xref: /linux/arch/powerpc/include/asm/book3s/64/pgalloc.h (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
1 #ifndef _ASM_POWERPC_BOOK3S_64_PGALLOC_H
2 #define _ASM_POWERPC_BOOK3S_64_PGALLOC_H
3 /*
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version
7  * 2 of the License, or (at your option) any later version.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/cpumask.h>
12 #include <linux/kmemleak.h>
13 #include <linux/percpu.h>
14 
15 struct vmemmap_backing {
16 	struct vmemmap_backing *list;
17 	unsigned long phys;
18 	unsigned long virt_addr;
19 };
20 extern struct vmemmap_backing *vmemmap_list;
21 
22 /*
23  * Functions that deal with pagetables that could be at any level of
24  * the table need to be passed an "index_size" so they know how to
25  * handle allocation.  For PTE pages (which are linked to a struct
26  * page for now, and drawn from the main get_free_pages() pool), the
27  * allocation size will be (2^index_size * sizeof(pointer)) and
28  * allocations are drawn from the kmem_cache in PGT_CACHE(index_size).
29  *
30  * The maximum index size needs to be big enough to allow any
31  * pagetable sizes we need, but small enough to fit in the low bits of
32  * any page table pointer.  In other words all pagetables, even tiny
33  * ones, must be aligned to allow at least enough low 0 bits to
34  * contain this value.  This value is also used as a mask, so it must
35  * be one less than a power of two.
36  */
37 #define MAX_PGTABLE_INDEX_SIZE	0xf
38 
39 extern struct kmem_cache *pgtable_cache[];
40 #define PGT_CACHE(shift) pgtable_cache[shift]
41 
42 extern pte_t *pte_fragment_alloc(struct mm_struct *, int);
43 extern pmd_t *pmd_fragment_alloc(struct mm_struct *, unsigned long);
44 extern void pte_fragment_free(unsigned long *, int);
45 extern void pmd_fragment_free(unsigned long *);
46 extern void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift);
47 #ifdef CONFIG_SMP
48 extern void __tlb_remove_table(void *_table);
49 #endif
50 void pte_frag_destroy(void *pte_frag);
51 
52 static inline pgd_t *radix__pgd_alloc(struct mm_struct *mm)
53 {
54 #ifdef CONFIG_PPC_64K_PAGES
55 	return (pgd_t *)__get_free_page(pgtable_gfp_flags(mm, PGALLOC_GFP));
56 #else
57 	struct page *page;
58 	page = alloc_pages(pgtable_gfp_flags(mm, PGALLOC_GFP | __GFP_RETRY_MAYFAIL),
59 				4);
60 	if (!page)
61 		return NULL;
62 	return (pgd_t *) page_address(page);
63 #endif
64 }
65 
66 static inline void radix__pgd_free(struct mm_struct *mm, pgd_t *pgd)
67 {
68 #ifdef CONFIG_PPC_64K_PAGES
69 	free_page((unsigned long)pgd);
70 #else
71 	free_pages((unsigned long)pgd, 4);
72 #endif
73 }
74 
75 static inline pgd_t *pgd_alloc(struct mm_struct *mm)
76 {
77 	pgd_t *pgd;
78 
79 	if (radix_enabled())
80 		return radix__pgd_alloc(mm);
81 
82 	pgd = kmem_cache_alloc(PGT_CACHE(PGD_INDEX_SIZE),
83 			       pgtable_gfp_flags(mm, GFP_KERNEL));
84 	/*
85 	 * Don't scan the PGD for pointers, it contains references to PUDs but
86 	 * those references are not full pointers and so can't be recognised by
87 	 * kmemleak.
88 	 */
89 	kmemleak_no_scan(pgd);
90 
91 	/*
92 	 * With hugetlb, we don't clear the second half of the page table.
93 	 * If we share the same slab cache with the pmd or pud level table,
94 	 * we need to make sure we zero out the full table on alloc.
95 	 * With 4K we don't store slot in the second half. Hence we don't
96 	 * need to do this for 4k.
97 	 */
98 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_PPC_64K_PAGES) && \
99 	(H_PGD_INDEX_SIZE == H_PUD_CACHE_INDEX)
100 	memset(pgd, 0, PGD_TABLE_SIZE);
101 #endif
102 	return pgd;
103 }
104 
105 static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
106 {
107 	if (radix_enabled())
108 		return radix__pgd_free(mm, pgd);
109 	kmem_cache_free(PGT_CACHE(PGD_INDEX_SIZE), pgd);
110 }
111 
112 static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pud_t *pud)
113 {
114 	pgd_set(pgd, __pgtable_ptr_val(pud) | PGD_VAL_BITS);
115 }
116 
117 static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long addr)
118 {
119 	pud_t *pud;
120 
121 	pud = kmem_cache_alloc(PGT_CACHE(PUD_CACHE_INDEX),
122 			       pgtable_gfp_flags(mm, GFP_KERNEL));
123 	/*
124 	 * Tell kmemleak to ignore the PUD, that means don't scan it for
125 	 * pointers and don't consider it a leak. PUDs are typically only
126 	 * referred to by their PGD, but kmemleak is not able to recognise those
127 	 * as pointers, leading to false leak reports.
128 	 */
129 	kmemleak_ignore(pud);
130 
131 	return pud;
132 }
133 
134 static inline void pud_free(struct mm_struct *mm, pud_t *pud)
135 {
136 	kmem_cache_free(PGT_CACHE(PUD_CACHE_INDEX), pud);
137 }
138 
139 static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
140 {
141 	pud_set(pud, __pgtable_ptr_val(pmd) | PUD_VAL_BITS);
142 }
143 
144 static inline void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud,
145 				  unsigned long address)
146 {
147 	/*
148 	 * By now all the pud entries should be none entries. So go
149 	 * ahead and flush the page walk cache
150 	 */
151 	flush_tlb_pgtable(tlb, address);
152 	pgtable_free_tlb(tlb, pud, PUD_INDEX);
153 }
154 
155 static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
156 {
157 	return pmd_fragment_alloc(mm, addr);
158 }
159 
160 static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
161 {
162 	pmd_fragment_free((unsigned long *)pmd);
163 }
164 
165 static inline void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd,
166 				  unsigned long address)
167 {
168 	/*
169 	 * By now all the pud entries should be none entries. So go
170 	 * ahead and flush the page walk cache
171 	 */
172 	flush_tlb_pgtable(tlb, address);
173 	return pgtable_free_tlb(tlb, pmd, PMD_INDEX);
174 }
175 
176 static inline void pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd,
177 				       pte_t *pte)
178 {
179 	pmd_set(pmd, __pgtable_ptr_val(pte) | PMD_VAL_BITS);
180 }
181 
182 static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmd,
183 				pgtable_t pte_page)
184 {
185 	pmd_set(pmd, __pgtable_ptr_val(pte_page) | PMD_VAL_BITS);
186 }
187 
188 static inline pgtable_t pmd_pgtable(pmd_t pmd)
189 {
190 	return (pgtable_t)pmd_page_vaddr(pmd);
191 }
192 
193 static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
194 {
195 	return (pte_t *)pte_fragment_alloc(mm, 1);
196 }
197 
198 static inline pgtable_t pte_alloc_one(struct mm_struct *mm)
199 {
200 	return (pgtable_t)pte_fragment_alloc(mm, 0);
201 }
202 
203 static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
204 {
205 	pte_fragment_free((unsigned long *)pte, 1);
206 }
207 
208 static inline void pte_free(struct mm_struct *mm, pgtable_t ptepage)
209 {
210 	pte_fragment_free((unsigned long *)ptepage, 0);
211 }
212 
213 static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t table,
214 				  unsigned long address)
215 {
216 	/*
217 	 * By now all the pud entries should be none entries. So go
218 	 * ahead and flush the page walk cache
219 	 */
220 	flush_tlb_pgtable(tlb, address);
221 	pgtable_free_tlb(tlb, table, PTE_INDEX);
222 }
223 
224 #define check_pgt_cache()	do { } while (0)
225 
226 extern atomic_long_t direct_pages_count[MMU_PAGE_COUNT];
227 static inline void update_page_count(int psize, long count)
228 {
229 	if (IS_ENABLED(CONFIG_PROC_FS))
230 		atomic_long_add(count, &direct_pages_count[psize]);
231 }
232 
233 #endif /* _ASM_POWERPC_BOOK3S_64_PGALLOC_H */
234