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 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
21 #include <linux/module.h>
22 #include <linux/init.h>
23 #include <linux/kernel.h>
24 #include <linux/types.h>
25 #include <linux/errno.h>
26 #include <linux/acpi.h>
27 #include <linux/numa.h>
28 #include <linux/nodemask.h>
29 #include <linux/topology.h>
31 #define PREFIX "ACPI: "
33 #define ACPI_NUMA 0x80000000
34 #define _COMPONENT ACPI_NUMA
35 ACPI_MODULE_NAME("numa");
37 static nodemask_t nodes_found_map
= NODE_MASK_NONE
;
39 /* maps to convert between proximity domain and logical node ID */
40 static int pxm_to_node_map
[MAX_PXM_DOMAINS
]
41 = { [0 ... MAX_PXM_DOMAINS
- 1] = NUMA_NO_NODE
};
42 static int node_to_pxm_map
[MAX_NUMNODES
]
43 = { [0 ... MAX_NUMNODES
- 1] = PXM_INVAL
};
45 unsigned char acpi_srat_revision __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
)
132 ACPI_FUNCTION_NAME("acpi_table_print_srat_entry");
137 switch (header
->type
) {
139 case ACPI_SRAT_TYPE_CPU_AFFINITY
:
140 #ifdef ACPI_DEBUG_OUTPUT
142 struct acpi_srat_cpu_affinity
*p
=
143 (struct acpi_srat_cpu_affinity
*)header
;
144 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
145 "SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
146 p
->apic_id
, p
->local_sapic_eid
,
147 p
->proximity_domain_lo
,
148 (p
->flags
& ACPI_SRAT_CPU_ENABLED
)?
149 "enabled" : "disabled"));
151 #endif /* ACPI_DEBUG_OUTPUT */
154 case ACPI_SRAT_TYPE_MEMORY_AFFINITY
:
155 #ifdef ACPI_DEBUG_OUTPUT
157 struct acpi_srat_mem_affinity
*p
=
158 (struct acpi_srat_mem_affinity
*)header
;
159 ACPI_DEBUG_PRINT((ACPI_DB_INFO
,
160 "SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
161 (unsigned long)p
->base_address
,
162 (unsigned long)p
->length
,
164 (p
->flags
& ACPI_SRAT_MEM_ENABLED
)?
165 "enabled" : "disabled",
166 (p
->flags
& ACPI_SRAT_MEM_HOT_PLUGGABLE
)?
167 " hot-pluggable" : "",
168 (p
->flags
& ACPI_SRAT_MEM_NON_VOLATILE
)?
169 " non-volatile" : ""));
171 #endif /* ACPI_DEBUG_OUTPUT */
174 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY
:
175 #ifdef ACPI_DEBUG_OUTPUT
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",
184 (p
->flags
& ACPI_SRAT_CPU_ENABLED
) ?
185 "enabled" : "disabled"));
187 #endif /* ACPI_DEBUG_OUTPUT */
190 printk(KERN_WARNING PREFIX
191 "Found unsupported SRAT entry (type = 0x%x)\n",
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.
203 static int __init
slit_valid(struct acpi_table_slit
*slit
)
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
];
211 if (val
!= LOCAL_DISTANCE
)
213 } else if (val
<= LOCAL_DISTANCE
)
220 static int __init
acpi_parse_slit(struct acpi_table_header
*table
)
222 struct acpi_table_slit
*slit
= (struct acpi_table_slit
*)table
;
224 if (!slit_valid(slit
)) {
225 printk(KERN_INFO
"ACPI: SLIT table looks invalid. Not used.\n");
228 acpi_numa_slit_init(slit
);
234 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity
*pa
)
236 printk(KERN_WARNING PREFIX
237 "Found unsupported x2apic [0x%08x] SRAT entry\n", pa
->apic_id
);
243 acpi_parse_x2apic_affinity(struct acpi_subtable_header
*header
,
244 const unsigned long end
)
246 struct acpi_srat_x2apic_cpu_affinity
*processor_affinity
;
248 processor_affinity
= (struct acpi_srat_x2apic_cpu_affinity
*)header
;
249 if (!processor_affinity
)
252 acpi_table_print_srat_entry(header
);
254 /* let architecture-dependent part to do it */
255 acpi_numa_x2apic_affinity_init(processor_affinity
);
261 acpi_parse_processor_affinity(struct acpi_subtable_header
*header
,
262 const unsigned long end
)
264 struct acpi_srat_cpu_affinity
*processor_affinity
;
266 processor_affinity
= (struct acpi_srat_cpu_affinity
*)header
;
267 if (!processor_affinity
)
270 acpi_table_print_srat_entry(header
);
272 /* let architecture-dependent part to do it */
273 acpi_numa_processor_affinity_init(processor_affinity
);
278 static int __initdata parsed_numa_memblks
;
281 acpi_parse_memory_affinity(struct acpi_subtable_header
* header
,
282 const unsigned long end
)
284 struct acpi_srat_mem_affinity
*memory_affinity
;
286 memory_affinity
= (struct acpi_srat_mem_affinity
*)header
;
287 if (!memory_affinity
)
290 acpi_table_print_srat_entry(header
);
292 /* let architecture-dependent part to do it */
293 if (!acpi_numa_memory_affinity_init(memory_affinity
))
294 parsed_numa_memblks
++;
298 static int __init
acpi_parse_srat(struct acpi_table_header
*table
)
300 struct acpi_table_srat
*srat
= (struct acpi_table_srat
*)table
;
302 acpi_srat_revision
= srat
->header
.revision
;
304 /* Real work done in acpi_table_parse_srat below. */
310 acpi_table_parse_srat(enum acpi_srat_type id
,
311 acpi_tbl_entry_handler handler
, unsigned int max_entries
)
313 return acpi_table_parse_entries(ACPI_SIG_SRAT
,
314 sizeof(struct acpi_table_srat
), id
,
315 handler
, max_entries
);
318 int __init
acpi_numa_init(void)
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.
328 /* SRAT: Static Resource Affinity Table */
329 if (!acpi_table_parse(ACPI_SIG_SRAT
, acpi_parse_srat
)) {
330 struct acpi_subtable_proc srat_proc
[2];
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
;
338 acpi_table_parse_entries_array(ACPI_SIG_SRAT
,
339 sizeof(struct acpi_table_srat
),
340 srat_proc
, ARRAY_SIZE(srat_proc
), 0);
342 cnt
= acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY
,
343 acpi_parse_memory_affinity
,
347 /* SLIT: System Locality Information Table */
348 acpi_table_parse(ACPI_SIG_SLIT
, acpi_parse_slit
);
350 acpi_numa_arch_fixup();
354 else if (!parsed_numa_memblks
)
359 static int acpi_get_pxm(acpi_handle h
)
361 unsigned long long pxm
;
364 acpi_handle phandle
= h
;
368 status
= acpi_evaluate_integer(handle
, "_PXM", NULL
, &pxm
);
369 if (ACPI_SUCCESS(status
))
371 status
= acpi_get_parent(handle
, &phandle
);
372 } while (ACPI_SUCCESS(status
));
376 int acpi_get_node(acpi_handle handle
)
380 pxm
= acpi_get_pxm(handle
);
382 return acpi_map_pxm_to_node(pxm
);
384 EXPORT_SYMBOL(acpi_get_node
);