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_PTE_SAME
 166 static inline int pte_same(pte_t pte_a, pte_t pte_b)
 167 {
 168         return pte_val(pte_a) == pte_val(pte_b);
 169 }
 170 #endif
 171
 172 #ifndef __HAVE_ARCH_PMD_SAME
 173 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 174 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 175 {
 176         return pmd_val(pmd_a) == pmd_val(pmd_b);
 177 }
 178 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 179 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 180 {
 181         BUG();
 182         return 0;
 183 }
 184 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 185 #endif
 186
 187 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 188 #define page_test_and_clear_dirty(pfn, mapped)  (0)
 189 #endif
 190
 191 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
 192 #define pte_maybe_dirty(pte)            pte_dirty(pte)
 193 #else
 194 #define pte_maybe_dirty(pte)            (1)
 195 #endif
 196
 197 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
 198 #define page_test_and_clear_young(pfn) (0)
 199 #endif
 200
 201 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 202 #define pgd_offset_gate(mm, addr)       pgd_offset(mm, addr)
 203 #endif
 204
 205 #ifndef __HAVE_ARCH_MOVE_PTE
 206 #define move_pte(pte, prot, old_addr, new_addr) (pte)
 207 #endif
 208
 209 #ifndef flush_tlb_fix_spurious_fault
 210 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
 211 #endif
 212
 213 #ifndef pgprot_noncached
 214 #define pgprot_noncached(prot)  (prot)
 215 #endif
 216
 217 #ifndef pgprot_writecombine
 218 #define pgprot_writecombine pgprot_noncached
 219 #endif
 220
 221 /*
 222  * When walking page tables, get the address of the next boundary,
 223  * or the end address of the range if that comes earlier.  Although no
 224  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
 225  */
 226
 227 #define pgd_addr_end(addr, end)                                         \
 228 ({      unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;  \
 229         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 230 })
 231
 232 #ifndef pud_addr_end
 233 #define pud_addr_end(addr, end)                                         \
 234 ({      unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;      \
 235         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 236 })
 237 #endif
 238
 239 #ifndef pmd_addr_end
 240 #define pmd_addr_end(addr, end)                                         \
 241 ({      unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;      \
 242         (__boundary - 1 < (end) - 1)? __boundary: (end);                \
 243 })
 244 #endif
 245
 246 /*
 247  * When walking page tables, we usually want to skip any p?d_none entries;
 248  * and any p?d_bad entries - reporting the error before resetting to none.
 249  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
 250  */
 251 void pgd_clear_bad(pgd_t *);
 252 void pud_clear_bad(pud_t *);
 253 void pmd_clear_bad(pmd_t *);
 254
 255 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
 256 {
 257         if (pgd_none(*pgd))
 258                 return 1;
 259         if (unlikely(pgd_bad(*pgd))) {
 260                 pgd_clear_bad(pgd);
 261                 return 1;
 262         }
 263         return 0;
 264 }
 265
 266 static inline int pud_none_or_clear_bad(pud_t *pud)
 267 {
 268         if (pud_none(*pud))
 269                 return 1;
 270         if (unlikely(pud_bad(*pud))) {
 271                 pud_clear_bad(pud);
 272                 return 1;
 273         }
 274         return 0;
 275 }
 276
 277 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
 278 {
 279         if (pmd_none(*pmd))
 280                 return 1;
 281         if (unlikely(pmd_bad(*pmd))) {
 282                 pmd_clear_bad(pmd);
 283                 return 1;
 284         }
 285         return 0;
 286 }
 287
 288 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
 289                                              unsigned long addr,
 290                                              pte_t *ptep)
 291 {
 292         /*
 293          * Get the current pte state, but zero it out to make it
 294          * non-present, preventing the hardware from asynchronously
 295          * updating it.
 296          */
 297         return ptep_get_and_clear(mm, addr, ptep);
 298 }
 299
 300 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
 301                                              unsigned long addr,
 302                                              pte_t *ptep, pte_t pte)
 303 {
 304         /*
 305          * The pte is non-present, so there's no hardware state to
 306          * preserve.
 307          */
 308         set_pte_at(mm, addr, ptep, pte);
 309 }
 310
 311 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 312 /*
 313  * Start a pte protection read-modify-write transaction, which
 314  * protects against asynchronous hardware modifications to the pte.
 315  * The intention is not to prevent the hardware from making pte
 316  * updates, but to prevent any updates it may make from being lost.
 317  *
 318  * This does not protect against other software modifications of the
 319  * pte; the appropriate pte lock must be held over the transation.
 320  *
 321  * Note that this interface is intended to be batchable, meaning that
 322  * ptep_modify_prot_commit may not actually update the pte, but merely
 323  * queue the update to be done at some later time.  The update must be
 324  * actually committed before the pte lock is released, however.
 325  */
 326 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
 327                                            unsigned long addr,
 328                                            pte_t *ptep)
 329 {
 330         return __ptep_modify_prot_start(mm, addr, ptep);
 331 }
 332
 333 /*
 334  * Commit an update to a pte, leaving any hardware-controlled bits in
 335  * the PTE unmodified.
 336  */
 337 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
 338                                            unsigned long addr,
 339                                            pte_t *ptep, pte_t pte)
 340 {
 341         __ptep_modify_prot_commit(mm, addr, ptep, pte);
 342 }
 343 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
 344 #endif /* CONFIG_MMU */
 345
 346 /*
 347  * A facility to provide lazy MMU batching.  This allows PTE updates and
 348  * page invalidations to be delayed until a call to leave lazy MMU mode
 349  * is issued.  Some architectures may benefit from doing this, and it is
 350  * beneficial for both shadow and direct mode hypervisors, which may batch
 351  * the PTE updates which happen during this window.  Note that using this
 352  * interface requires that read hazards be removed from the code.  A read
 353  * hazard could result in the direct mode hypervisor case, since the actual
 354  * write to the page tables may not yet have taken place, so reads though
 355  * a raw PTE pointer after it has been modified are not guaranteed to be
 356  * up to date.  This mode can only be entered and left under the protection of
 357  * the page table locks for all page tables which may be modified.  In the UP
 358  * case, this is required so that preemption is disabled, and in the SMP case,
 359  * it must synchronize the delayed page table writes properly on other CPUs.
 360  */
 361 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 362 #define arch_enter_lazy_mmu_mode()      do {} while (0)
 363 #define arch_leave_lazy_mmu_mode()      do {} while (0)
 364 #define arch_flush_lazy_mmu_mode()      do {} while (0)
 365 #endif
 366
 367 /*
 368  * A facility to provide batching of the reload of page tables and
 369  * other process state with the actual context switch code for
 370  * paravirtualized guests.  By convention, only one of the batched
 371  * update (lazy) modes (CPU, MMU) should be active at any given time,
 372  * entry should never be nested, and entry and exits should always be
 373  * paired.  This is for sanity of maintaining and reasoning about the
 374  * kernel code.  In this case, the exit (end of the context switch) is
 375  * in architecture-specific code, and so doesn't need a generic
 376  * definition.
 377  */
 378 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
 379 #define arch_start_context_switch(prev) do {} while (0)
 380 #endif
 381
 382 #ifndef __HAVE_PFNMAP_TRACKING
 383 /*
 384  * Interface that can be used by architecture code to keep track of
 385  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 386  *
 387  * track_pfn_vma_new is called when a _new_ pfn mapping is being established
 388  * for physical range indicated by pfn and size.
 389  */
 390 static inline int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 391                                         unsigned long pfn, unsigned long size)
 392 {
 393         return 0;
 394 }
 395
 396 /*
 397  * Interface that can be used by architecture code to keep track of
 398  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 399  *
 400  * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
 401  * copied through copy_page_range().
 402  */
 403 static inline int track_pfn_vma_copy(struct vm_area_struct *vma)
 404 {
 405         return 0;
 406 }
 407
 408 /*
 409  * Interface that can be used by architecture code to keep track of
 410  * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
 411  *
 412  * untrack_pfn_vma is called while unmapping a pfnmap for a region.
 413  * untrack can be called for a specific region indicated by pfn and size or
 414  * can be for the entire vma (in which case size can be zero).
 415  */
 416 static inline void untrack_pfn_vma(struct vm_area_struct *vma,
 417                                         unsigned long pfn, unsigned long size)
 418 {
 419 }
 420 #else
 421 extern int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
 422                                 unsigned long pfn, unsigned long size);
 423 extern int track_pfn_vma_copy(struct vm_area_struct *vma);
 424 extern void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
 425                                 unsigned long size);
 426 #endif
 427
 428 #ifdef CONFIG_MMU
 429
 430 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
 431 static inline int pmd_trans_huge(pmd_t pmd)
 432 {
 433         return 0;
 434 }
 435 static inline int pmd_trans_splitting(pmd_t pmd)
 436 {
 437         return 0;
 438 }
 439 #ifndef __HAVE_ARCH_PMD_WRITE
 440 static inline int pmd_write(pmd_t pmd)
 441 {
 442         BUG();
 443         return 0;
 444 }
 445 #endif /* __HAVE_ARCH_PMD_WRITE */
 446 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 447
 448 #ifndef pmd_read_atomic
 449 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
 450 {
 451         /*
 452          * Depend on compiler for an atomic pmd read. NOTE: this is
 453          * only going to work, if the pmdval_t isn't larger than
 454          * an unsigned long.
 455          */
 456         return *pmdp;
 457 }
 458 #endif
 459
 460 /*
 461  * This function is meant to be used by sites walking pagetables with
 462  * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
 463  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
 464  * into a null pmd and the transhuge page fault can convert a null pmd
 465  * into an hugepmd or into a regular pmd (if the hugepage allocation
 466  * fails). While holding the mmap_sem in read mode the pmd becomes
 467  * stable and stops changing under us only if it's not null and not a
 468  * transhuge pmd. When those races occurs and this function makes a
 469  * difference vs the standard pmd_none_or_clear_bad, the result is
 470  * undefined so behaving like if the pmd was none is safe (because it
 471  * can return none anyway). The compiler level barrier() is critically
 472  * important to compute the two checks atomically on the same pmdval.
 473  *
 474  * For 32bit kernels with a 64bit large pmd_t this automatically takes
 475  * care of reading the pmd atomically to avoid SMP race conditions
 476  * against pmd_populate() when the mmap_sem is hold for reading by the
 477  * caller (a special atomic read not done by "gcc" as in the generic
 478  * version above, is also needed when THP is disabled because the page
 479  * fault can populate the pmd from under us).
 480  */
 481 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
 482 {
 483         pmd_t pmdval = pmd_read_atomic(pmd);
 484         /*
 485          * The barrier will stabilize the pmdval in a register or on
 486          * the stack so that it will stop changing under the code.
 487          */
 488 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 489         barrier();
 490 #endif
 491         if (pmd_none(pmdval))
 492                 return 1;
 493         if (unlikely(pmd_bad(pmdval))) {
 494                 if (!pmd_trans_huge(pmdval))
 495                         pmd_clear_bad(pmd);
 496                 return 1;
 497         }
 498         return 0;
 499 }
 500
 501 /*
 502  * This is a noop if Transparent Hugepage Support is not built into
 503  * the kernel. Otherwise it is equivalent to
 504  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
 505  * places that already verified the pmd is not none and they want to
 506  * walk ptes while holding the mmap sem in read mode (write mode don't
 507  * need this). If THP is not enabled, the pmd can't go away under the
 508  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
 509  * run a pmd_trans_unstable before walking the ptes after
 510  * split_huge_page_pmd returns (because it may have run when the pmd
 511  * become null, but then a page fault can map in a THP and not a
 512  * regular page).
 513  */
 514 static inline int pmd_trans_unstable(pmd_t *pmd)
 515 {
 516 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 517         return pmd_none_or_trans_huge_or_clear_bad(pmd);
 518 #else
 519         return 0;
 520 #endif
 521 }
 522
 523 #endif /* CONFIG_MMU */
 524
 525 #endif /* !__ASSEMBLY__ */
 526
 527 #endif /* _ASM_GENERIC_PGTABLE_H */