Loading...
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 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
19 *
20 */
21
22#define pr_fmt(fmt) "ACPI: " fmt
23
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/bootmem.h>
31#include <linux/memblock.h>
32#include <linux/numa.h>
33#include <linux/nodemask.h>
34#include <linux/topology.h>
35
36static nodemask_t nodes_found_map = NODE_MASK_NONE;
37
38/* maps to convert between proximity domain and logical node ID */
39static int pxm_to_node_map[MAX_PXM_DOMAINS]
40 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
41static int node_to_pxm_map[MAX_NUMNODES]
42 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
43
44unsigned char acpi_srat_revision __initdata;
45int acpi_numa __initdata;
46
47int pxm_to_node(int pxm)
48{
49 if (pxm < 0)
50 return NUMA_NO_NODE;
51 return pxm_to_node_map[pxm];
52}
53
54int node_to_pxm(int node)
55{
56 if (node < 0)
57 return PXM_INVAL;
58 return node_to_pxm_map[node];
59}
60
61static void __acpi_map_pxm_to_node(int pxm, int node)
62{
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;
67}
68
69int acpi_map_pxm_to_node(int pxm)
70{
71 int node;
72
73 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
74 return NUMA_NO_NODE;
75
76 node = pxm_to_node_map[pxm];
77
78 if (node == NUMA_NO_NODE) {
79 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
80 return NUMA_NO_NODE;
81 node = first_unset_node(nodes_found_map);
82 __acpi_map_pxm_to_node(pxm, node);
83 node_set(node, nodes_found_map);
84 }
85
86 return node;
87}
88
89/**
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
128static void __init
129acpi_table_print_srat_entry(struct acpi_subtable_header *header)
130{
131 switch (header->type) {
132 case ACPI_SRAT_TYPE_CPU_AFFINITY:
133 {
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");
141 }
142 break;
143
144 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
145 {
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,
151 p->proximity_domain,
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" : "");
158 }
159 break;
160
161 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
162 {
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",
166 p->apic_id,
167 p->proximity_domain,
168 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
169 "enabled" : "disabled");
170 }
171 break;
172
173 case ACPI_SRAT_TYPE_GICC_AFFINITY:
174 {
175 struct acpi_srat_gicc_affinity *p =
176 (struct acpi_srat_gicc_affinity *)header;
177 pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
178 p->acpi_processor_uid,
179 p->proximity_domain,
180 (p->flags & ACPI_SRAT_GICC_ENABLED) ?
181 "enabled" : "disabled");
182 }
183 break;
184
185 default:
186 pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
187 header->type);
188 break;
189 }
190}
191
192/*
193 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
194 * up the NUMA heuristics which wants the local node to have a smaller
195 * distance than the others.
196 * Do some quick checks here and only use the SLIT if it passes.
197 */
198static int __init slit_valid(struct acpi_table_slit *slit)
199{
200 int i, j;
201 int d = slit->locality_count;
202 for (i = 0; i < d; i++) {
203 for (j = 0; j < d; j++) {
204 u8 val = slit->entry[d*i + j];
205 if (i == j) {
206 if (val != LOCAL_DISTANCE)
207 return 0;
208 } else if (val <= LOCAL_DISTANCE)
209 return 0;
210 }
211 }
212 return 1;
213}
214
215void __init bad_srat(void)
216{
217 pr_err("SRAT: SRAT not used.\n");
218 acpi_numa = -1;
219}
220
221int __init srat_disabled(void)
222{
223 return acpi_numa < 0;
224}
225
226#if defined(CONFIG_X86) || defined(CONFIG_ARM64)
227/*
228 * Callback for SLIT parsing. pxm_to_node() returns NUMA_NO_NODE for
229 * I/O localities since SRAT does not list them. I/O localities are
230 * not supported at this point.
231 */
232void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
233{
234 int i, j;
235
236 for (i = 0; i < slit->locality_count; i++) {
237 const int from_node = pxm_to_node(i);
238
239 if (from_node == NUMA_NO_NODE)
240 continue;
241
242 for (j = 0; j < slit->locality_count; j++) {
243 const int to_node = pxm_to_node(j);
244
245 if (to_node == NUMA_NO_NODE)
246 continue;
247
248 numa_set_distance(from_node, to_node,
249 slit->entry[slit->locality_count * i + j]);
250 }
251 }
252}
253
254/*
255 * Default callback for parsing of the Proximity Domain <-> Memory
256 * Area mappings
257 */
258int __init
259acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
260{
261 u64 start, end;
262 u32 hotpluggable;
263 int node, pxm;
264
265 if (srat_disabled())
266 goto out_err;
267 if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
268 pr_err("SRAT: Unexpected header length: %d\n",
269 ma->header.length);
270 goto out_err_bad_srat;
271 }
272 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
273 goto out_err;
274 hotpluggable = ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE;
275 if (hotpluggable && !IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
276 goto out_err;
277
278 start = ma->base_address;
279 end = start + ma->length;
280 pxm = ma->proximity_domain;
281 if (acpi_srat_revision <= 1)
282 pxm &= 0xff;
283
284 node = acpi_map_pxm_to_node(pxm);
285 if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
286 pr_err("SRAT: Too many proximity domains.\n");
287 goto out_err_bad_srat;
288 }
289
290 if (numa_add_memblk(node, start, end) < 0) {
291 pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
292 node, (unsigned long long) start,
293 (unsigned long long) end - 1);
294 goto out_err_bad_srat;
295 }
296
297 node_set(node, numa_nodes_parsed);
298
299 pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
300 node, pxm,
301 (unsigned long long) start, (unsigned long long) end - 1,
302 hotpluggable ? " hotplug" : "",
303 ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
304
305 /* Mark hotplug range in memblock. */
306 if (hotpluggable && memblock_mark_hotplug(start, ma->length))
307 pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
308 (unsigned long long)start, (unsigned long long)end - 1);
309
310 max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
311
312 return 0;
313out_err_bad_srat:
314 bad_srat();
315out_err:
316 return -EINVAL;
317}
318#endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
319
320static int __init acpi_parse_slit(struct acpi_table_header *table)
321{
322 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
323
324 if (!slit_valid(slit)) {
325 pr_info("SLIT table looks invalid. Not used.\n");
326 return -EINVAL;
327 }
328 acpi_numa_slit_init(slit);
329
330 return 0;
331}
332
333void __init __weak
334acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
335{
336 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
337}
338
339static int __init
340acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
341 const unsigned long end)
342{
343 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
344
345 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
346 if (!processor_affinity)
347 return -EINVAL;
348
349 acpi_table_print_srat_entry(header);
350
351 /* let architecture-dependent part to do it */
352 acpi_numa_x2apic_affinity_init(processor_affinity);
353
354 return 0;
355}
356
357static int __init
358acpi_parse_processor_affinity(struct acpi_subtable_header *header,
359 const unsigned long end)
360{
361 struct acpi_srat_cpu_affinity *processor_affinity;
362
363 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
364 if (!processor_affinity)
365 return -EINVAL;
366
367 acpi_table_print_srat_entry(header);
368
369 /* let architecture-dependent part to do it */
370 acpi_numa_processor_affinity_init(processor_affinity);
371
372 return 0;
373}
374
375static int __init
376acpi_parse_gicc_affinity(struct acpi_subtable_header *header,
377 const unsigned long end)
378{
379 struct acpi_srat_gicc_affinity *processor_affinity;
380
381 processor_affinity = (struct acpi_srat_gicc_affinity *)header;
382 if (!processor_affinity)
383 return -EINVAL;
384
385 acpi_table_print_srat_entry(header);
386
387 /* let architecture-dependent part to do it */
388 acpi_numa_gicc_affinity_init(processor_affinity);
389
390 return 0;
391}
392
393static int __initdata parsed_numa_memblks;
394
395static int __init
396acpi_parse_memory_affinity(struct acpi_subtable_header * header,
397 const unsigned long end)
398{
399 struct acpi_srat_mem_affinity *memory_affinity;
400
401 memory_affinity = (struct acpi_srat_mem_affinity *)header;
402 if (!memory_affinity)
403 return -EINVAL;
404
405 acpi_table_print_srat_entry(header);
406
407 /* let architecture-dependent part to do it */
408 if (!acpi_numa_memory_affinity_init(memory_affinity))
409 parsed_numa_memblks++;
410 return 0;
411}
412
413static int __init acpi_parse_srat(struct acpi_table_header *table)
414{
415 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
416
417 acpi_srat_revision = srat->header.revision;
418
419 /* Real work done in acpi_table_parse_srat below. */
420
421 return 0;
422}
423
424static int __init
425acpi_table_parse_srat(enum acpi_srat_type id,
426 acpi_tbl_entry_handler handler, unsigned int max_entries)
427{
428 return acpi_table_parse_entries(ACPI_SIG_SRAT,
429 sizeof(struct acpi_table_srat), id,
430 handler, max_entries);
431}
432
433int __init acpi_numa_init(void)
434{
435 int cnt = 0;
436
437 if (acpi_disabled)
438 return -EINVAL;
439
440 /*
441 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
442 * SRAT cpu entries could have different order with that in MADT.
443 * So go over all cpu entries in SRAT to get apicid to node mapping.
444 */
445
446 /* SRAT: Static Resource Affinity Table */
447 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
448 struct acpi_subtable_proc srat_proc[3];
449
450 memset(srat_proc, 0, sizeof(srat_proc));
451 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
452 srat_proc[0].handler = acpi_parse_processor_affinity;
453 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
454 srat_proc[1].handler = acpi_parse_x2apic_affinity;
455 srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
456 srat_proc[2].handler = acpi_parse_gicc_affinity;
457
458 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
459 sizeof(struct acpi_table_srat),
460 srat_proc, ARRAY_SIZE(srat_proc), 0);
461
462 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
463 acpi_parse_memory_affinity,
464 NR_NODE_MEMBLKS);
465 }
466
467 /* SLIT: System Locality Information Table */
468 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
469
470 if (cnt < 0)
471 return cnt;
472 else if (!parsed_numa_memblks)
473 return -ENOENT;
474 return 0;
475}
476
477static int acpi_get_pxm(acpi_handle h)
478{
479 unsigned long long pxm;
480 acpi_status status;
481 acpi_handle handle;
482 acpi_handle phandle = h;
483
484 do {
485 handle = phandle;
486 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
487 if (ACPI_SUCCESS(status))
488 return pxm;
489 status = acpi_get_parent(handle, &phandle);
490 } while (ACPI_SUCCESS(status));
491 return -1;
492}
493
494int acpi_get_node(acpi_handle handle)
495{
496 int pxm;
497
498 pxm = acpi_get_pxm(handle);
499
500 return acpi_map_pxm_to_node(pxm);
501}
502EXPORT_SYMBOL(acpi_get_node);