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, min_node;
107
108	node = acpi_map_pxm_to_node(pxm);
109
110	if (node == NUMA_NO_NODE)
111		node = 0;
112
113	min_node = node;
114	if (!node_online(node)) {
115		int min_dist = INT_MAX, dist, n;
116
117		for_each_online_node(n) {
118			dist = node_distance(node, n);
119			if (dist < min_dist) {
120				min_dist = dist;
121				min_node = n;
122			}
123		}
124	}
125
126	return min_node;
127}
128EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
129
130static void __init
131acpi_table_print_srat_entry(struct acpi_subtable_header *header)
132{
133	switch (header->type) {
134	case ACPI_SRAT_TYPE_CPU_AFFINITY:
135		{
136			struct acpi_srat_cpu_affinity *p =
137			    (struct acpi_srat_cpu_affinity *)header;
138			pr_debug("SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
139				 p->apic_id, p->local_sapic_eid,
140				 p->proximity_domain_lo,
141				 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
142				 "enabled" : "disabled");
143		}
144		break;
145
146	case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
147		{
148			struct acpi_srat_mem_affinity *p =
149			    (struct acpi_srat_mem_affinity *)header;
150			pr_debug("SRAT Memory (0x%lx length 0x%lx) in proximity domain %d %s%s%s\n",
151				 (unsigned long)p->base_address,
152				 (unsigned long)p->length,
153				 p->proximity_domain,
154				 (p->flags & ACPI_SRAT_MEM_ENABLED) ?
155				 "enabled" : "disabled",
156				 (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
157				 " hot-pluggable" : "",
158				 (p->flags & ACPI_SRAT_MEM_NON_VOLATILE) ?
159				 " non-volatile" : "");
160		}
161		break;
162
163	case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
164		{
165			struct acpi_srat_x2apic_cpu_affinity *p =
166			    (struct acpi_srat_x2apic_cpu_affinity *)header;
167			pr_debug("SRAT Processor (x2apicid[0x%08x]) in proximity domain %d %s\n",
168				 p->apic_id,
169				 p->proximity_domain,
170				 (p->flags & ACPI_SRAT_CPU_ENABLED) ?
171				 "enabled" : "disabled");
172		}
173		break;
174
175	case ACPI_SRAT_TYPE_GICC_AFFINITY:
176		{
177			struct acpi_srat_gicc_affinity *p =
178			    (struct acpi_srat_gicc_affinity *)header;
179			pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
180				 p->acpi_processor_uid,
181				 p->proximity_domain,
182				 (p->flags & ACPI_SRAT_GICC_ENABLED) ?
183				 "enabled" : "disabled");
184		}
185		break;
186
187	default:
188		pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
189			header->type);
190		break;
191	}
192}
193
194/*
195 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
196 * up the NUMA heuristics which wants the local node to have a smaller
197 * distance than the others.
198 * Do some quick checks here and only use the SLIT if it passes.
199 */
200static int __init slit_valid(struct acpi_table_slit *slit)
201{
202	int i, j;
203	int d = slit->locality_count;
204	for (i = 0; i < d; i++) {
205		for (j = 0; j < d; j++)  {
206			u8 val = slit->entry[d*i + j];
207			if (i == j) {
208				if (val != LOCAL_DISTANCE)
209					return 0;
210			} else if (val <= LOCAL_DISTANCE)
211				return 0;
212		}
213	}
214	return 1;
215}
216
217void __init bad_srat(void)
218{
219	pr_err("SRAT: SRAT not used.\n");
220	acpi_numa = -1;
221}
222
223int __init srat_disabled(void)
224{
225	return acpi_numa < 0;
226}
227
228#if defined(CONFIG_X86) || defined(CONFIG_ARM64)
229/*
230 * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
231 * I/O localities since SRAT does not list them.  I/O localities are
232 * not supported at this point.
233 */
234void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
235{
236	int i, j;
237
238	for (i = 0; i < slit->locality_count; i++) {
239		const int from_node = pxm_to_node(i);
240
241		if (from_node == NUMA_NO_NODE)
242			continue;
243
244		for (j = 0; j < slit->locality_count; j++) {
245			const int to_node = pxm_to_node(j);
246
247			if (to_node == NUMA_NO_NODE)
248				continue;
249
250			numa_set_distance(from_node, to_node,
251				slit->entry[slit->locality_count * i + j]);
252		}
253	}
254}
255
256/*
257 * Default callback for parsing of the Proximity Domain <-> Memory
258 * Area mappings
259 */
260int __init
261acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
262{
263	u64 start, end;
264	u32 hotpluggable;
265	int node, pxm;
266
267	if (srat_disabled())
268		goto out_err;
269	if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
270		pr_err("SRAT: Unexpected header length: %d\n",
271		       ma->header.length);
272		goto out_err_bad_srat;
273	}
274	if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
275		goto out_err;
276	hotpluggable = ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE;
277	if (hotpluggable && !IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
278		goto out_err;
279
280	start = ma->base_address;
281	end = start + ma->length;
282	pxm = ma->proximity_domain;
283	if (acpi_srat_revision <= 1)
284		pxm &= 0xff;
285
286	node = acpi_map_pxm_to_node(pxm);
287	if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
288		pr_err("SRAT: Too many proximity domains.\n");
289		goto out_err_bad_srat;
290	}
291
292	if (numa_add_memblk(node, start, end) < 0) {
293		pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
294		       node, (unsigned long long) start,
295		       (unsigned long long) end - 1);
296		goto out_err_bad_srat;
297	}
298
299	node_set(node, numa_nodes_parsed);
300
301	pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
302		node, pxm,
303		(unsigned long long) start, (unsigned long long) end - 1,
304		hotpluggable ? " hotplug" : "",
305		ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
306
307	/* Mark hotplug range in memblock. */
308	if (hotpluggable && memblock_mark_hotplug(start, ma->length))
309		pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
310			(unsigned long long)start, (unsigned long long)end - 1);
311
312	max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
313
314	return 0;
315out_err_bad_srat:
316	bad_srat();
317out_err:
318	return -EINVAL;
319}
320#endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
321
322static int __init acpi_parse_slit(struct acpi_table_header *table)
323{
324	struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
325
326	if (!slit_valid(slit)) {
327		pr_info("SLIT table looks invalid. Not used.\n");
328		return -EINVAL;
329	}
330	acpi_numa_slit_init(slit);
331
332	return 0;
333}
334
335void __init __weak
336acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
337{
338	pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
339}
340
341static int __init
342acpi_parse_x2apic_affinity(struct acpi_subtable_header *header,
343			   const unsigned long end)
344{
345	struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
346
347	processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
348	if (!processor_affinity)
349		return -EINVAL;
350
351	acpi_table_print_srat_entry(header);
352
353	/* let architecture-dependent part to do it */
354	acpi_numa_x2apic_affinity_init(processor_affinity);
355
356	return 0;
357}
358
359static int __init
360acpi_parse_processor_affinity(struct acpi_subtable_header *header,
361			      const unsigned long end)
362{
363	struct acpi_srat_cpu_affinity *processor_affinity;
364
365	processor_affinity = (struct acpi_srat_cpu_affinity *)header;
366	if (!processor_affinity)
367		return -EINVAL;
368
369	acpi_table_print_srat_entry(header);
370
371	/* let architecture-dependent part to do it */
372	acpi_numa_processor_affinity_init(processor_affinity);
373
374	return 0;
375}
376
377static int __init
378acpi_parse_gicc_affinity(struct acpi_subtable_header *header,
379			 const unsigned long end)
380{
381	struct acpi_srat_gicc_affinity *processor_affinity;
382
383	processor_affinity = (struct acpi_srat_gicc_affinity *)header;
384	if (!processor_affinity)
385		return -EINVAL;
386
387	acpi_table_print_srat_entry(header);
388
389	/* let architecture-dependent part to do it */
390	acpi_numa_gicc_affinity_init(processor_affinity);
391
392	return 0;
393}
394
395static int __initdata parsed_numa_memblks;
396
397static int __init
398acpi_parse_memory_affinity(struct acpi_subtable_header * header,
399			   const unsigned long end)
400{
401	struct acpi_srat_mem_affinity *memory_affinity;
402
403	memory_affinity = (struct acpi_srat_mem_affinity *)header;
404	if (!memory_affinity)
405		return -EINVAL;
406
407	acpi_table_print_srat_entry(header);
408
409	/* let architecture-dependent part to do it */
410	if (!acpi_numa_memory_affinity_init(memory_affinity))
411		parsed_numa_memblks++;
412	return 0;
413}
414
415static int __init acpi_parse_srat(struct acpi_table_header *table)
416{
417	struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
418
419	acpi_srat_revision = srat->header.revision;
420
421	/* Real work done in acpi_table_parse_srat below. */
422
423	return 0;
424}
425
426static int __init
427acpi_table_parse_srat(enum acpi_srat_type id,
428		      acpi_tbl_entry_handler handler, unsigned int max_entries)
429{
430	return acpi_table_parse_entries(ACPI_SIG_SRAT,
431					    sizeof(struct acpi_table_srat), id,
432					    handler, max_entries);
433}
434
435int __init acpi_numa_init(void)
436{
437	int cnt = 0;
438
439	if (acpi_disabled)
440		return -EINVAL;
441
442	/*
443	 * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
444	 * SRAT cpu entries could have different order with that in MADT.
445	 * So go over all cpu entries in SRAT to get apicid to node mapping.
446	 */
447
448	/* SRAT: System Resource Affinity Table */
449	if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
450		struct acpi_subtable_proc srat_proc[3];
451
452		memset(srat_proc, 0, sizeof(srat_proc));
453		srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
454		srat_proc[0].handler = acpi_parse_processor_affinity;
455		srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
456		srat_proc[1].handler = acpi_parse_x2apic_affinity;
457		srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
458		srat_proc[2].handler = acpi_parse_gicc_affinity;
459
460		acpi_table_parse_entries_array(ACPI_SIG_SRAT,
461					sizeof(struct acpi_table_srat),
462					srat_proc, ARRAY_SIZE(srat_proc), 0);
463
464		cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
465					    acpi_parse_memory_affinity, 0);
466	}
467
468	/* SLIT: System Locality Information Table */
469	acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
470
471	if (cnt < 0)
472		return cnt;
473	else if (!parsed_numa_memblks)
474		return -ENOENT;
475	return 0;
476}
477
478static int acpi_get_pxm(acpi_handle h)
479{
480	unsigned long long pxm;
481	acpi_status status;
482	acpi_handle handle;
483	acpi_handle phandle = h;
484
485	do {
486		handle = phandle;
487		status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
488		if (ACPI_SUCCESS(status))
489			return pxm;
490		status = acpi_get_parent(handle, &phandle);
491	} while (ACPI_SUCCESS(status));
492	return -1;
493}
494
495int acpi_get_node(acpi_handle handle)
496{
497	int pxm;
498
499	pxm = acpi_get_pxm(handle);
500
501	return acpi_map_pxm_to_node(pxm);
502}
503EXPORT_SYMBOL(acpi_get_node);