1 // SPDX-License-Identifier: GPL-2.0-only
3 * Stand-alone page-table allocator for hyp stage-1 and guest stage-2.
4 * No bombay mix was harmed in the writing of this file.
6 * Copyright (C) 2020 Google LLC
7 * Author: Will Deacon <will@kernel.org>
10 #include <linux/bitfield.h>
11 #include <asm/kvm_pgtable.h>
12 #include <asm/stage2_pgtable.h>
15 #define KVM_PTE_TYPE BIT(1)
16 #define KVM_PTE_TYPE_BLOCK 0
17 #define KVM_PTE_TYPE_PAGE 1
18 #define KVM_PTE_TYPE_TABLE 1
20 #define KVM_PTE_LEAF_ATTR_LO GENMASK(11, 2)
22 #define KVM_PTE_LEAF_ATTR_LO_S1_ATTRIDX GENMASK(4, 2)
23 #define KVM_PTE_LEAF_ATTR_LO_S1_AP GENMASK(7, 6)
24 #define KVM_PTE_LEAF_ATTR_LO_S1_AP_RO 3
25 #define KVM_PTE_LEAF_ATTR_LO_S1_AP_RW 1
26 #define KVM_PTE_LEAF_ATTR_LO_S1_SH GENMASK(9, 8)
27 #define KVM_PTE_LEAF_ATTR_LO_S1_SH_IS 3
28 #define KVM_PTE_LEAF_ATTR_LO_S1_AF BIT(10)
30 #define KVM_PTE_LEAF_ATTR_LO_S2_MEMATTR GENMASK(5, 2)
31 #define KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R BIT(6)
32 #define KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W BIT(7)
33 #define KVM_PTE_LEAF_ATTR_LO_S2_SH GENMASK(9, 8)
34 #define KVM_PTE_LEAF_ATTR_LO_S2_SH_IS 3
35 #define KVM_PTE_LEAF_ATTR_LO_S2_AF BIT(10)
37 #define KVM_PTE_LEAF_ATTR_HI GENMASK(63, 51)
39 #define KVM_PTE_LEAF_ATTR_HI_SW GENMASK(58, 55)
41 #define KVM_PTE_LEAF_ATTR_HI_S1_XN BIT(54)
43 #define KVM_PTE_LEAF_ATTR_HI_S2_XN BIT(54)
45 #define KVM_PTE_LEAF_ATTR_S2_PERMS (KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R | \
46 KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W | \
47 KVM_PTE_LEAF_ATTR_HI_S2_XN)
49 #define KVM_INVALID_PTE_OWNER_MASK GENMASK(9, 2)
50 #define KVM_MAX_OWNER_ID 1
52 struct kvm_pgtable_walk_data {
53 struct kvm_pgtable *pgt;
54 struct kvm_pgtable_walker *walker;
60 #define KVM_PHYS_INVALID (-1ULL)
62 static bool kvm_phys_is_valid(u64 phys)
64 return phys < BIT(id_aa64mmfr0_parange_to_phys_shift(ID_AA64MMFR0_PARANGE_MAX));
67 static bool kvm_block_mapping_supported(u64 addr, u64 end, u64 phys, u32 level)
69 u64 granule = kvm_granule_size(level);
71 if (!kvm_level_supports_block_mapping(level))
74 if (granule > (end - addr))
77 if (kvm_phys_is_valid(phys) && !IS_ALIGNED(phys, granule))
80 return IS_ALIGNED(addr, granule);
83 static u32 kvm_pgtable_idx(struct kvm_pgtable_walk_data *data, u32 level)
85 u64 shift = kvm_granule_shift(level);
86 u64 mask = BIT(PAGE_SHIFT - 3) - 1;
88 return (data->addr >> shift) & mask;
91 static u32 __kvm_pgd_page_idx(struct kvm_pgtable *pgt, u64 addr)
93 u64 shift = kvm_granule_shift(pgt->start_level - 1); /* May underflow */
94 u64 mask = BIT(pgt->ia_bits) - 1;
96 return (addr & mask) >> shift;
99 static u32 kvm_pgd_page_idx(struct kvm_pgtable_walk_data *data)
101 return __kvm_pgd_page_idx(data->pgt, data->addr);
104 static u32 kvm_pgd_pages(u32 ia_bits, u32 start_level)
106 struct kvm_pgtable pgt = {
108 .start_level = start_level,
111 return __kvm_pgd_page_idx(&pgt, -1ULL) + 1;
114 static bool kvm_pte_table(kvm_pte_t pte, u32 level)
116 if (level == KVM_PGTABLE_MAX_LEVELS - 1)
119 if (!kvm_pte_valid(pte))
122 return FIELD_GET(KVM_PTE_TYPE, pte) == KVM_PTE_TYPE_TABLE;
125 static kvm_pte_t kvm_phys_to_pte(u64 pa)
127 kvm_pte_t pte = pa & KVM_PTE_ADDR_MASK;
129 if (PAGE_SHIFT == 16)
130 pte |= FIELD_PREP(KVM_PTE_ADDR_51_48, pa >> 48);
135 static kvm_pte_t *kvm_pte_follow(kvm_pte_t pte, struct kvm_pgtable_mm_ops *mm_ops)
137 return mm_ops->phys_to_virt(kvm_pte_to_phys(pte));
140 static void kvm_clear_pte(kvm_pte_t *ptep)
142 WRITE_ONCE(*ptep, 0);
145 static void kvm_set_table_pte(kvm_pte_t *ptep, kvm_pte_t *childp,
146 struct kvm_pgtable_mm_ops *mm_ops)
148 kvm_pte_t old = *ptep, pte = kvm_phys_to_pte(mm_ops->virt_to_phys(childp));
150 pte |= FIELD_PREP(KVM_PTE_TYPE, KVM_PTE_TYPE_TABLE);
151 pte |= KVM_PTE_VALID;
153 WARN_ON(kvm_pte_valid(old));
154 smp_store_release(ptep, pte);
157 static kvm_pte_t kvm_init_valid_leaf_pte(u64 pa, kvm_pte_t attr, u32 level)
159 kvm_pte_t pte = kvm_phys_to_pte(pa);
160 u64 type = (level == KVM_PGTABLE_MAX_LEVELS - 1) ? KVM_PTE_TYPE_PAGE :
163 pte |= attr & (KVM_PTE_LEAF_ATTR_LO | KVM_PTE_LEAF_ATTR_HI);
164 pte |= FIELD_PREP(KVM_PTE_TYPE, type);
165 pte |= KVM_PTE_VALID;
170 static kvm_pte_t kvm_init_invalid_leaf_owner(u8 owner_id)
172 return FIELD_PREP(KVM_INVALID_PTE_OWNER_MASK, owner_id);
175 static int kvm_pgtable_visitor_cb(struct kvm_pgtable_walk_data *data, u64 addr,
176 u32 level, kvm_pte_t *ptep,
177 enum kvm_pgtable_walk_flags flag)
179 struct kvm_pgtable_walker *walker = data->walker;
180 return walker->cb(addr, data->end, level, ptep, flag, walker->arg);
183 static int __kvm_pgtable_walk(struct kvm_pgtable_walk_data *data,
184 kvm_pte_t *pgtable, u32 level);
186 static inline int __kvm_pgtable_visit(struct kvm_pgtable_walk_data *data,
187 kvm_pte_t *ptep, u32 level)
190 u64 addr = data->addr;
191 kvm_pte_t *childp, pte = *ptep;
192 bool table = kvm_pte_table(pte, level);
193 enum kvm_pgtable_walk_flags flags = data->walker->flags;
195 if (table && (flags & KVM_PGTABLE_WALK_TABLE_PRE)) {
196 ret = kvm_pgtable_visitor_cb(data, addr, level, ptep,
197 KVM_PGTABLE_WALK_TABLE_PRE);
200 if (!table && (flags & KVM_PGTABLE_WALK_LEAF)) {
201 ret = kvm_pgtable_visitor_cb(data, addr, level, ptep,
202 KVM_PGTABLE_WALK_LEAF);
204 table = kvm_pte_table(pte, level);
211 data->addr = ALIGN_DOWN(data->addr, kvm_granule_size(level));
212 data->addr += kvm_granule_size(level);
216 childp = kvm_pte_follow(pte, data->pgt->mm_ops);
217 ret = __kvm_pgtable_walk(data, childp, level + 1);
221 if (flags & KVM_PGTABLE_WALK_TABLE_POST) {
222 ret = kvm_pgtable_visitor_cb(data, addr, level, ptep,
223 KVM_PGTABLE_WALK_TABLE_POST);
230 static int __kvm_pgtable_walk(struct kvm_pgtable_walk_data *data,
231 kvm_pte_t *pgtable, u32 level)
236 if (WARN_ON_ONCE(level >= KVM_PGTABLE_MAX_LEVELS))
239 for (idx = kvm_pgtable_idx(data, level); idx < PTRS_PER_PTE; ++idx) {
240 kvm_pte_t *ptep = &pgtable[idx];
242 if (data->addr >= data->end)
245 ret = __kvm_pgtable_visit(data, ptep, level);
253 static int _kvm_pgtable_walk(struct kvm_pgtable_walk_data *data)
257 struct kvm_pgtable *pgt = data->pgt;
258 u64 limit = BIT(pgt->ia_bits);
260 if (data->addr > limit || data->end > limit)
266 for (idx = kvm_pgd_page_idx(data); data->addr < data->end; ++idx) {
267 kvm_pte_t *ptep = &pgt->pgd[idx * PTRS_PER_PTE];
269 ret = __kvm_pgtable_walk(data, ptep, pgt->start_level);
277 int kvm_pgtable_walk(struct kvm_pgtable *pgt, u64 addr, u64 size,
278 struct kvm_pgtable_walker *walker)
280 struct kvm_pgtable_walk_data walk_data = {
282 .addr = ALIGN_DOWN(addr, PAGE_SIZE),
283 .end = PAGE_ALIGN(walk_data.addr + size),
287 return _kvm_pgtable_walk(&walk_data);
290 struct leaf_walk_data {
295 static int leaf_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
296 enum kvm_pgtable_walk_flags flag, void * const arg)
298 struct leaf_walk_data *data = arg;
306 int kvm_pgtable_get_leaf(struct kvm_pgtable *pgt, u64 addr,
307 kvm_pte_t *ptep, u32 *level)
309 struct leaf_walk_data data;
310 struct kvm_pgtable_walker walker = {
312 .flags = KVM_PGTABLE_WALK_LEAF,
317 ret = kvm_pgtable_walk(pgt, ALIGN_DOWN(addr, PAGE_SIZE),
329 struct hyp_map_data {
332 struct kvm_pgtable_mm_ops *mm_ops;
335 static int hyp_set_prot_attr(enum kvm_pgtable_prot prot, kvm_pte_t *ptep)
337 bool device = prot & KVM_PGTABLE_PROT_DEVICE;
338 u32 mtype = device ? MT_DEVICE_nGnRE : MT_NORMAL;
339 kvm_pte_t attr = FIELD_PREP(KVM_PTE_LEAF_ATTR_LO_S1_ATTRIDX, mtype);
340 u32 sh = KVM_PTE_LEAF_ATTR_LO_S1_SH_IS;
341 u32 ap = (prot & KVM_PGTABLE_PROT_W) ? KVM_PTE_LEAF_ATTR_LO_S1_AP_RW :
342 KVM_PTE_LEAF_ATTR_LO_S1_AP_RO;
344 if (!(prot & KVM_PGTABLE_PROT_R))
347 if (prot & KVM_PGTABLE_PROT_X) {
348 if (prot & KVM_PGTABLE_PROT_W)
354 attr |= KVM_PTE_LEAF_ATTR_HI_S1_XN;
357 attr |= FIELD_PREP(KVM_PTE_LEAF_ATTR_LO_S1_AP, ap);
358 attr |= FIELD_PREP(KVM_PTE_LEAF_ATTR_LO_S1_SH, sh);
359 attr |= KVM_PTE_LEAF_ATTR_LO_S1_AF;
360 attr |= prot & KVM_PTE_LEAF_ATTR_HI_SW;
366 enum kvm_pgtable_prot kvm_pgtable_hyp_pte_prot(kvm_pte_t pte)
368 enum kvm_pgtable_prot prot = pte & KVM_PTE_LEAF_ATTR_HI_SW;
371 if (!kvm_pte_valid(pte))
374 if (!(pte & KVM_PTE_LEAF_ATTR_HI_S1_XN))
375 prot |= KVM_PGTABLE_PROT_X;
377 ap = FIELD_GET(KVM_PTE_LEAF_ATTR_LO_S1_AP, pte);
378 if (ap == KVM_PTE_LEAF_ATTR_LO_S1_AP_RO)
379 prot |= KVM_PGTABLE_PROT_R;
380 else if (ap == KVM_PTE_LEAF_ATTR_LO_S1_AP_RW)
381 prot |= KVM_PGTABLE_PROT_RW;
386 static bool hyp_map_walker_try_leaf(u64 addr, u64 end, u32 level,
387 kvm_pte_t *ptep, struct hyp_map_data *data)
389 kvm_pte_t new, old = *ptep;
390 u64 granule = kvm_granule_size(level), phys = data->phys;
392 if (!kvm_block_mapping_supported(addr, end, phys, level))
395 data->phys += granule;
396 new = kvm_init_valid_leaf_pte(phys, data->attr, level);
399 if (!kvm_pte_valid(old))
400 data->mm_ops->get_page(ptep);
401 else if (WARN_ON((old ^ new) & ~KVM_PTE_LEAF_ATTR_HI_SW))
404 smp_store_release(ptep, new);
408 static int hyp_map_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
409 enum kvm_pgtable_walk_flags flag, void * const arg)
412 struct hyp_map_data *data = arg;
413 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
415 if (hyp_map_walker_try_leaf(addr, end, level, ptep, arg))
418 if (WARN_ON(level == KVM_PGTABLE_MAX_LEVELS - 1))
421 childp = (kvm_pte_t *)mm_ops->zalloc_page(NULL);
425 kvm_set_table_pte(ptep, childp, mm_ops);
426 mm_ops->get_page(ptep);
430 int kvm_pgtable_hyp_map(struct kvm_pgtable *pgt, u64 addr, u64 size, u64 phys,
431 enum kvm_pgtable_prot prot)
434 struct hyp_map_data map_data = {
435 .phys = ALIGN_DOWN(phys, PAGE_SIZE),
436 .mm_ops = pgt->mm_ops,
438 struct kvm_pgtable_walker walker = {
439 .cb = hyp_map_walker,
440 .flags = KVM_PGTABLE_WALK_LEAF,
444 ret = hyp_set_prot_attr(prot, &map_data.attr);
448 ret = kvm_pgtable_walk(pgt, addr, size, &walker);
454 struct hyp_unmap_data {
456 struct kvm_pgtable_mm_ops *mm_ops;
459 static int hyp_unmap_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
460 enum kvm_pgtable_walk_flags flag, void * const arg)
462 kvm_pte_t pte = *ptep, *childp = NULL;
463 u64 granule = kvm_granule_size(level);
464 struct hyp_unmap_data *data = arg;
465 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
467 if (!kvm_pte_valid(pte))
470 if (kvm_pte_table(pte, level)) {
471 childp = kvm_pte_follow(pte, mm_ops);
473 if (mm_ops->page_count(childp) != 1)
478 __tlbi_level(vae2is, __TLBI_VADDR(addr, 0), level);
480 if (end - addr < granule)
485 __tlbi_level(vale2is, __TLBI_VADDR(addr, 0), level);
486 data->unmapped += granule;
491 mm_ops->put_page(ptep);
494 mm_ops->put_page(childp);
499 u64 kvm_pgtable_hyp_unmap(struct kvm_pgtable *pgt, u64 addr, u64 size)
501 struct hyp_unmap_data unmap_data = {
502 .mm_ops = pgt->mm_ops,
504 struct kvm_pgtable_walker walker = {
505 .cb = hyp_unmap_walker,
507 .flags = KVM_PGTABLE_WALK_LEAF | KVM_PGTABLE_WALK_TABLE_POST,
510 if (!pgt->mm_ops->page_count)
513 kvm_pgtable_walk(pgt, addr, size, &walker);
514 return unmap_data.unmapped;
517 int kvm_pgtable_hyp_init(struct kvm_pgtable *pgt, u32 va_bits,
518 struct kvm_pgtable_mm_ops *mm_ops)
520 u64 levels = ARM64_HW_PGTABLE_LEVELS(va_bits);
522 pgt->pgd = (kvm_pte_t *)mm_ops->zalloc_page(NULL);
526 pgt->ia_bits = va_bits;
527 pgt->start_level = KVM_PGTABLE_MAX_LEVELS - levels;
528 pgt->mm_ops = mm_ops;
530 pgt->force_pte_cb = NULL;
535 static int hyp_free_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
536 enum kvm_pgtable_walk_flags flag, void * const arg)
538 struct kvm_pgtable_mm_ops *mm_ops = arg;
539 kvm_pte_t pte = *ptep;
541 if (!kvm_pte_valid(pte))
544 mm_ops->put_page(ptep);
546 if (kvm_pte_table(pte, level))
547 mm_ops->put_page(kvm_pte_follow(pte, mm_ops));
552 void kvm_pgtable_hyp_destroy(struct kvm_pgtable *pgt)
554 struct kvm_pgtable_walker walker = {
555 .cb = hyp_free_walker,
556 .flags = KVM_PGTABLE_WALK_LEAF | KVM_PGTABLE_WALK_TABLE_POST,
560 WARN_ON(kvm_pgtable_walk(pgt, 0, BIT(pgt->ia_bits), &walker));
561 pgt->mm_ops->put_page(pgt->pgd);
565 struct stage2_map_data {
573 struct kvm_s2_mmu *mmu;
576 struct kvm_pgtable_mm_ops *mm_ops;
578 /* Force mappings to page granularity */
582 u64 kvm_get_vtcr(u64 mmfr0, u64 mmfr1, u32 phys_shift)
584 u64 vtcr = VTCR_EL2_FLAGS;
587 vtcr |= kvm_get_parange(mmfr0) << VTCR_EL2_PS_SHIFT;
588 vtcr |= VTCR_EL2_T0SZ(phys_shift);
590 * Use a minimum 2 level page table to prevent splitting
591 * host PMD huge pages at stage2.
593 lvls = stage2_pgtable_levels(phys_shift);
596 vtcr |= VTCR_EL2_LVLS_TO_SL0(lvls);
599 * Enable the Hardware Access Flag management, unconditionally
600 * on all CPUs. The features is RES0 on CPUs without the support
601 * and must be ignored by the CPUs.
605 /* Set the vmid bits */
606 vtcr |= (get_vmid_bits(mmfr1) == 16) ?
613 static bool stage2_has_fwb(struct kvm_pgtable *pgt)
615 if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
618 return !(pgt->flags & KVM_PGTABLE_S2_NOFWB);
621 #define KVM_S2_MEMATTR(pgt, attr) PAGE_S2_MEMATTR(attr, stage2_has_fwb(pgt))
623 static int stage2_set_prot_attr(struct kvm_pgtable *pgt, enum kvm_pgtable_prot prot,
626 bool device = prot & KVM_PGTABLE_PROT_DEVICE;
627 kvm_pte_t attr = device ? KVM_S2_MEMATTR(pgt, DEVICE_nGnRE) :
628 KVM_S2_MEMATTR(pgt, NORMAL);
629 u32 sh = KVM_PTE_LEAF_ATTR_LO_S2_SH_IS;
631 if (!(prot & KVM_PGTABLE_PROT_X))
632 attr |= KVM_PTE_LEAF_ATTR_HI_S2_XN;
636 if (prot & KVM_PGTABLE_PROT_R)
637 attr |= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R;
639 if (prot & KVM_PGTABLE_PROT_W)
640 attr |= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W;
642 attr |= FIELD_PREP(KVM_PTE_LEAF_ATTR_LO_S2_SH, sh);
643 attr |= KVM_PTE_LEAF_ATTR_LO_S2_AF;
644 attr |= prot & KVM_PTE_LEAF_ATTR_HI_SW;
650 enum kvm_pgtable_prot kvm_pgtable_stage2_pte_prot(kvm_pte_t pte)
652 enum kvm_pgtable_prot prot = pte & KVM_PTE_LEAF_ATTR_HI_SW;
654 if (!kvm_pte_valid(pte))
657 if (pte & KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R)
658 prot |= KVM_PGTABLE_PROT_R;
659 if (pte & KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W)
660 prot |= KVM_PGTABLE_PROT_W;
661 if (!(pte & KVM_PTE_LEAF_ATTR_HI_S2_XN))
662 prot |= KVM_PGTABLE_PROT_X;
667 static bool stage2_pte_needs_update(kvm_pte_t old, kvm_pte_t new)
669 if (!kvm_pte_valid(old) || !kvm_pte_valid(new))
672 return ((old ^ new) & (~KVM_PTE_LEAF_ATTR_S2_PERMS));
675 static bool stage2_pte_is_counted(kvm_pte_t pte)
678 * The refcount tracks valid entries as well as invalid entries if they
679 * encode ownership of a page to another entity than the page-table
680 * owner, whose id is 0.
685 static void stage2_put_pte(kvm_pte_t *ptep, struct kvm_s2_mmu *mmu, u64 addr,
686 u32 level, struct kvm_pgtable_mm_ops *mm_ops)
689 * Clear the existing PTE, and perform break-before-make with
690 * TLB maintenance if it was valid.
692 if (kvm_pte_valid(*ptep)) {
694 kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, mmu, addr, level);
697 mm_ops->put_page(ptep);
700 static bool stage2_pte_cacheable(struct kvm_pgtable *pgt, kvm_pte_t pte)
702 u64 memattr = pte & KVM_PTE_LEAF_ATTR_LO_S2_MEMATTR;
703 return memattr == KVM_S2_MEMATTR(pgt, NORMAL);
706 static bool stage2_pte_executable(kvm_pte_t pte)
708 return !(pte & KVM_PTE_LEAF_ATTR_HI_S2_XN);
711 static bool stage2_leaf_mapping_allowed(u64 addr, u64 end, u32 level,
712 struct stage2_map_data *data)
714 if (data->force_pte && (level < (KVM_PGTABLE_MAX_LEVELS - 1)))
717 return kvm_block_mapping_supported(addr, end, data->phys, level);
720 static int stage2_map_walker_try_leaf(u64 addr, u64 end, u32 level,
722 struct stage2_map_data *data)
724 kvm_pte_t new, old = *ptep;
725 u64 granule = kvm_granule_size(level), phys = data->phys;
726 struct kvm_pgtable *pgt = data->mmu->pgt;
727 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
729 if (!stage2_leaf_mapping_allowed(addr, end, level, data))
732 if (kvm_phys_is_valid(phys))
733 new = kvm_init_valid_leaf_pte(phys, data->attr, level);
735 new = kvm_init_invalid_leaf_owner(data->owner_id);
737 if (stage2_pte_is_counted(old)) {
739 * Skip updating the PTE if we are trying to recreate the exact
740 * same mapping or only change the access permissions. Instead,
741 * the vCPU will exit one more time from guest if still needed
742 * and then go through the path of relaxing permissions.
744 if (!stage2_pte_needs_update(old, new))
747 stage2_put_pte(ptep, data->mmu, addr, level, mm_ops);
750 /* Perform CMOs before installation of the guest stage-2 PTE */
751 if (mm_ops->dcache_clean_inval_poc && stage2_pte_cacheable(pgt, new))
752 mm_ops->dcache_clean_inval_poc(kvm_pte_follow(new, mm_ops),
755 if (mm_ops->icache_inval_pou && stage2_pte_executable(new))
756 mm_ops->icache_inval_pou(kvm_pte_follow(new, mm_ops), granule);
758 smp_store_release(ptep, new);
759 if (stage2_pte_is_counted(new))
760 mm_ops->get_page(ptep);
761 if (kvm_phys_is_valid(phys))
762 data->phys += granule;
766 static int stage2_map_walk_table_pre(u64 addr, u64 end, u32 level,
768 struct stage2_map_data *data)
773 if (!stage2_leaf_mapping_allowed(addr, end, level, data))
776 data->childp = kvm_pte_follow(*ptep, data->mm_ops);
780 * Invalidate the whole stage-2, as we may have numerous leaf
781 * entries below us which would otherwise need invalidating
784 kvm_call_hyp(__kvm_tlb_flush_vmid, data->mmu);
789 static int stage2_map_walk_leaf(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
790 struct stage2_map_data *data)
792 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
793 kvm_pte_t *childp, pte = *ptep;
797 if (stage2_pte_is_counted(pte))
798 mm_ops->put_page(ptep);
803 ret = stage2_map_walker_try_leaf(addr, end, level, ptep, data);
807 if (WARN_ON(level == KVM_PGTABLE_MAX_LEVELS - 1))
813 childp = mm_ops->zalloc_page(data->memcache);
818 * If we've run into an existing block mapping then replace it with
819 * a table. Accesses beyond 'end' that fall within the new table
820 * will be mapped lazily.
822 if (stage2_pte_is_counted(pte))
823 stage2_put_pte(ptep, data->mmu, addr, level, mm_ops);
825 kvm_set_table_pte(ptep, childp, mm_ops);
826 mm_ops->get_page(ptep);
831 static int stage2_map_walk_table_post(u64 addr, u64 end, u32 level,
833 struct stage2_map_data *data)
835 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
842 if (data->anchor == ptep) {
843 childp = data->childp;
846 ret = stage2_map_walk_leaf(addr, end, level, ptep, data);
848 childp = kvm_pte_follow(*ptep, mm_ops);
851 mm_ops->put_page(childp);
852 mm_ops->put_page(ptep);
858 * This is a little fiddly, as we use all three of the walk flags. The idea
859 * is that the TABLE_PRE callback runs for table entries on the way down,
860 * looking for table entries which we could conceivably replace with a
861 * block entry for this mapping. If it finds one, then it sets the 'anchor'
862 * field in 'struct stage2_map_data' to point at the table entry, before
863 * clearing the entry to zero and descending into the now detached table.
865 * The behaviour of the LEAF callback then depends on whether or not the
866 * anchor has been set. If not, then we're not using a block mapping higher
867 * up the table and we perform the mapping at the existing leaves instead.
868 * If, on the other hand, the anchor _is_ set, then we drop references to
869 * all valid leaves so that the pages beneath the anchor can be freed.
871 * Finally, the TABLE_POST callback does nothing if the anchor has not
872 * been set, but otherwise frees the page-table pages while walking back up
873 * the page-table, installing the block entry when it revisits the anchor
874 * pointer and clearing the anchor to NULL.
876 static int stage2_map_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
877 enum kvm_pgtable_walk_flags flag, void * const arg)
879 struct stage2_map_data *data = arg;
882 case KVM_PGTABLE_WALK_TABLE_PRE:
883 return stage2_map_walk_table_pre(addr, end, level, ptep, data);
884 case KVM_PGTABLE_WALK_LEAF:
885 return stage2_map_walk_leaf(addr, end, level, ptep, data);
886 case KVM_PGTABLE_WALK_TABLE_POST:
887 return stage2_map_walk_table_post(addr, end, level, ptep, data);
893 int kvm_pgtable_stage2_map(struct kvm_pgtable *pgt, u64 addr, u64 size,
894 u64 phys, enum kvm_pgtable_prot prot,
898 struct stage2_map_data map_data = {
899 .phys = ALIGN_DOWN(phys, PAGE_SIZE),
902 .mm_ops = pgt->mm_ops,
903 .force_pte = pgt->force_pte_cb && pgt->force_pte_cb(addr, addr + size, prot),
905 struct kvm_pgtable_walker walker = {
906 .cb = stage2_map_walker,
907 .flags = KVM_PGTABLE_WALK_TABLE_PRE |
908 KVM_PGTABLE_WALK_LEAF |
909 KVM_PGTABLE_WALK_TABLE_POST,
913 if (WARN_ON((pgt->flags & KVM_PGTABLE_S2_IDMAP) && (addr != phys)))
916 ret = stage2_set_prot_attr(pgt, prot, &map_data.attr);
920 ret = kvm_pgtable_walk(pgt, addr, size, &walker);
925 int kvm_pgtable_stage2_set_owner(struct kvm_pgtable *pgt, u64 addr, u64 size,
926 void *mc, u8 owner_id)
929 struct stage2_map_data map_data = {
930 .phys = KVM_PHYS_INVALID,
933 .mm_ops = pgt->mm_ops,
934 .owner_id = owner_id,
937 struct kvm_pgtable_walker walker = {
938 .cb = stage2_map_walker,
939 .flags = KVM_PGTABLE_WALK_TABLE_PRE |
940 KVM_PGTABLE_WALK_LEAF |
941 KVM_PGTABLE_WALK_TABLE_POST,
945 if (owner_id > KVM_MAX_OWNER_ID)
948 ret = kvm_pgtable_walk(pgt, addr, size, &walker);
952 static int stage2_unmap_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
953 enum kvm_pgtable_walk_flags flag,
956 struct kvm_pgtable *pgt = arg;
957 struct kvm_s2_mmu *mmu = pgt->mmu;
958 struct kvm_pgtable_mm_ops *mm_ops = pgt->mm_ops;
959 kvm_pte_t pte = *ptep, *childp = NULL;
960 bool need_flush = false;
962 if (!kvm_pte_valid(pte)) {
963 if (stage2_pte_is_counted(pte)) {
965 mm_ops->put_page(ptep);
970 if (kvm_pte_table(pte, level)) {
971 childp = kvm_pte_follow(pte, mm_ops);
973 if (mm_ops->page_count(childp) != 1)
975 } else if (stage2_pte_cacheable(pgt, pte)) {
976 need_flush = !stage2_has_fwb(pgt);
980 * This is similar to the map() path in that we unmap the entire
981 * block entry and rely on the remaining portions being faulted
984 stage2_put_pte(ptep, mmu, addr, level, mm_ops);
986 if (need_flush && mm_ops->dcache_clean_inval_poc)
987 mm_ops->dcache_clean_inval_poc(kvm_pte_follow(pte, mm_ops),
988 kvm_granule_size(level));
991 mm_ops->put_page(childp);
996 int kvm_pgtable_stage2_unmap(struct kvm_pgtable *pgt, u64 addr, u64 size)
998 struct kvm_pgtable_walker walker = {
999 .cb = stage2_unmap_walker,
1001 .flags = KVM_PGTABLE_WALK_LEAF | KVM_PGTABLE_WALK_TABLE_POST,
1004 return kvm_pgtable_walk(pgt, addr, size, &walker);
1007 struct stage2_attr_data {
1012 struct kvm_pgtable_mm_ops *mm_ops;
1015 static int stage2_attr_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
1016 enum kvm_pgtable_walk_flags flag,
1019 kvm_pte_t pte = *ptep;
1020 struct stage2_attr_data *data = arg;
1021 struct kvm_pgtable_mm_ops *mm_ops = data->mm_ops;
1023 if (!kvm_pte_valid(pte))
1026 data->level = level;
1028 pte &= ~data->attr_clr;
1029 pte |= data->attr_set;
1032 * We may race with the CPU trying to set the access flag here,
1033 * but worst-case the access flag update gets lost and will be
1034 * set on the next access instead.
1036 if (data->pte != pte) {
1038 * Invalidate instruction cache before updating the guest
1039 * stage-2 PTE if we are going to add executable permission.
1041 if (mm_ops->icache_inval_pou &&
1042 stage2_pte_executable(pte) && !stage2_pte_executable(*ptep))
1043 mm_ops->icache_inval_pou(kvm_pte_follow(pte, mm_ops),
1044 kvm_granule_size(level));
1045 WRITE_ONCE(*ptep, pte);
1051 static int stage2_update_leaf_attrs(struct kvm_pgtable *pgt, u64 addr,
1052 u64 size, kvm_pte_t attr_set,
1053 kvm_pte_t attr_clr, kvm_pte_t *orig_pte,
1057 kvm_pte_t attr_mask = KVM_PTE_LEAF_ATTR_LO | KVM_PTE_LEAF_ATTR_HI;
1058 struct stage2_attr_data data = {
1059 .attr_set = attr_set & attr_mask,
1060 .attr_clr = attr_clr & attr_mask,
1061 .mm_ops = pgt->mm_ops,
1063 struct kvm_pgtable_walker walker = {
1064 .cb = stage2_attr_walker,
1066 .flags = KVM_PGTABLE_WALK_LEAF,
1069 ret = kvm_pgtable_walk(pgt, addr, size, &walker);
1074 *orig_pte = data.pte;
1077 *level = data.level;
1081 int kvm_pgtable_stage2_wrprotect(struct kvm_pgtable *pgt, u64 addr, u64 size)
1083 return stage2_update_leaf_attrs(pgt, addr, size, 0,
1084 KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W,
1088 kvm_pte_t kvm_pgtable_stage2_mkyoung(struct kvm_pgtable *pgt, u64 addr)
1091 stage2_update_leaf_attrs(pgt, addr, 1, KVM_PTE_LEAF_ATTR_LO_S2_AF, 0,
1097 kvm_pte_t kvm_pgtable_stage2_mkold(struct kvm_pgtable *pgt, u64 addr)
1100 stage2_update_leaf_attrs(pgt, addr, 1, 0, KVM_PTE_LEAF_ATTR_LO_S2_AF,
1103 * "But where's the TLBI?!", you scream.
1104 * "Over in the core code", I sigh.
1106 * See the '->clear_flush_young()' callback on the KVM mmu notifier.
1111 bool kvm_pgtable_stage2_is_young(struct kvm_pgtable *pgt, u64 addr)
1114 stage2_update_leaf_attrs(pgt, addr, 1, 0, 0, &pte, NULL);
1115 return pte & KVM_PTE_LEAF_ATTR_LO_S2_AF;
1118 int kvm_pgtable_stage2_relax_perms(struct kvm_pgtable *pgt, u64 addr,
1119 enum kvm_pgtable_prot prot)
1123 kvm_pte_t set = 0, clr = 0;
1125 if (prot & KVM_PTE_LEAF_ATTR_HI_SW)
1128 if (prot & KVM_PGTABLE_PROT_R)
1129 set |= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R;
1131 if (prot & KVM_PGTABLE_PROT_W)
1132 set |= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W;
1134 if (prot & KVM_PGTABLE_PROT_X)
1135 clr |= KVM_PTE_LEAF_ATTR_HI_S2_XN;
1137 ret = stage2_update_leaf_attrs(pgt, addr, 1, set, clr, NULL, &level);
1139 kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, pgt->mmu, addr, level);
1143 static int stage2_flush_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
1144 enum kvm_pgtable_walk_flags flag,
1147 struct kvm_pgtable *pgt = arg;
1148 struct kvm_pgtable_mm_ops *mm_ops = pgt->mm_ops;
1149 kvm_pte_t pte = *ptep;
1151 if (!kvm_pte_valid(pte) || !stage2_pte_cacheable(pgt, pte))
1154 if (mm_ops->dcache_clean_inval_poc)
1155 mm_ops->dcache_clean_inval_poc(kvm_pte_follow(pte, mm_ops),
1156 kvm_granule_size(level));
1160 int kvm_pgtable_stage2_flush(struct kvm_pgtable *pgt, u64 addr, u64 size)
1162 struct kvm_pgtable_walker walker = {
1163 .cb = stage2_flush_walker,
1164 .flags = KVM_PGTABLE_WALK_LEAF,
1168 if (stage2_has_fwb(pgt))
1171 return kvm_pgtable_walk(pgt, addr, size, &walker);
1175 int __kvm_pgtable_stage2_init(struct kvm_pgtable *pgt, struct kvm_s2_mmu *mmu,
1176 struct kvm_pgtable_mm_ops *mm_ops,
1177 enum kvm_pgtable_stage2_flags flags,
1178 kvm_pgtable_force_pte_cb_t force_pte_cb)
1181 u64 vtcr = mmu->arch->vtcr;
1182 u32 ia_bits = VTCR_EL2_IPA(vtcr);
1183 u32 sl0 = FIELD_GET(VTCR_EL2_SL0_MASK, vtcr);
1184 u32 start_level = VTCR_EL2_TGRAN_SL0_BASE - sl0;
1186 pgd_sz = kvm_pgd_pages(ia_bits, start_level) * PAGE_SIZE;
1187 pgt->pgd = mm_ops->zalloc_pages_exact(pgd_sz);
1191 pgt->ia_bits = ia_bits;
1192 pgt->start_level = start_level;
1193 pgt->mm_ops = mm_ops;
1196 pgt->force_pte_cb = force_pte_cb;
1198 /* Ensure zeroed PGD pages are visible to the hardware walker */
1203 static int stage2_free_walker(u64 addr, u64 end, u32 level, kvm_pte_t *ptep,
1204 enum kvm_pgtable_walk_flags flag,
1207 struct kvm_pgtable_mm_ops *mm_ops = arg;
1208 kvm_pte_t pte = *ptep;
1210 if (!stage2_pte_is_counted(pte))
1213 mm_ops->put_page(ptep);
1215 if (kvm_pte_table(pte, level))
1216 mm_ops->put_page(kvm_pte_follow(pte, mm_ops));
1221 void kvm_pgtable_stage2_destroy(struct kvm_pgtable *pgt)
1224 struct kvm_pgtable_walker walker = {
1225 .cb = stage2_free_walker,
1226 .flags = KVM_PGTABLE_WALK_LEAF |
1227 KVM_PGTABLE_WALK_TABLE_POST,
1231 WARN_ON(kvm_pgtable_walk(pgt, 0, BIT(pgt->ia_bits), &walker));
1232 pgd_sz = kvm_pgd_pages(pgt->ia_bits, pgt->start_level) * PAGE_SIZE;
1233 pgt->mm_ops->free_pages_exact(pgt->pgd, pgd_sz);