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