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/numa.h>
31 #include <linux/nodemask.h>
32 #include <linux/topology.h>
34 static nodemask_t nodes_found_map = NODE_MASK_NONE;
36 /* maps to convert between proximity domain and logical node ID */
37 static int pxm_to_node_map[MAX_PXM_DOMAINS]
38 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
39 static int node_to_pxm_map[MAX_NUMNODES]
40 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
42 unsigned char acpi_srat_revision __initdata;
44 int pxm_to_node(int pxm)
48 return pxm_to_node_map[pxm];
51 int node_to_pxm(int node)
55 return node_to_pxm_map[node];
58 static void __acpi_map_pxm_to_node(int pxm, int node)
60 if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
61 pxm_to_node_map[pxm] = node;
62 if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
63 node_to_pxm_map[node] = pxm;
66 int acpi_map_pxm_to_node(int pxm)
70 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS)
73 node = pxm_to_node_map[pxm];
75 if (node == NUMA_NO_NODE) {
76 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
78 node = first_unset_node(nodes_found_map);
79 __acpi_map_pxm_to_node(pxm, node);
80 node_set(node, nodes_found_map);
87 * acpi_map_pxm_to_online_node - Map proximity ID to online node
88 * @pxm: ACPI proximity ID
90 * This is similar to acpi_map_pxm_to_node(), but always returns an online
91 * node. When the mapped node from a given proximity ID is offline, it
92 * looks up the node distance table and returns the nearest online node.
94 * ACPI device drivers, which are called after the NUMA initialization has
95 * completed in the kernel, can call this interface to obtain their device
96 * NUMA topology from ACPI tables. Such drivers do not have to deal with
97 * offline nodes. A node may be offline when a device proximity ID is
98 * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
101 int acpi_map_pxm_to_online_node(int pxm)
103 int node, n, dist, min_dist;
105 node = acpi_map_pxm_to_node(pxm);
107 if (node == NUMA_NO_NODE)
110 if (!node_online(node)) {
112 for_each_online_node(n) {
113 dist = node_distance(node, n);
114 if (dist < min_dist) {
123 EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
126 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 static int __init acpi_parse_slit(struct acpi_table_header *table)
205 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
207 if (!slit_valid(slit)) {
208 pr_info("SLIT table looks invalid. Not used.\n");
211 acpi_numa_slit_init(slit);
217 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
219 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
223 acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
224 const unsigned long end)
226 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
228 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
229 if (!processor_affinity)
232 acpi_table_print_srat_entry(header);
234 /* let architecture-dependent part to do it */
235 acpi_numa_x2apic_affinity_init(processor_affinity);
241 acpi_parse_processor_affinity(struct acpi_subtable_header *header,
242 const unsigned long end)
244 struct acpi_srat_cpu_affinity *processor_affinity;
246 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
247 if (!processor_affinity)
250 acpi_table_print_srat_entry(header);
252 /* let architecture-dependent part to do it */
253 acpi_numa_processor_affinity_init(processor_affinity);
258 static int __initdata parsed_numa_memblks;
261 acpi_parse_memory_affinity(struct acpi_subtable_header * header,
262 const unsigned long end)
264 struct acpi_srat_mem_affinity *memory_affinity;
266 memory_affinity = (struct acpi_srat_mem_affinity *)header;
267 if (!memory_affinity)
270 acpi_table_print_srat_entry(header);
272 /* let architecture-dependent part to do it */
273 if (!acpi_numa_memory_affinity_init(memory_affinity))
274 parsed_numa_memblks++;
278 static int __init acpi_parse_srat(struct acpi_table_header *table)
280 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
282 acpi_srat_revision = srat->header.revision;
284 /* Real work done in acpi_table_parse_srat below. */
290 acpi_table_parse_srat(enum acpi_srat_type id,
291 acpi_tbl_entry_handler handler, unsigned int max_entries)
293 return acpi_table_parse_entries(ACPI_SIG_SRAT,
294 sizeof(struct acpi_table_srat), id,
295 handler, max_entries);
298 int __init acpi_numa_init(void)
303 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
304 * SRAT cpu entries could have different order with that in MADT.
305 * So go over all cpu entries in SRAT to get apicid to node mapping.
308 /* SRAT: Static Resource Affinity Table */
309 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
310 struct acpi_subtable_proc srat_proc[2];
312 memset(srat_proc, 0, sizeof(srat_proc));
313 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
314 srat_proc[0].handler = acpi_parse_processor_affinity;
315 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
316 srat_proc[1].handler = acpi_parse_x2apic_affinity;
318 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
319 sizeof(struct acpi_table_srat),
320 srat_proc, ARRAY_SIZE(srat_proc), 0);
322 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
323 acpi_parse_memory_affinity,
327 /* SLIT: System Locality Information Table */
328 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
330 acpi_numa_arch_fixup();
334 else if (!parsed_numa_memblks)
339 static int acpi_get_pxm(acpi_handle h)
341 unsigned long long pxm;
344 acpi_handle phandle = h;
348 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
349 if (ACPI_SUCCESS(status))
351 status = acpi_get_parent(handle, &phandle);
352 } while (ACPI_SUCCESS(status));
356 int acpi_get_node(acpi_handle handle)
360 pxm = acpi_get_pxm(handle);
362 return acpi_map_pxm_to_node(pxm);
364 EXPORT_SYMBOL(acpi_get_node);