Commit | Line | Data |
---|---|---|
1a2db300 GK |
1 | /* |
2 | * NUMA support, based on the x86 implementation. | |
3 | * | |
4 | * Copyright (C) 2015 Cavium Inc. | |
5 | * Author: Ganapatrao Kulkarni <gkulkarni@cavium.com> | |
6 | * | |
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. | |
10 | * | |
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. | |
15 | * | |
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/>. | |
18 | */ | |
19 | ||
20 | #include <linux/bootmem.h> | |
21 | #include <linux/memblock.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/of.h> | |
24 | ||
25 | struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; | |
26 | EXPORT_SYMBOL(node_data); | |
27 | nodemask_t numa_nodes_parsed __initdata; | |
28 | static int cpu_to_node_map[NR_CPUS] = { [0 ... NR_CPUS-1] = NUMA_NO_NODE }; | |
29 | ||
30 | static int numa_distance_cnt; | |
31 | static u8 *numa_distance; | |
32 | static int numa_off; | |
33 | ||
34 | static __init int numa_parse_early_param(char *opt) | |
35 | { | |
36 | if (!opt) | |
37 | return -EINVAL; | |
38 | if (!strncmp(opt, "off", 3)) { | |
39 | pr_info("%s\n", "NUMA turned off"); | |
40 | numa_off = 1; | |
41 | } | |
42 | return 0; | |
43 | } | |
44 | early_param("numa", numa_parse_early_param); | |
45 | ||
46 | cpumask_var_t node_to_cpumask_map[MAX_NUMNODES]; | |
47 | EXPORT_SYMBOL(node_to_cpumask_map); | |
48 | ||
49 | #ifdef CONFIG_DEBUG_PER_CPU_MAPS | |
50 | ||
51 | /* | |
52 | * Returns a pointer to the bitmask of CPUs on Node 'node'. | |
53 | */ | |
54 | const struct cpumask *cpumask_of_node(int node) | |
55 | { | |
56 | if (WARN_ON(node >= nr_node_ids)) | |
57 | return cpu_none_mask; | |
58 | ||
59 | if (WARN_ON(node_to_cpumask_map[node] == NULL)) | |
60 | return cpu_online_mask; | |
61 | ||
62 | return node_to_cpumask_map[node]; | |
63 | } | |
64 | EXPORT_SYMBOL(cpumask_of_node); | |
65 | ||
66 | #endif | |
67 | ||
68 | static void map_cpu_to_node(unsigned int cpu, int nid) | |
69 | { | |
70 | set_cpu_numa_node(cpu, nid); | |
71 | if (nid >= 0) | |
72 | cpumask_set_cpu(cpu, node_to_cpumask_map[nid]); | |
73 | } | |
74 | ||
75 | void numa_clear_node(unsigned int cpu) | |
76 | { | |
77 | int nid = cpu_to_node(cpu); | |
78 | ||
79 | if (nid >= 0) | |
80 | cpumask_clear_cpu(cpu, node_to_cpumask_map[nid]); | |
81 | set_cpu_numa_node(cpu, NUMA_NO_NODE); | |
82 | } | |
83 | ||
84 | /* | |
85 | * Allocate node_to_cpumask_map based on number of available nodes | |
86 | * Requires node_possible_map to be valid. | |
87 | * | |
88 | * Note: cpumask_of_node() is not valid until after this is done. | |
89 | * (Use CONFIG_DEBUG_PER_CPU_MAPS to check this.) | |
90 | */ | |
91 | static void __init setup_node_to_cpumask_map(void) | |
92 | { | |
93 | unsigned int cpu; | |
94 | int node; | |
95 | ||
96 | /* setup nr_node_ids if not done yet */ | |
97 | if (nr_node_ids == MAX_NUMNODES) | |
98 | setup_nr_node_ids(); | |
99 | ||
100 | /* allocate and clear the mapping */ | |
101 | for (node = 0; node < nr_node_ids; node++) { | |
102 | alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]); | |
103 | cpumask_clear(node_to_cpumask_map[node]); | |
104 | } | |
105 | ||
106 | for_each_possible_cpu(cpu) | |
107 | set_cpu_numa_node(cpu, NUMA_NO_NODE); | |
108 | ||
109 | /* cpumask_of_node() will now work */ | |
110 | pr_debug("NUMA: Node to cpumask map for %d nodes\n", nr_node_ids); | |
111 | } | |
112 | ||
113 | /* | |
114 | * Set the cpu to node and mem mapping | |
115 | */ | |
116 | void numa_store_cpu_info(unsigned int cpu) | |
117 | { | |
118 | map_cpu_to_node(cpu, numa_off ? 0 : cpu_to_node_map[cpu]); | |
119 | } | |
120 | ||
121 | void __init early_map_cpu_to_node(unsigned int cpu, int nid) | |
122 | { | |
123 | /* fallback to node 0 */ | |
124 | if (nid < 0 || nid >= MAX_NUMNODES) | |
125 | nid = 0; | |
126 | ||
127 | cpu_to_node_map[cpu] = nid; | |
128 | } | |
129 | ||
130 | /** | |
131 | * numa_add_memblk - Set node id to memblk | |
132 | * @nid: NUMA node ID of the new memblk | |
133 | * @start: Start address of the new memblk | |
134 | * @size: Size of the new memblk | |
135 | * | |
136 | * RETURNS: | |
137 | * 0 on success, -errno on failure. | |
138 | */ | |
139 | int __init numa_add_memblk(int nid, u64 start, u64 size) | |
140 | { | |
141 | int ret; | |
142 | ||
143 | ret = memblock_set_node(start, size, &memblock.memory, nid); | |
144 | if (ret < 0) { | |
145 | pr_err("NUMA: memblock [0x%llx - 0x%llx] failed to add on node %d\n", | |
146 | start, (start + size - 1), nid); | |
147 | return ret; | |
148 | } | |
149 | ||
150 | node_set(nid, numa_nodes_parsed); | |
151 | pr_info("NUMA: Adding memblock [0x%llx - 0x%llx] on node %d\n", | |
152 | start, (start + size - 1), nid); | |
153 | return ret; | |
154 | } | |
155 | ||
156 | /** | |
157 | * Initialize NODE_DATA for a node on the local memory | |
158 | */ | |
159 | static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) | |
160 | { | |
161 | const size_t nd_size = roundup(sizeof(pg_data_t), SMP_CACHE_BYTES); | |
162 | u64 nd_pa; | |
163 | void *nd; | |
164 | int tnid; | |
165 | ||
166 | pr_info("NUMA: Initmem setup node %d [mem %#010Lx-%#010Lx]\n", | |
167 | nid, start_pfn << PAGE_SHIFT, | |
168 | (end_pfn << PAGE_SHIFT) - 1); | |
169 | ||
170 | nd_pa = memblock_alloc_try_nid(nd_size, SMP_CACHE_BYTES, nid); | |
171 | nd = __va(nd_pa); | |
172 | ||
173 | /* report and initialize */ | |
174 | pr_info("NUMA: NODE_DATA [mem %#010Lx-%#010Lx]\n", | |
175 | nd_pa, nd_pa + nd_size - 1); | |
176 | tnid = early_pfn_to_nid(nd_pa >> PAGE_SHIFT); | |
177 | if (tnid != nid) | |
178 | pr_info("NUMA: NODE_DATA(%d) on node %d\n", nid, tnid); | |
179 | ||
180 | node_data[nid] = nd; | |
181 | memset(NODE_DATA(nid), 0, sizeof(pg_data_t)); | |
182 | NODE_DATA(nid)->node_id = nid; | |
183 | NODE_DATA(nid)->node_start_pfn = start_pfn; | |
184 | NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn; | |
185 | } | |
186 | ||
187 | /** | |
188 | * numa_free_distance | |
189 | * | |
190 | * The current table is freed. | |
191 | */ | |
192 | void __init numa_free_distance(void) | |
193 | { | |
194 | size_t size; | |
195 | ||
196 | if (!numa_distance) | |
197 | return; | |
198 | ||
199 | size = numa_distance_cnt * numa_distance_cnt * | |
200 | sizeof(numa_distance[0]); | |
201 | ||
202 | memblock_free(__pa(numa_distance), size); | |
203 | numa_distance_cnt = 0; | |
204 | numa_distance = NULL; | |
205 | } | |
206 | ||
207 | /** | |
208 | * | |
209 | * Create a new NUMA distance table. | |
210 | * | |
211 | */ | |
212 | static int __init numa_alloc_distance(void) | |
213 | { | |
214 | size_t size; | |
215 | u64 phys; | |
216 | int i, j; | |
217 | ||
218 | size = nr_node_ids * nr_node_ids * sizeof(numa_distance[0]); | |
219 | phys = memblock_find_in_range(0, PFN_PHYS(max_pfn), | |
220 | size, PAGE_SIZE); | |
221 | if (WARN_ON(!phys)) | |
222 | return -ENOMEM; | |
223 | ||
224 | memblock_reserve(phys, size); | |
225 | ||
226 | numa_distance = __va(phys); | |
227 | numa_distance_cnt = nr_node_ids; | |
228 | ||
229 | /* fill with the default distances */ | |
230 | for (i = 0; i < numa_distance_cnt; i++) | |
231 | for (j = 0; j < numa_distance_cnt; j++) | |
232 | numa_distance[i * numa_distance_cnt + j] = i == j ? | |
233 | LOCAL_DISTANCE : REMOTE_DISTANCE; | |
234 | ||
235 | pr_debug("NUMA: Initialized distance table, cnt=%d\n", | |
236 | numa_distance_cnt); | |
237 | ||
238 | return 0; | |
239 | } | |
240 | ||
241 | /** | |
242 | * numa_set_distance - Set inter node NUMA distance from node to node. | |
243 | * @from: the 'from' node to set distance | |
244 | * @to: the 'to' node to set distance | |
245 | * @distance: NUMA distance | |
246 | * | |
247 | * Set the distance from node @from to @to to @distance. | |
248 | * If distance table doesn't exist, a warning is printed. | |
249 | * | |
250 | * If @from or @to is higher than the highest known node or lower than zero | |
251 | * or @distance doesn't make sense, the call is ignored. | |
252 | * | |
253 | */ | |
254 | void __init numa_set_distance(int from, int to, int distance) | |
255 | { | |
256 | if (!numa_distance) { | |
257 | pr_warn_once("NUMA: Warning: distance table not allocated yet\n"); | |
258 | return; | |
259 | } | |
260 | ||
261 | if (from >= numa_distance_cnt || to >= numa_distance_cnt || | |
262 | from < 0 || to < 0) { | |
263 | pr_warn_once("NUMA: Warning: node ids are out of bound, from=%d to=%d distance=%d\n", | |
264 | from, to, distance); | |
265 | return; | |
266 | } | |
267 | ||
268 | if ((u8)distance != distance || | |
269 | (from == to && distance != LOCAL_DISTANCE)) { | |
270 | pr_warn_once("NUMA: Warning: invalid distance parameter, from=%d to=%d distance=%d\n", | |
271 | from, to, distance); | |
272 | return; | |
273 | } | |
274 | ||
275 | numa_distance[from * numa_distance_cnt + to] = distance; | |
276 | } | |
277 | ||
278 | /** | |
279 | * Return NUMA distance @from to @to | |
280 | */ | |
281 | int __node_distance(int from, int to) | |
282 | { | |
283 | if (from >= numa_distance_cnt || to >= numa_distance_cnt) | |
284 | return from == to ? LOCAL_DISTANCE : REMOTE_DISTANCE; | |
285 | return numa_distance[from * numa_distance_cnt + to]; | |
286 | } | |
287 | EXPORT_SYMBOL(__node_distance); | |
288 | ||
289 | static int __init numa_register_nodes(void) | |
290 | { | |
291 | int nid; | |
292 | struct memblock_region *mblk; | |
293 | ||
294 | /* Check that valid nid is set to memblks */ | |
295 | for_each_memblock(memory, mblk) | |
296 | if (mblk->nid == NUMA_NO_NODE || mblk->nid >= MAX_NUMNODES) { | |
297 | pr_warn("NUMA: Warning: invalid memblk node %d [mem %#010Lx-%#010Lx]\n", | |
298 | mblk->nid, mblk->base, | |
299 | mblk->base + mblk->size - 1); | |
300 | return -EINVAL; | |
301 | } | |
302 | ||
303 | /* Finally register nodes. */ | |
304 | for_each_node_mask(nid, numa_nodes_parsed) { | |
305 | unsigned long start_pfn, end_pfn; | |
306 | ||
307 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
308 | setup_node_data(nid, start_pfn, end_pfn); | |
309 | node_set_online(nid); | |
310 | } | |
311 | ||
312 | /* Setup online nodes to actual nodes*/ | |
313 | node_possible_map = numa_nodes_parsed; | |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
318 | static int __init numa_init(int (*init_func)(void)) | |
319 | { | |
320 | int ret; | |
321 | ||
322 | nodes_clear(numa_nodes_parsed); | |
323 | nodes_clear(node_possible_map); | |
324 | nodes_clear(node_online_map); | |
325 | numa_free_distance(); | |
326 | ||
327 | ret = numa_alloc_distance(); | |
328 | if (ret < 0) | |
329 | return ret; | |
330 | ||
331 | ret = init_func(); | |
332 | if (ret < 0) | |
333 | return ret; | |
334 | ||
335 | if (nodes_empty(numa_nodes_parsed)) | |
336 | return -EINVAL; | |
337 | ||
338 | ret = numa_register_nodes(); | |
339 | if (ret < 0) | |
340 | return ret; | |
341 | ||
342 | setup_node_to_cpumask_map(); | |
343 | ||
344 | /* init boot processor */ | |
345 | cpu_to_node_map[0] = 0; | |
346 | map_cpu_to_node(0, 0); | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
351 | /** | |
352 | * dummy_numa_init - Fallback dummy NUMA init | |
353 | * | |
354 | * Used if there's no underlying NUMA architecture, NUMA initialization | |
355 | * fails, or NUMA is disabled on the command line. | |
356 | * | |
357 | * Must online at least one node (node 0) and add memory blocks that cover all | |
358 | * allowed memory. It is unlikely that this function fails. | |
359 | */ | |
360 | static int __init dummy_numa_init(void) | |
361 | { | |
362 | int ret; | |
363 | struct memblock_region *mblk; | |
364 | ||
365 | pr_info("%s\n", "No NUMA configuration found"); | |
366 | pr_info("NUMA: Faking a node at [mem %#018Lx-%#018Lx]\n", | |
367 | 0LLU, PFN_PHYS(max_pfn) - 1); | |
368 | ||
369 | for_each_memblock(memory, mblk) { | |
370 | ret = numa_add_memblk(0, mblk->base, mblk->size); | |
371 | if (!ret) | |
372 | continue; | |
373 | ||
374 | pr_err("NUMA init failed\n"); | |
375 | return ret; | |
376 | } | |
377 | ||
378 | numa_off = 1; | |
379 | return 0; | |
380 | } | |
381 | ||
382 | /** | |
383 | * arm64_numa_init - Initialize NUMA | |
384 | * | |
385 | * Try each configured NUMA initialization method until one succeeds. The | |
386 | * last fallback is dummy single node config encomapssing whole memory. | |
387 | */ | |
388 | void __init arm64_numa_init(void) | |
389 | { | |
390 | if (!numa_off) { | |
391 | if (!numa_init(of_numa_init)) | |
392 | return; | |
393 | } | |
394 | ||
395 | numa_init(dummy_numa_init); | |
396 | } |