blob: 00c23433b39f2ad868aefc6f414d2e728398bd15 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _ASM_GENERIC_PGTABLE_H
2#define _ASM_GENERIC_PGTABLE_H
3
Rusty Russell673eae82006-09-25 23:32:29 -07004#ifndef __ASSEMBLY__
5
Linus Torvalds1da177e2005-04-16 15:20:36 -07006#ifndef __HAVE_ARCH_PTEP_ESTABLISH
7/*
8 * Establish a new mapping:
9 * - flush the old one
10 * - update the page tables
11 * - inform the TLB about the new one
12 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070013 * We hold the mm semaphore for reading, and the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
15 * Note: the old pte is known to not be writable, so we don't need to
16 * worry about dirty bits etc getting lost.
17 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#define ptep_establish(__vma, __address, __ptep, __entry) \
19do { \
20 set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
21 flush_tlb_page(__vma, __address); \
22} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#endif
24
25#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
26/*
27 * Largely same as above, but only sets the access flags (dirty,
28 * accessed, and writable). Furthermore, we know it always gets set
29 * to a "more permissive" setting, which allows most architectures
30 * to optimize this.
31 */
32#define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
33do { \
34 set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
35 flush_tlb_page(__vma, __address); \
36} while (0)
37#endif
38
39#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
40#define ptep_test_and_clear_young(__vma, __address, __ptep) \
41({ \
42 pte_t __pte = *(__ptep); \
43 int r = 1; \
44 if (!pte_young(__pte)) \
45 r = 0; \
46 else \
47 set_pte_at((__vma)->vm_mm, (__address), \
48 (__ptep), pte_mkold(__pte)); \
49 r; \
50})
51#endif
52
53#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
54#define ptep_clear_flush_young(__vma, __address, __ptep) \
55({ \
56 int __young; \
57 __young = ptep_test_and_clear_young(__vma, __address, __ptep); \
58 if (__young) \
59 flush_tlb_page(__vma, __address); \
60 __young; \
61})
62#endif
63
64#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
65#define ptep_test_and_clear_dirty(__vma, __address, __ptep) \
66({ \
67 pte_t __pte = *__ptep; \
68 int r = 1; \
69 if (!pte_dirty(__pte)) \
70 r = 0; \
71 else \
72 set_pte_at((__vma)->vm_mm, (__address), (__ptep), \
73 pte_mkclean(__pte)); \
74 r; \
75})
76#endif
77
78#ifndef __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
79#define ptep_clear_flush_dirty(__vma, __address, __ptep) \
80({ \
81 int __dirty; \
82 __dirty = ptep_test_and_clear_dirty(__vma, __address, __ptep); \
83 if (__dirty) \
84 flush_tlb_page(__vma, __address); \
85 __dirty; \
86})
87#endif
88
89#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
90#define ptep_get_and_clear(__mm, __address, __ptep) \
91({ \
92 pte_t __pte = *(__ptep); \
93 pte_clear((__mm), (__address), (__ptep)); \
94 __pte; \
95})
96#endif
97
Zachary Amsdena6003882005-09-03 15:55:04 -070098#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
99#define ptep_get_and_clear_full(__mm, __address, __ptep, __full) \
100({ \
101 pte_t __pte; \
102 __pte = ptep_get_and_clear((__mm), (__address), (__ptep)); \
103 __pte; \
104})
105#endif
106
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700107/*
108 * Some architectures may be able to avoid expensive synchronization
109 * primitives when modifications are made to PTE's which are already
110 * not present, or in the process of an address space destruction.
111 */
112#ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
113#define pte_clear_not_present_full(__mm, __address, __ptep, __full) \
Zachary Amsdena6003882005-09-03 15:55:04 -0700114do { \
115 pte_clear((__mm), (__address), (__ptep)); \
116} while (0)
117#endif
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
120#define ptep_clear_flush(__vma, __address, __ptep) \
121({ \
122 pte_t __pte; \
123 __pte = ptep_get_and_clear((__vma)->vm_mm, __address, __ptep); \
124 flush_tlb_page(__vma, __address); \
125 __pte; \
126})
127#endif
128
129#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
Tim Schmielau8c65b4a2005-11-07 00:59:43 -0800130struct mm_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
132{
133 pte_t old_pte = *ptep;
134 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
135}
136#endif
137
138#ifndef __HAVE_ARCH_PTE_SAME
139#define pte_same(A,B) (pte_val(A) == pte_val(B))
140#endif
141
142#ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
143#define page_test_and_clear_dirty(page) (0)
Abhijit Karmarkarb4955ce2005-06-21 17:15:13 -0700144#define pte_maybe_dirty(pte) pte_dirty(pte)
145#else
146#define pte_maybe_dirty(pte) (1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147#endif
148
149#ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
150#define page_test_and_clear_young(page) (0)
151#endif
152
153#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
154#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
155#endif
156
157#ifndef __HAVE_ARCH_LAZY_MMU_PROT_UPDATE
158#define lazy_mmu_prot_update(pte) do { } while (0)
159#endif
160
David S. Miller0b0968a2006-06-01 17:47:25 -0700161#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700162#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700163#endif
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165/*
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700166 * A facility to provide lazy MMU batching. This allows PTE updates and
167 * page invalidations to be delayed until a call to leave lazy MMU mode
168 * is issued. Some architectures may benefit from doing this, and it is
169 * beneficial for both shadow and direct mode hypervisors, which may batch
170 * the PTE updates which happen during this window. Note that using this
171 * interface requires that read hazards be removed from the code. A read
172 * hazard could result in the direct mode hypervisor case, since the actual
173 * write to the page tables may not yet have taken place, so reads though
174 * a raw PTE pointer after it has been modified are not guaranteed to be
175 * up to date. This mode can only be entered and left under the protection of
176 * the page table locks for all page tables which may be modified. In the UP
177 * case, this is required so that preemption is disabled, and in the SMP case,
178 * it must synchronize the delayed page table writes properly on other CPUs.
179 */
180#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
181#define arch_enter_lazy_mmu_mode() do {} while (0)
182#define arch_leave_lazy_mmu_mode() do {} while (0)
183#endif
184
185/*
Zachary Amsden9226d122007-02-13 13:26:21 +0100186 * A facility to provide batching of the reload of page tables with the
187 * actual context switch code for paravirtualized guests. By convention,
188 * only one of the lazy modes (CPU, MMU) should be active at any given
189 * time, entry should never be nested, and entry and exits should always
190 * be paired. This is for sanity of maintaining and reasoning about the
191 * kernel code.
192 */
193#ifndef __HAVE_ARCH_ENTER_LAZY_CPU_MODE
194#define arch_enter_lazy_cpu_mode() do {} while (0)
195#define arch_leave_lazy_cpu_mode() do {} while (0)
196#endif
197
198/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700199 * When walking page tables, get the address of the next boundary,
200 * or the end address of the range if that comes earlier. Although no
201 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 */
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204#define pgd_addr_end(addr, end) \
205({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
206 (__boundary - 1 < (end) - 1)? __boundary: (end); \
207})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
209#ifndef pud_addr_end
210#define pud_addr_end(addr, end) \
211({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
212 (__boundary - 1 < (end) - 1)? __boundary: (end); \
213})
214#endif
215
216#ifndef pmd_addr_end
217#define pmd_addr_end(addr, end) \
218({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
219 (__boundary - 1 < (end) - 1)? __boundary: (end); \
220})
221#endif
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223/*
224 * When walking page tables, we usually want to skip any p?d_none entries;
225 * and any p?d_bad entries - reporting the error before resetting to none.
226 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
227 */
228void pgd_clear_bad(pgd_t *);
229void pud_clear_bad(pud_t *);
230void pmd_clear_bad(pmd_t *);
231
232static inline int pgd_none_or_clear_bad(pgd_t *pgd)
233{
234 if (pgd_none(*pgd))
235 return 1;
236 if (unlikely(pgd_bad(*pgd))) {
237 pgd_clear_bad(pgd);
238 return 1;
239 }
240 return 0;
241}
242
243static inline int pud_none_or_clear_bad(pud_t *pud)
244{
245 if (pud_none(*pud))
246 return 1;
247 if (unlikely(pud_bad(*pud))) {
248 pud_clear_bad(pud);
249 return 1;
250 }
251 return 0;
252}
253
254static inline int pmd_none_or_clear_bad(pmd_t *pmd)
255{
256 if (pmd_none(*pmd))
257 return 1;
258 if (unlikely(pmd_bad(*pmd))) {
259 pmd_clear_bad(pmd);
260 return 1;
261 }
262 return 0;
263}
264#endif /* !__ASSEMBLY__ */
265
266#endif /* _ASM_GENERIC_PGTABLE_H */