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 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
  14         defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
  15 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,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 pte_clear_savedwrite
 237 #define pte_clear_savedwrite pte_wrprotect
 238 #endif
 239
 240 #ifndef pmd_savedwrite
 241 #define pmd_savedwrite pmd_write
 242 #endif
 243
 244 #ifndef pmd_mk_savedwrite
 245 #define pmd_mk_savedwrite pmd_mkwrite
 246 #endif
 247
 248 #ifndef pmd_clear_savedwrite
 249 #define pmd_clear_savedwrite pmd_wrprotect
 250 #endif
 251
 252 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 253 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 254 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 255                                       unsigned long address, pmd_t *pmdp)
 256 {
 257         pmd_t old_pmd = *pmdp;
 258         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 259 }
 260 #else
 261 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 262                                       unsigned long address, pmd_t *pmdp)
 263 {
 264         BUILD_BUG();
 265 }
 266 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 267 #endif
 268 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
 269 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 270 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 271                                       unsigned long address, pud_t *pudp)
 272 {
 273         pud_t old_pud = *pudp;
 274
 275         set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
 276 }
 277 #else
 278 static inline void pudp_set_wrprotect(struct mm_struct *mm,
 279                                       unsigned long address, pud_t *pudp)
 280 {
 281         BUILD_BUG();
 282 }
 283 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
 284 #endif
 285
 286 #ifndef pmdp_collapse_flush
 287 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 288 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 289                                  unsigned long address, pmd_t *pmdp);
 290 #else
 291 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
 292                                         unsigned long address,
 293                                         pmd_t *pmdp)
 294 {
 295         BUILD_BUG();
 296         return *pmdp;
 297 }
 298 #define pmdp_collapse_flush pmdp_collapse_flush
 299 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 300 #endif
 301
 302 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 303 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
 304                                        pgtable_t pgtable);
 305 #endif
 306
 307 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 308 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
 309 #endif
 310
 311 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
 312 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 313                             pmd_t *pmdp);
 314 #endif
 315
 316 #ifndef __HAVE_ARCH_PMDP_HUGE_SPLIT_PREPARE
 317 static inline void pmdp_huge_split_prepare(struct vm_area_struct *vma,
 318                                            unsigned long address, pmd_t *pmdp)
 319 {
 320
 321 }
 322 #endif
 323
 324 #ifndef __HAVE_ARCH_PTE_SAME
 325 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 326 {
 327         return pte_val(pte_a) == pte_val(pte_b);
 328 }
 329 #endif
 330
 331 #ifndef __HAVE_ARCH_PTE_UNUSED
 332 /*
 333  * Some architectures provide facilities to virtualization guests
 334  * so that they can flag allocated pages as unused. This allows the
 335  * host to transparently reclaim unused pages. This function returns
 336  * whether the pte's page is unused.
 337  */
 338 static inline int pte_unused(pte_t pte)
 339 {
 340         return 0;
 341 }
 342 #endif
 343
 344 #ifndef __HAVE_ARCH_PMD_SAME
 345 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 346 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 347 {
 348         return pmd_val(pmd_a) == pmd_val(pmd_b);
 349 }
 350
 351 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 352 {
 353         return pud_val(pud_a) == pud_val(pud_b);
 354 }
 355 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 356 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 357 {
 358         BUILD_BUG();
 359         return 0;
 360 }
 361
 362 static inline int pud_same(pud_t pud_a, pud_t pud_b)
 363 {
 364         BUILD_BUG();
 365         return 0;
 366 }
 367 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 368 #endif
 369
 370 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 371 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 372 #endif
 373
 374 #ifndef __HAVE_ARCH_MOVE_PTE
 375 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 376 #endif
 377
 378 #ifndef pte_accessible
 379 # define pte_accessible(mm, pte)        ((void)(pte), 1)
 380 #endif
 381
 382 #ifndef flush_tlb_fix_spurious_fault
 383 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 384 #endif
 385
 386 #ifndef pgprot_noncached
 387 #define pgprot_noncached(prot)  (prot)
 388 #endif
 389
 390 #ifndef pgprot_writecombine
 391 #define pgprot_writecombine pgprot_noncached
 392 #endif
 393
 394 #ifndef pgprot_writethrough
 395 #define pgprot_writethrough pgprot_noncached
 396 #endif
 397
 398 #ifndef pgprot_device
 399 #define pgprot_device pgprot_noncached
 400 #endif
 401
 402 #ifndef pgprot_modify
 403 #define pgprot_modify pgprot_modify
 404 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 405 {
 406         if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
 407                 newprot = pgprot_noncached(newprot);
 408         if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
 409                 newprot = pgprot_writecombine(newprot);
 410         if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
 411                 newprot = pgprot_device(newprot);
 412         return newprot;
 413 }
 414 #endif
 415
 416 /*
 417  * When walking page tables, get the address of the next boundary,
 418  * or the end address of the range if that comes earlier.  Although no
 419  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 420  */
 421
 422 #define pgd_addr_end(addr, end)                                         \
 423 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 424         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 425 })
 426
 427 #ifndef p4d_addr_end
 428 #define p4d_addr_end(addr, end)                                         \
 429 ({      unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;      \
 430         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 431 })
 432 #endif
 433
 434 #ifndef pud_addr_end
 435 #define pud_addr_end(addr, end)                                         \
 436 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 437         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 438 })
 439 #endif
 440
 441 #ifndef pmd_addr_end
 442 #define pmd_addr_end(addr, end)                                         \
 443 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 444         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 445 })
 446 #endif
 447
 448 /*
 449  * When walking page tables, we usually want to skip any p?d_none entries;
 450  * and any p?d_bad entries - reporting the error before resetting to none.
 451  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 452  */
 453 void pgd_clear_bad(pgd_t *);
 454 void p4d_clear_bad(p4d_t *);
 455 void pud_clear_bad(pud_t *);
 456 void pmd_clear_bad(pmd_t *);
 457
 458 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 459 {
 460         if (pgd_none(*pgd))
 461                 return 1;
 462         if (unlikely(pgd_bad(*pgd))) {
 463                 pgd_clear_bad(pgd);
 464                 return 1;
 465         }
 466         return 0;
 467 }
 468
 469 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
 470 {
 471         if (p4d_none(*p4d))
 472                 return 1;
 473         if (unlikely(p4d_bad(*p4d))) {
 474                 p4d_clear_bad(p4d);
 475                 return 1;
 476         }
 477         return 0;
 478 }
 479
 480 static inline int pud_none_or_clear_bad(pud_t *pud)
 481 {
 482         if (pud_none(*pud))
 483                 return 1;
 484         if (unlikely(pud_bad(*pud))) {
 485                 pud_clear_bad(pud);
 486                 return 1;
 487         }
 488         return 0;
 489 }
 490
 491 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 492 {
 493         if (pmd_none(*pmd))
 494                 return 1;
 495         if (unlikely(pmd_bad(*pmd))) {
 496                 pmd_clear_bad(pmd);
 497                 return 1;
 498         }
 499         return 0;
 500 }
 501
 502 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 503                                              unsigned long addr,
 504                                              pte_t *ptep)
 505 {
 506         /*
 507          * Get the current pte state, but zero it out to make it
 508          * non-present, preventing the hardware from asynchronously
 509          * updating it.
 510          */
 511         return ptep_get_and_clear(mm, addr, ptep);
 512 }
 513
 514 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 515                                              unsigned long addr,
 516                                              pte_t *ptep, pte_t pte)
 517 {
 518         /*
 519          * The pte is non-present, so there's no hardware state to
 520          * preserve.
 521          */
 522         set_pte_at(mm, addr, ptep, pte);
 523 }
 524
 525 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 526 /*
 527  * Start a pte protection read-modify-write transaction, which
 528  * protects against asynchronous hardware modifications to the pte.
 529  * The intention is not to prevent the hardware from making pte
 530  * updates, but to prevent any updates it may make from being lost.
 531  *
 532  * This does not protect against other software modifications of the
 533  * pte; the appropriate pte lock must be held over the transation.
 534  *
 535  * Note that this interface is intended to be batchable, meaning that
 536  * ptep_modify_prot_commit may not actually update the pte, but merely
 537  * queue the update to be done at some later time.  The update must be
 538  * actually committed before the pte lock is released, however.
 539  */
 540 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 541                                            unsigned long addr,
 542                                            pte_t *ptep)
 543 {
 544         return __ptep_modify_prot_start(mm, addr, ptep);
 545 }
 546
 547 /*
 548  * Commit an update to a pte, leaving any hardware-controlled bits in
 549  * the PTE unmodified.
 550  */
 551 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 552                                            unsigned long addr,
 553                                            pte_t *ptep, pte_t pte)
 554 {
 555         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 556 }
 557 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 558 #endif /* CONFIG_MMU */
 559
 560 /*
 561  * A facility to provide lazy MMU batching.  This allows PTE updates and
 562  * page invalidations to be delayed until a call to leave lazy MMU mode
 563  * is issued.  Some architectures may benefit from doing this, and it is
 564  * beneficial for both shadow and direct mode hypervisors, which may batch
 565  * the PTE updates which happen during this window.  Note that using this
 566  * interface requires that read hazards be removed from the code.  A read
 567  * hazard could result in the direct mode hypervisor case, since the actual
 568  * write to the page tables may not yet have taken place, so reads though
 569  * a raw PTE pointer after it has been modified are not guaranteed to be
 570  * up to date.  This mode can only be entered and left under the protection of
 571  * the page table locks for all page tables which may be modified.  In the UP
 572  * case, this is required so that preemption is disabled, and in the SMP case,
 573  * it must synchronize the delayed page table writes properly on other CPUs.
 574  */
 575 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 576 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 577 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 578 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 579 #endif
 580
 581 /*
 582  * A facility to provide batching of the reload of page tables and
 583  * other process state with the actual context switch code for
 584  * paravirtualized guests.  By convention, only one of the batched
 585  * update (lazy) modes (CPU, MMU) should be active at any given time,
 586  * entry should never be nested, and entry and exits should always be
 587  * paired.  This is for sanity of maintaining and reasoning about the
 588  * kernel code.  In this case, the exit (end of the context switch) is
 589  * in architecture-specific code, and so doesn't need a generic
 590  * definition.
 591  */
 592 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 593 #define arch_start_context_switch(prev) do {} while (0)
 594 #endif
 595
 596 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
 597 static inline int pte_soft_dirty(pte_t pte)
 598 {
 599         return 0;
 600 }
 601
 602 static inline int pmd_soft_dirty(pmd_t pmd)
 603 {
 604         return 0;
 605 }
 606
 607 static inline pte_t pte_mksoft_dirty(pte_t pte)
 608 {
 609         return pte;
 610 }
 611
 612 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
 613 {
 614         return pmd;
 615 }
 616
 617 static inline pte_t pte_clear_soft_dirty(pte_t pte)
 618 {
 619         return pte;
 620 }
 621
 622 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
 623 {
 624         return pmd;
 625 }
 626
 627 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
 628 {
 629         return pte;
 630 }
 631
 632 static inline int pte_swp_soft_dirty(pte_t pte)
 633 {
 634         return 0;
 635 }
 636
 637 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
 638 {
 639         return pte;
 640 }
 641 #endif
 642
 643 #ifndef __HAVE_PFNMAP_TRACKING
 644 /*
 645  * Interfaces that can be used by architecture code to keep track of
 646  * memory type of pfn mappings specified by the remap_pfn_range,
 647  * vm_insert_pfn.
 648  */
 649
 650 /*
 651  * track_pfn_remap is called when a _new_ pfn mapping is being established
 652  * by remap_pfn_range() for physical range indicated by pfn and size.
 653  */
 654 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 655                                   unsigned long pfn, unsigned long addr,
 656                                   unsigned long size)
 657 {
 658         return 0;
 659 }
 660
 661 /*
 662  * track_pfn_insert is called when a _new_ single pfn is established
 663  * by vm_insert_pfn().
 664  */
 665 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 666                                     pfn_t pfn)
 667 {
 668 }
 669
 670 /*
 671  * track_pfn_copy is called when vma that is covering the pfnmap gets
 672  * copied through copy_page_range().
 673  */
 674 static inline int track_pfn_copy(struct vm_area_struct *vma)
 675 {
 676         return 0;
 677 }
 678
 679 /*
 680  * untrack_pfn is called while unmapping a pfnmap for a region.
 681  * untrack can be called for a specific region indicated by pfn and size or
 682  * can be for the entire vma (in which case pfn, size are zero).
 683  */
 684 static inline void untrack_pfn(struct vm_area_struct *vma,
 685                                unsigned long pfn, unsigned long size)
 686 {
 687 }
 688
 689 /*
 690  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
 691  */
 692 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
 693 {
 694 }
 695 #else
 696 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 697                            unsigned long pfn, unsigned long addr,
 698                            unsigned long size);
 699 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 700                              pfn_t pfn);
 701 extern int track_pfn_copy(struct vm_area_struct *vma);
 702 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 703                         unsigned long size);
 704 extern void untrack_pfn_moved(struct vm_area_struct *vma);
 705 #endif
 706
 707 #ifdef __HAVE_COLOR_ZERO_PAGE
 708 static inline int is_zero_pfn(unsigned long pfn)
 709 {
 710         extern unsigned long zero_pfn;
 711         unsigned long offset_from_zero_pfn = pfn - zero_pfn;
 712         return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
 713 }
 714
 715 #define my_zero_pfn(addr)       page_to_pfn(ZERO_PAGE(addr))
 716
 717 #else
 718 static inline int is_zero_pfn(unsigned long pfn)
 719 {
 720         extern unsigned long zero_pfn;
 721         return pfn == zero_pfn;
 722 }
 723
 724 static inline unsigned long my_zero_pfn(unsigned long addr)
 725 {
 726         extern unsigned long zero_pfn;
 727         return zero_pfn;
 728 }
 729 #endif
 730
 731 #ifdef CONFIG_MMU
 732
 733 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 734 static inline int pmd_trans_huge(pmd_t pmd)
 735 {
 736         return 0;
 737 }
 738 #ifndef __HAVE_ARCH_PMD_WRITE
 739 static inline int pmd_write(pmd_t pmd)
 740 {
 741         BUG();
 742         return 0;
 743 }
 744 #endif /* __HAVE_ARCH_PMD_WRITE */
 745 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 746
 747 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
 748         (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
 749          !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
 750 static inline int pud_trans_huge(pud_t pud)
 751 {
 752         return 0;
 753 }
 754 #endif
 755
 756 #ifndef pmd_read_atomic
 757 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 758 {
 759         /*
 760          * Depend on compiler for an atomic pmd read. NOTE: this is
 761          * only going to work, if the pmdval_t isn't larger than
 762          * an unsigned long.
 763          */
 764         return *pmdp;
 765 }
 766 #endif
 767
 768 #ifndef arch_needs_pgtable_deposit
 769 #define arch_needs_pgtable_deposit() (false)
 770 #endif
 771 /*
 772  * This function is meant to be used by sites walking pagetables with
 773  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 774  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 775  * into a null pmd and the transhuge page fault can convert a null pmd
 776  * into an hugepmd or into a regular pmd (if the hugepage allocation
 777  * fails). While holding the mmap_sem in read mode the pmd becomes
 778  * stable and stops changing under us only if it's not null and not a
 779  * transhuge pmd. When those races occurs and this function makes a
 780  * difference vs the standard pmd_none_or_clear_bad, the result is
 781  * undefined so behaving like if the pmd was none is safe (because it
 782  * can return none anyway). The compiler level barrier() is critically
 783  * important to compute the two checks atomically on the same pmdval.
 784  *
 785  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 786  * care of reading the pmd atomically to avoid SMP race conditions
 787  * against pmd_populate() when the mmap_sem is hold for reading by the
 788  * caller (a special atomic read not done by "gcc" as in the generic
 789  * version above, is also needed when THP is disabled because the page
 790  * fault can populate the pmd from under us).
 791  */
 792 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 793 {
 794         pmd_t pmdval = pmd_read_atomic(pmd);
 795         /*
 796          * The barrier will stabilize the pmdval in a register or on
 797          * the stack so that it will stop changing under the code.
 798          *
 799          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 800          * pmd_read_atomic is allowed to return a not atomic pmdval
 801          * (for example pointing to an hugepage that has never been
 802          * mapped in the pmd). The below checks will only care about
 803          * the low part of the pmd with 32bit PAE x86 anyway, with the
 804          * exception of pmd_none(). So the important thing is that if
 805          * the low part of the pmd is found null, the high part will
 806          * be also null or the pmd_none() check below would be
 807          * confused.
 808          */
 809 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 810         barrier();
 811 #endif
 812         if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
 813                 return 1;
 814         if (unlikely(pmd_bad(pmdval))) {
 815                 pmd_clear_bad(pmd);
 816                 return 1;
 817         }
 818         return 0;
 819 }
 820
 821 /*
 822  * This is a noop if Transparent Hugepage Support is not built into
 823  * the kernel. Otherwise it is equivalent to
 824  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 825  * places that already verified the pmd is not none and they want to
 826  * walk ptes while holding the mmap sem in read mode (write mode don't
 827  * need this). If THP is not enabled, the pmd can't go away under the
 828  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 829  * run a pmd_trans_unstable before walking the ptes after
 830  * split_huge_page_pmd returns (because it may have run when the pmd
 831  * become null, but then a page fault can map in a THP and not a
 832  * regular page).
 833  */
 834 static inline int pmd_trans_unstable(pmd_t *pmd)
 835 {
 836 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 837         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 838 #else
 839         return 0;
 840 #endif
 841 }
 842
 843 #ifndef CONFIG_NUMA_BALANCING
 844 /*
 845  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
 846  * the only case the kernel cares is for NUMA balancing and is only ever set
 847  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
 848  * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
 849  * is the responsibility of the caller to distinguish between PROT_NONE
 850  * protections and NUMA hinting fault protections.
 851  */
 852 static inline int pte_protnone(pte_t pte)
 853 {
 854         return 0;
 855 }
 856
 857 static inline int pmd_protnone(pmd_t pmd)
 858 {
 859         return 0;
 860 }
 861 #endif /* CONFIG_NUMA_BALANCING */
 862
 863 #endif /* CONFIG_MMU */
 864
 865 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
 866
 867 #ifndef __PAGETABLE_P4D_FOLDED
 868 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
 869 int p4d_clear_huge(p4d_t *p4d);
 870 #else
 871 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 872 {
 873         return 0;
 874 }
 875 static inline int p4d_clear_huge(p4d_t *p4d)
 876 {
 877         return 0;
 878 }
 879 #endif /* !__PAGETABLE_P4D_FOLDED */
 880
 881 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
 882 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
 883 int pud_clear_huge(pud_t *pud);
 884 int pmd_clear_huge(pmd_t *pmd);
 885 #else   /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
 886 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
 887 {
 888         return 0;
 889 }
 890 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
 891 {
 892         return 0;
 893 }
 894 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
 895 {
 896         return 0;
 897 }
 898 static inline int p4d_clear_huge(p4d_t *p4d)
 899 {
 900         return 0;
 901 }
 902 static inline int pud_clear_huge(pud_t *pud)
 903 {
 904         return 0;
 905 }
 906 static inline int pmd_clear_huge(pmd_t *pmd)
 907 {
 908         return 0;
 909 }
 910 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
 911
 912 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
 913 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 914 /*
 915  * ARCHes with special requirements for evicting THP backing TLB entries can
 916  * implement this. Otherwise also, it can help optimize normal TLB flush in
 917  * THP regime. stock flush_tlb_range() typically has optimization to nuke the
 918  * entire TLB TLB if flush span is greater than a threshold, which will
 919  * likely be true for a single huge page. Thus a single thp flush will
 920  * invalidate the entire TLB which is not desitable.
 921  * e.g. see arch/arc: flush_pmd_tlb_range
 922  */
 923 #define flush_pmd_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
 924 #define flush_pud_tlb_range(vma, addr, end)     flush_tlb_range(vma, addr, end)
 925 #else
 926 #define flush_pmd_tlb_range(vma, addr, end)     BUILD_BUG()
 927 #define flush_pud_tlb_range(vma, addr, end)     BUILD_BUG()
 928 #endif
 929 #endif
 930
 931 struct file;
 932 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
 933                         unsigned long size, pgprot_t *vma_prot);
 934 #endif /* !__ASSEMBLY__ */
 935
 936 #ifndef io_remap_pfn_range
 937 #define io_remap_pfn_range remap_pfn_range
 938 #endif
 939
 940 #ifndef has_transparent_hugepage
 941 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 942 #define has_transparent_hugepage() 1
 943 #else
 944 #define has_transparent_hugepage() 0
 945 #endif
 946 #endif
 947
 948 #endif /* _ASM_GENERIC_PGTABLE_H */