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 #define IOPM_SIZE PAGE_SIZE * 3
27 #define MSRPM_SIZE PAGE_SIZE * 2
29 #define MAX_DIRECT_ACCESS_MSRS 20
30 #define MSRPM_OFFSETS 16
31 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
32 extern bool npt_enabled;
35 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
37 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
39 VMCB_INTR, /* int_ctl, int_vector */
40 VMCB_NPT, /* npt_en, nCR3, gPAT */
41 VMCB_CR, /* CR0, CR3, CR4, EFER */
42 VMCB_DR, /* DR6, DR7 */
43 VMCB_DT, /* GDT, IDT */
44 VMCB_SEG, /* CS, DS, SS, ES, CPL */
45 VMCB_CR2, /* CR2 only */
46 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
47 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
48 * AVIC PHYSICAL_TABLE pointer,
49 * AVIC LOGICAL_TABLE pointer
54 /* TPR and CR2 are always written before VMRUN */
55 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
58 bool active; /* SEV enabled guest */
59 bool es_active; /* SEV-ES enabled guest */
60 unsigned int asid; /* ASID used for this guest */
61 unsigned int handle; /* SEV firmware handle */
62 int fd; /* SEV device fd */
63 unsigned long pages_locked; /* Number of pages locked */
64 struct list_head regions_list; /* List of registered regions */
65 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
66 struct kvm *enc_context_owner; /* Owner of copied encryption context */
67 struct misc_cg *misc_cg; /* For misc cgroup accounting */
73 /* Struct members for AVIC */
75 struct page *avic_logical_id_table_page;
76 struct page *avic_physical_id_table_page;
77 struct hlist_node hnode;
79 struct kvm_sev_info sev_info;
84 struct kvm_vmcb_info {
88 uint64_t asid_generation;
91 struct svm_nested_state {
92 struct kvm_vmcb_info vmcb02;
98 /* These are the merged vectors */
101 /* A VMRUN has started but has not yet been performed, so
102 * we cannot inject a nested vmexit yet. */
103 bool nested_run_pending;
105 /* cache for control fields of the guest */
106 struct vmcb_control_area ctl;
112 struct kvm_vcpu vcpu;
113 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
115 struct kvm_vmcb_info vmcb01;
116 struct kvm_vmcb_info *current_vmcb;
117 struct svm_cpu_data *svm_data;
129 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
130 * translated into the appropriate L2_CFG bits on the host to
131 * perform speculative control.
139 struct svm_nested_state nested;
142 u64 nmi_singlestep_guest_rflags;
144 unsigned int3_injected;
145 unsigned long int3_rip;
147 /* cached guest cpuid flags for faster access */
148 bool nrips_enabled : 1;
152 struct page *avic_backing_page;
153 u64 *avic_physical_id_cache;
154 bool avic_is_running;
157 * Per-vcpu list of struct amd_svm_iommu_ir:
158 * This is used mainly to store interrupt remapping information used
159 * when update the vcpu affinity. This avoids the need to scan for
160 * IRTE and try to match ga_tag in the IOMMU driver.
162 struct list_head ir_list;
163 spinlock_t ir_list_lock;
165 /* Save desired MSR intercept (read: pass-through) state */
167 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
168 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
169 } shadow_msr_intercept;
172 struct vmcb_save_area *vmsa;
174 struct kvm_host_map ghcb_map;
175 bool received_first_sipi;
177 /* SEV-ES scratch area support */
183 bool guest_state_loaded;
186 struct svm_cpu_data {
193 struct kvm_ldttss_desc *tss_desc;
195 struct page *save_area;
196 struct vmcb *current_vmcb;
198 /* index = sev_asid, value = vmcb pointer */
199 struct vmcb **sev_vmcbs;
202 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
204 void recalc_intercepts(struct vcpu_svm *svm);
206 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
208 return container_of(kvm, struct kvm_svm, kvm);
211 static inline bool sev_guest(struct kvm *kvm)
213 #ifdef CONFIG_KVM_AMD_SEV
214 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
222 static inline bool sev_es_guest(struct kvm *kvm)
224 #ifdef CONFIG_KVM_AMD_SEV
225 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
227 return sev_guest(kvm) && sev->es_active;
233 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
235 vmcb->control.clean = 0;
238 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
240 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
241 & ~VMCB_ALWAYS_DIRTY_MASK;
244 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
246 vmcb->control.clean &= ~(1 << bit);
249 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
251 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
254 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
256 return container_of(vcpu, struct vcpu_svm, vcpu);
259 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
261 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
262 __set_bit(bit, (unsigned long *)&control->intercepts);
265 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
267 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
268 __clear_bit(bit, (unsigned long *)&control->intercepts);
271 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
273 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
274 return test_bit(bit, (unsigned long *)&control->intercepts);
277 static inline void set_dr_intercepts(struct vcpu_svm *svm)
279 struct vmcb *vmcb = svm->vmcb01.ptr;
281 if (!sev_es_guest(svm->vcpu.kvm)) {
282 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
283 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
284 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
285 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
286 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
287 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
288 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
289 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
290 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
291 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
292 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
293 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
294 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
295 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
298 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
299 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
301 recalc_intercepts(svm);
304 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
306 struct vmcb *vmcb = svm->vmcb01.ptr;
308 vmcb->control.intercepts[INTERCEPT_DR] = 0;
310 /* DR7 access must remain intercepted for an SEV-ES guest */
311 if (sev_es_guest(svm->vcpu.kvm)) {
312 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
313 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
316 recalc_intercepts(svm);
319 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
321 struct vmcb *vmcb = svm->vmcb01.ptr;
323 WARN_ON_ONCE(bit >= 32);
324 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
326 recalc_intercepts(svm);
329 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
331 struct vmcb *vmcb = svm->vmcb01.ptr;
333 WARN_ON_ONCE(bit >= 32);
334 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
336 recalc_intercepts(svm);
339 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
341 struct vmcb *vmcb = svm->vmcb01.ptr;
343 vmcb_set_intercept(&vmcb->control, bit);
345 recalc_intercepts(svm);
348 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
350 struct vmcb *vmcb = svm->vmcb01.ptr;
352 vmcb_clr_intercept(&vmcb->control, bit);
354 recalc_intercepts(svm);
357 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
359 return vmcb_is_intercept(&svm->vmcb->control, bit);
362 static inline bool vgif_enabled(struct vcpu_svm *svm)
364 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
367 static inline void enable_gif(struct vcpu_svm *svm)
369 if (vgif_enabled(svm))
370 svm->vmcb->control.int_ctl |= V_GIF_MASK;
372 svm->vcpu.arch.hflags |= HF_GIF_MASK;
375 static inline void disable_gif(struct vcpu_svm *svm)
377 if (vgif_enabled(svm))
378 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
380 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
383 static inline bool gif_set(struct vcpu_svm *svm)
385 if (vgif_enabled(svm))
386 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
388 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
392 #define MSR_INVALID 0xffffffffU
394 extern bool dump_invalid_vmcb;
396 u32 svm_msrpm_offset(u32 msr);
397 u32 *svm_vcpu_alloc_msrpm(void);
398 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
399 void svm_vcpu_free_msrpm(u32 *msrpm);
401 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
402 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
403 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
404 void svm_flush_tlb(struct kvm_vcpu *vcpu);
405 void disable_nmi_singlestep(struct vcpu_svm *svm);
406 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
407 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
408 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
409 void svm_set_gif(struct vcpu_svm *svm, bool value);
410 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
411 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
412 int read, int write);
416 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
417 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
418 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
420 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
422 struct vcpu_svm *svm = to_svm(vcpu);
424 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
427 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
429 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
432 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
434 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
437 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
439 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
442 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, struct vmcb *vmcb12);
443 void svm_leave_nested(struct vcpu_svm *svm);
444 void svm_free_nested(struct vcpu_svm *svm);
445 int svm_allocate_nested(struct vcpu_svm *svm);
446 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
447 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
448 int nested_svm_vmexit(struct vcpu_svm *svm);
450 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
452 svm->vmcb->control.exit_code = exit_code;
453 svm->vmcb->control.exit_info_1 = 0;
454 svm->vmcb->control.exit_info_2 = 0;
455 return nested_svm_vmexit(svm);
458 int nested_svm_exit_handled(struct vcpu_svm *svm);
459 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
460 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
461 bool has_error_code, u32 error_code);
462 int nested_svm_exit_special(struct vcpu_svm *svm);
463 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
464 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
465 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
467 extern struct kvm_x86_nested_ops svm_nested_ops;
471 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
472 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
473 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
475 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
476 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
477 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
478 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
480 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
484 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
486 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
487 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
490 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
492 struct vcpu_svm *svm = to_svm(vcpu);
493 u64 *entry = svm->avic_physical_id_cache;
498 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
501 int avic_ga_log_notifier(u32 ga_tag);
502 void avic_vm_destroy(struct kvm *kvm);
503 int avic_vm_init(struct kvm *kvm);
504 void avic_init_vmcb(struct vcpu_svm *svm);
505 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
506 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
507 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
508 int avic_init_vcpu(struct vcpu_svm *svm);
509 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
510 void avic_vcpu_put(struct kvm_vcpu *vcpu);
511 void avic_post_state_restore(struct kvm_vcpu *vcpu);
512 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
513 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
514 bool svm_check_apicv_inhibit_reasons(ulong bit);
515 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
516 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
517 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
518 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
519 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
520 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
521 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
522 uint32_t guest_irq, bool set);
523 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
524 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
528 #define GHCB_VERSION_MAX 1ULL
529 #define GHCB_VERSION_MIN 1ULL
531 #define GHCB_MSR_INFO_POS 0
532 #define GHCB_MSR_INFO_MASK (BIT_ULL(12) - 1)
534 #define GHCB_MSR_SEV_INFO_RESP 0x001
535 #define GHCB_MSR_SEV_INFO_REQ 0x002
536 #define GHCB_MSR_VER_MAX_POS 48
537 #define GHCB_MSR_VER_MAX_MASK 0xffff
538 #define GHCB_MSR_VER_MIN_POS 32
539 #define GHCB_MSR_VER_MIN_MASK 0xffff
540 #define GHCB_MSR_CBIT_POS 24
541 #define GHCB_MSR_CBIT_MASK 0xff
542 #define GHCB_MSR_SEV_INFO(_max, _min, _cbit) \
543 ((((_max) & GHCB_MSR_VER_MAX_MASK) << GHCB_MSR_VER_MAX_POS) | \
544 (((_min) & GHCB_MSR_VER_MIN_MASK) << GHCB_MSR_VER_MIN_POS) | \
545 (((_cbit) & GHCB_MSR_CBIT_MASK) << GHCB_MSR_CBIT_POS) | \
546 GHCB_MSR_SEV_INFO_RESP)
548 #define GHCB_MSR_CPUID_REQ 0x004
549 #define GHCB_MSR_CPUID_RESP 0x005
550 #define GHCB_MSR_CPUID_FUNC_POS 32
551 #define GHCB_MSR_CPUID_FUNC_MASK 0xffffffff
552 #define GHCB_MSR_CPUID_VALUE_POS 32
553 #define GHCB_MSR_CPUID_VALUE_MASK 0xffffffff
554 #define GHCB_MSR_CPUID_REG_POS 30
555 #define GHCB_MSR_CPUID_REG_MASK 0x3
557 #define GHCB_MSR_TERM_REQ 0x100
558 #define GHCB_MSR_TERM_REASON_SET_POS 12
559 #define GHCB_MSR_TERM_REASON_SET_MASK 0xf
560 #define GHCB_MSR_TERM_REASON_POS 16
561 #define GHCB_MSR_TERM_REASON_MASK 0xff
563 extern unsigned int max_sev_asid;
565 void sev_vm_destroy(struct kvm *kvm);
566 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
567 int svm_register_enc_region(struct kvm *kvm,
568 struct kvm_enc_region *range);
569 int svm_unregister_enc_region(struct kvm *kvm,
570 struct kvm_enc_region *range);
571 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
572 void pre_sev_run(struct vcpu_svm *svm, int cpu);
573 void __init sev_set_cpu_caps(void);
574 void __init sev_hardware_setup(void);
575 void sev_hardware_teardown(void);
576 int sev_cpu_init(struct svm_cpu_data *sd);
577 void sev_free_vcpu(struct kvm_vcpu *vcpu);
578 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
579 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
580 void sev_es_init_vmcb(struct vcpu_svm *svm);
581 void sev_es_create_vcpu(struct vcpu_svm *svm);
582 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
583 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
587 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
588 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);