Commit | Line | Data |
---|---|---|
883b0a91 JR |
1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* | |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * AMD SVM support | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. | |
9 | * | |
10 | * Authors: | |
11 | * Yaniv Kamay <yaniv@qumranet.com> | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | */ | |
14 | ||
15 | #ifndef __SVM_SVM_H | |
16 | #define __SVM_SVM_H | |
17 | ||
18 | #include <linux/kvm_types.h> | |
19 | #include <linux/kvm_host.h> | |
20 | ||
21 | #include <asm/svm.h> | |
22 | ||
23 | static const u32 host_save_user_msrs[] = { | |
24 | #ifdef CONFIG_X86_64 | |
25 | MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE, | |
26 | MSR_FS_BASE, | |
27 | #endif | |
28 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
29 | MSR_TSC_AUX, | |
30 | }; | |
31 | ||
32 | #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs) | |
33 | ||
34 | #define MSRPM_OFFSETS 16 | |
35 | extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly; | |
36 | extern bool npt_enabled; | |
37 | ||
38 | enum { | |
39 | VMCB_INTERCEPTS, /* Intercept vectors, TSC offset, | |
40 | pause filter count */ | |
41 | VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */ | |
42 | VMCB_ASID, /* ASID */ | |
43 | VMCB_INTR, /* int_ctl, int_vector */ | |
44 | VMCB_NPT, /* npt_en, nCR3, gPAT */ | |
45 | VMCB_CR, /* CR0, CR3, CR4, EFER */ | |
46 | VMCB_DR, /* DR6, DR7 */ | |
47 | VMCB_DT, /* GDT, IDT */ | |
48 | VMCB_SEG, /* CS, DS, SS, ES, CPL */ | |
49 | VMCB_CR2, /* CR2 only */ | |
50 | VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */ | |
51 | VMCB_AVIC, /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE, | |
52 | * AVIC PHYSICAL_TABLE pointer, | |
53 | * AVIC LOGICAL_TABLE pointer | |
54 | */ | |
55 | VMCB_DIRTY_MAX, | |
56 | }; | |
57 | ||
58 | /* TPR and CR2 are always written before VMRUN */ | |
59 | #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2)) | |
60 | ||
61 | struct kvm_sev_info { | |
62 | bool active; /* SEV enabled guest */ | |
63 | unsigned int asid; /* ASID used for this guest */ | |
64 | unsigned int handle; /* SEV firmware handle */ | |
65 | int fd; /* SEV device fd */ | |
66 | unsigned long pages_locked; /* Number of pages locked */ | |
67 | struct list_head regions_list; /* List of registered regions */ | |
68 | }; | |
69 | ||
70 | struct kvm_svm { | |
71 | struct kvm kvm; | |
72 | ||
73 | /* Struct members for AVIC */ | |
74 | u32 avic_vm_id; | |
75 | struct page *avic_logical_id_table_page; | |
76 | struct page *avic_physical_id_table_page; | |
77 | struct hlist_node hnode; | |
78 | ||
79 | struct kvm_sev_info sev_info; | |
80 | }; | |
81 | ||
82 | struct kvm_vcpu; | |
83 | ||
7693b3eb | 84 | struct svm_nested_state { |
883b0a91 JR |
85 | struct vmcb *hsave; |
86 | u64 hsave_msr; | |
87 | u64 vm_cr_msr; | |
88 | u64 vmcb; | |
7c86663b | 89 | u32 host_intercept_exceptions; |
883b0a91 JR |
90 | |
91 | /* These are the merged vectors */ | |
92 | u32 *msrpm; | |
93 | ||
f74f9414 PB |
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; | |
97 | ||
e670bf68 PB |
98 | /* cache for control fields of the guest */ |
99 | struct vmcb_control_area ctl; | |
883b0a91 JR |
100 | }; |
101 | ||
102 | struct vcpu_svm { | |
103 | struct kvm_vcpu vcpu; | |
104 | struct vmcb *vmcb; | |
105 | unsigned long vmcb_pa; | |
106 | struct svm_cpu_data *svm_data; | |
107 | uint64_t asid_generation; | |
108 | uint64_t sysenter_esp; | |
109 | uint64_t sysenter_eip; | |
110 | uint64_t tsc_aux; | |
111 | ||
112 | u64 msr_decfg; | |
113 | ||
114 | u64 next_rip; | |
115 | ||
116 | u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS]; | |
117 | struct { | |
118 | u16 fs; | |
119 | u16 gs; | |
120 | u16 ldt; | |
121 | u64 gs_base; | |
122 | } host; | |
123 | ||
124 | u64 spec_ctrl; | |
125 | /* | |
126 | * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be | |
127 | * translated into the appropriate L2_CFG bits on the host to | |
128 | * perform speculative control. | |
129 | */ | |
130 | u64 virt_spec_ctrl; | |
131 | ||
132 | u32 *msrpm; | |
133 | ||
134 | ulong nmi_iret_rip; | |
135 | ||
7693b3eb | 136 | struct svm_nested_state nested; |
883b0a91 JR |
137 | |
138 | bool nmi_singlestep; | |
139 | u64 nmi_singlestep_guest_rflags; | |
140 | ||
141 | unsigned int3_injected; | |
142 | unsigned long int3_rip; | |
143 | ||
144 | /* cached guest cpuid flags for faster access */ | |
145 | bool nrips_enabled : 1; | |
146 | ||
147 | u32 ldr_reg; | |
148 | u32 dfr_reg; | |
149 | struct page *avic_backing_page; | |
150 | u64 *avic_physical_id_cache; | |
151 | bool avic_is_running; | |
152 | ||
153 | /* | |
154 | * Per-vcpu list of struct amd_svm_iommu_ir: | |
155 | * This is used mainly to store interrupt remapping information used | |
156 | * when update the vcpu affinity. This avoids the need to scan for | |
157 | * IRTE and try to match ga_tag in the IOMMU driver. | |
158 | */ | |
159 | struct list_head ir_list; | |
160 | spinlock_t ir_list_lock; | |
883b0a91 JR |
161 | }; |
162 | ||
eaf78265 JR |
163 | struct svm_cpu_data { |
164 | int cpu; | |
165 | ||
166 | u64 asid_generation; | |
167 | u32 max_asid; | |
168 | u32 next_asid; | |
169 | u32 min_asid; | |
170 | struct kvm_ldttss_desc *tss_desc; | |
171 | ||
172 | struct page *save_area; | |
173 | struct vmcb *current_vmcb; | |
174 | ||
175 | /* index = sev_asid, value = vmcb pointer */ | |
176 | struct vmcb **sev_vmcbs; | |
177 | }; | |
178 | ||
179 | DECLARE_PER_CPU(struct svm_cpu_data *, svm_data); | |
180 | ||
883b0a91 JR |
181 | void recalc_intercepts(struct vcpu_svm *svm); |
182 | ||
ef0f6496 JR |
183 | static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm) |
184 | { | |
185 | return container_of(kvm, struct kvm_svm, kvm); | |
186 | } | |
187 | ||
883b0a91 JR |
188 | static inline void mark_all_dirty(struct vmcb *vmcb) |
189 | { | |
190 | vmcb->control.clean = 0; | |
191 | } | |
192 | ||
193 | static inline void mark_all_clean(struct vmcb *vmcb) | |
194 | { | |
195 | vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1) | |
196 | & ~VMCB_ALWAYS_DIRTY_MASK; | |
197 | } | |
198 | ||
199 | static inline void mark_dirty(struct vmcb *vmcb, int bit) | |
200 | { | |
201 | vmcb->control.clean &= ~(1 << bit); | |
202 | } | |
203 | ||
204 | static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) | |
205 | { | |
206 | return container_of(vcpu, struct vcpu_svm, vcpu); | |
207 | } | |
208 | ||
209 | static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm) | |
210 | { | |
211 | if (is_guest_mode(&svm->vcpu)) | |
212 | return svm->nested.hsave; | |
213 | else | |
214 | return svm->vmcb; | |
215 | } | |
216 | ||
217 | static inline void set_cr_intercept(struct vcpu_svm *svm, int bit) | |
218 | { | |
219 | struct vmcb *vmcb = get_host_vmcb(svm); | |
220 | ||
221 | vmcb->control.intercept_cr |= (1U << bit); | |
222 | ||
223 | recalc_intercepts(svm); | |
224 | } | |
225 | ||
226 | static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit) | |
227 | { | |
228 | struct vmcb *vmcb = get_host_vmcb(svm); | |
229 | ||
230 | vmcb->control.intercept_cr &= ~(1U << bit); | |
231 | ||
232 | recalc_intercepts(svm); | |
233 | } | |
234 | ||
235 | static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit) | |
236 | { | |
237 | struct vmcb *vmcb = get_host_vmcb(svm); | |
238 | ||
239 | return vmcb->control.intercept_cr & (1U << bit); | |
240 | } | |
241 | ||
242 | static inline void set_dr_intercepts(struct vcpu_svm *svm) | |
243 | { | |
244 | struct vmcb *vmcb = get_host_vmcb(svm); | |
245 | ||
246 | vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ) | |
247 | | (1 << INTERCEPT_DR1_READ) | |
248 | | (1 << INTERCEPT_DR2_READ) | |
249 | | (1 << INTERCEPT_DR3_READ) | |
250 | | (1 << INTERCEPT_DR4_READ) | |
251 | | (1 << INTERCEPT_DR5_READ) | |
252 | | (1 << INTERCEPT_DR6_READ) | |
253 | | (1 << INTERCEPT_DR7_READ) | |
254 | | (1 << INTERCEPT_DR0_WRITE) | |
255 | | (1 << INTERCEPT_DR1_WRITE) | |
256 | | (1 << INTERCEPT_DR2_WRITE) | |
257 | | (1 << INTERCEPT_DR3_WRITE) | |
258 | | (1 << INTERCEPT_DR4_WRITE) | |
259 | | (1 << INTERCEPT_DR5_WRITE) | |
260 | | (1 << INTERCEPT_DR6_WRITE) | |
261 | | (1 << INTERCEPT_DR7_WRITE); | |
262 | ||
263 | recalc_intercepts(svm); | |
264 | } | |
265 | ||
266 | static inline void clr_dr_intercepts(struct vcpu_svm *svm) | |
267 | { | |
268 | struct vmcb *vmcb = get_host_vmcb(svm); | |
269 | ||
270 | vmcb->control.intercept_dr = 0; | |
271 | ||
272 | recalc_intercepts(svm); | |
273 | } | |
274 | ||
275 | static inline void set_exception_intercept(struct vcpu_svm *svm, int bit) | |
276 | { | |
277 | struct vmcb *vmcb = get_host_vmcb(svm); | |
278 | ||
279 | vmcb->control.intercept_exceptions |= (1U << bit); | |
280 | ||
281 | recalc_intercepts(svm); | |
282 | } | |
283 | ||
284 | static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit) | |
285 | { | |
286 | struct vmcb *vmcb = get_host_vmcb(svm); | |
287 | ||
288 | vmcb->control.intercept_exceptions &= ~(1U << bit); | |
289 | ||
290 | recalc_intercepts(svm); | |
291 | } | |
292 | ||
293 | static inline void set_intercept(struct vcpu_svm *svm, int bit) | |
294 | { | |
295 | struct vmcb *vmcb = get_host_vmcb(svm); | |
296 | ||
297 | vmcb->control.intercept |= (1ULL << bit); | |
298 | ||
299 | recalc_intercepts(svm); | |
300 | } | |
301 | ||
302 | static inline void clr_intercept(struct vcpu_svm *svm, int bit) | |
303 | { | |
304 | struct vmcb *vmcb = get_host_vmcb(svm); | |
305 | ||
306 | vmcb->control.intercept &= ~(1ULL << bit); | |
307 | ||
308 | recalc_intercepts(svm); | |
309 | } | |
310 | ||
311 | static inline bool is_intercept(struct vcpu_svm *svm, int bit) | |
312 | { | |
313 | return (svm->vmcb->control.intercept & (1ULL << bit)) != 0; | |
314 | } | |
315 | ||
316 | static inline bool vgif_enabled(struct vcpu_svm *svm) | |
317 | { | |
318 | return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK); | |
319 | } | |
320 | ||
321 | static inline void enable_gif(struct vcpu_svm *svm) | |
322 | { | |
323 | if (vgif_enabled(svm)) | |
324 | svm->vmcb->control.int_ctl |= V_GIF_MASK; | |
325 | else | |
326 | svm->vcpu.arch.hflags |= HF_GIF_MASK; | |
327 | } | |
328 | ||
329 | static inline void disable_gif(struct vcpu_svm *svm) | |
330 | { | |
331 | if (vgif_enabled(svm)) | |
332 | svm->vmcb->control.int_ctl &= ~V_GIF_MASK; | |
333 | else | |
334 | svm->vcpu.arch.hflags &= ~HF_GIF_MASK; | |
335 | } | |
336 | ||
337 | static inline bool gif_set(struct vcpu_svm *svm) | |
338 | { | |
339 | if (vgif_enabled(svm)) | |
340 | return !!(svm->vmcb->control.int_ctl & V_GIF_MASK); | |
341 | else | |
342 | return !!(svm->vcpu.arch.hflags & HF_GIF_MASK); | |
343 | } | |
344 | ||
345 | /* svm.c */ | |
346 | #define MSR_INVALID 0xffffffffU | |
347 | ||
348 | u32 svm_msrpm_offset(u32 msr); | |
349 | void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer); | |
350 | void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0); | |
351 | int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4); | |
f55ac304 | 352 | void svm_flush_tlb(struct kvm_vcpu *vcpu); |
883b0a91 | 353 | void disable_nmi_singlestep(struct vcpu_svm *svm); |
cae96af1 PB |
354 | bool svm_smi_blocked(struct kvm_vcpu *vcpu); |
355 | bool svm_nmi_blocked(struct kvm_vcpu *vcpu); | |
356 | bool svm_interrupt_blocked(struct kvm_vcpu *vcpu); | |
ffdf7f9e | 357 | void svm_set_gif(struct vcpu_svm *svm, bool value); |
883b0a91 JR |
358 | |
359 | /* nested.c */ | |
360 | ||
361 | #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ | |
362 | #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ | |
363 | #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ | |
364 | ||
883b0a91 JR |
365 | static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu) |
366 | { | |
e9fd761a PB |
367 | struct vcpu_svm *svm = to_svm(vcpu); |
368 | ||
369 | return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK); | |
883b0a91 JR |
370 | } |
371 | ||
55714cdd PB |
372 | static inline bool nested_exit_on_smi(struct vcpu_svm *svm) |
373 | { | |
e670bf68 | 374 | return (svm->nested.ctl.intercept & (1ULL << INTERCEPT_SMI)); |
55714cdd PB |
375 | } |
376 | ||
fc6f7c03 PB |
377 | static inline bool nested_exit_on_intr(struct vcpu_svm *svm) |
378 | { | |
e670bf68 | 379 | return (svm->nested.ctl.intercept & (1ULL << INTERCEPT_INTR)); |
fc6f7c03 PB |
380 | } |
381 | ||
bbdad0b5 PB |
382 | static inline bool nested_exit_on_nmi(struct vcpu_svm *svm) |
383 | { | |
e670bf68 | 384 | return (svm->nested.ctl.intercept & (1ULL << INTERCEPT_NMI)); |
bbdad0b5 PB |
385 | } |
386 | ||
883b0a91 | 387 | void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa, |
69c9dfa2 | 388 | struct vmcb *nested_vmcb); |
c513f484 | 389 | void svm_leave_nested(struct vcpu_svm *svm); |
883b0a91 JR |
390 | int nested_svm_vmrun(struct vcpu_svm *svm); |
391 | void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb); | |
392 | int nested_svm_vmexit(struct vcpu_svm *svm); | |
393 | int nested_svm_exit_handled(struct vcpu_svm *svm); | |
394 | int nested_svm_check_permissions(struct vcpu_svm *svm); | |
395 | int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, | |
396 | bool has_error_code, u32 error_code); | |
883b0a91 | 397 | int nested_svm_exit_special(struct vcpu_svm *svm); |
2d8a42be | 398 | void sync_nested_vmcb_control(struct vcpu_svm *svm); |
883b0a91 | 399 | |
33b22172 PB |
400 | extern struct kvm_x86_nested_ops svm_nested_ops; |
401 | ||
ef0f6496 JR |
402 | /* avic.c */ |
403 | ||
404 | #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK (0xFF) | |
405 | #define AVIC_LOGICAL_ID_ENTRY_VALID_BIT 31 | |
406 | #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK (1 << 31) | |
407 | ||
408 | #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK (0xFFULL) | |
409 | #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK (0xFFFFFFFFFFULL << 12) | |
410 | #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK (1ULL << 62) | |
411 | #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK (1ULL << 63) | |
412 | ||
413 | #define VMCB_AVIC_APIC_BAR_MASK 0xFFFFFFFFFF000ULL | |
414 | ||
415 | extern int avic; | |
416 | ||
417 | static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data) | |
418 | { | |
419 | svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK; | |
420 | mark_dirty(svm->vmcb, VMCB_AVIC); | |
421 | } | |
422 | ||
423 | static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu) | |
424 | { | |
425 | struct vcpu_svm *svm = to_svm(vcpu); | |
426 | u64 *entry = svm->avic_physical_id_cache; | |
427 | ||
428 | if (!entry) | |
429 | return false; | |
430 | ||
431 | return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK); | |
432 | } | |
433 | ||
434 | int avic_ga_log_notifier(u32 ga_tag); | |
435 | void avic_vm_destroy(struct kvm *kvm); | |
436 | int avic_vm_init(struct kvm *kvm); | |
437 | void avic_init_vmcb(struct vcpu_svm *svm); | |
438 | void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate); | |
439 | int avic_incomplete_ipi_interception(struct vcpu_svm *svm); | |
440 | int avic_unaccelerated_access_interception(struct vcpu_svm *svm); | |
441 | int avic_init_vcpu(struct vcpu_svm *svm); | |
442 | void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu); | |
443 | void avic_vcpu_put(struct kvm_vcpu *vcpu); | |
444 | void avic_post_state_restore(struct kvm_vcpu *vcpu); | |
445 | void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu); | |
446 | void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu); | |
447 | bool svm_check_apicv_inhibit_reasons(ulong bit); | |
448 | void svm_pre_update_apicv_exec_ctrl(struct kvm *kvm, bool activate); | |
449 | void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap); | |
450 | void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr); | |
451 | void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr); | |
452 | int svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec); | |
453 | bool svm_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu); | |
454 | int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq, | |
455 | uint32_t guest_irq, bool set); | |
456 | void svm_vcpu_blocking(struct kvm_vcpu *vcpu); | |
457 | void svm_vcpu_unblocking(struct kvm_vcpu *vcpu); | |
458 | ||
eaf78265 JR |
459 | /* sev.c */ |
460 | ||
461 | extern unsigned int max_sev_asid; | |
462 | ||
463 | static inline bool sev_guest(struct kvm *kvm) | |
464 | { | |
465 | #ifdef CONFIG_KVM_AMD_SEV | |
466 | struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; | |
467 | ||
468 | return sev->active; | |
469 | #else | |
470 | return false; | |
471 | #endif | |
472 | } | |
473 | ||
474 | static inline bool svm_sev_enabled(void) | |
475 | { | |
476 | return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0; | |
477 | } | |
478 | ||
479 | void sev_vm_destroy(struct kvm *kvm); | |
480 | int svm_mem_enc_op(struct kvm *kvm, void __user *argp); | |
481 | int svm_register_enc_region(struct kvm *kvm, | |
482 | struct kvm_enc_region *range); | |
483 | int svm_unregister_enc_region(struct kvm *kvm, | |
484 | struct kvm_enc_region *range); | |
485 | void pre_sev_run(struct vcpu_svm *svm, int cpu); | |
486 | int __init sev_hardware_setup(void); | |
487 | void sev_hardware_teardown(void); | |
488 | ||
883b0a91 | 489 | #endif |