init.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746
  1. /*
  2. * Initialize MMU support.
  3. *
  4. * Copyright (C) 1998-2003 Hewlett-Packard Co
  5. * David Mosberger-Tang <davidm@hpl.hp.com>
  6. */
  7. #include <linux/kernel.h>
  8. #include <linux/init.h>
  9. #include <linux/bootmem.h>
  10. #include <linux/efi.h>
  11. #include <linux/elf.h>
  12. #include <linux/mm.h>
  13. #include <linux/mmzone.h>
  14. #include <linux/module.h>
  15. #include <linux/personality.h>
  16. #include <linux/reboot.h>
  17. #include <linux/slab.h>
  18. #include <linux/swap.h>
  19. #include <linux/proc_fs.h>
  20. #include <linux/bitops.h>
  21. #include <linux/kexec.h>
  22. #include <asm/a.out.h>
  23. #include <asm/dma.h>
  24. #include <asm/ia32.h>
  25. #include <asm/io.h>
  26. #include <asm/machvec.h>
  27. #include <asm/numa.h>
  28. #include <asm/patch.h>
  29. #include <asm/pgalloc.h>
  30. #include <asm/sal.h>
  31. #include <asm/sections.h>
  32. #include <asm/system.h>
  33. #include <asm/tlb.h>
  34. #include <asm/uaccess.h>
  35. #include <asm/unistd.h>
  36. #include <asm/mca.h>
  37. DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  38. extern void ia64_tlb_init (void);
  39. unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
  40. #ifdef CONFIG_VIRTUAL_MEM_MAP
  41. unsigned long vmalloc_end = VMALLOC_END_INIT;
  42. EXPORT_SYMBOL(vmalloc_end);
  43. struct page *vmem_map;
  44. EXPORT_SYMBOL(vmem_map);
  45. #endif
  46. struct page *zero_page_memmap_ptr; /* map entry for zero page */
  47. EXPORT_SYMBOL(zero_page_memmap_ptr);
  48. void
  49. __ia64_sync_icache_dcache (pte_t pte)
  50. {
  51. unsigned long addr;
  52. struct page *page;
  53. page = pte_page(pte);
  54. addr = (unsigned long) page_address(page);
  55. if (test_bit(PG_arch_1, &page->flags))
  56. return; /* i-cache is already coherent with d-cache */
  57. flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
  58. set_bit(PG_arch_1, &page->flags); /* mark page as clean */
  59. }
  60. /*
  61. * Since DMA is i-cache coherent, any (complete) pages that were written via
  62. * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
  63. * flush them when they get mapped into an executable vm-area.
  64. */
  65. void
  66. dma_mark_clean(void *addr, size_t size)
  67. {
  68. unsigned long pg_addr, end;
  69. pg_addr = PAGE_ALIGN((unsigned long) addr);
  70. end = (unsigned long) addr + size;
  71. while (pg_addr + PAGE_SIZE <= end) {
  72. struct page *page = virt_to_page(pg_addr);
  73. set_bit(PG_arch_1, &page->flags);
  74. pg_addr += PAGE_SIZE;
  75. }
  76. }
  77. inline void
  78. ia64_set_rbs_bot (void)
  79. {
  80. unsigned long stack_size = current->signal->rlim[RLIMIT_STACK].rlim_max & -16;
  81. if (stack_size > MAX_USER_STACK_SIZE)
  82. stack_size = MAX_USER_STACK_SIZE;
  83. current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
  84. }
  85. /*
  86. * This performs some platform-dependent address space initialization.
  87. * On IA-64, we want to setup the VM area for the register backing
  88. * store (which grows upwards) and install the gateway page which is
  89. * used for signal trampolines, etc.
  90. */
  91. void
  92. ia64_init_addr_space (void)
  93. {
  94. struct vm_area_struct *vma;
  95. ia64_set_rbs_bot();
  96. /*
  97. * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
  98. * the problem. When the process attempts to write to the register backing store
  99. * for the first time, it will get a SEGFAULT in this case.
  100. */
  101. vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
  102. if (vma) {
  103. vma->vm_mm = current->mm;
  104. vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
  105. vma->vm_end = vma->vm_start + PAGE_SIZE;
  106. vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
  107. vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
  108. down_write(&current->mm->mmap_sem);
  109. if (insert_vm_struct(current->mm, vma)) {
  110. up_write(&current->mm->mmap_sem);
  111. kmem_cache_free(vm_area_cachep, vma);
  112. return;
  113. }
  114. up_write(&current->mm->mmap_sem);
  115. }
  116. /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
  117. if (!(current->personality & MMAP_PAGE_ZERO)) {
  118. vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
  119. if (vma) {
  120. vma->vm_mm = current->mm;
  121. vma->vm_end = PAGE_SIZE;
  122. vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
  123. vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
  124. down_write(&current->mm->mmap_sem);
  125. if (insert_vm_struct(current->mm, vma)) {
  126. up_write(&current->mm->mmap_sem);
  127. kmem_cache_free(vm_area_cachep, vma);
  128. return;
  129. }
  130. up_write(&current->mm->mmap_sem);
  131. }
  132. }
  133. }
  134. void
  135. free_initmem (void)
  136. {
  137. unsigned long addr, eaddr;
  138. addr = (unsigned long) ia64_imva(__init_begin);
  139. eaddr = (unsigned long) ia64_imva(__init_end);
  140. while (addr < eaddr) {
  141. ClearPageReserved(virt_to_page(addr));
  142. init_page_count(virt_to_page(addr));
  143. free_page(addr);
  144. ++totalram_pages;
  145. addr += PAGE_SIZE;
  146. }
  147. printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
  148. (__init_end - __init_begin) >> 10);
  149. }
  150. void __init
  151. free_initrd_mem (unsigned long start, unsigned long end)
  152. {
  153. struct page *page;
  154. /*
  155. * EFI uses 4KB pages while the kernel can use 4KB or bigger.
  156. * Thus EFI and the kernel may have different page sizes. It is
  157. * therefore possible to have the initrd share the same page as
  158. * the end of the kernel (given current setup).
  159. *
  160. * To avoid freeing/using the wrong page (kernel sized) we:
  161. * - align up the beginning of initrd
  162. * - align down the end of initrd
  163. *
  164. * | |
  165. * |=============| a000
  166. * | |
  167. * | |
  168. * | | 9000
  169. * |/////////////|
  170. * |/////////////|
  171. * |=============| 8000
  172. * |///INITRD////|
  173. * |/////////////|
  174. * |/////////////| 7000
  175. * | |
  176. * |KKKKKKKKKKKKK|
  177. * |=============| 6000
  178. * |KKKKKKKKKKKKK|
  179. * |KKKKKKKKKKKKK|
  180. * K=kernel using 8KB pages
  181. *
  182. * In this example, we must free page 8000 ONLY. So we must align up
  183. * initrd_start and keep initrd_end as is.
  184. */
  185. start = PAGE_ALIGN(start);
  186. end = end & PAGE_MASK;
  187. if (start < end)
  188. printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
  189. for (; start < end; start += PAGE_SIZE) {
  190. if (!virt_addr_valid(start))
  191. continue;
  192. page = virt_to_page(start);
  193. ClearPageReserved(page);
  194. init_page_count(page);
  195. free_page(start);
  196. ++totalram_pages;
  197. }
  198. }
  199. /*
  200. * This installs a clean page in the kernel's page table.
  201. */
  202. static struct page * __init
  203. put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
  204. {
  205. pgd_t *pgd;
  206. pud_t *pud;
  207. pmd_t *pmd;
  208. pte_t *pte;
  209. if (!PageReserved(page))
  210. printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
  211. page_address(page));
  212. pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
  213. {
  214. pud = pud_alloc(&init_mm, pgd, address);
  215. if (!pud)
  216. goto out;
  217. pmd = pmd_alloc(&init_mm, pud, address);
  218. if (!pmd)
  219. goto out;
  220. pte = pte_alloc_kernel(pmd, address);
  221. if (!pte)
  222. goto out;
  223. if (!pte_none(*pte))
  224. goto out;
  225. set_pte(pte, mk_pte(page, pgprot));
  226. }
  227. out:
  228. /* no need for flush_tlb */
  229. return page;
  230. }
  231. static void __init
  232. setup_gate (void)
  233. {
  234. struct page *page;
  235. /*
  236. * Map the gate page twice: once read-only to export the ELF
  237. * headers etc. and once execute-only page to enable
  238. * privilege-promotion via "epc":
  239. */
  240. page = virt_to_page(ia64_imva(__start_gate_section));
  241. put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
  242. #ifdef HAVE_BUGGY_SEGREL
  243. page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
  244. put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
  245. #else
  246. put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
  247. /* Fill in the holes (if any) with read-only zero pages: */
  248. {
  249. unsigned long addr;
  250. for (addr = GATE_ADDR + PAGE_SIZE;
  251. addr < GATE_ADDR + PERCPU_PAGE_SIZE;
  252. addr += PAGE_SIZE)
  253. {
  254. put_kernel_page(ZERO_PAGE(0), addr,
  255. PAGE_READONLY);
  256. put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
  257. PAGE_READONLY);
  258. }
  259. }
  260. #endif
  261. ia64_patch_gate();
  262. }
  263. void __devinit
  264. ia64_mmu_init (void *my_cpu_data)
  265. {
  266. unsigned long pta, impl_va_bits;
  267. extern void __devinit tlb_init (void);
  268. #ifdef CONFIG_DISABLE_VHPT
  269. # define VHPT_ENABLE_BIT 0
  270. #else
  271. # define VHPT_ENABLE_BIT 1
  272. #endif
  273. /*
  274. * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
  275. * address space. The IA-64 architecture guarantees that at least 50 bits of
  276. * virtual address space are implemented but if we pick a large enough page size
  277. * (e.g., 64KB), the mapped address space is big enough that it will overlap with
  278. * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
  279. * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
  280. * problem in practice. Alternatively, we could truncate the top of the mapped
  281. * address space to not permit mappings that would overlap with the VMLPT.
  282. * --davidm 00/12/06
  283. */
  284. # define pte_bits 3
  285. # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
  286. /*
  287. * The virtual page table has to cover the entire implemented address space within
  288. * a region even though not all of this space may be mappable. The reason for
  289. * this is that the Access bit and Dirty bit fault handlers perform
  290. * non-speculative accesses to the virtual page table, so the address range of the
  291. * virtual page table itself needs to be covered by virtual page table.
  292. */
  293. # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
  294. # define POW2(n) (1ULL << (n))
  295. impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
  296. if (impl_va_bits < 51 || impl_va_bits > 61)
  297. panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
  298. /*
  299. * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
  300. * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
  301. * the test makes sure that our mapped space doesn't overlap the
  302. * unimplemented hole in the middle of the region.
  303. */
  304. if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
  305. (mapped_space_bits > impl_va_bits - 1))
  306. panic("Cannot build a big enough virtual-linear page table"
  307. " to cover mapped address space.\n"
  308. " Try using a smaller page size.\n");
  309. /* place the VMLPT at the end of each page-table mapped region: */
  310. pta = POW2(61) - POW2(vmlpt_bits);
  311. /*
  312. * Set the (virtually mapped linear) page table address. Bit
  313. * 8 selects between the short and long format, bits 2-7 the
  314. * size of the table, and bit 0 whether the VHPT walker is
  315. * enabled.
  316. */
  317. ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
  318. ia64_tlb_init();
  319. #ifdef CONFIG_HUGETLB_PAGE
  320. ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
  321. ia64_srlz_d();
  322. #endif
  323. }
  324. #ifdef CONFIG_VIRTUAL_MEM_MAP
  325. int vmemmap_find_next_valid_pfn(int node, int i)
  326. {
  327. unsigned long end_address, hole_next_pfn;
  328. unsigned long stop_address;
  329. pg_data_t *pgdat = NODE_DATA(node);
  330. end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
  331. end_address = PAGE_ALIGN(end_address);
  332. stop_address = (unsigned long) &vmem_map[
  333. pgdat->node_start_pfn + pgdat->node_spanned_pages];
  334. do {
  335. pgd_t *pgd;
  336. pud_t *pud;
  337. pmd_t *pmd;
  338. pte_t *pte;
  339. pgd = pgd_offset_k(end_address);
  340. if (pgd_none(*pgd)) {
  341. end_address += PGDIR_SIZE;
  342. continue;
  343. }
  344. pud = pud_offset(pgd, end_address);
  345. if (pud_none(*pud)) {
  346. end_address += PUD_SIZE;
  347. continue;
  348. }
  349. pmd = pmd_offset(pud, end_address);
  350. if (pmd_none(*pmd)) {
  351. end_address += PMD_SIZE;
  352. continue;
  353. }
  354. pte = pte_offset_kernel(pmd, end_address);
  355. retry_pte:
  356. if (pte_none(*pte)) {
  357. end_address += PAGE_SIZE;
  358. pte++;
  359. if ((end_address < stop_address) &&
  360. (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
  361. goto retry_pte;
  362. continue;
  363. }
  364. /* Found next valid vmem_map page */
  365. break;
  366. } while (end_address < stop_address);
  367. end_address = min(end_address, stop_address);
  368. end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
  369. hole_next_pfn = end_address / sizeof(struct page);
  370. return hole_next_pfn - pgdat->node_start_pfn;
  371. }
  372. int __init
  373. create_mem_map_page_table (u64 start, u64 end, void *arg)
  374. {
  375. unsigned long address, start_page, end_page;
  376. struct page *map_start, *map_end;
  377. int node;
  378. pgd_t *pgd;
  379. pud_t *pud;
  380. pmd_t *pmd;
  381. pte_t *pte;
  382. map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
  383. map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
  384. start_page = (unsigned long) map_start & PAGE_MASK;
  385. end_page = PAGE_ALIGN((unsigned long) map_end);
  386. node = paddr_to_nid(__pa(start));
  387. for (address = start_page; address < end_page; address += PAGE_SIZE) {
  388. pgd = pgd_offset_k(address);
  389. if (pgd_none(*pgd))
  390. pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  391. pud = pud_offset(pgd, address);
  392. if (pud_none(*pud))
  393. pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  394. pmd = pmd_offset(pud, address);
  395. if (pmd_none(*pmd))
  396. pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  397. pte = pte_offset_kernel(pmd, address);
  398. if (pte_none(*pte))
  399. set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
  400. PAGE_KERNEL));
  401. }
  402. return 0;
  403. }
  404. struct memmap_init_callback_data {
  405. struct page *start;
  406. struct page *end;
  407. int nid;
  408. unsigned long zone;
  409. };
  410. static int __meminit
  411. virtual_memmap_init (u64 start, u64 end, void *arg)
  412. {
  413. struct memmap_init_callback_data *args;
  414. struct page *map_start, *map_end;
  415. args = (struct memmap_init_callback_data *) arg;
  416. map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
  417. map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
  418. if (map_start < args->start)
  419. map_start = args->start;
  420. if (map_end > args->end)
  421. map_end = args->end;
  422. /*
  423. * We have to initialize "out of bounds" struct page elements that fit completely
  424. * on the same pages that were allocated for the "in bounds" elements because they
  425. * may be referenced later (and found to be "reserved").
  426. */
  427. map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
  428. map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
  429. / sizeof(struct page));
  430. if (map_start < map_end)
  431. memmap_init_zone((unsigned long)(map_end - map_start),
  432. args->nid, args->zone, page_to_pfn(map_start),
  433. MEMMAP_EARLY);
  434. return 0;
  435. }
  436. void __meminit
  437. memmap_init (unsigned long size, int nid, unsigned long zone,
  438. unsigned long start_pfn)
  439. {
  440. if (!vmem_map)
  441. memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
  442. else {
  443. struct page *start;
  444. struct memmap_init_callback_data args;
  445. start = pfn_to_page(start_pfn);
  446. args.start = start;
  447. args.end = start + size;
  448. args.nid = nid;
  449. args.zone = zone;
  450. efi_memmap_walk(virtual_memmap_init, &args);
  451. }
  452. }
  453. int
  454. ia64_pfn_valid (unsigned long pfn)
  455. {
  456. char byte;
  457. struct page *pg = pfn_to_page(pfn);
  458. return (__get_user(byte, (char __user *) pg) == 0)
  459. && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
  460. || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
  461. }
  462. EXPORT_SYMBOL(ia64_pfn_valid);
  463. int __init
  464. find_largest_hole (u64 start, u64 end, void *arg)
  465. {
  466. u64 *max_gap = arg;
  467. static u64 last_end = PAGE_OFFSET;
  468. /* NOTE: this algorithm assumes efi memmap table is ordered */
  469. if (*max_gap < (start - last_end))
  470. *max_gap = start - last_end;
  471. last_end = end;
  472. return 0;
  473. }
  474. #endif /* CONFIG_VIRTUAL_MEM_MAP */
  475. int __init
  476. register_active_ranges(u64 start, u64 len, int nid)
  477. {
  478. u64 end = start + len;
  479. #ifdef CONFIG_KEXEC
  480. if (start > crashk_res.start && start < crashk_res.end)
  481. start = crashk_res.end;
  482. if (end > crashk_res.start && end < crashk_res.end)
  483. end = crashk_res.start;
  484. #endif
  485. if (start < end)
  486. add_active_range(nid, __pa(start) >> PAGE_SHIFT,
  487. __pa(end) >> PAGE_SHIFT);
  488. return 0;
  489. }
  490. static int __init
  491. count_reserved_pages (u64 start, u64 end, void *arg)
  492. {
  493. unsigned long num_reserved = 0;
  494. unsigned long *count = arg;
  495. for (; start < end; start += PAGE_SIZE)
  496. if (PageReserved(virt_to_page(start)))
  497. ++num_reserved;
  498. *count += num_reserved;
  499. return 0;
  500. }
  501. int
  502. find_max_min_low_pfn (unsigned long start, unsigned long end, void *arg)
  503. {
  504. unsigned long pfn_start, pfn_end;
  505. #ifdef CONFIG_FLATMEM
  506. pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
  507. pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
  508. #else
  509. pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
  510. pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
  511. #endif
  512. min_low_pfn = min(min_low_pfn, pfn_start);
  513. max_low_pfn = max(max_low_pfn, pfn_end);
  514. return 0;
  515. }
  516. /*
  517. * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
  518. * system call handler. When this option is in effect, all fsyscalls will end up bubbling
  519. * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
  520. * useful for performance testing, but conceivably could also come in handy for debugging
  521. * purposes.
  522. */
  523. static int nolwsys __initdata;
  524. static int __init
  525. nolwsys_setup (char *s)
  526. {
  527. nolwsys = 1;
  528. return 1;
  529. }
  530. __setup("nolwsys", nolwsys_setup);
  531. void __init
  532. mem_init (void)
  533. {
  534. long reserved_pages, codesize, datasize, initsize;
  535. pg_data_t *pgdat;
  536. int i;
  537. static struct kcore_list kcore_mem, kcore_vmem, kcore_kernel;
  538. BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
  539. BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
  540. BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
  541. #ifdef CONFIG_PCI
  542. /*
  543. * This needs to be called _after_ the command line has been parsed but _before_
  544. * any drivers that may need the PCI DMA interface are initialized or bootmem has
  545. * been freed.
  546. */
  547. platform_dma_init();
  548. #endif
  549. #ifdef CONFIG_FLATMEM
  550. if (!mem_map)
  551. BUG();
  552. max_mapnr = max_low_pfn;
  553. #endif
  554. high_memory = __va(max_low_pfn * PAGE_SIZE);
  555. kclist_add(&kcore_mem, __va(0), max_low_pfn * PAGE_SIZE);
  556. kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
  557. kclist_add(&kcore_kernel, _stext, _end - _stext);
  558. for_each_online_pgdat(pgdat)
  559. if (pgdat->bdata->node_bootmem_map)
  560. totalram_pages += free_all_bootmem_node(pgdat);
  561. reserved_pages = 0;
  562. efi_memmap_walk(count_reserved_pages, &reserved_pages);
  563. codesize = (unsigned long) _etext - (unsigned long) _stext;
  564. datasize = (unsigned long) _edata - (unsigned long) _etext;
  565. initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
  566. printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
  567. "%luk data, %luk init)\n", (unsigned long) nr_free_pages() << (PAGE_SHIFT - 10),
  568. num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
  569. reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
  570. /*
  571. * For fsyscall entrpoints with no light-weight handler, use the ordinary
  572. * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
  573. * code can tell them apart.
  574. */
  575. for (i = 0; i < NR_syscalls; ++i) {
  576. extern unsigned long fsyscall_table[NR_syscalls];
  577. extern unsigned long sys_call_table[NR_syscalls];
  578. if (!fsyscall_table[i] || nolwsys)
  579. fsyscall_table[i] = sys_call_table[i] | 1;
  580. }
  581. setup_gate();
  582. #ifdef CONFIG_IA32_SUPPORT
  583. ia32_mem_init();
  584. #endif
  585. }
  586. #ifdef CONFIG_MEMORY_HOTPLUG
  587. int arch_add_memory(int nid, u64 start, u64 size)
  588. {
  589. pg_data_t *pgdat;
  590. struct zone *zone;
  591. unsigned long start_pfn = start >> PAGE_SHIFT;
  592. unsigned long nr_pages = size >> PAGE_SHIFT;
  593. int ret;
  594. pgdat = NODE_DATA(nid);
  595. zone = pgdat->node_zones + ZONE_NORMAL;
  596. ret = __add_pages(zone, start_pfn, nr_pages);
  597. if (ret)
  598. printk("%s: Problem encountered in __add_pages() as ret=%d\n",
  599. __func__, ret);
  600. return ret;
  601. }
  602. #ifdef CONFIG_MEMORY_HOTREMOVE
  603. int remove_memory(u64 start, u64 size)
  604. {
  605. unsigned long start_pfn, end_pfn;
  606. unsigned long timeout = 120 * HZ;
  607. int ret;
  608. start_pfn = start >> PAGE_SHIFT;
  609. end_pfn = start_pfn + (size >> PAGE_SHIFT);
  610. ret = offline_pages(start_pfn, end_pfn, timeout);
  611. if (ret)
  612. goto out;
  613. /* we can free mem_map at this point */
  614. out:
  615. return ret;
  616. }
  617. EXPORT_SYMBOL_GPL(remove_memory);
  618. #endif /* CONFIG_MEMORY_HOTREMOVE */
  619. #endif
  620. /*
  621. * Even when CONFIG_IA32_SUPPORT is not enabled it is
  622. * useful to have the Linux/x86 domain registered to
  623. * avoid an attempted module load when emulators call
  624. * personality(PER_LINUX32). This saves several milliseconds
  625. * on each such call.
  626. */
  627. static struct exec_domain ia32_exec_domain;
  628. static int __init
  629. per_linux32_init(void)
  630. {
  631. ia32_exec_domain.name = "Linux/x86";
  632. ia32_exec_domain.handler = NULL;
  633. ia32_exec_domain.pers_low = PER_LINUX32;
  634. ia32_exec_domain.pers_high = PER_LINUX32;
  635. ia32_exec_domain.signal_map = default_exec_domain.signal_map;
  636. ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
  637. register_exec_domain(&ia32_exec_domain);
  638. return 0;
  639. }
  640. __initcall(per_linux32_init);