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, mapped)  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 flush_tlb_fix_spurious_fault
 133 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 134 #endif
 135
 136 #ifndef pgprot_noncached
 137 #define pgprot_noncached(prot)  (prot)
 138 #endif
 139
 140 #ifndef pgprot_writecombine
 141 #define pgprot_writecombine pgprot_noncached
 142 #endif
 143
 144 /*
 145  * When walking page tables, get the address of the next boundary,
 146  * or the end address of the range if that comes earlier.  Although no
 147  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 148  */
 149
 150 #define pgd_addr_end(addr, end)                                         \
 151 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 152         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 153 })
 154
 155 #ifndef pud_addr_end
 156 #define pud_addr_end(addr, end)                                         \
 157 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 158         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 159 })
 160 #endif
 161
 162 #ifndef pmd_addr_end
 163 #define pmd_addr_end(addr, end)                                         \
 164 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 165         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 166 })
 167 #endif
 168
 169 /*
 170  * When walking page tables, we usually want to skip any p?d_none entries;
 171  * and any p?d_bad entries - reporting the error before resetting to none.
 172  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 173  */
 174 void pgd_clear_bad(pgd_t *);
 175 void pud_clear_bad(pud_t *);
 176 void pmd_clear_bad(pmd_t *);
 177
 178 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 179 {
 180         if (pgd_none(*pgd))
 181                 return 1;
 182         if (unlikely(pgd_bad(*pgd))) {
 183                 pgd_clear_bad(pgd);
 184                 return 1;
 185         }
 186         return 0;
 187 }
 188
 189 static inline int pud_none_or_clear_bad(pud_t *pud)
 190 {
 191         if (pud_none(*pud))
 192                 return 1;
 193         if (unlikely(pud_bad(*pud))) {
 194                 pud_clear_bad(pud);
 195                 return 1;
 196         }
 197         return 0;
 198 }
 199
 200 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 201 {
 202         if (pmd_none(*pmd))
 203                 return 1;
 204         if (unlikely(pmd_bad(*pmd))) {
 205                 pmd_clear_bad(pmd);
 206                 return 1;
 207         }
 208         return 0;
 209 }
 210
 211 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 212                                              unsigned long addr,
 213                                              pte_t *ptep)
 214 {
 215         /*
 216          * Get the current pte state, but zero it out to make it
 217          * non-present, preventing the hardware from asynchronously
 218          * updating it.
 219          */
 220         return ptep_get_and_clear(mm, addr, ptep);
 221 }
 222
 223 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 224                                              unsigned long addr,
 225                                              pte_t *ptep, pte_t pte)
 226 {
 227         /*
 228          * The pte is non-present, so there's no hardware state to
 229          * preserve.
 230          */
 231         set_pte_at(mm, addr, ptep, pte);
 232 }
 233
 234 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 235 /*
 236  * Start a pte protection read-modify-write transaction, which
 237  * protects against asynchronous hardware modifications to the pte.
 238  * The intention is not to prevent the hardware from making pte
 239  * updates, but to prevent any updates it may make from being lost.
 240  *
 241  * This does not protect against other software modifications of the
 242  * pte; the appropriate pte lock must be held over the transation.
 243  *
 244  * Note that this interface is intended to be batchable, meaning that
 245  * ptep_modify_prot_commit may not actually update the pte, but merely
 246  * queue the update to be done at some later time.  The update must be
 247  * actually committed before the pte lock is released, however.
 248  */
 249 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 250                                            unsigned long addr,
 251                                            pte_t *ptep)
 252 {
 253         return __ptep_modify_prot_start(mm, addr, ptep);
 254 }
 255
 256 /*
 257  * Commit an update to a pte, leaving any hardware-controlled bits in
 258  * the PTE unmodified.
 259  */
 260 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 261                                            unsigned long addr,
 262                                            pte_t *ptep, pte_t pte)
 263 {
 264         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 265 }
 266 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 267 #endif /* CONFIG_MMU */
 268
 269 /*
 270  * A facility to provide lazy MMU batching.  This allows PTE updates and
 271  * page invalidations to be delayed until a call to leave lazy MMU mode
 272  * is issued.  Some architectures may benefit from doing this, and it is
 273  * beneficial for both shadow and direct mode hypervisors, which may batch
 274  * the PTE updates which happen during this window.  Note that using this
 275  * interface requires that read hazards be removed from the code.  A read
 276  * hazard could result in the direct mode hypervisor case, since the actual
 277  * write to the page tables may not yet have taken place, so reads though
 278  * a raw PTE pointer after it has been modified are not guaranteed to be
 279  * up to date.  This mode can only be entered and left under the protection of
 280  * the page table locks for all page tables which may be modified.  In the UP
 281  * case, this is required so that preemption is disabled, and in the SMP case,
 282  * it must synchronize the delayed page table writes properly on other CPUs.
 283  */
 284 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 285 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 286 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 287 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 288 #endif
 289
 290 /*
 291  * A facility to provide batching of the reload of page tables and
 292  * other process state with the actual context switch code for
 293  * paravirtualized guests.  By convention, only one of the batched
 294  * update (lazy) modes (CPU, MMU) should be active at any given time,
 295  * entry should never be nested, and entry and exits should always be
 296  * paired.  This is for sanity of maintaining and reasoning about the
 297  * kernel code.  In this case, the exit (end of the context switch) is
 298  * in architecture-specific code, and so doesn't need a generic
 299  * definition.
 300  */
 301 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 302 #define arch_start_context_switch(prev) do {} while (0)
 303 #endif
 304
 305 #ifndef __HAVE_PFNMAP_TRACKING
 306 /*
 307  * Interface that can be used by architecture code to keep track of
 308  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 309  *
 310  * track_pfn_vma_new is called when a _new_ pfn mapping is being established
 311  * for physical range indicated by pfn and size.
 312  */
 313 static inline int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 314                                         unsigned long pfn, unsigned long size)
 315 {
 316         return 0;
 317 }
 318
 319 /*
 320  * Interface that can be used by architecture code to keep track of
 321  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 322  *
 323  * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
 324  * copied through copy_page_range().
 325  */
 326 static inline int track_pfn_vma_copy(struct vm_area_struct *vma)
 327 {
 328         return 0;
 329 }
 330
 331 /*
 332  * Interface that can be used by architecture code to keep track of
 333  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 334  *
 335  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 336  * untrack can be called for a specific region indicated by pfn and size or
 337  * can be for the entire vma (in which case size can be zero).
 338  */
 339 static inline void untrack_pfn_vma(struct vm_area_struct *vma,
 340                                         unsigned long pfn, unsigned long size)
 341 {
 342 }
 343 #else
 344 extern int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 345                                 unsigned long pfn, unsigned long size);
 346 extern int track_pfn_vma_copy(struct vm_area_struct *vma);
 347 extern void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
 348                                 unsigned long size);
 349 #endif
 350
 351 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 352 static inline int pmd_trans_huge(pmd_t pmd)
 353 {
 354         return 0;
 355 }
 356 static inline int pmd_trans_splitting(pmd_t pmd)
 357 {
 358         return 0;
 359 }
 360 #endif
 361
 362 #endif /* !__ASSEMBLY__ */
 363
 364 #endif /* _ASM_GENERIC_PGTABLE_H */