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 #include "kvm_cache_regs.h"
27 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
29 #define IOPM_SIZE PAGE_SIZE * 3
30 #define MSRPM_SIZE PAGE_SIZE * 2
32 #define MAX_DIRECT_ACCESS_MSRS 20
33 #define MSRPM_OFFSETS 16
34 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
35 extern bool npt_enabled;
36 extern bool intercept_smi;
40 * VMCB_ALL_CLEAN_MASK might also need to
41 * be updated if this enum is modified.
44 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
46 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
48 VMCB_INTR, /* int_ctl, int_vector */
49 VMCB_NPT, /* npt_en, nCR3, gPAT */
50 VMCB_CR, /* CR0, CR3, CR4, EFER */
51 VMCB_DR, /* DR6, DR7 */
52 VMCB_DT, /* GDT, IDT */
53 VMCB_SEG, /* CS, DS, SS, ES, CPL */
54 VMCB_CR2, /* CR2 only */
55 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
56 VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
57 * AVIC PHYSICAL_TABLE pointer,
58 * AVIC LOGICAL_TABLE pointer
60 VMCB_SW = 31, /* Reserved for hypervisor/software use */
63 #define VMCB_ALL_CLEAN_MASK ( \
64 (1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) | \
65 (1U << VMCB_ASID) | (1U << VMCB_INTR) | \
66 (1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) | \
67 (1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) | \
68 (1U << VMCB_LBR) | (1U << VMCB_AVIC) | \
71 /* TPR and CR2 are always written before VMRUN */
72 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
75 bool active; /* SEV enabled guest */
76 bool es_active; /* SEV-ES enabled guest */
77 unsigned int asid; /* ASID used for this guest */
78 unsigned int handle; /* SEV firmware handle */
79 int fd; /* SEV device fd */
80 unsigned long pages_locked; /* Number of pages locked */
81 struct list_head regions_list; /* List of registered regions */
82 u64 ap_jump_table; /* SEV-ES AP Jump Table address */
83 struct kvm *enc_context_owner; /* Owner of copied encryption context */
84 struct misc_cg *misc_cg; /* For misc cgroup accounting */
90 /* Struct members for AVIC */
92 struct page *avic_logical_id_table_page;
93 struct page *avic_physical_id_table_page;
94 struct hlist_node hnode;
96 struct kvm_sev_info sev_info;
101 struct kvm_vmcb_info {
105 uint64_t asid_generation;
108 struct svm_nested_state {
109 struct kvm_vmcb_info vmcb02;
115 /* These are the merged vectors */
118 /* A VMRUN has started but has not yet been performed, so
119 * we cannot inject a nested vmexit yet. */
120 bool nested_run_pending;
122 /* cache for control fields of the guest */
123 struct vmcb_control_area ctl;
129 struct kvm_vcpu vcpu;
130 /* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
132 struct kvm_vmcb_info vmcb01;
133 struct kvm_vmcb_info *current_vmcb;
134 struct svm_cpu_data *svm_data;
146 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
147 * translated into the appropriate L2_CFG bits on the host to
148 * perform speculative control.
156 struct svm_nested_state nested;
159 u64 nmi_singlestep_guest_rflags;
161 unsigned int3_injected;
162 unsigned long int3_rip;
164 /* cached guest cpuid flags for faster access */
165 bool nrips_enabled : 1;
169 struct page *avic_backing_page;
170 u64 *avic_physical_id_cache;
171 bool avic_is_running;
174 * Per-vcpu list of struct amd_svm_iommu_ir:
175 * This is used mainly to store interrupt remapping information used
176 * when update the vcpu affinity. This avoids the need to scan for
177 * IRTE and try to match ga_tag in the IOMMU driver.
179 struct list_head ir_list;
180 spinlock_t ir_list_lock;
182 /* Save desired MSR intercept (read: pass-through) state */
184 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
185 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
186 } shadow_msr_intercept;
189 struct vmcb_save_area *vmsa;
191 struct kvm_host_map ghcb_map;
192 bool received_first_sipi;
194 /* SEV-ES scratch area support */
200 bool guest_state_loaded;
203 struct svm_cpu_data {
210 struct kvm_ldttss_desc *tss_desc;
212 struct page *save_area;
213 struct vmcb *current_vmcb;
215 /* index = sev_asid, value = vmcb pointer */
216 struct vmcb **sev_vmcbs;
219 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
221 void recalc_intercepts(struct vcpu_svm *svm);
223 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
225 return container_of(kvm, struct kvm_svm, kvm);
228 static inline bool sev_guest(struct kvm *kvm)
230 #ifdef CONFIG_KVM_AMD_SEV
231 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
239 static inline bool sev_es_guest(struct kvm *kvm)
241 #ifdef CONFIG_KVM_AMD_SEV
242 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
244 return sev_guest(kvm) && sev->es_active;
250 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
252 vmcb->control.clean = 0;
255 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
257 vmcb->control.clean = VMCB_ALL_CLEAN_MASK
258 & ~VMCB_ALWAYS_DIRTY_MASK;
261 static inline bool vmcb_is_clean(struct vmcb *vmcb, int bit)
263 return (vmcb->control.clean & (1 << bit));
266 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
268 vmcb->control.clean &= ~(1 << bit);
271 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
273 return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
276 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
278 return container_of(vcpu, struct vcpu_svm, vcpu);
281 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
283 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
284 __set_bit(bit, (unsigned long *)&control->intercepts);
287 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
289 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
290 __clear_bit(bit, (unsigned long *)&control->intercepts);
293 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
295 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
296 return test_bit(bit, (unsigned long *)&control->intercepts);
299 static inline void set_dr_intercepts(struct vcpu_svm *svm)
301 struct vmcb *vmcb = svm->vmcb01.ptr;
303 if (!sev_es_guest(svm->vcpu.kvm)) {
304 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
305 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
306 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
307 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
308 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
309 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
310 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
311 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
312 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
313 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
314 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
315 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
316 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
317 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
320 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
321 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
323 recalc_intercepts(svm);
326 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
328 struct vmcb *vmcb = svm->vmcb01.ptr;
330 vmcb->control.intercepts[INTERCEPT_DR] = 0;
332 /* DR7 access must remain intercepted for an SEV-ES guest */
333 if (sev_es_guest(svm->vcpu.kvm)) {
334 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
335 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
338 recalc_intercepts(svm);
341 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
343 struct vmcb *vmcb = svm->vmcb01.ptr;
345 WARN_ON_ONCE(bit >= 32);
346 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
348 recalc_intercepts(svm);
351 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
353 struct vmcb *vmcb = svm->vmcb01.ptr;
355 WARN_ON_ONCE(bit >= 32);
356 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
358 recalc_intercepts(svm);
361 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
363 struct vmcb *vmcb = svm->vmcb01.ptr;
365 vmcb_set_intercept(&vmcb->control, bit);
367 recalc_intercepts(svm);
370 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
372 struct vmcb *vmcb = svm->vmcb01.ptr;
374 vmcb_clr_intercept(&vmcb->control, bit);
376 recalc_intercepts(svm);
379 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
381 return vmcb_is_intercept(&svm->vmcb->control, bit);
384 static inline bool vgif_enabled(struct vcpu_svm *svm)
386 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
389 static inline void enable_gif(struct vcpu_svm *svm)
391 if (vgif_enabled(svm))
392 svm->vmcb->control.int_ctl |= V_GIF_MASK;
394 svm->vcpu.arch.hflags |= HF_GIF_MASK;
397 static inline void disable_gif(struct vcpu_svm *svm)
399 if (vgif_enabled(svm))
400 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
402 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
405 static inline bool gif_set(struct vcpu_svm *svm)
407 if (vgif_enabled(svm))
408 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
410 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
414 #define MSR_INVALID 0xffffffffU
416 extern bool dump_invalid_vmcb;
418 u32 svm_msrpm_offset(u32 msr);
419 u32 *svm_vcpu_alloc_msrpm(void);
420 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
421 void svm_vcpu_free_msrpm(u32 *msrpm);
423 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
424 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
425 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
426 void svm_flush_tlb(struct kvm_vcpu *vcpu);
427 void disable_nmi_singlestep(struct vcpu_svm *svm);
428 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
429 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
430 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
431 void svm_set_gif(struct vcpu_svm *svm, bool value);
432 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
433 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
434 int read, int write);
438 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
439 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
440 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
442 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
444 struct vcpu_svm *svm = to_svm(vcpu);
446 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
449 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
451 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
454 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
456 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
459 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
461 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
464 int enter_svm_guest_mode(struct kvm_vcpu *vcpu,
465 u64 vmcb_gpa, struct vmcb *vmcb12, bool from_vmrun);
466 void svm_leave_nested(struct kvm_vcpu *vcpu);
467 void svm_free_nested(struct vcpu_svm *svm);
468 int svm_allocate_nested(struct vcpu_svm *svm);
469 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
470 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
471 struct vmcb_save_area *from_save);
472 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
473 int nested_svm_vmexit(struct vcpu_svm *svm);
475 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
477 svm->vmcb->control.exit_code = exit_code;
478 svm->vmcb->control.exit_info_1 = 0;
479 svm->vmcb->control.exit_info_2 = 0;
480 return nested_svm_vmexit(svm);
483 int nested_svm_exit_handled(struct vcpu_svm *svm);
484 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
485 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
486 bool has_error_code, u32 error_code);
487 int nested_svm_exit_special(struct vcpu_svm *svm);
488 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
489 struct vmcb_control_area *control);
490 void __svm_write_tsc_multiplier(u64 multiplier);
491 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
492 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
493 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
495 extern struct kvm_x86_nested_ops svm_nested_ops;
499 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
500 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
501 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
503 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK GENMASK_ULL(11, 0)
504 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
505 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
506 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
508 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
510 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
512 struct vcpu_svm *svm = to_svm(vcpu);
513 u64 *entry = svm->avic_physical_id_cache;
518 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
521 int avic_ga_log_notifier(u32 ga_tag);
522 void avic_vm_destroy(struct kvm *kvm);
523 int avic_vm_init(struct kvm *kvm);
524 void avic_init_vmcb(struct vcpu_svm *svm);
525 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
526 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
527 int avic_init_vcpu(struct vcpu_svm *svm);
528 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
529 void avic_vcpu_put(struct kvm_vcpu *vcpu);
530 void avic_post_state_restore(struct kvm_vcpu *vcpu);
531 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
532 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
533 bool svm_check_apicv_inhibit_reasons(ulong bit);
534 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
535 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
536 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
537 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
538 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
539 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
540 uint32_t guest_irq, bool set);
541 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
542 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
546 #define GHCB_VERSION_MAX 1ULL
547 #define GHCB_VERSION_MIN 1ULL
550 extern unsigned int max_sev_asid;
552 void sev_vm_destroy(struct kvm *kvm);
553 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
554 int svm_register_enc_region(struct kvm *kvm,
555 struct kvm_enc_region *range);
556 int svm_unregister_enc_region(struct kvm *kvm,
557 struct kvm_enc_region *range);
558 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd);
559 void sev_guest_memory_reclaimed(struct kvm *kvm);
561 void pre_sev_run(struct vcpu_svm *svm, int cpu);
562 void __init sev_set_cpu_caps(void);
563 void __init sev_hardware_setup(void);
564 void sev_hardware_teardown(void);
565 int sev_cpu_init(struct svm_cpu_data *sd);
566 void sev_free_vcpu(struct kvm_vcpu *vcpu);
567 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
568 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
569 void sev_es_init_vmcb(struct vcpu_svm *svm);
570 void sev_es_create_vcpu(struct vcpu_svm *svm);
571 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
572 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu);
573 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
577 void __svm_sev_es_vcpu_run(unsigned long vmcb_pa);
578 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);