include/asm-generic/pgtable.h

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