1/*
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 *
18 *  You should have received a copy of the GNU General Public License
19 *  along with this program; if not, write to the Free Software
20 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
21 *
22 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
23 *
24 */
25#include <linux/module.h>
26#include <linux/init.h>
27#include <linux/kernel.h>
28#include <linux/types.h>
29#include <linux/errno.h>
30#include <linux/acpi.h>
31#include <linux/numa.h>
32
33#define PREFIX "ACPI: "
34
35#define ACPI_NUMA	0x80000000
36#define _COMPONENT	ACPI_NUMA
37ACPI_MODULE_NAME("numa");
38
39static nodemask_t nodes_found_map = NODE_MASK_NONE;
40
41/* maps to convert between proximity domain and logical node ID */
42static int pxm_to_node_map[MAX_PXM_DOMAINS]
43			= { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
44static int node_to_pxm_map[MAX_NUMNODES]
45			= { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
46
47unsigned char acpi_srat_revision __initdata;
48
49int pxm_to_node(int pxm)
50{
51	if (pxm < 0)
52		return NUMA_NO_NODE;
53	return pxm_to_node_map[pxm];
54}
55
56int node_to_pxm(int node)
57{
58	if (node < 0)
59		return PXM_INVAL;
60	return node_to_pxm_map[node];
61}
62
63static void __acpi_map_pxm_to_node(int pxm, int node)
64{
65	if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
66		pxm_to_node_map[pxm] = node;
67	if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
68		node_to_pxm_map[node] = pxm;
69}
70
71int acpi_map_pxm_to_node(int pxm)
72{
73	int node = pxm_to_node_map[pxm];
74
75	if (node == NUMA_NO_NODE) {
76		if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
77			return NUMA_NO_NODE;
78		node = first_unset_node(nodes_found_map);
79		__acpi_map_pxm_to_node(pxm, node);
80		node_set(node, nodes_found_map);
81	}
82
83	return node;
84}
85
86static void __init
87acpi_table_print_srat_entry(struct acpi_subtable_header *header)
88{
89
90	ACPI_FUNCTION_NAME("acpi_table_print_srat_entry");
91
92	if (!header)
93		return;
94
95	switch (header->type) {
96
97	case ACPI_SRAT_TYPE_CPU_AFFINITY:
98#ifdef ACPI_DEBUG_OUTPUT
99		{
100			struct acpi_srat_cpu_affinity *p =
101			    (struct acpi_srat_cpu_affinity *)header;
102			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
103					  "SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
104					  p->apic_id, p->local_sapic_eid,
105					  p->proximity_domain_lo,
106					  (p->flags & ACPI_SRAT_CPU_ENABLED)?
107					  "enabled" : "disabled"));
108		}
109#endif				/* ACPI_DEBUG_OUTPUT */
110		break;
111
112	case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
113#ifdef ACPI_DEBUG_OUTPUT
114		{
115			struct acpi_srat_mem_affinity *p =
116			    (struct acpi_srat_mem_affinity *)header;
117			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
118					  "SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
119					  (unsigned long)p->base_address,
120					  (unsigned long)p->length,
121					  p->proximity_domain,
122					  (p->flags & ACPI_SRAT_MEM_ENABLED)?
123					  "enabled" : "disabled",
124					  (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE)?
125					  " hot-pluggable" : "",
126					  (p->flags & ACPI_SRAT_MEM_NON_VOLATILE)?
127					  " non-volatile" : ""));
128		}
129#endif				/* ACPI_DEBUG_OUTPUT */
130		break;
131
132	case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
133#ifdef ACPI_DEBUG_OUTPUT
134		{
135			struct acpi_srat_x2apic_cpu_affinity *p =
136			    (struct acpi_srat_x2apic_cpu_affinity *)header;
137			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
138					  "SRAT Processor (x2apicid[0x%08x]) in"
139					  " proximity domain %d %s\n",
140					  p->apic_id,
141					  p->proximity_domain,
142					  (p->flags & ACPI_SRAT_CPU_ENABLED) ?
143					  "enabled" : "disabled"));
144		}
145#endif				/* ACPI_DEBUG_OUTPUT */
146		break;
147	default:
148		printk(KERN_WARNING PREFIX
149		       "Found unsupported SRAT entry (type = 0x%x)\n",
150		       header->type);
151		break;
152	}
153}
154
155/*
156 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
157 * up the NUMA heuristics which wants the local node to have a smaller
158 * distance than the others.
159 * Do some quick checks here and only use the SLIT if it passes.
160 */
161static int __init slit_valid(struct acpi_table_slit *slit)
162{
163	int i, j;
164	int d = slit->locality_count;
165	for (i = 0; i < d; i++) {
166		for (j = 0; j < d; j++)  {
167			u8 val = slit->entry[d*i + j];
168			if (i == j) {
169				if (val != LOCAL_DISTANCE)
170					return 0;
171			} else if (val <= LOCAL_DISTANCE)
172				return 0;
173		}
174	}
175	return 1;
176}
177
178static int __init acpi_parse_slit(struct acpi_table_header *table)
179{
180	struct acpi_table_slit *slit;
181
182	if (!table)
183		return -EINVAL;
184
185	slit = (struct acpi_table_slit *)table;
186
187	if (!slit_valid(slit)) {
188		printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
189		return -EINVAL;
190	}
191	acpi_numa_slit_init(slit);
192
193	return 0;
194}
195
196void __init __weak
197acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
198{
199	printk(KERN_WARNING PREFIX
200	       "Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
201	return;
202}
203
204
205static int __init
206acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
207			   const unsigned long end)
208{
209	struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
210
211	processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
212	if (!processor_affinity)
213		return -EINVAL;
214
215	acpi_table_print_srat_entry(header);
216
217	/* let architecture-dependent part to do it */
218	acpi_numa_x2apic_affinity_init(processor_affinity);
219
220	return 0;
221}
222
223static int __init
224acpi_parse_processor_affinity(struct acpi_subtable_header *header,
225			      const unsigned long end)
226{
227	struct acpi_srat_cpu_affinity *processor_affinity;
228
229	processor_affinity = (struct acpi_srat_cpu_affinity *)header;
230	if (!processor_affinity)
231		return -EINVAL;
232
233	acpi_table_print_srat_entry(header);
234
235	/* let architecture-dependent part to do it */
236	acpi_numa_processor_affinity_init(processor_affinity);
237
238	return 0;
239}
240
241static int __initdata parsed_numa_memblks;
242
243static int __init
244acpi_parse_memory_affinity(struct acpi_subtable_header * header,
245			   const unsigned long end)
246{
247	struct acpi_srat_mem_affinity *memory_affinity;
248
249	memory_affinity = (struct acpi_srat_mem_affinity *)header;
250	if (!memory_affinity)
251		return -EINVAL;
252
253	acpi_table_print_srat_entry(header);
254
255	/* let architecture-dependent part to do it */
256	if (!acpi_numa_memory_affinity_init(memory_affinity))
257		parsed_numa_memblks++;
258	return 0;
259}
260
261static int __init acpi_parse_srat(struct acpi_table_header *table)
262{
263	struct acpi_table_srat *srat;
264	if (!table)
265		return -EINVAL;
266
267	srat = (struct acpi_table_srat *)table;
268	acpi_srat_revision = srat->header.revision;
269
270	/* Real work done in acpi_table_parse_srat below. */
271
272	return 0;
273}
274
275static int __init
276acpi_table_parse_srat(enum acpi_srat_type id,
277		      acpi_tbl_entry_handler handler, unsigned int max_entries)
278{
279	return acpi_table_parse_entries(ACPI_SIG_SRAT,
280					    sizeof(struct acpi_table_srat), id,
281					    handler, max_entries);
282}
283
284int __init acpi_numa_init(void)
285{
286	int cnt = 0;
287
288	/*
289	 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
290	 * SRAT cpu entries could have different order with that in MADT.
291	 * So go over all cpu entries in SRAT to get apicid to node mapping.
292	 */
293
294	/* SRAT: Static Resource Affinity Table */
295	if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
296		acpi_table_parse_srat(ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY,
297				     acpi_parse_x2apic_affinity, 0);
298		acpi_table_parse_srat(ACPI_SRAT_TYPE_CPU_AFFINITY,
299				     acpi_parse_processor_affinity, 0);
300		cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
301					    acpi_parse_memory_affinity,
302					    NR_NODE_MEMBLKS);
303	}
304
305	/* SLIT: System Locality Information Table */
306	acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
307
308	acpi_numa_arch_fixup();
309
310	if (cnt < 0)
311		return cnt;
312	else if (!parsed_numa_memblks)
313		return -ENOENT;
314	return 0;
315}
316
317static int acpi_get_pxm(acpi_handle h)
318{
319	unsigned long long pxm;
320	acpi_status status;
321	acpi_handle handle;
322	acpi_handle phandle = h;
323
324	do {
325		handle = phandle;
326		status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
327		if (ACPI_SUCCESS(status))
328			return pxm;
329		status = acpi_get_parent(handle, &phandle);
330	} while (ACPI_SUCCESS(status));
331	return -1;
332}
333
334int acpi_get_node(acpi_handle handle)
335{
336	int pxm;
337
338	pxm = acpi_get_pxm(handle);
339	if (pxm < 0 || pxm >= MAX_PXM_DOMAINS)
340		return NUMA_NO_NODE;
341
342	return acpi_map_pxm_to_node(pxm);
343}
344EXPORT_SYMBOL(acpi_get_node);
345