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#include <asm/smp.h> 34 35/* 36 * proximity macros and definitions 37 */ 38#define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */ 39#define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */ 40#define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit)) 41#define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit))) 42/* bitmap length; _PXM is at most 255 */ 43#define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8) 44static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */ 45 46#define MAX_CHUNKS_PER_NODE 3 47#define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES) 48struct node_memory_chunk_s { 49 unsigned long start_pfn; 50 unsigned long end_pfn; 51 u8 pxm; // proximity domain of node 52 u8 nid; // which cnode contains this chunk? 53 u8 bank; // which mem bank on this node 54}; 55static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS]; 56 57static int num_memory_chunks; /* total number of memory chunks */ 58static u8 __initdata apicid_to_pxm[MAX_APICID]; 59 60extern void * boot_ioremap(unsigned long, unsigned long); 61 62/* Identify CPU proximity domains */ 63static void __init parse_cpu_affinity_structure(char *p) 64{ 65 struct acpi_srat_cpu_affinity *cpu_affinity = 66 (struct acpi_srat_cpu_affinity *) p; 67 68 if ((cpu_affinity->flags & ACPI_SRAT_CPU_ENABLED) == 0) 69 return; /* empty entry */ 70 71 /* mark this node as "seen" in node bitmap */ 72 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain_lo); 73 74 apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain_lo; 75 76 printk("CPU 0x%02X in proximity domain 0x%02X\n", 77 cpu_affinity->apic_id, cpu_affinity->proximity_domain_lo); 78} 79 80/* 81 * Identify memory proximity domains and hot-remove capabilities. 82 * Fill node memory chunk list structure. 83 */ 84static void __init parse_memory_affinity_structure (char *sratp) 85{ 86 unsigned long long paddr, size; 87 unsigned long start_pfn, end_pfn; 88 u8 pxm; 89 struct node_memory_chunk_s *p, *q, *pend; 90 struct acpi_srat_mem_affinity *memory_affinity = 91 (struct acpi_srat_mem_affinity *) sratp; 92 93 if ((memory_affinity->flags & ACPI_SRAT_MEM_ENABLED) == 0) 94 return; /* empty entry */ 95 96 pxm = memory_affinity->proximity_domain & 0xff; 97 98 /* mark this node as "seen" in node bitmap */ 99 BMAP_SET(pxm_bitmap, pxm); 100 101 /* calculate info for memory chunk structure */ 102 paddr = memory_affinity->base_address; 103 size = memory_affinity->length; 104 105 start_pfn = paddr >> PAGE_SHIFT; 106 end_pfn = (paddr + size) >> PAGE_SHIFT; 107 108 109 if (num_memory_chunks >= MAXCHUNKS) { 110 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n", 111 size/(1024*1024), paddr); 112 return; 113 } 114 115 /* Insertion sort based on base address */ 116 pend = &node_memory_chunk[num_memory_chunks]; 117 for (p = &node_memory_chunk[0]; p < pend; p++) { 118 if (start_pfn < p->start_pfn) 119 break; 120 } 121 if (p < pend) { 122 for (q = pend; q >= p; q--) 123 *(q + 1) = *q; 124 } 125 p->start_pfn = start_pfn; 126 p->end_pfn = end_pfn; 127 p->pxm = pxm; 128 129 num_memory_chunks++; 130 131 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n", 132 start_pfn, end_pfn, 133 memory_affinity->memory_type, 134 pxm, 135 ((memory_affinity->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ? 136 "enabled and removable" : "enabled" ) ); 137} 138 139/* 140 * The SRAT table always lists ascending addresses, so can always 141 * assume that the first "start" address that you see is the real 142 * start of the node, and that the current "end" address is after 143 * the previous one. 144 */ 145static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk) 146{ 147 /* 148 * Only add present memory as told by the e820. 149 * There is no guarantee from the SRAT that the memory it 150 * enumerates is present at boot time because it represents 151 * *possible* memory hotplug areas the same as normal RAM. 152 */ 153 if (memory_chunk->start_pfn >= max_pfn) { 154 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n", 155 memory_chunk->start_pfn, memory_chunk->end_pfn); 156 return; 157 } 158 if (memory_chunk->nid != nid) 159 return; 160 161 if (!node_has_online_mem(nid)) 162 node_start_pfn[nid] = memory_chunk->start_pfn; 163 164 if (node_start_pfn[nid] > memory_chunk->start_pfn) 165 node_start_pfn[nid] = memory_chunk->start_pfn; 166 167 if (node_end_pfn[nid] < memory_chunk->end_pfn) 168 node_end_pfn[nid] = memory_chunk->end_pfn; 169} 170 171/* Parse the ACPI Static Resource Affinity Table */ 172static int __init acpi20_parse_srat(struct acpi_table_srat *sratp) 173{ 174 u8 *start, *end, *p; 175 int i, j, nid; 176 177 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */ 178 p = start; 179 end = (u8 *)sratp + sratp->header.length; 180 181 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */ 182 memset(node_memory_chunk, 0, sizeof(node_memory_chunk)); 183 184 num_memory_chunks = 0; 185 while (p < end) { 186 switch (*p) { 187 case ACPI_SRAT_TYPE_CPU_AFFINITY: 188 parse_cpu_affinity_structure(p); 189 break; 190 case ACPI_SRAT_TYPE_MEMORY_AFFINITY: 191 parse_memory_affinity_structure(p); 192 break; 193 default: 194 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]); 195 break; 196 } 197 p += p[1]; 198 if (p[1] == 0) { 199 printk("acpi20_parse_srat: Entry length value is zero;" 200 " can't parse any further!\n"); 201 break; 202 } 203 } 204 205 if (num_memory_chunks == 0) { 206 printk("could not finy any ACPI SRAT memory areas.\n"); 207 goto out_fail; 208 } 209 210 /* Calculate total number of nodes in system from PXM bitmap and create 211 * a set of sequential node IDs starting at zero. (ACPI doesn't seem 212 * to specify the range of _PXM values.) 213 */ 214 /* 215 * MCD - we no longer HAVE to number nodes sequentially. PXM domain 216 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically 217 * 32, so we will continue numbering them in this manner until MAX_NUMNODES 218 * approaches MAX_PXM_DOMAINS for i386. 219 */ 220 nodes_clear(node_online_map); 221 for (i = 0; i < MAX_PXM_DOMAINS; i++) { 222 if (BMAP_TEST(pxm_bitmap, i)) { 223 int nid = acpi_map_pxm_to_node(i); 224 node_set_online(nid); 225 } 226 } 227 BUG_ON(num_online_nodes() == 0); 228 229 /* set cnode id in memory chunk structure */ 230 for (i = 0; i < num_memory_chunks; i++) 231 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm); 232 233 printk("pxm bitmap: "); 234 for (i = 0; i < sizeof(pxm_bitmap); i++) { 235 printk("%02X ", pxm_bitmap[i]); 236 } 237 printk("\n"); 238 printk("Number of logical nodes in system = %d\n", num_online_nodes()); 239 printk("Number of memory chunks in system = %d\n", num_memory_chunks); 240 241 for (i = 0; i < MAX_APICID; i++) 242 apicid_2_node[i] = pxm_to_node(apicid_to_pxm[i]); 243 244 for (j = 0; j < num_memory_chunks; j++){ 245 struct node_memory_chunk_s * chunk = &node_memory_chunk[j]; 246 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n", 247 j, chunk->nid, chunk->start_pfn, chunk->end_pfn); 248 node_read_chunk(chunk->nid, chunk); 249 add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn); 250 } 251 252 for_each_online_node(nid) { 253 unsigned long start = node_start_pfn[nid]; 254 unsigned long end = node_end_pfn[nid]; 255 256 memory_present(nid, start, end); 257 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end); 258 } 259 return 1; 260out_fail: 261 return 0; 262} 263 264struct acpi_static_rsdt { 265 struct acpi_table_rsdt table; 266 u32 padding[7]; /* Allow for 7 more table entries */ 267}; 268 269int __init get_memcfg_from_srat(void) 270{ 271 struct acpi_table_header *header = NULL; 272 struct acpi_table_rsdp *rsdp = NULL; 273 struct acpi_table_rsdt *rsdt = NULL; 274 acpi_native_uint rsdp_address = 0; 275 struct acpi_static_rsdt saved_rsdt; 276 int tables = 0; 277 int i = 0; 278 279 rsdp_address = acpi_find_rsdp(); 280 if (!rsdp_address) { 281 printk("%s: System description tables not found\n", 282 __FUNCTION__); 283 goto out_err; 284 } 285 286 printk("%s: assigning address to rsdp\n", __FUNCTION__); 287 rsdp = (struct acpi_table_rsdp *)(u32)rsdp_address; 288 if (!rsdp) { 289 printk("%s: Didn't find ACPI root!\n", __FUNCTION__); 290 goto out_err; 291 } 292 293 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision, 294 rsdp->oem_id); 295 296 if (strncmp(rsdp->signature, ACPI_SIG_RSDP,strlen(ACPI_SIG_RSDP))) { 297 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__); 298 goto out_err; 299 } 300 301 rsdt = (struct acpi_table_rsdt *) 302 boot_ioremap(rsdp->rsdt_physical_address, sizeof(struct acpi_table_rsdt)); 303 304 if (!rsdt) { 305 printk(KERN_WARNING 306 "%s: ACPI: Invalid root system description tables (RSDT)\n", 307 __FUNCTION__); 308 goto out_err; 309 } 310 311 header = &rsdt->header; 312 313 if (strncmp(header->signature, ACPI_SIG_RSDT, strlen(ACPI_SIG_RSDT))) { 314 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n"); 315 goto out_err; 316 } 317 318 /* 319 * The number of tables is computed by taking the 320 * size of all entries (header size minus total 321 * size of RSDT) divided by the size of each entry 322 * (4-byte table pointers). 323 */ 324 tables = (header->length - sizeof(struct acpi_table_header)) / 4; 325 326 if (!tables) 327 goto out_err; 328 329 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt)); 330 331 if (saved_rsdt.table.header.length > sizeof(saved_rsdt)) { 332 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n", 333 saved_rsdt.table.header.length); 334 goto out_err; 335 } 336 337 printk("Begin SRAT table scan....\n"); 338 339 for (i = 0; i < tables; i++) { 340 /* Map in header, then map in full table length. */ 341 header = (struct acpi_table_header *) 342 boot_ioremap(saved_rsdt.table.table_offset_entry[i], sizeof(struct acpi_table_header)); 343 if (!header) 344 break; 345 header = (struct acpi_table_header *) 346 boot_ioremap(saved_rsdt.table.table_offset_entry[i], header->length); 347 if (!header) 348 break; 349 350 if (strncmp((char *) &header->signature, ACPI_SIG_SRAT, 4)) 351 continue; 352 353 /* we've found the srat table. don't need to look at any more tables */ 354 return acpi20_parse_srat((struct acpi_table_srat *)header); 355 } 356out_err: 357 remove_all_active_ranges(); 358 printk("failed to get NUMA memory information from SRAT table\n"); 359 return 0; 360} 361