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 extern int pudp_set_access_flags(struct vm_area_struct *vma,
  40                                  unsigned long address, pud_t *pudp,
  41                                  pud_t entry, int dirty);
  42 #else
  43 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
  44                                         unsigned long address, pmd_t *pmdp,
  45                                         pmd_t entry, int dirty)
  46 {
  47         BUILD_BUG();
  48         return 0;
  49 }
  50 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
  51                                         unsigned long address, pud_t *pudp,
  52                                         pud_t entry, int dirty)
  53 {
  54         BUILD_BUG();
  55         return 0;
  56 }
  57 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  58 #endif
  59
  60 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  61 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  62                                             unsigned long address,
  63                                             pte_t *ptep)
  64 {
  65         pte_t pte = *ptep;
  66         int r = 1;
  67         if (!pte_young(pte))
  68                 r = 0;
  69         else
  70                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  71         return r;
  72 }
  73 #endif
  74
  75 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  76 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  77 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  78                                             unsigned long address,
  79                                             pmd_t *pmdp)
  80 {
  81         pmd_t pmd = *pmdp;
  82         int r = 1;
  83         if (!pmd_young(pmd))
  84                 r = 0;
  85         else
  86                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  87         return r;
  88 }
  89 #else
  90 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  91                                             unsigned long address,
  92                                             pmd_t *pmdp)
  93 {
  94         BUILD_BUG();
  95         return 0;
  96 }
  97 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  98 #endif
  99
 100 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
 101 int ptep_clear_flush_young(struct vm_area_struct *vma,
 102                            unsigned long address, pte_t *ptep);
 103 #endif
 104
 105 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
 106 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 107 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
 108                                   unsigned long address, pmd_t *pmdp);
 109 #else
 110 /*
 111  * Despite relevant to THP only, this API is called from generic rmap code
 112  * under PageTransHuge(), hence needs a dummy implementation for !THP
 113  */
 114 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
 115                                          unsigned long address, pmd_t *pmdp)
 116 {
 117         BUILD_BUG();
 118         return 0;
 119 }
 120 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
 124 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
 125                                        unsigned long address,
 126                                        pte_t *ptep)
 127 {
 128         pte_t pte = *ptep;
 129         pte_clear(mm, address, ptep);
 130         return pte;
 131 }
 132 #endif
 133
 134 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 135 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
 136 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
 137                                             unsigned long address,
 138                                             pmd_t *pmdp)
 139 {
 140         pmd_t pmd = *pmdp;
 141         pmd_clear(pmdp);
 142         return pmd;
 143 }
 144 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
 145 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
 146 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
 147                                             unsigned long address,
 148                                             pud_t *pudp)
 149 {
 150         pud_t pud = *pudp;
 151
 152         pud_clear(pudp);
 153         return pud;
 154 }
 155 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
 156 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 157
 158 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 159 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
 160 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
 161                                             unsigned long address, pmd_t *pmdp,
 162                                             int full)
 163 {
 164         return pmdp_huge_get_and_clear(mm, address, pmdp);
 165 }
 166 #endif
 167
 168 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
 169 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
 170                                             unsigned long address, pud_t *pudp,
 171                                             int full)
 172 {
 173         return pudp_huge_get_and_clear(mm, address, pudp);
 174 }
 175 #endif
 176 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 177
 178 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
 179 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 180                                             unsigned long address, pte_t *ptep,
 181                                             int full)
 182 {
 183         pte_t pte;
 184         pte = ptep_get_and_clear(mm, address, ptep);
 185         return pte;
 186 }
 187 #endif
 188
 189 /*
 190  * Some architectures may be able to avoid expensive synchronization
 191  * primitives when modifications are made to PTE's which are already
 192  * not present, or in the process of an address space destruction.
 193  */
 194 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 195 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 196                                               unsigned long address,
 197                                               pte_t *ptep,
 198                                               int full)
 199 {
 200         pte_clear(mm, address, ptep);
 201 }
 202 #endif
 203
 204 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 205 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 206                               unsigned long address,
 207                               pte_t *ptep);
 208 #endif
 209
 210 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
 211 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
 212                               unsigned long address,
 213                               pmd_t *pmdp);
 214 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
 215                               unsigned long address,
 216                               pud_t *pudp);
 217 #endif
 218
 219 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 220 struct mm_struct;
 221 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 222 {
 223         pte_t old_pte = *ptep;
 224         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 225 }
 226 #endif
 227
 228 #ifndef pte_savedwrite
 229 #define pte_savedwrite pte_write
 230 #endif
 231
 232 #ifndef pte_mk_savedwrite
 233 #define pte_mk_savedwrite pte_mkwrite
 234 #endif
 235
 236 #ifndef pmd_savedwrite
 237 #define pmd_savedwrite pmd_write
 238 #endif
 239
 240 #ifndef pmd_mk_savedwrite
 241 #define pmd_mk_savedwrite pmd_mkwrite
 242 #endif
 243
 244 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 246 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 247                                       unsigned long address, pmd_t *pmdp)
 248 {
 249         pmd_t old_pmd = *pmdp;
 250         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 251 }
 252 #else
 253 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 254                                       unsigned long address, pmd_t *pmdp)
 255 {
 256         BUILD_BUG();
 257 }
 258 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 259 #endif
 260 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
 261 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 262 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 263                                       unsigned long address, pud_t *pudp)
 264 {
 265         pud_t old_pud = *pudp;
 266
 267         set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
 268 }
 269 #else
 270 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 271                                       unsigned long address, pud_t *pudp)
 272 {
 273         BUILD_BUG();
 274 }
 275 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 276 #endif
 277
 278 #ifndef pmdp_collapse_flush
 279 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 280 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 281                                  unsigned long address, pmd_t *pmdp);
 282 #else
 283 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 284                                         unsigned long address,
 285                                         pmd_t *pmdp)
 286 {
 287         BUILD_BUG();
 288         return *pmdp;
 289 }
 290 #define pmdp_collapse_flush pmdp_collapse_flush
 291 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 292 #endif
 293
 294 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 295 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 296                                        pgtable_t pgtable);
 297 #endif
 298
 299 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 300 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 301 #endif
 302
 303 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 304 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 305                             pmd_t *pmdp);
 306 #endif
 307
 308 #ifndef __HAVE_ARCH_PMDP_HUGE_SPLIT_PREPARE
 309 static inline void pmdp_huge_split_prepare(struct vm_area_struct *vma,
 310                                            unsigned long address, pmd_t *pmdp)
 311 {
 312
 313 }
 314 #endif
 315
 316 #ifndef __HAVE_ARCH_PTE_SAME
 317 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 318 {
 319         return pte_val(pte_a) == pte_val(pte_b);
 320 }
 321 #endif
 322
 323 #ifndef __HAVE_ARCH_PTE_UNUSED
 324 /*
 325  * Some architectures provide facilities to virtualization guests
 326  * so that they can flag allocated pages as unused. This allows the
 327  * host to transparently reclaim unused pages. This function returns
 328  * whether the pte's page is unused.
 329  */
 330 static inline int pte_unused(pte_t pte)
 331 {
 332         return 0;
 333 }
 334 #endif
 335
 336 #ifndef __HAVE_ARCH_PMD_SAME
 337 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 338 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 339 {
 340         return pmd_val(pmd_a) == pmd_val(pmd_b);
 341 }
 342
 343 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 344 {
 345         return pud_val(pud_a) == pud_val(pud_b);
 346 }
 347 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 348 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 349 {
 350         BUILD_BUG();
 351         return 0;
 352 }
 353
 354 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 355 {
 356         BUILD_BUG();
 357         return 0;
 358 }
 359 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 360 #endif
 361
 362 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 363 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 364 #endif
 365
 366 #ifndef __HAVE_ARCH_MOVE_PTE
 367 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 368 #endif
 369
 370 #ifndef pte_accessible
 371 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 372 #endif
 373
 374 #ifndef flush_tlb_fix_spurious_fault
 375 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 376 #endif
 377
 378 #ifndef pgprot_noncached
 379 #define pgprot_noncached(prot)  (prot)
 380 #endif
 381
 382 #ifndef pgprot_writecombine
 383 #define pgprot_writecombine pgprot_noncached
 384 #endif
 385
 386 #ifndef pgprot_writethrough
 387 #define pgprot_writethrough pgprot_noncached
 388 #endif
 389
 390 #ifndef pgprot_device
 391 #define pgprot_device pgprot_noncached
 392 #endif
 393
 394 #ifndef pgprot_modify
 395 #define pgprot_modify pgprot_modify
 396 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 397 {
 398         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 399                 newprot = pgprot_noncached(newprot);
 400         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 401                 newprot = pgprot_writecombine(newprot);
 402         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 403                 newprot = pgprot_device(newprot);
 404         return newprot;
 405 }
 406 #endif
 407
 408 /*
 409  * When walking page tables, get the address of the next boundary,
 410  * or the end address of the range if that comes earlier.  Although no
 411  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 412  */
 413
 414 #define pgd_addr_end(addr, end)                                         \
 415 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 416         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 417 })
 418
 419 #ifndef pud_addr_end
 420 #define pud_addr_end(addr, end)                                         \
 421 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 422         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 423 })
 424 #endif
 425
 426 #ifndef pmd_addr_end
 427 #define pmd_addr_end(addr, end)                                         \
 428 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 429         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 430 })
 431 #endif
 432
 433 /*
 434  * When walking page tables, we usually want to skip any p?d_none entries;
 435  * and any p?d_bad entries - reporting the error before resetting to none.
 436  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 437  */
 438 void pgd_clear_bad(pgd_t *);
 439 void pud_clear_bad(pud_t *);
 440 void pmd_clear_bad(pmd_t *);
 441
 442 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 443 {
 444         if (pgd_none(*pgd))
 445                 return 1;
 446         if (unlikely(pgd_bad(*pgd))) {
 447                 pgd_clear_bad(pgd);
 448                 return 1;
 449         }
 450         return 0;
 451 }
 452
 453 static inline int pud_none_or_clear_bad(pud_t *pud)
 454 {
 455         if (pud_none(*pud))
 456                 return 1;
 457         if (unlikely(pud_bad(*pud))) {
 458                 pud_clear_bad(pud);
 459                 return 1;
 460         }
 461         return 0;
 462 }
 463
 464 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 465 {
 466         if (pmd_none(*pmd))
 467                 return 1;
 468         if (unlikely(pmd_bad(*pmd))) {
 469                 pmd_clear_bad(pmd);
 470                 return 1;
 471         }
 472         return 0;
 473 }
 474
 475 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 476                                              unsigned long addr,
 477                                              pte_t *ptep)
 478 {
 479         /*
 480          * Get the current pte state, but zero it out to make it
 481          * non-present, preventing the hardware from asynchronously
 482          * updating it.
 483          */
 484         return ptep_get_and_clear(mm, addr, ptep);
 485 }
 486
 487 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 488                                              unsigned long addr,
 489                                              pte_t *ptep, pte_t pte)
 490 {
 491         /*
 492          * The pte is non-present, so there's no hardware state to
 493          * preserve.
 494          */
 495         set_pte_at(mm, addr, ptep, pte);
 496 }
 497
 498 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 499 /*
 500  * Start a pte protection read-modify-write transaction, which
 501  * protects against asynchronous hardware modifications to the pte.
 502  * The intention is not to prevent the hardware from making pte
 503  * updates, but to prevent any updates it may make from being lost.
 504  *
 505  * This does not protect against other software modifications of the
 506  * pte; the appropriate pte lock must be held over the transation.
 507  *
 508  * Note that this interface is intended to be batchable, meaning that
 509  * ptep_modify_prot_commit may not actually update the pte, but merely
 510  * queue the update to be done at some later time.  The update must be
 511  * actually committed before the pte lock is released, however.
 512  */
 513 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 514                                            unsigned long addr,
 515                                            pte_t *ptep)
 516 {
 517         return __ptep_modify_prot_start(mm, addr, ptep);
 518 }
 519
 520 /*
 521  * Commit an update to a pte, leaving any hardware-controlled bits in
 522  * the PTE unmodified.
 523  */
 524 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 525                                            unsigned long addr,
 526                                            pte_t *ptep, pte_t pte)
 527 {
 528         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 529 }
 530 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 531 #endif /* CONFIG_MMU */
 532
 533 /*
 534  * A facility to provide lazy MMU batching.  This allows PTE updates and
 535  * page invalidations to be delayed until a call to leave lazy MMU mode
 536  * is issued.  Some architectures may benefit from doing this, and it is
 537  * beneficial for both shadow and direct mode hypervisors, which may batch
 538  * the PTE updates which happen during this window.  Note that using this
 539  * interface requires that read hazards be removed from the code.  A read
 540  * hazard could result in the direct mode hypervisor case, since the actual
 541  * write to the page tables may not yet have taken place, so reads though
 542  * a raw PTE pointer after it has been modified are not guaranteed to be
 543  * up to date.  This mode can only be entered and left under the protection of
 544  * the page table locks for all page tables which may be modified.  In the UP
 545  * case, this is required so that preemption is disabled, and in the SMP case,
 546  * it must synchronize the delayed page table writes properly on other CPUs.
 547  */
 548 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 549 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 550 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 551 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 552 #endif
 553
 554 /*
 555  * A facility to provide batching of the reload of page tables and
 556  * other process state with the actual context switch code for
 557  * paravirtualized guests.  By convention, only one of the batched
 558  * update (lazy) modes (CPU, MMU) should be active at any given time,
 559  * entry should never be nested, and entry and exits should always be
 560  * paired.  This is for sanity of maintaining and reasoning about the
 561  * kernel code.  In this case, the exit (end of the context switch) is
 562  * in architecture-specific code, and so doesn't need a generic
 563  * definition.
 564  */
 565 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 566 #define arch_start_context_switch(prev) do {} while (0)
 567 #endif
 568
 569 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
 570 static inline int pte_soft_dirty(pte_t pte)
 571 {
 572         return 0;
 573 }
 574
 575 static inline int pmd_soft_dirty(pmd_t pmd)
 576 {
 577         return 0;
 578 }
 579
 580 static inline pte_t pte_mksoft_dirty(pte_t pte)
 581 {
 582         return pte;
 583 }
 584
 585 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 586 {
 587         return pmd;
 588 }
 589
 590 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 591 {
 592         return pte;
 593 }
 594
 595 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 596 {
 597         return pmd;
 598 }
 599
 600 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 601 {
 602         return pte;
 603 }
 604
 605 static inline int pte_swp_soft_dirty(pte_t pte)
 606 {
 607         return 0;
 608 }
 609
 610 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 611 {
 612         return pte;
 613 }
 614 #endif
 615
 616 #ifndef __HAVE_PFNMAP_TRACKING
 617 /*
 618  * Interfaces that can be used by architecture code to keep track of
 619  * memory type of pfn mappings specified by the remap_pfn_range,
 620  * vm_insert_pfn.
 621  */
 622
 623 /*
 624  * track_pfn_remap is called when a _new_ pfn mapping is being established
 625  * by remap_pfn_range() for physical range indicated by pfn and size.
 626  */
 627 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 628                                   unsigned long pfn, unsigned long addr,
 629                                   unsigned long size)
 630 {
 631         return 0;
 632 }
 633
 634 /*
 635  * track_pfn_insert is called when a _new_ single pfn is established
 636  * by vm_insert_pfn().
 637  */
 638 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 639                                     pfn_t pfn)
 640 {
 641 }
 642
 643 /*
 644  * track_pfn_copy is called when vma that is covering the pfnmap gets
 645  * copied through copy_page_range().
 646  */
 647 static inline int track_pfn_copy(struct vm_area_struct *vma)
 648 {
 649         return 0;
 650 }
 651
 652 /*
 653  * untrack_pfn is called while unmapping a pfnmap for a region.
 654  * untrack can be called for a specific region indicated by pfn and size or
 655  * can be for the entire vma (in which case pfn, size are zero).
 656  */
 657 static inline void untrack_pfn(struct vm_area_struct *vma,
 658                                unsigned long pfn, unsigned long size)
 659 {
 660 }
 661
 662 /*
 663  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
 664  */
 665 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
 666 {
 667 }
 668 #else
 669 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 670                            unsigned long pfn, unsigned long addr,
 671                            unsigned long size);
 672 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 673                              pfn_t pfn);
 674 extern int track_pfn_copy(struct vm_area_struct *vma);
 675 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 676                         unsigned long size);
 677 extern void untrack_pfn_moved(struct vm_area_struct *vma);
 678 #endif
 679
 680 #ifdef __HAVE_COLOR_ZERO_PAGE
 681 static inline int is_zero_pfn(unsigned long pfn)
 682 {
 683         extern unsigned long zero_pfn;
 684         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 685         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 686 }
 687
 688 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 689
 690 #else
 691 static inline int is_zero_pfn(unsigned long pfn)
 692 {
 693         extern unsigned long zero_pfn;
 694         return pfn == zero_pfn;
 695 }
 696
 697 static inline unsigned long my_zero_pfn(unsigned long addr)
 698 {
 699         extern unsigned long zero_pfn;
 700         return zero_pfn;
 701 }
 702 #endif
 703
 704 #ifdef CONFIG_MMU
 705
 706 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 707 static inline int pmd_trans_huge(pmd_t pmd)
 708 {
 709         return 0;
 710 }
 711 #ifndef __HAVE_ARCH_PMD_WRITE
 712 static inline int pmd_write(pmd_t pmd)
 713 {
 714         BUG();
 715         return 0;
 716 }
 717 #endif /* __HAVE_ARCH_PMD_WRITE */
 718 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 719
 720 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
 721         (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
 722          !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
 723 static inline int pud_trans_huge(pud_t pud)
 724 {
 725         return 0;
 726 }
 727 #endif
 728
 729 #ifndef pmd_read_atomic
 730 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 731 {
 732         /*
 733          * Depend on compiler for an atomic pmd read. NOTE: this is
 734          * only going to work, if the pmdval_t isn't larger than
 735          * an unsigned long.
 736          */
 737         return *pmdp;
 738 }
 739 #endif
 740
 741 #ifndef arch_needs_pgtable_deposit
 742 #define arch_needs_pgtable_deposit() (false)
 743 #endif
 744 /*
 745  * This function is meant to be used by sites walking pagetables with
 746  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 747  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 748  * into a null pmd and the transhuge page fault can convert a null pmd
 749  * into an hugepmd or into a regular pmd (if the hugepage allocation
 750  * fails). While holding the mmap_sem in read mode the pmd becomes
 751  * stable and stops changing under us only if it's not null and not a
 752  * transhuge pmd. When those races occurs and this function makes a
 753  * difference vs the standard pmd_none_or_clear_bad, the result is
 754  * undefined so behaving like if the pmd was none is safe (because it
 755  * can return none anyway). The compiler level barrier() is critically
 756  * important to compute the two checks atomically on the same pmdval.
 757  *
 758  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 759  * care of reading the pmd atomically to avoid SMP race conditions
 760  * against pmd_populate() when the mmap_sem is hold for reading by the
 761  * caller (a special atomic read not done by "gcc" as in the generic
 762  * version above, is also needed when THP is disabled because the page
 763  * fault can populate the pmd from under us).
 764  */
 765 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 766 {
 767         pmd_t pmdval = pmd_read_atomic(pmd);
 768         /*
 769          * The barrier will stabilize the pmdval in a register or on
 770          * the stack so that it will stop changing under the code.
 771          *
 772          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 773          * pmd_read_atomic is allowed to return a not atomic pmdval
 774          * (for example pointing to an hugepage that has never been
 775          * mapped in the pmd). The below checks will only care about
 776          * the low part of the pmd with 32bit PAE x86 anyway, with the
 777          * exception of pmd_none(). So the important thing is that if
 778          * the low part of the pmd is found null, the high part will
 779          * be also null or the pmd_none() check below would be
 780          * confused.
 781          */
 782 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 783         barrier();
 784 #endif
 785         if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
 786                 return 1;
 787         if (unlikely(pmd_bad(pmdval))) {
 788                 pmd_clear_bad(pmd);
 789                 return 1;
 790         }
 791         return 0;
 792 }
 793
 794 /*
 795  * This is a noop if Transparent Hugepage Support is not built into
 796  * the kernel. Otherwise it is equivalent to
 797  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 798  * places that already verified the pmd is not none and they want to
 799  * walk ptes while holding the mmap sem in read mode (write mode don't
 800  * need this). If THP is not enabled, the pmd can't go away under the
 801  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 802  * run a pmd_trans_unstable before walking the ptes after
 803  * split_huge_page_pmd returns (because it may have run when the pmd
 804  * become null, but then a page fault can map in a THP and not a
 805  * regular page).
 806  */
 807 static inline int pmd_trans_unstable(pmd_t *pmd)
 808 {
 809 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 810         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 811 #else
 812         return 0;
 813 #endif
 814 }
 815
 816 #ifndef CONFIG_NUMA_BALANCING
 817 /*
 818  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 819  * the only case the kernel cares is for NUMA balancing and is only ever set
 820  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 821  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 822  * is the responsibility of the caller to distinguish between PROT_NONE
 823  * protections and NUMA hinting fault protections.
 824  */
 825 static inline int pte_protnone(pte_t pte)
 826 {
 827         return 0;
 828 }
 829
 830 static inline int pmd_protnone(pmd_t pmd)
 831 {
 832         return 0;
 833 }
 834 #endif /* CONFIG_NUMA_BALANCING */
 835
 836 #endif /* CONFIG_MMU */
 837
 838 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 839 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 840 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 841 int pud_clear_huge(pud_t *pud);
 842 int pmd_clear_huge(pmd_t *pmd);
 843 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 844 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 845 {
 846         return 0;
 847 }
 848 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 849 {
 850         return 0;
 851 }
 852 static inline int pud_clear_huge(pud_t *pud)
 853 {
 854         return 0;
 855 }
 856 static inline int pmd_clear_huge(pmd_t *pmd)
 857 {
 858         return 0;
 859 }
 860 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
 861
 862 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
 863 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 864 /*
 865  * ARCHes with special requirements for evicting THP backing TLB entries can
 866  * implement this. Otherwise also, it can help optimize normal TLB flush in
 867  * THP regime. stock flush_tlb_range() typically has optimization to nuke the
 868  * entire TLB TLB if flush span is greater than a threshold, which will
 869  * likely be true for a single huge page. Thus a single thp flush will
 870  * invalidate the entire TLB which is not desitable.
 871  * e.g. see arch/arc: flush_pmd_tlb_range
 872  */
 873 #define flush_pmd_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
 874 #define flush_pud_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
 875 #else
 876 #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
 877 #define flush_pud_tlb_range(vma, addr, end)     BUILD_BUG()
 878 #endif
 879 #endif
 880
 881 struct file;
 882 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
 883                         unsigned long size, pgprot_t *vma_prot);
 884 #endif /* !__ASSEMBLY__ */
 885
 886 #ifndef io_remap_pfn_range
 887 #define io_remap_pfn_range remap_pfn_range
 888 #endif
 889
 890 #ifndef has_transparent_hugepage
 891 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 892 #define has_transparent_hugepage() 1
 893 #else
 894 #define has_transparent_hugepage() 0
 895 #endif
 896 #endif
 897
 898 #endif /* _ASM_GENERIC_PGTABLE_H */