srat.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355
  1. /*
  2. * Some of the code in this file has been gleaned from the 64 bit
  3. * discontigmem support code base.
  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. * Send feedback to Pat Gaughen <gone@us.ibm.com>
  25. */
  26. #include <linux/mm.h>
  27. #include <linux/bootmem.h>
  28. #include <linux/mmzone.h>
  29. #include <linux/acpi.h>
  30. #include <linux/nodemask.h>
  31. #include <asm/srat.h>
  32. #include <asm/topology.h>
  33. /*
  34. * proximity macros and definitions
  35. */
  36. #define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
  37. #define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
  38. #define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
  39. #define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
  40. /* bitmap length; _PXM is at most 255 */
  41. #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
  42. static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
  43. #define MAX_CHUNKS_PER_NODE 3
  44. #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
  45. struct node_memory_chunk_s {
  46. unsigned long start_pfn;
  47. unsigned long end_pfn;
  48. u8 pxm; // proximity domain of node
  49. u8 nid; // which cnode contains this chunk?
  50. u8 bank; // which mem bank on this node
  51. };
  52. static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
  53. static int num_memory_chunks; /* total number of memory chunks */
  54. extern void * boot_ioremap(unsigned long, unsigned long);
  55. /* Identify CPU proximity domains */
  56. static void __init parse_cpu_affinity_structure(char *p)
  57. {
  58. struct acpi_table_processor_affinity *cpu_affinity =
  59. (struct acpi_table_processor_affinity *) p;
  60. if (!cpu_affinity->flags.enabled)
  61. return; /* empty entry */
  62. /* mark this node as "seen" in node bitmap */
  63. BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
  64. printk("CPU 0x%02X in proximity domain 0x%02X\n",
  65. cpu_affinity->apic_id, cpu_affinity->proximity_domain);
  66. }
  67. /*
  68. * Identify memory proximity domains and hot-remove capabilities.
  69. * Fill node memory chunk list structure.
  70. */
  71. static void __init parse_memory_affinity_structure (char *sratp)
  72. {
  73. unsigned long long paddr, size;
  74. unsigned long start_pfn, end_pfn;
  75. u8 pxm;
  76. struct node_memory_chunk_s *p, *q, *pend;
  77. struct acpi_table_memory_affinity *memory_affinity =
  78. (struct acpi_table_memory_affinity *) sratp;
  79. if (!memory_affinity->flags.enabled)
  80. return; /* empty entry */
  81. /* mark this node as "seen" in node bitmap */
  82. BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
  83. /* calculate info for memory chunk structure */
  84. paddr = memory_affinity->base_addr_hi;
  85. paddr = (paddr << 32) | memory_affinity->base_addr_lo;
  86. size = memory_affinity->length_hi;
  87. size = (size << 32) | memory_affinity->length_lo;
  88. start_pfn = paddr >> PAGE_SHIFT;
  89. end_pfn = (paddr + size) >> PAGE_SHIFT;
  90. pxm = memory_affinity->proximity_domain;
  91. if (num_memory_chunks >= MAXCHUNKS) {
  92. printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
  93. size/(1024*1024), paddr);
  94. return;
  95. }
  96. /* Insertion sort based on base address */
  97. pend = &node_memory_chunk[num_memory_chunks];
  98. for (p = &node_memory_chunk[0]; p < pend; p++) {
  99. if (start_pfn < p->start_pfn)
  100. break;
  101. }
  102. if (p < pend) {
  103. for (q = pend; q >= p; q--)
  104. *(q + 1) = *q;
  105. }
  106. p->start_pfn = start_pfn;
  107. p->end_pfn = end_pfn;
  108. p->pxm = pxm;
  109. num_memory_chunks++;
  110. printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
  111. start_pfn, end_pfn,
  112. memory_affinity->memory_type,
  113. memory_affinity->proximity_domain,
  114. (memory_affinity->flags.hot_pluggable ?
  115. "enabled and removable" : "enabled" ) );
  116. }
  117. /*
  118. * The SRAT table always lists ascending addresses, so can always
  119. * assume that the first "start" address that you see is the real
  120. * start of the node, and that the current "end" address is after
  121. * the previous one.
  122. */
  123. static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
  124. {
  125. /*
  126. * Only add present memory as told by the e820.
  127. * There is no guarantee from the SRAT that the memory it
  128. * enumerates is present at boot time because it represents
  129. * *possible* memory hotplug areas the same as normal RAM.
  130. */
  131. if (memory_chunk->start_pfn >= max_pfn) {
  132. printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
  133. memory_chunk->start_pfn, memory_chunk->end_pfn);
  134. return;
  135. }
  136. if (memory_chunk->nid != nid)
  137. return;
  138. if (!node_has_online_mem(nid))
  139. node_start_pfn[nid] = memory_chunk->start_pfn;
  140. if (node_start_pfn[nid] > memory_chunk->start_pfn)
  141. node_start_pfn[nid] = memory_chunk->start_pfn;
  142. if (node_end_pfn[nid] < memory_chunk->end_pfn)
  143. node_end_pfn[nid] = memory_chunk->end_pfn;
  144. }
  145. /* Parse the ACPI Static Resource Affinity Table */
  146. static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
  147. {
  148. u8 *start, *end, *p;
  149. int i, j, nid;
  150. start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
  151. p = start;
  152. end = (u8 *)sratp + sratp->header.length;
  153. memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
  154. memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
  155. num_memory_chunks = 0;
  156. while (p < end) {
  157. switch (*p) {
  158. case ACPI_SRAT_PROCESSOR_AFFINITY:
  159. parse_cpu_affinity_structure(p);
  160. break;
  161. case ACPI_SRAT_MEMORY_AFFINITY:
  162. parse_memory_affinity_structure(p);
  163. break;
  164. default:
  165. printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
  166. break;
  167. }
  168. p += p[1];
  169. if (p[1] == 0) {
  170. printk("acpi20_parse_srat: Entry length value is zero;"
  171. " can't parse any further!\n");
  172. break;
  173. }
  174. }
  175. if (num_memory_chunks == 0) {
  176. printk("could not finy any ACPI SRAT memory areas.\n");
  177. goto out_fail;
  178. }
  179. /* Calculate total number of nodes in system from PXM bitmap and create
  180. * a set of sequential node IDs starting at zero. (ACPI doesn't seem
  181. * to specify the range of _PXM values.)
  182. */
  183. /*
  184. * MCD - we no longer HAVE to number nodes sequentially. PXM domain
  185. * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
  186. * 32, so we will continue numbering them in this manner until MAX_NUMNODES
  187. * approaches MAX_PXM_DOMAINS for i386.
  188. */
  189. nodes_clear(node_online_map);
  190. for (i = 0; i < MAX_PXM_DOMAINS; i++) {
  191. if (BMAP_TEST(pxm_bitmap, i)) {
  192. int nid = acpi_map_pxm_to_node(i);
  193. node_set_online(nid);
  194. }
  195. }
  196. BUG_ON(num_online_nodes() == 0);
  197. /* set cnode id in memory chunk structure */
  198. for (i = 0; i < num_memory_chunks; i++)
  199. node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
  200. printk("pxm bitmap: ");
  201. for (i = 0; i < sizeof(pxm_bitmap); i++) {
  202. printk("%02X ", pxm_bitmap[i]);
  203. }
  204. printk("\n");
  205. printk("Number of logical nodes in system = %d\n", num_online_nodes());
  206. printk("Number of memory chunks in system = %d\n", num_memory_chunks);
  207. for (j = 0; j < num_memory_chunks; j++){
  208. struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
  209. printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
  210. j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
  211. node_read_chunk(chunk->nid, chunk);
  212. add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn);
  213. }
  214. for_each_online_node(nid) {
  215. unsigned long start = node_start_pfn[nid];
  216. unsigned long end = node_end_pfn[nid];
  217. memory_present(nid, start, end);
  218. node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
  219. }
  220. return 1;
  221. out_fail:
  222. return 0;
  223. }
  224. int __init get_memcfg_from_srat(void)
  225. {
  226. struct acpi_table_header *header = NULL;
  227. struct acpi_table_rsdp *rsdp = NULL;
  228. struct acpi_table_rsdt *rsdt = NULL;
  229. struct acpi_pointer *rsdp_address = NULL;
  230. struct acpi_table_rsdt saved_rsdt;
  231. int tables = 0;
  232. int i = 0;
  233. if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
  234. rsdp_address))) {
  235. printk("%s: System description tables not found\n",
  236. __FUNCTION__);
  237. goto out_err;
  238. }
  239. if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
  240. printk("%s: assigning address to rsdp\n", __FUNCTION__);
  241. rsdp = (struct acpi_table_rsdp *)
  242. (u32)rsdp_address->pointer.physical;
  243. } else {
  244. printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
  245. goto out_err;
  246. }
  247. if (!rsdp) {
  248. printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
  249. goto out_err;
  250. }
  251. printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
  252. rsdp->oem_id);
  253. if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
  254. printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
  255. goto out_err;
  256. }
  257. rsdt = (struct acpi_table_rsdt *)
  258. boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
  259. if (!rsdt) {
  260. printk(KERN_WARNING
  261. "%s: ACPI: Invalid root system description tables (RSDT)\n",
  262. __FUNCTION__);
  263. goto out_err;
  264. }
  265. header = & rsdt->header;
  266. if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
  267. printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
  268. goto out_err;
  269. }
  270. /*
  271. * The number of tables is computed by taking the
  272. * size of all entries (header size minus total
  273. * size of RSDT) divided by the size of each entry
  274. * (4-byte table pointers).
  275. */
  276. tables = (header->length - sizeof(struct acpi_table_header)) / 4;
  277. if (!tables)
  278. goto out_err;
  279. memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
  280. if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
  281. printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
  282. saved_rsdt.header.length);
  283. goto out_err;
  284. }
  285. printk("Begin SRAT table scan....\n");
  286. for (i = 0; i < tables; i++) {
  287. /* Map in header, then map in full table length. */
  288. header = (struct acpi_table_header *)
  289. boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
  290. if (!header)
  291. break;
  292. header = (struct acpi_table_header *)
  293. boot_ioremap(saved_rsdt.entry[i], header->length);
  294. if (!header)
  295. break;
  296. if (strncmp((char *) &header->signature, "SRAT", 4))
  297. continue;
  298. /* we've found the srat table. don't need to look at any more tables */
  299. return acpi20_parse_srat((struct acpi_table_srat *)header);
  300. }
  301. out_err:
  302. remove_all_active_ranges();
  303. printk("failed to get NUMA memory information from SRAT table\n");
  304. return 0;
  305. }