1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * arch/arm/include/asm/pgtable-3level.h
4  *
5  * Copyright (C) 2011 ARM Ltd.
6  * Author: Catalin Marinas <catalin.marinas@arm.com>
7  */
8 #ifndef _ASM_PGTABLE_3LEVEL_H
9 #define _ASM_PGTABLE_3LEVEL_H
10 
11 /*
12  * With LPAE, there are 3 levels of page tables. Each level has 512 entries of
13  * 8 bytes each, occupying a 4K page. The first level table covers a range of
14  * 512GB, each entry representing 1GB. Since we are limited to 4GB input
15  * address range, only 4 entries in the PGD are used.
16  *
17  * There are enough spare bits in a page table entry for the kernel specific
18  * state.
19  */
20 #define PTRS_PER_PTE		512
21 #define PTRS_PER_PMD		512
22 #define PTRS_PER_PGD		4
23 
24 #define PTE_HWTABLE_PTRS	(0)
25 #define PTE_HWTABLE_OFF		(0)
26 #define PTE_HWTABLE_SIZE	(PTRS_PER_PTE * sizeof(u64))
27 
28 /*
29  * PGDIR_SHIFT determines the size a top-level page table entry can map.
30  */
31 #define PGDIR_SHIFT		30
32 
33 /*
34  * PMD_SHIFT determines the size a middle-level page table entry can map.
35  */
36 #define PMD_SHIFT		21
37 
38 #define PMD_SIZE		(1UL << PMD_SHIFT)
39 #define PMD_MASK		(~((1 << PMD_SHIFT) - 1))
40 #define PGDIR_SIZE		(1UL << PGDIR_SHIFT)
41 #define PGDIR_MASK		(~((1 << PGDIR_SHIFT) - 1))
42 
43 /*
44  * section address mask and size definitions.
45  */
46 #define SECTION_SHIFT		21
47 #define SECTION_SIZE		(1UL << SECTION_SHIFT)
48 #define SECTION_MASK		(~((1 << SECTION_SHIFT) - 1))
49 
50 #define USER_PTRS_PER_PGD	(PAGE_OFFSET / PGDIR_SIZE)
51 
52 /*
53  * Hugetlb definitions.
54  */
55 #define HPAGE_SHIFT		PMD_SHIFT
56 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
57 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
58 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
59 
60 /*
61  * "Linux" PTE definitions for LPAE.
62  *
63  * These bits overlap with the hardware bits but the naming is preserved for
64  * consistency with the classic page table format.
65  */
66 #define L_PTE_VALID		(_AT(pteval_t, 1) << 0)		/* Valid */
67 #define L_PTE_PRESENT		(_AT(pteval_t, 3) << 0)		/* Present */
68 #define L_PTE_USER		(_AT(pteval_t, 1) << 6)		/* AP[1] */
69 #define L_PTE_SHARED		(_AT(pteval_t, 3) << 8)		/* SH[1:0], inner shareable */
70 #define L_PTE_YOUNG		(_AT(pteval_t, 1) << 10)	/* AF */
71 #define L_PTE_XN		(_AT(pteval_t, 1) << 54)	/* XN */
72 #define L_PTE_DIRTY		(_AT(pteval_t, 1) << 55)
73 #define L_PTE_SPECIAL		(_AT(pteval_t, 1) << 56)
74 #define L_PTE_NONE		(_AT(pteval_t, 1) << 57)	/* PROT_NONE */
75 #define L_PTE_RDONLY		(_AT(pteval_t, 1) << 58)	/* READ ONLY */
76 
77 #define L_PMD_SECT_VALID	(_AT(pmdval_t, 1) << 0)
78 #define L_PMD_SECT_DIRTY	(_AT(pmdval_t, 1) << 55)
79 #define L_PMD_SECT_NONE		(_AT(pmdval_t, 1) << 57)
80 #define L_PMD_SECT_RDONLY	(_AT(pteval_t, 1) << 58)
81 
82 /*
83  * To be used in assembly code with the upper page attributes.
84  */
85 #define L_PTE_XN_HIGH		(1 << (54 - 32))
86 #define L_PTE_DIRTY_HIGH	(1 << (55 - 32))
87 
88 /*
89  * AttrIndx[2:0] encoding (mapping attributes defined in the MAIR* registers).
90  */
91 #define L_PTE_MT_UNCACHED	(_AT(pteval_t, 0) << 2)	/* strongly ordered */
92 #define L_PTE_MT_BUFFERABLE	(_AT(pteval_t, 1) << 2)	/* normal non-cacheable */
93 #define L_PTE_MT_WRITETHROUGH	(_AT(pteval_t, 2) << 2)	/* normal inner write-through */
94 #define L_PTE_MT_WRITEBACK	(_AT(pteval_t, 3) << 2)	/* normal inner write-back */
95 #define L_PTE_MT_WRITEALLOC	(_AT(pteval_t, 7) << 2)	/* normal inner write-alloc */
96 #define L_PTE_MT_DEV_SHARED	(_AT(pteval_t, 4) << 2)	/* device */
97 #define L_PTE_MT_DEV_NONSHARED	(_AT(pteval_t, 4) << 2)	/* device */
98 #define L_PTE_MT_DEV_WC		(_AT(pteval_t, 1) << 2)	/* normal non-cacheable */
99 #define L_PTE_MT_DEV_CACHED	(_AT(pteval_t, 3) << 2)	/* normal inner write-back */
100 #define L_PTE_MT_MASK		(_AT(pteval_t, 7) << 2)
101 
102 /*
103  * Software PGD flags.
104  */
105 #define L_PGD_SWAPPER		(_AT(pgdval_t, 1) << 55)	/* swapper_pg_dir entry */
106 
107 /*
108  * 2nd stage PTE definitions for LPAE.
109  */
110 #define L_PTE_S2_MT_UNCACHED		(_AT(pteval_t, 0x0) << 2) /* strongly ordered */
111 #define L_PTE_S2_MT_WRITETHROUGH	(_AT(pteval_t, 0xa) << 2) /* normal inner write-through */
112 #define L_PTE_S2_MT_WRITEBACK		(_AT(pteval_t, 0xf) << 2) /* normal inner write-back */
113 #define L_PTE_S2_MT_DEV_SHARED		(_AT(pteval_t, 0x1) << 2) /* device */
114 #define L_PTE_S2_MT_MASK		(_AT(pteval_t, 0xf) << 2)
115 
116 #define L_PTE_S2_RDONLY			(_AT(pteval_t, 1) << 6)   /* HAP[1]   */
117 #define L_PTE_S2_RDWR			(_AT(pteval_t, 3) << 6)   /* HAP[2:1] */
118 
119 #define L_PMD_S2_RDONLY			(_AT(pmdval_t, 1) << 6)   /* HAP[1]   */
120 #define L_PMD_S2_RDWR			(_AT(pmdval_t, 3) << 6)   /* HAP[2:1] */
121 
122 /*
123  * Hyp-mode PL2 PTE definitions for LPAE.
124  */
125 #define L_PTE_HYP		L_PTE_USER
126 
127 #ifndef __ASSEMBLY__
128 
129 #define pud_none(pud)		(!pud_val(pud))
130 #define pud_bad(pud)		(!(pud_val(pud) & 2))
131 #define pud_present(pud)	(pud_val(pud))
132 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
133 						 PMD_TYPE_TABLE)
134 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
135 						 PMD_TYPE_SECT)
136 #define pmd_large(pmd)		pmd_sect(pmd)
137 
138 #define pud_clear(pudp)			\
139 	do {				\
140 		*pudp = __pud(0);	\
141 		clean_pmd_entry(pudp);	\
142 	} while (0)
143 
144 #define set_pud(pudp, pud)		\
145 	do {				\
146 		*pudp = pud;		\
147 		flush_pmd_entry(pudp);	\
148 	} while (0)
149 
150 static inline pmd_t *pud_page_vaddr(pud_t pud)
151 {
152 	return __va(pud_val(pud) & PHYS_MASK & (s32)PAGE_MASK);
153 }
154 
155 /* Find an entry in the second-level page table.. */
156 #define pmd_index(addr)		(((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
157 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
158 {
159 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(addr);
160 }
161 
162 #define pmd_bad(pmd)		(!(pmd_val(pmd) & 2))
163 
164 #define copy_pmd(pmdpd,pmdps)		\
165 	do {				\
166 		*pmdpd = *pmdps;	\
167 		flush_pmd_entry(pmdpd);	\
168 	} while (0)
169 
170 #define pmd_clear(pmdp)			\
171 	do {				\
172 		*pmdp = __pmd(0);	\
173 		clean_pmd_entry(pmdp);	\
174 	} while (0)
175 
176 /*
177  * For 3 levels of paging the PTE_EXT_NG bit will be set for user address ptes
178  * that are written to a page table but not for ptes created with mk_pte.
179  *
180  * In hugetlb_no_page, a new huge pte (new_pte) is generated and passed to
181  * hugetlb_cow, where it is compared with an entry in a page table.
182  * This comparison test fails erroneously leading ultimately to a memory leak.
183  *
184  * To correct this behaviour, we mask off PTE_EXT_NG for any pte that is
185  * present before running the comparison.
186  */
187 #define __HAVE_ARCH_PTE_SAME
188 #define pte_same(pte_a,pte_b)	((pte_present(pte_a) ? pte_val(pte_a) & ~PTE_EXT_NG	\
189 					: pte_val(pte_a))				\
190 				== (pte_present(pte_b) ? pte_val(pte_b) & ~PTE_EXT_NG	\
191 					: pte_val(pte_b)))
192 
193 #define set_pte_ext(ptep,pte,ext) cpu_set_pte_ext(ptep,__pte(pte_val(pte)|(ext)))
194 
195 #define pte_huge(pte)		(pte_val(pte) && !(pte_val(pte) & PTE_TABLE_BIT))
196 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
197 
198 #define pmd_isset(pmd, val)	((u32)(val) == (val) ? pmd_val(pmd) & (val) \
199 						: !!(pmd_val(pmd) & (val)))
200 #define pmd_isclear(pmd, val)	(!(pmd_val(pmd) & (val)))
201 
202 #define pmd_present(pmd)	(pmd_isset((pmd), L_PMD_SECT_VALID))
203 #define pmd_young(pmd)		(pmd_isset((pmd), PMD_SECT_AF))
204 #define pte_special(pte)	(pte_isset((pte), L_PTE_SPECIAL))
205 static inline pte_t pte_mkspecial(pte_t pte)
206 {
207 	pte_val(pte) |= L_PTE_SPECIAL;
208 	return pte;
209 }
210 
211 #define pmd_write(pmd)		(pmd_isclear((pmd), L_PMD_SECT_RDONLY))
212 #define pmd_dirty(pmd)		(pmd_isset((pmd), L_PMD_SECT_DIRTY))
213 #define pud_page(pud)		pmd_page(__pmd(pud_val(pud)))
214 #define pud_write(pud)		pmd_write(__pmd(pud_val(pud)))
215 
216 #define pmd_hugewillfault(pmd)	(!pmd_young(pmd) || !pmd_write(pmd))
217 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
218 
219 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
220 #define pmd_trans_huge(pmd)	(pmd_val(pmd) && !pmd_table(pmd))
221 #endif
222 
223 #define PMD_BIT_FUNC(fn,op) \
224 static inline pmd_t pmd_##fn(pmd_t pmd) { pmd_val(pmd) op; return pmd; }
225 
226 PMD_BIT_FUNC(wrprotect,	|= L_PMD_SECT_RDONLY);
227 PMD_BIT_FUNC(mkold,	&= ~PMD_SECT_AF);
228 PMD_BIT_FUNC(mkwrite,   &= ~L_PMD_SECT_RDONLY);
229 PMD_BIT_FUNC(mkdirty,   |= L_PMD_SECT_DIRTY);
230 PMD_BIT_FUNC(mkclean,   &= ~L_PMD_SECT_DIRTY);
231 PMD_BIT_FUNC(mkyoung,   |= PMD_SECT_AF);
232 
233 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
234 
235 #define pmd_pfn(pmd)		(((pmd_val(pmd) & PMD_MASK) & PHYS_MASK) >> PAGE_SHIFT)
236 #define pfn_pmd(pfn,prot)	(__pmd(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
237 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
238 
239 /* No hardware dirty/accessed bits -- generic_pmdp_establish() fits */
240 #define pmdp_establish generic_pmdp_establish
241 
242 /* represent a notpresent pmd by faulting entry, this is used by pmdp_invalidate */
243 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
244 {
245 	return __pmd(pmd_val(pmd) & ~L_PMD_SECT_VALID);
246 }
247 
248 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
249 {
250 	const pmdval_t mask = PMD_SECT_USER | PMD_SECT_XN | L_PMD_SECT_RDONLY |
251 				L_PMD_SECT_VALID | L_PMD_SECT_NONE;
252 	pmd_val(pmd) = (pmd_val(pmd) & ~mask) | (pgprot_val(newprot) & mask);
253 	return pmd;
254 }
255 
256 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
257 			      pmd_t *pmdp, pmd_t pmd)
258 {
259 	BUG_ON(addr >= TASK_SIZE);
260 
261 	/* create a faulting entry if PROT_NONE protected */
262 	if (pmd_val(pmd) & L_PMD_SECT_NONE)
263 		pmd_val(pmd) &= ~L_PMD_SECT_VALID;
264 
265 	if (pmd_write(pmd) && pmd_dirty(pmd))
266 		pmd_val(pmd) &= ~PMD_SECT_AP2;
267 	else
268 		pmd_val(pmd) |= PMD_SECT_AP2;
269 
270 	*pmdp = __pmd(pmd_val(pmd) | PMD_SECT_nG);
271 	flush_pmd_entry(pmdp);
272 }
273 
274 #endif /* __ASSEMBLY__ */
275 
276 #endif /* _ASM_PGTABLE_3LEVEL_H */
277