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 #include <linux/errno.h>
  10
  11 #if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
  12         CONFIG_PGTABLE_LEVELS
  13 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
  14 #endif
  15
  16 /*
  17  * On almost all architectures and configurations, 0 can be used as the
  18  * upper ceiling to free_pgtables(): on many architectures it has the same
  19  * effect as using TASK_SIZE.  However, there is one configuration which
  20  * must impose a more careful limit, to avoid freeing kernel pgtables.
  21  */
  22 #ifndef USER_PGTABLES_CEILING
  23 #define USER_PGTABLES_CEILING   0UL
  24 #endif
  25
  26 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  27 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  28                                  unsigned long address, pte_t *ptep,
  29                                  pte_t entry, int dirty);
  30 #endif
  31
  32 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  33 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  34                                  unsigned long address, pmd_t *pmdp,
  35                                  pmd_t entry, int dirty);
  36 #endif
  37
  38 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  39 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  40                                             unsigned long address,
  41                                             pte_t *ptep)
  42 {
  43         pte_t pte = *ptep;
  44         int r = 1;
  45         if (!pte_young(pte))
  46                 r = 0;
  47         else
  48                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  49         return r;
  50 }
  51 #endif
  52
  53 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  54 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  55 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  56                                             unsigned long address,
  57                                             pmd_t *pmdp)
  58 {
  59         pmd_t pmd = *pmdp;
  60         int r = 1;
  61         if (!pmd_young(pmd))
  62                 r = 0;
  63         else
  64                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  65         return r;
  66 }
  67 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  68 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  69                                             unsigned long address,
  70                                             pmd_t *pmdp)
  71 {
  72         BUG();
  73         return 0;
  74 }
  75 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  76 #endif
  77
  78 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  79 int ptep_clear_flush_young(struct vm_area_struct *vma,
  80                            unsigned long address, pte_t *ptep);
  81 #endif
  82
  83 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  84 int pmdp_clear_flush_young(struct vm_area_struct *vma,
  85                            unsigned long address, pmd_t *pmdp);
  86 #endif
  87
  88 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  89 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  90                                        unsigned long address,
  91                                        pte_t *ptep)
  92 {
  93         pte_t pte = *ptep;
  94         pte_clear(mm, address, ptep);
  95         return pte;
  96 }
  97 #endif
  98
  99 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
 100 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 101 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
 102                                             unsigned long address,
 103                                             pmd_t *pmdp)
 104 {
 105         pmd_t pmd = *pmdp;
 106         pmd_clear(pmdp);
 107         return pmd;
 108 }
 109 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 110 #endif
 111
 112 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
 113 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 114 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
 115                                             unsigned long address, pmd_t *pmdp,
 116                                             int full)
 117 {
 118         return pmdp_huge_get_and_clear(mm, address, pmdp);
 119 }
 120 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 124 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 125                                             unsigned long address, pte_t *ptep,
 126                                             int full)
 127 {
 128         pte_t pte;
 129         pte = ptep_get_and_clear(mm, address, ptep);
 130         return pte;
 131 }
 132 #endif
 133
 134 /*
 135  * Some architectures may be able to avoid expensive synchronization
 136  * primitives when modifications are made to PTE's which are already
 137  * not present, or in the process of an address space destruction.
 138  */
 139 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 140 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 141                                               unsigned long address,
 142                                               pte_t *ptep,
 143                                               int full)
 144 {
 145         pte_clear(mm, address, ptep);
 146 }
 147 #endif
 148
 149 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 150 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 151                               unsigned long address,
 152                               pte_t *ptep);
 153 #endif
 154
 155 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
 156 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
 157                               unsigned long address,
 158                               pmd_t *pmdp);
 159 #endif
 160
 161 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 162 struct mm_struct;
 163 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 164 {
 165         pte_t old_pte = *ptep;
 166         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 167 }
 168 #endif
 169
 170 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 171 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 172 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 173                                       unsigned long address, pmd_t *pmdp)
 174 {
 175         pmd_t old_pmd = *pmdp;
 176         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 177 }
 178 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 179 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 180                                       unsigned long address, pmd_t *pmdp)
 181 {
 182         BUG();
 183 }
 184 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 185 #endif
 186
 187 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
 188 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
 189                                  unsigned long address, pmd_t *pmdp);
 190 #endif
 191
 192 #ifndef pmdp_collapse_flush
 193 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 194 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 195                                  unsigned long address, pmd_t *pmdp);
 196 #else
 197 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 198                                         unsigned long address,
 199                                         pmd_t *pmdp)
 200 {
 201         BUILD_BUG();
 202         return *pmdp;
 203 }
 204 #define pmdp_collapse_flush pmdp_collapse_flush
 205 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 206 #endif
 207
 208 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 209 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 210                                        pgtable_t pgtable);
 211 #endif
 212
 213 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 214 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 215 #endif
 216
 217 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 218 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 219                             pmd_t *pmdp);
 220 #endif
 221
 222 #ifndef __HAVE_ARCH_PTE_SAME
 223 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 224 {
 225         return pte_val(pte_a) == pte_val(pte_b);
 226 }
 227 #endif
 228
 229 #ifndef __HAVE_ARCH_PTE_UNUSED
 230 /*
 231  * Some architectures provide facilities to virtualization guests
 232  * so that they can flag allocated pages as unused. This allows the
 233  * host to transparently reclaim unused pages. This function returns
 234  * whether the pte's page is unused.
 235  */
 236 static inline int pte_unused(pte_t pte)
 237 {
 238         return 0;
 239 }
 240 #endif
 241
 242 #ifndef __HAVE_ARCH_PMD_SAME
 243 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 244 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 245 {
 246         return pmd_val(pmd_a) == pmd_val(pmd_b);
 247 }
 248 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 249 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 250 {
 251         BUG();
 252         return 0;
 253 }
 254 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 255 #endif
 256
 257 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 258 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 259 #endif
 260
 261 #ifndef __HAVE_ARCH_MOVE_PTE
 262 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 263 #endif
 264
 265 #ifndef pte_accessible
 266 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 267 #endif
 268
 269 #ifndef flush_tlb_fix_spurious_fault
 270 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 271 #endif
 272
 273 #ifndef pgprot_noncached
 274 #define pgprot_noncached(prot)  (prot)
 275 #endif
 276
 277 #ifndef pgprot_writecombine
 278 #define pgprot_writecombine pgprot_noncached
 279 #endif
 280
 281 #ifndef pgprot_writethrough
 282 #define pgprot_writethrough pgprot_noncached
 283 #endif
 284
 285 #ifndef pgprot_device
 286 #define pgprot_device pgprot_noncached
 287 #endif
 288
 289 #ifndef pgprot_modify
 290 #define pgprot_modify pgprot_modify
 291 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 292 {
 293         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 294                 newprot = pgprot_noncached(newprot);
 295         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 296                 newprot = pgprot_writecombine(newprot);
 297         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 298                 newprot = pgprot_device(newprot);
 299         return newprot;
 300 }
 301 #endif
 302
 303 /*
 304  * When walking page tables, get the address of the next boundary,
 305  * or the end address of the range if that comes earlier.  Although no
 306  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 307  */
 308
 309 #define pgd_addr_end(addr, end)                                         \
 310 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 311         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 312 })
 313
 314 #ifndef pud_addr_end
 315 #define pud_addr_end(addr, end)                                         \
 316 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 317         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 318 })
 319 #endif
 320
 321 #ifndef pmd_addr_end
 322 #define pmd_addr_end(addr, end)                                         \
 323 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 324         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 325 })
 326 #endif
 327
 328 /*
 329  * When walking page tables, we usually want to skip any p?d_none entries;
 330  * and any p?d_bad entries - reporting the error before resetting to none.
 331  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 332  */
 333 void pgd_clear_bad(pgd_t *);
 334 void pud_clear_bad(pud_t *);
 335 void pmd_clear_bad(pmd_t *);
 336
 337 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 338 {
 339         if (pgd_none(*pgd))
 340                 return 1;
 341         if (unlikely(pgd_bad(*pgd))) {
 342                 pgd_clear_bad(pgd);
 343                 return 1;
 344         }
 345         return 0;
 346 }
 347
 348 static inline int pud_none_or_clear_bad(pud_t *pud)
 349 {
 350         if (pud_none(*pud))
 351                 return 1;
 352         if (unlikely(pud_bad(*pud))) {
 353                 pud_clear_bad(pud);
 354                 return 1;
 355         }
 356         return 0;
 357 }
 358
 359 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 360 {
 361         if (pmd_none(*pmd))
 362                 return 1;
 363         if (unlikely(pmd_bad(*pmd))) {
 364                 pmd_clear_bad(pmd);
 365                 return 1;
 366         }
 367         return 0;
 368 }
 369
 370 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 371                                              unsigned long addr,
 372                                              pte_t *ptep)
 373 {
 374         /*
 375          * Get the current pte state, but zero it out to make it
 376          * non-present, preventing the hardware from asynchronously
 377          * updating it.
 378          */
 379         return ptep_get_and_clear(mm, addr, ptep);
 380 }
 381
 382 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 383                                              unsigned long addr,
 384                                              pte_t *ptep, pte_t pte)
 385 {
 386         /*
 387          * The pte is non-present, so there's no hardware state to
 388          * preserve.
 389          */
 390         set_pte_at(mm, addr, ptep, pte);
 391 }
 392
 393 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 394 /*
 395  * Start a pte protection read-modify-write transaction, which
 396  * protects against asynchronous hardware modifications to the pte.
 397  * The intention is not to prevent the hardware from making pte
 398  * updates, but to prevent any updates it may make from being lost.
 399  *
 400  * This does not protect against other software modifications of the
 401  * pte; the appropriate pte lock must be held over the transation.
 402  *
 403  * Note that this interface is intended to be batchable, meaning that
 404  * ptep_modify_prot_commit may not actually update the pte, but merely
 405  * queue the update to be done at some later time.  The update must be
 406  * actually committed before the pte lock is released, however.
 407  */
 408 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 409                                            unsigned long addr,
 410                                            pte_t *ptep)
 411 {
 412         return __ptep_modify_prot_start(mm, addr, ptep);
 413 }
 414
 415 /*
 416  * Commit an update to a pte, leaving any hardware-controlled bits in
 417  * the PTE unmodified.
 418  */
 419 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 420                                            unsigned long addr,
 421                                            pte_t *ptep, pte_t pte)
 422 {
 423         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 424 }
 425 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 426 #endif /* CONFIG_MMU */
 427
 428 /*
 429  * A facility to provide lazy MMU batching.  This allows PTE updates and
 430  * page invalidations to be delayed until a call to leave lazy MMU mode
 431  * is issued.  Some architectures may benefit from doing this, and it is
 432  * beneficial for both shadow and direct mode hypervisors, which may batch
 433  * the PTE updates which happen during this window.  Note that using this
 434  * interface requires that read hazards be removed from the code.  A read
 435  * hazard could result in the direct mode hypervisor case, since the actual
 436  * write to the page tables may not yet have taken place, so reads though
 437  * a raw PTE pointer after it has been modified are not guaranteed to be
 438  * up to date.  This mode can only be entered and left under the protection of
 439  * the page table locks for all page tables which may be modified.  In the UP
 440  * case, this is required so that preemption is disabled, and in the SMP case,
 441  * it must synchronize the delayed page table writes properly on other CPUs.
 442  */
 443 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 444 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 445 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 446 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 447 #endif
 448
 449 /*
 450  * A facility to provide batching of the reload of page tables and
 451  * other process state with the actual context switch code for
 452  * paravirtualized guests.  By convention, only one of the batched
 453  * update (lazy) modes (CPU, MMU) should be active at any given time,
 454  * entry should never be nested, and entry and exits should always be
 455  * paired.  This is for sanity of maintaining and reasoning about the
 456  * kernel code.  In this case, the exit (end of the context switch) is
 457  * in architecture-specific code, and so doesn't need a generic
 458  * definition.
 459  */
 460 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 461 #define arch_start_context_switch(prev) do {} while (0)
 462 #endif
 463
 464 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
 465 static inline int pte_soft_dirty(pte_t pte)
 466 {
 467         return 0;
 468 }
 469
 470 static inline int pmd_soft_dirty(pmd_t pmd)
 471 {
 472         return 0;
 473 }
 474
 475 static inline pte_t pte_mksoft_dirty(pte_t pte)
 476 {
 477         return pte;
 478 }
 479
 480 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 481 {
 482         return pmd;
 483 }
 484
 485 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 486 {
 487         return pte;
 488 }
 489
 490 static inline int pte_swp_soft_dirty(pte_t pte)
 491 {
 492         return 0;
 493 }
 494
 495 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 496 {
 497         return pte;
 498 }
 499 #endif
 500
 501 #ifndef __HAVE_PFNMAP_TRACKING
 502 /*
 503  * Interfaces that can be used by architecture code to keep track of
 504  * memory type of pfn mappings specified by the remap_pfn_range,
 505  * vm_insert_pfn.
 506  */
 507
 508 /*
 509  * track_pfn_remap is called when a _new_ pfn mapping is being established
 510  * by remap_pfn_range() for physical range indicated by pfn and size.
 511  */
 512 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 513                                   unsigned long pfn, unsigned long addr,
 514                                   unsigned long size)
 515 {
 516         return 0;
 517 }
 518
 519 /*
 520  * track_pfn_insert is called when a _new_ single pfn is established
 521  * by vm_insert_pfn().
 522  */
 523 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 524                                    unsigned long pfn)
 525 {
 526         return 0;
 527 }
 528
 529 /*
 530  * track_pfn_copy is called when vma that is covering the pfnmap gets
 531  * copied through copy_page_range().
 532  */
 533 static inline int track_pfn_copy(struct vm_area_struct *vma)
 534 {
 535         return 0;
 536 }
 537
 538 /*
 539  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 540  * untrack can be called for a specific region indicated by pfn and size or
 541  * can be for the entire vma (in which case pfn, size are zero).
 542  */
 543 static inline void untrack_pfn(struct vm_area_struct *vma,
 544                                unsigned long pfn, unsigned long size)
 545 {
 546 }
 547 #else
 548 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 549                            unsigned long pfn, unsigned long addr,
 550                            unsigned long size);
 551 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 552                             unsigned long pfn);
 553 extern int track_pfn_copy(struct vm_area_struct *vma);
 554 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 555                         unsigned long size);
 556 #endif
 557
 558 #ifdef __HAVE_COLOR_ZERO_PAGE
 559 static inline int is_zero_pfn(unsigned long pfn)
 560 {
 561         extern unsigned long zero_pfn;
 562         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 563         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 564 }
 565
 566 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 567
 568 #else
 569 static inline int is_zero_pfn(unsigned long pfn)
 570 {
 571         extern unsigned long zero_pfn;
 572         return pfn == zero_pfn;
 573 }
 574
 575 static inline unsigned long my_zero_pfn(unsigned long addr)
 576 {
 577         extern unsigned long zero_pfn;
 578         return zero_pfn;
 579 }
 580 #endif
 581
 582 #ifdef CONFIG_MMU
 583
 584 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 585 static inline int pmd_trans_huge(pmd_t pmd)
 586 {
 587         return 0;
 588 }
 589 static inline int pmd_trans_splitting(pmd_t pmd)
 590 {
 591         return 0;
 592 }
 593 #ifndef __HAVE_ARCH_PMD_WRITE
 594 static inline int pmd_write(pmd_t pmd)
 595 {
 596         BUG();
 597         return 0;
 598 }
 599 #endif /* __HAVE_ARCH_PMD_WRITE */
 600 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 601
 602 #ifndef pmd_read_atomic
 603 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 604 {
 605         /*
 606          * Depend on compiler for an atomic pmd read. NOTE: this is
 607          * only going to work, if the pmdval_t isn't larger than
 608          * an unsigned long.
 609          */
 610         return *pmdp;
 611 }
 612 #endif
 613
 614 #ifndef pmd_move_must_withdraw
 615 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
 616                                          spinlock_t *old_pmd_ptl)
 617 {
 618         /*
 619          * With split pmd lock we also need to move preallocated
 620          * PTE page table if new_pmd is on different PMD page table.
 621          */
 622         return new_pmd_ptl != old_pmd_ptl;
 623 }
 624 #endif
 625
 626 /*
 627  * This function is meant to be used by sites walking pagetables with
 628  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 629  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 630  * into a null pmd and the transhuge page fault can convert a null pmd
 631  * into an hugepmd or into a regular pmd (if the hugepage allocation
 632  * fails). While holding the mmap_sem in read mode the pmd becomes
 633  * stable and stops changing under us only if it's not null and not a
 634  * transhuge pmd. When those races occurs and this function makes a
 635  * difference vs the standard pmd_none_or_clear_bad, the result is
 636  * undefined so behaving like if the pmd was none is safe (because it
 637  * can return none anyway). The compiler level barrier() is critically
 638  * important to compute the two checks atomically on the same pmdval.
 639  *
 640  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 641  * care of reading the pmd atomically to avoid SMP race conditions
 642  * against pmd_populate() when the mmap_sem is hold for reading by the
 643  * caller (a special atomic read not done by "gcc" as in the generic
 644  * version above, is also needed when THP is disabled because the page
 645  * fault can populate the pmd from under us).
 646  */
 647 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 648 {
 649         pmd_t pmdval = pmd_read_atomic(pmd);
 650         /*
 651          * The barrier will stabilize the pmdval in a register or on
 652          * the stack so that it will stop changing under the code.
 653          *
 654          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 655          * pmd_read_atomic is allowed to return a not atomic pmdval
 656          * (for example pointing to an hugepage that has never been
 657          * mapped in the pmd). The below checks will only care about
 658          * the low part of the pmd with 32bit PAE x86 anyway, with the
 659          * exception of pmd_none(). So the important thing is that if
 660          * the low part of the pmd is found null, the high part will
 661          * be also null or the pmd_none() check below would be
 662          * confused.
 663          */
 664 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 665         barrier();
 666 #endif
 667         if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
 668                 return 1;
 669         if (unlikely(pmd_bad(pmdval))) {
 670                 pmd_clear_bad(pmd);
 671                 return 1;
 672         }
 673         return 0;
 674 }
 675
 676 /*
 677  * This is a noop if Transparent Hugepage Support is not built into
 678  * the kernel. Otherwise it is equivalent to
 679  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 680  * places that already verified the pmd is not none and they want to
 681  * walk ptes while holding the mmap sem in read mode (write mode don't
 682  * need this). If THP is not enabled, the pmd can't go away under the
 683  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 684  * run a pmd_trans_unstable before walking the ptes after
 685  * split_huge_page_pmd returns (because it may have run when the pmd
 686  * become null, but then a page fault can map in a THP and not a
 687  * regular page).
 688  */
 689 static inline int pmd_trans_unstable(pmd_t *pmd)
 690 {
 691 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 692         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 693 #else
 694         return 0;
 695 #endif
 696 }
 697
 698 #ifndef CONFIG_NUMA_BALANCING
 699 /*
 700  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 701  * the only case the kernel cares is for NUMA balancing and is only ever set
 702  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 703  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 704  * is the responsibility of the caller to distinguish between PROT_NONE
 705  * protections and NUMA hinting fault protections.
 706  */
 707 static inline int pte_protnone(pte_t pte)
 708 {
 709         return 0;
 710 }
 711
 712 static inline int pmd_protnone(pmd_t pmd)
 713 {
 714         return 0;
 715 }
 716 #endif /* CONFIG_NUMA_BALANCING */
 717
 718 #endif /* CONFIG_MMU */
 719
 720 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 721 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 722 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 723 int pud_clear_huge(pud_t *pud);
 724 int pmd_clear_huge(pmd_t *pmd);
 725 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 726 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 727 {
 728         return 0;
 729 }
 730 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 731 {
 732         return 0;
 733 }
 734 static inline int pud_clear_huge(pud_t *pud)
 735 {
 736         return 0;
 737 }
 738 static inline int pmd_clear_huge(pmd_t *pmd)
 739 {
 740         return 0;
 741 }
 742 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
 743
 744 #endif /* !__ASSEMBLY__ */
 745
 746 #ifndef io_remap_pfn_range
 747 #define io_remap_pfn_range remap_pfn_range
 748 #endif
 749
 750 #endif /* _ASM_GENERIC_PGTABLE_H */