include/asm-generic/pgtable.h

   1 #ifndef _ASM_GENERIC_PGTABLE_H
   2 #define _ASM_GENERIC_PGTABLE_H
   3
   4 #ifndef __ASSEMBLY__
   5 #ifdef CONFIG_MMU
   6
   7 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
   8 /*
   9  * Largely same as above, but only sets the access flags (dirty,
  10  * accessed, and writable). Furthermore, we know it always gets set
  11  * to a "more permissive" setting, which allows most architectures
  12  * to optimize this. We return whether the PTE actually changed, which
  13  * in turn instructs the caller to do things like update__mmu_cache.
  14  * This used to be done in the caller, but sparc needs minor faults to
  15  * force that call on sun4c so we changed this macro slightly
  16  */
  17 #define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
  18 ({                                                                        \
  19         int __changed = !pte_same(*(__ptep), __entry);                    \
  20         if (__changed) {                                                  \
  21                 set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
  22                 flush_tlb_page(__vma, __address);                         \
  23         }                                                                 \
  24         __changed;                                                        \
  25 })
  26 #endif
  27
  28 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  29 #define ptep_test_and_clear_young(__vma, __address, __ptep)             \
  30 ({                                                                      \
  31         pte_t __pte = *(__ptep);                                        \
  32         int r = 1;                                                      \
  33         if (!pte_young(__pte))                                          \
  34                 r = 0;                                                  \
  35         else                                                            \
  36                 set_pte_at((__vma)->vm_mm, (__address),                 \
  37                            (__ptep), pte_mkold(__pte));                 \
  38         r;                                                              \
  39 })
  40 #endif
  41
  42 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  43 #define ptep_clear_flush_young(__vma, __address, __ptep)                \
  44 ({                                                                      \
  45         int __young;                                                    \
  46         __young = ptep_test_and_clear_young(__vma, __address, __ptep);  \
  47         if (__young)                                                    \
  48                 flush_tlb_page(__vma, __address);                       \
  49         __young;                                                        \
  50 })
  51 #endif
  52
  53 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  54 #define ptep_get_and_clear(__mm, __address, __ptep)                     \
  55 ({                                                                      \
  56         pte_t __pte = *(__ptep);                                        \
  57         pte_clear((__mm), (__address), (__ptep));                       \
  58         __pte;                                                          \
  59 })
  60 #endif
  61
  62 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
  63 #define ptep_get_and_clear_full(__mm, __address, __ptep, __full)        \
  64 ({                                                                      \
  65         pte_t __pte;                                                    \
  66         __pte = ptep_get_and_clear((__mm), (__address), (__ptep));      \
  67         __pte;                                                          \
  68 })
  69 #endif
  70
  71 /*
  72  * Some architectures may be able to avoid expensive synchronization
  73  * primitives when modifications are made to PTE's which are already
  74  * not present, or in the process of an address space destruction.
  75  */
  76 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
  77 #define pte_clear_not_present_full(__mm, __address, __ptep, __full)     \
  78 do {                                                                    \
  79         pte_clear((__mm), (__address), (__ptep));                       \
  80 } while (0)
  81 #endif
  82
  83 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
  84 #define ptep_clear_flush(__vma, __address, __ptep)                      \
  85 ({                                                                      \
  86         pte_t __pte;                                                    \
  87         __pte = ptep_get_and_clear((__vma)->vm_mm, __address, __ptep);  \
  88         flush_tlb_page(__vma, __address);                               \
  89         __pte;                                                          \
  90 })
  91 #endif
  92
  93 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
  94 struct mm_struct;
  95 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
  96 {
  97         pte_t old_pte = *ptep;
  98         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
  99 }
 100 #endif
 101
 102 #ifndef __HAVE_ARCH_PTE_SAME
 103 #define pte_same(A,B)   (pte_val(A) == pte_val(B))
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
 107 #define page_test_dirty(page)           (0)
 108 #endif
 109
 110 #ifndef __HAVE_ARCH_PAGE_CLEAR_DIRTY
 111 #define page_clear_dirty(page)          do { } while (0)
 112 #endif
 113
 114 #ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
 115 #define pte_maybe_dirty(pte)            pte_dirty(pte)
 116 #else
 117 #define pte_maybe_dirty(pte)            (1)
 118 #endif
 119
 120 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
 121 #define page_test_and_clear_young(page) (0)
 122 #endif
 123
 124 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 125 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 126 #endif
 127
 128 #ifndef __HAVE_ARCH_MOVE_PTE
 129 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 130 #endif
 131
 132 #ifndef pgprot_noncached
 133 #define pgprot_noncached(prot)  (prot)
 134 #endif
 135
 136 #ifndef pgprot_writecombine
 137 #define pgprot_writecombine pgprot_noncached
 138 #endif
 139
 140 /*
 141  * When walking page tables, get the address of the next boundary,
 142  * or the end address of the range if that comes earlier.  Although no
 143  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 144  */
 145
 146 #define pgd_addr_end(addr, end)                                         \
 147 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 148         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 149 })
 150
 151 #ifndef pud_addr_end
 152 #define pud_addr_end(addr, end)                                         \
 153 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 154         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 155 })
 156 #endif
 157
 158 #ifndef pmd_addr_end
 159 #define pmd_addr_end(addr, end)                                         \
 160 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 161         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 162 })
 163 #endif
 164
 165 /*
 166  * When walking page tables, we usually want to skip any p?d_none entries;
 167  * and any p?d_bad entries - reporting the error before resetting to none.
 168  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 169  */
 170 void pgd_clear_bad(pgd_t *);
 171 void pud_clear_bad(pud_t *);
 172 void pmd_clear_bad(pmd_t *);
 173
 174 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 175 {
 176         if (pgd_none(*pgd))
 177                 return 1;
 178         if (unlikely(pgd_bad(*pgd))) {
 179                 pgd_clear_bad(pgd);
 180                 return 1;
 181         }
 182         return 0;
 183 }
 184
 185 static inline int pud_none_or_clear_bad(pud_t *pud)
 186 {
 187         if (pud_none(*pud))
 188                 return 1;
 189         if (unlikely(pud_bad(*pud))) {
 190                 pud_clear_bad(pud);
 191                 return 1;
 192         }
 193         return 0;
 194 }
 195
 196 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 197 {
 198         if (pmd_none(*pmd))
 199                 return 1;
 200         if (unlikely(pmd_bad(*pmd))) {
 201                 pmd_clear_bad(pmd);
 202                 return 1;
 203         }
 204         return 0;
 205 }
 206
 207 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 208                                              unsigned long addr,
 209                                              pte_t *ptep)
 210 {
 211         /*
 212          * Get the current pte state, but zero it out to make it
 213          * non-present, preventing the hardware from asynchronously
 214          * updating it.
 215          */
 216         return ptep_get_and_clear(mm, addr, ptep);
 217 }
 218
 219 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 220                                              unsigned long addr,
 221                                              pte_t *ptep, pte_t pte)
 222 {
 223         /*
 224          * The pte is non-present, so there's no hardware state to
 225          * preserve.
 226          */
 227         set_pte_at(mm, addr, ptep, pte);
 228 }
 229
 230 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 231 /*
 232  * Start a pte protection read-modify-write transaction, which
 233  * protects against asynchronous hardware modifications to the pte.
 234  * The intention is not to prevent the hardware from making pte
 235  * updates, but to prevent any updates it may make from being lost.
 236  *
 237  * This does not protect against other software modifications of the
 238  * pte; the appropriate pte lock must be held over the transation.
 239  *
 240  * Note that this interface is intended to be batchable, meaning that
 241  * ptep_modify_prot_commit may not actually update the pte, but merely
 242  * queue the update to be done at some later time.  The update must be
 243  * actually committed before the pte lock is released, however.
 244  */
 245 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 246                                            unsigned long addr,
 247                                            pte_t *ptep)
 248 {
 249         return __ptep_modify_prot_start(mm, addr, ptep);
 250 }
 251
 252 /*
 253  * Commit an update to a pte, leaving any hardware-controlled bits in
 254  * the PTE unmodified.
 255  */
 256 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 257                                            unsigned long addr,
 258                                            pte_t *ptep, pte_t pte)
 259 {
 260         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 261 }
 262 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 263 #endif /* CONFIG_MMU */
 264
 265 /*
 266  * A facility to provide lazy MMU batching.  This allows PTE updates and
 267  * page invalidations to be delayed until a call to leave lazy MMU mode
 268  * is issued.  Some architectures may benefit from doing this, and it is
 269  * beneficial for both shadow and direct mode hypervisors, which may batch
 270  * the PTE updates which happen during this window.  Note that using this
 271  * interface requires that read hazards be removed from the code.  A read
 272  * hazard could result in the direct mode hypervisor case, since the actual
 273  * write to the page tables may not yet have taken place, so reads though
 274  * a raw PTE pointer after it has been modified are not guaranteed to be
 275  * up to date.  This mode can only be entered and left under the protection of
 276  * the page table locks for all page tables which may be modified.  In the UP
 277  * case, this is required so that preemption is disabled, and in the SMP case,
 278  * it must synchronize the delayed page table writes properly on other CPUs.
 279  */
 280 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 281 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 282 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 283 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 284 #endif
 285
 286 /*
 287  * A facility to provide batching of the reload of page tables and
 288  * other process state with the actual context switch code for
 289  * paravirtualized guests.  By convention, only one of the batched
 290  * update (lazy) modes (CPU, MMU) should be active at any given time,
 291  * entry should never be nested, and entry and exits should always be
 292  * paired.  This is for sanity of maintaining and reasoning about the
 293  * kernel code.  In this case, the exit (end of the context switch) is
 294  * in architecture-specific code, and so doesn't need a generic
 295  * definition.
 296  */
 297 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 298 #define arch_start_context_switch(prev) do {} while (0)
 299 #endif
 300
 301 #ifndef __HAVE_PFNMAP_TRACKING
 302 /*
 303  * Interface that can be used by architecture code to keep track of
 304  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 305  *
 306  * track_pfn_vma_new is called when a _new_ pfn mapping is being established
 307  * for physical range indicated by pfn and size.
 308  */
 309 static inline int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 310                                         unsigned long pfn, unsigned long size)
 311 {
 312         return 0;
 313 }
 314
 315 /*
 316  * Interface that can be used by architecture code to keep track of
 317  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 318  *
 319  * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
 320  * copied through copy_page_range().
 321  */
 322 static inline int track_pfn_vma_copy(struct vm_area_struct *vma)
 323 {
 324         return 0;
 325 }
 326
 327 /*
 328  * Interface that can be used by architecture code to keep track of
 329  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 330  *
 331  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 332  * untrack can be called for a specific region indicated by pfn and size or
 333  * can be for the entire vma (in which case size can be zero).
 334  */
 335 static inline void untrack_pfn_vma(struct vm_area_struct *vma,
 336                                         unsigned long pfn, unsigned long size)
 337 {
 338 }
 339 #else
 340 extern int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 341                                 unsigned long pfn, unsigned long size);
 342 extern int track_pfn_vma_copy(struct vm_area_struct *vma);
 343 extern void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
 344                                 unsigned long size);
 345 #endif
 346
 347 #endif /* !__ASSEMBLY__ */
 348
 349 #endif /* _ASM_GENERIC_PGTABLE_H */