discontig.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435
  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 <asm/e820.h>
  34. #include <asm/setup.h>
  35. #include <asm/mmzone.h>
  36. #include <bios_ebda.h>
  37. struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
  38. EXPORT_SYMBOL(node_data);
  39. bootmem_data_t node0_bdata;
  40. /*
  41. * numa interface - we expect the numa architecture specfic code to have
  42. * populated the following initialisation.
  43. *
  44. * 1) node_online_map - the map of all nodes configured (online) in the system
  45. * 2) node_start_pfn - the starting page frame number for a node
  46. * 3) node_end_pfn - the ending page fram number for a node
  47. */
  48. unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
  49. unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
  50. #ifdef CONFIG_DISCONTIGMEM
  51. /*
  52. * 4) physnode_map - the mapping between a pfn and owning node
  53. * physnode_map keeps track of the physical memory layout of a generic
  54. * numa node on a 256Mb break (each element of the array will
  55. * represent 256Mb of memory and will be marked by the node id. so,
  56. * if the first gig is on node 0, and the second gig is on node 1
  57. * physnode_map will contain:
  58. *
  59. * physnode_map[0-3] = 0;
  60. * physnode_map[4-7] = 1;
  61. * physnode_map[8- ] = -1;
  62. */
  63. s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
  64. EXPORT_SYMBOL(physnode_map);
  65. void memory_present(int nid, unsigned long start, unsigned long end)
  66. {
  67. unsigned long pfn;
  68. printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
  69. nid, start, end);
  70. printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
  71. printk(KERN_DEBUG " ");
  72. for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
  73. physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
  74. printk("%ld ", pfn);
  75. }
  76. printk("\n");
  77. }
  78. unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
  79. unsigned long end_pfn)
  80. {
  81. unsigned long nr_pages = end_pfn - start_pfn;
  82. if (!nr_pages)
  83. return 0;
  84. return (nr_pages + 1) * sizeof(struct page);
  85. }
  86. #endif
  87. extern unsigned long find_max_low_pfn(void);
  88. extern void find_max_pfn(void);
  89. extern void one_highpage_init(struct page *, int, int);
  90. extern struct e820map e820;
  91. extern unsigned long init_pg_tables_end;
  92. extern unsigned long highend_pfn, highstart_pfn;
  93. extern unsigned long max_low_pfn;
  94. extern unsigned long totalram_pages;
  95. extern unsigned long totalhigh_pages;
  96. #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
  97. unsigned long node_remap_start_pfn[MAX_NUMNODES];
  98. unsigned long node_remap_size[MAX_NUMNODES];
  99. unsigned long node_remap_offset[MAX_NUMNODES];
  100. void *node_remap_start_vaddr[MAX_NUMNODES];
  101. void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
  102. void *node_remap_end_vaddr[MAX_NUMNODES];
  103. void *node_remap_alloc_vaddr[MAX_NUMNODES];
  104. /*
  105. * FLAT - support for basic PC memory model with discontig enabled, essentially
  106. * a single node with all available processors in it with a flat
  107. * memory map.
  108. */
  109. int __init get_memcfg_numa_flat(void)
  110. {
  111. printk("NUMA - single node, flat memory mode\n");
  112. /* Run the memory configuration and find the top of memory. */
  113. find_max_pfn();
  114. node_start_pfn[0] = 0;
  115. node_end_pfn[0] = max_pfn;
  116. memory_present(0, 0, max_pfn);
  117. /* Indicate there is one node available. */
  118. nodes_clear(node_online_map);
  119. node_set_online(0);
  120. return 1;
  121. }
  122. /*
  123. * Find the highest page frame number we have available for the node
  124. */
  125. static void __init find_max_pfn_node(int nid)
  126. {
  127. if (node_end_pfn[nid] > max_pfn)
  128. node_end_pfn[nid] = max_pfn;
  129. /*
  130. * if a user has given mem=XXXX, then we need to make sure
  131. * that the node _starts_ before that, too, not just ends
  132. */
  133. if (node_start_pfn[nid] > max_pfn)
  134. node_start_pfn[nid] = max_pfn;
  135. if (node_start_pfn[nid] > node_end_pfn[nid])
  136. BUG();
  137. }
  138. /* Find the owning node for a pfn. */
  139. int early_pfn_to_nid(unsigned long pfn)
  140. {
  141. int nid;
  142. for_each_node(nid) {
  143. if (node_end_pfn[nid] == 0)
  144. break;
  145. if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
  146. return nid;
  147. }
  148. return 0;
  149. }
  150. /*
  151. * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
  152. * method. For node zero take this from the bottom of memory, for
  153. * subsequent nodes place them at node_remap_start_vaddr which contains
  154. * node local data in physically node local memory. See setup_memory()
  155. * for details.
  156. */
  157. static void __init allocate_pgdat(int nid)
  158. {
  159. if (nid && node_has_online_mem(nid))
  160. NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
  161. else {
  162. NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
  163. min_low_pfn += PFN_UP(sizeof(pg_data_t));
  164. }
  165. }
  166. void *alloc_remap(int nid, unsigned long size)
  167. {
  168. void *allocation = node_remap_alloc_vaddr[nid];
  169. size = ALIGN(size, L1_CACHE_BYTES);
  170. if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
  171. return 0;
  172. node_remap_alloc_vaddr[nid] += size;
  173. memset(allocation, 0, size);
  174. return allocation;
  175. }
  176. void __init remap_numa_kva(void)
  177. {
  178. void *vaddr;
  179. unsigned long pfn;
  180. int node;
  181. for_each_online_node(node) {
  182. for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
  183. vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
  184. set_pmd_pfn((ulong) vaddr,
  185. node_remap_start_pfn[node] + pfn,
  186. PAGE_KERNEL_LARGE);
  187. }
  188. }
  189. }
  190. static unsigned long calculate_numa_remap_pages(void)
  191. {
  192. int nid;
  193. unsigned long size, reserve_pages = 0;
  194. unsigned long pfn;
  195. for_each_online_node(nid) {
  196. /*
  197. * The acpi/srat node info can show hot-add memroy zones
  198. * where memory could be added but not currently present.
  199. */
  200. if (node_start_pfn[nid] > max_pfn)
  201. continue;
  202. if (node_end_pfn[nid] > max_pfn)
  203. node_end_pfn[nid] = max_pfn;
  204. /* ensure the remap includes space for the pgdat. */
  205. size = node_remap_size[nid] + sizeof(pg_data_t);
  206. /* convert size to large (pmd size) pages, rounding up */
  207. size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
  208. /* now the roundup is correct, convert to PAGE_SIZE pages */
  209. size = size * PTRS_PER_PTE;
  210. /*
  211. * Validate the region we are allocating only contains valid
  212. * pages.
  213. */
  214. for (pfn = node_end_pfn[nid] - size;
  215. pfn < node_end_pfn[nid]; pfn++)
  216. if (!page_is_ram(pfn))
  217. break;
  218. if (pfn != node_end_pfn[nid])
  219. size = 0;
  220. printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
  221. size, nid);
  222. node_remap_size[nid] = size;
  223. node_remap_offset[nid] = reserve_pages;
  224. reserve_pages += size;
  225. printk("Shrinking node %d from %ld pages to %ld pages\n",
  226. nid, node_end_pfn[nid], node_end_pfn[nid] - size);
  227. if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
  228. /*
  229. * Align node_end_pfn[] and node_remap_start_pfn[] to
  230. * pmd boundary. remap_numa_kva will barf otherwise.
  231. */
  232. printk("Shrinking node %d further by %ld pages for proper alignment\n",
  233. nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
  234. size += node_end_pfn[nid] & (PTRS_PER_PTE-1);
  235. }
  236. node_end_pfn[nid] -= size;
  237. node_remap_start_pfn[nid] = node_end_pfn[nid];
  238. }
  239. printk("Reserving total of %ld pages for numa KVA remap\n",
  240. reserve_pages);
  241. return reserve_pages;
  242. }
  243. extern void setup_bootmem_allocator(void);
  244. unsigned long __init setup_memory(void)
  245. {
  246. int nid;
  247. unsigned long system_start_pfn, system_max_low_pfn;
  248. unsigned long reserve_pages;
  249. /*
  250. * When mapping a NUMA machine we allocate the node_mem_map arrays
  251. * from node local memory. They are then mapped directly into KVA
  252. * between zone normal and vmalloc space. Calculate the size of
  253. * this space and use it to adjust the boundry between ZONE_NORMAL
  254. * and ZONE_HIGHMEM.
  255. */
  256. find_max_pfn();
  257. get_memcfg_numa();
  258. reserve_pages = calculate_numa_remap_pages();
  259. /* partially used pages are not usable - thus round upwards */
  260. system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
  261. system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
  262. printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
  263. reserve_pages, max_low_pfn + reserve_pages);
  264. printk("max_pfn = %ld\n", max_pfn);
  265. #ifdef CONFIG_HIGHMEM
  266. highstart_pfn = highend_pfn = max_pfn;
  267. if (max_pfn > system_max_low_pfn)
  268. highstart_pfn = system_max_low_pfn;
  269. printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
  270. pages_to_mb(highend_pfn - highstart_pfn));
  271. #endif
  272. printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
  273. pages_to_mb(system_max_low_pfn));
  274. printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
  275. min_low_pfn, max_low_pfn, highstart_pfn);
  276. printk("Low memory ends at vaddr %08lx\n",
  277. (ulong) pfn_to_kaddr(max_low_pfn));
  278. for_each_online_node(nid) {
  279. node_remap_start_vaddr[nid] = pfn_to_kaddr(
  280. highstart_pfn + node_remap_offset[nid]);
  281. /* Init the node remap allocator */
  282. node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
  283. (node_remap_size[nid] * PAGE_SIZE);
  284. node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
  285. ALIGN(sizeof(pg_data_t), PAGE_SIZE);
  286. allocate_pgdat(nid);
  287. printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
  288. (ulong) node_remap_start_vaddr[nid],
  289. (ulong) pfn_to_kaddr(highstart_pfn
  290. + node_remap_offset[nid] + node_remap_size[nid]));
  291. }
  292. printk("High memory starts at vaddr %08lx\n",
  293. (ulong) pfn_to_kaddr(highstart_pfn));
  294. vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
  295. for_each_online_node(nid)
  296. find_max_pfn_node(nid);
  297. memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
  298. NODE_DATA(0)->bdata = &node0_bdata;
  299. setup_bootmem_allocator();
  300. return max_low_pfn;
  301. }
  302. void __init zone_sizes_init(void)
  303. {
  304. int nid;
  305. /*
  306. * Insert nodes into pgdat_list backward so they appear in order.
  307. * Clobber node 0's links and NULL out pgdat_list before starting.
  308. */
  309. pgdat_list = NULL;
  310. for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) {
  311. if (!node_online(nid))
  312. continue;
  313. NODE_DATA(nid)->pgdat_next = pgdat_list;
  314. pgdat_list = NODE_DATA(nid);
  315. }
  316. for_each_online_node(nid) {
  317. unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
  318. unsigned long *zholes_size;
  319. unsigned int max_dma;
  320. unsigned long low = max_low_pfn;
  321. unsigned long start = node_start_pfn[nid];
  322. unsigned long high = node_end_pfn[nid];
  323. max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
  324. if (node_has_online_mem(nid)){
  325. if (start > low) {
  326. #ifdef CONFIG_HIGHMEM
  327. BUG_ON(start > high);
  328. zones_size[ZONE_HIGHMEM] = high - start;
  329. #endif
  330. } else {
  331. if (low < max_dma)
  332. zones_size[ZONE_DMA] = low;
  333. else {
  334. BUG_ON(max_dma > low);
  335. BUG_ON(low > high);
  336. zones_size[ZONE_DMA] = max_dma;
  337. zones_size[ZONE_NORMAL] = low - max_dma;
  338. #ifdef CONFIG_HIGHMEM
  339. zones_size[ZONE_HIGHMEM] = high - low;
  340. #endif
  341. }
  342. }
  343. }
  344. zholes_size = get_zholes_size(nid);
  345. free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
  346. zholes_size);
  347. }
  348. return;
  349. }
  350. void __init set_highmem_pages_init(int bad_ppro)
  351. {
  352. #ifdef CONFIG_HIGHMEM
  353. struct zone *zone;
  354. struct page *page;
  355. for_each_zone(zone) {
  356. unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
  357. if (!is_highmem(zone))
  358. continue;
  359. zone_start_pfn = zone->zone_start_pfn;
  360. zone_end_pfn = zone_start_pfn + zone->spanned_pages;
  361. printk("Initializing %s for node %d (%08lx:%08lx)\n",
  362. zone->name, zone->zone_pgdat->node_id,
  363. zone_start_pfn, zone_end_pfn);
  364. for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
  365. if (!pfn_valid(node_pfn))
  366. continue;
  367. page = pfn_to_page(node_pfn);
  368. one_highpage_init(page, node_pfn, bad_ppro);
  369. }
  370. }
  371. totalram_pages += totalhigh_pages;
  372. #endif
  373. }