2 * acpi_numa.c - ACPI NUMA support
4 * Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
6 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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.
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.
18 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
22 #define pr_fmt(fmt) "ACPI: " fmt
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/bootmem.h>
31 #include <linux/memblock.h>
32 #include <linux/numa.h>
33 #include <linux/nodemask.h>
34 #include <linux/topology.h>
36 static nodemask_t nodes_found_map = NODE_MASK_NONE;
38 /* maps to convert between proximity domain and logical node ID */
39 static int pxm_to_node_map[MAX_PXM_DOMAINS]
40 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
41 static int node_to_pxm_map[MAX_NUMNODES]
42 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
44 unsigned char acpi_srat_revision __initdata;
45 int acpi_numa __initdata;
47 int pxm_to_node(int pxm)
51 return pxm_to_node_map[pxm];
54 int node_to_pxm(int node)
58 return node_to_pxm_map[node];
61 static void __acpi_map_pxm_to_node(int pxm, int node)
63 if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
64 pxm_to_node_map[pxm] = node;
65 if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
66 node_to_pxm_map[node] = pxm;
69 int acpi_map_pxm_to_node(int pxm)
73 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS)
76 node = pxm_to_node_map[pxm];
78 if (node == NUMA_NO_NODE) {
79 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
81 node = first_unset_node(nodes_found_map);
82 __acpi_map_pxm_to_node(pxm, node);
83 node_set(node, nodes_found_map);
90 * acpi_map_pxm_to_online_node - Map proximity ID to online node
91 * @pxm: ACPI proximity ID
93 * This is similar to acpi_map_pxm_to_node(), but always returns an online
94 * node. When the mapped node from a given proximity ID is offline, it
95 * looks up the node distance table and returns the nearest online node.
97 * ACPI device drivers, which are called after the NUMA initialization has
98 * completed in the kernel, can call this interface to obtain their device
99 * NUMA topology from ACPI tables. Such drivers do not have to deal with
100 * offline nodes. A node may be offline when a device proximity ID is
101 * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
104 int acpi_map_pxm_to_online_node(int pxm)
106 int node, n, dist, min_dist;
108 node = acpi_map_pxm_to_node(pxm);
110 if (node == NUMA_NO_NODE)
113 if (!node_online(node)) {
115 for_each_online_node(n) {
116 dist = node_distance(node, n);
117 if (dist < min_dist) {
126 EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
129 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
131 switch (header->type) {
132 case ACPI_SRAT_TYPE_CPU_AFFINITY:
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");
144 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
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,
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" : "");
161 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
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",
168 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
169 "enabled" : "disabled");
174 pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
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.
186 static int __init slit_valid(struct acpi_table_slit *slit)
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];
194 if (val != LOCAL_DISTANCE)
196 } else if (val <= LOCAL_DISTANCE)
203 void __init bad_srat(void)
205 pr_err("SRAT: SRAT not used.\n");
209 int __init srat_disabled(void)
211 return acpi_numa < 0;
214 #if defined(CONFIG_X86) || defined(CONFIG_ARM64)
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.
220 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
224 for (i = 0; i < slit->locality_count; i++) {
225 const int from_node = pxm_to_node(i);
227 if (from_node == NUMA_NO_NODE)
230 for (j = 0; j < slit->locality_count; j++) {
231 const int to_node = pxm_to_node(j);
233 if (to_node == NUMA_NO_NODE)
236 numa_set_distance(from_node, to_node,
237 slit->entry[slit->locality_count * i + j]);
243 * Default callback for parsing of the Proximity Domain <-> Memory
247 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
255 if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
256 pr_err("SRAT: Unexpected header length: %d\n",
258 goto out_err_bad_srat;
260 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
262 hotpluggable = ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE;
263 if (hotpluggable && !IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
266 start = ma->base_address;
267 end = start + ma->length;
268 pxm = ma->proximity_domain;
269 if (acpi_srat_revision <= 1)
272 node = acpi_map_pxm_to_node(pxm);
273 if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
274 pr_err("SRAT: Too many proximity domains.\n");
275 goto out_err_bad_srat;
278 if (numa_add_memblk(node, start, end) < 0) {
279 pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
280 node, (unsigned long long) start,
281 (unsigned long long) end - 1);
282 goto out_err_bad_srat;
285 node_set(node, numa_nodes_parsed);
287 pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
289 (unsigned long long) start, (unsigned long long) end - 1,
290 hotpluggable ? " hotplug" : "",
291 ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
293 /* Mark hotplug range in memblock. */
294 if (hotpluggable && memblock_mark_hotplug(start, ma->length))
295 pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
296 (unsigned long long)start, (unsigned long long)end - 1);
298 max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
306 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
308 static int __init acpi_parse_slit(struct acpi_table_header *table)
310 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
312 if (!slit_valid(slit)) {
313 pr_info("SLIT table looks invalid. Not used.\n");
316 acpi_numa_slit_init(slit);
322 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
324 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
328 acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
329 const unsigned long end)
331 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
333 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
334 if (!processor_affinity)
337 acpi_table_print_srat_entry(header);
339 /* let architecture-dependent part to do it */
340 acpi_numa_x2apic_affinity_init(processor_affinity);
346 acpi_parse_processor_affinity(struct acpi_subtable_header *header,
347 const unsigned long end)
349 struct acpi_srat_cpu_affinity *processor_affinity;
351 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
352 if (!processor_affinity)
355 acpi_table_print_srat_entry(header);
357 /* let architecture-dependent part to do it */
358 acpi_numa_processor_affinity_init(processor_affinity);
363 static int __initdata parsed_numa_memblks;
366 acpi_parse_memory_affinity(struct acpi_subtable_header * header,
367 const unsigned long end)
369 struct acpi_srat_mem_affinity *memory_affinity;
371 memory_affinity = (struct acpi_srat_mem_affinity *)header;
372 if (!memory_affinity)
375 acpi_table_print_srat_entry(header);
377 /* let architecture-dependent part to do it */
378 if (!acpi_numa_memory_affinity_init(memory_affinity))
379 parsed_numa_memblks++;
383 static int __init acpi_parse_srat(struct acpi_table_header *table)
385 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
387 acpi_srat_revision = srat->header.revision;
389 /* Real work done in acpi_table_parse_srat below. */
395 acpi_table_parse_srat(enum acpi_srat_type id,
396 acpi_tbl_entry_handler handler, unsigned int max_entries)
398 return acpi_table_parse_entries(ACPI_SIG_SRAT,
399 sizeof(struct acpi_table_srat), id,
400 handler, max_entries);
403 int __init acpi_numa_init(void)
408 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
409 * SRAT cpu entries could have different order with that in MADT.
410 * So go over all cpu entries in SRAT to get apicid to node mapping.
413 /* SRAT: Static Resource Affinity Table */
414 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
415 struct acpi_subtable_proc srat_proc[2];
417 memset(srat_proc, 0, sizeof(srat_proc));
418 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
419 srat_proc[0].handler = acpi_parse_processor_affinity;
420 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
421 srat_proc[1].handler = acpi_parse_x2apic_affinity;
423 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
424 sizeof(struct acpi_table_srat),
425 srat_proc, ARRAY_SIZE(srat_proc), 0);
427 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
428 acpi_parse_memory_affinity,
432 /* SLIT: System Locality Information Table */
433 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
437 else if (!parsed_numa_memblks)
442 static int acpi_get_pxm(acpi_handle h)
444 unsigned long long pxm;
447 acpi_handle phandle = h;
451 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
452 if (ACPI_SUCCESS(status))
454 status = acpi_get_parent(handle, &phandle);
455 } while (ACPI_SUCCESS(status));
459 int acpi_get_node(acpi_handle handle)
463 pxm = acpi_get_pxm(handle);
465 return acpi_map_pxm_to_node(pxm);
467 EXPORT_SYMBOL(acpi_get_node);