xref: /linux/arch/x86/kvm/svm/svm.h (revision 01f217fa8a8c7878d28df90233f68c20bea9bdc7)
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  */
34 static inline unsigned long __sme_page_pa(struct page *page)
35 {
36 	return __sme_set(page_to_pfn(page) << PAGE_SHIFT);
37 }
38 
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 extern bool npt_enabled;
48 extern int nrips;
49 extern int vgif;
50 extern bool intercept_smi;
51 extern bool vnmi;
52 extern int lbrv;
53 
54 extern int tsc_aux_uret_slot __ro_after_init;
55 
56 extern struct kvm_x86_ops svm_x86_ops __initdata;
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_CET,	 /* S_CET, SSP, ISST_ADDR */
81 	VMCB_SW = 31,    /* Reserved for hypervisor/software use */
82 };
83 
84 #define VMCB_ALL_CLEAN_MASK (					\
85 	(1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) |	\
86 	(1U << VMCB_ASID) | (1U << VMCB_INTR) |			\
87 	(1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) |	\
88 	(1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) |	\
89 	(1U << VMCB_LBR) | (1U << VMCB_AVIC) | (1U << VMCB_CET) | \
90 	(1U << VMCB_SW))
91 
92 /* TPR and CR2 are always written before VMRUN */
93 #define VMCB_ALWAYS_DIRTY_MASK	((1U << VMCB_INTR) | (1U << VMCB_CR2))
94 
95 #ifdef CONFIG_KVM_AMD_SEV
96 struct kvm_sev_info {
97 	bool active;		/* SEV enabled guest */
98 	bool es_active;		/* SEV-ES enabled guest */
99 	bool need_init;		/* waiting for SEV_INIT2 */
100 	unsigned int asid;	/* ASID used for this guest */
101 	unsigned int handle;	/* SEV firmware handle */
102 	int fd;			/* SEV device fd */
103 	unsigned long policy;
104 	unsigned long pages_locked; /* Number of pages locked */
105 	struct list_head regions_list;  /* List of registered regions */
106 	u64 ap_jump_table;	/* SEV-ES AP Jump Table address */
107 	u64 vmsa_features;
108 	u16 ghcb_version;	/* Highest guest GHCB protocol version allowed */
109 	struct kvm *enc_context_owner; /* Owner of copied encryption context */
110 	struct list_head mirror_vms; /* List of VMs mirroring */
111 	struct list_head mirror_entry; /* Use as a list entry of mirrors */
112 	struct misc_cg *misc_cg; /* For misc cgroup accounting */
113 	atomic_t migration_in_progress;
114 	void *snp_context;      /* SNP guest context page */
115 	void *guest_req_buf;    /* Bounce buffer for SNP Guest Request input */
116 	void *guest_resp_buf;   /* Bounce buffer for SNP Guest Request output */
117 	struct mutex guest_req_mutex; /* Must acquire before using bounce buffers */
118 	cpumask_var_t have_run_cpus; /* CPUs that have done VMRUN for this VM. */
119 	bool snp_certs_enabled;	/* SNP certificate-fetching support. */
120 };
121 #endif
122 
123 struct kvm_svm {
124 	struct kvm kvm;
125 
126 	/* Struct members for AVIC */
127 	u32 avic_vm_id;
128 	u32 *avic_logical_id_table;
129 	u64 *avic_physical_id_table;
130 	struct hlist_node hnode;
131 
132 #ifdef CONFIG_KVM_AMD_SEV
133 	struct kvm_sev_info sev_info;
134 #endif
135 };
136 
137 struct kvm_vcpu;
138 
139 struct kvm_vmcb_info {
140 	struct vmcb *ptr;
141 	unsigned long pa;
142 	int cpu;
143 	uint64_t asid_generation;
144 };
145 
146 struct vmcb_save_area_cached {
147 	struct vmcb_seg es;
148 	struct vmcb_seg cs;
149 	struct vmcb_seg ss;
150 	struct vmcb_seg ds;
151 	struct vmcb_seg gdtr;
152 	struct vmcb_seg idtr;
153 	u8 cpl;
154 	u64 efer;
155 	u64 cr4;
156 	u64 cr3;
157 	u64 cr0;
158 	u64 dr7;
159 	u64 dr6;
160 	u64 rflags;
161 	u64 rip;
162 	u64 rsp;
163 	u64 s_cet;
164 	u64 ssp;
165 	u64 isst_addr;
166 	u64 rax;
167 	u64 cr2;
168 	u64 dbgctl;
169 	u64 br_from;
170 	u64 br_to;
171 	u64 last_excp_from;
172 	u64 last_excp_to;
173 };
174 
175 struct vmcb_ctrl_area_cached {
176 	u32 intercepts[MAX_INTERCEPT];
177 	u16 pause_filter_thresh;
178 	u16 pause_filter_count;
179 	u64 iopm_base_pa;
180 	u64 msrpm_base_pa;
181 	u64 tsc_offset;
182 	u32 asid;
183 	u8 tlb_ctl;
184 	u8 erap_ctl;
185 	u32 int_ctl;
186 	u32 int_vector;
187 	u32 int_state;
188 	u64 exit_code;
189 	u64 exit_info_1;
190 	u64 exit_info_2;
191 	u32 exit_int_info;
192 	u32 exit_int_info_err;
193 	u64 misc_ctl;
194 	u32 event_inj;
195 	u32 event_inj_err;
196 	u64 next_rip;
197 	u64 nested_cr3;
198 	u64 misc_ctl2;
199 	u32 clean;
200 	union {
201 #if IS_ENABLED(CONFIG_HYPERV) || IS_ENABLED(CONFIG_KVM_HYPERV)
202 		struct hv_vmcb_enlightenments hv_enlightenments;
203 #endif
204 		u8 reserved_sw[32];
205 	};
206 };
207 
208 struct svm_nested_state {
209 	struct kvm_vmcb_info vmcb02;
210 	u64 hsave_msr;
211 	u64 vm_cr_msr;
212 	u64 vmcb12_gpa;
213 	u64 last_vmcb12_gpa;
214 	u64 last_bus_lock_rip;
215 
216 	/*
217 	 * The MSR permissions map used for vmcb02, which is the merge result
218 	 * of vmcb01 and vmcb12
219 	 */
220 	void *msrpm;
221 
222 	/* cache for control fields of the guest */
223 	struct vmcb_ctrl_area_cached ctl;
224 
225 	/*
226 	 * Note: this struct is not kept up-to-date while L2 runs; it is only
227 	 * valid within nested_svm_vmrun.
228 	 */
229 	struct vmcb_save_area_cached save;
230 
231 	bool initialized;
232 
233 	/*
234 	 * Indicates whether MSR bitmap for L2 needs to be rebuilt due to
235 	 * changes in MSR bitmap for L1 or switching to a different L2. Note,
236 	 * this flag can only be used reliably in conjunction with a paravirt L1
237 	 * which informs L0 whether any changes to MSR bitmap for L2 were done
238 	 * on its side.
239 	 */
240 	bool force_msr_bitmap_recalc;
241 };
242 
243 struct vcpu_sev_es_state {
244 	/* SEV-ES support */
245 	struct sev_es_save_area *vmsa;
246 	struct ghcb *ghcb;
247 	u8 valid_bitmap[16];
248 	struct kvm_host_map ghcb_map;
249 	bool received_first_sipi;
250 	unsigned int ap_reset_hold_type;
251 
252 	/* SEV-ES scratch area support */
253 	u64 sw_scratch;
254 	void *ghcb_sa;
255 	u32 ghcb_sa_len;
256 	bool ghcb_sa_sync;
257 	bool ghcb_sa_free;
258 
259 	/* SNP Page-State-Change buffer entries currently being processed */
260 	u16 psc_idx;
261 	u16 psc_inflight;
262 	bool psc_2m;
263 
264 	u64 ghcb_registered_gpa;
265 
266 	struct mutex snp_vmsa_mutex; /* Used to handle concurrent updates of VMSA. */
267 	gpa_t snp_vmsa_gpa;
268 	bool snp_ap_waiting_for_reset;
269 	bool snp_has_guest_vmsa;
270 };
271 
272 struct vcpu_svm {
273 	struct kvm_vcpu vcpu;
274 	/* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
275 	struct vmcb *vmcb;
276 	struct kvm_vmcb_info vmcb01;
277 	struct kvm_vmcb_info *current_vmcb;
278 	u32 asid;
279 	u32 sysenter_esp_hi;
280 	u32 sysenter_eip_hi;
281 	uint64_t tsc_aux;
282 
283 	u64 msr_decfg;
284 
285 	u64 next_rip;
286 
287 	u64 spec_ctrl;
288 
289 	u64 tsc_ratio_msr;
290 	/*
291 	 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
292 	 * translated into the appropriate L2_CFG bits on the host to
293 	 * perform speculative control.
294 	 */
295 	u64 virt_spec_ctrl;
296 
297 	void *msrpm;
298 
299 	ulong nmi_iret_rip;
300 
301 	struct svm_nested_state nested;
302 
303 	/* NMI mask value, used when vNMI is not enabled */
304 	bool nmi_masked;
305 
306 	/*
307 	 * True when NMIs are still masked but guest IRET was just intercepted
308 	 * and KVM is waiting for RIP to change, which will signal that the
309 	 * intercepted IRET was retired and thus NMI can be unmasked.
310 	 */
311 	bool awaiting_iret_completion;
312 
313 	/*
314 	 * Set when KVM is awaiting IRET completion and needs to inject NMIs as
315 	 * soon as the IRET completes (e.g. NMI is pending injection).  KVM
316 	 * temporarily steals RFLAGS.TF to single-step the guest in this case
317 	 * in order to regain control as soon as the NMI-blocking condition
318 	 * goes away.
319 	 */
320 	bool nmi_singlestep;
321 	u64 nmi_singlestep_guest_rflags;
322 
323 	bool nmi_l1_to_l2;
324 
325 	unsigned long soft_int_csbase;
326 	unsigned long soft_int_old_rip;
327 	unsigned long soft_int_next_rip;
328 	bool soft_int_injected;
329 
330 	u32 ldr_reg;
331 	u32 dfr_reg;
332 
333 	/* This is essentially a shadow of the vCPU's actual entry in the
334 	 * Physical ID table that is programmed into the VMCB, i.e. that is
335 	 * seen by the CPU.  If IPI virtualization is disabled, IsRunning is
336 	 * only ever set in the shadow, i.e. is never propagated to the "real"
337 	 * table, so that hardware never sees IsRunning=1.
338 	 */
339 	u64 avic_physical_id_entry;
340 
341 	/*
342 	 * Per-vCPU list of irqfds that are eligible to post IRQs directly to
343 	 * the vCPU (a.k.a. device posted IRQs, a.k.a. IRQ bypass).  The list
344 	 * is used to reconfigure IRTEs when the vCPU is loaded/put (to set the
345 	 * target pCPU), when AVIC is toggled on/off (to (de)activate bypass),
346 	 * and if the irqfd becomes ineligible for posting (to put the IRTE
347 	 * back into remapped mode).
348 	 */
349 	struct list_head ir_list;
350 	raw_spinlock_t ir_list_lock;
351 
352 	struct vcpu_sev_es_state sev_es;
353 
354 	bool guest_state_loaded;
355 
356 	bool avic_irq_window;
357 	bool x2avic_msrs_intercepted;
358 	bool lbr_msrs_intercepted;
359 
360 	/* Guest GIF value, used when vGIF is not enabled */
361 	bool guest_gif;
362 };
363 
364 struct svm_cpu_data {
365 	u64 asid_generation;
366 	u32 max_asid;
367 	u32 next_asid;
368 	u32 min_asid;
369 
370 	bool bp_spec_reduce_set;
371 
372 	struct vmcb *save_area;
373 	unsigned long save_area_pa;
374 
375 	/* index = sev_asid, value = vmcb pointer */
376 	struct vmcb **sev_vmcbs;
377 };
378 
379 DECLARE_PER_CPU(struct svm_cpu_data, svm_data);
380 
381 static __always_inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
382 {
383 	return container_of(kvm, struct kvm_svm, kvm);
384 }
385 
386 #ifdef CONFIG_KVM_AMD_SEV
387 static __always_inline struct kvm_sev_info *to_kvm_sev_info(struct kvm *kvm)
388 {
389 	return &to_kvm_svm(kvm)->sev_info;
390 }
391 
392 static __always_inline bool ____sev_guest(struct kvm *kvm)
393 {
394 	return to_kvm_sev_info(kvm)->active;
395 }
396 static __always_inline bool ____sev_es_guest(struct kvm *kvm)
397 {
398 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
399 
400 	return sev->es_active && !WARN_ON_ONCE(!sev->active);
401 }
402 
403 static __always_inline bool ____sev_snp_guest(struct kvm *kvm)
404 {
405 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
406 
407 	return (sev->vmsa_features & SVM_SEV_FEAT_SNP_ACTIVE) &&
408 	       !WARN_ON_ONCE(!____sev_es_guest(kvm));
409 }
410 
411 static __always_inline bool is_sev_guest(struct kvm_vcpu *vcpu)
412 {
413 	return ____sev_guest(vcpu->kvm);
414 }
415 static __always_inline bool is_sev_es_guest(struct kvm_vcpu *vcpu)
416 {
417 	return ____sev_es_guest(vcpu->kvm);
418 }
419 
420 static __always_inline bool is_sev_snp_guest(struct kvm_vcpu *vcpu)
421 {
422 	return ____sev_snp_guest(vcpu->kvm);
423 }
424 #else
425 static __always_inline bool is_sev_guest(struct kvm_vcpu *vcpu)
426 {
427 	return false;
428 }
429 static __always_inline bool is_sev_es_guest(struct kvm_vcpu *vcpu)
430 {
431 	return false;
432 }
433 
434 static __always_inline bool is_sev_snp_guest(struct kvm_vcpu *vcpu)
435 {
436 	return false;
437 }
438 #endif
439 
440 static inline bool ghcb_gpa_is_registered(struct vcpu_svm *svm, u64 val)
441 {
442 	return svm->sev_es.ghcb_registered_gpa == val;
443 }
444 
445 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
446 {
447 	vmcb->control.clean = 0;
448 }
449 
450 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
451 {
452 	vmcb->control.clean = VMCB_ALL_CLEAN_MASK
453 			       & ~VMCB_ALWAYS_DIRTY_MASK;
454 }
455 
456 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
457 {
458 	vmcb->control.clean &= ~(1 << bit);
459 }
460 
461 static inline bool vmcb12_is_dirty(struct vmcb_ctrl_area_cached *control, int bit)
462 {
463 	return !test_bit(bit, (unsigned long *)&control->clean);
464 }
465 
466 static __always_inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
467 {
468 	return container_of(vcpu, struct vcpu_svm, vcpu);
469 }
470 
471 static inline bool svm_is_vmrun_failure(u64 exit_code)
472 {
473 	if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
474 		return (u32)exit_code == (u32)SVM_EXIT_ERR;
475 
476 	return exit_code == SVM_EXIT_ERR;
477 }
478 
479 /*
480  * Only the PDPTRs are loaded on demand into the shadow MMU.  All other
481  * fields are synchronized on VM-Exit, because accessing the VMCB is cheap.
482  *
483  * CR3 might be out of date in the VMCB but it is not marked dirty; instead,
484  * KVM_REQ_LOAD_MMU_PGD is always requested when the cached vcpu->arch.cr3
485  * is changed.  svm_load_mmu_pgd() then syncs the new CR3 value into the VMCB.
486  */
487 #define SVM_REGS_LAZY_LOAD_SET	(1 << VCPU_EXREG_PDPTR)
488 
489 static inline void __vmcb_set_intercept(unsigned long *intercepts, u32 bit)
490 {
491 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
492 	__set_bit(bit, intercepts);
493 }
494 
495 static inline void __vmcb_clr_intercept(unsigned long *intercepts, u32 bit)
496 {
497 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
498 	__clear_bit(bit, intercepts);
499 }
500 
501 static inline bool __vmcb_is_intercept(unsigned long *intercepts, u32 bit)
502 {
503 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
504 	return test_bit(bit, intercepts);
505 }
506 
507 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
508 {
509 	__vmcb_set_intercept((unsigned long *)&control->intercepts, bit);
510 }
511 
512 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
513 {
514 	__vmcb_clr_intercept((unsigned long *)&control->intercepts, bit);
515 }
516 
517 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
518 {
519 	return __vmcb_is_intercept((unsigned long *)&control->intercepts, bit);
520 }
521 
522 static inline void vmcb12_clr_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
523 {
524 	__vmcb_clr_intercept((unsigned long *)&control->intercepts, bit);
525 }
526 
527 static inline bool vmcb12_is_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
528 {
529 	return __vmcb_is_intercept((unsigned long *)&control->intercepts, bit);
530 }
531 
532 void nested_vmcb02_recalc_intercepts(struct vcpu_svm *svm);
533 
534 static inline void svm_mark_intercepts_dirty(struct vcpu_svm *svm)
535 {
536 	vmcb_mark_dirty(svm->vmcb01.ptr, VMCB_INTERCEPTS);
537 
538 	/*
539 	 * If L2 is active, recalculate the intercepts for vmcb02 to account
540 	 * for the changes made to vmcb01.  All intercept configuration is done
541 	 * for vmcb01 and then propagated to vmcb02 to combine KVM's intercepts
542 	 * with L1's intercepts (from the vmcb12 snapshot).
543 	 */
544 	if (is_guest_mode(&svm->vcpu))
545 		nested_vmcb02_recalc_intercepts(svm);
546 }
547 
548 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
549 {
550 	struct vmcb *vmcb = svm->vmcb01.ptr;
551 
552 	WARN_ON_ONCE(bit >= 32);
553 	vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
554 
555 	svm_mark_intercepts_dirty(svm);
556 }
557 
558 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
559 {
560 	struct vmcb *vmcb = svm->vmcb01.ptr;
561 
562 	WARN_ON_ONCE(bit >= 32);
563 	vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
564 
565 	svm_mark_intercepts_dirty(svm);
566 }
567 
568 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
569 {
570 	struct vmcb *vmcb = svm->vmcb01.ptr;
571 
572 	vmcb_set_intercept(&vmcb->control, bit);
573 
574 	svm_mark_intercepts_dirty(svm);
575 }
576 
577 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
578 {
579 	struct vmcb *vmcb = svm->vmcb01.ptr;
580 
581 	vmcb_clr_intercept(&vmcb->control, bit);
582 
583 	svm_mark_intercepts_dirty(svm);
584 }
585 
586 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
587 {
588 	return vmcb_is_intercept(&svm->vmcb->control, bit);
589 }
590 
591 static inline bool nested_vgif_enabled(struct vcpu_svm *svm)
592 {
593 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VGIF) &&
594 	       (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK);
595 }
596 
597 static inline struct vmcb *get_vgif_vmcb(struct vcpu_svm *svm)
598 {
599 	if (!vgif)
600 		return NULL;
601 
602 	if (is_guest_mode(&svm->vcpu) && !nested_vgif_enabled(svm))
603 		return svm->nested.vmcb02.ptr;
604 	else
605 		return svm->vmcb01.ptr;
606 }
607 
608 static inline void enable_gif(struct vcpu_svm *svm)
609 {
610 	struct vmcb *vmcb = get_vgif_vmcb(svm);
611 
612 	if (vmcb)
613 		vmcb->control.int_ctl |= V_GIF_MASK;
614 	else
615 		svm->guest_gif = true;
616 }
617 
618 static inline void disable_gif(struct vcpu_svm *svm)
619 {
620 	struct vmcb *vmcb = get_vgif_vmcb(svm);
621 
622 	if (vmcb)
623 		vmcb->control.int_ctl &= ~V_GIF_MASK;
624 	else
625 		svm->guest_gif = false;
626 }
627 
628 static inline bool gif_set(struct vcpu_svm *svm)
629 {
630 	struct vmcb *vmcb = get_vgif_vmcb(svm);
631 
632 	if (vmcb)
633 		return !!(vmcb->control.int_ctl & V_GIF_MASK);
634 	else
635 		return svm->guest_gif;
636 }
637 
638 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
639 {
640 	return svm->nested.ctl.misc_ctl & SVM_MISC_ENABLE_NP;
641 }
642 
643 static inline bool nested_vnmi_enabled(struct vcpu_svm *svm)
644 {
645 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VNMI) &&
646 	       (svm->nested.ctl.int_ctl & V_NMI_ENABLE_MASK);
647 }
648 
649 static inline bool is_x2apic_msrpm_offset(u32 offset)
650 {
651 	/* 4 msrs per u8, and 4 u8 in u32 */
652 	u32 msr = offset * 16;
653 
654 	return (msr >= APIC_BASE_MSR) &&
655 	       (msr < (APIC_BASE_MSR + 0x100));
656 }
657 
658 static inline struct vmcb *get_vnmi_vmcb_l1(struct vcpu_svm *svm)
659 {
660 	if (!vnmi)
661 		return NULL;
662 
663 	if (is_guest_mode(&svm->vcpu))
664 		return NULL;
665 	else
666 		return svm->vmcb01.ptr;
667 }
668 
669 static inline bool is_vnmi_enabled(struct vcpu_svm *svm)
670 {
671 	struct vmcb *vmcb = get_vnmi_vmcb_l1(svm);
672 
673 	if (vmcb)
674 		return !!(vmcb->control.int_ctl & V_NMI_ENABLE_MASK);
675 	else
676 		return false;
677 }
678 
679 static inline void svm_vmgexit_set_return_code(struct vcpu_svm *svm,
680 						u64 response, u64 data)
681 {
682 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, response);
683 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, data);
684 }
685 
686 static inline void svm_vmgexit_inject_exception(struct vcpu_svm *svm, u8 vector)
687 {
688 	u64 data = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT | vector;
689 
690 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_ISSUE_EXCEPTION, data);
691 }
692 
693 static inline void svm_vmgexit_bad_input(struct vcpu_svm *svm, u64 suberror)
694 {
695 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_MALFORMED_INPUT, suberror);
696 }
697 
698 static inline void svm_vmgexit_success(struct vcpu_svm *svm, u64 data)
699 {
700 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
701 }
702 
703 static inline void svm_vmgexit_no_action(struct vcpu_svm *svm, u64 data)
704 {
705 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
706 }
707 
708 /*
709  * The MSRPM is 8KiB in size, divided into four 2KiB ranges (the fourth range
710  * is reserved).  Each MSR within a range is covered by two bits, one each for
711  * read (bit 0) and write (bit 1), where a bit value of '1' means intercepted.
712  */
713 #define SVM_MSRPM_BYTES_PER_RANGE 2048
714 #define SVM_BITS_PER_MSR 2
715 #define SVM_MSRS_PER_BYTE (BITS_PER_BYTE / SVM_BITS_PER_MSR)
716 #define SVM_MSRS_PER_RANGE (SVM_MSRPM_BYTES_PER_RANGE * SVM_MSRS_PER_BYTE)
717 static_assert(SVM_MSRS_PER_RANGE == 8192);
718 #define SVM_MSRPM_OFFSET_MASK (SVM_MSRS_PER_RANGE - 1)
719 
720 static __always_inline int svm_msrpm_bit_nr(u32 msr)
721 {
722 	int range_nr;
723 
724 	switch (msr & ~SVM_MSRPM_OFFSET_MASK) {
725 	case 0:
726 		range_nr = 0;
727 		break;
728 	case 0xc0000000:
729 		range_nr = 1;
730 		break;
731 	case 0xc0010000:
732 		range_nr = 2;
733 		break;
734 	default:
735 		return -EINVAL;
736 	}
737 
738 	return range_nr * SVM_MSRPM_BYTES_PER_RANGE * BITS_PER_BYTE +
739 	       (msr & SVM_MSRPM_OFFSET_MASK) * SVM_BITS_PER_MSR;
740 }
741 
742 #define __BUILD_SVM_MSR_BITMAP_HELPER(rtype, action, bitop, access, bit_rw)	\
743 static inline rtype svm_##action##_msr_bitmap_##access(unsigned long *bitmap,	\
744 						       u32 msr)			\
745 {										\
746 	int bit_nr;								\
747 										\
748 	bit_nr = svm_msrpm_bit_nr(msr);						\
749 	if (bit_nr < 0)								\
750 		return (rtype)true;						\
751 										\
752 	return bitop##_bit(bit_nr + bit_rw, bitmap);				\
753 }
754 
755 #define BUILD_SVM_MSR_BITMAP_HELPERS(ret_type, action, bitop)			\
756 	__BUILD_SVM_MSR_BITMAP_HELPER(ret_type, action, bitop, read,  0)	\
757 	__BUILD_SVM_MSR_BITMAP_HELPER(ret_type, action, bitop, write, 1)
758 
759 BUILD_SVM_MSR_BITMAP_HELPERS(bool, test, test)
760 BUILD_SVM_MSR_BITMAP_HELPERS(void, clear, __clear)
761 BUILD_SVM_MSR_BITMAP_HELPERS(void, set, __set)
762 
763 #define DEBUGCTL_RESERVED_BITS (~DEBUGCTLMSR_LBR)
764 
765 /* svm.c */
766 extern bool dump_invalid_vmcb;
767 
768 void *svm_alloc_permissions_map(unsigned long size, gfp_t gfp_mask);
769 
770 static inline void *svm_vcpu_alloc_msrpm(void)
771 {
772 	return svm_alloc_permissions_map(MSRPM_SIZE, GFP_KERNEL_ACCOUNT);
773 }
774 
775 #define svm_copy_lbrs(to, from)					\
776 do {								\
777 	(to)->dbgctl		= (from)->dbgctl;		\
778 	(to)->br_from		= (from)->br_from;		\
779 	(to)->br_to		= (from)->br_to;		\
780 	(to)->last_excp_from	= (from)->last_excp_from;	\
781 	(to)->last_excp_to	= (from)->last_excp_to;		\
782 } while (0)
783 
784 void svm_vcpu_free_msrpm(void *msrpm);
785 void svm_enable_lbrv(struct kvm_vcpu *vcpu);
786 void svm_update_lbrv(struct kvm_vcpu *vcpu);
787 
788 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
789 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
790 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
791 void disable_nmi_singlestep(struct vcpu_svm *svm);
792 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
793 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
794 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
795 void svm_set_gif(struct vcpu_svm *svm, bool value);
796 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
797 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
798 			  int read, int write);
799 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
800 				     int trig_mode, int vec);
801 
802 void svm_set_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type, bool set);
803 
804 static inline void svm_disable_intercept_for_msr(struct kvm_vcpu *vcpu,
805 						 u32 msr, int type)
806 {
807 	svm_set_intercept_for_msr(vcpu, msr, type, false);
808 }
809 
810 static inline void svm_enable_intercept_for_msr(struct kvm_vcpu *vcpu,
811 						u32 msr, int type)
812 {
813 	svm_set_intercept_for_msr(vcpu, msr, type, true);
814 }
815 
816 /* nested.c */
817 
818 #define NESTED_EXIT_HOST	0	/* Exit handled on host level */
819 #define NESTED_EXIT_DONE	1	/* Exit caused nested vmexit  */
820 #define NESTED_EXIT_CONTINUE	2	/* Further checks needed      */
821 
822 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
823 {
824 	struct vcpu_svm *svm = to_svm(vcpu);
825 
826 	return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
827 }
828 
829 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
830 {
831 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
832 }
833 
834 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
835 {
836 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
837 }
838 
839 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
840 {
841 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
842 }
843 
844 int __init nested_svm_init_msrpm_merge_offsets(void);
845 
846 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, bool from_vmrun);
847 void svm_leave_nested(struct kvm_vcpu *vcpu);
848 void svm_free_nested(struct vcpu_svm *svm);
849 int svm_allocate_nested(struct vcpu_svm *svm);
850 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
851 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
852 			  struct vmcb_save_area *from_save);
853 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
854 void nested_svm_vmexit(struct vcpu_svm *svm);
855 
856 static inline void nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
857 {
858 	svm->vmcb->control.exit_code	= exit_code;
859 	svm->vmcb->control.exit_info_1	= 0;
860 	svm->vmcb->control.exit_info_2	= 0;
861 	nested_svm_vmexit(svm);
862 }
863 
864 int nested_svm_exit_handled(struct vcpu_svm *svm);
865 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
866 int nested_svm_check_cached_vmcb12(struct kvm_vcpu *vcpu);
867 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
868 			       bool has_error_code, u32 error_code);
869 int nested_svm_exit_special(struct vcpu_svm *svm);
870 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
871 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu);
872 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
873 				       struct vmcb_control_area *control);
874 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
875 				    struct vmcb_save_area *save);
876 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
877 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm);
878 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
879 
880 extern struct kvm_x86_nested_ops svm_nested_ops;
881 
882 /* avic.c */
883 #define AVIC_REQUIRED_APICV_INHIBITS			\
884 (							\
885 	BIT(APICV_INHIBIT_REASON_DISABLED) |		\
886 	BIT(APICV_INHIBIT_REASON_ABSENT) |		\
887 	BIT(APICV_INHIBIT_REASON_HYPERV) |		\
888 	BIT(APICV_INHIBIT_REASON_NESTED) |		\
889 	BIT(APICV_INHIBIT_REASON_IRQWIN) |		\
890 	BIT(APICV_INHIBIT_REASON_PIT_REINJ) |		\
891 	BIT(APICV_INHIBIT_REASON_BLOCKIRQ) |		\
892 	BIT(APICV_INHIBIT_REASON_SEV)      |		\
893 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
894 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
895 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED) |	\
896 	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED) |	\
897 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_TOO_BIG)	\
898 )
899 
900 bool __init avic_hardware_setup(void);
901 void avic_hardware_unsetup(void);
902 int avic_alloc_physical_id_table(struct kvm *kvm);
903 void avic_vm_destroy(struct kvm *kvm);
904 int avic_vm_init(struct kvm *kvm);
905 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb);
906 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
907 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
908 int avic_init_vcpu(struct vcpu_svm *svm);
909 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
910 void avic_vcpu_put(struct kvm_vcpu *vcpu);
911 void avic_apicv_post_state_restore(struct kvm_vcpu *vcpu);
912 void avic_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
913 int avic_pi_update_irte(struct kvm_kernel_irqfd *irqfd, struct kvm *kvm,
914 			unsigned int host_irq, uint32_t guest_irq,
915 			struct kvm_vcpu *vcpu, u32 vector);
916 void avic_vcpu_blocking(struct kvm_vcpu *vcpu);
917 void avic_vcpu_unblocking(struct kvm_vcpu *vcpu);
918 void avic_ring_doorbell(struct kvm_vcpu *vcpu);
919 unsigned long avic_vcpu_get_apicv_inhibit_reasons(struct kvm_vcpu *vcpu);
920 void avic_refresh_virtual_apic_mode(struct kvm_vcpu *vcpu);
921 
922 
923 /* sev.c */
924 
925 int pre_sev_run(struct vcpu_svm *svm, int cpu);
926 void sev_init_vmcb(struct vcpu_svm *svm, bool init_event);
927 void sev_vcpu_after_set_cpuid(struct vcpu_svm *svm);
928 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
929 void sev_es_recalc_msr_intercepts(struct kvm_vcpu *vcpu);
930 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
931 void sev_es_prepare_switch_to_guest(struct vcpu_svm *svm, struct sev_es_save_area *hostsa);
932 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
933 
934 #ifdef CONFIG_KVM_AMD_SEV
935 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp);
936 int sev_mem_enc_register_region(struct kvm *kvm,
937 				struct kvm_enc_region *range);
938 int sev_mem_enc_unregister_region(struct kvm *kvm,
939 				  struct kvm_enc_region *range);
940 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd);
941 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd);
942 void sev_guest_memory_reclaimed(struct kvm *kvm);
943 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
944 
945 /* These symbols are used in common code and are stubbed below.  */
946 
947 struct page *snp_safe_alloc_page_node(int node, gfp_t gfp);
948 static inline struct page *snp_safe_alloc_page(void)
949 {
950 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
951 }
952 
953 int sev_vcpu_create(struct kvm_vcpu *vcpu);
954 void sev_free_vcpu(struct kvm_vcpu *vcpu);
955 void sev_vm_init(struct kvm *kvm);
956 void sev_vm_destroy(struct kvm *kvm);
957 void __init sev_set_cpu_caps(void);
958 void __init sev_hardware_setup(void);
959 void sev_hardware_unsetup(void);
960 int sev_cpu_init(struct svm_cpu_data *sd);
961 int sev_dev_get_attr(u32 group, u64 attr, u64 *val);
962 extern unsigned int max_sev_asid;
963 void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code);
964 int sev_gmem_prepare(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order);
965 void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
966 int sev_gmem_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn, bool is_private);
967 struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu);
968 void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa);
969 #else
970 static inline struct page *snp_safe_alloc_page_node(int node, gfp_t gfp)
971 {
972 	return alloc_pages_node(node, gfp | __GFP_ZERO, 0);
973 }
974 
975 static inline struct page *snp_safe_alloc_page(void)
976 {
977 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
978 }
979 
980 static inline int sev_vcpu_create(struct kvm_vcpu *vcpu) { return 0; }
981 static inline void sev_free_vcpu(struct kvm_vcpu *vcpu) {}
982 static inline void sev_vm_init(struct kvm *kvm) {}
983 static inline void sev_vm_destroy(struct kvm *kvm) {}
984 static inline void __init sev_set_cpu_caps(void) {}
985 static inline void __init sev_hardware_setup(void) {}
986 static inline void sev_hardware_unsetup(void) {}
987 static inline int sev_cpu_init(struct svm_cpu_data *sd) { return 0; }
988 static inline int sev_dev_get_attr(u32 group, u64 attr, u64 *val) { return -ENXIO; }
989 #define max_sev_asid 0
990 static inline void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code) {}
991 static inline int sev_gmem_prepare(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order)
992 {
993 	return 0;
994 }
995 static inline void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end) {}
996 static inline int sev_gmem_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn, bool is_private)
997 {
998 	return 0;
999 }
1000 
1001 static inline struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu)
1002 {
1003 	return NULL;
1004 }
1005 static inline void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa) {}
1006 #endif
1007 
1008 /* vmenter.S */
1009 
1010 void __svm_sev_es_vcpu_run(struct vcpu_svm *svm, bool spec_ctrl_intercepted,
1011 			   struct sev_es_save_area *hostsa);
1012 void __svm_vcpu_run(struct vcpu_svm *svm, bool spec_ctrl_intercepted);
1013 
1014 #define DEFINE_KVM_GHCB_ACCESSORS(field)						\
1015 static __always_inline u64 kvm_ghcb_get_##field(struct vcpu_svm *svm)			\
1016 {											\
1017 	return READ_ONCE(svm->sev_es.ghcb->save.field);					\
1018 }											\
1019 											\
1020 static __always_inline bool kvm_ghcb_##field##_is_valid(const struct vcpu_svm *svm)	\
1021 {											\
1022 	return test_bit(GHCB_BITMAP_IDX(field),						\
1023 			(unsigned long *)&svm->sev_es.valid_bitmap);			\
1024 }											\
1025 											\
1026 static __always_inline u64 kvm_ghcb_get_##field##_if_valid(struct vcpu_svm *svm)	\
1027 {											\
1028 	return kvm_ghcb_##field##_is_valid(svm) ? kvm_ghcb_get_##field(svm) : 0;	\
1029 }
1030 
1031 DEFINE_KVM_GHCB_ACCESSORS(cpl)
1032 DEFINE_KVM_GHCB_ACCESSORS(rax)
1033 DEFINE_KVM_GHCB_ACCESSORS(rcx)
1034 DEFINE_KVM_GHCB_ACCESSORS(rdx)
1035 DEFINE_KVM_GHCB_ACCESSORS(rbx)
1036 DEFINE_KVM_GHCB_ACCESSORS(rsi)
1037 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_code)
1038 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_1)
1039 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_2)
1040 DEFINE_KVM_GHCB_ACCESSORS(sw_scratch)
1041 DEFINE_KVM_GHCB_ACCESSORS(xcr0)
1042 DEFINE_KVM_GHCB_ACCESSORS(xss)
1043 
1044 #endif
1045