include/asm-generic/pgtable.h

   1 #ifndef _ASM_GENERIC_PGTABLE_H
   2 #define _ASM_GENERIC_PGTABLE_H
   3
   4 #include <linux/pfn.h>
   5
   6 #ifndef __ASSEMBLY__
   7 #ifdef CONFIG_MMU
   8
   9 #include <linux/mm_types.h>
  10 #include <linux/bug.h>
  11 #include <linux/errno.h>
  12
  13 #if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
  14         CONFIG_PGTABLE_LEVELS
  15 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
  16 #endif
  17
  18 /*
  19  * On almost all architectures and configurations, 0 can be used as the
  20  * upper ceiling to free_pgtables(): on many architectures it has the same
  21  * effect as using TASK_SIZE.  However, there is one configuration which
  22  * must impose a more careful limit, to avoid freeing kernel pgtables.
  23  */
  24 #ifndef USER_PGTABLES_CEILING
  25 #define USER_PGTABLES_CEILING   0UL
  26 #endif
  27
  28 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  29 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  30                                  unsigned long address, pte_t *ptep,
  31                                  pte_t entry, int dirty);
  32 #endif
  33
  34 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  35 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  36 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  37                                  unsigned long address, pmd_t *pmdp,
  38                                  pmd_t entry, int dirty);
  39 #else
  40 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
  41                                         unsigned long address, pmd_t *pmdp,
  42                                         pmd_t entry, int dirty)
  43 {
  44         BUILD_BUG();
  45         return 0;
  46 }
  47 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  48 #endif
  49
  50 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  51 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  52                                             unsigned long address,
  53                                             pte_t *ptep)
  54 {
  55         pte_t pte = *ptep;
  56         int r = 1;
  57         if (!pte_young(pte))
  58                 r = 0;
  59         else
  60                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  61         return r;
  62 }
  63 #endif
  64
  65 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  66 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  67 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  68                                             unsigned long address,
  69                                             pmd_t *pmdp)
  70 {
  71         pmd_t pmd = *pmdp;
  72         int r = 1;
  73         if (!pmd_young(pmd))
  74                 r = 0;
  75         else
  76                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  77         return r;
  78 }
  79 #else
  80 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  81                                             unsigned long address,
  82                                             pmd_t *pmdp)
  83 {
  84         BUILD_BUG();
  85         return 0;
  86 }
  87 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  88 #endif
  89
  90 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  91 int ptep_clear_flush_young(struct vm_area_struct *vma,
  92                            unsigned long address, pte_t *ptep);
  93 #endif
  94
  95 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  96 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  97 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
  98                                   unsigned long address, pmd_t *pmdp);
  99 #else
 100 /*
 101  * Despite relevant to THP only, this API is called from generic rmap code
 102  * under PageTransHuge(), hence needs a dummy implementation for !THP
 103  */
 104 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
 105                                          unsigned long address, pmd_t *pmdp)
 106 {
 107         BUILD_BUG();
 108         return 0;
 109 }
 110 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 111 #endif
 112
 113 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
 114 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
 115                                        unsigned long address,
 116                                        pte_t *ptep)
 117 {
 118         pte_t pte = *ptep;
 119         pte_clear(mm, address, ptep);
 120         return pte;
 121 }
 122 #endif
 123
 124 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
 125 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 126 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
 127                                             unsigned long address,
 128                                             pmd_t *pmdp)
 129 {
 130         pmd_t pmd = *pmdp;
 131         pmd_clear(pmdp);
 132         return pmd;
 133 }
 134 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 135 #endif
 136
 137 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
 138 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 139 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
 140                                             unsigned long address, pmd_t *pmdp,
 141                                             int full)
 142 {
 143         return pmdp_huge_get_and_clear(mm, address, pmdp);
 144 }
 145 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 146 #endif
 147
 148 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 149 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 150                                             unsigned long address, pte_t *ptep,
 151                                             int full)
 152 {
 153         pte_t pte;
 154         pte = ptep_get_and_clear(mm, address, ptep);
 155         return pte;
 156 }
 157 #endif
 158
 159 /*
 160  * Some architectures may be able to avoid expensive synchronization
 161  * primitives when modifications are made to PTE's which are already
 162  * not present, or in the process of an address space destruction.
 163  */
 164 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 165 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 166                                               unsigned long address,
 167                                               pte_t *ptep,
 168                                               int full)
 169 {
 170         pte_clear(mm, address, ptep);
 171 }
 172 #endif
 173
 174 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 175 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 176                               unsigned long address,
 177                               pte_t *ptep);
 178 #endif
 179
 180 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
 181 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
 182                               unsigned long address,
 183                               pmd_t *pmdp);
 184 #endif
 185
 186 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 187 struct mm_struct;
 188 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 189 {
 190         pte_t old_pte = *ptep;
 191         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 192 }
 193 #endif
 194
 195 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 197 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 198                                       unsigned long address, pmd_t *pmdp)
 199 {
 200         pmd_t old_pmd = *pmdp;
 201         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 202 }
 203 #else
 204 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 205                                       unsigned long address, pmd_t *pmdp)
 206 {
 207         BUILD_BUG();
 208 }
 209 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 210 #endif
 211
 212 #ifndef pmdp_collapse_flush
 213 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 214 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 215                                  unsigned long address, pmd_t *pmdp);
 216 #else
 217 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 218                                         unsigned long address,
 219                                         pmd_t *pmdp)
 220 {
 221         BUILD_BUG();
 222         return *pmdp;
 223 }
 224 #define pmdp_collapse_flush pmdp_collapse_flush
 225 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 226 #endif
 227
 228 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 229 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 230                                        pgtable_t pgtable);
 231 #endif
 232
 233 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 234 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 235 #endif
 236
 237 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 238 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 239                             pmd_t *pmdp);
 240 #endif
 241
 242 #ifndef __HAVE_ARCH_PMDP_HUGE_SPLIT_PREPARE
 243 static inline void pmdp_huge_split_prepare(struct vm_area_struct *vma,
 244                                            unsigned long address, pmd_t *pmdp)
 245 {
 246
 247 }
 248 #endif
 249
 250 #ifndef __HAVE_ARCH_PTE_SAME
 251 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 252 {
 253         return pte_val(pte_a) == pte_val(pte_b);
 254 }
 255 #endif
 256
 257 #ifndef __HAVE_ARCH_PTE_UNUSED
 258 /*
 259  * Some architectures provide facilities to virtualization guests
 260  * so that they can flag allocated pages as unused. This allows the
 261  * host to transparently reclaim unused pages. This function returns
 262  * whether the pte's page is unused.
 263  */
 264 static inline int pte_unused(pte_t pte)
 265 {
 266         return 0;
 267 }
 268 #endif
 269
 270 #ifndef __HAVE_ARCH_PMD_SAME
 271 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 272 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 273 {
 274         return pmd_val(pmd_a) == pmd_val(pmd_b);
 275 }
 276 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 277 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 278 {
 279         BUILD_BUG();
 280         return 0;
 281 }
 282 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 283 #endif
 284
 285 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 286 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 287 #endif
 288
 289 #ifndef __HAVE_ARCH_MOVE_PTE
 290 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 291 #endif
 292
 293 #ifndef pte_accessible
 294 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 295 #endif
 296
 297 #ifndef flush_tlb_fix_spurious_fault
 298 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 299 #endif
 300
 301 #ifndef pgprot_noncached
 302 #define pgprot_noncached(prot)  (prot)
 303 #endif
 304
 305 #ifndef pgprot_writecombine
 306 #define pgprot_writecombine pgprot_noncached
 307 #endif
 308
 309 #ifndef pgprot_writethrough
 310 #define pgprot_writethrough pgprot_noncached
 311 #endif
 312
 313 #ifndef pgprot_device
 314 #define pgprot_device pgprot_noncached
 315 #endif
 316
 317 #ifndef pgprot_modify
 318 #define pgprot_modify pgprot_modify
 319 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 320 {
 321         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 322                 newprot = pgprot_noncached(newprot);
 323         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 324                 newprot = pgprot_writecombine(newprot);
 325         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 326                 newprot = pgprot_device(newprot);
 327         return newprot;
 328 }
 329 #endif
 330
 331 /*
 332  * When walking page tables, get the address of the next boundary,
 333  * or the end address of the range if that comes earlier.  Although no
 334  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 335  */
 336
 337 #define pgd_addr_end(addr, end)                                         \
 338 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 339         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 340 })
 341
 342 #ifndef pud_addr_end
 343 #define pud_addr_end(addr, end)                                         \
 344 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 345         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 346 })
 347 #endif
 348
 349 #ifndef pmd_addr_end
 350 #define pmd_addr_end(addr, end)                                         \
 351 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 352         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 353 })
 354 #endif
 355
 356 /*
 357  * When walking page tables, we usually want to skip any p?d_none entries;
 358  * and any p?d_bad entries - reporting the error before resetting to none.
 359  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 360  */
 361 void pgd_clear_bad(pgd_t *);
 362 void pud_clear_bad(pud_t *);
 363 void pmd_clear_bad(pmd_t *);
 364
 365 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 366 {
 367         if (pgd_none(*pgd))
 368                 return 1;
 369         if (unlikely(pgd_bad(*pgd))) {
 370                 pgd_clear_bad(pgd);
 371                 return 1;
 372         }
 373         return 0;
 374 }
 375
 376 static inline int pud_none_or_clear_bad(pud_t *pud)
 377 {
 378         if (pud_none(*pud))
 379                 return 1;
 380         if (unlikely(pud_bad(*pud))) {
 381                 pud_clear_bad(pud);
 382                 return 1;
 383         }
 384         return 0;
 385 }
 386
 387 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 388 {
 389         if (pmd_none(*pmd))
 390                 return 1;
 391         if (unlikely(pmd_bad(*pmd))) {
 392                 pmd_clear_bad(pmd);
 393                 return 1;
 394         }
 395         return 0;
 396 }
 397
 398 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 399                                              unsigned long addr,
 400                                              pte_t *ptep)
 401 {
 402         /*
 403          * Get the current pte state, but zero it out to make it
 404          * non-present, preventing the hardware from asynchronously
 405          * updating it.
 406          */
 407         return ptep_get_and_clear(mm, addr, ptep);
 408 }
 409
 410 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 411                                              unsigned long addr,
 412                                              pte_t *ptep, pte_t pte)
 413 {
 414         /*
 415          * The pte is non-present, so there's no hardware state to
 416          * preserve.
 417          */
 418         set_pte_at(mm, addr, ptep, pte);
 419 }
 420
 421 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 422 /*
 423  * Start a pte protection read-modify-write transaction, which
 424  * protects against asynchronous hardware modifications to the pte.
 425  * The intention is not to prevent the hardware from making pte
 426  * updates, but to prevent any updates it may make from being lost.
 427  *
 428  * This does not protect against other software modifications of the
 429  * pte; the appropriate pte lock must be held over the transation.
 430  *
 431  * Note that this interface is intended to be batchable, meaning that
 432  * ptep_modify_prot_commit may not actually update the pte, but merely
 433  * queue the update to be done at some later time.  The update must be
 434  * actually committed before the pte lock is released, however.
 435  */
 436 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 437                                            unsigned long addr,
 438                                            pte_t *ptep)
 439 {
 440         return __ptep_modify_prot_start(mm, addr, ptep);
 441 }
 442
 443 /*
 444  * Commit an update to a pte, leaving any hardware-controlled bits in
 445  * the PTE unmodified.
 446  */
 447 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 448                                            unsigned long addr,
 449                                            pte_t *ptep, pte_t pte)
 450 {
 451         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 452 }
 453 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 454 #endif /* CONFIG_MMU */
 455
 456 /*
 457  * A facility to provide lazy MMU batching.  This allows PTE updates and
 458  * page invalidations to be delayed until a call to leave lazy MMU mode
 459  * is issued.  Some architectures may benefit from doing this, and it is
 460  * beneficial for both shadow and direct mode hypervisors, which may batch
 461  * the PTE updates which happen during this window.  Note that using this
 462  * interface requires that read hazards be removed from the code.  A read
 463  * hazard could result in the direct mode hypervisor case, since the actual
 464  * write to the page tables may not yet have taken place, so reads though
 465  * a raw PTE pointer after it has been modified are not guaranteed to be
 466  * up to date.  This mode can only be entered and left under the protection of
 467  * the page table locks for all page tables which may be modified.  In the UP
 468  * case, this is required so that preemption is disabled, and in the SMP case,
 469  * it must synchronize the delayed page table writes properly on other CPUs.
 470  */
 471 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 472 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 473 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 474 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 475 #endif
 476
 477 /*
 478  * A facility to provide batching of the reload of page tables and
 479  * other process state with the actual context switch code for
 480  * paravirtualized guests.  By convention, only one of the batched
 481  * update (lazy) modes (CPU, MMU) should be active at any given time,
 482  * entry should never be nested, and entry and exits should always be
 483  * paired.  This is for sanity of maintaining and reasoning about the
 484  * kernel code.  In this case, the exit (end of the context switch) is
 485  * in architecture-specific code, and so doesn't need a generic
 486  * definition.
 487  */
 488 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 489 #define arch_start_context_switch(prev) do {} while (0)
 490 #endif
 491
 492 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
 493 static inline int pte_soft_dirty(pte_t pte)
 494 {
 495         return 0;
 496 }
 497
 498 static inline int pmd_soft_dirty(pmd_t pmd)
 499 {
 500         return 0;
 501 }
 502
 503 static inline pte_t pte_mksoft_dirty(pte_t pte)
 504 {
 505         return pte;
 506 }
 507
 508 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 509 {
 510         return pmd;
 511 }
 512
 513 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 514 {
 515         return pte;
 516 }
 517
 518 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 519 {
 520         return pmd;
 521 }
 522
 523 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 524 {
 525         return pte;
 526 }
 527
 528 static inline int pte_swp_soft_dirty(pte_t pte)
 529 {
 530         return 0;
 531 }
 532
 533 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 534 {
 535         return pte;
 536 }
 537 #endif
 538
 539 #ifndef __HAVE_PFNMAP_TRACKING
 540 /*
 541  * Interfaces that can be used by architecture code to keep track of
 542  * memory type of pfn mappings specified by the remap_pfn_range,
 543  * vm_insert_pfn.
 544  */
 545
 546 /*
 547  * track_pfn_remap is called when a _new_ pfn mapping is being established
 548  * by remap_pfn_range() for physical range indicated by pfn and size.
 549  */
 550 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 551                                   unsigned long pfn, unsigned long addr,
 552                                   unsigned long size)
 553 {
 554         return 0;
 555 }
 556
 557 /*
 558  * track_pfn_insert is called when a _new_ single pfn is established
 559  * by vm_insert_pfn().
 560  */
 561 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 562                                     pfn_t pfn)
 563 {
 564 }
 565
 566 /*
 567  * track_pfn_copy is called when vma that is covering the pfnmap gets
 568  * copied through copy_page_range().
 569  */
 570 static inline int track_pfn_copy(struct vm_area_struct *vma)
 571 {
 572         return 0;
 573 }
 574
 575 /*
 576  * untrack_pfn is called while unmapping a pfnmap for a region.
 577  * untrack can be called for a specific region indicated by pfn and size or
 578  * can be for the entire vma (in which case pfn, size are zero).
 579  */
 580 static inline void untrack_pfn(struct vm_area_struct *vma,
 581                                unsigned long pfn, unsigned long size)
 582 {
 583 }
 584
 585 /*
 586  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
 587  */
 588 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
 589 {
 590 }
 591 #else
 592 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 593                            unsigned long pfn, unsigned long addr,
 594                            unsigned long size);
 595 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 596                              pfn_t pfn);
 597 extern int track_pfn_copy(struct vm_area_struct *vma);
 598 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 599                         unsigned long size);
 600 extern void untrack_pfn_moved(struct vm_area_struct *vma);
 601 #endif
 602
 603 #ifdef __HAVE_COLOR_ZERO_PAGE
 604 static inline int is_zero_pfn(unsigned long pfn)
 605 {
 606         extern unsigned long zero_pfn;
 607         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 608         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 609 }
 610
 611 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 612
 613 #else
 614 static inline int is_zero_pfn(unsigned long pfn)
 615 {
 616         extern unsigned long zero_pfn;
 617         return pfn == zero_pfn;
 618 }
 619
 620 static inline unsigned long my_zero_pfn(unsigned long addr)
 621 {
 622         extern unsigned long zero_pfn;
 623         return zero_pfn;
 624 }
 625 #endif
 626
 627 #ifdef CONFIG_MMU
 628
 629 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 630 static inline int pmd_trans_huge(pmd_t pmd)
 631 {
 632         return 0;
 633 }
 634 #ifndef __HAVE_ARCH_PMD_WRITE
 635 static inline int pmd_write(pmd_t pmd)
 636 {
 637         BUG();
 638         return 0;
 639 }
 640 #endif /* __HAVE_ARCH_PMD_WRITE */
 641 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 642
 643 #ifndef pmd_read_atomic
 644 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 645 {
 646         /*
 647          * Depend on compiler for an atomic pmd read. NOTE: this is
 648          * only going to work, if the pmdval_t isn't larger than
 649          * an unsigned long.
 650          */
 651         return *pmdp;
 652 }
 653 #endif
 654
 655 #ifndef arch_needs_pgtable_deposit
 656 #define arch_needs_pgtable_deposit() (false)
 657 #endif
 658 /*
 659  * This function is meant to be used by sites walking pagetables with
 660  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 661  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 662  * into a null pmd and the transhuge page fault can convert a null pmd
 663  * into an hugepmd or into a regular pmd (if the hugepage allocation
 664  * fails). While holding the mmap_sem in read mode the pmd becomes
 665  * stable and stops changing under us only if it's not null and not a
 666  * transhuge pmd. When those races occurs and this function makes a
 667  * difference vs the standard pmd_none_or_clear_bad, the result is
 668  * undefined so behaving like if the pmd was none is safe (because it
 669  * can return none anyway). The compiler level barrier() is critically
 670  * important to compute the two checks atomically on the same pmdval.
 671  *
 672  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 673  * care of reading the pmd atomically to avoid SMP race conditions
 674  * against pmd_populate() when the mmap_sem is hold for reading by the
 675  * caller (a special atomic read not done by "gcc" as in the generic
 676  * version above, is also needed when THP is disabled because the page
 677  * fault can populate the pmd from under us).
 678  */
 679 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 680 {
 681         pmd_t pmdval = pmd_read_atomic(pmd);
 682         /*
 683          * The barrier will stabilize the pmdval in a register or on
 684          * the stack so that it will stop changing under the code.
 685          *
 686          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 687          * pmd_read_atomic is allowed to return a not atomic pmdval
 688          * (for example pointing to an hugepage that has never been
 689          * mapped in the pmd). The below checks will only care about
 690          * the low part of the pmd with 32bit PAE x86 anyway, with the
 691          * exception of pmd_none(). So the important thing is that if
 692          * the low part of the pmd is found null, the high part will
 693          * be also null or the pmd_none() check below would be
 694          * confused.
 695          */
 696 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 697         barrier();
 698 #endif
 699         if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
 700                 return 1;
 701         if (unlikely(pmd_bad(pmdval))) {
 702                 pmd_clear_bad(pmd);
 703                 return 1;
 704         }
 705         return 0;
 706 }
 707
 708 /*
 709  * This is a noop if Transparent Hugepage Support is not built into
 710  * the kernel. Otherwise it is equivalent to
 711  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 712  * places that already verified the pmd is not none and they want to
 713  * walk ptes while holding the mmap sem in read mode (write mode don't
 714  * need this). If THP is not enabled, the pmd can't go away under the
 715  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 716  * run a pmd_trans_unstable before walking the ptes after
 717  * split_huge_page_pmd returns (because it may have run when the pmd
 718  * become null, but then a page fault can map in a THP and not a
 719  * regular page).
 720  */
 721 static inline int pmd_trans_unstable(pmd_t *pmd)
 722 {
 723 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 724         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 725 #else
 726         return 0;
 727 #endif
 728 }
 729
 730 #ifndef CONFIG_NUMA_BALANCING
 731 /*
 732  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 733  * the only case the kernel cares is for NUMA balancing and is only ever set
 734  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 735  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 736  * is the responsibility of the caller to distinguish between PROT_NONE
 737  * protections and NUMA hinting fault protections.
 738  */
 739 static inline int pte_protnone(pte_t pte)
 740 {
 741         return 0;
 742 }
 743
 744 static inline int pmd_protnone(pmd_t pmd)
 745 {
 746         return 0;
 747 }
 748 #endif /* CONFIG_NUMA_BALANCING */
 749
 750 #endif /* CONFIG_MMU */
 751
 752 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 753 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 754 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 755 int pud_clear_huge(pud_t *pud);
 756 int pmd_clear_huge(pmd_t *pmd);
 757 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 758 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 759 {
 760         return 0;
 761 }
 762 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 763 {
 764         return 0;
 765 }
 766 static inline int pud_clear_huge(pud_t *pud)
 767 {
 768         return 0;
 769 }
 770 static inline int pmd_clear_huge(pmd_t *pmd)
 771 {
 772         return 0;
 773 }
 774 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
 775
 776 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
 777 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 778 /*
 779  * ARCHes with special requirements for evicting THP backing TLB entries can
 780  * implement this. Otherwise also, it can help optimize normal TLB flush in
 781  * THP regime. stock flush_tlb_range() typically has optimization to nuke the
 782  * entire TLB TLB if flush span is greater than a threshold, which will
 783  * likely be true for a single huge page. Thus a single thp flush will
 784  * invalidate the entire TLB which is not desitable.
 785  * e.g. see arch/arc: flush_pmd_tlb_range
 786  */
 787 #define flush_pmd_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
 788 #else
 789 #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
 790 #endif
 791 #endif
 792
 793 struct file;
 794 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
 795                         unsigned long size, pgprot_t *vma_prot);
 796 #endif /* !__ASSEMBLY__ */
 797
 798 #ifndef io_remap_pfn_range
 799 #define io_remap_pfn_range remap_pfn_range
 800 #endif
 801
 802 #ifndef has_transparent_hugepage
 803 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 804 #define has_transparent_hugepage() 1
 805 #else
 806 #define has_transparent_hugepage() 0
 807 #endif
 808 #endif
 809
 810 #endif /* _ASM_GENERIC_PGTABLE_H */