[PATCH] paravirt: lazy mmu mode hooks.patch
[linux-3.10.git] / include / asm-generic / pgtable.h
1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
3
4 #ifndef __ASSEMBLY__
5
6 #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  *
13  * We hold the mm semaphore for reading, and the pte lock.
14  *
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  */
18 #ifndef __HAVE_ARCH_SET_PTE_ATOMIC
19 #define ptep_establish(__vma, __address, __ptep, __entry)               \
20 do {                                                                    \
21         set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry);       \
22         flush_tlb_page(__vma, __address);                               \
23 } while (0)
24 #else /* __HAVE_ARCH_SET_PTE_ATOMIC */
25 #define ptep_establish(__vma, __address, __ptep, __entry)               \
26 do {                                                                    \
27         set_pte_atomic(__ptep, __entry);                                \
28         flush_tlb_page(__vma, __address);                               \
29 } while (0)
30 #endif /* __HAVE_ARCH_SET_PTE_ATOMIC */
31 #endif
32
33 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
34 /*
35  * Largely same as above, but only sets the access flags (dirty,
36  * accessed, and writable). Furthermore, we know it always gets set
37  * to a "more permissive" setting, which allows most architectures
38  * to optimize this.
39  */
40 #define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
41 do {                                                                      \
42         set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry);         \
43         flush_tlb_page(__vma, __address);                                 \
44 } while (0)
45 #endif
46
47 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
48 #define ptep_test_and_clear_young(__vma, __address, __ptep)             \
49 ({                                                                      \
50         pte_t __pte = *(__ptep);                                        \
51         int r = 1;                                                      \
52         if (!pte_young(__pte))                                          \
53                 r = 0;                                                  \
54         else                                                            \
55                 set_pte_at((__vma)->vm_mm, (__address),                 \
56                            (__ptep), pte_mkold(__pte));                 \
57         r;                                                              \
58 })
59 #endif
60
61 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
62 #define ptep_clear_flush_young(__vma, __address, __ptep)                \
63 ({                                                                      \
64         int __young;                                                    \
65         __young = ptep_test_and_clear_young(__vma, __address, __ptep);  \
66         if (__young)                                                    \
67                 flush_tlb_page(__vma, __address);                       \
68         __young;                                                        \
69 })
70 #endif
71
72 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
73 #define ptep_test_and_clear_dirty(__vma, __address, __ptep)             \
74 ({                                                                      \
75         pte_t __pte = *__ptep;                                          \
76         int r = 1;                                                      \
77         if (!pte_dirty(__pte))                                          \
78                 r = 0;                                                  \
79         else                                                            \
80                 set_pte_at((__vma)->vm_mm, (__address), (__ptep),       \
81                            pte_mkclean(__pte));                         \
82         r;                                                              \
83 })
84 #endif
85
86 #ifndef __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
87 #define ptep_clear_flush_dirty(__vma, __address, __ptep)                \
88 ({                                                                      \
89         int __dirty;                                                    \
90         __dirty = ptep_test_and_clear_dirty(__vma, __address, __ptep);  \
91         if (__dirty)                                                    \
92                 flush_tlb_page(__vma, __address);                       \
93         __dirty;                                                        \
94 })
95 #endif
96
97 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
98 #define ptep_get_and_clear(__mm, __address, __ptep)                     \
99 ({                                                                      \
100         pte_t __pte = *(__ptep);                                        \
101         pte_clear((__mm), (__address), (__ptep));                       \
102         __pte;                                                          \
103 })
104 #endif
105
106 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
107 #define ptep_get_and_clear_full(__mm, __address, __ptep, __full)        \
108 ({                                                                      \
109         pte_t __pte;                                                    \
110         __pte = ptep_get_and_clear((__mm), (__address), (__ptep));      \
111         __pte;                                                          \
112 })
113 #endif
114
115 /*
116  * Some architectures may be able to avoid expensive synchronization
117  * primitives when modifications are made to PTE's which are already
118  * not present, or in the process of an address space destruction.
119  */
120 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
121 #define pte_clear_not_present_full(__mm, __address, __ptep, __full)     \
122 do {                                                                    \
123         pte_clear((__mm), (__address), (__ptep));                       \
124 } while (0)
125 #endif
126
127 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
128 #define ptep_clear_flush(__vma, __address, __ptep)                      \
129 ({                                                                      \
130         pte_t __pte;                                                    \
131         __pte = ptep_get_and_clear((__vma)->vm_mm, __address, __ptep);  \
132         flush_tlb_page(__vma, __address);                               \
133         __pte;                                                          \
134 })
135 #endif
136
137 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
138 struct mm_struct;
139 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
140 {
141         pte_t old_pte = *ptep;
142         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
143 }
144 #endif
145
146 #ifndef __HAVE_ARCH_PTE_SAME
147 #define pte_same(A,B)   (pte_val(A) == pte_val(B))
148 #endif
149
150 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
151 #define page_test_and_clear_dirty(page) (0)
152 #define pte_maybe_dirty(pte)            pte_dirty(pte)
153 #else
154 #define pte_maybe_dirty(pte)            (1)
155 #endif
156
157 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
158 #define page_test_and_clear_young(page) (0)
159 #endif
160
161 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
162 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
163 #endif
164
165 #ifndef __HAVE_ARCH_LAZY_MMU_PROT_UPDATE
166 #define lazy_mmu_prot_update(pte)       do { } while (0)
167 #endif
168
169 #ifndef __HAVE_ARCH_MOVE_PTE
170 #define move_pte(pte, prot, old_addr, new_addr) (pte)
171 #endif
172
173 /*
174  * A facility to provide lazy MMU batching.  This allows PTE updates and
175  * page invalidations to be delayed until a call to leave lazy MMU mode
176  * is issued.  Some architectures may benefit from doing this, and it is
177  * beneficial for both shadow and direct mode hypervisors, which may batch
178  * the PTE updates which happen during this window.  Note that using this
179  * interface requires that read hazards be removed from the code.  A read
180  * hazard could result in the direct mode hypervisor case, since the actual
181  * write to the page tables may not yet have taken place, so reads though
182  * a raw PTE pointer after it has been modified are not guaranteed to be
183  * up to date.  This mode can only be entered and left under the protection of
184  * the page table locks for all page tables which may be modified.  In the UP
185  * case, this is required so that preemption is disabled, and in the SMP case,
186  * it must synchronize the delayed page table writes properly on other CPUs.
187  */
188 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
189 #define arch_enter_lazy_mmu_mode()      do {} while (0)
190 #define arch_leave_lazy_mmu_mode()      do {} while (0)
191 #endif
192
193 /*
194  * When walking page tables, get the address of the next boundary,
195  * or the end address of the range if that comes earlier.  Although no
196  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
197  */
198
199 #define pgd_addr_end(addr, end)                                         \
200 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
201         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
202 })
203
204 #ifndef pud_addr_end
205 #define pud_addr_end(addr, end)                                         \
206 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
207         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
208 })
209 #endif
210
211 #ifndef pmd_addr_end
212 #define pmd_addr_end(addr, end)                                         \
213 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
214         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
215 })
216 #endif
217
218 /*
219  * When walking page tables, we usually want to skip any p?d_none entries;
220  * and any p?d_bad entries - reporting the error before resetting to none.
221  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
222  */
223 void pgd_clear_bad(pgd_t *);
224 void pud_clear_bad(pud_t *);
225 void pmd_clear_bad(pmd_t *);
226
227 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
228 {
229         if (pgd_none(*pgd))
230                 return 1;
231         if (unlikely(pgd_bad(*pgd))) {
232                 pgd_clear_bad(pgd);
233                 return 1;
234         }
235         return 0;
236 }
237
238 static inline int pud_none_or_clear_bad(pud_t *pud)
239 {
240         if (pud_none(*pud))
241                 return 1;
242         if (unlikely(pud_bad(*pud))) {
243                 pud_clear_bad(pud);
244                 return 1;
245         }
246         return 0;
247 }
248
249 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
250 {
251         if (pmd_none(*pmd))
252                 return 1;
253         if (unlikely(pmd_bad(*pmd))) {
254                 pmd_clear_bad(pmd);
255                 return 1;
256         }
257         return 0;
258 }
259 #endif /* !__ASSEMBLY__ */
260
261 #endif /* _ASM_GENERIC_PGTABLE_H */