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>
24 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
26 static const u32 host_save_user_msrs[] = {
29 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
31 #define IOPM_SIZE PAGE_SIZE * 3
32 #define MSRPM_SIZE PAGE_SIZE * 2
34 #define MAX_DIRECT_ACCESS_MSRS 20
35 #define MSRPM_OFFSETS 16
36 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
37 extern bool npt_enabled;
40 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
42 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
44 VMCB_INTR, /* int_ctl, int_vector */
45 VMCB_NPT, /* npt_en, nCR3, gPAT */
46 VMCB_CR, /* CR0, CR3, CR4, EFER */
47 VMCB_DR, /* DR6, DR7 */
48 VMCB_DT, /* GDT, IDT */
49 VMCB_SEG, /* CS, DS, SS, ES, CPL */
50 VMCB_CR2, /* CR2 only */
51 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
52 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
53 * AVIC PHYSICAL_TABLE pointer,
54 * AVIC LOGICAL_TABLE pointer
59 /* TPR and CR2 are always written before VMRUN */
60 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
63 bool active; /* SEV enabled guest */
64 bool es_active; /* SEV-ES enabled guest */
65 unsigned int asid; /* ASID used for this guest */
66 unsigned int handle; /* SEV firmware handle */
67 int fd; /* SEV device fd */
68 unsigned long pages_locked; /* Number of pages locked */
69 struct list_head regions_list; /* List of registered regions */
70 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
71 struct kvm *enc_context_owner; /* Owner of copied encryption context */
72 struct misc_cg *misc_cg; /* For misc cgroup accounting */
78 /* Struct members for AVIC */
80 struct page *avic_logical_id_table_page;
81 struct page *avic_physical_id_table_page;
82 struct hlist_node hnode;
84 struct kvm_sev_info sev_info;
89 struct kvm_vmcb_info {
93 uint64_t asid_generation;
96 struct svm_nested_state {
97 struct kvm_vmcb_info vmcb02;
103 /* These are the merged vectors */
106 /* A VMRUN has started but has not yet been performed, so
107 * we cannot inject a nested vmexit yet. */
108 bool nested_run_pending;
110 /* cache for control fields of the guest */
111 struct vmcb_control_area ctl;
117 struct kvm_vcpu vcpu;
118 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
120 struct kvm_vmcb_info vmcb01;
121 struct kvm_vmcb_info *current_vmcb;
122 struct svm_cpu_data *svm_data;
132 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
136 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
137 * translated into the appropriate L2_CFG bits on the host to
138 * perform speculative control.
146 struct svm_nested_state nested;
149 u64 nmi_singlestep_guest_rflags;
151 unsigned int3_injected;
152 unsigned long int3_rip;
154 /* cached guest cpuid flags for faster access */
155 bool nrips_enabled : 1;
159 struct page *avic_backing_page;
160 u64 *avic_physical_id_cache;
161 bool avic_is_running;
164 * Per-vcpu list of struct amd_svm_iommu_ir:
165 * This is used mainly to store interrupt remapping information used
166 * when update the vcpu affinity. This avoids the need to scan for
167 * IRTE and try to match ga_tag in the IOMMU driver.
169 struct list_head ir_list;
170 spinlock_t ir_list_lock;
172 /* Save desired MSR intercept (read: pass-through) state */
174 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
175 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
176 } shadow_msr_intercept;
179 struct vmcb_save_area *vmsa;
181 struct kvm_host_map ghcb_map;
182 bool received_first_sipi;
184 /* SEV-ES scratch area support */
190 bool guest_state_loaded;
193 struct svm_cpu_data {
200 struct kvm_ldttss_desc *tss_desc;
202 struct page *save_area;
203 struct vmcb *current_vmcb;
205 /* index = sev_asid, value = vmcb pointer */
206 struct vmcb **sev_vmcbs;
209 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
211 void recalc_intercepts(struct vcpu_svm *svm);
213 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
215 return container_of(kvm, struct kvm_svm, kvm);
218 static inline bool sev_guest(struct kvm *kvm)
220 #ifdef CONFIG_KVM_AMD_SEV
221 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
229 static inline bool sev_es_guest(struct kvm *kvm)
231 #ifdef CONFIG_KVM_AMD_SEV
232 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
234 return sev_guest(kvm) && sev->es_active;
240 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
242 vmcb->control.clean = 0;
245 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
247 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
248 & ~VMCB_ALWAYS_DIRTY_MASK;
251 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
253 vmcb->control.clean &= ~(1 << bit);
256 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
258 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
261 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
263 return container_of(vcpu, struct vcpu_svm, vcpu);
266 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
268 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
269 __set_bit(bit, (unsigned long *)&control->intercepts);
272 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
274 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
275 __clear_bit(bit, (unsigned long *)&control->intercepts);
278 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
280 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
281 return test_bit(bit, (unsigned long *)&control->intercepts);
284 static inline void set_dr_intercepts(struct vcpu_svm *svm)
286 struct vmcb *vmcb = svm->vmcb01.ptr;
288 if (!sev_es_guest(svm->vcpu.kvm)) {
289 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
290 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
291 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
292 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
293 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
294 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
295 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
296 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
297 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
298 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
299 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
300 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
301 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
302 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
305 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
306 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
308 recalc_intercepts(svm);
311 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
313 struct vmcb *vmcb = svm->vmcb01.ptr;
315 vmcb->control.intercepts[INTERCEPT_DR] = 0;
317 /* DR7 access must remain intercepted for an SEV-ES guest */
318 if (sev_es_guest(svm->vcpu.kvm)) {
319 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
320 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
323 recalc_intercepts(svm);
326 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
328 struct vmcb *vmcb = svm->vmcb01.ptr;
330 WARN_ON_ONCE(bit >= 32);
331 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
333 recalc_intercepts(svm);
336 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
338 struct vmcb *vmcb = svm->vmcb01.ptr;
340 WARN_ON_ONCE(bit >= 32);
341 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
343 recalc_intercepts(svm);
346 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
348 struct vmcb *vmcb = svm->vmcb01.ptr;
350 vmcb_set_intercept(&vmcb->control, bit);
352 recalc_intercepts(svm);
355 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
357 struct vmcb *vmcb = svm->vmcb01.ptr;
359 vmcb_clr_intercept(&vmcb->control, bit);
361 recalc_intercepts(svm);
364 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
366 return vmcb_is_intercept(&svm->vmcb->control, bit);
369 static inline bool vgif_enabled(struct vcpu_svm *svm)
371 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
374 static inline void enable_gif(struct vcpu_svm *svm)
376 if (vgif_enabled(svm))
377 svm->vmcb->control.int_ctl |= V_GIF_MASK;
379 svm->vcpu.arch.hflags |= HF_GIF_MASK;
382 static inline void disable_gif(struct vcpu_svm *svm)
384 if (vgif_enabled(svm))
385 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
387 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
390 static inline bool gif_set(struct vcpu_svm *svm)
392 if (vgif_enabled(svm))
393 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
395 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
399 #define MSR_INVALID 0xffffffffU
403 extern bool dump_invalid_vmcb;
405 u32 svm_msrpm_offset(u32 msr);
406 u32 *svm_vcpu_alloc_msrpm(void);
407 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
408 void svm_vcpu_free_msrpm(u32 *msrpm);
410 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
411 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
412 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
413 void svm_flush_tlb(struct kvm_vcpu *vcpu);
414 void disable_nmi_singlestep(struct vcpu_svm *svm);
415 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
416 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
417 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
418 void svm_set_gif(struct vcpu_svm *svm, bool value);
419 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
420 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
421 int read, int write);
425 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
426 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
427 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
429 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
431 struct vcpu_svm *svm = to_svm(vcpu);
433 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
436 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
438 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
441 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
443 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
446 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
448 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
451 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, struct vmcb *vmcb12);
452 void svm_leave_nested(struct vcpu_svm *svm);
453 void svm_free_nested(struct vcpu_svm *svm);
454 int svm_allocate_nested(struct vcpu_svm *svm);
455 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
456 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
457 int nested_svm_vmexit(struct vcpu_svm *svm);
459 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
461 svm->vmcb->control.exit_code = exit_code;
462 svm->vmcb->control.exit_info_1 = 0;
463 svm->vmcb->control.exit_info_2 = 0;
464 return nested_svm_vmexit(svm);
467 int nested_svm_exit_handled(struct vcpu_svm *svm);
468 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
469 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
470 bool has_error_code, u32 error_code);
471 int nested_svm_exit_special(struct vcpu_svm *svm);
472 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
473 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
474 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
476 extern struct kvm_x86_nested_ops svm_nested_ops;
480 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
481 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
482 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
484 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
485 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
486 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
487 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
489 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
493 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
495 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
496 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
499 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
501 struct vcpu_svm *svm = to_svm(vcpu);
502 u64 *entry = svm->avic_physical_id_cache;
507 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
510 int avic_ga_log_notifier(u32 ga_tag);
511 void avic_vm_destroy(struct kvm *kvm);
512 int avic_vm_init(struct kvm *kvm);
513 void avic_init_vmcb(struct vcpu_svm *svm);
514 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
515 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
516 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
517 int avic_init_vcpu(struct vcpu_svm *svm);
518 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
519 void avic_vcpu_put(struct kvm_vcpu *vcpu);
520 void avic_post_state_restore(struct kvm_vcpu *vcpu);
521 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
522 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
523 bool svm_check_apicv_inhibit_reasons(ulong bit);
524 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
525 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
526 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
527 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
528 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
529 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
530 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
531 uint32_t guest_irq, bool set);
532 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
533 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
537 #define GHCB_VERSION_MAX 1ULL
538 #define GHCB_VERSION_MIN 1ULL
540 #define GHCB_MSR_INFO_POS 0
541 #define GHCB_MSR_INFO_MASK (BIT_ULL(12) - 1)
543 #define GHCB_MSR_SEV_INFO_RESP 0x001
544 #define GHCB_MSR_SEV_INFO_REQ 0x002
545 #define GHCB_MSR_VER_MAX_POS 48
546 #define GHCB_MSR_VER_MAX_MASK 0xffff
547 #define GHCB_MSR_VER_MIN_POS 32
548 #define GHCB_MSR_VER_MIN_MASK 0xffff
549 #define GHCB_MSR_CBIT_POS 24
550 #define GHCB_MSR_CBIT_MASK 0xff
551 #define GHCB_MSR_SEV_INFO(_max, _min, _cbit) \
552 ((((_max) & GHCB_MSR_VER_MAX_MASK) << GHCB_MSR_VER_MAX_POS) | \
553 (((_min) & GHCB_MSR_VER_MIN_MASK) << GHCB_MSR_VER_MIN_POS) | \
554 (((_cbit) & GHCB_MSR_CBIT_MASK) << GHCB_MSR_CBIT_POS) | \
555 GHCB_MSR_SEV_INFO_RESP)
557 #define GHCB_MSR_CPUID_REQ 0x004
558 #define GHCB_MSR_CPUID_RESP 0x005
559 #define GHCB_MSR_CPUID_FUNC_POS 32
560 #define GHCB_MSR_CPUID_FUNC_MASK 0xffffffff
561 #define GHCB_MSR_CPUID_VALUE_POS 32
562 #define GHCB_MSR_CPUID_VALUE_MASK 0xffffffff
563 #define GHCB_MSR_CPUID_REG_POS 30
564 #define GHCB_MSR_CPUID_REG_MASK 0x3
566 #define GHCB_MSR_TERM_REQ 0x100
567 #define GHCB_MSR_TERM_REASON_SET_POS 12
568 #define GHCB_MSR_TERM_REASON_SET_MASK 0xf
569 #define GHCB_MSR_TERM_REASON_POS 16
570 #define GHCB_MSR_TERM_REASON_MASK 0xff
572 extern unsigned int max_sev_asid;
574 static inline bool svm_sev_enabled(void)
576 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
579 void sev_vm_destroy(struct kvm *kvm);
580 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
581 int svm_register_enc_region(struct kvm *kvm,
582 struct kvm_enc_region *range);
583 int svm_unregister_enc_region(struct kvm *kvm,
584 struct kvm_enc_region *range);
585 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
586 void pre_sev_run(struct vcpu_svm *svm, int cpu);
587 void __init sev_hardware_setup(void);
588 void sev_hardware_teardown(void);
589 void sev_free_vcpu(struct kvm_vcpu *vcpu);
590 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
591 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
592 void sev_es_init_vmcb(struct vcpu_svm *svm);
593 void sev_es_create_vcpu(struct vcpu_svm *svm);
594 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
595 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
599 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
600 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);