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;
36 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
38 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
40 VMCB_INTR, /* int_ctl, int_vector */
41 VMCB_NPT, /* npt_en, nCR3, gPAT */
42 VMCB_CR, /* CR0, CR3, CR4, EFER */
43 VMCB_DR, /* DR6, DR7 */
44 VMCB_DT, /* GDT, IDT */
45 VMCB_SEG, /* CS, DS, SS, ES, CPL */
46 VMCB_CR2, /* CR2 only */
47 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
48 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
49 * AVIC PHYSICAL_TABLE pointer,
50 * AVIC LOGICAL_TABLE pointer
55 /* TPR and CR2 are always written before VMRUN */
56 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
59 bool active; /* SEV enabled guest */
60 bool es_active; /* SEV-ES enabled guest */
61 unsigned int asid; /* ASID used for this guest */
62 unsigned int handle; /* SEV firmware handle */
63 int fd; /* SEV device fd */
64 unsigned long pages_locked; /* Number of pages locked */
65 struct list_head regions_list; /* List of registered regions */
66 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
67 struct kvm *enc_context_owner; /* Owner of copied encryption context */
68 struct misc_cg *misc_cg; /* For misc cgroup accounting */
74 /* Struct members for AVIC */
76 struct page *avic_logical_id_table_page;
77 struct page *avic_physical_id_table_page;
78 struct hlist_node hnode;
80 struct kvm_sev_info sev_info;
85 struct kvm_vmcb_info {
89 uint64_t asid_generation;
92 struct svm_nested_state {
93 struct kvm_vmcb_info vmcb02;
99 /* These are the merged vectors */
102 /* A VMRUN has started but has not yet been performed, so
103 * we cannot inject a nested vmexit yet. */
104 bool nested_run_pending;
106 /* cache for control fields of the guest */
107 struct vmcb_control_area ctl;
113 struct kvm_vcpu vcpu;
114 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
116 struct kvm_vmcb_info vmcb01;
117 struct kvm_vmcb_info *current_vmcb;
118 struct svm_cpu_data *svm_data;
130 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
131 * translated into the appropriate L2_CFG bits on the host to
132 * perform speculative control.
140 struct svm_nested_state nested;
143 u64 nmi_singlestep_guest_rflags;
145 unsigned int3_injected;
146 unsigned long int3_rip;
148 /* cached guest cpuid flags for faster access */
149 bool nrips_enabled : 1;
153 struct page *avic_backing_page;
154 u64 *avic_physical_id_cache;
155 bool avic_is_running;
158 * Per-vcpu list of struct amd_svm_iommu_ir:
159 * This is used mainly to store interrupt remapping information used
160 * when update the vcpu affinity. This avoids the need to scan for
161 * IRTE and try to match ga_tag in the IOMMU driver.
163 struct list_head ir_list;
164 spinlock_t ir_list_lock;
166 /* Save desired MSR intercept (read: pass-through) state */
168 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
169 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
170 } shadow_msr_intercept;
173 struct vmcb_save_area *vmsa;
175 struct kvm_host_map ghcb_map;
176 bool received_first_sipi;
178 /* SEV-ES scratch area support */
184 bool guest_state_loaded;
187 struct svm_cpu_data {
194 struct kvm_ldttss_desc *tss_desc;
196 struct page *save_area;
197 struct vmcb *current_vmcb;
199 /* index = sev_asid, value = vmcb pointer */
200 struct vmcb **sev_vmcbs;
203 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
205 void recalc_intercepts(struct vcpu_svm *svm);
207 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
209 return container_of(kvm, struct kvm_svm, kvm);
212 static inline bool sev_guest(struct kvm *kvm)
214 #ifdef CONFIG_KVM_AMD_SEV
215 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
223 static inline bool sev_es_guest(struct kvm *kvm)
225 #ifdef CONFIG_KVM_AMD_SEV
226 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
228 return sev_guest(kvm) && sev->es_active;
234 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
236 vmcb->control.clean = 0;
239 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
241 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
242 & ~VMCB_ALWAYS_DIRTY_MASK;
245 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
247 vmcb->control.clean &= ~(1 << bit);
250 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
252 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
255 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
257 return container_of(vcpu, struct vcpu_svm, vcpu);
260 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
262 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
263 __set_bit(bit, (unsigned long *)&control->intercepts);
266 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
268 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
269 __clear_bit(bit, (unsigned long *)&control->intercepts);
272 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
274 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
275 return test_bit(bit, (unsigned long *)&control->intercepts);
278 static inline void set_dr_intercepts(struct vcpu_svm *svm)
280 struct vmcb *vmcb = svm->vmcb01.ptr;
282 if (!sev_es_guest(svm->vcpu.kvm)) {
283 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
284 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
285 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
286 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
287 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
288 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
289 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
290 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
291 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
292 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
293 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
294 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
295 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
296 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
299 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
300 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
302 recalc_intercepts(svm);
305 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
307 struct vmcb *vmcb = svm->vmcb01.ptr;
309 vmcb->control.intercepts[INTERCEPT_DR] = 0;
311 /* DR7 access must remain intercepted for an SEV-ES guest */
312 if (sev_es_guest(svm->vcpu.kvm)) {
313 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
314 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
317 recalc_intercepts(svm);
320 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
322 struct vmcb *vmcb = svm->vmcb01.ptr;
324 WARN_ON_ONCE(bit >= 32);
325 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
327 recalc_intercepts(svm);
330 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
332 struct vmcb *vmcb = svm->vmcb01.ptr;
334 WARN_ON_ONCE(bit >= 32);
335 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
337 recalc_intercepts(svm);
340 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
342 struct vmcb *vmcb = svm->vmcb01.ptr;
344 vmcb_set_intercept(&vmcb->control, bit);
346 recalc_intercepts(svm);
349 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
351 struct vmcb *vmcb = svm->vmcb01.ptr;
353 vmcb_clr_intercept(&vmcb->control, bit);
355 recalc_intercepts(svm);
358 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
360 return vmcb_is_intercept(&svm->vmcb->control, bit);
363 static inline bool vgif_enabled(struct vcpu_svm *svm)
365 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
368 static inline void enable_gif(struct vcpu_svm *svm)
370 if (vgif_enabled(svm))
371 svm->vmcb->control.int_ctl |= V_GIF_MASK;
373 svm->vcpu.arch.hflags |= HF_GIF_MASK;
376 static inline void disable_gif(struct vcpu_svm *svm)
378 if (vgif_enabled(svm))
379 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
381 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
384 static inline bool gif_set(struct vcpu_svm *svm)
386 if (vgif_enabled(svm))
387 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
389 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
393 #define MSR_INVALID 0xffffffffU
395 extern bool dump_invalid_vmcb;
397 u32 svm_msrpm_offset(u32 msr);
398 u32 *svm_vcpu_alloc_msrpm(void);
399 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
400 void svm_vcpu_free_msrpm(u32 *msrpm);
402 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
403 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
404 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
405 void svm_flush_tlb(struct kvm_vcpu *vcpu);
406 void disable_nmi_singlestep(struct vcpu_svm *svm);
407 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
408 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
409 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
410 void svm_set_gif(struct vcpu_svm *svm, bool value);
411 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
412 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
413 int read, int write);
417 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
418 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
419 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
421 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
423 struct vcpu_svm *svm = to_svm(vcpu);
425 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
428 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
430 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
433 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
435 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
438 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
440 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
443 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, struct vmcb *vmcb12);
444 void svm_leave_nested(struct vcpu_svm *svm);
445 void svm_free_nested(struct vcpu_svm *svm);
446 int svm_allocate_nested(struct vcpu_svm *svm);
447 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
448 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
449 int nested_svm_vmexit(struct vcpu_svm *svm);
451 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
453 svm->vmcb->control.exit_code = exit_code;
454 svm->vmcb->control.exit_info_1 = 0;
455 svm->vmcb->control.exit_info_2 = 0;
456 return nested_svm_vmexit(svm);
459 int nested_svm_exit_handled(struct vcpu_svm *svm);
460 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
461 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
462 bool has_error_code, u32 error_code);
463 int nested_svm_exit_special(struct vcpu_svm *svm);
464 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
465 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
466 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
468 extern struct kvm_x86_nested_ops svm_nested_ops;
472 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
473 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
474 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
476 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
477 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
478 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
479 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
481 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
485 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
487 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
488 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
491 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
493 struct vcpu_svm *svm = to_svm(vcpu);
494 u64 *entry = svm->avic_physical_id_cache;
499 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
502 int avic_ga_log_notifier(u32 ga_tag);
503 void avic_vm_destroy(struct kvm *kvm);
504 int avic_vm_init(struct kvm *kvm);
505 void avic_init_vmcb(struct vcpu_svm *svm);
506 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
507 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
508 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
509 int avic_init_vcpu(struct vcpu_svm *svm);
510 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
511 void avic_vcpu_put(struct kvm_vcpu *vcpu);
512 void avic_post_state_restore(struct kvm_vcpu *vcpu);
513 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
514 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
515 bool svm_check_apicv_inhibit_reasons(ulong bit);
516 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
517 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
518 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
519 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
520 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
521 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
522 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
523 uint32_t guest_irq, bool set);
524 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
525 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
529 #define GHCB_VERSION_MAX 1ULL
530 #define GHCB_VERSION_MIN 1ULL
533 extern unsigned int max_sev_asid;
535 void sev_vm_destroy(struct kvm *kvm);
536 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
537 int svm_register_enc_region(struct kvm *kvm,
538 struct kvm_enc_region *range);
539 int svm_unregister_enc_region(struct kvm *kvm,
540 struct kvm_enc_region *range);
541 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
542 void pre_sev_run(struct vcpu_svm *svm, int cpu);
543 void __init sev_set_cpu_caps(void);
544 void __init sev_hardware_setup(void);
545 void sev_hardware_teardown(void);
546 int sev_cpu_init(struct svm_cpu_data *sd);
547 void sev_free_vcpu(struct kvm_vcpu *vcpu);
548 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
549 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
550 void sev_es_init_vmcb(struct vcpu_svm *svm);
551 void sev_es_create_vcpu(struct vcpu_svm *svm);
552 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
553 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
554 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
558 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
559 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);