linux/drivers/acpi/numa/srat.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  acpi_numa.c - ACPI NUMA support
   4 *
   5 *  Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
   6 */
   7
   8#define pr_fmt(fmt) "ACPI: " fmt
   9
  10#include <linux/module.h>
  11#include <linux/init.h>
  12#include <linux/kernel.h>
  13#include <linux/types.h>
  14#include <linux/errno.h>
  15#include <linux/acpi.h>
  16#include <linux/memblock.h>
  17#include <linux/numa.h>
  18#include <linux/nodemask.h>
  19#include <linux/topology.h>
  20
  21static nodemask_t nodes_found_map = NODE_MASK_NONE;
  22
  23/* maps to convert between proximity domain and logical node ID */
  24static int pxm_to_node_map[MAX_PXM_DOMAINS]
  25                        = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
  26static int node_to_pxm_map[MAX_NUMNODES]
  27                        = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
  28
  29unsigned char acpi_srat_revision __initdata;
  30static int acpi_numa __initdata;
  31
  32void __init disable_srat(void)
  33{
  34        acpi_numa = -1;
  35}
  36
  37int pxm_to_node(int pxm)
  38{
  39        if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
  40                return NUMA_NO_NODE;
  41        return pxm_to_node_map[pxm];
  42}
  43EXPORT_SYMBOL(pxm_to_node);
  44
  45int node_to_pxm(int node)
  46{
  47        if (node < 0)
  48                return PXM_INVAL;
  49        return node_to_pxm_map[node];
  50}
  51
  52static void __acpi_map_pxm_to_node(int pxm, int node)
  53{
  54        if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
  55                pxm_to_node_map[pxm] = node;
  56        if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
  57                node_to_pxm_map[node] = pxm;
  58}
  59
  60int acpi_map_pxm_to_node(int pxm)
  61{
  62        int node;
  63
  64        if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
  65                return NUMA_NO_NODE;
  66
  67        node = pxm_to_node_map[pxm];
  68
  69        if (node == NUMA_NO_NODE) {
  70                if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
  71                        return NUMA_NO_NODE;
  72                node = first_unset_node(nodes_found_map);
  73                __acpi_map_pxm_to_node(pxm, node);
  74                node_set(node, nodes_found_map);
  75        }
  76
  77        return node;
  78}
  79EXPORT_SYMBOL(acpi_map_pxm_to_node);
  80
  81static void __init
  82acpi_table_print_srat_entry(struct acpi_subtable_header *header)
  83{
  84        switch (header->type) {
  85        case ACPI_SRAT_TYPE_CPU_AFFINITY:
  86                {
  87                        struct acpi_srat_cpu_affinity *p =
  88                            (struct acpi_srat_cpu_affinity *)header;
  89                        pr_debug("SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
  90                                 p->apic_id, p->local_sapic_eid,
  91                                 p->proximity_domain_lo,
  92                                 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
  93                                 "enabled" : "disabled");
  94                }
  95                break;
  96
  97        case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
  98                {
  99                        struct acpi_srat_mem_affinity *p =
 100                            (struct acpi_srat_mem_affinity *)header;
 101                        pr_debug("SRAT Memory (0x%llx length 0x%llx) in proximity domain %d %s%s%s\n",
 102                                 (unsigned long long)p->base_address,
 103                                 (unsigned long long)p->length,
 104                                 p->proximity_domain,
 105                                 (p->flags & ACPI_SRAT_MEM_ENABLED) ?
 106                                 "enabled" : "disabled",
 107                                 (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
 108                                 " hot-pluggable" : "",
 109                                 (p->flags & ACPI_SRAT_MEM_NON_VOLATILE) ?
 110                                 " non-volatile" : "");
 111                }
 112                break;
 113
 114        case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
 115                {
 116                        struct acpi_srat_x2apic_cpu_affinity *p =
 117                            (struct acpi_srat_x2apic_cpu_affinity *)header;
 118                        pr_debug("SRAT Processor (x2apicid[0x%08x]) in proximity domain %d %s\n",
 119                                 p->apic_id,
 120                                 p->proximity_domain,
 121                                 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
 122                                 "enabled" : "disabled");
 123                }
 124                break;
 125
 126        case ACPI_SRAT_TYPE_GICC_AFFINITY:
 127                {
 128                        struct acpi_srat_gicc_affinity *p =
 129                            (struct acpi_srat_gicc_affinity *)header;
 130                        pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
 131                                 p->acpi_processor_uid,
 132                                 p->proximity_domain,
 133                                 (p->flags & ACPI_SRAT_GICC_ENABLED) ?
 134                                 "enabled" : "disabled");
 135                }
 136                break;
 137
 138        case ACPI_SRAT_TYPE_GENERIC_AFFINITY:
 139        {
 140                struct acpi_srat_generic_affinity *p =
 141                        (struct acpi_srat_generic_affinity *)header;
 142
 143                if (p->device_handle_type == 0) {
 144                        /*
 145                         * For pci devices this may be the only place they
 146                         * are assigned a proximity domain
 147                         */
 148                        pr_debug("SRAT Generic Initiator(Seg:%u BDF:%u) in proximity domain %d %s\n",
 149                                 *(u16 *)(&p->device_handle[0]),
 150                                 *(u16 *)(&p->device_handle[2]),
 151                                 p->proximity_domain,
 152                                 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
 153                                "enabled" : "disabled");
 154                } else {
 155                        /*
 156                         * In this case we can rely on the device having a
 157                         * proximity domain reference
 158                         */
 159                        pr_debug("SRAT Generic Initiator(HID=%.8s UID=%.4s) in proximity domain %d %s\n",
 160                                (char *)(&p->device_handle[0]),
 161                                (char *)(&p->device_handle[8]),
 162                                p->proximity_domain,
 163                                (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
 164                                "enabled" : "disabled");
 165                }
 166        }
 167        break;
 168        default:
 169                pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
 170                        header->type);
 171                break;
 172        }
 173}
 174
 175/*
 176 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
 177 * up the NUMA heuristics which wants the local node to have a smaller
 178 * distance than the others.
 179 * Do some quick checks here and only use the SLIT if it passes.
 180 */
 181static int __init slit_valid(struct acpi_table_slit *slit)
 182{
 183        int i, j;
 184        int d = slit->locality_count;
 185        for (i = 0; i < d; i++) {
 186                for (j = 0; j < d; j++)  {
 187                        u8 val = slit->entry[d*i + j];
 188                        if (i == j) {
 189                                if (val != LOCAL_DISTANCE)
 190                                        return 0;
 191                        } else if (val <= LOCAL_DISTANCE)
 192                                return 0;
 193                }
 194        }
 195        return 1;
 196}
 197
 198void __init bad_srat(void)
 199{
 200        pr_err("SRAT: SRAT not used.\n");
 201        disable_srat();
 202}
 203
 204int __init srat_disabled(void)
 205{
 206        return acpi_numa < 0;
 207}
 208
 209#if defined(CONFIG_X86) || defined(CONFIG_ARM64)
 210/*
 211 * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
 212 * I/O localities since SRAT does not list them.  I/O localities are
 213 * not supported at this point.
 214 */
 215void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
 216{
 217        int i, j;
 218
 219        for (i = 0; i < slit->locality_count; i++) {
 220                const int from_node = pxm_to_node(i);
 221
 222                if (from_node == NUMA_NO_NODE)
 223                        continue;
 224
 225                for (j = 0; j < slit->locality_count; j++) {
 226                        const int to_node = pxm_to_node(j);
 227
 228                        if (to_node == NUMA_NO_NODE)
 229                                continue;
 230
 231                        numa_set_distance(from_node, to_node,
 232                                slit->entry[slit->locality_count * i + j]);
 233                }
 234        }
 235}
 236
 237/*
 238 * Default callback for parsing of the Proximity Domain <-> Memory
 239 * Area mappings
 240 */
 241int __init
 242acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
 243{
 244        u64 start, end;
 245        u32 hotpluggable;
 246        int node, pxm;
 247
 248        if (srat_disabled())
 249                goto out_err;
 250        if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
 251                pr_err("SRAT: Unexpected header length: %d\n",
 252                       ma->header.length);
 253                goto out_err_bad_srat;
 254        }
 255        if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
 256                goto out_err;
 257        hotpluggable = ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE;
 258        if (hotpluggable && !IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
 259                goto out_err;
 260
 261        start = ma->base_address;
 262        end = start + ma->length;
 263        pxm = ma->proximity_domain;
 264        if (acpi_srat_revision <= 1)
 265                pxm &= 0xff;
 266
 267        node = acpi_map_pxm_to_node(pxm);
 268        if (node == NUMA_NO_NODE) {
 269                pr_err("SRAT: Too many proximity domains.\n");
 270                goto out_err_bad_srat;
 271        }
 272
 273        if (numa_add_memblk(node, start, end) < 0) {
 274                pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
 275                       node, (unsigned long long) start,
 276                       (unsigned long long) end - 1);
 277                goto out_err_bad_srat;
 278        }
 279
 280        node_set(node, numa_nodes_parsed);
 281
 282        pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
 283                node, pxm,
 284                (unsigned long long) start, (unsigned long long) end - 1,
 285                hotpluggable ? " hotplug" : "",
 286                ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
 287
 288        /* Mark hotplug range in memblock. */
 289        if (hotpluggable && memblock_mark_hotplug(start, ma->length))
 290                pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
 291                        (unsigned long long)start, (unsigned long long)end - 1);
 292
 293        max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
 294
 295        return 0;
 296out_err_bad_srat:
 297        bad_srat();
 298out_err:
 299        return -EINVAL;
 300}
 301#endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
 302
 303static int __init acpi_parse_slit(struct acpi_table_header *table)
 304{
 305        struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
 306
 307        if (!slit_valid(slit)) {
 308                pr_info("SLIT table looks invalid. Not used.\n");
 309                return -EINVAL;
 310        }
 311        acpi_numa_slit_init(slit);
 312
 313        return 0;
 314}
 315
 316void __init __weak
 317acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
 318{
 319        pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
 320}
 321
 322static int __init
 323acpi_parse_x2apic_affinity(union acpi_subtable_headers *header,
 324                           const unsigned long end)
 325{
 326        struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
 327
 328        processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
 329
 330        acpi_table_print_srat_entry(&header->common);
 331
 332        /* let architecture-dependent part to do it */
 333        acpi_numa_x2apic_affinity_init(processor_affinity);
 334
 335        return 0;
 336}
 337
 338static int __init
 339acpi_parse_processor_affinity(union acpi_subtable_headers *header,
 340                              const unsigned long end)
 341{
 342        struct acpi_srat_cpu_affinity *processor_affinity;
 343
 344        processor_affinity = (struct acpi_srat_cpu_affinity *)header;
 345
 346        acpi_table_print_srat_entry(&header->common);
 347
 348        /* let architecture-dependent part to do it */
 349        acpi_numa_processor_affinity_init(processor_affinity);
 350
 351        return 0;
 352}
 353
 354static int __init
 355acpi_parse_gicc_affinity(union acpi_subtable_headers *header,
 356                         const unsigned long end)
 357{
 358        struct acpi_srat_gicc_affinity *processor_affinity;
 359
 360        processor_affinity = (struct acpi_srat_gicc_affinity *)header;
 361
 362        acpi_table_print_srat_entry(&header->common);
 363
 364        /* let architecture-dependent part to do it */
 365        acpi_numa_gicc_affinity_init(processor_affinity);
 366
 367        return 0;
 368}
 369
 370#if defined(CONFIG_X86) || defined(CONFIG_ARM64)
 371static int __init
 372acpi_parse_gi_affinity(union acpi_subtable_headers *header,
 373                       const unsigned long end)
 374{
 375        struct acpi_srat_generic_affinity *gi_affinity;
 376        int node;
 377
 378        gi_affinity = (struct acpi_srat_generic_affinity *)header;
 379        if (!gi_affinity)
 380                return -EINVAL;
 381        acpi_table_print_srat_entry(&header->common);
 382
 383        if (!(gi_affinity->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED))
 384                return -EINVAL;
 385
 386        node = acpi_map_pxm_to_node(gi_affinity->proximity_domain);
 387        if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
 388                pr_err("SRAT: Too many proximity domains.\n");
 389                return -EINVAL;
 390        }
 391        node_set(node, numa_nodes_parsed);
 392        node_set_state(node, N_GENERIC_INITIATOR);
 393
 394        return 0;
 395}
 396#else
 397static int __init
 398acpi_parse_gi_affinity(union acpi_subtable_headers *header,
 399                       const unsigned long end)
 400{
 401        return 0;
 402}
 403#endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
 404
 405static int __initdata parsed_numa_memblks;
 406
 407static int __init
 408acpi_parse_memory_affinity(union acpi_subtable_headers * header,
 409                           const unsigned long end)
 410{
 411        struct acpi_srat_mem_affinity *memory_affinity;
 412
 413        memory_affinity = (struct acpi_srat_mem_affinity *)header;
 414
 415        acpi_table_print_srat_entry(&header->common);
 416
 417        /* let architecture-dependent part to do it */
 418        if (!acpi_numa_memory_affinity_init(memory_affinity))
 419                parsed_numa_memblks++;
 420        return 0;
 421}
 422
 423static int __init acpi_parse_srat(struct acpi_table_header *table)
 424{
 425        struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
 426
 427        acpi_srat_revision = srat->header.revision;
 428
 429        /* Real work done in acpi_table_parse_srat below. */
 430
 431        return 0;
 432}
 433
 434static int __init
 435acpi_table_parse_srat(enum acpi_srat_type id,
 436                      acpi_tbl_entry_handler handler, unsigned int max_entries)
 437{
 438        return acpi_table_parse_entries(ACPI_SIG_SRAT,
 439                                            sizeof(struct acpi_table_srat), id,
 440                                            handler, max_entries);
 441}
 442
 443int __init acpi_numa_init(void)
 444{
 445        int cnt = 0;
 446
 447        if (acpi_disabled)
 448                return -EINVAL;
 449
 450        /*
 451         * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
 452         * SRAT cpu entries could have different order with that in MADT.
 453         * So go over all cpu entries in SRAT to get apicid to node mapping.
 454         */
 455
 456        /* SRAT: System Resource Affinity Table */
 457        if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
 458                struct acpi_subtable_proc srat_proc[4];
 459
 460                memset(srat_proc, 0, sizeof(srat_proc));
 461                srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
 462                srat_proc[0].handler = acpi_parse_processor_affinity;
 463                srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
 464                srat_proc[1].handler = acpi_parse_x2apic_affinity;
 465                srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
 466                srat_proc[2].handler = acpi_parse_gicc_affinity;
 467                srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
 468                srat_proc[3].handler = acpi_parse_gi_affinity;
 469
 470                acpi_table_parse_entries_array(ACPI_SIG_SRAT,
 471                                        sizeof(struct acpi_table_srat),
 472                                        srat_proc, ARRAY_SIZE(srat_proc), 0);
 473
 474                cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
 475                                            acpi_parse_memory_affinity, 0);
 476        }
 477
 478        /* SLIT: System Locality Information Table */
 479        acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
 480
 481        if (cnt < 0)
 482                return cnt;
 483        else if (!parsed_numa_memblks)
 484                return -ENOENT;
 485        return 0;
 486}
 487
 488static int acpi_get_pxm(acpi_handle h)
 489{
 490        unsigned long long pxm;
 491        acpi_status status;
 492        acpi_handle handle;
 493        acpi_handle phandle = h;
 494
 495        do {
 496                handle = phandle;
 497                status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
 498                if (ACPI_SUCCESS(status))
 499                        return pxm;
 500                status = acpi_get_parent(handle, &phandle);
 501        } while (ACPI_SUCCESS(status));
 502        return -1;
 503}
 504
 505int acpi_get_node(acpi_handle handle)
 506{
 507        int pxm;
 508
 509        pxm = acpi_get_pxm(handle);
 510
 511        return pxm_to_node(pxm);
 512}
 513EXPORT_SYMBOL(acpi_get_node);
 514