discontig.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422
  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 <asm/e820.h>
  33. #include <asm/setup.h>
  34. #include <asm/mmzone.h>
  35. #include <bios_ebda.h>
  36. struct pglist_data *node_data[MAX_NUMNODES];
  37. EXPORT_SYMBOL(node_data);
  38. bootmem_data_t node0_bdata;
  39. /*
  40. * numa interface - we expect the numa architecture specfic code to have
  41. * populated the following initialisation.
  42. *
  43. * 1) node_online_map - the map of all nodes configured (online) in the system
  44. * 2) node_start_pfn - the starting page frame number for a node
  45. * 3) node_end_pfn - the ending page fram number for a node
  46. */
  47. unsigned long node_start_pfn[MAX_NUMNODES];
  48. unsigned long node_end_pfn[MAX_NUMNODES];
  49. #ifdef CONFIG_DISCONTIGMEM
  50. /*
  51. * 4) physnode_map - the mapping between a pfn and owning node
  52. * physnode_map keeps track of the physical memory layout of a generic
  53. * numa node on a 256Mb break (each element of the array will
  54. * represent 256Mb of memory and will be marked by the node id. so,
  55. * if the first gig is on node 0, and the second gig is on node 1
  56. * physnode_map will contain:
  57. *
  58. * physnode_map[0-3] = 0;
  59. * physnode_map[4-7] = 1;
  60. * physnode_map[8- ] = -1;
  61. */
  62. s8 physnode_map[MAX_ELEMENTS] = { [0 ... (MAX_ELEMENTS - 1)] = -1};
  63. EXPORT_SYMBOL(physnode_map);
  64. void memory_present(int nid, unsigned long start, unsigned long end)
  65. {
  66. unsigned long pfn;
  67. printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
  68. nid, start, end);
  69. printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
  70. printk(KERN_DEBUG " ");
  71. for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
  72. physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
  73. printk("%ld ", pfn);
  74. }
  75. printk("\n");
  76. }
  77. unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
  78. unsigned long end_pfn)
  79. {
  80. unsigned long nr_pages = end_pfn - start_pfn;
  81. if (!nr_pages)
  82. return 0;
  83. return (nr_pages + 1) * sizeof(struct page);
  84. }
  85. #endif
  86. extern unsigned long find_max_low_pfn(void);
  87. extern void find_max_pfn(void);
  88. extern void one_highpage_init(struct page *, int, int);
  89. extern struct e820map e820;
  90. extern unsigned long init_pg_tables_end;
  91. extern unsigned long highend_pfn, highstart_pfn;
  92. extern unsigned long max_low_pfn;
  93. extern unsigned long totalram_pages;
  94. extern unsigned long totalhigh_pages;
  95. #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
  96. unsigned long node_remap_start_pfn[MAX_NUMNODES];
  97. unsigned long node_remap_size[MAX_NUMNODES];
  98. unsigned long node_remap_offset[MAX_NUMNODES];
  99. void *node_remap_start_vaddr[MAX_NUMNODES];
  100. void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
  101. void *node_remap_end_vaddr[MAX_NUMNODES];
  102. void *node_remap_alloc_vaddr[MAX_NUMNODES];
  103. /*
  104. * FLAT - support for basic PC memory model with discontig enabled, essentially
  105. * a single node with all available processors in it with a flat
  106. * memory map.
  107. */
  108. int __init get_memcfg_numa_flat(void)
  109. {
  110. printk("NUMA - single node, flat memory mode\n");
  111. /* Run the memory configuration and find the top of memory. */
  112. find_max_pfn();
  113. node_start_pfn[0] = 0;
  114. node_end_pfn[0] = max_pfn;
  115. memory_present(0, 0, max_pfn);
  116. /* Indicate there is one node available. */
  117. nodes_clear(node_online_map);
  118. node_set_online(0);
  119. return 1;
  120. }
  121. /*
  122. * Find the highest page frame number we have available for the node
  123. */
  124. static void __init find_max_pfn_node(int nid)
  125. {
  126. if (node_end_pfn[nid] > max_pfn)
  127. node_end_pfn[nid] = max_pfn;
  128. /*
  129. * if a user has given mem=XXXX, then we need to make sure
  130. * that the node _starts_ before that, too, not just ends
  131. */
  132. if (node_start_pfn[nid] > max_pfn)
  133. node_start_pfn[nid] = max_pfn;
  134. if (node_start_pfn[nid] > node_end_pfn[nid])
  135. BUG();
  136. }
  137. /* Find the owning node for a pfn. */
  138. int early_pfn_to_nid(unsigned long pfn)
  139. {
  140. int nid;
  141. for_each_node(nid) {
  142. if (node_end_pfn[nid] == 0)
  143. break;
  144. if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
  145. return nid;
  146. }
  147. return 0;
  148. }
  149. /*
  150. * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
  151. * method. For node zero take this from the bottom of memory, for
  152. * subsequent nodes place them at node_remap_start_vaddr which contains
  153. * node local data in physically node local memory. See setup_memory()
  154. * for details.
  155. */
  156. static void __init allocate_pgdat(int nid)
  157. {
  158. if (nid && node_has_online_mem(nid))
  159. NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
  160. else {
  161. NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
  162. min_low_pfn += PFN_UP(sizeof(pg_data_t));
  163. }
  164. }
  165. void *alloc_remap(int nid, unsigned long size)
  166. {
  167. void *allocation = node_remap_alloc_vaddr[nid];
  168. size = ALIGN(size, L1_CACHE_BYTES);
  169. if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
  170. return 0;
  171. node_remap_alloc_vaddr[nid] += size;
  172. memset(allocation, 0, size);
  173. return allocation;
  174. }
  175. void __init remap_numa_kva(void)
  176. {
  177. void *vaddr;
  178. unsigned long pfn;
  179. int node;
  180. for_each_online_node(node) {
  181. for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
  182. vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
  183. set_pmd_pfn((ulong) vaddr,
  184. node_remap_start_pfn[node] + pfn,
  185. PAGE_KERNEL_LARGE);
  186. }
  187. }
  188. }
  189. static unsigned long calculate_numa_remap_pages(void)
  190. {
  191. int nid;
  192. unsigned long size, reserve_pages = 0;
  193. unsigned long pfn;
  194. for_each_online_node(nid) {
  195. /*
  196. * The acpi/srat node info can show hot-add memroy zones
  197. * where memory could be added but not currently present.
  198. */
  199. if (node_start_pfn[nid] > max_pfn)
  200. continue;
  201. if (node_end_pfn[nid] > max_pfn)
  202. node_end_pfn[nid] = max_pfn;
  203. /* ensure the remap includes space for the pgdat. */
  204. size = node_remap_size[nid] + sizeof(pg_data_t);
  205. /* convert size to large (pmd size) pages, rounding up */
  206. size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
  207. /* now the roundup is correct, convert to PAGE_SIZE pages */
  208. size = size * PTRS_PER_PTE;
  209. /*
  210. * Validate the region we are allocating only contains valid
  211. * pages.
  212. */
  213. for (pfn = node_end_pfn[nid] - size;
  214. pfn < node_end_pfn[nid]; pfn++)
  215. if (!page_is_ram(pfn))
  216. break;
  217. if (pfn != node_end_pfn[nid])
  218. size = 0;
  219. printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
  220. size, nid);
  221. node_remap_size[nid] = size;
  222. node_remap_offset[nid] = reserve_pages;
  223. reserve_pages += size;
  224. printk("Shrinking node %d from %ld pages to %ld pages\n",
  225. nid, node_end_pfn[nid], node_end_pfn[nid] - size);
  226. node_end_pfn[nid] -= size;
  227. node_remap_start_pfn[nid] = node_end_pfn[nid];
  228. }
  229. printk("Reserving total of %ld pages for numa KVA remap\n",
  230. reserve_pages);
  231. return reserve_pages;
  232. }
  233. extern void setup_bootmem_allocator(void);
  234. unsigned long __init setup_memory(void)
  235. {
  236. int nid;
  237. unsigned long system_start_pfn, system_max_low_pfn;
  238. unsigned long reserve_pages;
  239. /*
  240. * When mapping a NUMA machine we allocate the node_mem_map arrays
  241. * from node local memory. They are then mapped directly into KVA
  242. * between zone normal and vmalloc space. Calculate the size of
  243. * this space and use it to adjust the boundry between ZONE_NORMAL
  244. * and ZONE_HIGHMEM.
  245. */
  246. find_max_pfn();
  247. get_memcfg_numa();
  248. reserve_pages = calculate_numa_remap_pages();
  249. /* partially used pages are not usable - thus round upwards */
  250. system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
  251. system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
  252. printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
  253. reserve_pages, max_low_pfn + reserve_pages);
  254. printk("max_pfn = %ld\n", max_pfn);
  255. #ifdef CONFIG_HIGHMEM
  256. highstart_pfn = highend_pfn = max_pfn;
  257. if (max_pfn > system_max_low_pfn)
  258. highstart_pfn = system_max_low_pfn;
  259. printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
  260. pages_to_mb(highend_pfn - highstart_pfn));
  261. #endif
  262. printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
  263. pages_to_mb(system_max_low_pfn));
  264. printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
  265. min_low_pfn, max_low_pfn, highstart_pfn);
  266. printk("Low memory ends at vaddr %08lx\n",
  267. (ulong) pfn_to_kaddr(max_low_pfn));
  268. for_each_online_node(nid) {
  269. node_remap_start_vaddr[nid] = pfn_to_kaddr(
  270. highstart_pfn + node_remap_offset[nid]);
  271. /* Init the node remap allocator */
  272. node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
  273. (node_remap_size[nid] * PAGE_SIZE);
  274. node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
  275. ALIGN(sizeof(pg_data_t), PAGE_SIZE);
  276. allocate_pgdat(nid);
  277. printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
  278. (ulong) node_remap_start_vaddr[nid],
  279. (ulong) pfn_to_kaddr(highstart_pfn
  280. + node_remap_offset[nid] + node_remap_size[nid]));
  281. }
  282. printk("High memory starts at vaddr %08lx\n",
  283. (ulong) pfn_to_kaddr(highstart_pfn));
  284. vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
  285. for_each_online_node(nid)
  286. find_max_pfn_node(nid);
  287. memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
  288. NODE_DATA(0)->bdata = &node0_bdata;
  289. setup_bootmem_allocator();
  290. return max_low_pfn;
  291. }
  292. void __init zone_sizes_init(void)
  293. {
  294. int nid;
  295. /*
  296. * Insert nodes into pgdat_list backward so they appear in order.
  297. * Clobber node 0's links and NULL out pgdat_list before starting.
  298. */
  299. pgdat_list = NULL;
  300. for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) {
  301. if (!node_online(nid))
  302. continue;
  303. NODE_DATA(nid)->pgdat_next = pgdat_list;
  304. pgdat_list = NODE_DATA(nid);
  305. }
  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. one_highpage_init(page, node_pfn, bad_ppro);
  359. }
  360. }
  361. totalram_pages += totalhigh_pages;
  362. #endif
  363. }