Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-2.6-microblaze.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 <marc.zyngier@arm.com>
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 <c.dall@virtualopensystems.com>
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
55 #define KVM_DIRTY_LOG_MANUAL_CAPS   (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \
56                                      KVM_DIRTY_LOG_INITIALLY_SET)
57
58 #define KVM_HAVE_MMU_RWLOCK
59
60 /*
61  * Mode of operation configurable with kvm-arm.mode early param.
62  * See Documentation/admin-guide/kernel-parameters.txt for more information.
63  */
64 enum kvm_mode {
65         KVM_MODE_DEFAULT,
66         KVM_MODE_PROTECTED,
67         KVM_MODE_NV,
68         KVM_MODE_NONE,
69 };
70 #ifdef CONFIG_KVM
71 enum kvm_mode kvm_get_mode(void);
72 #else
73 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; };
74 #endif
75
76 DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
77
78 extern unsigned int __ro_after_init kvm_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 struct kvm_arch_memory_slot {
194 };
195
196 /**
197  * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests
198  *
199  * @std_bmap: Bitmap of standard secure service calls
200  * @std_hyp_bmap: Bitmap of standard hypervisor service calls
201  * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls
202  */
203 struct kvm_smccc_features {
204         unsigned long std_bmap;
205         unsigned long std_hyp_bmap;
206         unsigned long vendor_hyp_bmap;
207 };
208
209 typedef unsigned int pkvm_handle_t;
210
211 struct kvm_protected_vm {
212         pkvm_handle_t handle;
213         struct kvm_hyp_memcache teardown_mc;
214 };
215
216 struct kvm_mpidr_data {
217         u64                     mpidr_mask;
218         DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx);
219 };
220
221 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr)
222 {
223         unsigned long mask = data->mpidr_mask;
224         u64 aff = mpidr & MPIDR_HWID_BITMASK;
225         int nbits, bit, bit_idx = 0;
226         u16 index = 0;
227
228         /*
229          * If this looks like RISC-V's BEXT or x86's PEXT
230          * instructions, it isn't by accident.
231          */
232         nbits = fls(mask);
233         for_each_set_bit(bit, &mask, nbits) {
234                 index |= (aff & BIT(bit)) >> (bit - bit_idx);
235                 bit_idx++;
236         }
237
238         return index;
239 }
240
241 struct kvm_sysreg_masks;
242
243 enum fgt_group_id {
244         __NO_FGT_GROUP__,
245         HFGxTR_GROUP,
246         HDFGRTR_GROUP,
247         HDFGWTR_GROUP = HDFGRTR_GROUP,
248         HFGITR_GROUP,
249         HAFGRTR_GROUP,
250
251         /* Must be last */
252         __NR_FGT_GROUP_IDS__
253 };
254
255 struct kvm_arch {
256         struct kvm_s2_mmu mmu;
257
258         /*
259          * Fine-Grained UNDEF, mimicking the FGT layout defined by the
260          * architecture. We track them globally, as we present the
261          * same feature-set to all vcpus.
262          *
263          * Index 0 is currently spare.
264          */
265         u64 fgu[__NR_FGT_GROUP_IDS__];
266
267         /* Interrupt controller */
268         struct vgic_dist        vgic;
269
270         /* Timers */
271         struct arch_timer_vm_data timer_data;
272
273         /* Mandated version of PSCI */
274         u32 psci_version;
275
276         /* Protects VM-scoped configuration data */
277         struct mutex config_lock;
278
279         /*
280          * If we encounter a data abort without valid instruction syndrome
281          * information, report this to user space.  User space can (and
282          * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
283          * supported.
284          */
285 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER      0
286         /* Memory Tagging Extension enabled for the guest */
287 #define KVM_ARCH_FLAG_MTE_ENABLED                       1
288         /* At least one vCPU has ran in the VM */
289 #define KVM_ARCH_FLAG_HAS_RAN_ONCE                      2
290         /* The vCPU feature set for the VM is configured */
291 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED          3
292         /* PSCI SYSTEM_SUSPEND enabled for the guest */
293 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED            4
294         /* VM counter offset */
295 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET                 5
296         /* Timer PPIs made immutable */
297 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE              6
298         /* Initial ID reg values loaded */
299 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED               7
300         /* Fine-Grained UNDEF initialised */
301 #define KVM_ARCH_FLAG_FGU_INITIALIZED                   8
302         unsigned long flags;
303
304         /* VM-wide vCPU feature set */
305         DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
306
307         /* MPIDR to vcpu index mapping, optional */
308         struct kvm_mpidr_data *mpidr_data;
309
310         /*
311          * VM-wide PMU filter, implemented as a bitmap and big enough for
312          * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
313          */
314         unsigned long *pmu_filter;
315         struct arm_pmu *arm_pmu;
316
317         cpumask_var_t supported_cpus;
318
319         /* PMCR_EL0.N value for the guest */
320         u8 pmcr_n;
321
322         /* Iterator for idreg debugfs */
323         u8      idreg_debugfs_iter;
324
325         /* Hypercall features firmware registers' descriptor */
326         struct kvm_smccc_features smccc_feat;
327         struct maple_tree smccc_filter;
328
329         /*
330          * Emulated CPU ID registers per VM
331          * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
332          * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
333          *
334          * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
335          * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
336          */
337 #define IDREG_IDX(id)           (((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
338 #define IDX_IDREG(idx)          sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask)
339 #define IDREG(kvm, id)          ((kvm)->arch.id_regs[IDREG_IDX(id)])
340 #define KVM_ARM_ID_REG_NUM      (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
341         u64 id_regs[KVM_ARM_ID_REG_NUM];
342
343         /* Masks for VNCR-baked sysregs */
344         struct kvm_sysreg_masks *sysreg_masks;
345
346         /*
347          * For an untrusted host VM, 'pkvm.handle' is used to lookup
348          * the associated pKVM instance in the hypervisor.
349          */
350         struct kvm_protected_vm pkvm;
351 };
352
353 struct kvm_vcpu_fault_info {
354         u64 esr_el2;            /* Hyp Syndrom Register */
355         u64 far_el2;            /* Hyp Fault Address Register */
356         u64 hpfar_el2;          /* Hyp IPA Fault Address Register */
357         u64 disr_el1;           /* Deferred [SError] Status Register */
358 };
359
360 /*
361  * VNCR() just places the VNCR_capable registers in the enum after
362  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
363  * from the VNCR base. As we don't require the enum to be otherwise ordered,
364  * we need the terrible hack below to ensure that we correctly size the
365  * sys_regs array, no matter what.
366  *
367  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
368  * treasure trove of bit hacks:
369  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
370  */
371 #define __MAX__(x,y)    ((x) ^ (((x) ^ (y)) & -((x) < (y))))
372 #define VNCR(r)                                         \
373         __before_##r,                                   \
374         r = __VNCR_START__ + ((VNCR_ ## r) / 8),        \
375         __after_##r = __MAX__(__before_##r - 1, r)
376
377 enum vcpu_sysreg {
378         __INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
379         MPIDR_EL1,      /* MultiProcessor Affinity Register */
380         CLIDR_EL1,      /* Cache Level ID Register */
381         CSSELR_EL1,     /* Cache Size Selection Register */
382         TPIDR_EL0,      /* Thread ID, User R/W */
383         TPIDRRO_EL0,    /* Thread ID, User R/O */
384         TPIDR_EL1,      /* Thread ID, Privileged */
385         CNTKCTL_EL1,    /* Timer Control Register (EL1) */
386         PAR_EL1,        /* Physical Address Register */
387         MDCCINT_EL1,    /* Monitor Debug Comms Channel Interrupt Enable Reg */
388         OSLSR_EL1,      /* OS Lock Status Register */
389         DISR_EL1,       /* Deferred Interrupt Status Register */
390
391         /* Performance Monitors Registers */
392         PMCR_EL0,       /* Control Register */
393         PMSELR_EL0,     /* Event Counter Selection Register */
394         PMEVCNTR0_EL0,  /* Event Counter Register (0-30) */
395         PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
396         PMCCNTR_EL0,    /* Cycle Counter Register */
397         PMEVTYPER0_EL0, /* Event Type Register (0-30) */
398         PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
399         PMCCFILTR_EL0,  /* Cycle Count Filter Register */
400         PMCNTENSET_EL0, /* Count Enable Set Register */
401         PMINTENSET_EL1, /* Interrupt Enable Set Register */
402         PMOVSSET_EL0,   /* Overflow Flag Status Set Register */
403         PMUSERENR_EL0,  /* User Enable Register */
404
405         /* Pointer Authentication Registers in a strict increasing order. */
406         APIAKEYLO_EL1,
407         APIAKEYHI_EL1,
408         APIBKEYLO_EL1,
409         APIBKEYHI_EL1,
410         APDAKEYLO_EL1,
411         APDAKEYHI_EL1,
412         APDBKEYLO_EL1,
413         APDBKEYHI_EL1,
414         APGAKEYLO_EL1,
415         APGAKEYHI_EL1,
416
417         /* Memory Tagging Extension registers */
418         RGSR_EL1,       /* Random Allocation Tag Seed Register */
419         GCR_EL1,        /* Tag Control Register */
420         TFSRE0_EL1,     /* Tag Fault Status Register (EL0) */
421
422         /* 32bit specific registers. */
423         DACR32_EL2,     /* Domain Access Control Register */
424         IFSR32_EL2,     /* Instruction Fault Status Register */
425         FPEXC32_EL2,    /* Floating-Point Exception Control Register */
426         DBGVCR32_EL2,   /* Debug Vector Catch Register */
427
428         /* EL2 registers */
429         SCTLR_EL2,      /* System Control Register (EL2) */
430         ACTLR_EL2,      /* Auxiliary Control Register (EL2) */
431         MDCR_EL2,       /* Monitor Debug Configuration Register (EL2) */
432         CPTR_EL2,       /* Architectural Feature Trap Register (EL2) */
433         HACR_EL2,       /* Hypervisor Auxiliary Control Register */
434         TTBR0_EL2,      /* Translation Table Base Register 0 (EL2) */
435         TTBR1_EL2,      /* Translation Table Base Register 1 (EL2) */
436         TCR_EL2,        /* Translation Control Register (EL2) */
437         SPSR_EL2,       /* EL2 saved program status register */
438         ELR_EL2,        /* EL2 exception link register */
439         AFSR0_EL2,      /* Auxiliary Fault Status Register 0 (EL2) */
440         AFSR1_EL2,      /* Auxiliary Fault Status Register 1 (EL2) */
441         ESR_EL2,        /* Exception Syndrome Register (EL2) */
442         FAR_EL2,        /* Fault Address Register (EL2) */
443         HPFAR_EL2,      /* Hypervisor IPA Fault Address Register */
444         MAIR_EL2,       /* Memory Attribute Indirection Register (EL2) */
445         AMAIR_EL2,      /* Auxiliary Memory Attribute Indirection Register (EL2) */
446         VBAR_EL2,       /* Vector Base Address Register (EL2) */
447         RVBAR_EL2,      /* Reset Vector Base Address Register */
448         CONTEXTIDR_EL2, /* Context ID Register (EL2) */
449         CNTHCTL_EL2,    /* Counter-timer Hypervisor Control register */
450         SP_EL2,         /* EL2 Stack Pointer */
451         CNTHP_CTL_EL2,
452         CNTHP_CVAL_EL2,
453         CNTHV_CTL_EL2,
454         CNTHV_CVAL_EL2,
455
456         __VNCR_START__, /* Any VNCR-capable reg goes after this point */
457
458         VNCR(SCTLR_EL1),/* System Control Register */
459         VNCR(ACTLR_EL1),/* Auxiliary Control Register */
460         VNCR(CPACR_EL1),/* Coprocessor Access Control */
461         VNCR(ZCR_EL1),  /* SVE Control */
462         VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
463         VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
464         VNCR(TCR_EL1),  /* Translation Control Register */
465         VNCR(TCR2_EL1), /* Extended Translation Control Register */
466         VNCR(ESR_EL1),  /* Exception Syndrome Register */
467         VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
468         VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
469         VNCR(FAR_EL1),  /* Fault Address Register */
470         VNCR(MAIR_EL1), /* Memory Attribute Indirection Register */
471         VNCR(VBAR_EL1), /* Vector Base Address Register */
472         VNCR(CONTEXTIDR_EL1),   /* Context ID Register */
473         VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
474         VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
475         VNCR(ELR_EL1),
476         VNCR(SP_EL1),
477         VNCR(SPSR_EL1),
478         VNCR(TFSR_EL1), /* Tag Fault Status Register (EL1) */
479         VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
480         VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
481         VNCR(HCR_EL2),  /* Hypervisor Configuration Register */
482         VNCR(HSTR_EL2), /* Hypervisor System Trap Register */
483         VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
484         VNCR(VTCR_EL2), /* Virtualization Translation Control Register */
485         VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
486         VNCR(HCRX_EL2), /* Extended Hypervisor Configuration Register */
487
488         /* Permission Indirection Extension registers */
489         VNCR(PIR_EL1),   /* Permission Indirection Register 1 (EL1) */
490         VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
491
492         VNCR(HFGRTR_EL2),
493         VNCR(HFGWTR_EL2),
494         VNCR(HFGITR_EL2),
495         VNCR(HDFGRTR_EL2),
496         VNCR(HDFGWTR_EL2),
497         VNCR(HAFGRTR_EL2),
498
499         VNCR(CNTVOFF_EL2),
500         VNCR(CNTV_CVAL_EL0),
501         VNCR(CNTV_CTL_EL0),
502         VNCR(CNTP_CVAL_EL0),
503         VNCR(CNTP_CTL_EL0),
504
505         NR_SYS_REGS     /* Nothing after this line! */
506 };
507
508 struct kvm_sysreg_masks {
509         struct {
510                 u64     res0;
511                 u64     res1;
512         } mask[NR_SYS_REGS - __VNCR_START__];
513 };
514
515 struct kvm_cpu_context {
516         struct user_pt_regs regs;       /* sp = sp_el0 */
517
518         u64     spsr_abt;
519         u64     spsr_und;
520         u64     spsr_irq;
521         u64     spsr_fiq;
522
523         struct user_fpsimd_state fp_regs;
524
525         u64 sys_regs[NR_SYS_REGS];
526
527         struct kvm_vcpu *__hyp_running_vcpu;
528
529         /* This pointer has to be 4kB aligned. */
530         u64 *vncr_array;
531 };
532
533 struct kvm_host_data {
534         struct kvm_cpu_context host_ctxt;
535 };
536
537 struct kvm_host_psci_config {
538         /* PSCI version used by host. */
539         u32 version;
540         u32 smccc_version;
541
542         /* Function IDs used by host if version is v0.1. */
543         struct psci_0_1_function_ids function_ids_0_1;
544
545         bool psci_0_1_cpu_suspend_implemented;
546         bool psci_0_1_cpu_on_implemented;
547         bool psci_0_1_cpu_off_implemented;
548         bool psci_0_1_migrate_implemented;
549 };
550
551 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
552 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
553
554 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
555 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
556
557 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
558 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
559
560 struct vcpu_reset_state {
561         unsigned long   pc;
562         unsigned long   r0;
563         bool            be;
564         bool            reset;
565 };
566
567 struct kvm_vcpu_arch {
568         struct kvm_cpu_context ctxt;
569
570         /*
571          * Guest floating point state
572          *
573          * The architecture has two main floating point extensions,
574          * the original FPSIMD and SVE.  These have overlapping
575          * register views, with the FPSIMD V registers occupying the
576          * low 128 bits of the SVE Z registers.  When the core
577          * floating point code saves the register state of a task it
578          * records which view it saved in fp_type.
579          */
580         void *sve_state;
581         enum fp_type fp_type;
582         unsigned int sve_max_vl;
583         u64 svcr;
584         u64 fpmr;
585
586         /* Stage 2 paging state used by the hardware on next switch */
587         struct kvm_s2_mmu *hw_mmu;
588
589         /* Values of trap registers for the guest. */
590         u64 hcr_el2;
591         u64 hcrx_el2;
592         u64 mdcr_el2;
593         u64 cptr_el2;
594
595         /* Values of trap registers for the host before guest entry. */
596         u64 mdcr_el2_host;
597
598         /* Exception Information */
599         struct kvm_vcpu_fault_info fault;
600
601         /* Ownership of the FP regs */
602         enum {
603                 FP_STATE_FREE,
604                 FP_STATE_HOST_OWNED,
605                 FP_STATE_GUEST_OWNED,
606         } fp_state;
607
608         /* Configuration flags, set once and for all before the vcpu can run */
609         u8 cflags;
610
611         /* Input flags to the hypervisor code, potentially cleared after use */
612         u8 iflags;
613
614         /* State flags for kernel bookkeeping, unused by the hypervisor code */
615         u8 sflags;
616
617         /*
618          * Don't run the guest (internal implementation need).
619          *
620          * Contrary to the flags above, this is set/cleared outside of
621          * a vcpu context, and thus cannot be mixed with the flags
622          * themselves (or the flag accesses need to be made atomic).
623          */
624         bool pause;
625
626         /*
627          * We maintain more than a single set of debug registers to support
628          * debugging the guest from the host and to maintain separate host and
629          * guest state during world switches. vcpu_debug_state are the debug
630          * registers of the vcpu as the guest sees them.  host_debug_state are
631          * the host registers which are saved and restored during
632          * world switches. external_debug_state contains the debug
633          * values we want to debug the guest. This is set via the
634          * KVM_SET_GUEST_DEBUG ioctl.
635          *
636          * debug_ptr points to the set of debug registers that should be loaded
637          * onto the hardware when running the guest.
638          */
639         struct kvm_guest_debug_arch *debug_ptr;
640         struct kvm_guest_debug_arch vcpu_debug_state;
641         struct kvm_guest_debug_arch external_debug_state;
642
643         struct user_fpsimd_state *host_fpsimd_state;    /* hyp VA */
644         struct task_struct *parent_task;
645
646         struct {
647                 /* {Break,watch}point registers */
648                 struct kvm_guest_debug_arch regs;
649                 /* Statistical profiling extension */
650                 u64 pmscr_el1;
651                 /* Self-hosted trace */
652                 u64 trfcr_el1;
653         } host_debug_state;
654
655         /* VGIC state */
656         struct vgic_cpu vgic_cpu;
657         struct arch_timer_cpu timer_cpu;
658         struct kvm_pmu pmu;
659
660         /*
661          * Guest registers we preserve during guest debugging.
662          *
663          * These shadow registers are updated by the kvm_handle_sys_reg
664          * trap handler if the guest accesses or updates them while we
665          * are using guest debug.
666          */
667         struct {
668                 u32     mdscr_el1;
669                 bool    pstate_ss;
670         } guest_debug_preserved;
671
672         /* vcpu power state */
673         struct kvm_mp_state mp_state;
674         spinlock_t mp_state_lock;
675
676         /* Cache some mmu pages needed inside spinlock regions */
677         struct kvm_mmu_memory_cache mmu_page_cache;
678
679         /* Virtual SError ESR to restore when HCR_EL2.VSE is set */
680         u64 vsesr_el2;
681
682         /* Additional reset state */
683         struct vcpu_reset_state reset_state;
684
685         /* Guest PV state */
686         struct {
687                 u64 last_steal;
688                 gpa_t base;
689         } steal;
690
691         /* Per-vcpu CCSIDR override or NULL */
692         u32 *ccsidr;
693 };
694
695 /*
696  * Each 'flag' is composed of a comma-separated triplet:
697  *
698  * - the flag-set it belongs to in the vcpu->arch structure
699  * - the value for that flag
700  * - the mask for that flag
701  *
702  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
703  * unpack_vcpu_flag() extract the flag value from the triplet for
704  * direct use outside of the flag accessors.
705  */
706 #define __vcpu_single_flag(_set, _f)    _set, (_f), (_f)
707
708 #define __unpack_flag(_set, _f, _m)     _f
709 #define unpack_vcpu_flag(...)           __unpack_flag(__VA_ARGS__)
710
711 #define __build_check_flag(v, flagset, f, m)                    \
712         do {                                                    \
713                 typeof(v->arch.flagset) *_fset;                 \
714                                                                 \
715                 /* Check that the flags fit in the mask */      \
716                 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \
717                 /* Check that the flags fit in the type */      \
718                 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \
719         } while (0)
720
721 #define __vcpu_get_flag(v, flagset, f, m)                       \
722         ({                                                      \
723                 __build_check_flag(v, flagset, f, m);           \
724                                                                 \
725                 READ_ONCE(v->arch.flagset) & (m);               \
726         })
727
728 /*
729  * Note that the set/clear accessors must be preempt-safe in order to
730  * avoid nesting them with load/put which also manipulate flags...
731  */
732 #ifdef __KVM_NVHE_HYPERVISOR__
733 /* the nVHE hypervisor is always non-preemptible */
734 #define __vcpu_flags_preempt_disable()
735 #define __vcpu_flags_preempt_enable()
736 #else
737 #define __vcpu_flags_preempt_disable()  preempt_disable()
738 #define __vcpu_flags_preempt_enable()   preempt_enable()
739 #endif
740
741 #define __vcpu_set_flag(v, flagset, f, m)                       \
742         do {                                                    \
743                 typeof(v->arch.flagset) *fset;                  \
744                                                                 \
745                 __build_check_flag(v, flagset, f, m);           \
746                                                                 \
747                 fset = &v->arch.flagset;                        \
748                 __vcpu_flags_preempt_disable();                 \
749                 if (HWEIGHT(m) > 1)                             \
750                         *fset &= ~(m);                          \
751                 *fset |= (f);                                   \
752                 __vcpu_flags_preempt_enable();                  \
753         } while (0)
754
755 #define __vcpu_clear_flag(v, flagset, f, m)                     \
756         do {                                                    \
757                 typeof(v->arch.flagset) *fset;                  \
758                                                                 \
759                 __build_check_flag(v, flagset, f, m);           \
760                                                                 \
761                 fset = &v->arch.flagset;                        \
762                 __vcpu_flags_preempt_disable();                 \
763                 *fset &= ~(m);                                  \
764                 __vcpu_flags_preempt_enable();                  \
765         } while (0)
766
767 #define vcpu_get_flag(v, ...)   __vcpu_get_flag((v), __VA_ARGS__)
768 #define vcpu_set_flag(v, ...)   __vcpu_set_flag((v), __VA_ARGS__)
769 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__)
770
771 /* SVE exposed to guest */
772 #define GUEST_HAS_SVE           __vcpu_single_flag(cflags, BIT(0))
773 /* SVE config completed */
774 #define VCPU_SVE_FINALIZED      __vcpu_single_flag(cflags, BIT(1))
775 /* PTRAUTH exposed to guest */
776 #define GUEST_HAS_PTRAUTH       __vcpu_single_flag(cflags, BIT(2))
777 /* KVM_ARM_VCPU_INIT completed */
778 #define VCPU_INITIALIZED        __vcpu_single_flag(cflags, BIT(3))
779
780 /* Exception pending */
781 #define PENDING_EXCEPTION       __vcpu_single_flag(iflags, BIT(0))
782 /*
783  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
784  * be set together with an exception...
785  */
786 #define INCREMENT_PC            __vcpu_single_flag(iflags, BIT(1))
787 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
788 #define EXCEPT_MASK             __vcpu_single_flag(iflags, GENMASK(3, 1))
789
790 /* Helpers to encode exceptions with minimum fuss */
791 #define __EXCEPT_MASK_VAL       unpack_vcpu_flag(EXCEPT_MASK)
792 #define __EXCEPT_SHIFT          __builtin_ctzl(__EXCEPT_MASK_VAL)
793 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
794
795 /*
796  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
797  * values:
798  *
799  * For AArch32 EL1:
800  */
801 #define EXCEPT_AA32_UND         __vcpu_except_flags(0)
802 #define EXCEPT_AA32_IABT        __vcpu_except_flags(1)
803 #define EXCEPT_AA32_DABT        __vcpu_except_flags(2)
804 /* For AArch64: */
805 #define EXCEPT_AA64_EL1_SYNC    __vcpu_except_flags(0)
806 #define EXCEPT_AA64_EL1_IRQ     __vcpu_except_flags(1)
807 #define EXCEPT_AA64_EL1_FIQ     __vcpu_except_flags(2)
808 #define EXCEPT_AA64_EL1_SERR    __vcpu_except_flags(3)
809 /* For AArch64 with NV: */
810 #define EXCEPT_AA64_EL2_SYNC    __vcpu_except_flags(4)
811 #define EXCEPT_AA64_EL2_IRQ     __vcpu_except_flags(5)
812 #define EXCEPT_AA64_EL2_FIQ     __vcpu_except_flags(6)
813 #define EXCEPT_AA64_EL2_SERR    __vcpu_except_flags(7)
814 /* Guest debug is live */
815 #define DEBUG_DIRTY             __vcpu_single_flag(iflags, BIT(4))
816 /* Save SPE context if active  */
817 #define DEBUG_STATE_SAVE_SPE    __vcpu_single_flag(iflags, BIT(5))
818 /* Save TRBE context if active  */
819 #define DEBUG_STATE_SAVE_TRBE   __vcpu_single_flag(iflags, BIT(6))
820 /* vcpu running in HYP context */
821 #define VCPU_HYP_CONTEXT        __vcpu_single_flag(iflags, BIT(7))
822
823 /* SVE enabled for host EL0 */
824 #define HOST_SVE_ENABLED        __vcpu_single_flag(sflags, BIT(0))
825 /* SME enabled for EL0 */
826 #define HOST_SME_ENABLED        __vcpu_single_flag(sflags, BIT(1))
827 /* Physical CPU not in supported_cpus */
828 #define ON_UNSUPPORTED_CPU      __vcpu_single_flag(sflags, BIT(2))
829 /* WFIT instruction trapped */
830 #define IN_WFIT                 __vcpu_single_flag(sflags, BIT(3))
831 /* vcpu system registers loaded on physical CPU */
832 #define SYSREGS_ON_CPU          __vcpu_single_flag(sflags, BIT(4))
833 /* Software step state is Active-pending */
834 #define DBG_SS_ACTIVE_PENDING   __vcpu_single_flag(sflags, BIT(5))
835 /* PMUSERENR for the guest EL0 is on physical CPU */
836 #define PMUSERENR_ON_CPU        __vcpu_single_flag(sflags, BIT(6))
837 /* WFI instruction trapped */
838 #define IN_WFI                  __vcpu_single_flag(sflags, BIT(7))
839
840
841 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
842 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +      \
843                              sve_ffr_offset((vcpu)->arch.sve_max_vl))
844
845 #define vcpu_sve_max_vq(vcpu)   sve_vq_from_vl((vcpu)->arch.sve_max_vl)
846
847 #define vcpu_sve_state_size(vcpu) ({                                    \
848         size_t __size_ret;                                              \
849         unsigned int __vcpu_vq;                                         \
850                                                                         \
851         if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {          \
852                 __size_ret = 0;                                         \
853         } else {                                                        \
854                 __vcpu_vq = vcpu_sve_max_vq(vcpu);                      \
855                 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);              \
856         }                                                               \
857                                                                         \
858         __size_ret;                                                     \
859 })
860
861 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
862                                  KVM_GUESTDBG_USE_SW_BP | \
863                                  KVM_GUESTDBG_USE_HW | \
864                                  KVM_GUESTDBG_SINGLESTEP)
865
866 #define vcpu_has_sve(vcpu) (system_supports_sve() &&                    \
867                             vcpu_get_flag(vcpu, GUEST_HAS_SVE))
868
869 #ifdef CONFIG_ARM64_PTR_AUTH
870 #define vcpu_has_ptrauth(vcpu)                                          \
871         ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||                \
872           cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&               \
873           vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
874 #else
875 #define vcpu_has_ptrauth(vcpu)          false
876 #endif
877
878 #define vcpu_on_unsupported_cpu(vcpu)                                   \
879         vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
880
881 #define vcpu_set_on_unsupported_cpu(vcpu)                               \
882         vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
883
884 #define vcpu_clear_on_unsupported_cpu(vcpu)                             \
885         vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
886
887 #define vcpu_gp_regs(v)         (&(v)->arch.ctxt.regs)
888
889 /*
890  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
891  * memory backed version of a register, and not the one most recently
892  * accessed by a running VCPU.  For example, for userspace access or
893  * for system registers that are never context switched, but only
894  * emulated.
895  *
896  * Don't bother with VNCR-based accesses in the nVHE code, it has no
897  * business dealing with NV.
898  */
899 static inline u64 *__ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
900 {
901 #if !defined (__KVM_NVHE_HYPERVISOR__)
902         if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
903                      r >= __VNCR_START__ && ctxt->vncr_array))
904                 return &ctxt->vncr_array[r - __VNCR_START__];
905 #endif
906         return (u64 *)&ctxt->sys_regs[r];
907 }
908
909 #define ctxt_sys_reg(c,r)       (*__ctxt_sys_reg(c,r))
910
911 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg);
912 #define __vcpu_sys_reg(v,r)                                             \
913         (*({                                                            \
914                 const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt;   \
915                 u64 *__r = __ctxt_sys_reg(ctxt, (r));                   \
916                 if (vcpu_has_nv((v)) && (r) >= __VNCR_START__)          \
917                         *__r = kvm_vcpu_sanitise_vncr_reg((v), (r));    \
918                 __r;                                                    \
919         }))
920
921 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
922 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
923
924 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
925 {
926         /*
927          * *** VHE ONLY ***
928          *
929          * System registers listed in the switch are not saved on every
930          * exit from the guest but are only saved on vcpu_put.
931          *
932          * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
933          * should never be listed below, because the guest cannot modify its
934          * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
935          * thread when emulating cross-VCPU communication.
936          */
937         if (!has_vhe())
938                 return false;
939
940         switch (reg) {
941         case SCTLR_EL1:         *val = read_sysreg_s(SYS_SCTLR_EL12);   break;
942         case CPACR_EL1:         *val = read_sysreg_s(SYS_CPACR_EL12);   break;
943         case TTBR0_EL1:         *val = read_sysreg_s(SYS_TTBR0_EL12);   break;
944         case TTBR1_EL1:         *val = read_sysreg_s(SYS_TTBR1_EL12);   break;
945         case TCR_EL1:           *val = read_sysreg_s(SYS_TCR_EL12);     break;
946         case ESR_EL1:           *val = read_sysreg_s(SYS_ESR_EL12);     break;
947         case AFSR0_EL1:         *val = read_sysreg_s(SYS_AFSR0_EL12);   break;
948         case AFSR1_EL1:         *val = read_sysreg_s(SYS_AFSR1_EL12);   break;
949         case FAR_EL1:           *val = read_sysreg_s(SYS_FAR_EL12);     break;
950         case MAIR_EL1:          *val = read_sysreg_s(SYS_MAIR_EL12);    break;
951         case VBAR_EL1:          *val = read_sysreg_s(SYS_VBAR_EL12);    break;
952         case CONTEXTIDR_EL1:    *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
953         case TPIDR_EL0:         *val = read_sysreg_s(SYS_TPIDR_EL0);    break;
954         case TPIDRRO_EL0:       *val = read_sysreg_s(SYS_TPIDRRO_EL0);  break;
955         case TPIDR_EL1:         *val = read_sysreg_s(SYS_TPIDR_EL1);    break;
956         case AMAIR_EL1:         *val = read_sysreg_s(SYS_AMAIR_EL12);   break;
957         case CNTKCTL_EL1:       *val = read_sysreg_s(SYS_CNTKCTL_EL12); break;
958         case ELR_EL1:           *val = read_sysreg_s(SYS_ELR_EL12);     break;
959         case SPSR_EL1:          *val = read_sysreg_s(SYS_SPSR_EL12);    break;
960         case PAR_EL1:           *val = read_sysreg_par();               break;
961         case DACR32_EL2:        *val = read_sysreg_s(SYS_DACR32_EL2);   break;
962         case IFSR32_EL2:        *val = read_sysreg_s(SYS_IFSR32_EL2);   break;
963         case DBGVCR32_EL2:      *val = read_sysreg_s(SYS_DBGVCR32_EL2); break;
964         default:                return false;
965         }
966
967         return true;
968 }
969
970 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
971 {
972         /*
973          * *** VHE ONLY ***
974          *
975          * System registers listed in the switch are not restored on every
976          * entry to the guest but are only restored on vcpu_load.
977          *
978          * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
979          * should never be listed below, because the MPIDR should only be set
980          * once, before running the VCPU, and never changed later.
981          */
982         if (!has_vhe())
983                 return false;
984
985         switch (reg) {
986         case SCTLR_EL1:         write_sysreg_s(val, SYS_SCTLR_EL12);    break;
987         case CPACR_EL1:         write_sysreg_s(val, SYS_CPACR_EL12);    break;
988         case TTBR0_EL1:         write_sysreg_s(val, SYS_TTBR0_EL12);    break;
989         case TTBR1_EL1:         write_sysreg_s(val, SYS_TTBR1_EL12);    break;
990         case TCR_EL1:           write_sysreg_s(val, SYS_TCR_EL12);      break;
991         case ESR_EL1:           write_sysreg_s(val, SYS_ESR_EL12);      break;
992         case AFSR0_EL1:         write_sysreg_s(val, SYS_AFSR0_EL12);    break;
993         case AFSR1_EL1:         write_sysreg_s(val, SYS_AFSR1_EL12);    break;
994         case FAR_EL1:           write_sysreg_s(val, SYS_FAR_EL12);      break;
995         case MAIR_EL1:          write_sysreg_s(val, SYS_MAIR_EL12);     break;
996         case VBAR_EL1:          write_sysreg_s(val, SYS_VBAR_EL12);     break;
997         case CONTEXTIDR_EL1:    write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
998         case TPIDR_EL0:         write_sysreg_s(val, SYS_TPIDR_EL0);     break;
999         case TPIDRRO_EL0:       write_sysreg_s(val, SYS_TPIDRRO_EL0);   break;
1000         case TPIDR_EL1:         write_sysreg_s(val, SYS_TPIDR_EL1);     break;
1001         case AMAIR_EL1:         write_sysreg_s(val, SYS_AMAIR_EL12);    break;
1002         case CNTKCTL_EL1:       write_sysreg_s(val, SYS_CNTKCTL_EL12);  break;
1003         case ELR_EL1:           write_sysreg_s(val, SYS_ELR_EL12);      break;
1004         case SPSR_EL1:          write_sysreg_s(val, SYS_SPSR_EL12);     break;
1005         case PAR_EL1:           write_sysreg_s(val, SYS_PAR_EL1);       break;
1006         case DACR32_EL2:        write_sysreg_s(val, SYS_DACR32_EL2);    break;
1007         case IFSR32_EL2:        write_sysreg_s(val, SYS_IFSR32_EL2);    break;
1008         case DBGVCR32_EL2:      write_sysreg_s(val, SYS_DBGVCR32_EL2);  break;
1009         default:                return false;
1010         }
1011
1012         return true;
1013 }
1014
1015 struct kvm_vm_stat {
1016         struct kvm_vm_stat_generic generic;
1017 };
1018
1019 struct kvm_vcpu_stat {
1020         struct kvm_vcpu_stat_generic generic;
1021         u64 hvc_exit_stat;
1022         u64 wfe_exit_stat;
1023         u64 wfi_exit_stat;
1024         u64 mmio_exit_user;
1025         u64 mmio_exit_kernel;
1026         u64 signal_exits;
1027         u64 exits;
1028 };
1029
1030 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
1031 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
1032 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1033 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1034
1035 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
1036 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
1037
1038 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1039                               struct kvm_vcpu_events *events);
1040
1041 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1042                               struct kvm_vcpu_events *events);
1043
1044 void kvm_arm_halt_guest(struct kvm *kvm);
1045 void kvm_arm_resume_guest(struct kvm *kvm);
1046
1047 #define vcpu_has_run_once(vcpu) !!rcu_access_pointer((vcpu)->pid)
1048
1049 #ifndef __KVM_NVHE_HYPERVISOR__
1050 #define kvm_call_hyp_nvhe(f, ...)                                               \
1051         ({                                                              \
1052                 struct arm_smccc_res res;                               \
1053                                                                         \
1054                 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),               \
1055                                   ##__VA_ARGS__, &res);                 \
1056                 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);                   \
1057                                                                         \
1058                 res.a1;                                                 \
1059         })
1060
1061 /*
1062  * The couple of isb() below are there to guarantee the same behaviour
1063  * on VHE as on !VHE, where the eret to EL1 acts as a context
1064  * synchronization event.
1065  */
1066 #define kvm_call_hyp(f, ...)                                            \
1067         do {                                                            \
1068                 if (has_vhe()) {                                        \
1069                         f(__VA_ARGS__);                                 \
1070                         isb();                                          \
1071                 } else {                                                \
1072                         kvm_call_hyp_nvhe(f, ##__VA_ARGS__);            \
1073                 }                                                       \
1074         } while(0)
1075
1076 #define kvm_call_hyp_ret(f, ...)                                        \
1077         ({                                                              \
1078                 typeof(f(__VA_ARGS__)) ret;                             \
1079                                                                         \
1080                 if (has_vhe()) {                                        \
1081                         ret = f(__VA_ARGS__);                           \
1082                         isb();                                          \
1083                 } else {                                                \
1084                         ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);      \
1085                 }                                                       \
1086                                                                         \
1087                 ret;                                                    \
1088         })
1089 #else /* __KVM_NVHE_HYPERVISOR__ */
1090 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1091 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1092 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1093 #endif /* __KVM_NVHE_HYPERVISOR__ */
1094
1095 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1096 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1097
1098 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1099 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1100 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1101 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1102 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1103 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1104 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1105
1106 void kvm_sys_regs_create_debugfs(struct kvm *kvm);
1107 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1108
1109 int __init kvm_sys_reg_table_init(void);
1110 struct sys_reg_desc;
1111 int __init populate_sysreg_config(const struct sys_reg_desc *sr,
1112                                   unsigned int idx);
1113 int __init populate_nv_trap_config(void);
1114
1115 bool lock_all_vcpus(struct kvm *kvm);
1116 void unlock_all_vcpus(struct kvm *kvm);
1117
1118 void kvm_init_sysreg(struct kvm_vcpu *);
1119
1120 /* MMIO helpers */
1121 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1122 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1123
1124 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1125 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1126
1127 /*
1128  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1129  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1130  * loaded is considered to be "in guest".
1131  */
1132 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1133 {
1134         return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1135 }
1136
1137 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1138 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1139 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1140
1141 bool kvm_arm_pvtime_supported(void);
1142 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1143                             struct kvm_device_attr *attr);
1144 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1145                             struct kvm_device_attr *attr);
1146 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1147                             struct kvm_device_attr *attr);
1148
1149 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1150 int __init kvm_arm_vmid_alloc_init(void);
1151 void __init kvm_arm_vmid_alloc_free(void);
1152 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1153 void kvm_arm_vmid_clear_active(void);
1154
1155 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1156 {
1157         vcpu_arch->steal.base = INVALID_GPA;
1158 }
1159
1160 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1161 {
1162         return (vcpu_arch->steal.base != INVALID_GPA);
1163 }
1164
1165 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1166
1167 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1168
1169 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1170
1171 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1172 {
1173         /* The host's MPIDR is immutable, so let's set it up at boot time */
1174         ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1175 }
1176
1177 static inline bool kvm_system_needs_idmapped_vectors(void)
1178 {
1179         return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1180 }
1181
1182 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1183 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1184
1185 void kvm_arm_init_debug(void);
1186 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1187 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1188 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1189 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1190
1191 #define kvm_vcpu_os_lock_enabled(vcpu)          \
1192         (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1193
1194 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1195                                struct kvm_device_attr *attr);
1196 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1197                                struct kvm_device_attr *attr);
1198 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1199                                struct kvm_device_attr *attr);
1200
1201 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1202                                struct kvm_arm_copy_mte_tags *copy_tags);
1203 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1204                                     struct kvm_arm_counter_offset *offset);
1205 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1206                                         struct reg_mask_range *range);
1207
1208 /* Guest/host FPSIMD coordination helpers */
1209 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1210 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1211 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1212 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1213 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1214 void kvm_vcpu_unshare_task_fp(struct kvm_vcpu *vcpu);
1215
1216 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1217 {
1218         return (!has_vhe() && attr->exclude_host);
1219 }
1220
1221 /* Flags for host debug state */
1222 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1223 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1224
1225 #ifdef CONFIG_KVM
1226 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1227 void kvm_clr_pmu_events(u32 clr);
1228 bool kvm_set_pmuserenr(u64 val);
1229 #else
1230 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1231 static inline void kvm_clr_pmu_events(u32 clr) {}
1232 static inline bool kvm_set_pmuserenr(u64 val)
1233 {
1234         return false;
1235 }
1236 #endif
1237
1238 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1239 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1240
1241 int __init kvm_set_ipa_limit(void);
1242
1243 #define __KVM_HAVE_ARCH_VM_ALLOC
1244 struct kvm *kvm_arch_alloc_vm(void);
1245
1246 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1247
1248 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1249
1250 static inline bool kvm_vm_is_protected(struct kvm *kvm)
1251 {
1252         return false;
1253 }
1254
1255 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1256 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1257
1258 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1259
1260 #define kvm_has_mte(kvm)                                        \
1261         (system_supports_mte() &&                               \
1262          test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1263
1264 #define kvm_supports_32bit_el0()                                \
1265         (system_supports_32bit_el0() &&                         \
1266          !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1267
1268 #define kvm_vm_has_ran_once(kvm)                                        \
1269         (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1270
1271 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1272 {
1273         return test_bit(feature, ka->vcpu_features);
1274 }
1275
1276 #define vcpu_has_feature(v, f)  __vcpu_has_feature(&(v)->kvm->arch, (f))
1277
1278 int kvm_trng_call(struct kvm_vcpu *vcpu);
1279 #ifdef CONFIG_KVM
1280 extern phys_addr_t hyp_mem_base;
1281 extern phys_addr_t hyp_mem_size;
1282 void __init kvm_hyp_reserve(void);
1283 #else
1284 static inline void kvm_hyp_reserve(void) { }
1285 #endif
1286
1287 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1288 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1289
1290 #define __expand_field_sign_unsigned(id, fld, val)                      \
1291         ((u64)SYS_FIELD_VALUE(id, fld, val))
1292
1293 #define __expand_field_sign_signed(id, fld, val)                        \
1294         ({                                                              \
1295                 u64 __val = SYS_FIELD_VALUE(id, fld, val);              \
1296                 sign_extend64(__val, id##_##fld##_WIDTH - 1);           \
1297         })
1298
1299 #define expand_field_sign(id, fld, val)                                 \
1300         (id##_##fld##_SIGNED ?                                          \
1301          __expand_field_sign_signed(id, fld, val) :                     \
1302          __expand_field_sign_unsigned(id, fld, val))
1303
1304 #define get_idreg_field_unsigned(kvm, id, fld)                          \
1305         ({                                                              \
1306                 u64 __val = IDREG((kvm), SYS_##id);                     \
1307                 FIELD_GET(id##_##fld##_MASK, __val);                    \
1308         })
1309
1310 #define get_idreg_field_signed(kvm, id, fld)                            \
1311         ({                                                              \
1312                 u64 __val = get_idreg_field_unsigned(kvm, id, fld);     \
1313                 sign_extend64(__val, id##_##fld##_WIDTH - 1);           \
1314         })
1315
1316 #define get_idreg_field_enum(kvm, id, fld)                              \
1317         get_idreg_field_unsigned(kvm, id, fld)
1318
1319 #define get_idreg_field(kvm, id, fld)                                   \
1320         (id##_##fld##_SIGNED ?                                          \
1321          get_idreg_field_signed(kvm, id, fld) :                         \
1322          get_idreg_field_unsigned(kvm, id, fld))
1323
1324 #define kvm_has_feat(kvm, id, fld, limit)                               \
1325         (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit))
1326
1327 #define kvm_has_feat_enum(kvm, id, fld, val)                            \
1328         (get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val))
1329
1330 #define kvm_has_feat_range(kvm, id, fld, min, max)                      \
1331         (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \
1332          get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max))
1333
1334 #endif /* __ARM64_KVM_HOST_H__ */