]>
Commit | Line | Data |
---|---|---|
46aeb7e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Simple NUMA memory policy for the Linux kernel. | |
4 | * | |
5 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 6 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
7 | * |
8 | * NUMA policy allows the user to give hints in which node(s) memory should | |
9 | * be allocated. | |
10 | * | |
11 | * Support four policies per VMA and per process: | |
12 | * | |
13 | * The VMA policy has priority over the process policy for a page fault. | |
14 | * | |
15 | * interleave Allocate memory interleaved over a set of nodes, | |
16 | * with normal fallback if it fails. | |
17 | * For VMA based allocations this interleaves based on the | |
18 | * offset into the backing object or offset into the mapping | |
19 | * for anonymous memory. For process policy an process counter | |
20 | * is used. | |
8bccd85f | 21 | * |
1da177e4 LT |
22 | * bind Only allocate memory on a specific set of nodes, |
23 | * no fallback. | |
8bccd85f CL |
24 | * FIXME: memory is allocated starting with the first node |
25 | * to the last. It would be better if bind would truly restrict | |
26 | * the allocation to memory nodes instead | |
27 | * | |
1da177e4 | 28 | * preferred Try a specific node first before normal fallback. |
00ef2d2f | 29 | * As a special case NUMA_NO_NODE here means do the allocation |
1da177e4 LT |
30 | * on the local CPU. This is normally identical to default, |
31 | * but useful to set in a VMA when you have a non default | |
32 | * process policy. | |
8bccd85f | 33 | * |
b27abacc DH |
34 | * preferred many Try a set of nodes first before normal fallback. This is |
35 | * similar to preferred without the special case. | |
36 | * | |
1da177e4 LT |
37 | * default Allocate on the local node first, or when on a VMA |
38 | * use the process policy. This is what Linux always did | |
39 | * in a NUMA aware kernel and still does by, ahem, default. | |
40 | * | |
41 | * The process policy is applied for most non interrupt memory allocations | |
42 | * in that process' context. Interrupts ignore the policies and always | |
43 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
44 | * allocations for a VMA in the VM. | |
45 | * | |
46 | * Currently there are a few corner cases in swapping where the policy | |
47 | * is not applied, but the majority should be handled. When process policy | |
48 | * is used it is not remembered over swap outs/swap ins. | |
49 | * | |
50 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
51 | * requesting a lower zone just use default policy. This implies that | |
52 | * on systems with highmem kernel lowmem allocation don't get policied. | |
53 | * Same with GFP_DMA allocations. | |
54 | * | |
55 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between | |
56 | * all users and remembered even when nobody has memory mapped. | |
57 | */ | |
58 | ||
59 | /* Notebook: | |
60 | fix mmap readahead to honour policy and enable policy for any page cache | |
61 | object | |
62 | statistics for bigpages | |
63 | global policy for page cache? currently it uses process policy. Requires | |
64 | first item above. | |
65 | handle mremap for shared memory (currently ignored for the policy) | |
66 | grows down? | |
67 | make bind policy root only? It can trigger oom much faster and the | |
68 | kernel is not always grateful with that. | |
1da177e4 LT |
69 | */ |
70 | ||
b1de0d13 MH |
71 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
72 | ||
1da177e4 | 73 | #include <linux/mempolicy.h> |
a520110e | 74 | #include <linux/pagewalk.h> |
1da177e4 LT |
75 | #include <linux/highmem.h> |
76 | #include <linux/hugetlb.h> | |
77 | #include <linux/kernel.h> | |
78 | #include <linux/sched.h> | |
6e84f315 | 79 | #include <linux/sched/mm.h> |
6a3827d7 | 80 | #include <linux/sched/numa_balancing.h> |
f719ff9b | 81 | #include <linux/sched/task.h> |
1da177e4 LT |
82 | #include <linux/nodemask.h> |
83 | #include <linux/cpuset.h> | |
1da177e4 LT |
84 | #include <linux/slab.h> |
85 | #include <linux/string.h> | |
b95f1b31 | 86 | #include <linux/export.h> |
b488893a | 87 | #include <linux/nsproxy.h> |
1da177e4 LT |
88 | #include <linux/interrupt.h> |
89 | #include <linux/init.h> | |
90 | #include <linux/compat.h> | |
31367466 | 91 | #include <linux/ptrace.h> |
dc9aa5b9 | 92 | #include <linux/swap.h> |
1a75a6c8 CL |
93 | #include <linux/seq_file.h> |
94 | #include <linux/proc_fs.h> | |
b20a3503 | 95 | #include <linux/migrate.h> |
62b61f61 | 96 | #include <linux/ksm.h> |
95a402c3 | 97 | #include <linux/rmap.h> |
86c3a764 | 98 | #include <linux/security.h> |
dbcb0f19 | 99 | #include <linux/syscalls.h> |
095f1fc4 | 100 | #include <linux/ctype.h> |
6d9c285a | 101 | #include <linux/mm_inline.h> |
b24f53a0 | 102 | #include <linux/mmu_notifier.h> |
b1de0d13 | 103 | #include <linux/printk.h> |
c8633798 | 104 | #include <linux/swapops.h> |
dc9aa5b9 | 105 | |
1da177e4 | 106 | #include <asm/tlbflush.h> |
4a18419f | 107 | #include <asm/tlb.h> |
7c0f6ba6 | 108 | #include <linux/uaccess.h> |
1da177e4 | 109 | |
62695a84 NP |
110 | #include "internal.h" |
111 | ||
38e35860 | 112 | /* Internal flags */ |
dc9aa5b9 | 113 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 114 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 115 | |
fcc234f8 PE |
116 | static struct kmem_cache *policy_cache; |
117 | static struct kmem_cache *sn_cache; | |
1da177e4 | 118 | |
1da177e4 LT |
119 | /* Highest zone. An specific allocation for a zone below that is not |
120 | policied. */ | |
6267276f | 121 | enum zone_type policy_zone = 0; |
1da177e4 | 122 | |
bea904d5 LS |
123 | /* |
124 | * run-time system-wide default policy => local allocation | |
125 | */ | |
e754d79d | 126 | static struct mempolicy default_policy = { |
1da177e4 | 127 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
7858d7bc | 128 | .mode = MPOL_LOCAL, |
1da177e4 LT |
129 | }; |
130 | ||
5606e387 MG |
131 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
132 | ||
b2ca916c DW |
133 | /** |
134 | * numa_map_to_online_node - Find closest online node | |
f6e92f40 | 135 | * @node: Node id to start the search |
b2ca916c DW |
136 | * |
137 | * Lookup the next closest node by distance if @nid is not online. | |
dad5b023 RD |
138 | * |
139 | * Return: this @node if it is online, otherwise the closest node by distance | |
b2ca916c DW |
140 | */ |
141 | int numa_map_to_online_node(int node) | |
142 | { | |
4fcbe96e | 143 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 144 | |
4fcbe96e DW |
145 | if (node == NUMA_NO_NODE || node_online(node)) |
146 | return node; | |
b2ca916c DW |
147 | |
148 | min_node = node; | |
4fcbe96e DW |
149 | for_each_online_node(n) { |
150 | dist = node_distance(node, n); | |
151 | if (dist < min_dist) { | |
152 | min_dist = dist; | |
153 | min_node = n; | |
b2ca916c DW |
154 | } |
155 | } | |
156 | ||
157 | return min_node; | |
158 | } | |
159 | EXPORT_SYMBOL_GPL(numa_map_to_online_node); | |
160 | ||
74d2c3a0 | 161 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
162 | { |
163 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 164 | int node; |
5606e387 | 165 | |
f15ca78e ON |
166 | if (pol) |
167 | return pol; | |
5606e387 | 168 | |
f15ca78e ON |
169 | node = numa_node_id(); |
170 | if (node != NUMA_NO_NODE) { | |
171 | pol = &preferred_node_policy[node]; | |
172 | /* preferred_node_policy is not initialised early in boot */ | |
173 | if (pol->mode) | |
174 | return pol; | |
5606e387 MG |
175 | } |
176 | ||
f15ca78e | 177 | return &default_policy; |
5606e387 MG |
178 | } |
179 | ||
37012946 DR |
180 | static const struct mempolicy_operations { |
181 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 182 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
183 | } mpol_ops[MPOL_MAX]; |
184 | ||
f5b087b5 DR |
185 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
186 | { | |
6d556294 | 187 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
188 | } |
189 | ||
190 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
191 | const nodemask_t *rel) | |
192 | { | |
193 | nodemask_t tmp; | |
194 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
195 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
196 | } |
197 | ||
be897d48 | 198 | static int mpol_new_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
199 | { |
200 | if (nodes_empty(*nodes)) | |
201 | return -EINVAL; | |
269fbe72 | 202 | pol->nodes = *nodes; |
37012946 DR |
203 | return 0; |
204 | } | |
205 | ||
206 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
207 | { | |
7858d7bc FT |
208 | if (nodes_empty(*nodes)) |
209 | return -EINVAL; | |
269fbe72 BW |
210 | |
211 | nodes_clear(pol->nodes); | |
212 | node_set(first_node(*nodes), pol->nodes); | |
37012946 DR |
213 | return 0; |
214 | } | |
215 | ||
58568d2a MX |
216 | /* |
217 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
218 | * any, for the new policy. mpol_new() has already validated the nodes | |
7858d7bc | 219 | * parameter with respect to the policy mode and flags. |
58568d2a MX |
220 | * |
221 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
c1e8d7c6 | 222 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 223 | */ |
4bfc4495 KH |
224 | static int mpol_set_nodemask(struct mempolicy *pol, |
225 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 226 | { |
58568d2a MX |
227 | int ret; |
228 | ||
7858d7bc FT |
229 | /* |
230 | * Default (pol==NULL) resp. local memory policies are not a | |
231 | * subject of any remapping. They also do not need any special | |
232 | * constructor. | |
233 | */ | |
234 | if (!pol || pol->mode == MPOL_LOCAL) | |
58568d2a | 235 | return 0; |
7858d7bc | 236 | |
01f13bd6 | 237 | /* Check N_MEMORY */ |
4bfc4495 | 238 | nodes_and(nsc->mask1, |
01f13bd6 | 239 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
240 | |
241 | VM_BUG_ON(!nodes); | |
4bfc4495 | 242 | |
7858d7bc FT |
243 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
244 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); | |
245 | else | |
246 | nodes_and(nsc->mask2, *nodes, nsc->mask1); | |
58568d2a | 247 | |
7858d7bc FT |
248 | if (mpol_store_user_nodemask(pol)) |
249 | pol->w.user_nodemask = *nodes; | |
4bfc4495 | 250 | else |
7858d7bc FT |
251 | pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed; |
252 | ||
253 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
58568d2a MX |
254 | return ret; |
255 | } | |
256 | ||
257 | /* | |
258 | * This function just creates a new policy, does some check and simple | |
259 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
260 | */ | |
028fec41 DR |
261 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
262 | nodemask_t *nodes) | |
1da177e4 LT |
263 | { |
264 | struct mempolicy *policy; | |
265 | ||
028fec41 | 266 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 267 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 268 | |
3e1f0645 DR |
269 | if (mode == MPOL_DEFAULT) { |
270 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 271 | return ERR_PTR(-EINVAL); |
d3a71033 | 272 | return NULL; |
37012946 | 273 | } |
3e1f0645 DR |
274 | VM_BUG_ON(!nodes); |
275 | ||
276 | /* | |
277 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
278 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
279 | * All other modes require a valid pointer to a non-empty nodemask. | |
280 | */ | |
281 | if (mode == MPOL_PREFERRED) { | |
282 | if (nodes_empty(*nodes)) { | |
283 | if (((flags & MPOL_F_STATIC_NODES) || | |
284 | (flags & MPOL_F_RELATIVE_NODES))) | |
285 | return ERR_PTR(-EINVAL); | |
7858d7bc FT |
286 | |
287 | mode = MPOL_LOCAL; | |
3e1f0645 | 288 | } |
479e2802 | 289 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
290 | if (!nodes_empty(*nodes) || |
291 | (flags & MPOL_F_STATIC_NODES) || | |
292 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 | 293 | return ERR_PTR(-EINVAL); |
3e1f0645 DR |
294 | } else if (nodes_empty(*nodes)) |
295 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
296 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
297 | if (!policy) | |
298 | return ERR_PTR(-ENOMEM); | |
299 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 300 | policy->mode = mode; |
3e1f0645 | 301 | policy->flags = flags; |
c6018b4b | 302 | policy->home_node = NUMA_NO_NODE; |
37012946 | 303 | |
1da177e4 | 304 | return policy; |
37012946 DR |
305 | } |
306 | ||
52cd3b07 LS |
307 | /* Slow path of a mpol destructor. */ |
308 | void __mpol_put(struct mempolicy *p) | |
309 | { | |
310 | if (!atomic_dec_and_test(&p->refcnt)) | |
311 | return; | |
52cd3b07 LS |
312 | kmem_cache_free(policy_cache, p); |
313 | } | |
314 | ||
213980c0 | 315 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
316 | { |
317 | } | |
318 | ||
213980c0 | 319 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
320 | { |
321 | nodemask_t tmp; | |
322 | ||
323 | if (pol->flags & MPOL_F_STATIC_NODES) | |
324 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
325 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
326 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
327 | else { | |
269fbe72 | 328 | nodes_remap(tmp, pol->nodes, pol->w.cpuset_mems_allowed, |
213980c0 | 329 | *nodes); |
29b190fa | 330 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 331 | } |
f5b087b5 | 332 | |
708c1bbc MX |
333 | if (nodes_empty(tmp)) |
334 | tmp = *nodes; | |
335 | ||
269fbe72 | 336 | pol->nodes = tmp; |
37012946 DR |
337 | } |
338 | ||
339 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 340 | const nodemask_t *nodes) |
37012946 | 341 | { |
7858d7bc | 342 | pol->w.cpuset_mems_allowed = *nodes; |
1da177e4 LT |
343 | } |
344 | ||
708c1bbc MX |
345 | /* |
346 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
347 | * | |
c1e8d7c6 | 348 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
349 | * policies are protected by task->mems_allowed_seq to prevent a premature |
350 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 351 | */ |
213980c0 | 352 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 353 | { |
018160ad | 354 | if (!pol || pol->mode == MPOL_LOCAL) |
1d0d2680 | 355 | return; |
7858d7bc | 356 | if (!mpol_store_user_nodemask(pol) && |
1d0d2680 DR |
357 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
358 | return; | |
708c1bbc | 359 | |
213980c0 | 360 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
361 | } |
362 | ||
363 | /* | |
364 | * Wrapper for mpol_rebind_policy() that just requires task | |
365 | * pointer, and updates task mempolicy. | |
58568d2a MX |
366 | * |
367 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
368 | */ |
369 | ||
213980c0 | 370 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 371 | { |
213980c0 | 372 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
373 | } |
374 | ||
375 | /* | |
376 | * Rebind each vma in mm to new nodemask. | |
377 | * | |
c1e8d7c6 | 378 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 DR |
379 | */ |
380 | ||
381 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
382 | { | |
383 | struct vm_area_struct *vma; | |
66850be5 | 384 | VMA_ITERATOR(vmi, mm, 0); |
1d0d2680 | 385 | |
d8ed45c5 | 386 | mmap_write_lock(mm); |
6c21e066 JH |
387 | for_each_vma(vmi, vma) { |
388 | vma_start_write(vma); | |
213980c0 | 389 | mpol_rebind_policy(vma->vm_policy, new); |
6c21e066 | 390 | } |
d8ed45c5 | 391 | mmap_write_unlock(mm); |
1d0d2680 DR |
392 | } |
393 | ||
37012946 DR |
394 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
395 | [MPOL_DEFAULT] = { | |
396 | .rebind = mpol_rebind_default, | |
397 | }, | |
398 | [MPOL_INTERLEAVE] = { | |
be897d48 | 399 | .create = mpol_new_nodemask, |
37012946 DR |
400 | .rebind = mpol_rebind_nodemask, |
401 | }, | |
402 | [MPOL_PREFERRED] = { | |
403 | .create = mpol_new_preferred, | |
404 | .rebind = mpol_rebind_preferred, | |
405 | }, | |
406 | [MPOL_BIND] = { | |
be897d48 | 407 | .create = mpol_new_nodemask, |
37012946 DR |
408 | .rebind = mpol_rebind_nodemask, |
409 | }, | |
7858d7bc FT |
410 | [MPOL_LOCAL] = { |
411 | .rebind = mpol_rebind_default, | |
412 | }, | |
b27abacc | 413 | [MPOL_PREFERRED_MANY] = { |
be897d48 | 414 | .create = mpol_new_nodemask, |
b27abacc DH |
415 | .rebind = mpol_rebind_preferred, |
416 | }, | |
37012946 DR |
417 | }; |
418 | ||
4a64981d | 419 | static int migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
fc301289 | 420 | unsigned long flags); |
1a75a6c8 | 421 | |
6f4576e3 NH |
422 | struct queue_pages { |
423 | struct list_head *pagelist; | |
424 | unsigned long flags; | |
425 | nodemask_t *nmask; | |
f18da660 LX |
426 | unsigned long start; |
427 | unsigned long end; | |
428 | struct vm_area_struct *first; | |
24526268 | 429 | bool has_unmovable; |
6f4576e3 NH |
430 | }; |
431 | ||
88aaa2a1 | 432 | /* |
d451b89d | 433 | * Check if the folio's nid is in qp->nmask. |
88aaa2a1 NH |
434 | * |
435 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
436 | * in the invert of qp->nmask. | |
437 | */ | |
d451b89d | 438 | static inline bool queue_folio_required(struct folio *folio, |
88aaa2a1 NH |
439 | struct queue_pages *qp) |
440 | { | |
d451b89d | 441 | int nid = folio_nid(folio); |
88aaa2a1 NH |
442 | unsigned long flags = qp->flags; |
443 | ||
444 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
445 | } | |
446 | ||
a7f40cfe | 447 | /* |
de1f5055 VMO |
448 | * queue_folios_pmd() has three possible return values: |
449 | * 0 - folios are placed on the right node or queued successfully, or | |
24526268 YS |
450 | * special page is met, i.e. zero page, or unmovable page is found |
451 | * but continue walking (indicated by queue_pages.has_unmovable). | |
d8835445 | 452 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an |
de1f5055 | 453 | * existing folio was already on a node that does not follow the |
d8835445 | 454 | * policy. |
a7f40cfe | 455 | */ |
de1f5055 | 456 | static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
c8633798 | 457 | unsigned long end, struct mm_walk *walk) |
959a7e13 | 458 | __releases(ptl) |
c8633798 NH |
459 | { |
460 | int ret = 0; | |
de1f5055 | 461 | struct folio *folio; |
c8633798 NH |
462 | struct queue_pages *qp = walk->private; |
463 | unsigned long flags; | |
464 | ||
465 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 466 | ret = -EIO; |
c8633798 NH |
467 | goto unlock; |
468 | } | |
de1f5055 VMO |
469 | folio = pfn_folio(pmd_pfn(*pmd)); |
470 | if (is_huge_zero_page(&folio->page)) { | |
e5947d23 | 471 | walk->action = ACTION_CONTINUE; |
6d97cf88 | 472 | goto unlock; |
c8633798 | 473 | } |
d451b89d | 474 | if (!queue_folio_required(folio, qp)) |
c8633798 | 475 | goto unlock; |
c8633798 | 476 | |
c8633798 | 477 | flags = qp->flags; |
de1f5055 | 478 | /* go to folio migration */ |
a7f40cfe | 479 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 | 480 | if (!vma_migratable(walk->vma) || |
4a64981d | 481 | migrate_folio_add(folio, qp->pagelist, flags)) { |
24526268 | 482 | qp->has_unmovable = true; |
a7f40cfe YS |
483 | goto unlock; |
484 | } | |
a7f40cfe YS |
485 | } else |
486 | ret = -EIO; | |
c8633798 NH |
487 | unlock: |
488 | spin_unlock(ptl); | |
c8633798 NH |
489 | return ret; |
490 | } | |
491 | ||
98094945 NH |
492 | /* |
493 | * Scan through pages checking if pages follow certain conditions, | |
494 | * and move them to the pagelist if they do. | |
d8835445 | 495 | * |
3dae02bb VMO |
496 | * queue_folios_pte_range() has three possible return values: |
497 | * 0 - folios are placed on the right node or queued successfully, or | |
24526268 YS |
498 | * special page is met, i.e. zero page, or unmovable page is found |
499 | * but continue walking (indicated by queue_pages.has_unmovable). | |
3dae02bb | 500 | * -EIO - only MPOL_MF_STRICT was specified and an existing folio was already |
d8835445 | 501 | * on a node that does not follow the policy. |
98094945 | 502 | */ |
3dae02bb | 503 | static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr, |
6f4576e3 | 504 | unsigned long end, struct mm_walk *walk) |
1da177e4 | 505 | { |
6f4576e3 | 506 | struct vm_area_struct *vma = walk->vma; |
3dae02bb | 507 | struct folio *folio; |
6f4576e3 NH |
508 | struct queue_pages *qp = walk->private; |
509 | unsigned long flags = qp->flags; | |
3f088420 | 510 | pte_t *pte, *mapped_pte; |
c33c7948 | 511 | pte_t ptent; |
705e87c0 | 512 | spinlock_t *ptl; |
941150a3 | 513 | |
c8633798 | 514 | ptl = pmd_trans_huge_lock(pmd, vma); |
bc78b5ed | 515 | if (ptl) |
de1f5055 | 516 | return queue_folios_pmd(pmd, ptl, addr, end, walk); |
91612e0d | 517 | |
3f088420 | 518 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
7780d040 HD |
519 | if (!pte) { |
520 | walk->action = ACTION_AGAIN; | |
521 | return 0; | |
522 | } | |
6f4576e3 | 523 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
c33c7948 RR |
524 | ptent = ptep_get(pte); |
525 | if (!pte_present(ptent)) | |
1da177e4 | 526 | continue; |
c33c7948 | 527 | folio = vm_normal_folio(vma, addr, ptent); |
3dae02bb | 528 | if (!folio || folio_is_zone_device(folio)) |
1da177e4 | 529 | continue; |
053837fc | 530 | /* |
3dae02bb VMO |
531 | * vm_normal_folio() filters out zero pages, but there might |
532 | * still be reserved folios to skip, perhaps in a VDSO. | |
053837fc | 533 | */ |
3dae02bb | 534 | if (folio_test_reserved(folio)) |
f4598c8b | 535 | continue; |
d451b89d | 536 | if (!queue_folio_required(folio, qp)) |
38e35860 | 537 | continue; |
a7f40cfe | 538 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
24526268 YS |
539 | /* |
540 | * MPOL_MF_STRICT must be specified if we get here. | |
541 | * Continue walking vmas due to MPOL_MF_MOVE* flags. | |
542 | */ | |
543 | if (!vma_migratable(vma)) | |
544 | qp->has_unmovable = true; | |
a53190a4 YS |
545 | |
546 | /* | |
547 | * Do not abort immediately since there may be | |
548 | * temporary off LRU pages in the range. Still | |
549 | * need migrate other LRU pages. | |
550 | */ | |
4a64981d | 551 | if (migrate_folio_add(folio, qp->pagelist, flags)) |
24526268 | 552 | qp->has_unmovable = true; |
a7f40cfe YS |
553 | } else |
554 | break; | |
6f4576e3 | 555 | } |
3f088420 | 556 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 557 | cond_resched(); |
d8835445 | 558 | |
a7f40cfe | 559 | return addr != end ? -EIO : 0; |
91612e0d HD |
560 | } |
561 | ||
0a2c1e81 | 562 | static int queue_folios_hugetlb(pte_t *pte, unsigned long hmask, |
6f4576e3 NH |
563 | unsigned long addr, unsigned long end, |
564 | struct mm_walk *walk) | |
e2d8cf40 | 565 | { |
dcf17635 | 566 | int ret = 0; |
e2d8cf40 | 567 | #ifdef CONFIG_HUGETLB_PAGE |
6f4576e3 | 568 | struct queue_pages *qp = walk->private; |
dcf17635 | 569 | unsigned long flags = (qp->flags & MPOL_MF_VALID); |
0a2c1e81 | 570 | struct folio *folio; |
cb900f41 | 571 | spinlock_t *ptl; |
d4c54919 | 572 | pte_t entry; |
e2d8cf40 | 573 | |
6f4576e3 NH |
574 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
575 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
576 | if (!pte_present(entry)) |
577 | goto unlock; | |
0a2c1e81 | 578 | folio = pfn_folio(pte_pfn(entry)); |
d451b89d | 579 | if (!queue_folio_required(folio, qp)) |
e2d8cf40 | 580 | goto unlock; |
dcf17635 LX |
581 | |
582 | if (flags == MPOL_MF_STRICT) { | |
583 | /* | |
0a2c1e81 | 584 | * STRICT alone means only detecting misplaced folio and no |
dcf17635 LX |
585 | * need to further check other vma. |
586 | */ | |
587 | ret = -EIO; | |
588 | goto unlock; | |
589 | } | |
590 | ||
591 | if (!vma_migratable(walk->vma)) { | |
592 | /* | |
593 | * Must be STRICT with MOVE*, otherwise .test_walk() have | |
594 | * stopped walking current vma. | |
0a2c1e81 | 595 | * Detecting misplaced folio but allow migrating folios which |
dcf17635 LX |
596 | * have been queued. |
597 | */ | |
24526268 | 598 | qp->has_unmovable = true; |
dcf17635 LX |
599 | goto unlock; |
600 | } | |
601 | ||
0a2c1e81 VMO |
602 | /* |
603 | * With MPOL_MF_MOVE, we try to migrate only unshared folios. If it | |
604 | * is shared it is likely not worth migrating. | |
605 | * | |
606 | * To check if the folio is shared, ideally we want to make sure | |
607 | * every page is mapped to the same process. Doing that is very | |
608 | * expensive, so check the estimated mapcount of the folio instead. | |
609 | */ | |
e2d8cf40 | 610 | if (flags & (MPOL_MF_MOVE_ALL) || |
0a2c1e81 | 611 | (flags & MPOL_MF_MOVE && folio_estimated_sharers(folio) == 1 && |
73bdf65e | 612 | !hugetlb_pmd_shared(pte))) { |
9747b9e9 | 613 | if (!isolate_hugetlb(folio, qp->pagelist) && |
dcf17635 LX |
614 | (flags & MPOL_MF_STRICT)) |
615 | /* | |
0a2c1e81 | 616 | * Failed to isolate folio but allow migrating pages |
dcf17635 LX |
617 | * which have been queued. |
618 | */ | |
24526268 | 619 | qp->has_unmovable = true; |
dcf17635 | 620 | } |
e2d8cf40 | 621 | unlock: |
cb900f41 | 622 | spin_unlock(ptl); |
e2d8cf40 NH |
623 | #else |
624 | BUG(); | |
625 | #endif | |
dcf17635 | 626 | return ret; |
1da177e4 LT |
627 | } |
628 | ||
5877231f | 629 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 630 | /* |
4b10e7d5 MG |
631 | * This is used to mark a range of virtual addresses to be inaccessible. |
632 | * These are later cleared by a NUMA hinting fault. Depending on these | |
633 | * faults, pages may be migrated for better NUMA placement. | |
634 | * | |
635 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
636 | * an architecture makes a different choice, it will need further | |
637 | * changes to the core. | |
b24f53a0 | 638 | */ |
4b10e7d5 MG |
639 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
640 | unsigned long addr, unsigned long end) | |
b24f53a0 | 641 | { |
4a18419f | 642 | struct mmu_gather tlb; |
a79390f5 | 643 | long nr_updated; |
b24f53a0 | 644 | |
4a18419f NA |
645 | tlb_gather_mmu(&tlb, vma->vm_mm); |
646 | ||
1ef488ed | 647 | nr_updated = change_protection(&tlb, vma, addr, end, MM_CP_PROT_NUMA); |
d1751118 | 648 | if (nr_updated > 0) |
03c5a6e1 | 649 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); |
b24f53a0 | 650 | |
4a18419f NA |
651 | tlb_finish_mmu(&tlb); |
652 | ||
4b10e7d5 | 653 | return nr_updated; |
b24f53a0 LS |
654 | } |
655 | #else | |
656 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
657 | unsigned long addr, unsigned long end) | |
658 | { | |
659 | return 0; | |
660 | } | |
5877231f | 661 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 662 | |
6f4576e3 NH |
663 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
664 | struct mm_walk *walk) | |
665 | { | |
66850be5 | 666 | struct vm_area_struct *next, *vma = walk->vma; |
6f4576e3 NH |
667 | struct queue_pages *qp = walk->private; |
668 | unsigned long endvma = vma->vm_end; | |
669 | unsigned long flags = qp->flags; | |
670 | ||
a18b3ac2 | 671 | /* range check first */ |
ce33135c | 672 | VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma); |
f18da660 LX |
673 | |
674 | if (!qp->first) { | |
675 | qp->first = vma; | |
676 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
677 | (qp->start < vma->vm_start)) | |
678 | /* hole at head side of range */ | |
a18b3ac2 LX |
679 | return -EFAULT; |
680 | } | |
66850be5 | 681 | next = find_vma(vma->vm_mm, vma->vm_end); |
f18da660 LX |
682 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
683 | ((vma->vm_end < qp->end) && | |
66850be5 | 684 | (!next || vma->vm_end < next->vm_start))) |
f18da660 LX |
685 | /* hole at middle or tail of range */ |
686 | return -EFAULT; | |
a18b3ac2 | 687 | |
a7f40cfe YS |
688 | /* |
689 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
690 | * regardless of vma_migratable | |
691 | */ | |
692 | if (!vma_migratable(vma) && | |
693 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
694 | return 1; |
695 | ||
6f4576e3 NH |
696 | if (endvma > end) |
697 | endvma = end; | |
6f4576e3 | 698 | |
6f4576e3 NH |
699 | if (flags & MPOL_MF_LAZY) { |
700 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
3122e80e | 701 | if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) && |
4355c018 | 702 | !(vma->vm_flags & VM_MIXEDMAP)) |
6f4576e3 NH |
703 | change_prot_numa(vma, start, endvma); |
704 | return 1; | |
705 | } | |
706 | ||
77bf45e7 | 707 | /* queue pages from current vma */ |
a7f40cfe | 708 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
709 | return 0; |
710 | return 1; | |
711 | } | |
712 | ||
7b86ac33 | 713 | static const struct mm_walk_ops queue_pages_walk_ops = { |
0a2c1e81 | 714 | .hugetlb_entry = queue_folios_hugetlb, |
3dae02bb | 715 | .pmd_entry = queue_folios_pte_range, |
7b86ac33 | 716 | .test_walk = queue_pages_test_walk, |
49b06385 SB |
717 | .walk_lock = PGWALK_RDLOCK, |
718 | }; | |
719 | ||
720 | static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = { | |
721 | .hugetlb_entry = queue_folios_hugetlb, | |
722 | .pmd_entry = queue_folios_pte_range, | |
723 | .test_walk = queue_pages_test_walk, | |
724 | .walk_lock = PGWALK_WRLOCK, | |
7b86ac33 CH |
725 | }; |
726 | ||
dc9aa5b9 | 727 | /* |
98094945 NH |
728 | * Walk through page tables and collect pages to be migrated. |
729 | * | |
730 | * If pages found in a given range are on a set of nodes (determined by | |
731 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
732 | * passed via @private. |
733 | * | |
734 | * queue_pages_range() has three possible return values: | |
735 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
736 | * specified. | |
737 | * 0 - queue pages successfully or no misplaced page. | |
a85dfc30 YS |
738 | * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or |
739 | * memory range specified by nodemask and maxnode points outside | |
740 | * your accessible address space (-EFAULT) | |
dc9aa5b9 | 741 | */ |
d05f0cdc | 742 | static int |
98094945 | 743 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 | 744 | nodemask_t *nodes, unsigned long flags, |
49b06385 | 745 | struct list_head *pagelist, bool lock_vma) |
1da177e4 | 746 | { |
f18da660 | 747 | int err; |
6f4576e3 NH |
748 | struct queue_pages qp = { |
749 | .pagelist = pagelist, | |
750 | .flags = flags, | |
751 | .nmask = nodes, | |
f18da660 LX |
752 | .start = start, |
753 | .end = end, | |
754 | .first = NULL, | |
24526268 | 755 | .has_unmovable = false, |
6f4576e3 | 756 | }; |
49b06385 SB |
757 | const struct mm_walk_ops *ops = lock_vma ? |
758 | &queue_pages_lock_vma_walk_ops : &queue_pages_walk_ops; | |
6f4576e3 | 759 | |
49b06385 | 760 | err = walk_page_range(mm, start, end, ops, &qp); |
f18da660 | 761 | |
24526268 YS |
762 | if (qp.has_unmovable) |
763 | err = 1; | |
f18da660 LX |
764 | if (!qp.first) |
765 | /* whole range in hole */ | |
766 | err = -EFAULT; | |
767 | ||
768 | return err; | |
1da177e4 LT |
769 | } |
770 | ||
869833f2 KM |
771 | /* |
772 | * Apply policy to a single VMA | |
c1e8d7c6 | 773 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
774 | */ |
775 | static int vma_replace_policy(struct vm_area_struct *vma, | |
776 | struct mempolicy *pol) | |
8d34694c | 777 | { |
869833f2 KM |
778 | int err; |
779 | struct mempolicy *old; | |
780 | struct mempolicy *new; | |
8d34694c | 781 | |
6c21e066 JH |
782 | vma_assert_write_locked(vma); |
783 | ||
8d34694c KM |
784 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
785 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
786 | vma->vm_ops, vma->vm_file, | |
787 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
788 | ||
869833f2 KM |
789 | new = mpol_dup(pol); |
790 | if (IS_ERR(new)) | |
791 | return PTR_ERR(new); | |
792 | ||
793 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 794 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
795 | if (err) |
796 | goto err_out; | |
8d34694c | 797 | } |
869833f2 KM |
798 | |
799 | old = vma->vm_policy; | |
c1e8d7c6 | 800 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
801 | mpol_put(old); |
802 | ||
803 | return 0; | |
804 | err_out: | |
805 | mpol_put(new); | |
8d34694c KM |
806 | return err; |
807 | } | |
808 | ||
f4e9e0e6 LH |
809 | /* Split or merge the VMA (if required) and apply the new policy */ |
810 | static int mbind_range(struct vma_iterator *vmi, struct vm_area_struct *vma, | |
811 | struct vm_area_struct **prev, unsigned long start, | |
812 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 | 813 | { |
f4e9e0e6 LH |
814 | struct vm_area_struct *merged; |
815 | unsigned long vmstart, vmend; | |
e26a5114 | 816 | pgoff_t pgoff; |
f4e9e0e6 | 817 | int err; |
9d8cebd4 | 818 | |
f4e9e0e6 LH |
819 | vmend = min(end, vma->vm_end); |
820 | if (start > vma->vm_start) { | |
821 | *prev = vma; | |
822 | vmstart = start; | |
823 | } else { | |
824 | vmstart = vma->vm_start; | |
825 | } | |
826 | ||
00ca0f2e LS |
827 | if (mpol_equal(vma_policy(vma), new_pol)) { |
828 | *prev = vma; | |
7329e3eb | 829 | return 0; |
00ca0f2e | 830 | } |
7329e3eb | 831 | |
f4e9e0e6 LH |
832 | pgoff = vma->vm_pgoff + ((vmstart - vma->vm_start) >> PAGE_SHIFT); |
833 | merged = vma_merge(vmi, vma->vm_mm, *prev, vmstart, vmend, vma->vm_flags, | |
834 | vma->anon_vma, vma->vm_file, pgoff, new_pol, | |
835 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); | |
836 | if (merged) { | |
837 | *prev = merged; | |
838 | return vma_replace_policy(merged, new_pol); | |
839 | } | |
840 | ||
841 | if (vma->vm_start != vmstart) { | |
842 | err = split_vma(vmi, vma, vmstart, 1); | |
8d34694c | 843 | if (err) |
f4e9e0e6 LH |
844 | return err; |
845 | } | |
9d8cebd4 | 846 | |
f4e9e0e6 LH |
847 | if (vma->vm_end != vmend) { |
848 | err = split_vma(vmi, vma, vmend, 0); | |
849 | if (err) | |
850 | return err; | |
851 | } | |
852 | ||
853 | *prev = vma; | |
854 | return vma_replace_policy(vma, new_pol); | |
1da177e4 LT |
855 | } |
856 | ||
1da177e4 | 857 | /* Set the process memory policy */ |
028fec41 DR |
858 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
859 | nodemask_t *nodes) | |
1da177e4 | 860 | { |
58568d2a | 861 | struct mempolicy *new, *old; |
4bfc4495 | 862 | NODEMASK_SCRATCH(scratch); |
58568d2a | 863 | int ret; |
1da177e4 | 864 | |
4bfc4495 KH |
865 | if (!scratch) |
866 | return -ENOMEM; | |
f4e53d91 | 867 | |
4bfc4495 KH |
868 | new = mpol_new(mode, flags, nodes); |
869 | if (IS_ERR(new)) { | |
870 | ret = PTR_ERR(new); | |
871 | goto out; | |
872 | } | |
2c7c3a7d | 873 | |
12c1dc8e | 874 | task_lock(current); |
4bfc4495 | 875 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 876 | if (ret) { |
12c1dc8e | 877 | task_unlock(current); |
58568d2a | 878 | mpol_put(new); |
4bfc4495 | 879 | goto out; |
58568d2a | 880 | } |
12c1dc8e | 881 | |
58568d2a | 882 | old = current->mempolicy; |
1da177e4 | 883 | current->mempolicy = new; |
45816682 VB |
884 | if (new && new->mode == MPOL_INTERLEAVE) |
885 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 886 | task_unlock(current); |
58568d2a | 887 | mpol_put(old); |
4bfc4495 KH |
888 | ret = 0; |
889 | out: | |
890 | NODEMASK_SCRATCH_FREE(scratch); | |
891 | return ret; | |
1da177e4 LT |
892 | } |
893 | ||
bea904d5 LS |
894 | /* |
895 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
896 | * |
897 | * Called with task's alloc_lock held | |
bea904d5 LS |
898 | */ |
899 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 900 | { |
dfcd3c0d | 901 | nodes_clear(*nodes); |
bea904d5 LS |
902 | if (p == &default_policy) |
903 | return; | |
904 | ||
45c4745a | 905 | switch (p->mode) { |
19770b32 | 906 | case MPOL_BIND: |
1da177e4 | 907 | case MPOL_INTERLEAVE: |
269fbe72 | 908 | case MPOL_PREFERRED: |
b27abacc | 909 | case MPOL_PREFERRED_MANY: |
269fbe72 | 910 | *nodes = p->nodes; |
1da177e4 | 911 | break; |
7858d7bc FT |
912 | case MPOL_LOCAL: |
913 | /* return empty node mask for local allocation */ | |
914 | break; | |
1da177e4 LT |
915 | default: |
916 | BUG(); | |
917 | } | |
918 | } | |
919 | ||
3b9aadf7 | 920 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 921 | { |
ba841078 | 922 | struct page *p = NULL; |
f728b9c4 | 923 | int ret; |
1da177e4 | 924 | |
f728b9c4 JH |
925 | ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p); |
926 | if (ret > 0) { | |
927 | ret = page_to_nid(p); | |
1da177e4 LT |
928 | put_page(p); |
929 | } | |
f728b9c4 | 930 | return ret; |
1da177e4 LT |
931 | } |
932 | ||
1da177e4 | 933 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
934 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
935 | unsigned long addr, unsigned long flags) | |
1da177e4 | 936 | { |
8bccd85f | 937 | int err; |
1da177e4 LT |
938 | struct mm_struct *mm = current->mm; |
939 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 940 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 941 | |
754af6f5 LS |
942 | if (flags & |
943 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 944 | return -EINVAL; |
754af6f5 LS |
945 | |
946 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
947 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
948 | return -EINVAL; | |
949 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 950 | task_lock(current); |
754af6f5 | 951 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 952 | task_unlock(current); |
754af6f5 LS |
953 | return 0; |
954 | } | |
955 | ||
1da177e4 | 956 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
957 | /* |
958 | * Do NOT fall back to task policy if the | |
959 | * vma/shared policy at addr is NULL. We | |
960 | * want to return MPOL_DEFAULT in this case. | |
961 | */ | |
d8ed45c5 | 962 | mmap_read_lock(mm); |
33e3575c | 963 | vma = vma_lookup(mm, addr); |
1da177e4 | 964 | if (!vma) { |
d8ed45c5 | 965 | mmap_read_unlock(mm); |
1da177e4 LT |
966 | return -EFAULT; |
967 | } | |
968 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
969 | pol = vma->vm_ops->get_policy(vma, addr); | |
970 | else | |
971 | pol = vma->vm_policy; | |
972 | } else if (addr) | |
973 | return -EINVAL; | |
974 | ||
975 | if (!pol) | |
bea904d5 | 976 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
977 | |
978 | if (flags & MPOL_F_NODE) { | |
979 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 | 980 | /* |
f728b9c4 JH |
981 | * Take a refcount on the mpol, because we are about to |
982 | * drop the mmap_lock, after which only "pol" remains | |
983 | * valid, "vma" is stale. | |
3b9aadf7 AA |
984 | */ |
985 | pol_refcount = pol; | |
986 | vma = NULL; | |
987 | mpol_get(pol); | |
f728b9c4 | 988 | mmap_read_unlock(mm); |
3b9aadf7 | 989 | err = lookup_node(mm, addr); |
1da177e4 LT |
990 | if (err < 0) |
991 | goto out; | |
8bccd85f | 992 | *policy = err; |
1da177e4 | 993 | } else if (pol == current->mempolicy && |
45c4745a | 994 | pol->mode == MPOL_INTERLEAVE) { |
269fbe72 | 995 | *policy = next_node_in(current->il_prev, pol->nodes); |
1da177e4 LT |
996 | } else { |
997 | err = -EINVAL; | |
998 | goto out; | |
999 | } | |
bea904d5 LS |
1000 | } else { |
1001 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
1002 | pol->mode; | |
d79df630 DR |
1003 | /* |
1004 | * Internal mempolicy flags must be masked off before exposing | |
1005 | * the policy to userspace. | |
1006 | */ | |
1007 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1008 | } |
1da177e4 | 1009 | |
1da177e4 | 1010 | err = 0; |
58568d2a | 1011 | if (nmask) { |
c6b6ef8b LS |
1012 | if (mpol_store_user_nodemask(pol)) { |
1013 | *nmask = pol->w.user_nodemask; | |
1014 | } else { | |
1015 | task_lock(current); | |
1016 | get_policy_nodemask(pol, nmask); | |
1017 | task_unlock(current); | |
1018 | } | |
58568d2a | 1019 | } |
1da177e4 LT |
1020 | |
1021 | out: | |
52cd3b07 | 1022 | mpol_cond_put(pol); |
1da177e4 | 1023 | if (vma) |
d8ed45c5 | 1024 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1025 | if (pol_refcount) |
1026 | mpol_put(pol_refcount); | |
1da177e4 LT |
1027 | return err; |
1028 | } | |
1029 | ||
b20a3503 | 1030 | #ifdef CONFIG_MIGRATION |
4a64981d | 1031 | static int migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
fc301289 | 1032 | unsigned long flags) |
6ce3c4c0 CL |
1033 | { |
1034 | /* | |
4a64981d VMO |
1035 | * We try to migrate only unshared folios. If it is shared it |
1036 | * is likely not worth migrating. | |
1037 | * | |
1038 | * To check if the folio is shared, ideally we want to make sure | |
1039 | * every page is mapped to the same process. Doing that is very | |
1040 | * expensive, so check the estimated mapcount of the folio instead. | |
6ce3c4c0 | 1041 | */ |
4a64981d | 1042 | if ((flags & MPOL_MF_MOVE_ALL) || folio_estimated_sharers(folio) == 1) { |
be2d5756 | 1043 | if (folio_isolate_lru(folio)) { |
4a64981d VMO |
1044 | list_add_tail(&folio->lru, foliolist); |
1045 | node_stat_mod_folio(folio, | |
1046 | NR_ISOLATED_ANON + folio_is_file_lru(folio), | |
1047 | folio_nr_pages(folio)); | |
a53190a4 YS |
1048 | } else if (flags & MPOL_MF_STRICT) { |
1049 | /* | |
4a64981d VMO |
1050 | * Non-movable folio may reach here. And, there may be |
1051 | * temporary off LRU folios or non-LRU movable folios. | |
1052 | * Treat them as unmovable folios since they can't be | |
a53190a4 YS |
1053 | * isolated, so they can't be moved at the moment. It |
1054 | * should return -EIO for this case too. | |
1055 | */ | |
1056 | return -EIO; | |
62695a84 NP |
1057 | } |
1058 | } | |
a53190a4 YS |
1059 | |
1060 | return 0; | |
7e2ab150 | 1061 | } |
6ce3c4c0 | 1062 | |
7e2ab150 CL |
1063 | /* |
1064 | * Migrate pages from one node to a target node. | |
1065 | * Returns error or the number of pages not migrated. | |
1066 | */ | |
dbcb0f19 AB |
1067 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1068 | int flags) | |
7e2ab150 CL |
1069 | { |
1070 | nodemask_t nmask; | |
66850be5 | 1071 | struct vm_area_struct *vma; |
7e2ab150 CL |
1072 | LIST_HEAD(pagelist); |
1073 | int err = 0; | |
a0976311 JK |
1074 | struct migration_target_control mtc = { |
1075 | .nid = dest, | |
1076 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1077 | }; | |
7e2ab150 CL |
1078 | |
1079 | nodes_clear(nmask); | |
1080 | node_set(source, nmask); | |
6ce3c4c0 | 1081 | |
08270807 MK |
1082 | /* |
1083 | * This does not "check" the range but isolates all pages that | |
1084 | * need migration. Between passing in the full user address | |
1085 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1086 | */ | |
66850be5 | 1087 | vma = find_vma(mm, 0); |
08270807 | 1088 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); |
66850be5 | 1089 | queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask, |
49b06385 | 1090 | flags | MPOL_MF_DISCONTIG_OK, &pagelist, false); |
7e2ab150 | 1091 | |
cf608ac1 | 1092 | if (!list_empty(&pagelist)) { |
a0976311 | 1093 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
5ac95884 | 1094 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
cf608ac1 | 1095 | if (err) |
e2d8cf40 | 1096 | putback_movable_pages(&pagelist); |
cf608ac1 | 1097 | } |
95a402c3 | 1098 | |
7e2ab150 | 1099 | return err; |
6ce3c4c0 CL |
1100 | } |
1101 | ||
39743889 | 1102 | /* |
7e2ab150 CL |
1103 | * Move pages between the two nodesets so as to preserve the physical |
1104 | * layout as much as possible. | |
39743889 CL |
1105 | * |
1106 | * Returns the number of page that could not be moved. | |
1107 | */ | |
0ce72d4f AM |
1108 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1109 | const nodemask_t *to, int flags) | |
39743889 | 1110 | { |
7e2ab150 | 1111 | int busy = 0; |
f555befd | 1112 | int err = 0; |
7e2ab150 | 1113 | nodemask_t tmp; |
39743889 | 1114 | |
361a2a22 | 1115 | lru_cache_disable(); |
0aedadf9 | 1116 | |
d8ed45c5 | 1117 | mmap_read_lock(mm); |
39743889 | 1118 | |
da0aa138 KM |
1119 | /* |
1120 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1121 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1122 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1123 | * The pair of nodemasks 'to' and 'from' define the map. | |
1124 | * | |
1125 | * If no pair of bits is found that way, fallback to picking some | |
1126 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1127 | * 'source' and 'dest' bits are the same, this represents a node | |
1128 | * that will be migrating to itself, so no pages need move. | |
1129 | * | |
1130 | * If no bits are left in 'tmp', or if all remaining bits left | |
1131 | * in 'tmp' correspond to the same bit in 'to', return false | |
1132 | * (nothing left to migrate). | |
1133 | * | |
1134 | * This lets us pick a pair of nodes to migrate between, such that | |
1135 | * if possible the dest node is not already occupied by some other | |
1136 | * source node, minimizing the risk of overloading the memory on a | |
1137 | * node that would happen if we migrated incoming memory to a node | |
1138 | * before migrating outgoing memory source that same node. | |
1139 | * | |
1140 | * A single scan of tmp is sufficient. As we go, we remember the | |
1141 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1142 | * that not only moved, but what's better, moved to an empty slot | |
1143 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1144 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1145 | * most recent <s, d> pair that moved. If we get all the way through |
1146 | * the scan of tmp without finding any node that moved, much less | |
1147 | * moved to an empty node, then there is nothing left worth migrating. | |
1148 | */ | |
d4984711 | 1149 | |
0ce72d4f | 1150 | tmp = *from; |
7e2ab150 | 1151 | while (!nodes_empty(tmp)) { |
68d68ff6 | 1152 | int s, d; |
b76ac7e7 | 1153 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1154 | int dest = 0; |
1155 | ||
1156 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1157 | |
1158 | /* | |
1159 | * do_migrate_pages() tries to maintain the relative | |
1160 | * node relationship of the pages established between | |
1161 | * threads and memory areas. | |
1162 | * | |
1163 | * However if the number of source nodes is not equal to | |
1164 | * the number of destination nodes we can not preserve | |
1165 | * this node relative relationship. In that case, skip | |
1166 | * copying memory from a node that is in the destination | |
1167 | * mask. | |
1168 | * | |
1169 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1170 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1171 | */ | |
1172 | ||
0ce72d4f AM |
1173 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1174 | (node_isset(s, *to))) | |
4a5b18cc LW |
1175 | continue; |
1176 | ||
0ce72d4f | 1177 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1178 | if (s == d) |
1179 | continue; | |
1180 | ||
1181 | source = s; /* Node moved. Memorize */ | |
1182 | dest = d; | |
1183 | ||
1184 | /* dest not in remaining from nodes? */ | |
1185 | if (!node_isset(dest, tmp)) | |
1186 | break; | |
1187 | } | |
b76ac7e7 | 1188 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1189 | break; |
1190 | ||
1191 | node_clear(source, tmp); | |
1192 | err = migrate_to_node(mm, source, dest, flags); | |
1193 | if (err > 0) | |
1194 | busy += err; | |
1195 | if (err < 0) | |
1196 | break; | |
39743889 | 1197 | } |
d8ed45c5 | 1198 | mmap_read_unlock(mm); |
d479960e | 1199 | |
361a2a22 | 1200 | lru_cache_enable(); |
7e2ab150 CL |
1201 | if (err < 0) |
1202 | return err; | |
1203 | return busy; | |
b20a3503 CL |
1204 | |
1205 | } | |
1206 | ||
3ad33b24 LS |
1207 | /* |
1208 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1209 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1210 | * Search forward from there, if not. N.B., this assumes that the |
1211 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1212 | * is in virtual address order. | |
1213 | */ | |
4e096ae1 | 1214 | static struct folio *new_folio(struct folio *src, unsigned long start) |
95a402c3 | 1215 | { |
d05f0cdc | 1216 | struct vm_area_struct *vma; |
3f649ab7 | 1217 | unsigned long address; |
66850be5 | 1218 | VMA_ITERATOR(vmi, current->mm, start); |
ec4858e0 | 1219 | gfp_t gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL; |
95a402c3 | 1220 | |
66850be5 | 1221 | for_each_vma(vmi, vma) { |
4e096ae1 | 1222 | address = page_address_in_vma(&src->page, vma); |
3ad33b24 LS |
1223 | if (address != -EFAULT) |
1224 | break; | |
3ad33b24 | 1225 | } |
11c731e8 | 1226 | |
d0ce0e47 | 1227 | if (folio_test_hugetlb(src)) { |
4e096ae1 | 1228 | return alloc_hugetlb_folio_vma(folio_hstate(src), |
389c8178 | 1229 | vma, address); |
d0ce0e47 | 1230 | } |
ec4858e0 MWO |
1231 | |
1232 | if (folio_test_large(src)) | |
1233 | gfp = GFP_TRANSHUGE; | |
1234 | ||
0bf598d8 | 1235 | /* |
ec4858e0 | 1236 | * if !vma, vma_alloc_folio() will use task or system default policy |
0bf598d8 | 1237 | */ |
4e096ae1 | 1238 | return vma_alloc_folio(gfp, folio_order(src), vma, address, |
ec4858e0 | 1239 | folio_test_large(src)); |
95a402c3 | 1240 | } |
b20a3503 CL |
1241 | #else |
1242 | ||
4a64981d | 1243 | static int migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
b20a3503 CL |
1244 | unsigned long flags) |
1245 | { | |
a53190a4 | 1246 | return -EIO; |
39743889 CL |
1247 | } |
1248 | ||
0ce72d4f AM |
1249 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1250 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1251 | { |
1252 | return -ENOSYS; | |
1253 | } | |
95a402c3 | 1254 | |
4e096ae1 | 1255 | static struct folio *new_folio(struct folio *src, unsigned long start) |
95a402c3 CL |
1256 | { |
1257 | return NULL; | |
1258 | } | |
b20a3503 CL |
1259 | #endif |
1260 | ||
dbcb0f19 | 1261 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1262 | unsigned short mode, unsigned short mode_flags, |
1263 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1264 | { |
6ce3c4c0 | 1265 | struct mm_struct *mm = current->mm; |
f4e9e0e6 LH |
1266 | struct vm_area_struct *vma, *prev; |
1267 | struct vma_iterator vmi; | |
6ce3c4c0 CL |
1268 | struct mempolicy *new; |
1269 | unsigned long end; | |
1270 | int err; | |
d8835445 | 1271 | int ret; |
6ce3c4c0 CL |
1272 | LIST_HEAD(pagelist); |
1273 | ||
b24f53a0 | 1274 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1275 | return -EINVAL; |
74c00241 | 1276 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1277 | return -EPERM; |
1278 | ||
1279 | if (start & ~PAGE_MASK) | |
1280 | return -EINVAL; | |
1281 | ||
1282 | if (mode == MPOL_DEFAULT) | |
1283 | flags &= ~MPOL_MF_STRICT; | |
1284 | ||
aaa31e05 | 1285 | len = PAGE_ALIGN(len); |
6ce3c4c0 CL |
1286 | end = start + len; |
1287 | ||
1288 | if (end < start) | |
1289 | return -EINVAL; | |
1290 | if (end == start) | |
1291 | return 0; | |
1292 | ||
028fec41 | 1293 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1294 | if (IS_ERR(new)) |
1295 | return PTR_ERR(new); | |
1296 | ||
b24f53a0 LS |
1297 | if (flags & MPOL_MF_LAZY) |
1298 | new->flags |= MPOL_F_MOF; | |
1299 | ||
6ce3c4c0 CL |
1300 | /* |
1301 | * If we are using the default policy then operation | |
1302 | * on discontinuous address spaces is okay after all | |
1303 | */ | |
1304 | if (!new) | |
1305 | flags |= MPOL_MF_DISCONTIG_OK; | |
1306 | ||
028fec41 DR |
1307 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1308 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1309 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1310 | |
0aedadf9 CL |
1311 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1312 | ||
361a2a22 | 1313 | lru_cache_disable(); |
0aedadf9 | 1314 | } |
4bfc4495 KH |
1315 | { |
1316 | NODEMASK_SCRATCH(scratch); | |
1317 | if (scratch) { | |
d8ed45c5 | 1318 | mmap_write_lock(mm); |
4bfc4495 | 1319 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1320 | if (err) |
d8ed45c5 | 1321 | mmap_write_unlock(mm); |
4bfc4495 KH |
1322 | } else |
1323 | err = -ENOMEM; | |
1324 | NODEMASK_SCRATCH_FREE(scratch); | |
1325 | } | |
b05ca738 KM |
1326 | if (err) |
1327 | goto mpol_out; | |
1328 | ||
6c21e066 JH |
1329 | /* |
1330 | * Lock the VMAs before scanning for pages to migrate, to ensure we don't | |
1331 | * miss a concurrently inserted page. | |
1332 | */ | |
d8835445 | 1333 | ret = queue_pages_range(mm, start, end, nmask, |
49b06385 | 1334 | flags | MPOL_MF_INVERT, &pagelist, true); |
d8835445 YS |
1335 | |
1336 | if (ret < 0) { | |
a85dfc30 | 1337 | err = ret; |
d8835445 YS |
1338 | goto up_out; |
1339 | } | |
1340 | ||
f4e9e0e6 LH |
1341 | vma_iter_init(&vmi, mm, start); |
1342 | prev = vma_prev(&vmi); | |
1343 | for_each_vma_range(vmi, vma, end) { | |
1344 | err = mbind_range(&vmi, vma, &prev, start, end, new); | |
1345 | if (err) | |
1346 | break; | |
1347 | } | |
7e2ab150 | 1348 | |
b24f53a0 LS |
1349 | if (!err) { |
1350 | int nr_failed = 0; | |
1351 | ||
cf608ac1 | 1352 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1353 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
4e096ae1 | 1354 | nr_failed = migrate_pages(&pagelist, new_folio, NULL, |
5ac95884 | 1355 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL); |
cf608ac1 | 1356 | if (nr_failed) |
74060e4d | 1357 | putback_movable_pages(&pagelist); |
cf608ac1 | 1358 | } |
6ce3c4c0 | 1359 | |
24526268 | 1360 | if (((ret > 0) || nr_failed) && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1361 | err = -EIO; |
a85dfc30 | 1362 | } else { |
d8835445 | 1363 | up_out: |
a85dfc30 YS |
1364 | if (!list_empty(&pagelist)) |
1365 | putback_movable_pages(&pagelist); | |
1366 | } | |
1367 | ||
d8ed45c5 | 1368 | mmap_write_unlock(mm); |
d8835445 | 1369 | mpol_out: |
f0be3d32 | 1370 | mpol_put(new); |
d479960e | 1371 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1372 | lru_cache_enable(); |
6ce3c4c0 CL |
1373 | return err; |
1374 | } | |
1375 | ||
8bccd85f CL |
1376 | /* |
1377 | * User space interface with variable sized bitmaps for nodelists. | |
1378 | */ | |
e130242d AB |
1379 | static int get_bitmap(unsigned long *mask, const unsigned long __user *nmask, |
1380 | unsigned long maxnode) | |
1381 | { | |
1382 | unsigned long nlongs = BITS_TO_LONGS(maxnode); | |
1383 | int ret; | |
1384 | ||
1385 | if (in_compat_syscall()) | |
1386 | ret = compat_get_bitmap(mask, | |
1387 | (const compat_ulong_t __user *)nmask, | |
1388 | maxnode); | |
1389 | else | |
1390 | ret = copy_from_user(mask, nmask, | |
1391 | nlongs * sizeof(unsigned long)); | |
1392 | ||
1393 | if (ret) | |
1394 | return -EFAULT; | |
1395 | ||
1396 | if (maxnode % BITS_PER_LONG) | |
1397 | mask[nlongs - 1] &= (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1398 | ||
1399 | return 0; | |
1400 | } | |
8bccd85f CL |
1401 | |
1402 | /* Copy a node mask from user space. */ | |
39743889 | 1403 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1404 | unsigned long maxnode) |
1405 | { | |
8bccd85f CL |
1406 | --maxnode; |
1407 | nodes_clear(*nodes); | |
1408 | if (maxnode == 0 || !nmask) | |
1409 | return 0; | |
a9c930ba | 1410 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1411 | return -EINVAL; |
8bccd85f | 1412 | |
56521e7a YX |
1413 | /* |
1414 | * When the user specified more nodes than supported just check | |
e130242d AB |
1415 | * if the non supported part is all zero, one word at a time, |
1416 | * starting at the end. | |
56521e7a | 1417 | */ |
e130242d AB |
1418 | while (maxnode > MAX_NUMNODES) { |
1419 | unsigned long bits = min_t(unsigned long, maxnode, BITS_PER_LONG); | |
1420 | unsigned long t; | |
8bccd85f | 1421 | |
000eca5d | 1422 | if (get_bitmap(&t, &nmask[(maxnode - 1) / BITS_PER_LONG], bits)) |
56521e7a | 1423 | return -EFAULT; |
e130242d AB |
1424 | |
1425 | if (maxnode - bits >= MAX_NUMNODES) { | |
1426 | maxnode -= bits; | |
1427 | } else { | |
1428 | maxnode = MAX_NUMNODES; | |
1429 | t &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1430 | } | |
1431 | if (t) | |
56521e7a YX |
1432 | return -EINVAL; |
1433 | } | |
1434 | ||
e130242d | 1435 | return get_bitmap(nodes_addr(*nodes), nmask, maxnode); |
8bccd85f CL |
1436 | } |
1437 | ||
1438 | /* Copy a kernel node mask to user space */ | |
1439 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1440 | nodemask_t *nodes) | |
1441 | { | |
1442 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1443 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
e130242d AB |
1444 | bool compat = in_compat_syscall(); |
1445 | ||
1446 | if (compat) | |
1447 | nbytes = BITS_TO_COMPAT_LONGS(nr_node_ids) * sizeof(compat_long_t); | |
8bccd85f CL |
1448 | |
1449 | if (copy > nbytes) { | |
1450 | if (copy > PAGE_SIZE) | |
1451 | return -EINVAL; | |
1452 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1453 | return -EFAULT; | |
1454 | copy = nbytes; | |
e130242d | 1455 | maxnode = nr_node_ids; |
8bccd85f | 1456 | } |
e130242d AB |
1457 | |
1458 | if (compat) | |
1459 | return compat_put_bitmap((compat_ulong_t __user *)mask, | |
1460 | nodes_addr(*nodes), maxnode); | |
1461 | ||
8bccd85f CL |
1462 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; |
1463 | } | |
1464 | ||
95837924 FT |
1465 | /* Basic parameter sanity check used by both mbind() and set_mempolicy() */ |
1466 | static inline int sanitize_mpol_flags(int *mode, unsigned short *flags) | |
1467 | { | |
1468 | *flags = *mode & MPOL_MODE_FLAGS; | |
1469 | *mode &= ~MPOL_MODE_FLAGS; | |
b27abacc | 1470 | |
a38a59fd | 1471 | if ((unsigned int)(*mode) >= MPOL_MAX) |
95837924 FT |
1472 | return -EINVAL; |
1473 | if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES)) | |
1474 | return -EINVAL; | |
6d2aec9e ED |
1475 | if (*flags & MPOL_F_NUMA_BALANCING) { |
1476 | if (*mode != MPOL_BIND) | |
1477 | return -EINVAL; | |
1478 | *flags |= (MPOL_F_MOF | MPOL_F_MORON); | |
1479 | } | |
95837924 FT |
1480 | return 0; |
1481 | } | |
1482 | ||
e7dc9ad6 DB |
1483 | static long kernel_mbind(unsigned long start, unsigned long len, |
1484 | unsigned long mode, const unsigned long __user *nmask, | |
1485 | unsigned long maxnode, unsigned int flags) | |
8bccd85f | 1486 | { |
95837924 | 1487 | unsigned short mode_flags; |
8bccd85f | 1488 | nodemask_t nodes; |
95837924 | 1489 | int lmode = mode; |
8bccd85f CL |
1490 | int err; |
1491 | ||
057d3389 | 1492 | start = untagged_addr(start); |
95837924 FT |
1493 | err = sanitize_mpol_flags(&lmode, &mode_flags); |
1494 | if (err) | |
1495 | return err; | |
1496 | ||
8bccd85f CL |
1497 | err = get_nodes(&nodes, nmask, maxnode); |
1498 | if (err) | |
1499 | return err; | |
95837924 FT |
1500 | |
1501 | return do_mbind(start, len, lmode, mode_flags, &nodes, flags); | |
8bccd85f CL |
1502 | } |
1503 | ||
c6018b4b AK |
1504 | SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, len, |
1505 | unsigned long, home_node, unsigned long, flags) | |
1506 | { | |
1507 | struct mm_struct *mm = current->mm; | |
f4e9e0e6 | 1508 | struct vm_area_struct *vma, *prev; |
e976936c | 1509 | struct mempolicy *new, *old; |
c6018b4b AK |
1510 | unsigned long end; |
1511 | int err = -ENOENT; | |
66850be5 | 1512 | VMA_ITERATOR(vmi, mm, start); |
c6018b4b AK |
1513 | |
1514 | start = untagged_addr(start); | |
1515 | if (start & ~PAGE_MASK) | |
1516 | return -EINVAL; | |
1517 | /* | |
1518 | * flags is used for future extension if any. | |
1519 | */ | |
1520 | if (flags != 0) | |
1521 | return -EINVAL; | |
1522 | ||
1523 | /* | |
1524 | * Check home_node is online to avoid accessing uninitialized | |
1525 | * NODE_DATA. | |
1526 | */ | |
1527 | if (home_node >= MAX_NUMNODES || !node_online(home_node)) | |
1528 | return -EINVAL; | |
1529 | ||
aaa31e05 | 1530 | len = PAGE_ALIGN(len); |
c6018b4b AK |
1531 | end = start + len; |
1532 | ||
1533 | if (end < start) | |
1534 | return -EINVAL; | |
1535 | if (end == start) | |
1536 | return 0; | |
1537 | mmap_write_lock(mm); | |
f4e9e0e6 | 1538 | prev = vma_prev(&vmi); |
66850be5 | 1539 | for_each_vma_range(vmi, vma, end) { |
c6018b4b AK |
1540 | /* |
1541 | * If any vma in the range got policy other than MPOL_BIND | |
1542 | * or MPOL_PREFERRED_MANY we return error. We don't reset | |
1543 | * the home node for vmas we already updated before. | |
1544 | */ | |
e976936c | 1545 | old = vma_policy(vma); |
51f62537 LH |
1546 | if (!old) { |
1547 | prev = vma; | |
e976936c | 1548 | continue; |
51f62537 | 1549 | } |
e976936c | 1550 | if (old->mode != MPOL_BIND && old->mode != MPOL_PREFERRED_MANY) { |
c6018b4b AK |
1551 | err = -EOPNOTSUPP; |
1552 | break; | |
1553 | } | |
e976936c MH |
1554 | new = mpol_dup(old); |
1555 | if (IS_ERR(new)) { | |
1556 | err = PTR_ERR(new); | |
1557 | break; | |
1558 | } | |
c6018b4b | 1559 | |
6c21e066 | 1560 | vma_start_write(vma); |
c6018b4b | 1561 | new->home_node = home_node; |
f4e9e0e6 | 1562 | err = mbind_range(&vmi, vma, &prev, start, end, new); |
c6018b4b AK |
1563 | mpol_put(new); |
1564 | if (err) | |
1565 | break; | |
1566 | } | |
1567 | mmap_write_unlock(mm); | |
1568 | return err; | |
1569 | } | |
1570 | ||
e7dc9ad6 DB |
1571 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1572 | unsigned long, mode, const unsigned long __user *, nmask, | |
1573 | unsigned long, maxnode, unsigned int, flags) | |
1574 | { | |
1575 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1576 | } | |
1577 | ||
8bccd85f | 1578 | /* Set the process memory policy */ |
af03c4ac DB |
1579 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1580 | unsigned long maxnode) | |
8bccd85f | 1581 | { |
95837924 | 1582 | unsigned short mode_flags; |
8bccd85f | 1583 | nodemask_t nodes; |
95837924 FT |
1584 | int lmode = mode; |
1585 | int err; | |
1586 | ||
1587 | err = sanitize_mpol_flags(&lmode, &mode_flags); | |
1588 | if (err) | |
1589 | return err; | |
8bccd85f | 1590 | |
8bccd85f CL |
1591 | err = get_nodes(&nodes, nmask, maxnode); |
1592 | if (err) | |
1593 | return err; | |
95837924 FT |
1594 | |
1595 | return do_set_mempolicy(lmode, mode_flags, &nodes); | |
8bccd85f CL |
1596 | } |
1597 | ||
af03c4ac DB |
1598 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1599 | unsigned long, maxnode) | |
1600 | { | |
1601 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1602 | } | |
1603 | ||
b6e9b0ba DB |
1604 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1605 | const unsigned long __user *old_nodes, | |
1606 | const unsigned long __user *new_nodes) | |
39743889 | 1607 | { |
596d7cfa | 1608 | struct mm_struct *mm = NULL; |
39743889 | 1609 | struct task_struct *task; |
39743889 CL |
1610 | nodemask_t task_nodes; |
1611 | int err; | |
596d7cfa KM |
1612 | nodemask_t *old; |
1613 | nodemask_t *new; | |
1614 | NODEMASK_SCRATCH(scratch); | |
1615 | ||
1616 | if (!scratch) | |
1617 | return -ENOMEM; | |
39743889 | 1618 | |
596d7cfa KM |
1619 | old = &scratch->mask1; |
1620 | new = &scratch->mask2; | |
1621 | ||
1622 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1623 | if (err) |
596d7cfa | 1624 | goto out; |
39743889 | 1625 | |
596d7cfa | 1626 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1627 | if (err) |
596d7cfa | 1628 | goto out; |
39743889 CL |
1629 | |
1630 | /* Find the mm_struct */ | |
55cfaa3c | 1631 | rcu_read_lock(); |
228ebcbe | 1632 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1633 | if (!task) { |
55cfaa3c | 1634 | rcu_read_unlock(); |
596d7cfa KM |
1635 | err = -ESRCH; |
1636 | goto out; | |
39743889 | 1637 | } |
3268c63e | 1638 | get_task_struct(task); |
39743889 | 1639 | |
596d7cfa | 1640 | err = -EINVAL; |
39743889 CL |
1641 | |
1642 | /* | |
31367466 OE |
1643 | * Check if this process has the right to modify the specified process. |
1644 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1645 | */ |
31367466 | 1646 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1647 | rcu_read_unlock(); |
39743889 | 1648 | err = -EPERM; |
3268c63e | 1649 | goto out_put; |
39743889 | 1650 | } |
c69e8d9c | 1651 | rcu_read_unlock(); |
39743889 CL |
1652 | |
1653 | task_nodes = cpuset_mems_allowed(task); | |
1654 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1655 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1656 | err = -EPERM; |
3268c63e | 1657 | goto out_put; |
39743889 CL |
1658 | } |
1659 | ||
0486a38b YX |
1660 | task_nodes = cpuset_mems_allowed(current); |
1661 | nodes_and(*new, *new, task_nodes); | |
1662 | if (nodes_empty(*new)) | |
1663 | goto out_put; | |
1664 | ||
86c3a764 DQ |
1665 | err = security_task_movememory(task); |
1666 | if (err) | |
3268c63e | 1667 | goto out_put; |
86c3a764 | 1668 | |
3268c63e CL |
1669 | mm = get_task_mm(task); |
1670 | put_task_struct(task); | |
f2a9ef88 SL |
1671 | |
1672 | if (!mm) { | |
3268c63e | 1673 | err = -EINVAL; |
f2a9ef88 SL |
1674 | goto out; |
1675 | } | |
1676 | ||
1677 | err = do_migrate_pages(mm, old, new, | |
1678 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1679 | |
1680 | mmput(mm); | |
1681 | out: | |
596d7cfa KM |
1682 | NODEMASK_SCRATCH_FREE(scratch); |
1683 | ||
39743889 | 1684 | return err; |
3268c63e CL |
1685 | |
1686 | out_put: | |
1687 | put_task_struct(task); | |
1688 | goto out; | |
1689 | ||
39743889 CL |
1690 | } |
1691 | ||
b6e9b0ba DB |
1692 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1693 | const unsigned long __user *, old_nodes, | |
1694 | const unsigned long __user *, new_nodes) | |
1695 | { | |
1696 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1697 | } | |
1698 | ||
39743889 | 1699 | |
8bccd85f | 1700 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1701 | static int kernel_get_mempolicy(int __user *policy, |
1702 | unsigned long __user *nmask, | |
1703 | unsigned long maxnode, | |
1704 | unsigned long addr, | |
1705 | unsigned long flags) | |
8bccd85f | 1706 | { |
dbcb0f19 | 1707 | int err; |
3f649ab7 | 1708 | int pval; |
8bccd85f CL |
1709 | nodemask_t nodes; |
1710 | ||
050c17f2 | 1711 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1712 | return -EINVAL; |
1713 | ||
4605f057 WH |
1714 | addr = untagged_addr(addr); |
1715 | ||
8bccd85f CL |
1716 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1717 | ||
1718 | if (err) | |
1719 | return err; | |
1720 | ||
1721 | if (policy && put_user(pval, policy)) | |
1722 | return -EFAULT; | |
1723 | ||
1724 | if (nmask) | |
1725 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1726 | ||
1727 | return err; | |
1728 | } | |
1729 | ||
af03c4ac DB |
1730 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1731 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1732 | unsigned long, addr, unsigned long, flags) | |
1733 | { | |
1734 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1735 | } | |
1736 | ||
20ca87f2 LX |
1737 | bool vma_migratable(struct vm_area_struct *vma) |
1738 | { | |
1739 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1740 | return false; | |
1741 | ||
1742 | /* | |
1743 | * DAX device mappings require predictable access latency, so avoid | |
1744 | * incurring periodic faults. | |
1745 | */ | |
1746 | if (vma_is_dax(vma)) | |
1747 | return false; | |
1748 | ||
1749 | if (is_vm_hugetlb_page(vma) && | |
1750 | !hugepage_migration_supported(hstate_vma(vma))) | |
1751 | return false; | |
1752 | ||
1753 | /* | |
1754 | * Migration allocates pages in the highest zone. If we cannot | |
1755 | * do so then migration (at least from node to node) is not | |
1756 | * possible. | |
1757 | */ | |
1758 | if (vma->vm_file && | |
1759 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1760 | < policy_zone) | |
1761 | return false; | |
1762 | return true; | |
1763 | } | |
1764 | ||
74d2c3a0 ON |
1765 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1766 | unsigned long addr) | |
1da177e4 | 1767 | { |
8d90274b | 1768 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1769 | |
1770 | if (vma) { | |
480eccf9 | 1771 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1772 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1773 | } else if (vma->vm_policy) { |
1da177e4 | 1774 | pol = vma->vm_policy; |
00442ad0 MG |
1775 | |
1776 | /* | |
1777 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1778 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1779 | * count on these policies which will be dropped by | |
1780 | * mpol_cond_put() later | |
1781 | */ | |
1782 | if (mpol_needs_cond_ref(pol)) | |
1783 | mpol_get(pol); | |
1784 | } | |
1da177e4 | 1785 | } |
f15ca78e | 1786 | |
74d2c3a0 ON |
1787 | return pol; |
1788 | } | |
1789 | ||
1790 | /* | |
dd6eecb9 | 1791 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1792 | * @vma: virtual memory area whose policy is sought |
1793 | * @addr: address in @vma for shared policy lookup | |
1794 | * | |
1795 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1796 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1797 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1798 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1799 | * freeing by another task. It is the caller's responsibility to free the | |
1800 | * extra reference for shared policies. | |
1801 | */ | |
ac79f78d | 1802 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1803 | unsigned long addr) |
74d2c3a0 ON |
1804 | { |
1805 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1806 | ||
8d90274b | 1807 | if (!pol) |
dd6eecb9 | 1808 | pol = get_task_policy(current); |
8d90274b | 1809 | |
1da177e4 LT |
1810 | return pol; |
1811 | } | |
1812 | ||
6b6482bb | 1813 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1814 | { |
6b6482bb | 1815 | struct mempolicy *pol; |
fc314724 | 1816 | |
6b6482bb ON |
1817 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1818 | bool ret = false; | |
fc314724 | 1819 | |
6b6482bb ON |
1820 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1821 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1822 | ret = true; | |
1823 | mpol_cond_put(pol); | |
8d90274b | 1824 | |
6b6482bb | 1825 | return ret; |
fc314724 MG |
1826 | } |
1827 | ||
6b6482bb | 1828 | pol = vma->vm_policy; |
8d90274b | 1829 | if (!pol) |
6b6482bb | 1830 | pol = get_task_policy(current); |
8d90274b | 1831 | |
fc314724 MG |
1832 | return pol->flags & MPOL_F_MOF; |
1833 | } | |
1834 | ||
d2226ebd | 1835 | bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
d3eb1570 LJ |
1836 | { |
1837 | enum zone_type dynamic_policy_zone = policy_zone; | |
1838 | ||
1839 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1840 | ||
1841 | /* | |
269fbe72 | 1842 | * if policy->nodes has movable memory only, |
d3eb1570 LJ |
1843 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. |
1844 | * | |
269fbe72 | 1845 | * policy->nodes is intersect with node_states[N_MEMORY]. |
f0953a1b | 1846 | * so if the following test fails, it implies |
269fbe72 | 1847 | * policy->nodes has movable memory only. |
d3eb1570 | 1848 | */ |
269fbe72 | 1849 | if (!nodes_intersects(policy->nodes, node_states[N_HIGH_MEMORY])) |
d3eb1570 LJ |
1850 | dynamic_policy_zone = ZONE_MOVABLE; |
1851 | ||
1852 | return zone >= dynamic_policy_zone; | |
1853 | } | |
1854 | ||
52cd3b07 LS |
1855 | /* |
1856 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1857 | * page allocation | |
1858 | */ | |
8ca39e68 | 1859 | nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
19770b32 | 1860 | { |
b27abacc DH |
1861 | int mode = policy->mode; |
1862 | ||
19770b32 | 1863 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ |
b27abacc DH |
1864 | if (unlikely(mode == MPOL_BIND) && |
1865 | apply_policy_zone(policy, gfp_zone(gfp)) && | |
1866 | cpuset_nodemask_valid_mems_allowed(&policy->nodes)) | |
1867 | return &policy->nodes; | |
1868 | ||
1869 | if (mode == MPOL_PREFERRED_MANY) | |
269fbe72 | 1870 | return &policy->nodes; |
19770b32 MG |
1871 | |
1872 | return NULL; | |
1873 | } | |
1874 | ||
b27abacc DH |
1875 | /* |
1876 | * Return the preferred node id for 'prefer' mempolicy, and return | |
1877 | * the given id for all other policies. | |
1878 | * | |
1879 | * policy_node() is always coupled with policy_nodemask(), which | |
1880 | * secures the nodemask limit for 'bind' and 'prefer-many' policy. | |
1881 | */ | |
f8fd5253 | 1882 | static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd) |
1da177e4 | 1883 | { |
7858d7bc | 1884 | if (policy->mode == MPOL_PREFERRED) { |
269fbe72 | 1885 | nd = first_node(policy->nodes); |
7858d7bc | 1886 | } else { |
19770b32 | 1887 | /* |
6d840958 MH |
1888 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1889 | * because we might easily break the expectation to stay on the | |
1890 | * requested node and not break the policy. | |
19770b32 | 1891 | */ |
6d840958 | 1892 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1893 | } |
6d840958 | 1894 | |
c6018b4b AK |
1895 | if ((policy->mode == MPOL_BIND || |
1896 | policy->mode == MPOL_PREFERRED_MANY) && | |
1897 | policy->home_node != NUMA_NO_NODE) | |
1898 | return policy->home_node; | |
1899 | ||
04ec6264 | 1900 | return nd; |
1da177e4 LT |
1901 | } |
1902 | ||
1903 | /* Do dynamic interleaving for a process */ | |
1904 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1905 | { | |
45816682 | 1906 | unsigned next; |
1da177e4 LT |
1907 | struct task_struct *me = current; |
1908 | ||
269fbe72 | 1909 | next = next_node_in(me->il_prev, policy->nodes); |
f5b087b5 | 1910 | if (next < MAX_NUMNODES) |
45816682 VB |
1911 | me->il_prev = next; |
1912 | return next; | |
1da177e4 LT |
1913 | } |
1914 | ||
dc85da15 CL |
1915 | /* |
1916 | * Depending on the memory policy provide a node from which to allocate the | |
1917 | * next slab entry. | |
1918 | */ | |
2a389610 | 1919 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1920 | { |
e7b691b0 | 1921 | struct mempolicy *policy; |
2a389610 | 1922 | int node = numa_mem_id(); |
e7b691b0 | 1923 | |
38b031dd | 1924 | if (!in_task()) |
2a389610 | 1925 | return node; |
e7b691b0 AK |
1926 | |
1927 | policy = current->mempolicy; | |
7858d7bc | 1928 | if (!policy) |
2a389610 | 1929 | return node; |
bea904d5 LS |
1930 | |
1931 | switch (policy->mode) { | |
1932 | case MPOL_PREFERRED: | |
269fbe72 | 1933 | return first_node(policy->nodes); |
765c4507 | 1934 | |
dc85da15 CL |
1935 | case MPOL_INTERLEAVE: |
1936 | return interleave_nodes(policy); | |
1937 | ||
b27abacc DH |
1938 | case MPOL_BIND: |
1939 | case MPOL_PREFERRED_MANY: | |
1940 | { | |
c33d6c06 MG |
1941 | struct zoneref *z; |
1942 | ||
dc85da15 CL |
1943 | /* |
1944 | * Follow bind policy behavior and start allocation at the | |
1945 | * first node. | |
1946 | */ | |
19770b32 | 1947 | struct zonelist *zonelist; |
19770b32 | 1948 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1949 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 | 1950 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
269fbe72 | 1951 | &policy->nodes); |
c1093b74 | 1952 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1953 | } |
7858d7bc FT |
1954 | case MPOL_LOCAL: |
1955 | return node; | |
dc85da15 | 1956 | |
dc85da15 | 1957 | default: |
bea904d5 | 1958 | BUG(); |
dc85da15 CL |
1959 | } |
1960 | } | |
1961 | ||
fee83b3a AM |
1962 | /* |
1963 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
269fbe72 | 1964 | * node in pol->nodes (starting from n=0), wrapping around if n exceeds the |
fee83b3a AM |
1965 | * number of present nodes. |
1966 | */ | |
98c70baa | 1967 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1968 | { |
276aeee1 | 1969 | nodemask_t nodemask = pol->nodes; |
1970 | unsigned int target, nnodes; | |
fee83b3a AM |
1971 | int i; |
1972 | int nid; | |
276aeee1 | 1973 | /* |
1974 | * The barrier will stabilize the nodemask in a register or on | |
1975 | * the stack so that it will stop changing under the code. | |
1976 | * | |
1977 | * Between first_node() and next_node(), pol->nodes could be changed | |
1978 | * by other threads. So we put pol->nodes in a local stack. | |
1979 | */ | |
1980 | barrier(); | |
1da177e4 | 1981 | |
276aeee1 | 1982 | nnodes = nodes_weight(nodemask); |
f5b087b5 DR |
1983 | if (!nnodes) |
1984 | return numa_node_id(); | |
fee83b3a | 1985 | target = (unsigned int)n % nnodes; |
276aeee1 | 1986 | nid = first_node(nodemask); |
fee83b3a | 1987 | for (i = 0; i < target; i++) |
276aeee1 | 1988 | nid = next_node(nid, nodemask); |
1da177e4 LT |
1989 | return nid; |
1990 | } | |
1991 | ||
5da7ca86 CL |
1992 | /* Determine a node number for interleave */ |
1993 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1994 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1995 | { | |
1996 | if (vma) { | |
1997 | unsigned long off; | |
1998 | ||
3b98b087 NA |
1999 | /* |
2000 | * for small pages, there is no difference between | |
2001 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
2002 | * for huge pages, since vm_pgoff is in units of small | |
2003 | * pages, we need to shift off the always 0 bits to get | |
2004 | * a useful offset. | |
2005 | */ | |
2006 | BUG_ON(shift < PAGE_SHIFT); | |
2007 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 2008 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 2009 | return offset_il_node(pol, off); |
5da7ca86 CL |
2010 | } else |
2011 | return interleave_nodes(pol); | |
2012 | } | |
2013 | ||
00ac59ad | 2014 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2015 | /* |
04ec6264 | 2016 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2017 | * @vma: virtual memory area whose policy is sought |
2018 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2019 | * @gfp_flags: for requested zone | |
2020 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
b27abacc | 2021 | * @nodemask: pointer to nodemask pointer for 'bind' and 'prefer-many' policy |
480eccf9 | 2022 | * |
04ec6264 | 2023 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 | 2024 | * to the struct mempolicy for conditional unref after allocation. |
b27abacc DH |
2025 | * If the effective policy is 'bind' or 'prefer-many', returns a pointer |
2026 | * to the mempolicy's @nodemask for filtering the zonelist. | |
c0ff7453 | 2027 | * |
d26914d1 | 2028 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2029 | */ |
04ec6264 VB |
2030 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2031 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2032 | { |
04ec6264 | 2033 | int nid; |
b27abacc | 2034 | int mode; |
5da7ca86 | 2035 | |
dd6eecb9 | 2036 | *mpol = get_vma_policy(vma, addr); |
b27abacc DH |
2037 | *nodemask = NULL; |
2038 | mode = (*mpol)->mode; | |
5da7ca86 | 2039 | |
b27abacc | 2040 | if (unlikely(mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2041 | nid = interleave_nid(*mpol, vma, addr, |
2042 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2043 | } else { |
04ec6264 | 2044 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
b27abacc | 2045 | if (mode == MPOL_BIND || mode == MPOL_PREFERRED_MANY) |
269fbe72 | 2046 | *nodemask = &(*mpol)->nodes; |
480eccf9 | 2047 | } |
04ec6264 | 2048 | return nid; |
5da7ca86 | 2049 | } |
06808b08 LS |
2050 | |
2051 | /* | |
2052 | * init_nodemask_of_mempolicy | |
2053 | * | |
2054 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2055 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2056 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2057 | * initialize the argument nodemask to contain the single node for | |
2058 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2059 | * of non-default mempolicy. | |
2060 | * | |
2061 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2062 | * because the current task is examining it's own mempolicy and a task's | |
2063 | * mempolicy is only ever changed by the task itself. | |
2064 | * | |
2065 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2066 | */ | |
2067 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2068 | { | |
2069 | struct mempolicy *mempolicy; | |
06808b08 LS |
2070 | |
2071 | if (!(mask && current->mempolicy)) | |
2072 | return false; | |
2073 | ||
c0ff7453 | 2074 | task_lock(current); |
06808b08 LS |
2075 | mempolicy = current->mempolicy; |
2076 | switch (mempolicy->mode) { | |
2077 | case MPOL_PREFERRED: | |
b27abacc | 2078 | case MPOL_PREFERRED_MANY: |
06808b08 | 2079 | case MPOL_BIND: |
06808b08 | 2080 | case MPOL_INTERLEAVE: |
269fbe72 | 2081 | *mask = mempolicy->nodes; |
7858d7bc FT |
2082 | break; |
2083 | ||
2084 | case MPOL_LOCAL: | |
269fbe72 | 2085 | init_nodemask_of_node(mask, numa_node_id()); |
06808b08 LS |
2086 | break; |
2087 | ||
2088 | default: | |
2089 | BUG(); | |
2090 | } | |
c0ff7453 | 2091 | task_unlock(current); |
06808b08 LS |
2092 | |
2093 | return true; | |
2094 | } | |
00ac59ad | 2095 | #endif |
5da7ca86 | 2096 | |
6f48d0eb | 2097 | /* |
b26e517a | 2098 | * mempolicy_in_oom_domain |
6f48d0eb | 2099 | * |
b26e517a FT |
2100 | * If tsk's mempolicy is "bind", check for intersection between mask and |
2101 | * the policy nodemask. Otherwise, return true for all other policies | |
2102 | * including "interleave", as a tsk with "interleave" policy may have | |
2103 | * memory allocated from all nodes in system. | |
6f48d0eb DR |
2104 | * |
2105 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2106 | */ | |
b26e517a | 2107 | bool mempolicy_in_oom_domain(struct task_struct *tsk, |
6f48d0eb DR |
2108 | const nodemask_t *mask) |
2109 | { | |
2110 | struct mempolicy *mempolicy; | |
2111 | bool ret = true; | |
2112 | ||
2113 | if (!mask) | |
2114 | return ret; | |
b26e517a | 2115 | |
6f48d0eb DR |
2116 | task_lock(tsk); |
2117 | mempolicy = tsk->mempolicy; | |
b26e517a | 2118 | if (mempolicy && mempolicy->mode == MPOL_BIND) |
269fbe72 | 2119 | ret = nodes_intersects(mempolicy->nodes, *mask); |
6f48d0eb | 2120 | task_unlock(tsk); |
b26e517a | 2121 | |
6f48d0eb DR |
2122 | return ret; |
2123 | } | |
2124 | ||
1da177e4 LT |
2125 | /* Allocate a page in interleaved policy. |
2126 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2127 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2128 | unsigned nid) | |
1da177e4 | 2129 | { |
1da177e4 LT |
2130 | struct page *page; |
2131 | ||
84172f4b | 2132 | page = __alloc_pages(gfp, order, nid, NULL); |
4518085e KW |
2133 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2134 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2135 | return page; | |
de55c8b2 AR |
2136 | if (page && page_to_nid(page) == nid) { |
2137 | preempt_disable(); | |
f19298b9 | 2138 | __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT); |
de55c8b2 AR |
2139 | preempt_enable(); |
2140 | } | |
1da177e4 LT |
2141 | return page; |
2142 | } | |
2143 | ||
4c54d949 FT |
2144 | static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order, |
2145 | int nid, struct mempolicy *pol) | |
2146 | { | |
2147 | struct page *page; | |
2148 | gfp_t preferred_gfp; | |
2149 | ||
2150 | /* | |
2151 | * This is a two pass approach. The first pass will only try the | |
2152 | * preferred nodes but skip the direct reclaim and allow the | |
2153 | * allocation to fail, while the second pass will try all the | |
2154 | * nodes in system. | |
2155 | */ | |
2156 | preferred_gfp = gfp | __GFP_NOWARN; | |
2157 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2158 | page = __alloc_pages(preferred_gfp, order, nid, &pol->nodes); | |
2159 | if (!page) | |
c0455116 | 2160 | page = __alloc_pages(gfp, order, nid, NULL); |
4c54d949 FT |
2161 | |
2162 | return page; | |
2163 | } | |
2164 | ||
1da177e4 | 2165 | /** |
adf88aa8 | 2166 | * vma_alloc_folio - Allocate a folio for a VMA. |
eb350739 | 2167 | * @gfp: GFP flags. |
adf88aa8 | 2168 | * @order: Order of the folio. |
eb350739 MWO |
2169 | * @vma: Pointer to VMA or NULL if not available. |
2170 | * @addr: Virtual address of the allocation. Must be inside @vma. | |
eb350739 | 2171 | * @hugepage: For hugepages try only the preferred node if possible. |
1da177e4 | 2172 | * |
adf88aa8 | 2173 | * Allocate a folio for a specific address in @vma, using the appropriate |
eb350739 MWO |
2174 | * NUMA policy. When @vma is not NULL the caller must hold the mmap_lock |
2175 | * of the mm_struct of the VMA to prevent it from going away. Should be | |
adf88aa8 | 2176 | * used for all allocations for folios that will be mapped into user space. |
1da177e4 | 2177 | * |
adf88aa8 | 2178 | * Return: The folio on success or NULL if allocation fails. |
1da177e4 | 2179 | */ |
adf88aa8 | 2180 | struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma, |
be1a13eb | 2181 | unsigned long addr, bool hugepage) |
1da177e4 | 2182 | { |
cc9a6c87 | 2183 | struct mempolicy *pol; |
be1a13eb | 2184 | int node = numa_node_id(); |
adf88aa8 | 2185 | struct folio *folio; |
04ec6264 | 2186 | int preferred_nid; |
be97a41b | 2187 | nodemask_t *nmask; |
cc9a6c87 | 2188 | |
dd6eecb9 | 2189 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2190 | |
0867a57c | 2191 | if (pol->mode == MPOL_INTERLEAVE) { |
adf88aa8 | 2192 | struct page *page; |
0867a57c VB |
2193 | unsigned nid; |
2194 | ||
2195 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2196 | mpol_cond_put(pol); | |
adf88aa8 | 2197 | gfp |= __GFP_COMP; |
0867a57c | 2198 | page = alloc_page_interleave(gfp, order, nid); |
adf88aa8 | 2199 | folio = (struct folio *)page; |
da6e7bf3 MWO |
2200 | if (folio && order > 1) |
2201 | folio_prep_large_rmappable(folio); | |
0867a57c | 2202 | goto out; |
19deb769 DR |
2203 | } |
2204 | ||
4c54d949 | 2205 | if (pol->mode == MPOL_PREFERRED_MANY) { |
adf88aa8 MWO |
2206 | struct page *page; |
2207 | ||
c0455116 | 2208 | node = policy_node(gfp, pol, node); |
adf88aa8 | 2209 | gfp |= __GFP_COMP; |
4c54d949 FT |
2210 | page = alloc_pages_preferred_many(gfp, order, node, pol); |
2211 | mpol_cond_put(pol); | |
adf88aa8 | 2212 | folio = (struct folio *)page; |
da6e7bf3 MWO |
2213 | if (folio && order > 1) |
2214 | folio_prep_large_rmappable(folio); | |
4c54d949 FT |
2215 | goto out; |
2216 | } | |
2217 | ||
19deb769 DR |
2218 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { |
2219 | int hpage_node = node; | |
2220 | ||
2221 | /* | |
2222 | * For hugepage allocation and non-interleave policy which | |
2223 | * allows the current node (or other explicitly preferred | |
2224 | * node) we only try to allocate from the current/preferred | |
2225 | * node and don't fall back to other nodes, as the cost of | |
2226 | * remote accesses would likely offset THP benefits. | |
2227 | * | |
b27abacc | 2228 | * If the policy is interleave or does not allow the current |
19deb769 DR |
2229 | * node in its nodemask, we allocate the standard way. |
2230 | */ | |
7858d7bc | 2231 | if (pol->mode == MPOL_PREFERRED) |
269fbe72 | 2232 | hpage_node = first_node(pol->nodes); |
19deb769 DR |
2233 | |
2234 | nmask = policy_nodemask(gfp, pol); | |
2235 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2236 | mpol_cond_put(pol); | |
cc638f32 VB |
2237 | /* |
2238 | * First, try to allocate THP only on local node, but | |
2239 | * don't reclaim unnecessarily, just compact. | |
2240 | */ | |
adf88aa8 MWO |
2241 | folio = __folio_alloc_node(gfp | __GFP_THISNODE | |
2242 | __GFP_NORETRY, order, hpage_node); | |
76e654cc DR |
2243 | |
2244 | /* | |
2245 | * If hugepage allocations are configured to always | |
2246 | * synchronous compact or the vma has been madvised | |
2247 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2248 | * memory with both reclaim and compact as well. |
76e654cc | 2249 | */ |
adf88aa8 MWO |
2250 | if (!folio && (gfp & __GFP_DIRECT_RECLAIM)) |
2251 | folio = __folio_alloc(gfp, order, hpage_node, | |
2252 | nmask); | |
76e654cc | 2253 | |
19deb769 DR |
2254 | goto out; |
2255 | } | |
356ff8a9 DR |
2256 | } |
2257 | ||
be97a41b | 2258 | nmask = policy_nodemask(gfp, pol); |
04ec6264 | 2259 | preferred_nid = policy_node(gfp, pol, node); |
adf88aa8 | 2260 | folio = __folio_alloc(gfp, order, preferred_nid, nmask); |
d51e9894 | 2261 | mpol_cond_put(pol); |
be97a41b | 2262 | out: |
f584b680 MWO |
2263 | return folio; |
2264 | } | |
adf88aa8 | 2265 | EXPORT_SYMBOL(vma_alloc_folio); |
f584b680 | 2266 | |
1da177e4 | 2267 | /** |
6421ec76 MWO |
2268 | * alloc_pages - Allocate pages. |
2269 | * @gfp: GFP flags. | |
2270 | * @order: Power of two of number of pages to allocate. | |
1da177e4 | 2271 | * |
6421ec76 MWO |
2272 | * Allocate 1 << @order contiguous pages. The physical address of the |
2273 | * first page is naturally aligned (eg an order-3 allocation will be aligned | |
2274 | * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current | |
2275 | * process is honoured when in process context. | |
1da177e4 | 2276 | * |
6421ec76 MWO |
2277 | * Context: Can be called from any context, providing the appropriate GFP |
2278 | * flags are used. | |
2279 | * Return: The page on success or NULL if allocation fails. | |
1da177e4 | 2280 | */ |
d7f946d0 | 2281 | struct page *alloc_pages(gfp_t gfp, unsigned order) |
1da177e4 | 2282 | { |
8d90274b | 2283 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2284 | struct page *page; |
1da177e4 | 2285 | |
8d90274b ON |
2286 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2287 | pol = get_task_policy(current); | |
52cd3b07 LS |
2288 | |
2289 | /* | |
2290 | * No reference counting needed for current->mempolicy | |
2291 | * nor system default_policy | |
2292 | */ | |
45c4745a | 2293 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 | 2294 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
4c54d949 FT |
2295 | else if (pol->mode == MPOL_PREFERRED_MANY) |
2296 | page = alloc_pages_preferred_many(gfp, order, | |
c0455116 | 2297 | policy_node(gfp, pol, numa_node_id()), pol); |
c0ff7453 | 2298 | else |
84172f4b | 2299 | page = __alloc_pages(gfp, order, |
04ec6264 | 2300 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2301 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2302 | |
c0ff7453 | 2303 | return page; |
1da177e4 | 2304 | } |
d7f946d0 | 2305 | EXPORT_SYMBOL(alloc_pages); |
1da177e4 | 2306 | |
cc09cb13 MWO |
2307 | struct folio *folio_alloc(gfp_t gfp, unsigned order) |
2308 | { | |
2309 | struct page *page = alloc_pages(gfp | __GFP_COMP, order); | |
da6e7bf3 | 2310 | struct folio *folio = (struct folio *)page; |
cc09cb13 | 2311 | |
da6e7bf3 MWO |
2312 | if (folio && order > 1) |
2313 | folio_prep_large_rmappable(folio); | |
2314 | return folio; | |
cc09cb13 MWO |
2315 | } |
2316 | EXPORT_SYMBOL(folio_alloc); | |
2317 | ||
c00b6b96 CW |
2318 | static unsigned long alloc_pages_bulk_array_interleave(gfp_t gfp, |
2319 | struct mempolicy *pol, unsigned long nr_pages, | |
2320 | struct page **page_array) | |
2321 | { | |
2322 | int nodes; | |
2323 | unsigned long nr_pages_per_node; | |
2324 | int delta; | |
2325 | int i; | |
2326 | unsigned long nr_allocated; | |
2327 | unsigned long total_allocated = 0; | |
2328 | ||
2329 | nodes = nodes_weight(pol->nodes); | |
2330 | nr_pages_per_node = nr_pages / nodes; | |
2331 | delta = nr_pages - nodes * nr_pages_per_node; | |
2332 | ||
2333 | for (i = 0; i < nodes; i++) { | |
2334 | if (delta) { | |
2335 | nr_allocated = __alloc_pages_bulk(gfp, | |
2336 | interleave_nodes(pol), NULL, | |
2337 | nr_pages_per_node + 1, NULL, | |
2338 | page_array); | |
2339 | delta--; | |
2340 | } else { | |
2341 | nr_allocated = __alloc_pages_bulk(gfp, | |
2342 | interleave_nodes(pol), NULL, | |
2343 | nr_pages_per_node, NULL, page_array); | |
2344 | } | |
2345 | ||
2346 | page_array += nr_allocated; | |
2347 | total_allocated += nr_allocated; | |
2348 | } | |
2349 | ||
2350 | return total_allocated; | |
2351 | } | |
2352 | ||
2353 | static unsigned long alloc_pages_bulk_array_preferred_many(gfp_t gfp, int nid, | |
2354 | struct mempolicy *pol, unsigned long nr_pages, | |
2355 | struct page **page_array) | |
2356 | { | |
2357 | gfp_t preferred_gfp; | |
2358 | unsigned long nr_allocated = 0; | |
2359 | ||
2360 | preferred_gfp = gfp | __GFP_NOWARN; | |
2361 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2362 | ||
2363 | nr_allocated = __alloc_pages_bulk(preferred_gfp, nid, &pol->nodes, | |
2364 | nr_pages, NULL, page_array); | |
2365 | ||
2366 | if (nr_allocated < nr_pages) | |
2367 | nr_allocated += __alloc_pages_bulk(gfp, numa_node_id(), NULL, | |
2368 | nr_pages - nr_allocated, NULL, | |
2369 | page_array + nr_allocated); | |
2370 | return nr_allocated; | |
2371 | } | |
2372 | ||
2373 | /* alloc pages bulk and mempolicy should be considered at the | |
2374 | * same time in some situation such as vmalloc. | |
2375 | * | |
2376 | * It can accelerate memory allocation especially interleaving | |
2377 | * allocate memory. | |
2378 | */ | |
2379 | unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, | |
2380 | unsigned long nr_pages, struct page **page_array) | |
2381 | { | |
2382 | struct mempolicy *pol = &default_policy; | |
2383 | ||
2384 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) | |
2385 | pol = get_task_policy(current); | |
2386 | ||
2387 | if (pol->mode == MPOL_INTERLEAVE) | |
2388 | return alloc_pages_bulk_array_interleave(gfp, pol, | |
2389 | nr_pages, page_array); | |
2390 | ||
2391 | if (pol->mode == MPOL_PREFERRED_MANY) | |
2392 | return alloc_pages_bulk_array_preferred_many(gfp, | |
2393 | numa_node_id(), pol, nr_pages, page_array); | |
2394 | ||
2395 | return __alloc_pages_bulk(gfp, policy_node(gfp, pol, numa_node_id()), | |
2396 | policy_nodemask(gfp, pol), nr_pages, NULL, | |
2397 | page_array); | |
2398 | } | |
2399 | ||
ef0855d3 ON |
2400 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2401 | { | |
2402 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2403 | ||
2404 | if (IS_ERR(pol)) | |
2405 | return PTR_ERR(pol); | |
2406 | dst->vm_policy = pol; | |
2407 | return 0; | |
2408 | } | |
2409 | ||
4225399a | 2410 | /* |
846a16bf | 2411 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2412 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2413 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2414 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2415 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2416 | * |
2417 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2418 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2419 | */ |
4225399a | 2420 | |
846a16bf LS |
2421 | /* Slow path of a mempolicy duplicate */ |
2422 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2423 | { |
2424 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2425 | ||
2426 | if (!new) | |
2427 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2428 | |
2429 | /* task's mempolicy is protected by alloc_lock */ | |
2430 | if (old == current->mempolicy) { | |
2431 | task_lock(current); | |
2432 | *new = *old; | |
2433 | task_unlock(current); | |
2434 | } else | |
2435 | *new = *old; | |
2436 | ||
4225399a PJ |
2437 | if (current_cpuset_is_being_rebound()) { |
2438 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2439 | mpol_rebind_policy(new, &mems); |
4225399a | 2440 | } |
1da177e4 | 2441 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2442 | return new; |
2443 | } | |
2444 | ||
2445 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2446 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2447 | { |
2448 | if (!a || !b) | |
fcfb4dcc | 2449 | return false; |
45c4745a | 2450 | if (a->mode != b->mode) |
fcfb4dcc | 2451 | return false; |
19800502 | 2452 | if (a->flags != b->flags) |
fcfb4dcc | 2453 | return false; |
c6018b4b AK |
2454 | if (a->home_node != b->home_node) |
2455 | return false; | |
19800502 BL |
2456 | if (mpol_store_user_nodemask(a)) |
2457 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2458 | return false; |
19800502 | 2459 | |
45c4745a | 2460 | switch (a->mode) { |
19770b32 | 2461 | case MPOL_BIND: |
1da177e4 | 2462 | case MPOL_INTERLEAVE: |
1da177e4 | 2463 | case MPOL_PREFERRED: |
b27abacc | 2464 | case MPOL_PREFERRED_MANY: |
269fbe72 | 2465 | return !!nodes_equal(a->nodes, b->nodes); |
7858d7bc FT |
2466 | case MPOL_LOCAL: |
2467 | return true; | |
1da177e4 LT |
2468 | default: |
2469 | BUG(); | |
fcfb4dcc | 2470 | return false; |
1da177e4 LT |
2471 | } |
2472 | } | |
2473 | ||
1da177e4 LT |
2474 | /* |
2475 | * Shared memory backing store policy support. | |
2476 | * | |
2477 | * Remember policies even when nobody has shared memory mapped. | |
2478 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2479 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2480 | * for any accesses to the tree. |
2481 | */ | |
2482 | ||
4a8c7bb5 NZ |
2483 | /* |
2484 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2485 | * reading or for writing | |
2486 | */ | |
1da177e4 LT |
2487 | static struct sp_node * |
2488 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2489 | { | |
2490 | struct rb_node *n = sp->root.rb_node; | |
2491 | ||
2492 | while (n) { | |
2493 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2494 | ||
2495 | if (start >= p->end) | |
2496 | n = n->rb_right; | |
2497 | else if (end <= p->start) | |
2498 | n = n->rb_left; | |
2499 | else | |
2500 | break; | |
2501 | } | |
2502 | if (!n) | |
2503 | return NULL; | |
2504 | for (;;) { | |
2505 | struct sp_node *w = NULL; | |
2506 | struct rb_node *prev = rb_prev(n); | |
2507 | if (!prev) | |
2508 | break; | |
2509 | w = rb_entry(prev, struct sp_node, nd); | |
2510 | if (w->end <= start) | |
2511 | break; | |
2512 | n = prev; | |
2513 | } | |
2514 | return rb_entry(n, struct sp_node, nd); | |
2515 | } | |
2516 | ||
4a8c7bb5 NZ |
2517 | /* |
2518 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2519 | * writing. | |
2520 | */ | |
1da177e4 LT |
2521 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2522 | { | |
2523 | struct rb_node **p = &sp->root.rb_node; | |
2524 | struct rb_node *parent = NULL; | |
2525 | struct sp_node *nd; | |
2526 | ||
2527 | while (*p) { | |
2528 | parent = *p; | |
2529 | nd = rb_entry(parent, struct sp_node, nd); | |
2530 | if (new->start < nd->start) | |
2531 | p = &(*p)->rb_left; | |
2532 | else if (new->end > nd->end) | |
2533 | p = &(*p)->rb_right; | |
2534 | else | |
2535 | BUG(); | |
2536 | } | |
2537 | rb_link_node(&new->nd, parent, p); | |
2538 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2539 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2540 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2541 | } |
2542 | ||
2543 | /* Find shared policy intersecting idx */ | |
2544 | struct mempolicy * | |
2545 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2546 | { | |
2547 | struct mempolicy *pol = NULL; | |
2548 | struct sp_node *sn; | |
2549 | ||
2550 | if (!sp->root.rb_node) | |
2551 | return NULL; | |
4a8c7bb5 | 2552 | read_lock(&sp->lock); |
1da177e4 LT |
2553 | sn = sp_lookup(sp, idx, idx+1); |
2554 | if (sn) { | |
2555 | mpol_get(sn->policy); | |
2556 | pol = sn->policy; | |
2557 | } | |
4a8c7bb5 | 2558 | read_unlock(&sp->lock); |
1da177e4 LT |
2559 | return pol; |
2560 | } | |
2561 | ||
63f74ca2 KM |
2562 | static void sp_free(struct sp_node *n) |
2563 | { | |
2564 | mpol_put(n->policy); | |
2565 | kmem_cache_free(sn_cache, n); | |
2566 | } | |
2567 | ||
771fb4d8 LS |
2568 | /** |
2569 | * mpol_misplaced - check whether current page node is valid in policy | |
2570 | * | |
b46e14ac FF |
2571 | * @page: page to be checked |
2572 | * @vma: vm area where page mapped | |
2573 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2574 | * |
2575 | * Lookup current policy node id for vma,addr and "compare to" page's | |
5f076944 | 2576 | * node id. Policy determination "mimics" alloc_page_vma(). |
771fb4d8 | 2577 | * Called from fault path where we know the vma and faulting address. |
5f076944 | 2578 | * |
062db293 BW |
2579 | * Return: NUMA_NO_NODE if the page is in a node that is valid for this |
2580 | * policy, or a suitable node ID to allocate a replacement page from. | |
771fb4d8 LS |
2581 | */ |
2582 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2583 | { | |
2584 | struct mempolicy *pol; | |
c33d6c06 | 2585 | struct zoneref *z; |
771fb4d8 LS |
2586 | int curnid = page_to_nid(page); |
2587 | unsigned long pgoff; | |
90572890 PZ |
2588 | int thiscpu = raw_smp_processor_id(); |
2589 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2590 | int polnid = NUMA_NO_NODE; |
062db293 | 2591 | int ret = NUMA_NO_NODE; |
771fb4d8 | 2592 | |
dd6eecb9 | 2593 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2594 | if (!(pol->flags & MPOL_F_MOF)) |
2595 | goto out; | |
2596 | ||
2597 | switch (pol->mode) { | |
2598 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2599 | pgoff = vma->vm_pgoff; |
2600 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2601 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2602 | break; |
2603 | ||
2604 | case MPOL_PREFERRED: | |
b27abacc DH |
2605 | if (node_isset(curnid, pol->nodes)) |
2606 | goto out; | |
269fbe72 | 2607 | polnid = first_node(pol->nodes); |
7858d7bc FT |
2608 | break; |
2609 | ||
2610 | case MPOL_LOCAL: | |
2611 | polnid = numa_node_id(); | |
771fb4d8 LS |
2612 | break; |
2613 | ||
2614 | case MPOL_BIND: | |
bda420b9 HY |
2615 | /* Optimize placement among multiple nodes via NUMA balancing */ |
2616 | if (pol->flags & MPOL_F_MORON) { | |
269fbe72 | 2617 | if (node_isset(thisnid, pol->nodes)) |
bda420b9 HY |
2618 | break; |
2619 | goto out; | |
2620 | } | |
b27abacc | 2621 | fallthrough; |
c33d6c06 | 2622 | |
b27abacc | 2623 | case MPOL_PREFERRED_MANY: |
771fb4d8 | 2624 | /* |
771fb4d8 LS |
2625 | * use current page if in policy nodemask, |
2626 | * else select nearest allowed node, if any. | |
2627 | * If no allowed nodes, use current [!misplaced]. | |
2628 | */ | |
269fbe72 | 2629 | if (node_isset(curnid, pol->nodes)) |
771fb4d8 | 2630 | goto out; |
c33d6c06 | 2631 | z = first_zones_zonelist( |
771fb4d8 LS |
2632 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2633 | gfp_zone(GFP_HIGHUSER), | |
269fbe72 | 2634 | &pol->nodes); |
c1093b74 | 2635 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2636 | break; |
2637 | ||
2638 | default: | |
2639 | BUG(); | |
2640 | } | |
5606e387 MG |
2641 | |
2642 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2643 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2644 | polnid = thisnid; |
5606e387 | 2645 | |
10f39042 | 2646 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2647 | goto out; |
e42c8ff2 MG |
2648 | } |
2649 | ||
771fb4d8 LS |
2650 | if (curnid != polnid) |
2651 | ret = polnid; | |
2652 | out: | |
2653 | mpol_cond_put(pol); | |
2654 | ||
2655 | return ret; | |
2656 | } | |
2657 | ||
c11600e4 DR |
2658 | /* |
2659 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2660 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2661 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2662 | * policy. | |
2663 | */ | |
2664 | void mpol_put_task_policy(struct task_struct *task) | |
2665 | { | |
2666 | struct mempolicy *pol; | |
2667 | ||
2668 | task_lock(task); | |
2669 | pol = task->mempolicy; | |
2670 | task->mempolicy = NULL; | |
2671 | task_unlock(task); | |
2672 | mpol_put(pol); | |
2673 | } | |
2674 | ||
1da177e4 LT |
2675 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2676 | { | |
140d5a49 | 2677 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2678 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2679 | sp_free(n); |
1da177e4 LT |
2680 | } |
2681 | ||
42288fe3 MG |
2682 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2683 | unsigned long end, struct mempolicy *pol) | |
2684 | { | |
2685 | node->start = start; | |
2686 | node->end = end; | |
2687 | node->policy = pol; | |
2688 | } | |
2689 | ||
dbcb0f19 AB |
2690 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2691 | struct mempolicy *pol) | |
1da177e4 | 2692 | { |
869833f2 KM |
2693 | struct sp_node *n; |
2694 | struct mempolicy *newpol; | |
1da177e4 | 2695 | |
869833f2 | 2696 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2697 | if (!n) |
2698 | return NULL; | |
869833f2 KM |
2699 | |
2700 | newpol = mpol_dup(pol); | |
2701 | if (IS_ERR(newpol)) { | |
2702 | kmem_cache_free(sn_cache, n); | |
2703 | return NULL; | |
2704 | } | |
2705 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2706 | sp_node_init(n, start, end, newpol); |
869833f2 | 2707 | |
1da177e4 LT |
2708 | return n; |
2709 | } | |
2710 | ||
2711 | /* Replace a policy range. */ | |
2712 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2713 | unsigned long end, struct sp_node *new) | |
2714 | { | |
b22d127a | 2715 | struct sp_node *n; |
42288fe3 MG |
2716 | struct sp_node *n_new = NULL; |
2717 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2718 | int ret = 0; |
1da177e4 | 2719 | |
42288fe3 | 2720 | restart: |
4a8c7bb5 | 2721 | write_lock(&sp->lock); |
1da177e4 LT |
2722 | n = sp_lookup(sp, start, end); |
2723 | /* Take care of old policies in the same range. */ | |
2724 | while (n && n->start < end) { | |
2725 | struct rb_node *next = rb_next(&n->nd); | |
2726 | if (n->start >= start) { | |
2727 | if (n->end <= end) | |
2728 | sp_delete(sp, n); | |
2729 | else | |
2730 | n->start = end; | |
2731 | } else { | |
2732 | /* Old policy spanning whole new range. */ | |
2733 | if (n->end > end) { | |
42288fe3 MG |
2734 | if (!n_new) |
2735 | goto alloc_new; | |
2736 | ||
2737 | *mpol_new = *n->policy; | |
2738 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2739 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2740 | n->end = start; |
5ca39575 | 2741 | sp_insert(sp, n_new); |
42288fe3 MG |
2742 | n_new = NULL; |
2743 | mpol_new = NULL; | |
1da177e4 LT |
2744 | break; |
2745 | } else | |
2746 | n->end = start; | |
2747 | } | |
2748 | if (!next) | |
2749 | break; | |
2750 | n = rb_entry(next, struct sp_node, nd); | |
2751 | } | |
2752 | if (new) | |
2753 | sp_insert(sp, new); | |
4a8c7bb5 | 2754 | write_unlock(&sp->lock); |
42288fe3 MG |
2755 | ret = 0; |
2756 | ||
2757 | err_out: | |
2758 | if (mpol_new) | |
2759 | mpol_put(mpol_new); | |
2760 | if (n_new) | |
2761 | kmem_cache_free(sn_cache, n_new); | |
2762 | ||
b22d127a | 2763 | return ret; |
42288fe3 MG |
2764 | |
2765 | alloc_new: | |
4a8c7bb5 | 2766 | write_unlock(&sp->lock); |
42288fe3 MG |
2767 | ret = -ENOMEM; |
2768 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2769 | if (!n_new) | |
2770 | goto err_out; | |
2771 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2772 | if (!mpol_new) | |
2773 | goto err_out; | |
4ad09955 | 2774 | atomic_set(&mpol_new->refcnt, 1); |
42288fe3 | 2775 | goto restart; |
1da177e4 LT |
2776 | } |
2777 | ||
71fe804b LS |
2778 | /** |
2779 | * mpol_shared_policy_init - initialize shared policy for inode | |
2780 | * @sp: pointer to inode shared policy | |
2781 | * @mpol: struct mempolicy to install | |
2782 | * | |
2783 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2784 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2785 | * This must be released on exit. | |
4bfc4495 | 2786 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2787 | */ |
2788 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2789 | { | |
58568d2a MX |
2790 | int ret; |
2791 | ||
71fe804b | 2792 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2793 | rwlock_init(&sp->lock); |
71fe804b LS |
2794 | |
2795 | if (mpol) { | |
2796 | struct vm_area_struct pvma; | |
2797 | struct mempolicy *new; | |
4bfc4495 | 2798 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2799 | |
4bfc4495 | 2800 | if (!scratch) |
5c0c1654 | 2801 | goto put_mpol; |
71fe804b LS |
2802 | /* contextualize the tmpfs mount point mempolicy */ |
2803 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2804 | if (IS_ERR(new)) |
0cae3457 | 2805 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2806 | |
2807 | task_lock(current); | |
4bfc4495 | 2808 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2809 | task_unlock(current); |
15d77835 | 2810 | if (ret) |
5c0c1654 | 2811 | goto put_new; |
71fe804b LS |
2812 | |
2813 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2814 | vma_init(&pvma, NULL); |
71fe804b LS |
2815 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2816 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2817 | |
5c0c1654 | 2818 | put_new: |
71fe804b | 2819 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2820 | free_scratch: |
4bfc4495 | 2821 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2822 | put_mpol: |
2823 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2824 | } |
2825 | } | |
2826 | ||
1da177e4 LT |
2827 | int mpol_set_shared_policy(struct shared_policy *info, |
2828 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2829 | { | |
2830 | int err; | |
2831 | struct sp_node *new = NULL; | |
2832 | unsigned long sz = vma_pages(vma); | |
2833 | ||
028fec41 | 2834 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2835 | vma->vm_pgoff, |
45c4745a | 2836 | sz, npol ? npol->mode : -1, |
028fec41 | 2837 | npol ? npol->flags : -1, |
269fbe72 | 2838 | npol ? nodes_addr(npol->nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2839 | |
2840 | if (npol) { | |
2841 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2842 | if (!new) | |
2843 | return -ENOMEM; | |
2844 | } | |
2845 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2846 | if (err && new) | |
63f74ca2 | 2847 | sp_free(new); |
1da177e4 LT |
2848 | return err; |
2849 | } | |
2850 | ||
2851 | /* Free a backing policy store on inode delete. */ | |
2852 | void mpol_free_shared_policy(struct shared_policy *p) | |
2853 | { | |
2854 | struct sp_node *n; | |
2855 | struct rb_node *next; | |
2856 | ||
2857 | if (!p->root.rb_node) | |
2858 | return; | |
4a8c7bb5 | 2859 | write_lock(&p->lock); |
1da177e4 LT |
2860 | next = rb_first(&p->root); |
2861 | while (next) { | |
2862 | n = rb_entry(next, struct sp_node, nd); | |
2863 | next = rb_next(&n->nd); | |
63f74ca2 | 2864 | sp_delete(p, n); |
1da177e4 | 2865 | } |
4a8c7bb5 | 2866 | write_unlock(&p->lock); |
1da177e4 LT |
2867 | } |
2868 | ||
1a687c2e | 2869 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2870 | static int __initdata numabalancing_override; |
1a687c2e MG |
2871 | |
2872 | static void __init check_numabalancing_enable(void) | |
2873 | { | |
2874 | bool numabalancing_default = false; | |
2875 | ||
2876 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2877 | numabalancing_default = true; | |
2878 | ||
c297663c MG |
2879 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2880 | if (numabalancing_override) | |
2881 | set_numabalancing_state(numabalancing_override == 1); | |
2882 | ||
b0dc2b9b | 2883 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2884 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2885 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2886 | set_numabalancing_state(numabalancing_default); |
2887 | } | |
2888 | } | |
2889 | ||
2890 | static int __init setup_numabalancing(char *str) | |
2891 | { | |
2892 | int ret = 0; | |
2893 | if (!str) | |
2894 | goto out; | |
1a687c2e MG |
2895 | |
2896 | if (!strcmp(str, "enable")) { | |
c297663c | 2897 | numabalancing_override = 1; |
1a687c2e MG |
2898 | ret = 1; |
2899 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2900 | numabalancing_override = -1; |
1a687c2e MG |
2901 | ret = 1; |
2902 | } | |
2903 | out: | |
2904 | if (!ret) | |
4a404bea | 2905 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2906 | |
2907 | return ret; | |
2908 | } | |
2909 | __setup("numa_balancing=", setup_numabalancing); | |
2910 | #else | |
2911 | static inline void __init check_numabalancing_enable(void) | |
2912 | { | |
2913 | } | |
2914 | #endif /* CONFIG_NUMA_BALANCING */ | |
2915 | ||
1da177e4 LT |
2916 | /* assumes fs == KERNEL_DS */ |
2917 | void __init numa_policy_init(void) | |
2918 | { | |
b71636e2 PM |
2919 | nodemask_t interleave_nodes; |
2920 | unsigned long largest = 0; | |
2921 | int nid, prefer = 0; | |
2922 | ||
1da177e4 LT |
2923 | policy_cache = kmem_cache_create("numa_policy", |
2924 | sizeof(struct mempolicy), | |
20c2df83 | 2925 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2926 | |
2927 | sn_cache = kmem_cache_create("shared_policy_node", | |
2928 | sizeof(struct sp_node), | |
20c2df83 | 2929 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2930 | |
5606e387 MG |
2931 | for_each_node(nid) { |
2932 | preferred_node_policy[nid] = (struct mempolicy) { | |
2933 | .refcnt = ATOMIC_INIT(1), | |
2934 | .mode = MPOL_PREFERRED, | |
2935 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
269fbe72 | 2936 | .nodes = nodemask_of_node(nid), |
5606e387 MG |
2937 | }; |
2938 | } | |
2939 | ||
b71636e2 PM |
2940 | /* |
2941 | * Set interleaving policy for system init. Interleaving is only | |
2942 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2943 | * fall back to the largest node if they're all smaller. | |
2944 | */ | |
2945 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2946 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2947 | unsigned long total_pages = node_present_pages(nid); |
2948 | ||
2949 | /* Preserve the largest node */ | |
2950 | if (largest < total_pages) { | |
2951 | largest = total_pages; | |
2952 | prefer = nid; | |
2953 | } | |
2954 | ||
2955 | /* Interleave this node? */ | |
2956 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2957 | node_set(nid, interleave_nodes); | |
2958 | } | |
2959 | ||
2960 | /* All too small, use the largest */ | |
2961 | if (unlikely(nodes_empty(interleave_nodes))) | |
2962 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2963 | |
028fec41 | 2964 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2965 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2966 | |
2967 | check_numabalancing_enable(); | |
1da177e4 LT |
2968 | } |
2969 | ||
8bccd85f | 2970 | /* Reset policy of current process to default */ |
1da177e4 LT |
2971 | void numa_default_policy(void) |
2972 | { | |
028fec41 | 2973 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2974 | } |
68860ec1 | 2975 | |
095f1fc4 LS |
2976 | /* |
2977 | * Parse and format mempolicy from/to strings | |
2978 | */ | |
2979 | ||
345ace9c LS |
2980 | static const char * const policy_modes[] = |
2981 | { | |
2982 | [MPOL_DEFAULT] = "default", | |
2983 | [MPOL_PREFERRED] = "prefer", | |
2984 | [MPOL_BIND] = "bind", | |
2985 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2986 | [MPOL_LOCAL] = "local", |
b27abacc | 2987 | [MPOL_PREFERRED_MANY] = "prefer (many)", |
345ace9c | 2988 | }; |
1a75a6c8 | 2989 | |
095f1fc4 LS |
2990 | |
2991 | #ifdef CONFIG_TMPFS | |
2992 | /** | |
f2a07f40 | 2993 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2994 | * @str: string containing mempolicy to parse |
71fe804b | 2995 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2996 | * |
2997 | * Format of input: | |
2998 | * <mode>[=<flags>][:<nodelist>] | |
2999 | * | |
dad5b023 | 3000 | * Return: %0 on success, else %1 |
095f1fc4 | 3001 | */ |
a7a88b23 | 3002 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 3003 | { |
71fe804b | 3004 | struct mempolicy *new = NULL; |
f2a07f40 | 3005 | unsigned short mode_flags; |
71fe804b | 3006 | nodemask_t nodes; |
095f1fc4 LS |
3007 | char *nodelist = strchr(str, ':'); |
3008 | char *flags = strchr(str, '='); | |
dedf2c73 | 3009 | int err = 1, mode; |
095f1fc4 | 3010 | |
c7a91bc7 DC |
3011 | if (flags) |
3012 | *flags++ = '\0'; /* terminate mode string */ | |
3013 | ||
095f1fc4 LS |
3014 | if (nodelist) { |
3015 | /* NUL-terminate mode or flags string */ | |
3016 | *nodelist++ = '\0'; | |
71fe804b | 3017 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 3018 | goto out; |
01f13bd6 | 3019 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 3020 | goto out; |
71fe804b LS |
3021 | } else |
3022 | nodes_clear(nodes); | |
3023 | ||
dedf2c73 | 3024 | mode = match_string(policy_modes, MPOL_MAX, str); |
3025 | if (mode < 0) | |
095f1fc4 LS |
3026 | goto out; |
3027 | ||
71fe804b | 3028 | switch (mode) { |
095f1fc4 | 3029 | case MPOL_PREFERRED: |
71fe804b | 3030 | /* |
aa9f7d51 RD |
3031 | * Insist on a nodelist of one node only, although later |
3032 | * we use first_node(nodes) to grab a single node, so here | |
3033 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 3034 | */ |
095f1fc4 LS |
3035 | if (nodelist) { |
3036 | char *rest = nodelist; | |
3037 | while (isdigit(*rest)) | |
3038 | rest++; | |
926f2ae0 KM |
3039 | if (*rest) |
3040 | goto out; | |
aa9f7d51 RD |
3041 | if (nodes_empty(nodes)) |
3042 | goto out; | |
095f1fc4 LS |
3043 | } |
3044 | break; | |
095f1fc4 LS |
3045 | case MPOL_INTERLEAVE: |
3046 | /* | |
3047 | * Default to online nodes with memory if no nodelist | |
3048 | */ | |
3049 | if (!nodelist) | |
01f13bd6 | 3050 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 3051 | break; |
71fe804b | 3052 | case MPOL_LOCAL: |
3f226aa1 | 3053 | /* |
71fe804b | 3054 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 3055 | */ |
71fe804b | 3056 | if (nodelist) |
3f226aa1 | 3057 | goto out; |
3f226aa1 | 3058 | break; |
413b43de RT |
3059 | case MPOL_DEFAULT: |
3060 | /* | |
3061 | * Insist on a empty nodelist | |
3062 | */ | |
3063 | if (!nodelist) | |
3064 | err = 0; | |
3065 | goto out; | |
b27abacc | 3066 | case MPOL_PREFERRED_MANY: |
d69b2e63 KM |
3067 | case MPOL_BIND: |
3068 | /* | |
3069 | * Insist on a nodelist | |
3070 | */ | |
3071 | if (!nodelist) | |
3072 | goto out; | |
095f1fc4 LS |
3073 | } |
3074 | ||
71fe804b | 3075 | mode_flags = 0; |
095f1fc4 LS |
3076 | if (flags) { |
3077 | /* | |
3078 | * Currently, we only support two mutually exclusive | |
3079 | * mode flags. | |
3080 | */ | |
3081 | if (!strcmp(flags, "static")) | |
71fe804b | 3082 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 3083 | else if (!strcmp(flags, "relative")) |
71fe804b | 3084 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 3085 | else |
926f2ae0 | 3086 | goto out; |
095f1fc4 | 3087 | } |
71fe804b LS |
3088 | |
3089 | new = mpol_new(mode, mode_flags, &nodes); | |
3090 | if (IS_ERR(new)) | |
926f2ae0 KM |
3091 | goto out; |
3092 | ||
f2a07f40 HD |
3093 | /* |
3094 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
3095 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
3096 | */ | |
269fbe72 BW |
3097 | if (mode != MPOL_PREFERRED) { |
3098 | new->nodes = nodes; | |
3099 | } else if (nodelist) { | |
3100 | nodes_clear(new->nodes); | |
3101 | node_set(first_node(nodes), new->nodes); | |
3102 | } else { | |
7858d7bc | 3103 | new->mode = MPOL_LOCAL; |
269fbe72 | 3104 | } |
f2a07f40 HD |
3105 | |
3106 | /* | |
3107 | * Save nodes for contextualization: this will be used to "clone" | |
3108 | * the mempolicy in a specific context [cpuset] at a later time. | |
3109 | */ | |
3110 | new->w.user_nodemask = nodes; | |
3111 | ||
926f2ae0 | 3112 | err = 0; |
71fe804b | 3113 | |
095f1fc4 LS |
3114 | out: |
3115 | /* Restore string for error message */ | |
3116 | if (nodelist) | |
3117 | *--nodelist = ':'; | |
3118 | if (flags) | |
3119 | *--flags = '='; | |
71fe804b LS |
3120 | if (!err) |
3121 | *mpol = new; | |
095f1fc4 LS |
3122 | return err; |
3123 | } | |
3124 | #endif /* CONFIG_TMPFS */ | |
3125 | ||
71fe804b LS |
3126 | /** |
3127 | * mpol_to_str - format a mempolicy structure for printing | |
3128 | * @buffer: to contain formatted mempolicy string | |
3129 | * @maxlen: length of @buffer | |
3130 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3131 | * |
948927ee DR |
3132 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3133 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3134 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3135 | */ |
948927ee | 3136 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3137 | { |
3138 | char *p = buffer; | |
948927ee DR |
3139 | nodemask_t nodes = NODE_MASK_NONE; |
3140 | unsigned short mode = MPOL_DEFAULT; | |
3141 | unsigned short flags = 0; | |
2291990a | 3142 | |
8790c71a | 3143 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3144 | mode = pol->mode; |
948927ee DR |
3145 | flags = pol->flags; |
3146 | } | |
bea904d5 | 3147 | |
1a75a6c8 CL |
3148 | switch (mode) { |
3149 | case MPOL_DEFAULT: | |
7858d7bc | 3150 | case MPOL_LOCAL: |
1a75a6c8 | 3151 | break; |
1a75a6c8 | 3152 | case MPOL_PREFERRED: |
b27abacc | 3153 | case MPOL_PREFERRED_MANY: |
1a75a6c8 | 3154 | case MPOL_BIND: |
1a75a6c8 | 3155 | case MPOL_INTERLEAVE: |
269fbe72 | 3156 | nodes = pol->nodes; |
1a75a6c8 | 3157 | break; |
1a75a6c8 | 3158 | default: |
948927ee DR |
3159 | WARN_ON_ONCE(1); |
3160 | snprintf(p, maxlen, "unknown"); | |
3161 | return; | |
1a75a6c8 CL |
3162 | } |
3163 | ||
b7a9f420 | 3164 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3165 | |
fc36b8d3 | 3166 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3167 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3168 | |
2291990a LS |
3169 | /* |
3170 | * Currently, the only defined flags are mutually exclusive | |
3171 | */ | |
f5b087b5 | 3172 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3173 | p += snprintf(p, buffer + maxlen - p, "static"); |
3174 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3175 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3176 | } |
3177 | ||
9e763e0f TH |
3178 | if (!nodes_empty(nodes)) |
3179 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3180 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3181 | } |