrpm(u32 *msrpm);
429 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
430 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
431 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
432 void svm_flush_tlb(struct kvm_vcpu *vcpu);
433 void disable_nmi_singlestep(struct vcpu_svm *svm);
434 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
435 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
436 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
437 void svm_set_gif(struct vcpu_svm *svm, bool value);
438 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
439 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
440 int read, int write);
444 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
445 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
446 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
448 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
450 struct vcpu_svm *svm = to_svm(vcpu);
452 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
455 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
457 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
460 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
462 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
465 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
467 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
470 int enter_svm_guest_mode(struct kvm_vcpu *vcpu,
471 u64 vmcb_gpa, struct vmcb *vmcb12, bool from_vmrun);
472 void svm_leave_nested(struct vcpu_svm *svm);
473 void svm_free_nested(struct vcpu_svm *svm);
474 int svm_allocate_nested(struct vcpu_svm *svm);
475 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
476 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
477 struct vmcb_save_area *from_save);
478 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
479 int nested_svm_vmexit(struct vcpu_svm *svm);
481 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
483 svm->vmcb->control.exit_code = exit_code;
484 svm->vmcb->control.exit_info_1 = 0;
485 svm->vmcb->control.exit_info_2 = 0;
486 return nested_svm_vmexit(svm);
489 int nested_svm_exit_handled(struct vcpu_svm *svm);
490 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
491 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
492 bool has_error_code, u32 error_code);
493 int nested_svm_exit_special(struct vcpu_svm *svm);
494 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
495 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier);
496 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
497 struct vmcb_control_area *control);
498 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
499 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
500 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
502 extern struct kvm_x86_nested_ops svm_nested_ops;
506 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
507 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
508 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
510 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
511 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
512 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
513 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
515 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
517 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
519 struct vcpu_svm *svm = to_svm(vcpu);
520 u64 *entry = svm->avic_physical_id_cache;
525 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
528 int avic_ga_log_notifier(u32 ga_tag);
529 void avic_vm_destroy(struct kvm *kvm);
530 int avic_vm_init(struct kvm *kvm);
531 void avic_init_vmcb(struct vcpu_svm *svm);
532 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
533 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
534 int avic_init_vcpu(struct vcpu_svm *svm);
535 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
536 void avic_vcpu_put(struct kvm_vcpu *vcpu);
537 void avic_post_state_restore(struct kvm_vcpu *vcpu);
538 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
539 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
540 bool svm_check_apicv_inhibit_reasons(ulong bit);
541 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
542 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
543 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
544 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
545 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
546 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
547 uint32_t guest_irq, bool set);
548 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
549 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
553 #define GHCB_VERSION_MAX 1ULL
554 #define GHCB_VERSION_MIN 1ULL
557 extern unsigned int max_sev_asid;
559 void sev_vm_destroy(struct kvm *kvm);
560 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
561 int svm_register_enc_region(struct kvm *kvm,
562 struct kvm_enc_region *range);
563 int svm_unregister_enc_region(struct kvm *kvm,
564 struct kvm_enc_region *range);
565 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
566 int svm_vm_migrate_from(struct kvm *kvm, unsigned int source_fd);
567 void pre_sev_run(struct vcpu_svm *svm, int cpu);
568 void __init sev_set_cpu_caps(void);
569 void __init sev_hardware_setup(void);
570 void sev_hardware_teardown(void);
571 int sev_cpu_init(struct svm_cpu_data *sd);
572 void sev_free_vcpu(struct kvm_vcpu *vcpu);
573 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
574 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
575 void sev_es_init_vmcb(struct vcpu_svm *svm);
576 void sev_es_vcpu_reset(struct vcpu_svm *svm);
577 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
578 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
579 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
583 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
584 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);