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