2 * NUMA support, based on the x86 implementation.
4 * Copyright (C) 2015 Cavium Inc.
5 * Author: Ganapatrao Kulkarni <gkulkarni@cavium.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 #define pr_fmt(fmt) "NUMA: " fmt
22 #include <linux/acpi.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/module.h>
29 #include <asm/sections.h>
31 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
32 EXPORT_SYMBOL(node_data);
33 nodemask_t numa_nodes_parsed __initdata;
34 static int cpu_to_node_map[NR_CPUS] = { [0 ... NR_CPUS-1] = NUMA_NO_NODE };
36 static int numa_distance_cnt;
37 static u8 *numa_distance;
40 static __init int numa_parse_early_param(char *opt)
44 if (!strncmp(opt, "off", 3))
49 early_param("numa", numa_parse_early_param);
51 cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
52 EXPORT_SYMBOL(node_to_cpumask_map);
54 #ifdef CONFIG_DEBUG_PER_CPU_MAPS
57 * Returns a pointer to the bitmask of CPUs on Node 'node'.
59 const struct cpumask *cpumask_of_node(int node)
61 if (WARN_ON(node >= nr_node_ids))
64 if (WARN_ON(node_to_cpumask_map[node] == NULL))
65 return cpu_online_mask;
67 return node_to_cpumask_map[node];
69 EXPORT_SYMBOL(cpumask_of_node);
73 static void numa_update_cpu(unsigned int cpu, bool remove)
75 int nid = cpu_to_node(cpu);
77 if (nid == NUMA_NO_NODE)
81 cpumask_clear_cpu(cpu, node_to_cpumask_map[nid]);
83 cpumask_set_cpu(cpu, node_to_cpumask_map[nid]);
86 void numa_add_cpu(unsigned int cpu)
88 numa_update_cpu(cpu, false);
91 void numa_remove_cpu(unsigned int cpu)
93 numa_update_cpu(cpu, true);
96 void numa_clear_node(unsigned int cpu)
99 set_cpu_numa_node(cpu, NUMA_NO_NODE);
103 * Allocate node_to_cpumask_map based on number of available nodes
104 * Requires node_possible_map to be valid.
106 * Note: cpumask_of_node() is not valid until after this is done.
107 * (Use CONFIG_DEBUG_PER_CPU_MAPS to check this.)
109 static void __init setup_node_to_cpumask_map(void)
113 /* setup nr_node_ids if not done yet */
114 if (nr_node_ids == MAX_NUMNODES)
117 /* allocate and clear the mapping */
118 for (node = 0; node < nr_node_ids; node++) {
119 alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]);
120 cpumask_clear(node_to_cpumask_map[node]);
123 /* cpumask_of_node() will now work */
124 pr_debug("Node to cpumask map for %d nodes\n", nr_node_ids);
128 * Set the cpu to node and mem mapping
130 void numa_store_cpu_info(unsigned int cpu)
132 set_cpu_numa_node(cpu, cpu_to_node_map[cpu]);
135 void __init early_map_cpu_to_node(unsigned int cpu, int nid)
137 /* fallback to node 0 */
138 if (nid < 0 || nid >= MAX_NUMNODES || numa_off)
141 cpu_to_node_map[cpu] = nid;
144 * We should set the numa node of cpu0 as soon as possible, because it
145 * has already been set up online before. cpu_to_node(0) will soon be
149 set_cpu_numa_node(cpu, nid);
152 #ifdef CONFIG_HAVE_SETUP_PER_CPU_AREA
153 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
154 EXPORT_SYMBOL(__per_cpu_offset);
156 static int __init early_cpu_to_node(int cpu)
158 return cpu_to_node_map[cpu];
161 static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
163 return node_distance(early_cpu_to_node(from), early_cpu_to_node(to));
166 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size,
169 int nid = early_cpu_to_node(cpu);
171 return memblock_virt_alloc_try_nid(size, align,
172 __pa(MAX_DMA_ADDRESS), MEMBLOCK_ALLOC_ACCESSIBLE, nid);
175 static void __init pcpu_fc_free(void *ptr, size_t size)
177 memblock_free_early(__pa(ptr), size);
180 void __init setup_per_cpu_areas(void)
187 * Always reserve area for module percpu variables. That's
188 * what the legacy allocator did.
190 rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
191 PERCPU_DYNAMIC_RESERVE, PAGE_SIZE,
193 pcpu_fc_alloc, pcpu_fc_free);
195 panic("Failed to initialize percpu areas.");
197 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
198 for_each_possible_cpu(cpu)
199 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
204 * numa_add_memblk - Set node id to memblk
205 * @nid: NUMA node ID of the new memblk
206 * @start: Start address of the new memblk
207 * @end: End address of the new memblk
210 * 0 on success, -errno on failure.
212 int __init numa_add_memblk(int nid, u64 start, u64 end)
216 ret = memblock_set_node(start, (end - start), &memblock.memory, nid);
218 pr_err("memblock [0x%llx - 0x%llx] failed to add on node %d\n",
219 start, (end - 1), nid);
223 node_set(nid, numa_nodes_parsed);
228 * Initialize NODE_DATA for a node on the local memory
230 static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn)
232 const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES);
237 if (start_pfn >= end_pfn)
238 pr_info("Initmem setup node %d [<memory-less node>]\n", nid);
240 nd_pa = memblock_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid);
243 /* report and initialize */
244 pr_info("NODE_DATA [mem %#010Lx-%#010Lx]\n",
245 nd_pa, nd_pa + nd_size - 1);
246 tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT);
248 pr_info("NODE_DATA(%d) on node %d\n", nid, tnid);
251 memset(NODE_DATA(nid), 0, sizeof(pg_data_t));
252 NODE_DATA(nid)->node_id = nid;
253 NODE_DATA(nid)->node_start_pfn = start_pfn;
254 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
260 * The current table is freed.
262 void __init numa_free_distance(void)
269 size = numa_distance_cnt * numa_distance_cnt *
270 sizeof(numa_distance[0]);
272 memblock_free(__pa(numa_distance), size);
273 numa_distance_cnt = 0;
274 numa_distance = NULL;
279 * Create a new NUMA distance table.
282 static int __init numa_alloc_distance(void)
288 size = nr_node_ids * nr_node_ids * sizeof(numa_distance[0]);
289 phys = memblock_find_in_range(0, PFN_PHYS(max_pfn),
294 memblock_reserve(phys, size);
296 numa_distance = __va(phys);
297 numa_distance_cnt = nr_node_ids;
299 /* fill with the default distances */
300 for (i = 0; i < numa_distance_cnt; i++)
301 for (j = 0; j < numa_distance_cnt; j++)
302 numa_distance[i * numa_distance_cnt + j] = i == j ?
303 LOCAL_DISTANCE : REMOTE_DISTANCE;
305 pr_debug("Initialized distance table, cnt=%d\n", numa_distance_cnt);
311 * numa_set_distance - Set inter node NUMA distance from node to node.
312 * @from: the 'from' node to set distance
313 * @to: the 'to' node to set distance
314 * @distance: NUMA distance
316 * Set the distance from node @from to @to to @distance.
317 * If distance table doesn't exist, a warning is printed.
319 * If @from or @to is higher than the highest known node or lower than zero
320 * or @distance doesn't make sense, the call is ignored.
323 void __init numa_set_distance(int from, int to, int distance)
325 if (!numa_distance) {
326 pr_warn_once("Warning: distance table not allocated yet\n");
330 if (from >= numa_distance_cnt || to >= numa_distance_cnt ||
331 from < 0 || to < 0) {
332 pr_warn_once("Warning: node ids are out of bound, from=%d to=%d distance=%d\n",
337 if ((u8)distance != distance ||
338 (from == to && distance != LOCAL_DISTANCE)) {
339 pr_warn_once("Warning: invalid distance parameter, from=%d to=%d distance=%d\n",
344 numa_distance[from * numa_distance_cnt + to] = distance;
348 * Return NUMA distance @from to @to
350 int __node_distance(int from, int to)
352 if (from >= numa_distance_cnt || to >= numa_distance_cnt)
353 return from == to ? LOCAL_DISTANCE : REMOTE_DISTANCE;
354 return numa_distance[from * numa_distance_cnt + to];
356 EXPORT_SYMBOL(__node_distance);
358 static int __init numa_register_nodes(void)
361 struct memblock_region *mblk;
363 /* Check that valid nid is set to memblks */
364 for_each_memblock(memory, mblk)
365 if (mblk->nid == NUMA_NO_NODE || mblk->nid >= MAX_NUMNODES) {
366 pr_warn("Warning: invalid memblk node %d [mem %#010Lx-%#010Lx]\n",
367 mblk->nid, mblk->base,
368 mblk->base + mblk->size - 1);
372 /* Finally register nodes. */
373 for_each_node_mask(nid, numa_nodes_parsed) {
374 unsigned long start_pfn, end_pfn;
376 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
377 setup_node_data(nid, start_pfn, end_pfn);
378 node_set_online(nid);
381 /* Setup online nodes to actual nodes*/
382 node_possible_map = numa_nodes_parsed;
387 static int __init numa_init(int (*init_func)(void))
391 nodes_clear(numa_nodes_parsed);
392 nodes_clear(node_possible_map);
393 nodes_clear(node_online_map);
395 ret = numa_alloc_distance();
401 goto out_free_distance;
403 if (nodes_empty(numa_nodes_parsed)) {
404 pr_info("No NUMA configuration found\n");
406 goto out_free_distance;
409 ret = numa_register_nodes();
411 goto out_free_distance;
413 setup_node_to_cpumask_map();
417 numa_free_distance();
422 * dummy_numa_init - Fallback dummy NUMA init
424 * Used if there's no underlying NUMA architecture, NUMA initialization
425 * fails, or NUMA is disabled on the command line.
427 * Must online at least one node (node 0) and add memory blocks that cover all
428 * allowed memory. It is unlikely that this function fails.
430 static int __init dummy_numa_init(void)
433 struct memblock_region *mblk;
436 pr_info("NUMA disabled\n"); /* Forced off on command line. */
437 pr_info("Faking a node at [mem %#018Lx-%#018Lx]\n",
438 memblock_start_of_DRAM(), memblock_end_of_DRAM() - 1);
440 for_each_memblock(memory, mblk) {
441 ret = numa_add_memblk(0, mblk->base, mblk->base + mblk->size);
445 pr_err("NUMA init failed\n");
454 * arm64_numa_init - Initialize NUMA
456 * Try each configured NUMA initialization method until one succeeds. The
457 * last fallback is dummy single node config encomapssing whole memory.
459 void __init arm64_numa_init(void)
462 if (!acpi_disabled && !numa_init(arm64_acpi_numa_init))
464 if (acpi_disabled && !numa_init(of_numa_init))
468 numa_init(dummy_numa_init);