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 21 static nodemask_t nodes_found_map = NODE_MASK_NONE; 22 23 /* maps to convert between proximity domain and logical node ID */ 24 static int pxm_to_node_map[MAX_PXM_DOMAINS] 25 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE }; 26 static int node_to_pxm_map[MAX_NUMNODES] 27 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL }; 28 29 unsigned char acpi_srat_revision __initdata; 30 static int acpi_numa __initdata; 31 32 void __init disable_srat(void) 33 { 34 acpi_numa = -1; 35 } 36 37 int 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 } 43 EXPORT_SYMBOL(pxm_to_node); 44 45 int node_to_pxm(int node) 46 { 47 if (node < 0) 48 return PXM_INVAL; 49 return node_to_pxm_map[node]; 50 } 51 52 static 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 60 int 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 } 79 EXPORT_SYMBOL(acpi_map_pxm_to_node); 80 81 static void __init 82 acpi_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 */ 181 static 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 } 197 198 void __init bad_srat(void) 199 { 200 pr_err("SRAT: SRAT not used.\n"); 201 disable_srat(); 202 } 203 204 int __init srat_disabled(void) 205 { 206 return acpi_numa < 0; 207 } 208 209 __weak int __init numa_fill_memblks(u64 start, u64 end) 210 { 211 return NUMA_NO_MEMBLK; 212 } 213 214 #if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH) 215 /* 216 * Callback for SLIT parsing. pxm_to_node() returns NUMA_NO_NODE for 217 * I/O localities since SRAT does not list them. I/O localities are 218 * not supported at this point. 219 */ 220 void __init acpi_numa_slit_init(struct acpi_table_slit *slit) 221 { 222 int i, j; 223 224 for (i = 0; i < slit->locality_count; i++) { 225 const int from_node = pxm_to_node(i); 226 227 if (from_node == NUMA_NO_NODE) 228 continue; 229 230 for (j = 0; j < slit->locality_count; j++) { 231 const int to_node = pxm_to_node(j); 232 233 if (to_node == NUMA_NO_NODE) 234 continue; 235 236 numa_set_distance(from_node, to_node, 237 slit->entry[slit->locality_count * i + j]); 238 } 239 } 240 } 241 242 /* 243 * Default callback for parsing of the Proximity Domain <-> Memory 244 * Area mappings 245 */ 246 int __init 247 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma) 248 { 249 u64 start, end; 250 u32 hotpluggable; 251 int node, pxm; 252 253 if (srat_disabled()) 254 goto out_err; 255 if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) { 256 pr_err("SRAT: Unexpected header length: %d\n", 257 ma->header.length); 258 goto out_err_bad_srat; 259 } 260 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0) 261 goto out_err; 262 hotpluggable = IS_ENABLED(CONFIG_MEMORY_HOTPLUG) && 263 (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE); 264 265 start = ma->base_address; 266 end = start + ma->length; 267 pxm = ma->proximity_domain; 268 if (acpi_srat_revision <= 1) 269 pxm &= 0xff; 270 271 node = acpi_map_pxm_to_node(pxm); 272 if (node == NUMA_NO_NODE) { 273 pr_err("SRAT: Too many proximity domains.\n"); 274 goto out_err_bad_srat; 275 } 276 277 if (numa_add_memblk(node, start, end) < 0) { 278 pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n", 279 node, (unsigned long long) start, 280 (unsigned long long) end - 1); 281 goto out_err_bad_srat; 282 } 283 284 node_set(node, numa_nodes_parsed); 285 286 pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n", 287 node, pxm, 288 (unsigned long long) start, (unsigned long long) end - 1, 289 hotpluggable ? " hotplug" : "", 290 ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : ""); 291 292 /* Mark hotplug range in memblock. */ 293 if (hotpluggable && memblock_mark_hotplug(start, ma->length)) 294 pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n", 295 (unsigned long long)start, (unsigned long long)end - 1); 296 297 max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1)); 298 299 return 0; 300 out_err_bad_srat: 301 bad_srat(); 302 out_err: 303 return -EINVAL; 304 } 305 306 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header, 307 void *arg, const unsigned long table_end) 308 { 309 struct acpi_cedt_cfmws *cfmws; 310 int *fake_pxm = arg; 311 u64 start, end; 312 int node; 313 314 cfmws = (struct acpi_cedt_cfmws *)header; 315 start = cfmws->base_hpa; 316 end = cfmws->base_hpa + cfmws->window_size; 317 318 /* 319 * The SRAT may have already described NUMA details for all, 320 * or a portion of, this CFMWS HPA range. Extend the memblks 321 * found for any portion of the window to cover the entire 322 * window. 323 */ 324 if (!numa_fill_memblks(start, end)) 325 return 0; 326 327 /* No SRAT description. Create a new node. */ 328 node = acpi_map_pxm_to_node(*fake_pxm); 329 330 if (node == NUMA_NO_NODE) { 331 pr_err("ACPI NUMA: Too many proximity domains while processing CFMWS.\n"); 332 return -EINVAL; 333 } 334 335 if (numa_add_memblk(node, start, end) < 0) { 336 /* CXL driver must handle the NUMA_NO_NODE case */ 337 pr_warn("ACPI NUMA: Failed to add memblk for CFMWS node %d [mem %#llx-%#llx]\n", 338 node, start, end); 339 } 340 node_set(node, numa_nodes_parsed); 341 342 /* Set the next available fake_pxm value */ 343 (*fake_pxm)++; 344 return 0; 345 } 346 #else 347 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header, 348 void *arg, const unsigned long table_end) 349 { 350 return 0; 351 } 352 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */ 353 354 static int __init acpi_parse_slit(struct acpi_table_header *table) 355 { 356 struct acpi_table_slit *slit = (struct acpi_table_slit *)table; 357 358 if (!slit_valid(slit)) { 359 pr_info("SLIT table looks invalid. Not used.\n"); 360 return -EINVAL; 361 } 362 acpi_numa_slit_init(slit); 363 364 return 0; 365 } 366 367 void __init __weak 368 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa) 369 { 370 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id); 371 } 372 373 static int __init 374 acpi_parse_x2apic_affinity(union acpi_subtable_headers *header, 375 const unsigned long end) 376 { 377 struct acpi_srat_x2apic_cpu_affinity *processor_affinity; 378 379 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header; 380 381 acpi_table_print_srat_entry(&header->common); 382 383 /* let architecture-dependent part to do it */ 384 acpi_numa_x2apic_affinity_init(processor_affinity); 385 386 return 0; 387 } 388 389 static int __init 390 acpi_parse_processor_affinity(union acpi_subtable_headers *header, 391 const unsigned long end) 392 { 393 struct acpi_srat_cpu_affinity *processor_affinity; 394 395 processor_affinity = (struct acpi_srat_cpu_affinity *)header; 396 397 acpi_table_print_srat_entry(&header->common); 398 399 /* let architecture-dependent part to do it */ 400 acpi_numa_processor_affinity_init(processor_affinity); 401 402 return 0; 403 } 404 405 static int __init 406 acpi_parse_gicc_affinity(union acpi_subtable_headers *header, 407 const unsigned long end) 408 { 409 struct acpi_srat_gicc_affinity *processor_affinity; 410 411 processor_affinity = (struct acpi_srat_gicc_affinity *)header; 412 413 acpi_table_print_srat_entry(&header->common); 414 415 /* let architecture-dependent part to do it */ 416 acpi_numa_gicc_affinity_init(processor_affinity); 417 418 return 0; 419 } 420 421 #if defined(CONFIG_X86) || defined(CONFIG_ARM64) 422 static int __init 423 acpi_parse_gi_affinity(union acpi_subtable_headers *header, 424 const unsigned long end) 425 { 426 struct acpi_srat_generic_affinity *gi_affinity; 427 int node; 428 429 gi_affinity = (struct acpi_srat_generic_affinity *)header; 430 if (!gi_affinity) 431 return -EINVAL; 432 acpi_table_print_srat_entry(&header->common); 433 434 if (!(gi_affinity->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED)) 435 return -EINVAL; 436 437 node = acpi_map_pxm_to_node(gi_affinity->proximity_domain); 438 if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) { 439 pr_err("SRAT: Too many proximity domains.\n"); 440 return -EINVAL; 441 } 442 node_set(node, numa_nodes_parsed); 443 node_set_state(node, N_GENERIC_INITIATOR); 444 445 return 0; 446 } 447 #else 448 static int __init 449 acpi_parse_gi_affinity(union acpi_subtable_headers *header, 450 const unsigned long end) 451 { 452 return 0; 453 } 454 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */ 455 456 static int __initdata parsed_numa_memblks; 457 458 static int __init 459 acpi_parse_memory_affinity(union acpi_subtable_headers * header, 460 const unsigned long end) 461 { 462 struct acpi_srat_mem_affinity *memory_affinity; 463 464 memory_affinity = (struct acpi_srat_mem_affinity *)header; 465 466 acpi_table_print_srat_entry(&header->common); 467 468 /* let architecture-dependent part to do it */ 469 if (!acpi_numa_memory_affinity_init(memory_affinity)) 470 parsed_numa_memblks++; 471 return 0; 472 } 473 474 static int __init acpi_parse_srat(struct acpi_table_header *table) 475 { 476 struct acpi_table_srat *srat = (struct acpi_table_srat *)table; 477 478 acpi_srat_revision = srat->header.revision; 479 480 /* Real work done in acpi_table_parse_srat below. */ 481 482 return 0; 483 } 484 485 static int __init 486 acpi_table_parse_srat(enum acpi_srat_type id, 487 acpi_tbl_entry_handler handler, unsigned int max_entries) 488 { 489 return acpi_table_parse_entries(ACPI_SIG_SRAT, 490 sizeof(struct acpi_table_srat), id, 491 handler, max_entries); 492 } 493 494 int __init acpi_numa_init(void) 495 { 496 int i, fake_pxm, cnt = 0; 497 498 if (acpi_disabled) 499 return -EINVAL; 500 501 /* 502 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus= 503 * SRAT cpu entries could have different order with that in MADT. 504 * So go over all cpu entries in SRAT to get apicid to node mapping. 505 */ 506 507 /* SRAT: System Resource Affinity Table */ 508 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) { 509 struct acpi_subtable_proc srat_proc[4]; 510 511 memset(srat_proc, 0, sizeof(srat_proc)); 512 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY; 513 srat_proc[0].handler = acpi_parse_processor_affinity; 514 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY; 515 srat_proc[1].handler = acpi_parse_x2apic_affinity; 516 srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY; 517 srat_proc[2].handler = acpi_parse_gicc_affinity; 518 srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY; 519 srat_proc[3].handler = acpi_parse_gi_affinity; 520 521 acpi_table_parse_entries_array(ACPI_SIG_SRAT, 522 sizeof(struct acpi_table_srat), 523 srat_proc, ARRAY_SIZE(srat_proc), 0); 524 525 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY, 526 acpi_parse_memory_affinity, 0); 527 } 528 529 /* SLIT: System Locality Information Table */ 530 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit); 531 532 /* 533 * CXL Fixed Memory Window Structures (CFMWS) must be parsed 534 * after the SRAT. Create NUMA Nodes for CXL memory ranges that 535 * are defined in the CFMWS and not already defined in the SRAT. 536 * Initialize a fake_pxm as the first available PXM to emulate. 537 */ 538 539 /* fake_pxm is the next unused PXM value after SRAT parsing */ 540 for (i = 0, fake_pxm = -1; i < MAX_NUMNODES; i++) { 541 if (node_to_pxm_map[i] > fake_pxm) 542 fake_pxm = node_to_pxm_map[i]; 543 } 544 fake_pxm++; 545 acpi_table_parse_cedt(ACPI_CEDT_TYPE_CFMWS, acpi_parse_cfmws, 546 &fake_pxm); 547 548 if (cnt < 0) 549 return cnt; 550 else if (!parsed_numa_memblks) 551 return -ENOENT; 552 return 0; 553 } 554 555 static int acpi_get_pxm(acpi_handle h) 556 { 557 unsigned long long pxm; 558 acpi_status status; 559 acpi_handle handle; 560 acpi_handle phandle = h; 561 562 do { 563 handle = phandle; 564 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm); 565 if (ACPI_SUCCESS(status)) 566 return pxm; 567 status = acpi_get_parent(handle, &phandle); 568 } while (ACPI_SUCCESS(status)); 569 return -1; 570 } 571 572 int acpi_get_node(acpi_handle handle) 573 { 574 int pxm; 575 576 pxm = acpi_get_pxm(handle); 577 578 return pxm_to_node(pxm); 579 } 580 EXPORT_SYMBOL(acpi_get_node); 581