xref: /linux/arch/x86/kvm/svm/svm.h (revision 3349e275067f94ffb4141989aed9cbae7409429b)
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 #include <linux/bits.h>
21 
22 #include <asm/svm.h>
23 #include <asm/sev-common.h>
24 
25 #include "cpuid.h"
26 #include "kvm_cache_regs.h"
27 
28 /*
29  * Helpers to convert to/from physical addresses for pages whose address is
30  * consumed directly by hardware.  Even though it's a physical address, SVM
31  * often restricts the address to the natural width, hence 'unsigned long'
32  * instead of 'hpa_t'.
33  */
__sme_page_pa(struct page * page)34 static inline unsigned long __sme_page_pa(struct page *page)
35 {
36 	return __sme_set(page_to_pfn(page) << PAGE_SHIFT);
37 }
38 
__sme_pa_to_page(unsigned long pa)39 static inline struct page *__sme_pa_to_page(unsigned long pa)
40 {
41 	return pfn_to_page(__sme_clr(pa) >> PAGE_SHIFT);
42 }
43 
44 #define	IOPM_SIZE PAGE_SIZE * 3
45 #define	MSRPM_SIZE PAGE_SIZE * 2
46 
47 #define MAX_DIRECT_ACCESS_MSRS	48
48 #define MSRPM_OFFSETS	32
49 extern u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
50 extern bool npt_enabled;
51 extern int nrips;
52 extern int vgif;
53 extern bool intercept_smi;
54 extern bool x2avic_enabled;
55 extern bool vnmi;
56 extern int lbrv;
57 
58 /*
59  * Clean bits in VMCB.
60  * VMCB_ALL_CLEAN_MASK might also need to
61  * be updated if this enum is modified.
62  */
63 enum {
64 	VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
65 			    pause filter count */
66 	VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
67 	VMCB_ASID,	 /* ASID */
68 	VMCB_INTR,	 /* int_ctl, int_vector */
69 	VMCB_NPT,        /* npt_en, nCR3, gPAT */
70 	VMCB_CR,	 /* CR0, CR3, CR4, EFER */
71 	VMCB_DR,         /* DR6, DR7 */
72 	VMCB_DT,         /* GDT, IDT */
73 	VMCB_SEG,        /* CS, DS, SS, ES, CPL */
74 	VMCB_CR2,        /* CR2 only */
75 	VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
76 	VMCB_AVIC,       /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
77 			  * AVIC PHYSICAL_TABLE pointer,
78 			  * AVIC LOGICAL_TABLE pointer
79 			  */
80 	VMCB_SW = 31,    /* Reserved for hypervisor/software use */
81 };
82 
83 #define VMCB_ALL_CLEAN_MASK (					\
84 	(1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) |	\
85 	(1U << VMCB_ASID) | (1U << VMCB_INTR) |			\
86 	(1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) |	\
87 	(1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) |	\
88 	(1U << VMCB_LBR) | (1U << VMCB_AVIC) |			\
89 	(1U << VMCB_SW))
90 
91 /* TPR and CR2 are always written before VMRUN */
92 #define VMCB_ALWAYS_DIRTY_MASK	((1U << VMCB_INTR) | (1U << VMCB_CR2))
93 
94 struct kvm_sev_info {
95 	bool active;		/* SEV enabled guest */
96 	bool es_active;		/* SEV-ES enabled guest */
97 	bool need_init;		/* waiting for SEV_INIT2 */
98 	unsigned int asid;	/* ASID used for this guest */
99 	unsigned int handle;	/* SEV firmware handle */
100 	int fd;			/* SEV device fd */
101 	unsigned long policy;
102 	unsigned long pages_locked; /* Number of pages locked */
103 	struct list_head regions_list;  /* List of registered regions */
104 	u64 ap_jump_table;	/* SEV-ES AP Jump Table address */
105 	u64 vmsa_features;
106 	u16 ghcb_version;	/* Highest guest GHCB protocol version allowed */
107 	struct kvm *enc_context_owner; /* Owner of copied encryption context */
108 	struct list_head mirror_vms; /* List of VMs mirroring */
109 	struct list_head mirror_entry; /* Use as a list entry of mirrors */
110 	struct misc_cg *misc_cg; /* For misc cgroup accounting */
111 	atomic_t migration_in_progress;
112 	void *snp_context;      /* SNP guest context page */
113 	void *guest_req_buf;    /* Bounce buffer for SNP Guest Request input */
114 	void *guest_resp_buf;   /* Bounce buffer for SNP Guest Request output */
115 	struct mutex guest_req_mutex; /* Must acquire before using bounce buffers */
116 };
117 
118 #define SEV_POLICY_NODBG	BIT_ULL(0)
119 #define SNP_POLICY_DEBUG	BIT_ULL(19)
120 
121 struct kvm_svm {
122 	struct kvm kvm;
123 
124 	/* Struct members for AVIC */
125 	u32 avic_vm_id;
126 	struct page *avic_logical_id_table_page;
127 	struct page *avic_physical_id_table_page;
128 	struct hlist_node hnode;
129 
130 	struct kvm_sev_info sev_info;
131 };
132 
133 struct kvm_vcpu;
134 
135 struct kvm_vmcb_info {
136 	struct vmcb *ptr;
137 	unsigned long pa;
138 	int cpu;
139 	uint64_t asid_generation;
140 };
141 
142 struct vmcb_save_area_cached {
143 	u64 efer;
144 	u64 cr4;
145 	u64 cr3;
146 	u64 cr0;
147 	u64 dr7;
148 	u64 dr6;
149 };
150 
151 struct vmcb_ctrl_area_cached {
152 	u32 intercepts[MAX_INTERCEPT];
153 	u16 pause_filter_thresh;
154 	u16 pause_filter_count;
155 	u64 iopm_base_pa;
156 	u64 msrpm_base_pa;
157 	u64 tsc_offset;
158 	u32 asid;
159 	u8 tlb_ctl;
160 	u32 int_ctl;
161 	u32 int_vector;
162 	u32 int_state;
163 	u32 exit_code;
164 	u32 exit_code_hi;
165 	u64 exit_info_1;
166 	u64 exit_info_2;
167 	u32 exit_int_info;
168 	u32 exit_int_info_err;
169 	u64 nested_ctl;
170 	u32 event_inj;
171 	u32 event_inj_err;
172 	u64 next_rip;
173 	u64 nested_cr3;
174 	u64 virt_ext;
175 	u32 clean;
176 	u64 bus_lock_rip;
177 	union {
178 #if IS_ENABLED(CONFIG_HYPERV) || IS_ENABLED(CONFIG_KVM_HYPERV)
179 		struct hv_vmcb_enlightenments hv_enlightenments;
180 #endif
181 		u8 reserved_sw[32];
182 	};
183 };
184 
185 struct svm_nested_state {
186 	struct kvm_vmcb_info vmcb02;
187 	u64 hsave_msr;
188 	u64 vm_cr_msr;
189 	u64 vmcb12_gpa;
190 	u64 last_vmcb12_gpa;
191 
192 	/* These are the merged vectors */
193 	u32 *msrpm;
194 
195 	/* A VMRUN has started but has not yet been performed, so
196 	 * we cannot inject a nested vmexit yet.  */
197 	bool nested_run_pending;
198 
199 	/* cache for control fields of the guest */
200 	struct vmcb_ctrl_area_cached ctl;
201 
202 	/*
203 	 * Note: this struct is not kept up-to-date while L2 runs; it is only
204 	 * valid within nested_svm_vmrun.
205 	 */
206 	struct vmcb_save_area_cached save;
207 
208 	bool initialized;
209 
210 	/*
211 	 * Indicates whether MSR bitmap for L2 needs to be rebuilt due to
212 	 * changes in MSR bitmap for L1 or switching to a different L2. Note,
213 	 * this flag can only be used reliably in conjunction with a paravirt L1
214 	 * which informs L0 whether any changes to MSR bitmap for L2 were done
215 	 * on its side.
216 	 */
217 	bool force_msr_bitmap_recalc;
218 };
219 
220 struct vcpu_sev_es_state {
221 	/* SEV-ES support */
222 	struct sev_es_save_area *vmsa;
223 	struct ghcb *ghcb;
224 	u8 valid_bitmap[16];
225 	struct kvm_host_map ghcb_map;
226 	bool received_first_sipi;
227 	unsigned int ap_reset_hold_type;
228 
229 	/* SEV-ES scratch area support */
230 	u64 sw_scratch;
231 	void *ghcb_sa;
232 	u32 ghcb_sa_len;
233 	bool ghcb_sa_sync;
234 	bool ghcb_sa_free;
235 
236 	/* SNP Page-State-Change buffer entries currently being processed */
237 	u16 psc_idx;
238 	u16 psc_inflight;
239 	bool psc_2m;
240 
241 	u64 ghcb_registered_gpa;
242 
243 	struct mutex snp_vmsa_mutex; /* Used to handle concurrent updates of VMSA. */
244 	gpa_t snp_vmsa_gpa;
245 	bool snp_ap_waiting_for_reset;
246 	bool snp_has_guest_vmsa;
247 };
248 
249 struct vcpu_svm {
250 	struct kvm_vcpu vcpu;
251 	/* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
252 	struct vmcb *vmcb;
253 	struct kvm_vmcb_info vmcb01;
254 	struct kvm_vmcb_info *current_vmcb;
255 	u32 asid;
256 	u32 sysenter_esp_hi;
257 	u32 sysenter_eip_hi;
258 	uint64_t tsc_aux;
259 
260 	u64 msr_decfg;
261 
262 	u64 next_rip;
263 
264 	u64 spec_ctrl;
265 
266 	u64 tsc_ratio_msr;
267 	/*
268 	 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
269 	 * translated into the appropriate L2_CFG bits on the host to
270 	 * perform speculative control.
271 	 */
272 	u64 virt_spec_ctrl;
273 
274 	u32 *msrpm;
275 
276 	ulong nmi_iret_rip;
277 
278 	struct svm_nested_state nested;
279 
280 	/* NMI mask value, used when vNMI is not enabled */
281 	bool nmi_masked;
282 
283 	/*
284 	 * True when NMIs are still masked but guest IRET was just intercepted
285 	 * and KVM is waiting for RIP to change, which will signal that the
286 	 * intercepted IRET was retired and thus NMI can be unmasked.
287 	 */
288 	bool awaiting_iret_completion;
289 
290 	/*
291 	 * Set when KVM is awaiting IRET completion and needs to inject NMIs as
292 	 * soon as the IRET completes (e.g. NMI is pending injection).  KVM
293 	 * temporarily steals RFLAGS.TF to single-step the guest in this case
294 	 * in order to regain control as soon as the NMI-blocking condition
295 	 * goes away.
296 	 */
297 	bool nmi_singlestep;
298 	u64 nmi_singlestep_guest_rflags;
299 
300 	bool nmi_l1_to_l2;
301 
302 	unsigned long soft_int_csbase;
303 	unsigned long soft_int_old_rip;
304 	unsigned long soft_int_next_rip;
305 	bool soft_int_injected;
306 
307 	u32 ldr_reg;
308 	u32 dfr_reg;
309 	struct page *avic_backing_page;
310 	u64 *avic_physical_id_cache;
311 
312 	/*
313 	 * Per-vcpu list of struct amd_svm_iommu_ir:
314 	 * This is used mainly to store interrupt remapping information used
315 	 * when update the vcpu affinity. This avoids the need to scan for
316 	 * IRTE and try to match ga_tag in the IOMMU driver.
317 	 */
318 	struct list_head ir_list;
319 	spinlock_t ir_list_lock;
320 
321 	/* Save desired MSR intercept (read: pass-through) state */
322 	struct {
323 		DECLARE_BITMAP(read, MAX_DIRECT_ACCESS_MSRS);
324 		DECLARE_BITMAP(write, MAX_DIRECT_ACCESS_MSRS);
325 	} shadow_msr_intercept;
326 
327 	struct vcpu_sev_es_state sev_es;
328 
329 	bool guest_state_loaded;
330 
331 	bool x2avic_msrs_intercepted;
332 
333 	/* Guest GIF value, used when vGIF is not enabled */
334 	bool guest_gif;
335 };
336 
337 struct svm_cpu_data {
338 	u64 asid_generation;
339 	u32 max_asid;
340 	u32 next_asid;
341 	u32 min_asid;
342 
343 	bool bp_spec_reduce_set;
344 
345 	struct vmcb *save_area;
346 	unsigned long save_area_pa;
347 
348 	/* index = sev_asid, value = vmcb pointer */
349 	struct vmcb **sev_vmcbs;
350 };
351 
352 DECLARE_PER_CPU(struct svm_cpu_data, svm_data);
353 
354 void recalc_intercepts(struct vcpu_svm *svm);
355 
to_kvm_svm(struct kvm * kvm)356 static __always_inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
357 {
358 	return container_of(kvm, struct kvm_svm, kvm);
359 }
360 
to_kvm_sev_info(struct kvm * kvm)361 static __always_inline struct kvm_sev_info *to_kvm_sev_info(struct kvm *kvm)
362 {
363 	return &to_kvm_svm(kvm)->sev_info;
364 }
365 
366 #ifdef CONFIG_KVM_AMD_SEV
sev_guest(struct kvm * kvm)367 static __always_inline bool sev_guest(struct kvm *kvm)
368 {
369 	return to_kvm_sev_info(kvm)->active;
370 }
sev_es_guest(struct kvm * kvm)371 static __always_inline bool sev_es_guest(struct kvm *kvm)
372 {
373 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
374 
375 	return sev->es_active && !WARN_ON_ONCE(!sev->active);
376 }
377 
sev_snp_guest(struct kvm * kvm)378 static __always_inline bool sev_snp_guest(struct kvm *kvm)
379 {
380 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
381 
382 	return (sev->vmsa_features & SVM_SEV_FEAT_SNP_ACTIVE) &&
383 	       !WARN_ON_ONCE(!sev_es_guest(kvm));
384 }
385 #else
386 #define sev_guest(kvm) false
387 #define sev_es_guest(kvm) false
388 #define sev_snp_guest(kvm) false
389 #endif
390 
ghcb_gpa_is_registered(struct vcpu_svm * svm,u64 val)391 static inline bool ghcb_gpa_is_registered(struct vcpu_svm *svm, u64 val)
392 {
393 	return svm->sev_es.ghcb_registered_gpa == val;
394 }
395 
vmcb_mark_all_dirty(struct vmcb * vmcb)396 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
397 {
398 	vmcb->control.clean = 0;
399 }
400 
vmcb_mark_all_clean(struct vmcb * vmcb)401 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
402 {
403 	vmcb->control.clean = VMCB_ALL_CLEAN_MASK
404 			       & ~VMCB_ALWAYS_DIRTY_MASK;
405 }
406 
vmcb_mark_dirty(struct vmcb * vmcb,int bit)407 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
408 {
409 	vmcb->control.clean &= ~(1 << bit);
410 }
411 
vmcb_is_dirty(struct vmcb * vmcb,int bit)412 static inline bool vmcb_is_dirty(struct vmcb *vmcb, int bit)
413 {
414         return !test_bit(bit, (unsigned long *)&vmcb->control.clean);
415 }
416 
to_svm(struct kvm_vcpu * vcpu)417 static __always_inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
418 {
419 	return container_of(vcpu, struct vcpu_svm, vcpu);
420 }
421 
422 /*
423  * Only the PDPTRs are loaded on demand into the shadow MMU.  All other
424  * fields are synchronized on VM-Exit, because accessing the VMCB is cheap.
425  *
426  * CR3 might be out of date in the VMCB but it is not marked dirty; instead,
427  * KVM_REQ_LOAD_MMU_PGD is always requested when the cached vcpu->arch.cr3
428  * is changed.  svm_load_mmu_pgd() then syncs the new CR3 value into the VMCB.
429  */
430 #define SVM_REGS_LAZY_LOAD_SET	(1 << VCPU_EXREG_PDPTR)
431 
vmcb_set_intercept(struct vmcb_control_area * control,u32 bit)432 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
433 {
434 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
435 	__set_bit(bit, (unsigned long *)&control->intercepts);
436 }
437 
vmcb_clr_intercept(struct vmcb_control_area * control,u32 bit)438 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
439 {
440 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
441 	__clear_bit(bit, (unsigned long *)&control->intercepts);
442 }
443 
vmcb_is_intercept(struct vmcb_control_area * control,u32 bit)444 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
445 {
446 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
447 	return test_bit(bit, (unsigned long *)&control->intercepts);
448 }
449 
vmcb12_is_intercept(struct vmcb_ctrl_area_cached * control,u32 bit)450 static inline bool vmcb12_is_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
451 {
452 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
453 	return test_bit(bit, (unsigned long *)&control->intercepts);
454 }
455 
set_exception_intercept(struct vcpu_svm * svm,u32 bit)456 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
457 {
458 	struct vmcb *vmcb = svm->vmcb01.ptr;
459 
460 	WARN_ON_ONCE(bit >= 32);
461 	vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
462 
463 	recalc_intercepts(svm);
464 }
465 
clr_exception_intercept(struct vcpu_svm * svm,u32 bit)466 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
467 {
468 	struct vmcb *vmcb = svm->vmcb01.ptr;
469 
470 	WARN_ON_ONCE(bit >= 32);
471 	vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
472 
473 	recalc_intercepts(svm);
474 }
475 
svm_set_intercept(struct vcpu_svm * svm,int bit)476 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
477 {
478 	struct vmcb *vmcb = svm->vmcb01.ptr;
479 
480 	vmcb_set_intercept(&vmcb->control, bit);
481 
482 	recalc_intercepts(svm);
483 }
484 
svm_clr_intercept(struct vcpu_svm * svm,int bit)485 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
486 {
487 	struct vmcb *vmcb = svm->vmcb01.ptr;
488 
489 	vmcb_clr_intercept(&vmcb->control, bit);
490 
491 	recalc_intercepts(svm);
492 }
493 
svm_is_intercept(struct vcpu_svm * svm,int bit)494 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
495 {
496 	return vmcb_is_intercept(&svm->vmcb->control, bit);
497 }
498 
nested_vgif_enabled(struct vcpu_svm * svm)499 static inline bool nested_vgif_enabled(struct vcpu_svm *svm)
500 {
501 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VGIF) &&
502 	       (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK);
503 }
504 
get_vgif_vmcb(struct vcpu_svm * svm)505 static inline struct vmcb *get_vgif_vmcb(struct vcpu_svm *svm)
506 {
507 	if (!vgif)
508 		return NULL;
509 
510 	if (is_guest_mode(&svm->vcpu) && !nested_vgif_enabled(svm))
511 		return svm->nested.vmcb02.ptr;
512 	else
513 		return svm->vmcb01.ptr;
514 }
515 
enable_gif(struct vcpu_svm * svm)516 static inline void enable_gif(struct vcpu_svm *svm)
517 {
518 	struct vmcb *vmcb = get_vgif_vmcb(svm);
519 
520 	if (vmcb)
521 		vmcb->control.int_ctl |= V_GIF_MASK;
522 	else
523 		svm->guest_gif = true;
524 }
525 
disable_gif(struct vcpu_svm * svm)526 static inline void disable_gif(struct vcpu_svm *svm)
527 {
528 	struct vmcb *vmcb = get_vgif_vmcb(svm);
529 
530 	if (vmcb)
531 		vmcb->control.int_ctl &= ~V_GIF_MASK;
532 	else
533 		svm->guest_gif = false;
534 }
535 
gif_set(struct vcpu_svm * svm)536 static inline bool gif_set(struct vcpu_svm *svm)
537 {
538 	struct vmcb *vmcb = get_vgif_vmcb(svm);
539 
540 	if (vmcb)
541 		return !!(vmcb->control.int_ctl & V_GIF_MASK);
542 	else
543 		return svm->guest_gif;
544 }
545 
nested_npt_enabled(struct vcpu_svm * svm)546 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
547 {
548 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
549 }
550 
nested_vnmi_enabled(struct vcpu_svm * svm)551 static inline bool nested_vnmi_enabled(struct vcpu_svm *svm)
552 {
553 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VNMI) &&
554 	       (svm->nested.ctl.int_ctl & V_NMI_ENABLE_MASK);
555 }
556 
is_x2apic_msrpm_offset(u32 offset)557 static inline bool is_x2apic_msrpm_offset(u32 offset)
558 {
559 	/* 4 msrs per u8, and 4 u8 in u32 */
560 	u32 msr = offset * 16;
561 
562 	return (msr >= APIC_BASE_MSR) &&
563 	       (msr < (APIC_BASE_MSR + 0x100));
564 }
565 
get_vnmi_vmcb_l1(struct vcpu_svm * svm)566 static inline struct vmcb *get_vnmi_vmcb_l1(struct vcpu_svm *svm)
567 {
568 	if (!vnmi)
569 		return NULL;
570 
571 	if (is_guest_mode(&svm->vcpu))
572 		return NULL;
573 	else
574 		return svm->vmcb01.ptr;
575 }
576 
is_vnmi_enabled(struct vcpu_svm * svm)577 static inline bool is_vnmi_enabled(struct vcpu_svm *svm)
578 {
579 	struct vmcb *vmcb = get_vnmi_vmcb_l1(svm);
580 
581 	if (vmcb)
582 		return !!(vmcb->control.int_ctl & V_NMI_ENABLE_MASK);
583 	else
584 		return false;
585 }
586 
svm_vmgexit_set_return_code(struct vcpu_svm * svm,u64 response,u64 data)587 static inline void svm_vmgexit_set_return_code(struct vcpu_svm *svm,
588 						u64 response, u64 data)
589 {
590 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, response);
591 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, data);
592 }
593 
svm_vmgexit_inject_exception(struct vcpu_svm * svm,u8 vector)594 static inline void svm_vmgexit_inject_exception(struct vcpu_svm *svm, u8 vector)
595 {
596 	u64 data = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT | vector;
597 
598 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_ISSUE_EXCEPTION, data);
599 }
600 
svm_vmgexit_bad_input(struct vcpu_svm * svm,u64 suberror)601 static inline void svm_vmgexit_bad_input(struct vcpu_svm *svm, u64 suberror)
602 {
603 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_MALFORMED_INPUT, suberror);
604 }
605 
svm_vmgexit_success(struct vcpu_svm * svm,u64 data)606 static inline void svm_vmgexit_success(struct vcpu_svm *svm, u64 data)
607 {
608 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
609 }
610 
svm_vmgexit_no_action(struct vcpu_svm * svm,u64 data)611 static inline void svm_vmgexit_no_action(struct vcpu_svm *svm, u64 data)
612 {
613 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
614 }
615 
616 /* svm.c */
617 #define MSR_INVALID				0xffffffffU
618 
619 #define DEBUGCTL_RESERVED_BITS (~DEBUGCTLMSR_LBR)
620 
621 extern bool dump_invalid_vmcb;
622 
623 u32 svm_msrpm_offset(u32 msr);
624 u32 *svm_vcpu_alloc_msrpm(void);
625 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm);
626 void svm_vcpu_free_msrpm(u32 *msrpm);
627 void svm_copy_lbrs(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
628 void svm_enable_lbrv(struct kvm_vcpu *vcpu);
629 void svm_update_lbrv(struct kvm_vcpu *vcpu);
630 
631 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
632 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
633 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
634 void disable_nmi_singlestep(struct vcpu_svm *svm);
635 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
636 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
637 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
638 void svm_set_gif(struct vcpu_svm *svm, bool value);
639 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
640 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
641 			  int read, int write);
642 void svm_set_x2apic_msr_interception(struct vcpu_svm *svm, bool disable);
643 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
644 				     int trig_mode, int vec);
645 
646 /* nested.c */
647 
648 #define NESTED_EXIT_HOST	0	/* Exit handled on host level */
649 #define NESTED_EXIT_DONE	1	/* Exit caused nested vmexit  */
650 #define NESTED_EXIT_CONTINUE	2	/* Further checks needed      */
651 
nested_svm_virtualize_tpr(struct kvm_vcpu * vcpu)652 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
653 {
654 	struct vcpu_svm *svm = to_svm(vcpu);
655 
656 	return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
657 }
658 
nested_exit_on_smi(struct vcpu_svm * svm)659 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
660 {
661 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
662 }
663 
nested_exit_on_intr(struct vcpu_svm * svm)664 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
665 {
666 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
667 }
668 
nested_exit_on_nmi(struct vcpu_svm * svm)669 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
670 {
671 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
672 }
673 
674 int enter_svm_guest_mode(struct kvm_vcpu *vcpu,
675 			 u64 vmcb_gpa, struct vmcb *vmcb12, bool from_vmrun);
676 void svm_leave_nested(struct kvm_vcpu *vcpu);
677 void svm_free_nested(struct vcpu_svm *svm);
678 int svm_allocate_nested(struct vcpu_svm *svm);
679 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
680 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
681 			  struct vmcb_save_area *from_save);
682 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
683 int nested_svm_vmexit(struct vcpu_svm *svm);
684 
nested_svm_simple_vmexit(struct vcpu_svm * svm,u32 exit_code)685 static inline int nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
686 {
687 	svm->vmcb->control.exit_code   = exit_code;
688 	svm->vmcb->control.exit_info_1 = 0;
689 	svm->vmcb->control.exit_info_2 = 0;
690 	return nested_svm_vmexit(svm);
691 }
692 
693 int nested_svm_exit_handled(struct vcpu_svm *svm);
694 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
695 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
696 			       bool has_error_code, u32 error_code);
697 int nested_svm_exit_special(struct vcpu_svm *svm);
698 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
699 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu);
700 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
701 				       struct vmcb_control_area *control);
702 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
703 				    struct vmcb_save_area *save);
704 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
705 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
706 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
707 
708 extern struct kvm_x86_nested_ops svm_nested_ops;
709 
710 /* avic.c */
711 #define AVIC_REQUIRED_APICV_INHIBITS			\
712 (							\
713 	BIT(APICV_INHIBIT_REASON_DISABLED) |		\
714 	BIT(APICV_INHIBIT_REASON_ABSENT) |		\
715 	BIT(APICV_INHIBIT_REASON_HYPERV) |		\
716 	BIT(APICV_INHIBIT_REASON_NESTED) |		\
717 	BIT(APICV_INHIBIT_REASON_IRQWIN) |		\
718 	BIT(APICV_INHIBIT_REASON_PIT_REINJ) |		\
719 	BIT(APICV_INHIBIT_REASON_BLOCKIRQ) |		\
720 	BIT(APICV_INHIBIT_REASON_SEV)      |		\
721 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
722 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
723 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED) |	\
724 	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED)	\
725 )
726 
727 bool avic_hardware_setup(void);
728 int avic_ga_log_notifier(u32 ga_tag);
729 void avic_vm_destroy(struct kvm *kvm);
730 int avic_vm_init(struct kvm *kvm);
731 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb);
732 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
733 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
734 int avic_init_vcpu(struct vcpu_svm *svm);
735 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
736 void avic_vcpu_put(struct kvm_vcpu *vcpu);
737 void avic_apicv_post_state_restore(struct kvm_vcpu *vcpu);
738 void avic_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
739 int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
740 			uint32_t guest_irq, bool set);
741 void avic_vcpu_blocking(struct kvm_vcpu *vcpu);
742 void avic_vcpu_unblocking(struct kvm_vcpu *vcpu);
743 void avic_ring_doorbell(struct kvm_vcpu *vcpu);
744 unsigned long avic_vcpu_get_apicv_inhibit_reasons(struct kvm_vcpu *vcpu);
745 void avic_refresh_virtual_apic_mode(struct kvm_vcpu *vcpu);
746 
747 
748 /* sev.c */
749 
750 int pre_sev_run(struct vcpu_svm *svm, int cpu);
751 void sev_init_vmcb(struct vcpu_svm *svm);
752 void sev_vcpu_after_set_cpuid(struct vcpu_svm *svm);
753 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
754 void sev_es_vcpu_reset(struct vcpu_svm *svm);
755 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
756 void sev_es_prepare_switch_to_guest(struct vcpu_svm *svm, struct sev_es_save_area *hostsa);
757 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
758 
759 #ifdef CONFIG_KVM_AMD_SEV
760 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp);
761 int sev_mem_enc_register_region(struct kvm *kvm,
762 				struct kvm_enc_region *range);
763 int sev_mem_enc_unregister_region(struct kvm *kvm,
764 				  struct kvm_enc_region *range);
765 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd);
766 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd);
767 void sev_guest_memory_reclaimed(struct kvm *kvm);
768 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
769 
770 /* These symbols are used in common code and are stubbed below.  */
771 
772 struct page *snp_safe_alloc_page_node(int node, gfp_t gfp);
snp_safe_alloc_page(void)773 static inline struct page *snp_safe_alloc_page(void)
774 {
775 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
776 }
777 
778 void sev_free_vcpu(struct kvm_vcpu *vcpu);
779 void sev_vm_destroy(struct kvm *kvm);
780 void __init sev_set_cpu_caps(void);
781 void __init sev_hardware_setup(void);
782 void sev_hardware_unsetup(void);
783 int sev_cpu_init(struct svm_cpu_data *sd);
784 int sev_dev_get_attr(u32 group, u64 attr, u64 *val);
785 extern unsigned int max_sev_asid;
786 void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code);
787 void sev_snp_init_protected_guest_state(struct kvm_vcpu *vcpu);
788 int sev_gmem_prepare(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order);
789 void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
790 int sev_private_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn);
791 struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu);
792 void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa);
793 #else
snp_safe_alloc_page_node(int node,gfp_t gfp)794 static inline struct page *snp_safe_alloc_page_node(int node, gfp_t gfp)
795 {
796 	return alloc_pages_node(node, gfp | __GFP_ZERO, 0);
797 }
798 
snp_safe_alloc_page(void)799 static inline struct page *snp_safe_alloc_page(void)
800 {
801 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
802 }
803 
sev_free_vcpu(struct kvm_vcpu * vcpu)804 static inline void sev_free_vcpu(struct kvm_vcpu *vcpu) {}
sev_vm_destroy(struct kvm * kvm)805 static inline void sev_vm_destroy(struct kvm *kvm) {}
sev_set_cpu_caps(void)806 static inline void __init sev_set_cpu_caps(void) {}
sev_hardware_setup(void)807 static inline void __init sev_hardware_setup(void) {}
sev_hardware_unsetup(void)808 static inline void sev_hardware_unsetup(void) {}
sev_cpu_init(struct svm_cpu_data * sd)809 static inline int sev_cpu_init(struct svm_cpu_data *sd) { return 0; }
sev_dev_get_attr(u32 group,u64 attr,u64 * val)810 static inline int sev_dev_get_attr(u32 group, u64 attr, u64 *val) { return -ENXIO; }
811 #define max_sev_asid 0
sev_handle_rmp_fault(struct kvm_vcpu * vcpu,gpa_t gpa,u64 error_code)812 static inline void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code) {}
sev_snp_init_protected_guest_state(struct kvm_vcpu * vcpu)813 static inline void sev_snp_init_protected_guest_state(struct kvm_vcpu *vcpu) {}
sev_gmem_prepare(struct kvm * kvm,kvm_pfn_t pfn,gfn_t gfn,int max_order)814 static inline int sev_gmem_prepare(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order)
815 {
816 	return 0;
817 }
sev_gmem_invalidate(kvm_pfn_t start,kvm_pfn_t end)818 static inline void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end) {}
sev_private_max_mapping_level(struct kvm * kvm,kvm_pfn_t pfn)819 static inline int sev_private_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn)
820 {
821 	return 0;
822 }
823 
sev_decrypt_vmsa(struct kvm_vcpu * vcpu)824 static inline struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu)
825 {
826 	return NULL;
827 }
sev_free_decrypted_vmsa(struct kvm_vcpu * vcpu,struct vmcb_save_area * vmsa)828 static inline void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa) {}
829 #endif
830 
831 /* vmenter.S */
832 
833 void __svm_sev_es_vcpu_run(struct vcpu_svm *svm, bool spec_ctrl_intercepted,
834 			   struct sev_es_save_area *hostsa);
835 void __svm_vcpu_run(struct vcpu_svm *svm, bool spec_ctrl_intercepted);
836 
837 #define DEFINE_KVM_GHCB_ACCESSORS(field)						\
838 	static __always_inline bool kvm_ghcb_##field##_is_valid(const struct vcpu_svm *svm) \
839 	{									\
840 		return test_bit(GHCB_BITMAP_IDX(field),				\
841 				(unsigned long *)&svm->sev_es.valid_bitmap);	\
842 	}									\
843 										\
844 	static __always_inline u64 kvm_ghcb_get_##field##_if_valid(struct vcpu_svm *svm, struct ghcb *ghcb) \
845 	{									\
846 		return kvm_ghcb_##field##_is_valid(svm) ? ghcb->save.field : 0;	\
847 	}									\
848 
849 DEFINE_KVM_GHCB_ACCESSORS(cpl)
850 DEFINE_KVM_GHCB_ACCESSORS(rax)
851 DEFINE_KVM_GHCB_ACCESSORS(rcx)
852 DEFINE_KVM_GHCB_ACCESSORS(rdx)
853 DEFINE_KVM_GHCB_ACCESSORS(rbx)
854 DEFINE_KVM_GHCB_ACCESSORS(rsi)
855 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_code)
856 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_1)
857 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_2)
858 DEFINE_KVM_GHCB_ACCESSORS(sw_scratch)
859 DEFINE_KVM_GHCB_ACCESSORS(xcr0)
860 
861 #endif
862