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 /*
  11  * On almost all architectures and configurations, 0 can be used as the
  12  * upper ceiling to free_pgtables(): on many architectures it has the same
  13  * effect as using TASK_SIZE.  However, there is one configuration which
  14  * must impose a more careful limit, to avoid freeing kernel pgtables.
  15  */
  16 #ifndef USER_PGTABLES_CEILING
  17 #define USER_PGTABLES_CEILING   0UL
  18 #endif
  19
  20 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  21 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  22                                  unsigned long address, pte_t *ptep,
  23                                  pte_t entry, int dirty);
  24 #endif
  25
  26 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  27 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  28                                  unsigned long address, pmd_t *pmdp,
  29                                  pmd_t entry, int dirty);
  30 #endif
  31
  32 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  33 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  34                                             unsigned long address,
  35                                             pte_t *ptep)
  36 {
  37         pte_t pte = *ptep;
  38         int r = 1;
  39         if (!pte_young(pte))
  40                 r = 0;
  41         else
  42                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  43         return r;
  44 }
  45 #endif
  46
  47 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  48 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  49 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  50                                             unsigned long address,
  51                                             pmd_t *pmdp)
  52 {
  53         pmd_t pmd = *pmdp;
  54         int r = 1;
  55         if (!pmd_young(pmd))
  56                 r = 0;
  57         else
  58                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  59         return r;
  60 }
  61 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  62 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  63                                             unsigned long address,
  64                                             pmd_t *pmdp)
  65 {
  66         BUG();
  67         return 0;
  68 }
  69 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  70 #endif
  71
  72 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  73 int ptep_clear_flush_young(struct vm_area_struct *vma,
  74                            unsigned long address, pte_t *ptep);
  75 #endif
  76
  77 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  78 int pmdp_clear_flush_young(struct vm_area_struct *vma,
  79                            unsigned long address, pmd_t *pmdp);
  80 #endif
  81
  82 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  83 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  84                                        unsigned long address,
  85                                        pte_t *ptep)
  86 {
  87         pte_t pte = *ptep;
  88         pte_clear(mm, address, ptep);
  89         return pte;
  90 }
  91 #endif
  92
  93 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
  94 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  95 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
  96                                        unsigned long address,
  97                                        pmd_t *pmdp)
  98 {
  99         pmd_t pmd = *pmdp;
 100         pmd_clear(pmdp);
 101         return pmd;
 102 }
 103 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 107 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 108                                             unsigned long address, pte_t *ptep,
 109                                             int full)
 110 {
 111         pte_t pte;
 112         pte = ptep_get_and_clear(mm, address, ptep);
 113         return pte;
 114 }
 115 #endif
 116
 117 /*
 118  * Some architectures may be able to avoid expensive synchronization
 119  * primitives when modifications are made to PTE's which are already
 120  * not present, or in the process of an address space destruction.
 121  */
 122 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 123 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 124                                               unsigned long address,
 125                                               pte_t *ptep,
 126                                               int full)
 127 {
 128         pte_clear(mm, address, ptep);
 129 }
 130 #endif
 131
 132 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 133 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 134                               unsigned long address,
 135                               pte_t *ptep);
 136 #endif
 137
 138 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
 139 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
 140                               unsigned long address,
 141                               pmd_t *pmdp);
 142 #endif
 143
 144 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 145 struct mm_struct;
 146 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 147 {
 148         pte_t old_pte = *ptep;
 149         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 150 }
 151 #endif
 152
 153 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 154 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 155 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 156                                       unsigned long address, pmd_t *pmdp)
 157 {
 158         pmd_t old_pmd = *pmdp;
 159         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 160 }
 161 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 162 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 163                                       unsigned long address, pmd_t *pmdp)
 164 {
 165         BUG();
 166 }
 167 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 168 #endif
 169
 170 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
 171 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
 172                                  unsigned long address, pmd_t *pmdp);
 173 #endif
 174
 175 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 176 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 177                                        pgtable_t pgtable);
 178 #endif
 179
 180 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 181 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 182 #endif
 183
 184 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 185 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 186                             pmd_t *pmdp);
 187 #endif
 188
 189 #ifndef __HAVE_ARCH_PTE_SAME
 190 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 191 {
 192         return pte_val(pte_a) == pte_val(pte_b);
 193 }
 194 #endif
 195
 196 #ifndef __HAVE_ARCH_PTE_UNUSED
 197 /*
 198  * Some architectures provide facilities to virtualization guests
 199  * so that they can flag allocated pages as unused. This allows the
 200  * host to transparently reclaim unused pages. This function returns
 201  * whether the pte's page is unused.
 202  */
 203 static inline int pte_unused(pte_t pte)
 204 {
 205         return 0;
 206 }
 207 #endif
 208
 209 #ifndef __HAVE_ARCH_PMD_SAME
 210 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 211 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 212 {
 213         return pmd_val(pmd_a) == pmd_val(pmd_b);
 214 }
 215 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 216 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 217 {
 218         BUG();
 219         return 0;
 220 }
 221 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 222 #endif
 223
 224 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 225 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 226 #endif
 227
 228 #ifndef __HAVE_ARCH_MOVE_PTE
 229 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 230 #endif
 231
 232 #ifndef pte_accessible
 233 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 234 #endif
 235
 236 #ifndef pte_present_nonuma
 237 #define pte_present_nonuma(pte) pte_present(pte)
 238 #endif
 239
 240 #ifndef flush_tlb_fix_spurious_fault
 241 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 242 #endif
 243
 244 #ifndef pgprot_noncached
 245 #define pgprot_noncached(prot)  (prot)
 246 #endif
 247
 248 #ifndef pgprot_writecombine
 249 #define pgprot_writecombine pgprot_noncached
 250 #endif
 251
 252 #ifndef pgprot_device
 253 #define pgprot_device pgprot_noncached
 254 #endif
 255
 256 #ifndef pgprot_modify
 257 #define pgprot_modify pgprot_modify
 258 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 259 {
 260         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 261                 newprot = pgprot_noncached(newprot);
 262         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 263                 newprot = pgprot_writecombine(newprot);
 264         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 265                 newprot = pgprot_device(newprot);
 266         return newprot;
 267 }
 268 #endif
 269
 270 /*
 271  * When walking page tables, get the address of the next boundary,
 272  * or the end address of the range if that comes earlier.  Although no
 273  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 274  */
 275
 276 #define pgd_addr_end(addr, end)                                         \
 277 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 278         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 279 })
 280
 281 #ifndef pud_addr_end
 282 #define pud_addr_end(addr, end)                                         \
 283 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 284         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 285 })
 286 #endif
 287
 288 #ifndef pmd_addr_end
 289 #define pmd_addr_end(addr, end)                                         \
 290 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 291         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 292 })
 293 #endif
 294
 295 /*
 296  * When walking page tables, we usually want to skip any p?d_none entries;
 297  * and any p?d_bad entries - reporting the error before resetting to none.
 298  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 299  */
 300 void pgd_clear_bad(pgd_t *);
 301 void pud_clear_bad(pud_t *);
 302 void pmd_clear_bad(pmd_t *);
 303
 304 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 305 {
 306         if (pgd_none(*pgd))
 307                 return 1;
 308         if (unlikely(pgd_bad(*pgd))) {
 309                 pgd_clear_bad(pgd);
 310                 return 1;
 311         }
 312         return 0;
 313 }
 314
 315 static inline int pud_none_or_clear_bad(pud_t *pud)
 316 {
 317         if (pud_none(*pud))
 318                 return 1;
 319         if (unlikely(pud_bad(*pud))) {
 320                 pud_clear_bad(pud);
 321                 return 1;
 322         }
 323         return 0;
 324 }
 325
 326 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 327 {
 328         if (pmd_none(*pmd))
 329                 return 1;
 330         if (unlikely(pmd_bad(*pmd))) {
 331                 pmd_clear_bad(pmd);
 332                 return 1;
 333         }
 334         return 0;
 335 }
 336
 337 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 338                                              unsigned long addr,
 339                                              pte_t *ptep)
 340 {
 341         /*
 342          * Get the current pte state, but zero it out to make it
 343          * non-present, preventing the hardware from asynchronously
 344          * updating it.
 345          */
 346         return ptep_get_and_clear(mm, addr, ptep);
 347 }
 348
 349 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 350                                              unsigned long addr,
 351                                              pte_t *ptep, pte_t pte)
 352 {
 353         /*
 354          * The pte is non-present, so there's no hardware state to
 355          * preserve.
 356          */
 357         set_pte_at(mm, addr, ptep, pte);
 358 }
 359
 360 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 361 /*
 362  * Start a pte protection read-modify-write transaction, which
 363  * protects against asynchronous hardware modifications to the pte.
 364  * The intention is not to prevent the hardware from making pte
 365  * updates, but to prevent any updates it may make from being lost.
 366  *
 367  * This does not protect against other software modifications of the
 368  * pte; the appropriate pte lock must be held over the transation.
 369  *
 370  * Note that this interface is intended to be batchable, meaning that
 371  * ptep_modify_prot_commit may not actually update the pte, but merely
 372  * queue the update to be done at some later time.  The update must be
 373  * actually committed before the pte lock is released, however.
 374  */
 375 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 376                                            unsigned long addr,
 377                                            pte_t *ptep)
 378 {
 379         return __ptep_modify_prot_start(mm, addr, ptep);
 380 }
 381
 382 /*
 383  * Commit an update to a pte, leaving any hardware-controlled bits in
 384  * the PTE unmodified.
 385  */
 386 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 387                                            unsigned long addr,
 388                                            pte_t *ptep, pte_t pte)
 389 {
 390         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 391 }
 392 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 393 #endif /* CONFIG_MMU */
 394
 395 /*
 396  * A facility to provide lazy MMU batching.  This allows PTE updates and
 397  * page invalidations to be delayed until a call to leave lazy MMU mode
 398  * is issued.  Some architectures may benefit from doing this, and it is
 399  * beneficial for both shadow and direct mode hypervisors, which may batch
 400  * the PTE updates which happen during this window.  Note that using this
 401  * interface requires that read hazards be removed from the code.  A read
 402  * hazard could result in the direct mode hypervisor case, since the actual
 403  * write to the page tables may not yet have taken place, so reads though
 404  * a raw PTE pointer after it has been modified are not guaranteed to be
 405  * up to date.  This mode can only be entered and left under the protection of
 406  * the page table locks for all page tables which may be modified.  In the UP
 407  * case, this is required so that preemption is disabled, and in the SMP case,
 408  * it must synchronize the delayed page table writes properly on other CPUs.
 409  */
 410 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 411 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 412 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 413 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 414 #endif
 415
 416 /*
 417  * A facility to provide batching of the reload of page tables and
 418  * other process state with the actual context switch code for
 419  * paravirtualized guests.  By convention, only one of the batched
 420  * update (lazy) modes (CPU, MMU) should be active at any given time,
 421  * entry should never be nested, and entry and exits should always be
 422  * paired.  This is for sanity of maintaining and reasoning about the
 423  * kernel code.  In this case, the exit (end of the context switch) is
 424  * in architecture-specific code, and so doesn't need a generic
 425  * definition.
 426  */
 427 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 428 #define arch_start_context_switch(prev) do {} while (0)
 429 #endif
 430
 431 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
 432 static inline int pte_soft_dirty(pte_t pte)
 433 {
 434         return 0;
 435 }
 436
 437 static inline int pmd_soft_dirty(pmd_t pmd)
 438 {
 439         return 0;
 440 }
 441
 442 static inline pte_t pte_mksoft_dirty(pte_t pte)
 443 {
 444         return pte;
 445 }
 446
 447 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 448 {
 449         return pmd;
 450 }
 451
 452 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 453 {
 454         return pte;
 455 }
 456
 457 static inline int pte_swp_soft_dirty(pte_t pte)
 458 {
 459         return 0;
 460 }
 461
 462 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 463 {
 464         return pte;
 465 }
 466
 467 static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
 468 {
 469        return pte;
 470 }
 471
 472 static inline pte_t pte_file_mksoft_dirty(pte_t pte)
 473 {
 474        return pte;
 475 }
 476
 477 static inline int pte_file_soft_dirty(pte_t pte)
 478 {
 479        return 0;
 480 }
 481 #endif
 482
 483 #ifndef __HAVE_PFNMAP_TRACKING
 484 /*
 485  * Interfaces that can be used by architecture code to keep track of
 486  * memory type of pfn mappings specified by the remap_pfn_range,
 487  * vm_insert_pfn.
 488  */
 489
 490 /*
 491  * track_pfn_remap is called when a _new_ pfn mapping is being established
 492  * by remap_pfn_range() for physical range indicated by pfn and size.
 493  */
 494 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 495                                   unsigned long pfn, unsigned long addr,
 496                                   unsigned long size)
 497 {
 498         return 0;
 499 }
 500
 501 /*
 502  * track_pfn_insert is called when a _new_ single pfn is established
 503  * by vm_insert_pfn().
 504  */
 505 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 506                                    unsigned long pfn)
 507 {
 508         return 0;
 509 }
 510
 511 /*
 512  * track_pfn_copy is called when vma that is covering the pfnmap gets
 513  * copied through copy_page_range().
 514  */
 515 static inline int track_pfn_copy(struct vm_area_struct *vma)
 516 {
 517         return 0;
 518 }
 519
 520 /*
 521  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 522  * untrack can be called for a specific region indicated by pfn and size or
 523  * can be for the entire vma (in which case pfn, size are zero).
 524  */
 525 static inline void untrack_pfn(struct vm_area_struct *vma,
 526                                unsigned long pfn, unsigned long size)
 527 {
 528 }
 529 #else
 530 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 531                            unsigned long pfn, unsigned long addr,
 532                            unsigned long size);
 533 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 534                             unsigned long pfn);
 535 extern int track_pfn_copy(struct vm_area_struct *vma);
 536 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 537                         unsigned long size);
 538 #endif
 539
 540 #ifdef __HAVE_COLOR_ZERO_PAGE
 541 static inline int is_zero_pfn(unsigned long pfn)
 542 {
 543         extern unsigned long zero_pfn;
 544         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 545         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 546 }
 547
 548 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 549
 550 #else
 551 static inline int is_zero_pfn(unsigned long pfn)
 552 {
 553         extern unsigned long zero_pfn;
 554         return pfn == zero_pfn;
 555 }
 556
 557 static inline unsigned long my_zero_pfn(unsigned long addr)
 558 {
 559         extern unsigned long zero_pfn;
 560         return zero_pfn;
 561 }
 562 #endif
 563
 564 #ifdef CONFIG_MMU
 565
 566 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 567 static inline int pmd_trans_huge(pmd_t pmd)
 568 {
 569         return 0;
 570 }
 571 static inline int pmd_trans_splitting(pmd_t pmd)
 572 {
 573         return 0;
 574 }
 575 #ifndef __HAVE_ARCH_PMD_WRITE
 576 static inline int pmd_write(pmd_t pmd)
 577 {
 578         BUG();
 579         return 0;
 580 }
 581 #endif /* __HAVE_ARCH_PMD_WRITE */
 582 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 583
 584 #ifndef pmd_read_atomic
 585 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 586 {
 587         /*
 588          * Depend on compiler for an atomic pmd read. NOTE: this is
 589          * only going to work, if the pmdval_t isn't larger than
 590          * an unsigned long.
 591          */
 592         return *pmdp;
 593 }
 594 #endif
 595
 596 #ifndef pmd_move_must_withdraw
 597 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
 598                                          spinlock_t *old_pmd_ptl)
 599 {
 600         /*
 601          * With split pmd lock we also need to move preallocated
 602          * PTE page table if new_pmd is on different PMD page table.
 603          */
 604         return new_pmd_ptl != old_pmd_ptl;
 605 }
 606 #endif
 607
 608 /*
 609  * This function is meant to be used by sites walking pagetables with
 610  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 611  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 612  * into a null pmd and the transhuge page fault can convert a null pmd
 613  * into an hugepmd or into a regular pmd (if the hugepage allocation
 614  * fails). While holding the mmap_sem in read mode the pmd becomes
 615  * stable and stops changing under us only if it's not null and not a
 616  * transhuge pmd. When those races occurs and this function makes a
 617  * difference vs the standard pmd_none_or_clear_bad, the result is
 618  * undefined so behaving like if the pmd was none is safe (because it
 619  * can return none anyway). The compiler level barrier() is critically
 620  * important to compute the two checks atomically on the same pmdval.
 621  *
 622  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 623  * care of reading the pmd atomically to avoid SMP race conditions
 624  * against pmd_populate() when the mmap_sem is hold for reading by the
 625  * caller (a special atomic read not done by "gcc" as in the generic
 626  * version above, is also needed when THP is disabled because the page
 627  * fault can populate the pmd from under us).
 628  */
 629 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 630 {
 631         pmd_t pmdval = pmd_read_atomic(pmd);
 632         /*
 633          * The barrier will stabilize the pmdval in a register or on
 634          * the stack so that it will stop changing under the code.
 635          *
 636          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 637          * pmd_read_atomic is allowed to return a not atomic pmdval
 638          * (for example pointing to an hugepage that has never been
 639          * mapped in the pmd). The below checks will only care about
 640          * the low part of the pmd with 32bit PAE x86 anyway, with the
 641          * exception of pmd_none(). So the important thing is that if
 642          * the low part of the pmd is found null, the high part will
 643          * be also null or the pmd_none() check below would be
 644          * confused.
 645          */
 646 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 647         barrier();
 648 #endif
 649         if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
 650                 return 1;
 651         if (unlikely(pmd_bad(pmdval))) {
 652                 pmd_clear_bad(pmd);
 653                 return 1;
 654         }
 655         return 0;
 656 }
 657
 658 /*
 659  * This is a noop if Transparent Hugepage Support is not built into
 660  * the kernel. Otherwise it is equivalent to
 661  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 662  * places that already verified the pmd is not none and they want to
 663  * walk ptes while holding the mmap sem in read mode (write mode don't
 664  * need this). If THP is not enabled, the pmd can't go away under the
 665  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 666  * run a pmd_trans_unstable before walking the ptes after
 667  * split_huge_page_pmd returns (because it may have run when the pmd
 668  * become null, but then a page fault can map in a THP and not a
 669  * regular page).
 670  */
 671 static inline int pmd_trans_unstable(pmd_t *pmd)
 672 {
 673 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 674         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 675 #else
 676         return 0;
 677 #endif
 678 }
 679
 680 #ifdef CONFIG_NUMA_BALANCING
 681 /*
 682  * _PAGE_NUMA distinguishes between an unmapped page table entry, an entry that
 683  * is protected for PROT_NONE and a NUMA hinting fault entry. If the
 684  * architecture defines __PAGE_PROTNONE then it should take that into account
 685  * but those that do not can rely on the fact that the NUMA hinting scanner
 686  * skips inaccessible VMAs.
 687  *
 688  * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
 689  * fault triggers on those regions if pte/pmd_numa returns true
 690  * (because _PAGE_PRESENT is not set).
 691  */
 692 #ifndef pte_numa
 693 static inline int pte_numa(pte_t pte)
 694 {
 695         return ptenuma_flags(pte) == _PAGE_NUMA;
 696 }
 697 #endif
 698
 699 #ifndef pmd_numa
 700 static inline int pmd_numa(pmd_t pmd)
 701 {
 702         return pmdnuma_flags(pmd) == _PAGE_NUMA;
 703 }
 704 #endif
 705
 706 /*
 707  * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
 708  * because they're called by the NUMA hinting minor page fault. If we
 709  * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
 710  * would be forced to set it later while filling the TLB after we
 711  * return to userland. That would trigger a second write to memory
 712  * that we optimize away by setting _PAGE_ACCESSED here.
 713  */
 714 #ifndef pte_mknonnuma
 715 static inline pte_t pte_mknonnuma(pte_t pte)
 716 {
 717         pteval_t val = pte_val(pte);
 718
 719         val &= ~_PAGE_NUMA;
 720         val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
 721         return __pte(val);
 722 }
 723 #endif
 724
 725 #ifndef pmd_mknonnuma
 726 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
 727 {
 728         pmdval_t val = pmd_val(pmd);
 729
 730         val &= ~_PAGE_NUMA;
 731         val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
 732
 733         return __pmd(val);
 734 }
 735 #endif
 736
 737 #ifndef pte_mknuma
 738 static inline pte_t pte_mknuma(pte_t pte)
 739 {
 740         pteval_t val = pte_val(pte);
 741
 742         VM_BUG_ON(!(val & _PAGE_PRESENT));
 743
 744         val &= ~_PAGE_PRESENT;
 745         val |= _PAGE_NUMA;
 746
 747         return __pte(val);
 748 }
 749 #endif
 750
 751 #ifndef ptep_set_numa
 752 static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
 753                                  pte_t *ptep)
 754 {
 755         pte_t ptent = *ptep;
 756
 757         ptent = pte_mknuma(ptent);
 758         set_pte_at(mm, addr, ptep, ptent);
 759         return;
 760 }
 761 #endif
 762
 763 #ifndef pmd_mknuma
 764 static inline pmd_t pmd_mknuma(pmd_t pmd)
 765 {
 766         pmdval_t val = pmd_val(pmd);
 767
 768         val &= ~_PAGE_PRESENT;
 769         val |= _PAGE_NUMA;
 770
 771         return __pmd(val);
 772 }
 773 #endif
 774
 775 #ifndef pmdp_set_numa
 776 static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
 777                                  pmd_t *pmdp)
 778 {
 779         pmd_t pmd = *pmdp;
 780
 781         pmd = pmd_mknuma(pmd);
 782         set_pmd_at(mm, addr, pmdp, pmd);
 783         return;
 784 }
 785 #endif
 786 #else
 787 static inline int pmd_numa(pmd_t pmd)
 788 {
 789         return 0;
 790 }
 791
 792 static inline int pte_numa(pte_t pte)
 793 {
 794         return 0;
 795 }
 796
 797 static inline pte_t pte_mknonnuma(pte_t pte)
 798 {
 799         return pte;
 800 }
 801
 802 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
 803 {
 804         return pmd;
 805 }
 806
 807 static inline pte_t pte_mknuma(pte_t pte)
 808 {
 809         return pte;
 810 }
 811
 812 static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
 813                                  pte_t *ptep)
 814 {
 815         return;
 816 }
 817
 818
 819 static inline pmd_t pmd_mknuma(pmd_t pmd)
 820 {
 821         return pmd;
 822 }
 823
 824 static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
 825                                  pmd_t *pmdp)
 826 {
 827         return ;
 828 }
 829 #endif /* CONFIG_NUMA_BALANCING */
 830
 831 #endif /* CONFIG_MMU */
 832
 833 #endif /* !__ASSEMBLY__ */
 834
 835 #ifndef io_remap_pfn_range
 836 #define io_remap_pfn_range remap_pfn_range
 837 #endif
 838
 839 #endif /* _ASM_GENERIC_PGTABLE_H */