xref: /linux/arch/arm64/include/asm/kvm_host.h (revision 8540bd1b990bad7f7e95b5bf1adf30bfaf2e38c9)
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 	bool enabled;
215 };
216 
217 struct kvm_mpidr_data {
218 	u64			mpidr_mask;
219 	DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx);
220 };
221 
222 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr)
223 {
224 	unsigned long mask = data->mpidr_mask;
225 	u64 aff = mpidr & MPIDR_HWID_BITMASK;
226 	int nbits, bit, bit_idx = 0;
227 	u16 index = 0;
228 
229 	/*
230 	 * If this looks like RISC-V's BEXT or x86's PEXT
231 	 * instructions, it isn't by accident.
232 	 */
233 	nbits = fls(mask);
234 	for_each_set_bit(bit, &mask, nbits) {
235 		index |= (aff & BIT(bit)) >> (bit - bit_idx);
236 		bit_idx++;
237 	}
238 
239 	return index;
240 }
241 
242 struct kvm_sysreg_masks;
243 
244 enum fgt_group_id {
245 	__NO_FGT_GROUP__,
246 	HFGxTR_GROUP,
247 	HDFGRTR_GROUP,
248 	HDFGWTR_GROUP = HDFGRTR_GROUP,
249 	HFGITR_GROUP,
250 	HAFGRTR_GROUP,
251 
252 	/* Must be last */
253 	__NR_FGT_GROUP_IDS__
254 };
255 
256 struct kvm_arch {
257 	struct kvm_s2_mmu mmu;
258 
259 	/*
260 	 * Fine-Grained UNDEF, mimicking the FGT layout defined by the
261 	 * architecture. We track them globally, as we present the
262 	 * same feature-set to all vcpus.
263 	 *
264 	 * Index 0 is currently spare.
265 	 */
266 	u64 fgu[__NR_FGT_GROUP_IDS__];
267 
268 	/* Interrupt controller */
269 	struct vgic_dist	vgic;
270 
271 	/* Timers */
272 	struct arch_timer_vm_data timer_data;
273 
274 	/* Mandated version of PSCI */
275 	u32 psci_version;
276 
277 	/* Protects VM-scoped configuration data */
278 	struct mutex config_lock;
279 
280 	/*
281 	 * If we encounter a data abort without valid instruction syndrome
282 	 * information, report this to user space.  User space can (and
283 	 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
284 	 * supported.
285 	 */
286 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER	0
287 	/* Memory Tagging Extension enabled for the guest */
288 #define KVM_ARCH_FLAG_MTE_ENABLED			1
289 	/* At least one vCPU has ran in the VM */
290 #define KVM_ARCH_FLAG_HAS_RAN_ONCE			2
291 	/* The vCPU feature set for the VM is configured */
292 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED		3
293 	/* PSCI SYSTEM_SUSPEND enabled for the guest */
294 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED		4
295 	/* VM counter offset */
296 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET			5
297 	/* Timer PPIs made immutable */
298 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE		6
299 	/* Initial ID reg values loaded */
300 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED		7
301 	/* Fine-Grained UNDEF initialised */
302 #define KVM_ARCH_FLAG_FGU_INITIALIZED			8
303 	unsigned long flags;
304 
305 	/* VM-wide vCPU feature set */
306 	DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
307 
308 	/* MPIDR to vcpu index mapping, optional */
309 	struct kvm_mpidr_data *mpidr_data;
310 
311 	/*
312 	 * VM-wide PMU filter, implemented as a bitmap and big enough for
313 	 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
314 	 */
315 	unsigned long *pmu_filter;
316 	struct arm_pmu *arm_pmu;
317 
318 	cpumask_var_t supported_cpus;
319 
320 	/* PMCR_EL0.N value for the guest */
321 	u8 pmcr_n;
322 
323 	/* Iterator for idreg debugfs */
324 	u8	idreg_debugfs_iter;
325 
326 	/* Hypercall features firmware registers' descriptor */
327 	struct kvm_smccc_features smccc_feat;
328 	struct maple_tree smccc_filter;
329 
330 	/*
331 	 * Emulated CPU ID registers per VM
332 	 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
333 	 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
334 	 *
335 	 * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
336 	 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
337 	 */
338 #define IDREG_IDX(id)		(((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
339 #define IDX_IDREG(idx)		sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask)
340 #define IDREG(kvm, id)		((kvm)->arch.id_regs[IDREG_IDX(id)])
341 #define KVM_ARM_ID_REG_NUM	(IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
342 	u64 id_regs[KVM_ARM_ID_REG_NUM];
343 
344 	/* Masks for VNCR-baked sysregs */
345 	struct kvm_sysreg_masks	*sysreg_masks;
346 
347 	/*
348 	 * For an untrusted host VM, 'pkvm.handle' is used to lookup
349 	 * the associated pKVM instance in the hypervisor.
350 	 */
351 	struct kvm_protected_vm pkvm;
352 };
353 
354 struct kvm_vcpu_fault_info {
355 	u64 esr_el2;		/* Hyp Syndrom Register */
356 	u64 far_el2;		/* Hyp Fault Address Register */
357 	u64 hpfar_el2;		/* Hyp IPA Fault Address Register */
358 	u64 disr_el1;		/* Deferred [SError] Status Register */
359 };
360 
361 /*
362  * VNCR() just places the VNCR_capable registers in the enum after
363  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
364  * from the VNCR base. As we don't require the enum to be otherwise ordered,
365  * we need the terrible hack below to ensure that we correctly size the
366  * sys_regs array, no matter what.
367  *
368  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
369  * treasure trove of bit hacks:
370  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
371  */
372 #define __MAX__(x,y)	((x) ^ (((x) ^ (y)) & -((x) < (y))))
373 #define VNCR(r)						\
374 	__before_##r,					\
375 	r = __VNCR_START__ + ((VNCR_ ## r) / 8),	\
376 	__after_##r = __MAX__(__before_##r - 1, r)
377 
378 enum vcpu_sysreg {
379 	__INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
380 	MPIDR_EL1,	/* MultiProcessor Affinity Register */
381 	CLIDR_EL1,	/* Cache Level ID Register */
382 	CSSELR_EL1,	/* Cache Size Selection Register */
383 	TPIDR_EL0,	/* Thread ID, User R/W */
384 	TPIDRRO_EL0,	/* Thread ID, User R/O */
385 	TPIDR_EL1,	/* Thread ID, Privileged */
386 	CNTKCTL_EL1,	/* Timer Control Register (EL1) */
387 	PAR_EL1,	/* Physical Address Register */
388 	MDCCINT_EL1,	/* Monitor Debug Comms Channel Interrupt Enable Reg */
389 	OSLSR_EL1,	/* OS Lock Status Register */
390 	DISR_EL1,	/* Deferred Interrupt Status Register */
391 
392 	/* Performance Monitors Registers */
393 	PMCR_EL0,	/* Control Register */
394 	PMSELR_EL0,	/* Event Counter Selection Register */
395 	PMEVCNTR0_EL0,	/* Event Counter Register (0-30) */
396 	PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
397 	PMCCNTR_EL0,	/* Cycle Counter Register */
398 	PMEVTYPER0_EL0,	/* Event Type Register (0-30) */
399 	PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
400 	PMCCFILTR_EL0,	/* Cycle Count Filter Register */
401 	PMCNTENSET_EL0,	/* Count Enable Set Register */
402 	PMINTENSET_EL1,	/* Interrupt Enable Set Register */
403 	PMOVSSET_EL0,	/* Overflow Flag Status Set Register */
404 	PMUSERENR_EL0,	/* User Enable Register */
405 
406 	/* Pointer Authentication Registers in a strict increasing order. */
407 	APIAKEYLO_EL1,
408 	APIAKEYHI_EL1,
409 	APIBKEYLO_EL1,
410 	APIBKEYHI_EL1,
411 	APDAKEYLO_EL1,
412 	APDAKEYHI_EL1,
413 	APDBKEYLO_EL1,
414 	APDBKEYHI_EL1,
415 	APGAKEYLO_EL1,
416 	APGAKEYHI_EL1,
417 
418 	/* Memory Tagging Extension registers */
419 	RGSR_EL1,	/* Random Allocation Tag Seed Register */
420 	GCR_EL1,	/* Tag Control Register */
421 	TFSRE0_EL1,	/* Tag Fault Status Register (EL0) */
422 
423 	/* 32bit specific registers. */
424 	DACR32_EL2,	/* Domain Access Control Register */
425 	IFSR32_EL2,	/* Instruction Fault Status Register */
426 	FPEXC32_EL2,	/* Floating-Point Exception Control Register */
427 	DBGVCR32_EL2,	/* Debug Vector Catch Register */
428 
429 	/* EL2 registers */
430 	SCTLR_EL2,	/* System Control Register (EL2) */
431 	ACTLR_EL2,	/* Auxiliary Control Register (EL2) */
432 	MDCR_EL2,	/* Monitor Debug Configuration Register (EL2) */
433 	CPTR_EL2,	/* Architectural Feature Trap Register (EL2) */
434 	HACR_EL2,	/* Hypervisor Auxiliary Control Register */
435 	TTBR0_EL2,	/* Translation Table Base Register 0 (EL2) */
436 	TTBR1_EL2,	/* Translation Table Base Register 1 (EL2) */
437 	TCR_EL2,	/* Translation Control Register (EL2) */
438 	SPSR_EL2,	/* EL2 saved program status register */
439 	ELR_EL2,	/* EL2 exception link register */
440 	AFSR0_EL2,	/* Auxiliary Fault Status Register 0 (EL2) */
441 	AFSR1_EL2,	/* Auxiliary Fault Status Register 1 (EL2) */
442 	ESR_EL2,	/* Exception Syndrome Register (EL2) */
443 	FAR_EL2,	/* Fault Address Register (EL2) */
444 	HPFAR_EL2,	/* Hypervisor IPA Fault Address Register */
445 	MAIR_EL2,	/* Memory Attribute Indirection Register (EL2) */
446 	AMAIR_EL2,	/* Auxiliary Memory Attribute Indirection Register (EL2) */
447 	VBAR_EL2,	/* Vector Base Address Register (EL2) */
448 	RVBAR_EL2,	/* Reset Vector Base Address Register */
449 	CONTEXTIDR_EL2,	/* Context ID Register (EL2) */
450 	CNTHCTL_EL2,	/* Counter-timer Hypervisor Control register */
451 	SP_EL2,		/* EL2 Stack Pointer */
452 	CNTHP_CTL_EL2,
453 	CNTHP_CVAL_EL2,
454 	CNTHV_CTL_EL2,
455 	CNTHV_CVAL_EL2,
456 
457 	__VNCR_START__,	/* Any VNCR-capable reg goes after this point */
458 
459 	VNCR(SCTLR_EL1),/* System Control Register */
460 	VNCR(ACTLR_EL1),/* Auxiliary Control Register */
461 	VNCR(CPACR_EL1),/* Coprocessor Access Control */
462 	VNCR(ZCR_EL1),	/* SVE Control */
463 	VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
464 	VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
465 	VNCR(TCR_EL1),	/* Translation Control Register */
466 	VNCR(TCR2_EL1),	/* Extended Translation Control Register */
467 	VNCR(ESR_EL1),	/* Exception Syndrome Register */
468 	VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
469 	VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
470 	VNCR(FAR_EL1),	/* Fault Address Register */
471 	VNCR(MAIR_EL1),	/* Memory Attribute Indirection Register */
472 	VNCR(VBAR_EL1),	/* Vector Base Address Register */
473 	VNCR(CONTEXTIDR_EL1),	/* Context ID Register */
474 	VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
475 	VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
476 	VNCR(ELR_EL1),
477 	VNCR(SP_EL1),
478 	VNCR(SPSR_EL1),
479 	VNCR(TFSR_EL1),	/* Tag Fault Status Register (EL1) */
480 	VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
481 	VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
482 	VNCR(HCR_EL2),	/* Hypervisor Configuration Register */
483 	VNCR(HSTR_EL2),	/* Hypervisor System Trap Register */
484 	VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
485 	VNCR(VTCR_EL2),	/* Virtualization Translation Control Register */
486 	VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
487 	VNCR(HCRX_EL2),	/* Extended Hypervisor Configuration Register */
488 
489 	/* Permission Indirection Extension registers */
490 	VNCR(PIR_EL1),	 /* Permission Indirection Register 1 (EL1) */
491 	VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
492 
493 	VNCR(HFGRTR_EL2),
494 	VNCR(HFGWTR_EL2),
495 	VNCR(HFGITR_EL2),
496 	VNCR(HDFGRTR_EL2),
497 	VNCR(HDFGWTR_EL2),
498 	VNCR(HAFGRTR_EL2),
499 
500 	VNCR(CNTVOFF_EL2),
501 	VNCR(CNTV_CVAL_EL0),
502 	VNCR(CNTV_CTL_EL0),
503 	VNCR(CNTP_CVAL_EL0),
504 	VNCR(CNTP_CTL_EL0),
505 
506 	NR_SYS_REGS	/* Nothing after this line! */
507 };
508 
509 struct kvm_sysreg_masks {
510 	struct {
511 		u64	res0;
512 		u64	res1;
513 	} mask[NR_SYS_REGS - __VNCR_START__];
514 };
515 
516 struct kvm_cpu_context {
517 	struct user_pt_regs regs;	/* sp = sp_el0 */
518 
519 	u64	spsr_abt;
520 	u64	spsr_und;
521 	u64	spsr_irq;
522 	u64	spsr_fiq;
523 
524 	struct user_fpsimd_state fp_regs;
525 
526 	u64 sys_regs[NR_SYS_REGS];
527 
528 	struct kvm_vcpu *__hyp_running_vcpu;
529 
530 	/* This pointer has to be 4kB aligned. */
531 	u64 *vncr_array;
532 };
533 
534 /*
535  * This structure is instantiated on a per-CPU basis, and contains
536  * data that is:
537  *
538  * - tied to a single physical CPU, and
539  * - either have a lifetime that does not extend past vcpu_put()
540  * - or is an invariant for the lifetime of the system
541  *
542  * Use host_data_ptr(field) as a way to access a pointer to such a
543  * field.
544  */
545 struct kvm_host_data {
546 	struct kvm_cpu_context host_ctxt;
547 	struct user_fpsimd_state *fpsimd_state;	/* hyp VA */
548 
549 	/* Ownership of the FP regs */
550 	enum {
551 		FP_STATE_FREE,
552 		FP_STATE_HOST_OWNED,
553 		FP_STATE_GUEST_OWNED,
554 	} fp_owner;
555 
556 	/*
557 	 * host_debug_state contains the host registers which are
558 	 * saved and restored during world switches.
559 	 */
560 	 struct {
561 		/* {Break,watch}point registers */
562 		struct kvm_guest_debug_arch regs;
563 		/* Statistical profiling extension */
564 		u64 pmscr_el1;
565 		/* Self-hosted trace */
566 		u64 trfcr_el1;
567 		/* Values of trap registers for the host before guest entry. */
568 		u64 mdcr_el2;
569 	} host_debug_state;
570 };
571 
572 struct kvm_host_psci_config {
573 	/* PSCI version used by host. */
574 	u32 version;
575 	u32 smccc_version;
576 
577 	/* Function IDs used by host if version is v0.1. */
578 	struct psci_0_1_function_ids function_ids_0_1;
579 
580 	bool psci_0_1_cpu_suspend_implemented;
581 	bool psci_0_1_cpu_on_implemented;
582 	bool psci_0_1_cpu_off_implemented;
583 	bool psci_0_1_migrate_implemented;
584 };
585 
586 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
587 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
588 
589 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
590 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
591 
592 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
593 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
594 
595 struct vcpu_reset_state {
596 	unsigned long	pc;
597 	unsigned long	r0;
598 	bool		be;
599 	bool		reset;
600 };
601 
602 struct kvm_vcpu_arch {
603 	struct kvm_cpu_context ctxt;
604 
605 	/*
606 	 * Guest floating point state
607 	 *
608 	 * The architecture has two main floating point extensions,
609 	 * the original FPSIMD and SVE.  These have overlapping
610 	 * register views, with the FPSIMD V registers occupying the
611 	 * low 128 bits of the SVE Z registers.  When the core
612 	 * floating point code saves the register state of a task it
613 	 * records which view it saved in fp_type.
614 	 */
615 	void *sve_state;
616 	enum fp_type fp_type;
617 	unsigned int sve_max_vl;
618 	u64 svcr;
619 	u64 fpmr;
620 
621 	/* Stage 2 paging state used by the hardware on next switch */
622 	struct kvm_s2_mmu *hw_mmu;
623 
624 	/* Values of trap registers for the guest. */
625 	u64 hcr_el2;
626 	u64 hcrx_el2;
627 	u64 mdcr_el2;
628 	u64 cptr_el2;
629 
630 	/* Exception Information */
631 	struct kvm_vcpu_fault_info fault;
632 
633 	/* Configuration flags, set once and for all before the vcpu can run */
634 	u8 cflags;
635 
636 	/* Input flags to the hypervisor code, potentially cleared after use */
637 	u8 iflags;
638 
639 	/* State flags for kernel bookkeeping, unused by the hypervisor code */
640 	u8 sflags;
641 
642 	/*
643 	 * Don't run the guest (internal implementation need).
644 	 *
645 	 * Contrary to the flags above, this is set/cleared outside of
646 	 * a vcpu context, and thus cannot be mixed with the flags
647 	 * themselves (or the flag accesses need to be made atomic).
648 	 */
649 	bool pause;
650 
651 	/*
652 	 * We maintain more than a single set of debug registers to support
653 	 * debugging the guest from the host and to maintain separate host and
654 	 * guest state during world switches. vcpu_debug_state are the debug
655 	 * registers of the vcpu as the guest sees them.
656 	 *
657 	 * external_debug_state contains the debug values we want to debug the
658 	 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl.
659 	 *
660 	 * debug_ptr points to the set of debug registers that should be loaded
661 	 * onto the hardware when running the guest.
662 	 */
663 	struct kvm_guest_debug_arch *debug_ptr;
664 	struct kvm_guest_debug_arch vcpu_debug_state;
665 	struct kvm_guest_debug_arch external_debug_state;
666 
667 	/* VGIC state */
668 	struct vgic_cpu vgic_cpu;
669 	struct arch_timer_cpu timer_cpu;
670 	struct kvm_pmu pmu;
671 
672 	/*
673 	 * Guest registers we preserve during guest debugging.
674 	 *
675 	 * These shadow registers are updated by the kvm_handle_sys_reg
676 	 * trap handler if the guest accesses or updates them while we
677 	 * are using guest debug.
678 	 */
679 	struct {
680 		u32	mdscr_el1;
681 		bool	pstate_ss;
682 	} guest_debug_preserved;
683 
684 	/* vcpu power state */
685 	struct kvm_mp_state mp_state;
686 	spinlock_t mp_state_lock;
687 
688 	/* Cache some mmu pages needed inside spinlock regions */
689 	struct kvm_mmu_memory_cache mmu_page_cache;
690 
691 	/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
692 	u64 vsesr_el2;
693 
694 	/* Additional reset state */
695 	struct vcpu_reset_state	reset_state;
696 
697 	/* Guest PV state */
698 	struct {
699 		u64 last_steal;
700 		gpa_t base;
701 	} steal;
702 
703 	/* Per-vcpu CCSIDR override or NULL */
704 	u32 *ccsidr;
705 };
706 
707 /*
708  * Each 'flag' is composed of a comma-separated triplet:
709  *
710  * - the flag-set it belongs to in the vcpu->arch structure
711  * - the value for that flag
712  * - the mask for that flag
713  *
714  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
715  * unpack_vcpu_flag() extract the flag value from the triplet for
716  * direct use outside of the flag accessors.
717  */
718 #define __vcpu_single_flag(_set, _f)	_set, (_f), (_f)
719 
720 #define __unpack_flag(_set, _f, _m)	_f
721 #define unpack_vcpu_flag(...)		__unpack_flag(__VA_ARGS__)
722 
723 #define __build_check_flag(v, flagset, f, m)			\
724 	do {							\
725 		typeof(v->arch.flagset) *_fset;			\
726 								\
727 		/* Check that the flags fit in the mask */	\
728 		BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m)));	\
729 		/* Check that the flags fit in the type */	\
730 		BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m));	\
731 	} while (0)
732 
733 #define __vcpu_get_flag(v, flagset, f, m)			\
734 	({							\
735 		__build_check_flag(v, flagset, f, m);		\
736 								\
737 		READ_ONCE(v->arch.flagset) & (m);		\
738 	})
739 
740 /*
741  * Note that the set/clear accessors must be preempt-safe in order to
742  * avoid nesting them with load/put which also manipulate flags...
743  */
744 #ifdef __KVM_NVHE_HYPERVISOR__
745 /* the nVHE hypervisor is always non-preemptible */
746 #define __vcpu_flags_preempt_disable()
747 #define __vcpu_flags_preempt_enable()
748 #else
749 #define __vcpu_flags_preempt_disable()	preempt_disable()
750 #define __vcpu_flags_preempt_enable()	preempt_enable()
751 #endif
752 
753 #define __vcpu_set_flag(v, flagset, f, m)			\
754 	do {							\
755 		typeof(v->arch.flagset) *fset;			\
756 								\
757 		__build_check_flag(v, flagset, f, m);		\
758 								\
759 		fset = &v->arch.flagset;			\
760 		__vcpu_flags_preempt_disable();			\
761 		if (HWEIGHT(m) > 1)				\
762 			*fset &= ~(m);				\
763 		*fset |= (f);					\
764 		__vcpu_flags_preempt_enable();			\
765 	} while (0)
766 
767 #define __vcpu_clear_flag(v, flagset, f, m)			\
768 	do {							\
769 		typeof(v->arch.flagset) *fset;			\
770 								\
771 		__build_check_flag(v, flagset, f, m);		\
772 								\
773 		fset = &v->arch.flagset;			\
774 		__vcpu_flags_preempt_disable();			\
775 		*fset &= ~(m);					\
776 		__vcpu_flags_preempt_enable();			\
777 	} while (0)
778 
779 #define vcpu_get_flag(v, ...)	__vcpu_get_flag((v), __VA_ARGS__)
780 #define vcpu_set_flag(v, ...)	__vcpu_set_flag((v), __VA_ARGS__)
781 #define vcpu_clear_flag(v, ...)	__vcpu_clear_flag((v), __VA_ARGS__)
782 
783 /* SVE exposed to guest */
784 #define GUEST_HAS_SVE		__vcpu_single_flag(cflags, BIT(0))
785 /* SVE config completed */
786 #define VCPU_SVE_FINALIZED	__vcpu_single_flag(cflags, BIT(1))
787 /* PTRAUTH exposed to guest */
788 #define GUEST_HAS_PTRAUTH	__vcpu_single_flag(cflags, BIT(2))
789 /* KVM_ARM_VCPU_INIT completed */
790 #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(3))
791 
792 /* Exception pending */
793 #define PENDING_EXCEPTION	__vcpu_single_flag(iflags, BIT(0))
794 /*
795  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
796  * be set together with an exception...
797  */
798 #define INCREMENT_PC		__vcpu_single_flag(iflags, BIT(1))
799 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
800 #define EXCEPT_MASK		__vcpu_single_flag(iflags, GENMASK(3, 1))
801 
802 /* Helpers to encode exceptions with minimum fuss */
803 #define __EXCEPT_MASK_VAL	unpack_vcpu_flag(EXCEPT_MASK)
804 #define __EXCEPT_SHIFT		__builtin_ctzl(__EXCEPT_MASK_VAL)
805 #define __vcpu_except_flags(_f)	iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
806 
807 /*
808  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
809  * values:
810  *
811  * For AArch32 EL1:
812  */
813 #define EXCEPT_AA32_UND		__vcpu_except_flags(0)
814 #define EXCEPT_AA32_IABT	__vcpu_except_flags(1)
815 #define EXCEPT_AA32_DABT	__vcpu_except_flags(2)
816 /* For AArch64: */
817 #define EXCEPT_AA64_EL1_SYNC	__vcpu_except_flags(0)
818 #define EXCEPT_AA64_EL1_IRQ	__vcpu_except_flags(1)
819 #define EXCEPT_AA64_EL1_FIQ	__vcpu_except_flags(2)
820 #define EXCEPT_AA64_EL1_SERR	__vcpu_except_flags(3)
821 /* For AArch64 with NV: */
822 #define EXCEPT_AA64_EL2_SYNC	__vcpu_except_flags(4)
823 #define EXCEPT_AA64_EL2_IRQ	__vcpu_except_flags(5)
824 #define EXCEPT_AA64_EL2_FIQ	__vcpu_except_flags(6)
825 #define EXCEPT_AA64_EL2_SERR	__vcpu_except_flags(7)
826 /* Guest debug is live */
827 #define DEBUG_DIRTY		__vcpu_single_flag(iflags, BIT(4))
828 /* Save SPE context if active  */
829 #define DEBUG_STATE_SAVE_SPE	__vcpu_single_flag(iflags, BIT(5))
830 /* Save TRBE context if active  */
831 #define DEBUG_STATE_SAVE_TRBE	__vcpu_single_flag(iflags, BIT(6))
832 
833 /* SVE enabled for host EL0 */
834 #define HOST_SVE_ENABLED	__vcpu_single_flag(sflags, BIT(0))
835 /* SME enabled for EL0 */
836 #define HOST_SME_ENABLED	__vcpu_single_flag(sflags, BIT(1))
837 /* Physical CPU not in supported_cpus */
838 #define ON_UNSUPPORTED_CPU	__vcpu_single_flag(sflags, BIT(2))
839 /* WFIT instruction trapped */
840 #define IN_WFIT			__vcpu_single_flag(sflags, BIT(3))
841 /* vcpu system registers loaded on physical CPU */
842 #define SYSREGS_ON_CPU		__vcpu_single_flag(sflags, BIT(4))
843 /* Software step state is Active-pending */
844 #define DBG_SS_ACTIVE_PENDING	__vcpu_single_flag(sflags, BIT(5))
845 /* PMUSERENR for the guest EL0 is on physical CPU */
846 #define PMUSERENR_ON_CPU	__vcpu_single_flag(sflags, BIT(6))
847 /* WFI instruction trapped */
848 #define IN_WFI			__vcpu_single_flag(sflags, BIT(7))
849 
850 
851 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
852 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +	\
853 			     sve_ffr_offset((vcpu)->arch.sve_max_vl))
854 
855 #define vcpu_sve_max_vq(vcpu)	sve_vq_from_vl((vcpu)->arch.sve_max_vl)
856 
857 #define vcpu_sve_state_size(vcpu) ({					\
858 	size_t __size_ret;						\
859 	unsigned int __vcpu_vq;						\
860 									\
861 	if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {		\
862 		__size_ret = 0;						\
863 	} else {							\
864 		__vcpu_vq = vcpu_sve_max_vq(vcpu);			\
865 		__size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);		\
866 	}								\
867 									\
868 	__size_ret;							\
869 })
870 
871 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
872 				 KVM_GUESTDBG_USE_SW_BP | \
873 				 KVM_GUESTDBG_USE_HW | \
874 				 KVM_GUESTDBG_SINGLESTEP)
875 
876 #define vcpu_has_sve(vcpu) (system_supports_sve() &&			\
877 			    vcpu_get_flag(vcpu, GUEST_HAS_SVE))
878 
879 #ifdef CONFIG_ARM64_PTR_AUTH
880 #define vcpu_has_ptrauth(vcpu)						\
881 	((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||		\
882 	  cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&		\
883 	  vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
884 #else
885 #define vcpu_has_ptrauth(vcpu)		false
886 #endif
887 
888 #define vcpu_on_unsupported_cpu(vcpu)					\
889 	vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
890 
891 #define vcpu_set_on_unsupported_cpu(vcpu)				\
892 	vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
893 
894 #define vcpu_clear_on_unsupported_cpu(vcpu)				\
895 	vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
896 
897 #define vcpu_gp_regs(v)		(&(v)->arch.ctxt.regs)
898 
899 /*
900  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
901  * memory backed version of a register, and not the one most recently
902  * accessed by a running VCPU.  For example, for userspace access or
903  * for system registers that are never context switched, but only
904  * emulated.
905  *
906  * Don't bother with VNCR-based accesses in the nVHE code, it has no
907  * business dealing with NV.
908  */
909 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
910 {
911 #if !defined (__KVM_NVHE_HYPERVISOR__)
912 	if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
913 		     r >= __VNCR_START__ && ctxt->vncr_array))
914 		return &ctxt->vncr_array[r - __VNCR_START__];
915 #endif
916 	return (u64 *)&ctxt->sys_regs[r];
917 }
918 
919 #define __ctxt_sys_reg(c,r)						\
920 	({								\
921 		BUILD_BUG_ON(__builtin_constant_p(r) &&			\
922 			     (r) >= NR_SYS_REGS);			\
923 		___ctxt_sys_reg(c, r);					\
924 	})
925 
926 #define ctxt_sys_reg(c,r)	(*__ctxt_sys_reg(c,r))
927 
928 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg);
929 #define __vcpu_sys_reg(v,r)						\
930 	(*({								\
931 		const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt;	\
932 		u64 *__r = __ctxt_sys_reg(ctxt, (r));			\
933 		if (vcpu_has_nv((v)) && (r) >= __VNCR_START__)		\
934 			*__r = kvm_vcpu_sanitise_vncr_reg((v), (r));	\
935 		__r;							\
936 	}))
937 
938 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
939 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
940 
941 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
942 {
943 	/*
944 	 * *** VHE ONLY ***
945 	 *
946 	 * System registers listed in the switch are not saved on every
947 	 * exit from the guest but are only saved on vcpu_put.
948 	 *
949 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
950 	 * should never be listed below, because the guest cannot modify its
951 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
952 	 * thread when emulating cross-VCPU communication.
953 	 */
954 	if (!has_vhe())
955 		return false;
956 
957 	switch (reg) {
958 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
959 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
960 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
961 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
962 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
963 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
964 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
965 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
966 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
967 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
968 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
969 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
970 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
971 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
972 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
973 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
974 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
975 	case ELR_EL1:		*val = read_sysreg_s(SYS_ELR_EL12);	break;
976 	case SPSR_EL1:		*val = read_sysreg_s(SYS_SPSR_EL12);	break;
977 	case PAR_EL1:		*val = read_sysreg_par();		break;
978 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
979 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
980 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
981 	default:		return false;
982 	}
983 
984 	return true;
985 }
986 
987 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
988 {
989 	/*
990 	 * *** VHE ONLY ***
991 	 *
992 	 * System registers listed in the switch are not restored on every
993 	 * entry to the guest but are only restored on vcpu_load.
994 	 *
995 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
996 	 * should never be listed below, because the MPIDR should only be set
997 	 * once, before running the VCPU, and never changed later.
998 	 */
999 	if (!has_vhe())
1000 		return false;
1001 
1002 	switch (reg) {
1003 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
1004 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
1005 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
1006 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
1007 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
1008 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
1009 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
1010 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
1011 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
1012 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
1013 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
1014 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
1015 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
1016 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
1017 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
1018 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
1019 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
1020 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
1021 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
1022 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
1023 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
1024 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
1025 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
1026 	default:		return false;
1027 	}
1028 
1029 	return true;
1030 }
1031 
1032 struct kvm_vm_stat {
1033 	struct kvm_vm_stat_generic generic;
1034 };
1035 
1036 struct kvm_vcpu_stat {
1037 	struct kvm_vcpu_stat_generic generic;
1038 	u64 hvc_exit_stat;
1039 	u64 wfe_exit_stat;
1040 	u64 wfi_exit_stat;
1041 	u64 mmio_exit_user;
1042 	u64 mmio_exit_kernel;
1043 	u64 signal_exits;
1044 	u64 exits;
1045 };
1046 
1047 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
1048 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
1049 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1050 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1051 
1052 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
1053 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
1054 
1055 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1056 			      struct kvm_vcpu_events *events);
1057 
1058 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1059 			      struct kvm_vcpu_events *events);
1060 
1061 void kvm_arm_halt_guest(struct kvm *kvm);
1062 void kvm_arm_resume_guest(struct kvm *kvm);
1063 
1064 #define vcpu_has_run_once(vcpu)	!!rcu_access_pointer((vcpu)->pid)
1065 
1066 #ifndef __KVM_NVHE_HYPERVISOR__
1067 #define kvm_call_hyp_nvhe(f, ...)						\
1068 	({								\
1069 		struct arm_smccc_res res;				\
1070 									\
1071 		arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),		\
1072 				  ##__VA_ARGS__, &res);			\
1073 		WARN_ON(res.a0 != SMCCC_RET_SUCCESS);			\
1074 									\
1075 		res.a1;							\
1076 	})
1077 
1078 /*
1079  * The couple of isb() below are there to guarantee the same behaviour
1080  * on VHE as on !VHE, where the eret to EL1 acts as a context
1081  * synchronization event.
1082  */
1083 #define kvm_call_hyp(f, ...)						\
1084 	do {								\
1085 		if (has_vhe()) {					\
1086 			f(__VA_ARGS__);					\
1087 			isb();						\
1088 		} else {						\
1089 			kvm_call_hyp_nvhe(f, ##__VA_ARGS__);		\
1090 		}							\
1091 	} while(0)
1092 
1093 #define kvm_call_hyp_ret(f, ...)					\
1094 	({								\
1095 		typeof(f(__VA_ARGS__)) ret;				\
1096 									\
1097 		if (has_vhe()) {					\
1098 			ret = f(__VA_ARGS__);				\
1099 			isb();						\
1100 		} else {						\
1101 			ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);	\
1102 		}							\
1103 									\
1104 		ret;							\
1105 	})
1106 #else /* __KVM_NVHE_HYPERVISOR__ */
1107 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1108 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1109 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1110 #endif /* __KVM_NVHE_HYPERVISOR__ */
1111 
1112 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1113 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1114 
1115 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1116 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1117 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1118 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1119 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1120 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1121 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1122 
1123 void kvm_sys_regs_create_debugfs(struct kvm *kvm);
1124 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1125 
1126 int __init kvm_sys_reg_table_init(void);
1127 struct sys_reg_desc;
1128 int __init populate_sysreg_config(const struct sys_reg_desc *sr,
1129 				  unsigned int idx);
1130 int __init populate_nv_trap_config(void);
1131 
1132 bool lock_all_vcpus(struct kvm *kvm);
1133 void unlock_all_vcpus(struct kvm *kvm);
1134 
1135 void kvm_init_sysreg(struct kvm_vcpu *);
1136 
1137 /* MMIO helpers */
1138 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1139 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1140 
1141 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1142 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1143 
1144 /*
1145  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1146  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1147  * loaded is considered to be "in guest".
1148  */
1149 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1150 {
1151 	return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1152 }
1153 
1154 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1155 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1156 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1157 
1158 bool kvm_arm_pvtime_supported(void);
1159 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1160 			    struct kvm_device_attr *attr);
1161 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1162 			    struct kvm_device_attr *attr);
1163 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1164 			    struct kvm_device_attr *attr);
1165 
1166 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1167 int __init kvm_arm_vmid_alloc_init(void);
1168 void __init kvm_arm_vmid_alloc_free(void);
1169 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1170 void kvm_arm_vmid_clear_active(void);
1171 
1172 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1173 {
1174 	vcpu_arch->steal.base = INVALID_GPA;
1175 }
1176 
1177 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1178 {
1179 	return (vcpu_arch->steal.base != INVALID_GPA);
1180 }
1181 
1182 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1183 
1184 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1185 
1186 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1187 
1188 /*
1189  * How we access per-CPU host data depends on the where we access it from,
1190  * and the mode we're in:
1191  *
1192  * - VHE and nVHE hypervisor bits use their locally defined instance
1193  *
1194  * - the rest of the kernel use either the VHE or nVHE one, depending on
1195  *   the mode we're running in.
1196  *
1197  *   Unless we're in protected mode, fully deprivileged, and the nVHE
1198  *   per-CPU stuff is exclusively accessible to the protected EL2 code.
1199  *   In this case, the EL1 code uses the *VHE* data as its private state
1200  *   (which makes sense in a way as there shouldn't be any shared state
1201  *   between the host and the hypervisor).
1202  *
1203  * Yes, this is all totally trivial. Shoot me now.
1204  */
1205 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__)
1206 #define host_data_ptr(f)	(&this_cpu_ptr(&kvm_host_data)->f)
1207 #else
1208 #define host_data_ptr(f)						\
1209 	(static_branch_unlikely(&kvm_protected_mode_initialized) ?	\
1210 	 &this_cpu_ptr(&kvm_host_data)->f :				\
1211 	 &this_cpu_ptr_hyp_sym(kvm_host_data)->f)
1212 #endif
1213 
1214 /* Check whether the FP regs are owned by the guest */
1215 static inline bool guest_owns_fp_regs(void)
1216 {
1217 	return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED;
1218 }
1219 
1220 /* Check whether the FP regs are owned by the host */
1221 static inline bool host_owns_fp_regs(void)
1222 {
1223 	return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED;
1224 }
1225 
1226 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1227 {
1228 	/* The host's MPIDR is immutable, so let's set it up at boot time */
1229 	ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1230 }
1231 
1232 static inline bool kvm_system_needs_idmapped_vectors(void)
1233 {
1234 	return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1235 }
1236 
1237 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1238 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1239 
1240 void kvm_arm_init_debug(void);
1241 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1242 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1243 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1244 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1245 
1246 #define kvm_vcpu_os_lock_enabled(vcpu)		\
1247 	(!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1248 
1249 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1250 			       struct kvm_device_attr *attr);
1251 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1252 			       struct kvm_device_attr *attr);
1253 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1254 			       struct kvm_device_attr *attr);
1255 
1256 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1257 			       struct kvm_arm_copy_mte_tags *copy_tags);
1258 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1259 				    struct kvm_arm_counter_offset *offset);
1260 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1261 					struct reg_mask_range *range);
1262 
1263 /* Guest/host FPSIMD coordination helpers */
1264 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1265 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1266 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1267 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1268 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1269 
1270 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1271 {
1272 	return (!has_vhe() && attr->exclude_host);
1273 }
1274 
1275 /* Flags for host debug state */
1276 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1277 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1278 
1279 #ifdef CONFIG_KVM
1280 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1281 void kvm_clr_pmu_events(u32 clr);
1282 bool kvm_set_pmuserenr(u64 val);
1283 #else
1284 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1285 static inline void kvm_clr_pmu_events(u32 clr) {}
1286 static inline bool kvm_set_pmuserenr(u64 val)
1287 {
1288 	return false;
1289 }
1290 #endif
1291 
1292 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1293 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1294 
1295 int __init kvm_set_ipa_limit(void);
1296 
1297 #define __KVM_HAVE_ARCH_VM_ALLOC
1298 struct kvm *kvm_arch_alloc_vm(void);
1299 
1300 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1301 
1302 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1303 
1304 #define kvm_vm_is_protected(kvm)	(is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled)
1305 
1306 #define vcpu_is_protected(vcpu)		kvm_vm_is_protected((vcpu)->kvm)
1307 
1308 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1309 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1310 
1311 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1312 
1313 #define kvm_has_mte(kvm)					\
1314 	(system_supports_mte() &&				\
1315 	 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1316 
1317 #define kvm_supports_32bit_el0()				\
1318 	(system_supports_32bit_el0() &&				\
1319 	 !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1320 
1321 #define kvm_vm_has_ran_once(kvm)					\
1322 	(test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1323 
1324 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1325 {
1326 	return test_bit(feature, ka->vcpu_features);
1327 }
1328 
1329 #define vcpu_has_feature(v, f)	__vcpu_has_feature(&(v)->kvm->arch, (f))
1330 
1331 int kvm_trng_call(struct kvm_vcpu *vcpu);
1332 #ifdef CONFIG_KVM
1333 extern phys_addr_t hyp_mem_base;
1334 extern phys_addr_t hyp_mem_size;
1335 void __init kvm_hyp_reserve(void);
1336 #else
1337 static inline void kvm_hyp_reserve(void) { }
1338 #endif
1339 
1340 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1341 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1342 
1343 #define __expand_field_sign_unsigned(id, fld, val)			\
1344 	((u64)SYS_FIELD_VALUE(id, fld, val))
1345 
1346 #define __expand_field_sign_signed(id, fld, val)			\
1347 	({								\
1348 		u64 __val = SYS_FIELD_VALUE(id, fld, val);		\
1349 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1350 	})
1351 
1352 #define expand_field_sign(id, fld, val)					\
1353 	(id##_##fld##_SIGNED ?						\
1354 	 __expand_field_sign_signed(id, fld, val) :			\
1355 	 __expand_field_sign_unsigned(id, fld, val))
1356 
1357 #define get_idreg_field_unsigned(kvm, id, fld)				\
1358 	({								\
1359 		u64 __val = IDREG((kvm), SYS_##id);			\
1360 		FIELD_GET(id##_##fld##_MASK, __val);			\
1361 	})
1362 
1363 #define get_idreg_field_signed(kvm, id, fld)				\
1364 	({								\
1365 		u64 __val = get_idreg_field_unsigned(kvm, id, fld);	\
1366 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1367 	})
1368 
1369 #define get_idreg_field_enum(kvm, id, fld)				\
1370 	get_idreg_field_unsigned(kvm, id, fld)
1371 
1372 #define get_idreg_field(kvm, id, fld)					\
1373 	(id##_##fld##_SIGNED ?						\
1374 	 get_idreg_field_signed(kvm, id, fld) :				\
1375 	 get_idreg_field_unsigned(kvm, id, fld))
1376 
1377 #define kvm_has_feat(kvm, id, fld, limit)				\
1378 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit))
1379 
1380 #define kvm_has_feat_enum(kvm, id, fld, val)				\
1381 	(get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val))
1382 
1383 #define kvm_has_feat_range(kvm, id, fld, min, max)			\
1384 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \
1385 	 get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max))
1386 
1387 /* Check for a given level of PAuth support */
1388 #define kvm_has_pauth(k, l)						\
1389 	({								\
1390 		bool pa, pi, pa3;					\
1391 									\
1392 		pa  = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l);	\
1393 		pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP);	\
1394 		pi  = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l);	\
1395 		pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP);	\
1396 		pa3  = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l);	\
1397 		pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP);	\
1398 									\
1399 		(pa + pi + pa3) == 1;					\
1400 	})
1401 
1402 #endif /* __ARM64_KVM_HOST_H__ */
1403