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 */
76 /* Struct members for AVIC */
78 struct page *avic_logical_id_table_page;
79 struct page *avic_physical_id_table_page;
80 struct hlist_node hnode;
82 struct kvm_sev_info sev_info;
87 struct kvm_vmcb_info {
91 uint64_t asid_generation;
94 struct svm_nested_state {
95 struct kvm_vmcb_info vmcb02;
101 /* These are the merged vectors */
104 /* A VMRUN has started but has not yet been performed, so
105 * we cannot inject a nested vmexit yet. */
106 bool nested_run_pending;
108 /* cache for control fields of the guest */
109 struct vmcb_control_area ctl;
115 struct kvm_vcpu vcpu;
116 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
118 struct kvm_vmcb_info vmcb01;
119 struct kvm_vmcb_info *current_vmcb;
120 struct svm_cpu_data *svm_data;
130 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
134 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
135 * translated into the appropriate L2_CFG bits on the host to
136 * perform speculative control.
144 struct svm_nested_state nested;
147 u64 nmi_singlestep_guest_rflags;
149 unsigned int3_injected;
150 unsigned long int3_rip;
152 /* cached guest cpuid flags for faster access */
153 bool nrips_enabled : 1;
157 struct page *avic_backing_page;
158 u64 *avic_physical_id_cache;
159 bool avic_is_running;
162 * Per-vcpu list of struct amd_svm_iommu_ir:
163 * This is used mainly to store interrupt remapping information used
164 * when update the vcpu affinity. This avoids the need to scan for
165 * IRTE and try to match ga_tag in the IOMMU driver.
167 struct list_head ir_list;
168 spinlock_t ir_list_lock;
170 /* Save desired MSR intercept (read: pass-through) state */
172 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
173 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
174 } shadow_msr_intercept;
177 struct vmcb_save_area *vmsa;
179 struct kvm_host_map ghcb_map;
180 bool received_first_sipi;
182 /* SEV-ES scratch area support */
188 bool guest_state_loaded;
191 struct svm_cpu_data {
198 struct kvm_ldttss_desc *tss_desc;
200 struct page *save_area;
201 struct vmcb *current_vmcb;
203 /* index = sev_asid, value = vmcb pointer */
204 struct vmcb **sev_vmcbs;
207 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
209 void recalc_intercepts(struct vcpu_svm *svm);
211 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
213 return container_of(kvm, struct kvm_svm, kvm);
216 static inline bool sev_guest(struct kvm *kvm)
218 #ifdef CONFIG_KVM_AMD_SEV
219 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
227 static inline bool sev_es_guest(struct kvm *kvm)
229 #ifdef CONFIG_KVM_AMD_SEV
230 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
232 return sev_guest(kvm) && sev->es_active;
238 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
240 vmcb->control.clean = 0;
243 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
245 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
246 & ~VMCB_ALWAYS_DIRTY_MASK;
249 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
251 vmcb->control.clean &= ~(1 << bit);
254 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
256 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
259 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
261 return container_of(vcpu, struct vcpu_svm, vcpu);
264 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
266 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
267 __set_bit(bit, (unsigned long *)&control->intercepts);
270 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
272 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
273 __clear_bit(bit, (unsigned long *)&control->intercepts);
276 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
278 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
279 return test_bit(bit, (unsigned long *)&control->intercepts);
282 static inline void set_dr_intercepts(struct vcpu_svm *svm)
284 struct vmcb *vmcb = svm->vmcb01.ptr;
286 if (!sev_es_guest(svm->vcpu.kvm)) {
287 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
288 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
289 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
290 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
291 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
292 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
293 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
294 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
295 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
296 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
297 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
298 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
299 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
300 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
303 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
304 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
306 recalc_intercepts(svm);
309 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
311 struct vmcb *vmcb = svm->vmcb01.ptr;
313 vmcb->control.intercepts[INTERCEPT_DR] = 0;
315 /* DR7 access must remain intercepted for an SEV-ES guest */
316 if (sev_es_guest(svm->vcpu.kvm)) {
317 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
318 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
321 recalc_intercepts(svm);
324 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
326 struct vmcb *vmcb = svm->vmcb01.ptr;
328 WARN_ON_ONCE(bit >= 32);
329 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
331 recalc_intercepts(svm);
334 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
336 struct vmcb *vmcb = svm->vmcb01.ptr;
338 WARN_ON_ONCE(bit >= 32);
339 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
341 recalc_intercepts(svm);
344 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
346 struct vmcb *vmcb = svm->vmcb01.ptr;
348 vmcb_set_intercept(&vmcb->control, bit);
350 recalc_intercepts(svm);
353 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
355 struct vmcb *vmcb = svm->vmcb01.ptr;
357 vmcb_clr_intercept(&vmcb->control, bit);
359 recalc_intercepts(svm);
362 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
364 return vmcb_is_intercept(&svm->vmcb->control, bit);
367 static inline bool vgif_enabled(struct vcpu_svm *svm)
369 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
372 static inline void enable_gif(struct vcpu_svm *svm)
374 if (vgif_enabled(svm))
375 svm->vmcb->control.int_ctl |= V_GIF_MASK;
377 svm->vcpu.arch.hflags |= HF_GIF_MASK;
380 static inline void disable_gif(struct vcpu_svm *svm)
382 if (vgif_enabled(svm))
383 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
385 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
388 static inline bool gif_set(struct vcpu_svm *svm)
390 if (vgif_enabled(svm))
391 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
393 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
397 #define MSR_INVALID 0xffffffffU
401 extern bool dump_invalid_vmcb;
403 u32 svm_msrpm_offset(u32 msr);
404 u32 *svm_vcpu_alloc_msrpm(void);
405 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
406 void svm_vcpu_free_msrpm(u32 *msrpm);
408 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
409 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
410 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
411 void svm_flush_tlb(struct kvm_vcpu *vcpu);
412 void disable_nmi_singlestep(struct vcpu_svm *svm);
413 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
414 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
415 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
416 void svm_set_gif(struct vcpu_svm *svm, bool value);
417 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
418 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
419 int read, int write);
423 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
424 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
425 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
427 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
429 struct vcpu_svm *svm = to_svm(vcpu);
431 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
434 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
436 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
439 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
441 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
444 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
446 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
449 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, struct vmcb *vmcb12);
450 void svm_leave_nested(struct vcpu_svm *svm);
451 void svm_free_nested(struct vcpu_svm *svm);
452 int svm_allocate_nested(struct vcpu_svm *svm);
453 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
454 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
455 int nested_svm_vmexit(struct vcpu_svm *svm);
457 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
459 svm->vmcb->control.exit_code = exit_code;
460 svm->vmcb->control.exit_info_1 = 0;
461 svm->vmcb->control.exit_info_2 = 0;
462 return nested_svm_vmexit(svm);
465 int nested_svm_exit_handled(struct vcpu_svm *svm);
466 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
467 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
468 bool has_error_code, u32 error_code);
469 int nested_svm_exit_special(struct vcpu_svm *svm);
470 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
471 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
472 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
474 extern struct kvm_x86_nested_ops svm_nested_ops;
478 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
479 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
480 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
482 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
483 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
484 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
485 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
487 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
491 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
493 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
494 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
497 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
499 struct vcpu_svm *svm = to_svm(vcpu);
500 u64 *entry = svm->avic_physical_id_cache;
505 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
508 int avic_ga_log_notifier(u32 ga_tag);
509 void avic_vm_destroy(struct kvm *kvm);
510 int avic_vm_init(struct kvm *kvm);
511 void avic_init_vmcb(struct vcpu_svm *svm);
512 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
513 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
514 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
515 int avic_init_vcpu(struct vcpu_svm *svm);
516 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
517 void avic_vcpu_put(struct kvm_vcpu *vcpu);
518 void avic_post_state_restore(struct kvm_vcpu *vcpu);
519 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
520 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
521 bool svm_check_apicv_inhibit_reasons(ulong bit);
522 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
523 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
524 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
525 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
526 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
527 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
528 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
529 uint32_t guest_irq, bool set);
530 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
531 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
535 #define GHCB_VERSION_MAX 1ULL
536 #define GHCB_VERSION_MIN 1ULL
538 #define GHCB_MSR_INFO_POS 0
539 #define GHCB_MSR_INFO_MASK (BIT_ULL(12) - 1)
541 #define GHCB_MSR_SEV_INFO_RESP 0x001
542 #define GHCB_MSR_SEV_INFO_REQ 0x002
543 #define GHCB_MSR_VER_MAX_POS 48
544 #define GHCB_MSR_VER_MAX_MASK 0xffff
545 #define GHCB_MSR_VER_MIN_POS 32
546 #define GHCB_MSR_VER_MIN_MASK 0xffff
547 #define GHCB_MSR_CBIT_POS 24
548 #define GHCB_MSR_CBIT_MASK 0xff
549 #define GHCB_MSR_SEV_INFO(_max, _min, _cbit) \
550 ((((_max) & GHCB_MSR_VER_MAX_MASK) << GHCB_MSR_VER_MAX_POS) | \
551 (((_min) & GHCB_MSR_VER_MIN_MASK) << GHCB_MSR_VER_MIN_POS) | \
552 (((_cbit) & GHCB_MSR_CBIT_MASK) << GHCB_MSR_CBIT_POS) | \
553 GHCB_MSR_SEV_INFO_RESP)
555 #define GHCB_MSR_CPUID_REQ 0x004
556 #define GHCB_MSR_CPUID_RESP 0x005
557 #define GHCB_MSR_CPUID_FUNC_POS 32
558 #define GHCB_MSR_CPUID_FUNC_MASK 0xffffffff
559 #define GHCB_MSR_CPUID_VALUE_POS 32
560 #define GHCB_MSR_CPUID_VALUE_MASK 0xffffffff
561 #define GHCB_MSR_CPUID_REG_POS 30
562 #define GHCB_MSR_CPUID_REG_MASK 0x3
564 #define GHCB_MSR_TERM_REQ 0x100
565 #define GHCB_MSR_TERM_REASON_SET_POS 12
566 #define GHCB_MSR_TERM_REASON_SET_MASK 0xf
567 #define GHCB_MSR_TERM_REASON_POS 16
568 #define GHCB_MSR_TERM_REASON_MASK 0xff
570 extern unsigned int max_sev_asid;
572 static inline bool svm_sev_enabled(void)
574 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
577 void sev_vm_destroy(struct kvm *kvm);
578 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
579 int svm_register_enc_region(struct kvm *kvm,
580 struct kvm_enc_region *range);
581 int svm_unregister_enc_region(struct kvm *kvm,
582 struct kvm_enc_region *range);
583 void pre_sev_run(struct vcpu_svm *svm, int cpu);
584 void __init sev_hardware_setup(void);
585 void sev_hardware_teardown(void);
586 void sev_free_vcpu(struct kvm_vcpu *vcpu);
587 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
588 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
589 void sev_es_init_vmcb(struct vcpu_svm *svm);
590 void sev_es_create_vcpu(struct vcpu_svm *svm);
591 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
592 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
596 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
597 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);