xref: /linux/arch/arm64/include/asm/kvm_host.h (revision 9e56ff53b4115875667760445b028357848b4748)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/include/asm/kvm_host.h:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
9  */
10 
11 #ifndef __ARM64_KVM_HOST_H__
12 #define __ARM64_KVM_HOST_H__
13 
14 #include <linux/arm-smccc.h>
15 #include <linux/bitmap.h>
16 #include <linux/types.h>
17 #include <linux/jump_label.h>
18 #include <linux/kvm_types.h>
19 #include <linux/maple_tree.h>
20 #include <linux/percpu.h>
21 #include <linux/psci.h>
22 #include <asm/arch_gicv3.h>
23 #include <asm/barrier.h>
24 #include <asm/cpufeature.h>
25 #include <asm/cputype.h>
26 #include <asm/daifflags.h>
27 #include <asm/fpsimd.h>
28 #include <asm/kvm.h>
29 #include <asm/kvm_asm.h>
30 #include <asm/vncr_mapping.h>
31 
32 #define __KVM_HAVE_ARCH_INTC_INITIALIZED
33 
34 #define KVM_HALT_POLL_NS_DEFAULT 500000
35 
36 #include <kvm/arm_vgic.h>
37 #include <kvm/arm_arch_timer.h>
38 #include <kvm/arm_pmu.h>
39 
40 #define KVM_MAX_VCPUS VGIC_V3_MAX_CPUS
41 
42 #define KVM_VCPU_MAX_FEATURES 7
43 #define KVM_VCPU_VALID_FEATURES	(BIT(KVM_VCPU_MAX_FEATURES) - 1)
44 
45 #define KVM_REQ_SLEEP \
46 	KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
47 #define KVM_REQ_IRQ_PENDING	KVM_ARCH_REQ(1)
48 #define KVM_REQ_VCPU_RESET	KVM_ARCH_REQ(2)
49 #define KVM_REQ_RECORD_STEAL	KVM_ARCH_REQ(3)
50 #define KVM_REQ_RELOAD_GICv4	KVM_ARCH_REQ(4)
51 #define KVM_REQ_RELOAD_PMU	KVM_ARCH_REQ(5)
52 #define KVM_REQ_SUSPEND		KVM_ARCH_REQ(6)
53 #define KVM_REQ_RESYNC_PMU_EL0	KVM_ARCH_REQ(7)
54 
55 #define KVM_DIRTY_LOG_MANUAL_CAPS   (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \
56 				     KVM_DIRTY_LOG_INITIALLY_SET)
57 
58 #define KVM_HAVE_MMU_RWLOCK
59 
60 /*
61  * Mode of operation configurable with kvm-arm.mode early param.
62  * See Documentation/admin-guide/kernel-parameters.txt for more information.
63  */
64 enum kvm_mode {
65 	KVM_MODE_DEFAULT,
66 	KVM_MODE_PROTECTED,
67 	KVM_MODE_NV,
68 	KVM_MODE_NONE,
69 };
70 #ifdef CONFIG_KVM
71 enum kvm_mode kvm_get_mode(void);
72 #else
73 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; };
74 #endif
75 
76 DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
77 
78 extern unsigned int __ro_after_init kvm_sve_max_vl;
79 int __init kvm_arm_init_sve(void);
80 
81 u32 __attribute_const__ kvm_target_cpu(void);
82 void kvm_reset_vcpu(struct kvm_vcpu *vcpu);
83 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu);
84 
85 struct kvm_hyp_memcache {
86 	phys_addr_t head;
87 	unsigned long nr_pages;
88 };
89 
90 static inline void push_hyp_memcache(struct kvm_hyp_memcache *mc,
91 				     phys_addr_t *p,
92 				     phys_addr_t (*to_pa)(void *virt))
93 {
94 	*p = mc->head;
95 	mc->head = to_pa(p);
96 	mc->nr_pages++;
97 }
98 
99 static inline void *pop_hyp_memcache(struct kvm_hyp_memcache *mc,
100 				     void *(*to_va)(phys_addr_t phys))
101 {
102 	phys_addr_t *p = to_va(mc->head);
103 
104 	if (!mc->nr_pages)
105 		return NULL;
106 
107 	mc->head = *p;
108 	mc->nr_pages--;
109 
110 	return p;
111 }
112 
113 static inline int __topup_hyp_memcache(struct kvm_hyp_memcache *mc,
114 				       unsigned long min_pages,
115 				       void *(*alloc_fn)(void *arg),
116 				       phys_addr_t (*to_pa)(void *virt),
117 				       void *arg)
118 {
119 	while (mc->nr_pages < min_pages) {
120 		phys_addr_t *p = alloc_fn(arg);
121 
122 		if (!p)
123 			return -ENOMEM;
124 		push_hyp_memcache(mc, p, to_pa);
125 	}
126 
127 	return 0;
128 }
129 
130 static inline void __free_hyp_memcache(struct kvm_hyp_memcache *mc,
131 				       void (*free_fn)(void *virt, void *arg),
132 				       void *(*to_va)(phys_addr_t phys),
133 				       void *arg)
134 {
135 	while (mc->nr_pages)
136 		free_fn(pop_hyp_memcache(mc, to_va), arg);
137 }
138 
139 void free_hyp_memcache(struct kvm_hyp_memcache *mc);
140 int topup_hyp_memcache(struct kvm_hyp_memcache *mc, unsigned long min_pages);
141 
142 struct kvm_vmid {
143 	atomic64_t id;
144 };
145 
146 struct kvm_s2_mmu {
147 	struct kvm_vmid vmid;
148 
149 	/*
150 	 * stage2 entry level table
151 	 *
152 	 * Two kvm_s2_mmu structures in the same VM can point to the same
153 	 * pgd here.  This happens when running a guest using a
154 	 * translation regime that isn't affected by its own stage-2
155 	 * translation, such as a non-VHE hypervisor running at vEL2, or
156 	 * for vEL1/EL0 with vHCR_EL2.VM == 0.  In that case, we use the
157 	 * canonical stage-2 page tables.
158 	 */
159 	phys_addr_t	pgd_phys;
160 	struct kvm_pgtable *pgt;
161 
162 	/*
163 	 * VTCR value used on the host. For a non-NV guest (or a NV
164 	 * guest that runs in a context where its own S2 doesn't
165 	 * apply), its T0SZ value reflects that of the IPA size.
166 	 *
167 	 * For a shadow S2 MMU, T0SZ reflects the PARange exposed to
168 	 * the guest.
169 	 */
170 	u64	vtcr;
171 
172 	/* The last vcpu id that ran on each physical CPU */
173 	int __percpu *last_vcpu_ran;
174 
175 #define KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT 0
176 	/*
177 	 * Memory cache used to split
178 	 * KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE worth of huge pages. It
179 	 * is used to allocate stage2 page tables while splitting huge
180 	 * pages. The choice of KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE
181 	 * influences both the capacity of the split page cache, and
182 	 * how often KVM reschedules. Be wary of raising CHUNK_SIZE
183 	 * too high.
184 	 *
185 	 * Protected by kvm->slots_lock.
186 	 */
187 	struct kvm_mmu_memory_cache split_page_cache;
188 	uint64_t split_page_chunk_size;
189 
190 	struct kvm_arch *arch;
191 };
192 
193 struct kvm_arch_memory_slot {
194 };
195 
196 /**
197  * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests
198  *
199  * @std_bmap: Bitmap of standard secure service calls
200  * @std_hyp_bmap: Bitmap of standard hypervisor service calls
201  * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls
202  */
203 struct kvm_smccc_features {
204 	unsigned long std_bmap;
205 	unsigned long std_hyp_bmap;
206 	unsigned long vendor_hyp_bmap;
207 };
208 
209 typedef unsigned int pkvm_handle_t;
210 
211 struct kvm_protected_vm {
212 	pkvm_handle_t handle;
213 	struct kvm_hyp_memcache teardown_mc;
214 };
215 
216 struct kvm_mpidr_data {
217 	u64			mpidr_mask;
218 	DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx);
219 };
220 
221 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr)
222 {
223 	unsigned long mask = data->mpidr_mask;
224 	u64 aff = mpidr & MPIDR_HWID_BITMASK;
225 	int nbits, bit, bit_idx = 0;
226 	u16 index = 0;
227 
228 	/*
229 	 * If this looks like RISC-V's BEXT or x86's PEXT
230 	 * instructions, it isn't by accident.
231 	 */
232 	nbits = fls(mask);
233 	for_each_set_bit(bit, &mask, nbits) {
234 		index |= (aff & BIT(bit)) >> (bit - bit_idx);
235 		bit_idx++;
236 	}
237 
238 	return index;
239 }
240 
241 struct kvm_arch {
242 	struct kvm_s2_mmu mmu;
243 
244 	/* Interrupt controller */
245 	struct vgic_dist	vgic;
246 
247 	/* Timers */
248 	struct arch_timer_vm_data timer_data;
249 
250 	/* Mandated version of PSCI */
251 	u32 psci_version;
252 
253 	/* Protects VM-scoped configuration data */
254 	struct mutex config_lock;
255 
256 	/*
257 	 * If we encounter a data abort without valid instruction syndrome
258 	 * information, report this to user space.  User space can (and
259 	 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
260 	 * supported.
261 	 */
262 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER	0
263 	/* Memory Tagging Extension enabled for the guest */
264 #define KVM_ARCH_FLAG_MTE_ENABLED			1
265 	/* At least one vCPU has ran in the VM */
266 #define KVM_ARCH_FLAG_HAS_RAN_ONCE			2
267 	/* The vCPU feature set for the VM is configured */
268 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED		3
269 	/* PSCI SYSTEM_SUSPEND enabled for the guest */
270 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED		4
271 	/* VM counter offset */
272 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET			5
273 	/* Timer PPIs made immutable */
274 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE		6
275 	/* Initial ID reg values loaded */
276 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED		7
277 	unsigned long flags;
278 
279 	/* VM-wide vCPU feature set */
280 	DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
281 
282 	/* MPIDR to vcpu index mapping, optional */
283 	struct kvm_mpidr_data *mpidr_data;
284 
285 	/*
286 	 * VM-wide PMU filter, implemented as a bitmap and big enough for
287 	 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
288 	 */
289 	unsigned long *pmu_filter;
290 	struct arm_pmu *arm_pmu;
291 
292 	cpumask_var_t supported_cpus;
293 
294 	/* PMCR_EL0.N value for the guest */
295 	u8 pmcr_n;
296 
297 	/* Hypercall features firmware registers' descriptor */
298 	struct kvm_smccc_features smccc_feat;
299 	struct maple_tree smccc_filter;
300 
301 	/*
302 	 * Emulated CPU ID registers per VM
303 	 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
304 	 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
305 	 *
306 	 * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
307 	 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
308 	 */
309 #define IDREG_IDX(id)		(((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
310 #define IDX_IDREG(idx)		sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask)
311 #define IDREG(kvm, id)		((kvm)->arch.id_regs[IDREG_IDX(id)])
312 #define KVM_ARM_ID_REG_NUM	(IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
313 	u64 id_regs[KVM_ARM_ID_REG_NUM];
314 
315 	/*
316 	 * For an untrusted host VM, 'pkvm.handle' is used to lookup
317 	 * the associated pKVM instance in the hypervisor.
318 	 */
319 	struct kvm_protected_vm pkvm;
320 };
321 
322 struct kvm_vcpu_fault_info {
323 	u64 esr_el2;		/* Hyp Syndrom Register */
324 	u64 far_el2;		/* Hyp Fault Address Register */
325 	u64 hpfar_el2;		/* Hyp IPA Fault Address Register */
326 	u64 disr_el1;		/* Deferred [SError] Status Register */
327 };
328 
329 /*
330  * VNCR() just places the VNCR_capable registers in the enum after
331  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
332  * from the VNCR base. As we don't require the enum to be otherwise ordered,
333  * we need the terrible hack below to ensure that we correctly size the
334  * sys_regs array, no matter what.
335  *
336  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
337  * treasure trove of bit hacks:
338  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
339  */
340 #define __MAX__(x,y)	((x) ^ (((x) ^ (y)) & -((x) < (y))))
341 #define VNCR(r)						\
342 	__before_##r,					\
343 	r = __VNCR_START__ + ((VNCR_ ## r) / 8),	\
344 	__after_##r = __MAX__(__before_##r - 1, r)
345 
346 enum vcpu_sysreg {
347 	__INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
348 	MPIDR_EL1,	/* MultiProcessor Affinity Register */
349 	CLIDR_EL1,	/* Cache Level ID Register */
350 	CSSELR_EL1,	/* Cache Size Selection Register */
351 	TPIDR_EL0,	/* Thread ID, User R/W */
352 	TPIDRRO_EL0,	/* Thread ID, User R/O */
353 	TPIDR_EL1,	/* Thread ID, Privileged */
354 	CNTKCTL_EL1,	/* Timer Control Register (EL1) */
355 	PAR_EL1,	/* Physical Address Register */
356 	MDCCINT_EL1,	/* Monitor Debug Comms Channel Interrupt Enable Reg */
357 	OSLSR_EL1,	/* OS Lock Status Register */
358 	DISR_EL1,	/* Deferred Interrupt Status Register */
359 
360 	/* Performance Monitors Registers */
361 	PMCR_EL0,	/* Control Register */
362 	PMSELR_EL0,	/* Event Counter Selection Register */
363 	PMEVCNTR0_EL0,	/* Event Counter Register (0-30) */
364 	PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
365 	PMCCNTR_EL0,	/* Cycle Counter Register */
366 	PMEVTYPER0_EL0,	/* Event Type Register (0-30) */
367 	PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
368 	PMCCFILTR_EL0,	/* Cycle Count Filter Register */
369 	PMCNTENSET_EL0,	/* Count Enable Set Register */
370 	PMINTENSET_EL1,	/* Interrupt Enable Set Register */
371 	PMOVSSET_EL0,	/* Overflow Flag Status Set Register */
372 	PMUSERENR_EL0,	/* User Enable Register */
373 
374 	/* Pointer Authentication Registers in a strict increasing order. */
375 	APIAKEYLO_EL1,
376 	APIAKEYHI_EL1,
377 	APIBKEYLO_EL1,
378 	APIBKEYHI_EL1,
379 	APDAKEYLO_EL1,
380 	APDAKEYHI_EL1,
381 	APDBKEYLO_EL1,
382 	APDBKEYHI_EL1,
383 	APGAKEYLO_EL1,
384 	APGAKEYHI_EL1,
385 
386 	/* Memory Tagging Extension registers */
387 	RGSR_EL1,	/* Random Allocation Tag Seed Register */
388 	GCR_EL1,	/* Tag Control Register */
389 	TFSRE0_EL1,	/* Tag Fault Status Register (EL0) */
390 
391 	/* 32bit specific registers. */
392 	DACR32_EL2,	/* Domain Access Control Register */
393 	IFSR32_EL2,	/* Instruction Fault Status Register */
394 	FPEXC32_EL2,	/* Floating-Point Exception Control Register */
395 	DBGVCR32_EL2,	/* Debug Vector Catch Register */
396 
397 	/* EL2 registers */
398 	SCTLR_EL2,	/* System Control Register (EL2) */
399 	ACTLR_EL2,	/* Auxiliary Control Register (EL2) */
400 	MDCR_EL2,	/* Monitor Debug Configuration Register (EL2) */
401 	CPTR_EL2,	/* Architectural Feature Trap Register (EL2) */
402 	HACR_EL2,	/* Hypervisor Auxiliary Control Register */
403 	TTBR0_EL2,	/* Translation Table Base Register 0 (EL2) */
404 	TTBR1_EL2,	/* Translation Table Base Register 1 (EL2) */
405 	TCR_EL2,	/* Translation Control Register (EL2) */
406 	SPSR_EL2,	/* EL2 saved program status register */
407 	ELR_EL2,	/* EL2 exception link register */
408 	AFSR0_EL2,	/* Auxiliary Fault Status Register 0 (EL2) */
409 	AFSR1_EL2,	/* Auxiliary Fault Status Register 1 (EL2) */
410 	ESR_EL2,	/* Exception Syndrome Register (EL2) */
411 	FAR_EL2,	/* Fault Address Register (EL2) */
412 	HPFAR_EL2,	/* Hypervisor IPA Fault Address Register */
413 	MAIR_EL2,	/* Memory Attribute Indirection Register (EL2) */
414 	AMAIR_EL2,	/* Auxiliary Memory Attribute Indirection Register (EL2) */
415 	VBAR_EL2,	/* Vector Base Address Register (EL2) */
416 	RVBAR_EL2,	/* Reset Vector Base Address Register */
417 	CONTEXTIDR_EL2,	/* Context ID Register (EL2) */
418 	CNTHCTL_EL2,	/* Counter-timer Hypervisor Control register */
419 	SP_EL2,		/* EL2 Stack Pointer */
420 	CNTHP_CTL_EL2,
421 	CNTHP_CVAL_EL2,
422 	CNTHV_CTL_EL2,
423 	CNTHV_CVAL_EL2,
424 
425 	__VNCR_START__,	/* Any VNCR-capable reg goes after this point */
426 
427 	VNCR(SCTLR_EL1),/* System Control Register */
428 	VNCR(ACTLR_EL1),/* Auxiliary Control Register */
429 	VNCR(CPACR_EL1),/* Coprocessor Access Control */
430 	VNCR(ZCR_EL1),	/* SVE Control */
431 	VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
432 	VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
433 	VNCR(TCR_EL1),	/* Translation Control Register */
434 	VNCR(TCR2_EL1),	/* Extended Translation Control Register */
435 	VNCR(ESR_EL1),	/* Exception Syndrome Register */
436 	VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
437 	VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
438 	VNCR(FAR_EL1),	/* Fault Address Register */
439 	VNCR(MAIR_EL1),	/* Memory Attribute Indirection Register */
440 	VNCR(VBAR_EL1),	/* Vector Base Address Register */
441 	VNCR(CONTEXTIDR_EL1),	/* Context ID Register */
442 	VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
443 	VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
444 	VNCR(ELR_EL1),
445 	VNCR(SP_EL1),
446 	VNCR(SPSR_EL1),
447 	VNCR(TFSR_EL1),	/* Tag Fault Status Register (EL1) */
448 	VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
449 	VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
450 	VNCR(HCR_EL2),	/* Hypervisor Configuration Register */
451 	VNCR(HSTR_EL2),	/* Hypervisor System Trap Register */
452 	VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
453 	VNCR(VTCR_EL2),	/* Virtualization Translation Control Register */
454 	VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
455 	VNCR(HCRX_EL2),	/* Extended Hypervisor Configuration Register */
456 
457 	/* Permission Indirection Extension registers */
458 	VNCR(PIR_EL1),	 /* Permission Indirection Register 1 (EL1) */
459 	VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
460 
461 	VNCR(HFGRTR_EL2),
462 	VNCR(HFGWTR_EL2),
463 	VNCR(HFGITR_EL2),
464 	VNCR(HDFGRTR_EL2),
465 	VNCR(HDFGWTR_EL2),
466 	VNCR(HAFGRTR_EL2),
467 
468 	VNCR(CNTVOFF_EL2),
469 	VNCR(CNTV_CVAL_EL0),
470 	VNCR(CNTV_CTL_EL0),
471 	VNCR(CNTP_CVAL_EL0),
472 	VNCR(CNTP_CTL_EL0),
473 
474 	NR_SYS_REGS	/* Nothing after this line! */
475 };
476 
477 struct kvm_cpu_context {
478 	struct user_pt_regs regs;	/* sp = sp_el0 */
479 
480 	u64	spsr_abt;
481 	u64	spsr_und;
482 	u64	spsr_irq;
483 	u64	spsr_fiq;
484 
485 	struct user_fpsimd_state fp_regs;
486 
487 	u64 sys_regs[NR_SYS_REGS];
488 
489 	struct kvm_vcpu *__hyp_running_vcpu;
490 
491 	/* This pointer has to be 4kB aligned. */
492 	u64 *vncr_array;
493 };
494 
495 struct kvm_host_data {
496 	struct kvm_cpu_context host_ctxt;
497 };
498 
499 struct kvm_host_psci_config {
500 	/* PSCI version used by host. */
501 	u32 version;
502 	u32 smccc_version;
503 
504 	/* Function IDs used by host if version is v0.1. */
505 	struct psci_0_1_function_ids function_ids_0_1;
506 
507 	bool psci_0_1_cpu_suspend_implemented;
508 	bool psci_0_1_cpu_on_implemented;
509 	bool psci_0_1_cpu_off_implemented;
510 	bool psci_0_1_migrate_implemented;
511 };
512 
513 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
514 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
515 
516 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
517 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
518 
519 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
520 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
521 
522 struct vcpu_reset_state {
523 	unsigned long	pc;
524 	unsigned long	r0;
525 	bool		be;
526 	bool		reset;
527 };
528 
529 struct kvm_vcpu_arch {
530 	struct kvm_cpu_context ctxt;
531 
532 	/*
533 	 * Guest floating point state
534 	 *
535 	 * The architecture has two main floating point extensions,
536 	 * the original FPSIMD and SVE.  These have overlapping
537 	 * register views, with the FPSIMD V registers occupying the
538 	 * low 128 bits of the SVE Z registers.  When the core
539 	 * floating point code saves the register state of a task it
540 	 * records which view it saved in fp_type.
541 	 */
542 	void *sve_state;
543 	enum fp_type fp_type;
544 	unsigned int sve_max_vl;
545 	u64 svcr;
546 
547 	/* Stage 2 paging state used by the hardware on next switch */
548 	struct kvm_s2_mmu *hw_mmu;
549 
550 	/* Values of trap registers for the guest. */
551 	u64 hcr_el2;
552 	u64 mdcr_el2;
553 	u64 cptr_el2;
554 
555 	/* Values of trap registers for the host before guest entry. */
556 	u64 mdcr_el2_host;
557 
558 	/* Exception Information */
559 	struct kvm_vcpu_fault_info fault;
560 
561 	/* Ownership of the FP regs */
562 	enum {
563 		FP_STATE_FREE,
564 		FP_STATE_HOST_OWNED,
565 		FP_STATE_GUEST_OWNED,
566 	} fp_state;
567 
568 	/* Configuration flags, set once and for all before the vcpu can run */
569 	u8 cflags;
570 
571 	/* Input flags to the hypervisor code, potentially cleared after use */
572 	u8 iflags;
573 
574 	/* State flags for kernel bookkeeping, unused by the hypervisor code */
575 	u8 sflags;
576 
577 	/*
578 	 * Don't run the guest (internal implementation need).
579 	 *
580 	 * Contrary to the flags above, this is set/cleared outside of
581 	 * a vcpu context, and thus cannot be mixed with the flags
582 	 * themselves (or the flag accesses need to be made atomic).
583 	 */
584 	bool pause;
585 
586 	/*
587 	 * We maintain more than a single set of debug registers to support
588 	 * debugging the guest from the host and to maintain separate host and
589 	 * guest state during world switches. vcpu_debug_state are the debug
590 	 * registers of the vcpu as the guest sees them.  host_debug_state are
591 	 * the host registers which are saved and restored during
592 	 * world switches. external_debug_state contains the debug
593 	 * values we want to debug the guest. This is set via the
594 	 * KVM_SET_GUEST_DEBUG ioctl.
595 	 *
596 	 * debug_ptr points to the set of debug registers that should be loaded
597 	 * onto the hardware when running the guest.
598 	 */
599 	struct kvm_guest_debug_arch *debug_ptr;
600 	struct kvm_guest_debug_arch vcpu_debug_state;
601 	struct kvm_guest_debug_arch external_debug_state;
602 
603 	struct user_fpsimd_state *host_fpsimd_state;	/* hyp VA */
604 	struct task_struct *parent_task;
605 
606 	struct {
607 		/* {Break,watch}point registers */
608 		struct kvm_guest_debug_arch regs;
609 		/* Statistical profiling extension */
610 		u64 pmscr_el1;
611 		/* Self-hosted trace */
612 		u64 trfcr_el1;
613 	} host_debug_state;
614 
615 	/* VGIC state */
616 	struct vgic_cpu vgic_cpu;
617 	struct arch_timer_cpu timer_cpu;
618 	struct kvm_pmu pmu;
619 
620 	/*
621 	 * Guest registers we preserve during guest debugging.
622 	 *
623 	 * These shadow registers are updated by the kvm_handle_sys_reg
624 	 * trap handler if the guest accesses or updates them while we
625 	 * are using guest debug.
626 	 */
627 	struct {
628 		u32	mdscr_el1;
629 		bool	pstate_ss;
630 	} guest_debug_preserved;
631 
632 	/* vcpu power state */
633 	struct kvm_mp_state mp_state;
634 	spinlock_t mp_state_lock;
635 
636 	/* Cache some mmu pages needed inside spinlock regions */
637 	struct kvm_mmu_memory_cache mmu_page_cache;
638 
639 	/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
640 	u64 vsesr_el2;
641 
642 	/* Additional reset state */
643 	struct vcpu_reset_state	reset_state;
644 
645 	/* Guest PV state */
646 	struct {
647 		u64 last_steal;
648 		gpa_t base;
649 	} steal;
650 
651 	/* Per-vcpu CCSIDR override or NULL */
652 	u32 *ccsidr;
653 };
654 
655 /*
656  * Each 'flag' is composed of a comma-separated triplet:
657  *
658  * - the flag-set it belongs to in the vcpu->arch structure
659  * - the value for that flag
660  * - the mask for that flag
661  *
662  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
663  * unpack_vcpu_flag() extract the flag value from the triplet for
664  * direct use outside of the flag accessors.
665  */
666 #define __vcpu_single_flag(_set, _f)	_set, (_f), (_f)
667 
668 #define __unpack_flag(_set, _f, _m)	_f
669 #define unpack_vcpu_flag(...)		__unpack_flag(__VA_ARGS__)
670 
671 #define __build_check_flag(v, flagset, f, m)			\
672 	do {							\
673 		typeof(v->arch.flagset) *_fset;			\
674 								\
675 		/* Check that the flags fit in the mask */	\
676 		BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m)));	\
677 		/* Check that the flags fit in the type */	\
678 		BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m));	\
679 	} while (0)
680 
681 #define __vcpu_get_flag(v, flagset, f, m)			\
682 	({							\
683 		__build_check_flag(v, flagset, f, m);		\
684 								\
685 		READ_ONCE(v->arch.flagset) & (m);		\
686 	})
687 
688 /*
689  * Note that the set/clear accessors must be preempt-safe in order to
690  * avoid nesting them with load/put which also manipulate flags...
691  */
692 #ifdef __KVM_NVHE_HYPERVISOR__
693 /* the nVHE hypervisor is always non-preemptible */
694 #define __vcpu_flags_preempt_disable()
695 #define __vcpu_flags_preempt_enable()
696 #else
697 #define __vcpu_flags_preempt_disable()	preempt_disable()
698 #define __vcpu_flags_preempt_enable()	preempt_enable()
699 #endif
700 
701 #define __vcpu_set_flag(v, flagset, f, m)			\
702 	do {							\
703 		typeof(v->arch.flagset) *fset;			\
704 								\
705 		__build_check_flag(v, flagset, f, m);		\
706 								\
707 		fset = &v->arch.flagset;			\
708 		__vcpu_flags_preempt_disable();			\
709 		if (HWEIGHT(m) > 1)				\
710 			*fset &= ~(m);				\
711 		*fset |= (f);					\
712 		__vcpu_flags_preempt_enable();			\
713 	} while (0)
714 
715 #define __vcpu_clear_flag(v, flagset, f, m)			\
716 	do {							\
717 		typeof(v->arch.flagset) *fset;			\
718 								\
719 		__build_check_flag(v, flagset, f, m);		\
720 								\
721 		fset = &v->arch.flagset;			\
722 		__vcpu_flags_preempt_disable();			\
723 		*fset &= ~(m);					\
724 		__vcpu_flags_preempt_enable();			\
725 	} while (0)
726 
727 #define vcpu_get_flag(v, ...)	__vcpu_get_flag((v), __VA_ARGS__)
728 #define vcpu_set_flag(v, ...)	__vcpu_set_flag((v), __VA_ARGS__)
729 #define vcpu_clear_flag(v, ...)	__vcpu_clear_flag((v), __VA_ARGS__)
730 
731 /* SVE exposed to guest */
732 #define GUEST_HAS_SVE		__vcpu_single_flag(cflags, BIT(0))
733 /* SVE config completed */
734 #define VCPU_SVE_FINALIZED	__vcpu_single_flag(cflags, BIT(1))
735 /* PTRAUTH exposed to guest */
736 #define GUEST_HAS_PTRAUTH	__vcpu_single_flag(cflags, BIT(2))
737 /* KVM_ARM_VCPU_INIT completed */
738 #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(3))
739 
740 /* Exception pending */
741 #define PENDING_EXCEPTION	__vcpu_single_flag(iflags, BIT(0))
742 /*
743  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
744  * be set together with an exception...
745  */
746 #define INCREMENT_PC		__vcpu_single_flag(iflags, BIT(1))
747 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
748 #define EXCEPT_MASK		__vcpu_single_flag(iflags, GENMASK(3, 1))
749 
750 /* Helpers to encode exceptions with minimum fuss */
751 #define __EXCEPT_MASK_VAL	unpack_vcpu_flag(EXCEPT_MASK)
752 #define __EXCEPT_SHIFT		__builtin_ctzl(__EXCEPT_MASK_VAL)
753 #define __vcpu_except_flags(_f)	iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
754 
755 /*
756  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
757  * values:
758  *
759  * For AArch32 EL1:
760  */
761 #define EXCEPT_AA32_UND		__vcpu_except_flags(0)
762 #define EXCEPT_AA32_IABT	__vcpu_except_flags(1)
763 #define EXCEPT_AA32_DABT	__vcpu_except_flags(2)
764 /* For AArch64: */
765 #define EXCEPT_AA64_EL1_SYNC	__vcpu_except_flags(0)
766 #define EXCEPT_AA64_EL1_IRQ	__vcpu_except_flags(1)
767 #define EXCEPT_AA64_EL1_FIQ	__vcpu_except_flags(2)
768 #define EXCEPT_AA64_EL1_SERR	__vcpu_except_flags(3)
769 /* For AArch64 with NV: */
770 #define EXCEPT_AA64_EL2_SYNC	__vcpu_except_flags(4)
771 #define EXCEPT_AA64_EL2_IRQ	__vcpu_except_flags(5)
772 #define EXCEPT_AA64_EL2_FIQ	__vcpu_except_flags(6)
773 #define EXCEPT_AA64_EL2_SERR	__vcpu_except_flags(7)
774 /* Guest debug is live */
775 #define DEBUG_DIRTY		__vcpu_single_flag(iflags, BIT(4))
776 /* Save SPE context if active  */
777 #define DEBUG_STATE_SAVE_SPE	__vcpu_single_flag(iflags, BIT(5))
778 /* Save TRBE context if active  */
779 #define DEBUG_STATE_SAVE_TRBE	__vcpu_single_flag(iflags, BIT(6))
780 /* vcpu running in HYP context */
781 #define VCPU_HYP_CONTEXT	__vcpu_single_flag(iflags, BIT(7))
782 
783 /* SVE enabled for host EL0 */
784 #define HOST_SVE_ENABLED	__vcpu_single_flag(sflags, BIT(0))
785 /* SME enabled for EL0 */
786 #define HOST_SME_ENABLED	__vcpu_single_flag(sflags, BIT(1))
787 /* Physical CPU not in supported_cpus */
788 #define ON_UNSUPPORTED_CPU	__vcpu_single_flag(sflags, BIT(2))
789 /* WFIT instruction trapped */
790 #define IN_WFIT			__vcpu_single_flag(sflags, BIT(3))
791 /* vcpu system registers loaded on physical CPU */
792 #define SYSREGS_ON_CPU		__vcpu_single_flag(sflags, BIT(4))
793 /* Software step state is Active-pending */
794 #define DBG_SS_ACTIVE_PENDING	__vcpu_single_flag(sflags, BIT(5))
795 /* PMUSERENR for the guest EL0 is on physical CPU */
796 #define PMUSERENR_ON_CPU	__vcpu_single_flag(sflags, BIT(6))
797 /* WFI instruction trapped */
798 #define IN_WFI			__vcpu_single_flag(sflags, BIT(7))
799 
800 
801 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
802 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +	\
803 			     sve_ffr_offset((vcpu)->arch.sve_max_vl))
804 
805 #define vcpu_sve_max_vq(vcpu)	sve_vq_from_vl((vcpu)->arch.sve_max_vl)
806 
807 #define vcpu_sve_state_size(vcpu) ({					\
808 	size_t __size_ret;						\
809 	unsigned int __vcpu_vq;						\
810 									\
811 	if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {		\
812 		__size_ret = 0;						\
813 	} else {							\
814 		__vcpu_vq = vcpu_sve_max_vq(vcpu);			\
815 		__size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);		\
816 	}								\
817 									\
818 	__size_ret;							\
819 })
820 
821 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
822 				 KVM_GUESTDBG_USE_SW_BP | \
823 				 KVM_GUESTDBG_USE_HW | \
824 				 KVM_GUESTDBG_SINGLESTEP)
825 
826 #define vcpu_has_sve(vcpu) (system_supports_sve() &&			\
827 			    vcpu_get_flag(vcpu, GUEST_HAS_SVE))
828 
829 #ifdef CONFIG_ARM64_PTR_AUTH
830 #define vcpu_has_ptrauth(vcpu)						\
831 	((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||		\
832 	  cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&		\
833 	  vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
834 #else
835 #define vcpu_has_ptrauth(vcpu)		false
836 #endif
837 
838 #define vcpu_on_unsupported_cpu(vcpu)					\
839 	vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
840 
841 #define vcpu_set_on_unsupported_cpu(vcpu)				\
842 	vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
843 
844 #define vcpu_clear_on_unsupported_cpu(vcpu)				\
845 	vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
846 
847 #define vcpu_gp_regs(v)		(&(v)->arch.ctxt.regs)
848 
849 /*
850  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
851  * memory backed version of a register, and not the one most recently
852  * accessed by a running VCPU.  For example, for userspace access or
853  * for system registers that are never context switched, but only
854  * emulated.
855  *
856  * Don't bother with VNCR-based accesses in the nVHE code, it has no
857  * business dealing with NV.
858  */
859 static inline u64 *__ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
860 {
861 #if !defined (__KVM_NVHE_HYPERVISOR__)
862 	if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
863 		     r >= __VNCR_START__ && ctxt->vncr_array))
864 		return &ctxt->vncr_array[r - __VNCR_START__];
865 #endif
866 	return (u64 *)&ctxt->sys_regs[r];
867 }
868 
869 #define ctxt_sys_reg(c,r)	(*__ctxt_sys_reg(c,r))
870 
871 #define __vcpu_sys_reg(v,r)	(ctxt_sys_reg(&(v)->arch.ctxt, (r)))
872 
873 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
874 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
875 
876 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
877 {
878 	/*
879 	 * *** VHE ONLY ***
880 	 *
881 	 * System registers listed in the switch are not saved on every
882 	 * exit from the guest but are only saved on vcpu_put.
883 	 *
884 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
885 	 * should never be listed below, because the guest cannot modify its
886 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
887 	 * thread when emulating cross-VCPU communication.
888 	 */
889 	if (!has_vhe())
890 		return false;
891 
892 	switch (reg) {
893 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
894 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
895 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
896 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
897 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
898 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
899 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
900 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
901 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
902 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
903 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
904 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
905 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
906 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
907 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
908 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
909 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
910 	case ELR_EL1:		*val = read_sysreg_s(SYS_ELR_EL12);	break;
911 	case SPSR_EL1:		*val = read_sysreg_s(SYS_SPSR_EL12);	break;
912 	case PAR_EL1:		*val = read_sysreg_par();		break;
913 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
914 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
915 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
916 	default:		return false;
917 	}
918 
919 	return true;
920 }
921 
922 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
923 {
924 	/*
925 	 * *** VHE ONLY ***
926 	 *
927 	 * System registers listed in the switch are not restored on every
928 	 * entry to the guest but are only restored on vcpu_load.
929 	 *
930 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
931 	 * should never be listed below, because the MPIDR should only be set
932 	 * once, before running the VCPU, and never changed later.
933 	 */
934 	if (!has_vhe())
935 		return false;
936 
937 	switch (reg) {
938 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
939 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
940 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
941 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
942 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
943 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
944 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
945 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
946 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
947 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
948 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
949 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
950 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
951 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
952 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
953 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
954 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
955 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
956 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
957 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
958 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
959 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
960 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
961 	default:		return false;
962 	}
963 
964 	return true;
965 }
966 
967 struct kvm_vm_stat {
968 	struct kvm_vm_stat_generic generic;
969 };
970 
971 struct kvm_vcpu_stat {
972 	struct kvm_vcpu_stat_generic generic;
973 	u64 hvc_exit_stat;
974 	u64 wfe_exit_stat;
975 	u64 wfi_exit_stat;
976 	u64 mmio_exit_user;
977 	u64 mmio_exit_kernel;
978 	u64 signal_exits;
979 	u64 exits;
980 };
981 
982 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
983 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
984 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
985 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
986 
987 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
988 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
989 
990 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
991 			      struct kvm_vcpu_events *events);
992 
993 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
994 			      struct kvm_vcpu_events *events);
995 
996 void kvm_arm_halt_guest(struct kvm *kvm);
997 void kvm_arm_resume_guest(struct kvm *kvm);
998 
999 #define vcpu_has_run_once(vcpu)	!!rcu_access_pointer((vcpu)->pid)
1000 
1001 #ifndef __KVM_NVHE_HYPERVISOR__
1002 #define kvm_call_hyp_nvhe(f, ...)						\
1003 	({								\
1004 		struct arm_smccc_res res;				\
1005 									\
1006 		arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),		\
1007 				  ##__VA_ARGS__, &res);			\
1008 		WARN_ON(res.a0 != SMCCC_RET_SUCCESS);			\
1009 									\
1010 		res.a1;							\
1011 	})
1012 
1013 /*
1014  * The couple of isb() below are there to guarantee the same behaviour
1015  * on VHE as on !VHE, where the eret to EL1 acts as a context
1016  * synchronization event.
1017  */
1018 #define kvm_call_hyp(f, ...)						\
1019 	do {								\
1020 		if (has_vhe()) {					\
1021 			f(__VA_ARGS__);					\
1022 			isb();						\
1023 		} else {						\
1024 			kvm_call_hyp_nvhe(f, ##__VA_ARGS__);		\
1025 		}							\
1026 	} while(0)
1027 
1028 #define kvm_call_hyp_ret(f, ...)					\
1029 	({								\
1030 		typeof(f(__VA_ARGS__)) ret;				\
1031 									\
1032 		if (has_vhe()) {					\
1033 			ret = f(__VA_ARGS__);				\
1034 			isb();						\
1035 		} else {						\
1036 			ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);	\
1037 		}							\
1038 									\
1039 		ret;							\
1040 	})
1041 #else /* __KVM_NVHE_HYPERVISOR__ */
1042 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1043 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1044 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1045 #endif /* __KVM_NVHE_HYPERVISOR__ */
1046 
1047 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1048 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1049 
1050 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1051 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1052 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1053 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1054 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1055 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1056 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1057 
1058 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1059 
1060 int __init kvm_sys_reg_table_init(void);
1061 int __init populate_nv_trap_config(void);
1062 
1063 bool lock_all_vcpus(struct kvm *kvm);
1064 void unlock_all_vcpus(struct kvm *kvm);
1065 
1066 /* MMIO helpers */
1067 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1068 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1069 
1070 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1071 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1072 
1073 /*
1074  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1075  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1076  * loaded is considered to be "in guest".
1077  */
1078 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1079 {
1080 	return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1081 }
1082 
1083 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1084 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1085 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1086 
1087 bool kvm_arm_pvtime_supported(void);
1088 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1089 			    struct kvm_device_attr *attr);
1090 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1091 			    struct kvm_device_attr *attr);
1092 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1093 			    struct kvm_device_attr *attr);
1094 
1095 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1096 int __init kvm_arm_vmid_alloc_init(void);
1097 void __init kvm_arm_vmid_alloc_free(void);
1098 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1099 void kvm_arm_vmid_clear_active(void);
1100 
1101 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1102 {
1103 	vcpu_arch->steal.base = INVALID_GPA;
1104 }
1105 
1106 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1107 {
1108 	return (vcpu_arch->steal.base != INVALID_GPA);
1109 }
1110 
1111 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1112 
1113 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1114 
1115 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1116 
1117 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1118 {
1119 	/* The host's MPIDR is immutable, so let's set it up at boot time */
1120 	ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1121 }
1122 
1123 static inline bool kvm_system_needs_idmapped_vectors(void)
1124 {
1125 	return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1126 }
1127 
1128 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1129 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1130 
1131 void kvm_arm_init_debug(void);
1132 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1133 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1134 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1135 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1136 
1137 #define kvm_vcpu_os_lock_enabled(vcpu)		\
1138 	(!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1139 
1140 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1141 			       struct kvm_device_attr *attr);
1142 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1143 			       struct kvm_device_attr *attr);
1144 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1145 			       struct kvm_device_attr *attr);
1146 
1147 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1148 			       struct kvm_arm_copy_mte_tags *copy_tags);
1149 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1150 				    struct kvm_arm_counter_offset *offset);
1151 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1152 					struct reg_mask_range *range);
1153 
1154 /* Guest/host FPSIMD coordination helpers */
1155 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1156 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1157 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1158 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1159 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1160 void kvm_vcpu_unshare_task_fp(struct kvm_vcpu *vcpu);
1161 
1162 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1163 {
1164 	return (!has_vhe() && attr->exclude_host);
1165 }
1166 
1167 /* Flags for host debug state */
1168 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1169 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1170 
1171 #ifdef CONFIG_KVM
1172 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1173 void kvm_clr_pmu_events(u32 clr);
1174 bool kvm_set_pmuserenr(u64 val);
1175 #else
1176 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1177 static inline void kvm_clr_pmu_events(u32 clr) {}
1178 static inline bool kvm_set_pmuserenr(u64 val)
1179 {
1180 	return false;
1181 }
1182 #endif
1183 
1184 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1185 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1186 
1187 int __init kvm_set_ipa_limit(void);
1188 
1189 #define __KVM_HAVE_ARCH_VM_ALLOC
1190 struct kvm *kvm_arch_alloc_vm(void);
1191 
1192 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1193 
1194 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1195 
1196 static inline bool kvm_vm_is_protected(struct kvm *kvm)
1197 {
1198 	return false;
1199 }
1200 
1201 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1202 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1203 
1204 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1205 
1206 #define kvm_has_mte(kvm)					\
1207 	(system_supports_mte() &&				\
1208 	 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1209 
1210 #define kvm_supports_32bit_el0()				\
1211 	(system_supports_32bit_el0() &&				\
1212 	 !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1213 
1214 #define kvm_vm_has_ran_once(kvm)					\
1215 	(test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1216 
1217 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1218 {
1219 	return test_bit(feature, ka->vcpu_features);
1220 }
1221 
1222 #define vcpu_has_feature(v, f)	__vcpu_has_feature(&(v)->kvm->arch, (f))
1223 
1224 int kvm_trng_call(struct kvm_vcpu *vcpu);
1225 #ifdef CONFIG_KVM
1226 extern phys_addr_t hyp_mem_base;
1227 extern phys_addr_t hyp_mem_size;
1228 void __init kvm_hyp_reserve(void);
1229 #else
1230 static inline void kvm_hyp_reserve(void) { }
1231 #endif
1232 
1233 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1234 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1235 
1236 #endif /* __ARM64_KVM_HOST_H__ */
1237