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
  10 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  11 extern int ptep_set_access_flags(struct vm_area_struct *vma,
  12                                  unsigned long address, pte_t *ptep,
  13                                  pte_t entry, int dirty);
  14 #endif
  15
  16 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  17 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
  18                                  unsigned long address, pmd_t *pmdp,
  19                                  pmd_t entry, int dirty);
  20 #endif
  21
  22 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  23 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  24                                             unsigned long address,
  25                                             pte_t *ptep)
  26 {
  27         pte_t pte = *ptep;
  28         int r = 1;
  29         if (!pte_young(pte))
  30                 r = 0;
  31         else
  32                 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  33         return r;
  34 }
  35 #endif
  36
  37 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  38 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  39 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  40                                             unsigned long address,
  41                                             pmd_t *pmdp)
  42 {
  43         pmd_t pmd = *pmdp;
  44         int r = 1;
  45         if (!pmd_young(pmd))
  46                 r = 0;
  47         else
  48                 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  49         return r;
  50 }
  51 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  52 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  53                                             unsigned long address,
  54                                             pmd_t *pmdp)
  55 {
  56         BUG();
  57         return 0;
  58 }
  59 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  60 #endif
  61
  62 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  63 int ptep_clear_flush_young(struct vm_area_struct *vma,
  64                            unsigned long address, pte_t *ptep);
  65 #endif
  66
  67 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  68 int pmdp_clear_flush_young(struct vm_area_struct *vma,
  69                            unsigned long address, pmd_t *pmdp);
  70 #endif
  71
  72 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  73 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  74                                        unsigned long address,
  75                                        pte_t *ptep)
  76 {
  77         pte_t pte = *ptep;
  78         pte_clear(mm, address, ptep);
  79         return pte;
  80 }
  81 #endif
  82
  83 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
  84 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  85 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
  86                                        unsigned long address,
  87                                        pmd_t *pmdp)
  88 {
  89         pmd_t pmd = *pmdp;
  90         pmd_clear(mm, address, pmdp);
  91         return pmd;
  92 }
  93 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  94 #endif
  95
  96 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
  97 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
  98                                             unsigned long address, pte_t *ptep,
  99                                             int full)
 100 {
 101         pte_t pte;
 102         pte = ptep_get_and_clear(mm, address, ptep);
 103         return pte;
 104 }
 105 #endif
 106
 107 /*
 108  * Some architectures may be able to avoid expensive synchronization
 109  * primitives when modifications are made to PTE's which are already
 110  * not present, or in the process of an address space destruction.
 111  */
 112 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
 113 static inline void pte_clear_not_present_full(struct mm_struct *mm,
 114                                               unsigned long address,
 115                                               pte_t *ptep,
 116                                               int full)
 117 {
 118         pte_clear(mm, address, ptep);
 119 }
 120 #endif
 121
 122 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
 123 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
 124                               unsigned long address,
 125                               pte_t *ptep);
 126 #endif
 127
 128 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
 129 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
 130                               unsigned long address,
 131                               pmd_t *pmdp);
 132 #endif
 133
 134 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
 135 struct mm_struct;
 136 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
 137 {
 138         pte_t old_pte = *ptep;
 139         set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
 140 }
 141 #endif
 142
 143 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
 144 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 145 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 146                                       unsigned long address, pmd_t *pmdp)
 147 {
 148         pmd_t old_pmd = *pmdp;
 149         set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
 150 }
 151 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 152 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
 153                                       unsigned long address, pmd_t *pmdp)
 154 {
 155         BUG();
 156 }
 157 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 158 #endif
 159
 160 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
 161 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
 162                                  unsigned long address, pmd_t *pmdp);
 163 #endif
 164
 165 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
 166 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
 167 #endif
 168
 169 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
 170 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
 171 #endif
 172
 173 #ifndef __HAVE_ARCH_PTE_SAME
 174 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 175 {
 176         return pte_val(pte_a) == pte_val(pte_b);
 177 }
 178 #endif
 179
 180 #ifndef __HAVE_ARCH_PMD_SAME
 181 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 182 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 183 {
 184         return pmd_val(pmd_a) == pmd_val(pmd_b);
 185 }
 186 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 187 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 188 {
 189         BUG();
 190         return 0;
 191 }
 192 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 193 #endif
 194
 195 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 196 #define page_test_and_clear_dirty(pfn, mapped)  (0)
 197 #endif
 198
 199 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 200 #define pte_maybe_dirty(pte)            pte_dirty(pte)
 201 #else
 202 #define pte_maybe_dirty(pte)            (1)
 203 #endif
 204
 205 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
 206 #define page_test_and_clear_young(pfn) (0)
 207 #endif
 208
 209 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 210 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 211 #endif
 212
 213 #ifndef __HAVE_ARCH_MOVE_PTE
 214 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 215 #endif
 216
 217 #ifndef flush_tlb_fix_spurious_fault
 218 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 219 #endif
 220
 221 #ifndef pgprot_noncached
 222 #define pgprot_noncached(prot)  (prot)
 223 #endif
 224
 225 #ifndef pgprot_writecombine
 226 #define pgprot_writecombine pgprot_noncached
 227 #endif
 228
 229 /*
 230  * When walking page tables, get the address of the next boundary,
 231  * or the end address of the range if that comes earlier.  Although no
 232  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 233  */
 234
 235 #define pgd_addr_end(addr, end)                                         \
 236 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 237         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 238 })
 239
 240 #ifndef pud_addr_end
 241 #define pud_addr_end(addr, end)                                         \
 242 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 243         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 244 })
 245 #endif
 246
 247 #ifndef pmd_addr_end
 248 #define pmd_addr_end(addr, end)                                         \
 249 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 250         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 251 })
 252 #endif
 253
 254 /*
 255  * When walking page tables, we usually want to skip any p?d_none entries;
 256  * and any p?d_bad entries - reporting the error before resetting to none.
 257  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 258  */
 259 void pgd_clear_bad(pgd_t *);
 260 void pud_clear_bad(pud_t *);
 261 void pmd_clear_bad(pmd_t *);
 262
 263 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 264 {
 265         if (pgd_none(*pgd))
 266                 return 1;
 267         if (unlikely(pgd_bad(*pgd))) {
 268                 pgd_clear_bad(pgd);
 269                 return 1;
 270         }
 271         return 0;
 272 }
 273
 274 static inline int pud_none_or_clear_bad(pud_t *pud)
 275 {
 276         if (pud_none(*pud))
 277                 return 1;
 278         if (unlikely(pud_bad(*pud))) {
 279                 pud_clear_bad(pud);
 280                 return 1;
 281         }
 282         return 0;
 283 }
 284
 285 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 286 {
 287         if (pmd_none(*pmd))
 288                 return 1;
 289         if (unlikely(pmd_bad(*pmd))) {
 290                 pmd_clear_bad(pmd);
 291                 return 1;
 292         }
 293         return 0;
 294 }
 295
 296 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 297                                              unsigned long addr,
 298                                              pte_t *ptep)
 299 {
 300         /*
 301          * Get the current pte state, but zero it out to make it
 302          * non-present, preventing the hardware from asynchronously
 303          * updating it.
 304          */
 305         return ptep_get_and_clear(mm, addr, ptep);
 306 }
 307
 308 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 309                                              unsigned long addr,
 310                                              pte_t *ptep, pte_t pte)
 311 {
 312         /*
 313          * The pte is non-present, so there's no hardware state to
 314          * preserve.
 315          */
 316         set_pte_at(mm, addr, ptep, pte);
 317 }
 318
 319 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 320 /*
 321  * Start a pte protection read-modify-write transaction, which
 322  * protects against asynchronous hardware modifications to the pte.
 323  * The intention is not to prevent the hardware from making pte
 324  * updates, but to prevent any updates it may make from being lost.
 325  *
 326  * This does not protect against other software modifications of the
 327  * pte; the appropriate pte lock must be held over the transation.
 328  *
 329  * Note that this interface is intended to be batchable, meaning that
 330  * ptep_modify_prot_commit may not actually update the pte, but merely
 331  * queue the update to be done at some later time.  The update must be
 332  * actually committed before the pte lock is released, however.
 333  */
 334 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 335                                            unsigned long addr,
 336                                            pte_t *ptep)
 337 {
 338         return __ptep_modify_prot_start(mm, addr, ptep);
 339 }
 340
 341 /*
 342  * Commit an update to a pte, leaving any hardware-controlled bits in
 343  * the PTE unmodified.
 344  */
 345 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 346                                            unsigned long addr,
 347                                            pte_t *ptep, pte_t pte)
 348 {
 349         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 350 }
 351 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 352 #endif /* CONFIG_MMU */
 353
 354 /*
 355  * A facility to provide lazy MMU batching.  This allows PTE updates and
 356  * page invalidations to be delayed until a call to leave lazy MMU mode
 357  * is issued.  Some architectures may benefit from doing this, and it is
 358  * beneficial for both shadow and direct mode hypervisors, which may batch
 359  * the PTE updates which happen during this window.  Note that using this
 360  * interface requires that read hazards be removed from the code.  A read
 361  * hazard could result in the direct mode hypervisor case, since the actual
 362  * write to the page tables may not yet have taken place, so reads though
 363  * a raw PTE pointer after it has been modified are not guaranteed to be
 364  * up to date.  This mode can only be entered and left under the protection of
 365  * the page table locks for all page tables which may be modified.  In the UP
 366  * case, this is required so that preemption is disabled, and in the SMP case,
 367  * it must synchronize the delayed page table writes properly on other CPUs.
 368  */
 369 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 370 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 371 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 372 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 373 #endif
 374
 375 /*
 376  * A facility to provide batching of the reload of page tables and
 377  * other process state with the actual context switch code for
 378  * paravirtualized guests.  By convention, only one of the batched
 379  * update (lazy) modes (CPU, MMU) should be active at any given time,
 380  * entry should never be nested, and entry and exits should always be
 381  * paired.  This is for sanity of maintaining and reasoning about the
 382  * kernel code.  In this case, the exit (end of the context switch) is
 383  * in architecture-specific code, and so doesn't need a generic
 384  * definition.
 385  */
 386 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 387 #define arch_start_context_switch(prev) do {} while (0)
 388 #endif
 389
 390 #ifndef __HAVE_PFNMAP_TRACKING
 391 /*
 392  * Interfaces that can be used by architecture code to keep track of
 393  * memory type of pfn mappings specified by the remap_pfn_range,
 394  * vm_insert_pfn.
 395  */
 396
 397 /*
 398  * track_pfn_remap is called when a _new_ pfn mapping is being established
 399  * by remap_pfn_range() for physical range indicated by pfn and size.
 400  */
 401 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 402                                   unsigned long pfn, unsigned long addr,
 403                                   unsigned long size)
 404 {
 405         return 0;
 406 }
 407
 408 /*
 409  * track_pfn_insert is called when a _new_ single pfn is established
 410  * by vm_insert_pfn().
 411  */
 412 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 413                                    unsigned long pfn)
 414 {
 415         return 0;
 416 }
 417
 418 /*
 419  * track_pfn_copy is called when vma that is covering the pfnmap gets
 420  * copied through copy_page_range().
 421  */
 422 static inline int track_pfn_copy(struct vm_area_struct *vma)
 423 {
 424         return 0;
 425 }
 426
 427 /*
 428  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 429  * untrack can be called for a specific region indicated by pfn and size or
 430  * can be for the entire vma (in which case pfn, size are zero).
 431  */
 432 static inline void untrack_pfn(struct vm_area_struct *vma,
 433                                unsigned long pfn, unsigned long size)
 434 {
 435 }
 436 #else
 437 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 438                            unsigned long pfn, unsigned long addr,
 439                            unsigned long size);
 440 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
 441                             unsigned long pfn);
 442 extern int track_pfn_copy(struct vm_area_struct *vma);
 443 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 444                         unsigned long size);
 445 #endif
 446
 447 #ifdef CONFIG_MMU
 448
 449 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 450 static inline int pmd_trans_huge(pmd_t pmd)
 451 {
 452         return 0;
 453 }
 454 static inline int pmd_trans_splitting(pmd_t pmd)
 455 {
 456         return 0;
 457 }
 458 #ifndef __HAVE_ARCH_PMD_WRITE
 459 static inline int pmd_write(pmd_t pmd)
 460 {
 461         BUG();
 462         return 0;
 463 }
 464 #endif /* __HAVE_ARCH_PMD_WRITE */
 465 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 466
 467 #ifndef pmd_read_atomic
 468 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 469 {
 470         /*
 471          * Depend on compiler for an atomic pmd read. NOTE: this is
 472          * only going to work, if the pmdval_t isn't larger than
 473          * an unsigned long.
 474          */
 475         return *pmdp;
 476 }
 477 #endif
 478
 479 /*
 480  * This function is meant to be used by sites walking pagetables with
 481  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 482  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 483  * into a null pmd and the transhuge page fault can convert a null pmd
 484  * into an hugepmd or into a regular pmd (if the hugepage allocation
 485  * fails). While holding the mmap_sem in read mode the pmd becomes
 486  * stable and stops changing under us only if it's not null and not a
 487  * transhuge pmd. When those races occurs and this function makes a
 488  * difference vs the standard pmd_none_or_clear_bad, the result is
 489  * undefined so behaving like if the pmd was none is safe (because it
 490  * can return none anyway). The compiler level barrier() is critically
 491  * important to compute the two checks atomically on the same pmdval.
 492  *
 493  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 494  * care of reading the pmd atomically to avoid SMP race conditions
 495  * against pmd_populate() when the mmap_sem is hold for reading by the
 496  * caller (a special atomic read not done by "gcc" as in the generic
 497  * version above, is also needed when THP is disabled because the page
 498  * fault can populate the pmd from under us).
 499  */
 500 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 501 {
 502         pmd_t pmdval = pmd_read_atomic(pmd);
 503         /*
 504          * The barrier will stabilize the pmdval in a register or on
 505          * the stack so that it will stop changing under the code.
 506          *
 507          * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
 508          * pmd_read_atomic is allowed to return a not atomic pmdval
 509          * (for example pointing to an hugepage that has never been
 510          * mapped in the pmd). The below checks will only care about
 511          * the low part of the pmd with 32bit PAE x86 anyway, with the
 512          * exception of pmd_none(). So the important thing is that if
 513          * the low part of the pmd is found null, the high part will
 514          * be also null or the pmd_none() check below would be
 515          * confused.
 516          */
 517 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 518         barrier();
 519 #endif
 520         if (pmd_none(pmdval))
 521                 return 1;
 522         if (unlikely(pmd_bad(pmdval))) {
 523                 if (!pmd_trans_huge(pmdval))
 524                         pmd_clear_bad(pmd);
 525                 return 1;
 526         }
 527         return 0;
 528 }
 529
 530 /*
 531  * This is a noop if Transparent Hugepage Support is not built into
 532  * the kernel. Otherwise it is equivalent to
 533  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 534  * places that already verified the pmd is not none and they want to
 535  * walk ptes while holding the mmap sem in read mode (write mode don't
 536  * need this). If THP is not enabled, the pmd can't go away under the
 537  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 538  * run a pmd_trans_unstable before walking the ptes after
 539  * split_huge_page_pmd returns (because it may have run when the pmd
 540  * become null, but then a page fault can map in a THP and not a
 541  * regular page).
 542  */
 543 static inline int pmd_trans_unstable(pmd_t *pmd)
 544 {
 545 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 546         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 547 #else
 548         return 0;
 549 #endif
 550 }
 551
 552 #endif /* CONFIG_MMU */
 553
 554 #endif /* !__ASSEMBLY__ */
 555
 556 #endif /* _ASM_GENERIC_PGTABLE_H */