init.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774
  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. DEFINE_PER_CPU(unsigned long *, __pgtable_quicklist);
  39. DEFINE_PER_CPU(long, __pgtable_quicklist_size);
  40. extern void ia64_tlb_init (void);
  41. unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
  42. #ifdef CONFIG_VIRTUAL_MEM_MAP
  43. unsigned long vmalloc_end = VMALLOC_END_INIT;
  44. EXPORT_SYMBOL(vmalloc_end);
  45. struct page *vmem_map;
  46. EXPORT_SYMBOL(vmem_map);
  47. #endif
  48. struct page *zero_page_memmap_ptr; /* map entry for zero page */
  49. EXPORT_SYMBOL(zero_page_memmap_ptr);
  50. #define MIN_PGT_PAGES 25UL
  51. #define MAX_PGT_FREES_PER_PASS 16L
  52. #define PGT_FRACTION_OF_NODE_MEM 16
  53. static inline long
  54. max_pgt_pages(void)
  55. {
  56. u64 node_free_pages, max_pgt_pages;
  57. #ifndef CONFIG_NUMA
  58. node_free_pages = nr_free_pages();
  59. #else
  60. node_free_pages = node_page_state(numa_node_id(), NR_FREE_PAGES);
  61. #endif
  62. max_pgt_pages = node_free_pages / PGT_FRACTION_OF_NODE_MEM;
  63. max_pgt_pages = max(max_pgt_pages, MIN_PGT_PAGES);
  64. return max_pgt_pages;
  65. }
  66. static inline long
  67. min_pages_to_free(void)
  68. {
  69. long pages_to_free;
  70. pages_to_free = pgtable_quicklist_size - max_pgt_pages();
  71. pages_to_free = min(pages_to_free, MAX_PGT_FREES_PER_PASS);
  72. return pages_to_free;
  73. }
  74. void
  75. check_pgt_cache(void)
  76. {
  77. long pages_to_free;
  78. if (unlikely(pgtable_quicklist_size <= MIN_PGT_PAGES))
  79. return;
  80. preempt_disable();
  81. while (unlikely((pages_to_free = min_pages_to_free()) > 0)) {
  82. while (pages_to_free--) {
  83. free_page((unsigned long)pgtable_quicklist_alloc());
  84. }
  85. preempt_enable();
  86. preempt_disable();
  87. }
  88. preempt_enable();
  89. }
  90. void
  91. lazy_mmu_prot_update (pte_t pte)
  92. {
  93. unsigned long addr;
  94. struct page *page;
  95. unsigned long order;
  96. if (!pte_exec(pte))
  97. return; /* not an executable page... */
  98. page = pte_page(pte);
  99. addr = (unsigned long) page_address(page);
  100. if (test_bit(PG_arch_1, &page->flags))
  101. return; /* i-cache is already coherent with d-cache */
  102. if (PageCompound(page)) {
  103. order = (unsigned long) (page[1].lru.prev);
  104. flush_icache_range(addr, addr + (1UL << order << PAGE_SHIFT));
  105. }
  106. else
  107. flush_icache_range(addr, addr + PAGE_SIZE);
  108. set_bit(PG_arch_1, &page->flags); /* mark page as clean */
  109. }
  110. /*
  111. * Since DMA is i-cache coherent, any (complete) pages that were written via
  112. * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
  113. * flush them when they get mapped into an executable vm-area.
  114. */
  115. void
  116. dma_mark_clean(void *addr, size_t size)
  117. {
  118. unsigned long pg_addr, end;
  119. pg_addr = PAGE_ALIGN((unsigned long) addr);
  120. end = (unsigned long) addr + size;
  121. while (pg_addr + PAGE_SIZE <= end) {
  122. struct page *page = virt_to_page(pg_addr);
  123. set_bit(PG_arch_1, &page->flags);
  124. pg_addr += PAGE_SIZE;
  125. }
  126. }
  127. inline void
  128. ia64_set_rbs_bot (void)
  129. {
  130. unsigned long stack_size = current->signal->rlim[RLIMIT_STACK].rlim_max & -16;
  131. if (stack_size > MAX_USER_STACK_SIZE)
  132. stack_size = MAX_USER_STACK_SIZE;
  133. current->thread.rbs_bot = STACK_TOP - stack_size;
  134. }
  135. /*
  136. * This performs some platform-dependent address space initialization.
  137. * On IA-64, we want to setup the VM area for the register backing
  138. * store (which grows upwards) and install the gateway page which is
  139. * used for signal trampolines, etc.
  140. */
  141. void
  142. ia64_init_addr_space (void)
  143. {
  144. struct vm_area_struct *vma;
  145. ia64_set_rbs_bot();
  146. /*
  147. * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
  148. * the problem. When the process attempts to write to the register backing store
  149. * for the first time, it will get a SEGFAULT in this case.
  150. */
  151. vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
  152. if (vma) {
  153. vma->vm_mm = current->mm;
  154. vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
  155. vma->vm_end = vma->vm_start + PAGE_SIZE;
  156. vma->vm_page_prot = protection_map[VM_DATA_DEFAULT_FLAGS & 0x7];
  157. vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
  158. down_write(&current->mm->mmap_sem);
  159. if (insert_vm_struct(current->mm, vma)) {
  160. up_write(&current->mm->mmap_sem);
  161. kmem_cache_free(vm_area_cachep, vma);
  162. return;
  163. }
  164. up_write(&current->mm->mmap_sem);
  165. }
  166. /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
  167. if (!(current->personality & MMAP_PAGE_ZERO)) {
  168. vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
  169. if (vma) {
  170. vma->vm_mm = current->mm;
  171. vma->vm_end = PAGE_SIZE;
  172. vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
  173. vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
  174. down_write(&current->mm->mmap_sem);
  175. if (insert_vm_struct(current->mm, vma)) {
  176. up_write(&current->mm->mmap_sem);
  177. kmem_cache_free(vm_area_cachep, vma);
  178. return;
  179. }
  180. up_write(&current->mm->mmap_sem);
  181. }
  182. }
  183. }
  184. void
  185. free_initmem (void)
  186. {
  187. unsigned long addr, eaddr;
  188. addr = (unsigned long) ia64_imva(__init_begin);
  189. eaddr = (unsigned long) ia64_imva(__init_end);
  190. while (addr < eaddr) {
  191. ClearPageReserved(virt_to_page(addr));
  192. init_page_count(virt_to_page(addr));
  193. free_page(addr);
  194. ++totalram_pages;
  195. addr += PAGE_SIZE;
  196. }
  197. printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
  198. (__init_end - __init_begin) >> 10);
  199. }
  200. void __init
  201. free_initrd_mem (unsigned long start, unsigned long end)
  202. {
  203. struct page *page;
  204. /*
  205. * EFI uses 4KB pages while the kernel can use 4KB or bigger.
  206. * Thus EFI and the kernel may have different page sizes. It is
  207. * therefore possible to have the initrd share the same page as
  208. * the end of the kernel (given current setup).
  209. *
  210. * To avoid freeing/using the wrong page (kernel sized) we:
  211. * - align up the beginning of initrd
  212. * - align down the end of initrd
  213. *
  214. * | |
  215. * |=============| a000
  216. * | |
  217. * | |
  218. * | | 9000
  219. * |/////////////|
  220. * |/////////////|
  221. * |=============| 8000
  222. * |///INITRD////|
  223. * |/////////////|
  224. * |/////////////| 7000
  225. * | |
  226. * |KKKKKKKKKKKKK|
  227. * |=============| 6000
  228. * |KKKKKKKKKKKKK|
  229. * |KKKKKKKKKKKKK|
  230. * K=kernel using 8KB pages
  231. *
  232. * In this example, we must free page 8000 ONLY. So we must align up
  233. * initrd_start and keep initrd_end as is.
  234. */
  235. start = PAGE_ALIGN(start);
  236. end = end & PAGE_MASK;
  237. if (start < end)
  238. printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
  239. for (; start < end; start += PAGE_SIZE) {
  240. if (!virt_addr_valid(start))
  241. continue;
  242. page = virt_to_page(start);
  243. ClearPageReserved(page);
  244. init_page_count(page);
  245. free_page(start);
  246. ++totalram_pages;
  247. }
  248. }
  249. /*
  250. * This installs a clean page in the kernel's page table.
  251. */
  252. static struct page * __init
  253. put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
  254. {
  255. pgd_t *pgd;
  256. pud_t *pud;
  257. pmd_t *pmd;
  258. pte_t *pte;
  259. if (!PageReserved(page))
  260. printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
  261. page_address(page));
  262. pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
  263. {
  264. pud = pud_alloc(&init_mm, pgd, address);
  265. if (!pud)
  266. goto out;
  267. pmd = pmd_alloc(&init_mm, pud, address);
  268. if (!pmd)
  269. goto out;
  270. pte = pte_alloc_kernel(pmd, address);
  271. if (!pte)
  272. goto out;
  273. if (!pte_none(*pte))
  274. goto out;
  275. set_pte(pte, mk_pte(page, pgprot));
  276. }
  277. out:
  278. /* no need for flush_tlb */
  279. return page;
  280. }
  281. static void __init
  282. setup_gate (void)
  283. {
  284. struct page *page;
  285. /*
  286. * Map the gate page twice: once read-only to export the ELF
  287. * headers etc. and once execute-only page to enable
  288. * privilege-promotion via "epc":
  289. */
  290. page = virt_to_page(ia64_imva(__start_gate_section));
  291. put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
  292. #ifdef HAVE_BUGGY_SEGREL
  293. page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
  294. put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
  295. #else
  296. put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
  297. /* Fill in the holes (if any) with read-only zero pages: */
  298. {
  299. unsigned long addr;
  300. for (addr = GATE_ADDR + PAGE_SIZE;
  301. addr < GATE_ADDR + PERCPU_PAGE_SIZE;
  302. addr += PAGE_SIZE)
  303. {
  304. put_kernel_page(ZERO_PAGE(0), addr,
  305. PAGE_READONLY);
  306. put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
  307. PAGE_READONLY);
  308. }
  309. }
  310. #endif
  311. ia64_patch_gate();
  312. }
  313. void __devinit
  314. ia64_mmu_init (void *my_cpu_data)
  315. {
  316. unsigned long psr, pta, impl_va_bits;
  317. extern void __devinit tlb_init (void);
  318. #ifdef CONFIG_DISABLE_VHPT
  319. # define VHPT_ENABLE_BIT 0
  320. #else
  321. # define VHPT_ENABLE_BIT 1
  322. #endif
  323. /* Pin mapping for percpu area into TLB */
  324. psr = ia64_clear_ic();
  325. ia64_itr(0x2, IA64_TR_PERCPU_DATA, PERCPU_ADDR,
  326. pte_val(pfn_pte(__pa(my_cpu_data) >> PAGE_SHIFT, PAGE_KERNEL)),
  327. PERCPU_PAGE_SHIFT);
  328. ia64_set_psr(psr);
  329. ia64_srlz_i();
  330. /*
  331. * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
  332. * address space. The IA-64 architecture guarantees that at least 50 bits of
  333. * virtual address space are implemented but if we pick a large enough page size
  334. * (e.g., 64KB), the mapped address space is big enough that it will overlap with
  335. * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
  336. * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
  337. * problem in practice. Alternatively, we could truncate the top of the mapped
  338. * address space to not permit mappings that would overlap with the VMLPT.
  339. * --davidm 00/12/06
  340. */
  341. # define pte_bits 3
  342. # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
  343. /*
  344. * The virtual page table has to cover the entire implemented address space within
  345. * a region even though not all of this space may be mappable. The reason for
  346. * this is that the Access bit and Dirty bit fault handlers perform
  347. * non-speculative accesses to the virtual page table, so the address range of the
  348. * virtual page table itself needs to be covered by virtual page table.
  349. */
  350. # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
  351. # define POW2(n) (1ULL << (n))
  352. impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
  353. if (impl_va_bits < 51 || impl_va_bits > 61)
  354. panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
  355. /*
  356. * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
  357. * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
  358. * the test makes sure that our mapped space doesn't overlap the
  359. * unimplemented hole in the middle of the region.
  360. */
  361. if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
  362. (mapped_space_bits > impl_va_bits - 1))
  363. panic("Cannot build a big enough virtual-linear page table"
  364. " to cover mapped address space.\n"
  365. " Try using a smaller page size.\n");
  366. /* place the VMLPT at the end of each page-table mapped region: */
  367. pta = POW2(61) - POW2(vmlpt_bits);
  368. /*
  369. * Set the (virtually mapped linear) page table address. Bit
  370. * 8 selects between the short and long format, bits 2-7 the
  371. * size of the table, and bit 0 whether the VHPT walker is
  372. * enabled.
  373. */
  374. ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
  375. ia64_tlb_init();
  376. #ifdef CONFIG_HUGETLB_PAGE
  377. ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
  378. ia64_srlz_d();
  379. #endif
  380. }
  381. #ifdef CONFIG_VIRTUAL_MEM_MAP
  382. int vmemmap_find_next_valid_pfn(int node, int i)
  383. {
  384. unsigned long end_address, hole_next_pfn;
  385. unsigned long stop_address;
  386. pg_data_t *pgdat = NODE_DATA(node);
  387. end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
  388. end_address = PAGE_ALIGN(end_address);
  389. stop_address = (unsigned long) &vmem_map[
  390. pgdat->node_start_pfn + pgdat->node_spanned_pages];
  391. do {
  392. pgd_t *pgd;
  393. pud_t *pud;
  394. pmd_t *pmd;
  395. pte_t *pte;
  396. pgd = pgd_offset_k(end_address);
  397. if (pgd_none(*pgd)) {
  398. end_address += PGDIR_SIZE;
  399. continue;
  400. }
  401. pud = pud_offset(pgd, end_address);
  402. if (pud_none(*pud)) {
  403. end_address += PUD_SIZE;
  404. continue;
  405. }
  406. pmd = pmd_offset(pud, end_address);
  407. if (pmd_none(*pmd)) {
  408. end_address += PMD_SIZE;
  409. continue;
  410. }
  411. pte = pte_offset_kernel(pmd, end_address);
  412. retry_pte:
  413. if (pte_none(*pte)) {
  414. end_address += PAGE_SIZE;
  415. pte++;
  416. if ((end_address < stop_address) &&
  417. (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
  418. goto retry_pte;
  419. continue;
  420. }
  421. /* Found next valid vmem_map page */
  422. break;
  423. } while (end_address < stop_address);
  424. end_address = min(end_address, stop_address);
  425. end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
  426. hole_next_pfn = end_address / sizeof(struct page);
  427. return hole_next_pfn - pgdat->node_start_pfn;
  428. }
  429. int __init
  430. create_mem_map_page_table (u64 start, u64 end, void *arg)
  431. {
  432. unsigned long address, start_page, end_page;
  433. struct page *map_start, *map_end;
  434. int node;
  435. pgd_t *pgd;
  436. pud_t *pud;
  437. pmd_t *pmd;
  438. pte_t *pte;
  439. map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
  440. map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
  441. start_page = (unsigned long) map_start & PAGE_MASK;
  442. end_page = PAGE_ALIGN((unsigned long) map_end);
  443. node = paddr_to_nid(__pa(start));
  444. for (address = start_page; address < end_page; address += PAGE_SIZE) {
  445. pgd = pgd_offset_k(address);
  446. if (pgd_none(*pgd))
  447. pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  448. pud = pud_offset(pgd, address);
  449. if (pud_none(*pud))
  450. pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  451. pmd = pmd_offset(pud, address);
  452. if (pmd_none(*pmd))
  453. pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
  454. pte = pte_offset_kernel(pmd, address);
  455. if (pte_none(*pte))
  456. set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
  457. PAGE_KERNEL));
  458. }
  459. return 0;
  460. }
  461. struct memmap_init_callback_data {
  462. struct page *start;
  463. struct page *end;
  464. int nid;
  465. unsigned long zone;
  466. };
  467. static int
  468. virtual_memmap_init (u64 start, u64 end, void *arg)
  469. {
  470. struct memmap_init_callback_data *args;
  471. struct page *map_start, *map_end;
  472. args = (struct memmap_init_callback_data *) arg;
  473. map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
  474. map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
  475. if (map_start < args->start)
  476. map_start = args->start;
  477. if (map_end > args->end)
  478. map_end = args->end;
  479. /*
  480. * We have to initialize "out of bounds" struct page elements that fit completely
  481. * on the same pages that were allocated for the "in bounds" elements because they
  482. * may be referenced later (and found to be "reserved").
  483. */
  484. map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
  485. map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
  486. / sizeof(struct page));
  487. if (map_start < map_end)
  488. memmap_init_zone((unsigned long)(map_end - map_start),
  489. args->nid, args->zone, page_to_pfn(map_start),
  490. MEMMAP_EARLY);
  491. return 0;
  492. }
  493. void
  494. memmap_init (unsigned long size, int nid, unsigned long zone,
  495. unsigned long start_pfn)
  496. {
  497. if (!vmem_map)
  498. memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
  499. else {
  500. struct page *start;
  501. struct memmap_init_callback_data args;
  502. start = pfn_to_page(start_pfn);
  503. args.start = start;
  504. args.end = start + size;
  505. args.nid = nid;
  506. args.zone = zone;
  507. efi_memmap_walk(virtual_memmap_init, &args);
  508. }
  509. }
  510. int
  511. ia64_pfn_valid (unsigned long pfn)
  512. {
  513. char byte;
  514. struct page *pg = pfn_to_page(pfn);
  515. return (__get_user(byte, (char __user *) pg) == 0)
  516. && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
  517. || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
  518. }
  519. EXPORT_SYMBOL(ia64_pfn_valid);
  520. int __init
  521. find_largest_hole (u64 start, u64 end, void *arg)
  522. {
  523. u64 *max_gap = arg;
  524. static u64 last_end = PAGE_OFFSET;
  525. /* NOTE: this algorithm assumes efi memmap table is ordered */
  526. if (*max_gap < (start - last_end))
  527. *max_gap = start - last_end;
  528. last_end = end;
  529. return 0;
  530. }
  531. #endif /* CONFIG_VIRTUAL_MEM_MAP */
  532. int __init
  533. register_active_ranges(u64 start, u64 end, void *arg)
  534. {
  535. int nid = paddr_to_nid(__pa(start));
  536. if (nid < 0)
  537. nid = 0;
  538. #ifdef CONFIG_KEXEC
  539. if (start > crashk_res.start && start < crashk_res.end)
  540. start = crashk_res.end;
  541. if (end > crashk_res.start && end < crashk_res.end)
  542. end = crashk_res.start;
  543. #endif
  544. if (start < end)
  545. add_active_range(nid, __pa(start) >> PAGE_SHIFT,
  546. __pa(end) >> PAGE_SHIFT);
  547. return 0;
  548. }
  549. static int __init
  550. count_reserved_pages (u64 start, u64 end, void *arg)
  551. {
  552. unsigned long num_reserved = 0;
  553. unsigned long *count = arg;
  554. for (; start < end; start += PAGE_SIZE)
  555. if (PageReserved(virt_to_page(start)))
  556. ++num_reserved;
  557. *count += num_reserved;
  558. return 0;
  559. }
  560. /*
  561. * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
  562. * system call handler. When this option is in effect, all fsyscalls will end up bubbling
  563. * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
  564. * useful for performance testing, but conceivably could also come in handy for debugging
  565. * purposes.
  566. */
  567. static int nolwsys __initdata;
  568. static int __init
  569. nolwsys_setup (char *s)
  570. {
  571. nolwsys = 1;
  572. return 1;
  573. }
  574. __setup("nolwsys", nolwsys_setup);
  575. void __init
  576. mem_init (void)
  577. {
  578. long reserved_pages, codesize, datasize, initsize;
  579. pg_data_t *pgdat;
  580. int i;
  581. static struct kcore_list kcore_mem, kcore_vmem, kcore_kernel;
  582. BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
  583. BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
  584. BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
  585. #ifdef CONFIG_PCI
  586. /*
  587. * This needs to be called _after_ the command line has been parsed but _before_
  588. * any drivers that may need the PCI DMA interface are initialized or bootmem has
  589. * been freed.
  590. */
  591. platform_dma_init();
  592. #endif
  593. #ifdef CONFIG_FLATMEM
  594. if (!mem_map)
  595. BUG();
  596. max_mapnr = max_low_pfn;
  597. #endif
  598. high_memory = __va(max_low_pfn * PAGE_SIZE);
  599. kclist_add(&kcore_mem, __va(0), max_low_pfn * PAGE_SIZE);
  600. kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
  601. kclist_add(&kcore_kernel, _stext, _end - _stext);
  602. for_each_online_pgdat(pgdat)
  603. if (pgdat->bdata->node_bootmem_map)
  604. totalram_pages += free_all_bootmem_node(pgdat);
  605. reserved_pages = 0;
  606. efi_memmap_walk(count_reserved_pages, &reserved_pages);
  607. codesize = (unsigned long) _etext - (unsigned long) _stext;
  608. datasize = (unsigned long) _edata - (unsigned long) _etext;
  609. initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
  610. printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
  611. "%luk data, %luk init)\n", (unsigned long) nr_free_pages() << (PAGE_SHIFT - 10),
  612. num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
  613. reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
  614. /*
  615. * For fsyscall entrpoints with no light-weight handler, use the ordinary
  616. * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
  617. * code can tell them apart.
  618. */
  619. for (i = 0; i < NR_syscalls; ++i) {
  620. extern unsigned long fsyscall_table[NR_syscalls];
  621. extern unsigned long sys_call_table[NR_syscalls];
  622. if (!fsyscall_table[i] || nolwsys)
  623. fsyscall_table[i] = sys_call_table[i] | 1;
  624. }
  625. setup_gate();
  626. #ifdef CONFIG_IA32_SUPPORT
  627. ia32_mem_init();
  628. #endif
  629. }
  630. #ifdef CONFIG_MEMORY_HOTPLUG
  631. void online_page(struct page *page)
  632. {
  633. ClearPageReserved(page);
  634. init_page_count(page);
  635. __free_page(page);
  636. totalram_pages++;
  637. num_physpages++;
  638. }
  639. int arch_add_memory(int nid, u64 start, u64 size)
  640. {
  641. pg_data_t *pgdat;
  642. struct zone *zone;
  643. unsigned long start_pfn = start >> PAGE_SHIFT;
  644. unsigned long nr_pages = size >> PAGE_SHIFT;
  645. int ret;
  646. pgdat = NODE_DATA(nid);
  647. zone = pgdat->node_zones + ZONE_NORMAL;
  648. ret = __add_pages(zone, start_pfn, nr_pages);
  649. if (ret)
  650. printk("%s: Problem encountered in __add_pages() as ret=%d\n",
  651. __FUNCTION__, ret);
  652. return ret;
  653. }
  654. int remove_memory(u64 start, u64 size)
  655. {
  656. return -EINVAL;
  657. }
  658. EXPORT_SYMBOL_GPL(remove_memory);
  659. #endif