xref: /linux/arch/arm64/include/asm/kvm_host.h (revision a674fefd17324fc467f043568e738b80ca22f2b4)
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 extern unsigned int __ro_after_init kvm_host_sve_max_vl;
80 int __init kvm_arm_init_sve(void);
81 
82 u32 __attribute_const__ kvm_target_cpu(void);
83 void kvm_reset_vcpu(struct kvm_vcpu *vcpu);
84 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu);
85 
86 struct kvm_hyp_memcache {
87 	phys_addr_t head;
88 	unsigned long nr_pages;
89 };
90 
91 static inline void push_hyp_memcache(struct kvm_hyp_memcache *mc,
92 				     phys_addr_t *p,
93 				     phys_addr_t (*to_pa)(void *virt))
94 {
95 	*p = mc->head;
96 	mc->head = to_pa(p);
97 	mc->nr_pages++;
98 }
99 
100 static inline void *pop_hyp_memcache(struct kvm_hyp_memcache *mc,
101 				     void *(*to_va)(phys_addr_t phys))
102 {
103 	phys_addr_t *p = to_va(mc->head);
104 
105 	if (!mc->nr_pages)
106 		return NULL;
107 
108 	mc->head = *p;
109 	mc->nr_pages--;
110 
111 	return p;
112 }
113 
114 static inline int __topup_hyp_memcache(struct kvm_hyp_memcache *mc,
115 				       unsigned long min_pages,
116 				       void *(*alloc_fn)(void *arg),
117 				       phys_addr_t (*to_pa)(void *virt),
118 				       void *arg)
119 {
120 	while (mc->nr_pages < min_pages) {
121 		phys_addr_t *p = alloc_fn(arg);
122 
123 		if (!p)
124 			return -ENOMEM;
125 		push_hyp_memcache(mc, p, to_pa);
126 	}
127 
128 	return 0;
129 }
130 
131 static inline void __free_hyp_memcache(struct kvm_hyp_memcache *mc,
132 				       void (*free_fn)(void *virt, void *arg),
133 				       void *(*to_va)(phys_addr_t phys),
134 				       void *arg)
135 {
136 	while (mc->nr_pages)
137 		free_fn(pop_hyp_memcache(mc, to_va), arg);
138 }
139 
140 void free_hyp_memcache(struct kvm_hyp_memcache *mc);
141 int topup_hyp_memcache(struct kvm_hyp_memcache *mc, unsigned long min_pages);
142 
143 struct kvm_vmid {
144 	atomic64_t id;
145 };
146 
147 struct kvm_s2_mmu {
148 	struct kvm_vmid vmid;
149 
150 	/*
151 	 * stage2 entry level table
152 	 *
153 	 * Two kvm_s2_mmu structures in the same VM can point to the same
154 	 * pgd here.  This happens when running a guest using a
155 	 * translation regime that isn't affected by its own stage-2
156 	 * translation, such as a non-VHE hypervisor running at vEL2, or
157 	 * for vEL1/EL0 with vHCR_EL2.VM == 0.  In that case, we use the
158 	 * canonical stage-2 page tables.
159 	 */
160 	phys_addr_t	pgd_phys;
161 	struct kvm_pgtable *pgt;
162 
163 	/*
164 	 * VTCR value used on the host. For a non-NV guest (or a NV
165 	 * guest that runs in a context where its own S2 doesn't
166 	 * apply), its T0SZ value reflects that of the IPA size.
167 	 *
168 	 * For a shadow S2 MMU, T0SZ reflects the PARange exposed to
169 	 * the guest.
170 	 */
171 	u64	vtcr;
172 
173 	/* The last vcpu id that ran on each physical CPU */
174 	int __percpu *last_vcpu_ran;
175 
176 #define KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT 0
177 	/*
178 	 * Memory cache used to split
179 	 * KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE worth of huge pages. It
180 	 * is used to allocate stage2 page tables while splitting huge
181 	 * pages. The choice of KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE
182 	 * influences both the capacity of the split page cache, and
183 	 * how often KVM reschedules. Be wary of raising CHUNK_SIZE
184 	 * too high.
185 	 *
186 	 * Protected by kvm->slots_lock.
187 	 */
188 	struct kvm_mmu_memory_cache split_page_cache;
189 	uint64_t split_page_chunk_size;
190 
191 	struct kvm_arch *arch;
192 };
193 
194 struct kvm_arch_memory_slot {
195 };
196 
197 /**
198  * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests
199  *
200  * @std_bmap: Bitmap of standard secure service calls
201  * @std_hyp_bmap: Bitmap of standard hypervisor service calls
202  * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls
203  */
204 struct kvm_smccc_features {
205 	unsigned long std_bmap;
206 	unsigned long std_hyp_bmap;
207 	unsigned long vendor_hyp_bmap;
208 };
209 
210 typedef unsigned int pkvm_handle_t;
211 
212 struct kvm_protected_vm {
213 	pkvm_handle_t handle;
214 	struct kvm_hyp_memcache teardown_mc;
215 	bool enabled;
216 };
217 
218 struct kvm_mpidr_data {
219 	u64			mpidr_mask;
220 	DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx);
221 };
222 
223 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr)
224 {
225 	unsigned long index = 0, mask = data->mpidr_mask;
226 	unsigned long aff = mpidr & MPIDR_HWID_BITMASK;
227 
228 	bitmap_gather(&index, &aff, &mask, fls(mask));
229 
230 	return index;
231 }
232 
233 struct kvm_sysreg_masks;
234 
235 enum fgt_group_id {
236 	__NO_FGT_GROUP__,
237 	HFGxTR_GROUP,
238 	HDFGRTR_GROUP,
239 	HDFGWTR_GROUP = HDFGRTR_GROUP,
240 	HFGITR_GROUP,
241 	HAFGRTR_GROUP,
242 
243 	/* Must be last */
244 	__NR_FGT_GROUP_IDS__
245 };
246 
247 struct kvm_arch {
248 	struct kvm_s2_mmu mmu;
249 
250 	/*
251 	 * Fine-Grained UNDEF, mimicking the FGT layout defined by the
252 	 * architecture. We track them globally, as we present the
253 	 * same feature-set to all vcpus.
254 	 *
255 	 * Index 0 is currently spare.
256 	 */
257 	u64 fgu[__NR_FGT_GROUP_IDS__];
258 
259 	/* Interrupt controller */
260 	struct vgic_dist	vgic;
261 
262 	/* Timers */
263 	struct arch_timer_vm_data timer_data;
264 
265 	/* Mandated version of PSCI */
266 	u32 psci_version;
267 
268 	/* Protects VM-scoped configuration data */
269 	struct mutex config_lock;
270 
271 	/*
272 	 * If we encounter a data abort without valid instruction syndrome
273 	 * information, report this to user space.  User space can (and
274 	 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
275 	 * supported.
276 	 */
277 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER	0
278 	/* Memory Tagging Extension enabled for the guest */
279 #define KVM_ARCH_FLAG_MTE_ENABLED			1
280 	/* At least one vCPU has ran in the VM */
281 #define KVM_ARCH_FLAG_HAS_RAN_ONCE			2
282 	/* The vCPU feature set for the VM is configured */
283 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED		3
284 	/* PSCI SYSTEM_SUSPEND enabled for the guest */
285 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED		4
286 	/* VM counter offset */
287 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET			5
288 	/* Timer PPIs made immutable */
289 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE		6
290 	/* Initial ID reg values loaded */
291 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED		7
292 	/* Fine-Grained UNDEF initialised */
293 #define KVM_ARCH_FLAG_FGU_INITIALIZED			8
294 	unsigned long flags;
295 
296 	/* VM-wide vCPU feature set */
297 	DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
298 
299 	/* MPIDR to vcpu index mapping, optional */
300 	struct kvm_mpidr_data *mpidr_data;
301 
302 	/*
303 	 * VM-wide PMU filter, implemented as a bitmap and big enough for
304 	 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
305 	 */
306 	unsigned long *pmu_filter;
307 	struct arm_pmu *arm_pmu;
308 
309 	cpumask_var_t supported_cpus;
310 
311 	/* PMCR_EL0.N value for the guest */
312 	u8 pmcr_n;
313 
314 	/* Iterator for idreg debugfs */
315 	u8	idreg_debugfs_iter;
316 
317 	/* Hypercall features firmware registers' descriptor */
318 	struct kvm_smccc_features smccc_feat;
319 	struct maple_tree smccc_filter;
320 
321 	/*
322 	 * Emulated CPU ID registers per VM
323 	 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
324 	 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
325 	 *
326 	 * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
327 	 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
328 	 */
329 #define IDREG_IDX(id)		(((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
330 #define IDX_IDREG(idx)		sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask)
331 #define IDREG(kvm, id)		((kvm)->arch.id_regs[IDREG_IDX(id)])
332 #define KVM_ARM_ID_REG_NUM	(IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
333 	u64 id_regs[KVM_ARM_ID_REG_NUM];
334 
335 	/* Masks for VNCR-baked sysregs */
336 	struct kvm_sysreg_masks	*sysreg_masks;
337 
338 	/*
339 	 * For an untrusted host VM, 'pkvm.handle' is used to lookup
340 	 * the associated pKVM instance in the hypervisor.
341 	 */
342 	struct kvm_protected_vm pkvm;
343 };
344 
345 struct kvm_vcpu_fault_info {
346 	u64 esr_el2;		/* Hyp Syndrom Register */
347 	u64 far_el2;		/* Hyp Fault Address Register */
348 	u64 hpfar_el2;		/* Hyp IPA Fault Address Register */
349 	u64 disr_el1;		/* Deferred [SError] Status Register */
350 };
351 
352 /*
353  * VNCR() just places the VNCR_capable registers in the enum after
354  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
355  * from the VNCR base. As we don't require the enum to be otherwise ordered,
356  * we need the terrible hack below to ensure that we correctly size the
357  * sys_regs array, no matter what.
358  *
359  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
360  * treasure trove of bit hacks:
361  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
362  */
363 #define __MAX__(x,y)	((x) ^ (((x) ^ (y)) & -((x) < (y))))
364 #define VNCR(r)						\
365 	__before_##r,					\
366 	r = __VNCR_START__ + ((VNCR_ ## r) / 8),	\
367 	__after_##r = __MAX__(__before_##r - 1, r)
368 
369 enum vcpu_sysreg {
370 	__INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
371 	MPIDR_EL1,	/* MultiProcessor Affinity Register */
372 	CLIDR_EL1,	/* Cache Level ID Register */
373 	CSSELR_EL1,	/* Cache Size Selection Register */
374 	TPIDR_EL0,	/* Thread ID, User R/W */
375 	TPIDRRO_EL0,	/* Thread ID, User R/O */
376 	TPIDR_EL1,	/* Thread ID, Privileged */
377 	CNTKCTL_EL1,	/* Timer Control Register (EL1) */
378 	PAR_EL1,	/* Physical Address Register */
379 	MDCCINT_EL1,	/* Monitor Debug Comms Channel Interrupt Enable Reg */
380 	OSLSR_EL1,	/* OS Lock Status Register */
381 	DISR_EL1,	/* Deferred Interrupt Status Register */
382 
383 	/* Performance Monitors Registers */
384 	PMCR_EL0,	/* Control Register */
385 	PMSELR_EL0,	/* Event Counter Selection Register */
386 	PMEVCNTR0_EL0,	/* Event Counter Register (0-30) */
387 	PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
388 	PMCCNTR_EL0,	/* Cycle Counter Register */
389 	PMEVTYPER0_EL0,	/* Event Type Register (0-30) */
390 	PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
391 	PMCCFILTR_EL0,	/* Cycle Count Filter Register */
392 	PMCNTENSET_EL0,	/* Count Enable Set Register */
393 	PMINTENSET_EL1,	/* Interrupt Enable Set Register */
394 	PMOVSSET_EL0,	/* Overflow Flag Status Set Register */
395 	PMUSERENR_EL0,	/* User Enable Register */
396 
397 	/* Pointer Authentication Registers in a strict increasing order. */
398 	APIAKEYLO_EL1,
399 	APIAKEYHI_EL1,
400 	APIBKEYLO_EL1,
401 	APIBKEYHI_EL1,
402 	APDAKEYLO_EL1,
403 	APDAKEYHI_EL1,
404 	APDBKEYLO_EL1,
405 	APDBKEYHI_EL1,
406 	APGAKEYLO_EL1,
407 	APGAKEYHI_EL1,
408 
409 	/* Memory Tagging Extension registers */
410 	RGSR_EL1,	/* Random Allocation Tag Seed Register */
411 	GCR_EL1,	/* Tag Control Register */
412 	TFSRE0_EL1,	/* Tag Fault Status Register (EL0) */
413 
414 	/* 32bit specific registers. */
415 	DACR32_EL2,	/* Domain Access Control Register */
416 	IFSR32_EL2,	/* Instruction Fault Status Register */
417 	FPEXC32_EL2,	/* Floating-Point Exception Control Register */
418 	DBGVCR32_EL2,	/* Debug Vector Catch Register */
419 
420 	/* EL2 registers */
421 	SCTLR_EL2,	/* System Control Register (EL2) */
422 	ACTLR_EL2,	/* Auxiliary Control Register (EL2) */
423 	MDCR_EL2,	/* Monitor Debug Configuration Register (EL2) */
424 	CPTR_EL2,	/* Architectural Feature Trap Register (EL2) */
425 	HACR_EL2,	/* Hypervisor Auxiliary Control Register */
426 	TTBR0_EL2,	/* Translation Table Base Register 0 (EL2) */
427 	TTBR1_EL2,	/* Translation Table Base Register 1 (EL2) */
428 	TCR_EL2,	/* Translation Control Register (EL2) */
429 	SPSR_EL2,	/* EL2 saved program status register */
430 	ELR_EL2,	/* EL2 exception link register */
431 	AFSR0_EL2,	/* Auxiliary Fault Status Register 0 (EL2) */
432 	AFSR1_EL2,	/* Auxiliary Fault Status Register 1 (EL2) */
433 	ESR_EL2,	/* Exception Syndrome Register (EL2) */
434 	FAR_EL2,	/* Fault Address Register (EL2) */
435 	HPFAR_EL2,	/* Hypervisor IPA Fault Address Register */
436 	MAIR_EL2,	/* Memory Attribute Indirection Register (EL2) */
437 	AMAIR_EL2,	/* Auxiliary Memory Attribute Indirection Register (EL2) */
438 	VBAR_EL2,	/* Vector Base Address Register (EL2) */
439 	RVBAR_EL2,	/* Reset Vector Base Address Register */
440 	CONTEXTIDR_EL2,	/* Context ID Register (EL2) */
441 	CNTHCTL_EL2,	/* Counter-timer Hypervisor Control register */
442 	SP_EL2,		/* EL2 Stack Pointer */
443 	CNTHP_CTL_EL2,
444 	CNTHP_CVAL_EL2,
445 	CNTHV_CTL_EL2,
446 	CNTHV_CVAL_EL2,
447 
448 	__VNCR_START__,	/* Any VNCR-capable reg goes after this point */
449 
450 	VNCR(SCTLR_EL1),/* System Control Register */
451 	VNCR(ACTLR_EL1),/* Auxiliary Control Register */
452 	VNCR(CPACR_EL1),/* Coprocessor Access Control */
453 	VNCR(ZCR_EL1),	/* SVE Control */
454 	VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
455 	VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
456 	VNCR(TCR_EL1),	/* Translation Control Register */
457 	VNCR(TCR2_EL1),	/* Extended Translation Control Register */
458 	VNCR(ESR_EL1),	/* Exception Syndrome Register */
459 	VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
460 	VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
461 	VNCR(FAR_EL1),	/* Fault Address Register */
462 	VNCR(MAIR_EL1),	/* Memory Attribute Indirection Register */
463 	VNCR(VBAR_EL1),	/* Vector Base Address Register */
464 	VNCR(CONTEXTIDR_EL1),	/* Context ID Register */
465 	VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
466 	VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
467 	VNCR(ELR_EL1),
468 	VNCR(SP_EL1),
469 	VNCR(SPSR_EL1),
470 	VNCR(TFSR_EL1),	/* Tag Fault Status Register (EL1) */
471 	VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
472 	VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
473 	VNCR(HCR_EL2),	/* Hypervisor Configuration Register */
474 	VNCR(HSTR_EL2),	/* Hypervisor System Trap Register */
475 	VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
476 	VNCR(VTCR_EL2),	/* Virtualization Translation Control Register */
477 	VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
478 	VNCR(HCRX_EL2),	/* Extended Hypervisor Configuration Register */
479 
480 	/* Permission Indirection Extension registers */
481 	VNCR(PIR_EL1),	 /* Permission Indirection Register 1 (EL1) */
482 	VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
483 
484 	VNCR(HFGRTR_EL2),
485 	VNCR(HFGWTR_EL2),
486 	VNCR(HFGITR_EL2),
487 	VNCR(HDFGRTR_EL2),
488 	VNCR(HDFGWTR_EL2),
489 	VNCR(HAFGRTR_EL2),
490 
491 	VNCR(CNTVOFF_EL2),
492 	VNCR(CNTV_CVAL_EL0),
493 	VNCR(CNTV_CTL_EL0),
494 	VNCR(CNTP_CVAL_EL0),
495 	VNCR(CNTP_CTL_EL0),
496 
497 	NR_SYS_REGS	/* Nothing after this line! */
498 };
499 
500 struct kvm_sysreg_masks {
501 	struct {
502 		u64	res0;
503 		u64	res1;
504 	} mask[NR_SYS_REGS - __VNCR_START__];
505 };
506 
507 struct kvm_cpu_context {
508 	struct user_pt_regs regs;	/* sp = sp_el0 */
509 
510 	u64	spsr_abt;
511 	u64	spsr_und;
512 	u64	spsr_irq;
513 	u64	spsr_fiq;
514 
515 	struct user_fpsimd_state fp_regs;
516 
517 	u64 sys_regs[NR_SYS_REGS];
518 
519 	struct kvm_vcpu *__hyp_running_vcpu;
520 
521 	/* This pointer has to be 4kB aligned. */
522 	u64 *vncr_array;
523 };
524 
525 struct cpu_sve_state {
526 	__u64 zcr_el1;
527 
528 	/*
529 	 * Ordering is important since __sve_save_state/__sve_restore_state
530 	 * relies on it.
531 	 */
532 	__u32 fpsr;
533 	__u32 fpcr;
534 
535 	/* Must be SVE_VQ_BYTES (128 bit) aligned. */
536 	__u8 sve_regs[];
537 };
538 
539 /*
540  * This structure is instantiated on a per-CPU basis, and contains
541  * data that is:
542  *
543  * - tied to a single physical CPU, and
544  * - either have a lifetime that does not extend past vcpu_put()
545  * - or is an invariant for the lifetime of the system
546  *
547  * Use host_data_ptr(field) as a way to access a pointer to such a
548  * field.
549  */
550 struct kvm_host_data {
551 	struct kvm_cpu_context host_ctxt;
552 
553 	/*
554 	 * All pointers in this union are hyp VA.
555 	 * sve_state is only used in pKVM and if system_supports_sve().
556 	 */
557 	union {
558 		struct user_fpsimd_state *fpsimd_state;
559 		struct cpu_sve_state *sve_state;
560 	};
561 
562 	/* Ownership of the FP regs */
563 	enum {
564 		FP_STATE_FREE,
565 		FP_STATE_HOST_OWNED,
566 		FP_STATE_GUEST_OWNED,
567 	} fp_owner;
568 
569 	/*
570 	 * host_debug_state contains the host registers which are
571 	 * saved and restored during world switches.
572 	 */
573 	 struct {
574 		/* {Break,watch}point registers */
575 		struct kvm_guest_debug_arch regs;
576 		/* Statistical profiling extension */
577 		u64 pmscr_el1;
578 		/* Self-hosted trace */
579 		u64 trfcr_el1;
580 		/* Values of trap registers for the host before guest entry. */
581 		u64 mdcr_el2;
582 	} host_debug_state;
583 };
584 
585 struct kvm_host_psci_config {
586 	/* PSCI version used by host. */
587 	u32 version;
588 	u32 smccc_version;
589 
590 	/* Function IDs used by host if version is v0.1. */
591 	struct psci_0_1_function_ids function_ids_0_1;
592 
593 	bool psci_0_1_cpu_suspend_implemented;
594 	bool psci_0_1_cpu_on_implemented;
595 	bool psci_0_1_cpu_off_implemented;
596 	bool psci_0_1_migrate_implemented;
597 };
598 
599 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
600 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
601 
602 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
603 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
604 
605 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
606 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
607 
608 struct vcpu_reset_state {
609 	unsigned long	pc;
610 	unsigned long	r0;
611 	bool		be;
612 	bool		reset;
613 };
614 
615 struct kvm_vcpu_arch {
616 	struct kvm_cpu_context ctxt;
617 
618 	/*
619 	 * Guest floating point state
620 	 *
621 	 * The architecture has two main floating point extensions,
622 	 * the original FPSIMD and SVE.  These have overlapping
623 	 * register views, with the FPSIMD V registers occupying the
624 	 * low 128 bits of the SVE Z registers.  When the core
625 	 * floating point code saves the register state of a task it
626 	 * records which view it saved in fp_type.
627 	 */
628 	void *sve_state;
629 	enum fp_type fp_type;
630 	unsigned int sve_max_vl;
631 	u64 svcr;
632 	u64 fpmr;
633 
634 	/* Stage 2 paging state used by the hardware on next switch */
635 	struct kvm_s2_mmu *hw_mmu;
636 
637 	/* Values of trap registers for the guest. */
638 	u64 hcr_el2;
639 	u64 hcrx_el2;
640 	u64 mdcr_el2;
641 	u64 cptr_el2;
642 
643 	/* Exception Information */
644 	struct kvm_vcpu_fault_info fault;
645 
646 	/* Configuration flags, set once and for all before the vcpu can run */
647 	u8 cflags;
648 
649 	/* Input flags to the hypervisor code, potentially cleared after use */
650 	u8 iflags;
651 
652 	/* State flags for kernel bookkeeping, unused by the hypervisor code */
653 	u8 sflags;
654 
655 	/*
656 	 * Don't run the guest (internal implementation need).
657 	 *
658 	 * Contrary to the flags above, this is set/cleared outside of
659 	 * a vcpu context, and thus cannot be mixed with the flags
660 	 * themselves (or the flag accesses need to be made atomic).
661 	 */
662 	bool pause;
663 
664 	/*
665 	 * We maintain more than a single set of debug registers to support
666 	 * debugging the guest from the host and to maintain separate host and
667 	 * guest state during world switches. vcpu_debug_state are the debug
668 	 * registers of the vcpu as the guest sees them.
669 	 *
670 	 * external_debug_state contains the debug values we want to debug the
671 	 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl.
672 	 *
673 	 * debug_ptr points to the set of debug registers that should be loaded
674 	 * onto the hardware when running the guest.
675 	 */
676 	struct kvm_guest_debug_arch *debug_ptr;
677 	struct kvm_guest_debug_arch vcpu_debug_state;
678 	struct kvm_guest_debug_arch external_debug_state;
679 
680 	/* VGIC state */
681 	struct vgic_cpu vgic_cpu;
682 	struct arch_timer_cpu timer_cpu;
683 	struct kvm_pmu pmu;
684 
685 	/*
686 	 * Guest registers we preserve during guest debugging.
687 	 *
688 	 * These shadow registers are updated by the kvm_handle_sys_reg
689 	 * trap handler if the guest accesses or updates them while we
690 	 * are using guest debug.
691 	 */
692 	struct {
693 		u32	mdscr_el1;
694 		bool	pstate_ss;
695 	} guest_debug_preserved;
696 
697 	/* vcpu power state */
698 	struct kvm_mp_state mp_state;
699 	spinlock_t mp_state_lock;
700 
701 	/* Cache some mmu pages needed inside spinlock regions */
702 	struct kvm_mmu_memory_cache mmu_page_cache;
703 
704 	/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
705 	u64 vsesr_el2;
706 
707 	/* Additional reset state */
708 	struct vcpu_reset_state	reset_state;
709 
710 	/* Guest PV state */
711 	struct {
712 		u64 last_steal;
713 		gpa_t base;
714 	} steal;
715 
716 	/* Per-vcpu CCSIDR override or NULL */
717 	u32 *ccsidr;
718 };
719 
720 /*
721  * Each 'flag' is composed of a comma-separated triplet:
722  *
723  * - the flag-set it belongs to in the vcpu->arch structure
724  * - the value for that flag
725  * - the mask for that flag
726  *
727  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
728  * unpack_vcpu_flag() extract the flag value from the triplet for
729  * direct use outside of the flag accessors.
730  */
731 #define __vcpu_single_flag(_set, _f)	_set, (_f), (_f)
732 
733 #define __unpack_flag(_set, _f, _m)	_f
734 #define unpack_vcpu_flag(...)		__unpack_flag(__VA_ARGS__)
735 
736 #define __build_check_flag(v, flagset, f, m)			\
737 	do {							\
738 		typeof(v->arch.flagset) *_fset;			\
739 								\
740 		/* Check that the flags fit in the mask */	\
741 		BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m)));	\
742 		/* Check that the flags fit in the type */	\
743 		BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m));	\
744 	} while (0)
745 
746 #define __vcpu_get_flag(v, flagset, f, m)			\
747 	({							\
748 		__build_check_flag(v, flagset, f, m);		\
749 								\
750 		READ_ONCE(v->arch.flagset) & (m);		\
751 	})
752 
753 /*
754  * Note that the set/clear accessors must be preempt-safe in order to
755  * avoid nesting them with load/put which also manipulate flags...
756  */
757 #ifdef __KVM_NVHE_HYPERVISOR__
758 /* the nVHE hypervisor is always non-preemptible */
759 #define __vcpu_flags_preempt_disable()
760 #define __vcpu_flags_preempt_enable()
761 #else
762 #define __vcpu_flags_preempt_disable()	preempt_disable()
763 #define __vcpu_flags_preempt_enable()	preempt_enable()
764 #endif
765 
766 #define __vcpu_set_flag(v, flagset, f, m)			\
767 	do {							\
768 		typeof(v->arch.flagset) *fset;			\
769 								\
770 		__build_check_flag(v, flagset, f, m);		\
771 								\
772 		fset = &v->arch.flagset;			\
773 		__vcpu_flags_preempt_disable();			\
774 		if (HWEIGHT(m) > 1)				\
775 			*fset &= ~(m);				\
776 		*fset |= (f);					\
777 		__vcpu_flags_preempt_enable();			\
778 	} while (0)
779 
780 #define __vcpu_clear_flag(v, flagset, f, m)			\
781 	do {							\
782 		typeof(v->arch.flagset) *fset;			\
783 								\
784 		__build_check_flag(v, flagset, f, m);		\
785 								\
786 		fset = &v->arch.flagset;			\
787 		__vcpu_flags_preempt_disable();			\
788 		*fset &= ~(m);					\
789 		__vcpu_flags_preempt_enable();			\
790 	} while (0)
791 
792 #define vcpu_get_flag(v, ...)	__vcpu_get_flag((v), __VA_ARGS__)
793 #define vcpu_set_flag(v, ...)	__vcpu_set_flag((v), __VA_ARGS__)
794 #define vcpu_clear_flag(v, ...)	__vcpu_clear_flag((v), __VA_ARGS__)
795 
796 /* SVE exposed to guest */
797 #define GUEST_HAS_SVE		__vcpu_single_flag(cflags, BIT(0))
798 /* SVE config completed */
799 #define VCPU_SVE_FINALIZED	__vcpu_single_flag(cflags, BIT(1))
800 /* PTRAUTH exposed to guest */
801 #define GUEST_HAS_PTRAUTH	__vcpu_single_flag(cflags, BIT(2))
802 /* KVM_ARM_VCPU_INIT completed */
803 #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(3))
804 
805 /* Exception pending */
806 #define PENDING_EXCEPTION	__vcpu_single_flag(iflags, BIT(0))
807 /*
808  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
809  * be set together with an exception...
810  */
811 #define INCREMENT_PC		__vcpu_single_flag(iflags, BIT(1))
812 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
813 #define EXCEPT_MASK		__vcpu_single_flag(iflags, GENMASK(3, 1))
814 
815 /* Helpers to encode exceptions with minimum fuss */
816 #define __EXCEPT_MASK_VAL	unpack_vcpu_flag(EXCEPT_MASK)
817 #define __EXCEPT_SHIFT		__builtin_ctzl(__EXCEPT_MASK_VAL)
818 #define __vcpu_except_flags(_f)	iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
819 
820 /*
821  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
822  * values:
823  *
824  * For AArch32 EL1:
825  */
826 #define EXCEPT_AA32_UND		__vcpu_except_flags(0)
827 #define EXCEPT_AA32_IABT	__vcpu_except_flags(1)
828 #define EXCEPT_AA32_DABT	__vcpu_except_flags(2)
829 /* For AArch64: */
830 #define EXCEPT_AA64_EL1_SYNC	__vcpu_except_flags(0)
831 #define EXCEPT_AA64_EL1_IRQ	__vcpu_except_flags(1)
832 #define EXCEPT_AA64_EL1_FIQ	__vcpu_except_flags(2)
833 #define EXCEPT_AA64_EL1_SERR	__vcpu_except_flags(3)
834 /* For AArch64 with NV: */
835 #define EXCEPT_AA64_EL2_SYNC	__vcpu_except_flags(4)
836 #define EXCEPT_AA64_EL2_IRQ	__vcpu_except_flags(5)
837 #define EXCEPT_AA64_EL2_FIQ	__vcpu_except_flags(6)
838 #define EXCEPT_AA64_EL2_SERR	__vcpu_except_flags(7)
839 /* Guest debug is live */
840 #define DEBUG_DIRTY		__vcpu_single_flag(iflags, BIT(4))
841 /* Save SPE context if active  */
842 #define DEBUG_STATE_SAVE_SPE	__vcpu_single_flag(iflags, BIT(5))
843 /* Save TRBE context if active  */
844 #define DEBUG_STATE_SAVE_TRBE	__vcpu_single_flag(iflags, BIT(6))
845 
846 /* SVE enabled for host EL0 */
847 #define HOST_SVE_ENABLED	__vcpu_single_flag(sflags, BIT(0))
848 /* SME enabled for EL0 */
849 #define HOST_SME_ENABLED	__vcpu_single_flag(sflags, BIT(1))
850 /* Physical CPU not in supported_cpus */
851 #define ON_UNSUPPORTED_CPU	__vcpu_single_flag(sflags, BIT(2))
852 /* WFIT instruction trapped */
853 #define IN_WFIT			__vcpu_single_flag(sflags, BIT(3))
854 /* vcpu system registers loaded on physical CPU */
855 #define SYSREGS_ON_CPU		__vcpu_single_flag(sflags, BIT(4))
856 /* Software step state is Active-pending */
857 #define DBG_SS_ACTIVE_PENDING	__vcpu_single_flag(sflags, BIT(5))
858 /* PMUSERENR for the guest EL0 is on physical CPU */
859 #define PMUSERENR_ON_CPU	__vcpu_single_flag(sflags, BIT(6))
860 /* WFI instruction trapped */
861 #define IN_WFI			__vcpu_single_flag(sflags, BIT(7))
862 
863 
864 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
865 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +	\
866 			     sve_ffr_offset((vcpu)->arch.sve_max_vl))
867 
868 #define vcpu_sve_max_vq(vcpu)	sve_vq_from_vl((vcpu)->arch.sve_max_vl)
869 
870 #define vcpu_sve_state_size(vcpu) ({					\
871 	size_t __size_ret;						\
872 	unsigned int __vcpu_vq;						\
873 									\
874 	if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {		\
875 		__size_ret = 0;						\
876 	} else {							\
877 		__vcpu_vq = vcpu_sve_max_vq(vcpu);			\
878 		__size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);		\
879 	}								\
880 									\
881 	__size_ret;							\
882 })
883 
884 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
885 				 KVM_GUESTDBG_USE_SW_BP | \
886 				 KVM_GUESTDBG_USE_HW | \
887 				 KVM_GUESTDBG_SINGLESTEP)
888 
889 #define vcpu_has_sve(vcpu) (system_supports_sve() &&			\
890 			    vcpu_get_flag(vcpu, GUEST_HAS_SVE))
891 
892 #ifdef CONFIG_ARM64_PTR_AUTH
893 #define vcpu_has_ptrauth(vcpu)						\
894 	((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||		\
895 	  cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&		\
896 	  vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
897 #else
898 #define vcpu_has_ptrauth(vcpu)		false
899 #endif
900 
901 #define vcpu_on_unsupported_cpu(vcpu)					\
902 	vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
903 
904 #define vcpu_set_on_unsupported_cpu(vcpu)				\
905 	vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
906 
907 #define vcpu_clear_on_unsupported_cpu(vcpu)				\
908 	vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
909 
910 #define vcpu_gp_regs(v)		(&(v)->arch.ctxt.regs)
911 
912 /*
913  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
914  * memory backed version of a register, and not the one most recently
915  * accessed by a running VCPU.  For example, for userspace access or
916  * for system registers that are never context switched, but only
917  * emulated.
918  *
919  * Don't bother with VNCR-based accesses in the nVHE code, it has no
920  * business dealing with NV.
921  */
922 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
923 {
924 #if !defined (__KVM_NVHE_HYPERVISOR__)
925 	if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
926 		     r >= __VNCR_START__ && ctxt->vncr_array))
927 		return &ctxt->vncr_array[r - __VNCR_START__];
928 #endif
929 	return (u64 *)&ctxt->sys_regs[r];
930 }
931 
932 #define __ctxt_sys_reg(c,r)						\
933 	({								\
934 		BUILD_BUG_ON(__builtin_constant_p(r) &&			\
935 			     (r) >= NR_SYS_REGS);			\
936 		___ctxt_sys_reg(c, r);					\
937 	})
938 
939 #define ctxt_sys_reg(c,r)	(*__ctxt_sys_reg(c,r))
940 
941 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg);
942 #define __vcpu_sys_reg(v,r)						\
943 	(*({								\
944 		const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt;	\
945 		u64 *__r = __ctxt_sys_reg(ctxt, (r));			\
946 		if (vcpu_has_nv((v)) && (r) >= __VNCR_START__)		\
947 			*__r = kvm_vcpu_sanitise_vncr_reg((v), (r));	\
948 		__r;							\
949 	}))
950 
951 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
952 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
953 
954 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
955 {
956 	/*
957 	 * *** VHE ONLY ***
958 	 *
959 	 * System registers listed in the switch are not saved on every
960 	 * exit from the guest but are only saved on vcpu_put.
961 	 *
962 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
963 	 * should never be listed below, because the guest cannot modify its
964 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
965 	 * thread when emulating cross-VCPU communication.
966 	 */
967 	if (!has_vhe())
968 		return false;
969 
970 	switch (reg) {
971 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
972 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
973 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
974 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
975 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
976 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
977 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
978 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
979 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
980 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
981 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
982 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
983 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
984 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
985 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
986 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
987 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
988 	case ELR_EL1:		*val = read_sysreg_s(SYS_ELR_EL12);	break;
989 	case SPSR_EL1:		*val = read_sysreg_s(SYS_SPSR_EL12);	break;
990 	case PAR_EL1:		*val = read_sysreg_par();		break;
991 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
992 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
993 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
994 	default:		return false;
995 	}
996 
997 	return true;
998 }
999 
1000 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
1001 {
1002 	/*
1003 	 * *** VHE ONLY ***
1004 	 *
1005 	 * System registers listed in the switch are not restored on every
1006 	 * entry to the guest but are only restored on vcpu_load.
1007 	 *
1008 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
1009 	 * should never be listed below, because the MPIDR should only be set
1010 	 * once, before running the VCPU, and never changed later.
1011 	 */
1012 	if (!has_vhe())
1013 		return false;
1014 
1015 	switch (reg) {
1016 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
1017 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
1018 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
1019 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
1020 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
1021 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
1022 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
1023 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
1024 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
1025 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
1026 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
1027 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
1028 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
1029 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
1030 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
1031 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
1032 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
1033 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
1034 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
1035 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
1036 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
1037 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
1038 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
1039 	default:		return false;
1040 	}
1041 
1042 	return true;
1043 }
1044 
1045 struct kvm_vm_stat {
1046 	struct kvm_vm_stat_generic generic;
1047 };
1048 
1049 struct kvm_vcpu_stat {
1050 	struct kvm_vcpu_stat_generic generic;
1051 	u64 hvc_exit_stat;
1052 	u64 wfe_exit_stat;
1053 	u64 wfi_exit_stat;
1054 	u64 mmio_exit_user;
1055 	u64 mmio_exit_kernel;
1056 	u64 signal_exits;
1057 	u64 exits;
1058 };
1059 
1060 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
1061 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
1062 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1063 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1064 
1065 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
1066 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
1067 
1068 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1069 			      struct kvm_vcpu_events *events);
1070 
1071 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1072 			      struct kvm_vcpu_events *events);
1073 
1074 void kvm_arm_halt_guest(struct kvm *kvm);
1075 void kvm_arm_resume_guest(struct kvm *kvm);
1076 
1077 #define vcpu_has_run_once(vcpu)	!!rcu_access_pointer((vcpu)->pid)
1078 
1079 #ifndef __KVM_NVHE_HYPERVISOR__
1080 #define kvm_call_hyp_nvhe(f, ...)						\
1081 	({								\
1082 		struct arm_smccc_res res;				\
1083 									\
1084 		arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),		\
1085 				  ##__VA_ARGS__, &res);			\
1086 		WARN_ON(res.a0 != SMCCC_RET_SUCCESS);			\
1087 									\
1088 		res.a1;							\
1089 	})
1090 
1091 /*
1092  * The couple of isb() below are there to guarantee the same behaviour
1093  * on VHE as on !VHE, where the eret to EL1 acts as a context
1094  * synchronization event.
1095  */
1096 #define kvm_call_hyp(f, ...)						\
1097 	do {								\
1098 		if (has_vhe()) {					\
1099 			f(__VA_ARGS__);					\
1100 			isb();						\
1101 		} else {						\
1102 			kvm_call_hyp_nvhe(f, ##__VA_ARGS__);		\
1103 		}							\
1104 	} while(0)
1105 
1106 #define kvm_call_hyp_ret(f, ...)					\
1107 	({								\
1108 		typeof(f(__VA_ARGS__)) ret;				\
1109 									\
1110 		if (has_vhe()) {					\
1111 			ret = f(__VA_ARGS__);				\
1112 			isb();						\
1113 		} else {						\
1114 			ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);	\
1115 		}							\
1116 									\
1117 		ret;							\
1118 	})
1119 #else /* __KVM_NVHE_HYPERVISOR__ */
1120 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1121 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1122 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1123 #endif /* __KVM_NVHE_HYPERVISOR__ */
1124 
1125 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1126 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1127 
1128 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1129 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1130 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1131 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1132 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1133 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1134 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1135 
1136 void kvm_sys_regs_create_debugfs(struct kvm *kvm);
1137 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1138 
1139 int __init kvm_sys_reg_table_init(void);
1140 struct sys_reg_desc;
1141 int __init populate_sysreg_config(const struct sys_reg_desc *sr,
1142 				  unsigned int idx);
1143 int __init populate_nv_trap_config(void);
1144 
1145 bool lock_all_vcpus(struct kvm *kvm);
1146 void unlock_all_vcpus(struct kvm *kvm);
1147 
1148 void kvm_init_sysreg(struct kvm_vcpu *);
1149 
1150 /* MMIO helpers */
1151 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1152 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1153 
1154 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1155 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1156 
1157 /*
1158  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1159  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1160  * loaded is considered to be "in guest".
1161  */
1162 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1163 {
1164 	return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1165 }
1166 
1167 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1168 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1169 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1170 
1171 bool kvm_arm_pvtime_supported(void);
1172 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1173 			    struct kvm_device_attr *attr);
1174 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1175 			    struct kvm_device_attr *attr);
1176 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1177 			    struct kvm_device_attr *attr);
1178 
1179 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1180 int __init kvm_arm_vmid_alloc_init(void);
1181 void __init kvm_arm_vmid_alloc_free(void);
1182 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1183 void kvm_arm_vmid_clear_active(void);
1184 
1185 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1186 {
1187 	vcpu_arch->steal.base = INVALID_GPA;
1188 }
1189 
1190 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1191 {
1192 	return (vcpu_arch->steal.base != INVALID_GPA);
1193 }
1194 
1195 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1196 
1197 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1198 
1199 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1200 
1201 /*
1202  * How we access per-CPU host data depends on the where we access it from,
1203  * and the mode we're in:
1204  *
1205  * - VHE and nVHE hypervisor bits use their locally defined instance
1206  *
1207  * - the rest of the kernel use either the VHE or nVHE one, depending on
1208  *   the mode we're running in.
1209  *
1210  *   Unless we're in protected mode, fully deprivileged, and the nVHE
1211  *   per-CPU stuff is exclusively accessible to the protected EL2 code.
1212  *   In this case, the EL1 code uses the *VHE* data as its private state
1213  *   (which makes sense in a way as there shouldn't be any shared state
1214  *   between the host and the hypervisor).
1215  *
1216  * Yes, this is all totally trivial. Shoot me now.
1217  */
1218 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__)
1219 #define host_data_ptr(f)	(&this_cpu_ptr(&kvm_host_data)->f)
1220 #else
1221 #define host_data_ptr(f)						\
1222 	(static_branch_unlikely(&kvm_protected_mode_initialized) ?	\
1223 	 &this_cpu_ptr(&kvm_host_data)->f :				\
1224 	 &this_cpu_ptr_hyp_sym(kvm_host_data)->f)
1225 #endif
1226 
1227 /* Check whether the FP regs are owned by the guest */
1228 static inline bool guest_owns_fp_regs(void)
1229 {
1230 	return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED;
1231 }
1232 
1233 /* Check whether the FP regs are owned by the host */
1234 static inline bool host_owns_fp_regs(void)
1235 {
1236 	return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED;
1237 }
1238 
1239 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1240 {
1241 	/* The host's MPIDR is immutable, so let's set it up at boot time */
1242 	ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1243 }
1244 
1245 static inline bool kvm_system_needs_idmapped_vectors(void)
1246 {
1247 	return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1248 }
1249 
1250 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1251 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1252 
1253 void kvm_arm_init_debug(void);
1254 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1255 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1256 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1257 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1258 
1259 #define kvm_vcpu_os_lock_enabled(vcpu)		\
1260 	(!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1261 
1262 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1263 			       struct kvm_device_attr *attr);
1264 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1265 			       struct kvm_device_attr *attr);
1266 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1267 			       struct kvm_device_attr *attr);
1268 
1269 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1270 			       struct kvm_arm_copy_mte_tags *copy_tags);
1271 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1272 				    struct kvm_arm_counter_offset *offset);
1273 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1274 					struct reg_mask_range *range);
1275 
1276 /* Guest/host FPSIMD coordination helpers */
1277 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1278 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1279 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1280 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1281 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1282 
1283 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1284 {
1285 	return (!has_vhe() && attr->exclude_host);
1286 }
1287 
1288 /* Flags for host debug state */
1289 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1290 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1291 
1292 #ifdef CONFIG_KVM
1293 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1294 void kvm_clr_pmu_events(u32 clr);
1295 bool kvm_set_pmuserenr(u64 val);
1296 #else
1297 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1298 static inline void kvm_clr_pmu_events(u32 clr) {}
1299 static inline bool kvm_set_pmuserenr(u64 val)
1300 {
1301 	return false;
1302 }
1303 #endif
1304 
1305 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1306 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1307 
1308 int __init kvm_set_ipa_limit(void);
1309 
1310 #define __KVM_HAVE_ARCH_VM_ALLOC
1311 struct kvm *kvm_arch_alloc_vm(void);
1312 
1313 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1314 
1315 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1316 
1317 #define kvm_vm_is_protected(kvm)	(is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled)
1318 
1319 #define vcpu_is_protected(vcpu)		kvm_vm_is_protected((vcpu)->kvm)
1320 
1321 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1322 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1323 
1324 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1325 
1326 #define kvm_has_mte(kvm)					\
1327 	(system_supports_mte() &&				\
1328 	 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1329 
1330 #define kvm_supports_32bit_el0()				\
1331 	(system_supports_32bit_el0() &&				\
1332 	 !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1333 
1334 #define kvm_vm_has_ran_once(kvm)					\
1335 	(test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1336 
1337 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1338 {
1339 	return test_bit(feature, ka->vcpu_features);
1340 }
1341 
1342 #define vcpu_has_feature(v, f)	__vcpu_has_feature(&(v)->kvm->arch, (f))
1343 
1344 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED)
1345 
1346 int kvm_trng_call(struct kvm_vcpu *vcpu);
1347 #ifdef CONFIG_KVM
1348 extern phys_addr_t hyp_mem_base;
1349 extern phys_addr_t hyp_mem_size;
1350 void __init kvm_hyp_reserve(void);
1351 #else
1352 static inline void kvm_hyp_reserve(void) { }
1353 #endif
1354 
1355 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1356 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1357 
1358 #define __expand_field_sign_unsigned(id, fld, val)			\
1359 	((u64)SYS_FIELD_VALUE(id, fld, val))
1360 
1361 #define __expand_field_sign_signed(id, fld, val)			\
1362 	({								\
1363 		u64 __val = SYS_FIELD_VALUE(id, fld, val);		\
1364 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1365 	})
1366 
1367 #define expand_field_sign(id, fld, val)					\
1368 	(id##_##fld##_SIGNED ?						\
1369 	 __expand_field_sign_signed(id, fld, val) :			\
1370 	 __expand_field_sign_unsigned(id, fld, val))
1371 
1372 #define get_idreg_field_unsigned(kvm, id, fld)				\
1373 	({								\
1374 		u64 __val = IDREG((kvm), SYS_##id);			\
1375 		FIELD_GET(id##_##fld##_MASK, __val);			\
1376 	})
1377 
1378 #define get_idreg_field_signed(kvm, id, fld)				\
1379 	({								\
1380 		u64 __val = get_idreg_field_unsigned(kvm, id, fld);	\
1381 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1382 	})
1383 
1384 #define get_idreg_field_enum(kvm, id, fld)				\
1385 	get_idreg_field_unsigned(kvm, id, fld)
1386 
1387 #define get_idreg_field(kvm, id, fld)					\
1388 	(id##_##fld##_SIGNED ?						\
1389 	 get_idreg_field_signed(kvm, id, fld) :				\
1390 	 get_idreg_field_unsigned(kvm, id, fld))
1391 
1392 #define kvm_has_feat(kvm, id, fld, limit)				\
1393 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit))
1394 
1395 #define kvm_has_feat_enum(kvm, id, fld, val)				\
1396 	(get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val))
1397 
1398 #define kvm_has_feat_range(kvm, id, fld, min, max)			\
1399 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \
1400 	 get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max))
1401 
1402 /* Check for a given level of PAuth support */
1403 #define kvm_has_pauth(k, l)						\
1404 	({								\
1405 		bool pa, pi, pa3;					\
1406 									\
1407 		pa  = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l);	\
1408 		pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP);	\
1409 		pi  = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l);	\
1410 		pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP);	\
1411 		pa3  = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l);	\
1412 		pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP);	\
1413 									\
1414 		(pa + pi + pa3) == 1;					\
1415 	})
1416 
1417 #endif /* __ARM64_KVM_HOST_H__ */
1418