xref: /openbmc/linux/arch/arm64/kvm/guest.c (revision 6fa24b41)
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/kvm/guest.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
9  */
10 
11 #include <linux/bits.h>
12 #include <linux/errno.h>
13 #include <linux/err.h>
14 #include <linux/nospec.h>
15 #include <linux/kvm_host.h>
16 #include <linux/module.h>
17 #include <linux/stddef.h>
18 #include <linux/string.h>
19 #include <linux/vmalloc.h>
20 #include <linux/fs.h>
21 #include <kvm/arm_hypercalls.h>
22 #include <asm/cputype.h>
23 #include <linux/uaccess.h>
24 #include <asm/fpsimd.h>
25 #include <asm/kvm.h>
26 #include <asm/kvm_emulate.h>
27 #include <asm/kvm_nested.h>
28 #include <asm/sigcontext.h>
29 
30 #include "trace.h"
31 
32 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
33 	KVM_GENERIC_VM_STATS()
34 };
35 
36 const struct kvm_stats_header kvm_vm_stats_header = {
37 	.name_size = KVM_STATS_NAME_SIZE,
38 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
39 	.id_offset =  sizeof(struct kvm_stats_header),
40 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
41 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
42 		       sizeof(kvm_vm_stats_desc),
43 };
44 
45 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
46 	KVM_GENERIC_VCPU_STATS(),
47 	STATS_DESC_COUNTER(VCPU, hvc_exit_stat),
48 	STATS_DESC_COUNTER(VCPU, wfe_exit_stat),
49 	STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
50 	STATS_DESC_COUNTER(VCPU, mmio_exit_user),
51 	STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
52 	STATS_DESC_COUNTER(VCPU, signal_exits),
53 	STATS_DESC_COUNTER(VCPU, exits)
54 };
55 
56 const struct kvm_stats_header kvm_vcpu_stats_header = {
57 	.name_size = KVM_STATS_NAME_SIZE,
58 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
59 	.id_offset = sizeof(struct kvm_stats_header),
60 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
61 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
62 		       sizeof(kvm_vcpu_stats_desc),
63 };
64 
65 static bool core_reg_offset_is_vreg(u64 off)
66 {
67 	return off >= KVM_REG_ARM_CORE_REG(fp_regs.vregs) &&
68 		off < KVM_REG_ARM_CORE_REG(fp_regs.fpsr);
69 }
70 
71 static u64 core_reg_offset_from_id(u64 id)
72 {
73 	return id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK | KVM_REG_ARM_CORE);
74 }
75 
76 static int core_reg_size_from_offset(const struct kvm_vcpu *vcpu, u64 off)
77 {
78 	int size;
79 
80 	switch (off) {
81 	case KVM_REG_ARM_CORE_REG(regs.regs[0]) ...
82 	     KVM_REG_ARM_CORE_REG(regs.regs[30]):
83 	case KVM_REG_ARM_CORE_REG(regs.sp):
84 	case KVM_REG_ARM_CORE_REG(regs.pc):
85 	case KVM_REG_ARM_CORE_REG(regs.pstate):
86 	case KVM_REG_ARM_CORE_REG(sp_el1):
87 	case KVM_REG_ARM_CORE_REG(elr_el1):
88 	case KVM_REG_ARM_CORE_REG(spsr[0]) ...
89 	     KVM_REG_ARM_CORE_REG(spsr[KVM_NR_SPSR - 1]):
90 		size = sizeof(__u64);
91 		break;
92 
93 	case KVM_REG_ARM_CORE_REG(fp_regs.vregs[0]) ...
94 	     KVM_REG_ARM_CORE_REG(fp_regs.vregs[31]):
95 		size = sizeof(__uint128_t);
96 		break;
97 
98 	case KVM_REG_ARM_CORE_REG(fp_regs.fpsr):
99 	case KVM_REG_ARM_CORE_REG(fp_regs.fpcr):
100 		size = sizeof(__u32);
101 		break;
102 
103 	default:
104 		return -EINVAL;
105 	}
106 
107 	if (!IS_ALIGNED(off, size / sizeof(__u32)))
108 		return -EINVAL;
109 
110 	/*
111 	 * The KVM_REG_ARM64_SVE regs must be used instead of
112 	 * KVM_REG_ARM_CORE for accessing the FPSIMD V-registers on
113 	 * SVE-enabled vcpus:
114 	 */
115 	if (vcpu_has_sve(vcpu) && core_reg_offset_is_vreg(off))
116 		return -EINVAL;
117 
118 	return size;
119 }
120 
121 static void *core_reg_addr(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
122 {
123 	u64 off = core_reg_offset_from_id(reg->id);
124 	int size = core_reg_size_from_offset(vcpu, off);
125 
126 	if (size < 0)
127 		return NULL;
128 
129 	if (KVM_REG_SIZE(reg->id) != size)
130 		return NULL;
131 
132 	switch (off) {
133 	case KVM_REG_ARM_CORE_REG(regs.regs[0]) ...
134 	     KVM_REG_ARM_CORE_REG(regs.regs[30]):
135 		off -= KVM_REG_ARM_CORE_REG(regs.regs[0]);
136 		off /= 2;
137 		return &vcpu->arch.ctxt.regs.regs[off];
138 
139 	case KVM_REG_ARM_CORE_REG(regs.sp):
140 		return &vcpu->arch.ctxt.regs.sp;
141 
142 	case KVM_REG_ARM_CORE_REG(regs.pc):
143 		return &vcpu->arch.ctxt.regs.pc;
144 
145 	case KVM_REG_ARM_CORE_REG(regs.pstate):
146 		return &vcpu->arch.ctxt.regs.pstate;
147 
148 	case KVM_REG_ARM_CORE_REG(sp_el1):
149 		return __ctxt_sys_reg(&vcpu->arch.ctxt, SP_EL1);
150 
151 	case KVM_REG_ARM_CORE_REG(elr_el1):
152 		return __ctxt_sys_reg(&vcpu->arch.ctxt, ELR_EL1);
153 
154 	case KVM_REG_ARM_CORE_REG(spsr[KVM_SPSR_EL1]):
155 		return __ctxt_sys_reg(&vcpu->arch.ctxt, SPSR_EL1);
156 
157 	case KVM_REG_ARM_CORE_REG(spsr[KVM_SPSR_ABT]):
158 		return &vcpu->arch.ctxt.spsr_abt;
159 
160 	case KVM_REG_ARM_CORE_REG(spsr[KVM_SPSR_UND]):
161 		return &vcpu->arch.ctxt.spsr_und;
162 
163 	case KVM_REG_ARM_CORE_REG(spsr[KVM_SPSR_IRQ]):
164 		return &vcpu->arch.ctxt.spsr_irq;
165 
166 	case KVM_REG_ARM_CORE_REG(spsr[KVM_SPSR_FIQ]):
167 		return &vcpu->arch.ctxt.spsr_fiq;
168 
169 	case KVM_REG_ARM_CORE_REG(fp_regs.vregs[0]) ...
170 	     KVM_REG_ARM_CORE_REG(fp_regs.vregs[31]):
171 		off -= KVM_REG_ARM_CORE_REG(fp_regs.vregs[0]);
172 		off /= 4;
173 		return &vcpu->arch.ctxt.fp_regs.vregs[off];
174 
175 	case KVM_REG_ARM_CORE_REG(fp_regs.fpsr):
176 		return &vcpu->arch.ctxt.fp_regs.fpsr;
177 
178 	case KVM_REG_ARM_CORE_REG(fp_regs.fpcr):
179 		return &vcpu->arch.ctxt.fp_regs.fpcr;
180 
181 	default:
182 		return NULL;
183 	}
184 }
185 
186 static int get_core_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
187 {
188 	/*
189 	 * Because the kvm_regs structure is a mix of 32, 64 and
190 	 * 128bit fields, we index it as if it was a 32bit
191 	 * array. Hence below, nr_regs is the number of entries, and
192 	 * off the index in the "array".
193 	 */
194 	__u32 __user *uaddr = (__u32 __user *)(unsigned long)reg->addr;
195 	int nr_regs = sizeof(struct kvm_regs) / sizeof(__u32);
196 	void *addr;
197 	u32 off;
198 
199 	/* Our ID is an index into the kvm_regs struct. */
200 	off = core_reg_offset_from_id(reg->id);
201 	if (off >= nr_regs ||
202 	    (off + (KVM_REG_SIZE(reg->id) / sizeof(__u32))) >= nr_regs)
203 		return -ENOENT;
204 
205 	addr = core_reg_addr(vcpu, reg);
206 	if (!addr)
207 		return -EINVAL;
208 
209 	if (copy_to_user(uaddr, addr, KVM_REG_SIZE(reg->id)))
210 		return -EFAULT;
211 
212 	return 0;
213 }
214 
215 static int set_core_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
216 {
217 	__u32 __user *uaddr = (__u32 __user *)(unsigned long)reg->addr;
218 	int nr_regs = sizeof(struct kvm_regs) / sizeof(__u32);
219 	__uint128_t tmp;
220 	void *valp = &tmp, *addr;
221 	u64 off;
222 	int err = 0;
223 
224 	/* Our ID is an index into the kvm_regs struct. */
225 	off = core_reg_offset_from_id(reg->id);
226 	if (off >= nr_regs ||
227 	    (off + (KVM_REG_SIZE(reg->id) / sizeof(__u32))) >= nr_regs)
228 		return -ENOENT;
229 
230 	addr = core_reg_addr(vcpu, reg);
231 	if (!addr)
232 		return -EINVAL;
233 
234 	if (KVM_REG_SIZE(reg->id) > sizeof(tmp))
235 		return -EINVAL;
236 
237 	if (copy_from_user(valp, uaddr, KVM_REG_SIZE(reg->id))) {
238 		err = -EFAULT;
239 		goto out;
240 	}
241 
242 	if (off == KVM_REG_ARM_CORE_REG(regs.pstate)) {
243 		u64 mode = (*(u64 *)valp) & PSR_AA32_MODE_MASK;
244 		switch (mode) {
245 		case PSR_AA32_MODE_USR:
246 			if (!kvm_supports_32bit_el0())
247 				return -EINVAL;
248 			break;
249 		case PSR_AA32_MODE_FIQ:
250 		case PSR_AA32_MODE_IRQ:
251 		case PSR_AA32_MODE_SVC:
252 		case PSR_AA32_MODE_ABT:
253 		case PSR_AA32_MODE_UND:
254 		case PSR_AA32_MODE_SYS:
255 			if (!vcpu_el1_is_32bit(vcpu))
256 				return -EINVAL;
257 			break;
258 		case PSR_MODE_EL2h:
259 		case PSR_MODE_EL2t:
260 			if (!vcpu_has_nv(vcpu))
261 				return -EINVAL;
262 			fallthrough;
263 		case PSR_MODE_EL0t:
264 		case PSR_MODE_EL1t:
265 		case PSR_MODE_EL1h:
266 			if (vcpu_el1_is_32bit(vcpu))
267 				return -EINVAL;
268 			break;
269 		default:
270 			err = -EINVAL;
271 			goto out;
272 		}
273 	}
274 
275 	memcpy(addr, valp, KVM_REG_SIZE(reg->id));
276 
277 	if (*vcpu_cpsr(vcpu) & PSR_MODE32_BIT) {
278 		int i, nr_reg;
279 
280 		switch (*vcpu_cpsr(vcpu) & PSR_AA32_MODE_MASK) {
281 		/*
282 		 * Either we are dealing with user mode, and only the
283 		 * first 15 registers (+ PC) must be narrowed to 32bit.
284 		 * AArch32 r0-r14 conveniently map to AArch64 x0-x14.
285 		 */
286 		case PSR_AA32_MODE_USR:
287 		case PSR_AA32_MODE_SYS:
288 			nr_reg = 15;
289 			break;
290 
291 		/*
292 		 * Otherwise, this is a privileged mode, and *all* the
293 		 * registers must be narrowed to 32bit.
294 		 */
295 		default:
296 			nr_reg = 31;
297 			break;
298 		}
299 
300 		for (i = 0; i < nr_reg; i++)
301 			vcpu_set_reg(vcpu, i, (u32)vcpu_get_reg(vcpu, i));
302 
303 		*vcpu_pc(vcpu) = (u32)*vcpu_pc(vcpu);
304 	}
305 out:
306 	return err;
307 }
308 
309 #define vq_word(vq) (((vq) - SVE_VQ_MIN) / 64)
310 #define vq_mask(vq) ((u64)1 << ((vq) - SVE_VQ_MIN) % 64)
311 #define vq_present(vqs, vq) (!!((vqs)[vq_word(vq)] & vq_mask(vq)))
312 
313 static int get_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
314 {
315 	unsigned int max_vq, vq;
316 	u64 vqs[KVM_ARM64_SVE_VLS_WORDS];
317 
318 	if (!vcpu_has_sve(vcpu))
319 		return -ENOENT;
320 
321 	if (WARN_ON(!sve_vl_valid(vcpu->arch.sve_max_vl)))
322 		return -EINVAL;
323 
324 	memset(vqs, 0, sizeof(vqs));
325 
326 	max_vq = vcpu_sve_max_vq(vcpu);
327 	for (vq = SVE_VQ_MIN; vq <= max_vq; ++vq)
328 		if (sve_vq_available(vq))
329 			vqs[vq_word(vq)] |= vq_mask(vq);
330 
331 	if (copy_to_user((void __user *)reg->addr, vqs, sizeof(vqs)))
332 		return -EFAULT;
333 
334 	return 0;
335 }
336 
337 static int set_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
338 {
339 	unsigned int max_vq, vq;
340 	u64 vqs[KVM_ARM64_SVE_VLS_WORDS];
341 
342 	if (!vcpu_has_sve(vcpu))
343 		return -ENOENT;
344 
345 	if (kvm_arm_vcpu_sve_finalized(vcpu))
346 		return -EPERM; /* too late! */
347 
348 	if (WARN_ON(vcpu->arch.sve_state))
349 		return -EINVAL;
350 
351 	if (copy_from_user(vqs, (const void __user *)reg->addr, sizeof(vqs)))
352 		return -EFAULT;
353 
354 	max_vq = 0;
355 	for (vq = SVE_VQ_MIN; vq <= SVE_VQ_MAX; ++vq)
356 		if (vq_present(vqs, vq))
357 			max_vq = vq;
358 
359 	if (max_vq > sve_vq_from_vl(kvm_sve_max_vl))
360 		return -EINVAL;
361 
362 	/*
363 	 * Vector lengths supported by the host can't currently be
364 	 * hidden from the guest individually: instead we can only set a
365 	 * maximum via ZCR_EL2.LEN.  So, make sure the available vector
366 	 * lengths match the set requested exactly up to the requested
367 	 * maximum:
368 	 */
369 	for (vq = SVE_VQ_MIN; vq <= max_vq; ++vq)
370 		if (vq_present(vqs, vq) != sve_vq_available(vq))
371 			return -EINVAL;
372 
373 	/* Can't run with no vector lengths at all: */
374 	if (max_vq < SVE_VQ_MIN)
375 		return -EINVAL;
376 
377 	/* vcpu->arch.sve_state will be alloc'd by kvm_vcpu_finalize_sve() */
378 	vcpu->arch.sve_max_vl = sve_vl_from_vq(max_vq);
379 
380 	return 0;
381 }
382 
383 #define SVE_REG_SLICE_SHIFT	0
384 #define SVE_REG_SLICE_BITS	5
385 #define SVE_REG_ID_SHIFT	(SVE_REG_SLICE_SHIFT + SVE_REG_SLICE_BITS)
386 #define SVE_REG_ID_BITS		5
387 
388 #define SVE_REG_SLICE_MASK					\
389 	GENMASK(SVE_REG_SLICE_SHIFT + SVE_REG_SLICE_BITS - 1,	\
390 		SVE_REG_SLICE_SHIFT)
391 #define SVE_REG_ID_MASK							\
392 	GENMASK(SVE_REG_ID_SHIFT + SVE_REG_ID_BITS - 1, SVE_REG_ID_SHIFT)
393 
394 #define SVE_NUM_SLICES (1 << SVE_REG_SLICE_BITS)
395 
396 #define KVM_SVE_ZREG_SIZE KVM_REG_SIZE(KVM_REG_ARM64_SVE_ZREG(0, 0))
397 #define KVM_SVE_PREG_SIZE KVM_REG_SIZE(KVM_REG_ARM64_SVE_PREG(0, 0))
398 
399 /*
400  * Number of register slices required to cover each whole SVE register.
401  * NOTE: Only the first slice every exists, for now.
402  * If you are tempted to modify this, you must also rework sve_reg_to_region()
403  * to match:
404  */
405 #define vcpu_sve_slices(vcpu) 1
406 
407 /* Bounds of a single SVE register slice within vcpu->arch.sve_state */
408 struct sve_state_reg_region {
409 	unsigned int koffset;	/* offset into sve_state in kernel memory */
410 	unsigned int klen;	/* length in kernel memory */
411 	unsigned int upad;	/* extra trailing padding in user memory */
412 };
413 
414 /*
415  * Validate SVE register ID and get sanitised bounds for user/kernel SVE
416  * register copy
417  */
418 static int sve_reg_to_region(struct sve_state_reg_region *region,
419 			     struct kvm_vcpu *vcpu,
420 			     const struct kvm_one_reg *reg)
421 {
422 	/* reg ID ranges for Z- registers */
423 	const u64 zreg_id_min = KVM_REG_ARM64_SVE_ZREG(0, 0);
424 	const u64 zreg_id_max = KVM_REG_ARM64_SVE_ZREG(SVE_NUM_ZREGS - 1,
425 						       SVE_NUM_SLICES - 1);
426 
427 	/* reg ID ranges for P- registers and FFR (which are contiguous) */
428 	const u64 preg_id_min = KVM_REG_ARM64_SVE_PREG(0, 0);
429 	const u64 preg_id_max = KVM_REG_ARM64_SVE_FFR(SVE_NUM_SLICES - 1);
430 
431 	unsigned int vq;
432 	unsigned int reg_num;
433 
434 	unsigned int reqoffset, reqlen; /* User-requested offset and length */
435 	unsigned int maxlen; /* Maximum permitted length */
436 
437 	size_t sve_state_size;
438 
439 	const u64 last_preg_id = KVM_REG_ARM64_SVE_PREG(SVE_NUM_PREGS - 1,
440 							SVE_NUM_SLICES - 1);
441 
442 	/* Verify that the P-regs and FFR really do have contiguous IDs: */
443 	BUILD_BUG_ON(KVM_REG_ARM64_SVE_FFR(0) != last_preg_id + 1);
444 
445 	/* Verify that we match the UAPI header: */
446 	BUILD_BUG_ON(SVE_NUM_SLICES != KVM_ARM64_SVE_MAX_SLICES);
447 
448 	reg_num = (reg->id & SVE_REG_ID_MASK) >> SVE_REG_ID_SHIFT;
449 
450 	if (reg->id >= zreg_id_min && reg->id <= zreg_id_max) {
451 		if (!vcpu_has_sve(vcpu) || (reg->id & SVE_REG_SLICE_MASK) > 0)
452 			return -ENOENT;
453 
454 		vq = vcpu_sve_max_vq(vcpu);
455 
456 		reqoffset = SVE_SIG_ZREG_OFFSET(vq, reg_num) -
457 				SVE_SIG_REGS_OFFSET;
458 		reqlen = KVM_SVE_ZREG_SIZE;
459 		maxlen = SVE_SIG_ZREG_SIZE(vq);
460 	} else if (reg->id >= preg_id_min && reg->id <= preg_id_max) {
461 		if (!vcpu_has_sve(vcpu) || (reg->id & SVE_REG_SLICE_MASK) > 0)
462 			return -ENOENT;
463 
464 		vq = vcpu_sve_max_vq(vcpu);
465 
466 		reqoffset = SVE_SIG_PREG_OFFSET(vq, reg_num) -
467 				SVE_SIG_REGS_OFFSET;
468 		reqlen = KVM_SVE_PREG_SIZE;
469 		maxlen = SVE_SIG_PREG_SIZE(vq);
470 	} else {
471 		return -EINVAL;
472 	}
473 
474 	sve_state_size = vcpu_sve_state_size(vcpu);
475 	if (WARN_ON(!sve_state_size))
476 		return -EINVAL;
477 
478 	region->koffset = array_index_nospec(reqoffset, sve_state_size);
479 	region->klen = min(maxlen, reqlen);
480 	region->upad = reqlen - region->klen;
481 
482 	return 0;
483 }
484 
485 static int get_sve_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
486 {
487 	int ret;
488 	struct sve_state_reg_region region;
489 	char __user *uptr = (char __user *)reg->addr;
490 
491 	/* Handle the KVM_REG_ARM64_SVE_VLS pseudo-reg as a special case: */
492 	if (reg->id == KVM_REG_ARM64_SVE_VLS)
493 		return get_sve_vls(vcpu, reg);
494 
495 	/* Try to interpret reg ID as an architectural SVE register... */
496 	ret = sve_reg_to_region(&region, vcpu, reg);
497 	if (ret)
498 		return ret;
499 
500 	if (!kvm_arm_vcpu_sve_finalized(vcpu))
501 		return -EPERM;
502 
503 	if (copy_to_user(uptr, vcpu->arch.sve_state + region.koffset,
504 			 region.klen) ||
505 	    clear_user(uptr + region.klen, region.upad))
506 		return -EFAULT;
507 
508 	return 0;
509 }
510 
511 static int set_sve_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
512 {
513 	int ret;
514 	struct sve_state_reg_region region;
515 	const char __user *uptr = (const char __user *)reg->addr;
516 
517 	/* Handle the KVM_REG_ARM64_SVE_VLS pseudo-reg as a special case: */
518 	if (reg->id == KVM_REG_ARM64_SVE_VLS)
519 		return set_sve_vls(vcpu, reg);
520 
521 	/* Try to interpret reg ID as an architectural SVE register... */
522 	ret = sve_reg_to_region(&region, vcpu, reg);
523 	if (ret)
524 		return ret;
525 
526 	if (!kvm_arm_vcpu_sve_finalized(vcpu))
527 		return -EPERM;
528 
529 	if (copy_from_user(vcpu->arch.sve_state + region.koffset, uptr,
530 			   region.klen))
531 		return -EFAULT;
532 
533 	return 0;
534 }
535 
536 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
537 {
538 	return -EINVAL;
539 }
540 
541 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
542 {
543 	return -EINVAL;
544 }
545 
546 static int copy_core_reg_indices(const struct kvm_vcpu *vcpu,
547 				 u64 __user *uindices)
548 {
549 	unsigned int i;
550 	int n = 0;
551 
552 	for (i = 0; i < sizeof(struct kvm_regs) / sizeof(__u32); i++) {
553 		u64 reg = KVM_REG_ARM64 | KVM_REG_ARM_CORE | i;
554 		int size = core_reg_size_from_offset(vcpu, i);
555 
556 		if (size < 0)
557 			continue;
558 
559 		switch (size) {
560 		case sizeof(__u32):
561 			reg |= KVM_REG_SIZE_U32;
562 			break;
563 
564 		case sizeof(__u64):
565 			reg |= KVM_REG_SIZE_U64;
566 			break;
567 
568 		case sizeof(__uint128_t):
569 			reg |= KVM_REG_SIZE_U128;
570 			break;
571 
572 		default:
573 			WARN_ON(1);
574 			continue;
575 		}
576 
577 		if (uindices) {
578 			if (put_user(reg, uindices))
579 				return -EFAULT;
580 			uindices++;
581 		}
582 
583 		n++;
584 	}
585 
586 	return n;
587 }
588 
589 static unsigned long num_core_regs(const struct kvm_vcpu *vcpu)
590 {
591 	return copy_core_reg_indices(vcpu, NULL);
592 }
593 
594 static const u64 timer_reg_list[] = {
595 	KVM_REG_ARM_TIMER_CTL,
596 	KVM_REG_ARM_TIMER_CNT,
597 	KVM_REG_ARM_TIMER_CVAL,
598 	KVM_REG_ARM_PTIMER_CTL,
599 	KVM_REG_ARM_PTIMER_CNT,
600 	KVM_REG_ARM_PTIMER_CVAL,
601 };
602 
603 #define NUM_TIMER_REGS ARRAY_SIZE(timer_reg_list)
604 
605 static bool is_timer_reg(u64 index)
606 {
607 	switch (index) {
608 	case KVM_REG_ARM_TIMER_CTL:
609 	case KVM_REG_ARM_TIMER_CNT:
610 	case KVM_REG_ARM_TIMER_CVAL:
611 	case KVM_REG_ARM_PTIMER_CTL:
612 	case KVM_REG_ARM_PTIMER_CNT:
613 	case KVM_REG_ARM_PTIMER_CVAL:
614 		return true;
615 	}
616 	return false;
617 }
618 
619 static int copy_timer_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
620 {
621 	for (int i = 0; i < NUM_TIMER_REGS; i++) {
622 		if (put_user(timer_reg_list[i], uindices))
623 			return -EFAULT;
624 		uindices++;
625 	}
626 
627 	return 0;
628 }
629 
630 static int set_timer_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
631 {
632 	void __user *uaddr = (void __user *)(long)reg->addr;
633 	u64 val;
634 	int ret;
635 
636 	ret = copy_from_user(&val, uaddr, KVM_REG_SIZE(reg->id));
637 	if (ret != 0)
638 		return -EFAULT;
639 
640 	return kvm_arm_timer_set_reg(vcpu, reg->id, val);
641 }
642 
643 static int get_timer_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
644 {
645 	void __user *uaddr = (void __user *)(long)reg->addr;
646 	u64 val;
647 
648 	val = kvm_arm_timer_get_reg(vcpu, reg->id);
649 	return copy_to_user(uaddr, &val, KVM_REG_SIZE(reg->id)) ? -EFAULT : 0;
650 }
651 
652 static unsigned long num_sve_regs(const struct kvm_vcpu *vcpu)
653 {
654 	const unsigned int slices = vcpu_sve_slices(vcpu);
655 
656 	if (!vcpu_has_sve(vcpu))
657 		return 0;
658 
659 	/* Policed by KVM_GET_REG_LIST: */
660 	WARN_ON(!kvm_arm_vcpu_sve_finalized(vcpu));
661 
662 	return slices * (SVE_NUM_PREGS + SVE_NUM_ZREGS + 1 /* FFR */)
663 		+ 1; /* KVM_REG_ARM64_SVE_VLS */
664 }
665 
666 static int copy_sve_reg_indices(const struct kvm_vcpu *vcpu,
667 				u64 __user *uindices)
668 {
669 	const unsigned int slices = vcpu_sve_slices(vcpu);
670 	u64 reg;
671 	unsigned int i, n;
672 	int num_regs = 0;
673 
674 	if (!vcpu_has_sve(vcpu))
675 		return 0;
676 
677 	/* Policed by KVM_GET_REG_LIST: */
678 	WARN_ON(!kvm_arm_vcpu_sve_finalized(vcpu));
679 
680 	/*
681 	 * Enumerate this first, so that userspace can save/restore in
682 	 * the order reported by KVM_GET_REG_LIST:
683 	 */
684 	reg = KVM_REG_ARM64_SVE_VLS;
685 	if (put_user(reg, uindices++))
686 		return -EFAULT;
687 	++num_regs;
688 
689 	for (i = 0; i < slices; i++) {
690 		for (n = 0; n < SVE_NUM_ZREGS; n++) {
691 			reg = KVM_REG_ARM64_SVE_ZREG(n, i);
692 			if (put_user(reg, uindices++))
693 				return -EFAULT;
694 			num_regs++;
695 		}
696 
697 		for (n = 0; n < SVE_NUM_PREGS; n++) {
698 			reg = KVM_REG_ARM64_SVE_PREG(n, i);
699 			if (put_user(reg, uindices++))
700 				return -EFAULT;
701 			num_regs++;
702 		}
703 
704 		reg = KVM_REG_ARM64_SVE_FFR(i);
705 		if (put_user(reg, uindices++))
706 			return -EFAULT;
707 		num_regs++;
708 	}
709 
710 	return num_regs;
711 }
712 
713 /**
714  * kvm_arm_num_regs - how many registers do we present via KVM_GET_ONE_REG
715  *
716  * This is for all registers.
717  */
718 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu)
719 {
720 	unsigned long res = 0;
721 
722 	res += num_core_regs(vcpu);
723 	res += num_sve_regs(vcpu);
724 	res += kvm_arm_num_sys_reg_descs(vcpu);
725 	res += kvm_arm_get_fw_num_regs(vcpu);
726 	res += NUM_TIMER_REGS;
727 
728 	return res;
729 }
730 
731 /**
732  * kvm_arm_copy_reg_indices - get indices of all registers.
733  *
734  * We do core registers right here, then we append system regs.
735  */
736 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
737 {
738 	int ret;
739 
740 	ret = copy_core_reg_indices(vcpu, uindices);
741 	if (ret < 0)
742 		return ret;
743 	uindices += ret;
744 
745 	ret = copy_sve_reg_indices(vcpu, uindices);
746 	if (ret < 0)
747 		return ret;
748 	uindices += ret;
749 
750 	ret = kvm_arm_copy_fw_reg_indices(vcpu, uindices);
751 	if (ret < 0)
752 		return ret;
753 	uindices += kvm_arm_get_fw_num_regs(vcpu);
754 
755 	ret = copy_timer_indices(vcpu, uindices);
756 	if (ret < 0)
757 		return ret;
758 	uindices += NUM_TIMER_REGS;
759 
760 	return kvm_arm_copy_sys_reg_indices(vcpu, uindices);
761 }
762 
763 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
764 {
765 	/* We currently use nothing arch-specific in upper 32 bits */
766 	if ((reg->id & ~KVM_REG_SIZE_MASK) >> 32 != KVM_REG_ARM64 >> 32)
767 		return -EINVAL;
768 
769 	switch (reg->id & KVM_REG_ARM_COPROC_MASK) {
770 	case KVM_REG_ARM_CORE:	return get_core_reg(vcpu, reg);
771 	case KVM_REG_ARM_FW:
772 	case KVM_REG_ARM_FW_FEAT_BMAP:
773 		return kvm_arm_get_fw_reg(vcpu, reg);
774 	case KVM_REG_ARM64_SVE:	return get_sve_reg(vcpu, reg);
775 	}
776 
777 	if (is_timer_reg(reg->id))
778 		return get_timer_reg(vcpu, reg);
779 
780 	return kvm_arm_sys_reg_get_reg(vcpu, reg);
781 }
782 
783 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
784 {
785 	/* We currently use nothing arch-specific in upper 32 bits */
786 	if ((reg->id & ~KVM_REG_SIZE_MASK) >> 32 != KVM_REG_ARM64 >> 32)
787 		return -EINVAL;
788 
789 	switch (reg->id & KVM_REG_ARM_COPROC_MASK) {
790 	case KVM_REG_ARM_CORE:	return set_core_reg(vcpu, reg);
791 	case KVM_REG_ARM_FW:
792 	case KVM_REG_ARM_FW_FEAT_BMAP:
793 		return kvm_arm_set_fw_reg(vcpu, reg);
794 	case KVM_REG_ARM64_SVE:	return set_sve_reg(vcpu, reg);
795 	}
796 
797 	if (is_timer_reg(reg->id))
798 		return set_timer_reg(vcpu, reg);
799 
800 	return kvm_arm_sys_reg_set_reg(vcpu, reg);
801 }
802 
803 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
804 				  struct kvm_sregs *sregs)
805 {
806 	return -EINVAL;
807 }
808 
809 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
810 				  struct kvm_sregs *sregs)
811 {
812 	return -EINVAL;
813 }
814 
815 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
816 			      struct kvm_vcpu_events *events)
817 {
818 	events->exception.serror_pending = !!(vcpu->arch.hcr_el2 & HCR_VSE);
819 	events->exception.serror_has_esr = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
820 
821 	if (events->exception.serror_pending && events->exception.serror_has_esr)
822 		events->exception.serror_esr = vcpu_get_vsesr(vcpu);
823 
824 	/*
825 	 * We never return a pending ext_dabt here because we deliver it to
826 	 * the virtual CPU directly when setting the event and it's no longer
827 	 * 'pending' at this point.
828 	 */
829 
830 	return 0;
831 }
832 
833 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
834 			      struct kvm_vcpu_events *events)
835 {
836 	bool serror_pending = events->exception.serror_pending;
837 	bool has_esr = events->exception.serror_has_esr;
838 	bool ext_dabt_pending = events->exception.ext_dabt_pending;
839 
840 	if (serror_pending && has_esr) {
841 		if (!cpus_have_const_cap(ARM64_HAS_RAS_EXTN))
842 			return -EINVAL;
843 
844 		if (!((events->exception.serror_esr) & ~ESR_ELx_ISS_MASK))
845 			kvm_set_sei_esr(vcpu, events->exception.serror_esr);
846 		else
847 			return -EINVAL;
848 	} else if (serror_pending) {
849 		kvm_inject_vabt(vcpu);
850 	}
851 
852 	if (ext_dabt_pending)
853 		kvm_inject_dabt(vcpu, kvm_vcpu_get_hfar(vcpu));
854 
855 	return 0;
856 }
857 
858 u32 __attribute_const__ kvm_target_cpu(void)
859 {
860 	unsigned long implementor = read_cpuid_implementor();
861 	unsigned long part_number = read_cpuid_part_number();
862 
863 	switch (implementor) {
864 	case ARM_CPU_IMP_ARM:
865 		switch (part_number) {
866 		case ARM_CPU_PART_AEM_V8:
867 			return KVM_ARM_TARGET_AEM_V8;
868 		case ARM_CPU_PART_FOUNDATION:
869 			return KVM_ARM_TARGET_FOUNDATION_V8;
870 		case ARM_CPU_PART_CORTEX_A53:
871 			return KVM_ARM_TARGET_CORTEX_A53;
872 		case ARM_CPU_PART_CORTEX_A57:
873 			return KVM_ARM_TARGET_CORTEX_A57;
874 		}
875 		break;
876 	case ARM_CPU_IMP_APM:
877 		switch (part_number) {
878 		case APM_CPU_PART_XGENE:
879 			return KVM_ARM_TARGET_XGENE_POTENZA;
880 		}
881 		break;
882 	}
883 
884 	/* Return a default generic target */
885 	return KVM_ARM_TARGET_GENERIC_V8;
886 }
887 
888 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
889 {
890 	return -EINVAL;
891 }
892 
893 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
894 {
895 	return -EINVAL;
896 }
897 
898 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
899 				  struct kvm_translation *tr)
900 {
901 	return -EINVAL;
902 }
903 
904 /**
905  * kvm_arch_vcpu_ioctl_set_guest_debug - set up guest debugging
906  * @kvm:	pointer to the KVM struct
907  * @kvm_guest_debug: the ioctl data buffer
908  *
909  * This sets up and enables the VM for guest debugging. Userspace
910  * passes in a control flag to enable different debug types and
911  * potentially other architecture specific information in the rest of
912  * the structure.
913  */
914 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
915 					struct kvm_guest_debug *dbg)
916 {
917 	int ret = 0;
918 
919 	trace_kvm_set_guest_debug(vcpu, dbg->control);
920 
921 	if (dbg->control & ~KVM_GUESTDBG_VALID_MASK) {
922 		ret = -EINVAL;
923 		goto out;
924 	}
925 
926 	if (dbg->control & KVM_GUESTDBG_ENABLE) {
927 		vcpu->guest_debug = dbg->control;
928 
929 		/* Hardware assisted Break and Watch points */
930 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW) {
931 			vcpu->arch.external_debug_state = dbg->arch;
932 		}
933 
934 	} else {
935 		/* If not enabled clear all flags */
936 		vcpu->guest_debug = 0;
937 		vcpu_clear_flag(vcpu, DBG_SS_ACTIVE_PENDING);
938 	}
939 
940 out:
941 	return ret;
942 }
943 
944 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
945 			       struct kvm_device_attr *attr)
946 {
947 	int ret;
948 
949 	switch (attr->group) {
950 	case KVM_ARM_VCPU_PMU_V3_CTRL:
951 		mutex_lock(&vcpu->kvm->arch.config_lock);
952 		ret = kvm_arm_pmu_v3_set_attr(vcpu, attr);
953 		mutex_unlock(&vcpu->kvm->arch.config_lock);
954 		break;
955 	case KVM_ARM_VCPU_TIMER_CTRL:
956 		ret = kvm_arm_timer_set_attr(vcpu, attr);
957 		break;
958 	case KVM_ARM_VCPU_PVTIME_CTRL:
959 		ret = kvm_arm_pvtime_set_attr(vcpu, attr);
960 		break;
961 	default:
962 		ret = -ENXIO;
963 		break;
964 	}
965 
966 	return ret;
967 }
968 
969 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
970 			       struct kvm_device_attr *attr)
971 {
972 	int ret;
973 
974 	switch (attr->group) {
975 	case KVM_ARM_VCPU_PMU_V3_CTRL:
976 		ret = kvm_arm_pmu_v3_get_attr(vcpu, attr);
977 		break;
978 	case KVM_ARM_VCPU_TIMER_CTRL:
979 		ret = kvm_arm_timer_get_attr(vcpu, attr);
980 		break;
981 	case KVM_ARM_VCPU_PVTIME_CTRL:
982 		ret = kvm_arm_pvtime_get_attr(vcpu, attr);
983 		break;
984 	default:
985 		ret = -ENXIO;
986 		break;
987 	}
988 
989 	return ret;
990 }
991 
992 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
993 			       struct kvm_device_attr *attr)
994 {
995 	int ret;
996 
997 	switch (attr->group) {
998 	case KVM_ARM_VCPU_PMU_V3_CTRL:
999 		ret = kvm_arm_pmu_v3_has_attr(vcpu, attr);
1000 		break;
1001 	case KVM_ARM_VCPU_TIMER_CTRL:
1002 		ret = kvm_arm_timer_has_attr(vcpu, attr);
1003 		break;
1004 	case KVM_ARM_VCPU_PVTIME_CTRL:
1005 		ret = kvm_arm_pvtime_has_attr(vcpu, attr);
1006 		break;
1007 	default:
1008 		ret = -ENXIO;
1009 		break;
1010 	}
1011 
1012 	return ret;
1013 }
1014 
1015 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1016 			       struct kvm_arm_copy_mte_tags *copy_tags)
1017 {
1018 	gpa_t guest_ipa = copy_tags->guest_ipa;
1019 	size_t length = copy_tags->length;
1020 	void __user *tags = copy_tags->addr;
1021 	gpa_t gfn;
1022 	bool write = !(copy_tags->flags & KVM_ARM_TAGS_FROM_GUEST);
1023 	int ret = 0;
1024 
1025 	if (!kvm_has_mte(kvm))
1026 		return -EINVAL;
1027 
1028 	if (copy_tags->reserved[0] || copy_tags->reserved[1])
1029 		return -EINVAL;
1030 
1031 	if (copy_tags->flags & ~KVM_ARM_TAGS_FROM_GUEST)
1032 		return -EINVAL;
1033 
1034 	if (length & ~PAGE_MASK || guest_ipa & ~PAGE_MASK)
1035 		return -EINVAL;
1036 
1037 	/* Lengths above INT_MAX cannot be represented in the return value */
1038 	if (length > INT_MAX)
1039 		return -EINVAL;
1040 
1041 	gfn = gpa_to_gfn(guest_ipa);
1042 
1043 	mutex_lock(&kvm->slots_lock);
1044 
1045 	while (length > 0) {
1046 		kvm_pfn_t pfn = gfn_to_pfn_prot(kvm, gfn, write, NULL);
1047 		void *maddr;
1048 		unsigned long num_tags;
1049 		struct page *page;
1050 
1051 		if (is_error_noslot_pfn(pfn)) {
1052 			ret = -EFAULT;
1053 			goto out;
1054 		}
1055 
1056 		page = pfn_to_online_page(pfn);
1057 		if (!page) {
1058 			/* Reject ZONE_DEVICE memory */
1059 			ret = -EFAULT;
1060 			goto out;
1061 		}
1062 		maddr = page_address(page);
1063 
1064 		if (!write) {
1065 			if (page_mte_tagged(page))
1066 				num_tags = mte_copy_tags_to_user(tags, maddr,
1067 							MTE_GRANULES_PER_PAGE);
1068 			else
1069 				/* No tags in memory, so write zeros */
1070 				num_tags = MTE_GRANULES_PER_PAGE -
1071 					clear_user(tags, MTE_GRANULES_PER_PAGE);
1072 			kvm_release_pfn_clean(pfn);
1073 		} else {
1074 			/*
1075 			 * Only locking to serialise with a concurrent
1076 			 * set_pte_at() in the VMM but still overriding the
1077 			 * tags, hence ignoring the return value.
1078 			 */
1079 			try_page_mte_tagging(page);
1080 			num_tags = mte_copy_tags_from_user(maddr, tags,
1081 							MTE_GRANULES_PER_PAGE);
1082 
1083 			/* uaccess failed, don't leave stale tags */
1084 			if (num_tags != MTE_GRANULES_PER_PAGE)
1085 				mte_clear_page_tags(maddr);
1086 			set_page_mte_tagged(page);
1087 
1088 			kvm_release_pfn_dirty(pfn);
1089 		}
1090 
1091 		if (num_tags != MTE_GRANULES_PER_PAGE) {
1092 			ret = -EFAULT;
1093 			goto out;
1094 		}
1095 
1096 		gfn++;
1097 		tags += num_tags;
1098 		length -= PAGE_SIZE;
1099 	}
1100 
1101 out:
1102 	mutex_unlock(&kvm->slots_lock);
1103 	/* If some data has been copied report the number of bytes copied */
1104 	if (length != copy_tags->length)
1105 		return copy_tags->length - length;
1106 	return ret;
1107 }
1108