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 #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. We return whether the PTE actually changed, which
  31  * in turn instructs the caller to do things like update__mmu_cache.
  32  * This used to be done in the caller, but sparc needs minor faults to
  33  * force that call on sun4c so we changed this macro slightly
  34  */
  35 #define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
  36 ({                                                                        \
  37         int __changed = !pte_same(*(__ptep), __entry);                    \
  38         if (__changed) {                                                  \
  39                 set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
  40                 flush_tlb_page(__vma, __address);                         \
  41         }                                                                 \
  42         __changed;                                                        \
  43 })
  44 #endif
  45
  46 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  47 #define ptep_test_and_clear_young(__vma, __address, __ptep)             \
  48 ({                                                                      \
  49         pte_t __pte = *(__ptep);                                        \
  50         int r = 1;                                                      \
  51         if (!pte_young(__pte))                                          \
  52                 r = 0;                                                  \
  53         else                                                            \
  54                 set_pte_at((__vma)->vm_mm, (__address),                 \
  55                            (__ptep), pte_mkold(__pte));                 \
  56         r;                                                              \
  57 })
  58 #endif
  59
  60 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  61 #define ptep_clear_flush_young(__vma, __address, __ptep)                \
  62 ({                                                                      \
  63         int __young;                                                    \
  64         __young = ptep_test_and_clear_young(__vma, __address, __ptep);  \
  65         if (__young)                                                    \
  66                 flush_tlb_page(__vma, __address);                       \
  67         __young;                                                        \
  68 })
  69 #endif
  70
  71 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
  72 #define ptep_test_and_clear_dirty(__vma, __address, __ptep)             \
  73 ({                                                                      \
  74         pte_t __pte = *__ptep;                                          \
  75         int r = 1;                                                      \
  76         if (!pte_dirty(__pte))                                          \
  77                 r = 0;                                                  \
  78         else                                                            \
  79                 set_pte_at((__vma)->vm_mm, (__address), (__ptep),       \
  80                            pte_mkclean(__pte));                         \
  81         r;                                                              \
  82 })
  83 #endif
  84
  85 #ifndef __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
  86 #define ptep_clear_flush_dirty(__vma, __address, __ptep)                \
  87 ({                                                                      \
  88         int __dirty;                                                    \
  89         __dirty = ptep_test_and_clear_dirty(__vma, __address, __ptep);  \
  90         if (__dirty)                                                    \
  91                 flush_tlb_page(__vma, __address);                       \
  92         __dirty;                                                        \
  93 })
  94 #endif
  95
  96 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  97 #define ptep_get_and_clear(__mm, __address, __ptep)                     \
  98 ({                                                                      \
  99         pte_t __pte = *(__ptep);                                        \
 100         pte_clear((__mm), (__address), (__ptep));                       \
 101         __pte;                                                          \
 102 })
 103 #endif
 104
 105 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 106 #define ptep_get_and_clear_full(__mm, __address, __ptep, __full)        \
 107 ({                                                                      \
 108         pte_t __pte;                                                    \
 109         __pte = ptep_get_and_clear((__mm), (__address), (__ptep));      \
 110         __pte;                                                          \
 111 })
 112 #endif
 113
 114 /*
 115  * Some architectures may be able to avoid expensive synchronization
 116  * primitives when modifications are made to PTE's which are already
 117  * not present, or in the process of an address space destruction.
 118  */
 119 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 120 #define pte_clear_not_present_full(__mm, __address, __ptep, __full)     \
 121 do {                                                                    \
 122         pte_clear((__mm), (__address), (__ptep));                       \
 123 } while (0)
 124 #endif
 125
 126 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 127 #define ptep_clear_flush(__vma, __address, __ptep)                      \
 128 ({                                                                      \
 129         pte_t __pte;                                                    \
 130         __pte = ptep_get_and_clear((__vma)->vm_mm, __address, __ptep);  \
 131         flush_tlb_page(__vma, __address);                               \
 132         __pte;                                                          \
 133 })
 134 #endif
 135
 136 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 137 struct mm_struct;
 138 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 139 {
 140         pte_t old_pte = *ptep;
 141         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 142 }
 143 #endif
 144
 145 #ifndef __HAVE_ARCH_PTE_SAME
 146 #define pte_same(A,B)   (pte_val(A) == pte_val(B))
 147 #endif
 148
 149 #ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
 150 #define page_test_dirty(page)           (0)
 151 #endif
 152
 153 #ifndef __HAVE_ARCH_PAGE_CLEAR_DIRTY
 154 #define page_clear_dirty(page)          do { } while (0)
 155 #endif
 156
 157 #ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
 158 #define pte_maybe_dirty(pte)            pte_dirty(pte)
 159 #else
 160 #define pte_maybe_dirty(pte)            (1)
 161 #endif
 162
 163 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
 164 #define page_test_and_clear_young(page) (0)
 165 #endif
 166
 167 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 168 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 169 #endif
 170
 171 #ifndef __HAVE_ARCH_LAZY_MMU_PROT_UPDATE
 172 #define lazy_mmu_prot_update(pte)       do { } while (0)
 173 #endif
 174
 175 #ifndef __HAVE_ARCH_MOVE_PTE
 176 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 177 #endif
 178
 179 /*
 180  * A facility to provide lazy MMU batching.  This allows PTE updates and
 181  * page invalidations to be delayed until a call to leave lazy MMU mode
 182  * is issued.  Some architectures may benefit from doing this, and it is
 183  * beneficial for both shadow and direct mode hypervisors, which may batch
 184  * the PTE updates which happen during this window.  Note that using this
 185  * interface requires that read hazards be removed from the code.  A read
 186  * hazard could result in the direct mode hypervisor case, since the actual
 187  * write to the page tables may not yet have taken place, so reads though
 188  * a raw PTE pointer after it has been modified are not guaranteed to be
 189  * up to date.  This mode can only be entered and left under the protection of
 190  * the page table locks for all page tables which may be modified.  In the UP
 191  * case, this is required so that preemption is disabled, and in the SMP case,
 192  * it must synchronize the delayed page table writes properly on other CPUs.
 193  */
 194 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 195 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 196 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 197 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 198 #endif
 199
 200 /*
 201  * A facility to provide batching of the reload of page tables with the
 202  * actual context switch code for paravirtualized guests.  By convention,
 203  * only one of the lazy modes (CPU, MMU) should be active at any given
 204  * time, entry should never be nested, and entry and exits should always
 205  * be paired.  This is for sanity of maintaining and reasoning about the
 206  * kernel code.
 207  */
 208 #ifndef __HAVE_ARCH_ENTER_LAZY_CPU_MODE
 209 #define arch_enter_lazy_cpu_mode()      do {} while (0)
 210 #define arch_leave_lazy_cpu_mode()      do {} while (0)
 211 #define arch_flush_lazy_cpu_mode()      do {} while (0)
 212 #endif
 213
 214 /*
 215  * When walking page tables, get the address of the next boundary,
 216  * or the end address of the range if that comes earlier.  Although no
 217  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 218  */
 219
 220 #define pgd_addr_end(addr, end)                                         \
 221 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 222         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 223 })
 224
 225 #ifndef pud_addr_end
 226 #define pud_addr_end(addr, end)                                         \
 227 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 228         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 229 })
 230 #endif
 231
 232 #ifndef pmd_addr_end
 233 #define pmd_addr_end(addr, end)                                         \
 234 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 235         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 236 })
 237 #endif
 238
 239 /*
 240  * When walking page tables, we usually want to skip any p?d_none entries;
 241  * and any p?d_bad entries - reporting the error before resetting to none.
 242  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 243  */
 244 void pgd_clear_bad(pgd_t *);
 245 void pud_clear_bad(pud_t *);
 246 void pmd_clear_bad(pmd_t *);
 247
 248 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 249 {
 250         if (pgd_none(*pgd))
 251                 return 1;
 252         if (unlikely(pgd_bad(*pgd))) {
 253                 pgd_clear_bad(pgd);
 254                 return 1;
 255         }
 256         return 0;
 257 }
 258
 259 static inline int pud_none_or_clear_bad(pud_t *pud)
 260 {
 261         if (pud_none(*pud))
 262                 return 1;
 263         if (unlikely(pud_bad(*pud))) {
 264                 pud_clear_bad(pud);
 265                 return 1;
 266         }
 267         return 0;
 268 }
 269
 270 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 271 {
 272         if (pmd_none(*pmd))
 273                 return 1;
 274         if (unlikely(pmd_bad(*pmd))) {
 275                 pmd_clear_bad(pmd);
 276                 return 1;
 277         }
 278         return 0;
 279 }
 280 #endif /* !__ASSEMBLY__ */
 281
 282 #endif /* _ASM_GENERIC_PGTABLE_H */