pgtable.h 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730
  1. #ifndef _ASM_GENERIC_PGTABLE_H
  2. #define _ASM_GENERIC_PGTABLE_H
  3. #ifndef __ASSEMBLY__
  4. #ifdef CONFIG_MMU
  5. #include <linux/mm_types.h>
  6. #include <linux/bug.h>
  7. #include <linux/errno.h>
  8. #if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
  9. CONFIG_PGTABLE_LEVELS
  10. #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
  11. #endif
  12. /*
  13. * On almost all architectures and configurations, 0 can be used as the
  14. * upper ceiling to free_pgtables(): on many architectures it has the same
  15. * effect as using TASK_SIZE. However, there is one configuration which
  16. * must impose a more careful limit, to avoid freeing kernel pgtables.
  17. */
  18. #ifndef USER_PGTABLES_CEILING
  19. #define USER_PGTABLES_CEILING 0UL
  20. #endif
  21. #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
  22. extern int ptep_set_access_flags(struct vm_area_struct *vma,
  23. unsigned long address, pte_t *ptep,
  24. pte_t entry, int dirty);
  25. #endif
  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. #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
  32. static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
  33. unsigned long address,
  34. pte_t *ptep)
  35. {
  36. pte_t pte = *ptep;
  37. int r = 1;
  38. if (!pte_young(pte))
  39. r = 0;
  40. else
  41. set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
  42. return r;
  43. }
  44. #endif
  45. #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
  46. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  47. static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  48. unsigned long address,
  49. pmd_t *pmdp)
  50. {
  51. pmd_t pmd = *pmdp;
  52. int r = 1;
  53. if (!pmd_young(pmd))
  54. r = 0;
  55. else
  56. set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
  57. return r;
  58. }
  59. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  60. static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
  61. unsigned long address,
  62. pmd_t *pmdp)
  63. {
  64. BUG();
  65. return 0;
  66. }
  67. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  68. #endif
  69. #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
  70. int ptep_clear_flush_young(struct vm_area_struct *vma,
  71. unsigned long address, pte_t *ptep);
  72. #endif
  73. #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
  74. int pmdp_clear_flush_young(struct vm_area_struct *vma,
  75. unsigned long address, pmd_t *pmdp);
  76. #endif
  77. #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
  78. static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
  79. unsigned long address,
  80. pte_t *ptep)
  81. {
  82. pte_t pte = *ptep;
  83. pte_clear(mm, address, ptep);
  84. return pte;
  85. }
  86. #endif
  87. #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
  88. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  89. static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
  90. unsigned long address,
  91. pmd_t *pmdp)
  92. {
  93. pmd_t pmd = *pmdp;
  94. pmd_clear(pmdp);
  95. return pmd;
  96. }
  97. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  98. #endif
  99. #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
  100. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  101. static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
  102. unsigned long address, pmd_t *pmdp,
  103. int full)
  104. {
  105. return pmdp_get_and_clear(mm, address, pmdp);
  106. }
  107. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  108. #endif
  109. #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
  110. static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
  111. unsigned long address, pte_t *ptep,
  112. int full)
  113. {
  114. pte_t pte;
  115. pte = ptep_get_and_clear(mm, address, ptep);
  116. return pte;
  117. }
  118. #endif
  119. /*
  120. * Some architectures may be able to avoid expensive synchronization
  121. * primitives when modifications are made to PTE's which are already
  122. * not present, or in the process of an address space destruction.
  123. */
  124. #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
  125. static inline void pte_clear_not_present_full(struct mm_struct *mm,
  126. unsigned long address,
  127. pte_t *ptep,
  128. int full)
  129. {
  130. pte_clear(mm, address, ptep);
  131. }
  132. #endif
  133. #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
  134. extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
  135. unsigned long address,
  136. pte_t *ptep);
  137. #endif
  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. #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
  144. struct mm_struct;
  145. static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
  146. {
  147. pte_t old_pte = *ptep;
  148. set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
  149. }
  150. #endif
  151. #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
  152. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  153. static inline void pmdp_set_wrprotect(struct mm_struct *mm,
  154. unsigned long address, pmd_t *pmdp)
  155. {
  156. pmd_t old_pmd = *pmdp;
  157. set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
  158. }
  159. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  160. static inline void pmdp_set_wrprotect(struct mm_struct *mm,
  161. unsigned long address, pmd_t *pmdp)
  162. {
  163. BUG();
  164. }
  165. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  166. #endif
  167. #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
  168. extern void pmdp_splitting_flush(struct vm_area_struct *vma,
  169. unsigned long address, pmd_t *pmdp);
  170. #endif
  171. #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
  172. extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
  173. pgtable_t pgtable);
  174. #endif
  175. #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
  176. extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
  177. #endif
  178. #ifndef __HAVE_ARCH_PMDP_INVALIDATE
  179. extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
  180. pmd_t *pmdp);
  181. #endif
  182. #ifndef __HAVE_ARCH_PTE_SAME
  183. static inline int pte_same(pte_t pte_a, pte_t pte_b)
  184. {
  185. return pte_val(pte_a) == pte_val(pte_b);
  186. }
  187. #endif
  188. #ifndef __HAVE_ARCH_PTE_UNUSED
  189. /*
  190. * Some architectures provide facilities to virtualization guests
  191. * so that they can flag allocated pages as unused. This allows the
  192. * host to transparently reclaim unused pages. This function returns
  193. * whether the pte's page is unused.
  194. */
  195. static inline int pte_unused(pte_t pte)
  196. {
  197. return 0;
  198. }
  199. #endif
  200. #ifndef __HAVE_ARCH_PMD_SAME
  201. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  202. static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
  203. {
  204. return pmd_val(pmd_a) == pmd_val(pmd_b);
  205. }
  206. #else /* CONFIG_TRANSPARENT_HUGEPAGE */
  207. static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
  208. {
  209. BUG();
  210. return 0;
  211. }
  212. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  213. #endif
  214. #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
  215. #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
  216. #endif
  217. #ifndef __HAVE_ARCH_MOVE_PTE
  218. #define move_pte(pte, prot, old_addr, new_addr) (pte)
  219. #endif
  220. #ifndef pte_accessible
  221. # define pte_accessible(mm, pte) ((void)(pte), 1)
  222. #endif
  223. #ifndef flush_tlb_fix_spurious_fault
  224. #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
  225. #endif
  226. #ifndef pgprot_noncached
  227. #define pgprot_noncached(prot) (prot)
  228. #endif
  229. #ifndef pgprot_writecombine
  230. #define pgprot_writecombine pgprot_noncached
  231. #endif
  232. #ifndef pgprot_device
  233. #define pgprot_device pgprot_noncached
  234. #endif
  235. #ifndef pgprot_modify
  236. #define pgprot_modify pgprot_modify
  237. static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
  238. {
  239. if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
  240. newprot = pgprot_noncached(newprot);
  241. if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
  242. newprot = pgprot_writecombine(newprot);
  243. if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
  244. newprot = pgprot_device(newprot);
  245. return newprot;
  246. }
  247. #endif
  248. /*
  249. * When walking page tables, get the address of the next boundary,
  250. * or the end address of the range if that comes earlier. Although no
  251. * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
  252. */
  253. #define pgd_addr_end(addr, end) \
  254. ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
  255. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  256. })
  257. #ifndef pud_addr_end
  258. #define pud_addr_end(addr, end) \
  259. ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
  260. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  261. })
  262. #endif
  263. #ifndef pmd_addr_end
  264. #define pmd_addr_end(addr, end) \
  265. ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
  266. (__boundary - 1 < (end) - 1)? __boundary: (end); \
  267. })
  268. #endif
  269. /*
  270. * When walking page tables, we usually want to skip any p?d_none entries;
  271. * and any p?d_bad entries - reporting the error before resetting to none.
  272. * Do the tests inline, but report and clear the bad entry in mm/memory.c.
  273. */
  274. void pgd_clear_bad(pgd_t *);
  275. void pud_clear_bad(pud_t *);
  276. void pmd_clear_bad(pmd_t *);
  277. static inline int pgd_none_or_clear_bad(pgd_t *pgd)
  278. {
  279. if (pgd_none(*pgd))
  280. return 1;
  281. if (unlikely(pgd_bad(*pgd))) {
  282. pgd_clear_bad(pgd);
  283. return 1;
  284. }
  285. return 0;
  286. }
  287. static inline int pud_none_or_clear_bad(pud_t *pud)
  288. {
  289. if (pud_none(*pud))
  290. return 1;
  291. if (unlikely(pud_bad(*pud))) {
  292. pud_clear_bad(pud);
  293. return 1;
  294. }
  295. return 0;
  296. }
  297. static inline int pmd_none_or_clear_bad(pmd_t *pmd)
  298. {
  299. if (pmd_none(*pmd))
  300. return 1;
  301. if (unlikely(pmd_bad(*pmd))) {
  302. pmd_clear_bad(pmd);
  303. return 1;
  304. }
  305. return 0;
  306. }
  307. static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
  308. unsigned long addr,
  309. pte_t *ptep)
  310. {
  311. /*
  312. * Get the current pte state, but zero it out to make it
  313. * non-present, preventing the hardware from asynchronously
  314. * updating it.
  315. */
  316. return ptep_get_and_clear(mm, addr, ptep);
  317. }
  318. static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
  319. unsigned long addr,
  320. pte_t *ptep, pte_t pte)
  321. {
  322. /*
  323. * The pte is non-present, so there's no hardware state to
  324. * preserve.
  325. */
  326. set_pte_at(mm, addr, ptep, pte);
  327. }
  328. #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
  329. /*
  330. * Start a pte protection read-modify-write transaction, which
  331. * protects against asynchronous hardware modifications to the pte.
  332. * The intention is not to prevent the hardware from making pte
  333. * updates, but to prevent any updates it may make from being lost.
  334. *
  335. * This does not protect against other software modifications of the
  336. * pte; the appropriate pte lock must be held over the transation.
  337. *
  338. * Note that this interface is intended to be batchable, meaning that
  339. * ptep_modify_prot_commit may not actually update the pte, but merely
  340. * queue the update to be done at some later time. The update must be
  341. * actually committed before the pte lock is released, however.
  342. */
  343. static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
  344. unsigned long addr,
  345. pte_t *ptep)
  346. {
  347. return __ptep_modify_prot_start(mm, addr, ptep);
  348. }
  349. /*
  350. * Commit an update to a pte, leaving any hardware-controlled bits in
  351. * the PTE unmodified.
  352. */
  353. static inline void ptep_modify_prot_commit(struct mm_struct *mm,
  354. unsigned long addr,
  355. pte_t *ptep, pte_t pte)
  356. {
  357. __ptep_modify_prot_commit(mm, addr, ptep, pte);
  358. }
  359. #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
  360. #endif /* CONFIG_MMU */
  361. /*
  362. * A facility to provide lazy MMU batching. This allows PTE updates and
  363. * page invalidations to be delayed until a call to leave lazy MMU mode
  364. * is issued. Some architectures may benefit from doing this, and it is
  365. * beneficial for both shadow and direct mode hypervisors, which may batch
  366. * the PTE updates which happen during this window. Note that using this
  367. * interface requires that read hazards be removed from the code. A read
  368. * hazard could result in the direct mode hypervisor case, since the actual
  369. * write to the page tables may not yet have taken place, so reads though
  370. * a raw PTE pointer after it has been modified are not guaranteed to be
  371. * up to date. This mode can only be entered and left under the protection of
  372. * the page table locks for all page tables which may be modified. In the UP
  373. * case, this is required so that preemption is disabled, and in the SMP case,
  374. * it must synchronize the delayed page table writes properly on other CPUs.
  375. */
  376. #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
  377. #define arch_enter_lazy_mmu_mode() do {} while (0)
  378. #define arch_leave_lazy_mmu_mode() do {} while (0)
  379. #define arch_flush_lazy_mmu_mode() do {} while (0)
  380. #endif
  381. /*
  382. * A facility to provide batching of the reload of page tables and
  383. * other process state with the actual context switch code for
  384. * paravirtualized guests. By convention, only one of the batched
  385. * update (lazy) modes (CPU, MMU) should be active at any given time,
  386. * entry should never be nested, and entry and exits should always be
  387. * paired. This is for sanity of maintaining and reasoning about the
  388. * kernel code. In this case, the exit (end of the context switch) is
  389. * in architecture-specific code, and so doesn't need a generic
  390. * definition.
  391. */
  392. #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
  393. #define arch_start_context_switch(prev) do {} while (0)
  394. #endif
  395. #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
  396. static inline int pte_soft_dirty(pte_t pte)
  397. {
  398. return 0;
  399. }
  400. static inline int pmd_soft_dirty(pmd_t pmd)
  401. {
  402. return 0;
  403. }
  404. static inline pte_t pte_mksoft_dirty(pte_t pte)
  405. {
  406. return pte;
  407. }
  408. static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
  409. {
  410. return pmd;
  411. }
  412. static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
  413. {
  414. return pte;
  415. }
  416. static inline int pte_swp_soft_dirty(pte_t pte)
  417. {
  418. return 0;
  419. }
  420. static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
  421. {
  422. return pte;
  423. }
  424. #endif
  425. #ifndef __HAVE_PFNMAP_TRACKING
  426. /*
  427. * Interfaces that can be used by architecture code to keep track of
  428. * memory type of pfn mappings specified by the remap_pfn_range,
  429. * vm_insert_pfn.
  430. */
  431. /*
  432. * track_pfn_remap is called when a _new_ pfn mapping is being established
  433. * by remap_pfn_range() for physical range indicated by pfn and size.
  434. */
  435. static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
  436. unsigned long pfn, unsigned long addr,
  437. unsigned long size)
  438. {
  439. return 0;
  440. }
  441. /*
  442. * track_pfn_insert is called when a _new_ single pfn is established
  443. * by vm_insert_pfn().
  444. */
  445. static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
  446. unsigned long pfn)
  447. {
  448. return 0;
  449. }
  450. /*
  451. * track_pfn_copy is called when vma that is covering the pfnmap gets
  452. * copied through copy_page_range().
  453. */
  454. static inline int track_pfn_copy(struct vm_area_struct *vma)
  455. {
  456. return 0;
  457. }
  458. /*
  459. * untrack_pfn_vma is called while unmapping a pfnmap for a region.
  460. * untrack can be called for a specific region indicated by pfn and size or
  461. * can be for the entire vma (in which case pfn, size are zero).
  462. */
  463. static inline void untrack_pfn(struct vm_area_struct *vma,
  464. unsigned long pfn, unsigned long size)
  465. {
  466. }
  467. #else
  468. extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
  469. unsigned long pfn, unsigned long addr,
  470. unsigned long size);
  471. extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
  472. unsigned long pfn);
  473. extern int track_pfn_copy(struct vm_area_struct *vma);
  474. extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
  475. unsigned long size);
  476. #endif
  477. #ifdef __HAVE_COLOR_ZERO_PAGE
  478. static inline int is_zero_pfn(unsigned long pfn)
  479. {
  480. extern unsigned long zero_pfn;
  481. unsigned long offset_from_zero_pfn = pfn - zero_pfn;
  482. return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
  483. }
  484. #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
  485. #else
  486. static inline int is_zero_pfn(unsigned long pfn)
  487. {
  488. extern unsigned long zero_pfn;
  489. return pfn == zero_pfn;
  490. }
  491. static inline unsigned long my_zero_pfn(unsigned long addr)
  492. {
  493. extern unsigned long zero_pfn;
  494. return zero_pfn;
  495. }
  496. #endif
  497. #ifdef CONFIG_MMU
  498. #ifndef CONFIG_TRANSPARENT_HUGEPAGE
  499. static inline int pmd_trans_huge(pmd_t pmd)
  500. {
  501. return 0;
  502. }
  503. static inline int pmd_trans_splitting(pmd_t pmd)
  504. {
  505. return 0;
  506. }
  507. #ifndef __HAVE_ARCH_PMD_WRITE
  508. static inline int pmd_write(pmd_t pmd)
  509. {
  510. BUG();
  511. return 0;
  512. }
  513. #endif /* __HAVE_ARCH_PMD_WRITE */
  514. #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  515. #ifndef pmd_read_atomic
  516. static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
  517. {
  518. /*
  519. * Depend on compiler for an atomic pmd read. NOTE: this is
  520. * only going to work, if the pmdval_t isn't larger than
  521. * an unsigned long.
  522. */
  523. return *pmdp;
  524. }
  525. #endif
  526. #ifndef pmd_move_must_withdraw
  527. static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
  528. spinlock_t *old_pmd_ptl)
  529. {
  530. /*
  531. * With split pmd lock we also need to move preallocated
  532. * PTE page table if new_pmd is on different PMD page table.
  533. */
  534. return new_pmd_ptl != old_pmd_ptl;
  535. }
  536. #endif
  537. /*
  538. * This function is meant to be used by sites walking pagetables with
  539. * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
  540. * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
  541. * into a null pmd and the transhuge page fault can convert a null pmd
  542. * into an hugepmd or into a regular pmd (if the hugepage allocation
  543. * fails). While holding the mmap_sem in read mode the pmd becomes
  544. * stable and stops changing under us only if it's not null and not a
  545. * transhuge pmd. When those races occurs and this function makes a
  546. * difference vs the standard pmd_none_or_clear_bad, the result is
  547. * undefined so behaving like if the pmd was none is safe (because it
  548. * can return none anyway). The compiler level barrier() is critically
  549. * important to compute the two checks atomically on the same pmdval.
  550. *
  551. * For 32bit kernels with a 64bit large pmd_t this automatically takes
  552. * care of reading the pmd atomically to avoid SMP race conditions
  553. * against pmd_populate() when the mmap_sem is hold for reading by the
  554. * caller (a special atomic read not done by "gcc" as in the generic
  555. * version above, is also needed when THP is disabled because the page
  556. * fault can populate the pmd from under us).
  557. */
  558. static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
  559. {
  560. pmd_t pmdval = pmd_read_atomic(pmd);
  561. /*
  562. * The barrier will stabilize the pmdval in a register or on
  563. * the stack so that it will stop changing under the code.
  564. *
  565. * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
  566. * pmd_read_atomic is allowed to return a not atomic pmdval
  567. * (for example pointing to an hugepage that has never been
  568. * mapped in the pmd). The below checks will only care about
  569. * the low part of the pmd with 32bit PAE x86 anyway, with the
  570. * exception of pmd_none(). So the important thing is that if
  571. * the low part of the pmd is found null, the high part will
  572. * be also null or the pmd_none() check below would be
  573. * confused.
  574. */
  575. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  576. barrier();
  577. #endif
  578. if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
  579. return 1;
  580. if (unlikely(pmd_bad(pmdval))) {
  581. pmd_clear_bad(pmd);
  582. return 1;
  583. }
  584. return 0;
  585. }
  586. /*
  587. * This is a noop if Transparent Hugepage Support is not built into
  588. * the kernel. Otherwise it is equivalent to
  589. * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
  590. * places that already verified the pmd is not none and they want to
  591. * walk ptes while holding the mmap sem in read mode (write mode don't
  592. * need this). If THP is not enabled, the pmd can't go away under the
  593. * code even if MADV_DONTNEED runs, but if THP is enabled we need to
  594. * run a pmd_trans_unstable before walking the ptes after
  595. * split_huge_page_pmd returns (because it may have run when the pmd
  596. * become null, but then a page fault can map in a THP and not a
  597. * regular page).
  598. */
  599. static inline int pmd_trans_unstable(pmd_t *pmd)
  600. {
  601. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  602. return pmd_none_or_trans_huge_or_clear_bad(pmd);
  603. #else
  604. return 0;
  605. #endif
  606. }
  607. #ifndef CONFIG_NUMA_BALANCING
  608. /*
  609. * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
  610. * the only case the kernel cares is for NUMA balancing and is only ever set
  611. * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
  612. * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
  613. * is the responsibility of the caller to distinguish between PROT_NONE
  614. * protections and NUMA hinting fault protections.
  615. */
  616. static inline int pte_protnone(pte_t pte)
  617. {
  618. return 0;
  619. }
  620. static inline int pmd_protnone(pmd_t pmd)
  621. {
  622. return 0;
  623. }
  624. #endif /* CONFIG_NUMA_BALANCING */
  625. #endif /* CONFIG_MMU */
  626. #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
  627. int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
  628. int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
  629. int pud_clear_huge(pud_t *pud);
  630. int pmd_clear_huge(pmd_t *pmd);
  631. #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
  632. static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
  633. {
  634. return 0;
  635. }
  636. static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
  637. {
  638. return 0;
  639. }
  640. static inline int pud_clear_huge(pud_t *pud)
  641. {
  642. return 0;
  643. }
  644. static inline int pmd_clear_huge(pmd_t *pmd)
  645. {
  646. return 0;
  647. }
  648. #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
  649. #endif /* !__ASSEMBLY__ */
  650. #ifndef io_remap_pfn_range
  651. #define io_remap_pfn_range remap_pfn_range
  652. #endif
  653. #endif /* _ASM_GENERIC_PGTABLE_H */