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 */
77 /* Struct members for AVIC */
79 struct page *avic_logical_id_table_page;
80 struct page *avic_physical_id_table_page;
81 struct hlist_node hnode;
83 struct kvm_sev_info sev_info;
88 struct kvm_vmcb_info {
92 uint64_t asid_generation;
95 struct svm_nested_state {
96 struct kvm_vmcb_info vmcb02;
102 /* These are the merged vectors */
105 /* A VMRUN has started but has not yet been performed, so
106 * we cannot inject a nested vmexit yet. */
107 bool nested_run_pending;
109 /* cache for control fields of the guest */
110 struct vmcb_control_area ctl;
116 struct kvm_vcpu vcpu;
117 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
119 struct kvm_vmcb_info vmcb01;
120 struct kvm_vmcb_info *current_vmcb;
121 struct svm_cpu_data *svm_data;
131 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
135 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
136 * translated into the appropriate L2_CFG bits on the host to
137 * perform speculative control.
145 struct svm_nested_state nested;
148 u64 nmi_singlestep_guest_rflags;
150 unsigned int3_injected;
151 unsigned long int3_rip;
153 /* cached guest cpuid flags for faster access */
154 bool nrips_enabled : 1;
158 struct page *avic_backing_page;
159 u64 *avic_physical_id_cache;
160 bool avic_is_running;
163 * Per-vcpu list of struct amd_svm_iommu_ir:
164 * This is used mainly to store interrupt remapping information used
165 * when update the vcpu affinity. This avoids the need to scan for
166 * IRTE and try to match ga_tag in the IOMMU driver.
168 struct list_head ir_list;
169 spinlock_t ir_list_lock;
171 /* Save desired MSR intercept (read: pass-through) state */
173 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
174 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
175 } shadow_msr_intercept;
178 struct vmcb_save_area *vmsa;
180 struct kvm_host_map ghcb_map;
181 bool received_first_sipi;
183 /* SEV-ES scratch area support */
189 bool guest_state_loaded;
192 struct svm_cpu_data {
199 struct kvm_ldttss_desc *tss_desc;
201 struct page *save_area;
202 struct vmcb *current_vmcb;
204 /* index = sev_asid, value = vmcb pointer */
205 struct vmcb **sev_vmcbs;
208 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
210 void recalc_intercepts(struct vcpu_svm *svm);
212 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
214 return container_of(kvm, struct kvm_svm, kvm);
217 static inline bool sev_guest(struct kvm *kvm)
219 #ifdef CONFIG_KVM_AMD_SEV
220 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
228 static inline bool sev_es_guest(struct kvm *kvm)
230 #ifdef CONFIG_KVM_AMD_SEV
231 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
233 return sev_guest(kvm) && sev->es_active;
239 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
241 vmcb->control.clean = 0;
244 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
246 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
247 & ~VMCB_ALWAYS_DIRTY_MASK;
250 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
252 vmcb->control.clean &= ~(1 << bit);
255 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
257 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
260 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
262 return container_of(vcpu, struct vcpu_svm, vcpu);
265 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
267 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
268 __set_bit(bit, (unsigned long *)&control->intercepts);
271 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
273 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
274 __clear_bit(bit, (unsigned long *)&control->intercepts);
277 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
279 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
280 return test_bit(bit, (unsigned long *)&control->intercepts);
283 static inline void set_dr_intercepts(struct vcpu_svm *svm)
285 struct vmcb *vmcb = svm->vmcb01.ptr;
287 if (!sev_es_guest(svm->vcpu.kvm)) {
288 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
289 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
290 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
291 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
292 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
293 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
294 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
295 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
296 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
297 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
298 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
299 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
300 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
301 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
304 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
305 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
307 recalc_intercepts(svm);
310 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
312 struct vmcb *vmcb = svm->vmcb01.ptr;
314 vmcb->control.intercepts[INTERCEPT_DR] = 0;
316 /* DR7 access must remain intercepted for an SEV-ES guest */
317 if (sev_es_guest(svm->vcpu.kvm)) {
318 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
319 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
322 recalc_intercepts(svm);
325 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
327 struct vmcb *vmcb = svm->vmcb01.ptr;
329 WARN_ON_ONCE(bit >= 32);
330 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
332 recalc_intercepts(svm);
335 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
337 struct vmcb *vmcb = svm->vmcb01.ptr;
339 WARN_ON_ONCE(bit >= 32);
340 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
342 recalc_intercepts(svm);
345 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
347 struct vmcb *vmcb = svm->vmcb01.ptr;
349 vmcb_set_intercept(&vmcb->control, bit);
351 recalc_intercepts(svm);
354 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
356 struct vmcb *vmcb = svm->vmcb01.ptr;
358 vmcb_clr_intercept(&vmcb->control, bit);
360 recalc_intercepts(svm);
363 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
365 return vmcb_is_intercept(&svm->vmcb->control, bit);
368 static inline bool vgif_enabled(struct vcpu_svm *svm)
370 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
373 static inline void enable_gif(struct vcpu_svm *svm)
375 if (vgif_enabled(svm))
376 svm->vmcb->control.int_ctl |= V_GIF_MASK;
378 svm->vcpu.arch.hflags |= HF_GIF_MASK;
381 static inline void disable_gif(struct vcpu_svm *svm)
383 if (vgif_enabled(svm))
384 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
386 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
389 static inline bool gif_set(struct vcpu_svm *svm)
391 if (vgif_enabled(svm))
392 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
394 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
398 #define MSR_INVALID 0xffffffffU
402 extern bool dump_invalid_vmcb;
404 u32 svm_msrpm_offset(u32 msr);
405 u32 *svm_vcpu_alloc_msrpm(void);
406 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
407 void svm_vcpu_free_msrpm(u32 *msrpm);
409 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
410 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
411 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
412 void svm_flush_tlb(struct kvm_vcpu *vcpu);
413 void disable_nmi_singlestep(struct vcpu_svm *svm);
414 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
415 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
416 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
417 void svm_set_gif(struct vcpu_svm *svm, bool value);
418 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
419 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
420 int read, int write);
424 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
425 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
426 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
428 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
430 struct vcpu_svm *svm = to_svm(vcpu);
432 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
435 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
437 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
440 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
442 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
445 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
447 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
450 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, struct vmcb *vmcb12);
451 void svm_leave_nested(struct vcpu_svm *svm);
452 void svm_free_nested(struct vcpu_svm *svm);
453 int svm_allocate_nested(struct vcpu_svm *svm);
454 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
455 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
456 int nested_svm_vmexit(struct vcpu_svm *svm);
458 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
460 svm->vmcb->control.exit_code = exit_code;
461 svm->vmcb->control.exit_info_1 = 0;
462 svm->vmcb->control.exit_info_2 = 0;
463 return nested_svm_vmexit(svm);
466 int nested_svm_exit_handled(struct vcpu_svm *svm);
467 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
468 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
469 bool has_error_code, u32 error_code);
470 int nested_svm_exit_special(struct vcpu_svm *svm);
471 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
472 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
473 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
475 extern struct kvm_x86_nested_ops svm_nested_ops;
479 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
480 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
481 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
483 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
484 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
485 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
486 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
488 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
492 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
494 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
495 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
498 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
500 struct vcpu_svm *svm = to_svm(vcpu);
501 u64 *entry = svm->avic_physical_id_cache;
506 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
509 int avic_ga_log_notifier(u32 ga_tag);
510 void avic_vm_destroy(struct kvm *kvm);
511 int avic_vm_init(struct kvm *kvm);
512 void avic_init_vmcb(struct vcpu_svm *svm);
513 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
514 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
515 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
516 int avic_init_vcpu(struct vcpu_svm *svm);
517 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
518 void avic_vcpu_put(struct kvm_vcpu *vcpu);
519 void avic_post_state_restore(struct kvm_vcpu *vcpu);
520 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
521 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
522 bool svm_check_apicv_inhibit_reasons(ulong bit);
523 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
524 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
525 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
526 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
527 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
528 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
529 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
530 uint32_t guest_irq, bool set);
531 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
532 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
536 #define GHCB_VERSION_MAX 1ULL
537 #define GHCB_VERSION_MIN 1ULL
539 #define GHCB_MSR_INFO_POS 0
540 #define GHCB_MSR_INFO_MASK (BIT_ULL(12) - 1)
542 #define GHCB_MSR_SEV_INFO_RESP 0x001
543 #define GHCB_MSR_SEV_INFO_REQ 0x002
544 #define GHCB_MSR_VER_MAX_POS 48
545 #define GHCB_MSR_VER_MAX_MASK 0xffff
546 #define GHCB_MSR_VER_MIN_POS 32
547 #define GHCB_MSR_VER_MIN_MASK 0xffff
548 #define GHCB_MSR_CBIT_POS 24
549 #define GHCB_MSR_CBIT_MASK 0xff
550 #define GHCB_MSR_SEV_INFO(_max, _min, _cbit) \
551 ((((_max) & GHCB_MSR_VER_MAX_MASK) << GHCB_MSR_VER_MAX_POS) | \
552 (((_min) & GHCB_MSR_VER_MIN_MASK) << GHCB_MSR_VER_MIN_POS) | \
553 (((_cbit) & GHCB_MSR_CBIT_MASK) << GHCB_MSR_CBIT_POS) | \
554 GHCB_MSR_SEV_INFO_RESP)
556 #define GHCB_MSR_CPUID_REQ 0x004
557 #define GHCB_MSR_CPUID_RESP 0x005
558 #define GHCB_MSR_CPUID_FUNC_POS 32
559 #define GHCB_MSR_CPUID_FUNC_MASK 0xffffffff
560 #define GHCB_MSR_CPUID_VALUE_POS 32
561 #define GHCB_MSR_CPUID_VALUE_MASK 0xffffffff
562 #define GHCB_MSR_CPUID_REG_POS 30
563 #define GHCB_MSR_CPUID_REG_MASK 0x3
565 #define GHCB_MSR_TERM_REQ 0x100
566 #define GHCB_MSR_TERM_REASON_SET_POS 12
567 #define GHCB_MSR_TERM_REASON_SET_MASK 0xf
568 #define GHCB_MSR_TERM_REASON_POS 16
569 #define GHCB_MSR_TERM_REASON_MASK 0xff
571 extern unsigned int max_sev_asid;
573 static inline bool svm_sev_enabled(void)
575 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
578 void sev_vm_destroy(struct kvm *kvm);
579 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
580 int svm_register_enc_region(struct kvm *kvm,
581 struct kvm_enc_region *range);
582 int svm_unregister_enc_region(struct kvm *kvm,
583 struct kvm_enc_region *range);
584 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
585 void pre_sev_run(struct vcpu_svm *svm, int cpu);
586 void __init sev_hardware_setup(void);
587 void sev_hardware_teardown(void);
588 void sev_free_vcpu(struct kvm_vcpu *vcpu);
589 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
590 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
591 void sev_es_init_vmcb(struct vcpu_svm *svm);
592 void sev_es_create_vcpu(struct vcpu_svm *svm);
593 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
594 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
598 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
599 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);