1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * acpi_numa.c - ACPI NUMA support
4 *
5 * Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
6 */
7
8 #define pr_fmt(fmt) "ACPI: " fmt
9
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/errno.h>
15 #include <linux/acpi.h>
16 #include <linux/memblock.h>
17 #include <linux/numa.h>
18 #include <linux/nodemask.h>
19 #include <linux/topology.h>
20
21 static nodemask_t nodes_found_map = NODE_MASK_NONE;
22
23 /* maps to convert between proximity domain and logical node ID */
24 static int pxm_to_node_map[MAX_PXM_DOMAINS]
25 = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
26 static int node_to_pxm_map[MAX_NUMNODES]
27 = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
28
29 unsigned char acpi_srat_revision __initdata;
30 static int acpi_numa __initdata;
31
disable_srat(void)32 void __init disable_srat(void)
33 {
34 acpi_numa = -1;
35 }
36
pxm_to_node(int pxm)37 int pxm_to_node(int pxm)
38 {
39 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
40 return NUMA_NO_NODE;
41 return pxm_to_node_map[pxm];
42 }
43 EXPORT_SYMBOL(pxm_to_node);
44
node_to_pxm(int node)45 int node_to_pxm(int node)
46 {
47 if (node < 0)
48 return PXM_INVAL;
49 return node_to_pxm_map[node];
50 }
51
__acpi_map_pxm_to_node(int pxm,int node)52 static void __acpi_map_pxm_to_node(int pxm, int node)
53 {
54 if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
55 pxm_to_node_map[pxm] = node;
56 if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
57 node_to_pxm_map[node] = pxm;
58 }
59
acpi_map_pxm_to_node(int pxm)60 int acpi_map_pxm_to_node(int pxm)
61 {
62 int node;
63
64 if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
65 return NUMA_NO_NODE;
66
67 node = pxm_to_node_map[pxm];
68
69 if (node == NUMA_NO_NODE) {
70 if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
71 return NUMA_NO_NODE;
72 node = first_unset_node(nodes_found_map);
73 __acpi_map_pxm_to_node(pxm, node);
74 node_set(node, nodes_found_map);
75 }
76
77 return node;
78 }
79 EXPORT_SYMBOL(acpi_map_pxm_to_node);
80
81 static void __init
acpi_table_print_srat_entry(struct acpi_subtable_header * header)82 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
83 {
84 switch (header->type) {
85 case ACPI_SRAT_TYPE_CPU_AFFINITY:
86 {
87 struct acpi_srat_cpu_affinity *p =
88 (struct acpi_srat_cpu_affinity *)header;
89 pr_debug("SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
90 p->apic_id, p->local_sapic_eid,
91 p->proximity_domain_lo,
92 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
93 "enabled" : "disabled");
94 }
95 break;
96
97 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
98 {
99 struct acpi_srat_mem_affinity *p =
100 (struct acpi_srat_mem_affinity *)header;
101 pr_debug("SRAT Memory (0x%llx length 0x%llx) in proximity domain %d %s%s%s\n",
102 (unsigned long long)p->base_address,
103 (unsigned long long)p->length,
104 p->proximity_domain,
105 (p->flags & ACPI_SRAT_MEM_ENABLED) ?
106 "enabled" : "disabled",
107 (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
108 " hot-pluggable" : "",
109 (p->flags & ACPI_SRAT_MEM_NON_VOLATILE) ?
110 " non-volatile" : "");
111 }
112 break;
113
114 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
115 {
116 struct acpi_srat_x2apic_cpu_affinity *p =
117 (struct acpi_srat_x2apic_cpu_affinity *)header;
118 pr_debug("SRAT Processor (x2apicid[0x%08x]) in proximity domain %d %s\n",
119 p->apic_id,
120 p->proximity_domain,
121 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
122 "enabled" : "disabled");
123 }
124 break;
125
126 case ACPI_SRAT_TYPE_GICC_AFFINITY:
127 {
128 struct acpi_srat_gicc_affinity *p =
129 (struct acpi_srat_gicc_affinity *)header;
130 pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
131 p->acpi_processor_uid,
132 p->proximity_domain,
133 (p->flags & ACPI_SRAT_GICC_ENABLED) ?
134 "enabled" : "disabled");
135 }
136 break;
137
138 case ACPI_SRAT_TYPE_GENERIC_AFFINITY:
139 {
140 struct acpi_srat_generic_affinity *p =
141 (struct acpi_srat_generic_affinity *)header;
142
143 if (p->device_handle_type == 0) {
144 /*
145 * For pci devices this may be the only place they
146 * are assigned a proximity domain
147 */
148 pr_debug("SRAT Generic Initiator(Seg:%u BDF:%u) in proximity domain %d %s\n",
149 *(u16 *)(&p->device_handle[0]),
150 *(u16 *)(&p->device_handle[2]),
151 p->proximity_domain,
152 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
153 "enabled" : "disabled");
154 } else {
155 /*
156 * In this case we can rely on the device having a
157 * proximity domain reference
158 */
159 pr_debug("SRAT Generic Initiator(HID=%.8s UID=%.4s) in proximity domain %d %s\n",
160 (char *)(&p->device_handle[0]),
161 (char *)(&p->device_handle[8]),
162 p->proximity_domain,
163 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
164 "enabled" : "disabled");
165 }
166 }
167 break;
168 default:
169 pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
170 header->type);
171 break;
172 }
173 }
174
175 /*
176 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
177 * up the NUMA heuristics which wants the local node to have a smaller
178 * distance than the others.
179 * Do some quick checks here and only use the SLIT if it passes.
180 */
slit_valid(struct acpi_table_slit * slit)181 static int __init slit_valid(struct acpi_table_slit *slit)
182 {
183 int i, j;
184 int d = slit->locality_count;
185 for (i = 0; i < d; i++) {
186 for (j = 0; j < d; j++) {
187 u8 val = slit->entry[d*i + j];
188 if (i == j) {
189 if (val != LOCAL_DISTANCE)
190 return 0;
191 } else if (val <= LOCAL_DISTANCE)
192 return 0;
193 }
194 }
195 return 1;
196 }
197
bad_srat(void)198 void __init bad_srat(void)
199 {
200 pr_err("SRAT: SRAT not used.\n");
201 disable_srat();
202 }
203
srat_disabled(void)204 int __init srat_disabled(void)
205 {
206 return acpi_numa < 0;
207 }
208
209 #if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
210 /*
211 * Callback for SLIT parsing. pxm_to_node() returns NUMA_NO_NODE for
212 * I/O localities since SRAT does not list them. I/O localities are
213 * not supported at this point.
214 */
acpi_numa_slit_init(struct acpi_table_slit * slit)215 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
216 {
217 int i, j;
218
219 for (i = 0; i < slit->locality_count; i++) {
220 const int from_node = pxm_to_node(i);
221
222 if (from_node == NUMA_NO_NODE)
223 continue;
224
225 for (j = 0; j < slit->locality_count; j++) {
226 const int to_node = pxm_to_node(j);
227
228 if (to_node == NUMA_NO_NODE)
229 continue;
230
231 numa_set_distance(from_node, to_node,
232 slit->entry[slit->locality_count * i + j]);
233 }
234 }
235 }
236
237 /*
238 * Default callback for parsing of the Proximity Domain <-> Memory
239 * Area mappings
240 */
241 int __init
acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity * ma)242 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
243 {
244 u64 start, end;
245 u32 hotpluggable;
246 int node, pxm;
247
248 if (srat_disabled())
249 goto out_err;
250 if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
251 pr_err("SRAT: Unexpected header length: %d\n",
252 ma->header.length);
253 goto out_err_bad_srat;
254 }
255 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
256 goto out_err;
257 hotpluggable = IS_ENABLED(CONFIG_MEMORY_HOTPLUG) &&
258 (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE);
259
260 start = ma->base_address;
261 end = start + ma->length;
262 pxm = ma->proximity_domain;
263 if (acpi_srat_revision <= 1)
264 pxm &= 0xff;
265
266 node = acpi_map_pxm_to_node(pxm);
267 if (node == NUMA_NO_NODE) {
268 pr_err("SRAT: Too many proximity domains.\n");
269 goto out_err_bad_srat;
270 }
271
272 if (numa_add_memblk(node, start, end) < 0) {
273 pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
274 node, (unsigned long long) start,
275 (unsigned long long) end - 1);
276 goto out_err_bad_srat;
277 }
278
279 node_set(node, numa_nodes_parsed);
280
281 pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
282 node, pxm,
283 (unsigned long long) start, (unsigned long long) end - 1,
284 hotpluggable ? " hotplug" : "",
285 ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
286
287 /* Mark hotplug range in memblock. */
288 if (hotpluggable && memblock_mark_hotplug(start, ma->length))
289 pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
290 (unsigned long long)start, (unsigned long long)end - 1);
291
292 max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
293
294 return 0;
295 out_err_bad_srat:
296 bad_srat();
297 out_err:
298 return -EINVAL;
299 }
300
acpi_parse_cfmws(union acpi_subtable_headers * header,void * arg,const unsigned long table_end)301 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
302 void *arg, const unsigned long table_end)
303 {
304 struct acpi_cedt_cfmws *cfmws;
305 int *fake_pxm = arg;
306 u64 start, end;
307 int node;
308
309 cfmws = (struct acpi_cedt_cfmws *)header;
310 start = cfmws->base_hpa;
311 end = cfmws->base_hpa + cfmws->window_size;
312
313 /* Skip if the SRAT already described the NUMA details for this HPA */
314 node = phys_to_target_node(start);
315 if (node != NUMA_NO_NODE)
316 return 0;
317
318 node = acpi_map_pxm_to_node(*fake_pxm);
319
320 if (node == NUMA_NO_NODE) {
321 pr_err("ACPI NUMA: Too many proximity domains while processing CFMWS.\n");
322 return -EINVAL;
323 }
324
325 if (numa_add_memblk(node, start, end) < 0) {
326 /* CXL driver must handle the NUMA_NO_NODE case */
327 pr_warn("ACPI NUMA: Failed to add memblk for CFMWS node %d [mem %#llx-%#llx]\n",
328 node, start, end);
329 }
330 node_set(node, numa_nodes_parsed);
331
332 /* Set the next available fake_pxm value */
333 (*fake_pxm)++;
334 return 0;
335 }
336 #else
acpi_parse_cfmws(union acpi_subtable_headers * header,void * arg,const unsigned long table_end)337 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
338 void *arg, const unsigned long table_end)
339 {
340 return 0;
341 }
342 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
343
acpi_parse_slit(struct acpi_table_header * table)344 static int __init acpi_parse_slit(struct acpi_table_header *table)
345 {
346 struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
347
348 if (!slit_valid(slit)) {
349 pr_info("SLIT table looks invalid. Not used.\n");
350 return -EINVAL;
351 }
352 acpi_numa_slit_init(slit);
353
354 return 0;
355 }
356
357 void __init __weak
acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity * pa)358 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
359 {
360 pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
361 }
362
363 static int __init
acpi_parse_x2apic_affinity(union acpi_subtable_headers * header,const unsigned long end)364 acpi_parse_x2apic_affinity(union acpi_subtable_headers *header,
365 const unsigned long end)
366 {
367 struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
368
369 processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
370
371 acpi_table_print_srat_entry(&header->common);
372
373 /* let architecture-dependent part to do it */
374 acpi_numa_x2apic_affinity_init(processor_affinity);
375
376 return 0;
377 }
378
379 static int __init
acpi_parse_processor_affinity(union acpi_subtable_headers * header,const unsigned long end)380 acpi_parse_processor_affinity(union acpi_subtable_headers *header,
381 const unsigned long end)
382 {
383 struct acpi_srat_cpu_affinity *processor_affinity;
384
385 processor_affinity = (struct acpi_srat_cpu_affinity *)header;
386
387 acpi_table_print_srat_entry(&header->common);
388
389 /* let architecture-dependent part to do it */
390 acpi_numa_processor_affinity_init(processor_affinity);
391
392 return 0;
393 }
394
395 static int __init
acpi_parse_gicc_affinity(union acpi_subtable_headers * header,const unsigned long end)396 acpi_parse_gicc_affinity(union acpi_subtable_headers *header,
397 const unsigned long end)
398 {
399 struct acpi_srat_gicc_affinity *processor_affinity;
400
401 processor_affinity = (struct acpi_srat_gicc_affinity *)header;
402
403 acpi_table_print_srat_entry(&header->common);
404
405 /* let architecture-dependent part to do it */
406 acpi_numa_gicc_affinity_init(processor_affinity);
407
408 return 0;
409 }
410
411 #if defined(CONFIG_X86) || defined(CONFIG_ARM64)
412 static int __init
acpi_parse_gi_affinity(union acpi_subtable_headers * header,const unsigned long end)413 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
414 const unsigned long end)
415 {
416 struct acpi_srat_generic_affinity *gi_affinity;
417 int node;
418
419 gi_affinity = (struct acpi_srat_generic_affinity *)header;
420 if (!gi_affinity)
421 return -EINVAL;
422 acpi_table_print_srat_entry(&header->common);
423
424 if (!(gi_affinity->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED))
425 return -EINVAL;
426
427 node = acpi_map_pxm_to_node(gi_affinity->proximity_domain);
428 if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
429 pr_err("SRAT: Too many proximity domains.\n");
430 return -EINVAL;
431 }
432 node_set(node, numa_nodes_parsed);
433 node_set_state(node, N_GENERIC_INITIATOR);
434
435 return 0;
436 }
437 #else
438 static int __init
acpi_parse_gi_affinity(union acpi_subtable_headers * header,const unsigned long end)439 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
440 const unsigned long end)
441 {
442 return 0;
443 }
444 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
445
446 static int __initdata parsed_numa_memblks;
447
448 static int __init
acpi_parse_memory_affinity(union acpi_subtable_headers * header,const unsigned long end)449 acpi_parse_memory_affinity(union acpi_subtable_headers * header,
450 const unsigned long end)
451 {
452 struct acpi_srat_mem_affinity *memory_affinity;
453
454 memory_affinity = (struct acpi_srat_mem_affinity *)header;
455
456 acpi_table_print_srat_entry(&header->common);
457
458 /* let architecture-dependent part to do it */
459 if (!acpi_numa_memory_affinity_init(memory_affinity))
460 parsed_numa_memblks++;
461 return 0;
462 }
463
acpi_parse_srat(struct acpi_table_header * table)464 static int __init acpi_parse_srat(struct acpi_table_header *table)
465 {
466 struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
467
468 acpi_srat_revision = srat->header.revision;
469
470 /* Real work done in acpi_table_parse_srat below. */
471
472 return 0;
473 }
474
475 static int __init
acpi_table_parse_srat(enum acpi_srat_type id,acpi_tbl_entry_handler handler,unsigned int max_entries)476 acpi_table_parse_srat(enum acpi_srat_type id,
477 acpi_tbl_entry_handler handler, unsigned int max_entries)
478 {
479 return acpi_table_parse_entries(ACPI_SIG_SRAT,
480 sizeof(struct acpi_table_srat), id,
481 handler, max_entries);
482 }
483
acpi_numa_init(void)484 int __init acpi_numa_init(void)
485 {
486 int i, fake_pxm, cnt = 0;
487
488 if (acpi_disabled)
489 return -EINVAL;
490
491 /*
492 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
493 * SRAT cpu entries could have different order with that in MADT.
494 * So go over all cpu entries in SRAT to get apicid to node mapping.
495 */
496
497 /* SRAT: System Resource Affinity Table */
498 if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
499 struct acpi_subtable_proc srat_proc[4];
500
501 memset(srat_proc, 0, sizeof(srat_proc));
502 srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
503 srat_proc[0].handler = acpi_parse_processor_affinity;
504 srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
505 srat_proc[1].handler = acpi_parse_x2apic_affinity;
506 srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
507 srat_proc[2].handler = acpi_parse_gicc_affinity;
508 srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
509 srat_proc[3].handler = acpi_parse_gi_affinity;
510
511 acpi_table_parse_entries_array(ACPI_SIG_SRAT,
512 sizeof(struct acpi_table_srat),
513 srat_proc, ARRAY_SIZE(srat_proc), 0);
514
515 cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
516 acpi_parse_memory_affinity, 0);
517 }
518
519 /* SLIT: System Locality Information Table */
520 acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
521
522 /*
523 * CXL Fixed Memory Window Structures (CFMWS) must be parsed
524 * after the SRAT. Create NUMA Nodes for CXL memory ranges that
525 * are defined in the CFMWS and not already defined in the SRAT.
526 * Initialize a fake_pxm as the first available PXM to emulate.
527 */
528
529 /* fake_pxm is the next unused PXM value after SRAT parsing */
530 for (i = 0, fake_pxm = -1; i < MAX_NUMNODES - 1; i++) {
531 if (node_to_pxm_map[i] > fake_pxm)
532 fake_pxm = node_to_pxm_map[i];
533 }
534 fake_pxm++;
535 acpi_table_parse_cedt(ACPI_CEDT_TYPE_CFMWS, acpi_parse_cfmws,
536 &fake_pxm);
537
538 if (cnt < 0)
539 return cnt;
540 else if (!parsed_numa_memblks)
541 return -ENOENT;
542 return 0;
543 }
544
acpi_get_pxm(acpi_handle h)545 static int acpi_get_pxm(acpi_handle h)
546 {
547 unsigned long long pxm;
548 acpi_status status;
549 acpi_handle handle;
550 acpi_handle phandle = h;
551
552 do {
553 handle = phandle;
554 status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
555 if (ACPI_SUCCESS(status))
556 return pxm;
557 status = acpi_get_parent(handle, &phandle);
558 } while (ACPI_SUCCESS(status));
559 return -1;
560 }
561
acpi_get_node(acpi_handle handle)562 int acpi_get_node(acpi_handle handle)
563 {
564 int pxm;
565
566 pxm = acpi_get_pxm(handle);
567
568 return pxm_to_node(pxm);
569 }
570 EXPORT_SYMBOL(acpi_get_node);
571