]> Git Repo - J-linux.git/blob - arch/arm64/include/asm/kvm_host.h
Merge tag 'vfs-6.13-rc7.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
[J-linux.git] / arch / arm64 / include / asm / kvm_host.h
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <[email protected]>
5  *
6  * Derived from arch/arm/include/asm/kvm_host.h:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Author: Christoffer Dall <[email protected]>
9  */
10
11 #ifndef __ARM64_KVM_HOST_H__
12 #define __ARM64_KVM_HOST_H__
13
14 #include <linux/arm-smccc.h>
15 #include <linux/bitmap.h>
16 #include <linux/types.h>
17 #include <linux/jump_label.h>
18 #include <linux/kvm_types.h>
19 #include <linux/maple_tree.h>
20 #include <linux/percpu.h>
21 #include <linux/psci.h>
22 #include <asm/arch_gicv3.h>
23 #include <asm/barrier.h>
24 #include <asm/cpufeature.h>
25 #include <asm/cputype.h>
26 #include <asm/daifflags.h>
27 #include <asm/fpsimd.h>
28 #include <asm/kvm.h>
29 #include <asm/kvm_asm.h>
30 #include <asm/vncr_mapping.h>
31
32 #define __KVM_HAVE_ARCH_INTC_INITIALIZED
33
34 #define KVM_HALT_POLL_NS_DEFAULT 500000
35
36 #include <kvm/arm_vgic.h>
37 #include <kvm/arm_arch_timer.h>
38 #include <kvm/arm_pmu.h>
39
40 #define KVM_MAX_VCPUS VGIC_V3_MAX_CPUS
41
42 #define KVM_VCPU_MAX_FEATURES 7
43 #define KVM_VCPU_VALID_FEATURES (BIT(KVM_VCPU_MAX_FEATURES) - 1)
44
45 #define KVM_REQ_SLEEP \
46         KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
47 #define KVM_REQ_IRQ_PENDING     KVM_ARCH_REQ(1)
48 #define KVM_REQ_VCPU_RESET      KVM_ARCH_REQ(2)
49 #define KVM_REQ_RECORD_STEAL    KVM_ARCH_REQ(3)
50 #define KVM_REQ_RELOAD_GICv4    KVM_ARCH_REQ(4)
51 #define KVM_REQ_RELOAD_PMU      KVM_ARCH_REQ(5)
52 #define KVM_REQ_SUSPEND         KVM_ARCH_REQ(6)
53 #define KVM_REQ_RESYNC_PMU_EL0  KVM_ARCH_REQ(7)
54 #define KVM_REQ_NESTED_S2_UNMAP KVM_ARCH_REQ(8)
55
56 #define KVM_DIRTY_LOG_MANUAL_CAPS   (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \
57                                      KVM_DIRTY_LOG_INITIALLY_SET)
58
59 #define KVM_HAVE_MMU_RWLOCK
60
61 /*
62  * Mode of operation configurable with kvm-arm.mode early param.
63  * See Documentation/admin-guide/kernel-parameters.txt for more information.
64  */
65 enum kvm_mode {
66         KVM_MODE_DEFAULT,
67         KVM_MODE_PROTECTED,
68         KVM_MODE_NV,
69         KVM_MODE_NONE,
70 };
71 #ifdef CONFIG_KVM
72 enum kvm_mode kvm_get_mode(void);
73 #else
74 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; };
75 #endif
76
77 extern unsigned int __ro_after_init kvm_sve_max_vl;
78 extern unsigned int __ro_after_init kvm_host_sve_max_vl;
79 int __init kvm_arm_init_sve(void);
80
81 u32 __attribute_const__ kvm_target_cpu(void);
82 void kvm_reset_vcpu(struct kvm_vcpu *vcpu);
83 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu);
84
85 struct kvm_hyp_memcache {
86         phys_addr_t head;
87         unsigned long nr_pages;
88 };
89
90 static inline void push_hyp_memcache(struct kvm_hyp_memcache *mc,
91                                      phys_addr_t *p,
92                                      phys_addr_t (*to_pa)(void *virt))
93 {
94         *p = mc->head;
95         mc->head = to_pa(p);
96         mc->nr_pages++;
97 }
98
99 static inline void *pop_hyp_memcache(struct kvm_hyp_memcache *mc,
100                                      void *(*to_va)(phys_addr_t phys))
101 {
102         phys_addr_t *p = to_va(mc->head);
103
104         if (!mc->nr_pages)
105                 return NULL;
106
107         mc->head = *p;
108         mc->nr_pages--;
109
110         return p;
111 }
112
113 static inline int __topup_hyp_memcache(struct kvm_hyp_memcache *mc,
114                                        unsigned long min_pages,
115                                        void *(*alloc_fn)(void *arg),
116                                        phys_addr_t (*to_pa)(void *virt),
117                                        void *arg)
118 {
119         while (mc->nr_pages < min_pages) {
120                 phys_addr_t *p = alloc_fn(arg);
121
122                 if (!p)
123                         return -ENOMEM;
124                 push_hyp_memcache(mc, p, to_pa);
125         }
126
127         return 0;
128 }
129
130 static inline void __free_hyp_memcache(struct kvm_hyp_memcache *mc,
131                                        void (*free_fn)(void *virt, void *arg),
132                                        void *(*to_va)(phys_addr_t phys),
133                                        void *arg)
134 {
135         while (mc->nr_pages)
136                 free_fn(pop_hyp_memcache(mc, to_va), arg);
137 }
138
139 void free_hyp_memcache(struct kvm_hyp_memcache *mc);
140 int topup_hyp_memcache(struct kvm_hyp_memcache *mc, unsigned long min_pages);
141
142 struct kvm_vmid {
143         atomic64_t id;
144 };
145
146 struct kvm_s2_mmu {
147         struct kvm_vmid vmid;
148
149         /*
150          * stage2 entry level table
151          *
152          * Two kvm_s2_mmu structures in the same VM can point to the same
153          * pgd here.  This happens when running a guest using a
154          * translation regime that isn't affected by its own stage-2
155          * translation, such as a non-VHE hypervisor running at vEL2, or
156          * for vEL1/EL0 with vHCR_EL2.VM == 0.  In that case, we use the
157          * canonical stage-2 page tables.
158          */
159         phys_addr_t     pgd_phys;
160         struct kvm_pgtable *pgt;
161
162         /*
163          * VTCR value used on the host. For a non-NV guest (or a NV
164          * guest that runs in a context where its own S2 doesn't
165          * apply), its T0SZ value reflects that of the IPA size.
166          *
167          * For a shadow S2 MMU, T0SZ reflects the PARange exposed to
168          * the guest.
169          */
170         u64     vtcr;
171
172         /* The last vcpu id that ran on each physical CPU */
173         int __percpu *last_vcpu_ran;
174
175 #define KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT 0
176         /*
177          * Memory cache used to split
178          * KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE worth of huge pages. It
179          * is used to allocate stage2 page tables while splitting huge
180          * pages. The choice of KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE
181          * influences both the capacity of the split page cache, and
182          * how often KVM reschedules. Be wary of raising CHUNK_SIZE
183          * too high.
184          *
185          * Protected by kvm->slots_lock.
186          */
187         struct kvm_mmu_memory_cache split_page_cache;
188         uint64_t split_page_chunk_size;
189
190         struct kvm_arch *arch;
191
192         /*
193          * For a shadow stage-2 MMU, the virtual vttbr used by the
194          * host to parse the guest S2.
195          * This either contains:
196          * - the virtual VTTBR programmed by the guest hypervisor with
197          *   CnP cleared
198          * - The value 1 (VMID=0, BADDR=0, CnP=1) if invalid
199          *
200          * We also cache the full VTCR which gets used for TLB invalidation,
201          * taking the ARM ARM's "Any of the bits in VTCR_EL2 are permitted
202          * to be cached in a TLB" to the letter.
203          */
204         u64     tlb_vttbr;
205         u64     tlb_vtcr;
206
207         /*
208          * true when this represents a nested context where virtual
209          * HCR_EL2.VM == 1
210          */
211         bool    nested_stage2_enabled;
212
213         /*
214          * true when this MMU needs to be unmapped before being used for a new
215          * purpose.
216          */
217         bool    pending_unmap;
218
219         /*
220          *  0: Nobody is currently using this, check vttbr for validity
221          * >0: Somebody is actively using this.
222          */
223         atomic_t refcnt;
224 };
225
226 struct kvm_arch_memory_slot {
227 };
228
229 /**
230  * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests
231  *
232  * @std_bmap: Bitmap of standard secure service calls
233  * @std_hyp_bmap: Bitmap of standard hypervisor service calls
234  * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls
235  */
236 struct kvm_smccc_features {
237         unsigned long std_bmap;
238         unsigned long std_hyp_bmap;
239         unsigned long vendor_hyp_bmap;
240 };
241
242 typedef unsigned int pkvm_handle_t;
243
244 struct kvm_protected_vm {
245         pkvm_handle_t handle;
246         struct kvm_hyp_memcache teardown_mc;
247         bool enabled;
248 };
249
250 struct kvm_mpidr_data {
251         u64                     mpidr_mask;
252         DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx);
253 };
254
255 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr)
256 {
257         unsigned long index = 0, mask = data->mpidr_mask;
258         unsigned long aff = mpidr & MPIDR_HWID_BITMASK;
259
260         bitmap_gather(&index, &aff, &mask, fls(mask));
261
262         return index;
263 }
264
265 struct kvm_sysreg_masks;
266
267 enum fgt_group_id {
268         __NO_FGT_GROUP__,
269         HFGxTR_GROUP,
270         HDFGRTR_GROUP,
271         HDFGWTR_GROUP = HDFGRTR_GROUP,
272         HFGITR_GROUP,
273         HAFGRTR_GROUP,
274
275         /* Must be last */
276         __NR_FGT_GROUP_IDS__
277 };
278
279 struct kvm_arch {
280         struct kvm_s2_mmu mmu;
281
282         /*
283          * Fine-Grained UNDEF, mimicking the FGT layout defined by the
284          * architecture. We track them globally, as we present the
285          * same feature-set to all vcpus.
286          *
287          * Index 0 is currently spare.
288          */
289         u64 fgu[__NR_FGT_GROUP_IDS__];
290
291         /*
292          * Stage 2 paging state for VMs with nested S2 using a virtual
293          * VMID.
294          */
295         struct kvm_s2_mmu *nested_mmus;
296         size_t nested_mmus_size;
297         int nested_mmus_next;
298
299         /* Interrupt controller */
300         struct vgic_dist        vgic;
301
302         /* Timers */
303         struct arch_timer_vm_data timer_data;
304
305         /* Mandated version of PSCI */
306         u32 psci_version;
307
308         /* Protects VM-scoped configuration data */
309         struct mutex config_lock;
310
311         /*
312          * If we encounter a data abort without valid instruction syndrome
313          * information, report this to user space.  User space can (and
314          * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
315          * supported.
316          */
317 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER      0
318         /* Memory Tagging Extension enabled for the guest */
319 #define KVM_ARCH_FLAG_MTE_ENABLED                       1
320         /* At least one vCPU has ran in the VM */
321 #define KVM_ARCH_FLAG_HAS_RAN_ONCE                      2
322         /* The vCPU feature set for the VM is configured */
323 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED          3
324         /* PSCI SYSTEM_SUSPEND enabled for the guest */
325 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED            4
326         /* VM counter offset */
327 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET                 5
328         /* Timer PPIs made immutable */
329 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE              6
330         /* Initial ID reg values loaded */
331 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED               7
332         /* Fine-Grained UNDEF initialised */
333 #define KVM_ARCH_FLAG_FGU_INITIALIZED                   8
334         unsigned long flags;
335
336         /* VM-wide vCPU feature set */
337         DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
338
339         /* MPIDR to vcpu index mapping, optional */
340         struct kvm_mpidr_data *mpidr_data;
341
342         /*
343          * VM-wide PMU filter, implemented as a bitmap and big enough for
344          * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
345          */
346         unsigned long *pmu_filter;
347         struct arm_pmu *arm_pmu;
348
349         cpumask_var_t supported_cpus;
350
351         /* PMCR_EL0.N value for the guest */
352         u8 pmcr_n;
353
354         /* Iterator for idreg debugfs */
355         u8      idreg_debugfs_iter;
356
357         /* Hypercall features firmware registers' descriptor */
358         struct kvm_smccc_features smccc_feat;
359         struct maple_tree smccc_filter;
360
361         /*
362          * Emulated CPU ID registers per VM
363          * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
364          * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
365          *
366          * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
367          * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
368          */
369 #define IDREG_IDX(id)           (((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
370 #define KVM_ARM_ID_REG_NUM      (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
371         u64 id_regs[KVM_ARM_ID_REG_NUM];
372
373         u64 ctr_el0;
374
375         /* Masks for VNCR-backed and general EL2 sysregs */
376         struct kvm_sysreg_masks *sysreg_masks;
377
378         /*
379          * For an untrusted host VM, 'pkvm.handle' is used to lookup
380          * the associated pKVM instance in the hypervisor.
381          */
382         struct kvm_protected_vm pkvm;
383 };
384
385 struct kvm_vcpu_fault_info {
386         u64 esr_el2;            /* Hyp Syndrom Register */
387         u64 far_el2;            /* Hyp Fault Address Register */
388         u64 hpfar_el2;          /* Hyp IPA Fault Address Register */
389         u64 disr_el1;           /* Deferred [SError] Status Register */
390 };
391
392 /*
393  * VNCR() just places the VNCR_capable registers in the enum after
394  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
395  * from the VNCR base. As we don't require the enum to be otherwise ordered,
396  * we need the terrible hack below to ensure that we correctly size the
397  * sys_regs array, no matter what.
398  *
399  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
400  * treasure trove of bit hacks:
401  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
402  */
403 #define __MAX__(x,y)    ((x) ^ (((x) ^ (y)) & -((x) < (y))))
404 #define VNCR(r)                                         \
405         __before_##r,                                   \
406         r = __VNCR_START__ + ((VNCR_ ## r) / 8),        \
407         __after_##r = __MAX__(__before_##r - 1, r)
408
409 #define MARKER(m)                               \
410         m, __after_##m = m - 1
411
412 enum vcpu_sysreg {
413         __INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
414         MPIDR_EL1,      /* MultiProcessor Affinity Register */
415         CLIDR_EL1,      /* Cache Level ID Register */
416         CSSELR_EL1,     /* Cache Size Selection Register */
417         TPIDR_EL0,      /* Thread ID, User R/W */
418         TPIDRRO_EL0,    /* Thread ID, User R/O */
419         TPIDR_EL1,      /* Thread ID, Privileged */
420         CNTKCTL_EL1,    /* Timer Control Register (EL1) */
421         PAR_EL1,        /* Physical Address Register */
422         MDCCINT_EL1,    /* Monitor Debug Comms Channel Interrupt Enable Reg */
423         OSLSR_EL1,      /* OS Lock Status Register */
424         DISR_EL1,       /* Deferred Interrupt Status Register */
425
426         /* Performance Monitors Registers */
427         PMCR_EL0,       /* Control Register */
428         PMSELR_EL0,     /* Event Counter Selection Register */
429         PMEVCNTR0_EL0,  /* Event Counter Register (0-30) */
430         PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
431         PMCCNTR_EL0,    /* Cycle Counter Register */
432         PMEVTYPER0_EL0, /* Event Type Register (0-30) */
433         PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
434         PMCCFILTR_EL0,  /* Cycle Count Filter Register */
435         PMCNTENSET_EL0, /* Count Enable Set Register */
436         PMINTENSET_EL1, /* Interrupt Enable Set Register */
437         PMOVSSET_EL0,   /* Overflow Flag Status Set Register */
438         PMUSERENR_EL0,  /* User Enable Register */
439
440         /* Pointer Authentication Registers in a strict increasing order. */
441         APIAKEYLO_EL1,
442         APIAKEYHI_EL1,
443         APIBKEYLO_EL1,
444         APIBKEYHI_EL1,
445         APDAKEYLO_EL1,
446         APDAKEYHI_EL1,
447         APDBKEYLO_EL1,
448         APDBKEYHI_EL1,
449         APGAKEYLO_EL1,
450         APGAKEYHI_EL1,
451
452         /* Memory Tagging Extension registers */
453         RGSR_EL1,       /* Random Allocation Tag Seed Register */
454         GCR_EL1,        /* Tag Control Register */
455         TFSRE0_EL1,     /* Tag Fault Status Register (EL0) */
456
457         POR_EL0,        /* Permission Overlay Register 0 (EL0) */
458
459         /* FP/SIMD/SVE */
460         SVCR,
461         FPMR,
462
463         /* 32bit specific registers. */
464         DACR32_EL2,     /* Domain Access Control Register */
465         IFSR32_EL2,     /* Instruction Fault Status Register */
466         FPEXC32_EL2,    /* Floating-Point Exception Control Register */
467         DBGVCR32_EL2,   /* Debug Vector Catch Register */
468
469         /* EL2 registers */
470         SCTLR_EL2,      /* System Control Register (EL2) */
471         ACTLR_EL2,      /* Auxiliary Control Register (EL2) */
472         CPTR_EL2,       /* Architectural Feature Trap Register (EL2) */
473         HACR_EL2,       /* Hypervisor Auxiliary Control Register */
474         ZCR_EL2,        /* SVE Control Register (EL2) */
475         TTBR0_EL2,      /* Translation Table Base Register 0 (EL2) */
476         TTBR1_EL2,      /* Translation Table Base Register 1 (EL2) */
477         TCR_EL2,        /* Translation Control Register (EL2) */
478         PIRE0_EL2,      /* Permission Indirection Register 0 (EL2) */
479         PIR_EL2,        /* Permission Indirection Register 1 (EL2) */
480         POR_EL2,        /* Permission Overlay Register 2 (EL2) */
481         SPSR_EL2,       /* EL2 saved program status register */
482         ELR_EL2,        /* EL2 exception link register */
483         AFSR0_EL2,      /* Auxiliary Fault Status Register 0 (EL2) */
484         AFSR1_EL2,      /* Auxiliary Fault Status Register 1 (EL2) */
485         ESR_EL2,        /* Exception Syndrome Register (EL2) */
486         FAR_EL2,        /* Fault Address Register (EL2) */
487         HPFAR_EL2,      /* Hypervisor IPA Fault Address Register */
488         MAIR_EL2,       /* Memory Attribute Indirection Register (EL2) */
489         AMAIR_EL2,      /* Auxiliary Memory Attribute Indirection Register (EL2) */
490         VBAR_EL2,       /* Vector Base Address Register (EL2) */
491         RVBAR_EL2,      /* Reset Vector Base Address Register */
492         CONTEXTIDR_EL2, /* Context ID Register (EL2) */
493         CNTHCTL_EL2,    /* Counter-timer Hypervisor Control register */
494         SP_EL2,         /* EL2 Stack Pointer */
495         CNTHP_CTL_EL2,
496         CNTHP_CVAL_EL2,
497         CNTHV_CTL_EL2,
498         CNTHV_CVAL_EL2,
499
500         /* Anything from this can be RES0/RES1 sanitised */
501         MARKER(__SANITISED_REG_START__),
502         TCR2_EL2,       /* Extended Translation Control Register (EL2) */
503         MDCR_EL2,       /* Monitor Debug Configuration Register (EL2) */
504
505         /* Any VNCR-capable reg goes after this point */
506         MARKER(__VNCR_START__),
507
508         VNCR(SCTLR_EL1),/* System Control Register */
509         VNCR(ACTLR_EL1),/* Auxiliary Control Register */
510         VNCR(CPACR_EL1),/* Coprocessor Access Control */
511         VNCR(ZCR_EL1),  /* SVE Control */
512         VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
513         VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
514         VNCR(TCR_EL1),  /* Translation Control Register */
515         VNCR(TCR2_EL1), /* Extended Translation Control Register */
516         VNCR(ESR_EL1),  /* Exception Syndrome Register */
517         VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
518         VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
519         VNCR(FAR_EL1),  /* Fault Address Register */
520         VNCR(MAIR_EL1), /* Memory Attribute Indirection Register */
521         VNCR(VBAR_EL1), /* Vector Base Address Register */
522         VNCR(CONTEXTIDR_EL1),   /* Context ID Register */
523         VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
524         VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
525         VNCR(ELR_EL1),
526         VNCR(SP_EL1),
527         VNCR(SPSR_EL1),
528         VNCR(TFSR_EL1), /* Tag Fault Status Register (EL1) */
529         VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
530         VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
531         VNCR(HCR_EL2),  /* Hypervisor Configuration Register */
532         VNCR(HSTR_EL2), /* Hypervisor System Trap Register */
533         VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
534         VNCR(VTCR_EL2), /* Virtualization Translation Control Register */
535         VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
536         VNCR(HCRX_EL2), /* Extended Hypervisor Configuration Register */
537
538         /* Permission Indirection Extension registers */
539         VNCR(PIR_EL1),   /* Permission Indirection Register 1 (EL1) */
540         VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
541
542         VNCR(POR_EL1),  /* Permission Overlay Register 1 (EL1) */
543
544         VNCR(HFGRTR_EL2),
545         VNCR(HFGWTR_EL2),
546         VNCR(HFGITR_EL2),
547         VNCR(HDFGRTR_EL2),
548         VNCR(HDFGWTR_EL2),
549         VNCR(HAFGRTR_EL2),
550
551         VNCR(CNTVOFF_EL2),
552         VNCR(CNTV_CVAL_EL0),
553         VNCR(CNTV_CTL_EL0),
554         VNCR(CNTP_CVAL_EL0),
555         VNCR(CNTP_CTL_EL0),
556
557         VNCR(ICH_HCR_EL2),
558
559         NR_SYS_REGS     /* Nothing after this line! */
560 };
561
562 struct kvm_sysreg_masks {
563         struct {
564                 u64     res0;
565                 u64     res1;
566         } mask[NR_SYS_REGS - __SANITISED_REG_START__];
567 };
568
569 struct kvm_cpu_context {
570         struct user_pt_regs regs;       /* sp = sp_el0 */
571
572         u64     spsr_abt;
573         u64     spsr_und;
574         u64     spsr_irq;
575         u64     spsr_fiq;
576
577         struct user_fpsimd_state fp_regs;
578
579         u64 sys_regs[NR_SYS_REGS];
580
581         struct kvm_vcpu *__hyp_running_vcpu;
582
583         /* This pointer has to be 4kB aligned. */
584         u64 *vncr_array;
585 };
586
587 struct cpu_sve_state {
588         __u64 zcr_el1;
589
590         /*
591          * Ordering is important since __sve_save_state/__sve_restore_state
592          * relies on it.
593          */
594         __u32 fpsr;
595         __u32 fpcr;
596
597         /* Must be SVE_VQ_BYTES (128 bit) aligned. */
598         __u8 sve_regs[];
599 };
600
601 /*
602  * This structure is instantiated on a per-CPU basis, and contains
603  * data that is:
604  *
605  * - tied to a single physical CPU, and
606  * - either have a lifetime that does not extend past vcpu_put()
607  * - or is an invariant for the lifetime of the system
608  *
609  * Use host_data_ptr(field) as a way to access a pointer to such a
610  * field.
611  */
612 struct kvm_host_data {
613         struct kvm_cpu_context host_ctxt;
614
615         /*
616          * All pointers in this union are hyp VA.
617          * sve_state is only used in pKVM and if system_supports_sve().
618          */
619         union {
620                 struct user_fpsimd_state *fpsimd_state;
621                 struct cpu_sve_state *sve_state;
622         };
623
624         union {
625                 /* HYP VA pointer to the host storage for FPMR */
626                 u64     *fpmr_ptr;
627                 /*
628                  * Used by pKVM only, as it needs to provide storage
629                  * for the host
630                  */
631                 u64     fpmr;
632         };
633
634         /* Ownership of the FP regs */
635         enum {
636                 FP_STATE_FREE,
637                 FP_STATE_HOST_OWNED,
638                 FP_STATE_GUEST_OWNED,
639         } fp_owner;
640
641         /*
642          * host_debug_state contains the host registers which are
643          * saved and restored during world switches.
644          */
645          struct {
646                 /* {Break,watch}point registers */
647                 struct kvm_guest_debug_arch regs;
648                 /* Statistical profiling extension */
649                 u64 pmscr_el1;
650                 /* Self-hosted trace */
651                 u64 trfcr_el1;
652                 /* Values of trap registers for the host before guest entry. */
653                 u64 mdcr_el2;
654         } host_debug_state;
655 };
656
657 struct kvm_host_psci_config {
658         /* PSCI version used by host. */
659         u32 version;
660         u32 smccc_version;
661
662         /* Function IDs used by host if version is v0.1. */
663         struct psci_0_1_function_ids function_ids_0_1;
664
665         bool psci_0_1_cpu_suspend_implemented;
666         bool psci_0_1_cpu_on_implemented;
667         bool psci_0_1_cpu_off_implemented;
668         bool psci_0_1_migrate_implemented;
669 };
670
671 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
672 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
673
674 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
675 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
676
677 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
678 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
679
680 struct vcpu_reset_state {
681         unsigned long   pc;
682         unsigned long   r0;
683         bool            be;
684         bool            reset;
685 };
686
687 struct kvm_vcpu_arch {
688         struct kvm_cpu_context ctxt;
689
690         /*
691          * Guest floating point state
692          *
693          * The architecture has two main floating point extensions,
694          * the original FPSIMD and SVE.  These have overlapping
695          * register views, with the FPSIMD V registers occupying the
696          * low 128 bits of the SVE Z registers.  When the core
697          * floating point code saves the register state of a task it
698          * records which view it saved in fp_type.
699          */
700         void *sve_state;
701         enum fp_type fp_type;
702         unsigned int sve_max_vl;
703
704         /* Stage 2 paging state used by the hardware on next switch */
705         struct kvm_s2_mmu *hw_mmu;
706
707         /* Values of trap registers for the guest. */
708         u64 hcr_el2;
709         u64 hcrx_el2;
710         u64 mdcr_el2;
711         u64 cptr_el2;
712
713         /* Exception Information */
714         struct kvm_vcpu_fault_info fault;
715
716         /* Configuration flags, set once and for all before the vcpu can run */
717         u8 cflags;
718
719         /* Input flags to the hypervisor code, potentially cleared after use */
720         u8 iflags;
721
722         /* State flags for kernel bookkeeping, unused by the hypervisor code */
723         u8 sflags;
724
725         /*
726          * Don't run the guest (internal implementation need).
727          *
728          * Contrary to the flags above, this is set/cleared outside of
729          * a vcpu context, and thus cannot be mixed with the flags
730          * themselves (or the flag accesses need to be made atomic).
731          */
732         bool pause;
733
734         /*
735          * We maintain more than a single set of debug registers to support
736          * debugging the guest from the host and to maintain separate host and
737          * guest state during world switches. vcpu_debug_state are the debug
738          * registers of the vcpu as the guest sees them.
739          *
740          * external_debug_state contains the debug values we want to debug the
741          * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl.
742          *
743          * debug_ptr points to the set of debug registers that should be loaded
744          * onto the hardware when running the guest.
745          */
746         struct kvm_guest_debug_arch *debug_ptr;
747         struct kvm_guest_debug_arch vcpu_debug_state;
748         struct kvm_guest_debug_arch external_debug_state;
749
750         /* VGIC state */
751         struct vgic_cpu vgic_cpu;
752         struct arch_timer_cpu timer_cpu;
753         struct kvm_pmu pmu;
754
755         /*
756          * Guest registers we preserve during guest debugging.
757          *
758          * These shadow registers are updated by the kvm_handle_sys_reg
759          * trap handler if the guest accesses or updates them while we
760          * are using guest debug.
761          */
762         struct {
763                 u32     mdscr_el1;
764                 bool    pstate_ss;
765         } guest_debug_preserved;
766
767         /* vcpu power state */
768         struct kvm_mp_state mp_state;
769         spinlock_t mp_state_lock;
770
771         /* Cache some mmu pages needed inside spinlock regions */
772         struct kvm_mmu_memory_cache mmu_page_cache;
773
774         /* Virtual SError ESR to restore when HCR_EL2.VSE is set */
775         u64 vsesr_el2;
776
777         /* Additional reset state */
778         struct vcpu_reset_state reset_state;
779
780         /* Guest PV state */
781         struct {
782                 u64 last_steal;
783                 gpa_t base;
784         } steal;
785
786         /* Per-vcpu CCSIDR override or NULL */
787         u32 *ccsidr;
788 };
789
790 /*
791  * Each 'flag' is composed of a comma-separated triplet:
792  *
793  * - the flag-set it belongs to in the vcpu->arch structure
794  * - the value for that flag
795  * - the mask for that flag
796  *
797  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
798  * unpack_vcpu_flag() extract the flag value from the triplet for
799  * direct use outside of the flag accessors.
800  */
801 #define __vcpu_single_flag(_set, _f)    _set, (_f), (_f)
802
803 #define __unpack_flag(_set, _f, _m)     _f
804 #define unpack_vcpu_flag(...)           __unpack_flag(__VA_ARGS__)
805
806 #define __build_check_flag(v, flagset, f, m)                    \
807         do {                                                    \
808                 typeof(v->arch.flagset) *_fset;                 \
809                                                                 \
810                 /* Check that the flags fit in the mask */      \
811                 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \
812                 /* Check that the flags fit in the type */      \
813                 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \
814         } while (0)
815
816 #define __vcpu_get_flag(v, flagset, f, m)                       \
817         ({                                                      \
818                 __build_check_flag(v, flagset, f, m);           \
819                                                                 \
820                 READ_ONCE(v->arch.flagset) & (m);               \
821         })
822
823 /*
824  * Note that the set/clear accessors must be preempt-safe in order to
825  * avoid nesting them with load/put which also manipulate flags...
826  */
827 #ifdef __KVM_NVHE_HYPERVISOR__
828 /* the nVHE hypervisor is always non-preemptible */
829 #define __vcpu_flags_preempt_disable()
830 #define __vcpu_flags_preempt_enable()
831 #else
832 #define __vcpu_flags_preempt_disable()  preempt_disable()
833 #define __vcpu_flags_preempt_enable()   preempt_enable()
834 #endif
835
836 #define __vcpu_set_flag(v, flagset, f, m)                       \
837         do {                                                    \
838                 typeof(v->arch.flagset) *fset;                  \
839                                                                 \
840                 __build_check_flag(v, flagset, f, m);           \
841                                                                 \
842                 fset = &v->arch.flagset;                        \
843                 __vcpu_flags_preempt_disable();                 \
844                 if (HWEIGHT(m) > 1)                             \
845                         *fset &= ~(m);                          \
846                 *fset |= (f);                                   \
847                 __vcpu_flags_preempt_enable();                  \
848         } while (0)
849
850 #define __vcpu_clear_flag(v, flagset, f, m)                     \
851         do {                                                    \
852                 typeof(v->arch.flagset) *fset;                  \
853                                                                 \
854                 __build_check_flag(v, flagset, f, m);           \
855                                                                 \
856                 fset = &v->arch.flagset;                        \
857                 __vcpu_flags_preempt_disable();                 \
858                 *fset &= ~(m);                                  \
859                 __vcpu_flags_preempt_enable();                  \
860         } while (0)
861
862 #define vcpu_get_flag(v, ...)   __vcpu_get_flag((v), __VA_ARGS__)
863 #define vcpu_set_flag(v, ...)   __vcpu_set_flag((v), __VA_ARGS__)
864 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__)
865
866 /* SVE exposed to guest */
867 #define GUEST_HAS_SVE           __vcpu_single_flag(cflags, BIT(0))
868 /* SVE config completed */
869 #define VCPU_SVE_FINALIZED      __vcpu_single_flag(cflags, BIT(1))
870 /* PTRAUTH exposed to guest */
871 #define GUEST_HAS_PTRAUTH       __vcpu_single_flag(cflags, BIT(2))
872 /* KVM_ARM_VCPU_INIT completed */
873 #define VCPU_INITIALIZED        __vcpu_single_flag(cflags, BIT(3))
874
875 /* Exception pending */
876 #define PENDING_EXCEPTION       __vcpu_single_flag(iflags, BIT(0))
877 /*
878  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
879  * be set together with an exception...
880  */
881 #define INCREMENT_PC            __vcpu_single_flag(iflags, BIT(1))
882 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
883 #define EXCEPT_MASK             __vcpu_single_flag(iflags, GENMASK(3, 1))
884
885 /* Helpers to encode exceptions with minimum fuss */
886 #define __EXCEPT_MASK_VAL       unpack_vcpu_flag(EXCEPT_MASK)
887 #define __EXCEPT_SHIFT          __builtin_ctzl(__EXCEPT_MASK_VAL)
888 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
889
890 /*
891  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
892  * values:
893  *
894  * For AArch32 EL1:
895  */
896 #define EXCEPT_AA32_UND         __vcpu_except_flags(0)
897 #define EXCEPT_AA32_IABT        __vcpu_except_flags(1)
898 #define EXCEPT_AA32_DABT        __vcpu_except_flags(2)
899 /* For AArch64: */
900 #define EXCEPT_AA64_EL1_SYNC    __vcpu_except_flags(0)
901 #define EXCEPT_AA64_EL1_IRQ     __vcpu_except_flags(1)
902 #define EXCEPT_AA64_EL1_FIQ     __vcpu_except_flags(2)
903 #define EXCEPT_AA64_EL1_SERR    __vcpu_except_flags(3)
904 /* For AArch64 with NV: */
905 #define EXCEPT_AA64_EL2_SYNC    __vcpu_except_flags(4)
906 #define EXCEPT_AA64_EL2_IRQ     __vcpu_except_flags(5)
907 #define EXCEPT_AA64_EL2_FIQ     __vcpu_except_flags(6)
908 #define EXCEPT_AA64_EL2_SERR    __vcpu_except_flags(7)
909 /* Guest debug is live */
910 #define DEBUG_DIRTY             __vcpu_single_flag(iflags, BIT(4))
911 /* Save SPE context if active  */
912 #define DEBUG_STATE_SAVE_SPE    __vcpu_single_flag(iflags, BIT(5))
913 /* Save TRBE context if active  */
914 #define DEBUG_STATE_SAVE_TRBE   __vcpu_single_flag(iflags, BIT(6))
915
916 /* SVE enabled for host EL0 */
917 #define HOST_SVE_ENABLED        __vcpu_single_flag(sflags, BIT(0))
918 /* SME enabled for EL0 */
919 #define HOST_SME_ENABLED        __vcpu_single_flag(sflags, BIT(1))
920 /* Physical CPU not in supported_cpus */
921 #define ON_UNSUPPORTED_CPU      __vcpu_single_flag(sflags, BIT(2))
922 /* WFIT instruction trapped */
923 #define IN_WFIT                 __vcpu_single_flag(sflags, BIT(3))
924 /* vcpu system registers loaded on physical CPU */
925 #define SYSREGS_ON_CPU          __vcpu_single_flag(sflags, BIT(4))
926 /* Software step state is Active-pending */
927 #define DBG_SS_ACTIVE_PENDING   __vcpu_single_flag(sflags, BIT(5))
928 /* PMUSERENR for the guest EL0 is on physical CPU */
929 #define PMUSERENR_ON_CPU        __vcpu_single_flag(sflags, BIT(6))
930 /* WFI instruction trapped */
931 #define IN_WFI                  __vcpu_single_flag(sflags, BIT(7))
932
933
934 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
935 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +      \
936                              sve_ffr_offset((vcpu)->arch.sve_max_vl))
937
938 #define vcpu_sve_max_vq(vcpu)   sve_vq_from_vl((vcpu)->arch.sve_max_vl)
939
940 #define vcpu_sve_zcr_elx(vcpu)                                          \
941         (unlikely(is_hyp_ctxt(vcpu)) ? ZCR_EL2 : ZCR_EL1)
942
943 #define vcpu_sve_state_size(vcpu) ({                                    \
944         size_t __size_ret;                                              \
945         unsigned int __vcpu_vq;                                         \
946                                                                         \
947         if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {          \
948                 __size_ret = 0;                                         \
949         } else {                                                        \
950                 __vcpu_vq = vcpu_sve_max_vq(vcpu);                      \
951                 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);              \
952         }                                                               \
953                                                                         \
954         __size_ret;                                                     \
955 })
956
957 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
958                                  KVM_GUESTDBG_USE_SW_BP | \
959                                  KVM_GUESTDBG_USE_HW | \
960                                  KVM_GUESTDBG_SINGLESTEP)
961
962 #define vcpu_has_sve(vcpu) (system_supports_sve() &&                    \
963                             vcpu_get_flag(vcpu, GUEST_HAS_SVE))
964
965 #ifdef CONFIG_ARM64_PTR_AUTH
966 #define vcpu_has_ptrauth(vcpu)                                          \
967         ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||                \
968           cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&               \
969           vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
970 #else
971 #define vcpu_has_ptrauth(vcpu)          false
972 #endif
973
974 #define vcpu_on_unsupported_cpu(vcpu)                                   \
975         vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
976
977 #define vcpu_set_on_unsupported_cpu(vcpu)                               \
978         vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
979
980 #define vcpu_clear_on_unsupported_cpu(vcpu)                             \
981         vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
982
983 #define vcpu_gp_regs(v)         (&(v)->arch.ctxt.regs)
984
985 /*
986  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
987  * memory backed version of a register, and not the one most recently
988  * accessed by a running VCPU.  For example, for userspace access or
989  * for system registers that are never context switched, but only
990  * emulated.
991  *
992  * Don't bother with VNCR-based accesses in the nVHE code, it has no
993  * business dealing with NV.
994  */
995 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
996 {
997 #if !defined (__KVM_NVHE_HYPERVISOR__)
998         if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
999                      r >= __VNCR_START__ && ctxt->vncr_array))
1000                 return &ctxt->vncr_array[r - __VNCR_START__];
1001 #endif
1002         return (u64 *)&ctxt->sys_regs[r];
1003 }
1004
1005 #define __ctxt_sys_reg(c,r)                                             \
1006         ({                                                              \
1007                 BUILD_BUG_ON(__builtin_constant_p(r) &&                 \
1008                              (r) >= NR_SYS_REGS);                       \
1009                 ___ctxt_sys_reg(c, r);                                  \
1010         })
1011
1012 #define ctxt_sys_reg(c,r)       (*__ctxt_sys_reg(c,r))
1013
1014 u64 kvm_vcpu_apply_reg_masks(const struct kvm_vcpu *, enum vcpu_sysreg, u64);
1015 #define __vcpu_sys_reg(v,r)                                             \
1016         (*({                                                            \
1017                 const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt;   \
1018                 u64 *__r = __ctxt_sys_reg(ctxt, (r));                   \
1019                 if (vcpu_has_nv((v)) && (r) >= __SANITISED_REG_START__) \
1020                         *__r = kvm_vcpu_apply_reg_masks((v), (r), *__r);\
1021                 __r;                                                    \
1022         }))
1023
1024 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
1025 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
1026
1027 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
1028 {
1029         /*
1030          * *** VHE ONLY ***
1031          *
1032          * System registers listed in the switch are not saved on every
1033          * exit from the guest but are only saved on vcpu_put.
1034          *
1035          * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
1036          * should never be listed below, because the guest cannot modify its
1037          * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
1038          * thread when emulating cross-VCPU communication.
1039          */
1040         if (!has_vhe())
1041                 return false;
1042
1043         switch (reg) {
1044         case SCTLR_EL1:         *val = read_sysreg_s(SYS_SCTLR_EL12);   break;
1045         case CPACR_EL1:         *val = read_sysreg_s(SYS_CPACR_EL12);   break;
1046         case TTBR0_EL1:         *val = read_sysreg_s(SYS_TTBR0_EL12);   break;
1047         case TTBR1_EL1:         *val = read_sysreg_s(SYS_TTBR1_EL12);   break;
1048         case TCR_EL1:           *val = read_sysreg_s(SYS_TCR_EL12);     break;
1049         case TCR2_EL1:          *val = read_sysreg_s(SYS_TCR2_EL12);    break;
1050         case PIR_EL1:           *val = read_sysreg_s(SYS_PIR_EL12);     break;
1051         case PIRE0_EL1:         *val = read_sysreg_s(SYS_PIRE0_EL12);   break;
1052         case POR_EL1:           *val = read_sysreg_s(SYS_POR_EL12);     break;
1053         case ESR_EL1:           *val = read_sysreg_s(SYS_ESR_EL12);     break;
1054         case AFSR0_EL1:         *val = read_sysreg_s(SYS_AFSR0_EL12);   break;
1055         case AFSR1_EL1:         *val = read_sysreg_s(SYS_AFSR1_EL12);   break;
1056         case FAR_EL1:           *val = read_sysreg_s(SYS_FAR_EL12);     break;
1057         case MAIR_EL1:          *val = read_sysreg_s(SYS_MAIR_EL12);    break;
1058         case VBAR_EL1:          *val = read_sysreg_s(SYS_VBAR_EL12);    break;
1059         case CONTEXTIDR_EL1:    *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
1060         case TPIDR_EL0:         *val = read_sysreg_s(SYS_TPIDR_EL0);    break;
1061         case TPIDRRO_EL0:       *val = read_sysreg_s(SYS_TPIDRRO_EL0);  break;
1062         case TPIDR_EL1:         *val = read_sysreg_s(SYS_TPIDR_EL1);    break;
1063         case AMAIR_EL1:         *val = read_sysreg_s(SYS_AMAIR_EL12);   break;
1064         case CNTKCTL_EL1:       *val = read_sysreg_s(SYS_CNTKCTL_EL12); break;
1065         case ELR_EL1:           *val = read_sysreg_s(SYS_ELR_EL12);     break;
1066         case SPSR_EL1:          *val = read_sysreg_s(SYS_SPSR_EL12);    break;
1067         case PAR_EL1:           *val = read_sysreg_par();               break;
1068         case DACR32_EL2:        *val = read_sysreg_s(SYS_DACR32_EL2);   break;
1069         case IFSR32_EL2:        *val = read_sysreg_s(SYS_IFSR32_EL2);   break;
1070         case DBGVCR32_EL2:      *val = read_sysreg_s(SYS_DBGVCR32_EL2); break;
1071         case ZCR_EL1:           *val = read_sysreg_s(SYS_ZCR_EL12);     break;
1072         default:                return false;
1073         }
1074
1075         return true;
1076 }
1077
1078 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
1079 {
1080         /*
1081          * *** VHE ONLY ***
1082          *
1083          * System registers listed in the switch are not restored on every
1084          * entry to the guest but are only restored on vcpu_load.
1085          *
1086          * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
1087          * should never be listed below, because the MPIDR should only be set
1088          * once, before running the VCPU, and never changed later.
1089          */
1090         if (!has_vhe())
1091                 return false;
1092
1093         switch (reg) {
1094         case SCTLR_EL1:         write_sysreg_s(val, SYS_SCTLR_EL12);    break;
1095         case CPACR_EL1:         write_sysreg_s(val, SYS_CPACR_EL12);    break;
1096         case TTBR0_EL1:         write_sysreg_s(val, SYS_TTBR0_EL12);    break;
1097         case TTBR1_EL1:         write_sysreg_s(val, SYS_TTBR1_EL12);    break;
1098         case TCR_EL1:           write_sysreg_s(val, SYS_TCR_EL12);      break;
1099         case TCR2_EL1:          write_sysreg_s(val, SYS_TCR2_EL12);     break;
1100         case PIR_EL1:           write_sysreg_s(val, SYS_PIR_EL12);      break;
1101         case PIRE0_EL1:         write_sysreg_s(val, SYS_PIRE0_EL12);    break;
1102         case POR_EL1:           write_sysreg_s(val, SYS_POR_EL12);      break;
1103         case ESR_EL1:           write_sysreg_s(val, SYS_ESR_EL12);      break;
1104         case AFSR0_EL1:         write_sysreg_s(val, SYS_AFSR0_EL12);    break;
1105         case AFSR1_EL1:         write_sysreg_s(val, SYS_AFSR1_EL12);    break;
1106         case FAR_EL1:           write_sysreg_s(val, SYS_FAR_EL12);      break;
1107         case MAIR_EL1:          write_sysreg_s(val, SYS_MAIR_EL12);     break;
1108         case VBAR_EL1:          write_sysreg_s(val, SYS_VBAR_EL12);     break;
1109         case CONTEXTIDR_EL1:    write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
1110         case TPIDR_EL0:         write_sysreg_s(val, SYS_TPIDR_EL0);     break;
1111         case TPIDRRO_EL0:       write_sysreg_s(val, SYS_TPIDRRO_EL0);   break;
1112         case TPIDR_EL1:         write_sysreg_s(val, SYS_TPIDR_EL1);     break;
1113         case AMAIR_EL1:         write_sysreg_s(val, SYS_AMAIR_EL12);    break;
1114         case CNTKCTL_EL1:       write_sysreg_s(val, SYS_CNTKCTL_EL12);  break;
1115         case ELR_EL1:           write_sysreg_s(val, SYS_ELR_EL12);      break;
1116         case SPSR_EL1:          write_sysreg_s(val, SYS_SPSR_EL12);     break;
1117         case PAR_EL1:           write_sysreg_s(val, SYS_PAR_EL1);       break;
1118         case DACR32_EL2:        write_sysreg_s(val, SYS_DACR32_EL2);    break;
1119         case IFSR32_EL2:        write_sysreg_s(val, SYS_IFSR32_EL2);    break;
1120         case DBGVCR32_EL2:      write_sysreg_s(val, SYS_DBGVCR32_EL2);  break;
1121         case ZCR_EL1:           write_sysreg_s(val, SYS_ZCR_EL12);      break;
1122         default:                return false;
1123         }
1124
1125         return true;
1126 }
1127
1128 struct kvm_vm_stat {
1129         struct kvm_vm_stat_generic generic;
1130 };
1131
1132 struct kvm_vcpu_stat {
1133         struct kvm_vcpu_stat_generic generic;
1134         u64 hvc_exit_stat;
1135         u64 wfe_exit_stat;
1136         u64 wfi_exit_stat;
1137         u64 mmio_exit_user;
1138         u64 mmio_exit_kernel;
1139         u64 signal_exits;
1140         u64 exits;
1141 };
1142
1143 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
1144 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
1145 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1146 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1147
1148 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
1149 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
1150
1151 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1152                               struct kvm_vcpu_events *events);
1153
1154 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1155                               struct kvm_vcpu_events *events);
1156
1157 void kvm_arm_halt_guest(struct kvm *kvm);
1158 void kvm_arm_resume_guest(struct kvm *kvm);
1159
1160 #define vcpu_has_run_once(vcpu) (!!READ_ONCE((vcpu)->pid))
1161
1162 #ifndef __KVM_NVHE_HYPERVISOR__
1163 #define kvm_call_hyp_nvhe(f, ...)                                               \
1164         ({                                                              \
1165                 struct arm_smccc_res res;                               \
1166                                                                         \
1167                 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),               \
1168                                   ##__VA_ARGS__, &res);                 \
1169                 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);                   \
1170                                                                         \
1171                 res.a1;                                                 \
1172         })
1173
1174 /*
1175  * The couple of isb() below are there to guarantee the same behaviour
1176  * on VHE as on !VHE, where the eret to EL1 acts as a context
1177  * synchronization event.
1178  */
1179 #define kvm_call_hyp(f, ...)                                            \
1180         do {                                                            \
1181                 if (has_vhe()) {                                        \
1182                         f(__VA_ARGS__);                                 \
1183                         isb();                                          \
1184                 } else {                                                \
1185                         kvm_call_hyp_nvhe(f, ##__VA_ARGS__);            \
1186                 }                                                       \
1187         } while(0)
1188
1189 #define kvm_call_hyp_ret(f, ...)                                        \
1190         ({                                                              \
1191                 typeof(f(__VA_ARGS__)) ret;                             \
1192                                                                         \
1193                 if (has_vhe()) {                                        \
1194                         ret = f(__VA_ARGS__);                           \
1195                         isb();                                          \
1196                 } else {                                                \
1197                         ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);      \
1198                 }                                                       \
1199                                                                         \
1200                 ret;                                                    \
1201         })
1202 #else /* __KVM_NVHE_HYPERVISOR__ */
1203 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1204 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1205 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1206 #endif /* __KVM_NVHE_HYPERVISOR__ */
1207
1208 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1209 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1210
1211 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1212 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1213 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1214 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1215 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1216 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1217 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1218
1219 void kvm_sys_regs_create_debugfs(struct kvm *kvm);
1220 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1221
1222 int __init kvm_sys_reg_table_init(void);
1223 struct sys_reg_desc;
1224 int __init populate_sysreg_config(const struct sys_reg_desc *sr,
1225                                   unsigned int idx);
1226 int __init populate_nv_trap_config(void);
1227
1228 bool lock_all_vcpus(struct kvm *kvm);
1229 void unlock_all_vcpus(struct kvm *kvm);
1230
1231 void kvm_calculate_traps(struct kvm_vcpu *vcpu);
1232
1233 /* MMIO helpers */
1234 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1235 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1236
1237 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1238 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1239
1240 /*
1241  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1242  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1243  * loaded is considered to be "in guest".
1244  */
1245 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1246 {
1247         return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1248 }
1249
1250 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1251 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1252 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1253
1254 bool kvm_arm_pvtime_supported(void);
1255 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1256                             struct kvm_device_attr *attr);
1257 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1258                             struct kvm_device_attr *attr);
1259 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1260                             struct kvm_device_attr *attr);
1261
1262 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1263 int __init kvm_arm_vmid_alloc_init(void);
1264 void __init kvm_arm_vmid_alloc_free(void);
1265 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1266 void kvm_arm_vmid_clear_active(void);
1267
1268 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1269 {
1270         vcpu_arch->steal.base = INVALID_GPA;
1271 }
1272
1273 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1274 {
1275         return (vcpu_arch->steal.base != INVALID_GPA);
1276 }
1277
1278 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1279
1280 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1281
1282 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1283
1284 /*
1285  * How we access per-CPU host data depends on the where we access it from,
1286  * and the mode we're in:
1287  *
1288  * - VHE and nVHE hypervisor bits use their locally defined instance
1289  *
1290  * - the rest of the kernel use either the VHE or nVHE one, depending on
1291  *   the mode we're running in.
1292  *
1293  *   Unless we're in protected mode, fully deprivileged, and the nVHE
1294  *   per-CPU stuff is exclusively accessible to the protected EL2 code.
1295  *   In this case, the EL1 code uses the *VHE* data as its private state
1296  *   (which makes sense in a way as there shouldn't be any shared state
1297  *   between the host and the hypervisor).
1298  *
1299  * Yes, this is all totally trivial. Shoot me now.
1300  */
1301 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__)
1302 #define host_data_ptr(f)        (&this_cpu_ptr(&kvm_host_data)->f)
1303 #else
1304 #define host_data_ptr(f)                                                \
1305         (static_branch_unlikely(&kvm_protected_mode_initialized) ?      \
1306          &this_cpu_ptr(&kvm_host_data)->f :                             \
1307          &this_cpu_ptr_hyp_sym(kvm_host_data)->f)
1308 #endif
1309
1310 /* Check whether the FP regs are owned by the guest */
1311 static inline bool guest_owns_fp_regs(void)
1312 {
1313         return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED;
1314 }
1315
1316 /* Check whether the FP regs are owned by the host */
1317 static inline bool host_owns_fp_regs(void)
1318 {
1319         return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED;
1320 }
1321
1322 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1323 {
1324         /* The host's MPIDR is immutable, so let's set it up at boot time */
1325         ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1326 }
1327
1328 static inline bool kvm_system_needs_idmapped_vectors(void)
1329 {
1330         return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1331 }
1332
1333 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1334
1335 void kvm_arm_init_debug(void);
1336 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1337 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1338 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1339 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1340
1341 #define kvm_vcpu_os_lock_enabled(vcpu)          \
1342         (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1343
1344 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1345                                struct kvm_device_attr *attr);
1346 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1347                                struct kvm_device_attr *attr);
1348 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1349                                struct kvm_device_attr *attr);
1350
1351 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1352                                struct kvm_arm_copy_mte_tags *copy_tags);
1353 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1354                                     struct kvm_arm_counter_offset *offset);
1355 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1356                                         struct reg_mask_range *range);
1357
1358 /* Guest/host FPSIMD coordination helpers */
1359 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1360 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1361 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1362 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1363 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1364
1365 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1366 {
1367         return (!has_vhe() && attr->exclude_host);
1368 }
1369
1370 /* Flags for host debug state */
1371 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1372 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1373
1374 #ifdef CONFIG_KVM
1375 void kvm_set_pmu_events(u64 set, struct perf_event_attr *attr);
1376 void kvm_clr_pmu_events(u64 clr);
1377 bool kvm_set_pmuserenr(u64 val);
1378 #else
1379 static inline void kvm_set_pmu_events(u64 set, struct perf_event_attr *attr) {}
1380 static inline void kvm_clr_pmu_events(u64 clr) {}
1381 static inline bool kvm_set_pmuserenr(u64 val)
1382 {
1383         return false;
1384 }
1385 #endif
1386
1387 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1388 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1389
1390 int __init kvm_set_ipa_limit(void);
1391 u32 kvm_get_pa_bits(struct kvm *kvm);
1392
1393 #define __KVM_HAVE_ARCH_VM_ALLOC
1394 struct kvm *kvm_arch_alloc_vm(void);
1395
1396 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1397
1398 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1399
1400 #define kvm_vm_is_protected(kvm)        (is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled)
1401
1402 #define vcpu_is_protected(vcpu)         kvm_vm_is_protected((vcpu)->kvm)
1403
1404 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1405 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1406
1407 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1408
1409 #define kvm_has_mte(kvm)                                        \
1410         (system_supports_mte() &&                               \
1411          test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1412
1413 #define kvm_supports_32bit_el0()                                \
1414         (system_supports_32bit_el0() &&                         \
1415          !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1416
1417 #define kvm_vm_has_ran_once(kvm)                                        \
1418         (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1419
1420 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1421 {
1422         return test_bit(feature, ka->vcpu_features);
1423 }
1424
1425 #define vcpu_has_feature(v, f)  __vcpu_has_feature(&(v)->kvm->arch, (f))
1426
1427 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED)
1428
1429 int kvm_trng_call(struct kvm_vcpu *vcpu);
1430 #ifdef CONFIG_KVM
1431 extern phys_addr_t hyp_mem_base;
1432 extern phys_addr_t hyp_mem_size;
1433 void __init kvm_hyp_reserve(void);
1434 #else
1435 static inline void kvm_hyp_reserve(void) { }
1436 #endif
1437
1438 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1439 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1440
1441 static inline u64 *__vm_id_reg(struct kvm_arch *ka, u32 reg)
1442 {
1443         switch (reg) {
1444         case sys_reg(3, 0, 0, 1, 0) ... sys_reg(3, 0, 0, 7, 7):
1445                 return &ka->id_regs[IDREG_IDX(reg)];
1446         case SYS_CTR_EL0:
1447                 return &ka->ctr_el0;
1448         default:
1449                 WARN_ON_ONCE(1);
1450                 return NULL;
1451         }
1452 }
1453
1454 #define kvm_read_vm_id_reg(kvm, reg)                                    \
1455         ({ u64 __val = *__vm_id_reg(&(kvm)->arch, reg); __val; })
1456
1457 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val);
1458
1459 #define __expand_field_sign_unsigned(id, fld, val)                      \
1460         ((u64)SYS_FIELD_VALUE(id, fld, val))
1461
1462 #define __expand_field_sign_signed(id, fld, val)                        \
1463         ({                                                              \
1464                 u64 __val = SYS_FIELD_VALUE(id, fld, val);              \
1465                 sign_extend64(__val, id##_##fld##_WIDTH - 1);           \
1466         })
1467
1468 #define get_idreg_field_unsigned(kvm, id, fld)                          \
1469         ({                                                              \
1470                 u64 __val = kvm_read_vm_id_reg((kvm), SYS_##id);        \
1471                 FIELD_GET(id##_##fld##_MASK, __val);                    \
1472         })
1473
1474 #define get_idreg_field_signed(kvm, id, fld)                            \
1475         ({                                                              \
1476                 u64 __val = get_idreg_field_unsigned(kvm, id, fld);     \
1477                 sign_extend64(__val, id##_##fld##_WIDTH - 1);           \
1478         })
1479
1480 #define get_idreg_field_enum(kvm, id, fld)                              \
1481         get_idreg_field_unsigned(kvm, id, fld)
1482
1483 #define kvm_cmp_feat_signed(kvm, id, fld, op, limit)                    \
1484         (get_idreg_field_signed((kvm), id, fld) op __expand_field_sign_signed(id, fld, limit))
1485
1486 #define kvm_cmp_feat_unsigned(kvm, id, fld, op, limit)                  \
1487         (get_idreg_field_unsigned((kvm), id, fld) op __expand_field_sign_unsigned(id, fld, limit))
1488
1489 #define kvm_cmp_feat(kvm, id, fld, op, limit)                           \
1490         (id##_##fld##_SIGNED ?                                          \
1491          kvm_cmp_feat_signed(kvm, id, fld, op, limit) :                 \
1492          kvm_cmp_feat_unsigned(kvm, id, fld, op, limit))
1493
1494 #define kvm_has_feat(kvm, id, fld, limit)                               \
1495         kvm_cmp_feat(kvm, id, fld, >=, limit)
1496
1497 #define kvm_has_feat_enum(kvm, id, fld, val)                            \
1498         kvm_cmp_feat_unsigned(kvm, id, fld, ==, val)
1499
1500 #define kvm_has_feat_range(kvm, id, fld, min, max)                      \
1501         (kvm_cmp_feat(kvm, id, fld, >=, min) &&                         \
1502         kvm_cmp_feat(kvm, id, fld, <=, max))
1503
1504 /* Check for a given level of PAuth support */
1505 #define kvm_has_pauth(k, l)                                             \
1506         ({                                                              \
1507                 bool pa, pi, pa3;                                       \
1508                                                                         \
1509                 pa  = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l);      \
1510                 pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP);    \
1511                 pi  = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l);      \
1512                 pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP);    \
1513                 pa3  = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l);    \
1514                 pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP);  \
1515                                                                         \
1516                 (pa + pi + pa3) == 1;                                   \
1517         })
1518
1519 #define kvm_has_fpmr(k)                                 \
1520         (system_supports_fpmr() &&                      \
1521          kvm_has_feat((k), ID_AA64PFR2_EL1, FPMR, IMP))
1522
1523 #define kvm_has_tcr2(k)                         \
1524         (kvm_has_feat((k), ID_AA64MMFR3_EL1, TCRX, IMP))
1525
1526 #define kvm_has_s1pie(k)                                \
1527         (kvm_has_feat((k), ID_AA64MMFR3_EL1, S1PIE, IMP))
1528
1529 #define kvm_has_s1poe(k)                                \
1530         (kvm_has_feat((k), ID_AA64MMFR3_EL1, S1POE, IMP))
1531
1532 #endif /* __ARM64_KVM_HOST_H__ */
This page took 0.113409 seconds and 4 git commands to generate.