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 #include <linux/mm_types.h>
   8 #include <linux/bug.h>
   9
  10 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  11 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  12                                  unsigned long address, pte_t *ptep,
  13                                  pte_t entry, int dirty);
  14 #endif
  15
  16 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  17 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  18                                  unsigned long address, pmd_t *pmdp,
  19                                  pmd_t entry, int dirty);
  20 #endif
  21
  22 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  23 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  24                                             unsigned long address,
  25                                             pte_t *ptep)
  26 {
  27         pte_t pte = *ptep;
  28         int r = 1;
  29         if (!pte_young(pte))
  30                 r = 0;
  31         else
  32                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  33         return r;
  34 }
  35 #endif
  36
  37 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  38 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  39 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  40                                             unsigned long address,
  41                                             pmd_t *pmdp)
  42 {
  43         pmd_t pmd = *pmdp;
  44         int r = 1;
  45         if (!pmd_young(pmd))
  46                 r = 0;
  47         else
  48                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  49         return r;
  50 }
  51 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  52 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  53                                             unsigned long address,
  54                                             pmd_t *pmdp)
  55 {
  56         BUG();
  57         return 0;
  58 }
  59 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  60 #endif
  61
  62 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  63 int ptep_clear_flush_young(struct vm_area_struct *vma,
  64                            unsigned long address, pte_t *ptep);
  65 #endif
  66
  67 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  68 int pmdp_clear_flush_young(struct vm_area_struct *vma,
  69                            unsigned long address, pmd_t *pmdp);
  70 #endif
  71
  72 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  73 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  74                                        unsigned long address,
  75                                        pte_t *ptep)
  76 {
  77         pte_t pte = *ptep;
  78         pte_clear(mm, address, ptep);
  79         return pte;
  80 }
  81 #endif
  82
  83 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
  84 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  85 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
  86                                        unsigned long address,
  87                                        pmd_t *pmdp)
  88 {
  89         pmd_t pmd = *pmdp;
  90         pmd_clear(pmdp);
  91         return pmd;
  92 }
  93 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  94 #endif
  95
  96 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
  97 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
  98                                             unsigned long address, pte_t *ptep,
  99                                             int full)
 100 {
 101         pte_t pte;
 102         pte = ptep_get_and_clear(mm, address, ptep);
 103         return 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 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 114                                               unsigned long address,
 115                                               pte_t *ptep,
 116                                               int full)
 117 {
 118         pte_clear(mm, address, ptep);
 119 }
 120 #endif
 121
 122 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 123 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 124                               unsigned long address,
 125                               pte_t *ptep);
 126 #endif
 127
 128 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
 129 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
 130                               unsigned long address,
 131                               pmd_t *pmdp);
 132 #endif
 133
 134 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 135 struct mm_struct;
 136 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 137 {
 138         pte_t old_pte = *ptep;
 139         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 140 }
 141 #endif
 142
 143 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 144 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 145 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 146                                       unsigned long address, pmd_t *pmdp)
 147 {
 148         pmd_t old_pmd = *pmdp;
 149         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 150 }
 151 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 152 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 153                                       unsigned long address, pmd_t *pmdp)
 154 {
 155         BUG();
 156 }
 157 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 158 #endif
 159
 160 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
 161 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
 162                                  unsigned long address, pmd_t *pmdp);
 163 #endif
 164
 165 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 166 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
 167 #endif
 168
 169 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 170 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
 171 #endif
 172
 173 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 174 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 175                             pmd_t *pmdp);
 176 #endif
 177
 178 #ifndef __HAVE_ARCH_PTE_SAME
 179 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 180 {
 181         return pte_val(pte_a) == pte_val(pte_b);
 182 }
 183 #endif
 184
 185 #ifndef __HAVE_ARCH_PMD_SAME
 186 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 187 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 188 {
 189         return pmd_val(pmd_a) == pmd_val(pmd_b);
 190 }
 191 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 192 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 193 {
 194         BUG();
 195         return 0;
 196 }
 197 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 198 #endif
 199
 200 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 201 #define page_test_and_clear_dirty(pfn, mapped)  (0)
 202 #endif
 203
 204 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 205 #define pte_maybe_dirty(pte)            pte_dirty(pte)
 206 #else
 207 #define pte_maybe_dirty(pte)            (1)
 208 #endif
 209
 210 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
 211 #define page_test_and_clear_young(pfn) (0)
 212 #endif
 213
 214 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 215 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 216 #endif
 217
 218 #ifndef __HAVE_ARCH_MOVE_PTE
 219 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 220 #endif
 221
 222 #ifndef pte_accessible
 223 # define pte_accessible(pte)            ((void)(pte),1)
 224 #endif
 225
 226 #ifndef flush_tlb_fix_spurious_fault
 227 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 228 #endif
 229
 230 #ifndef pgprot_noncached
 231 #define pgprot_noncached(prot)  (prot)
 232 #endif
 233
 234 #ifndef pgprot_writecombine
 235 #define pgprot_writecombine pgprot_noncached
 236 #endif
 237
 238 /*
 239  * When walking page tables, get the address of the next boundary,
 240  * or the end address of the range if that comes earlier.  Although no
 241  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 242  */
 243
 244 #define pgd_addr_end(addr, end)                                         \
 245 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 246         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 247 })
 248
 249 #ifndef pud_addr_end
 250 #define pud_addr_end(addr, end)                                         \
 251 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 252         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 253 })
 254 #endif
 255
 256 #ifndef pmd_addr_end
 257 #define pmd_addr_end(addr, end)                                         \
 258 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 259         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 260 })
 261 #endif
 262
 263 /*
 264  * When walking page tables, we usually want to skip any p?d_none entries;
 265  * and any p?d_bad entries - reporting the error before resetting to none.
 266  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 267  */
 268 void pgd_clear_bad(pgd_t *);
 269 void pud_clear_bad(pud_t *);
 270 void pmd_clear_bad(pmd_t *);
 271
 272 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 273 {
 274         if (pgd_none(*pgd))
 275                 return 1;
 276         if (unlikely(pgd_bad(*pgd))) {
 277                 pgd_clear_bad(pgd);
 278                 return 1;
 279         }
 280         return 0;
 281 }
 282
 283 static inline int pud_none_or_clear_bad(pud_t *pud)
 284 {
 285         if (pud_none(*pud))
 286                 return 1;
 287         if (unlikely(pud_bad(*pud))) {
 288                 pud_clear_bad(pud);
 289                 return 1;
 290         }
 291         return 0;
 292 }
 293
 294 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 295 {
 296         if (pmd_none(*pmd))
 297                 return 1;
 298         if (unlikely(pmd_bad(*pmd))) {
 299                 pmd_clear_bad(pmd);
 300                 return 1;
 301         }
 302         return 0;
 303 }
 304
 305 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 306                                              unsigned long addr,
 307                                              pte_t *ptep)
 308 {
 309         /*
 310          * Get the current pte state, but zero it out to make it
 311          * non-present, preventing the hardware from asynchronously
 312          * updating it.
 313          */
 314         return ptep_get_and_clear(mm, addr, ptep);
 315 }
 316
 317 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 318                                              unsigned long addr,
 319                                              pte_t *ptep, pte_t pte)
 320 {
 321         /*
 322          * The pte is non-present, so there's no hardware state to
 323          * preserve.
 324          */
 325         set_pte_at(mm, addr, ptep, pte);
 326 }
 327
 328 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 329 /*
 330  * Start a pte protection read-modify-write transaction, which
 331  * protects against asynchronous hardware modifications to the pte.
 332  * The intention is not to prevent the hardware from making pte
 333  * updates, but to prevent any updates it may make from being lost.
 334  *
 335  * This does not protect against other software modifications of the
 336  * pte; the appropriate pte lock must be held over the transation.
 337  *
 338  * Note that this interface is intended to be batchable, meaning that
 339  * ptep_modify_prot_commit may not actually update the pte, but merely
 340  * queue the update to be done at some later time.  The update must be
 341  * actually committed before the pte lock is released, however.
 342  */
 343 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 344                                            unsigned long addr,
 345                                            pte_t *ptep)
 346 {
 347         return __ptep_modify_prot_start(mm, addr, ptep);
 348 }
 349
 350 /*
 351  * Commit an update to a pte, leaving any hardware-controlled bits in
 352  * the PTE unmodified.
 353  */
 354 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 355                                            unsigned long addr,
 356                                            pte_t *ptep, pte_t pte)
 357 {
 358         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 359 }
 360 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 361 #endif /* CONFIG_MMU */
 362
 363 /*
 364  * A facility to provide lazy MMU batching.  This allows PTE updates and
 365  * page invalidations to be delayed until a call to leave lazy MMU mode
 366  * is issued.  Some architectures may benefit from doing this, and it is
 367  * beneficial for both shadow and direct mode hypervisors, which may batch
 368  * the PTE updates which happen during this window.  Note that using this
 369  * interface requires that read hazards be removed from the code.  A read
 370  * hazard could result in the direct mode hypervisor case, since the actual
 371  * write to the page tables may not yet have taken place, so reads though
 372  * a raw PTE pointer after it has been modified are not guaranteed to be
 373  * up to date.  This mode can only be entered and left under the protection of
 374  * the page table locks for all page tables which may be modified.  In the UP
 375  * case, this is required so that preemption is disabled, and in the SMP case,
 376  * it must synchronize the delayed page table writes properly on other CPUs.
 377  */
 378 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 379 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 380 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 381 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 382 #endif
 383
 384 /*
 385  * A facility to provide batching of the reload of page tables and
 386  * other process state with the actual context switch code for
 387  * paravirtualized guests.  By convention, only one of the batched
 388  * update (lazy) modes (CPU, MMU) should be active at any given time,
 389  * entry should never be nested, and entry and exits should always be
 390  * paired.  This is for sanity of maintaining and reasoning about the
 391  * kernel code.  In this case, the exit (end of the context switch) is
 392  * in architecture-specific code, and so doesn't need a generic
 393  * definition.
 394  */
 395 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 396 #define arch_start_context_switch(prev) do {} while (0)
 397 #endif
 398
 399 #ifndef __HAVE_PFNMAP_TRACKING
 400 /*
 401  * Interfaces that can be used by architecture code to keep track of
 402  * memory type of pfn mappings specified by the remap_pfn_range,
 403  * vm_insert_pfn.
 404  */
 405
 406 /*
 407  * track_pfn_remap is called when a _new_ pfn mapping is being established
 408  * by remap_pfn_range() for physical range indicated by pfn and size.
 409  */
 410 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 411                                   unsigned long pfn, unsigned long addr,
 412                                   unsigned long size)
 413 {
 414         return 0;
 415 }
 416
 417 /*
 418  * track_pfn_insert is called when a _new_ single pfn is established
 419  * by vm_insert_pfn().
 420  */
 421 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 422                                    unsigned long pfn)
 423 {
 424         return 0;
 425 }
 426
 427 /*
 428  * track_pfn_copy is called when vma that is covering the pfnmap gets
 429  * copied through copy_page_range().
 430  */
 431 static inline int track_pfn_copy(struct vm_area_struct *vma)
 432 {
 433         return 0;
 434 }
 435
 436 /*
 437  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 438  * untrack can be called for a specific region indicated by pfn and size or
 439  * can be for the entire vma (in which case pfn, size are zero).
 440  */
 441 static inline void untrack_pfn(struct vm_area_struct *vma,
 442                                unsigned long pfn, unsigned long size)
 443 {
 444 }
 445 #else
 446 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 447                            unsigned long pfn, unsigned long addr,
 448                            unsigned long size);
 449 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 450                             unsigned long pfn);
 451 extern int track_pfn_copy(struct vm_area_struct *vma);
 452 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 453                         unsigned long size);
 454 #endif
 455
 456 #ifdef __HAVE_COLOR_ZERO_PAGE
 457 static inline int is_zero_pfn(unsigned long pfn)
 458 {
 459         extern unsigned long zero_pfn;
 460         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 461         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 462 }
 463
 464 static inline unsigned long my_zero_pfn(unsigned long addr)
 465 {
 466         return page_to_pfn(ZERO_PAGE(addr));
 467 }
 468 #else
 469 static inline int is_zero_pfn(unsigned long pfn)
 470 {
 471         extern unsigned long zero_pfn;
 472         return pfn == zero_pfn;
 473 }
 474
 475 static inline unsigned long my_zero_pfn(unsigned long addr)
 476 {
 477         extern unsigned long zero_pfn;
 478         return zero_pfn;
 479 }
 480 #endif
 481
 482 #ifdef CONFIG_MMU
 483
 484 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 485 static inline int pmd_trans_huge(pmd_t pmd)
 486 {
 487         return 0;
 488 }
 489 static inline int pmd_trans_splitting(pmd_t pmd)
 490 {
 491         return 0;
 492 }
 493 #ifndef __HAVE_ARCH_PMD_WRITE
 494 static inline int pmd_write(pmd_t pmd)
 495 {
 496         BUG();
 497         return 0;
 498 }
 499 #endif /* __HAVE_ARCH_PMD_WRITE */
 500 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 501
 502 #ifndef pmd_read_atomic
 503 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 504 {
 505         /*
 506          * Depend on compiler for an atomic pmd read. NOTE: this is
 507          * only going to work, if the pmdval_t isn't larger than
 508          * an unsigned long.
 509          */
 510         return *pmdp;
 511 }
 512 #endif
 513
 514 /*
 515  * This function is meant to be used by sites walking pagetables with
 516  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 517  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 518  * into a null pmd and the transhuge page fault can convert a null pmd
 519  * into an hugepmd or into a regular pmd (if the hugepage allocation
 520  * fails). While holding the mmap_sem in read mode the pmd becomes
 521  * stable and stops changing under us only if it's not null and not a
 522  * transhuge pmd. When those races occurs and this function makes a
 523  * difference vs the standard pmd_none_or_clear_bad, the result is
 524  * undefined so behaving like if the pmd was none is safe (because it
 525  * can return none anyway). The compiler level barrier() is critically
 526  * important to compute the two checks atomically on the same pmdval.
 527  *
 528  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 529  * care of reading the pmd atomically to avoid SMP race conditions
 530  * against pmd_populate() when the mmap_sem is hold for reading by the
 531  * caller (a special atomic read not done by "gcc" as in the generic
 532  * version above, is also needed when THP is disabled because the page
 533  * fault can populate the pmd from under us).
 534  */
 535 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 536 {
 537         pmd_t pmdval = pmd_read_atomic(pmd);
 538         /*
 539          * The barrier will stabilize the pmdval in a register or on
 540          * the stack so that it will stop changing under the code.
 541          *
 542          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 543          * pmd_read_atomic is allowed to return a not atomic pmdval
 544          * (for example pointing to an hugepage that has never been
 545          * mapped in the pmd). The below checks will only care about
 546          * the low part of the pmd with 32bit PAE x86 anyway, with the
 547          * exception of pmd_none(). So the important thing is that if
 548          * the low part of the pmd is found null, the high part will
 549          * be also null or the pmd_none() check below would be
 550          * confused.
 551          */
 552 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 553         barrier();
 554 #endif
 555         if (pmd_none(pmdval))
 556                 return 1;
 557         if (unlikely(pmd_bad(pmdval))) {
 558                 if (!pmd_trans_huge(pmdval))
 559                         pmd_clear_bad(pmd);
 560                 return 1;
 561         }
 562         return 0;
 563 }
 564
 565 /*
 566  * This is a noop if Transparent Hugepage Support is not built into
 567  * the kernel. Otherwise it is equivalent to
 568  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 569  * places that already verified the pmd is not none and they want to
 570  * walk ptes while holding the mmap sem in read mode (write mode don't
 571  * need this). If THP is not enabled, the pmd can't go away under the
 572  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 573  * run a pmd_trans_unstable before walking the ptes after
 574  * split_huge_page_pmd returns (because it may have run when the pmd
 575  * become null, but then a page fault can map in a THP and not a
 576  * regular page).
 577  */
 578 static inline int pmd_trans_unstable(pmd_t *pmd)
 579 {
 580 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 581         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 582 #else
 583         return 0;
 584 #endif
 585 }
 586
 587 #ifdef CONFIG_NUMA_BALANCING
 588 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
 589 /*
 590  * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
 591  * same bit too). It's set only when _PAGE_PRESET is not set and it's
 592  * never set if _PAGE_PRESENT is set.
 593  *
 594  * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
 595  * fault triggers on those regions if pte/pmd_numa returns true
 596  * (because _PAGE_PRESENT is not set).
 597  */
 598 #ifndef pte_numa
 599 static inline int pte_numa(pte_t pte)
 600 {
 601         return (pte_flags(pte) &
 602                 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
 603 }
 604 #endif
 605
 606 #ifndef pmd_numa
 607 static inline int pmd_numa(pmd_t pmd)
 608 {
 609         return (pmd_flags(pmd) &
 610                 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
 611 }
 612 #endif
 613
 614 /*
 615  * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
 616  * because they're called by the NUMA hinting minor page fault. If we
 617  * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
 618  * would be forced to set it later while filling the TLB after we
 619  * return to userland. That would trigger a second write to memory
 620  * that we optimize away by setting _PAGE_ACCESSED here.
 621  */
 622 #ifndef pte_mknonnuma
 623 static inline pte_t pte_mknonnuma(pte_t pte)
 624 {
 625         pte = pte_clear_flags(pte, _PAGE_NUMA);
 626         return pte_set_flags(pte, _PAGE_PRESENT|_PAGE_ACCESSED);
 627 }
 628 #endif
 629
 630 #ifndef pmd_mknonnuma
 631 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
 632 {
 633         pmd = pmd_clear_flags(pmd, _PAGE_NUMA);
 634         return pmd_set_flags(pmd, _PAGE_PRESENT|_PAGE_ACCESSED);
 635 }
 636 #endif
 637
 638 #ifndef pte_mknuma
 639 static inline pte_t pte_mknuma(pte_t pte)
 640 {
 641         pte = pte_set_flags(pte, _PAGE_NUMA);
 642         return pte_clear_flags(pte, _PAGE_PRESENT);
 643 }
 644 #endif
 645
 646 #ifndef pmd_mknuma
 647 static inline pmd_t pmd_mknuma(pmd_t pmd)
 648 {
 649         pmd = pmd_set_flags(pmd, _PAGE_NUMA);
 650         return pmd_clear_flags(pmd, _PAGE_PRESENT);
 651 }
 652 #endif
 653 #else
 654 extern int pte_numa(pte_t pte);
 655 extern int pmd_numa(pmd_t pmd);
 656 extern pte_t pte_mknonnuma(pte_t pte);
 657 extern pmd_t pmd_mknonnuma(pmd_t pmd);
 658 extern pte_t pte_mknuma(pte_t pte);
 659 extern pmd_t pmd_mknuma(pmd_t pmd);
 660 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
 661 #else
 662 static inline int pmd_numa(pmd_t pmd)
 663 {
 664         return 0;
 665 }
 666
 667 static inline int pte_numa(pte_t pte)
 668 {
 669         return 0;
 670 }
 671
 672 static inline pte_t pte_mknonnuma(pte_t pte)
 673 {
 674         return pte;
 675 }
 676
 677 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
 678 {
 679         return pmd;
 680 }
 681
 682 static inline pte_t pte_mknuma(pte_t pte)
 683 {
 684         return pte;
 685 }
 686
 687 static inline pmd_t pmd_mknuma(pmd_t pmd)
 688 {
 689         return pmd;
 690 }
 691 #endif /* CONFIG_NUMA_BALANCING */
 692
 693 #endif /* CONFIG_MMU */
 694
 695 #endif /* !__ASSEMBLY__ */
 696
 697 #endif /* _ASM_GENERIC_PGTABLE_H */