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>
23 static const u32 host_save_user_msrs[] = {
25 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
28 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
32 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
34 #define MAX_DIRECT_ACCESS_MSRS 15
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 unsigned int asid; /* ASID used for this guest */
65 unsigned int handle; /* SEV firmware handle */
66 int fd; /* SEV device fd */
67 unsigned long pages_locked; /* Number of pages locked */
68 struct list_head regions_list; /* List of registered regions */
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 svm_nested_state {
91 /* These are the merged vectors */
94 /* A VMRUN has started but has not yet been performed, so
95 * we cannot inject a nested vmexit yet. */
96 bool nested_run_pending;
98 /* cache for control fields of the guest */
99 struct vmcb_control_area ctl;
105 struct kvm_vcpu vcpu;
107 unsigned long vmcb_pa;
108 struct svm_cpu_data *svm_data;
109 uint64_t asid_generation;
110 uint64_t sysenter_esp;
111 uint64_t sysenter_eip;
118 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
128 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
129 * translated into the appropriate L2_CFG bits on the host to
130 * perform speculative control.
138 struct svm_nested_state nested;
141 u64 nmi_singlestep_guest_rflags;
143 unsigned int3_injected;
144 unsigned long int3_rip;
146 /* cached guest cpuid flags for faster access */
147 bool nrips_enabled : 1;
151 struct page *avic_backing_page;
152 u64 *avic_physical_id_cache;
153 bool avic_is_running;
156 * Per-vcpu list of struct amd_svm_iommu_ir:
157 * This is used mainly to store interrupt remapping information used
158 * when update the vcpu affinity. This avoids the need to scan for
159 * IRTE and try to match ga_tag in the IOMMU driver.
161 struct list_head ir_list;
162 spinlock_t ir_list_lock;
164 /* Save desired MSR intercept (read: pass-through) state */
166 DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
167 DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
168 } shadow_msr_intercept;
171 struct svm_cpu_data {
178 struct kvm_ldttss_desc *tss_desc;
180 struct page *save_area;
181 struct vmcb *current_vmcb;
183 /* index = sev_asid, value = vmcb pointer */
184 struct vmcb **sev_vmcbs;
187 DECLARE_PER_CPU(struct svm_cpu_data *, svm_data);
189 void recalc_intercepts(struct vcpu_svm *svm);
191 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
193 return container_of(kvm, struct kvm_svm, kvm);
196 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
198 vmcb->control.clean = 0;
201 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
203 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
204 & ~VMCB_ALWAYS_DIRTY_MASK;
207 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
209 vmcb->control.clean &= ~(1 << bit);
212 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
214 return container_of(vcpu, struct vcpu_svm, vcpu);
217 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
219 if (is_guest_mode(&svm->vcpu))
220 return svm->nested.hsave;
225 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
227 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
228 __set_bit(bit, (unsigned long *)&control->intercepts);
231 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
233 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
234 __clear_bit(bit, (unsigned long *)&control->intercepts);
237 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
239 WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
240 return test_bit(bit, (unsigned long *)&control->intercepts);
243 static inline void set_dr_intercepts(struct vcpu_svm *svm)
245 struct vmcb *vmcb = get_host_vmcb(svm);
247 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_READ);
248 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_READ);
249 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_READ);
250 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_READ);
251 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_READ);
252 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_READ);
253 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_READ);
254 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
255 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR0_WRITE);
256 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR1_WRITE);
257 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR2_WRITE);
258 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR3_WRITE);
259 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR4_WRITE);
260 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR5_WRITE);
261 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
262 vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
264 recalc_intercepts(svm);
267 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
269 struct vmcb *vmcb = get_host_vmcb(svm);
271 vmcb->control.intercepts[INTERCEPT_DR] = 0;
273 recalc_intercepts(svm);
276 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
278 struct vmcb *vmcb = get_host_vmcb(svm);
280 WARN_ON_ONCE(bit >= 32);
281 vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
283 recalc_intercepts(svm);
286 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
288 struct vmcb *vmcb = get_host_vmcb(svm);
290 WARN_ON_ONCE(bit >= 32);
291 vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
293 recalc_intercepts(svm);
296 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
298 struct vmcb *vmcb = get_host_vmcb(svm);
300 vmcb_set_intercept(&vmcb->control, bit);
302 recalc_intercepts(svm);
305 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
307 struct vmcb *vmcb = get_host_vmcb(svm);
309 vmcb_clr_intercept(&vmcb->control, bit);
311 recalc_intercepts(svm);
314 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
316 return vmcb_is_intercept(&svm->vmcb->control, bit);
319 static inline bool vgif_enabled(struct vcpu_svm *svm)
321 return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
324 static inline void enable_gif(struct vcpu_svm *svm)
326 if (vgif_enabled(svm))
327 svm->vmcb->control.int_ctl |= V_GIF_MASK;
329 svm->vcpu.arch.hflags |= HF_GIF_MASK;
332 static inline void disable_gif(struct vcpu_svm *svm)
334 if (vgif_enabled(svm))
335 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
337 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
340 static inline bool gif_set(struct vcpu_svm *svm)
342 if (vgif_enabled(svm))
343 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
345 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
349 #define MSR_CR3_LEGACY_RESERVED_MASK 0xfe7U
350 #define MSR_CR3_LEGACY_PAE_RESERVED_MASK 0x7U
351 #define MSR_CR3_LONG_MBZ_MASK 0xfff0000000000000U
352 #define MSR_INVALID 0xffffffffU
354 u32 svm_msrpm_offset(u32 msr);
355 u32 *svm_vcpu_alloc_msrpm(void);
356 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
357 void svm_vcpu_free_msrpm(u32 *msrpm);
359 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
360 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
361 int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
362 void svm_flush_tlb(struct kvm_vcpu *vcpu);
363 void disable_nmi_singlestep(struct vcpu_svm *svm);
364 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
365 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
366 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
367 void svm_set_gif(struct vcpu_svm *svm, bool value);
371 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
372 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
373 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
375 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
377 struct vcpu_svm *svm = to_svm(vcpu);
379 return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
382 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
384 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
387 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
389 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
392 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
394 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
397 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
398 struct vmcb *nested_vmcb);
399 void svm_leave_nested(struct vcpu_svm *svm);
400 void svm_free_nested(struct vcpu_svm *svm);
401 int svm_allocate_nested(struct vcpu_svm *svm);
402 int nested_svm_vmrun(struct vcpu_svm *svm);
403 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb);
404 int nested_svm_vmexit(struct vcpu_svm *svm);
405 int nested_svm_exit_handled(struct vcpu_svm *svm);
406 int nested_svm_check_permissions(struct vcpu_svm *svm);
407 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
408 bool has_error_code, u32 error_code);
409 int nested_svm_exit_special(struct vcpu_svm *svm);
410 void sync_nested_vmcb_control(struct vcpu_svm *svm);
412 extern struct kvm_x86_nested_ops svm_nested_ops;
416 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF)
417 #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31
418 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31)
420 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL)
421 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12)
422 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62)
423 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63)
425 #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL
429 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
431 svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
432 vmcb_mark_dirty(svm->vmcb, VMCB_AVIC);
435 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
437 struct vcpu_svm *svm = to_svm(vcpu);
438 u64 *entry = svm->avic_physical_id_cache;
443 return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
446 int avic_ga_log_notifier(u32 ga_tag);
447 void avic_vm_destroy(struct kvm *kvm);
448 int avic_vm_init(struct kvm *kvm);
449 void avic_init_vmcb(struct vcpu_svm *svm);
450 void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
451 int avic_incomplete_ipi_interception(struct vcpu_svm *svm);
452 int avic_unaccelerated_access_interception(struct vcpu_svm *svm);
453 int avic_init_vcpu(struct vcpu_svm *svm);
454 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
455 void avic_vcpu_put(struct kvm_vcpu *vcpu);
456 void avic_post_state_restore(struct kvm_vcpu *vcpu);
457 void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu);
458 void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
459 bool svm_check_apicv_inhibit_reasons(ulong bit);
460 void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate);
461 void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap);
462 void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr);
463 void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr);
464 int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec);
465 bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu);
466 int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
467 uint32_t guest_irq, bool set);
468 void svm_vcpu_blocking(struct kvm_vcpu *vcpu);
469 void svm_vcpu_unblocking(struct kvm_vcpu *vcpu);
473 extern unsigned int max_sev_asid;
475 static inline bool sev_guest(struct kvm *kvm)
477 #ifdef CONFIG_KVM_AMD_SEV
478 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
486 static inline bool svm_sev_enabled(void)
488 return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
491 void sev_vm_destroy(struct kvm *kvm);
492 int svm_mem_enc_op(struct kvm *kvm, void __user *argp);
493 int svm_register_enc_region(struct kvm *kvm,
494 struct kvm_enc_region *range);
495 int svm_unregister_enc_region(struct kvm *kvm,
496 struct kvm_enc_region *range);
497 void pre_sev_run(struct vcpu_svm *svm, int cpu);
498 int __init sev_hardware_setup(void);
499 void sev_hardware_teardown(void);