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)
128 switch (header->type) {
129 case ACPI_SRAT_TYPE_CPU_AFFINITY:
131 struct acpi_srat_cpu_affinity *p =
132 (struct acpi_srat_cpu_affinity *)header;
133 pr_debug("SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
134 p->apic_id, p->local_sapic_eid,
135 p->proximity_domain_lo,
136 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
137 "enabled" : "disabled");
141 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
143 struct acpi_srat_mem_affinity *p =
144 (struct acpi_srat_mem_affinity *)header;
145 pr_debug("SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
146 (unsigned long)p->base_address,
147 (unsigned long)p->length,
149 (p->flags & ACPI_SRAT_MEM_ENABLED) ?
150 "enabled" : "disabled",
151 (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
152 " hot-pluggable" : "",
153 (p->flags & ACPI_SRAT_MEM_NON_VOLATILE) ?
154 " non-volatile" : "");
158 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
160 struct acpi_srat_x2apic_cpu_affinity *p =
161 (struct acpi_srat_x2apic_cpu_affinity *)header;
162 pr_debug("SRAT Processor (x2apicid[0x%08x]) in proximity domain %d %s\n",
165 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
166 "enabled" : "disabled");
171 pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
178 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
179 * up the NUMA heuristics which wants the local node to have a smaller
180 * distance than the others.
181 * Do some quick checks here and only use the SLIT if it passes.
183 static int __init slit_valid(struct acpi_table_slit *slit)
186 int d = slit->locality_count;
187 for (i = 0; i < d; i++) {
188 for (j = 0; j < d; j++) {
189 u8 val = slit->entry[d*i + j];
191 if (val != LOCAL_DISTANCE)
193 } else if (val <= LOCAL_DISTANCE)
200 #if defined(CONFIG_X86) || defined(CONFIG_ARM64)
202 * Callback for SLIT parsing. pxm_to_node() returns NUMA_NO_NODE for
203 * I/O localities since SRAT does not list them. I/O localities are
204 * not supported at this point.
206 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
210 for (i = 0; i < slit->locality_count; i++) {
211 const int from_node = pxm_to_node(i);
213 if (from_node == NUMA_NO_NODE)
216 for (j = 0; j < slit->locality_count; j++) {
217 const int to_node = pxm_to_node(j);
219 if (to_node == NUMA_NO_NODE)
222 numa_set_distance(from_node, to_node,
223 slit->entry[slit->locality_count * i + j]);
227 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
229 static int __init acpi_parse_slit(struct acpi_table_header *table)
231 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
233 if (!slit_valid(slit)) {
234 pr_info("SLIT table looks invalid. Not used.\n");
237 acpi_numa_slit_init(slit);
243 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
245 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
249 acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
250 const unsigned long end)
252 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
254 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
255 if (!processor_affinity)
258 acpi_table_print_srat_entry(header);
260 /* let architecture-dependent part to do it */
261 acpi_numa_x2apic_affinity_init(processor_affinity);
267 acpi_parse_processor_affinity(struct acpi_subtable_header *header,
268 const unsigned long end)
270 struct acpi_srat_cpu_affinity *processor_affinity;
272 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
273 if (!processor_affinity)
276 acpi_table_print_srat_entry(header);
278 /* let architecture-dependent part to do it */
279 acpi_numa_processor_affinity_init(processor_affinity);
284 static int __initdata parsed_numa_memblks;
287 acpi_parse_memory_affinity(struct acpi_subtable_header * header,
288 const unsigned long end)
290 struct acpi_srat_mem_affinity *memory_affinity;
292 memory_affinity = (struct acpi_srat_mem_affinity *)header;
293 if (!memory_affinity)
296 acpi_table_print_srat_entry(header);
298 /* let architecture-dependent part to do it */
299 if (!acpi_numa_memory_affinity_init(memory_affinity))
300 parsed_numa_memblks++;
304 static int __init acpi_parse_srat(struct acpi_table_header *table)
306 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
308 acpi_srat_revision = srat->header.revision;
310 /* Real work done in acpi_table_parse_srat below. */
316 acpi_table_parse_srat(enum acpi_srat_type id,
317 acpi_tbl_entry_handler handler, unsigned int max_entries)
319 return acpi_table_parse_entries(ACPI_SIG_SRAT,
320 sizeof(struct acpi_table_srat), id,
321 handler, max_entries);
324 int __init acpi_numa_init(void)
329 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
330 * SRAT cpu entries could have different order with that in MADT.
331 * So go over all cpu entries in SRAT to get apicid to node mapping.
334 /* SRAT: Static Resource Affinity Table */
335 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
336 struct acpi_subtable_proc srat_proc[2];
338 memset(srat_proc, 0, sizeof(srat_proc));
339 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
340 srat_proc[0].handler = acpi_parse_processor_affinity;
341 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
342 srat_proc[1].handler = acpi_parse_x2apic_affinity;
344 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
345 sizeof(struct acpi_table_srat),
346 srat_proc, ARRAY_SIZE(srat_proc), 0);
348 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
349 acpi_parse_memory_affinity,
353 /* SLIT: System Locality Information Table */
354 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
358 else if (!parsed_numa_memblks)
363 static int acpi_get_pxm(acpi_handle h)
365 unsigned long long pxm;
368 acpi_handle phandle = h;
372 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
373 if (ACPI_SUCCESS(status))
375 status = acpi_get_parent(handle, &phandle);
376 } while (ACPI_SUCCESS(status));
380 int acpi_get_node(acpi_handle handle)
384 pxm = acpi_get_pxm(handle);
386 return acpi_map_pxm_to_node(pxm);
388 EXPORT_SYMBOL(acpi_get_node);