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