| 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 | #define ptep_establish(__vma, __address, __ptep, __entry) \
|
| 19 | do { \
|
| 20 | set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
|
| 21 | flush_tlb_page(__vma, __address); \
|
| 22 | } while (0)
|
| 23 | #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) \
|
| 33 | do { \
|
| 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 |
|
| 98 | #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 |
|
| 107 | |
| 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) \
|
| 114 | do { \
|
| 115 | pte_clear((__mm), (__address), (__ptep)); \
|
| 116 | } while (0)
|
| 117 | #endif
|
| 118 |
|
| 119 | #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
|
| 130 | struct mm_struct;
|
| 131 | static 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)
|
| 144 | #define pte_maybe_dirty(pte) pte_dirty(pte)
|
| 145 | #else
|
| 146 | #define pte_maybe_dirty(pte) (1)
|
| 147 | #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 |
|
| 161 | #ifndef __HAVE_ARCH_MOVE_PTE
|
| 162 | #define move_pte(pte, prot, old_addr, new_addr) (pte)
|
| 163 | #endif
|
| 164 |
|
| 165 | |
| 166 | * 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 | #define arch_flush_lazy_mmu_mode() do {} while (0)
|
| 184 | #endif
|
| 185 |
|
| 186 | |
| 187 | * A facility to provide batching of the reload of page tables with the
|
| 188 | * actual context switch code for paravirtualized guests. By convention,
|
| 189 | * only one of the lazy modes (CPU, MMU) should be active at any given
|
| 190 | * time, entry should never be nested, and entry and exits should always
|
| 191 | * be paired. This is for sanity of maintaining and reasoning about the
|
| 192 | * kernel code.
|
| 193 |
|
| 194 | #ifndef __HAVE_ARCH_ENTER_LAZY_CPU_MODE
|
| 195 | #define arch_enter_lazy_cpu_mode() do {} while (0)
|
| 196 | #define arch_leave_lazy_cpu_mode() do {} while (0)
|
| 197 | #define arch_flush_lazy_cpu_mode() do {} while (0)
|
| 198 | #endif
|
| 199 |
|
| 200 | |
| 201 | * When walking page tables, get the address of the next boundary,
|
| 202 | * or the end address of the range if that comes earlier. Although no
|
| 203 | * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
|
| 204 |
|
| 205 |
|
| 206 | #define pgd_addr_end(addr, end) \
|
| 207 | ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
|
| 208 | (__boundary - 1 < (end) - 1)? __boundary: (end); \
|
| 209 | })
|
| 210 |
|
| 211 | #ifndef pud_addr_end
|
| 212 | #define pud_addr_end(addr, end) \
|
| 213 | ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
|
| 214 | (__boundary - 1 < (end) - 1)? __boundary: (end); \
|
| 215 | })
|
| 216 | #endif
|
| 217 |
|
| 218 | #ifndef pmd_addr_end
|
| 219 | #define pmd_addr_end(addr, end) \
|
| 220 | ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
|
| 221 | (__boundary - 1 < (end) - 1)? __boundary: (end); \
|
| 222 | })
|
| 223 | #endif
|
| 224 |
|
| 225 | |
| 226 | * When walking page tables, we usually want to skip any p?d_none entries;
|
| 227 | * and any p?d_bad entries - reporting the error before resetting to none.
|
| 228 | * Do the tests inline, but report and clear the bad entry in mm/memory.c.
|
| 229 |
|
| 230 | void pgd_clear_bad(pgd_t *);
|
| 231 | void pud_clear_bad(pud_t *);
|
| 232 | void pmd_clear_bad(pmd_t *);
|
| 233 |
|
| 234 | static inline int pgd_none_or_clear_bad(pgd_t *pgd)
|
| 235 | {
|
| 236 | if (pgd_none(*pgd))
|
| 237 | return 1;
|
| 238 | if (unlikely(pgd_bad(*pgd))) {
|
| 239 | pgd_clear_bad(pgd);
|
| 240 | return 1;
|
| 241 | }
|
| 242 | return 0;
|
| 243 | }
|
| 244 |
|
| 245 | static inline int pud_none_or_clear_bad(pud_t *pud)
|
| 246 | {
|
| 247 | if (pud_none(*pud))
|
| 248 | return 1;
|
| 249 | if (unlikely(pud_bad(*pud))) {
|
| 250 | pud_clear_bad(pud);
|
| 251 | return 1;
|
| 252 | }
|
| 253 | return 0;
|
| 254 | }
|
| 255 |
|
| 256 | static inline int pmd_none_or_clear_bad(pmd_t *pmd)
|
| 257 | {
|
| 258 | if (pmd_none(*pmd))
|
| 259 | return 1;
|
| 260 | if (unlikely(pmd_bad(*pmd))) {
|
| 261 | pmd_clear_bad(pmd);
|
| 262 | return 1;
|
| 263 | }
|
| 264 | return 0;
|
| 265 | }
|
| 266 | #endif
|
| 267 |
|
| 268 | #endif
|