xref: /linux/arch/arm64/kvm/nested.c (revision 55d0969c451159cff86949b38c39171cab962069)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017 - Columbia University and Linaro Ltd.
4  * Author: Jintack Lim <jintack.lim@linaro.org>
5  */
6 
7 #include <linux/bitfield.h>
8 #include <linux/kvm.h>
9 #include <linux/kvm_host.h>
10 
11 #include <asm/kvm_arm.h>
12 #include <asm/kvm_emulate.h>
13 #include <asm/kvm_mmu.h>
14 #include <asm/kvm_nested.h>
15 #include <asm/sysreg.h>
16 
17 #include "sys_regs.h"
18 
19 /* Protection against the sysreg repainting madness... */
20 #define NV_FTR(r, f)		ID_AA64##r##_EL1_##f
21 
22 /*
23  * Ratio of live shadow S2 MMU per vcpu. This is a trade-off between
24  * memory usage and potential number of different sets of S2 PTs in
25  * the guests. Running out of S2 MMUs only affects performance (we
26  * will invalidate them more often).
27  */
28 #define S2_MMU_PER_VCPU		2
29 
30 void kvm_init_nested(struct kvm *kvm)
31 {
32 	kvm->arch.nested_mmus = NULL;
33 	kvm->arch.nested_mmus_size = 0;
34 }
35 
36 static int init_nested_s2_mmu(struct kvm *kvm, struct kvm_s2_mmu *mmu)
37 {
38 	/*
39 	 * We only initialise the IPA range on the canonical MMU, which
40 	 * defines the contract between KVM and userspace on where the
41 	 * "hardware" is in the IPA space. This affects the validity of MMIO
42 	 * exits forwarded to userspace, for example.
43 	 *
44 	 * For nested S2s, we use the PARange as exposed to the guest, as it
45 	 * is allowed to use it at will to expose whatever memory map it
46 	 * wants to its own guests as it would be on real HW.
47 	 */
48 	return kvm_init_stage2_mmu(kvm, mmu, kvm_get_pa_bits(kvm));
49 }
50 
51 int kvm_vcpu_init_nested(struct kvm_vcpu *vcpu)
52 {
53 	struct kvm *kvm = vcpu->kvm;
54 	struct kvm_s2_mmu *tmp;
55 	int num_mmus, ret = 0;
56 
57 	/*
58 	 * Let's treat memory allocation failures as benign: If we fail to
59 	 * allocate anything, return an error and keep the allocated array
60 	 * alive. Userspace may try to recover by intializing the vcpu
61 	 * again, and there is no reason to affect the whole VM for this.
62 	 */
63 	num_mmus = atomic_read(&kvm->online_vcpus) * S2_MMU_PER_VCPU;
64 	tmp = kvrealloc(kvm->arch.nested_mmus,
65 			size_mul(sizeof(*kvm->arch.nested_mmus), num_mmus),
66 			GFP_KERNEL_ACCOUNT | __GFP_ZERO);
67 	if (!tmp)
68 		return -ENOMEM;
69 
70 	/*
71 	 * If we went through a realocation, adjust the MMU back-pointers in
72 	 * the previously initialised kvm_pgtable structures.
73 	 */
74 	if (kvm->arch.nested_mmus != tmp)
75 		for (int i = 0; i < kvm->arch.nested_mmus_size; i++)
76 			tmp[i].pgt->mmu = &tmp[i];
77 
78 	for (int i = kvm->arch.nested_mmus_size; !ret && i < num_mmus; i++)
79 		ret = init_nested_s2_mmu(kvm, &tmp[i]);
80 
81 	if (ret) {
82 		for (int i = kvm->arch.nested_mmus_size; i < num_mmus; i++)
83 			kvm_free_stage2_pgd(&tmp[i]);
84 
85 		return ret;
86 	}
87 
88 	kvm->arch.nested_mmus_size = num_mmus;
89 	kvm->arch.nested_mmus = tmp;
90 
91 	return 0;
92 }
93 
94 struct s2_walk_info {
95 	int	     (*read_desc)(phys_addr_t pa, u64 *desc, void *data);
96 	void	     *data;
97 	u64	     baddr;
98 	unsigned int max_oa_bits;
99 	unsigned int pgshift;
100 	unsigned int sl;
101 	unsigned int t0sz;
102 	bool	     be;
103 };
104 
105 static u32 compute_fsc(int level, u32 fsc)
106 {
107 	return fsc | (level & 0x3);
108 }
109 
110 static int esr_s2_fault(struct kvm_vcpu *vcpu, int level, u32 fsc)
111 {
112 	u32 esr;
113 
114 	esr = kvm_vcpu_get_esr(vcpu) & ~ESR_ELx_FSC;
115 	esr |= compute_fsc(level, fsc);
116 	return esr;
117 }
118 
119 static int get_ia_size(struct s2_walk_info *wi)
120 {
121 	return 64 - wi->t0sz;
122 }
123 
124 static int check_base_s2_limits(struct s2_walk_info *wi,
125 				int level, int input_size, int stride)
126 {
127 	int start_size, ia_size;
128 
129 	ia_size = get_ia_size(wi);
130 
131 	/* Check translation limits */
132 	switch (BIT(wi->pgshift)) {
133 	case SZ_64K:
134 		if (level == 0 || (level == 1 && ia_size <= 42))
135 			return -EFAULT;
136 		break;
137 	case SZ_16K:
138 		if (level == 0 || (level == 1 && ia_size <= 40))
139 			return -EFAULT;
140 		break;
141 	case SZ_4K:
142 		if (level < 0 || (level == 0 && ia_size <= 42))
143 			return -EFAULT;
144 		break;
145 	}
146 
147 	/* Check input size limits */
148 	if (input_size > ia_size)
149 		return -EFAULT;
150 
151 	/* Check number of entries in starting level table */
152 	start_size = input_size - ((3 - level) * stride + wi->pgshift);
153 	if (start_size < 1 || start_size > stride + 4)
154 		return -EFAULT;
155 
156 	return 0;
157 }
158 
159 /* Check if output is within boundaries */
160 static int check_output_size(struct s2_walk_info *wi, phys_addr_t output)
161 {
162 	unsigned int output_size = wi->max_oa_bits;
163 
164 	if (output_size != 48 && (output & GENMASK_ULL(47, output_size)))
165 		return -1;
166 
167 	return 0;
168 }
169 
170 /*
171  * This is essentially a C-version of the pseudo code from the ARM ARM
172  * AArch64.TranslationTableWalk  function.  I strongly recommend looking at
173  * that pseudocode in trying to understand this.
174  *
175  * Must be called with the kvm->srcu read lock held
176  */
177 static int walk_nested_s2_pgd(phys_addr_t ipa,
178 			      struct s2_walk_info *wi, struct kvm_s2_trans *out)
179 {
180 	int first_block_level, level, stride, input_size, base_lower_bound;
181 	phys_addr_t base_addr;
182 	unsigned int addr_top, addr_bottom;
183 	u64 desc;  /* page table entry */
184 	int ret;
185 	phys_addr_t paddr;
186 
187 	switch (BIT(wi->pgshift)) {
188 	default:
189 	case SZ_64K:
190 	case SZ_16K:
191 		level = 3 - wi->sl;
192 		first_block_level = 2;
193 		break;
194 	case SZ_4K:
195 		level = 2 - wi->sl;
196 		first_block_level = 1;
197 		break;
198 	}
199 
200 	stride = wi->pgshift - 3;
201 	input_size = get_ia_size(wi);
202 	if (input_size > 48 || input_size < 25)
203 		return -EFAULT;
204 
205 	ret = check_base_s2_limits(wi, level, input_size, stride);
206 	if (WARN_ON(ret))
207 		return ret;
208 
209 	base_lower_bound = 3 + input_size - ((3 - level) * stride +
210 			   wi->pgshift);
211 	base_addr = wi->baddr & GENMASK_ULL(47, base_lower_bound);
212 
213 	if (check_output_size(wi, base_addr)) {
214 		out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
215 		return 1;
216 	}
217 
218 	addr_top = input_size - 1;
219 
220 	while (1) {
221 		phys_addr_t index;
222 
223 		addr_bottom = (3 - level) * stride + wi->pgshift;
224 		index = (ipa & GENMASK_ULL(addr_top, addr_bottom))
225 			>> (addr_bottom - 3);
226 
227 		paddr = base_addr | index;
228 		ret = wi->read_desc(paddr, &desc, wi->data);
229 		if (ret < 0)
230 			return ret;
231 
232 		/*
233 		 * Handle reversedescriptors if endianness differs between the
234 		 * host and the guest hypervisor.
235 		 */
236 		if (wi->be)
237 			desc = be64_to_cpu((__force __be64)desc);
238 		else
239 			desc = le64_to_cpu((__force __le64)desc);
240 
241 		/* Check for valid descriptor at this point */
242 		if (!(desc & 1) || ((desc & 3) == 1 && level == 3)) {
243 			out->esr = compute_fsc(level, ESR_ELx_FSC_FAULT);
244 			out->desc = desc;
245 			return 1;
246 		}
247 
248 		/* We're at the final level or block translation level */
249 		if ((desc & 3) == 1 || level == 3)
250 			break;
251 
252 		if (check_output_size(wi, desc)) {
253 			out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
254 			out->desc = desc;
255 			return 1;
256 		}
257 
258 		base_addr = desc & GENMASK_ULL(47, wi->pgshift);
259 
260 		level += 1;
261 		addr_top = addr_bottom - 1;
262 	}
263 
264 	if (level < first_block_level) {
265 		out->esr = compute_fsc(level, ESR_ELx_FSC_FAULT);
266 		out->desc = desc;
267 		return 1;
268 	}
269 
270 	if (check_output_size(wi, desc)) {
271 		out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
272 		out->desc = desc;
273 		return 1;
274 	}
275 
276 	if (!(desc & BIT(10))) {
277 		out->esr = compute_fsc(level, ESR_ELx_FSC_ACCESS);
278 		out->desc = desc;
279 		return 1;
280 	}
281 
282 	addr_bottom += contiguous_bit_shift(desc, wi, level);
283 
284 	/* Calculate and return the result */
285 	paddr = (desc & GENMASK_ULL(47, addr_bottom)) |
286 		(ipa & GENMASK_ULL(addr_bottom - 1, 0));
287 	out->output = paddr;
288 	out->block_size = 1UL << ((3 - level) * stride + wi->pgshift);
289 	out->readable = desc & (0b01 << 6);
290 	out->writable = desc & (0b10 << 6);
291 	out->level = level;
292 	out->desc = desc;
293 	return 0;
294 }
295 
296 static int read_guest_s2_desc(phys_addr_t pa, u64 *desc, void *data)
297 {
298 	struct kvm_vcpu *vcpu = data;
299 
300 	return kvm_read_guest(vcpu->kvm, pa, desc, sizeof(*desc));
301 }
302 
303 static void vtcr_to_walk_info(u64 vtcr, struct s2_walk_info *wi)
304 {
305 	wi->t0sz = vtcr & TCR_EL2_T0SZ_MASK;
306 
307 	switch (vtcr & VTCR_EL2_TG0_MASK) {
308 	case VTCR_EL2_TG0_4K:
309 		wi->pgshift = 12;	 break;
310 	case VTCR_EL2_TG0_16K:
311 		wi->pgshift = 14;	 break;
312 	case VTCR_EL2_TG0_64K:
313 	default:	    /* IMPDEF: treat any other value as 64k */
314 		wi->pgshift = 16;	 break;
315 	}
316 
317 	wi->sl = FIELD_GET(VTCR_EL2_SL0_MASK, vtcr);
318 	/* Global limit for now, should eventually be per-VM */
319 	wi->max_oa_bits = min(get_kvm_ipa_limit(),
320 			      ps_to_output_size(FIELD_GET(VTCR_EL2_PS_MASK, vtcr)));
321 }
322 
323 int kvm_walk_nested_s2(struct kvm_vcpu *vcpu, phys_addr_t gipa,
324 		       struct kvm_s2_trans *result)
325 {
326 	u64 vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
327 	struct s2_walk_info wi;
328 	int ret;
329 
330 	result->esr = 0;
331 
332 	if (!vcpu_has_nv(vcpu))
333 		return 0;
334 
335 	wi.read_desc = read_guest_s2_desc;
336 	wi.data = vcpu;
337 	wi.baddr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
338 
339 	vtcr_to_walk_info(vtcr, &wi);
340 
341 	wi.be = vcpu_read_sys_reg(vcpu, SCTLR_EL2) & SCTLR_ELx_EE;
342 
343 	ret = walk_nested_s2_pgd(gipa, &wi, result);
344 	if (ret)
345 		result->esr |= (kvm_vcpu_get_esr(vcpu) & ~ESR_ELx_FSC);
346 
347 	return ret;
348 }
349 
350 static unsigned int ttl_to_size(u8 ttl)
351 {
352 	int level = ttl & 3;
353 	int gran = (ttl >> 2) & 3;
354 	unsigned int max_size = 0;
355 
356 	switch (gran) {
357 	case TLBI_TTL_TG_4K:
358 		switch (level) {
359 		case 0:
360 			break;
361 		case 1:
362 			max_size = SZ_1G;
363 			break;
364 		case 2:
365 			max_size = SZ_2M;
366 			break;
367 		case 3:
368 			max_size = SZ_4K;
369 			break;
370 		}
371 		break;
372 	case TLBI_TTL_TG_16K:
373 		switch (level) {
374 		case 0:
375 		case 1:
376 			break;
377 		case 2:
378 			max_size = SZ_32M;
379 			break;
380 		case 3:
381 			max_size = SZ_16K;
382 			break;
383 		}
384 		break;
385 	case TLBI_TTL_TG_64K:
386 		switch (level) {
387 		case 0:
388 		case 1:
389 			/* No 52bit IPA support */
390 			break;
391 		case 2:
392 			max_size = SZ_512M;
393 			break;
394 		case 3:
395 			max_size = SZ_64K;
396 			break;
397 		}
398 		break;
399 	default:			/* No size information */
400 		break;
401 	}
402 
403 	return max_size;
404 }
405 
406 /*
407  * Compute the equivalent of the TTL field by parsing the shadow PT.  The
408  * granule size is extracted from the cached VTCR_EL2.TG0 while the level is
409  * retrieved from first entry carrying the level as a tag.
410  */
411 static u8 get_guest_mapping_ttl(struct kvm_s2_mmu *mmu, u64 addr)
412 {
413 	u64 tmp, sz = 0, vtcr = mmu->tlb_vtcr;
414 	kvm_pte_t pte;
415 	u8 ttl, level;
416 
417 	lockdep_assert_held_write(&kvm_s2_mmu_to_kvm(mmu)->mmu_lock);
418 
419 	switch (vtcr & VTCR_EL2_TG0_MASK) {
420 	case VTCR_EL2_TG0_4K:
421 		ttl = (TLBI_TTL_TG_4K << 2);
422 		break;
423 	case VTCR_EL2_TG0_16K:
424 		ttl = (TLBI_TTL_TG_16K << 2);
425 		break;
426 	case VTCR_EL2_TG0_64K:
427 	default:	    /* IMPDEF: treat any other value as 64k */
428 		ttl = (TLBI_TTL_TG_64K << 2);
429 		break;
430 	}
431 
432 	tmp = addr;
433 
434 again:
435 	/* Iteratively compute the block sizes for a particular granule size */
436 	switch (vtcr & VTCR_EL2_TG0_MASK) {
437 	case VTCR_EL2_TG0_4K:
438 		if	(sz < SZ_4K)	sz = SZ_4K;
439 		else if (sz < SZ_2M)	sz = SZ_2M;
440 		else if (sz < SZ_1G)	sz = SZ_1G;
441 		else			sz = 0;
442 		break;
443 	case VTCR_EL2_TG0_16K:
444 		if	(sz < SZ_16K)	sz = SZ_16K;
445 		else if (sz < SZ_32M)	sz = SZ_32M;
446 		else			sz = 0;
447 		break;
448 	case VTCR_EL2_TG0_64K:
449 	default:	    /* IMPDEF: treat any other value as 64k */
450 		if	(sz < SZ_64K)	sz = SZ_64K;
451 		else if (sz < SZ_512M)	sz = SZ_512M;
452 		else			sz = 0;
453 		break;
454 	}
455 
456 	if (sz == 0)
457 		return 0;
458 
459 	tmp &= ~(sz - 1);
460 	if (kvm_pgtable_get_leaf(mmu->pgt, tmp, &pte, NULL))
461 		goto again;
462 	if (!(pte & PTE_VALID))
463 		goto again;
464 	level = FIELD_GET(KVM_NV_GUEST_MAP_SZ, pte);
465 	if (!level)
466 		goto again;
467 
468 	ttl |= level;
469 
470 	/*
471 	 * We now have found some level information in the shadow S2. Check
472 	 * that the resulting range is actually including the original IPA.
473 	 */
474 	sz = ttl_to_size(ttl);
475 	if (addr < (tmp + sz))
476 		return ttl;
477 
478 	return 0;
479 }
480 
481 unsigned long compute_tlb_inval_range(struct kvm_s2_mmu *mmu, u64 val)
482 {
483 	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
484 	unsigned long max_size;
485 	u8 ttl;
486 
487 	ttl = FIELD_GET(TLBI_TTL_MASK, val);
488 
489 	if (!ttl || !kvm_has_feat(kvm, ID_AA64MMFR2_EL1, TTL, IMP)) {
490 		/* No TTL, check the shadow S2 for a hint */
491 		u64 addr = (val & GENMASK_ULL(35, 0)) << 12;
492 		ttl = get_guest_mapping_ttl(mmu, addr);
493 	}
494 
495 	max_size = ttl_to_size(ttl);
496 
497 	if (!max_size) {
498 		/* Compute the maximum extent of the invalidation */
499 		switch (mmu->tlb_vtcr & VTCR_EL2_TG0_MASK) {
500 		case VTCR_EL2_TG0_4K:
501 			max_size = SZ_1G;
502 			break;
503 		case VTCR_EL2_TG0_16K:
504 			max_size = SZ_32M;
505 			break;
506 		case VTCR_EL2_TG0_64K:
507 		default:    /* IMPDEF: treat any other value as 64k */
508 			/*
509 			 * No, we do not support 52bit IPA in nested yet. Once
510 			 * we do, this should be 4TB.
511 			 */
512 			max_size = SZ_512M;
513 			break;
514 		}
515 	}
516 
517 	WARN_ON(!max_size);
518 	return max_size;
519 }
520 
521 /*
522  * We can have multiple *different* MMU contexts with the same VMID:
523  *
524  * - S2 being enabled or not, hence differing by the HCR_EL2.VM bit
525  *
526  * - Multiple vcpus using private S2s (huh huh...), hence differing by the
527  *   VBBTR_EL2.BADDR address
528  *
529  * - A combination of the above...
530  *
531  * We can always identify which MMU context to pick at run-time.  However,
532  * TLB invalidation involving a VMID must take action on all the TLBs using
533  * this particular VMID. This translates into applying the same invalidation
534  * operation to all the contexts that are using this VMID. Moar phun!
535  */
536 void kvm_s2_mmu_iterate_by_vmid(struct kvm *kvm, u16 vmid,
537 				const union tlbi_info *info,
538 				void (*tlbi_callback)(struct kvm_s2_mmu *,
539 						      const union tlbi_info *))
540 {
541 	write_lock(&kvm->mmu_lock);
542 
543 	for (int i = 0; i < kvm->arch.nested_mmus_size; i++) {
544 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
545 
546 		if (!kvm_s2_mmu_valid(mmu))
547 			continue;
548 
549 		if (vmid == get_vmid(mmu->tlb_vttbr))
550 			tlbi_callback(mmu, info);
551 	}
552 
553 	write_unlock(&kvm->mmu_lock);
554 }
555 
556 struct kvm_s2_mmu *lookup_s2_mmu(struct kvm_vcpu *vcpu)
557 {
558 	struct kvm *kvm = vcpu->kvm;
559 	bool nested_stage2_enabled;
560 	u64 vttbr, vtcr, hcr;
561 
562 	lockdep_assert_held_write(&kvm->mmu_lock);
563 
564 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
565 	vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
566 	hcr = vcpu_read_sys_reg(vcpu, HCR_EL2);
567 
568 	nested_stage2_enabled = hcr & HCR_VM;
569 
570 	/* Don't consider the CnP bit for the vttbr match */
571 	vttbr &= ~VTTBR_CNP_BIT;
572 
573 	/*
574 	 * Two possibilities when looking up a S2 MMU context:
575 	 *
576 	 * - either S2 is enabled in the guest, and we need a context that is
577 	 *   S2-enabled and matches the full VTTBR (VMID+BADDR) and VTCR,
578 	 *   which makes it safe from a TLB conflict perspective (a broken
579 	 *   guest won't be able to generate them),
580 	 *
581 	 * - or S2 is disabled, and we need a context that is S2-disabled
582 	 *   and matches the VMID only, as all TLBs are tagged by VMID even
583 	 *   if S2 translation is disabled.
584 	 */
585 	for (int i = 0; i < kvm->arch.nested_mmus_size; i++) {
586 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
587 
588 		if (!kvm_s2_mmu_valid(mmu))
589 			continue;
590 
591 		if (nested_stage2_enabled &&
592 		    mmu->nested_stage2_enabled &&
593 		    vttbr == mmu->tlb_vttbr &&
594 		    vtcr == mmu->tlb_vtcr)
595 			return mmu;
596 
597 		if (!nested_stage2_enabled &&
598 		    !mmu->nested_stage2_enabled &&
599 		    get_vmid(vttbr) == get_vmid(mmu->tlb_vttbr))
600 			return mmu;
601 	}
602 	return NULL;
603 }
604 
605 static struct kvm_s2_mmu *get_s2_mmu_nested(struct kvm_vcpu *vcpu)
606 {
607 	struct kvm *kvm = vcpu->kvm;
608 	struct kvm_s2_mmu *s2_mmu;
609 	int i;
610 
611 	lockdep_assert_held_write(&vcpu->kvm->mmu_lock);
612 
613 	s2_mmu = lookup_s2_mmu(vcpu);
614 	if (s2_mmu)
615 		goto out;
616 
617 	/*
618 	 * Make sure we don't always search from the same point, or we
619 	 * will always reuse a potentially active context, leaving
620 	 * free contexts unused.
621 	 */
622 	for (i = kvm->arch.nested_mmus_next;
623 	     i < (kvm->arch.nested_mmus_size + kvm->arch.nested_mmus_next);
624 	     i++) {
625 		s2_mmu = &kvm->arch.nested_mmus[i % kvm->arch.nested_mmus_size];
626 
627 		if (atomic_read(&s2_mmu->refcnt) == 0)
628 			break;
629 	}
630 	BUG_ON(atomic_read(&s2_mmu->refcnt)); /* We have struct MMUs to spare */
631 
632 	/* Set the scene for the next search */
633 	kvm->arch.nested_mmus_next = (i + 1) % kvm->arch.nested_mmus_size;
634 
635 	/* Clear the old state */
636 	if (kvm_s2_mmu_valid(s2_mmu))
637 		kvm_stage2_unmap_range(s2_mmu, 0, kvm_phys_size(s2_mmu));
638 
639 	/*
640 	 * The virtual VMID (modulo CnP) will be used as a key when matching
641 	 * an existing kvm_s2_mmu.
642 	 *
643 	 * We cache VTCR at allocation time, once and for all. It'd be great
644 	 * if the guest didn't screw that one up, as this is not very
645 	 * forgiving...
646 	 */
647 	s2_mmu->tlb_vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2) & ~VTTBR_CNP_BIT;
648 	s2_mmu->tlb_vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
649 	s2_mmu->nested_stage2_enabled = vcpu_read_sys_reg(vcpu, HCR_EL2) & HCR_VM;
650 
651 out:
652 	atomic_inc(&s2_mmu->refcnt);
653 	return s2_mmu;
654 }
655 
656 void kvm_init_nested_s2_mmu(struct kvm_s2_mmu *mmu)
657 {
658 	/* CnP being set denotes an invalid entry */
659 	mmu->tlb_vttbr = VTTBR_CNP_BIT;
660 	mmu->nested_stage2_enabled = false;
661 	atomic_set(&mmu->refcnt, 0);
662 }
663 
664 void kvm_vcpu_load_hw_mmu(struct kvm_vcpu *vcpu)
665 {
666 	if (is_hyp_ctxt(vcpu)) {
667 		vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
668 	} else {
669 		write_lock(&vcpu->kvm->mmu_lock);
670 		vcpu->arch.hw_mmu = get_s2_mmu_nested(vcpu);
671 		write_unlock(&vcpu->kvm->mmu_lock);
672 	}
673 }
674 
675 void kvm_vcpu_put_hw_mmu(struct kvm_vcpu *vcpu)
676 {
677 	if (kvm_is_nested_s2_mmu(vcpu->kvm, vcpu->arch.hw_mmu)) {
678 		atomic_dec(&vcpu->arch.hw_mmu->refcnt);
679 		vcpu->arch.hw_mmu = NULL;
680 	}
681 }
682 
683 /*
684  * Returns non-zero if permission fault is handled by injecting it to the next
685  * level hypervisor.
686  */
687 int kvm_s2_handle_perm_fault(struct kvm_vcpu *vcpu, struct kvm_s2_trans *trans)
688 {
689 	bool forward_fault = false;
690 
691 	trans->esr = 0;
692 
693 	if (!kvm_vcpu_trap_is_permission_fault(vcpu))
694 		return 0;
695 
696 	if (kvm_vcpu_trap_is_iabt(vcpu)) {
697 		forward_fault = !kvm_s2_trans_executable(trans);
698 	} else {
699 		bool write_fault = kvm_is_write_fault(vcpu);
700 
701 		forward_fault = ((write_fault && !trans->writable) ||
702 				 (!write_fault && !trans->readable));
703 	}
704 
705 	if (forward_fault)
706 		trans->esr = esr_s2_fault(vcpu, trans->level, ESR_ELx_FSC_PERM);
707 
708 	return forward_fault;
709 }
710 
711 int kvm_inject_s2_fault(struct kvm_vcpu *vcpu, u64 esr_el2)
712 {
713 	vcpu_write_sys_reg(vcpu, vcpu->arch.fault.far_el2, FAR_EL2);
714 	vcpu_write_sys_reg(vcpu, vcpu->arch.fault.hpfar_el2, HPFAR_EL2);
715 
716 	return kvm_inject_nested_sync(vcpu, esr_el2);
717 }
718 
719 void kvm_nested_s2_wp(struct kvm *kvm)
720 {
721 	int i;
722 
723 	lockdep_assert_held_write(&kvm->mmu_lock);
724 
725 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
726 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
727 
728 		if (kvm_s2_mmu_valid(mmu))
729 			kvm_stage2_wp_range(mmu, 0, kvm_phys_size(mmu));
730 	}
731 }
732 
733 void kvm_nested_s2_unmap(struct kvm *kvm)
734 {
735 	int i;
736 
737 	lockdep_assert_held_write(&kvm->mmu_lock);
738 
739 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
740 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
741 
742 		if (kvm_s2_mmu_valid(mmu))
743 			kvm_stage2_unmap_range(mmu, 0, kvm_phys_size(mmu));
744 	}
745 }
746 
747 void kvm_nested_s2_flush(struct kvm *kvm)
748 {
749 	int i;
750 
751 	lockdep_assert_held_write(&kvm->mmu_lock);
752 
753 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
754 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
755 
756 		if (kvm_s2_mmu_valid(mmu))
757 			kvm_stage2_flush_range(mmu, 0, kvm_phys_size(mmu));
758 	}
759 }
760 
761 void kvm_arch_flush_shadow_all(struct kvm *kvm)
762 {
763 	int i;
764 
765 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
766 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
767 
768 		if (!WARN_ON(atomic_read(&mmu->refcnt)))
769 			kvm_free_stage2_pgd(mmu);
770 	}
771 	kvfree(kvm->arch.nested_mmus);
772 	kvm->arch.nested_mmus = NULL;
773 	kvm->arch.nested_mmus_size = 0;
774 	kvm_uninit_stage2_mmu(kvm);
775 }
776 
777 /*
778  * Our emulated CPU doesn't support all the possible features. For the
779  * sake of simplicity (and probably mental sanity), wipe out a number
780  * of feature bits we don't intend to support for the time being.
781  * This list should get updated as new features get added to the NV
782  * support, and new extension to the architecture.
783  */
784 static void limit_nv_id_regs(struct kvm *kvm)
785 {
786 	u64 val, tmp;
787 
788 	/* Support everything but TME */
789 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64ISAR0_EL1);
790 	val &= ~NV_FTR(ISAR0, TME);
791 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64ISAR0_EL1, val);
792 
793 	/* Support everything but Spec Invalidation and LS64 */
794 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64ISAR1_EL1);
795 	val &= ~(NV_FTR(ISAR1, LS64)	|
796 		 NV_FTR(ISAR1, SPECRES));
797 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64ISAR1_EL1, val);
798 
799 	/* No AMU, MPAM, S-EL2, or RAS */
800 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1);
801 	val &= ~(GENMASK_ULL(55, 52)	|
802 		 NV_FTR(PFR0, AMU)	|
803 		 NV_FTR(PFR0, MPAM)	|
804 		 NV_FTR(PFR0, SEL2)	|
805 		 NV_FTR(PFR0, RAS)	|
806 		 NV_FTR(PFR0, EL3)	|
807 		 NV_FTR(PFR0, EL2)	|
808 		 NV_FTR(PFR0, EL1));
809 	/* 64bit EL1/EL2/EL3 only */
810 	val |= FIELD_PREP(NV_FTR(PFR0, EL1), 0b0001);
811 	val |= FIELD_PREP(NV_FTR(PFR0, EL2), 0b0001);
812 	val |= FIELD_PREP(NV_FTR(PFR0, EL3), 0b0001);
813 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1, val);
814 
815 	/* Only support BTI, SSBS, CSV2_frac */
816 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR1_EL1);
817 	val &= (NV_FTR(PFR1, BT)	|
818 		NV_FTR(PFR1, SSBS)	|
819 		NV_FTR(PFR1, CSV2_frac));
820 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR1_EL1, val);
821 
822 	/* Hide ECV, ExS, Secure Memory */
823 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR0_EL1);
824 	val &= ~(NV_FTR(MMFR0, ECV)		|
825 		 NV_FTR(MMFR0, EXS)		|
826 		 NV_FTR(MMFR0, TGRAN4_2)	|
827 		 NV_FTR(MMFR0, TGRAN16_2)	|
828 		 NV_FTR(MMFR0, TGRAN64_2)	|
829 		 NV_FTR(MMFR0, SNSMEM));
830 
831 	/* Disallow unsupported S2 page sizes */
832 	switch (PAGE_SIZE) {
833 	case SZ_64K:
834 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN16_2), 0b0001);
835 		fallthrough;
836 	case SZ_16K:
837 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN4_2), 0b0001);
838 		fallthrough;
839 	case SZ_4K:
840 		/* Support everything */
841 		break;
842 	}
843 	/*
844 	 * Since we can't support a guest S2 page size smaller than
845 	 * the host's own page size (due to KVM only populating its
846 	 * own S2 using the kernel's page size), advertise the
847 	 * limitation using FEAT_GTG.
848 	 */
849 	switch (PAGE_SIZE) {
850 	case SZ_4K:
851 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN4_2), 0b0010);
852 		fallthrough;
853 	case SZ_16K:
854 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN16_2), 0b0010);
855 		fallthrough;
856 	case SZ_64K:
857 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN64_2), 0b0010);
858 		break;
859 	}
860 	/* Cap PARange to 48bits */
861 	tmp = FIELD_GET(NV_FTR(MMFR0, PARANGE), val);
862 	if (tmp > 0b0101) {
863 		val &= ~NV_FTR(MMFR0, PARANGE);
864 		val |= FIELD_PREP(NV_FTR(MMFR0, PARANGE), 0b0101);
865 	}
866 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR0_EL1, val);
867 
868 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR1_EL1);
869 	val &= (NV_FTR(MMFR1, HCX)	|
870 		NV_FTR(MMFR1, PAN)	|
871 		NV_FTR(MMFR1, LO)	|
872 		NV_FTR(MMFR1, HPDS)	|
873 		NV_FTR(MMFR1, VH)	|
874 		NV_FTR(MMFR1, VMIDBits));
875 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR1_EL1, val);
876 
877 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR2_EL1);
878 	val &= ~(NV_FTR(MMFR2, BBM)	|
879 		 NV_FTR(MMFR2, TTL)	|
880 		 GENMASK_ULL(47, 44)	|
881 		 NV_FTR(MMFR2, ST)	|
882 		 NV_FTR(MMFR2, CCIDX)	|
883 		 NV_FTR(MMFR2, VARange));
884 
885 	/* Force TTL support */
886 	val |= FIELD_PREP(NV_FTR(MMFR2, TTL), 0b0001);
887 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR2_EL1, val);
888 
889 	val = 0;
890 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
891 		val |= FIELD_PREP(NV_FTR(MMFR4, E2H0),
892 				  ID_AA64MMFR4_EL1_E2H0_NI_NV1);
893 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR4_EL1, val);
894 
895 	/* Only limited support for PMU, Debug, BPs and WPs */
896 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64DFR0_EL1);
897 	val &= (NV_FTR(DFR0, PMUVer)	|
898 		NV_FTR(DFR0, WRPs)	|
899 		NV_FTR(DFR0, BRPs)	|
900 		NV_FTR(DFR0, DebugVer));
901 
902 	/* Cap Debug to ARMv8.1 */
903 	tmp = FIELD_GET(NV_FTR(DFR0, DebugVer), val);
904 	if (tmp > 0b0111) {
905 		val &= ~NV_FTR(DFR0, DebugVer);
906 		val |= FIELD_PREP(NV_FTR(DFR0, DebugVer), 0b0111);
907 	}
908 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64DFR0_EL1, val);
909 }
910 
911 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg sr)
912 {
913 	u64 v = ctxt_sys_reg(&vcpu->arch.ctxt, sr);
914 	struct kvm_sysreg_masks *masks;
915 
916 	masks = vcpu->kvm->arch.sysreg_masks;
917 
918 	if (masks) {
919 		sr -= __VNCR_START__;
920 
921 		v &= ~masks->mask[sr].res0;
922 		v |= masks->mask[sr].res1;
923 	}
924 
925 	return v;
926 }
927 
928 static void set_sysreg_masks(struct kvm *kvm, int sr, u64 res0, u64 res1)
929 {
930 	int i = sr - __VNCR_START__;
931 
932 	kvm->arch.sysreg_masks->mask[i].res0 = res0;
933 	kvm->arch.sysreg_masks->mask[i].res1 = res1;
934 }
935 
936 int kvm_init_nv_sysregs(struct kvm *kvm)
937 {
938 	u64 res0, res1;
939 
940 	lockdep_assert_held(&kvm->arch.config_lock);
941 
942 	if (kvm->arch.sysreg_masks)
943 		return 0;
944 
945 	kvm->arch.sysreg_masks = kzalloc(sizeof(*(kvm->arch.sysreg_masks)),
946 					 GFP_KERNEL_ACCOUNT);
947 	if (!kvm->arch.sysreg_masks)
948 		return -ENOMEM;
949 
950 	limit_nv_id_regs(kvm);
951 
952 	/* VTTBR_EL2 */
953 	res0 = res1 = 0;
954 	if (!kvm_has_feat_enum(kvm, ID_AA64MMFR1_EL1, VMIDBits, 16))
955 		res0 |= GENMASK(63, 56);
956 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, CnP, IMP))
957 		res0 |= VTTBR_CNP_BIT;
958 	set_sysreg_masks(kvm, VTTBR_EL2, res0, res1);
959 
960 	/* VTCR_EL2 */
961 	res0 = GENMASK(63, 32) | GENMASK(30, 20);
962 	res1 = BIT(31);
963 	set_sysreg_masks(kvm, VTCR_EL2, res0, res1);
964 
965 	/* VMPIDR_EL2 */
966 	res0 = GENMASK(63, 40) | GENMASK(30, 24);
967 	res1 = BIT(31);
968 	set_sysreg_masks(kvm, VMPIDR_EL2, res0, res1);
969 
970 	/* HCR_EL2 */
971 	res0 = BIT(48);
972 	res1 = HCR_RW;
973 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, TWED, IMP))
974 		res0 |= GENMASK(63, 59);
975 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, MTE, MTE2))
976 		res0 |= (HCR_TID5 | HCR_DCT | HCR_ATA);
977 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, EVT, TTLBxS))
978 		res0 |= (HCR_TTLBIS | HCR_TTLBOS);
979 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, CSV2, CSV2_2) &&
980 	    !kvm_has_feat(kvm, ID_AA64PFR1_EL1, CSV2_frac, CSV2_1p2))
981 		res0 |= HCR_ENSCXT;
982 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, EVT, IMP))
983 		res0 |= (HCR_TOCU | HCR_TICAB | HCR_TID4);
984 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, V1P1))
985 		res0 |= HCR_AMVOFFEN;
986 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, V1P1))
987 		res0 |= HCR_FIEN;
988 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, FWB, IMP))
989 		res0 |= HCR_FWB;
990 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, NV, NV2))
991 		res0 |= HCR_NV2;
992 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, NV, IMP))
993 		res0 |= (HCR_AT | HCR_NV1 | HCR_NV);
994 	if (!(__vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_ADDRESS) &&
995 	      __vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_GENERIC)))
996 		res0 |= (HCR_API | HCR_APK);
997 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TME, IMP))
998 		res0 |= BIT(39);
999 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP))
1000 		res0 |= (HCR_TEA | HCR_TERR);
1001 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, LO, IMP))
1002 		res0 |= HCR_TLOR;
1003 	if (!kvm_has_feat(kvm, ID_AA64MMFR4_EL1, E2H0, IMP))
1004 		res1 |= HCR_E2H;
1005 	set_sysreg_masks(kvm, HCR_EL2, res0, res1);
1006 
1007 	/* HCRX_EL2 */
1008 	res0 = HCRX_EL2_RES0;
1009 	res1 = HCRX_EL2_RES1;
1010 	if (!kvm_has_feat(kvm, ID_AA64ISAR3_EL1, PACM, TRIVIAL_IMP))
1011 		res0 |= HCRX_EL2_PACMEn;
1012 	if (!kvm_has_feat(kvm, ID_AA64PFR2_EL1, FPMR, IMP))
1013 		res0 |= HCRX_EL2_EnFPM;
1014 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1015 		res0 |= HCRX_EL2_GCSEn;
1016 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, SYSREG_128, IMP))
1017 		res0 |= HCRX_EL2_EnIDCP128;
1018 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, ADERR, DEV_ASYNC))
1019 		res0 |= (HCRX_EL2_EnSDERR | HCRX_EL2_EnSNERR);
1020 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, DF2, IMP))
1021 		res0 |= HCRX_EL2_TMEA;
1022 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, D128, IMP))
1023 		res0 |= HCRX_EL2_D128En;
1024 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, THE, IMP))
1025 		res0 |= HCRX_EL2_PTTWI;
1026 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, SCTLRX, IMP))
1027 		res0 |= HCRX_EL2_SCTLR2En;
1028 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
1029 		res0 |= HCRX_EL2_TCR2En;
1030 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, MOPS, IMP))
1031 		res0 |= (HCRX_EL2_MSCEn | HCRX_EL2_MCE2);
1032 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, CMOW, IMP))
1033 		res0 |= HCRX_EL2_CMOW;
1034 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, NMI, IMP))
1035 		res0 |= (HCRX_EL2_VFNMI | HCRX_EL2_VINMI | HCRX_EL2_TALLINT);
1036 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, SME, IMP) ||
1037 	    !(read_sysreg_s(SYS_SMIDR_EL1) & SMIDR_EL1_SMPS))
1038 		res0 |= HCRX_EL2_SMPME;
1039 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
1040 		res0 |= (HCRX_EL2_FGTnXS | HCRX_EL2_FnXS);
1041 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_V))
1042 		res0 |= HCRX_EL2_EnASR;
1043 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64))
1044 		res0 |= HCRX_EL2_EnALS;
1045 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_ACCDATA))
1046 		res0 |= HCRX_EL2_EnAS0;
1047 	set_sysreg_masks(kvm, HCRX_EL2, res0, res1);
1048 
1049 	/* HFG[RW]TR_EL2 */
1050 	res0 = res1 = 0;
1051 	if (!(__vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_ADDRESS) &&
1052 	      __vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_GENERIC)))
1053 		res0 |= (HFGxTR_EL2_APDAKey | HFGxTR_EL2_APDBKey |
1054 			 HFGxTR_EL2_APGAKey | HFGxTR_EL2_APIAKey |
1055 			 HFGxTR_EL2_APIBKey);
1056 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, LO, IMP))
1057 		res0 |= (HFGxTR_EL2_LORC_EL1 | HFGxTR_EL2_LOREA_EL1 |
1058 			 HFGxTR_EL2_LORID_EL1 | HFGxTR_EL2_LORN_EL1 |
1059 			 HFGxTR_EL2_LORSA_EL1);
1060 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, CSV2, CSV2_2) &&
1061 	    !kvm_has_feat(kvm, ID_AA64PFR1_EL1, CSV2_frac, CSV2_1p2))
1062 		res0 |= (HFGxTR_EL2_SCXTNUM_EL1 | HFGxTR_EL2_SCXTNUM_EL0);
1063 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, GIC, IMP))
1064 		res0 |= HFGxTR_EL2_ICC_IGRPENn_EL1;
1065 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP))
1066 		res0 |= (HFGxTR_EL2_ERRIDR_EL1 | HFGxTR_EL2_ERRSELR_EL1 |
1067 			 HFGxTR_EL2_ERXFR_EL1 | HFGxTR_EL2_ERXCTLR_EL1 |
1068 			 HFGxTR_EL2_ERXSTATUS_EL1 | HFGxTR_EL2_ERXMISCn_EL1 |
1069 			 HFGxTR_EL2_ERXPFGF_EL1 | HFGxTR_EL2_ERXPFGCTL_EL1 |
1070 			 HFGxTR_EL2_ERXPFGCDN_EL1 | HFGxTR_EL2_ERXADDR_EL1);
1071 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_ACCDATA))
1072 		res0 |= HFGxTR_EL2_nACCDATA_EL1;
1073 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1074 		res0 |= (HFGxTR_EL2_nGCS_EL0 | HFGxTR_EL2_nGCS_EL1);
1075 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, SME, IMP))
1076 		res0 |= (HFGxTR_EL2_nSMPRI_EL1 | HFGxTR_EL2_nTPIDR2_EL0);
1077 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, THE, IMP))
1078 		res0 |= HFGxTR_EL2_nRCWMASK_EL1;
1079 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1PIE, IMP))
1080 		res0 |= (HFGxTR_EL2_nPIRE0_EL1 | HFGxTR_EL2_nPIR_EL1);
1081 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
1082 		res0 |= (HFGxTR_EL2_nPOR_EL0 | HFGxTR_EL2_nPOR_EL1);
1083 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S2POE, IMP))
1084 		res0 |= HFGxTR_EL2_nS2POR_EL1;
1085 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, AIE, IMP))
1086 		res0 |= (HFGxTR_EL2_nMAIR2_EL1 | HFGxTR_EL2_nAMAIR2_EL1);
1087 	set_sysreg_masks(kvm, HFGRTR_EL2, res0 | __HFGRTR_EL2_RES0, res1);
1088 	set_sysreg_masks(kvm, HFGWTR_EL2, res0 | __HFGWTR_EL2_RES0, res1);
1089 
1090 	/* HDFG[RW]TR_EL2 */
1091 	res0 = res1 = 0;
1092 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, DoubleLock, IMP))
1093 		res0 |= HDFGRTR_EL2_OSDLR_EL1;
1094 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
1095 		res0 |= (HDFGRTR_EL2_PMEVCNTRn_EL0 | HDFGRTR_EL2_PMEVTYPERn_EL0 |
1096 			 HDFGRTR_EL2_PMCCFILTR_EL0 | HDFGRTR_EL2_PMCCNTR_EL0 |
1097 			 HDFGRTR_EL2_PMCNTEN | HDFGRTR_EL2_PMINTEN |
1098 			 HDFGRTR_EL2_PMOVS | HDFGRTR_EL2_PMSELR_EL0 |
1099 			 HDFGRTR_EL2_PMMIR_EL1 | HDFGRTR_EL2_PMUSERENR_EL0 |
1100 			 HDFGRTR_EL2_PMCEIDn_EL0);
1101 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMSVer, IMP))
1102 		res0 |= (HDFGRTR_EL2_PMBLIMITR_EL1 | HDFGRTR_EL2_PMBPTR_EL1 |
1103 			 HDFGRTR_EL2_PMBSR_EL1 | HDFGRTR_EL2_PMSCR_EL1 |
1104 			 HDFGRTR_EL2_PMSEVFR_EL1 | HDFGRTR_EL2_PMSFCR_EL1 |
1105 			 HDFGRTR_EL2_PMSICR_EL1 | HDFGRTR_EL2_PMSIDR_EL1 |
1106 			 HDFGRTR_EL2_PMSIRR_EL1 | HDFGRTR_EL2_PMSLATFR_EL1 |
1107 			 HDFGRTR_EL2_PMBIDR_EL1);
1108 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceVer, IMP))
1109 		res0 |= (HDFGRTR_EL2_TRC | HDFGRTR_EL2_TRCAUTHSTATUS |
1110 			 HDFGRTR_EL2_TRCAUXCTLR | HDFGRTR_EL2_TRCCLAIM |
1111 			 HDFGRTR_EL2_TRCCNTVRn | HDFGRTR_EL2_TRCID |
1112 			 HDFGRTR_EL2_TRCIMSPECn | HDFGRTR_EL2_TRCOSLSR |
1113 			 HDFGRTR_EL2_TRCPRGCTLR | HDFGRTR_EL2_TRCSEQSTR |
1114 			 HDFGRTR_EL2_TRCSSCSRn | HDFGRTR_EL2_TRCSTATR |
1115 			 HDFGRTR_EL2_TRCVICTLR);
1116 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceBuffer, IMP))
1117 		res0 |= (HDFGRTR_EL2_TRBBASER_EL1 | HDFGRTR_EL2_TRBIDR_EL1 |
1118 			 HDFGRTR_EL2_TRBLIMITR_EL1 | HDFGRTR_EL2_TRBMAR_EL1 |
1119 			 HDFGRTR_EL2_TRBPTR_EL1 | HDFGRTR_EL2_TRBSR_EL1 |
1120 			 HDFGRTR_EL2_TRBTRG_EL1);
1121 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, BRBE, IMP))
1122 		res0 |= (HDFGRTR_EL2_nBRBIDR | HDFGRTR_EL2_nBRBCTL |
1123 			 HDFGRTR_EL2_nBRBDATA);
1124 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMSVer, V1P2))
1125 		res0 |= HDFGRTR_EL2_nPMSNEVFR_EL1;
1126 	set_sysreg_masks(kvm, HDFGRTR_EL2, res0 | HDFGRTR_EL2_RES0, res1);
1127 
1128 	/* Reuse the bits from the read-side and add the write-specific stuff */
1129 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
1130 		res0 |= (HDFGWTR_EL2_PMCR_EL0 | HDFGWTR_EL2_PMSWINC_EL0);
1131 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceVer, IMP))
1132 		res0 |= HDFGWTR_EL2_TRCOSLAR;
1133 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceFilt, IMP))
1134 		res0 |= HDFGWTR_EL2_TRFCR_EL1;
1135 	set_sysreg_masks(kvm, HFGWTR_EL2, res0 | HDFGWTR_EL2_RES0, res1);
1136 
1137 	/* HFGITR_EL2 */
1138 	res0 = HFGITR_EL2_RES0;
1139 	res1 = HFGITR_EL2_RES1;
1140 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, DPB, DPB2))
1141 		res0 |= HFGITR_EL2_DCCVADP;
1142 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN2))
1143 		res0 |= (HFGITR_EL2_ATS1E1RP | HFGITR_EL2_ATS1E1WP);
1144 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
1145 		res0 |= (HFGITR_EL2_TLBIRVAALE1OS | HFGITR_EL2_TLBIRVALE1OS |
1146 			 HFGITR_EL2_TLBIRVAAE1OS | HFGITR_EL2_TLBIRVAE1OS |
1147 			 HFGITR_EL2_TLBIVAALE1OS | HFGITR_EL2_TLBIVALE1OS |
1148 			 HFGITR_EL2_TLBIVAAE1OS | HFGITR_EL2_TLBIASIDE1OS |
1149 			 HFGITR_EL2_TLBIVAE1OS | HFGITR_EL2_TLBIVMALLE1OS);
1150 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
1151 		res0 |= (HFGITR_EL2_TLBIRVAALE1 | HFGITR_EL2_TLBIRVALE1 |
1152 			 HFGITR_EL2_TLBIRVAAE1 | HFGITR_EL2_TLBIRVAE1 |
1153 			 HFGITR_EL2_TLBIRVAALE1IS | HFGITR_EL2_TLBIRVALE1IS |
1154 			 HFGITR_EL2_TLBIRVAAE1IS | HFGITR_EL2_TLBIRVAE1IS |
1155 			 HFGITR_EL2_TLBIRVAALE1OS | HFGITR_EL2_TLBIRVALE1OS |
1156 			 HFGITR_EL2_TLBIRVAAE1OS | HFGITR_EL2_TLBIRVAE1OS);
1157 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, SPECRES, IMP))
1158 		res0 |= (HFGITR_EL2_CFPRCTX | HFGITR_EL2_DVPRCTX |
1159 			 HFGITR_EL2_CPPRCTX);
1160 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, BRBE, IMP))
1161 		res0 |= (HFGITR_EL2_nBRBINJ | HFGITR_EL2_nBRBIALL);
1162 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1163 		res0 |= (HFGITR_EL2_nGCSPUSHM_EL1 | HFGITR_EL2_nGCSSTR_EL1 |
1164 			 HFGITR_EL2_nGCSEPP);
1165 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, SPECRES, COSP_RCTX))
1166 		res0 |= HFGITR_EL2_COSPRCTX;
1167 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, ATS1A, IMP))
1168 		res0 |= HFGITR_EL2_ATS1E1A;
1169 	set_sysreg_masks(kvm, HFGITR_EL2, res0, res1);
1170 
1171 	/* HAFGRTR_EL2 - not a lot to see here */
1172 	res0 = HAFGRTR_EL2_RES0;
1173 	res1 = HAFGRTR_EL2_RES1;
1174 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, V1P1))
1175 		res0 |= ~(res0 | res1);
1176 	set_sysreg_masks(kvm, HAFGRTR_EL2, res0, res1);
1177 
1178 	/* SCTLR_EL1 */
1179 	res0 = SCTLR_EL1_RES0;
1180 	res1 = SCTLR_EL1_RES1;
1181 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN3))
1182 		res0 |= SCTLR_EL1_EPAN;
1183 	set_sysreg_masks(kvm, SCTLR_EL1, res0, res1);
1184 
1185 	return 0;
1186 }
1187