2 * Lockless get_user_pages_fast for s390
4 * Copyright IBM Corp. 2010
5 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
7 #include <linux/sched.h>
9 #include <linux/hugetlb.h>
10 #include <linux/vmstat.h>
11 #include <linux/pagemap.h>
12 #include <linux/rwsem.h>
13 #include <asm/pgtable.h>
16 * The performance critical leaf functions are made noinline otherwise gcc
17 * inlines everything into a single function which results in too much
20 static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
21 unsigned long end, int write, struct page **pages, int *nr)
27 mask = (write ? _PAGE_RO : 0) | _PAGE_INVALID | _PAGE_SPECIAL;
29 ptep = ((pte_t *) pmd_deref(pmd)) + pte_index(addr);
33 if ((pte_val(pte) & mask) != 0)
35 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
37 if (!page_cache_get_speculative(page))
39 if (unlikely(pte_val(pte) != pte_val(*ptep))) {
46 } while (ptep++, addr += PAGE_SIZE, addr != end);
51 static inline void get_huge_page_tail(struct page *page)
54 * __split_huge_page_refcount() cannot run
57 VM_BUG_ON(page_mapcount(page) < 0);
58 VM_BUG_ON(atomic_read(&page->_count) != 0);
59 atomic_inc(&page->_mapcount);
62 static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
63 unsigned long end, int write, struct page **pages, int *nr)
65 unsigned long mask, result;
66 struct page *head, *page, *tail;
69 result = write ? 0 : _SEGMENT_ENTRY_RO;
70 mask = result | _SEGMENT_ENTRY_INV;
71 if ((pmd_val(pmd) & mask) != result)
73 VM_BUG_ON(!pfn_valid(pmd_val(pmd) >> PAGE_SHIFT));
77 page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
80 VM_BUG_ON(compound_head(page) != head);
85 } while (addr += PAGE_SIZE, addr != end);
87 if (!page_cache_add_speculative(head, refs)) {
92 if (unlikely(pmd_val(pmd) != pmd_val(*pmdp))) {
100 * Any tail page need their mapcount reference taken before we
105 get_huge_page_tail(tail);
113 static inline int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr,
114 unsigned long end, int write, struct page **pages, int *nr)
119 pmdp = (pmd_t *) pudp;
121 if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
122 pmdp = (pmd_t *) pud_deref(pud);
123 pmdp += pmd_index(addr);
128 next = pmd_addr_end(addr, end);
131 if (unlikely(pmd_huge(pmd))) {
132 if (!gup_huge_pmd(pmdp, pmd, addr, next,
135 } else if (!gup_pte_range(pmdp, pmd, addr, next,
138 } while (pmdp++, addr = next, addr != end);
143 static inline int gup_pud_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr,
144 unsigned long end, int write, struct page **pages, int *nr)
149 pudp = (pud_t *) pgdp;
151 if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
152 pudp = (pud_t *) pgd_deref(pgd);
153 pudp += pud_index(addr);
158 next = pud_addr_end(addr, end);
161 if (!gup_pmd_range(pudp, pud, addr, next, write, pages, nr))
163 } while (pudp++, addr = next, addr != end);
169 * get_user_pages_fast() - pin user pages in memory
170 * @start: starting user address
171 * @nr_pages: number of pages from start to pin
172 * @write: whether pages will be written to
173 * @pages: array that receives pointers to the pages pinned.
174 * Should be at least nr_pages long.
176 * Attempt to pin user pages in memory without taking mm->mmap_sem.
177 * If not successful, it will fall back to taking the lock and
178 * calling get_user_pages().
180 * Returns number of pages pinned. This may be fewer than the number
181 * requested. If nr_pages is 0 or negative, returns 0. If no pages
182 * were pinned, returns -errno.
184 int get_user_pages_fast(unsigned long start, int nr_pages, int write,
187 struct mm_struct *mm = current->mm;
188 unsigned long addr, len, end;
195 len = (unsigned long) nr_pages << PAGE_SHIFT;
201 * local_irq_disable() doesn't prevent pagetable teardown, but does
202 * prevent the pagetables from being freed on s390.
204 * So long as we atomically load page table pointers versus teardown,
205 * we can follow the address down to the the page and take a ref on it.
208 pgdp = pgd_offset(mm, addr);
212 next = pgd_addr_end(addr, end);
215 if (!gup_pud_range(pgdp, pgd, addr, next, write, pages, &nr))
217 } while (pgdp++, addr = next, addr != end);
220 VM_BUG_ON(nr != (end - start) >> PAGE_SHIFT);
228 /* Try to get the remaining pages with get_user_pages */
229 start += nr << PAGE_SHIFT;
232 down_read(&mm->mmap_sem);
233 ret = get_user_pages(current, mm, start,
234 (end - start) >> PAGE_SHIFT, write, 0, pages, NULL);
235 up_read(&mm->mmap_sem);
237 /* Have to be a bit careful with return values */