xref: /linux/arch/arm64/kvm/nested.c (revision 2a6b6c9a226279b4f6668450ddb21ae655558087)
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), kvm->arch.nested_mmus_size),
66 			size_mul(sizeof(*kvm->arch.nested_mmus), num_mmus),
67 			GFP_KERNEL_ACCOUNT | __GFP_ZERO);
68 	if (!tmp)
69 		return -ENOMEM;
70 
71 	/*
72 	 * If we went through a realocation, adjust the MMU back-pointers in
73 	 * the previously initialised kvm_pgtable structures.
74 	 */
75 	if (kvm->arch.nested_mmus != tmp)
76 		for (int i = 0; i < kvm->arch.nested_mmus_size; i++)
77 			tmp[i].pgt->mmu = &tmp[i];
78 
79 	for (int i = kvm->arch.nested_mmus_size; !ret && i < num_mmus; i++)
80 		ret = init_nested_s2_mmu(kvm, &tmp[i]);
81 
82 	if (ret) {
83 		for (int i = kvm->arch.nested_mmus_size; i < num_mmus; i++)
84 			kvm_free_stage2_pgd(&tmp[i]);
85 
86 		return ret;
87 	}
88 
89 	kvm->arch.nested_mmus_size = num_mmus;
90 	kvm->arch.nested_mmus = tmp;
91 
92 	return 0;
93 }
94 
95 struct s2_walk_info {
96 	int	     (*read_desc)(phys_addr_t pa, u64 *desc, void *data);
97 	void	     *data;
98 	u64	     baddr;
99 	unsigned int max_oa_bits;
100 	unsigned int pgshift;
101 	unsigned int sl;
102 	unsigned int t0sz;
103 	bool	     be;
104 };
105 
106 static unsigned int ps_to_output_size(unsigned int ps)
107 {
108 	switch (ps) {
109 	case 0: return 32;
110 	case 1: return 36;
111 	case 2: return 40;
112 	case 3: return 42;
113 	case 4: return 44;
114 	case 5:
115 	default:
116 		return 48;
117 	}
118 }
119 
120 static u32 compute_fsc(int level, u32 fsc)
121 {
122 	return fsc | (level & 0x3);
123 }
124 
125 static int esr_s2_fault(struct kvm_vcpu *vcpu, int level, u32 fsc)
126 {
127 	u32 esr;
128 
129 	esr = kvm_vcpu_get_esr(vcpu) & ~ESR_ELx_FSC;
130 	esr |= compute_fsc(level, fsc);
131 	return esr;
132 }
133 
134 static int get_ia_size(struct s2_walk_info *wi)
135 {
136 	return 64 - wi->t0sz;
137 }
138 
139 static int check_base_s2_limits(struct s2_walk_info *wi,
140 				int level, int input_size, int stride)
141 {
142 	int start_size, ia_size;
143 
144 	ia_size = get_ia_size(wi);
145 
146 	/* Check translation limits */
147 	switch (BIT(wi->pgshift)) {
148 	case SZ_64K:
149 		if (level == 0 || (level == 1 && ia_size <= 42))
150 			return -EFAULT;
151 		break;
152 	case SZ_16K:
153 		if (level == 0 || (level == 1 && ia_size <= 40))
154 			return -EFAULT;
155 		break;
156 	case SZ_4K:
157 		if (level < 0 || (level == 0 && ia_size <= 42))
158 			return -EFAULT;
159 		break;
160 	}
161 
162 	/* Check input size limits */
163 	if (input_size > ia_size)
164 		return -EFAULT;
165 
166 	/* Check number of entries in starting level table */
167 	start_size = input_size - ((3 - level) * stride + wi->pgshift);
168 	if (start_size < 1 || start_size > stride + 4)
169 		return -EFAULT;
170 
171 	return 0;
172 }
173 
174 /* Check if output is within boundaries */
175 static int check_output_size(struct s2_walk_info *wi, phys_addr_t output)
176 {
177 	unsigned int output_size = wi->max_oa_bits;
178 
179 	if (output_size != 48 && (output & GENMASK_ULL(47, output_size)))
180 		return -1;
181 
182 	return 0;
183 }
184 
185 /*
186  * This is essentially a C-version of the pseudo code from the ARM ARM
187  * AArch64.TranslationTableWalk  function.  I strongly recommend looking at
188  * that pseudocode in trying to understand this.
189  *
190  * Must be called with the kvm->srcu read lock held
191  */
192 static int walk_nested_s2_pgd(phys_addr_t ipa,
193 			      struct s2_walk_info *wi, struct kvm_s2_trans *out)
194 {
195 	int first_block_level, level, stride, input_size, base_lower_bound;
196 	phys_addr_t base_addr;
197 	unsigned int addr_top, addr_bottom;
198 	u64 desc;  /* page table entry */
199 	int ret;
200 	phys_addr_t paddr;
201 
202 	switch (BIT(wi->pgshift)) {
203 	default:
204 	case SZ_64K:
205 	case SZ_16K:
206 		level = 3 - wi->sl;
207 		first_block_level = 2;
208 		break;
209 	case SZ_4K:
210 		level = 2 - wi->sl;
211 		first_block_level = 1;
212 		break;
213 	}
214 
215 	stride = wi->pgshift - 3;
216 	input_size = get_ia_size(wi);
217 	if (input_size > 48 || input_size < 25)
218 		return -EFAULT;
219 
220 	ret = check_base_s2_limits(wi, level, input_size, stride);
221 	if (WARN_ON(ret))
222 		return ret;
223 
224 	base_lower_bound = 3 + input_size - ((3 - level) * stride +
225 			   wi->pgshift);
226 	base_addr = wi->baddr & GENMASK_ULL(47, base_lower_bound);
227 
228 	if (check_output_size(wi, base_addr)) {
229 		out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
230 		return 1;
231 	}
232 
233 	addr_top = input_size - 1;
234 
235 	while (1) {
236 		phys_addr_t index;
237 
238 		addr_bottom = (3 - level) * stride + wi->pgshift;
239 		index = (ipa & GENMASK_ULL(addr_top, addr_bottom))
240 			>> (addr_bottom - 3);
241 
242 		paddr = base_addr | index;
243 		ret = wi->read_desc(paddr, &desc, wi->data);
244 		if (ret < 0)
245 			return ret;
246 
247 		/*
248 		 * Handle reversedescriptors if endianness differs between the
249 		 * host and the guest hypervisor.
250 		 */
251 		if (wi->be)
252 			desc = be64_to_cpu((__force __be64)desc);
253 		else
254 			desc = le64_to_cpu((__force __le64)desc);
255 
256 		/* Check for valid descriptor at this point */
257 		if (!(desc & 1) || ((desc & 3) == 1 && level == 3)) {
258 			out->esr = compute_fsc(level, ESR_ELx_FSC_FAULT);
259 			out->upper_attr = desc;
260 			return 1;
261 		}
262 
263 		/* We're at the final level or block translation level */
264 		if ((desc & 3) == 1 || level == 3)
265 			break;
266 
267 		if (check_output_size(wi, desc)) {
268 			out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
269 			out->upper_attr = desc;
270 			return 1;
271 		}
272 
273 		base_addr = desc & GENMASK_ULL(47, wi->pgshift);
274 
275 		level += 1;
276 		addr_top = addr_bottom - 1;
277 	}
278 
279 	if (level < first_block_level) {
280 		out->esr = compute_fsc(level, ESR_ELx_FSC_FAULT);
281 		out->upper_attr = desc;
282 		return 1;
283 	}
284 
285 	/*
286 	 * We don't use the contiguous bit in the stage-2 ptes, so skip check
287 	 * for misprogramming of the contiguous bit.
288 	 */
289 
290 	if (check_output_size(wi, desc)) {
291 		out->esr = compute_fsc(level, ESR_ELx_FSC_ADDRSZ);
292 		out->upper_attr = desc;
293 		return 1;
294 	}
295 
296 	if (!(desc & BIT(10))) {
297 		out->esr = compute_fsc(level, ESR_ELx_FSC_ACCESS);
298 		out->upper_attr = desc;
299 		return 1;
300 	}
301 
302 	/* Calculate and return the result */
303 	paddr = (desc & GENMASK_ULL(47, addr_bottom)) |
304 		(ipa & GENMASK_ULL(addr_bottom - 1, 0));
305 	out->output = paddr;
306 	out->block_size = 1UL << ((3 - level) * stride + wi->pgshift);
307 	out->readable = desc & (0b01 << 6);
308 	out->writable = desc & (0b10 << 6);
309 	out->level = level;
310 	out->upper_attr = desc & GENMASK_ULL(63, 52);
311 	return 0;
312 }
313 
314 static int read_guest_s2_desc(phys_addr_t pa, u64 *desc, void *data)
315 {
316 	struct kvm_vcpu *vcpu = data;
317 
318 	return kvm_read_guest(vcpu->kvm, pa, desc, sizeof(*desc));
319 }
320 
321 static void vtcr_to_walk_info(u64 vtcr, struct s2_walk_info *wi)
322 {
323 	wi->t0sz = vtcr & TCR_EL2_T0SZ_MASK;
324 
325 	switch (vtcr & VTCR_EL2_TG0_MASK) {
326 	case VTCR_EL2_TG0_4K:
327 		wi->pgshift = 12;	 break;
328 	case VTCR_EL2_TG0_16K:
329 		wi->pgshift = 14;	 break;
330 	case VTCR_EL2_TG0_64K:
331 	default:	    /* IMPDEF: treat any other value as 64k */
332 		wi->pgshift = 16;	 break;
333 	}
334 
335 	wi->sl = FIELD_GET(VTCR_EL2_SL0_MASK, vtcr);
336 	/* Global limit for now, should eventually be per-VM */
337 	wi->max_oa_bits = min(get_kvm_ipa_limit(),
338 			      ps_to_output_size(FIELD_GET(VTCR_EL2_PS_MASK, vtcr)));
339 }
340 
341 int kvm_walk_nested_s2(struct kvm_vcpu *vcpu, phys_addr_t gipa,
342 		       struct kvm_s2_trans *result)
343 {
344 	u64 vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
345 	struct s2_walk_info wi;
346 	int ret;
347 
348 	result->esr = 0;
349 
350 	if (!vcpu_has_nv(vcpu))
351 		return 0;
352 
353 	wi.read_desc = read_guest_s2_desc;
354 	wi.data = vcpu;
355 	wi.baddr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
356 
357 	vtcr_to_walk_info(vtcr, &wi);
358 
359 	wi.be = vcpu_read_sys_reg(vcpu, SCTLR_EL2) & SCTLR_ELx_EE;
360 
361 	ret = walk_nested_s2_pgd(gipa, &wi, result);
362 	if (ret)
363 		result->esr |= (kvm_vcpu_get_esr(vcpu) & ~ESR_ELx_FSC);
364 
365 	return ret;
366 }
367 
368 static unsigned int ttl_to_size(u8 ttl)
369 {
370 	int level = ttl & 3;
371 	int gran = (ttl >> 2) & 3;
372 	unsigned int max_size = 0;
373 
374 	switch (gran) {
375 	case TLBI_TTL_TG_4K:
376 		switch (level) {
377 		case 0:
378 			break;
379 		case 1:
380 			max_size = SZ_1G;
381 			break;
382 		case 2:
383 			max_size = SZ_2M;
384 			break;
385 		case 3:
386 			max_size = SZ_4K;
387 			break;
388 		}
389 		break;
390 	case TLBI_TTL_TG_16K:
391 		switch (level) {
392 		case 0:
393 		case 1:
394 			break;
395 		case 2:
396 			max_size = SZ_32M;
397 			break;
398 		case 3:
399 			max_size = SZ_16K;
400 			break;
401 		}
402 		break;
403 	case TLBI_TTL_TG_64K:
404 		switch (level) {
405 		case 0:
406 		case 1:
407 			/* No 52bit IPA support */
408 			break;
409 		case 2:
410 			max_size = SZ_512M;
411 			break;
412 		case 3:
413 			max_size = SZ_64K;
414 			break;
415 		}
416 		break;
417 	default:			/* No size information */
418 		break;
419 	}
420 
421 	return max_size;
422 }
423 
424 /*
425  * Compute the equivalent of the TTL field by parsing the shadow PT.  The
426  * granule size is extracted from the cached VTCR_EL2.TG0 while the level is
427  * retrieved from first entry carrying the level as a tag.
428  */
429 static u8 get_guest_mapping_ttl(struct kvm_s2_mmu *mmu, u64 addr)
430 {
431 	u64 tmp, sz = 0, vtcr = mmu->tlb_vtcr;
432 	kvm_pte_t pte;
433 	u8 ttl, level;
434 
435 	lockdep_assert_held_write(&kvm_s2_mmu_to_kvm(mmu)->mmu_lock);
436 
437 	switch (vtcr & VTCR_EL2_TG0_MASK) {
438 	case VTCR_EL2_TG0_4K:
439 		ttl = (TLBI_TTL_TG_4K << 2);
440 		break;
441 	case VTCR_EL2_TG0_16K:
442 		ttl = (TLBI_TTL_TG_16K << 2);
443 		break;
444 	case VTCR_EL2_TG0_64K:
445 	default:	    /* IMPDEF: treat any other value as 64k */
446 		ttl = (TLBI_TTL_TG_64K << 2);
447 		break;
448 	}
449 
450 	tmp = addr;
451 
452 again:
453 	/* Iteratively compute the block sizes for a particular granule size */
454 	switch (vtcr & VTCR_EL2_TG0_MASK) {
455 	case VTCR_EL2_TG0_4K:
456 		if	(sz < SZ_4K)	sz = SZ_4K;
457 		else if (sz < SZ_2M)	sz = SZ_2M;
458 		else if (sz < SZ_1G)	sz = SZ_1G;
459 		else			sz = 0;
460 		break;
461 	case VTCR_EL2_TG0_16K:
462 		if	(sz < SZ_16K)	sz = SZ_16K;
463 		else if (sz < SZ_32M)	sz = SZ_32M;
464 		else			sz = 0;
465 		break;
466 	case VTCR_EL2_TG0_64K:
467 	default:	    /* IMPDEF: treat any other value as 64k */
468 		if	(sz < SZ_64K)	sz = SZ_64K;
469 		else if (sz < SZ_512M)	sz = SZ_512M;
470 		else			sz = 0;
471 		break;
472 	}
473 
474 	if (sz == 0)
475 		return 0;
476 
477 	tmp &= ~(sz - 1);
478 	if (kvm_pgtable_get_leaf(mmu->pgt, tmp, &pte, NULL))
479 		goto again;
480 	if (!(pte & PTE_VALID))
481 		goto again;
482 	level = FIELD_GET(KVM_NV_GUEST_MAP_SZ, pte);
483 	if (!level)
484 		goto again;
485 
486 	ttl |= level;
487 
488 	/*
489 	 * We now have found some level information in the shadow S2. Check
490 	 * that the resulting range is actually including the original IPA.
491 	 */
492 	sz = ttl_to_size(ttl);
493 	if (addr < (tmp + sz))
494 		return ttl;
495 
496 	return 0;
497 }
498 
499 unsigned long compute_tlb_inval_range(struct kvm_s2_mmu *mmu, u64 val)
500 {
501 	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
502 	unsigned long max_size;
503 	u8 ttl;
504 
505 	ttl = FIELD_GET(TLBI_TTL_MASK, val);
506 
507 	if (!ttl || !kvm_has_feat(kvm, ID_AA64MMFR2_EL1, TTL, IMP)) {
508 		/* No TTL, check the shadow S2 for a hint */
509 		u64 addr = (val & GENMASK_ULL(35, 0)) << 12;
510 		ttl = get_guest_mapping_ttl(mmu, addr);
511 	}
512 
513 	max_size = ttl_to_size(ttl);
514 
515 	if (!max_size) {
516 		/* Compute the maximum extent of the invalidation */
517 		switch (mmu->tlb_vtcr & VTCR_EL2_TG0_MASK) {
518 		case VTCR_EL2_TG0_4K:
519 			max_size = SZ_1G;
520 			break;
521 		case VTCR_EL2_TG0_16K:
522 			max_size = SZ_32M;
523 			break;
524 		case VTCR_EL2_TG0_64K:
525 		default:    /* IMPDEF: treat any other value as 64k */
526 			/*
527 			 * No, we do not support 52bit IPA in nested yet. Once
528 			 * we do, this should be 4TB.
529 			 */
530 			max_size = SZ_512M;
531 			break;
532 		}
533 	}
534 
535 	WARN_ON(!max_size);
536 	return max_size;
537 }
538 
539 /*
540  * We can have multiple *different* MMU contexts with the same VMID:
541  *
542  * - S2 being enabled or not, hence differing by the HCR_EL2.VM bit
543  *
544  * - Multiple vcpus using private S2s (huh huh...), hence differing by the
545  *   VBBTR_EL2.BADDR address
546  *
547  * - A combination of the above...
548  *
549  * We can always identify which MMU context to pick at run-time.  However,
550  * TLB invalidation involving a VMID must take action on all the TLBs using
551  * this particular VMID. This translates into applying the same invalidation
552  * operation to all the contexts that are using this VMID. Moar phun!
553  */
554 void kvm_s2_mmu_iterate_by_vmid(struct kvm *kvm, u16 vmid,
555 				const union tlbi_info *info,
556 				void (*tlbi_callback)(struct kvm_s2_mmu *,
557 						      const union tlbi_info *))
558 {
559 	write_lock(&kvm->mmu_lock);
560 
561 	for (int i = 0; i < kvm->arch.nested_mmus_size; i++) {
562 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
563 
564 		if (!kvm_s2_mmu_valid(mmu))
565 			continue;
566 
567 		if (vmid == get_vmid(mmu->tlb_vttbr))
568 			tlbi_callback(mmu, info);
569 	}
570 
571 	write_unlock(&kvm->mmu_lock);
572 }
573 
574 struct kvm_s2_mmu *lookup_s2_mmu(struct kvm_vcpu *vcpu)
575 {
576 	struct kvm *kvm = vcpu->kvm;
577 	bool nested_stage2_enabled;
578 	u64 vttbr, vtcr, hcr;
579 
580 	lockdep_assert_held_write(&kvm->mmu_lock);
581 
582 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
583 	vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
584 	hcr = vcpu_read_sys_reg(vcpu, HCR_EL2);
585 
586 	nested_stage2_enabled = hcr & HCR_VM;
587 
588 	/* Don't consider the CnP bit for the vttbr match */
589 	vttbr &= ~VTTBR_CNP_BIT;
590 
591 	/*
592 	 * Two possibilities when looking up a S2 MMU context:
593 	 *
594 	 * - either S2 is enabled in the guest, and we need a context that is
595 	 *   S2-enabled and matches the full VTTBR (VMID+BADDR) and VTCR,
596 	 *   which makes it safe from a TLB conflict perspective (a broken
597 	 *   guest won't be able to generate them),
598 	 *
599 	 * - or S2 is disabled, and we need a context that is S2-disabled
600 	 *   and matches the VMID only, as all TLBs are tagged by VMID even
601 	 *   if S2 translation is disabled.
602 	 */
603 	for (int i = 0; i < kvm->arch.nested_mmus_size; i++) {
604 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
605 
606 		if (!kvm_s2_mmu_valid(mmu))
607 			continue;
608 
609 		if (nested_stage2_enabled &&
610 		    mmu->nested_stage2_enabled &&
611 		    vttbr == mmu->tlb_vttbr &&
612 		    vtcr == mmu->tlb_vtcr)
613 			return mmu;
614 
615 		if (!nested_stage2_enabled &&
616 		    !mmu->nested_stage2_enabled &&
617 		    get_vmid(vttbr) == get_vmid(mmu->tlb_vttbr))
618 			return mmu;
619 	}
620 	return NULL;
621 }
622 
623 static struct kvm_s2_mmu *get_s2_mmu_nested(struct kvm_vcpu *vcpu)
624 {
625 	struct kvm *kvm = vcpu->kvm;
626 	struct kvm_s2_mmu *s2_mmu;
627 	int i;
628 
629 	lockdep_assert_held_write(&vcpu->kvm->mmu_lock);
630 
631 	s2_mmu = lookup_s2_mmu(vcpu);
632 	if (s2_mmu)
633 		goto out;
634 
635 	/*
636 	 * Make sure we don't always search from the same point, or we
637 	 * will always reuse a potentially active context, leaving
638 	 * free contexts unused.
639 	 */
640 	for (i = kvm->arch.nested_mmus_next;
641 	     i < (kvm->arch.nested_mmus_size + kvm->arch.nested_mmus_next);
642 	     i++) {
643 		s2_mmu = &kvm->arch.nested_mmus[i % kvm->arch.nested_mmus_size];
644 
645 		if (atomic_read(&s2_mmu->refcnt) == 0)
646 			break;
647 	}
648 	BUG_ON(atomic_read(&s2_mmu->refcnt)); /* We have struct MMUs to spare */
649 
650 	/* Set the scene for the next search */
651 	kvm->arch.nested_mmus_next = (i + 1) % kvm->arch.nested_mmus_size;
652 
653 	/* Clear the old state */
654 	if (kvm_s2_mmu_valid(s2_mmu))
655 		kvm_stage2_unmap_range(s2_mmu, 0, kvm_phys_size(s2_mmu));
656 
657 	/*
658 	 * The virtual VMID (modulo CnP) will be used as a key when matching
659 	 * an existing kvm_s2_mmu.
660 	 *
661 	 * We cache VTCR at allocation time, once and for all. It'd be great
662 	 * if the guest didn't screw that one up, as this is not very
663 	 * forgiving...
664 	 */
665 	s2_mmu->tlb_vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2) & ~VTTBR_CNP_BIT;
666 	s2_mmu->tlb_vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
667 	s2_mmu->nested_stage2_enabled = vcpu_read_sys_reg(vcpu, HCR_EL2) & HCR_VM;
668 
669 out:
670 	atomic_inc(&s2_mmu->refcnt);
671 	return s2_mmu;
672 }
673 
674 void kvm_init_nested_s2_mmu(struct kvm_s2_mmu *mmu)
675 {
676 	/* CnP being set denotes an invalid entry */
677 	mmu->tlb_vttbr = VTTBR_CNP_BIT;
678 	mmu->nested_stage2_enabled = false;
679 	atomic_set(&mmu->refcnt, 0);
680 }
681 
682 void kvm_vcpu_load_hw_mmu(struct kvm_vcpu *vcpu)
683 {
684 	if (is_hyp_ctxt(vcpu)) {
685 		vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
686 	} else {
687 		write_lock(&vcpu->kvm->mmu_lock);
688 		vcpu->arch.hw_mmu = get_s2_mmu_nested(vcpu);
689 		write_unlock(&vcpu->kvm->mmu_lock);
690 	}
691 }
692 
693 void kvm_vcpu_put_hw_mmu(struct kvm_vcpu *vcpu)
694 {
695 	if (kvm_is_nested_s2_mmu(vcpu->kvm, vcpu->arch.hw_mmu)) {
696 		atomic_dec(&vcpu->arch.hw_mmu->refcnt);
697 		vcpu->arch.hw_mmu = NULL;
698 	}
699 }
700 
701 /*
702  * Returns non-zero if permission fault is handled by injecting it to the next
703  * level hypervisor.
704  */
705 int kvm_s2_handle_perm_fault(struct kvm_vcpu *vcpu, struct kvm_s2_trans *trans)
706 {
707 	bool forward_fault = false;
708 
709 	trans->esr = 0;
710 
711 	if (!kvm_vcpu_trap_is_permission_fault(vcpu))
712 		return 0;
713 
714 	if (kvm_vcpu_trap_is_iabt(vcpu)) {
715 		forward_fault = !kvm_s2_trans_executable(trans);
716 	} else {
717 		bool write_fault = kvm_is_write_fault(vcpu);
718 
719 		forward_fault = ((write_fault && !trans->writable) ||
720 				 (!write_fault && !trans->readable));
721 	}
722 
723 	if (forward_fault)
724 		trans->esr = esr_s2_fault(vcpu, trans->level, ESR_ELx_FSC_PERM);
725 
726 	return forward_fault;
727 }
728 
729 int kvm_inject_s2_fault(struct kvm_vcpu *vcpu, u64 esr_el2)
730 {
731 	vcpu_write_sys_reg(vcpu, vcpu->arch.fault.far_el2, FAR_EL2);
732 	vcpu_write_sys_reg(vcpu, vcpu->arch.fault.hpfar_el2, HPFAR_EL2);
733 
734 	return kvm_inject_nested_sync(vcpu, esr_el2);
735 }
736 
737 void kvm_nested_s2_wp(struct kvm *kvm)
738 {
739 	int i;
740 
741 	lockdep_assert_held_write(&kvm->mmu_lock);
742 
743 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
744 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
745 
746 		if (kvm_s2_mmu_valid(mmu))
747 			kvm_stage2_wp_range(mmu, 0, kvm_phys_size(mmu));
748 	}
749 }
750 
751 void kvm_nested_s2_unmap(struct kvm *kvm)
752 {
753 	int i;
754 
755 	lockdep_assert_held_write(&kvm->mmu_lock);
756 
757 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
758 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
759 
760 		if (kvm_s2_mmu_valid(mmu))
761 			kvm_stage2_unmap_range(mmu, 0, kvm_phys_size(mmu));
762 	}
763 }
764 
765 void kvm_nested_s2_flush(struct kvm *kvm)
766 {
767 	int i;
768 
769 	lockdep_assert_held_write(&kvm->mmu_lock);
770 
771 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
772 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
773 
774 		if (kvm_s2_mmu_valid(mmu))
775 			kvm_stage2_flush_range(mmu, 0, kvm_phys_size(mmu));
776 	}
777 }
778 
779 void kvm_arch_flush_shadow_all(struct kvm *kvm)
780 {
781 	int i;
782 
783 	for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
784 		struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
785 
786 		if (!WARN_ON(atomic_read(&mmu->refcnt)))
787 			kvm_free_stage2_pgd(mmu);
788 	}
789 	kfree(kvm->arch.nested_mmus);
790 	kvm->arch.nested_mmus = NULL;
791 	kvm->arch.nested_mmus_size = 0;
792 	kvm_uninit_stage2_mmu(kvm);
793 }
794 
795 /*
796  * Our emulated CPU doesn't support all the possible features. For the
797  * sake of simplicity (and probably mental sanity), wipe out a number
798  * of feature bits we don't intend to support for the time being.
799  * This list should get updated as new features get added to the NV
800  * support, and new extension to the architecture.
801  */
802 static void limit_nv_id_regs(struct kvm *kvm)
803 {
804 	u64 val, tmp;
805 
806 	/* Support everything but TME */
807 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64ISAR0_EL1);
808 	val &= ~NV_FTR(ISAR0, TME);
809 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64ISAR0_EL1, val);
810 
811 	/* Support everything but Spec Invalidation and LS64 */
812 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64ISAR1_EL1);
813 	val &= ~(NV_FTR(ISAR1, LS64)	|
814 		 NV_FTR(ISAR1, SPECRES));
815 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64ISAR1_EL1, val);
816 
817 	/* No AMU, MPAM, S-EL2, or RAS */
818 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1);
819 	val &= ~(GENMASK_ULL(55, 52)	|
820 		 NV_FTR(PFR0, AMU)	|
821 		 NV_FTR(PFR0, MPAM)	|
822 		 NV_FTR(PFR0, SEL2)	|
823 		 NV_FTR(PFR0, RAS)	|
824 		 NV_FTR(PFR0, EL3)	|
825 		 NV_FTR(PFR0, EL2)	|
826 		 NV_FTR(PFR0, EL1));
827 	/* 64bit EL1/EL2/EL3 only */
828 	val |= FIELD_PREP(NV_FTR(PFR0, EL1), 0b0001);
829 	val |= FIELD_PREP(NV_FTR(PFR0, EL2), 0b0001);
830 	val |= FIELD_PREP(NV_FTR(PFR0, EL3), 0b0001);
831 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1, val);
832 
833 	/* Only support BTI, SSBS, CSV2_frac */
834 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR1_EL1);
835 	val &= (NV_FTR(PFR1, BT)	|
836 		NV_FTR(PFR1, SSBS)	|
837 		NV_FTR(PFR1, CSV2_frac));
838 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR1_EL1, val);
839 
840 	/* Hide ECV, ExS, Secure Memory */
841 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR0_EL1);
842 	val &= ~(NV_FTR(MMFR0, ECV)		|
843 		 NV_FTR(MMFR0, EXS)		|
844 		 NV_FTR(MMFR0, TGRAN4_2)	|
845 		 NV_FTR(MMFR0, TGRAN16_2)	|
846 		 NV_FTR(MMFR0, TGRAN64_2)	|
847 		 NV_FTR(MMFR0, SNSMEM));
848 
849 	/* Disallow unsupported S2 page sizes */
850 	switch (PAGE_SIZE) {
851 	case SZ_64K:
852 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN16_2), 0b0001);
853 		fallthrough;
854 	case SZ_16K:
855 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN4_2), 0b0001);
856 		fallthrough;
857 	case SZ_4K:
858 		/* Support everything */
859 		break;
860 	}
861 	/*
862 	 * Since we can't support a guest S2 page size smaller than
863 	 * the host's own page size (due to KVM only populating its
864 	 * own S2 using the kernel's page size), advertise the
865 	 * limitation using FEAT_GTG.
866 	 */
867 	switch (PAGE_SIZE) {
868 	case SZ_4K:
869 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN4_2), 0b0010);
870 		fallthrough;
871 	case SZ_16K:
872 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN16_2), 0b0010);
873 		fallthrough;
874 	case SZ_64K:
875 		val |= FIELD_PREP(NV_FTR(MMFR0, TGRAN64_2), 0b0010);
876 		break;
877 	}
878 	/* Cap PARange to 48bits */
879 	tmp = FIELD_GET(NV_FTR(MMFR0, PARANGE), val);
880 	if (tmp > 0b0101) {
881 		val &= ~NV_FTR(MMFR0, PARANGE);
882 		val |= FIELD_PREP(NV_FTR(MMFR0, PARANGE), 0b0101);
883 	}
884 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR0_EL1, val);
885 
886 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR1_EL1);
887 	val &= (NV_FTR(MMFR1, HCX)	|
888 		NV_FTR(MMFR1, PAN)	|
889 		NV_FTR(MMFR1, LO)	|
890 		NV_FTR(MMFR1, HPDS)	|
891 		NV_FTR(MMFR1, VH)	|
892 		NV_FTR(MMFR1, VMIDBits));
893 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR1_EL1, val);
894 
895 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64MMFR2_EL1);
896 	val &= ~(NV_FTR(MMFR2, BBM)	|
897 		 NV_FTR(MMFR2, TTL)	|
898 		 GENMASK_ULL(47, 44)	|
899 		 NV_FTR(MMFR2, ST)	|
900 		 NV_FTR(MMFR2, CCIDX)	|
901 		 NV_FTR(MMFR2, VARange));
902 
903 	/* Force TTL support */
904 	val |= FIELD_PREP(NV_FTR(MMFR2, TTL), 0b0001);
905 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR2_EL1, val);
906 
907 	val = 0;
908 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
909 		val |= FIELD_PREP(NV_FTR(MMFR4, E2H0),
910 				  ID_AA64MMFR4_EL1_E2H0_NI_NV1);
911 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64MMFR4_EL1, val);
912 
913 	/* Only limited support for PMU, Debug, BPs and WPs */
914 	val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64DFR0_EL1);
915 	val &= (NV_FTR(DFR0, PMUVer)	|
916 		NV_FTR(DFR0, WRPs)	|
917 		NV_FTR(DFR0, BRPs)	|
918 		NV_FTR(DFR0, DebugVer));
919 
920 	/* Cap Debug to ARMv8.1 */
921 	tmp = FIELD_GET(NV_FTR(DFR0, DebugVer), val);
922 	if (tmp > 0b0111) {
923 		val &= ~NV_FTR(DFR0, DebugVer);
924 		val |= FIELD_PREP(NV_FTR(DFR0, DebugVer), 0b0111);
925 	}
926 	kvm_set_vm_id_reg(kvm, SYS_ID_AA64DFR0_EL1, val);
927 }
928 
929 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg sr)
930 {
931 	u64 v = ctxt_sys_reg(&vcpu->arch.ctxt, sr);
932 	struct kvm_sysreg_masks *masks;
933 
934 	masks = vcpu->kvm->arch.sysreg_masks;
935 
936 	if (masks) {
937 		sr -= __VNCR_START__;
938 
939 		v &= ~masks->mask[sr].res0;
940 		v |= masks->mask[sr].res1;
941 	}
942 
943 	return v;
944 }
945 
946 static void set_sysreg_masks(struct kvm *kvm, int sr, u64 res0, u64 res1)
947 {
948 	int i = sr - __VNCR_START__;
949 
950 	kvm->arch.sysreg_masks->mask[i].res0 = res0;
951 	kvm->arch.sysreg_masks->mask[i].res1 = res1;
952 }
953 
954 int kvm_init_nv_sysregs(struct kvm *kvm)
955 {
956 	u64 res0, res1;
957 	int ret = 0;
958 
959 	mutex_lock(&kvm->arch.config_lock);
960 
961 	if (kvm->arch.sysreg_masks)
962 		goto out;
963 
964 	kvm->arch.sysreg_masks = kzalloc(sizeof(*(kvm->arch.sysreg_masks)),
965 					 GFP_KERNEL_ACCOUNT);
966 	if (!kvm->arch.sysreg_masks) {
967 		ret = -ENOMEM;
968 		goto out;
969 	}
970 
971 	limit_nv_id_regs(kvm);
972 
973 	/* VTTBR_EL2 */
974 	res0 = res1 = 0;
975 	if (!kvm_has_feat_enum(kvm, ID_AA64MMFR1_EL1, VMIDBits, 16))
976 		res0 |= GENMASK(63, 56);
977 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, CnP, IMP))
978 		res0 |= VTTBR_CNP_BIT;
979 	set_sysreg_masks(kvm, VTTBR_EL2, res0, res1);
980 
981 	/* VTCR_EL2 */
982 	res0 = GENMASK(63, 32) | GENMASK(30, 20);
983 	res1 = BIT(31);
984 	set_sysreg_masks(kvm, VTCR_EL2, res0, res1);
985 
986 	/* VMPIDR_EL2 */
987 	res0 = GENMASK(63, 40) | GENMASK(30, 24);
988 	res1 = BIT(31);
989 	set_sysreg_masks(kvm, VMPIDR_EL2, res0, res1);
990 
991 	/* HCR_EL2 */
992 	res0 = BIT(48);
993 	res1 = HCR_RW;
994 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, TWED, IMP))
995 		res0 |= GENMASK(63, 59);
996 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, MTE, MTE2))
997 		res0 |= (HCR_TID5 | HCR_DCT | HCR_ATA);
998 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, EVT, TTLBxS))
999 		res0 |= (HCR_TTLBIS | HCR_TTLBOS);
1000 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, CSV2, CSV2_2) &&
1001 	    !kvm_has_feat(kvm, ID_AA64PFR1_EL1, CSV2_frac, CSV2_1p2))
1002 		res0 |= HCR_ENSCXT;
1003 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, EVT, IMP))
1004 		res0 |= (HCR_TOCU | HCR_TICAB | HCR_TID4);
1005 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, V1P1))
1006 		res0 |= HCR_AMVOFFEN;
1007 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, V1P1))
1008 		res0 |= HCR_FIEN;
1009 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, FWB, IMP))
1010 		res0 |= HCR_FWB;
1011 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, NV, NV2))
1012 		res0 |= HCR_NV2;
1013 	if (!kvm_has_feat(kvm, ID_AA64MMFR2_EL1, NV, IMP))
1014 		res0 |= (HCR_AT | HCR_NV1 | HCR_NV);
1015 	if (!(__vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_ADDRESS) &&
1016 	      __vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_GENERIC)))
1017 		res0 |= (HCR_API | HCR_APK);
1018 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TME, IMP))
1019 		res0 |= BIT(39);
1020 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP))
1021 		res0 |= (HCR_TEA | HCR_TERR);
1022 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, LO, IMP))
1023 		res0 |= HCR_TLOR;
1024 	if (!kvm_has_feat(kvm, ID_AA64MMFR4_EL1, E2H0, IMP))
1025 		res1 |= HCR_E2H;
1026 	set_sysreg_masks(kvm, HCR_EL2, res0, res1);
1027 
1028 	/* HCRX_EL2 */
1029 	res0 = HCRX_EL2_RES0;
1030 	res1 = HCRX_EL2_RES1;
1031 	if (!kvm_has_feat(kvm, ID_AA64ISAR3_EL1, PACM, TRIVIAL_IMP))
1032 		res0 |= HCRX_EL2_PACMEn;
1033 	if (!kvm_has_feat(kvm, ID_AA64PFR2_EL1, FPMR, IMP))
1034 		res0 |= HCRX_EL2_EnFPM;
1035 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1036 		res0 |= HCRX_EL2_GCSEn;
1037 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, SYSREG_128, IMP))
1038 		res0 |= HCRX_EL2_EnIDCP128;
1039 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, ADERR, DEV_ASYNC))
1040 		res0 |= (HCRX_EL2_EnSDERR | HCRX_EL2_EnSNERR);
1041 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, DF2, IMP))
1042 		res0 |= HCRX_EL2_TMEA;
1043 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, D128, IMP))
1044 		res0 |= HCRX_EL2_D128En;
1045 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, THE, IMP))
1046 		res0 |= HCRX_EL2_PTTWI;
1047 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, SCTLRX, IMP))
1048 		res0 |= HCRX_EL2_SCTLR2En;
1049 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
1050 		res0 |= HCRX_EL2_TCR2En;
1051 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, MOPS, IMP))
1052 		res0 |= (HCRX_EL2_MSCEn | HCRX_EL2_MCE2);
1053 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, CMOW, IMP))
1054 		res0 |= HCRX_EL2_CMOW;
1055 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, NMI, IMP))
1056 		res0 |= (HCRX_EL2_VFNMI | HCRX_EL2_VINMI | HCRX_EL2_TALLINT);
1057 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, SME, IMP) ||
1058 	    !(read_sysreg_s(SYS_SMIDR_EL1) & SMIDR_EL1_SMPS))
1059 		res0 |= HCRX_EL2_SMPME;
1060 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
1061 		res0 |= (HCRX_EL2_FGTnXS | HCRX_EL2_FnXS);
1062 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_V))
1063 		res0 |= HCRX_EL2_EnASR;
1064 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64))
1065 		res0 |= HCRX_EL2_EnALS;
1066 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_ACCDATA))
1067 		res0 |= HCRX_EL2_EnAS0;
1068 	set_sysreg_masks(kvm, HCRX_EL2, res0, res1);
1069 
1070 	/* HFG[RW]TR_EL2 */
1071 	res0 = res1 = 0;
1072 	if (!(__vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_ADDRESS) &&
1073 	      __vcpu_has_feature(&kvm->arch, KVM_ARM_VCPU_PTRAUTH_GENERIC)))
1074 		res0 |= (HFGxTR_EL2_APDAKey | HFGxTR_EL2_APDBKey |
1075 			 HFGxTR_EL2_APGAKey | HFGxTR_EL2_APIAKey |
1076 			 HFGxTR_EL2_APIBKey);
1077 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, LO, IMP))
1078 		res0 |= (HFGxTR_EL2_LORC_EL1 | HFGxTR_EL2_LOREA_EL1 |
1079 			 HFGxTR_EL2_LORID_EL1 | HFGxTR_EL2_LORN_EL1 |
1080 			 HFGxTR_EL2_LORSA_EL1);
1081 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, CSV2, CSV2_2) &&
1082 	    !kvm_has_feat(kvm, ID_AA64PFR1_EL1, CSV2_frac, CSV2_1p2))
1083 		res0 |= (HFGxTR_EL2_SCXTNUM_EL1 | HFGxTR_EL2_SCXTNUM_EL0);
1084 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, GIC, IMP))
1085 		res0 |= HFGxTR_EL2_ICC_IGRPENn_EL1;
1086 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP))
1087 		res0 |= (HFGxTR_EL2_ERRIDR_EL1 | HFGxTR_EL2_ERRSELR_EL1 |
1088 			 HFGxTR_EL2_ERXFR_EL1 | HFGxTR_EL2_ERXCTLR_EL1 |
1089 			 HFGxTR_EL2_ERXSTATUS_EL1 | HFGxTR_EL2_ERXMISCn_EL1 |
1090 			 HFGxTR_EL2_ERXPFGF_EL1 | HFGxTR_EL2_ERXPFGCTL_EL1 |
1091 			 HFGxTR_EL2_ERXPFGCDN_EL1 | HFGxTR_EL2_ERXADDR_EL1);
1092 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, LS64, LS64_ACCDATA))
1093 		res0 |= HFGxTR_EL2_nACCDATA_EL1;
1094 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1095 		res0 |= (HFGxTR_EL2_nGCS_EL0 | HFGxTR_EL2_nGCS_EL1);
1096 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, SME, IMP))
1097 		res0 |= (HFGxTR_EL2_nSMPRI_EL1 | HFGxTR_EL2_nTPIDR2_EL0);
1098 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, THE, IMP))
1099 		res0 |= HFGxTR_EL2_nRCWMASK_EL1;
1100 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1PIE, IMP))
1101 		res0 |= (HFGxTR_EL2_nPIRE0_EL1 | HFGxTR_EL2_nPIR_EL1);
1102 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
1103 		res0 |= (HFGxTR_EL2_nPOR_EL0 | HFGxTR_EL2_nPOR_EL1);
1104 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S2POE, IMP))
1105 		res0 |= HFGxTR_EL2_nS2POR_EL1;
1106 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, AIE, IMP))
1107 		res0 |= (HFGxTR_EL2_nMAIR2_EL1 | HFGxTR_EL2_nAMAIR2_EL1);
1108 	set_sysreg_masks(kvm, HFGRTR_EL2, res0 | __HFGRTR_EL2_RES0, res1);
1109 	set_sysreg_masks(kvm, HFGWTR_EL2, res0 | __HFGWTR_EL2_RES0, res1);
1110 
1111 	/* HDFG[RW]TR_EL2 */
1112 	res0 = res1 = 0;
1113 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, DoubleLock, IMP))
1114 		res0 |= HDFGRTR_EL2_OSDLR_EL1;
1115 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
1116 		res0 |= (HDFGRTR_EL2_PMEVCNTRn_EL0 | HDFGRTR_EL2_PMEVTYPERn_EL0 |
1117 			 HDFGRTR_EL2_PMCCFILTR_EL0 | HDFGRTR_EL2_PMCCNTR_EL0 |
1118 			 HDFGRTR_EL2_PMCNTEN | HDFGRTR_EL2_PMINTEN |
1119 			 HDFGRTR_EL2_PMOVS | HDFGRTR_EL2_PMSELR_EL0 |
1120 			 HDFGRTR_EL2_PMMIR_EL1 | HDFGRTR_EL2_PMUSERENR_EL0 |
1121 			 HDFGRTR_EL2_PMCEIDn_EL0);
1122 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMSVer, IMP))
1123 		res0 |= (HDFGRTR_EL2_PMBLIMITR_EL1 | HDFGRTR_EL2_PMBPTR_EL1 |
1124 			 HDFGRTR_EL2_PMBSR_EL1 | HDFGRTR_EL2_PMSCR_EL1 |
1125 			 HDFGRTR_EL2_PMSEVFR_EL1 | HDFGRTR_EL2_PMSFCR_EL1 |
1126 			 HDFGRTR_EL2_PMSICR_EL1 | HDFGRTR_EL2_PMSIDR_EL1 |
1127 			 HDFGRTR_EL2_PMSIRR_EL1 | HDFGRTR_EL2_PMSLATFR_EL1 |
1128 			 HDFGRTR_EL2_PMBIDR_EL1);
1129 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceVer, IMP))
1130 		res0 |= (HDFGRTR_EL2_TRC | HDFGRTR_EL2_TRCAUTHSTATUS |
1131 			 HDFGRTR_EL2_TRCAUXCTLR | HDFGRTR_EL2_TRCCLAIM |
1132 			 HDFGRTR_EL2_TRCCNTVRn | HDFGRTR_EL2_TRCID |
1133 			 HDFGRTR_EL2_TRCIMSPECn | HDFGRTR_EL2_TRCOSLSR |
1134 			 HDFGRTR_EL2_TRCPRGCTLR | HDFGRTR_EL2_TRCSEQSTR |
1135 			 HDFGRTR_EL2_TRCSSCSRn | HDFGRTR_EL2_TRCSTATR |
1136 			 HDFGRTR_EL2_TRCVICTLR);
1137 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceBuffer, IMP))
1138 		res0 |= (HDFGRTR_EL2_TRBBASER_EL1 | HDFGRTR_EL2_TRBIDR_EL1 |
1139 			 HDFGRTR_EL2_TRBLIMITR_EL1 | HDFGRTR_EL2_TRBMAR_EL1 |
1140 			 HDFGRTR_EL2_TRBPTR_EL1 | HDFGRTR_EL2_TRBSR_EL1 |
1141 			 HDFGRTR_EL2_TRBTRG_EL1);
1142 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, BRBE, IMP))
1143 		res0 |= (HDFGRTR_EL2_nBRBIDR | HDFGRTR_EL2_nBRBCTL |
1144 			 HDFGRTR_EL2_nBRBDATA);
1145 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMSVer, V1P2))
1146 		res0 |= HDFGRTR_EL2_nPMSNEVFR_EL1;
1147 	set_sysreg_masks(kvm, HDFGRTR_EL2, res0 | HDFGRTR_EL2_RES0, res1);
1148 
1149 	/* Reuse the bits from the read-side and add the write-specific stuff */
1150 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
1151 		res0 |= (HDFGWTR_EL2_PMCR_EL0 | HDFGWTR_EL2_PMSWINC_EL0);
1152 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceVer, IMP))
1153 		res0 |= HDFGWTR_EL2_TRCOSLAR;
1154 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, TraceFilt, IMP))
1155 		res0 |= HDFGWTR_EL2_TRFCR_EL1;
1156 	set_sysreg_masks(kvm, HFGWTR_EL2, res0 | HDFGWTR_EL2_RES0, res1);
1157 
1158 	/* HFGITR_EL2 */
1159 	res0 = HFGITR_EL2_RES0;
1160 	res1 = HFGITR_EL2_RES1;
1161 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, DPB, DPB2))
1162 		res0 |= HFGITR_EL2_DCCVADP;
1163 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN2))
1164 		res0 |= (HFGITR_EL2_ATS1E1RP | HFGITR_EL2_ATS1E1WP);
1165 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
1166 		res0 |= (HFGITR_EL2_TLBIRVAALE1OS | HFGITR_EL2_TLBIRVALE1OS |
1167 			 HFGITR_EL2_TLBIRVAAE1OS | HFGITR_EL2_TLBIRVAE1OS |
1168 			 HFGITR_EL2_TLBIVAALE1OS | HFGITR_EL2_TLBIVALE1OS |
1169 			 HFGITR_EL2_TLBIVAAE1OS | HFGITR_EL2_TLBIASIDE1OS |
1170 			 HFGITR_EL2_TLBIVAE1OS | HFGITR_EL2_TLBIVMALLE1OS);
1171 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
1172 		res0 |= (HFGITR_EL2_TLBIRVAALE1 | HFGITR_EL2_TLBIRVALE1 |
1173 			 HFGITR_EL2_TLBIRVAAE1 | HFGITR_EL2_TLBIRVAE1 |
1174 			 HFGITR_EL2_TLBIRVAALE1IS | HFGITR_EL2_TLBIRVALE1IS |
1175 			 HFGITR_EL2_TLBIRVAAE1IS | HFGITR_EL2_TLBIRVAE1IS |
1176 			 HFGITR_EL2_TLBIRVAALE1OS | HFGITR_EL2_TLBIRVALE1OS |
1177 			 HFGITR_EL2_TLBIRVAAE1OS | HFGITR_EL2_TLBIRVAE1OS);
1178 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, SPECRES, IMP))
1179 		res0 |= (HFGITR_EL2_CFPRCTX | HFGITR_EL2_DVPRCTX |
1180 			 HFGITR_EL2_CPPRCTX);
1181 	if (!kvm_has_feat(kvm, ID_AA64DFR0_EL1, BRBE, IMP))
1182 		res0 |= (HFGITR_EL2_nBRBINJ | HFGITR_EL2_nBRBIALL);
1183 	if (!kvm_has_feat(kvm, ID_AA64PFR1_EL1, GCS, IMP))
1184 		res0 |= (HFGITR_EL2_nGCSPUSHM_EL1 | HFGITR_EL2_nGCSSTR_EL1 |
1185 			 HFGITR_EL2_nGCSEPP);
1186 	if (!kvm_has_feat(kvm, ID_AA64ISAR1_EL1, SPECRES, COSP_RCTX))
1187 		res0 |= HFGITR_EL2_COSPRCTX;
1188 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, ATS1A, IMP))
1189 		res0 |= HFGITR_EL2_ATS1E1A;
1190 	set_sysreg_masks(kvm, HFGITR_EL2, res0, res1);
1191 
1192 	/* HAFGRTR_EL2 - not a lot to see here */
1193 	res0 = HAFGRTR_EL2_RES0;
1194 	res1 = HAFGRTR_EL2_RES1;
1195 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, V1P1))
1196 		res0 |= ~(res0 | res1);
1197 	set_sysreg_masks(kvm, HAFGRTR_EL2, res0, res1);
1198 out:
1199 	mutex_unlock(&kvm->arch.config_lock);
1200 
1201 	return ret;
1202 }
1203