xref: /linux/arch/x86/kvm/svm/svm.h (revision 3a2c4d55e32ad65efebdb6de44eef3bfa08bb49d)
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 "regs.h"
27 #include "x86.h"
28 #include "pmu.h"
29 
30 /*
31  * Helpers to convert to/from physical addresses for pages whose address is
32  * consumed directly by hardware.  Even though it's a physical address, SVM
33  * often restricts the address to the natural width, hence 'unsigned long'
34  * instead of 'hpa_t'.
35  */
36 static inline unsigned long __sme_page_pa(struct page *page)
37 {
38 	return __sme_set(page_to_pfn(page) << PAGE_SHIFT);
39 }
40 
41 static inline struct page *__sme_pa_to_page(unsigned long pa)
42 {
43 	return pfn_to_page(__sme_clr(pa) >> PAGE_SHIFT);
44 }
45 
46 #define	IOPM_SIZE PAGE_SIZE * 3
47 #define	MSRPM_SIZE PAGE_SIZE * 2
48 
49 extern bool gmet_enabled;
50 extern bool npt_enabled;
51 extern int nrips;
52 extern int vgif;
53 extern bool intercept_smi;
54 extern bool vnmi;
55 extern int lbrv;
56 
57 extern int tsc_aux_uret_slot __ro_after_init;
58 
59 extern struct kvm_x86_ops svm_x86_ops __initdata;
60 
61 /*
62  * Clean bits in VMCB.
63  * VMCB_ALL_CLEAN_MASK might also need to
64  * be updated if this enum is modified.
65  */
66 enum {
67 	VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
68 			    pause filter count */
69 	VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
70 	VMCB_ASID,	 /* ASID */
71 	VMCB_INTR,	 /* int_ctl, int_vector */
72 	VMCB_NPT,        /* npt_en, nCR3, gPAT */
73 	VMCB_CR,	 /* CR0, CR3, CR4, EFER */
74 	VMCB_DR,         /* DR6, DR7 */
75 	VMCB_DT,         /* GDT, IDT */
76 	VMCB_SEG,        /* CS, DS, SS, ES, CPL */
77 	VMCB_CR2,        /* CR2 only */
78 	VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
79 	VMCB_AVIC,       /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
80 			  * AVIC PHYSICAL_TABLE pointer,
81 			  * AVIC LOGICAL_TABLE pointer
82 			  */
83 	VMCB_CET,	 /* S_CET, SSP, ISST_ADDR */
84 	VMCB_SW = 31,    /* Reserved for hypervisor/software use */
85 };
86 
87 #define VMCB_ALL_CLEAN_MASK (					\
88 	(1U << VMCB_INTERCEPTS) | (1U << VMCB_PERM_MAP) |	\
89 	(1U << VMCB_ASID) | (1U << VMCB_INTR) |			\
90 	(1U << VMCB_NPT) | (1U << VMCB_CR) | (1U << VMCB_DR) |	\
91 	(1U << VMCB_DT) | (1U << VMCB_SEG) | (1U << VMCB_CR2) |	\
92 	(1U << VMCB_LBR) | (1U << VMCB_AVIC) | (1U << VMCB_CET) | \
93 	(1U << VMCB_SW))
94 
95 /* TPR and CR2 are always written before VMRUN */
96 #define VMCB_ALWAYS_DIRTY_MASK	((1U << VMCB_INTR) | (1U << VMCB_CR2))
97 
98 #ifdef CONFIG_KVM_AMD_SEV
99 struct kvm_sev_info {
100 	bool active;		/* SEV enabled guest */
101 	bool es_active;		/* SEV-ES enabled guest */
102 	bool need_init;		/* waiting for SEV_INIT2 */
103 	unsigned int asid;	/* ASID used for this guest */
104 	unsigned int handle;	/* SEV firmware handle */
105 	int fd;			/* SEV device fd */
106 	unsigned long policy;
107 	unsigned long pages_locked; /* Number of pages locked */
108 	struct list_head regions_list;  /* List of registered regions */
109 	u64 ap_jump_table;	/* SEV-ES AP Jump Table address */
110 	u64 vmsa_features;
111 	u16 ghcb_version;	/* Highest guest GHCB protocol version allowed */
112 	/* The three fields below are protected by sev_mirror_lock */
113 	struct kvm *enc_context_owner; /* Owner of copied encryption context */
114 	struct list_head mirror_vms; /* List of VMs mirroring */
115 	struct list_head mirror_entry; /* Use as a list entry of mirrors */
116 	struct misc_cg *misc_cg; /* For misc cgroup accounting */
117 	atomic_t migration_in_progress;
118 	void *snp_context;      /* SNP guest context page */
119 	void *guest_req_buf;    /* Bounce buffer for SNP Guest Request input */
120 	void *guest_resp_buf;   /* Bounce buffer for SNP Guest Request output */
121 	struct mutex guest_req_mutex; /* Must acquire before using bounce buffers */
122 	cpumask_var_t have_run_cpus; /* CPUs that have done VMRUN for this VM. */
123 	bool snp_certs_enabled;	/* SNP certificate-fetching support. */
124 };
125 #endif
126 
127 struct kvm_svm {
128 	struct kvm kvm;
129 
130 	/* Struct members for AVIC */
131 	u32 avic_vm_id;
132 	u32 *avic_logical_id_table;
133 	u64 *avic_physical_id_table;
134 	struct hlist_node hnode;
135 
136 #ifdef CONFIG_KVM_AMD_SEV
137 	struct kvm_sev_info sev_info;
138 #endif
139 };
140 
141 struct kvm_vcpu;
142 
143 struct kvm_vmcb_info {
144 	struct vmcb *ptr;
145 	unsigned long pa;
146 	int cpu;
147 	uint64_t asid_generation;
148 };
149 
150 struct vmcb_save_area_cached {
151 	struct vmcb_seg es;
152 	struct vmcb_seg cs;
153 	struct vmcb_seg ss;
154 	struct vmcb_seg ds;
155 	struct vmcb_seg gdtr;
156 	struct vmcb_seg idtr;
157 	u8 cpl;
158 	u64 efer;
159 	u64 cr4;
160 	u64 cr3;
161 	u64 cr0;
162 	u64 dr7;
163 	u64 dr6;
164 	u64 rflags;
165 	u64 rip;
166 	u64 rsp;
167 	u64 s_cet;
168 	u64 ssp;
169 	u64 isst_addr;
170 	u64 rax;
171 	u64 cr2;
172 	u64 g_pat;
173 	u64 dbgctl;
174 	u64 br_from;
175 	u64 br_to;
176 	u64 last_excp_from;
177 	u64 last_excp_to;
178 };
179 
180 struct vmcb_ctrl_area_cached {
181 	u32 intercepts[MAX_INTERCEPT];
182 	u16 pause_filter_thresh;
183 	u16 pause_filter_count;
184 	u64 iopm_base_pa;
185 	u64 msrpm_base_pa;
186 	u64 tsc_offset;
187 	u32 asid;
188 	u8 tlb_ctl;
189 	u8 erap_ctl;
190 	u32 int_ctl;
191 	u32 int_vector;
192 	u32 int_state;
193 	u64 exit_code;
194 	u64 exit_info_1;
195 	u64 exit_info_2;
196 	u32 exit_int_info;
197 	u32 exit_int_info_err;
198 	u64 misc_ctl;
199 	u32 event_inj;
200 	u32 event_inj_err;
201 	u64 next_rip;
202 	u64 nested_cr3;
203 	u64 misc_ctl2;
204 	u32 clean;
205 	union {
206 #if IS_ENABLED(CONFIG_HYPERV) || IS_ENABLED(CONFIG_KVM_HYPERV)
207 		struct hv_vmcb_enlightenments hv_enlightenments;
208 #endif
209 		u8 reserved_sw[32];
210 	};
211 };
212 
213 struct svm_nested_state {
214 	struct kvm_vmcb_info vmcb02;
215 	u64 hsave_msr;
216 	u64 vm_cr_msr;
217 	u64 vmcb12_gpa;
218 	u64 last_vmcb12_gpa;
219 	u64 last_bus_lock_rip;
220 
221 	/*
222 	 * The MSR permissions map used for vmcb02, which is the merge result
223 	 * of vmcb01 and vmcb12
224 	 */
225 	void *msrpm;
226 
227 	/* cache for control fields of the guest */
228 	struct vmcb_ctrl_area_cached ctl;
229 
230 	/*
231 	 * Note: this struct is not kept up-to-date while L2 runs; it is only
232 	 * valid within nested_svm_vmrun.
233 	 */
234 	struct vmcb_save_area_cached save;
235 
236 	bool initialized;
237 
238 	/*
239 	 * Indicates whether MSR bitmap for L2 needs to be rebuilt due to
240 	 * changes in MSR bitmap for L1 or switching to a different L2. Note,
241 	 * this flag can only be used reliably in conjunction with a paravirt L1
242 	 * which informs L0 whether any changes to MSR bitmap for L2 were done
243 	 * on its side.
244 	 */
245 	bool force_msr_bitmap_recalc;
246 };
247 
248 struct vcpu_sev_es_state {
249 	/* SEV-ES support */
250 	struct sev_es_save_area *vmsa;
251 	struct ghcb *ghcb;
252 	u8 valid_bitmap[16];
253 	struct kvm_host_map ghcb_map;
254 	bool received_first_sipi;
255 	unsigned int ap_reset_hold_type;
256 
257 	/* SEV-ES scratch area support */
258 	u64 sw_scratch;
259 	void *ghcb_sa;
260 	u32 ghcb_sa_len;
261 	bool ghcb_sa_sync;
262 	bool ghcb_sa_free;
263 
264 	/* SNP Page-State-Change buffer entries currently being processed */
265 	struct {
266 		u16 cur_idx;
267 		u16 end_idx;
268 		u16 batch_size;
269 		bool is_2m;
270 	} psc;
271 
272 	u64 ghcb_registered_gpa;
273 
274 	struct mutex snp_vmsa_mutex; /* Used to handle concurrent updates of VMSA. */
275 	gpa_t snp_pending_vmsa_gpa;
276 	gpa_t snp_guest_vmsa_gpa;
277 	bool snp_ap_waiting_for_reset;
278 	bool snp_has_guest_vmsa;
279 };
280 
281 struct vcpu_svm {
282 	struct kvm_vcpu vcpu;
283 	/* vmcb always points at current_vmcb->ptr, it's purely a shorthand. */
284 	struct vmcb *vmcb;
285 	struct kvm_vmcb_info vmcb01;
286 	struct kvm_vmcb_info *current_vmcb;
287 	u32 asid;
288 	u32 sysenter_esp_hi;
289 	u32 sysenter_eip_hi;
290 	uint64_t tsc_aux;
291 
292 	u64 msr_decfg;
293 
294 	u64 next_rip;
295 
296 	u64 spec_ctrl;
297 
298 	u64 tsc_ratio_msr;
299 	/*
300 	 * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
301 	 * translated into the appropriate L2_CFG bits on the host to
302 	 * perform speculative control.
303 	 */
304 	u64 virt_spec_ctrl;
305 
306 	void *msrpm;
307 
308 	ulong nmi_iret_rip;
309 
310 	struct svm_nested_state nested;
311 
312 	/* NMI mask value, used when vNMI is not enabled */
313 	bool nmi_masked;
314 
315 	/*
316 	 * True when NMIs are still masked but guest IRET was just intercepted
317 	 * and KVM is waiting for RIP to change, which will signal that the
318 	 * intercepted IRET was retired and thus NMI can be unmasked.
319 	 */
320 	bool awaiting_iret_completion;
321 
322 	/*
323 	 * Set when KVM is awaiting IRET completion and needs to inject NMIs as
324 	 * soon as the IRET completes (e.g. NMI is pending injection).  KVM
325 	 * temporarily steals RFLAGS.TF to single-step the guest in this case
326 	 * in order to regain control as soon as the NMI-blocking condition
327 	 * goes away.
328 	 */
329 	bool nmi_singlestep;
330 	u64 nmi_singlestep_guest_rflags;
331 
332 	bool nmi_l1_to_l2;
333 
334 	unsigned long soft_int_csbase;
335 	unsigned long soft_int_old_rip;
336 	unsigned long soft_int_next_rip;
337 	bool soft_int_injected;
338 
339 	u32 ldr_reg;
340 	u32 dfr_reg;
341 
342 	/* This is essentially a shadow of the vCPU's actual entry in the
343 	 * Physical ID table that is programmed into the VMCB, i.e. that is
344 	 * seen by the CPU.  If IPI virtualization is disabled, IsRunning is
345 	 * only ever set in the shadow, i.e. is never propagated to the "real"
346 	 * table, so that hardware never sees IsRunning=1.
347 	 */
348 	u64 avic_physical_id_entry;
349 
350 	/*
351 	 * Per-vCPU list of irqfds that are eligible to post IRQs directly to
352 	 * the vCPU (a.k.a. device posted IRQs, a.k.a. IRQ bypass).  The list
353 	 * is used to reconfigure IRTEs when the vCPU is loaded/put (to set the
354 	 * target pCPU), when AVIC is toggled on/off (to (de)activate bypass),
355 	 * and if the irqfd becomes ineligible for posting (to put the IRTE
356 	 * back into remapped mode).
357 	 */
358 	struct list_head ir_list;
359 	raw_spinlock_t ir_list_lock;
360 
361 	struct vcpu_sev_es_state sev_es;
362 
363 	bool guest_state_loaded;
364 
365 	bool avic_irq_window;
366 	bool x2avic_msrs_intercepted;
367 	bool lbr_msrs_intercepted;
368 
369 	/* Guest GIF value, used when vGIF is not enabled */
370 	bool guest_gif;
371 };
372 
373 struct svm_cpu_data {
374 	u64 asid_generation;
375 	u32 max_asid;
376 	u32 next_asid;
377 	u32 min_asid;
378 
379 	bool bp_spec_reduce_set;
380 
381 	struct vmcb *save_area;
382 	unsigned long save_area_pa;
383 
384 	/* index = sev_asid, value = vmcb pointer */
385 	struct vmcb **sev_vmcbs;
386 };
387 
388 DECLARE_PER_CPU(struct svm_cpu_data, svm_data);
389 
390 static __always_inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
391 {
392 	return container_of(kvm, struct kvm_svm, kvm);
393 }
394 
395 #ifdef CONFIG_KVM_AMD_SEV
396 static __always_inline struct kvm_sev_info *to_kvm_sev_info(struct kvm *kvm)
397 {
398 	return &to_kvm_svm(kvm)->sev_info;
399 }
400 
401 static __always_inline bool ____sev_guest(struct kvm *kvm)
402 {
403 	return to_kvm_sev_info(kvm)->active;
404 }
405 static __always_inline bool ____sev_es_guest(struct kvm *kvm)
406 {
407 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
408 
409 	return sev->es_active && !WARN_ON_ONCE(!sev->active);
410 }
411 
412 static __always_inline bool ____sev_snp_guest(struct kvm *kvm)
413 {
414 	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
415 
416 	return (sev->vmsa_features & SVM_SEV_FEAT_SNP_ACTIVE) &&
417 	       !WARN_ON_ONCE(!____sev_es_guest(kvm));
418 }
419 
420 static __always_inline bool is_sev_guest(struct kvm_vcpu *vcpu)
421 {
422 	return ____sev_guest(vcpu->kvm);
423 }
424 static __always_inline bool is_sev_es_guest(struct kvm_vcpu *vcpu)
425 {
426 	return ____sev_es_guest(vcpu->kvm);
427 }
428 
429 static __always_inline bool is_sev_snp_guest(struct kvm_vcpu *vcpu)
430 {
431 	return ____sev_snp_guest(vcpu->kvm);
432 }
433 #else
434 static __always_inline bool is_sev_guest(struct kvm_vcpu *vcpu)
435 {
436 	return false;
437 }
438 static __always_inline bool is_sev_es_guest(struct kvm_vcpu *vcpu)
439 {
440 	return false;
441 }
442 
443 static __always_inline bool is_sev_snp_guest(struct kvm_vcpu *vcpu)
444 {
445 	return false;
446 }
447 #endif
448 
449 static inline bool ghcb_gpa_is_registered(struct vcpu_svm *svm, u64 val)
450 {
451 	return svm->sev_es.ghcb_registered_gpa == val;
452 }
453 
454 static inline void vmcb_mark_all_dirty(struct vmcb *vmcb)
455 {
456 	vmcb->control.clean = 0;
457 }
458 
459 static inline void vmcb_mark_all_clean(struct vmcb *vmcb)
460 {
461 	vmcb->control.clean = VMCB_ALL_CLEAN_MASK
462 			       & ~VMCB_ALWAYS_DIRTY_MASK;
463 }
464 
465 static inline void vmcb_mark_dirty(struct vmcb *vmcb, int bit)
466 {
467 	vmcb->control.clean &= ~(1 << bit);
468 }
469 
470 static inline bool vmcb12_is_dirty(struct vmcb_ctrl_area_cached *control, int bit)
471 {
472 	return !test_bit(bit, (unsigned long *)&control->clean);
473 }
474 
475 static inline void vmcb_set_gpat(struct vmcb *vmcb, u64 data)
476 {
477 	vmcb->save.g_pat = data;
478 	vmcb_mark_dirty(vmcb, VMCB_NPT);
479 }
480 
481 static __always_inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
482 {
483 	return container_of(vcpu, struct vcpu_svm, vcpu);
484 }
485 
486 static inline bool svm_is_vmrun_failure(u64 exit_code)
487 {
488 	if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
489 		return (u32)exit_code == (u32)SVM_EXIT_ERR;
490 
491 	return exit_code == SVM_EXIT_ERR;
492 }
493 
494 /*
495  * Only the PDPTRs are loaded on demand into the shadow MMU.  All other
496  * fields are synchronized on VM-Exit, because accessing the VMCB is cheap.
497  *
498  * CR3 might be out of date in the VMCB but it is not marked dirty; instead,
499  * KVM_REQ_LOAD_MMU_PGD is always requested when the cached vcpu->arch.cr3
500  * is changed.  svm_load_mmu_pgd() then syncs the new CR3 value into the VMCB.
501  */
502 #define SVM_REGS_LAZY_LOAD_SET	(BIT(VCPU_REG_PDPTR))
503 
504 static inline void __vmcb_set_intercept(unsigned long *intercepts, u32 bit)
505 {
506 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
507 	__set_bit(bit, intercepts);
508 }
509 
510 static inline void __vmcb_clr_intercept(unsigned long *intercepts, u32 bit)
511 {
512 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
513 	__clear_bit(bit, intercepts);
514 }
515 
516 static inline bool __vmcb_is_intercept(unsigned long *intercepts, u32 bit)
517 {
518 	WARN_ON_ONCE(bit >= 32 * MAX_INTERCEPT);
519 	return test_bit(bit, intercepts);
520 }
521 
522 static inline void vmcb_set_intercept(struct vmcb_control_area *control, u32 bit)
523 {
524 	__vmcb_set_intercept((unsigned long *)&control->intercepts, bit);
525 }
526 
527 static inline void vmcb_clr_intercept(struct vmcb_control_area *control, u32 bit)
528 {
529 	__vmcb_clr_intercept((unsigned long *)&control->intercepts, bit);
530 }
531 
532 static inline bool vmcb_is_intercept(struct vmcb_control_area *control, u32 bit)
533 {
534 	return __vmcb_is_intercept((unsigned long *)&control->intercepts, bit);
535 }
536 
537 static inline void vmcb12_clr_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
538 {
539 	__vmcb_clr_intercept((unsigned long *)&control->intercepts, bit);
540 }
541 
542 static inline bool vmcb12_is_intercept(struct vmcb_ctrl_area_cached *control, u32 bit)
543 {
544 	return __vmcb_is_intercept((unsigned long *)&control->intercepts, bit);
545 }
546 
547 void nested_vmcb02_recalc_intercepts(struct vcpu_svm *svm);
548 
549 static inline void svm_mark_intercepts_dirty(struct vcpu_svm *svm)
550 {
551 	vmcb_mark_dirty(svm->vmcb01.ptr, VMCB_INTERCEPTS);
552 
553 	/*
554 	 * If L2 is active, recalculate the intercepts for vmcb02 to account
555 	 * for the changes made to vmcb01.  All intercept configuration is done
556 	 * for vmcb01 and then propagated to vmcb02 to combine KVM's intercepts
557 	 * with L1's intercepts (from the vmcb12 snapshot).
558 	 */
559 	if (is_guest_mode(&svm->vcpu))
560 		nested_vmcb02_recalc_intercepts(svm);
561 }
562 
563 static inline void set_exception_intercept(struct vcpu_svm *svm, u32 bit)
564 {
565 	struct vmcb *vmcb = svm->vmcb01.ptr;
566 
567 	WARN_ON_ONCE(bit >= 32);
568 	vmcb_set_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
569 
570 	svm_mark_intercepts_dirty(svm);
571 }
572 
573 static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
574 {
575 	struct vmcb *vmcb = svm->vmcb01.ptr;
576 
577 	WARN_ON_ONCE(bit >= 32);
578 	vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
579 
580 	svm_mark_intercepts_dirty(svm);
581 }
582 
583 static inline void svm_set_intercept(struct vcpu_svm *svm, int bit)
584 {
585 	struct vmcb *vmcb = svm->vmcb01.ptr;
586 
587 	vmcb_set_intercept(&vmcb->control, bit);
588 
589 	svm_mark_intercepts_dirty(svm);
590 }
591 
592 static inline void svm_clr_intercept(struct vcpu_svm *svm, int bit)
593 {
594 	struct vmcb *vmcb = svm->vmcb01.ptr;
595 
596 	vmcb_clr_intercept(&vmcb->control, bit);
597 
598 	svm_mark_intercepts_dirty(svm);
599 }
600 
601 static inline bool svm_is_intercept(struct vcpu_svm *svm, int bit)
602 {
603 	return vmcb_is_intercept(&svm->vmcb->control, bit);
604 }
605 
606 static inline bool nested_vgif_enabled(struct vcpu_svm *svm)
607 {
608 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VGIF) &&
609 	       (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK);
610 }
611 
612 static inline struct vmcb *get_vgif_vmcb(struct vcpu_svm *svm)
613 {
614 	if (!vgif)
615 		return NULL;
616 
617 	if (is_guest_mode(&svm->vcpu) && !nested_vgif_enabled(svm))
618 		return svm->nested.vmcb02.ptr;
619 	else
620 		return svm->vmcb01.ptr;
621 }
622 
623 static inline void enable_gif(struct vcpu_svm *svm)
624 {
625 	struct vmcb *vmcb = get_vgif_vmcb(svm);
626 
627 	if (vmcb)
628 		vmcb->control.int_ctl |= V_GIF_MASK;
629 	else
630 		svm->guest_gif = true;
631 }
632 
633 static inline void disable_gif(struct vcpu_svm *svm)
634 {
635 	struct vmcb *vmcb = get_vgif_vmcb(svm);
636 
637 	if (vmcb)
638 		vmcb->control.int_ctl &= ~V_GIF_MASK;
639 	else
640 		svm->guest_gif = false;
641 }
642 
643 static inline bool gif_set(struct vcpu_svm *svm)
644 {
645 	struct vmcb *vmcb = get_vgif_vmcb(svm);
646 
647 	if (vmcb)
648 		return !!(vmcb->control.int_ctl & V_GIF_MASK);
649 	else
650 		return svm->guest_gif;
651 }
652 
653 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
654 {
655 	return svm->nested.ctl.misc_ctl & SVM_MISC_ENABLE_NP;
656 }
657 
658 static inline bool l2_has_separate_pat(struct kvm_vcpu *vcpu)
659 {
660 	/*
661 	 * If KVM_X86_QUIRK_NESTED_SVM_SHARED_PAT is disabled while a vCPU
662 	 * is running, the L2 IA32_PAT semantics for that vCPU are undefined.
663 	 */
664 	return nested_npt_enabled(to_svm(vcpu)) &&
665 	       !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_NESTED_SVM_SHARED_PAT);
666 }
667 
668 static inline bool nested_vnmi_enabled(struct vcpu_svm *svm)
669 {
670 	return guest_cpu_cap_has(&svm->vcpu, X86_FEATURE_VNMI) &&
671 	       (svm->nested.ctl.int_ctl & V_NMI_ENABLE_MASK);
672 }
673 
674 static inline bool is_x2apic_msrpm_offset(u32 offset)
675 {
676 	/* 4 msrs per u8, and 4 u8 in u32 */
677 	u32 msr = offset * 16;
678 
679 	return (msr >= APIC_BASE_MSR) &&
680 	       (msr < (APIC_BASE_MSR + 0x100));
681 }
682 
683 static inline struct vmcb *get_vnmi_vmcb_l1(struct vcpu_svm *svm)
684 {
685 	if (!vnmi)
686 		return NULL;
687 
688 	if (is_guest_mode(&svm->vcpu))
689 		return NULL;
690 	else
691 		return svm->vmcb01.ptr;
692 }
693 
694 static inline bool is_vnmi_enabled(struct vcpu_svm *svm)
695 {
696 	struct vmcb *vmcb = get_vnmi_vmcb_l1(svm);
697 
698 	if (vmcb)
699 		return !!(vmcb->control.int_ctl & V_NMI_ENABLE_MASK);
700 	else
701 		return false;
702 }
703 
704 static inline void svm_vmgexit_set_return_code(struct vcpu_svm *svm,
705 						u64 response, u64 data)
706 {
707 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, response);
708 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, data);
709 }
710 
711 static inline void svm_vmgexit_inject_exception(struct vcpu_svm *svm, u8 vector)
712 {
713 	u64 data = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT | vector;
714 
715 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_ISSUE_EXCEPTION, data);
716 }
717 
718 static inline void svm_vmgexit_bad_input(struct vcpu_svm *svm, u64 suberror)
719 {
720 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_MALFORMED_INPUT, suberror);
721 }
722 
723 static inline void svm_vmgexit_success(struct vcpu_svm *svm, u64 data)
724 {
725 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
726 }
727 
728 static inline void svm_vmgexit_no_action(struct vcpu_svm *svm, u64 data)
729 {
730 	svm_vmgexit_set_return_code(svm, GHCB_HV_RESP_NO_ACTION, data);
731 }
732 
733 /*
734  * The MSRPM is 8KiB in size, divided into four 2KiB ranges (the fourth range
735  * is reserved).  Each MSR within a range is covered by two bits, one each for
736  * read (bit 0) and write (bit 1), where a bit value of '1' means intercepted.
737  */
738 #define SVM_MSRPM_BYTES_PER_RANGE 2048
739 #define SVM_BITS_PER_MSR 2
740 #define SVM_MSRS_PER_BYTE (BITS_PER_BYTE / SVM_BITS_PER_MSR)
741 #define SVM_MSRS_PER_RANGE (SVM_MSRPM_BYTES_PER_RANGE * SVM_MSRS_PER_BYTE)
742 static_assert(SVM_MSRS_PER_RANGE == 8192);
743 #define SVM_MSRPM_OFFSET_MASK (SVM_MSRS_PER_RANGE - 1)
744 
745 static __always_inline int svm_msrpm_bit_nr(u32 msr)
746 {
747 	int range_nr;
748 
749 	switch (msr & ~SVM_MSRPM_OFFSET_MASK) {
750 	case 0:
751 		range_nr = 0;
752 		break;
753 	case 0xc0000000:
754 		range_nr = 1;
755 		break;
756 	case 0xc0010000:
757 		range_nr = 2;
758 		break;
759 	default:
760 		return -EINVAL;
761 	}
762 
763 	return range_nr * SVM_MSRPM_BYTES_PER_RANGE * BITS_PER_BYTE +
764 	       (msr & SVM_MSRPM_OFFSET_MASK) * SVM_BITS_PER_MSR;
765 }
766 
767 #define __BUILD_SVM_MSR_BITMAP_HELPER(rtype, action, bitop, access, bit_rw)	\
768 static inline rtype svm_##action##_msr_bitmap_##access(unsigned long *bitmap,	\
769 						       u32 msr)			\
770 {										\
771 	int bit_nr;								\
772 										\
773 	bit_nr = svm_msrpm_bit_nr(msr);						\
774 	if (bit_nr < 0)								\
775 		return (rtype)true;						\
776 										\
777 	return bitop##_bit(bit_nr + bit_rw, bitmap);				\
778 }
779 
780 #define BUILD_SVM_MSR_BITMAP_HELPERS(ret_type, action, bitop)			\
781 	__BUILD_SVM_MSR_BITMAP_HELPER(ret_type, action, bitop, read,  0)	\
782 	__BUILD_SVM_MSR_BITMAP_HELPER(ret_type, action, bitop, write, 1)
783 
784 BUILD_SVM_MSR_BITMAP_HELPERS(bool, test, test)
785 BUILD_SVM_MSR_BITMAP_HELPERS(void, clear, __clear)
786 BUILD_SVM_MSR_BITMAP_HELPERS(void, set, __set)
787 
788 #define DEBUGCTL_RESERVED_BITS (~DEBUGCTLMSR_LBR)
789 
790 /* svm.c */
791 extern bool dump_invalid_vmcb;
792 
793 void *svm_alloc_permissions_map(unsigned long size, gfp_t gfp_mask);
794 
795 static inline void *svm_vcpu_alloc_msrpm(void)
796 {
797 	return svm_alloc_permissions_map(MSRPM_SIZE, GFP_KERNEL_ACCOUNT);
798 }
799 
800 #define svm_copy_lbrs(to, from)					\
801 do {								\
802 	(to)->dbgctl		= (from)->dbgctl;		\
803 	(to)->br_from		= (from)->br_from;		\
804 	(to)->br_to		= (from)->br_to;		\
805 	(to)->last_excp_from	= (from)->last_excp_from;	\
806 	(to)->last_excp_to	= (from)->last_excp_to;		\
807 } while (0)
808 
809 void svm_vcpu_free_msrpm(void *msrpm);
810 void svm_enable_lbrv(struct kvm_vcpu *vcpu);
811 void svm_update_lbrv(struct kvm_vcpu *vcpu);
812 
813 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer);
814 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
815 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
816 void disable_nmi_singlestep(struct vcpu_svm *svm);
817 bool svm_smi_blocked(struct kvm_vcpu *vcpu);
818 bool svm_nmi_blocked(struct kvm_vcpu *vcpu);
819 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu);
820 void svm_set_gif(struct vcpu_svm *svm, bool value);
821 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code);
822 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
823 			  int read, int write);
824 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
825 				     int trig_mode, int vec);
826 
827 void svm_set_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type, bool set);
828 
829 static inline void svm_disable_intercept_for_msr(struct kvm_vcpu *vcpu,
830 						 u32 msr, int type)
831 {
832 	svm_set_intercept_for_msr(vcpu, msr, type, false);
833 }
834 
835 static inline void svm_enable_intercept_for_msr(struct kvm_vcpu *vcpu,
836 						u32 msr, int type)
837 {
838 	svm_set_intercept_for_msr(vcpu, msr, type, true);
839 }
840 
841 int svm_skip_emulated_instruction(struct kvm_vcpu *vcpu);
842 
843 /* nested.c */
844 
845 #define NESTED_EXIT_HOST	0	/* Exit handled on host level */
846 #define NESTED_EXIT_DONE	1	/* Exit caused nested vmexit  */
847 #define NESTED_EXIT_CONTINUE	2	/* Further checks needed      */
848 
849 static inline bool nested_svm_virtualize_tpr(struct kvm_vcpu *vcpu)
850 {
851 	struct vcpu_svm *svm = to_svm(vcpu);
852 
853 	return is_guest_mode(vcpu) && (svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK);
854 }
855 
856 static inline bool nested_exit_on_smi(struct vcpu_svm *svm)
857 {
858 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SMI);
859 }
860 
861 static inline bool nested_exit_on_intr(struct vcpu_svm *svm)
862 {
863 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INTR);
864 }
865 
866 static inline bool nested_exit_on_nmi(struct vcpu_svm *svm)
867 {
868 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_NMI);
869 }
870 
871 int __init nested_svm_init_msrpm_merge_offsets(void);
872 
873 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb_gpa, bool from_vmrun);
874 void svm_leave_nested(struct kvm_vcpu *vcpu);
875 void svm_free_nested(struct vcpu_svm *svm);
876 int svm_allocate_nested(struct vcpu_svm *svm);
877 int nested_svm_vmrun(struct kvm_vcpu *vcpu);
878 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
879 			  struct vmcb_save_area *from_save);
880 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb);
881 void nested_svm_vmexit(struct vcpu_svm *svm);
882 
883 static inline void nested_svm_simple_vmexit(struct vcpu_svm *svm, u32 exit_code)
884 {
885 	svm->vmcb->control.exit_code	= exit_code;
886 	svm->vmcb->control.exit_info_1	= 0;
887 	svm->vmcb->control.exit_info_2	= 0;
888 	nested_svm_vmexit(svm);
889 }
890 
891 int nested_svm_exit_handled(struct vcpu_svm *svm);
892 int nested_svm_check_permissions(struct kvm_vcpu *vcpu);
893 int nested_svm_check_cached_vmcb12(struct kvm_vcpu *vcpu);
894 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
895 			       bool has_error_code, u32 error_code);
896 int nested_svm_exit_special(struct vcpu_svm *svm);
897 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu);
898 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu);
899 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
900 				       struct vmcb_control_area *control);
901 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
902 				    struct vmcb_save_area *save);
903 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm);
904 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb);
905 
906 
907 static inline void __svm_pmu_handle_nested_transition(struct vcpu_svm *svm,
908 						      bool defer)
909 {
910 	struct kvm_pmu *pmu = vcpu_to_pmu(&svm->vcpu);
911 	u64 counters = *(u64 *)pmu->pmc_has_mode_specific_enables;
912 
913 	__kvm_pmu_reprogram_counters(pmu, counters, defer);
914 }
915 
916 static inline void svm_pmu_handle_nested_transition(struct vcpu_svm *svm)
917 {
918 	/*
919 	 * Do NOT defer reprogramming the counters by default.  Instructions
920 	 * causing a state change are counted based on the _new_ CPU state
921 	 * (e.g. a successful VMRUN is counted in guest mode). Hence, the
922 	 * counters should be reprogrammed with the new state _before_ the
923 	 * instruction is potentially counted upon emulation completion.
924 	 */
925 	__svm_pmu_handle_nested_transition(svm, false);
926 }
927 
928 extern struct kvm_x86_nested_ops svm_nested_ops;
929 
930 /* avic.c */
931 #define AVIC_REQUIRED_APICV_INHIBITS			\
932 (							\
933 	BIT(APICV_INHIBIT_REASON_DISABLED) |		\
934 	BIT(APICV_INHIBIT_REASON_ABSENT) |		\
935 	BIT(APICV_INHIBIT_REASON_HYPERV) |		\
936 	BIT(APICV_INHIBIT_REASON_NESTED) |		\
937 	BIT(APICV_INHIBIT_REASON_IRQWIN) |		\
938 	BIT(APICV_INHIBIT_REASON_PIT_REINJ) |		\
939 	BIT(APICV_INHIBIT_REASON_BLOCKIRQ) |		\
940 	BIT(APICV_INHIBIT_REASON_SEV)      |		\
941 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
942 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
943 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED) |	\
944 	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED) |	\
945 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_TOO_BIG)	\
946 )
947 
948 bool __init avic_hardware_setup(void);
949 void avic_hardware_unsetup(void);
950 int avic_vcpu_precreate(struct kvm *kvm);
951 void avic_vm_pre_destroy(struct kvm *kvm);
952 void avic_vm_destroy(struct kvm *kvm);
953 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb);
954 int avic_incomplete_ipi_interception(struct kvm_vcpu *vcpu);
955 int avic_unaccelerated_access_interception(struct kvm_vcpu *vcpu);
956 int avic_init_vcpu(struct vcpu_svm *svm);
957 void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
958 void avic_vcpu_put(struct kvm_vcpu *vcpu);
959 void avic_apicv_post_state_restore(struct kvm_vcpu *vcpu);
960 void avic_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu);
961 int avic_pi_update_irte(struct kvm_kernel_irqfd *irqfd, struct kvm *kvm,
962 			unsigned int host_irq, uint32_t guest_irq,
963 			struct kvm_vcpu *vcpu, u32 vector);
964 void avic_vcpu_blocking(struct kvm_vcpu *vcpu);
965 void avic_vcpu_unblocking(struct kvm_vcpu *vcpu);
966 void avic_ring_doorbell(struct kvm_vcpu *vcpu);
967 unsigned long avic_vcpu_get_apicv_inhibit_reasons(struct kvm_vcpu *vcpu);
968 void avic_refresh_virtual_apic_mode(struct kvm_vcpu *vcpu);
969 
970 
971 /* sev.c */
972 
973 int pre_sev_run(struct vcpu_svm *svm, int cpu);
974 void sev_init_vmcb(struct vcpu_svm *svm, bool init_event);
975 void sev_vcpu_after_set_cpuid(struct vcpu_svm *svm);
976 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in);
977 void sev_es_recalc_msr_intercepts(struct kvm_vcpu *vcpu);
978 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
979 void sev_es_prepare_switch_to_guest(struct vcpu_svm *svm, struct sev_es_save_area *hostsa);
980 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
981 
982 #ifdef CONFIG_KVM_AMD_SEV
983 bool sev_vcpu_needs_initialization(struct kvm_vcpu *vcpu);
984 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp);
985 int sev_mem_enc_register_region(struct kvm *kvm,
986 				struct kvm_enc_region *range);
987 int sev_mem_enc_unregister_region(struct kvm *kvm,
988 				  struct kvm_enc_region *range);
989 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd);
990 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd);
991 void sev_guest_memory_reclaimed(struct kvm *kvm);
992 int sev_handle_vmgexit(struct kvm_vcpu *vcpu);
993 
994 /* These symbols are used in common code and are stubbed below.  */
995 
996 struct page *snp_safe_alloc_page_node(int node, gfp_t gfp);
997 static inline struct page *snp_safe_alloc_page(void)
998 {
999 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
1000 }
1001 void sev_snp_reload_vmsa(struct kvm_vcpu *vcpu);
1002 
1003 int sev_vcpu_create(struct kvm_vcpu *vcpu);
1004 void sev_free_vcpu(struct kvm_vcpu *vcpu);
1005 void sev_vm_init(struct kvm *kvm);
1006 void sev_vm_destroy(struct kvm *kvm);
1007 void __init sev_set_cpu_caps(void);
1008 void __init sev_hardware_setup(void);
1009 void sev_hardware_unsetup(void);
1010 int sev_cpu_init(struct svm_cpu_data *sd);
1011 int sev_dev_get_attr(u32 group, u64 attr, u64 *val);
1012 extern unsigned int max_sev_asid;
1013 void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code);
1014 int sev_gmem_make_private(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn, kvm_pfn_t nr_pages);
1015 void sev_gmem_make_shared(kvm_pfn_t pfn, kvm_pfn_t nr_pages);
1016 void sev_gmem_invalidate_range(struct kvm *kvm, struct kvm_gfn_range *range);
1017 int sev_gmem_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn, bool is_private);
1018 struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu);
1019 void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa);
1020 #else
1021 static inline struct page *snp_safe_alloc_page_node(int node, gfp_t gfp)
1022 {
1023 	return alloc_pages_node(node, gfp | __GFP_ZERO, 0);
1024 }
1025 
1026 static inline struct page *snp_safe_alloc_page(void)
1027 {
1028 	return snp_safe_alloc_page_node(numa_node_id(), GFP_KERNEL_ACCOUNT);
1029 }
1030 
1031 static inline int sev_vcpu_create(struct kvm_vcpu *vcpu) { return 0; }
1032 static inline void sev_free_vcpu(struct kvm_vcpu *vcpu) {}
1033 static inline void sev_vm_init(struct kvm *kvm) {}
1034 static inline void sev_vm_destroy(struct kvm *kvm) {}
1035 static inline void __init sev_set_cpu_caps(void) {}
1036 static inline void __init sev_hardware_setup(void) {}
1037 static inline void sev_hardware_unsetup(void) {}
1038 static inline int sev_cpu_init(struct svm_cpu_data *sd) { return 0; }
1039 static inline int sev_dev_get_attr(u32 group, u64 attr, u64 *val) { return -ENXIO; }
1040 #define max_sev_asid 0
1041 static inline void sev_handle_rmp_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u64 error_code) {}
1042 static inline struct vmcb_save_area *sev_decrypt_vmsa(struct kvm_vcpu *vcpu)
1043 {
1044 	return NULL;
1045 }
1046 static inline void sev_free_decrypted_vmsa(struct kvm_vcpu *vcpu, struct vmcb_save_area *vmsa) {}
1047 #endif
1048 
1049 /* vmenter.S */
1050 
1051 void __svm_sev_es_vcpu_run(struct vcpu_svm *svm, unsigned int flags,
1052 			   struct sev_es_save_area *hostsa);
1053 void __svm_vcpu_run(struct vcpu_svm *svm, unsigned int flags);
1054 
1055 #define DEFINE_KVM_GHCB_ACCESSORS(field)						\
1056 static __always_inline u64 kvm_ghcb_get_##field(struct vcpu_svm *svm)			\
1057 {											\
1058 	return READ_ONCE(svm->sev_es.ghcb->save.field);					\
1059 }											\
1060 											\
1061 static __always_inline bool kvm_ghcb_##field##_is_valid(const struct vcpu_svm *svm)	\
1062 {											\
1063 	return test_bit(GHCB_BITMAP_IDX(field),						\
1064 			(unsigned long *)&svm->sev_es.valid_bitmap);			\
1065 }											\
1066 											\
1067 static __always_inline u64 kvm_ghcb_get_##field##_if_valid(struct vcpu_svm *svm)	\
1068 {											\
1069 	return kvm_ghcb_##field##_is_valid(svm) ? kvm_ghcb_get_##field(svm) : 0;	\
1070 }
1071 
1072 DEFINE_KVM_GHCB_ACCESSORS(cpl)
1073 DEFINE_KVM_GHCB_ACCESSORS(rax)
1074 DEFINE_KVM_GHCB_ACCESSORS(rcx)
1075 DEFINE_KVM_GHCB_ACCESSORS(rdx)
1076 DEFINE_KVM_GHCB_ACCESSORS(rbx)
1077 DEFINE_KVM_GHCB_ACCESSORS(rsi)
1078 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_code)
1079 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_1)
1080 DEFINE_KVM_GHCB_ACCESSORS(sw_exit_info_2)
1081 DEFINE_KVM_GHCB_ACCESSORS(sw_scratch)
1082 DEFINE_KVM_GHCB_ACCESSORS(xcr0)
1083 DEFINE_KVM_GHCB_ACCESSORS(xss)
1084 
1085 #endif
1086