2 * Lockless get_user_pages_fast for SuperH
4 * Copyright (C) 2009 - 2010 Paul Mundt
6 * Cloned from the x86 and PowerPC versions, by:
8 * Copyright (C) 2008 Nick Piggin
9 * Copyright (C) 2008 Novell Inc.
11 #include <linux/sched.h>
13 #include <linux/vmstat.h>
14 #include <linux/highmem.h>
15 #include <asm/pgtable.h>
17 static inline pte_t gup_get_pte(pte_t *ptep)
20 return ACCESS_ONCE(*ptep);
23 * With get_user_pages_fast, we walk down the pagetables without
24 * taking any locks. For this we would like to load the pointers
25 * atomically, but that is not possible with 64-bit PTEs. What
26 * we do have is the guarantee that a pte will only either go
27 * from not present to present, or present to not present or both
28 * -- it will not switch to a completely different present page
29 * without a TLB flush in between; something that we are blocking
30 * by holding interrupts off.
32 * Setting ptes from not present to present goes:
37 * And present to not present goes:
42 * We must ensure here that the load of pte_low sees l iff pte_high
43 * sees h. We load pte_high *after* loading pte_low, which ensures we
44 * don't see an older value of pte_high. *Then* we recheck pte_low,
45 * which ensures that we haven't picked up a changed pte high. We might
46 * have got rubbish values from pte_low and pte_high, but we are
47 * guaranteed that pte_low will not have the present bit set *unless*
48 * it is 'l'. And get_user_pages_fast only operates on present ptes, so
51 * gup_get_pte should not be used or copied outside gup.c without being
52 * very careful -- it does not atomically load the pte or anything that
53 * is likely to be useful for you.
58 pte.pte_low = ptep->pte_low;
60 pte.pte_high = ptep->pte_high;
62 if (unlikely(pte.pte_low != ptep->pte_low))
70 * The performance critical leaf functions are made noinline otherwise gcc
71 * inlines everything into a single function which results in too much
74 static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
75 unsigned long end, int write, struct page **pages, int *nr)
81 result = _PAGE_PRESENT | _PAGE_EXT(_PAGE_EXT_KERN_READ | _PAGE_EXT_USER_READ);
83 result |= _PAGE_EXT(_PAGE_EXT_KERN_WRITE | _PAGE_EXT_USER_WRITE);
84 #elif defined(CONFIG_SUPERH64)
85 result = _PAGE_PRESENT | _PAGE_USER | _PAGE_READ;
87 result |= _PAGE_WRITE;
89 result = _PAGE_PRESENT | _PAGE_USER;
94 mask = result | _PAGE_SPECIAL;
96 ptep = pte_offset_map(&pmd, addr);
98 pte_t pte = gup_get_pte(ptep);
101 if ((pte_val(pte) & mask) != result) {
105 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
106 page = pte_page(pte);
111 } while (ptep++, addr += PAGE_SIZE, addr != end);
117 static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end,
118 int write, struct page **pages, int *nr)
123 pmdp = pmd_offset(&pud, addr);
127 next = pmd_addr_end(addr, end);
130 if (!gup_pte_range(pmd, addr, next, write, pages, nr))
132 } while (pmdp++, addr = next, addr != end);
137 static int gup_pud_range(pgd_t pgd, unsigned long addr, unsigned long end,
138 int write, struct page **pages, int *nr)
143 pudp = pud_offset(&pgd, addr);
147 next = pud_addr_end(addr, end);
150 if (!gup_pmd_range(pud, addr, next, write, pages, nr))
152 } while (pudp++, addr = next, addr != end);
158 * Like get_user_pages_fast() except its IRQ-safe in that it won't fall
159 * back to the regular GUP.
161 int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
164 struct mm_struct *mm = current->mm;
165 unsigned long addr, len, end;
173 len = (unsigned long) nr_pages << PAGE_SHIFT;
175 if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
176 (void __user *)start, len)))
180 * This doesn't prevent pagetable teardown, but does prevent
181 * the pagetables and pages from being freed.
183 local_irq_save(flags);
184 pgdp = pgd_offset(mm, addr);
188 next = pgd_addr_end(addr, end);
191 if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
193 } while (pgdp++, addr = next, addr != end);
194 local_irq_restore(flags);
200 * get_user_pages_fast() - pin user pages in memory
201 * @start: starting user address
202 * @nr_pages: number of pages from start to pin
203 * @write: whether pages will be written to
204 * @pages: array that receives pointers to the pages pinned.
205 * Should be at least nr_pages long.
207 * Attempt to pin user pages in memory without taking mm->mmap_sem.
208 * If not successful, it will fall back to taking the lock and
209 * calling get_user_pages().
211 * Returns number of pages pinned. This may be fewer than the number
212 * requested. If nr_pages is 0 or negative, returns 0. If no pages
213 * were pinned, returns -errno.
215 int get_user_pages_fast(unsigned long start, int nr_pages, int write,
218 struct mm_struct *mm = current->mm;
219 unsigned long addr, len, end;
226 len = (unsigned long) nr_pages << PAGE_SHIFT;
233 pgdp = pgd_offset(mm, addr);
237 next = pgd_addr_end(addr, end);
240 if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
242 } while (pgdp++, addr = next, addr != end);
245 VM_BUG_ON(nr != (end - start) >> PAGE_SHIFT);
254 /* Try to get the remaining pages with get_user_pages */
255 start += nr << PAGE_SHIFT;
258 down_read(&mm->mmap_sem);
259 ret = get_user_pages(current, mm, start,
260 (end - start) >> PAGE_SHIFT, write, 0, pages, NULL);
261 up_read(&mm->mmap_sem);
263 /* Have to be a bit careful with return values */