pgtable.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466
  1. /*
  2. * Copyright 2010 Tilera Corporation. All Rights Reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public License
  6. * as published by the Free Software Foundation, version 2.
  7. *
  8. * This program is distributed in the hope that it will be useful, but
  9. * WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  11. * NON INFRINGEMENT. See the GNU General Public License for
  12. * more details.
  13. *
  14. * This file contains the functions and defines necessary to modify and use
  15. * the TILE page table tree.
  16. */
  17. #ifndef _ASM_TILE_PGTABLE_H
  18. #define _ASM_TILE_PGTABLE_H
  19. #include <hv/hypervisor.h>
  20. #ifndef __ASSEMBLY__
  21. #include <linux/bitops.h>
  22. #include <linux/threads.h>
  23. #include <linux/slab.h>
  24. #include <linux/list.h>
  25. #include <linux/spinlock.h>
  26. #include <asm/processor.h>
  27. #include <asm/fixmap.h>
  28. struct mm_struct;
  29. struct vm_area_struct;
  30. /*
  31. * ZERO_PAGE is a global shared page that is always zero: used
  32. * for zero-mapped memory areas etc..
  33. */
  34. extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
  35. #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
  36. extern pgd_t swapper_pg_dir[];
  37. extern pgprot_t swapper_pgprot;
  38. extern struct kmem_cache *pgd_cache;
  39. extern spinlock_t pgd_lock;
  40. extern struct list_head pgd_list;
  41. /*
  42. * The very last slots in the pgd_t are for addresses unusable by Linux
  43. * (pgd_addr_invalid() returns true). So we use them for the list structure.
  44. * The x86 code we are modelled on uses the page->private/index fields
  45. * (older 2.6 kernels) or the lru list (newer 2.6 kernels), but since
  46. * our pgds are so much smaller than a page, it seems a waste to
  47. * spend a whole page on each pgd.
  48. */
  49. #define PGD_LIST_OFFSET \
  50. ((PTRS_PER_PGD * sizeof(pgd_t)) - sizeof(struct list_head))
  51. #define pgd_to_list(pgd) \
  52. ((struct list_head *)((char *)(pgd) + PGD_LIST_OFFSET))
  53. #define list_to_pgd(list) \
  54. ((pgd_t *)((char *)(list) - PGD_LIST_OFFSET))
  55. extern void pgtable_cache_init(void);
  56. extern void paging_init(void);
  57. extern void set_page_homes(void);
  58. #define FIRST_USER_ADDRESS 0
  59. #define _PAGE_PRESENT HV_PTE_PRESENT
  60. #define _PAGE_HUGE_PAGE HV_PTE_PAGE
  61. #define _PAGE_READABLE HV_PTE_READABLE
  62. #define _PAGE_WRITABLE HV_PTE_WRITABLE
  63. #define _PAGE_EXECUTABLE HV_PTE_EXECUTABLE
  64. #define _PAGE_ACCESSED HV_PTE_ACCESSED
  65. #define _PAGE_DIRTY HV_PTE_DIRTY
  66. #define _PAGE_GLOBAL HV_PTE_GLOBAL
  67. #define _PAGE_USER HV_PTE_USER
  68. /*
  69. * All the "standard" bits. Cache-control bits are managed elsewhere.
  70. * This is used to test for valid level-2 page table pointers by checking
  71. * all the bits, and to mask away the cache control bits for mprotect.
  72. */
  73. #define _PAGE_ALL (\
  74. _PAGE_PRESENT | \
  75. _PAGE_HUGE_PAGE | \
  76. _PAGE_READABLE | \
  77. _PAGE_WRITABLE | \
  78. _PAGE_EXECUTABLE | \
  79. _PAGE_ACCESSED | \
  80. _PAGE_DIRTY | \
  81. _PAGE_GLOBAL | \
  82. _PAGE_USER \
  83. )
  84. #define PAGE_NONE \
  85. __pgprot(_PAGE_PRESENT | _PAGE_ACCESSED)
  86. #define PAGE_SHARED \
  87. __pgprot(_PAGE_PRESENT | _PAGE_READABLE | _PAGE_WRITABLE | \
  88. _PAGE_USER | _PAGE_ACCESSED)
  89. #define PAGE_SHARED_EXEC \
  90. __pgprot(_PAGE_PRESENT | _PAGE_READABLE | _PAGE_WRITABLE | \
  91. _PAGE_EXECUTABLE | _PAGE_USER | _PAGE_ACCESSED)
  92. #define PAGE_COPY_NOEXEC \
  93. __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_READABLE)
  94. #define PAGE_COPY_EXEC \
  95. __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | \
  96. _PAGE_READABLE | _PAGE_EXECUTABLE)
  97. #define PAGE_COPY \
  98. PAGE_COPY_NOEXEC
  99. #define PAGE_READONLY \
  100. __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_READABLE)
  101. #define PAGE_READONLY_EXEC \
  102. __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | \
  103. _PAGE_READABLE | _PAGE_EXECUTABLE)
  104. #define _PAGE_KERNEL_RO \
  105. (_PAGE_PRESENT | _PAGE_GLOBAL | _PAGE_READABLE | _PAGE_ACCESSED)
  106. #define _PAGE_KERNEL \
  107. (_PAGE_KERNEL_RO | _PAGE_WRITABLE | _PAGE_DIRTY)
  108. #define _PAGE_KERNEL_EXEC (_PAGE_KERNEL_RO | _PAGE_EXECUTABLE)
  109. #define PAGE_KERNEL __pgprot(_PAGE_KERNEL)
  110. #define PAGE_KERNEL_RO __pgprot(_PAGE_KERNEL_RO)
  111. #define PAGE_KERNEL_EXEC __pgprot(_PAGE_KERNEL_EXEC)
  112. #define page_to_kpgprot(p) PAGE_KERNEL
  113. /*
  114. * We could tighten these up, but for now writable or executable
  115. * implies readable.
  116. */
  117. #define __P000 PAGE_NONE
  118. #define __P001 PAGE_READONLY
  119. #define __P010 PAGE_COPY /* this is write-only, which we won't support */
  120. #define __P011 PAGE_COPY
  121. #define __P100 PAGE_READONLY_EXEC
  122. #define __P101 PAGE_READONLY_EXEC
  123. #define __P110 PAGE_COPY_EXEC
  124. #define __P111 PAGE_COPY_EXEC
  125. #define __S000 PAGE_NONE
  126. #define __S001 PAGE_READONLY
  127. #define __S010 PAGE_SHARED
  128. #define __S011 PAGE_SHARED
  129. #define __S100 PAGE_READONLY_EXEC
  130. #define __S101 PAGE_READONLY_EXEC
  131. #define __S110 PAGE_SHARED_EXEC
  132. #define __S111 PAGE_SHARED_EXEC
  133. /*
  134. * All the normal _PAGE_ALL bits are ignored for PMDs, except PAGE_PRESENT
  135. * and PAGE_HUGE_PAGE, which must be one and zero, respectively.
  136. * We set the ignored bits to zero.
  137. */
  138. #define _PAGE_TABLE _PAGE_PRESENT
  139. /* Inherit the caching flags from the old protection bits. */
  140. #define pgprot_modify(oldprot, newprot) \
  141. (pgprot_t) { ((oldprot).val & ~_PAGE_ALL) | (newprot).val }
  142. /* Just setting the PFN to zero suffices. */
  143. #define pte_pgprot(x) hv_pte_set_pfn((x), 0)
  144. /*
  145. * For PTEs and PDEs, we must clear the Present bit first when
  146. * clearing a page table entry, so clear the bottom half first and
  147. * enforce ordering with a barrier.
  148. */
  149. static inline void __pte_clear(pte_t *ptep)
  150. {
  151. #ifdef __tilegx__
  152. ptep->val = 0;
  153. #else
  154. u32 *tmp = (u32 *)ptep;
  155. tmp[0] = 0;
  156. barrier();
  157. tmp[1] = 0;
  158. #endif
  159. }
  160. #define pte_clear(mm, addr, ptep) __pte_clear(ptep)
  161. /*
  162. * The following only work if pte_present() is true.
  163. * Undefined behaviour if not..
  164. */
  165. #define pte_present hv_pte_get_present
  166. #define pte_user hv_pte_get_user
  167. #define pte_read hv_pte_get_readable
  168. #define pte_dirty hv_pte_get_dirty
  169. #define pte_young hv_pte_get_accessed
  170. #define pte_write hv_pte_get_writable
  171. #define pte_exec hv_pte_get_executable
  172. #define pte_huge hv_pte_get_page
  173. #define pte_rdprotect hv_pte_clear_readable
  174. #define pte_exprotect hv_pte_clear_executable
  175. #define pte_mkclean hv_pte_clear_dirty
  176. #define pte_mkold hv_pte_clear_accessed
  177. #define pte_wrprotect hv_pte_clear_writable
  178. #define pte_mksmall hv_pte_clear_page
  179. #define pte_mkread hv_pte_set_readable
  180. #define pte_mkexec hv_pte_set_executable
  181. #define pte_mkdirty hv_pte_set_dirty
  182. #define pte_mkyoung hv_pte_set_accessed
  183. #define pte_mkwrite hv_pte_set_writable
  184. #define pte_mkhuge hv_pte_set_page
  185. #define pte_special(pte) 0
  186. #define pte_mkspecial(pte) (pte)
  187. /*
  188. * Use some spare bits in the PTE for user-caching tags.
  189. */
  190. #define pte_set_forcecache hv_pte_set_client0
  191. #define pte_get_forcecache hv_pte_get_client0
  192. #define pte_clear_forcecache hv_pte_clear_client0
  193. #define pte_set_anyhome hv_pte_set_client1
  194. #define pte_get_anyhome hv_pte_get_client1
  195. #define pte_clear_anyhome hv_pte_clear_client1
  196. /*
  197. * A migrating PTE has PAGE_PRESENT clear but all the other bits preserved.
  198. */
  199. #define pte_migrating hv_pte_get_migrating
  200. #define pte_mkmigrate(x) hv_pte_set_migrating(hv_pte_clear_present(x))
  201. #define pte_donemigrate(x) hv_pte_set_present(hv_pte_clear_migrating(x))
  202. #define pte_ERROR(e) \
  203. pr_err("%s:%d: bad pte 0x%016llx.\n", __FILE__, __LINE__, pte_val(e))
  204. #define pgd_ERROR(e) \
  205. pr_err("%s:%d: bad pgd 0x%016llx.\n", __FILE__, __LINE__, pgd_val(e))
  206. /* Return PA and protection info for a given kernel VA. */
  207. int va_to_cpa_and_pte(void *va, phys_addr_t *cpa, pte_t *pte);
  208. /*
  209. * __set_pte() ensures we write the 64-bit PTE with 32-bit words in
  210. * the right order on 32-bit platforms and also allows us to write
  211. * hooks to check valid PTEs, etc., if we want.
  212. */
  213. void __set_pte(pte_t *ptep, pte_t pte);
  214. /*
  215. * set_pte() sets the given PTE and also sanity-checks the
  216. * requested PTE against the page homecaching. Unspecified parts
  217. * of the PTE are filled in when it is written to memory, i.e. all
  218. * caching attributes if "!forcecache", or the home cpu if "anyhome".
  219. */
  220. extern void set_pte(pte_t *ptep, pte_t pte);
  221. #define set_pte_at(mm, addr, ptep, pteval) set_pte(ptep, pteval)
  222. #define set_pte_atomic(pteptr, pteval) set_pte(pteptr, pteval)
  223. #define pte_page(x) pfn_to_page(pte_pfn(x))
  224. static inline int pte_none(pte_t pte)
  225. {
  226. return !pte.val;
  227. }
  228. static inline unsigned long pte_pfn(pte_t pte)
  229. {
  230. return hv_pte_get_pfn(pte);
  231. }
  232. /* Set or get the remote cache cpu in a pgprot with remote caching. */
  233. extern pgprot_t set_remote_cache_cpu(pgprot_t prot, int cpu);
  234. extern int get_remote_cache_cpu(pgprot_t prot);
  235. static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
  236. {
  237. return hv_pte_set_pfn(prot, pfn);
  238. }
  239. /* Support for priority mappings. */
  240. extern void start_mm_caching(struct mm_struct *mm);
  241. extern void check_mm_caching(struct mm_struct *prev, struct mm_struct *next);
  242. /*
  243. * Support non-linear file mappings (see sys_remap_file_pages).
  244. * This is defined by CLIENT1 set but CLIENT0 and _PAGE_PRESENT clear, and the
  245. * file offset in the 32 high bits.
  246. */
  247. #define _PAGE_FILE HV_PTE_CLIENT1
  248. #define PTE_FILE_MAX_BITS 32
  249. #define pte_file(pte) (hv_pte_get_client1(pte) && !hv_pte_get_client0(pte))
  250. #define pte_to_pgoff(pte) ((pte).val >> 32)
  251. #define pgoff_to_pte(off) ((pte_t) { (((long long)(off)) << 32) | _PAGE_FILE })
  252. /*
  253. * Encode and de-code a swap entry (see <linux/swapops.h>).
  254. * We put the swap file type+offset in the 32 high bits;
  255. * I believe we can just leave the low bits clear.
  256. */
  257. #define __swp_type(swp) ((swp).val & 0x1f)
  258. #define __swp_offset(swp) ((swp).val >> 5)
  259. #define __swp_entry(type, off) ((swp_entry_t) { (type) | ((off) << 5) })
  260. #define __pte_to_swp_entry(pte) ((swp_entry_t) { (pte).val >> 32 })
  261. #define __swp_entry_to_pte(swp) ((pte_t) { (((long long) ((swp).val)) << 32) })
  262. /*
  263. * Conversion functions: convert a page and protection to a page entry,
  264. * and a page entry and page directory to the page they refer to.
  265. */
  266. #define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
  267. /*
  268. * If we are doing an mprotect(), just accept the new vma->vm_page_prot
  269. * value and combine it with the PFN from the old PTE to get a new PTE.
  270. */
  271. static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
  272. {
  273. return pfn_pte(hv_pte_get_pfn(pte), newprot);
  274. }
  275. /*
  276. * The pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
  277. *
  278. * This macro returns the index of the entry in the pgd page which would
  279. * control the given virtual address.
  280. */
  281. #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
  282. /*
  283. * pgd_offset() returns a (pgd_t *)
  284. * pgd_index() is used get the offset into the pgd page's array of pgd_t's.
  285. */
  286. #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
  287. /*
  288. * A shortcut which implies the use of the kernel's pgd, instead
  289. * of a process's.
  290. */
  291. #define pgd_offset_k(address) pgd_offset(&init_mm, address)
  292. #if defined(CONFIG_HIGHPTE)
  293. extern pte_t *pte_offset_map(pmd_t *, unsigned long address);
  294. #define pte_unmap(pte) kunmap_atomic(pte)
  295. #else
  296. #define pte_offset_map(dir, address) pte_offset_kernel(dir, address)
  297. #define pte_unmap(pte) do { } while (0)
  298. #endif
  299. /* Clear a non-executable kernel PTE and flush it from the TLB. */
  300. #define kpte_clear_flush(ptep, vaddr) \
  301. do { \
  302. pte_clear(&init_mm, (vaddr), (ptep)); \
  303. local_flush_tlb_page(FLUSH_NONEXEC, (vaddr), PAGE_SIZE); \
  304. } while (0)
  305. /*
  306. * The kernel page tables contain what we need, and we flush when we
  307. * change specific page table entries.
  308. */
  309. #define update_mmu_cache(vma, address, pte) do { } while (0)
  310. #ifdef CONFIG_FLATMEM
  311. #define kern_addr_valid(addr) (1)
  312. #endif /* CONFIG_FLATMEM */
  313. #define io_remap_pfn_range(vma, vaddr, pfn, size, prot) \
  314. remap_pfn_range(vma, vaddr, pfn, size, prot)
  315. extern void vmalloc_sync_all(void);
  316. #endif /* !__ASSEMBLY__ */
  317. #ifdef __tilegx__
  318. #include <asm/pgtable_64.h>
  319. #else
  320. #include <asm/pgtable_32.h>
  321. #endif
  322. #ifndef __ASSEMBLY__
  323. static inline int pmd_none(pmd_t pmd)
  324. {
  325. /*
  326. * Only check low word on 32-bit platforms, since it might be
  327. * out of sync with upper half.
  328. */
  329. return (unsigned long)pmd_val(pmd) == 0;
  330. }
  331. static inline int pmd_present(pmd_t pmd)
  332. {
  333. return pmd_val(pmd) & _PAGE_PRESENT;
  334. }
  335. static inline int pmd_bad(pmd_t pmd)
  336. {
  337. return ((pmd_val(pmd) & _PAGE_ALL) != _PAGE_TABLE);
  338. }
  339. static inline unsigned long pages_to_mb(unsigned long npg)
  340. {
  341. return npg >> (20 - PAGE_SHIFT);
  342. }
  343. /*
  344. * The pmd can be thought of an array like this: pmd_t[PTRS_PER_PMD]
  345. *
  346. * This function returns the index of the entry in the pmd which would
  347. * control the given virtual address.
  348. */
  349. static inline unsigned long pmd_index(unsigned long address)
  350. {
  351. return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
  352. }
  353. /*
  354. * A given kernel pmd_t maps to a specific virtual address (either a
  355. * kernel huge page or a kernel pte_t table). Since kernel pte_t
  356. * tables can be aligned at sub-page granularity, this function can
  357. * return non-page-aligned pointers, despite its name.
  358. */
  359. static inline unsigned long pmd_page_vaddr(pmd_t pmd)
  360. {
  361. phys_addr_t pa =
  362. (phys_addr_t)pmd_ptfn(pmd) << HV_LOG2_PAGE_TABLE_ALIGN;
  363. return (unsigned long)__va(pa);
  364. }
  365. /*
  366. * A pmd_t points to the base of a huge page or to a pte_t array.
  367. * If a pte_t array, since we can have multiple per page, we don't
  368. * have a one-to-one mapping of pmd_t's to pages. However, this is
  369. * OK for pte_lockptr(), since we just end up with potentially one
  370. * lock being used for several pte_t arrays.
  371. */
  372. #define pmd_page(pmd) pfn_to_page(HV_PTFN_TO_PFN(pmd_ptfn(pmd)))
  373. /*
  374. * The pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
  375. *
  376. * This macro returns the index of the entry in the pte page which would
  377. * control the given virtual address.
  378. */
  379. static inline unsigned long pte_index(unsigned long address)
  380. {
  381. return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
  382. }
  383. static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
  384. {
  385. return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
  386. }
  387. static inline int pmd_huge_page(pmd_t pmd)
  388. {
  389. return pmd_val(pmd) & _PAGE_HUGE_PAGE;
  390. }
  391. #include <asm-generic/pgtable.h>
  392. /* Support /proc/NN/pgtable API. */
  393. struct seq_file;
  394. int arch_proc_pgtable_show(struct seq_file *m, struct mm_struct *mm,
  395. unsigned long vaddr, pte_t *ptep, void **datap);
  396. #endif /* !__ASSEMBLY__ */
  397. #endif /* _ASM_TILE_PGTABLE_H */