1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * acpi_numa.c - ACPI NUMA support
5 * Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
8 #define pr_fmt(fmt) "ACPI: " fmt
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>
21 static nodemask_t nodes_found_map = NODE_MASK_NONE;
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 };
29 unsigned char acpi_srat_revision __initdata;
30 static int acpi_numa __initdata;
32 void __init disable_srat(void)
37 int pxm_to_node(int pxm)
39 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
41 return pxm_to_node_map[pxm];
43 EXPORT_SYMBOL(pxm_to_node);
45 int node_to_pxm(int node)
49 return node_to_pxm_map[node];
52 static void __acpi_map_pxm_to_node(int pxm, int node)
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;
60 int acpi_map_pxm_to_node(int pxm)
64 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
67 node = pxm_to_node_map[pxm];
69 if (node == NUMA_NO_NODE) {
70 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
72 node = first_unset_node(nodes_found_map);
73 __acpi_map_pxm_to_node(pxm, node);
74 node_set(node, nodes_found_map);
79 EXPORT_SYMBOL(acpi_map_pxm_to_node);
82 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
84 switch (header->type) {
85 case ACPI_SRAT_TYPE_CPU_AFFINITY:
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");
97 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
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,
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" : "");
114 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
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",
121 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
122 "enabled" : "disabled");
126 case ACPI_SRAT_TYPE_GICC_AFFINITY:
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,
133 (p->flags & ACPI_SRAT_GICC_ENABLED) ?
134 "enabled" : "disabled");
138 case ACPI_SRAT_TYPE_GENERIC_AFFINITY:
140 struct acpi_srat_generic_affinity *p =
141 (struct acpi_srat_generic_affinity *)header;
143 if (p->device_handle_type == 0) {
145 * For pci devices this may be the only place they
146 * are assigned a proximity domain
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]),
152 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
153 "enabled" : "disabled");
156 * In this case we can rely on the device having a
157 * proximity domain reference
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]),
163 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
164 "enabled" : "disabled");
169 pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
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.
181 static int __init slit_valid(struct acpi_table_slit *slit)
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];
189 if (val != LOCAL_DISTANCE)
191 } else if (val <= LOCAL_DISTANCE)
198 void __init bad_srat(void)
200 pr_err("SRAT: SRAT not used.\n");
204 int __init srat_disabled(void)
206 return acpi_numa < 0;
209 #if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
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.
215 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
219 for (i = 0; i < slit->locality_count; i++) {
220 const int from_node = pxm_to_node(i);
222 if (from_node == NUMA_NO_NODE)
225 for (j = 0; j < slit->locality_count; j++) {
226 const int to_node = pxm_to_node(j);
228 if (to_node == NUMA_NO_NODE)
231 numa_set_distance(from_node, to_node,
232 slit->entry[slit->locality_count * i + j]);
238 * Default callback for parsing of the Proximity Domain <-> Memory
242 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
250 if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
251 pr_err("SRAT: Unexpected header length: %d\n",
253 goto out_err_bad_srat;
255 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
257 hotpluggable = IS_ENABLED(CONFIG_MEMORY_HOTPLUG) &&
258 (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE);
260 start = ma->base_address;
261 end = start + ma->length;
262 pxm = ma->proximity_domain;
263 if (acpi_srat_revision <= 1)
266 node = acpi_map_pxm_to_node(pxm);
267 if (node == NUMA_NO_NODE) {
268 pr_err("SRAT: Too many proximity domains.\n");
269 goto out_err_bad_srat;
272 if (numa_add_memblk(node, start, end) < 0) {
273 pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
274 node, (unsigned long long) start,
275 (unsigned long long) end - 1);
276 goto out_err_bad_srat;
279 node_set(node, numa_nodes_parsed);
281 pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
283 (unsigned long long) start, (unsigned long long) end - 1,
284 hotpluggable ? " hotplug" : "",
285 ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
287 /* Mark hotplug range in memblock. */
288 if (hotpluggable && memblock_mark_hotplug(start, ma->length))
289 pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
290 (unsigned long long)start, (unsigned long long)end - 1);
292 max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
301 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
302 void *arg, const unsigned long table_end)
304 struct acpi_cedt_cfmws *cfmws;
309 cfmws = (struct acpi_cedt_cfmws *)header;
310 start = cfmws->base_hpa;
311 end = cfmws->base_hpa + cfmws->window_size;
313 /* Skip if the SRAT already described the NUMA details for this HPA */
314 node = phys_to_target_node(start);
315 if (node != NUMA_NO_NODE)
318 node = acpi_map_pxm_to_node(*fake_pxm);
320 if (node == NUMA_NO_NODE) {
321 pr_err("ACPI NUMA: Too many proximity domains while processing CFMWS.\n");
325 if (numa_add_memblk(node, start, end) < 0) {
326 /* CXL driver must handle the NUMA_NO_NODE case */
327 pr_warn("ACPI NUMA: Failed to add memblk for CFMWS node %d [mem %#llx-%#llx]\n",
331 /* Set the next available fake_pxm value */
336 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
337 void *arg, const unsigned long table_end)
341 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
343 static int __init acpi_parse_slit(struct acpi_table_header *table)
345 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
347 if (!slit_valid(slit)) {
348 pr_info("SLIT table looks invalid. Not used.\n");
351 acpi_numa_slit_init(slit);
357 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
359 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
363 acpi_parse_x2apic_affinity(union acpi_subtable_headers *header,
364 const unsigned long end)
366 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
368 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
370 acpi_table_print_srat_entry(&header->common);
372 /* let architecture-dependent part to do it */
373 acpi_numa_x2apic_affinity_init(processor_affinity);
379 acpi_parse_processor_affinity(union acpi_subtable_headers *header,
380 const unsigned long end)
382 struct acpi_srat_cpu_affinity *processor_affinity;
384 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
386 acpi_table_print_srat_entry(&header->common);
388 /* let architecture-dependent part to do it */
389 acpi_numa_processor_affinity_init(processor_affinity);
395 acpi_parse_gicc_affinity(union acpi_subtable_headers *header,
396 const unsigned long end)
398 struct acpi_srat_gicc_affinity *processor_affinity;
400 processor_affinity = (struct acpi_srat_gicc_affinity *)header;
402 acpi_table_print_srat_entry(&header->common);
404 /* let architecture-dependent part to do it */
405 acpi_numa_gicc_affinity_init(processor_affinity);
410 #if defined(CONFIG_X86) || defined(CONFIG_ARM64)
412 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
413 const unsigned long end)
415 struct acpi_srat_generic_affinity *gi_affinity;
418 gi_affinity = (struct acpi_srat_generic_affinity *)header;
421 acpi_table_print_srat_entry(&header->common);
423 if (!(gi_affinity->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED))
426 node = acpi_map_pxm_to_node(gi_affinity->proximity_domain);
427 if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
428 pr_err("SRAT: Too many proximity domains.\n");
431 node_set(node, numa_nodes_parsed);
432 node_set_state(node, N_GENERIC_INITIATOR);
438 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
439 const unsigned long end)
443 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
445 static int __initdata parsed_numa_memblks;
448 acpi_parse_memory_affinity(union acpi_subtable_headers * header,
449 const unsigned long end)
451 struct acpi_srat_mem_affinity *memory_affinity;
453 memory_affinity = (struct acpi_srat_mem_affinity *)header;
455 acpi_table_print_srat_entry(&header->common);
457 /* let architecture-dependent part to do it */
458 if (!acpi_numa_memory_affinity_init(memory_affinity))
459 parsed_numa_memblks++;
463 static int __init acpi_parse_srat(struct acpi_table_header *table)
465 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
467 acpi_srat_revision = srat->header.revision;
469 /* Real work done in acpi_table_parse_srat below. */
475 acpi_table_parse_srat(enum acpi_srat_type id,
476 acpi_tbl_entry_handler handler, unsigned int max_entries)
478 return acpi_table_parse_entries(ACPI_SIG_SRAT,
479 sizeof(struct acpi_table_srat), id,
480 handler, max_entries);
483 int __init acpi_numa_init(void)
485 int i, fake_pxm, cnt = 0;
491 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
492 * SRAT cpu entries could have different order with that in MADT.
493 * So go over all cpu entries in SRAT to get apicid to node mapping.
496 /* SRAT: System Resource Affinity Table */
497 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
498 struct acpi_subtable_proc srat_proc[4];
500 memset(srat_proc, 0, sizeof(srat_proc));
501 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
502 srat_proc[0].handler = acpi_parse_processor_affinity;
503 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
504 srat_proc[1].handler = acpi_parse_x2apic_affinity;
505 srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
506 srat_proc[2].handler = acpi_parse_gicc_affinity;
507 srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
508 srat_proc[3].handler = acpi_parse_gi_affinity;
510 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
511 sizeof(struct acpi_table_srat),
512 srat_proc, ARRAY_SIZE(srat_proc), 0);
514 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
515 acpi_parse_memory_affinity, 0);
518 /* SLIT: System Locality Information Table */
519 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
522 * CXL Fixed Memory Window Structures (CFMWS) must be parsed
523 * after the SRAT. Create NUMA Nodes for CXL memory ranges that
524 * are defined in the CFMWS and not already defined in the SRAT.
525 * Initialize a fake_pxm as the first available PXM to emulate.
528 /* fake_pxm is the next unused PXM value after SRAT parsing */
529 for (i = 0, fake_pxm = -1; i < MAX_NUMNODES - 1; i++) {
530 if (node_to_pxm_map[i] > fake_pxm)
531 fake_pxm = node_to_pxm_map[i];
534 acpi_table_parse_cedt(ACPI_CEDT_TYPE_CFMWS, acpi_parse_cfmws,
539 else if (!parsed_numa_memblks)
544 static int acpi_get_pxm(acpi_handle h)
546 unsigned long long pxm;
549 acpi_handle phandle = h;
553 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
554 if (ACPI_SUCCESS(status))
556 status = acpi_get_parent(handle, &phandle);
557 } while (ACPI_SUCCESS(status));
561 int acpi_get_node(acpi_handle handle)
565 pxm = acpi_get_pxm(handle);
567 return pxm_to_node(pxm);
569 EXPORT_SYMBOL(acpi_get_node);