init.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726
  1. #include <linux/gfp.h>
  2. #include <linux/initrd.h>
  3. #include <linux/ioport.h>
  4. #include <linux/swap.h>
  5. #include <linux/memblock.h>
  6. #include <linux/bootmem.h> /* for max_low_pfn */
  7. #include <asm/cacheflush.h>
  8. #include <asm/e820.h>
  9. #include <asm/init.h>
  10. #include <asm/page.h>
  11. #include <asm/page_types.h>
  12. #include <asm/sections.h>
  13. #include <asm/setup.h>
  14. #include <asm/tlbflush.h>
  15. #include <asm/tlb.h>
  16. #include <asm/proto.h>
  17. #include <asm/dma.h> /* for MAX_DMA_PFN */
  18. #include <asm/microcode.h>
  19. /*
  20. * We need to define the tracepoints somewhere, and tlb.c
  21. * is only compied when SMP=y.
  22. */
  23. #define CREATE_TRACE_POINTS
  24. #include <trace/events/tlb.h>
  25. #include "mm_internal.h"
  26. /*
  27. * Tables translating between page_cache_type_t and pte encoding.
  28. * Minimal supported modes are defined statically, modified if more supported
  29. * cache modes are available.
  30. * Index into __cachemode2pte_tbl is the cachemode.
  31. * Index into __pte2cachemode_tbl are the caching attribute bits of the pte
  32. * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2.
  33. */
  34. uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = {
  35. [_PAGE_CACHE_MODE_WB] = 0,
  36. [_PAGE_CACHE_MODE_WC] = _PAGE_PWT,
  37. [_PAGE_CACHE_MODE_UC_MINUS] = _PAGE_PCD,
  38. [_PAGE_CACHE_MODE_UC] = _PAGE_PCD | _PAGE_PWT,
  39. [_PAGE_CACHE_MODE_WT] = _PAGE_PCD,
  40. [_PAGE_CACHE_MODE_WP] = _PAGE_PCD,
  41. };
  42. EXPORT_SYMBOL_GPL(__cachemode2pte_tbl);
  43. uint8_t __pte2cachemode_tbl[8] = {
  44. [__pte2cm_idx(0)] = _PAGE_CACHE_MODE_WB,
  45. [__pte2cm_idx(_PAGE_PWT)] = _PAGE_CACHE_MODE_WC,
  46. [__pte2cm_idx(_PAGE_PCD)] = _PAGE_CACHE_MODE_UC_MINUS,
  47. [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD)] = _PAGE_CACHE_MODE_UC,
  48. [__pte2cm_idx(_PAGE_PAT)] = _PAGE_CACHE_MODE_WB,
  49. [__pte2cm_idx(_PAGE_PWT | _PAGE_PAT)] = _PAGE_CACHE_MODE_WC,
  50. [__pte2cm_idx(_PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS,
  51. [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC,
  52. };
  53. EXPORT_SYMBOL_GPL(__pte2cachemode_tbl);
  54. static unsigned long __initdata pgt_buf_start;
  55. static unsigned long __initdata pgt_buf_end;
  56. static unsigned long __initdata pgt_buf_top;
  57. static unsigned long min_pfn_mapped;
  58. static bool __initdata can_use_brk_pgt = true;
  59. /*
  60. * Pages returned are already directly mapped.
  61. *
  62. * Changing that is likely to break Xen, see commit:
  63. *
  64. * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
  65. *
  66. * for detailed information.
  67. */
  68. __ref void *alloc_low_pages(unsigned int num)
  69. {
  70. unsigned long pfn;
  71. int i;
  72. if (after_bootmem) {
  73. unsigned int order;
  74. order = get_order((unsigned long)num << PAGE_SHIFT);
  75. return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK |
  76. __GFP_ZERO, order);
  77. }
  78. if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
  79. unsigned long ret;
  80. if (min_pfn_mapped >= max_pfn_mapped)
  81. panic("alloc_low_pages: ran out of memory");
  82. ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
  83. max_pfn_mapped << PAGE_SHIFT,
  84. PAGE_SIZE * num , PAGE_SIZE);
  85. if (!ret)
  86. panic("alloc_low_pages: can not alloc memory");
  87. memblock_reserve(ret, PAGE_SIZE * num);
  88. pfn = ret >> PAGE_SHIFT;
  89. } else {
  90. pfn = pgt_buf_end;
  91. pgt_buf_end += num;
  92. printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n",
  93. pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1);
  94. }
  95. for (i = 0; i < num; i++) {
  96. void *adr;
  97. adr = __va((pfn + i) << PAGE_SHIFT);
  98. clear_page(adr);
  99. }
  100. return __va(pfn << PAGE_SHIFT);
  101. }
  102. /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */
  103. #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE)
  104. RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE);
  105. void __init early_alloc_pgt_buf(void)
  106. {
  107. unsigned long tables = INIT_PGT_BUF_SIZE;
  108. phys_addr_t base;
  109. base = __pa(extend_brk(tables, PAGE_SIZE));
  110. pgt_buf_start = base >> PAGE_SHIFT;
  111. pgt_buf_end = pgt_buf_start;
  112. pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT);
  113. }
  114. int after_bootmem;
  115. int direct_gbpages
  116. #ifdef CONFIG_DIRECT_GBPAGES
  117. = 1
  118. #endif
  119. ;
  120. static void __init init_gbpages(void)
  121. {
  122. #ifdef CONFIG_X86_64
  123. if (direct_gbpages && cpu_has_gbpages)
  124. printk(KERN_INFO "Using GB pages for direct mapping\n");
  125. else
  126. direct_gbpages = 0;
  127. #endif
  128. }
  129. struct map_range {
  130. unsigned long start;
  131. unsigned long end;
  132. unsigned page_size_mask;
  133. };
  134. static int page_size_mask;
  135. static void __init probe_page_size_mask(void)
  136. {
  137. init_gbpages();
  138. #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
  139. /*
  140. * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
  141. * This will simplify cpa(), which otherwise needs to support splitting
  142. * large pages into small in interrupt context, etc.
  143. */
  144. if (direct_gbpages)
  145. page_size_mask |= 1 << PG_LEVEL_1G;
  146. if (cpu_has_pse)
  147. page_size_mask |= 1 << PG_LEVEL_2M;
  148. #endif
  149. /* Enable PSE if available */
  150. if (cpu_has_pse)
  151. set_in_cr4(X86_CR4_PSE);
  152. /* Enable PGE if available */
  153. if (cpu_has_pge) {
  154. set_in_cr4(X86_CR4_PGE);
  155. __supported_pte_mask |= _PAGE_GLOBAL;
  156. }
  157. }
  158. #ifdef CONFIG_X86_32
  159. #define NR_RANGE_MR 3
  160. #else /* CONFIG_X86_64 */
  161. #define NR_RANGE_MR 5
  162. #endif
  163. static int __meminit save_mr(struct map_range *mr, int nr_range,
  164. unsigned long start_pfn, unsigned long end_pfn,
  165. unsigned long page_size_mask)
  166. {
  167. if (start_pfn < end_pfn) {
  168. if (nr_range >= NR_RANGE_MR)
  169. panic("run out of range for init_memory_mapping\n");
  170. mr[nr_range].start = start_pfn<<PAGE_SHIFT;
  171. mr[nr_range].end = end_pfn<<PAGE_SHIFT;
  172. mr[nr_range].page_size_mask = page_size_mask;
  173. nr_range++;
  174. }
  175. return nr_range;
  176. }
  177. /*
  178. * adjust the page_size_mask for small range to go with
  179. * big page size instead small one if nearby are ram too.
  180. */
  181. static void __init_refok adjust_range_page_size_mask(struct map_range *mr,
  182. int nr_range)
  183. {
  184. int i;
  185. for (i = 0; i < nr_range; i++) {
  186. if ((page_size_mask & (1<<PG_LEVEL_2M)) &&
  187. !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) {
  188. unsigned long start = round_down(mr[i].start, PMD_SIZE);
  189. unsigned long end = round_up(mr[i].end, PMD_SIZE);
  190. #ifdef CONFIG_X86_32
  191. if ((end >> PAGE_SHIFT) > max_low_pfn)
  192. continue;
  193. #endif
  194. if (memblock_is_region_memory(start, end - start))
  195. mr[i].page_size_mask |= 1<<PG_LEVEL_2M;
  196. }
  197. if ((page_size_mask & (1<<PG_LEVEL_1G)) &&
  198. !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) {
  199. unsigned long start = round_down(mr[i].start, PUD_SIZE);
  200. unsigned long end = round_up(mr[i].end, PUD_SIZE);
  201. if (memblock_is_region_memory(start, end - start))
  202. mr[i].page_size_mask |= 1<<PG_LEVEL_1G;
  203. }
  204. }
  205. }
  206. static int __meminit split_mem_range(struct map_range *mr, int nr_range,
  207. unsigned long start,
  208. unsigned long end)
  209. {
  210. unsigned long start_pfn, end_pfn, limit_pfn;
  211. unsigned long pfn;
  212. int i;
  213. limit_pfn = PFN_DOWN(end);
  214. /* head if not big page alignment ? */
  215. pfn = start_pfn = PFN_DOWN(start);
  216. #ifdef CONFIG_X86_32
  217. /*
  218. * Don't use a large page for the first 2/4MB of memory
  219. * because there are often fixed size MTRRs in there
  220. * and overlapping MTRRs into large pages can cause
  221. * slowdowns.
  222. */
  223. if (pfn == 0)
  224. end_pfn = PFN_DOWN(PMD_SIZE);
  225. else
  226. end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
  227. #else /* CONFIG_X86_64 */
  228. end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
  229. #endif
  230. if (end_pfn > limit_pfn)
  231. end_pfn = limit_pfn;
  232. if (start_pfn < end_pfn) {
  233. nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
  234. pfn = end_pfn;
  235. }
  236. /* big page (2M) range */
  237. start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
  238. #ifdef CONFIG_X86_32
  239. end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
  240. #else /* CONFIG_X86_64 */
  241. end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
  242. if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE)))
  243. end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
  244. #endif
  245. if (start_pfn < end_pfn) {
  246. nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
  247. page_size_mask & (1<<PG_LEVEL_2M));
  248. pfn = end_pfn;
  249. }
  250. #ifdef CONFIG_X86_64
  251. /* big page (1G) range */
  252. start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE));
  253. end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE));
  254. if (start_pfn < end_pfn) {
  255. nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
  256. page_size_mask &
  257. ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
  258. pfn = end_pfn;
  259. }
  260. /* tail is not big page (1G) alignment */
  261. start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE));
  262. end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE));
  263. if (start_pfn < end_pfn) {
  264. nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
  265. page_size_mask & (1<<PG_LEVEL_2M));
  266. pfn = end_pfn;
  267. }
  268. #endif
  269. /* tail is not big page (2M) alignment */
  270. start_pfn = pfn;
  271. end_pfn = limit_pfn;
  272. nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
  273. if (!after_bootmem)
  274. adjust_range_page_size_mask(mr, nr_range);
  275. /* try to merge same page size and continuous */
  276. for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
  277. unsigned long old_start;
  278. if (mr[i].end != mr[i+1].start ||
  279. mr[i].page_size_mask != mr[i+1].page_size_mask)
  280. continue;
  281. /* move it */
  282. old_start = mr[i].start;
  283. memmove(&mr[i], &mr[i+1],
  284. (nr_range - 1 - i) * sizeof(struct map_range));
  285. mr[i--].start = old_start;
  286. nr_range--;
  287. }
  288. for (i = 0; i < nr_range; i++)
  289. printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n",
  290. mr[i].start, mr[i].end - 1,
  291. (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
  292. (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
  293. return nr_range;
  294. }
  295. struct range pfn_mapped[E820_X_MAX];
  296. int nr_pfn_mapped;
  297. static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn)
  298. {
  299. nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX,
  300. nr_pfn_mapped, start_pfn, end_pfn);
  301. nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX);
  302. max_pfn_mapped = max(max_pfn_mapped, end_pfn);
  303. if (start_pfn < (1UL<<(32-PAGE_SHIFT)))
  304. max_low_pfn_mapped = max(max_low_pfn_mapped,
  305. min(end_pfn, 1UL<<(32-PAGE_SHIFT)));
  306. }
  307. bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn)
  308. {
  309. int i;
  310. for (i = 0; i < nr_pfn_mapped; i++)
  311. if ((start_pfn >= pfn_mapped[i].start) &&
  312. (end_pfn <= pfn_mapped[i].end))
  313. return true;
  314. return false;
  315. }
  316. /*
  317. * Setup the direct mapping of the physical memory at PAGE_OFFSET.
  318. * This runs before bootmem is initialized and gets pages directly from
  319. * the physical memory. To access them they are temporarily mapped.
  320. */
  321. unsigned long __init_refok init_memory_mapping(unsigned long start,
  322. unsigned long end)
  323. {
  324. struct map_range mr[NR_RANGE_MR];
  325. unsigned long ret = 0;
  326. int nr_range, i;
  327. pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n",
  328. start, end - 1);
  329. memset(mr, 0, sizeof(mr));
  330. nr_range = split_mem_range(mr, 0, start, end);
  331. for (i = 0; i < nr_range; i++)
  332. ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
  333. mr[i].page_size_mask);
  334. add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT);
  335. return ret >> PAGE_SHIFT;
  336. }
  337. /*
  338. * We need to iterate through the E820 memory map and create direct mappings
  339. * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
  340. * create direct mappings for all pfns from [0 to max_low_pfn) and
  341. * [4GB to max_pfn) because of possible memory holes in high addresses
  342. * that cannot be marked as UC by fixed/variable range MTRRs.
  343. * Depending on the alignment of E820 ranges, this may possibly result
  344. * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
  345. *
  346. * init_mem_mapping() calls init_range_memory_mapping() with big range.
  347. * That range would have hole in the middle or ends, and only ram parts
  348. * will be mapped in init_range_memory_mapping().
  349. */
  350. static unsigned long __init init_range_memory_mapping(
  351. unsigned long r_start,
  352. unsigned long r_end)
  353. {
  354. unsigned long start_pfn, end_pfn;
  355. unsigned long mapped_ram_size = 0;
  356. int i;
  357. for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
  358. u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end);
  359. u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end);
  360. if (start >= end)
  361. continue;
  362. /*
  363. * if it is overlapping with brk pgt, we need to
  364. * alloc pgt buf from memblock instead.
  365. */
  366. can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >=
  367. min(end, (u64)pgt_buf_top<<PAGE_SHIFT);
  368. init_memory_mapping(start, end);
  369. mapped_ram_size += end - start;
  370. can_use_brk_pgt = true;
  371. }
  372. return mapped_ram_size;
  373. }
  374. static unsigned long __init get_new_step_size(unsigned long step_size)
  375. {
  376. /*
  377. * Explain why we shift by 5 and why we don't have to worry about
  378. * 'step_size << 5' overflowing:
  379. *
  380. * initial mapped size is PMD_SIZE (2M).
  381. * We can not set step_size to be PUD_SIZE (1G) yet.
  382. * In worse case, when we cross the 1G boundary, and
  383. * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
  384. * to map 1G range with PTE. Use 5 as shift for now.
  385. *
  386. * Don't need to worry about overflow, on 32bit, when step_size
  387. * is 0, round_down() returns 0 for start, and that turns it
  388. * into 0x100000000ULL.
  389. */
  390. return step_size << 5;
  391. }
  392. /**
  393. * memory_map_top_down - Map [map_start, map_end) top down
  394. * @map_start: start address of the target memory range
  395. * @map_end: end address of the target memory range
  396. *
  397. * This function will setup direct mapping for memory range
  398. * [map_start, map_end) in top-down. That said, the page tables
  399. * will be allocated at the end of the memory, and we map the
  400. * memory in top-down.
  401. */
  402. static void __init memory_map_top_down(unsigned long map_start,
  403. unsigned long map_end)
  404. {
  405. unsigned long real_end, start, last_start;
  406. unsigned long step_size;
  407. unsigned long addr;
  408. unsigned long mapped_ram_size = 0;
  409. unsigned long new_mapped_ram_size;
  410. /* xen has big range in reserved near end of ram, skip it at first.*/
  411. addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE);
  412. real_end = addr + PMD_SIZE;
  413. /* step_size need to be small so pgt_buf from BRK could cover it */
  414. step_size = PMD_SIZE;
  415. max_pfn_mapped = 0; /* will get exact value next */
  416. min_pfn_mapped = real_end >> PAGE_SHIFT;
  417. last_start = start = real_end;
  418. /*
  419. * We start from the top (end of memory) and go to the bottom.
  420. * The memblock_find_in_range() gets us a block of RAM from the
  421. * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
  422. * for page table.
  423. */
  424. while (last_start > map_start) {
  425. if (last_start > step_size) {
  426. start = round_down(last_start - 1, step_size);
  427. if (start < map_start)
  428. start = map_start;
  429. } else
  430. start = map_start;
  431. new_mapped_ram_size = init_range_memory_mapping(start,
  432. last_start);
  433. last_start = start;
  434. min_pfn_mapped = last_start >> PAGE_SHIFT;
  435. /* only increase step_size after big range get mapped */
  436. if (new_mapped_ram_size > mapped_ram_size)
  437. step_size = get_new_step_size(step_size);
  438. mapped_ram_size += new_mapped_ram_size;
  439. }
  440. if (real_end < map_end)
  441. init_range_memory_mapping(real_end, map_end);
  442. }
  443. /**
  444. * memory_map_bottom_up - Map [map_start, map_end) bottom up
  445. * @map_start: start address of the target memory range
  446. * @map_end: end address of the target memory range
  447. *
  448. * This function will setup direct mapping for memory range
  449. * [map_start, map_end) in bottom-up. Since we have limited the
  450. * bottom-up allocation above the kernel, the page tables will
  451. * be allocated just above the kernel and we map the memory
  452. * in [map_start, map_end) in bottom-up.
  453. */
  454. static void __init memory_map_bottom_up(unsigned long map_start,
  455. unsigned long map_end)
  456. {
  457. unsigned long next, new_mapped_ram_size, start;
  458. unsigned long mapped_ram_size = 0;
  459. /* step_size need to be small so pgt_buf from BRK could cover it */
  460. unsigned long step_size = PMD_SIZE;
  461. start = map_start;
  462. min_pfn_mapped = start >> PAGE_SHIFT;
  463. /*
  464. * We start from the bottom (@map_start) and go to the top (@map_end).
  465. * The memblock_find_in_range() gets us a block of RAM from the
  466. * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
  467. * for page table.
  468. */
  469. while (start < map_end) {
  470. if (map_end - start > step_size) {
  471. next = round_up(start + 1, step_size);
  472. if (next > map_end)
  473. next = map_end;
  474. } else
  475. next = map_end;
  476. new_mapped_ram_size = init_range_memory_mapping(start, next);
  477. start = next;
  478. if (new_mapped_ram_size > mapped_ram_size)
  479. step_size = get_new_step_size(step_size);
  480. mapped_ram_size += new_mapped_ram_size;
  481. }
  482. }
  483. void __init init_mem_mapping(void)
  484. {
  485. unsigned long end;
  486. probe_page_size_mask();
  487. #ifdef CONFIG_X86_64
  488. end = max_pfn << PAGE_SHIFT;
  489. #else
  490. end = max_low_pfn << PAGE_SHIFT;
  491. #endif
  492. /* the ISA range is always mapped regardless of memory holes */
  493. init_memory_mapping(0, ISA_END_ADDRESS);
  494. /*
  495. * If the allocation is in bottom-up direction, we setup direct mapping
  496. * in bottom-up, otherwise we setup direct mapping in top-down.
  497. */
  498. if (memblock_bottom_up()) {
  499. unsigned long kernel_end = __pa_symbol(_end);
  500. /*
  501. * we need two separate calls here. This is because we want to
  502. * allocate page tables above the kernel. So we first map
  503. * [kernel_end, end) to make memory above the kernel be mapped
  504. * as soon as possible. And then use page tables allocated above
  505. * the kernel to map [ISA_END_ADDRESS, kernel_end).
  506. */
  507. memory_map_bottom_up(kernel_end, end);
  508. memory_map_bottom_up(ISA_END_ADDRESS, kernel_end);
  509. } else {
  510. memory_map_top_down(ISA_END_ADDRESS, end);
  511. }
  512. #ifdef CONFIG_X86_64
  513. if (max_pfn > max_low_pfn) {
  514. /* can we preseve max_low_pfn ?*/
  515. max_low_pfn = max_pfn;
  516. }
  517. #else
  518. early_ioremap_page_table_range_init();
  519. #endif
  520. load_cr3(swapper_pg_dir);
  521. __flush_tlb_all();
  522. early_memtest(0, max_pfn_mapped << PAGE_SHIFT);
  523. }
  524. /*
  525. * devmem_is_allowed() checks to see if /dev/mem access to a certain address
  526. * is valid. The argument is a physical page number.
  527. *
  528. *
  529. * On x86, access has to be given to the first megabyte of ram because that area
  530. * contains bios code and data regions used by X and dosemu and similar apps.
  531. * Access has to be given to non-kernel-ram areas as well, these contain the PCI
  532. * mmio resources as well as potential bios/acpi data regions.
  533. */
  534. int devmem_is_allowed(unsigned long pagenr)
  535. {
  536. if (pagenr < 256)
  537. return 1;
  538. if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
  539. return 0;
  540. if (!page_is_ram(pagenr))
  541. return 1;
  542. return 0;
  543. }
  544. void free_init_pages(char *what, unsigned long begin, unsigned long end)
  545. {
  546. unsigned long begin_aligned, end_aligned;
  547. /* Make sure boundaries are page aligned */
  548. begin_aligned = PAGE_ALIGN(begin);
  549. end_aligned = end & PAGE_MASK;
  550. if (WARN_ON(begin_aligned != begin || end_aligned != end)) {
  551. begin = begin_aligned;
  552. end = end_aligned;
  553. }
  554. if (begin >= end)
  555. return;
  556. /*
  557. * If debugging page accesses then do not free this memory but
  558. * mark them not present - any buggy init-section access will
  559. * create a kernel page fault:
  560. */
  561. #ifdef CONFIG_DEBUG_PAGEALLOC
  562. printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
  563. begin, end - 1);
  564. set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
  565. #else
  566. /*
  567. * We just marked the kernel text read only above, now that
  568. * we are going to free part of that, we need to make that
  569. * writeable and non-executable first.
  570. */
  571. set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
  572. set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
  573. free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what);
  574. #endif
  575. }
  576. void free_initmem(void)
  577. {
  578. free_init_pages("unused kernel",
  579. (unsigned long)(&__init_begin),
  580. (unsigned long)(&__init_end));
  581. }
  582. #ifdef CONFIG_BLK_DEV_INITRD
  583. void __init free_initrd_mem(unsigned long start, unsigned long end)
  584. {
  585. #ifdef CONFIG_MICROCODE_EARLY
  586. /*
  587. * Remember, initrd memory may contain microcode or other useful things.
  588. * Before we lose initrd mem, we need to find a place to hold them
  589. * now that normal virtual memory is enabled.
  590. */
  591. save_microcode_in_initrd();
  592. #endif
  593. /*
  594. * end could be not aligned, and We can not align that,
  595. * decompresser could be confused by aligned initrd_end
  596. * We already reserve the end partial page before in
  597. * - i386_start_kernel()
  598. * - x86_64_start_kernel()
  599. * - relocate_initrd()
  600. * So here We can do PAGE_ALIGN() safely to get partial page to be freed
  601. */
  602. free_init_pages("initrd", start, PAGE_ALIGN(end));
  603. }
  604. #endif
  605. void __init zone_sizes_init(void)
  606. {
  607. unsigned long max_zone_pfns[MAX_NR_ZONES];
  608. memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
  609. #ifdef CONFIG_ZONE_DMA
  610. max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn);
  611. #endif
  612. #ifdef CONFIG_ZONE_DMA32
  613. max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn);
  614. #endif
  615. max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
  616. #ifdef CONFIG_HIGHMEM
  617. max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
  618. #endif
  619. free_area_init_nodes(max_zone_pfns);
  620. }
  621. void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache)
  622. {
  623. /* entry 0 MUST be WB (hardwired to speed up translations) */
  624. BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB);
  625. __cachemode2pte_tbl[cache] = __cm_idx2pte(entry);
  626. __pte2cachemode_tbl[entry] = cache;
  627. }