drivers/acpi/numa.c

   1 /*
   2  *  acpi_numa.c - ACPI NUMA support
   3  *
   4  *  Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
   5  *
   6  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   7  *
   8  *  This program is free software; you can redistribute it and/or modify
   9  *  it under the terms of the GNU General Public License as published by
  10  *  the Free Software Foundation; either version 2 of the License, or
  11  *  (at your option) any later version.
  12  *
  13  *  This program is distributed in the hope that it will be useful,
  14  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *  GNU General Public License for more details.
  17  *
  18  *  You should have received a copy of the GNU General Public License
  19  *  along with this program; if not, write to the Free Software
  20  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  *
  22  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  23  *
  24  */
  25 #include <linux/module.h>
  26 #include <linux/init.h>
  27 #include <linux/kernel.h>
  28 #include <linux/types.h>
  29 #include <linux/errno.h>
  30 #include <linux/acpi.h>
  31 #include <linux/numa.h>
  32 #include <linux/nodemask.h>
  33 #include <linux/topology.h>
  34
  35 #define PREFIX "ACPI: "
  36
  37 #define ACPI_NUMA       0x80000000
  38 #define _COMPONENT      ACPI_NUMA
  39 ACPI_MODULE_NAME("numa");
  40
  41 static nodemask_t nodes_found_map = NODE_MASK_NONE;
  42
  43 /* maps to convert between proximity domain and logical node ID */
  44 static int pxm_to_node_map[MAX_PXM_DOMAINS]
  45                         = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
  46 static int node_to_pxm_map[MAX_NUMNODES]
  47                         = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
  48
  49 unsigned char acpi_srat_revision __initdata;
  50
  51 int pxm_to_node(int pxm)
  52 {
  53         if (pxm < 0)
  54                 return NUMA_NO_NODE;
  55         return pxm_to_node_map[pxm];
  56 }
  57
  58 int node_to_pxm(int node)
  59 {
  60         if (node < 0)
  61                 return PXM_INVAL;
  62         return node_to_pxm_map[node];
  63 }
  64
  65 static void __acpi_map_pxm_to_node(int pxm, int node)
  66 {
  67         if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
  68                 pxm_to_node_map[pxm] = node;
  69         if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
  70                 node_to_pxm_map[node] = pxm;
  71 }
  72
  73 int acpi_map_pxm_to_node(int pxm)
  74 {
  75         int node;
  76
  77         if (pxm < 0 || pxm >= MAX_PXM_DOMAINS)
  78                 return NUMA_NO_NODE;
  79
  80         node = pxm_to_node_map[pxm];
  81
  82         if (node == NUMA_NO_NODE) {
  83                 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
  84                         return NUMA_NO_NODE;
  85                 node = first_unset_node(nodes_found_map);
  86                 __acpi_map_pxm_to_node(pxm, node);
  87                 node_set(node, nodes_found_map);
  88         }
  89
  90         return node;
  91 }
  92
  93 /**
  94  * acpi_map_pxm_to_online_node - Map proximity ID to online node
  95  * @pxm: ACPI proximity ID
  96  *
  97  * This is similar to acpi_map_pxm_to_node(), but always returns an online
  98  * node.  When the mapped node from a given proximity ID is offline, it
  99  * looks up the node distance table and returns the nearest online node.
 100  *
 101  * ACPI device drivers, which are called after the NUMA initialization has
 102  * completed in the kernel, can call this interface to obtain their device
 103  * NUMA topology from ACPI tables.  Such drivers do not have to deal with
 104  * offline nodes.  A node may be offline when a device proximity ID is
 105  * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
 106  * "numa=off" on x86.
 107  */
 108 int acpi_map_pxm_to_online_node(int pxm)
 109 {
 110         int node, n, dist, min_dist;
 111
 112         node = acpi_map_pxm_to_node(pxm);
 113
 114         if (node == NUMA_NO_NODE)
 115                 node = 0;
 116
 117         if (!node_online(node)) {
 118                 min_dist = INT_MAX;
 119                 for_each_online_node(n) {
 120                         dist = node_distance(node, n);
 121                         if (dist < min_dist) {
 122                                 min_dist = dist;
 123                                 node = n;
 124                         }
 125                 }
 126         }
 127
 128         return node;
 129 }
 130 EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
 131
 132 static void __init
 133 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
 134 {
 135
 136         ACPI_FUNCTION_NAME("acpi_table_print_srat_entry");
 137
 138         if (!header)
 139                 return;
 140
 141         switch (header->type) {
 142
 143         case ACPI_SRAT_TYPE_CPU_AFFINITY:
 144 #ifdef ACPI_DEBUG_OUTPUT
 145                 {
 146                         struct acpi_srat_cpu_affinity *p =
 147                             (struct acpi_srat_cpu_affinity *)header;
 148                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 149                                           "SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
 150                                           p->apic_id, p->local_sapic_eid,
 151                                           p->proximity_domain_lo,
 152                                           (p->flags & ACPI_SRAT_CPU_ENABLED)?
 153                                           "enabled" : "disabled"));
 154                 }
 155 #endif                          /* ACPI_DEBUG_OUTPUT */
 156                 break;
 157
 158         case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
 159 #ifdef ACPI_DEBUG_OUTPUT
 160                 {
 161                         struct acpi_srat_mem_affinity *p =
 162                             (struct acpi_srat_mem_affinity *)header;
 163                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 164                                           "SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
 165                                           (unsigned long)p->base_address,
 166                                           (unsigned long)p->length,
 167                                           p->proximity_domain,
 168                                           (p->flags & ACPI_SRAT_MEM_ENABLED)?
 169                                           "enabled" : "disabled",
 170                                           (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE)?
 171                                           " hot-pluggable" : "",
 172                                           (p->flags & ACPI_SRAT_MEM_NON_VOLATILE)?
 173                                           " non-volatile" : ""));
 174                 }
 175 #endif                          /* ACPI_DEBUG_OUTPUT */
 176                 break;
 177
 178         case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
 179 #ifdef ACPI_DEBUG_OUTPUT
 180                 {
 181                         struct acpi_srat_x2apic_cpu_affinity *p =
 182                             (struct acpi_srat_x2apic_cpu_affinity *)header;
 183                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 184                                           "SRAT Processor (x2apicid[0x%08x]) in"
 185                                           " proximity domain %d %s\n",
 186                                           p->apic_id,
 187                                           p->proximity_domain,
 188                                           (p->flags & ACPI_SRAT_CPU_ENABLED) ?
 189                                           "enabled" : "disabled"));
 190                 }
 191 #endif                          /* ACPI_DEBUG_OUTPUT */
 192                 break;
 193         default:
 194                 printk(KERN_WARNING PREFIX
 195                        "Found unsupported SRAT entry (type = 0x%x)\n",
 196                        header->type);
 197                 break;
 198         }
 199 }
 200
 201 /*
 202  * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
 203  * up the NUMA heuristics which wants the local node to have a smaller
 204  * distance than the others.
 205  * Do some quick checks here and only use the SLIT if it passes.
 206  */
 207 static int __init slit_valid(struct acpi_table_slit *slit)
 208 {
 209         int i, j;
 210         int d = slit->locality_count;
 211         for (i = 0; i < d; i++) {
 212                 for (j = 0; j < d; j++)  {
 213                         u8 val = slit->entry[d*i + j];
 214                         if (i == j) {
 215                                 if (val != LOCAL_DISTANCE)
 216                                         return 0;
 217                         } else if (val <= LOCAL_DISTANCE)
 218                                 return 0;
 219                 }
 220         }
 221         return 1;
 222 }
 223
 224 static int __init acpi_parse_slit(struct acpi_table_header *table)
 225 {
 226         struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
 227
 228         if (!slit_valid(slit)) {
 229                 printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
 230                 return -EINVAL;
 231         }
 232         acpi_numa_slit_init(slit);
 233
 234         return 0;
 235 }
 236
 237 void __init __weak
 238 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
 239 {
 240         printk(KERN_WARNING PREFIX
 241                "Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
 242         return;
 243 }
 244
 245
 246 static int __init
 247 acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
 248                            const unsigned long end)
 249 {
 250         struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
 251
 252         processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
 253         if (!processor_affinity)
 254                 return -EINVAL;
 255
 256         acpi_table_print_srat_entry(header);
 257
 258         /* let architecture-dependent part to do it */
 259         acpi_numa_x2apic_affinity_init(processor_affinity);
 260
 261         return 0;
 262 }
 263
 264 static int __init
 265 acpi_parse_processor_affinity(struct acpi_subtable_header *header,
 266                               const unsigned long end)
 267 {
 268         struct acpi_srat_cpu_affinity *processor_affinity;
 269
 270         processor_affinity = (struct acpi_srat_cpu_affinity *)header;
 271         if (!processor_affinity)
 272                 return -EINVAL;
 273
 274         acpi_table_print_srat_entry(header);
 275
 276         /* let architecture-dependent part to do it */
 277         acpi_numa_processor_affinity_init(processor_affinity);
 278
 279         return 0;
 280 }
 281
 282 static int __initdata parsed_numa_memblks;
 283
 284 static int __init
 285 acpi_parse_memory_affinity(struct acpi_subtable_header * header,
 286                            const unsigned long end)
 287 {
 288         struct acpi_srat_mem_affinity *memory_affinity;
 289
 290         memory_affinity = (struct acpi_srat_mem_affinity *)header;
 291         if (!memory_affinity)
 292                 return -EINVAL;
 293
 294         acpi_table_print_srat_entry(header);
 295
 296         /* let architecture-dependent part to do it */
 297         if (!acpi_numa_memory_affinity_init(memory_affinity))
 298                 parsed_numa_memblks++;
 299         return 0;
 300 }
 301
 302 static int __init acpi_parse_srat(struct acpi_table_header *table)
 303 {
 304         struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
 305
 306         acpi_srat_revision = srat->header.revision;
 307
 308         /* Real work done in acpi_table_parse_srat below. */
 309
 310         return 0;
 311 }
 312
 313 static int __init
 314 acpi_table_parse_srat(enum acpi_srat_type id,
 315                       acpi_tbl_entry_handler handler, unsigned int max_entries)
 316 {
 317         return acpi_table_parse_entries(ACPI_SIG_SRAT,
 318                                             sizeof(struct acpi_table_srat), id,
 319                                             handler, max_entries);
 320 }
 321
 322 int __init acpi_numa_init(void)
 323 {
 324         int cnt = 0;
 325
 326         /*
 327          * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
 328          * SRAT cpu entries could have different order with that in MADT.
 329          * So go over all cpu entries in SRAT to get apicid to node mapping.
 330          */
 331
 332         /* SRAT: Static Resource Affinity Table */
 333         if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
 334                 acpi_table_parse_srat(ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY,
 335                                      acpi_parse_x2apic_affinity, 0);
 336                 acpi_table_parse_srat(ACPI_SRAT_TYPE_CPU_AFFINITY,
 337                                      acpi_parse_processor_affinity, 0);
 338                 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
 339                                             acpi_parse_memory_affinity,
 340                                             NR_NODE_MEMBLKS);
 341         }
 342
 343         /* SLIT: System Locality Information Table */
 344         acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
 345
 346         acpi_numa_arch_fixup();
 347
 348         if (cnt < 0)
 349                 return cnt;
 350         else if (!parsed_numa_memblks)
 351                 return -ENOENT;
 352         return 0;
 353 }
 354
 355 static int acpi_get_pxm(acpi_handle h)
 356 {
 357         unsigned long long pxm;
 358         acpi_status status;
 359         acpi_handle handle;
 360         acpi_handle phandle = h;
 361
 362         do {
 363                 handle = phandle;
 364                 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
 365                 if (ACPI_SUCCESS(status))
 366                         return pxm;
 367                 status = acpi_get_parent(handle, &phandle);
 368         } while (ACPI_SUCCESS(status));
 369         return -1;
 370 }
 371
 372 int acpi_get_node(acpi_handle handle)
 373 {
 374         int pxm;
 375
 376         pxm = acpi_get_pxm(handle);
 377
 378         return acpi_map_pxm_to_node(pxm);
 379 }
 380 EXPORT_SYMBOL(acpi_get_node);