discontig.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425
  1. /*
  2. * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
  3. * August 2002: added remote node KVA remap - Martin J. Bligh
  4. *
  5. * Copyright (C) 2002, IBM Corp.
  6. *
  7. * All rights reserved.
  8. *
  9. * This program is free software; you can redistribute it and/or modify
  10. * it under the terms of the GNU General Public License as published by
  11. * the Free Software Foundation; either version 2 of the License, or
  12. * (at your option) any later version.
  13. *
  14. * This program is distributed in the hope that it will be useful, but
  15. * WITHOUT ANY WARRANTY; without even the implied warranty of
  16. * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  17. * NON INFRINGEMENT. See the GNU General Public License for more
  18. * details.
  19. *
  20. * You should have received a copy of the GNU General Public License
  21. * along with this program; if not, write to the Free Software
  22. * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23. */
  24. #include <linux/config.h>
  25. #include <linux/mm.h>
  26. #include <linux/bootmem.h>
  27. #include <linux/mmzone.h>
  28. #include <linux/highmem.h>
  29. #include <linux/initrd.h>
  30. #include <linux/nodemask.h>
  31. #include <linux/module.h>
  32. #include <linux/kexec.h>
  33. #include <linux/pfn.h>
  34. #include <asm/e820.h>
  35. #include <asm/setup.h>
  36. #include <asm/mmzone.h>
  37. #include <bios_ebda.h>
  38. struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
  39. EXPORT_SYMBOL(node_data);
  40. bootmem_data_t node0_bdata;
  41. /*
  42. * numa interface - we expect the numa architecture specfic code to have
  43. * populated the following initialisation.
  44. *
  45. * 1) node_online_map - the map of all nodes configured (online) in the system
  46. * 2) node_start_pfn - the starting page frame number for a node
  47. * 3) node_end_pfn - the ending page fram number for a node
  48. */
  49. unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
  50. unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
  51. #ifdef CONFIG_DISCONTIGMEM
  52. /*
  53. * 4) physnode_map - the mapping between a pfn and owning node
  54. * physnode_map keeps track of the physical memory layout of a generic
  55. * numa node on a 256Mb break (each element of the array will
  56. * represent 256Mb of memory and will be marked by the node id. so,
  57. * if the first gig is on node 0, and the second gig is on node 1
  58. * physnode_map will contain:
  59. *
  60. * physnode_map[0-3] = 0;
  61. * physnode_map[4-7] = 1;
  62. * physnode_map[8- ] = -1;
  63. */
  64. s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
  65. EXPORT_SYMBOL(physnode_map);
  66. void memory_present(int nid, unsigned long start, unsigned long end)
  67. {
  68. unsigned long pfn;
  69. printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
  70. nid, start, end);
  71. printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
  72. printk(KERN_DEBUG " ");
  73. for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
  74. physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
  75. printk("%ld ", pfn);
  76. }
  77. printk("\n");
  78. }
  79. unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
  80. unsigned long end_pfn)
  81. {
  82. unsigned long nr_pages = end_pfn - start_pfn;
  83. if (!nr_pages)
  84. return 0;
  85. return (nr_pages + 1) * sizeof(struct page);
  86. }
  87. #endif
  88. extern unsigned long find_max_low_pfn(void);
  89. extern void find_max_pfn(void);
  90. extern void add_one_highpage_init(struct page *, int, int);
  91. extern struct e820map e820;
  92. extern unsigned long init_pg_tables_end;
  93. extern unsigned long highend_pfn, highstart_pfn;
  94. extern unsigned long max_low_pfn;
  95. extern unsigned long totalram_pages;
  96. extern unsigned long totalhigh_pages;
  97. #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
  98. unsigned long node_remap_start_pfn[MAX_NUMNODES];
  99. unsigned long node_remap_size[MAX_NUMNODES];
  100. unsigned long node_remap_offset[MAX_NUMNODES];
  101. void *node_remap_start_vaddr[MAX_NUMNODES];
  102. void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
  103. void *node_remap_end_vaddr[MAX_NUMNODES];
  104. void *node_remap_alloc_vaddr[MAX_NUMNODES];
  105. /*
  106. * FLAT - support for basic PC memory model with discontig enabled, essentially
  107. * a single node with all available processors in it with a flat
  108. * memory map.
  109. */
  110. int __init get_memcfg_numa_flat(void)
  111. {
  112. printk("NUMA - single node, flat memory mode\n");
  113. /* Run the memory configuration and find the top of memory. */
  114. find_max_pfn();
  115. node_start_pfn[0] = 0;
  116. node_end_pfn[0] = max_pfn;
  117. memory_present(0, 0, max_pfn);
  118. /* Indicate there is one node available. */
  119. nodes_clear(node_online_map);
  120. node_set_online(0);
  121. return 1;
  122. }
  123. /*
  124. * Find the highest page frame number we have available for the node
  125. */
  126. static void __init find_max_pfn_node(int nid)
  127. {
  128. if (node_end_pfn[nid] > max_pfn)
  129. node_end_pfn[nid] = max_pfn;
  130. /*
  131. * if a user has given mem=XXXX, then we need to make sure
  132. * that the node _starts_ before that, too, not just ends
  133. */
  134. if (node_start_pfn[nid] > max_pfn)
  135. node_start_pfn[nid] = max_pfn;
  136. if (node_start_pfn[nid] > node_end_pfn[nid])
  137. BUG();
  138. }
  139. /* Find the owning node for a pfn. */
  140. int early_pfn_to_nid(unsigned long pfn)
  141. {
  142. int nid;
  143. for_each_node(nid) {
  144. if (node_end_pfn[nid] == 0)
  145. break;
  146. if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
  147. return nid;
  148. }
  149. return 0;
  150. }
  151. /*
  152. * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
  153. * method. For node zero take this from the bottom of memory, for
  154. * subsequent nodes place them at node_remap_start_vaddr which contains
  155. * node local data in physically node local memory. See setup_memory()
  156. * for details.
  157. */
  158. static void __init allocate_pgdat(int nid)
  159. {
  160. if (nid && node_has_online_mem(nid))
  161. NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
  162. else {
  163. NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
  164. min_low_pfn += PFN_UP(sizeof(pg_data_t));
  165. }
  166. }
  167. void *alloc_remap(int nid, unsigned long size)
  168. {
  169. void *allocation = node_remap_alloc_vaddr[nid];
  170. size = ALIGN(size, L1_CACHE_BYTES);
  171. if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
  172. return 0;
  173. node_remap_alloc_vaddr[nid] += size;
  174. memset(allocation, 0, size);
  175. return allocation;
  176. }
  177. void __init remap_numa_kva(void)
  178. {
  179. void *vaddr;
  180. unsigned long pfn;
  181. int node;
  182. for_each_online_node(node) {
  183. for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
  184. vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
  185. set_pmd_pfn((ulong) vaddr,
  186. node_remap_start_pfn[node] + pfn,
  187. PAGE_KERNEL_LARGE);
  188. }
  189. }
  190. }
  191. static unsigned long calculate_numa_remap_pages(void)
  192. {
  193. int nid;
  194. unsigned long size, reserve_pages = 0;
  195. unsigned long pfn;
  196. for_each_online_node(nid) {
  197. /*
  198. * The acpi/srat node info can show hot-add memroy zones
  199. * where memory could be added but not currently present.
  200. */
  201. if (node_start_pfn[nid] > max_pfn)
  202. continue;
  203. if (node_end_pfn[nid] > max_pfn)
  204. node_end_pfn[nid] = max_pfn;
  205. /* ensure the remap includes space for the pgdat. */
  206. size = node_remap_size[nid] + sizeof(pg_data_t);
  207. /* convert size to large (pmd size) pages, rounding up */
  208. size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
  209. /* now the roundup is correct, convert to PAGE_SIZE pages */
  210. size = size * PTRS_PER_PTE;
  211. /*
  212. * Validate the region we are allocating only contains valid
  213. * pages.
  214. */
  215. for (pfn = node_end_pfn[nid] - size;
  216. pfn < node_end_pfn[nid]; pfn++)
  217. if (!page_is_ram(pfn))
  218. break;
  219. if (pfn != node_end_pfn[nid])
  220. size = 0;
  221. printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
  222. size, nid);
  223. node_remap_size[nid] = size;
  224. node_remap_offset[nid] = reserve_pages;
  225. reserve_pages += size;
  226. printk("Shrinking node %d from %ld pages to %ld pages\n",
  227. nid, node_end_pfn[nid], node_end_pfn[nid] - size);
  228. if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
  229. /*
  230. * Align node_end_pfn[] and node_remap_start_pfn[] to
  231. * pmd boundary. remap_numa_kva will barf otherwise.
  232. */
  233. printk("Shrinking node %d further by %ld pages for proper alignment\n",
  234. nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
  235. size += node_end_pfn[nid] & (PTRS_PER_PTE-1);
  236. }
  237. node_end_pfn[nid] -= size;
  238. node_remap_start_pfn[nid] = node_end_pfn[nid];
  239. }
  240. printk("Reserving total of %ld pages for numa KVA remap\n",
  241. reserve_pages);
  242. return reserve_pages;
  243. }
  244. extern void setup_bootmem_allocator(void);
  245. unsigned long __init setup_memory(void)
  246. {
  247. int nid;
  248. unsigned long system_start_pfn, system_max_low_pfn;
  249. unsigned long reserve_pages;
  250. /*
  251. * When mapping a NUMA machine we allocate the node_mem_map arrays
  252. * from node local memory. They are then mapped directly into KVA
  253. * between zone normal and vmalloc space. Calculate the size of
  254. * this space and use it to adjust the boundry between ZONE_NORMAL
  255. * and ZONE_HIGHMEM.
  256. */
  257. find_max_pfn();
  258. get_memcfg_numa();
  259. reserve_pages = calculate_numa_remap_pages();
  260. /* partially used pages are not usable - thus round upwards */
  261. system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
  262. system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
  263. printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
  264. reserve_pages, max_low_pfn + reserve_pages);
  265. printk("max_pfn = %ld\n", max_pfn);
  266. #ifdef CONFIG_HIGHMEM
  267. highstart_pfn = highend_pfn = max_pfn;
  268. if (max_pfn > system_max_low_pfn)
  269. highstart_pfn = system_max_low_pfn;
  270. printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
  271. pages_to_mb(highend_pfn - highstart_pfn));
  272. #endif
  273. printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
  274. pages_to_mb(system_max_low_pfn));
  275. printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
  276. min_low_pfn, max_low_pfn, highstart_pfn);
  277. printk("Low memory ends at vaddr %08lx\n",
  278. (ulong) pfn_to_kaddr(max_low_pfn));
  279. for_each_online_node(nid) {
  280. node_remap_start_vaddr[nid] = pfn_to_kaddr(
  281. highstart_pfn + node_remap_offset[nid]);
  282. /* Init the node remap allocator */
  283. node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
  284. (node_remap_size[nid] * PAGE_SIZE);
  285. node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
  286. ALIGN(sizeof(pg_data_t), PAGE_SIZE);
  287. allocate_pgdat(nid);
  288. printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
  289. (ulong) node_remap_start_vaddr[nid],
  290. (ulong) pfn_to_kaddr(highstart_pfn
  291. + node_remap_offset[nid] + node_remap_size[nid]));
  292. }
  293. printk("High memory starts at vaddr %08lx\n",
  294. (ulong) pfn_to_kaddr(highstart_pfn));
  295. vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
  296. for_each_online_node(nid)
  297. find_max_pfn_node(nid);
  298. memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
  299. NODE_DATA(0)->bdata = &node0_bdata;
  300. setup_bootmem_allocator();
  301. return max_low_pfn;
  302. }
  303. void __init zone_sizes_init(void)
  304. {
  305. int nid;
  306. for_each_online_node(nid) {
  307. unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
  308. unsigned long *zholes_size;
  309. unsigned int max_dma;
  310. unsigned long low = max_low_pfn;
  311. unsigned long start = node_start_pfn[nid];
  312. unsigned long high = node_end_pfn[nid];
  313. max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
  314. if (node_has_online_mem(nid)){
  315. if (start > low) {
  316. #ifdef CONFIG_HIGHMEM
  317. BUG_ON(start > high);
  318. zones_size[ZONE_HIGHMEM] = high - start;
  319. #endif
  320. } else {
  321. if (low < max_dma)
  322. zones_size[ZONE_DMA] = low;
  323. else {
  324. BUG_ON(max_dma > low);
  325. BUG_ON(low > high);
  326. zones_size[ZONE_DMA] = max_dma;
  327. zones_size[ZONE_NORMAL] = low - max_dma;
  328. #ifdef CONFIG_HIGHMEM
  329. zones_size[ZONE_HIGHMEM] = high - low;
  330. #endif
  331. }
  332. }
  333. }
  334. zholes_size = get_zholes_size(nid);
  335. free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
  336. zholes_size);
  337. }
  338. return;
  339. }
  340. void __init set_highmem_pages_init(int bad_ppro)
  341. {
  342. #ifdef CONFIG_HIGHMEM
  343. struct zone *zone;
  344. struct page *page;
  345. for_each_zone(zone) {
  346. unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
  347. if (!is_highmem(zone))
  348. continue;
  349. zone_start_pfn = zone->zone_start_pfn;
  350. zone_end_pfn = zone_start_pfn + zone->spanned_pages;
  351. printk("Initializing %s for node %d (%08lx:%08lx)\n",
  352. zone->name, zone->zone_pgdat->node_id,
  353. zone_start_pfn, zone_end_pfn);
  354. for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
  355. if (!pfn_valid(node_pfn))
  356. continue;
  357. page = pfn_to_page(node_pfn);
  358. add_one_highpage_init(page, node_pfn, bad_ppro);
  359. }
  360. }
  361. totalram_pages += totalhigh_pages;
  362. #endif
  363. }