]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
667b279b PM |
2 | /* |
3 | * Lockless get_user_pages_fast for SuperH | |
4 | * | |
5 | * Copyright (C) 2009 - 2010 Paul Mundt | |
6 | * | |
7 | * Cloned from the x86 and PowerPC versions, by: | |
8 | * | |
9 | * Copyright (C) 2008 Nick Piggin | |
10 | * Copyright (C) 2008 Novell Inc. | |
11 | */ | |
12 | #include <linux/sched.h> | |
13 | #include <linux/mm.h> | |
14 | #include <linux/vmstat.h> | |
15 | #include <linux/highmem.h> | |
16 | #include <asm/pgtable.h> | |
17 | ||
18 | static inline pte_t gup_get_pte(pte_t *ptep) | |
19 | { | |
20 | #ifndef CONFIG_X2TLB | |
378af02b | 21 | return READ_ONCE(*ptep); |
667b279b PM |
22 | #else |
23 | /* | |
24 | * With get_user_pages_fast, we walk down the pagetables without | |
25 | * taking any locks. For this we would like to load the pointers | |
26 | * atomically, but that is not possible with 64-bit PTEs. What | |
27 | * we do have is the guarantee that a pte will only either go | |
28 | * from not present to present, or present to not present or both | |
29 | * -- it will not switch to a completely different present page | |
30 | * without a TLB flush in between; something that we are blocking | |
31 | * by holding interrupts off. | |
32 | * | |
33 | * Setting ptes from not present to present goes: | |
34 | * ptep->pte_high = h; | |
35 | * smp_wmb(); | |
36 | * ptep->pte_low = l; | |
37 | * | |
38 | * And present to not present goes: | |
39 | * ptep->pte_low = 0; | |
40 | * smp_wmb(); | |
41 | * ptep->pte_high = 0; | |
42 | * | |
43 | * We must ensure here that the load of pte_low sees l iff pte_high | |
44 | * sees h. We load pte_high *after* loading pte_low, which ensures we | |
45 | * don't see an older value of pte_high. *Then* we recheck pte_low, | |
46 | * which ensures that we haven't picked up a changed pte high. We might | |
47 | * have got rubbish values from pte_low and pte_high, but we are | |
48 | * guaranteed that pte_low will not have the present bit set *unless* | |
49 | * it is 'l'. And get_user_pages_fast only operates on present ptes, so | |
50 | * we're safe. | |
51 | * | |
52 | * gup_get_pte should not be used or copied outside gup.c without being | |
53 | * very careful -- it does not atomically load the pte or anything that | |
54 | * is likely to be useful for you. | |
55 | */ | |
56 | pte_t pte; | |
57 | ||
58 | retry: | |
59 | pte.pte_low = ptep->pte_low; | |
60 | smp_rmb(); | |
61 | pte.pte_high = ptep->pte_high; | |
62 | smp_rmb(); | |
63 | if (unlikely(pte.pte_low != ptep->pte_low)) | |
64 | goto retry; | |
65 | ||
66 | return pte; | |
67 | #endif | |
68 | } | |
69 | ||
70 | /* | |
71 | * The performance critical leaf functions are made noinline otherwise gcc | |
72 | * inlines everything into a single function which results in too much | |
73 | * register pressure. | |
74 | */ | |
75 | static noinline int gup_pte_range(pmd_t pmd, unsigned long addr, | |
76 | unsigned long end, int write, struct page **pages, int *nr) | |
77 | { | |
78 | u64 mask, result; | |
79 | pte_t *ptep; | |
80 | ||
81 | #ifdef CONFIG_X2TLB | |
82 | result = _PAGE_PRESENT | _PAGE_EXT(_PAGE_EXT_KERN_READ | _PAGE_EXT_USER_READ); | |
83 | if (write) | |
84 | result |= _PAGE_EXT(_PAGE_EXT_KERN_WRITE | _PAGE_EXT_USER_WRITE); | |
85 | #elif defined(CONFIG_SUPERH64) | |
86 | result = _PAGE_PRESENT | _PAGE_USER | _PAGE_READ; | |
87 | if (write) | |
88 | result |= _PAGE_WRITE; | |
89 | #else | |
90 | result = _PAGE_PRESENT | _PAGE_USER; | |
91 | if (write) | |
92 | result |= _PAGE_RW; | |
93 | #endif | |
94 | ||
95 | mask = result | _PAGE_SPECIAL; | |
96 | ||
97 | ptep = pte_offset_map(&pmd, addr); | |
98 | do { | |
99 | pte_t pte = gup_get_pte(ptep); | |
100 | struct page *page; | |
101 | ||
102 | if ((pte_val(pte) & mask) != result) { | |
103 | pte_unmap(ptep); | |
104 | return 0; | |
105 | } | |
106 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
107 | page = pte_page(pte); | |
108 | get_page(page); | |
caac7e6d SS |
109 | __flush_anon_page(page, addr); |
110 | flush_dcache_page(page); | |
667b279b PM |
111 | pages[*nr] = page; |
112 | (*nr)++; | |
113 | ||
114 | } while (ptep++, addr += PAGE_SIZE, addr != end); | |
115 | pte_unmap(ptep - 1); | |
116 | ||
117 | return 1; | |
118 | } | |
119 | ||
120 | static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end, | |
121 | int write, struct page **pages, int *nr) | |
122 | { | |
123 | unsigned long next; | |
124 | pmd_t *pmdp; | |
125 | ||
126 | pmdp = pmd_offset(&pud, addr); | |
127 | do { | |
128 | pmd_t pmd = *pmdp; | |
129 | ||
130 | next = pmd_addr_end(addr, end); | |
131 | if (pmd_none(pmd)) | |
132 | return 0; | |
133 | if (!gup_pte_range(pmd, addr, next, write, pages, nr)) | |
134 | return 0; | |
135 | } while (pmdp++, addr = next, addr != end); | |
136 | ||
137 | return 1; | |
138 | } | |
139 | ||
140 | static int gup_pud_range(pgd_t pgd, unsigned long addr, unsigned long end, | |
141 | int write, struct page **pages, int *nr) | |
142 | { | |
143 | unsigned long next; | |
144 | pud_t *pudp; | |
145 | ||
146 | pudp = pud_offset(&pgd, addr); | |
147 | do { | |
148 | pud_t pud = *pudp; | |
149 | ||
150 | next = pud_addr_end(addr, end); | |
151 | if (pud_none(pud)) | |
152 | return 0; | |
153 | if (!gup_pmd_range(pud, addr, next, write, pages, nr)) | |
154 | return 0; | |
155 | } while (pudp++, addr = next, addr != end); | |
156 | ||
157 | return 1; | |
158 | } | |
159 | ||
160 | /* | |
161 | * Like get_user_pages_fast() except its IRQ-safe in that it won't fall | |
162 | * back to the regular GUP. | |
d0811078 MT |
163 | * Note a difference with get_user_pages_fast: this always returns the |
164 | * number of pages pinned, 0 if no pages were pinned. | |
667b279b PM |
165 | */ |
166 | int __get_user_pages_fast(unsigned long start, int nr_pages, int write, | |
167 | struct page **pages) | |
168 | { | |
169 | struct mm_struct *mm = current->mm; | |
170 | unsigned long addr, len, end; | |
171 | unsigned long next; | |
172 | unsigned long flags; | |
173 | pgd_t *pgdp; | |
174 | int nr = 0; | |
175 | ||
176 | start &= PAGE_MASK; | |
177 | addr = start; | |
178 | len = (unsigned long) nr_pages << PAGE_SHIFT; | |
179 | end = start + len; | |
96d4f267 | 180 | if (unlikely(!access_ok((void __user *)start, len))) |
667b279b PM |
181 | return 0; |
182 | ||
183 | /* | |
184 | * This doesn't prevent pagetable teardown, but does prevent | |
185 | * the pagetables and pages from being freed. | |
186 | */ | |
187 | local_irq_save(flags); | |
188 | pgdp = pgd_offset(mm, addr); | |
189 | do { | |
190 | pgd_t pgd = *pgdp; | |
191 | ||
192 | next = pgd_addr_end(addr, end); | |
193 | if (pgd_none(pgd)) | |
194 | break; | |
195 | if (!gup_pud_range(pgd, addr, next, write, pages, &nr)) | |
196 | break; | |
197 | } while (pgdp++, addr = next, addr != end); | |
198 | local_irq_restore(flags); | |
199 | ||
200 | return nr; | |
201 | } | |
202 | ||
203 | /** | |
204 | * get_user_pages_fast() - pin user pages in memory | |
205 | * @start: starting user address | |
206 | * @nr_pages: number of pages from start to pin | |
73b0140b | 207 | * @gup_flags: flags modifying pin behaviour |
667b279b PM |
208 | * @pages: array that receives pointers to the pages pinned. |
209 | * Should be at least nr_pages long. | |
210 | * | |
211 | * Attempt to pin user pages in memory without taking mm->mmap_sem. | |
212 | * If not successful, it will fall back to taking the lock and | |
213 | * calling get_user_pages(). | |
214 | * | |
215 | * Returns number of pages pinned. This may be fewer than the number | |
216 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
217 | * were pinned, returns -errno. | |
218 | */ | |
73b0140b IW |
219 | int get_user_pages_fast(unsigned long start, int nr_pages, |
220 | unsigned int gup_flags, struct page **pages) | |
667b279b PM |
221 | { |
222 | struct mm_struct *mm = current->mm; | |
223 | unsigned long addr, len, end; | |
224 | unsigned long next; | |
225 | pgd_t *pgdp; | |
226 | int nr = 0; | |
227 | ||
228 | start &= PAGE_MASK; | |
229 | addr = start; | |
230 | len = (unsigned long) nr_pages << PAGE_SHIFT; | |
231 | ||
232 | end = start + len; | |
233 | if (end < start) | |
234 | goto slow_irqon; | |
235 | ||
236 | local_irq_disable(); | |
237 | pgdp = pgd_offset(mm, addr); | |
238 | do { | |
239 | pgd_t pgd = *pgdp; | |
240 | ||
241 | next = pgd_addr_end(addr, end); | |
242 | if (pgd_none(pgd)) | |
243 | goto slow; | |
73b0140b IW |
244 | if (!gup_pud_range(pgd, addr, next, gup_flags & FOLL_WRITE, |
245 | pages, &nr)) | |
667b279b PM |
246 | goto slow; |
247 | } while (pgdp++, addr = next, addr != end); | |
248 | local_irq_enable(); | |
249 | ||
250 | VM_BUG_ON(nr != (end - start) >> PAGE_SHIFT); | |
251 | return nr; | |
252 | ||
253 | { | |
254 | int ret; | |
255 | ||
256 | slow: | |
257 | local_irq_enable(); | |
258 | slow_irqon: | |
259 | /* Try to get the remaining pages with get_user_pages */ | |
260 | start += nr << PAGE_SHIFT; | |
261 | pages += nr; | |
262 | ||
d4edcf0d | 263 | ret = get_user_pages_unlocked(start, |
c164154f | 264 | (end - start) >> PAGE_SHIFT, pages, |
73b0140b | 265 | gup_flags); |
667b279b PM |
266 | |
267 | /* Have to be a bit careful with return values */ | |
268 | if (nr > 0) { | |
269 | if (ret < 0) | |
270 | ret = nr; | |
271 | else | |
272 | ret += nr; | |
273 | } | |
274 | ||
275 | return ret; | |
276 | } | |
277 | } |