discontig.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419
  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 <asm/e820.h>
  32. #include <asm/setup.h>
  33. #include <asm/mmzone.h>
  34. #include <bios_ebda.h>
  35. struct pglist_data *node_data[MAX_NUMNODES];
  36. bootmem_data_t node0_bdata;
  37. /*
  38. * numa interface - we expect the numa architecture specfic code to have
  39. * populated the following initialisation.
  40. *
  41. * 1) node_online_map - the map of all nodes configured (online) in the system
  42. * 2) node_start_pfn - the starting page frame number for a node
  43. * 3) node_end_pfn - the ending page fram number for a node
  44. */
  45. unsigned long node_start_pfn[MAX_NUMNODES];
  46. unsigned long node_end_pfn[MAX_NUMNODES];
  47. #ifdef CONFIG_DISCONTIGMEM
  48. /*
  49. * 4) physnode_map - the mapping between a pfn and owning node
  50. * physnode_map keeps track of the physical memory layout of a generic
  51. * numa node on a 256Mb break (each element of the array will
  52. * represent 256Mb of memory and will be marked by the node id. so,
  53. * if the first gig is on node 0, and the second gig is on node 1
  54. * physnode_map will contain:
  55. *
  56. * physnode_map[0-3] = 0;
  57. * physnode_map[4-7] = 1;
  58. * physnode_map[8- ] = -1;
  59. */
  60. s8 physnode_map[MAX_ELEMENTS] = { [0 ... (MAX_ELEMENTS - 1)] = -1};
  61. void memory_present(int nid, unsigned long start, unsigned long end)
  62. {
  63. unsigned long pfn;
  64. printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
  65. nid, start, end);
  66. printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
  67. printk(KERN_DEBUG " ");
  68. for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
  69. physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
  70. printk("%ld ", pfn);
  71. }
  72. printk("\n");
  73. }
  74. unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
  75. unsigned long end_pfn)
  76. {
  77. unsigned long nr_pages = end_pfn - start_pfn;
  78. if (!nr_pages)
  79. return 0;
  80. return (nr_pages + 1) * sizeof(struct page);
  81. }
  82. #endif
  83. extern unsigned long find_max_low_pfn(void);
  84. extern void find_max_pfn(void);
  85. extern void one_highpage_init(struct page *, int, int);
  86. extern struct e820map e820;
  87. extern unsigned long init_pg_tables_end;
  88. extern unsigned long highend_pfn, highstart_pfn;
  89. extern unsigned long max_low_pfn;
  90. extern unsigned long totalram_pages;
  91. extern unsigned long totalhigh_pages;
  92. #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
  93. unsigned long node_remap_start_pfn[MAX_NUMNODES];
  94. unsigned long node_remap_size[MAX_NUMNODES];
  95. unsigned long node_remap_offset[MAX_NUMNODES];
  96. void *node_remap_start_vaddr[MAX_NUMNODES];
  97. void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
  98. void *node_remap_end_vaddr[MAX_NUMNODES];
  99. void *node_remap_alloc_vaddr[MAX_NUMNODES];
  100. /*
  101. * FLAT - support for basic PC memory model with discontig enabled, essentially
  102. * a single node with all available processors in it with a flat
  103. * memory map.
  104. */
  105. int __init get_memcfg_numa_flat(void)
  106. {
  107. printk("NUMA - single node, flat memory mode\n");
  108. /* Run the memory configuration and find the top of memory. */
  109. find_max_pfn();
  110. node_start_pfn[0] = 0;
  111. node_end_pfn[0] = max_pfn;
  112. memory_present(0, 0, max_pfn);
  113. /* Indicate there is one node available. */
  114. nodes_clear(node_online_map);
  115. node_set_online(0);
  116. return 1;
  117. }
  118. /*
  119. * Find the highest page frame number we have available for the node
  120. */
  121. static void __init find_max_pfn_node(int nid)
  122. {
  123. if (node_end_pfn[nid] > max_pfn)
  124. node_end_pfn[nid] = max_pfn;
  125. /*
  126. * if a user has given mem=XXXX, then we need to make sure
  127. * that the node _starts_ before that, too, not just ends
  128. */
  129. if (node_start_pfn[nid] > max_pfn)
  130. node_start_pfn[nid] = max_pfn;
  131. if (node_start_pfn[nid] > node_end_pfn[nid])
  132. BUG();
  133. }
  134. /* Find the owning node for a pfn. */
  135. int early_pfn_to_nid(unsigned long pfn)
  136. {
  137. int nid;
  138. for_each_node(nid) {
  139. if (node_end_pfn[nid] == 0)
  140. break;
  141. if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
  142. return nid;
  143. }
  144. return 0;
  145. }
  146. /*
  147. * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
  148. * method. For node zero take this from the bottom of memory, for
  149. * subsequent nodes place them at node_remap_start_vaddr which contains
  150. * node local data in physically node local memory. See setup_memory()
  151. * for details.
  152. */
  153. static void __init allocate_pgdat(int nid)
  154. {
  155. if (nid && node_has_online_mem(nid))
  156. NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
  157. else {
  158. NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
  159. min_low_pfn += PFN_UP(sizeof(pg_data_t));
  160. }
  161. }
  162. void *alloc_remap(int nid, unsigned long size)
  163. {
  164. void *allocation = node_remap_alloc_vaddr[nid];
  165. size = ALIGN(size, L1_CACHE_BYTES);
  166. if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
  167. return 0;
  168. node_remap_alloc_vaddr[nid] += size;
  169. memset(allocation, 0, size);
  170. return allocation;
  171. }
  172. void __init remap_numa_kva(void)
  173. {
  174. void *vaddr;
  175. unsigned long pfn;
  176. int node;
  177. for_each_online_node(node) {
  178. for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
  179. vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
  180. set_pmd_pfn((ulong) vaddr,
  181. node_remap_start_pfn[node] + pfn,
  182. PAGE_KERNEL_LARGE);
  183. }
  184. }
  185. }
  186. static unsigned long calculate_numa_remap_pages(void)
  187. {
  188. int nid;
  189. unsigned long size, reserve_pages = 0;
  190. unsigned long pfn;
  191. for_each_online_node(nid) {
  192. /*
  193. * The acpi/srat node info can show hot-add memroy zones
  194. * where memory could be added but not currently present.
  195. */
  196. if (node_start_pfn[nid] > max_pfn)
  197. continue;
  198. if (node_end_pfn[nid] > max_pfn)
  199. node_end_pfn[nid] = max_pfn;
  200. /* ensure the remap includes space for the pgdat. */
  201. size = node_remap_size[nid] + sizeof(pg_data_t);
  202. /* convert size to large (pmd size) pages, rounding up */
  203. size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
  204. /* now the roundup is correct, convert to PAGE_SIZE pages */
  205. size = size * PTRS_PER_PTE;
  206. /*
  207. * Validate the region we are allocating only contains valid
  208. * pages.
  209. */
  210. for (pfn = node_end_pfn[nid] - size;
  211. pfn < node_end_pfn[nid]; pfn++)
  212. if (!page_is_ram(pfn))
  213. break;
  214. if (pfn != node_end_pfn[nid])
  215. size = 0;
  216. printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
  217. size, nid);
  218. node_remap_size[nid] = size;
  219. node_remap_offset[nid] = reserve_pages;
  220. reserve_pages += size;
  221. printk("Shrinking node %d from %ld pages to %ld pages\n",
  222. nid, node_end_pfn[nid], node_end_pfn[nid] - size);
  223. node_end_pfn[nid] -= size;
  224. node_remap_start_pfn[nid] = node_end_pfn[nid];
  225. }
  226. printk("Reserving total of %ld pages for numa KVA remap\n",
  227. reserve_pages);
  228. return reserve_pages;
  229. }
  230. extern void setup_bootmem_allocator(void);
  231. unsigned long __init setup_memory(void)
  232. {
  233. int nid;
  234. unsigned long system_start_pfn, system_max_low_pfn;
  235. unsigned long reserve_pages;
  236. /*
  237. * When mapping a NUMA machine we allocate the node_mem_map arrays
  238. * from node local memory. They are then mapped directly into KVA
  239. * between zone normal and vmalloc space. Calculate the size of
  240. * this space and use it to adjust the boundry between ZONE_NORMAL
  241. * and ZONE_HIGHMEM.
  242. */
  243. find_max_pfn();
  244. get_memcfg_numa();
  245. reserve_pages = calculate_numa_remap_pages();
  246. /* partially used pages are not usable - thus round upwards */
  247. system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
  248. system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
  249. printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
  250. reserve_pages, max_low_pfn + reserve_pages);
  251. printk("max_pfn = %ld\n", max_pfn);
  252. #ifdef CONFIG_HIGHMEM
  253. highstart_pfn = highend_pfn = max_pfn;
  254. if (max_pfn > system_max_low_pfn)
  255. highstart_pfn = system_max_low_pfn;
  256. printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
  257. pages_to_mb(highend_pfn - highstart_pfn));
  258. #endif
  259. printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
  260. pages_to_mb(system_max_low_pfn));
  261. printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
  262. min_low_pfn, max_low_pfn, highstart_pfn);
  263. printk("Low memory ends at vaddr %08lx\n",
  264. (ulong) pfn_to_kaddr(max_low_pfn));
  265. for_each_online_node(nid) {
  266. node_remap_start_vaddr[nid] = pfn_to_kaddr(
  267. highstart_pfn + node_remap_offset[nid]);
  268. /* Init the node remap allocator */
  269. node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
  270. (node_remap_size[nid] * PAGE_SIZE);
  271. node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
  272. ALIGN(sizeof(pg_data_t), PAGE_SIZE);
  273. allocate_pgdat(nid);
  274. printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
  275. (ulong) node_remap_start_vaddr[nid],
  276. (ulong) pfn_to_kaddr(highstart_pfn
  277. + node_remap_offset[nid] + node_remap_size[nid]));
  278. }
  279. printk("High memory starts at vaddr %08lx\n",
  280. (ulong) pfn_to_kaddr(highstart_pfn));
  281. vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
  282. for_each_online_node(nid)
  283. find_max_pfn_node(nid);
  284. memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
  285. NODE_DATA(0)->bdata = &node0_bdata;
  286. setup_bootmem_allocator();
  287. return max_low_pfn;
  288. }
  289. void __init zone_sizes_init(void)
  290. {
  291. int nid;
  292. /*
  293. * Insert nodes into pgdat_list backward so they appear in order.
  294. * Clobber node 0's links and NULL out pgdat_list before starting.
  295. */
  296. pgdat_list = NULL;
  297. for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) {
  298. if (!node_online(nid))
  299. continue;
  300. NODE_DATA(nid)->pgdat_next = pgdat_list;
  301. pgdat_list = NODE_DATA(nid);
  302. }
  303. for_each_online_node(nid) {
  304. unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
  305. unsigned long *zholes_size;
  306. unsigned int max_dma;
  307. unsigned long low = max_low_pfn;
  308. unsigned long start = node_start_pfn[nid];
  309. unsigned long high = node_end_pfn[nid];
  310. max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
  311. if (node_has_online_mem(nid)){
  312. if (start > low) {
  313. #ifdef CONFIG_HIGHMEM
  314. BUG_ON(start > high);
  315. zones_size[ZONE_HIGHMEM] = high - start;
  316. #endif
  317. } else {
  318. if (low < max_dma)
  319. zones_size[ZONE_DMA] = low;
  320. else {
  321. BUG_ON(max_dma > low);
  322. BUG_ON(low > high);
  323. zones_size[ZONE_DMA] = max_dma;
  324. zones_size[ZONE_NORMAL] = low - max_dma;
  325. #ifdef CONFIG_HIGHMEM
  326. zones_size[ZONE_HIGHMEM] = high - low;
  327. #endif
  328. }
  329. }
  330. }
  331. zholes_size = get_zholes_size(nid);
  332. free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
  333. zholes_size);
  334. }
  335. return;
  336. }
  337. void __init set_highmem_pages_init(int bad_ppro)
  338. {
  339. #ifdef CONFIG_HIGHMEM
  340. struct zone *zone;
  341. struct page *page;
  342. for_each_zone(zone) {
  343. unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
  344. if (!is_highmem(zone))
  345. continue;
  346. zone_start_pfn = zone->zone_start_pfn;
  347. zone_end_pfn = zone_start_pfn + zone->spanned_pages;
  348. printk("Initializing %s for node %d (%08lx:%08lx)\n",
  349. zone->name, zone->zone_pgdat->node_id,
  350. zone_start_pfn, zone_end_pfn);
  351. for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
  352. if (!pfn_valid(node_pfn))
  353. continue;
  354. page = pfn_to_page(node_pfn);
  355. one_highpage_init(page, node_pfn, bad_ppro);
  356. }
  357. }
  358. totalram_pages += totalhigh_pages;
  359. #endif
  360. }