blob: d176e0ece47041c851f287563f31d4a8fbae6288 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * acpi_numa.c - ACPI NUMA support
3 *
4 * Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
5 *
6 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
7 *
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.
12 *
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.
17 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
19 *
20 */
21#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/init.h>
23#include <linux/kernel.h>
24#include <linux/types.h>
25#include <linux/errno.h>
26#include <linux/acpi.h>
David Rientjesb552a8c2009-11-17 14:27:32 -080027#include <linux/numa.h>
Toshi Kani99759862015-06-19 17:14:15 -060028#include <linux/nodemask.h>
29#include <linux/topology.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Len Browna192a952009-07-28 16:45:54 -040031#define PREFIX "ACPI: "
32
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#define ACPI_NUMA 0x80000000
34#define _COMPONENT ACPI_NUMA
Len Brownf52fd662007-02-12 22:42:12 -050035ACPI_MODULE_NAME("numa");
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Yasunori Goto762834e2006-06-23 02:03:19 -070037static nodemask_t nodes_found_map = NODE_MASK_NONE;
Yasunori Goto762834e2006-06-23 02:03:19 -070038
39/* maps to convert between proximity domain and logical node ID */
Jan Beulichffada892007-12-13 08:32:26 +000040static int pxm_to_node_map[MAX_PXM_DOMAINS]
David Rientjesb552a8c2009-11-17 14:27:32 -080041 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
Jan Beulichffada892007-12-13 08:32:26 +000042static int node_to_pxm_map[MAX_NUMNODES]
David Rientjesb552a8c2009-11-17 14:27:32 -080043 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
Yasunori Goto762834e2006-06-23 02:03:19 -070044
Kurt Garloff8df0eb7c2012-01-17 04:18:02 -050045unsigned char acpi_srat_revision __initdata;
46
Aaron Durbinf363d162007-05-16 22:11:06 -070047int pxm_to_node(int pxm)
Yasunori Goto762834e2006-06-23 02:03:19 -070048{
49 if (pxm < 0)
David Rientjesb552a8c2009-11-17 14:27:32 -080050 return NUMA_NO_NODE;
Yasunori Goto762834e2006-06-23 02:03:19 -070051 return pxm_to_node_map[pxm];
52}
53
Aaron Durbinf363d162007-05-16 22:11:06 -070054int node_to_pxm(int node)
Yasunori Goto762834e2006-06-23 02:03:19 -070055{
56 if (node < 0)
57 return PXM_INVAL;
58 return node_to_pxm_map[node];
59}
60
Bjorn Helgaasd79ed242014-01-24 15:48:01 -070061static void __acpi_map_pxm_to_node(int pxm, int node)
David Rientjes3484d792007-07-21 17:10:32 +020062{
David Rientjes0f9b75e2010-03-25 16:33:04 -070063 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;
David Rientjes3484d792007-07-21 17:10:32 +020067}
68
Luck, Tony8ff6f482007-05-24 13:57:40 -070069int acpi_map_pxm_to_node(int pxm)
Yasunori Goto762834e2006-06-23 02:03:19 -070070{
Toshi Kani99759862015-06-19 17:14:15 -060071 int node;
72
73 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS)
74 return NUMA_NO_NODE;
75
76 node = pxm_to_node_map[pxm];
Yasunori Goto762834e2006-06-23 02:03:19 -070077
Jianguo Wu1bb25df2013-08-30 09:25:40 +080078 if (node == NUMA_NO_NODE) {
Yasunori Goto762834e2006-06-23 02:03:19 -070079 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
David Rientjesb552a8c2009-11-17 14:27:32 -080080 return NUMA_NO_NODE;
Yasunori Goto762834e2006-06-23 02:03:19 -070081 node = first_unset_node(nodes_found_map);
David Rientjes3484d792007-07-21 17:10:32 +020082 __acpi_map_pxm_to_node(pxm, node);
Yasunori Goto762834e2006-06-23 02:03:19 -070083 node_set(node, nodes_found_map);
84 }
85
86 return node;
87}
88
Toshi Kani99759862015-06-19 17:14:15 -060089/**
90 * acpi_map_pxm_to_online_node - Map proximity ID to online node
91 * @pxm: ACPI proximity ID
92 *
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.
96 *
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.
102 * "numa=off" on x86.
103 */
104int acpi_map_pxm_to_online_node(int pxm)
105{
106 int node, n, dist, min_dist;
107
108 node = acpi_map_pxm_to_node(pxm);
109
110 if (node == NUMA_NO_NODE)
111 node = 0;
112
113 if (!node_online(node)) {
114 min_dist = INT_MAX;
115 for_each_online_node(n) {
116 dist = node_distance(node, n);
117 if (dist < min_dist) {
118 min_dist = dist;
119 node = n;
120 }
121 }
122 }
123
124 return node;
125}
126EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
127
David Rientjesae2c6dc2007-07-21 17:09:56 +0200128static void __init
129acpi_table_print_srat_entry(struct acpi_subtable_header *header)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
131
Len Brown4be44fc2005-08-05 00:44:28 -0400132 ACPI_FUNCTION_NAME("acpi_table_print_srat_entry");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
134 if (!header)
135 return;
136
137 switch (header->type) {
138
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300139 case ACPI_SRAT_TYPE_CPU_AFFINITY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140#ifdef ACPI_DEBUG_OUTPUT
Len Brown4be44fc2005-08-05 00:44:28 -0400141 {
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300142 struct acpi_srat_cpu_affinity *p =
143 (struct acpi_srat_cpu_affinity *)header;
Len Brown4be44fc2005-08-05 00:44:28 -0400144 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
145 "SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300146 p->apic_id, p->local_sapic_eid,
147 p->proximity_domain_lo,
148 (p->flags & ACPI_SRAT_CPU_ENABLED)?
149 "enabled" : "disabled"));
Len Brown4be44fc2005-08-05 00:44:28 -0400150 }
151#endif /* ACPI_DEBUG_OUTPUT */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 break;
153
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300154 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155#ifdef ACPI_DEBUG_OUTPUT
Len Brown4be44fc2005-08-05 00:44:28 -0400156 {
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300157 struct acpi_srat_mem_affinity *p =
158 (struct acpi_srat_mem_affinity *)header;
Len Brown4be44fc2005-08-05 00:44:28 -0400159 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
Davidlohr Bueso208f6cc2013-01-09 22:06:08 +0000160 "SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300161 (unsigned long)p->base_address,
162 (unsigned long)p->length,
Bob Moore19d0cfe2008-06-10 15:54:40 +0800163 p->proximity_domain,
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300164 (p->flags & ACPI_SRAT_MEM_ENABLED)?
165 "enabled" : "disabled",
166 (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE)?
Davidlohr Bueso208f6cc2013-01-09 22:06:08 +0000167 " hot-pluggable" : "",
168 (p->flags & ACPI_SRAT_MEM_NON_VOLATILE)?
169 " non-volatile" : ""));
Len Brown4be44fc2005-08-05 00:44:28 -0400170 }
171#endif /* ACPI_DEBUG_OUTPUT */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 break;
173
Suresh Siddha7237d3d2009-03-30 13:55:30 -0800174 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
175#ifdef ACPI_DEBUG_OUTPUT
176 {
177 struct acpi_srat_x2apic_cpu_affinity *p =
178 (struct acpi_srat_x2apic_cpu_affinity *)header;
179 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
180 "SRAT Processor (x2apicid[0x%08x]) in"
181 " proximity domain %d %s\n",
182 p->apic_id,
183 p->proximity_domain,
184 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
185 "enabled" : "disabled"));
186 }
187#endif /* ACPI_DEBUG_OUTPUT */
188 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 default:
Len Brown4be44fc2005-08-05 00:44:28 -0400190 printk(KERN_WARNING PREFIX
191 "Found unsupported SRAT entry (type = 0x%x)\n",
192 header->type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 break;
194 }
195}
196
Fenghua Yu39b89312008-06-09 16:48:18 -0700197/*
198 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
199 * up the NUMA heuristics which wants the local node to have a smaller
200 * distance than the others.
201 * Do some quick checks here and only use the SLIT if it passes.
202 */
Hanjun Guo40e31852013-08-13 18:31:14 +0800203static int __init slit_valid(struct acpi_table_slit *slit)
Fenghua Yu39b89312008-06-09 16:48:18 -0700204{
205 int i, j;
206 int d = slit->locality_count;
207 for (i = 0; i < d; i++) {
208 for (j = 0; j < d; j++) {
209 u8 val = slit->entry[d*i + j];
210 if (i == j) {
211 if (val != LOCAL_DISTANCE)
212 return 0;
213 } else if (val <= LOCAL_DISTANCE)
214 return 0;
215 }
216 }
217 return 1;
218}
219
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300220static int __init acpi_parse_slit(struct acpi_table_header *table)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Hanjun Guo2fad9302015-02-05 17:33:14 +0800222 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Fenghua Yu39b89312008-06-09 16:48:18 -0700224 if (!slit_valid(slit)) {
225 printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
226 return -EINVAL;
227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 acpi_numa_slit_init(slit);
229
230 return 0;
231}
232
Bjorn Helgaasbeffbe52014-01-27 17:19:15 -0700233void __init __weak
Suresh Siddha7237d3d2009-03-30 13:55:30 -0800234acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
235{
236 printk(KERN_WARNING PREFIX
237 "Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
238 return;
239}
240
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242static int __init
Suresh Siddha7237d3d2009-03-30 13:55:30 -0800243acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
244 const unsigned long end)
245{
246 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
247
248 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
249 if (!processor_affinity)
250 return -EINVAL;
251
252 acpi_table_print_srat_entry(header);
253
254 /* let architecture-dependent part to do it */
255 acpi_numa_x2apic_affinity_init(processor_affinity);
256
257 return 0;
258}
259
260static int __init
261acpi_parse_processor_affinity(struct acpi_subtable_header *header,
Len Brown4be44fc2005-08-05 00:44:28 -0400262 const unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300264 struct acpi_srat_cpu_affinity *processor_affinity;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300266 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 if (!processor_affinity)
268 return -EINVAL;
269
270 acpi_table_print_srat_entry(header);
271
272 /* let architecture-dependent part to do it */
273 acpi_numa_processor_affinity_init(processor_affinity);
274
275 return 0;
276}
277
Thomas Renninger095adbb2012-07-31 17:41:09 +0200278static int __initdata parsed_numa_memblks;
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280static int __init
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300281acpi_parse_memory_affinity(struct acpi_subtable_header * header,
Len Brown4be44fc2005-08-05 00:44:28 -0400282 const unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300284 struct acpi_srat_mem_affinity *memory_affinity;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300286 memory_affinity = (struct acpi_srat_mem_affinity *)header;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (!memory_affinity)
288 return -EINVAL;
289
290 acpi_table_print_srat_entry(header);
291
292 /* let architecture-dependent part to do it */
Thomas Renninger095adbb2012-07-31 17:41:09 +0200293 if (!acpi_numa_memory_affinity_init(memory_affinity))
294 parsed_numa_memblks++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return 0;
296}
297
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300298static int __init acpi_parse_srat(struct acpi_table_header *table)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Hanjun Guo2fad9302015-02-05 17:33:14 +0800300 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
Kurt Garloff8df0eb7c2012-01-17 04:18:02 -0500302 acpi_srat_revision = srat->header.revision;
303
Andi Kleencfa806f2010-07-20 15:18:36 -0700304 /* Real work done in acpi_table_parse_srat below. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
306 return 0;
307}
308
David Rientjesae2c6dc2007-07-21 17:09:56 +0200309static int __init
Alexey Starikovskiy15a58ed2007-02-02 19:48:22 +0300310acpi_table_parse_srat(enum acpi_srat_type id,
Lv Zhengb43e1062013-01-12 15:29:38 +0000311 acpi_tbl_entry_handler handler, unsigned int max_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Len Brown6eb87fe2007-02-10 22:17:07 -0500313 return acpi_table_parse_entries(ACPI_SIG_SRAT,
Len Brown4be44fc2005-08-05 00:44:28 -0400314 sizeof(struct acpi_table_srat), id,
315 handler, max_entries);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
Yinghai Lu20e69262013-03-01 14:51:27 -0800318int __init acpi_numa_init(void)
Tang Chene8d19552013-02-22 16:33:44 -0800319{
Yinghai Lu20e69262013-03-01 14:51:27 -0800320 int cnt = 0;
321
Yinghai Lud3bd0582010-12-16 19:09:58 -0800322 /*
323 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
324 * SRAT cpu entries could have different order with that in MADT.
325 * So go over all cpu entries in SRAT to get apicid to node mapping.
326 */
David Rientjes87162732009-09-25 15:20:04 -0700327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 /* SRAT: Static Resource Affinity Table */
Len Brown7f8f97c2007-02-10 21:28:03 -0500329 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
Lukasz Anaczkowski702b07f2016-04-21 11:29:00 +0200330 struct acpi_subtable_proc srat_proc[2];
331
332 memset(srat_proc, 0, sizeof(srat_proc));
333 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
334 srat_proc[0].handler = acpi_parse_processor_affinity;
335 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
336 srat_proc[1].handler = acpi_parse_x2apic_affinity;
337
338 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
339 sizeof(struct acpi_table_srat),
340 srat_proc, ARRAY_SIZE(srat_proc), 0);
341
Yinghai Lu20e69262013-03-01 14:51:27 -0800342 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
343 acpi_parse_memory_affinity,
344 NR_NODE_MEMBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 }
346
347 /* SLIT: System Locality Information Table */
Len Brown7f8f97c2007-02-10 21:28:03 -0500348 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 acpi_numa_arch_fixup();
Tejun Heo940fed22011-02-16 12:13:06 +0100351
Yinghai Lu20e69262013-03-01 14:51:27 -0800352 if (cnt < 0)
353 return cnt;
Thomas Renninger095adbb2012-07-31 17:41:09 +0200354 else if (!parsed_numa_memblks)
Thomas Renningerf3946fb2012-07-31 17:41:08 +0200355 return -ENOENT;
Tejun Heo940fed22011-02-16 12:13:06 +0100356 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
Bjorn Helgaasd79ed242014-01-24 15:48:01 -0700359static int acpi_get_pxm(acpi_handle h)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
Matthew Wilcox27663c52008-10-10 02:22:59 -0400361 unsigned long long pxm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 acpi_status status;
363 acpi_handle handle;
364 acpi_handle phandle = h;
365
366 do {
367 handle = phandle;
368 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
369 if (ACPI_SUCCESS(status))
Jan Engelhardt50dd0962006-10-01 00:28:50 +0200370 return pxm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 status = acpi_get_parent(handle, &phandle);
Len Brown4be44fc2005-08-05 00:44:28 -0400372 } while (ACPI_SUCCESS(status));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 return -1;
374}
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700375
Bjorn Helgaas486c79b2014-01-24 15:25:10 -0700376int acpi_get_node(acpi_handle handle)
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700377{
Bjorn Helgaas962fe9c2014-01-24 15:42:17 -0700378 int pxm;
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700379
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700380 pxm = acpi_get_pxm(handle);
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700381
Bjorn Helgaas962fe9c2014-01-24 15:42:17 -0700382 return acpi_map_pxm_to_node(pxm);
Yasunori Goto1e3590e2006-06-27 02:53:31 -0700383}
384EXPORT_SYMBOL(acpi_get_node);