mem.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522
  1. /*
  2. * PowerPC version
  3. * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  4. *
  5. * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
  6. * and Cort Dougan (PReP) (cort@cs.nmt.edu)
  7. * Copyright (C) 1996 Paul Mackerras
  8. * PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
  9. *
  10. * Derived from "arch/i386/mm/init.c"
  11. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  12. *
  13. * This program is free software; you can redistribute it and/or
  14. * modify it under the terms of the GNU General Public License
  15. * as published by the Free Software Foundation; either version
  16. * 2 of the License, or (at your option) any later version.
  17. *
  18. */
  19. #include <linux/module.h>
  20. #include <linux/sched.h>
  21. #include <linux/kernel.h>
  22. #include <linux/errno.h>
  23. #include <linux/string.h>
  24. #include <linux/gfp.h>
  25. #include <linux/types.h>
  26. #include <linux/mm.h>
  27. #include <linux/stddef.h>
  28. #include <linux/init.h>
  29. #include <linux/bootmem.h>
  30. #include <linux/highmem.h>
  31. #include <linux/initrd.h>
  32. #include <linux/pagemap.h>
  33. #include <linux/suspend.h>
  34. #include <linux/memblock.h>
  35. #include <linux/hugetlb.h>
  36. #include <asm/pgalloc.h>
  37. #include <asm/prom.h>
  38. #include <asm/io.h>
  39. #include <asm/mmu_context.h>
  40. #include <asm/pgtable.h>
  41. #include <asm/mmu.h>
  42. #include <asm/smp.h>
  43. #include <asm/machdep.h>
  44. #include <asm/btext.h>
  45. #include <asm/tlb.h>
  46. #include <asm/sections.h>
  47. #include <asm/sparsemem.h>
  48. #include <asm/vdso.h>
  49. #include <asm/fixmap.h>
  50. #include <asm/swiotlb.h>
  51. #include "mmu_decl.h"
  52. #ifndef CPU_FTR_COHERENT_ICACHE
  53. #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
  54. #define CPU_FTR_NOEXECUTE 0
  55. #endif
  56. int init_bootmem_done;
  57. int mem_init_done;
  58. phys_addr_t memory_limit;
  59. #ifdef CONFIG_HIGHMEM
  60. pte_t *kmap_pte;
  61. pgprot_t kmap_prot;
  62. EXPORT_SYMBOL(kmap_prot);
  63. EXPORT_SYMBOL(kmap_pte);
  64. static inline pte_t *virt_to_kpte(unsigned long vaddr)
  65. {
  66. return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  67. vaddr), vaddr), vaddr);
  68. }
  69. #endif
  70. int page_is_ram(unsigned long pfn)
  71. {
  72. #ifndef CONFIG_PPC64 /* XXX for now */
  73. return pfn < max_pfn;
  74. #else
  75. unsigned long paddr = (pfn << PAGE_SHIFT);
  76. struct memblock_region *reg;
  77. for_each_memblock(memory, reg)
  78. if (paddr >= reg->base && paddr < (reg->base + reg->size))
  79. return 1;
  80. return 0;
  81. #endif
  82. }
  83. pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  84. unsigned long size, pgprot_t vma_prot)
  85. {
  86. if (ppc_md.phys_mem_access_prot)
  87. return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  88. if (!page_is_ram(pfn))
  89. vma_prot = pgprot_noncached(vma_prot);
  90. return vma_prot;
  91. }
  92. EXPORT_SYMBOL(phys_mem_access_prot);
  93. #ifdef CONFIG_MEMORY_HOTPLUG
  94. #ifdef CONFIG_NUMA
  95. int memory_add_physaddr_to_nid(u64 start)
  96. {
  97. return hot_add_scn_to_nid(start);
  98. }
  99. #endif
  100. int arch_add_memory(int nid, u64 start, u64 size)
  101. {
  102. struct pglist_data *pgdata;
  103. struct zone *zone;
  104. unsigned long start_pfn = start >> PAGE_SHIFT;
  105. unsigned long nr_pages = size >> PAGE_SHIFT;
  106. pgdata = NODE_DATA(nid);
  107. start = (unsigned long)__va(start);
  108. create_section_mapping(start, start + size);
  109. /* this should work for most non-highmem platforms */
  110. zone = pgdata->node_zones;
  111. return __add_pages(nid, zone, start_pfn, nr_pages);
  112. }
  113. #endif /* CONFIG_MEMORY_HOTPLUG */
  114. /*
  115. * walk_memory_resource() needs to make sure there is no holes in a given
  116. * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
  117. * Instead it maintains it in memblock.memory structures. Walk through the
  118. * memory regions, find holes and callback for contiguous regions.
  119. */
  120. int
  121. walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
  122. void *arg, int (*func)(unsigned long, unsigned long, void *))
  123. {
  124. struct memblock_region *reg;
  125. unsigned long end_pfn = start_pfn + nr_pages;
  126. unsigned long tstart, tend;
  127. int ret = -1;
  128. for_each_memblock(memory, reg) {
  129. tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
  130. tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
  131. if (tstart >= tend)
  132. continue;
  133. ret = (*func)(tstart, tend - tstart, arg);
  134. if (ret)
  135. break;
  136. }
  137. return ret;
  138. }
  139. EXPORT_SYMBOL_GPL(walk_system_ram_range);
  140. /*
  141. * Initialize the bootmem system and give it all the memory we
  142. * have available. If we are using highmem, we only put the
  143. * lowmem into the bootmem system.
  144. */
  145. #ifndef CONFIG_NEED_MULTIPLE_NODES
  146. void __init do_init_bootmem(void)
  147. {
  148. unsigned long start, bootmap_pages;
  149. unsigned long total_pages;
  150. struct memblock_region *reg;
  151. int boot_mapsize;
  152. max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
  153. total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
  154. #ifdef CONFIG_HIGHMEM
  155. total_pages = total_lowmem >> PAGE_SHIFT;
  156. max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
  157. #endif
  158. /*
  159. * Find an area to use for the bootmem bitmap. Calculate the size of
  160. * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
  161. * Add 1 additional page in case the address isn't page-aligned.
  162. */
  163. bootmap_pages = bootmem_bootmap_pages(total_pages);
  164. start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
  165. min_low_pfn = MEMORY_START >> PAGE_SHIFT;
  166. boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
  167. /* Add active regions with valid PFNs */
  168. for_each_memblock(memory, reg) {
  169. unsigned long start_pfn, end_pfn;
  170. start_pfn = memblock_region_memory_base_pfn(reg);
  171. end_pfn = memblock_region_memory_end_pfn(reg);
  172. add_active_range(0, start_pfn, end_pfn);
  173. }
  174. /* Add all physical memory to the bootmem map, mark each area
  175. * present.
  176. */
  177. #ifdef CONFIG_HIGHMEM
  178. free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
  179. /* reserve the sections we're already using */
  180. for_each_memblock(reserved, reg) {
  181. unsigned long top = reg->base + reg->size - 1;
  182. if (top < lowmem_end_addr)
  183. reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
  184. else if (reg->base < lowmem_end_addr) {
  185. unsigned long trunc_size = lowmem_end_addr - reg->base;
  186. reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
  187. }
  188. }
  189. #else
  190. free_bootmem_with_active_regions(0, max_pfn);
  191. /* reserve the sections we're already using */
  192. for_each_memblock(reserved, reg)
  193. reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
  194. #endif
  195. /* XXX need to clip this if using highmem? */
  196. sparse_memory_present_with_active_regions(0);
  197. init_bootmem_done = 1;
  198. }
  199. /* mark pages that don't exist as nosave */
  200. static int __init mark_nonram_nosave(void)
  201. {
  202. struct memblock_region *reg, *prev = NULL;
  203. for_each_memblock(memory, reg) {
  204. if (prev &&
  205. memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
  206. register_nosave_region(memblock_region_memory_end_pfn(prev),
  207. memblock_region_memory_base_pfn(reg));
  208. prev = reg;
  209. }
  210. return 0;
  211. }
  212. /*
  213. * paging_init() sets up the page tables - in fact we've already done this.
  214. */
  215. void __init paging_init(void)
  216. {
  217. unsigned long total_ram = memblock_phys_mem_size();
  218. phys_addr_t top_of_ram = memblock_end_of_DRAM();
  219. unsigned long max_zone_pfns[MAX_NR_ZONES];
  220. #ifdef CONFIG_PPC32
  221. unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
  222. unsigned long end = __fix_to_virt(FIX_HOLE);
  223. for (; v < end; v += PAGE_SIZE)
  224. map_page(v, 0, 0); /* XXX gross */
  225. #endif
  226. #ifdef CONFIG_HIGHMEM
  227. map_page(PKMAP_BASE, 0, 0); /* XXX gross */
  228. pkmap_page_table = virt_to_kpte(PKMAP_BASE);
  229. kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
  230. kmap_prot = PAGE_KERNEL;
  231. #endif /* CONFIG_HIGHMEM */
  232. printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
  233. (unsigned long long)top_of_ram, total_ram);
  234. printk(KERN_DEBUG "Memory hole size: %ldMB\n",
  235. (long int)((top_of_ram - total_ram) >> 20));
  236. memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
  237. #ifdef CONFIG_HIGHMEM
  238. max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
  239. max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
  240. #else
  241. max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
  242. #endif
  243. free_area_init_nodes(max_zone_pfns);
  244. mark_nonram_nosave();
  245. }
  246. #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
  247. void __init mem_init(void)
  248. {
  249. #ifdef CONFIG_NEED_MULTIPLE_NODES
  250. int nid;
  251. #endif
  252. pg_data_t *pgdat;
  253. unsigned long i;
  254. struct page *page;
  255. unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
  256. #ifdef CONFIG_SWIOTLB
  257. if (ppc_swiotlb_enable)
  258. swiotlb_init(1);
  259. #endif
  260. num_physpages = memblock_phys_mem_size() >> PAGE_SHIFT;
  261. high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
  262. #ifdef CONFIG_NEED_MULTIPLE_NODES
  263. for_each_online_node(nid) {
  264. if (NODE_DATA(nid)->node_spanned_pages != 0) {
  265. printk("freeing bootmem node %d\n", nid);
  266. totalram_pages +=
  267. free_all_bootmem_node(NODE_DATA(nid));
  268. }
  269. }
  270. #else
  271. max_mapnr = max_pfn;
  272. totalram_pages += free_all_bootmem();
  273. #endif
  274. for_each_online_pgdat(pgdat) {
  275. for (i = 0; i < pgdat->node_spanned_pages; i++) {
  276. if (!pfn_valid(pgdat->node_start_pfn + i))
  277. continue;
  278. page = pgdat_page_nr(pgdat, i);
  279. if (PageReserved(page))
  280. reservedpages++;
  281. }
  282. }
  283. codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
  284. datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
  285. initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
  286. bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
  287. #ifdef CONFIG_HIGHMEM
  288. {
  289. unsigned long pfn, highmem_mapnr;
  290. highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
  291. for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
  292. struct page *page = pfn_to_page(pfn);
  293. if (memblock_is_reserved(pfn << PAGE_SHIFT))
  294. continue;
  295. ClearPageReserved(page);
  296. init_page_count(page);
  297. __free_page(page);
  298. totalhigh_pages++;
  299. reservedpages--;
  300. }
  301. totalram_pages += totalhigh_pages;
  302. printk(KERN_DEBUG "High memory: %luk\n",
  303. totalhigh_pages << (PAGE_SHIFT-10));
  304. }
  305. #endif /* CONFIG_HIGHMEM */
  306. printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
  307. "%luk reserved, %luk data, %luk bss, %luk init)\n",
  308. nr_free_pages() << (PAGE_SHIFT-10),
  309. num_physpages << (PAGE_SHIFT-10),
  310. codesize >> 10,
  311. reservedpages << (PAGE_SHIFT-10),
  312. datasize >> 10,
  313. bsssize >> 10,
  314. initsize >> 10);
  315. #ifdef CONFIG_PPC32
  316. pr_info("Kernel virtual memory layout:\n");
  317. pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
  318. #ifdef CONFIG_HIGHMEM
  319. pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
  320. PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
  321. #endif /* CONFIG_HIGHMEM */
  322. #ifdef CONFIG_NOT_COHERENT_CACHE
  323. pr_info(" * 0x%08lx..0x%08lx : consistent mem\n",
  324. IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
  325. #endif /* CONFIG_NOT_COHERENT_CACHE */
  326. pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
  327. ioremap_bot, IOREMAP_TOP);
  328. pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
  329. VMALLOC_START, VMALLOC_END);
  330. #endif /* CONFIG_PPC32 */
  331. mem_init_done = 1;
  332. }
  333. #ifdef CONFIG_BLK_DEV_INITRD
  334. void __init free_initrd_mem(unsigned long start, unsigned long end)
  335. {
  336. if (start >= end)
  337. return;
  338. start = _ALIGN_DOWN(start, PAGE_SIZE);
  339. end = _ALIGN_UP(end, PAGE_SIZE);
  340. pr_info("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
  341. for (; start < end; start += PAGE_SIZE) {
  342. ClearPageReserved(virt_to_page(start));
  343. init_page_count(virt_to_page(start));
  344. free_page(start);
  345. totalram_pages++;
  346. }
  347. }
  348. #endif
  349. /*
  350. * This is called when a page has been modified by the kernel.
  351. * It just marks the page as not i-cache clean. We do the i-cache
  352. * flush later when the page is given to a user process, if necessary.
  353. */
  354. void flush_dcache_page(struct page *page)
  355. {
  356. if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
  357. return;
  358. /* avoid an atomic op if possible */
  359. if (test_bit(PG_arch_1, &page->flags))
  360. clear_bit(PG_arch_1, &page->flags);
  361. }
  362. EXPORT_SYMBOL(flush_dcache_page);
  363. void flush_dcache_icache_page(struct page *page)
  364. {
  365. #ifdef CONFIG_HUGETLB_PAGE
  366. if (PageCompound(page)) {
  367. flush_dcache_icache_hugepage(page);
  368. return;
  369. }
  370. #endif
  371. #ifdef CONFIG_BOOKE
  372. {
  373. void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
  374. __flush_dcache_icache(start);
  375. kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
  376. }
  377. #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
  378. /* On 8xx there is no need to kmap since highmem is not supported */
  379. __flush_dcache_icache(page_address(page));
  380. #else
  381. __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
  382. #endif
  383. }
  384. void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
  385. {
  386. clear_page(page);
  387. /*
  388. * We shouldn't have to do this, but some versions of glibc
  389. * require it (ld.so assumes zero filled pages are icache clean)
  390. * - Anton
  391. */
  392. flush_dcache_page(pg);
  393. }
  394. EXPORT_SYMBOL(clear_user_page);
  395. void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
  396. struct page *pg)
  397. {
  398. copy_page(vto, vfrom);
  399. /*
  400. * We should be able to use the following optimisation, however
  401. * there are two problems.
  402. * Firstly a bug in some versions of binutils meant PLT sections
  403. * were not marked executable.
  404. * Secondly the first word in the GOT section is blrl, used
  405. * to establish the GOT address. Until recently the GOT was
  406. * not marked executable.
  407. * - Anton
  408. */
  409. #if 0
  410. if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
  411. return;
  412. #endif
  413. flush_dcache_page(pg);
  414. }
  415. void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
  416. unsigned long addr, int len)
  417. {
  418. unsigned long maddr;
  419. maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
  420. flush_icache_range(maddr, maddr + len);
  421. kunmap(page);
  422. }
  423. EXPORT_SYMBOL(flush_icache_user_range);
  424. /*
  425. * This is called at the end of handling a user page fault, when the
  426. * fault has been handled by updating a PTE in the linux page tables.
  427. * We use it to preload an HPTE into the hash table corresponding to
  428. * the updated linux PTE.
  429. *
  430. * This must always be called with the pte lock held.
  431. */
  432. void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
  433. pte_t *ptep)
  434. {
  435. #ifdef CONFIG_PPC_STD_MMU
  436. unsigned long access = 0, trap;
  437. /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
  438. if (!pte_young(*ptep) || address >= TASK_SIZE)
  439. return;
  440. /* We try to figure out if we are coming from an instruction
  441. * access fault and pass that down to __hash_page so we avoid
  442. * double-faulting on execution of fresh text. We have to test
  443. * for regs NULL since init will get here first thing at boot
  444. *
  445. * We also avoid filling the hash if not coming from a fault
  446. */
  447. if (current->thread.regs == NULL)
  448. return;
  449. trap = TRAP(current->thread.regs);
  450. if (trap == 0x400)
  451. access |= _PAGE_EXEC;
  452. else if (trap != 0x300)
  453. return;
  454. hash_preload(vma->vm_mm, address, access, trap);
  455. #endif /* CONFIG_PPC_STD_MMU */
  456. }