pgtable_32.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470
  1. /*
  2. * This file contains the routines setting up the linux page tables.
  3. * -- paulus
  4. *
  5. * Derived from arch/ppc/mm/init.c:
  6. * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  7. *
  8. * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
  9. * and Cort Dougan (PReP) (cort@cs.nmt.edu)
  10. * Copyright (C) 1996 Paul Mackerras
  11. *
  12. * Derived from "arch/i386/mm/init.c"
  13. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  14. *
  15. * This program is free software; you can redistribute it and/or
  16. * modify it under the terms of the GNU General Public License
  17. * as published by the Free Software Foundation; either version
  18. * 2 of the License, or (at your option) any later version.
  19. *
  20. */
  21. #include <linux/kernel.h>
  22. #include <linux/module.h>
  23. #include <linux/types.h>
  24. #include <linux/mm.h>
  25. #include <linux/vmalloc.h>
  26. #include <linux/init.h>
  27. #include <linux/highmem.h>
  28. #include <linux/memblock.h>
  29. #include <linux/slab.h>
  30. #include <asm/pgtable.h>
  31. #include <asm/pgalloc.h>
  32. #include <asm/fixmap.h>
  33. #include <asm/io.h>
  34. #include <asm/setup.h>
  35. #include "mmu_decl.h"
  36. unsigned long ioremap_base;
  37. unsigned long ioremap_bot;
  38. EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */
  39. #ifdef CONFIG_6xx
  40. #define HAVE_BATS 1
  41. #endif
  42. #if defined(CONFIG_FSL_BOOKE)
  43. #define HAVE_TLBCAM 1
  44. #endif
  45. extern char etext[], _stext[];
  46. #ifdef HAVE_BATS
  47. extern phys_addr_t v_mapped_by_bats(unsigned long va);
  48. extern unsigned long p_mapped_by_bats(phys_addr_t pa);
  49. void setbat(int index, unsigned long virt, phys_addr_t phys,
  50. unsigned int size, int flags);
  51. #else /* !HAVE_BATS */
  52. #define v_mapped_by_bats(x) (0UL)
  53. #define p_mapped_by_bats(x) (0UL)
  54. #endif /* HAVE_BATS */
  55. #ifdef HAVE_TLBCAM
  56. extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
  57. extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
  58. #else /* !HAVE_TLBCAM */
  59. #define v_mapped_by_tlbcam(x) (0UL)
  60. #define p_mapped_by_tlbcam(x) (0UL)
  61. #endif /* HAVE_TLBCAM */
  62. #define PGDIR_ORDER (32 + PGD_T_LOG2 - PGDIR_SHIFT)
  63. #ifndef CONFIG_PPC_4K_PAGES
  64. static struct kmem_cache *pgtable_cache;
  65. void pgtable_cache_init(void)
  66. {
  67. pgtable_cache = kmem_cache_create("PGDIR cache", 1 << PGDIR_ORDER,
  68. 1 << PGDIR_ORDER, 0, NULL);
  69. if (pgtable_cache == NULL)
  70. panic("Couldn't allocate pgtable caches");
  71. }
  72. #endif
  73. pgd_t *pgd_alloc(struct mm_struct *mm)
  74. {
  75. pgd_t *ret;
  76. /* pgdir take page or two with 4K pages and a page fraction otherwise */
  77. #ifndef CONFIG_PPC_4K_PAGES
  78. ret = kmem_cache_alloc(pgtable_cache, GFP_KERNEL | __GFP_ZERO);
  79. #else
  80. ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
  81. PGDIR_ORDER - PAGE_SHIFT);
  82. #endif
  83. return ret;
  84. }
  85. void pgd_free(struct mm_struct *mm, pgd_t *pgd)
  86. {
  87. #ifndef CONFIG_PPC_4K_PAGES
  88. kmem_cache_free(pgtable_cache, (void *)pgd);
  89. #else
  90. free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
  91. #endif
  92. }
  93. __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  94. {
  95. pte_t *pte;
  96. extern int mem_init_done;
  97. if (mem_init_done) {
  98. pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
  99. } else {
  100. pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE));
  101. if (pte)
  102. clear_page(pte);
  103. }
  104. return pte;
  105. }
  106. pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  107. {
  108. struct page *ptepage;
  109. gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
  110. ptepage = alloc_pages(flags, 0);
  111. if (!ptepage)
  112. return NULL;
  113. if (!pgtable_page_ctor(ptepage)) {
  114. __free_page(ptepage);
  115. return NULL;
  116. }
  117. return ptepage;
  118. }
  119. void __iomem *
  120. ioremap(phys_addr_t addr, unsigned long size)
  121. {
  122. return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
  123. __builtin_return_address(0));
  124. }
  125. EXPORT_SYMBOL(ioremap);
  126. void __iomem *
  127. ioremap_wc(phys_addr_t addr, unsigned long size)
  128. {
  129. return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
  130. __builtin_return_address(0));
  131. }
  132. EXPORT_SYMBOL(ioremap_wc);
  133. void __iomem *
  134. ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
  135. {
  136. /* writeable implies dirty for kernel addresses */
  137. if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO)
  138. flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
  139. /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
  140. flags &= ~(_PAGE_USER | _PAGE_EXEC);
  141. #ifdef _PAGE_BAP_SR
  142. /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
  143. * which means that we just cleared supervisor access... oops ;-) This
  144. * restores it
  145. */
  146. flags |= _PAGE_BAP_SR;
  147. #endif
  148. return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
  149. }
  150. EXPORT_SYMBOL(ioremap_prot);
  151. void __iomem *
  152. __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
  153. {
  154. return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
  155. }
  156. void __iomem *
  157. __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
  158. void *caller)
  159. {
  160. unsigned long v, i;
  161. phys_addr_t p;
  162. int err;
  163. /* Make sure we have the base flags */
  164. if ((flags & _PAGE_PRESENT) == 0)
  165. flags |= PAGE_KERNEL;
  166. /* Non-cacheable page cannot be coherent */
  167. if (flags & _PAGE_NO_CACHE)
  168. flags &= ~_PAGE_COHERENT;
  169. /*
  170. * Choose an address to map it to.
  171. * Once the vmalloc system is running, we use it.
  172. * Before then, we use space going down from ioremap_base
  173. * (ioremap_bot records where we're up to).
  174. */
  175. p = addr & PAGE_MASK;
  176. size = PAGE_ALIGN(addr + size) - p;
  177. /*
  178. * If the address lies within the first 16 MB, assume it's in ISA
  179. * memory space
  180. */
  181. if (p < 16*1024*1024)
  182. p += _ISA_MEM_BASE;
  183. #ifndef CONFIG_CRASH_DUMP
  184. /*
  185. * Don't allow anybody to remap normal RAM that we're using.
  186. * mem_init() sets high_memory so only do the check after that.
  187. */
  188. if (mem_init_done && (p < virt_to_phys(high_memory)) &&
  189. !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
  190. printk("__ioremap(): phys addr 0x%llx is RAM lr %pf\n",
  191. (unsigned long long)p, __builtin_return_address(0));
  192. return NULL;
  193. }
  194. #endif
  195. if (size == 0)
  196. return NULL;
  197. /*
  198. * Is it already mapped? Perhaps overlapped by a previous
  199. * BAT mapping. If the whole area is mapped then we're done,
  200. * otherwise remap it since we want to keep the virt addrs for
  201. * each request contiguous.
  202. *
  203. * We make the assumption here that if the bottom and top
  204. * of the range we want are mapped then it's mapped to the
  205. * same virt address (and this is contiguous).
  206. * -- Cort
  207. */
  208. if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
  209. goto out;
  210. if ((v = p_mapped_by_tlbcam(p)))
  211. goto out;
  212. if (mem_init_done) {
  213. struct vm_struct *area;
  214. area = get_vm_area_caller(size, VM_IOREMAP, caller);
  215. if (area == 0)
  216. return NULL;
  217. area->phys_addr = p;
  218. v = (unsigned long) area->addr;
  219. } else {
  220. v = (ioremap_bot -= size);
  221. }
  222. /*
  223. * Should check if it is a candidate for a BAT mapping
  224. */
  225. err = 0;
  226. for (i = 0; i < size && err == 0; i += PAGE_SIZE)
  227. err = map_page(v+i, p+i, flags);
  228. if (err) {
  229. if (mem_init_done)
  230. vunmap((void *)v);
  231. return NULL;
  232. }
  233. out:
  234. return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
  235. }
  236. EXPORT_SYMBOL(__ioremap);
  237. void iounmap(volatile void __iomem *addr)
  238. {
  239. /*
  240. * If mapped by BATs then there is nothing to do.
  241. * Calling vfree() generates a benign warning.
  242. */
  243. if (v_mapped_by_bats((unsigned long)addr)) return;
  244. if (addr > high_memory && (unsigned long) addr < ioremap_bot)
  245. vunmap((void *) (PAGE_MASK & (unsigned long)addr));
  246. }
  247. EXPORT_SYMBOL(iounmap);
  248. int map_page(unsigned long va, phys_addr_t pa, int flags)
  249. {
  250. pmd_t *pd;
  251. pte_t *pg;
  252. int err = -ENOMEM;
  253. /* Use upper 10 bits of VA to index the first level map */
  254. pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
  255. /* Use middle 10 bits of VA to index the second-level map */
  256. pg = pte_alloc_kernel(pd, va);
  257. if (pg != 0) {
  258. err = 0;
  259. /* The PTE should never be already set nor present in the
  260. * hash table
  261. */
  262. BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
  263. flags);
  264. set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
  265. __pgprot(flags)));
  266. }
  267. smp_wmb();
  268. return err;
  269. }
  270. /*
  271. * Map in a chunk of physical memory starting at start.
  272. */
  273. void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
  274. {
  275. unsigned long v, s, f;
  276. phys_addr_t p;
  277. int ktext;
  278. s = offset;
  279. v = PAGE_OFFSET + s;
  280. p = memstart_addr + s;
  281. for (; s < top; s += PAGE_SIZE) {
  282. ktext = ((char *) v >= _stext && (char *) v < etext);
  283. f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL;
  284. map_page(v, p, f);
  285. #ifdef CONFIG_PPC_STD_MMU_32
  286. if (ktext)
  287. hash_preload(&init_mm, v, 0, 0x300);
  288. #endif
  289. v += PAGE_SIZE;
  290. p += PAGE_SIZE;
  291. }
  292. }
  293. void __init mapin_ram(void)
  294. {
  295. unsigned long s, top;
  296. #ifndef CONFIG_WII
  297. top = total_lowmem;
  298. s = mmu_mapin_ram(top);
  299. __mapin_ram_chunk(s, top);
  300. #else
  301. if (!wii_hole_size) {
  302. s = mmu_mapin_ram(total_lowmem);
  303. __mapin_ram_chunk(s, total_lowmem);
  304. } else {
  305. top = wii_hole_start;
  306. s = mmu_mapin_ram(top);
  307. __mapin_ram_chunk(s, top);
  308. top = memblock_end_of_DRAM();
  309. s = wii_mmu_mapin_mem2(top);
  310. __mapin_ram_chunk(s, top);
  311. }
  312. #endif
  313. }
  314. /* Scan the real Linux page tables and return a PTE pointer for
  315. * a virtual address in a context.
  316. * Returns true (1) if PTE was found, zero otherwise. The pointer to
  317. * the PTE pointer is unmodified if PTE is not found.
  318. */
  319. int
  320. get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
  321. {
  322. pgd_t *pgd;
  323. pud_t *pud;
  324. pmd_t *pmd;
  325. pte_t *pte;
  326. int retval = 0;
  327. pgd = pgd_offset(mm, addr & PAGE_MASK);
  328. if (pgd) {
  329. pud = pud_offset(pgd, addr & PAGE_MASK);
  330. if (pud && pud_present(*pud)) {
  331. pmd = pmd_offset(pud, addr & PAGE_MASK);
  332. if (pmd_present(*pmd)) {
  333. pte = pte_offset_map(pmd, addr & PAGE_MASK);
  334. if (pte) {
  335. retval = 1;
  336. *ptep = pte;
  337. if (pmdp)
  338. *pmdp = pmd;
  339. /* XXX caller needs to do pte_unmap, yuck */
  340. }
  341. }
  342. }
  343. }
  344. return(retval);
  345. }
  346. #ifdef CONFIG_DEBUG_PAGEALLOC
  347. static int __change_page_attr(struct page *page, pgprot_t prot)
  348. {
  349. pte_t *kpte;
  350. pmd_t *kpmd;
  351. unsigned long address;
  352. BUG_ON(PageHighMem(page));
  353. address = (unsigned long)page_address(page);
  354. if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
  355. return 0;
  356. if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
  357. return -EINVAL;
  358. __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
  359. wmb();
  360. flush_tlb_page(NULL, address);
  361. pte_unmap(kpte);
  362. return 0;
  363. }
  364. /*
  365. * Change the page attributes of an page in the linear mapping.
  366. *
  367. * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
  368. */
  369. static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
  370. {
  371. int i, err = 0;
  372. unsigned long flags;
  373. local_irq_save(flags);
  374. for (i = 0; i < numpages; i++, page++) {
  375. err = __change_page_attr(page, prot);
  376. if (err)
  377. break;
  378. }
  379. local_irq_restore(flags);
  380. return err;
  381. }
  382. void __kernel_map_pages(struct page *page, int numpages, int enable)
  383. {
  384. if (PageHighMem(page))
  385. return;
  386. change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
  387. }
  388. #endif /* CONFIG_DEBUG_PAGEALLOC */
  389. static int fixmaps;
  390. void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
  391. {
  392. unsigned long address = __fix_to_virt(idx);
  393. if (idx >= __end_of_fixed_addresses) {
  394. BUG();
  395. return;
  396. }
  397. map_page(address, phys, pgprot_val(flags));
  398. fixmaps++;
  399. }
  400. void __this_fixmap_does_not_exist(void)
  401. {
  402. WARN_ON(1);
  403. }