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