xref: /linux/arch/arm64/include/asm/kvm_host.h (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 index = 0, mask = data->mpidr_mask;
225 	unsigned long aff = mpidr & MPIDR_HWID_BITMASK;
226 
227 	bitmap_gather(&index, &aff, &mask, fls(mask));
228 
229 	return index;
230 }
231 
232 struct kvm_sysreg_masks;
233 
234 enum fgt_group_id {
235 	__NO_FGT_GROUP__,
236 	HFGxTR_GROUP,
237 	HDFGRTR_GROUP,
238 	HDFGWTR_GROUP = HDFGRTR_GROUP,
239 	HFGITR_GROUP,
240 	HAFGRTR_GROUP,
241 
242 	/* Must be last */
243 	__NR_FGT_GROUP_IDS__
244 };
245 
246 struct kvm_arch {
247 	struct kvm_s2_mmu mmu;
248 
249 	/*
250 	 * Fine-Grained UNDEF, mimicking the FGT layout defined by the
251 	 * architecture. We track them globally, as we present the
252 	 * same feature-set to all vcpus.
253 	 *
254 	 * Index 0 is currently spare.
255 	 */
256 	u64 fgu[__NR_FGT_GROUP_IDS__];
257 
258 	/* Interrupt controller */
259 	struct vgic_dist	vgic;
260 
261 	/* Timers */
262 	struct arch_timer_vm_data timer_data;
263 
264 	/* Mandated version of PSCI */
265 	u32 psci_version;
266 
267 	/* Protects VM-scoped configuration data */
268 	struct mutex config_lock;
269 
270 	/*
271 	 * If we encounter a data abort without valid instruction syndrome
272 	 * information, report this to user space.  User space can (and
273 	 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
274 	 * supported.
275 	 */
276 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER	0
277 	/* Memory Tagging Extension enabled for the guest */
278 #define KVM_ARCH_FLAG_MTE_ENABLED			1
279 	/* At least one vCPU has ran in the VM */
280 #define KVM_ARCH_FLAG_HAS_RAN_ONCE			2
281 	/* The vCPU feature set for the VM is configured */
282 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED		3
283 	/* PSCI SYSTEM_SUSPEND enabled for the guest */
284 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED		4
285 	/* VM counter offset */
286 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET			5
287 	/* Timer PPIs made immutable */
288 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE		6
289 	/* Initial ID reg values loaded */
290 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED		7
291 	/* Fine-Grained UNDEF initialised */
292 #define KVM_ARCH_FLAG_FGU_INITIALIZED			8
293 	unsigned long flags;
294 
295 	/* VM-wide vCPU feature set */
296 	DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
297 
298 	/* MPIDR to vcpu index mapping, optional */
299 	struct kvm_mpidr_data *mpidr_data;
300 
301 	/*
302 	 * VM-wide PMU filter, implemented as a bitmap and big enough for
303 	 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
304 	 */
305 	unsigned long *pmu_filter;
306 	struct arm_pmu *arm_pmu;
307 
308 	cpumask_var_t supported_cpus;
309 
310 	/* PMCR_EL0.N value for the guest */
311 	u8 pmcr_n;
312 
313 	/* Iterator for idreg debugfs */
314 	u8	idreg_debugfs_iter;
315 
316 	/* Hypercall features firmware registers' descriptor */
317 	struct kvm_smccc_features smccc_feat;
318 	struct maple_tree smccc_filter;
319 
320 	/*
321 	 * Emulated CPU ID registers per VM
322 	 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
323 	 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
324 	 *
325 	 * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
326 	 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
327 	 */
328 #define IDREG_IDX(id)		(((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
329 #define IDX_IDREG(idx)		sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask)
330 #define IDREG(kvm, id)		((kvm)->arch.id_regs[IDREG_IDX(id)])
331 #define KVM_ARM_ID_REG_NUM	(IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
332 	u64 id_regs[KVM_ARM_ID_REG_NUM];
333 
334 	/* Masks for VNCR-baked sysregs */
335 	struct kvm_sysreg_masks	*sysreg_masks;
336 
337 	/*
338 	 * For an untrusted host VM, 'pkvm.handle' is used to lookup
339 	 * the associated pKVM instance in the hypervisor.
340 	 */
341 	struct kvm_protected_vm pkvm;
342 };
343 
344 struct kvm_vcpu_fault_info {
345 	u64 esr_el2;		/* Hyp Syndrom Register */
346 	u64 far_el2;		/* Hyp Fault Address Register */
347 	u64 hpfar_el2;		/* Hyp IPA Fault Address Register */
348 	u64 disr_el1;		/* Deferred [SError] Status Register */
349 };
350 
351 /*
352  * VNCR() just places the VNCR_capable registers in the enum after
353  * __VNCR_START__, and the value (after correction) to be an 8-byte offset
354  * from the VNCR base. As we don't require the enum to be otherwise ordered,
355  * we need the terrible hack below to ensure that we correctly size the
356  * sys_regs array, no matter what.
357  *
358  * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful
359  * treasure trove of bit hacks:
360  * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
361  */
362 #define __MAX__(x,y)	((x) ^ (((x) ^ (y)) & -((x) < (y))))
363 #define VNCR(r)						\
364 	__before_##r,					\
365 	r = __VNCR_START__ + ((VNCR_ ## r) / 8),	\
366 	__after_##r = __MAX__(__before_##r - 1, r)
367 
368 enum vcpu_sysreg {
369 	__INVALID_SYSREG__,   /* 0 is reserved as an invalid value */
370 	MPIDR_EL1,	/* MultiProcessor Affinity Register */
371 	CLIDR_EL1,	/* Cache Level ID Register */
372 	CSSELR_EL1,	/* Cache Size Selection Register */
373 	TPIDR_EL0,	/* Thread ID, User R/W */
374 	TPIDRRO_EL0,	/* Thread ID, User R/O */
375 	TPIDR_EL1,	/* Thread ID, Privileged */
376 	CNTKCTL_EL1,	/* Timer Control Register (EL1) */
377 	PAR_EL1,	/* Physical Address Register */
378 	MDCCINT_EL1,	/* Monitor Debug Comms Channel Interrupt Enable Reg */
379 	OSLSR_EL1,	/* OS Lock Status Register */
380 	DISR_EL1,	/* Deferred Interrupt Status Register */
381 
382 	/* Performance Monitors Registers */
383 	PMCR_EL0,	/* Control Register */
384 	PMSELR_EL0,	/* Event Counter Selection Register */
385 	PMEVCNTR0_EL0,	/* Event Counter Register (0-30) */
386 	PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
387 	PMCCNTR_EL0,	/* Cycle Counter Register */
388 	PMEVTYPER0_EL0,	/* Event Type Register (0-30) */
389 	PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
390 	PMCCFILTR_EL0,	/* Cycle Count Filter Register */
391 	PMCNTENSET_EL0,	/* Count Enable Set Register */
392 	PMINTENSET_EL1,	/* Interrupt Enable Set Register */
393 	PMOVSSET_EL0,	/* Overflow Flag Status Set Register */
394 	PMUSERENR_EL0,	/* User Enable Register */
395 
396 	/* Pointer Authentication Registers in a strict increasing order. */
397 	APIAKEYLO_EL1,
398 	APIAKEYHI_EL1,
399 	APIBKEYLO_EL1,
400 	APIBKEYHI_EL1,
401 	APDAKEYLO_EL1,
402 	APDAKEYHI_EL1,
403 	APDBKEYLO_EL1,
404 	APDBKEYHI_EL1,
405 	APGAKEYLO_EL1,
406 	APGAKEYHI_EL1,
407 
408 	/* Memory Tagging Extension registers */
409 	RGSR_EL1,	/* Random Allocation Tag Seed Register */
410 	GCR_EL1,	/* Tag Control Register */
411 	TFSRE0_EL1,	/* Tag Fault Status Register (EL0) */
412 
413 	/* 32bit specific registers. */
414 	DACR32_EL2,	/* Domain Access Control Register */
415 	IFSR32_EL2,	/* Instruction Fault Status Register */
416 	FPEXC32_EL2,	/* Floating-Point Exception Control Register */
417 	DBGVCR32_EL2,	/* Debug Vector Catch Register */
418 
419 	/* EL2 registers */
420 	SCTLR_EL2,	/* System Control Register (EL2) */
421 	ACTLR_EL2,	/* Auxiliary Control Register (EL2) */
422 	MDCR_EL2,	/* Monitor Debug Configuration Register (EL2) */
423 	CPTR_EL2,	/* Architectural Feature Trap Register (EL2) */
424 	HACR_EL2,	/* Hypervisor Auxiliary Control Register */
425 	TTBR0_EL2,	/* Translation Table Base Register 0 (EL2) */
426 	TTBR1_EL2,	/* Translation Table Base Register 1 (EL2) */
427 	TCR_EL2,	/* Translation Control Register (EL2) */
428 	SPSR_EL2,	/* EL2 saved program status register */
429 	ELR_EL2,	/* EL2 exception link register */
430 	AFSR0_EL2,	/* Auxiliary Fault Status Register 0 (EL2) */
431 	AFSR1_EL2,	/* Auxiliary Fault Status Register 1 (EL2) */
432 	ESR_EL2,	/* Exception Syndrome Register (EL2) */
433 	FAR_EL2,	/* Fault Address Register (EL2) */
434 	HPFAR_EL2,	/* Hypervisor IPA Fault Address Register */
435 	MAIR_EL2,	/* Memory Attribute Indirection Register (EL2) */
436 	AMAIR_EL2,	/* Auxiliary Memory Attribute Indirection Register (EL2) */
437 	VBAR_EL2,	/* Vector Base Address Register (EL2) */
438 	RVBAR_EL2,	/* Reset Vector Base Address Register */
439 	CONTEXTIDR_EL2,	/* Context ID Register (EL2) */
440 	CNTHCTL_EL2,	/* Counter-timer Hypervisor Control register */
441 	SP_EL2,		/* EL2 Stack Pointer */
442 	CNTHP_CTL_EL2,
443 	CNTHP_CVAL_EL2,
444 	CNTHV_CTL_EL2,
445 	CNTHV_CVAL_EL2,
446 
447 	__VNCR_START__,	/* Any VNCR-capable reg goes after this point */
448 
449 	VNCR(SCTLR_EL1),/* System Control Register */
450 	VNCR(ACTLR_EL1),/* Auxiliary Control Register */
451 	VNCR(CPACR_EL1),/* Coprocessor Access Control */
452 	VNCR(ZCR_EL1),	/* SVE Control */
453 	VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */
454 	VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */
455 	VNCR(TCR_EL1),	/* Translation Control Register */
456 	VNCR(TCR2_EL1),	/* Extended Translation Control Register */
457 	VNCR(ESR_EL1),	/* Exception Syndrome Register */
458 	VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */
459 	VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */
460 	VNCR(FAR_EL1),	/* Fault Address Register */
461 	VNCR(MAIR_EL1),	/* Memory Attribute Indirection Register */
462 	VNCR(VBAR_EL1),	/* Vector Base Address Register */
463 	VNCR(CONTEXTIDR_EL1),	/* Context ID Register */
464 	VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */
465 	VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */
466 	VNCR(ELR_EL1),
467 	VNCR(SP_EL1),
468 	VNCR(SPSR_EL1),
469 	VNCR(TFSR_EL1),	/* Tag Fault Status Register (EL1) */
470 	VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */
471 	VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */
472 	VNCR(HCR_EL2),	/* Hypervisor Configuration Register */
473 	VNCR(HSTR_EL2),	/* Hypervisor System Trap Register */
474 	VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */
475 	VNCR(VTCR_EL2),	/* Virtualization Translation Control Register */
476 	VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */
477 	VNCR(HCRX_EL2),	/* Extended Hypervisor Configuration Register */
478 
479 	/* Permission Indirection Extension registers */
480 	VNCR(PIR_EL1),	 /* Permission Indirection Register 1 (EL1) */
481 	VNCR(PIRE0_EL1), /*  Permission Indirection Register 0 (EL1) */
482 
483 	VNCR(HFGRTR_EL2),
484 	VNCR(HFGWTR_EL2),
485 	VNCR(HFGITR_EL2),
486 	VNCR(HDFGRTR_EL2),
487 	VNCR(HDFGWTR_EL2),
488 	VNCR(HAFGRTR_EL2),
489 
490 	VNCR(CNTVOFF_EL2),
491 	VNCR(CNTV_CVAL_EL0),
492 	VNCR(CNTV_CTL_EL0),
493 	VNCR(CNTP_CVAL_EL0),
494 	VNCR(CNTP_CTL_EL0),
495 
496 	NR_SYS_REGS	/* Nothing after this line! */
497 };
498 
499 struct kvm_sysreg_masks {
500 	struct {
501 		u64	res0;
502 		u64	res1;
503 	} mask[NR_SYS_REGS - __VNCR_START__];
504 };
505 
506 struct kvm_cpu_context {
507 	struct user_pt_regs regs;	/* sp = sp_el0 */
508 
509 	u64	spsr_abt;
510 	u64	spsr_und;
511 	u64	spsr_irq;
512 	u64	spsr_fiq;
513 
514 	struct user_fpsimd_state fp_regs;
515 
516 	u64 sys_regs[NR_SYS_REGS];
517 
518 	struct kvm_vcpu *__hyp_running_vcpu;
519 
520 	/* This pointer has to be 4kB aligned. */
521 	u64 *vncr_array;
522 };
523 
524 /*
525  * This structure is instantiated on a per-CPU basis, and contains
526  * data that is:
527  *
528  * - tied to a single physical CPU, and
529  * - either have a lifetime that does not extend past vcpu_put()
530  * - or is an invariant for the lifetime of the system
531  *
532  * Use host_data_ptr(field) as a way to access a pointer to such a
533  * field.
534  */
535 struct kvm_host_data {
536 	struct kvm_cpu_context host_ctxt;
537 	struct user_fpsimd_state *fpsimd_state;	/* hyp VA */
538 
539 	/* Ownership of the FP regs */
540 	enum {
541 		FP_STATE_FREE,
542 		FP_STATE_HOST_OWNED,
543 		FP_STATE_GUEST_OWNED,
544 	} fp_owner;
545 
546 	/*
547 	 * host_debug_state contains the host registers which are
548 	 * saved and restored during world switches.
549 	 */
550 	 struct {
551 		/* {Break,watch}point registers */
552 		struct kvm_guest_debug_arch regs;
553 		/* Statistical profiling extension */
554 		u64 pmscr_el1;
555 		/* Self-hosted trace */
556 		u64 trfcr_el1;
557 		/* Values of trap registers for the host before guest entry. */
558 		u64 mdcr_el2;
559 	} host_debug_state;
560 };
561 
562 struct kvm_host_psci_config {
563 	/* PSCI version used by host. */
564 	u32 version;
565 	u32 smccc_version;
566 
567 	/* Function IDs used by host if version is v0.1. */
568 	struct psci_0_1_function_ids function_ids_0_1;
569 
570 	bool psci_0_1_cpu_suspend_implemented;
571 	bool psci_0_1_cpu_on_implemented;
572 	bool psci_0_1_cpu_off_implemented;
573 	bool psci_0_1_migrate_implemented;
574 };
575 
576 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
577 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
578 
579 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
580 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
581 
582 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
583 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
584 
585 struct vcpu_reset_state {
586 	unsigned long	pc;
587 	unsigned long	r0;
588 	bool		be;
589 	bool		reset;
590 };
591 
592 struct kvm_vcpu_arch {
593 	struct kvm_cpu_context ctxt;
594 
595 	/*
596 	 * Guest floating point state
597 	 *
598 	 * The architecture has two main floating point extensions,
599 	 * the original FPSIMD and SVE.  These have overlapping
600 	 * register views, with the FPSIMD V registers occupying the
601 	 * low 128 bits of the SVE Z registers.  When the core
602 	 * floating point code saves the register state of a task it
603 	 * records which view it saved in fp_type.
604 	 */
605 	void *sve_state;
606 	enum fp_type fp_type;
607 	unsigned int sve_max_vl;
608 	u64 svcr;
609 	u64 fpmr;
610 
611 	/* Stage 2 paging state used by the hardware on next switch */
612 	struct kvm_s2_mmu *hw_mmu;
613 
614 	/* Values of trap registers for the guest. */
615 	u64 hcr_el2;
616 	u64 hcrx_el2;
617 	u64 mdcr_el2;
618 	u64 cptr_el2;
619 
620 	/* Exception Information */
621 	struct kvm_vcpu_fault_info fault;
622 
623 	/* Configuration flags, set once and for all before the vcpu can run */
624 	u8 cflags;
625 
626 	/* Input flags to the hypervisor code, potentially cleared after use */
627 	u8 iflags;
628 
629 	/* State flags for kernel bookkeeping, unused by the hypervisor code */
630 	u8 sflags;
631 
632 	/*
633 	 * Don't run the guest (internal implementation need).
634 	 *
635 	 * Contrary to the flags above, this is set/cleared outside of
636 	 * a vcpu context, and thus cannot be mixed with the flags
637 	 * themselves (or the flag accesses need to be made atomic).
638 	 */
639 	bool pause;
640 
641 	/*
642 	 * We maintain more than a single set of debug registers to support
643 	 * debugging the guest from the host and to maintain separate host and
644 	 * guest state during world switches. vcpu_debug_state are the debug
645 	 * registers of the vcpu as the guest sees them.
646 	 *
647 	 * external_debug_state contains the debug values we want to debug the
648 	 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl.
649 	 *
650 	 * debug_ptr points to the set of debug registers that should be loaded
651 	 * onto the hardware when running the guest.
652 	 */
653 	struct kvm_guest_debug_arch *debug_ptr;
654 	struct kvm_guest_debug_arch vcpu_debug_state;
655 	struct kvm_guest_debug_arch external_debug_state;
656 
657 	/* VGIC state */
658 	struct vgic_cpu vgic_cpu;
659 	struct arch_timer_cpu timer_cpu;
660 	struct kvm_pmu pmu;
661 
662 	/*
663 	 * Guest registers we preserve during guest debugging.
664 	 *
665 	 * These shadow registers are updated by the kvm_handle_sys_reg
666 	 * trap handler if the guest accesses or updates them while we
667 	 * are using guest debug.
668 	 */
669 	struct {
670 		u32	mdscr_el1;
671 		bool	pstate_ss;
672 	} guest_debug_preserved;
673 
674 	/* vcpu power state */
675 	struct kvm_mp_state mp_state;
676 	spinlock_t mp_state_lock;
677 
678 	/* Cache some mmu pages needed inside spinlock regions */
679 	struct kvm_mmu_memory_cache mmu_page_cache;
680 
681 	/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
682 	u64 vsesr_el2;
683 
684 	/* Additional reset state */
685 	struct vcpu_reset_state	reset_state;
686 
687 	/* Guest PV state */
688 	struct {
689 		u64 last_steal;
690 		gpa_t base;
691 	} steal;
692 
693 	/* Per-vcpu CCSIDR override or NULL */
694 	u32 *ccsidr;
695 };
696 
697 /*
698  * Each 'flag' is composed of a comma-separated triplet:
699  *
700  * - the flag-set it belongs to in the vcpu->arch structure
701  * - the value for that flag
702  * - the mask for that flag
703  *
704  *  __vcpu_single_flag() builds such a triplet for a single-bit flag.
705  * unpack_vcpu_flag() extract the flag value from the triplet for
706  * direct use outside of the flag accessors.
707  */
708 #define __vcpu_single_flag(_set, _f)	_set, (_f), (_f)
709 
710 #define __unpack_flag(_set, _f, _m)	_f
711 #define unpack_vcpu_flag(...)		__unpack_flag(__VA_ARGS__)
712 
713 #define __build_check_flag(v, flagset, f, m)			\
714 	do {							\
715 		typeof(v->arch.flagset) *_fset;			\
716 								\
717 		/* Check that the flags fit in the mask */	\
718 		BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m)));	\
719 		/* Check that the flags fit in the type */	\
720 		BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m));	\
721 	} while (0)
722 
723 #define __vcpu_get_flag(v, flagset, f, m)			\
724 	({							\
725 		__build_check_flag(v, flagset, f, m);		\
726 								\
727 		READ_ONCE(v->arch.flagset) & (m);		\
728 	})
729 
730 /*
731  * Note that the set/clear accessors must be preempt-safe in order to
732  * avoid nesting them with load/put which also manipulate flags...
733  */
734 #ifdef __KVM_NVHE_HYPERVISOR__
735 /* the nVHE hypervisor is always non-preemptible */
736 #define __vcpu_flags_preempt_disable()
737 #define __vcpu_flags_preempt_enable()
738 #else
739 #define __vcpu_flags_preempt_disable()	preempt_disable()
740 #define __vcpu_flags_preempt_enable()	preempt_enable()
741 #endif
742 
743 #define __vcpu_set_flag(v, flagset, f, m)			\
744 	do {							\
745 		typeof(v->arch.flagset) *fset;			\
746 								\
747 		__build_check_flag(v, flagset, f, m);		\
748 								\
749 		fset = &v->arch.flagset;			\
750 		__vcpu_flags_preempt_disable();			\
751 		if (HWEIGHT(m) > 1)				\
752 			*fset &= ~(m);				\
753 		*fset |= (f);					\
754 		__vcpu_flags_preempt_enable();			\
755 	} while (0)
756 
757 #define __vcpu_clear_flag(v, flagset, f, m)			\
758 	do {							\
759 		typeof(v->arch.flagset) *fset;			\
760 								\
761 		__build_check_flag(v, flagset, f, m);		\
762 								\
763 		fset = &v->arch.flagset;			\
764 		__vcpu_flags_preempt_disable();			\
765 		*fset &= ~(m);					\
766 		__vcpu_flags_preempt_enable();			\
767 	} while (0)
768 
769 #define vcpu_get_flag(v, ...)	__vcpu_get_flag((v), __VA_ARGS__)
770 #define vcpu_set_flag(v, ...)	__vcpu_set_flag((v), __VA_ARGS__)
771 #define vcpu_clear_flag(v, ...)	__vcpu_clear_flag((v), __VA_ARGS__)
772 
773 /* SVE exposed to guest */
774 #define GUEST_HAS_SVE		__vcpu_single_flag(cflags, BIT(0))
775 /* SVE config completed */
776 #define VCPU_SVE_FINALIZED	__vcpu_single_flag(cflags, BIT(1))
777 /* PTRAUTH exposed to guest */
778 #define GUEST_HAS_PTRAUTH	__vcpu_single_flag(cflags, BIT(2))
779 /* KVM_ARM_VCPU_INIT completed */
780 #define VCPU_INITIALIZED	__vcpu_single_flag(cflags, BIT(3))
781 
782 /* Exception pending */
783 #define PENDING_EXCEPTION	__vcpu_single_flag(iflags, BIT(0))
784 /*
785  * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
786  * be set together with an exception...
787  */
788 #define INCREMENT_PC		__vcpu_single_flag(iflags, BIT(1))
789 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
790 #define EXCEPT_MASK		__vcpu_single_flag(iflags, GENMASK(3, 1))
791 
792 /* Helpers to encode exceptions with minimum fuss */
793 #define __EXCEPT_MASK_VAL	unpack_vcpu_flag(EXCEPT_MASK)
794 #define __EXCEPT_SHIFT		__builtin_ctzl(__EXCEPT_MASK_VAL)
795 #define __vcpu_except_flags(_f)	iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
796 
797 /*
798  * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
799  * values:
800  *
801  * For AArch32 EL1:
802  */
803 #define EXCEPT_AA32_UND		__vcpu_except_flags(0)
804 #define EXCEPT_AA32_IABT	__vcpu_except_flags(1)
805 #define EXCEPT_AA32_DABT	__vcpu_except_flags(2)
806 /* For AArch64: */
807 #define EXCEPT_AA64_EL1_SYNC	__vcpu_except_flags(0)
808 #define EXCEPT_AA64_EL1_IRQ	__vcpu_except_flags(1)
809 #define EXCEPT_AA64_EL1_FIQ	__vcpu_except_flags(2)
810 #define EXCEPT_AA64_EL1_SERR	__vcpu_except_flags(3)
811 /* For AArch64 with NV: */
812 #define EXCEPT_AA64_EL2_SYNC	__vcpu_except_flags(4)
813 #define EXCEPT_AA64_EL2_IRQ	__vcpu_except_flags(5)
814 #define EXCEPT_AA64_EL2_FIQ	__vcpu_except_flags(6)
815 #define EXCEPT_AA64_EL2_SERR	__vcpu_except_flags(7)
816 /* Guest debug is live */
817 #define DEBUG_DIRTY		__vcpu_single_flag(iflags, BIT(4))
818 /* Save SPE context if active  */
819 #define DEBUG_STATE_SAVE_SPE	__vcpu_single_flag(iflags, BIT(5))
820 /* Save TRBE context if active  */
821 #define DEBUG_STATE_SAVE_TRBE	__vcpu_single_flag(iflags, BIT(6))
822 
823 /* SVE enabled for host EL0 */
824 #define HOST_SVE_ENABLED	__vcpu_single_flag(sflags, BIT(0))
825 /* SME enabled for EL0 */
826 #define HOST_SME_ENABLED	__vcpu_single_flag(sflags, BIT(1))
827 /* Physical CPU not in supported_cpus */
828 #define ON_UNSUPPORTED_CPU	__vcpu_single_flag(sflags, BIT(2))
829 /* WFIT instruction trapped */
830 #define IN_WFIT			__vcpu_single_flag(sflags, BIT(3))
831 /* vcpu system registers loaded on physical CPU */
832 #define SYSREGS_ON_CPU		__vcpu_single_flag(sflags, BIT(4))
833 /* Software step state is Active-pending */
834 #define DBG_SS_ACTIVE_PENDING	__vcpu_single_flag(sflags, BIT(5))
835 /* PMUSERENR for the guest EL0 is on physical CPU */
836 #define PMUSERENR_ON_CPU	__vcpu_single_flag(sflags, BIT(6))
837 /* WFI instruction trapped */
838 #define IN_WFI			__vcpu_single_flag(sflags, BIT(7))
839 
840 
841 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
842 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +	\
843 			     sve_ffr_offset((vcpu)->arch.sve_max_vl))
844 
845 #define vcpu_sve_max_vq(vcpu)	sve_vq_from_vl((vcpu)->arch.sve_max_vl)
846 
847 #define vcpu_sve_state_size(vcpu) ({					\
848 	size_t __size_ret;						\
849 	unsigned int __vcpu_vq;						\
850 									\
851 	if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {		\
852 		__size_ret = 0;						\
853 	} else {							\
854 		__vcpu_vq = vcpu_sve_max_vq(vcpu);			\
855 		__size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);		\
856 	}								\
857 									\
858 	__size_ret;							\
859 })
860 
861 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
862 				 KVM_GUESTDBG_USE_SW_BP | \
863 				 KVM_GUESTDBG_USE_HW | \
864 				 KVM_GUESTDBG_SINGLESTEP)
865 
866 #define vcpu_has_sve(vcpu) (system_supports_sve() &&			\
867 			    vcpu_get_flag(vcpu, GUEST_HAS_SVE))
868 
869 #ifdef CONFIG_ARM64_PTR_AUTH
870 #define vcpu_has_ptrauth(vcpu)						\
871 	((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) ||		\
872 	  cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) &&		\
873 	  vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
874 #else
875 #define vcpu_has_ptrauth(vcpu)		false
876 #endif
877 
878 #define vcpu_on_unsupported_cpu(vcpu)					\
879 	vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
880 
881 #define vcpu_set_on_unsupported_cpu(vcpu)				\
882 	vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
883 
884 #define vcpu_clear_on_unsupported_cpu(vcpu)				\
885 	vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
886 
887 #define vcpu_gp_regs(v)		(&(v)->arch.ctxt.regs)
888 
889 /*
890  * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
891  * memory backed version of a register, and not the one most recently
892  * accessed by a running VCPU.  For example, for userspace access or
893  * for system registers that are never context switched, but only
894  * emulated.
895  *
896  * Don't bother with VNCR-based accesses in the nVHE code, it has no
897  * business dealing with NV.
898  */
899 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r)
900 {
901 #if !defined (__KVM_NVHE_HYPERVISOR__)
902 	if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) &&
903 		     r >= __VNCR_START__ && ctxt->vncr_array))
904 		return &ctxt->vncr_array[r - __VNCR_START__];
905 #endif
906 	return (u64 *)&ctxt->sys_regs[r];
907 }
908 
909 #define __ctxt_sys_reg(c,r)						\
910 	({								\
911 		BUILD_BUG_ON(__builtin_constant_p(r) &&			\
912 			     (r) >= NR_SYS_REGS);			\
913 		___ctxt_sys_reg(c, r);					\
914 	})
915 
916 #define ctxt_sys_reg(c,r)	(*__ctxt_sys_reg(c,r))
917 
918 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg);
919 #define __vcpu_sys_reg(v,r)						\
920 	(*({								\
921 		const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt;	\
922 		u64 *__r = __ctxt_sys_reg(ctxt, (r));			\
923 		if (vcpu_has_nv((v)) && (r) >= __VNCR_START__)		\
924 			*__r = kvm_vcpu_sanitise_vncr_reg((v), (r));	\
925 		__r;							\
926 	}))
927 
928 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
929 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
930 
931 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
932 {
933 	/*
934 	 * *** VHE ONLY ***
935 	 *
936 	 * System registers listed in the switch are not saved on every
937 	 * exit from the guest but are only saved on vcpu_put.
938 	 *
939 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
940 	 * should never be listed below, because the guest cannot modify its
941 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
942 	 * thread when emulating cross-VCPU communication.
943 	 */
944 	if (!has_vhe())
945 		return false;
946 
947 	switch (reg) {
948 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
949 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
950 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
951 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
952 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
953 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
954 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
955 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
956 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
957 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
958 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
959 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
960 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
961 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
962 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
963 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
964 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
965 	case ELR_EL1:		*val = read_sysreg_s(SYS_ELR_EL12);	break;
966 	case SPSR_EL1:		*val = read_sysreg_s(SYS_SPSR_EL12);	break;
967 	case PAR_EL1:		*val = read_sysreg_par();		break;
968 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
969 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
970 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
971 	default:		return false;
972 	}
973 
974 	return true;
975 }
976 
977 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
978 {
979 	/*
980 	 * *** VHE ONLY ***
981 	 *
982 	 * System registers listed in the switch are not restored on every
983 	 * entry to the guest but are only restored on vcpu_load.
984 	 *
985 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
986 	 * should never be listed below, because the MPIDR should only be set
987 	 * once, before running the VCPU, and never changed later.
988 	 */
989 	if (!has_vhe())
990 		return false;
991 
992 	switch (reg) {
993 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
994 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
995 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
996 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
997 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
998 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
999 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
1000 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
1001 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
1002 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
1003 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
1004 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
1005 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
1006 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
1007 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
1008 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
1009 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
1010 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
1011 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
1012 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
1013 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
1014 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
1015 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
1016 	default:		return false;
1017 	}
1018 
1019 	return true;
1020 }
1021 
1022 struct kvm_vm_stat {
1023 	struct kvm_vm_stat_generic generic;
1024 };
1025 
1026 struct kvm_vcpu_stat {
1027 	struct kvm_vcpu_stat_generic generic;
1028 	u64 hvc_exit_stat;
1029 	u64 wfe_exit_stat;
1030 	u64 wfi_exit_stat;
1031 	u64 mmio_exit_user;
1032 	u64 mmio_exit_kernel;
1033 	u64 signal_exits;
1034 	u64 exits;
1035 };
1036 
1037 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
1038 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
1039 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1040 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
1041 
1042 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
1043 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
1044 
1045 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1046 			      struct kvm_vcpu_events *events);
1047 
1048 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1049 			      struct kvm_vcpu_events *events);
1050 
1051 void kvm_arm_halt_guest(struct kvm *kvm);
1052 void kvm_arm_resume_guest(struct kvm *kvm);
1053 
1054 #define vcpu_has_run_once(vcpu)	!!rcu_access_pointer((vcpu)->pid)
1055 
1056 #ifndef __KVM_NVHE_HYPERVISOR__
1057 #define kvm_call_hyp_nvhe(f, ...)						\
1058 	({								\
1059 		struct arm_smccc_res res;				\
1060 									\
1061 		arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f),		\
1062 				  ##__VA_ARGS__, &res);			\
1063 		WARN_ON(res.a0 != SMCCC_RET_SUCCESS);			\
1064 									\
1065 		res.a1;							\
1066 	})
1067 
1068 /*
1069  * The couple of isb() below are there to guarantee the same behaviour
1070  * on VHE as on !VHE, where the eret to EL1 acts as a context
1071  * synchronization event.
1072  */
1073 #define kvm_call_hyp(f, ...)						\
1074 	do {								\
1075 		if (has_vhe()) {					\
1076 			f(__VA_ARGS__);					\
1077 			isb();						\
1078 		} else {						\
1079 			kvm_call_hyp_nvhe(f, ##__VA_ARGS__);		\
1080 		}							\
1081 	} while(0)
1082 
1083 #define kvm_call_hyp_ret(f, ...)					\
1084 	({								\
1085 		typeof(f(__VA_ARGS__)) ret;				\
1086 									\
1087 		if (has_vhe()) {					\
1088 			ret = f(__VA_ARGS__);				\
1089 			isb();						\
1090 		} else {						\
1091 			ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__);	\
1092 		}							\
1093 									\
1094 		ret;							\
1095 	})
1096 #else /* __KVM_NVHE_HYPERVISOR__ */
1097 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
1098 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
1099 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
1100 #endif /* __KVM_NVHE_HYPERVISOR__ */
1101 
1102 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
1103 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
1104 
1105 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
1106 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
1107 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
1108 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
1109 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
1110 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
1111 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
1112 
1113 void kvm_sys_regs_create_debugfs(struct kvm *kvm);
1114 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
1115 
1116 int __init kvm_sys_reg_table_init(void);
1117 struct sys_reg_desc;
1118 int __init populate_sysreg_config(const struct sys_reg_desc *sr,
1119 				  unsigned int idx);
1120 int __init populate_nv_trap_config(void);
1121 
1122 bool lock_all_vcpus(struct kvm *kvm);
1123 void unlock_all_vcpus(struct kvm *kvm);
1124 
1125 void kvm_init_sysreg(struct kvm_vcpu *);
1126 
1127 /* MMIO helpers */
1128 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
1129 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
1130 
1131 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
1132 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
1133 
1134 /*
1135  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
1136  * arrived in guest context.  For arm64, any event that arrives while a vCPU is
1137  * loaded is considered to be "in guest".
1138  */
1139 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
1140 {
1141 	return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1142 }
1143 
1144 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1145 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1146 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1147 
1148 bool kvm_arm_pvtime_supported(void);
1149 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1150 			    struct kvm_device_attr *attr);
1151 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1152 			    struct kvm_device_attr *attr);
1153 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1154 			    struct kvm_device_attr *attr);
1155 
1156 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1157 int __init kvm_arm_vmid_alloc_init(void);
1158 void __init kvm_arm_vmid_alloc_free(void);
1159 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1160 void kvm_arm_vmid_clear_active(void);
1161 
1162 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1163 {
1164 	vcpu_arch->steal.base = INVALID_GPA;
1165 }
1166 
1167 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1168 {
1169 	return (vcpu_arch->steal.base != INVALID_GPA);
1170 }
1171 
1172 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1173 
1174 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1175 
1176 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1177 
1178 /*
1179  * How we access per-CPU host data depends on the where we access it from,
1180  * and the mode we're in:
1181  *
1182  * - VHE and nVHE hypervisor bits use their locally defined instance
1183  *
1184  * - the rest of the kernel use either the VHE or nVHE one, depending on
1185  *   the mode we're running in.
1186  *
1187  *   Unless we're in protected mode, fully deprivileged, and the nVHE
1188  *   per-CPU stuff is exclusively accessible to the protected EL2 code.
1189  *   In this case, the EL1 code uses the *VHE* data as its private state
1190  *   (which makes sense in a way as there shouldn't be any shared state
1191  *   between the host and the hypervisor).
1192  *
1193  * Yes, this is all totally trivial. Shoot me now.
1194  */
1195 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__)
1196 #define host_data_ptr(f)	(&this_cpu_ptr(&kvm_host_data)->f)
1197 #else
1198 #define host_data_ptr(f)						\
1199 	(static_branch_unlikely(&kvm_protected_mode_initialized) ?	\
1200 	 &this_cpu_ptr(&kvm_host_data)->f :				\
1201 	 &this_cpu_ptr_hyp_sym(kvm_host_data)->f)
1202 #endif
1203 
1204 /* Check whether the FP regs are owned by the guest */
1205 static inline bool guest_owns_fp_regs(void)
1206 {
1207 	return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED;
1208 }
1209 
1210 /* Check whether the FP regs are owned by the host */
1211 static inline bool host_owns_fp_regs(void)
1212 {
1213 	return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED;
1214 }
1215 
1216 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1217 {
1218 	/* The host's MPIDR is immutable, so let's set it up at boot time */
1219 	ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1220 }
1221 
1222 static inline bool kvm_system_needs_idmapped_vectors(void)
1223 {
1224 	return cpus_have_final_cap(ARM64_SPECTRE_V3A);
1225 }
1226 
1227 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1228 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1229 
1230 void kvm_arm_init_debug(void);
1231 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1232 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1233 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1234 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1235 
1236 #define kvm_vcpu_os_lock_enabled(vcpu)		\
1237 	(!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK))
1238 
1239 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1240 			       struct kvm_device_attr *attr);
1241 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1242 			       struct kvm_device_attr *attr);
1243 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1244 			       struct kvm_device_attr *attr);
1245 
1246 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1247 			       struct kvm_arm_copy_mte_tags *copy_tags);
1248 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1249 				    struct kvm_arm_counter_offset *offset);
1250 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm,
1251 					struct reg_mask_range *range);
1252 
1253 /* Guest/host FPSIMD coordination helpers */
1254 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1255 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1256 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1257 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1258 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1259 
1260 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1261 {
1262 	return (!has_vhe() && attr->exclude_host);
1263 }
1264 
1265 /* Flags for host debug state */
1266 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1267 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1268 
1269 #ifdef CONFIG_KVM
1270 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1271 void kvm_clr_pmu_events(u32 clr);
1272 bool kvm_set_pmuserenr(u64 val);
1273 #else
1274 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1275 static inline void kvm_clr_pmu_events(u32 clr) {}
1276 static inline bool kvm_set_pmuserenr(u64 val)
1277 {
1278 	return false;
1279 }
1280 #endif
1281 
1282 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu);
1283 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu);
1284 
1285 int __init kvm_set_ipa_limit(void);
1286 
1287 #define __KVM_HAVE_ARCH_VM_ALLOC
1288 struct kvm *kvm_arch_alloc_vm(void);
1289 
1290 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS
1291 
1292 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE
1293 
1294 #define kvm_vm_is_protected(kvm)	(is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled)
1295 
1296 #define vcpu_is_protected(vcpu)		kvm_vm_is_protected((vcpu)->kvm)
1297 
1298 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1299 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1300 
1301 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1302 
1303 #define kvm_has_mte(kvm)					\
1304 	(system_supports_mte() &&				\
1305 	 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1306 
1307 #define kvm_supports_32bit_el0()				\
1308 	(system_supports_32bit_el0() &&				\
1309 	 !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1310 
1311 #define kvm_vm_has_ran_once(kvm)					\
1312 	(test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1313 
1314 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature)
1315 {
1316 	return test_bit(feature, ka->vcpu_features);
1317 }
1318 
1319 #define vcpu_has_feature(v, f)	__vcpu_has_feature(&(v)->kvm->arch, (f))
1320 
1321 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED)
1322 
1323 int kvm_trng_call(struct kvm_vcpu *vcpu);
1324 #ifdef CONFIG_KVM
1325 extern phys_addr_t hyp_mem_base;
1326 extern phys_addr_t hyp_mem_size;
1327 void __init kvm_hyp_reserve(void);
1328 #else
1329 static inline void kvm_hyp_reserve(void) { }
1330 #endif
1331 
1332 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1333 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1334 
1335 #define __expand_field_sign_unsigned(id, fld, val)			\
1336 	((u64)SYS_FIELD_VALUE(id, fld, val))
1337 
1338 #define __expand_field_sign_signed(id, fld, val)			\
1339 	({								\
1340 		u64 __val = SYS_FIELD_VALUE(id, fld, val);		\
1341 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1342 	})
1343 
1344 #define expand_field_sign(id, fld, val)					\
1345 	(id##_##fld##_SIGNED ?						\
1346 	 __expand_field_sign_signed(id, fld, val) :			\
1347 	 __expand_field_sign_unsigned(id, fld, val))
1348 
1349 #define get_idreg_field_unsigned(kvm, id, fld)				\
1350 	({								\
1351 		u64 __val = IDREG((kvm), SYS_##id);			\
1352 		FIELD_GET(id##_##fld##_MASK, __val);			\
1353 	})
1354 
1355 #define get_idreg_field_signed(kvm, id, fld)				\
1356 	({								\
1357 		u64 __val = get_idreg_field_unsigned(kvm, id, fld);	\
1358 		sign_extend64(__val, id##_##fld##_WIDTH - 1);		\
1359 	})
1360 
1361 #define get_idreg_field_enum(kvm, id, fld)				\
1362 	get_idreg_field_unsigned(kvm, id, fld)
1363 
1364 #define get_idreg_field(kvm, id, fld)					\
1365 	(id##_##fld##_SIGNED ?						\
1366 	 get_idreg_field_signed(kvm, id, fld) :				\
1367 	 get_idreg_field_unsigned(kvm, id, fld))
1368 
1369 #define kvm_has_feat(kvm, id, fld, limit)				\
1370 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit))
1371 
1372 #define kvm_has_feat_enum(kvm, id, fld, val)				\
1373 	(get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val))
1374 
1375 #define kvm_has_feat_range(kvm, id, fld, min, max)			\
1376 	(get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \
1377 	 get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max))
1378 
1379 /* Check for a given level of PAuth support */
1380 #define kvm_has_pauth(k, l)						\
1381 	({								\
1382 		bool pa, pi, pa3;					\
1383 									\
1384 		pa  = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l);	\
1385 		pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP);	\
1386 		pi  = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l);	\
1387 		pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP);	\
1388 		pa3  = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l);	\
1389 		pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP);	\
1390 									\
1391 		(pa + pi + pa3) == 1;					\
1392 	})
1393 
1394 #endif /* __ARM64_KVM_HOST_H__ */
1395