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) ({				\
41 			BUG_ON(!(shift));		\
42 			pgtable_cache[(shift) - 1];	\
43 		})
44 
45 extern pte_t *pte_fragment_alloc(struct mm_struct *, unsigned long, int);
46 extern pmd_t *pmd_fragment_alloc(struct mm_struct *, unsigned long);
47 extern void pte_fragment_free(unsigned long *, int);
48 extern void pmd_fragment_free(unsigned long *);
49 extern void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift);
50 #ifdef CONFIG_SMP
51 extern void __tlb_remove_table(void *_table);
52 #endif
53 
54 static inline pgd_t *radix__pgd_alloc(struct mm_struct *mm)
55 {
56 #ifdef CONFIG_PPC_64K_PAGES
57 	return (pgd_t *)__get_free_page(pgtable_gfp_flags(mm, PGALLOC_GFP));
58 #else
59 	struct page *page;
60 	page = alloc_pages(pgtable_gfp_flags(mm, PGALLOC_GFP | __GFP_RETRY_MAYFAIL),
61 				4);
62 	if (!page)
63 		return NULL;
64 	return (pgd_t *) page_address(page);
65 #endif
66 }
67 
68 static inline void radix__pgd_free(struct mm_struct *mm, pgd_t *pgd)
69 {
70 #ifdef CONFIG_PPC_64K_PAGES
71 	free_page((unsigned long)pgd);
72 #else
73 	free_pages((unsigned long)pgd, 4);
74 #endif
75 }
76 
77 static inline pgd_t *pgd_alloc(struct mm_struct *mm)
78 {
79 	pgd_t *pgd;
80 
81 	if (radix_enabled())
82 		return radix__pgd_alloc(mm);
83 
84 	pgd = kmem_cache_alloc(PGT_CACHE(PGD_INDEX_SIZE),
85 			       pgtable_gfp_flags(mm, GFP_KERNEL));
86 	/*
87 	 * Don't scan the PGD for pointers, it contains references to PUDs but
88 	 * those references are not full pointers and so can't be recognised by
89 	 * kmemleak.
90 	 */
91 	kmemleak_no_scan(pgd);
92 
93 	/*
94 	 * With hugetlb, we don't clear the second half of the page table.
95 	 * If we share the same slab cache with the pmd or pud level table,
96 	 * we need to make sure we zero out the full table on alloc.
97 	 * With 4K we don't store slot in the second half. Hence we don't
98 	 * need to do this for 4k.
99 	 */
100 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_PPC_64K_PAGES) && \
101 	(H_PGD_INDEX_SIZE == H_PUD_CACHE_INDEX)
102 	memset(pgd, 0, PGD_TABLE_SIZE);
103 #endif
104 	return pgd;
105 }
106 
107 static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
108 {
109 	if (radix_enabled())
110 		return radix__pgd_free(mm, pgd);
111 	kmem_cache_free(PGT_CACHE(PGD_INDEX_SIZE), pgd);
112 }
113 
114 static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pud_t *pud)
115 {
116 	pgd_set(pgd, __pgtable_ptr_val(pud) | PGD_VAL_BITS);
117 }
118 
119 static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long addr)
120 {
121 	pud_t *pud;
122 
123 	pud = kmem_cache_alloc(PGT_CACHE(PUD_CACHE_INDEX),
124 			       pgtable_gfp_flags(mm, GFP_KERNEL));
125 	/*
126 	 * Tell kmemleak to ignore the PUD, that means don't scan it for
127 	 * pointers and don't consider it a leak. PUDs are typically only
128 	 * referred to by their PGD, but kmemleak is not able to recognise those
129 	 * as pointers, leading to false leak reports.
130 	 */
131 	kmemleak_ignore(pud);
132 
133 	return pud;
134 }
135 
136 static inline void pud_free(struct mm_struct *mm, pud_t *pud)
137 {
138 	kmem_cache_free(PGT_CACHE(PUD_CACHE_INDEX), pud);
139 }
140 
141 static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
142 {
143 	pud_set(pud, __pgtable_ptr_val(pmd) | PUD_VAL_BITS);
144 }
145 
146 static inline void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud,
147 				  unsigned long address)
148 {
149 	/*
150 	 * By now all the pud entries should be none entries. So go
151 	 * ahead and flush the page walk cache
152 	 */
153 	flush_tlb_pgtable(tlb, address);
154 	pgtable_free_tlb(tlb, pud, PUD_INDEX);
155 }
156 
157 static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
158 {
159 	return pmd_fragment_alloc(mm, addr);
160 }
161 
162 static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
163 {
164 	pmd_fragment_free((unsigned long *)pmd);
165 }
166 
167 static inline void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd,
168 				  unsigned long address)
169 {
170 	/*
171 	 * By now all the pud entries should be none entries. So go
172 	 * ahead and flush the page walk cache
173 	 */
174 	flush_tlb_pgtable(tlb, address);
175 	return pgtable_free_tlb(tlb, pmd, PMD_INDEX);
176 }
177 
178 static inline void pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd,
179 				       pte_t *pte)
180 {
181 	pmd_set(pmd, __pgtable_ptr_val(pte) | PMD_VAL_BITS);
182 }
183 
184 static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmd,
185 				pgtable_t pte_page)
186 {
187 	pmd_set(pmd, __pgtable_ptr_val(pte_page) | PMD_VAL_BITS);
188 }
189 
190 static inline pgtable_t pmd_pgtable(pmd_t pmd)
191 {
192 	return (pgtable_t)pmd_page_vaddr(pmd);
193 }
194 
195 static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm,
196 					  unsigned long address)
197 {
198 	return (pte_t *)pte_fragment_alloc(mm, address, 1);
199 }
200 
201 static inline pgtable_t pte_alloc_one(struct mm_struct *mm,
202 				      unsigned long address)
203 {
204 	return (pgtable_t)pte_fragment_alloc(mm, address, 0);
205 }
206 
207 static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
208 {
209 	pte_fragment_free((unsigned long *)pte, 1);
210 }
211 
212 static inline void pte_free(struct mm_struct *mm, pgtable_t ptepage)
213 {
214 	pte_fragment_free((unsigned long *)ptepage, 0);
215 }
216 
217 static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t table,
218 				  unsigned long address)
219 {
220 	/*
221 	 * By now all the pud entries should be none entries. So go
222 	 * ahead and flush the page walk cache
223 	 */
224 	flush_tlb_pgtable(tlb, address);
225 	pgtable_free_tlb(tlb, table, PTE_INDEX);
226 }
227 
228 #define check_pgt_cache()	do { } while (0)
229 
230 extern atomic_long_t direct_pages_count[MMU_PAGE_COUNT];
231 static inline void update_page_count(int psize, long count)
232 {
233 	if (IS_ENABLED(CONFIG_PROC_FS))
234 		atomic_long_add(count, &direct_pages_count[psize]);
235 }
236 
237 #endif /* _ASM_POWERPC_BOOK3S_64_PGALLOC_H */
238