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 | * |
fa3bea4e GP |
22 | * weighted interleave |
23 | * Allocate memory interleaved over a set of nodes based on | |
24 | * a set of weights (per-node), with normal fallback if it | |
25 | * fails. Otherwise operates the same as interleave. | |
26 | * Example: nodeset(0,1) & weights (2,1) - 2 pages allocated | |
27 | * on node 0 for every 1 page allocated on node 1. | |
28 | * | |
1da177e4 LT |
29 | * bind Only allocate memory on a specific set of nodes, |
30 | * no fallback. | |
8bccd85f CL |
31 | * FIXME: memory is allocated starting with the first node |
32 | * to the last. It would be better if bind would truly restrict | |
33 | * the allocation to memory nodes instead | |
34 | * | |
c36f6e6d | 35 | * preferred Try a specific node first before normal fallback. |
00ef2d2f | 36 | * As a special case NUMA_NO_NODE here means do the allocation |
1da177e4 LT |
37 | * on the local CPU. This is normally identical to default, |
38 | * but useful to set in a VMA when you have a non default | |
39 | * process policy. | |
8bccd85f | 40 | * |
b27abacc DH |
41 | * preferred many Try a set of nodes first before normal fallback. This is |
42 | * similar to preferred without the special case. | |
43 | * | |
1da177e4 LT |
44 | * default Allocate on the local node first, or when on a VMA |
45 | * use the process policy. This is what Linux always did | |
46 | * in a NUMA aware kernel and still does by, ahem, default. | |
47 | * | |
48 | * The process policy is applied for most non interrupt memory allocations | |
49 | * in that process' context. Interrupts ignore the policies and always | |
50 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
51 | * allocations for a VMA in the VM. | |
52 | * | |
53 | * Currently there are a few corner cases in swapping where the policy | |
54 | * is not applied, but the majority should be handled. When process policy | |
55 | * is used it is not remembered over swap outs/swap ins. | |
56 | * | |
57 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
58 | * requesting a lower zone just use default policy. This implies that | |
59 | * on systems with highmem kernel lowmem allocation don't get policied. | |
60 | * Same with GFP_DMA allocations. | |
61 | * | |
c36f6e6d | 62 | * For shmem/tmpfs shared memory the policy is shared between |
1da177e4 LT |
63 | * all users and remembered even when nobody has memory mapped. |
64 | */ | |
65 | ||
66 | /* Notebook: | |
67 | fix mmap readahead to honour policy and enable policy for any page cache | |
68 | object | |
69 | statistics for bigpages | |
70 | global policy for page cache? currently it uses process policy. Requires | |
71 | first item above. | |
72 | handle mremap for shared memory (currently ignored for the policy) | |
73 | grows down? | |
74 | make bind policy root only? It can trigger oom much faster and the | |
75 | kernel is not always grateful with that. | |
1da177e4 LT |
76 | */ |
77 | ||
b1de0d13 MH |
78 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
79 | ||
1da177e4 | 80 | #include <linux/mempolicy.h> |
a520110e | 81 | #include <linux/pagewalk.h> |
1da177e4 LT |
82 | #include <linux/highmem.h> |
83 | #include <linux/hugetlb.h> | |
84 | #include <linux/kernel.h> | |
85 | #include <linux/sched.h> | |
6e84f315 | 86 | #include <linux/sched/mm.h> |
6a3827d7 | 87 | #include <linux/sched/numa_balancing.h> |
f719ff9b | 88 | #include <linux/sched/task.h> |
1da177e4 LT |
89 | #include <linux/nodemask.h> |
90 | #include <linux/cpuset.h> | |
1da177e4 LT |
91 | #include <linux/slab.h> |
92 | #include <linux/string.h> | |
b95f1b31 | 93 | #include <linux/export.h> |
b488893a | 94 | #include <linux/nsproxy.h> |
1da177e4 LT |
95 | #include <linux/interrupt.h> |
96 | #include <linux/init.h> | |
97 | #include <linux/compat.h> | |
31367466 | 98 | #include <linux/ptrace.h> |
dc9aa5b9 | 99 | #include <linux/swap.h> |
1a75a6c8 CL |
100 | #include <linux/seq_file.h> |
101 | #include <linux/proc_fs.h> | |
b20a3503 | 102 | #include <linux/migrate.h> |
62b61f61 | 103 | #include <linux/ksm.h> |
95a402c3 | 104 | #include <linux/rmap.h> |
86c3a764 | 105 | #include <linux/security.h> |
dbcb0f19 | 106 | #include <linux/syscalls.h> |
095f1fc4 | 107 | #include <linux/ctype.h> |
6d9c285a | 108 | #include <linux/mm_inline.h> |
b24f53a0 | 109 | #include <linux/mmu_notifier.h> |
b1de0d13 | 110 | #include <linux/printk.h> |
c8633798 | 111 | #include <linux/swapops.h> |
dc9aa5b9 | 112 | |
1da177e4 | 113 | #include <asm/tlbflush.h> |
4a18419f | 114 | #include <asm/tlb.h> |
7c0f6ba6 | 115 | #include <linux/uaccess.h> |
1da177e4 | 116 | |
62695a84 NP |
117 | #include "internal.h" |
118 | ||
38e35860 | 119 | /* Internal flags */ |
dc9aa5b9 | 120 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
1cb5d11a HD |
121 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
122 | #define MPOL_MF_WRLOCK (MPOL_MF_INTERNAL << 2) /* Write-lock walked vmas */ | |
dc9aa5b9 | 123 | |
fcc234f8 PE |
124 | static struct kmem_cache *policy_cache; |
125 | static struct kmem_cache *sn_cache; | |
1da177e4 | 126 | |
1da177e4 LT |
127 | /* Highest zone. An specific allocation for a zone below that is not |
128 | policied. */ | |
6267276f | 129 | enum zone_type policy_zone = 0; |
1da177e4 | 130 | |
bea904d5 LS |
131 | /* |
132 | * run-time system-wide default policy => local allocation | |
133 | */ | |
e754d79d | 134 | static struct mempolicy default_policy = { |
1da177e4 | 135 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
7858d7bc | 136 | .mode = MPOL_LOCAL, |
1da177e4 LT |
137 | }; |
138 | ||
5606e387 MG |
139 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
140 | ||
dce41f5a RK |
141 | /* |
142 | * iw_table is the sysfs-set interleave weight table, a value of 0 denotes | |
143 | * system-default value should be used. A NULL iw_table also denotes that | |
144 | * system-default values should be used. Until the system-default table | |
145 | * is implemented, the system-default is always 1. | |
146 | * | |
147 | * iw_table is RCU protected | |
148 | */ | |
149 | static u8 __rcu *iw_table; | |
150 | static DEFINE_MUTEX(iw_table_lock); | |
151 | ||
152 | static u8 get_il_weight(int node) | |
153 | { | |
154 | u8 *table; | |
155 | u8 weight; | |
156 | ||
157 | rcu_read_lock(); | |
158 | table = rcu_dereference(iw_table); | |
159 | /* if no iw_table, use system default */ | |
160 | weight = table ? table[node] : 1; | |
161 | /* if value in iw_table is 0, use system default */ | |
162 | weight = weight ? weight : 1; | |
163 | rcu_read_unlock(); | |
164 | return weight; | |
165 | } | |
166 | ||
b2ca916c | 167 | /** |
b1f099b1 | 168 | * numa_nearest_node - Find nearest node by state |
f6e92f40 | 169 | * @node: Node id to start the search |
b1f099b1 | 170 | * @state: State to filter the search |
b2ca916c | 171 | * |
b1f099b1 | 172 | * Lookup the closest node by distance if @nid is not in state. |
dad5b023 | 173 | * |
b1f099b1 | 174 | * Return: this @node if it is in state, otherwise the closest node by distance |
b2ca916c | 175 | */ |
b1f099b1 | 176 | int numa_nearest_node(int node, unsigned int state) |
b2ca916c | 177 | { |
4fcbe96e | 178 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 179 | |
b1f099b1 YN |
180 | if (state >= NR_NODE_STATES) |
181 | return -EINVAL; | |
182 | ||
183 | if (node == NUMA_NO_NODE || node_state(node, state)) | |
4fcbe96e | 184 | return node; |
b2ca916c DW |
185 | |
186 | min_node = node; | |
b1f099b1 | 187 | for_each_node_state(n, state) { |
4fcbe96e DW |
188 | dist = node_distance(node, n); |
189 | if (dist < min_dist) { | |
190 | min_dist = dist; | |
191 | min_node = n; | |
b2ca916c DW |
192 | } |
193 | } | |
194 | ||
195 | return min_node; | |
196 | } | |
b1f099b1 | 197 | EXPORT_SYMBOL_GPL(numa_nearest_node); |
b2ca916c | 198 | |
74d2c3a0 | 199 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
200 | { |
201 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 202 | int node; |
5606e387 | 203 | |
f15ca78e ON |
204 | if (pol) |
205 | return pol; | |
5606e387 | 206 | |
f15ca78e ON |
207 | node = numa_node_id(); |
208 | if (node != NUMA_NO_NODE) { | |
209 | pol = &preferred_node_policy[node]; | |
210 | /* preferred_node_policy is not initialised early in boot */ | |
211 | if (pol->mode) | |
212 | return pol; | |
5606e387 MG |
213 | } |
214 | ||
f15ca78e | 215 | return &default_policy; |
5606e387 MG |
216 | } |
217 | ||
37012946 DR |
218 | static const struct mempolicy_operations { |
219 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 220 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
221 | } mpol_ops[MPOL_MAX]; |
222 | ||
f5b087b5 DR |
223 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
224 | { | |
6d556294 | 225 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
226 | } |
227 | ||
228 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
229 | const nodemask_t *rel) | |
230 | { | |
231 | nodemask_t tmp; | |
232 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
233 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
234 | } |
235 | ||
be897d48 | 236 | static int mpol_new_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
237 | { |
238 | if (nodes_empty(*nodes)) | |
239 | return -EINVAL; | |
269fbe72 | 240 | pol->nodes = *nodes; |
37012946 DR |
241 | return 0; |
242 | } | |
243 | ||
244 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
245 | { | |
7858d7bc FT |
246 | if (nodes_empty(*nodes)) |
247 | return -EINVAL; | |
269fbe72 BW |
248 | |
249 | nodes_clear(pol->nodes); | |
250 | node_set(first_node(*nodes), pol->nodes); | |
37012946 DR |
251 | return 0; |
252 | } | |
253 | ||
58568d2a MX |
254 | /* |
255 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
256 | * any, for the new policy. mpol_new() has already validated the nodes | |
7858d7bc | 257 | * parameter with respect to the policy mode and flags. |
58568d2a MX |
258 | * |
259 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
c1e8d7c6 | 260 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 261 | */ |
4bfc4495 KH |
262 | static int mpol_set_nodemask(struct mempolicy *pol, |
263 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 264 | { |
58568d2a MX |
265 | int ret; |
266 | ||
7858d7bc FT |
267 | /* |
268 | * Default (pol==NULL) resp. local memory policies are not a | |
269 | * subject of any remapping. They also do not need any special | |
270 | * constructor. | |
271 | */ | |
272 | if (!pol || pol->mode == MPOL_LOCAL) | |
58568d2a | 273 | return 0; |
7858d7bc | 274 | |
01f13bd6 | 275 | /* Check N_MEMORY */ |
4bfc4495 | 276 | nodes_and(nsc->mask1, |
01f13bd6 | 277 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
278 | |
279 | VM_BUG_ON(!nodes); | |
4bfc4495 | 280 | |
7858d7bc FT |
281 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
282 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); | |
283 | else | |
284 | nodes_and(nsc->mask2, *nodes, nsc->mask1); | |
58568d2a | 285 | |
7858d7bc FT |
286 | if (mpol_store_user_nodemask(pol)) |
287 | pol->w.user_nodemask = *nodes; | |
4bfc4495 | 288 | else |
7858d7bc FT |
289 | pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed; |
290 | ||
291 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
58568d2a MX |
292 | return ret; |
293 | } | |
294 | ||
295 | /* | |
296 | * This function just creates a new policy, does some check and simple | |
297 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
298 | */ | |
028fec41 DR |
299 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
300 | nodemask_t *nodes) | |
1da177e4 LT |
301 | { |
302 | struct mempolicy *policy; | |
303 | ||
3e1f0645 DR |
304 | if (mode == MPOL_DEFAULT) { |
305 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 306 | return ERR_PTR(-EINVAL); |
d3a71033 | 307 | return NULL; |
37012946 | 308 | } |
3e1f0645 DR |
309 | VM_BUG_ON(!nodes); |
310 | ||
311 | /* | |
312 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
313 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
314 | * All other modes require a valid pointer to a non-empty nodemask. | |
315 | */ | |
316 | if (mode == MPOL_PREFERRED) { | |
317 | if (nodes_empty(*nodes)) { | |
318 | if (((flags & MPOL_F_STATIC_NODES) || | |
319 | (flags & MPOL_F_RELATIVE_NODES))) | |
320 | return ERR_PTR(-EINVAL); | |
7858d7bc FT |
321 | |
322 | mode = MPOL_LOCAL; | |
3e1f0645 | 323 | } |
479e2802 | 324 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
325 | if (!nodes_empty(*nodes) || |
326 | (flags & MPOL_F_STATIC_NODES) || | |
327 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 | 328 | return ERR_PTR(-EINVAL); |
3e1f0645 DR |
329 | } else if (nodes_empty(*nodes)) |
330 | return ERR_PTR(-EINVAL); | |
c36f6e6d | 331 | |
1da177e4 LT |
332 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
333 | if (!policy) | |
334 | return ERR_PTR(-ENOMEM); | |
335 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 336 | policy->mode = mode; |
3e1f0645 | 337 | policy->flags = flags; |
c6018b4b | 338 | policy->home_node = NUMA_NO_NODE; |
37012946 | 339 | |
1da177e4 | 340 | return policy; |
37012946 DR |
341 | } |
342 | ||
52cd3b07 | 343 | /* Slow path of a mpol destructor. */ |
c36f6e6d | 344 | void __mpol_put(struct mempolicy *pol) |
52cd3b07 | 345 | { |
c36f6e6d | 346 | if (!atomic_dec_and_test(&pol->refcnt)) |
52cd3b07 | 347 | return; |
c36f6e6d | 348 | kmem_cache_free(policy_cache, pol); |
52cd3b07 LS |
349 | } |
350 | ||
213980c0 | 351 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
352 | { |
353 | } | |
354 | ||
213980c0 | 355 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
356 | { |
357 | nodemask_t tmp; | |
358 | ||
359 | if (pol->flags & MPOL_F_STATIC_NODES) | |
360 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
361 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
362 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
363 | else { | |
269fbe72 | 364 | nodes_remap(tmp, pol->nodes, pol->w.cpuset_mems_allowed, |
213980c0 | 365 | *nodes); |
29b190fa | 366 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 367 | } |
f5b087b5 | 368 | |
708c1bbc MX |
369 | if (nodes_empty(tmp)) |
370 | tmp = *nodes; | |
371 | ||
269fbe72 | 372 | pol->nodes = tmp; |
37012946 DR |
373 | } |
374 | ||
375 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 376 | const nodemask_t *nodes) |
37012946 | 377 | { |
7858d7bc | 378 | pol->w.cpuset_mems_allowed = *nodes; |
1da177e4 LT |
379 | } |
380 | ||
708c1bbc MX |
381 | /* |
382 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
383 | * | |
c1e8d7c6 | 384 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
385 | * policies are protected by task->mems_allowed_seq to prevent a premature |
386 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 387 | */ |
213980c0 | 388 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 389 | { |
018160ad | 390 | if (!pol || pol->mode == MPOL_LOCAL) |
1d0d2680 | 391 | return; |
7858d7bc | 392 | if (!mpol_store_user_nodemask(pol) && |
1d0d2680 DR |
393 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
394 | return; | |
708c1bbc | 395 | |
213980c0 | 396 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
397 | } |
398 | ||
399 | /* | |
400 | * Wrapper for mpol_rebind_policy() that just requires task | |
401 | * pointer, and updates task mempolicy. | |
58568d2a MX |
402 | * |
403 | * Called with task's alloc_lock held. | |
1d0d2680 | 404 | */ |
213980c0 | 405 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 406 | { |
213980c0 | 407 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
408 | } |
409 | ||
410 | /* | |
411 | * Rebind each vma in mm to new nodemask. | |
412 | * | |
c1e8d7c6 | 413 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 | 414 | */ |
1d0d2680 DR |
415 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) |
416 | { | |
417 | struct vm_area_struct *vma; | |
66850be5 | 418 | VMA_ITERATOR(vmi, mm, 0); |
1d0d2680 | 419 | |
d8ed45c5 | 420 | mmap_write_lock(mm); |
6c21e066 JH |
421 | for_each_vma(vmi, vma) { |
422 | vma_start_write(vma); | |
213980c0 | 423 | mpol_rebind_policy(vma->vm_policy, new); |
6c21e066 | 424 | } |
d8ed45c5 | 425 | mmap_write_unlock(mm); |
1d0d2680 DR |
426 | } |
427 | ||
37012946 DR |
428 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
429 | [MPOL_DEFAULT] = { | |
430 | .rebind = mpol_rebind_default, | |
431 | }, | |
432 | [MPOL_INTERLEAVE] = { | |
be897d48 | 433 | .create = mpol_new_nodemask, |
37012946 DR |
434 | .rebind = mpol_rebind_nodemask, |
435 | }, | |
436 | [MPOL_PREFERRED] = { | |
437 | .create = mpol_new_preferred, | |
438 | .rebind = mpol_rebind_preferred, | |
439 | }, | |
440 | [MPOL_BIND] = { | |
be897d48 | 441 | .create = mpol_new_nodemask, |
37012946 DR |
442 | .rebind = mpol_rebind_nodemask, |
443 | }, | |
7858d7bc FT |
444 | [MPOL_LOCAL] = { |
445 | .rebind = mpol_rebind_default, | |
446 | }, | |
b27abacc | 447 | [MPOL_PREFERRED_MANY] = { |
be897d48 | 448 | .create = mpol_new_nodemask, |
b27abacc DH |
449 | .rebind = mpol_rebind_preferred, |
450 | }, | |
fa3bea4e GP |
451 | [MPOL_WEIGHTED_INTERLEAVE] = { |
452 | .create = mpol_new_nodemask, | |
453 | .rebind = mpol_rebind_nodemask, | |
454 | }, | |
37012946 DR |
455 | }; |
456 | ||
1cb5d11a | 457 | static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
fc301289 | 458 | unsigned long flags); |
72e315f7 HD |
459 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol, |
460 | pgoff_t ilx, int *nid); | |
1a75a6c8 | 461 | |
1cb5d11a HD |
462 | static bool strictly_unmovable(unsigned long flags) |
463 | { | |
464 | /* | |
465 | * STRICT without MOVE flags lets do_mbind() fail immediately with -EIO | |
466 | * if any misplaced page is found. | |
467 | */ | |
468 | return (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) == | |
469 | MPOL_MF_STRICT; | |
470 | } | |
471 | ||
88c91dc5 HD |
472 | struct migration_mpol { /* for alloc_migration_target_by_mpol() */ |
473 | struct mempolicy *pol; | |
474 | pgoff_t ilx; | |
475 | }; | |
1a75a6c8 | 476 | |
6f4576e3 NH |
477 | struct queue_pages { |
478 | struct list_head *pagelist; | |
479 | unsigned long flags; | |
480 | nodemask_t *nmask; | |
f18da660 LX |
481 | unsigned long start; |
482 | unsigned long end; | |
483 | struct vm_area_struct *first; | |
1cb5d11a HD |
484 | struct folio *large; /* note last large folio encountered */ |
485 | long nr_failed; /* could not be isolated at this time */ | |
6f4576e3 NH |
486 | }; |
487 | ||
88aaa2a1 | 488 | /* |
d451b89d | 489 | * Check if the folio's nid is in qp->nmask. |
88aaa2a1 NH |
490 | * |
491 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
492 | * in the invert of qp->nmask. | |
493 | */ | |
d451b89d | 494 | static inline bool queue_folio_required(struct folio *folio, |
88aaa2a1 NH |
495 | struct queue_pages *qp) |
496 | { | |
d451b89d | 497 | int nid = folio_nid(folio); |
88aaa2a1 NH |
498 | unsigned long flags = qp->flags; |
499 | ||
500 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
501 | } | |
502 | ||
1cb5d11a | 503 | static void queue_folios_pmd(pmd_t *pmd, struct mm_walk *walk) |
c8633798 | 504 | { |
de1f5055 | 505 | struct folio *folio; |
c8633798 | 506 | struct queue_pages *qp = walk->private; |
c8633798 NH |
507 | |
508 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
1cb5d11a HD |
509 | qp->nr_failed++; |
510 | return; | |
c8633798 | 511 | } |
de1f5055 VMO |
512 | folio = pfn_folio(pmd_pfn(*pmd)); |
513 | if (is_huge_zero_page(&folio->page)) { | |
e5947d23 | 514 | walk->action = ACTION_CONTINUE; |
1cb5d11a | 515 | return; |
c8633798 | 516 | } |
d451b89d | 517 | if (!queue_folio_required(folio, qp)) |
1cb5d11a HD |
518 | return; |
519 | if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) || | |
520 | !vma_migratable(walk->vma) || | |
521 | !migrate_folio_add(folio, qp->pagelist, qp->flags)) | |
522 | qp->nr_failed++; | |
c8633798 NH |
523 | } |
524 | ||
98094945 | 525 | /* |
1cb5d11a HD |
526 | * Scan through folios, checking if they satisfy the required conditions, |
527 | * moving them from LRU to local pagelist for migration if they do (or not). | |
d8835445 | 528 | * |
1cb5d11a HD |
529 | * queue_folios_pte_range() has two possible return values: |
530 | * 0 - continue walking to scan for more, even if an existing folio on the | |
531 | * wrong node could not be isolated and queued for migration. | |
532 | * -EIO - only MPOL_MF_STRICT was specified, without MPOL_MF_MOVE or ..._ALL, | |
533 | * and an existing folio was on a node that does not follow the policy. | |
98094945 | 534 | */ |
3dae02bb | 535 | static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr, |
6f4576e3 | 536 | unsigned long end, struct mm_walk *walk) |
1da177e4 | 537 | { |
6f4576e3 | 538 | struct vm_area_struct *vma = walk->vma; |
3dae02bb | 539 | struct folio *folio; |
6f4576e3 NH |
540 | struct queue_pages *qp = walk->private; |
541 | unsigned long flags = qp->flags; | |
3f088420 | 542 | pte_t *pte, *mapped_pte; |
c33c7948 | 543 | pte_t ptent; |
705e87c0 | 544 | spinlock_t *ptl; |
941150a3 | 545 | |
c8633798 | 546 | ptl = pmd_trans_huge_lock(pmd, vma); |
1cb5d11a HD |
547 | if (ptl) { |
548 | queue_folios_pmd(pmd, walk); | |
549 | spin_unlock(ptl); | |
550 | goto out; | |
551 | } | |
91612e0d | 552 | |
3f088420 | 553 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
7780d040 HD |
554 | if (!pte) { |
555 | walk->action = ACTION_AGAIN; | |
556 | return 0; | |
557 | } | |
6f4576e3 | 558 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
c33c7948 | 559 | ptent = ptep_get(pte); |
1cb5d11a | 560 | if (pte_none(ptent)) |
1da177e4 | 561 | continue; |
1cb5d11a HD |
562 | if (!pte_present(ptent)) { |
563 | if (is_migration_entry(pte_to_swp_entry(ptent))) | |
564 | qp->nr_failed++; | |
1da177e4 | 565 | continue; |
1cb5d11a | 566 | } |
c33c7948 | 567 | folio = vm_normal_folio(vma, addr, ptent); |
3dae02bb | 568 | if (!folio || folio_is_zone_device(folio)) |
1da177e4 | 569 | continue; |
053837fc | 570 | /* |
3dae02bb VMO |
571 | * vm_normal_folio() filters out zero pages, but there might |
572 | * still be reserved folios to skip, perhaps in a VDSO. | |
053837fc | 573 | */ |
3dae02bb | 574 | if (folio_test_reserved(folio)) |
f4598c8b | 575 | continue; |
d451b89d | 576 | if (!queue_folio_required(folio, qp)) |
38e35860 | 577 | continue; |
1cb5d11a | 578 | if (folio_test_large(folio)) { |
a53190a4 | 579 | /* |
1cb5d11a HD |
580 | * A large folio can only be isolated from LRU once, |
581 | * but may be mapped by many PTEs (and Copy-On-Write may | |
582 | * intersperse PTEs of other, order 0, folios). This is | |
583 | * a common case, so don't mistake it for failure (but | |
584 | * there can be other cases of multi-mapped pages which | |
585 | * this quick check does not help to filter out - and a | |
586 | * search of the pagelist might grow to be prohibitive). | |
587 | * | |
588 | * migrate_pages(&pagelist) returns nr_failed folios, so | |
589 | * check "large" now so that queue_pages_range() returns | |
590 | * a comparable nr_failed folios. This does imply that | |
591 | * if folio could not be isolated for some racy reason | |
592 | * at its first PTE, later PTEs will not give it another | |
593 | * chance of isolation; but keeps the accounting simple. | |
a53190a4 | 594 | */ |
1cb5d11a HD |
595 | if (folio == qp->large) |
596 | continue; | |
597 | qp->large = folio; | |
598 | } | |
599 | if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) || | |
600 | !vma_migratable(vma) || | |
601 | !migrate_folio_add(folio, qp->pagelist, flags)) { | |
602 | qp->nr_failed++; | |
603 | if (strictly_unmovable(flags)) | |
604 | break; | |
605 | } | |
6f4576e3 | 606 | } |
3f088420 | 607 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 608 | cond_resched(); |
1cb5d11a HD |
609 | out: |
610 | if (qp->nr_failed && strictly_unmovable(flags)) | |
611 | return -EIO; | |
612 | return 0; | |
91612e0d HD |
613 | } |
614 | ||
0a2c1e81 | 615 | static int queue_folios_hugetlb(pte_t *pte, unsigned long hmask, |
6f4576e3 NH |
616 | unsigned long addr, unsigned long end, |
617 | struct mm_walk *walk) | |
e2d8cf40 NH |
618 | { |
619 | #ifdef CONFIG_HUGETLB_PAGE | |
6f4576e3 | 620 | struct queue_pages *qp = walk->private; |
1cb5d11a | 621 | unsigned long flags = qp->flags; |
0a2c1e81 | 622 | struct folio *folio; |
cb900f41 | 623 | spinlock_t *ptl; |
d4c54919 | 624 | pte_t entry; |
e2d8cf40 | 625 | |
6f4576e3 NH |
626 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
627 | entry = huge_ptep_get(pte); | |
1cb5d11a HD |
628 | if (!pte_present(entry)) { |
629 | if (unlikely(is_hugetlb_entry_migration(entry))) | |
630 | qp->nr_failed++; | |
d4c54919 | 631 | goto unlock; |
1cb5d11a | 632 | } |
0a2c1e81 | 633 | folio = pfn_folio(pte_pfn(entry)); |
d451b89d | 634 | if (!queue_folio_required(folio, qp)) |
e2d8cf40 | 635 | goto unlock; |
1cb5d11a HD |
636 | if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) || |
637 | !vma_migratable(walk->vma)) { | |
638 | qp->nr_failed++; | |
dcf17635 LX |
639 | goto unlock; |
640 | } | |
0a2c1e81 | 641 | /* |
1cb5d11a HD |
642 | * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio. |
643 | * Choosing not to migrate a shared folio is not counted as a failure. | |
0a2c1e81 VMO |
644 | * |
645 | * To check if the folio is shared, ideally we want to make sure | |
646 | * every page is mapped to the same process. Doing that is very | |
1cb5d11a | 647 | * expensive, so check the estimated sharers of the folio instead. |
0a2c1e81 | 648 | */ |
1cb5d11a HD |
649 | if ((flags & MPOL_MF_MOVE_ALL) || |
650 | (folio_estimated_sharers(folio) == 1 && !hugetlb_pmd_shared(pte))) | |
651 | if (!isolate_hugetlb(folio, qp->pagelist)) | |
652 | qp->nr_failed++; | |
e2d8cf40 | 653 | unlock: |
cb900f41 | 654 | spin_unlock(ptl); |
1cb5d11a HD |
655 | if (qp->nr_failed && strictly_unmovable(flags)) |
656 | return -EIO; | |
e2d8cf40 | 657 | #endif |
1cb5d11a | 658 | return 0; |
1da177e4 LT |
659 | } |
660 | ||
5877231f | 661 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 662 | /* |
4b10e7d5 MG |
663 | * This is used to mark a range of virtual addresses to be inaccessible. |
664 | * These are later cleared by a NUMA hinting fault. Depending on these | |
665 | * faults, pages may be migrated for better NUMA placement. | |
666 | * | |
667 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
668 | * an architecture makes a different choice, it will need further | |
669 | * changes to the core. | |
b24f53a0 | 670 | */ |
4b10e7d5 MG |
671 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
672 | unsigned long addr, unsigned long end) | |
b24f53a0 | 673 | { |
4a18419f | 674 | struct mmu_gather tlb; |
a79390f5 | 675 | long nr_updated; |
b24f53a0 | 676 | |
4a18419f NA |
677 | tlb_gather_mmu(&tlb, vma->vm_mm); |
678 | ||
1ef488ed | 679 | nr_updated = change_protection(&tlb, vma, addr, end, MM_CP_PROT_NUMA); |
d1751118 | 680 | if (nr_updated > 0) |
03c5a6e1 | 681 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); |
b24f53a0 | 682 | |
4a18419f NA |
683 | tlb_finish_mmu(&tlb); |
684 | ||
4b10e7d5 | 685 | return nr_updated; |
b24f53a0 | 686 | } |
5877231f | 687 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 688 | |
6f4576e3 NH |
689 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
690 | struct mm_walk *walk) | |
691 | { | |
66850be5 | 692 | struct vm_area_struct *next, *vma = walk->vma; |
6f4576e3 | 693 | struct queue_pages *qp = walk->private; |
6f4576e3 NH |
694 | unsigned long flags = qp->flags; |
695 | ||
a18b3ac2 | 696 | /* range check first */ |
ce33135c | 697 | VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma); |
f18da660 LX |
698 | |
699 | if (!qp->first) { | |
700 | qp->first = vma; | |
701 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
702 | (qp->start < vma->vm_start)) | |
703 | /* hole at head side of range */ | |
a18b3ac2 LX |
704 | return -EFAULT; |
705 | } | |
66850be5 | 706 | next = find_vma(vma->vm_mm, vma->vm_end); |
f18da660 LX |
707 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
708 | ((vma->vm_end < qp->end) && | |
66850be5 | 709 | (!next || vma->vm_end < next->vm_start))) |
f18da660 LX |
710 | /* hole at middle or tail of range */ |
711 | return -EFAULT; | |
a18b3ac2 | 712 | |
a7f40cfe YS |
713 | /* |
714 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
715 | * regardless of vma_migratable | |
716 | */ | |
717 | if (!vma_migratable(vma) && | |
718 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
719 | return 1; |
720 | ||
1cb5d11a HD |
721 | /* |
722 | * Check page nodes, and queue pages to move, in the current vma. | |
723 | * But if no moving, and no strict checking, the scan can be skipped. | |
724 | */ | |
725 | if (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6f4576e3 NH |
726 | return 0; |
727 | return 1; | |
728 | } | |
729 | ||
7b86ac33 | 730 | static const struct mm_walk_ops queue_pages_walk_ops = { |
0a2c1e81 | 731 | .hugetlb_entry = queue_folios_hugetlb, |
3dae02bb | 732 | .pmd_entry = queue_folios_pte_range, |
7b86ac33 | 733 | .test_walk = queue_pages_test_walk, |
49b06385 SB |
734 | .walk_lock = PGWALK_RDLOCK, |
735 | }; | |
736 | ||
737 | static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = { | |
738 | .hugetlb_entry = queue_folios_hugetlb, | |
739 | .pmd_entry = queue_folios_pte_range, | |
740 | .test_walk = queue_pages_test_walk, | |
741 | .walk_lock = PGWALK_WRLOCK, | |
7b86ac33 CH |
742 | }; |
743 | ||
dc9aa5b9 | 744 | /* |
98094945 NH |
745 | * Walk through page tables and collect pages to be migrated. |
746 | * | |
1cb5d11a HD |
747 | * If pages found in a given range are not on the required set of @nodes, |
748 | * and migration is allowed, they are isolated and queued to @pagelist. | |
d8835445 | 749 | * |
1cb5d11a HD |
750 | * queue_pages_range() may return: |
751 | * 0 - all pages already on the right node, or successfully queued for moving | |
752 | * (or neither strict checking nor moving requested: only range checking). | |
753 | * >0 - this number of misplaced folios could not be queued for moving | |
754 | * (a hugetlbfs page or a transparent huge page being counted as 1). | |
755 | * -EIO - a misplaced page found, when MPOL_MF_STRICT specified without MOVEs. | |
756 | * -EFAULT - a hole in the memory range, when MPOL_MF_DISCONTIG_OK unspecified. | |
dc9aa5b9 | 757 | */ |
1cb5d11a | 758 | static long |
98094945 | 759 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 | 760 | nodemask_t *nodes, unsigned long flags, |
1cb5d11a | 761 | struct list_head *pagelist) |
1da177e4 | 762 | { |
f18da660 | 763 | int err; |
6f4576e3 NH |
764 | struct queue_pages qp = { |
765 | .pagelist = pagelist, | |
766 | .flags = flags, | |
767 | .nmask = nodes, | |
f18da660 LX |
768 | .start = start, |
769 | .end = end, | |
770 | .first = NULL, | |
6f4576e3 | 771 | }; |
1cb5d11a | 772 | const struct mm_walk_ops *ops = (flags & MPOL_MF_WRLOCK) ? |
49b06385 | 773 | &queue_pages_lock_vma_walk_ops : &queue_pages_walk_ops; |
6f4576e3 | 774 | |
49b06385 | 775 | err = walk_page_range(mm, start, end, ops, &qp); |
f18da660 LX |
776 | |
777 | if (!qp.first) | |
778 | /* whole range in hole */ | |
779 | err = -EFAULT; | |
780 | ||
1cb5d11a | 781 | return err ? : qp.nr_failed; |
1da177e4 LT |
782 | } |
783 | ||
869833f2 KM |
784 | /* |
785 | * Apply policy to a single VMA | |
c1e8d7c6 | 786 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
787 | */ |
788 | static int vma_replace_policy(struct vm_area_struct *vma, | |
c36f6e6d | 789 | struct mempolicy *pol) |
8d34694c | 790 | { |
869833f2 KM |
791 | int err; |
792 | struct mempolicy *old; | |
793 | struct mempolicy *new; | |
8d34694c | 794 | |
6c21e066 JH |
795 | vma_assert_write_locked(vma); |
796 | ||
869833f2 KM |
797 | new = mpol_dup(pol); |
798 | if (IS_ERR(new)) | |
799 | return PTR_ERR(new); | |
800 | ||
801 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 802 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
803 | if (err) |
804 | goto err_out; | |
8d34694c | 805 | } |
869833f2 KM |
806 | |
807 | old = vma->vm_policy; | |
c1e8d7c6 | 808 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
809 | mpol_put(old); |
810 | ||
811 | return 0; | |
812 | err_out: | |
813 | mpol_put(new); | |
8d34694c KM |
814 | return err; |
815 | } | |
816 | ||
f4e9e0e6 LH |
817 | /* Split or merge the VMA (if required) and apply the new policy */ |
818 | static int mbind_range(struct vma_iterator *vmi, struct vm_area_struct *vma, | |
819 | struct vm_area_struct **prev, unsigned long start, | |
820 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 | 821 | { |
f4e9e0e6 | 822 | unsigned long vmstart, vmend; |
9d8cebd4 | 823 | |
f4e9e0e6 LH |
824 | vmend = min(end, vma->vm_end); |
825 | if (start > vma->vm_start) { | |
826 | *prev = vma; | |
827 | vmstart = start; | |
828 | } else { | |
829 | vmstart = vma->vm_start; | |
830 | } | |
831 | ||
c36f6e6d | 832 | if (mpol_equal(vma->vm_policy, new_pol)) { |
00ca0f2e | 833 | *prev = vma; |
7329e3eb | 834 | return 0; |
00ca0f2e | 835 | } |
7329e3eb | 836 | |
94d7d923 LS |
837 | vma = vma_modify_policy(vmi, *prev, vma, vmstart, vmend, new_pol); |
838 | if (IS_ERR(vma)) | |
839 | return PTR_ERR(vma); | |
f4e9e0e6 LH |
840 | |
841 | *prev = vma; | |
842 | return vma_replace_policy(vma, new_pol); | |
1da177e4 LT |
843 | } |
844 | ||
1da177e4 | 845 | /* Set the process memory policy */ |
028fec41 DR |
846 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
847 | nodemask_t *nodes) | |
1da177e4 | 848 | { |
58568d2a | 849 | struct mempolicy *new, *old; |
4bfc4495 | 850 | NODEMASK_SCRATCH(scratch); |
58568d2a | 851 | int ret; |
1da177e4 | 852 | |
4bfc4495 KH |
853 | if (!scratch) |
854 | return -ENOMEM; | |
f4e53d91 | 855 | |
4bfc4495 KH |
856 | new = mpol_new(mode, flags, nodes); |
857 | if (IS_ERR(new)) { | |
858 | ret = PTR_ERR(new); | |
859 | goto out; | |
860 | } | |
2c7c3a7d | 861 | |
12c1dc8e | 862 | task_lock(current); |
4bfc4495 | 863 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 864 | if (ret) { |
12c1dc8e | 865 | task_unlock(current); |
58568d2a | 866 | mpol_put(new); |
4bfc4495 | 867 | goto out; |
58568d2a | 868 | } |
12c1dc8e | 869 | |
58568d2a | 870 | old = current->mempolicy; |
1da177e4 | 871 | current->mempolicy = new; |
fa3bea4e GP |
872 | if (new && (new->mode == MPOL_INTERLEAVE || |
873 | new->mode == MPOL_WEIGHTED_INTERLEAVE)) { | |
45816682 | 874 | current->il_prev = MAX_NUMNODES-1; |
fa3bea4e GP |
875 | current->il_weight = 0; |
876 | } | |
58568d2a | 877 | task_unlock(current); |
58568d2a | 878 | mpol_put(old); |
4bfc4495 KH |
879 | ret = 0; |
880 | out: | |
881 | NODEMASK_SCRATCH_FREE(scratch); | |
882 | return ret; | |
1da177e4 LT |
883 | } |
884 | ||
bea904d5 LS |
885 | /* |
886 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
887 | * |
888 | * Called with task's alloc_lock held | |
bea904d5 | 889 | */ |
c36f6e6d | 890 | static void get_policy_nodemask(struct mempolicy *pol, nodemask_t *nodes) |
1da177e4 | 891 | { |
dfcd3c0d | 892 | nodes_clear(*nodes); |
c36f6e6d | 893 | if (pol == &default_policy) |
bea904d5 LS |
894 | return; |
895 | ||
c36f6e6d | 896 | switch (pol->mode) { |
19770b32 | 897 | case MPOL_BIND: |
1da177e4 | 898 | case MPOL_INTERLEAVE: |
269fbe72 | 899 | case MPOL_PREFERRED: |
b27abacc | 900 | case MPOL_PREFERRED_MANY: |
fa3bea4e | 901 | case MPOL_WEIGHTED_INTERLEAVE: |
c36f6e6d | 902 | *nodes = pol->nodes; |
1da177e4 | 903 | break; |
7858d7bc FT |
904 | case MPOL_LOCAL: |
905 | /* return empty node mask for local allocation */ | |
906 | break; | |
1da177e4 LT |
907 | default: |
908 | BUG(); | |
909 | } | |
910 | } | |
911 | ||
3b9aadf7 | 912 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 913 | { |
ba841078 | 914 | struct page *p = NULL; |
f728b9c4 | 915 | int ret; |
1da177e4 | 916 | |
f728b9c4 JH |
917 | ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p); |
918 | if (ret > 0) { | |
919 | ret = page_to_nid(p); | |
1da177e4 LT |
920 | put_page(p); |
921 | } | |
f728b9c4 | 922 | return ret; |
1da177e4 LT |
923 | } |
924 | ||
1da177e4 | 925 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
926 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
927 | unsigned long addr, unsigned long flags) | |
1da177e4 | 928 | { |
8bccd85f | 929 | int err; |
1da177e4 LT |
930 | struct mm_struct *mm = current->mm; |
931 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 932 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 933 | |
754af6f5 LS |
934 | if (flags & |
935 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 936 | return -EINVAL; |
754af6f5 LS |
937 | |
938 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
939 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
940 | return -EINVAL; | |
941 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 942 | task_lock(current); |
754af6f5 | 943 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 944 | task_unlock(current); |
754af6f5 LS |
945 | return 0; |
946 | } | |
947 | ||
1da177e4 | 948 | if (flags & MPOL_F_ADDR) { |
ddc1a5cb | 949 | pgoff_t ilx; /* ignored here */ |
bea904d5 LS |
950 | /* |
951 | * Do NOT fall back to task policy if the | |
952 | * vma/shared policy at addr is NULL. We | |
953 | * want to return MPOL_DEFAULT in this case. | |
954 | */ | |
d8ed45c5 | 955 | mmap_read_lock(mm); |
33e3575c | 956 | vma = vma_lookup(mm, addr); |
1da177e4 | 957 | if (!vma) { |
d8ed45c5 | 958 | mmap_read_unlock(mm); |
1da177e4 LT |
959 | return -EFAULT; |
960 | } | |
ddc1a5cb | 961 | pol = __get_vma_policy(vma, addr, &ilx); |
1da177e4 LT |
962 | } else if (addr) |
963 | return -EINVAL; | |
964 | ||
965 | if (!pol) | |
bea904d5 | 966 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
967 | |
968 | if (flags & MPOL_F_NODE) { | |
969 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 | 970 | /* |
f728b9c4 JH |
971 | * Take a refcount on the mpol, because we are about to |
972 | * drop the mmap_lock, after which only "pol" remains | |
973 | * valid, "vma" is stale. | |
3b9aadf7 AA |
974 | */ |
975 | pol_refcount = pol; | |
976 | vma = NULL; | |
977 | mpol_get(pol); | |
f728b9c4 | 978 | mmap_read_unlock(mm); |
3b9aadf7 | 979 | err = lookup_node(mm, addr); |
1da177e4 LT |
980 | if (err < 0) |
981 | goto out; | |
8bccd85f | 982 | *policy = err; |
1da177e4 | 983 | } else if (pol == current->mempolicy && |
45c4745a | 984 | pol->mode == MPOL_INTERLEAVE) { |
269fbe72 | 985 | *policy = next_node_in(current->il_prev, pol->nodes); |
fa3bea4e GP |
986 | } else if (pol == current->mempolicy && |
987 | pol->mode == MPOL_WEIGHTED_INTERLEAVE) { | |
988 | if (current->il_weight) | |
989 | *policy = current->il_prev; | |
990 | else | |
991 | *policy = next_node_in(current->il_prev, | |
992 | pol->nodes); | |
1da177e4 LT |
993 | } else { |
994 | err = -EINVAL; | |
995 | goto out; | |
996 | } | |
bea904d5 LS |
997 | } else { |
998 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
999 | pol->mode; | |
d79df630 DR |
1000 | /* |
1001 | * Internal mempolicy flags must be masked off before exposing | |
1002 | * the policy to userspace. | |
1003 | */ | |
1004 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1005 | } |
1da177e4 | 1006 | |
1da177e4 | 1007 | err = 0; |
58568d2a | 1008 | if (nmask) { |
c6b6ef8b LS |
1009 | if (mpol_store_user_nodemask(pol)) { |
1010 | *nmask = pol->w.user_nodemask; | |
1011 | } else { | |
1012 | task_lock(current); | |
1013 | get_policy_nodemask(pol, nmask); | |
1014 | task_unlock(current); | |
1015 | } | |
58568d2a | 1016 | } |
1da177e4 LT |
1017 | |
1018 | out: | |
52cd3b07 | 1019 | mpol_cond_put(pol); |
1da177e4 | 1020 | if (vma) |
d8ed45c5 | 1021 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1022 | if (pol_refcount) |
1023 | mpol_put(pol_refcount); | |
1da177e4 LT |
1024 | return err; |
1025 | } | |
1026 | ||
b20a3503 | 1027 | #ifdef CONFIG_MIGRATION |
1cb5d11a | 1028 | static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
fc301289 | 1029 | unsigned long flags) |
6ce3c4c0 CL |
1030 | { |
1031 | /* | |
1cb5d11a HD |
1032 | * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio. |
1033 | * Choosing not to migrate a shared folio is not counted as a failure. | |
4a64981d VMO |
1034 | * |
1035 | * To check if the folio is shared, ideally we want to make sure | |
1036 | * every page is mapped to the same process. Doing that is very | |
1cb5d11a | 1037 | * expensive, so check the estimated sharers of the folio instead. |
6ce3c4c0 | 1038 | */ |
4a64981d | 1039 | if ((flags & MPOL_MF_MOVE_ALL) || folio_estimated_sharers(folio) == 1) { |
be2d5756 | 1040 | if (folio_isolate_lru(folio)) { |
4a64981d VMO |
1041 | list_add_tail(&folio->lru, foliolist); |
1042 | node_stat_mod_folio(folio, | |
1043 | NR_ISOLATED_ANON + folio_is_file_lru(folio), | |
1044 | folio_nr_pages(folio)); | |
1cb5d11a | 1045 | } else { |
a53190a4 | 1046 | /* |
4a64981d VMO |
1047 | * Non-movable folio may reach here. And, there may be |
1048 | * temporary off LRU folios or non-LRU movable folios. | |
1049 | * Treat them as unmovable folios since they can't be | |
1cb5d11a | 1050 | * isolated, so they can't be moved at the moment. |
a53190a4 | 1051 | */ |
1cb5d11a | 1052 | return false; |
62695a84 NP |
1053 | } |
1054 | } | |
1cb5d11a | 1055 | return true; |
7e2ab150 | 1056 | } |
6ce3c4c0 | 1057 | |
7e2ab150 CL |
1058 | /* |
1059 | * Migrate pages from one node to a target node. | |
1060 | * Returns error or the number of pages not migrated. | |
1061 | */ | |
1cb5d11a HD |
1062 | static long migrate_to_node(struct mm_struct *mm, int source, int dest, |
1063 | int flags) | |
7e2ab150 CL |
1064 | { |
1065 | nodemask_t nmask; | |
66850be5 | 1066 | struct vm_area_struct *vma; |
7e2ab150 | 1067 | LIST_HEAD(pagelist); |
1cb5d11a HD |
1068 | long nr_failed; |
1069 | long err = 0; | |
a0976311 JK |
1070 | struct migration_target_control mtc = { |
1071 | .nid = dest, | |
1072 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1073 | }; | |
7e2ab150 CL |
1074 | |
1075 | nodes_clear(nmask); | |
1076 | node_set(source, nmask); | |
6ce3c4c0 | 1077 | |
1cb5d11a | 1078 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); |
72e315f7 HD |
1079 | |
1080 | mmap_read_lock(mm); | |
1cb5d11a HD |
1081 | vma = find_vma(mm, 0); |
1082 | ||
08270807 | 1083 | /* |
1cb5d11a | 1084 | * This does not migrate the range, but isolates all pages that |
08270807 | 1085 | * need migration. Between passing in the full user address |
1cb5d11a HD |
1086 | * space range and MPOL_MF_DISCONTIG_OK, this call cannot fail, |
1087 | * but passes back the count of pages which could not be isolated. | |
08270807 | 1088 | */ |
1cb5d11a HD |
1089 | nr_failed = queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask, |
1090 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
72e315f7 | 1091 | mmap_read_unlock(mm); |
7e2ab150 | 1092 | |
cf608ac1 | 1093 | if (!list_empty(&pagelist)) { |
a0976311 | 1094 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
1cb5d11a | 1095 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
cf608ac1 | 1096 | if (err) |
e2d8cf40 | 1097 | putback_movable_pages(&pagelist); |
cf608ac1 | 1098 | } |
95a402c3 | 1099 | |
1cb5d11a HD |
1100 | if (err >= 0) |
1101 | err += nr_failed; | |
7e2ab150 | 1102 | return err; |
6ce3c4c0 CL |
1103 | } |
1104 | ||
39743889 | 1105 | /* |
7e2ab150 CL |
1106 | * Move pages between the two nodesets so as to preserve the physical |
1107 | * layout as much as possible. | |
39743889 CL |
1108 | * |
1109 | * Returns the number of page that could not be moved. | |
1110 | */ | |
0ce72d4f AM |
1111 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1112 | const nodemask_t *to, int flags) | |
39743889 | 1113 | { |
1cb5d11a HD |
1114 | long nr_failed = 0; |
1115 | long err = 0; | |
7e2ab150 | 1116 | nodemask_t tmp; |
39743889 | 1117 | |
361a2a22 | 1118 | lru_cache_disable(); |
0aedadf9 | 1119 | |
da0aa138 KM |
1120 | /* |
1121 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1122 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1123 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1124 | * The pair of nodemasks 'to' and 'from' define the map. | |
1125 | * | |
1126 | * If no pair of bits is found that way, fallback to picking some | |
1127 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1128 | * 'source' and 'dest' bits are the same, this represents a node | |
1129 | * that will be migrating to itself, so no pages need move. | |
1130 | * | |
1131 | * If no bits are left in 'tmp', or if all remaining bits left | |
1132 | * in 'tmp' correspond to the same bit in 'to', return false | |
1133 | * (nothing left to migrate). | |
1134 | * | |
1135 | * This lets us pick a pair of nodes to migrate between, such that | |
1136 | * if possible the dest node is not already occupied by some other | |
1137 | * source node, minimizing the risk of overloading the memory on a | |
1138 | * node that would happen if we migrated incoming memory to a node | |
1139 | * before migrating outgoing memory source that same node. | |
1140 | * | |
1141 | * A single scan of tmp is sufficient. As we go, we remember the | |
1142 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1143 | * that not only moved, but what's better, moved to an empty slot | |
1144 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1145 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1146 | * most recent <s, d> pair that moved. If we get all the way through |
1147 | * the scan of tmp without finding any node that moved, much less | |
1148 | * moved to an empty node, then there is nothing left worth migrating. | |
1149 | */ | |
d4984711 | 1150 | |
0ce72d4f | 1151 | tmp = *from; |
7e2ab150 | 1152 | while (!nodes_empty(tmp)) { |
68d68ff6 | 1153 | int s, d; |
b76ac7e7 | 1154 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1155 | int dest = 0; |
1156 | ||
1157 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1158 | |
1159 | /* | |
1160 | * do_migrate_pages() tries to maintain the relative | |
1161 | * node relationship of the pages established between | |
1162 | * threads and memory areas. | |
1163 | * | |
1164 | * However if the number of source nodes is not equal to | |
1165 | * the number of destination nodes we can not preserve | |
1166 | * this node relative relationship. In that case, skip | |
1167 | * copying memory from a node that is in the destination | |
1168 | * mask. | |
1169 | * | |
1170 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1171 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1172 | */ | |
1173 | ||
0ce72d4f AM |
1174 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1175 | (node_isset(s, *to))) | |
4a5b18cc LW |
1176 | continue; |
1177 | ||
0ce72d4f | 1178 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1179 | if (s == d) |
1180 | continue; | |
1181 | ||
1182 | source = s; /* Node moved. Memorize */ | |
1183 | dest = d; | |
1184 | ||
1185 | /* dest not in remaining from nodes? */ | |
1186 | if (!node_isset(dest, tmp)) | |
1187 | break; | |
1188 | } | |
b76ac7e7 | 1189 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1190 | break; |
1191 | ||
1192 | node_clear(source, tmp); | |
1193 | err = migrate_to_node(mm, source, dest, flags); | |
1194 | if (err > 0) | |
1cb5d11a | 1195 | nr_failed += err; |
7e2ab150 CL |
1196 | if (err < 0) |
1197 | break; | |
39743889 | 1198 | } |
d479960e | 1199 | |
361a2a22 | 1200 | lru_cache_enable(); |
7e2ab150 CL |
1201 | if (err < 0) |
1202 | return err; | |
1cb5d11a | 1203 | return (nr_failed < INT_MAX) ? nr_failed : INT_MAX; |
b20a3503 CL |
1204 | } |
1205 | ||
3ad33b24 | 1206 | /* |
72e315f7 | 1207 | * Allocate a new folio for page migration, according to NUMA mempolicy. |
3ad33b24 | 1208 | */ |
72e315f7 HD |
1209 | static struct folio *alloc_migration_target_by_mpol(struct folio *src, |
1210 | unsigned long private) | |
95a402c3 | 1211 | { |
88c91dc5 HD |
1212 | struct migration_mpol *mmpol = (struct migration_mpol *)private; |
1213 | struct mempolicy *pol = mmpol->pol; | |
1214 | pgoff_t ilx = mmpol->ilx; | |
72e315f7 HD |
1215 | struct page *page; |
1216 | unsigned int order; | |
1217 | int nid = numa_node_id(); | |
1218 | gfp_t gfp; | |
95a402c3 | 1219 | |
72e315f7 HD |
1220 | order = folio_order(src); |
1221 | ilx += src->index >> order; | |
11c731e8 | 1222 | |
d0ce0e47 | 1223 | if (folio_test_hugetlb(src)) { |
72e315f7 HD |
1224 | nodemask_t *nodemask; |
1225 | struct hstate *h; | |
1226 | ||
1227 | h = folio_hstate(src); | |
1228 | gfp = htlb_alloc_mask(h); | |
1229 | nodemask = policy_nodemask(gfp, pol, ilx, &nid); | |
1230 | return alloc_hugetlb_folio_nodemask(h, nid, nodemask, gfp); | |
d0ce0e47 | 1231 | } |
ec4858e0 MWO |
1232 | |
1233 | if (folio_test_large(src)) | |
1234 | gfp = GFP_TRANSHUGE; | |
72e315f7 HD |
1235 | else |
1236 | gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL | __GFP_COMP; | |
ec4858e0 | 1237 | |
72e315f7 HD |
1238 | page = alloc_pages_mpol(gfp, order, pol, ilx, nid); |
1239 | return page_rmappable_folio(page); | |
95a402c3 | 1240 | } |
b20a3503 CL |
1241 | #else |
1242 | ||
1cb5d11a | 1243 | static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist, |
b20a3503 CL |
1244 | unsigned long flags) |
1245 | { | |
1cb5d11a | 1246 | return false; |
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 | |
72e315f7 HD |
1255 | static struct folio *alloc_migration_target_by_mpol(struct folio *src, |
1256 | unsigned long private) | |
95a402c3 CL |
1257 | { |
1258 | return NULL; | |
1259 | } | |
b20a3503 CL |
1260 | #endif |
1261 | ||
dbcb0f19 | 1262 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1263 | unsigned short mode, unsigned short mode_flags, |
1264 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1265 | { |
6ce3c4c0 | 1266 | struct mm_struct *mm = current->mm; |
f4e9e0e6 LH |
1267 | struct vm_area_struct *vma, *prev; |
1268 | struct vma_iterator vmi; | |
88c91dc5 | 1269 | struct migration_mpol mmpol; |
6ce3c4c0 CL |
1270 | struct mempolicy *new; |
1271 | unsigned long end; | |
1cb5d11a HD |
1272 | long err; |
1273 | long nr_failed; | |
6ce3c4c0 CL |
1274 | LIST_HEAD(pagelist); |
1275 | ||
b24f53a0 | 1276 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1277 | return -EINVAL; |
74c00241 | 1278 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1279 | return -EPERM; |
1280 | ||
1281 | if (start & ~PAGE_MASK) | |
1282 | return -EINVAL; | |
1283 | ||
1284 | if (mode == MPOL_DEFAULT) | |
1285 | flags &= ~MPOL_MF_STRICT; | |
1286 | ||
aaa31e05 | 1287 | len = PAGE_ALIGN(len); |
6ce3c4c0 CL |
1288 | end = start + len; |
1289 | ||
1290 | if (end < start) | |
1291 | return -EINVAL; | |
1292 | if (end == start) | |
1293 | return 0; | |
1294 | ||
028fec41 | 1295 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1296 | if (IS_ERR(new)) |
1297 | return PTR_ERR(new); | |
1298 | ||
1299 | /* | |
1300 | * If we are using the default policy then operation | |
1301 | * on discontinuous address spaces is okay after all | |
1302 | */ | |
1303 | if (!new) | |
1304 | flags |= MPOL_MF_DISCONTIG_OK; | |
1305 | ||
1cb5d11a | 1306 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1307 | lru_cache_disable(); |
4bfc4495 KH |
1308 | { |
1309 | NODEMASK_SCRATCH(scratch); | |
1310 | if (scratch) { | |
d8ed45c5 | 1311 | mmap_write_lock(mm); |
4bfc4495 | 1312 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1313 | if (err) |
d8ed45c5 | 1314 | mmap_write_unlock(mm); |
4bfc4495 KH |
1315 | } else |
1316 | err = -ENOMEM; | |
1317 | NODEMASK_SCRATCH_FREE(scratch); | |
1318 | } | |
b05ca738 KM |
1319 | if (err) |
1320 | goto mpol_out; | |
1321 | ||
6c21e066 | 1322 | /* |
1cb5d11a HD |
1323 | * Lock the VMAs before scanning for pages to migrate, |
1324 | * to ensure we don't miss a concurrently inserted page. | |
6c21e066 | 1325 | */ |
1cb5d11a HD |
1326 | nr_failed = queue_pages_range(mm, start, end, nmask, |
1327 | flags | MPOL_MF_INVERT | MPOL_MF_WRLOCK, &pagelist); | |
d8835445 | 1328 | |
1cb5d11a HD |
1329 | if (nr_failed < 0) { |
1330 | err = nr_failed; | |
72e315f7 | 1331 | nr_failed = 0; |
1cb5d11a HD |
1332 | } else { |
1333 | vma_iter_init(&vmi, mm, start); | |
1334 | prev = vma_prev(&vmi); | |
1335 | for_each_vma_range(vmi, vma, end) { | |
1336 | err = mbind_range(&vmi, vma, &prev, start, end, new); | |
1337 | if (err) | |
1338 | break; | |
1339 | } | |
f4e9e0e6 | 1340 | } |
7e2ab150 | 1341 | |
72e315f7 HD |
1342 | if (!err && !list_empty(&pagelist)) { |
1343 | /* Convert MPOL_DEFAULT's NULL to task or default policy */ | |
1344 | if (!new) { | |
1345 | new = get_task_policy(current); | |
1346 | mpol_get(new); | |
cf608ac1 | 1347 | } |
88c91dc5 HD |
1348 | mmpol.pol = new; |
1349 | mmpol.ilx = 0; | |
6ce3c4c0 | 1350 | |
88c91dc5 HD |
1351 | /* |
1352 | * In the interleaved case, attempt to allocate on exactly the | |
1353 | * targeted nodes, for the first VMA to be migrated; for later | |
1354 | * VMAs, the nodes will still be interleaved from the targeted | |
1355 | * nodemask, but one by one may be selected differently. | |
1356 | */ | |
fa3bea4e GP |
1357 | if (new->mode == MPOL_INTERLEAVE || |
1358 | new->mode == MPOL_WEIGHTED_INTERLEAVE) { | |
f1cce6f7 | 1359 | struct folio *folio; |
88c91dc5 HD |
1360 | unsigned int order; |
1361 | unsigned long addr = -EFAULT; | |
1362 | ||
f1cce6f7 MWO |
1363 | list_for_each_entry(folio, &pagelist, lru) { |
1364 | if (!folio_test_ksm(folio)) | |
88c91dc5 HD |
1365 | break; |
1366 | } | |
f1cce6f7 | 1367 | if (!list_entry_is_head(folio, &pagelist, lru)) { |
88c91dc5 HD |
1368 | vma_iter_init(&vmi, mm, start); |
1369 | for_each_vma_range(vmi, vma, end) { | |
f1cce6f7 MWO |
1370 | addr = page_address_in_vma( |
1371 | folio_page(folio, 0), vma); | |
88c91dc5 HD |
1372 | if (addr != -EFAULT) |
1373 | break; | |
1374 | } | |
1375 | } | |
1376 | if (addr != -EFAULT) { | |
f1cce6f7 | 1377 | order = folio_order(folio); |
88c91dc5 HD |
1378 | /* We already know the pol, but not the ilx */ |
1379 | mpol_cond_put(get_vma_policy(vma, addr, order, | |
1380 | &mmpol.ilx)); | |
1381 | /* Set base from which to increment by index */ | |
f1cce6f7 | 1382 | mmpol.ilx -= folio->index >> order; |
88c91dc5 HD |
1383 | } |
1384 | } | |
a85dfc30 YS |
1385 | } |
1386 | ||
d8ed45c5 | 1387 | mmap_write_unlock(mm); |
88c91dc5 HD |
1388 | |
1389 | if (!err && !list_empty(&pagelist)) { | |
72e315f7 HD |
1390 | nr_failed |= migrate_pages(&pagelist, |
1391 | alloc_migration_target_by_mpol, NULL, | |
88c91dc5 | 1392 | (unsigned long)&mmpol, MIGRATE_SYNC, |
72e315f7 | 1393 | MR_MEMPOLICY_MBIND, NULL); |
a85dfc30 YS |
1394 | } |
1395 | ||
72e315f7 HD |
1396 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
1397 | err = -EIO; | |
1cb5d11a HD |
1398 | if (!list_empty(&pagelist)) |
1399 | putback_movable_pages(&pagelist); | |
d8835445 | 1400 | mpol_out: |
f0be3d32 | 1401 | mpol_put(new); |
d479960e | 1402 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1403 | lru_cache_enable(); |
6ce3c4c0 CL |
1404 | return err; |
1405 | } | |
1406 | ||
8bccd85f CL |
1407 | /* |
1408 | * User space interface with variable sized bitmaps for nodelists. | |
1409 | */ | |
e130242d AB |
1410 | static int get_bitmap(unsigned long *mask, const unsigned long __user *nmask, |
1411 | unsigned long maxnode) | |
1412 | { | |
1413 | unsigned long nlongs = BITS_TO_LONGS(maxnode); | |
1414 | int ret; | |
1415 | ||
1416 | if (in_compat_syscall()) | |
1417 | ret = compat_get_bitmap(mask, | |
1418 | (const compat_ulong_t __user *)nmask, | |
1419 | maxnode); | |
1420 | else | |
1421 | ret = copy_from_user(mask, nmask, | |
1422 | nlongs * sizeof(unsigned long)); | |
1423 | ||
1424 | if (ret) | |
1425 | return -EFAULT; | |
1426 | ||
1427 | if (maxnode % BITS_PER_LONG) | |
1428 | mask[nlongs - 1] &= (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1429 | ||
1430 | return 0; | |
1431 | } | |
8bccd85f CL |
1432 | |
1433 | /* Copy a node mask from user space. */ | |
39743889 | 1434 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1435 | unsigned long maxnode) |
1436 | { | |
8bccd85f CL |
1437 | --maxnode; |
1438 | nodes_clear(*nodes); | |
1439 | if (maxnode == 0 || !nmask) | |
1440 | return 0; | |
a9c930ba | 1441 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1442 | return -EINVAL; |
8bccd85f | 1443 | |
56521e7a YX |
1444 | /* |
1445 | * When the user specified more nodes than supported just check | |
e130242d AB |
1446 | * if the non supported part is all zero, one word at a time, |
1447 | * starting at the end. | |
56521e7a | 1448 | */ |
e130242d AB |
1449 | while (maxnode > MAX_NUMNODES) { |
1450 | unsigned long bits = min_t(unsigned long, maxnode, BITS_PER_LONG); | |
1451 | unsigned long t; | |
8bccd85f | 1452 | |
000eca5d | 1453 | if (get_bitmap(&t, &nmask[(maxnode - 1) / BITS_PER_LONG], bits)) |
56521e7a | 1454 | return -EFAULT; |
e130242d AB |
1455 | |
1456 | if (maxnode - bits >= MAX_NUMNODES) { | |
1457 | maxnode -= bits; | |
1458 | } else { | |
1459 | maxnode = MAX_NUMNODES; | |
1460 | t &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1461 | } | |
1462 | if (t) | |
56521e7a YX |
1463 | return -EINVAL; |
1464 | } | |
1465 | ||
e130242d | 1466 | return get_bitmap(nodes_addr(*nodes), nmask, maxnode); |
8bccd85f CL |
1467 | } |
1468 | ||
1469 | /* Copy a kernel node mask to user space */ | |
1470 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1471 | nodemask_t *nodes) | |
1472 | { | |
1473 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1474 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
e130242d AB |
1475 | bool compat = in_compat_syscall(); |
1476 | ||
1477 | if (compat) | |
1478 | nbytes = BITS_TO_COMPAT_LONGS(nr_node_ids) * sizeof(compat_long_t); | |
8bccd85f CL |
1479 | |
1480 | if (copy > nbytes) { | |
1481 | if (copy > PAGE_SIZE) | |
1482 | return -EINVAL; | |
1483 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1484 | return -EFAULT; | |
1485 | copy = nbytes; | |
e130242d | 1486 | maxnode = nr_node_ids; |
8bccd85f | 1487 | } |
e130242d AB |
1488 | |
1489 | if (compat) | |
1490 | return compat_put_bitmap((compat_ulong_t __user *)mask, | |
1491 | nodes_addr(*nodes), maxnode); | |
1492 | ||
8bccd85f CL |
1493 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; |
1494 | } | |
1495 | ||
95837924 FT |
1496 | /* Basic parameter sanity check used by both mbind() and set_mempolicy() */ |
1497 | static inline int sanitize_mpol_flags(int *mode, unsigned short *flags) | |
1498 | { | |
1499 | *flags = *mode & MPOL_MODE_FLAGS; | |
1500 | *mode &= ~MPOL_MODE_FLAGS; | |
b27abacc | 1501 | |
a38a59fd | 1502 | if ((unsigned int)(*mode) >= MPOL_MAX) |
95837924 FT |
1503 | return -EINVAL; |
1504 | if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES)) | |
1505 | return -EINVAL; | |
6d2aec9e ED |
1506 | if (*flags & MPOL_F_NUMA_BALANCING) { |
1507 | if (*mode != MPOL_BIND) | |
1508 | return -EINVAL; | |
1509 | *flags |= (MPOL_F_MOF | MPOL_F_MORON); | |
1510 | } | |
95837924 FT |
1511 | return 0; |
1512 | } | |
1513 | ||
e7dc9ad6 DB |
1514 | static long kernel_mbind(unsigned long start, unsigned long len, |
1515 | unsigned long mode, const unsigned long __user *nmask, | |
1516 | unsigned long maxnode, unsigned int flags) | |
8bccd85f | 1517 | { |
95837924 | 1518 | unsigned short mode_flags; |
8bccd85f | 1519 | nodemask_t nodes; |
95837924 | 1520 | int lmode = mode; |
8bccd85f CL |
1521 | int err; |
1522 | ||
057d3389 | 1523 | start = untagged_addr(start); |
95837924 FT |
1524 | err = sanitize_mpol_flags(&lmode, &mode_flags); |
1525 | if (err) | |
1526 | return err; | |
1527 | ||
8bccd85f CL |
1528 | err = get_nodes(&nodes, nmask, maxnode); |
1529 | if (err) | |
1530 | return err; | |
95837924 FT |
1531 | |
1532 | return do_mbind(start, len, lmode, mode_flags, &nodes, flags); | |
8bccd85f CL |
1533 | } |
1534 | ||
c6018b4b AK |
1535 | SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, len, |
1536 | unsigned long, home_node, unsigned long, flags) | |
1537 | { | |
1538 | struct mm_struct *mm = current->mm; | |
f4e9e0e6 | 1539 | struct vm_area_struct *vma, *prev; |
e976936c | 1540 | struct mempolicy *new, *old; |
c6018b4b AK |
1541 | unsigned long end; |
1542 | int err = -ENOENT; | |
66850be5 | 1543 | VMA_ITERATOR(vmi, mm, start); |
c6018b4b AK |
1544 | |
1545 | start = untagged_addr(start); | |
1546 | if (start & ~PAGE_MASK) | |
1547 | return -EINVAL; | |
1548 | /* | |
1549 | * flags is used for future extension if any. | |
1550 | */ | |
1551 | if (flags != 0) | |
1552 | return -EINVAL; | |
1553 | ||
1554 | /* | |
1555 | * Check home_node is online to avoid accessing uninitialized | |
1556 | * NODE_DATA. | |
1557 | */ | |
1558 | if (home_node >= MAX_NUMNODES || !node_online(home_node)) | |
1559 | return -EINVAL; | |
1560 | ||
aaa31e05 | 1561 | len = PAGE_ALIGN(len); |
c6018b4b AK |
1562 | end = start + len; |
1563 | ||
1564 | if (end < start) | |
1565 | return -EINVAL; | |
1566 | if (end == start) | |
1567 | return 0; | |
1568 | mmap_write_lock(mm); | |
f4e9e0e6 | 1569 | prev = vma_prev(&vmi); |
66850be5 | 1570 | for_each_vma_range(vmi, vma, end) { |
c6018b4b AK |
1571 | /* |
1572 | * If any vma in the range got policy other than MPOL_BIND | |
1573 | * or MPOL_PREFERRED_MANY we return error. We don't reset | |
1574 | * the home node for vmas we already updated before. | |
1575 | */ | |
e976936c | 1576 | old = vma_policy(vma); |
51f62537 LH |
1577 | if (!old) { |
1578 | prev = vma; | |
e976936c | 1579 | continue; |
51f62537 | 1580 | } |
e976936c | 1581 | if (old->mode != MPOL_BIND && old->mode != MPOL_PREFERRED_MANY) { |
c6018b4b AK |
1582 | err = -EOPNOTSUPP; |
1583 | break; | |
1584 | } | |
e976936c MH |
1585 | new = mpol_dup(old); |
1586 | if (IS_ERR(new)) { | |
1587 | err = PTR_ERR(new); | |
1588 | break; | |
1589 | } | |
c6018b4b | 1590 | |
6c21e066 | 1591 | vma_start_write(vma); |
c6018b4b | 1592 | new->home_node = home_node; |
f4e9e0e6 | 1593 | err = mbind_range(&vmi, vma, &prev, start, end, new); |
c6018b4b AK |
1594 | mpol_put(new); |
1595 | if (err) | |
1596 | break; | |
1597 | } | |
1598 | mmap_write_unlock(mm); | |
1599 | return err; | |
1600 | } | |
1601 | ||
e7dc9ad6 DB |
1602 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1603 | unsigned long, mode, const unsigned long __user *, nmask, | |
1604 | unsigned long, maxnode, unsigned int, flags) | |
1605 | { | |
1606 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1607 | } | |
1608 | ||
8bccd85f | 1609 | /* Set the process memory policy */ |
af03c4ac DB |
1610 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1611 | unsigned long maxnode) | |
8bccd85f | 1612 | { |
95837924 | 1613 | unsigned short mode_flags; |
8bccd85f | 1614 | nodemask_t nodes; |
95837924 FT |
1615 | int lmode = mode; |
1616 | int err; | |
1617 | ||
1618 | err = sanitize_mpol_flags(&lmode, &mode_flags); | |
1619 | if (err) | |
1620 | return err; | |
8bccd85f | 1621 | |
8bccd85f CL |
1622 | err = get_nodes(&nodes, nmask, maxnode); |
1623 | if (err) | |
1624 | return err; | |
95837924 FT |
1625 | |
1626 | return do_set_mempolicy(lmode, mode_flags, &nodes); | |
8bccd85f CL |
1627 | } |
1628 | ||
af03c4ac DB |
1629 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1630 | unsigned long, maxnode) | |
1631 | { | |
1632 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1633 | } | |
1634 | ||
b6e9b0ba DB |
1635 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1636 | const unsigned long __user *old_nodes, | |
1637 | const unsigned long __user *new_nodes) | |
39743889 | 1638 | { |
596d7cfa | 1639 | struct mm_struct *mm = NULL; |
39743889 | 1640 | struct task_struct *task; |
39743889 CL |
1641 | nodemask_t task_nodes; |
1642 | int err; | |
596d7cfa KM |
1643 | nodemask_t *old; |
1644 | nodemask_t *new; | |
1645 | NODEMASK_SCRATCH(scratch); | |
1646 | ||
1647 | if (!scratch) | |
1648 | return -ENOMEM; | |
39743889 | 1649 | |
596d7cfa KM |
1650 | old = &scratch->mask1; |
1651 | new = &scratch->mask2; | |
1652 | ||
1653 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1654 | if (err) |
596d7cfa | 1655 | goto out; |
39743889 | 1656 | |
596d7cfa | 1657 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1658 | if (err) |
596d7cfa | 1659 | goto out; |
39743889 CL |
1660 | |
1661 | /* Find the mm_struct */ | |
55cfaa3c | 1662 | rcu_read_lock(); |
228ebcbe | 1663 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1664 | if (!task) { |
55cfaa3c | 1665 | rcu_read_unlock(); |
596d7cfa KM |
1666 | err = -ESRCH; |
1667 | goto out; | |
39743889 | 1668 | } |
3268c63e | 1669 | get_task_struct(task); |
39743889 | 1670 | |
596d7cfa | 1671 | err = -EINVAL; |
39743889 CL |
1672 | |
1673 | /* | |
31367466 OE |
1674 | * Check if this process has the right to modify the specified process. |
1675 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1676 | */ |
31367466 | 1677 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1678 | rcu_read_unlock(); |
39743889 | 1679 | err = -EPERM; |
3268c63e | 1680 | goto out_put; |
39743889 | 1681 | } |
c69e8d9c | 1682 | rcu_read_unlock(); |
39743889 CL |
1683 | |
1684 | task_nodes = cpuset_mems_allowed(task); | |
1685 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1686 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1687 | err = -EPERM; |
3268c63e | 1688 | goto out_put; |
39743889 CL |
1689 | } |
1690 | ||
0486a38b YX |
1691 | task_nodes = cpuset_mems_allowed(current); |
1692 | nodes_and(*new, *new, task_nodes); | |
1693 | if (nodes_empty(*new)) | |
1694 | goto out_put; | |
1695 | ||
86c3a764 DQ |
1696 | err = security_task_movememory(task); |
1697 | if (err) | |
3268c63e | 1698 | goto out_put; |
86c3a764 | 1699 | |
3268c63e CL |
1700 | mm = get_task_mm(task); |
1701 | put_task_struct(task); | |
f2a9ef88 SL |
1702 | |
1703 | if (!mm) { | |
3268c63e | 1704 | err = -EINVAL; |
f2a9ef88 SL |
1705 | goto out; |
1706 | } | |
1707 | ||
1708 | err = do_migrate_pages(mm, old, new, | |
1709 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1710 | |
1711 | mmput(mm); | |
1712 | out: | |
596d7cfa KM |
1713 | NODEMASK_SCRATCH_FREE(scratch); |
1714 | ||
39743889 | 1715 | return err; |
3268c63e CL |
1716 | |
1717 | out_put: | |
1718 | put_task_struct(task); | |
1719 | goto out; | |
39743889 CL |
1720 | } |
1721 | ||
b6e9b0ba DB |
1722 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1723 | const unsigned long __user *, old_nodes, | |
1724 | const unsigned long __user *, new_nodes) | |
1725 | { | |
1726 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1727 | } | |
1728 | ||
8bccd85f | 1729 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1730 | static int kernel_get_mempolicy(int __user *policy, |
1731 | unsigned long __user *nmask, | |
1732 | unsigned long maxnode, | |
1733 | unsigned long addr, | |
1734 | unsigned long flags) | |
8bccd85f | 1735 | { |
dbcb0f19 | 1736 | int err; |
3f649ab7 | 1737 | int pval; |
8bccd85f CL |
1738 | nodemask_t nodes; |
1739 | ||
050c17f2 | 1740 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1741 | return -EINVAL; |
1742 | ||
4605f057 WH |
1743 | addr = untagged_addr(addr); |
1744 | ||
8bccd85f CL |
1745 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1746 | ||
1747 | if (err) | |
1748 | return err; | |
1749 | ||
1750 | if (policy && put_user(pval, policy)) | |
1751 | return -EFAULT; | |
1752 | ||
1753 | if (nmask) | |
1754 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1755 | ||
1756 | return err; | |
1757 | } | |
1758 | ||
af03c4ac DB |
1759 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1760 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1761 | unsigned long, addr, unsigned long, flags) | |
1762 | { | |
1763 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1764 | } | |
1765 | ||
20ca87f2 LX |
1766 | bool vma_migratable(struct vm_area_struct *vma) |
1767 | { | |
1768 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1769 | return false; | |
1770 | ||
1771 | /* | |
1772 | * DAX device mappings require predictable access latency, so avoid | |
1773 | * incurring periodic faults. | |
1774 | */ | |
1775 | if (vma_is_dax(vma)) | |
1776 | return false; | |
1777 | ||
1778 | if (is_vm_hugetlb_page(vma) && | |
1779 | !hugepage_migration_supported(hstate_vma(vma))) | |
1780 | return false; | |
1781 | ||
1782 | /* | |
1783 | * Migration allocates pages in the highest zone. If we cannot | |
1784 | * do so then migration (at least from node to node) is not | |
1785 | * possible. | |
1786 | */ | |
1787 | if (vma->vm_file && | |
1788 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1789 | < policy_zone) | |
1790 | return false; | |
1791 | return true; | |
1792 | } | |
1793 | ||
74d2c3a0 | 1794 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
ddc1a5cb | 1795 | unsigned long addr, pgoff_t *ilx) |
1da177e4 | 1796 | { |
ddc1a5cb HD |
1797 | *ilx = 0; |
1798 | return (vma->vm_ops && vma->vm_ops->get_policy) ? | |
1799 | vma->vm_ops->get_policy(vma, addr, ilx) : vma->vm_policy; | |
74d2c3a0 ON |
1800 | } |
1801 | ||
1802 | /* | |
ddc1a5cb | 1803 | * get_vma_policy(@vma, @addr, @order, @ilx) |
74d2c3a0 ON |
1804 | * @vma: virtual memory area whose policy is sought |
1805 | * @addr: address in @vma for shared policy lookup | |
ddc1a5cb | 1806 | * @order: 0, or appropriate huge_page_order for interleaving |
fa3bea4e GP |
1807 | * @ilx: interleave index (output), for use only when MPOL_INTERLEAVE or |
1808 | * MPOL_WEIGHTED_INTERLEAVE | |
74d2c3a0 ON |
1809 | * |
1810 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1811 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1812 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1813 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1814 | * freeing by another task. It is the caller's responsibility to free the | |
1815 | * extra reference for shared policies. | |
1816 | */ | |
ddc1a5cb HD |
1817 | struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
1818 | unsigned long addr, int order, pgoff_t *ilx) | |
74d2c3a0 | 1819 | { |
ddc1a5cb | 1820 | struct mempolicy *pol; |
74d2c3a0 | 1821 | |
ddc1a5cb | 1822 | pol = __get_vma_policy(vma, addr, ilx); |
8d90274b | 1823 | if (!pol) |
dd6eecb9 | 1824 | pol = get_task_policy(current); |
fa3bea4e GP |
1825 | if (pol->mode == MPOL_INTERLEAVE || |
1826 | pol->mode == MPOL_WEIGHTED_INTERLEAVE) { | |
ddc1a5cb HD |
1827 | *ilx += vma->vm_pgoff >> order; |
1828 | *ilx += (addr - vma->vm_start) >> (PAGE_SHIFT + order); | |
1829 | } | |
1da177e4 LT |
1830 | return pol; |
1831 | } | |
1832 | ||
6b6482bb | 1833 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1834 | { |
6b6482bb | 1835 | struct mempolicy *pol; |
fc314724 | 1836 | |
6b6482bb ON |
1837 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1838 | bool ret = false; | |
ddc1a5cb | 1839 | pgoff_t ilx; /* ignored here */ |
fc314724 | 1840 | |
ddc1a5cb | 1841 | pol = vma->vm_ops->get_policy(vma, vma->vm_start, &ilx); |
6b6482bb ON |
1842 | if (pol && (pol->flags & MPOL_F_MOF)) |
1843 | ret = true; | |
1844 | mpol_cond_put(pol); | |
8d90274b | 1845 | |
6b6482bb | 1846 | return ret; |
fc314724 MG |
1847 | } |
1848 | ||
6b6482bb | 1849 | pol = vma->vm_policy; |
8d90274b | 1850 | if (!pol) |
6b6482bb | 1851 | pol = get_task_policy(current); |
8d90274b | 1852 | |
fc314724 MG |
1853 | return pol->flags & MPOL_F_MOF; |
1854 | } | |
1855 | ||
d2226ebd | 1856 | bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
d3eb1570 LJ |
1857 | { |
1858 | enum zone_type dynamic_policy_zone = policy_zone; | |
1859 | ||
1860 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1861 | ||
1862 | /* | |
269fbe72 | 1863 | * if policy->nodes has movable memory only, |
d3eb1570 LJ |
1864 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. |
1865 | * | |
269fbe72 | 1866 | * policy->nodes is intersect with node_states[N_MEMORY]. |
f0953a1b | 1867 | * so if the following test fails, it implies |
269fbe72 | 1868 | * policy->nodes has movable memory only. |
d3eb1570 | 1869 | */ |
269fbe72 | 1870 | if (!nodes_intersects(policy->nodes, node_states[N_HIGH_MEMORY])) |
d3eb1570 LJ |
1871 | dynamic_policy_zone = ZONE_MOVABLE; |
1872 | ||
1873 | return zone >= dynamic_policy_zone; | |
1874 | } | |
1875 | ||
fa3bea4e GP |
1876 | static unsigned int weighted_interleave_nodes(struct mempolicy *policy) |
1877 | { | |
274519ed GP |
1878 | unsigned int node; |
1879 | unsigned int cpuset_mems_cookie; | |
fa3bea4e | 1880 | |
274519ed GP |
1881 | retry: |
1882 | /* to prevent miscount use tsk->mems_allowed_seq to detect rebind */ | |
1883 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
1884 | node = current->il_prev; | |
fa3bea4e GP |
1885 | if (!current->il_weight || !node_isset(node, policy->nodes)) { |
1886 | node = next_node_in(node, policy->nodes); | |
274519ed GP |
1887 | if (read_mems_allowed_retry(cpuset_mems_cookie)) |
1888 | goto retry; | |
fa3bea4e GP |
1889 | if (node == MAX_NUMNODES) |
1890 | return node; | |
1891 | current->il_prev = node; | |
1892 | current->il_weight = get_il_weight(node); | |
1893 | } | |
1894 | current->il_weight--; | |
1895 | return node; | |
1896 | } | |
1897 | ||
1da177e4 | 1898 | /* Do dynamic interleaving for a process */ |
c36f6e6d | 1899 | static unsigned int interleave_nodes(struct mempolicy *policy) |
1da177e4 | 1900 | { |
c36f6e6d | 1901 | unsigned int nid; |
274519ed GP |
1902 | unsigned int cpuset_mems_cookie; |
1903 | ||
1904 | /* to prevent miscount, use tsk->mems_allowed_seq to detect rebind */ | |
1905 | do { | |
1906 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
1907 | nid = next_node_in(current->il_prev, policy->nodes); | |
1908 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); | |
1da177e4 | 1909 | |
c36f6e6d HD |
1910 | if (nid < MAX_NUMNODES) |
1911 | current->il_prev = nid; | |
1912 | return nid; | |
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 | ||
fa3bea4e GP |
1938 | case MPOL_WEIGHTED_INTERLEAVE: |
1939 | return weighted_interleave_nodes(policy); | |
1940 | ||
b27abacc DH |
1941 | case MPOL_BIND: |
1942 | case MPOL_PREFERRED_MANY: | |
1943 | { | |
c33d6c06 MG |
1944 | struct zoneref *z; |
1945 | ||
dc85da15 CL |
1946 | /* |
1947 | * Follow bind policy behavior and start allocation at the | |
1948 | * first node. | |
1949 | */ | |
19770b32 | 1950 | struct zonelist *zonelist; |
19770b32 | 1951 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1952 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 | 1953 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
269fbe72 | 1954 | &policy->nodes); |
c1093b74 | 1955 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1956 | } |
7858d7bc FT |
1957 | case MPOL_LOCAL: |
1958 | return node; | |
dc85da15 | 1959 | |
dc85da15 | 1960 | default: |
bea904d5 | 1961 | BUG(); |
dc85da15 CL |
1962 | } |
1963 | } | |
1964 | ||
9685e6e3 GP |
1965 | static unsigned int read_once_policy_nodemask(struct mempolicy *pol, |
1966 | nodemask_t *mask) | |
1967 | { | |
1968 | /* | |
1969 | * barrier stabilizes the nodemask locally so that it can be iterated | |
1970 | * over safely without concern for changes. Allocators validate node | |
1971 | * selection does not violate mems_allowed, so this is safe. | |
1972 | */ | |
1973 | barrier(); | |
1974 | memcpy(mask, &pol->nodes, sizeof(nodemask_t)); | |
1975 | barrier(); | |
1976 | return nodes_weight(*mask); | |
1977 | } | |
1978 | ||
fa3bea4e GP |
1979 | static unsigned int weighted_interleave_nid(struct mempolicy *pol, pgoff_t ilx) |
1980 | { | |
1981 | nodemask_t nodemask; | |
1982 | unsigned int target, nr_nodes; | |
1983 | u8 *table; | |
1984 | unsigned int weight_total = 0; | |
1985 | u8 weight; | |
1986 | int nid; | |
1987 | ||
1988 | nr_nodes = read_once_policy_nodemask(pol, &nodemask); | |
1989 | if (!nr_nodes) | |
1990 | return numa_node_id(); | |
1991 | ||
1992 | rcu_read_lock(); | |
1993 | table = rcu_dereference(iw_table); | |
1994 | /* calculate the total weight */ | |
1995 | for_each_node_mask(nid, nodemask) { | |
1996 | /* detect system default usage */ | |
1997 | weight = table ? table[nid] : 1; | |
1998 | weight = weight ? weight : 1; | |
1999 | weight_total += weight; | |
2000 | } | |
2001 | ||
2002 | /* Calculate the node offset based on totals */ | |
2003 | target = ilx % weight_total; | |
2004 | nid = first_node(nodemask); | |
2005 | while (target) { | |
2006 | /* detect system default usage */ | |
2007 | weight = table ? table[nid] : 1; | |
2008 | weight = weight ? weight : 1; | |
2009 | if (target < weight) | |
2010 | break; | |
2011 | target -= weight; | |
2012 | nid = next_node_in(nid, nodemask); | |
2013 | } | |
2014 | rcu_read_unlock(); | |
2015 | return nid; | |
2016 | } | |
2017 | ||
fee83b3a | 2018 | /* |
ddc1a5cb HD |
2019 | * Do static interleaving for interleave index @ilx. Returns the ilx'th |
2020 | * node in pol->nodes (starting from ilx=0), wrapping around if ilx | |
2021 | * exceeds the number of present nodes. | |
fee83b3a | 2022 | */ |
ddc1a5cb | 2023 | static unsigned int interleave_nid(struct mempolicy *pol, pgoff_t ilx) |
1da177e4 | 2024 | { |
9685e6e3 | 2025 | nodemask_t nodemask; |
276aeee1 | 2026 | unsigned int target, nnodes; |
fee83b3a AM |
2027 | int i; |
2028 | int nid; | |
1da177e4 | 2029 | |
9685e6e3 | 2030 | nnodes = read_once_policy_nodemask(pol, &nodemask); |
f5b087b5 DR |
2031 | if (!nnodes) |
2032 | return numa_node_id(); | |
ddc1a5cb | 2033 | target = ilx % nnodes; |
276aeee1 | 2034 | nid = first_node(nodemask); |
fee83b3a | 2035 | for (i = 0; i < target; i++) |
276aeee1 | 2036 | nid = next_node(nid, nodemask); |
1da177e4 LT |
2037 | return nid; |
2038 | } | |
2039 | ||
ddc1a5cb HD |
2040 | /* |
2041 | * Return a nodemask representing a mempolicy for filtering nodes for | |
2042 | * page allocation, together with preferred node id (or the input node id). | |
2043 | */ | |
2044 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol, | |
2045 | pgoff_t ilx, int *nid) | |
5da7ca86 | 2046 | { |
ddc1a5cb | 2047 | nodemask_t *nodemask = NULL; |
5da7ca86 | 2048 | |
ddc1a5cb HD |
2049 | switch (pol->mode) { |
2050 | case MPOL_PREFERRED: | |
2051 | /* Override input node id */ | |
2052 | *nid = first_node(pol->nodes); | |
2053 | break; | |
2054 | case MPOL_PREFERRED_MANY: | |
2055 | nodemask = &pol->nodes; | |
2056 | if (pol->home_node != NUMA_NO_NODE) | |
2057 | *nid = pol->home_node; | |
2058 | break; | |
2059 | case MPOL_BIND: | |
2060 | /* Restrict to nodemask (but not on lower zones) */ | |
2061 | if (apply_policy_zone(pol, gfp_zone(gfp)) && | |
2062 | cpuset_nodemask_valid_mems_allowed(&pol->nodes)) | |
2063 | nodemask = &pol->nodes; | |
2064 | if (pol->home_node != NUMA_NO_NODE) | |
2065 | *nid = pol->home_node; | |
3b98b087 | 2066 | /* |
ddc1a5cb HD |
2067 | * __GFP_THISNODE shouldn't even be used with the bind policy |
2068 | * because we might easily break the expectation to stay on the | |
2069 | * requested node and not break the policy. | |
3b98b087 | 2070 | */ |
ddc1a5cb HD |
2071 | WARN_ON_ONCE(gfp & __GFP_THISNODE); |
2072 | break; | |
2073 | case MPOL_INTERLEAVE: | |
2074 | /* Override input node id */ | |
2075 | *nid = (ilx == NO_INTERLEAVE_INDEX) ? | |
2076 | interleave_nodes(pol) : interleave_nid(pol, ilx); | |
2077 | break; | |
fa3bea4e GP |
2078 | case MPOL_WEIGHTED_INTERLEAVE: |
2079 | *nid = (ilx == NO_INTERLEAVE_INDEX) ? | |
2080 | weighted_interleave_nodes(pol) : | |
2081 | weighted_interleave_nid(pol, ilx); | |
2082 | break; | |
ddc1a5cb HD |
2083 | } |
2084 | ||
2085 | return nodemask; | |
5da7ca86 CL |
2086 | } |
2087 | ||
00ac59ad | 2088 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2089 | /* |
04ec6264 | 2090 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2091 | * @vma: virtual memory area whose policy is sought |
2092 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2093 | * @gfp_flags: for requested zone | |
2094 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
b27abacc | 2095 | * @nodemask: pointer to nodemask pointer for 'bind' and 'prefer-many' policy |
480eccf9 | 2096 | * |
04ec6264 | 2097 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 | 2098 | * to the struct mempolicy for conditional unref after allocation. |
b27abacc DH |
2099 | * If the effective policy is 'bind' or 'prefer-many', returns a pointer |
2100 | * to the mempolicy's @nodemask for filtering the zonelist. | |
480eccf9 | 2101 | */ |
04ec6264 | 2102 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
ddc1a5cb | 2103 | struct mempolicy **mpol, nodemask_t **nodemask) |
5da7ca86 | 2104 | { |
ddc1a5cb | 2105 | pgoff_t ilx; |
04ec6264 | 2106 | int nid; |
5da7ca86 | 2107 | |
ddc1a5cb HD |
2108 | nid = numa_node_id(); |
2109 | *mpol = get_vma_policy(vma, addr, hstate_vma(vma)->order, &ilx); | |
2110 | *nodemask = policy_nodemask(gfp_flags, *mpol, ilx, &nid); | |
04ec6264 | 2111 | return nid; |
5da7ca86 | 2112 | } |
06808b08 LS |
2113 | |
2114 | /* | |
2115 | * init_nodemask_of_mempolicy | |
2116 | * | |
2117 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2118 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2119 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2120 | * initialize the argument nodemask to contain the single node for | |
2121 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2122 | * of non-default mempolicy. | |
2123 | * | |
2124 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2125 | * because the current task is examining it's own mempolicy and a task's | |
2126 | * mempolicy is only ever changed by the task itself. | |
2127 | * | |
2128 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2129 | */ | |
2130 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2131 | { | |
2132 | struct mempolicy *mempolicy; | |
06808b08 LS |
2133 | |
2134 | if (!(mask && current->mempolicy)) | |
2135 | return false; | |
2136 | ||
c0ff7453 | 2137 | task_lock(current); |
06808b08 LS |
2138 | mempolicy = current->mempolicy; |
2139 | switch (mempolicy->mode) { | |
2140 | case MPOL_PREFERRED: | |
b27abacc | 2141 | case MPOL_PREFERRED_MANY: |
06808b08 | 2142 | case MPOL_BIND: |
06808b08 | 2143 | case MPOL_INTERLEAVE: |
fa3bea4e | 2144 | case MPOL_WEIGHTED_INTERLEAVE: |
269fbe72 | 2145 | *mask = mempolicy->nodes; |
7858d7bc FT |
2146 | break; |
2147 | ||
2148 | case MPOL_LOCAL: | |
269fbe72 | 2149 | init_nodemask_of_node(mask, numa_node_id()); |
06808b08 LS |
2150 | break; |
2151 | ||
2152 | default: | |
2153 | BUG(); | |
2154 | } | |
c0ff7453 | 2155 | task_unlock(current); |
06808b08 LS |
2156 | |
2157 | return true; | |
2158 | } | |
00ac59ad | 2159 | #endif |
5da7ca86 | 2160 | |
6f48d0eb | 2161 | /* |
b26e517a | 2162 | * mempolicy_in_oom_domain |
6f48d0eb | 2163 | * |
b26e517a FT |
2164 | * If tsk's mempolicy is "bind", check for intersection between mask and |
2165 | * the policy nodemask. Otherwise, return true for all other policies | |
2166 | * including "interleave", as a tsk with "interleave" policy may have | |
2167 | * memory allocated from all nodes in system. | |
6f48d0eb DR |
2168 | * |
2169 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2170 | */ | |
b26e517a | 2171 | bool mempolicy_in_oom_domain(struct task_struct *tsk, |
6f48d0eb DR |
2172 | const nodemask_t *mask) |
2173 | { | |
2174 | struct mempolicy *mempolicy; | |
2175 | bool ret = true; | |
2176 | ||
2177 | if (!mask) | |
2178 | return ret; | |
b26e517a | 2179 | |
6f48d0eb DR |
2180 | task_lock(tsk); |
2181 | mempolicy = tsk->mempolicy; | |
b26e517a | 2182 | if (mempolicy && mempolicy->mode == MPOL_BIND) |
269fbe72 | 2183 | ret = nodes_intersects(mempolicy->nodes, *mask); |
6f48d0eb | 2184 | task_unlock(tsk); |
b26e517a | 2185 | |
6f48d0eb DR |
2186 | return ret; |
2187 | } | |
2188 | ||
4c54d949 | 2189 | static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order, |
ddc1a5cb | 2190 | int nid, nodemask_t *nodemask) |
4c54d949 FT |
2191 | { |
2192 | struct page *page; | |
2193 | gfp_t preferred_gfp; | |
2194 | ||
2195 | /* | |
2196 | * This is a two pass approach. The first pass will only try the | |
2197 | * preferred nodes but skip the direct reclaim and allow the | |
2198 | * allocation to fail, while the second pass will try all the | |
2199 | * nodes in system. | |
2200 | */ | |
2201 | preferred_gfp = gfp | __GFP_NOWARN; | |
2202 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
ddc1a5cb | 2203 | page = __alloc_pages(preferred_gfp, order, nid, nodemask); |
4c54d949 | 2204 | if (!page) |
c0455116 | 2205 | page = __alloc_pages(gfp, order, nid, NULL); |
4c54d949 FT |
2206 | |
2207 | return page; | |
2208 | } | |
2209 | ||
1da177e4 | 2210 | /** |
ddc1a5cb | 2211 | * alloc_pages_mpol - Allocate pages according to NUMA mempolicy. |
eb350739 | 2212 | * @gfp: GFP flags. |
ddc1a5cb HD |
2213 | * @order: Order of the page allocation. |
2214 | * @pol: Pointer to the NUMA mempolicy. | |
2215 | * @ilx: Index for interleave mempolicy (also distinguishes alloc_pages()). | |
2216 | * @nid: Preferred node (usually numa_node_id() but @mpol may override it). | |
1da177e4 | 2217 | * |
ddc1a5cb | 2218 | * Return: The page on success or NULL if allocation fails. |
1da177e4 | 2219 | */ |
ddc1a5cb HD |
2220 | struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order, |
2221 | struct mempolicy *pol, pgoff_t ilx, int nid) | |
1da177e4 | 2222 | { |
ddc1a5cb HD |
2223 | nodemask_t *nodemask; |
2224 | struct page *page; | |
adf88aa8 | 2225 | |
ddc1a5cb | 2226 | nodemask = policy_nodemask(gfp, pol, ilx, &nid); |
4c54d949 | 2227 | |
ddc1a5cb HD |
2228 | if (pol->mode == MPOL_PREFERRED_MANY) |
2229 | return alloc_pages_preferred_many(gfp, order, nid, nodemask); | |
19deb769 | 2230 | |
ddc1a5cb HD |
2231 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
2232 | /* filter "hugepage" allocation, unless from alloc_pages() */ | |
2233 | order == HPAGE_PMD_ORDER && ilx != NO_INTERLEAVE_INDEX) { | |
19deb769 DR |
2234 | /* |
2235 | * For hugepage allocation and non-interleave policy which | |
2236 | * allows the current node (or other explicitly preferred | |
2237 | * node) we only try to allocate from the current/preferred | |
2238 | * node and don't fall back to other nodes, as the cost of | |
2239 | * remote accesses would likely offset THP benefits. | |
2240 | * | |
b27abacc | 2241 | * If the policy is interleave or does not allow the current |
19deb769 DR |
2242 | * node in its nodemask, we allocate the standard way. |
2243 | */ | |
ddc1a5cb | 2244 | if (pol->mode != MPOL_INTERLEAVE && |
fa3bea4e | 2245 | pol->mode != MPOL_WEIGHTED_INTERLEAVE && |
ddc1a5cb | 2246 | (!nodemask || node_isset(nid, *nodemask))) { |
cc638f32 VB |
2247 | /* |
2248 | * First, try to allocate THP only on local node, but | |
2249 | * don't reclaim unnecessarily, just compact. | |
2250 | */ | |
ddc1a5cb HD |
2251 | page = __alloc_pages_node(nid, |
2252 | gfp | __GFP_THISNODE | __GFP_NORETRY, order); | |
2253 | if (page || !(gfp & __GFP_DIRECT_RECLAIM)) | |
2254 | return page; | |
76e654cc DR |
2255 | /* |
2256 | * If hugepage allocations are configured to always | |
2257 | * synchronous compact or the vma has been madvised | |
2258 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2259 | * memory with both reclaim and compact as well. |
76e654cc | 2260 | */ |
ddc1a5cb HD |
2261 | } |
2262 | } | |
76e654cc | 2263 | |
ddc1a5cb HD |
2264 | page = __alloc_pages(gfp, order, nid, nodemask); |
2265 | ||
2266 | if (unlikely(pol->mode == MPOL_INTERLEAVE) && page) { | |
2267 | /* skip NUMA_INTERLEAVE_HIT update if numa stats is disabled */ | |
2268 | if (static_branch_likely(&vm_numa_stat_key) && | |
2269 | page_to_nid(page) == nid) { | |
2270 | preempt_disable(); | |
2271 | __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT); | |
2272 | preempt_enable(); | |
19deb769 | 2273 | } |
356ff8a9 DR |
2274 | } |
2275 | ||
ddc1a5cb HD |
2276 | return page; |
2277 | } | |
2278 | ||
2279 | /** | |
2280 | * vma_alloc_folio - Allocate a folio for a VMA. | |
2281 | * @gfp: GFP flags. | |
2282 | * @order: Order of the folio. | |
2283 | * @vma: Pointer to VMA. | |
2284 | * @addr: Virtual address of the allocation. Must be inside @vma. | |
2285 | * @hugepage: Unused (was: For hugepages try only preferred node if possible). | |
2286 | * | |
2287 | * Allocate a folio for a specific address in @vma, using the appropriate | |
2288 | * NUMA policy. The caller must hold the mmap_lock of the mm_struct of the | |
2289 | * VMA to prevent it from going away. Should be used for all allocations | |
2290 | * for folios that will be mapped into user space, excepting hugetlbfs, and | |
2291 | * excepting where direct use of alloc_pages_mpol() is more appropriate. | |
2292 | * | |
2293 | * Return: The folio on success or NULL if allocation fails. | |
2294 | */ | |
2295 | struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma, | |
2296 | unsigned long addr, bool hugepage) | |
2297 | { | |
2298 | struct mempolicy *pol; | |
2299 | pgoff_t ilx; | |
2300 | struct page *page; | |
2301 | ||
2302 | pol = get_vma_policy(vma, addr, order, &ilx); | |
2303 | page = alloc_pages_mpol(gfp | __GFP_COMP, order, | |
2304 | pol, ilx, numa_node_id()); | |
d51e9894 | 2305 | mpol_cond_put(pol); |
ddc1a5cb | 2306 | return page_rmappable_folio(page); |
f584b680 | 2307 | } |
adf88aa8 | 2308 | EXPORT_SYMBOL(vma_alloc_folio); |
f584b680 | 2309 | |
1da177e4 | 2310 | /** |
6421ec76 MWO |
2311 | * alloc_pages - Allocate pages. |
2312 | * @gfp: GFP flags. | |
2313 | * @order: Power of two of number of pages to allocate. | |
1da177e4 | 2314 | * |
6421ec76 MWO |
2315 | * Allocate 1 << @order contiguous pages. The physical address of the |
2316 | * first page is naturally aligned (eg an order-3 allocation will be aligned | |
2317 | * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current | |
2318 | * process is honoured when in process context. | |
1da177e4 | 2319 | * |
6421ec76 MWO |
2320 | * Context: Can be called from any context, providing the appropriate GFP |
2321 | * flags are used. | |
2322 | * Return: The page on success or NULL if allocation fails. | |
1da177e4 | 2323 | */ |
ddc1a5cb | 2324 | struct page *alloc_pages(gfp_t gfp, unsigned int order) |
1da177e4 | 2325 | { |
8d90274b | 2326 | struct mempolicy *pol = &default_policy; |
52cd3b07 LS |
2327 | |
2328 | /* | |
2329 | * No reference counting needed for current->mempolicy | |
2330 | * nor system default_policy | |
2331 | */ | |
ddc1a5cb HD |
2332 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2333 | pol = get_task_policy(current); | |
cc9a6c87 | 2334 | |
ddc1a5cb HD |
2335 | return alloc_pages_mpol(gfp, order, |
2336 | pol, NO_INTERLEAVE_INDEX, numa_node_id()); | |
1da177e4 | 2337 | } |
d7f946d0 | 2338 | EXPORT_SYMBOL(alloc_pages); |
1da177e4 | 2339 | |
ddc1a5cb | 2340 | struct folio *folio_alloc(gfp_t gfp, unsigned int order) |
cc09cb13 | 2341 | { |
23e48832 | 2342 | return page_rmappable_folio(alloc_pages(gfp | __GFP_COMP, order)); |
cc09cb13 MWO |
2343 | } |
2344 | EXPORT_SYMBOL(folio_alloc); | |
2345 | ||
c00b6b96 CW |
2346 | static unsigned long alloc_pages_bulk_array_interleave(gfp_t gfp, |
2347 | struct mempolicy *pol, unsigned long nr_pages, | |
2348 | struct page **page_array) | |
2349 | { | |
2350 | int nodes; | |
2351 | unsigned long nr_pages_per_node; | |
2352 | int delta; | |
2353 | int i; | |
2354 | unsigned long nr_allocated; | |
2355 | unsigned long total_allocated = 0; | |
2356 | ||
2357 | nodes = nodes_weight(pol->nodes); | |
2358 | nr_pages_per_node = nr_pages / nodes; | |
2359 | delta = nr_pages - nodes * nr_pages_per_node; | |
2360 | ||
2361 | for (i = 0; i < nodes; i++) { | |
2362 | if (delta) { | |
2363 | nr_allocated = __alloc_pages_bulk(gfp, | |
2364 | interleave_nodes(pol), NULL, | |
2365 | nr_pages_per_node + 1, NULL, | |
2366 | page_array); | |
2367 | delta--; | |
2368 | } else { | |
2369 | nr_allocated = __alloc_pages_bulk(gfp, | |
2370 | interleave_nodes(pol), NULL, | |
2371 | nr_pages_per_node, NULL, page_array); | |
2372 | } | |
2373 | ||
2374 | page_array += nr_allocated; | |
2375 | total_allocated += nr_allocated; | |
2376 | } | |
2377 | ||
2378 | return total_allocated; | |
2379 | } | |
2380 | ||
fa3bea4e GP |
2381 | static unsigned long alloc_pages_bulk_array_weighted_interleave(gfp_t gfp, |
2382 | struct mempolicy *pol, unsigned long nr_pages, | |
2383 | struct page **page_array) | |
2384 | { | |
2385 | struct task_struct *me = current; | |
274519ed | 2386 | unsigned int cpuset_mems_cookie; |
fa3bea4e GP |
2387 | unsigned long total_allocated = 0; |
2388 | unsigned long nr_allocated = 0; | |
2389 | unsigned long rounds; | |
2390 | unsigned long node_pages, delta; | |
2391 | u8 *table, *weights, weight; | |
2392 | unsigned int weight_total = 0; | |
2393 | unsigned long rem_pages = nr_pages; | |
2394 | nodemask_t nodes; | |
2395 | int nnodes, node; | |
2396 | int resume_node = MAX_NUMNODES - 1; | |
2397 | u8 resume_weight = 0; | |
2398 | int prev_node; | |
2399 | int i; | |
2400 | ||
2401 | if (!nr_pages) | |
2402 | return 0; | |
2403 | ||
274519ed GP |
2404 | /* read the nodes onto the stack, retry if done during rebind */ |
2405 | do { | |
2406 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
2407 | nnodes = read_once_policy_nodemask(pol, &nodes); | |
2408 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); | |
2409 | ||
2410 | /* if the nodemask has become invalid, we cannot do anything */ | |
fa3bea4e GP |
2411 | if (!nnodes) |
2412 | return 0; | |
2413 | ||
2414 | /* Continue allocating from most recent node and adjust the nr_pages */ | |
2415 | node = me->il_prev; | |
2416 | weight = me->il_weight; | |
2417 | if (weight && node_isset(node, nodes)) { | |
2418 | node_pages = min(rem_pages, weight); | |
2419 | nr_allocated = __alloc_pages_bulk(gfp, node, NULL, node_pages, | |
2420 | NULL, page_array); | |
2421 | page_array += nr_allocated; | |
2422 | total_allocated += nr_allocated; | |
2423 | /* if that's all the pages, no need to interleave */ | |
2424 | if (rem_pages <= weight) { | |
2425 | me->il_weight -= rem_pages; | |
2426 | return total_allocated; | |
2427 | } | |
2428 | /* Otherwise we adjust remaining pages, continue from there */ | |
2429 | rem_pages -= weight; | |
2430 | } | |
2431 | /* clear active weight in case of an allocation failure */ | |
2432 | me->il_weight = 0; | |
2433 | prev_node = node; | |
2434 | ||
2435 | /* create a local copy of node weights to operate on outside rcu */ | |
2436 | weights = kzalloc(nr_node_ids, GFP_KERNEL); | |
2437 | if (!weights) | |
2438 | return total_allocated; | |
2439 | ||
2440 | rcu_read_lock(); | |
2441 | table = rcu_dereference(iw_table); | |
2442 | if (table) | |
2443 | memcpy(weights, table, nr_node_ids); | |
2444 | rcu_read_unlock(); | |
2445 | ||
2446 | /* calculate total, detect system default usage */ | |
2447 | for_each_node_mask(node, nodes) { | |
2448 | if (!weights[node]) | |
2449 | weights[node] = 1; | |
2450 | weight_total += weights[node]; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * Calculate rounds/partial rounds to minimize __alloc_pages_bulk calls. | |
2455 | * Track which node weighted interleave should resume from. | |
2456 | * | |
2457 | * if (rounds > 0) and (delta == 0), resume_node will always be | |
2458 | * the node following prev_node and its weight. | |
2459 | */ | |
2460 | rounds = rem_pages / weight_total; | |
2461 | delta = rem_pages % weight_total; | |
2462 | resume_node = next_node_in(prev_node, nodes); | |
2463 | resume_weight = weights[resume_node]; | |
2464 | for (i = 0; i < nnodes; i++) { | |
2465 | node = next_node_in(prev_node, nodes); | |
2466 | weight = weights[node]; | |
2467 | node_pages = weight * rounds; | |
2468 | /* If a delta exists, add this node's portion of the delta */ | |
2469 | if (delta > weight) { | |
2470 | node_pages += weight; | |
2471 | delta -= weight; | |
2472 | } else if (delta) { | |
2473 | /* when delta is depleted, resume from that node */ | |
2474 | node_pages += delta; | |
2475 | resume_node = node; | |
2476 | resume_weight = weight - delta; | |
2477 | delta = 0; | |
2478 | } | |
2479 | /* node_pages can be 0 if an allocation fails and rounds == 0 */ | |
2480 | if (!node_pages) | |
2481 | break; | |
2482 | nr_allocated = __alloc_pages_bulk(gfp, node, NULL, node_pages, | |
2483 | NULL, page_array); | |
2484 | page_array += nr_allocated; | |
2485 | total_allocated += nr_allocated; | |
2486 | if (total_allocated == nr_pages) | |
2487 | break; | |
2488 | prev_node = node; | |
2489 | } | |
2490 | me->il_prev = resume_node; | |
2491 | me->il_weight = resume_weight; | |
2492 | kfree(weights); | |
2493 | return total_allocated; | |
2494 | } | |
2495 | ||
c00b6b96 CW |
2496 | static unsigned long alloc_pages_bulk_array_preferred_many(gfp_t gfp, int nid, |
2497 | struct mempolicy *pol, unsigned long nr_pages, | |
2498 | struct page **page_array) | |
2499 | { | |
2500 | gfp_t preferred_gfp; | |
2501 | unsigned long nr_allocated = 0; | |
2502 | ||
2503 | preferred_gfp = gfp | __GFP_NOWARN; | |
2504 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2505 | ||
2506 | nr_allocated = __alloc_pages_bulk(preferred_gfp, nid, &pol->nodes, | |
2507 | nr_pages, NULL, page_array); | |
2508 | ||
2509 | if (nr_allocated < nr_pages) | |
2510 | nr_allocated += __alloc_pages_bulk(gfp, numa_node_id(), NULL, | |
2511 | nr_pages - nr_allocated, NULL, | |
2512 | page_array + nr_allocated); | |
2513 | return nr_allocated; | |
2514 | } | |
2515 | ||
2516 | /* alloc pages bulk and mempolicy should be considered at the | |
2517 | * same time in some situation such as vmalloc. | |
2518 | * | |
2519 | * It can accelerate memory allocation especially interleaving | |
2520 | * allocate memory. | |
2521 | */ | |
2522 | unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, | |
2523 | unsigned long nr_pages, struct page **page_array) | |
2524 | { | |
2525 | struct mempolicy *pol = &default_policy; | |
ddc1a5cb HD |
2526 | nodemask_t *nodemask; |
2527 | int nid; | |
c00b6b96 CW |
2528 | |
2529 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) | |
2530 | pol = get_task_policy(current); | |
2531 | ||
2532 | if (pol->mode == MPOL_INTERLEAVE) | |
2533 | return alloc_pages_bulk_array_interleave(gfp, pol, | |
2534 | nr_pages, page_array); | |
2535 | ||
fa3bea4e GP |
2536 | if (pol->mode == MPOL_WEIGHTED_INTERLEAVE) |
2537 | return alloc_pages_bulk_array_weighted_interleave( | |
2538 | gfp, pol, nr_pages, page_array); | |
2539 | ||
c00b6b96 CW |
2540 | if (pol->mode == MPOL_PREFERRED_MANY) |
2541 | return alloc_pages_bulk_array_preferred_many(gfp, | |
2542 | numa_node_id(), pol, nr_pages, page_array); | |
2543 | ||
ddc1a5cb HD |
2544 | nid = numa_node_id(); |
2545 | nodemask = policy_nodemask(gfp, pol, NO_INTERLEAVE_INDEX, &nid); | |
2546 | return __alloc_pages_bulk(gfp, nid, nodemask, | |
2547 | nr_pages, NULL, page_array); | |
c00b6b96 CW |
2548 | } |
2549 | ||
ef0855d3 ON |
2550 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2551 | { | |
c36f6e6d | 2552 | struct mempolicy *pol = mpol_dup(src->vm_policy); |
ef0855d3 ON |
2553 | |
2554 | if (IS_ERR(pol)) | |
2555 | return PTR_ERR(pol); | |
2556 | dst->vm_policy = pol; | |
2557 | return 0; | |
2558 | } | |
2559 | ||
4225399a | 2560 | /* |
846a16bf | 2561 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2562 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2563 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2564 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2565 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2566 | * |
2567 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2568 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2569 | */ |
4225399a | 2570 | |
846a16bf LS |
2571 | /* Slow path of a mempolicy duplicate */ |
2572 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2573 | { |
2574 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2575 | ||
2576 | if (!new) | |
2577 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2578 | |
2579 | /* task's mempolicy is protected by alloc_lock */ | |
2580 | if (old == current->mempolicy) { | |
2581 | task_lock(current); | |
2582 | *new = *old; | |
2583 | task_unlock(current); | |
2584 | } else | |
2585 | *new = *old; | |
2586 | ||
4225399a PJ |
2587 | if (current_cpuset_is_being_rebound()) { |
2588 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2589 | mpol_rebind_policy(new, &mems); |
4225399a | 2590 | } |
1da177e4 | 2591 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2592 | return new; |
2593 | } | |
2594 | ||
2595 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2596 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2597 | { |
2598 | if (!a || !b) | |
fcfb4dcc | 2599 | return false; |
45c4745a | 2600 | if (a->mode != b->mode) |
fcfb4dcc | 2601 | return false; |
19800502 | 2602 | if (a->flags != b->flags) |
fcfb4dcc | 2603 | return false; |
c6018b4b AK |
2604 | if (a->home_node != b->home_node) |
2605 | return false; | |
19800502 BL |
2606 | if (mpol_store_user_nodemask(a)) |
2607 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2608 | return false; |
19800502 | 2609 | |
45c4745a | 2610 | switch (a->mode) { |
19770b32 | 2611 | case MPOL_BIND: |
1da177e4 | 2612 | case MPOL_INTERLEAVE: |
1da177e4 | 2613 | case MPOL_PREFERRED: |
b27abacc | 2614 | case MPOL_PREFERRED_MANY: |
fa3bea4e | 2615 | case MPOL_WEIGHTED_INTERLEAVE: |
269fbe72 | 2616 | return !!nodes_equal(a->nodes, b->nodes); |
7858d7bc FT |
2617 | case MPOL_LOCAL: |
2618 | return true; | |
1da177e4 LT |
2619 | default: |
2620 | BUG(); | |
fcfb4dcc | 2621 | return false; |
1da177e4 LT |
2622 | } |
2623 | } | |
2624 | ||
1da177e4 LT |
2625 | /* |
2626 | * Shared memory backing store policy support. | |
2627 | * | |
2628 | * Remember policies even when nobody has shared memory mapped. | |
2629 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2630 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2631 | * for any accesses to the tree. |
2632 | */ | |
2633 | ||
4a8c7bb5 NZ |
2634 | /* |
2635 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2636 | * reading or for writing | |
2637 | */ | |
93397c3b HD |
2638 | static struct sp_node *sp_lookup(struct shared_policy *sp, |
2639 | pgoff_t start, pgoff_t end) | |
1da177e4 LT |
2640 | { |
2641 | struct rb_node *n = sp->root.rb_node; | |
2642 | ||
2643 | while (n) { | |
2644 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2645 | ||
2646 | if (start >= p->end) | |
2647 | n = n->rb_right; | |
2648 | else if (end <= p->start) | |
2649 | n = n->rb_left; | |
2650 | else | |
2651 | break; | |
2652 | } | |
2653 | if (!n) | |
2654 | return NULL; | |
2655 | for (;;) { | |
2656 | struct sp_node *w = NULL; | |
2657 | struct rb_node *prev = rb_prev(n); | |
2658 | if (!prev) | |
2659 | break; | |
2660 | w = rb_entry(prev, struct sp_node, nd); | |
2661 | if (w->end <= start) | |
2662 | break; | |
2663 | n = prev; | |
2664 | } | |
2665 | return rb_entry(n, struct sp_node, nd); | |
2666 | } | |
2667 | ||
4a8c7bb5 NZ |
2668 | /* |
2669 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2670 | * writing. | |
2671 | */ | |
1da177e4 LT |
2672 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2673 | { | |
2674 | struct rb_node **p = &sp->root.rb_node; | |
2675 | struct rb_node *parent = NULL; | |
2676 | struct sp_node *nd; | |
2677 | ||
2678 | while (*p) { | |
2679 | parent = *p; | |
2680 | nd = rb_entry(parent, struct sp_node, nd); | |
2681 | if (new->start < nd->start) | |
2682 | p = &(*p)->rb_left; | |
2683 | else if (new->end > nd->end) | |
2684 | p = &(*p)->rb_right; | |
2685 | else | |
2686 | BUG(); | |
2687 | } | |
2688 | rb_link_node(&new->nd, parent, p); | |
2689 | rb_insert_color(&new->nd, &sp->root); | |
1da177e4 LT |
2690 | } |
2691 | ||
2692 | /* Find shared policy intersecting idx */ | |
93397c3b HD |
2693 | struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp, |
2694 | pgoff_t idx) | |
1da177e4 LT |
2695 | { |
2696 | struct mempolicy *pol = NULL; | |
2697 | struct sp_node *sn; | |
2698 | ||
2699 | if (!sp->root.rb_node) | |
2700 | return NULL; | |
4a8c7bb5 | 2701 | read_lock(&sp->lock); |
1da177e4 LT |
2702 | sn = sp_lookup(sp, idx, idx+1); |
2703 | if (sn) { | |
2704 | mpol_get(sn->policy); | |
2705 | pol = sn->policy; | |
2706 | } | |
4a8c7bb5 | 2707 | read_unlock(&sp->lock); |
1da177e4 LT |
2708 | return pol; |
2709 | } | |
2710 | ||
63f74ca2 KM |
2711 | static void sp_free(struct sp_node *n) |
2712 | { | |
2713 | mpol_put(n->policy); | |
2714 | kmem_cache_free(sn_cache, n); | |
2715 | } | |
2716 | ||
771fb4d8 | 2717 | /** |
75c70128 | 2718 | * mpol_misplaced - check whether current folio node is valid in policy |
771fb4d8 | 2719 | * |
75c70128 | 2720 | * @folio: folio to be checked |
f8fd525b | 2721 | * @vmf: structure describing the fault |
75c70128 | 2722 | * @addr: virtual address in @vma for shared policy lookup and interleave policy |
771fb4d8 | 2723 | * |
75c70128 | 2724 | * Lookup current policy node id for vma,addr and "compare to" folio's |
5f076944 | 2725 | * node id. Policy determination "mimics" alloc_page_vma(). |
771fb4d8 | 2726 | * Called from fault path where we know the vma and faulting address. |
5f076944 | 2727 | * |
062db293 | 2728 | * Return: NUMA_NO_NODE if the page is in a node that is valid for this |
75c70128 | 2729 | * policy, or a suitable node ID to allocate a replacement folio from. |
771fb4d8 | 2730 | */ |
f8fd525b | 2731 | int mpol_misplaced(struct folio *folio, struct vm_fault *vmf, |
75c70128 | 2732 | unsigned long addr) |
771fb4d8 LS |
2733 | { |
2734 | struct mempolicy *pol; | |
ddc1a5cb | 2735 | pgoff_t ilx; |
c33d6c06 | 2736 | struct zoneref *z; |
75c70128 | 2737 | int curnid = folio_nid(folio); |
f8fd525b | 2738 | struct vm_area_struct *vma = vmf->vma; |
90572890 | 2739 | int thiscpu = raw_smp_processor_id(); |
f8fd525b | 2740 | int thisnid = numa_node_id(); |
98fa15f3 | 2741 | int polnid = NUMA_NO_NODE; |
062db293 | 2742 | int ret = NUMA_NO_NODE; |
771fb4d8 | 2743 | |
f8fd525b DT |
2744 | /* |
2745 | * Make sure ptl is held so that we don't preempt and we | |
2746 | * have a stable smp processor id | |
2747 | */ | |
2748 | lockdep_assert_held(vmf->ptl); | |
ddc1a5cb | 2749 | pol = get_vma_policy(vma, addr, folio_order(folio), &ilx); |
771fb4d8 LS |
2750 | if (!(pol->flags & MPOL_F_MOF)) |
2751 | goto out; | |
2752 | ||
2753 | switch (pol->mode) { | |
2754 | case MPOL_INTERLEAVE: | |
ddc1a5cb | 2755 | polnid = interleave_nid(pol, ilx); |
771fb4d8 LS |
2756 | break; |
2757 | ||
fa3bea4e GP |
2758 | case MPOL_WEIGHTED_INTERLEAVE: |
2759 | polnid = weighted_interleave_nid(pol, ilx); | |
2760 | break; | |
2761 | ||
771fb4d8 | 2762 | case MPOL_PREFERRED: |
b27abacc DH |
2763 | if (node_isset(curnid, pol->nodes)) |
2764 | goto out; | |
269fbe72 | 2765 | polnid = first_node(pol->nodes); |
7858d7bc FT |
2766 | break; |
2767 | ||
2768 | case MPOL_LOCAL: | |
2769 | polnid = numa_node_id(); | |
771fb4d8 LS |
2770 | break; |
2771 | ||
2772 | case MPOL_BIND: | |
bda420b9 HY |
2773 | /* Optimize placement among multiple nodes via NUMA balancing */ |
2774 | if (pol->flags & MPOL_F_MORON) { | |
269fbe72 | 2775 | if (node_isset(thisnid, pol->nodes)) |
bda420b9 HY |
2776 | break; |
2777 | goto out; | |
2778 | } | |
b27abacc | 2779 | fallthrough; |
c33d6c06 | 2780 | |
b27abacc | 2781 | case MPOL_PREFERRED_MANY: |
771fb4d8 | 2782 | /* |
771fb4d8 LS |
2783 | * use current page if in policy nodemask, |
2784 | * else select nearest allowed node, if any. | |
2785 | * If no allowed nodes, use current [!misplaced]. | |
2786 | */ | |
269fbe72 | 2787 | if (node_isset(curnid, pol->nodes)) |
771fb4d8 | 2788 | goto out; |
c33d6c06 | 2789 | z = first_zones_zonelist( |
f8fd525b | 2790 | node_zonelist(thisnid, GFP_HIGHUSER), |
771fb4d8 | 2791 | gfp_zone(GFP_HIGHUSER), |
269fbe72 | 2792 | &pol->nodes); |
c1093b74 | 2793 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2794 | break; |
2795 | ||
2796 | default: | |
2797 | BUG(); | |
2798 | } | |
5606e387 | 2799 | |
75c70128 | 2800 | /* Migrate the folio towards the node whose CPU is referencing it */ |
e42c8ff2 | 2801 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2802 | polnid = thisnid; |
5606e387 | 2803 | |
8c9ae56d | 2804 | if (!should_numa_migrate_memory(current, folio, curnid, |
75c70128 | 2805 | thiscpu)) |
de1c9ce6 | 2806 | goto out; |
e42c8ff2 MG |
2807 | } |
2808 | ||
771fb4d8 LS |
2809 | if (curnid != polnid) |
2810 | ret = polnid; | |
2811 | out: | |
2812 | mpol_cond_put(pol); | |
2813 | ||
2814 | return ret; | |
2815 | } | |
2816 | ||
c11600e4 DR |
2817 | /* |
2818 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2819 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2820 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2821 | * policy. | |
2822 | */ | |
2823 | void mpol_put_task_policy(struct task_struct *task) | |
2824 | { | |
2825 | struct mempolicy *pol; | |
2826 | ||
2827 | task_lock(task); | |
2828 | pol = task->mempolicy; | |
2829 | task->mempolicy = NULL; | |
2830 | task_unlock(task); | |
2831 | mpol_put(pol); | |
2832 | } | |
2833 | ||
1da177e4 LT |
2834 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2835 | { | |
1da177e4 | 2836 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2837 | sp_free(n); |
1da177e4 LT |
2838 | } |
2839 | ||
42288fe3 MG |
2840 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2841 | unsigned long end, struct mempolicy *pol) | |
2842 | { | |
2843 | node->start = start; | |
2844 | node->end = end; | |
2845 | node->policy = pol; | |
2846 | } | |
2847 | ||
dbcb0f19 AB |
2848 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2849 | struct mempolicy *pol) | |
1da177e4 | 2850 | { |
869833f2 KM |
2851 | struct sp_node *n; |
2852 | struct mempolicy *newpol; | |
1da177e4 | 2853 | |
869833f2 | 2854 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2855 | if (!n) |
2856 | return NULL; | |
869833f2 KM |
2857 | |
2858 | newpol = mpol_dup(pol); | |
2859 | if (IS_ERR(newpol)) { | |
2860 | kmem_cache_free(sn_cache, n); | |
2861 | return NULL; | |
2862 | } | |
2863 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2864 | sp_node_init(n, start, end, newpol); |
869833f2 | 2865 | |
1da177e4 LT |
2866 | return n; |
2867 | } | |
2868 | ||
2869 | /* Replace a policy range. */ | |
93397c3b HD |
2870 | static int shared_policy_replace(struct shared_policy *sp, pgoff_t start, |
2871 | pgoff_t end, struct sp_node *new) | |
1da177e4 | 2872 | { |
b22d127a | 2873 | struct sp_node *n; |
42288fe3 MG |
2874 | struct sp_node *n_new = NULL; |
2875 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2876 | int ret = 0; |
1da177e4 | 2877 | |
42288fe3 | 2878 | restart: |
4a8c7bb5 | 2879 | write_lock(&sp->lock); |
1da177e4 LT |
2880 | n = sp_lookup(sp, start, end); |
2881 | /* Take care of old policies in the same range. */ | |
2882 | while (n && n->start < end) { | |
2883 | struct rb_node *next = rb_next(&n->nd); | |
2884 | if (n->start >= start) { | |
2885 | if (n->end <= end) | |
2886 | sp_delete(sp, n); | |
2887 | else | |
2888 | n->start = end; | |
2889 | } else { | |
2890 | /* Old policy spanning whole new range. */ | |
2891 | if (n->end > end) { | |
42288fe3 MG |
2892 | if (!n_new) |
2893 | goto alloc_new; | |
2894 | ||
2895 | *mpol_new = *n->policy; | |
2896 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2897 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2898 | n->end = start; |
5ca39575 | 2899 | sp_insert(sp, n_new); |
42288fe3 MG |
2900 | n_new = NULL; |
2901 | mpol_new = NULL; | |
1da177e4 LT |
2902 | break; |
2903 | } else | |
2904 | n->end = start; | |
2905 | } | |
2906 | if (!next) | |
2907 | break; | |
2908 | n = rb_entry(next, struct sp_node, nd); | |
2909 | } | |
2910 | if (new) | |
2911 | sp_insert(sp, new); | |
4a8c7bb5 | 2912 | write_unlock(&sp->lock); |
42288fe3 MG |
2913 | ret = 0; |
2914 | ||
2915 | err_out: | |
2916 | if (mpol_new) | |
2917 | mpol_put(mpol_new); | |
2918 | if (n_new) | |
2919 | kmem_cache_free(sn_cache, n_new); | |
2920 | ||
b22d127a | 2921 | return ret; |
42288fe3 MG |
2922 | |
2923 | alloc_new: | |
4a8c7bb5 | 2924 | write_unlock(&sp->lock); |
42288fe3 MG |
2925 | ret = -ENOMEM; |
2926 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2927 | if (!n_new) | |
2928 | goto err_out; | |
2929 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2930 | if (!mpol_new) | |
2931 | goto err_out; | |
4ad09955 | 2932 | atomic_set(&mpol_new->refcnt, 1); |
42288fe3 | 2933 | goto restart; |
1da177e4 LT |
2934 | } |
2935 | ||
71fe804b LS |
2936 | /** |
2937 | * mpol_shared_policy_init - initialize shared policy for inode | |
2938 | * @sp: pointer to inode shared policy | |
2939 | * @mpol: struct mempolicy to install | |
2940 | * | |
2941 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2942 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2943 | * This must be released on exit. | |
4bfc4495 | 2944 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2945 | */ |
2946 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2947 | { | |
58568d2a MX |
2948 | int ret; |
2949 | ||
71fe804b | 2950 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2951 | rwlock_init(&sp->lock); |
71fe804b LS |
2952 | |
2953 | if (mpol) { | |
35ec8fa0 HD |
2954 | struct sp_node *sn; |
2955 | struct mempolicy *npol; | |
4bfc4495 | 2956 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2957 | |
4bfc4495 | 2958 | if (!scratch) |
5c0c1654 | 2959 | goto put_mpol; |
35ec8fa0 HD |
2960 | |
2961 | /* contextualize the tmpfs mount point mempolicy to this file */ | |
2962 | npol = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
2963 | if (IS_ERR(npol)) | |
0cae3457 | 2964 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2965 | |
2966 | task_lock(current); | |
35ec8fa0 | 2967 | ret = mpol_set_nodemask(npol, &mpol->w.user_nodemask, scratch); |
58568d2a | 2968 | task_unlock(current); |
15d77835 | 2969 | if (ret) |
35ec8fa0 HD |
2970 | goto put_npol; |
2971 | ||
2972 | /* alloc node covering entire file; adds ref to file's npol */ | |
2973 | sn = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, npol); | |
2974 | if (sn) | |
2975 | sp_insert(sp, sn); | |
2976 | put_npol: | |
2977 | mpol_put(npol); /* drop initial ref on file's npol */ | |
0cae3457 | 2978 | free_scratch: |
4bfc4495 | 2979 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2980 | put_mpol: |
2981 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2982 | } |
2983 | } | |
2984 | ||
c36f6e6d HD |
2985 | int mpol_set_shared_policy(struct shared_policy *sp, |
2986 | struct vm_area_struct *vma, struct mempolicy *pol) | |
1da177e4 LT |
2987 | { |
2988 | int err; | |
2989 | struct sp_node *new = NULL; | |
2990 | unsigned long sz = vma_pages(vma); | |
2991 | ||
c36f6e6d HD |
2992 | if (pol) { |
2993 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, pol); | |
1da177e4 LT |
2994 | if (!new) |
2995 | return -ENOMEM; | |
2996 | } | |
c36f6e6d | 2997 | err = shared_policy_replace(sp, vma->vm_pgoff, vma->vm_pgoff + sz, new); |
1da177e4 | 2998 | if (err && new) |
63f74ca2 | 2999 | sp_free(new); |
1da177e4 LT |
3000 | return err; |
3001 | } | |
3002 | ||
3003 | /* Free a backing policy store on inode delete. */ | |
c36f6e6d | 3004 | void mpol_free_shared_policy(struct shared_policy *sp) |
1da177e4 LT |
3005 | { |
3006 | struct sp_node *n; | |
3007 | struct rb_node *next; | |
3008 | ||
c36f6e6d | 3009 | if (!sp->root.rb_node) |
1da177e4 | 3010 | return; |
c36f6e6d HD |
3011 | write_lock(&sp->lock); |
3012 | next = rb_first(&sp->root); | |
1da177e4 LT |
3013 | while (next) { |
3014 | n = rb_entry(next, struct sp_node, nd); | |
3015 | next = rb_next(&n->nd); | |
c36f6e6d | 3016 | sp_delete(sp, n); |
1da177e4 | 3017 | } |
c36f6e6d | 3018 | write_unlock(&sp->lock); |
1da177e4 LT |
3019 | } |
3020 | ||
1a687c2e | 3021 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 3022 | static int __initdata numabalancing_override; |
1a687c2e MG |
3023 | |
3024 | static void __init check_numabalancing_enable(void) | |
3025 | { | |
3026 | bool numabalancing_default = false; | |
3027 | ||
3028 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
3029 | numabalancing_default = true; | |
3030 | ||
c297663c MG |
3031 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
3032 | if (numabalancing_override) | |
3033 | set_numabalancing_state(numabalancing_override == 1); | |
3034 | ||
b0dc2b9b | 3035 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 3036 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 3037 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
3038 | set_numabalancing_state(numabalancing_default); |
3039 | } | |
3040 | } | |
3041 | ||
3042 | static int __init setup_numabalancing(char *str) | |
3043 | { | |
3044 | int ret = 0; | |
3045 | if (!str) | |
3046 | goto out; | |
1a687c2e MG |
3047 | |
3048 | if (!strcmp(str, "enable")) { | |
c297663c | 3049 | numabalancing_override = 1; |
1a687c2e MG |
3050 | ret = 1; |
3051 | } else if (!strcmp(str, "disable")) { | |
c297663c | 3052 | numabalancing_override = -1; |
1a687c2e MG |
3053 | ret = 1; |
3054 | } | |
3055 | out: | |
3056 | if (!ret) | |
4a404bea | 3057 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
3058 | |
3059 | return ret; | |
3060 | } | |
3061 | __setup("numa_balancing=", setup_numabalancing); | |
3062 | #else | |
3063 | static inline void __init check_numabalancing_enable(void) | |
3064 | { | |
3065 | } | |
3066 | #endif /* CONFIG_NUMA_BALANCING */ | |
3067 | ||
1da177e4 LT |
3068 | void __init numa_policy_init(void) |
3069 | { | |
b71636e2 PM |
3070 | nodemask_t interleave_nodes; |
3071 | unsigned long largest = 0; | |
3072 | int nid, prefer = 0; | |
3073 | ||
1da177e4 LT |
3074 | policy_cache = kmem_cache_create("numa_policy", |
3075 | sizeof(struct mempolicy), | |
20c2df83 | 3076 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
3077 | |
3078 | sn_cache = kmem_cache_create("shared_policy_node", | |
3079 | sizeof(struct sp_node), | |
20c2df83 | 3080 | 0, SLAB_PANIC, NULL); |
1da177e4 | 3081 | |
5606e387 MG |
3082 | for_each_node(nid) { |
3083 | preferred_node_policy[nid] = (struct mempolicy) { | |
3084 | .refcnt = ATOMIC_INIT(1), | |
3085 | .mode = MPOL_PREFERRED, | |
3086 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
269fbe72 | 3087 | .nodes = nodemask_of_node(nid), |
5606e387 MG |
3088 | }; |
3089 | } | |
3090 | ||
b71636e2 PM |
3091 | /* |
3092 | * Set interleaving policy for system init. Interleaving is only | |
3093 | * enabled across suitably sized nodes (default is >= 16MB), or | |
3094 | * fall back to the largest node if they're all smaller. | |
3095 | */ | |
3096 | nodes_clear(interleave_nodes); | |
01f13bd6 | 3097 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
3098 | unsigned long total_pages = node_present_pages(nid); |
3099 | ||
3100 | /* Preserve the largest node */ | |
3101 | if (largest < total_pages) { | |
3102 | largest = total_pages; | |
3103 | prefer = nid; | |
3104 | } | |
3105 | ||
3106 | /* Interleave this node? */ | |
3107 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
3108 | node_set(nid, interleave_nodes); | |
3109 | } | |
3110 | ||
3111 | /* All too small, use the largest */ | |
3112 | if (unlikely(nodes_empty(interleave_nodes))) | |
3113 | node_set(prefer, interleave_nodes); | |
1da177e4 | 3114 | |
028fec41 | 3115 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 3116 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
3117 | |
3118 | check_numabalancing_enable(); | |
1da177e4 LT |
3119 | } |
3120 | ||
8bccd85f | 3121 | /* Reset policy of current process to default */ |
1da177e4 LT |
3122 | void numa_default_policy(void) |
3123 | { | |
028fec41 | 3124 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 3125 | } |
68860ec1 | 3126 | |
095f1fc4 LS |
3127 | /* |
3128 | * Parse and format mempolicy from/to strings | |
3129 | */ | |
345ace9c LS |
3130 | static const char * const policy_modes[] = |
3131 | { | |
3132 | [MPOL_DEFAULT] = "default", | |
3133 | [MPOL_PREFERRED] = "prefer", | |
3134 | [MPOL_BIND] = "bind", | |
3135 | [MPOL_INTERLEAVE] = "interleave", | |
fa3bea4e | 3136 | [MPOL_WEIGHTED_INTERLEAVE] = "weighted interleave", |
d3a71033 | 3137 | [MPOL_LOCAL] = "local", |
b27abacc | 3138 | [MPOL_PREFERRED_MANY] = "prefer (many)", |
345ace9c | 3139 | }; |
1a75a6c8 | 3140 | |
095f1fc4 LS |
3141 | #ifdef CONFIG_TMPFS |
3142 | /** | |
f2a07f40 | 3143 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 3144 | * @str: string containing mempolicy to parse |
71fe804b | 3145 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
3146 | * |
3147 | * Format of input: | |
3148 | * <mode>[=<flags>][:<nodelist>] | |
3149 | * | |
dad5b023 | 3150 | * Return: %0 on success, else %1 |
095f1fc4 | 3151 | */ |
a7a88b23 | 3152 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 3153 | { |
71fe804b | 3154 | struct mempolicy *new = NULL; |
f2a07f40 | 3155 | unsigned short mode_flags; |
71fe804b | 3156 | nodemask_t nodes; |
095f1fc4 LS |
3157 | char *nodelist = strchr(str, ':'); |
3158 | char *flags = strchr(str, '='); | |
dedf2c73 | 3159 | int err = 1, mode; |
095f1fc4 | 3160 | |
c7a91bc7 DC |
3161 | if (flags) |
3162 | *flags++ = '\0'; /* terminate mode string */ | |
3163 | ||
095f1fc4 LS |
3164 | if (nodelist) { |
3165 | /* NUL-terminate mode or flags string */ | |
3166 | *nodelist++ = '\0'; | |
71fe804b | 3167 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 3168 | goto out; |
01f13bd6 | 3169 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 3170 | goto out; |
71fe804b LS |
3171 | } else |
3172 | nodes_clear(nodes); | |
3173 | ||
dedf2c73 | 3174 | mode = match_string(policy_modes, MPOL_MAX, str); |
3175 | if (mode < 0) | |
095f1fc4 LS |
3176 | goto out; |
3177 | ||
71fe804b | 3178 | switch (mode) { |
095f1fc4 | 3179 | case MPOL_PREFERRED: |
71fe804b | 3180 | /* |
aa9f7d51 RD |
3181 | * Insist on a nodelist of one node only, although later |
3182 | * we use first_node(nodes) to grab a single node, so here | |
3183 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 3184 | */ |
095f1fc4 LS |
3185 | if (nodelist) { |
3186 | char *rest = nodelist; | |
3187 | while (isdigit(*rest)) | |
3188 | rest++; | |
926f2ae0 KM |
3189 | if (*rest) |
3190 | goto out; | |
aa9f7d51 RD |
3191 | if (nodes_empty(nodes)) |
3192 | goto out; | |
095f1fc4 LS |
3193 | } |
3194 | break; | |
095f1fc4 | 3195 | case MPOL_INTERLEAVE: |
fa3bea4e | 3196 | case MPOL_WEIGHTED_INTERLEAVE: |
095f1fc4 LS |
3197 | /* |
3198 | * Default to online nodes with memory if no nodelist | |
3199 | */ | |
3200 | if (!nodelist) | |
01f13bd6 | 3201 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 3202 | break; |
71fe804b | 3203 | case MPOL_LOCAL: |
3f226aa1 | 3204 | /* |
71fe804b | 3205 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 3206 | */ |
71fe804b | 3207 | if (nodelist) |
3f226aa1 | 3208 | goto out; |
3f226aa1 | 3209 | break; |
413b43de RT |
3210 | case MPOL_DEFAULT: |
3211 | /* | |
3212 | * Insist on a empty nodelist | |
3213 | */ | |
3214 | if (!nodelist) | |
3215 | err = 0; | |
3216 | goto out; | |
b27abacc | 3217 | case MPOL_PREFERRED_MANY: |
d69b2e63 KM |
3218 | case MPOL_BIND: |
3219 | /* | |
3220 | * Insist on a nodelist | |
3221 | */ | |
3222 | if (!nodelist) | |
3223 | goto out; | |
095f1fc4 LS |
3224 | } |
3225 | ||
71fe804b | 3226 | mode_flags = 0; |
095f1fc4 LS |
3227 | if (flags) { |
3228 | /* | |
3229 | * Currently, we only support two mutually exclusive | |
3230 | * mode flags. | |
3231 | */ | |
3232 | if (!strcmp(flags, "static")) | |
71fe804b | 3233 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 3234 | else if (!strcmp(flags, "relative")) |
71fe804b | 3235 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 3236 | else |
926f2ae0 | 3237 | goto out; |
095f1fc4 | 3238 | } |
71fe804b LS |
3239 | |
3240 | new = mpol_new(mode, mode_flags, &nodes); | |
3241 | if (IS_ERR(new)) | |
926f2ae0 KM |
3242 | goto out; |
3243 | ||
f2a07f40 HD |
3244 | /* |
3245 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
3246 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
3247 | */ | |
269fbe72 BW |
3248 | if (mode != MPOL_PREFERRED) { |
3249 | new->nodes = nodes; | |
3250 | } else if (nodelist) { | |
3251 | nodes_clear(new->nodes); | |
3252 | node_set(first_node(nodes), new->nodes); | |
3253 | } else { | |
7858d7bc | 3254 | new->mode = MPOL_LOCAL; |
269fbe72 | 3255 | } |
f2a07f40 HD |
3256 | |
3257 | /* | |
3258 | * Save nodes for contextualization: this will be used to "clone" | |
3259 | * the mempolicy in a specific context [cpuset] at a later time. | |
3260 | */ | |
3261 | new->w.user_nodemask = nodes; | |
3262 | ||
926f2ae0 | 3263 | err = 0; |
71fe804b | 3264 | |
095f1fc4 LS |
3265 | out: |
3266 | /* Restore string for error message */ | |
3267 | if (nodelist) | |
3268 | *--nodelist = ':'; | |
3269 | if (flags) | |
3270 | *--flags = '='; | |
71fe804b LS |
3271 | if (!err) |
3272 | *mpol = new; | |
095f1fc4 LS |
3273 | return err; |
3274 | } | |
3275 | #endif /* CONFIG_TMPFS */ | |
3276 | ||
71fe804b LS |
3277 | /** |
3278 | * mpol_to_str - format a mempolicy structure for printing | |
3279 | * @buffer: to contain formatted mempolicy string | |
3280 | * @maxlen: length of @buffer | |
3281 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3282 | * |
948927ee DR |
3283 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3284 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3285 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3286 | */ |
948927ee | 3287 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3288 | { |
3289 | char *p = buffer; | |
948927ee DR |
3290 | nodemask_t nodes = NODE_MASK_NONE; |
3291 | unsigned short mode = MPOL_DEFAULT; | |
3292 | unsigned short flags = 0; | |
2291990a | 3293 | |
8790c71a | 3294 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3295 | mode = pol->mode; |
948927ee DR |
3296 | flags = pol->flags; |
3297 | } | |
bea904d5 | 3298 | |
1a75a6c8 CL |
3299 | switch (mode) { |
3300 | case MPOL_DEFAULT: | |
7858d7bc | 3301 | case MPOL_LOCAL: |
1a75a6c8 | 3302 | break; |
1a75a6c8 | 3303 | case MPOL_PREFERRED: |
b27abacc | 3304 | case MPOL_PREFERRED_MANY: |
1a75a6c8 | 3305 | case MPOL_BIND: |
1a75a6c8 | 3306 | case MPOL_INTERLEAVE: |
fa3bea4e | 3307 | case MPOL_WEIGHTED_INTERLEAVE: |
269fbe72 | 3308 | nodes = pol->nodes; |
1a75a6c8 | 3309 | break; |
1a75a6c8 | 3310 | default: |
948927ee DR |
3311 | WARN_ON_ONCE(1); |
3312 | snprintf(p, maxlen, "unknown"); | |
3313 | return; | |
1a75a6c8 CL |
3314 | } |
3315 | ||
b7a9f420 | 3316 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3317 | |
fc36b8d3 | 3318 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3319 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3320 | |
2291990a LS |
3321 | /* |
3322 | * Currently, the only defined flags are mutually exclusive | |
3323 | */ | |
f5b087b5 | 3324 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3325 | p += snprintf(p, buffer + maxlen - p, "static"); |
3326 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3327 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3328 | } |
3329 | ||
9e763e0f TH |
3330 | if (!nodes_empty(nodes)) |
3331 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3332 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3333 | } |
dce41f5a RK |
3334 | |
3335 | #ifdef CONFIG_SYSFS | |
3336 | struct iw_node_attr { | |
3337 | struct kobj_attribute kobj_attr; | |
3338 | int nid; | |
3339 | }; | |
3340 | ||
3341 | static ssize_t node_show(struct kobject *kobj, struct kobj_attribute *attr, | |
3342 | char *buf) | |
3343 | { | |
3344 | struct iw_node_attr *node_attr; | |
3345 | u8 weight; | |
3346 | ||
3347 | node_attr = container_of(attr, struct iw_node_attr, kobj_attr); | |
3348 | weight = get_il_weight(node_attr->nid); | |
3349 | return sysfs_emit(buf, "%d\n", weight); | |
3350 | } | |
3351 | ||
3352 | static ssize_t node_store(struct kobject *kobj, struct kobj_attribute *attr, | |
3353 | const char *buf, size_t count) | |
3354 | { | |
3355 | struct iw_node_attr *node_attr; | |
3356 | u8 *new; | |
3357 | u8 *old; | |
3358 | u8 weight = 0; | |
3359 | ||
3360 | node_attr = container_of(attr, struct iw_node_attr, kobj_attr); | |
3361 | if (count == 0 || sysfs_streq(buf, "")) | |
3362 | weight = 0; | |
3363 | else if (kstrtou8(buf, 0, &weight)) | |
3364 | return -EINVAL; | |
3365 | ||
3366 | new = kzalloc(nr_node_ids, GFP_KERNEL); | |
3367 | if (!new) | |
3368 | return -ENOMEM; | |
3369 | ||
3370 | mutex_lock(&iw_table_lock); | |
3371 | old = rcu_dereference_protected(iw_table, | |
3372 | lockdep_is_held(&iw_table_lock)); | |
3373 | if (old) | |
3374 | memcpy(new, old, nr_node_ids); | |
3375 | new[node_attr->nid] = weight; | |
3376 | rcu_assign_pointer(iw_table, new); | |
3377 | mutex_unlock(&iw_table_lock); | |
3378 | synchronize_rcu(); | |
3379 | kfree(old); | |
3380 | return count; | |
3381 | } | |
3382 | ||
3383 | static struct iw_node_attr **node_attrs; | |
3384 | ||
3385 | static void sysfs_wi_node_release(struct iw_node_attr *node_attr, | |
3386 | struct kobject *parent) | |
3387 | { | |
3388 | if (!node_attr) | |
3389 | return; | |
3390 | sysfs_remove_file(parent, &node_attr->kobj_attr.attr); | |
3391 | kfree(node_attr->kobj_attr.attr.name); | |
3392 | kfree(node_attr); | |
3393 | } | |
3394 | ||
3395 | static void sysfs_wi_release(struct kobject *wi_kobj) | |
3396 | { | |
3397 | int i; | |
3398 | ||
3399 | for (i = 0; i < nr_node_ids; i++) | |
3400 | sysfs_wi_node_release(node_attrs[i], wi_kobj); | |
3401 | kobject_put(wi_kobj); | |
3402 | } | |
3403 | ||
3404 | static const struct kobj_type wi_ktype = { | |
3405 | .sysfs_ops = &kobj_sysfs_ops, | |
3406 | .release = sysfs_wi_release, | |
3407 | }; | |
3408 | ||
3409 | static int add_weight_node(int nid, struct kobject *wi_kobj) | |
3410 | { | |
3411 | struct iw_node_attr *node_attr; | |
3412 | char *name; | |
3413 | ||
3414 | node_attr = kzalloc(sizeof(*node_attr), GFP_KERNEL); | |
3415 | if (!node_attr) | |
3416 | return -ENOMEM; | |
3417 | ||
3418 | name = kasprintf(GFP_KERNEL, "node%d", nid); | |
3419 | if (!name) { | |
3420 | kfree(node_attr); | |
3421 | return -ENOMEM; | |
3422 | } | |
3423 | ||
3424 | sysfs_attr_init(&node_attr->kobj_attr.attr); | |
3425 | node_attr->kobj_attr.attr.name = name; | |
3426 | node_attr->kobj_attr.attr.mode = 0644; | |
3427 | node_attr->kobj_attr.show = node_show; | |
3428 | node_attr->kobj_attr.store = node_store; | |
3429 | node_attr->nid = nid; | |
3430 | ||
3431 | if (sysfs_create_file(wi_kobj, &node_attr->kobj_attr.attr)) { | |
3432 | kfree(node_attr->kobj_attr.attr.name); | |
3433 | kfree(node_attr); | |
3434 | pr_err("failed to add attribute to weighted_interleave\n"); | |
3435 | return -ENOMEM; | |
3436 | } | |
3437 | ||
3438 | node_attrs[nid] = node_attr; | |
3439 | return 0; | |
3440 | } | |
3441 | ||
3442 | static int add_weighted_interleave_group(struct kobject *root_kobj) | |
3443 | { | |
3444 | struct kobject *wi_kobj; | |
3445 | int nid, err; | |
3446 | ||
3447 | wi_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); | |
3448 | if (!wi_kobj) | |
3449 | return -ENOMEM; | |
3450 | ||
3451 | err = kobject_init_and_add(wi_kobj, &wi_ktype, root_kobj, | |
3452 | "weighted_interleave"); | |
3453 | if (err) { | |
3454 | kfree(wi_kobj); | |
3455 | return err; | |
3456 | } | |
3457 | ||
3458 | for_each_node_state(nid, N_POSSIBLE) { | |
3459 | err = add_weight_node(nid, wi_kobj); | |
3460 | if (err) { | |
3461 | pr_err("failed to add sysfs [node%d]\n", nid); | |
3462 | break; | |
3463 | } | |
3464 | } | |
3465 | if (err) | |
3466 | kobject_put(wi_kobj); | |
3467 | return 0; | |
3468 | } | |
3469 | ||
3470 | static void mempolicy_kobj_release(struct kobject *kobj) | |
3471 | { | |
3472 | u8 *old; | |
3473 | ||
3474 | mutex_lock(&iw_table_lock); | |
3475 | old = rcu_dereference_protected(iw_table, | |
3476 | lockdep_is_held(&iw_table_lock)); | |
3477 | rcu_assign_pointer(iw_table, NULL); | |
3478 | mutex_unlock(&iw_table_lock); | |
3479 | synchronize_rcu(); | |
3480 | kfree(old); | |
3481 | kfree(node_attrs); | |
3482 | kfree(kobj); | |
3483 | } | |
3484 | ||
3485 | static const struct kobj_type mempolicy_ktype = { | |
3486 | .release = mempolicy_kobj_release | |
3487 | }; | |
3488 | ||
3489 | static int __init mempolicy_sysfs_init(void) | |
3490 | { | |
3491 | int err; | |
3492 | static struct kobject *mempolicy_kobj; | |
3493 | ||
3494 | mempolicy_kobj = kzalloc(sizeof(*mempolicy_kobj), GFP_KERNEL); | |
3495 | if (!mempolicy_kobj) { | |
3496 | err = -ENOMEM; | |
3497 | goto err_out; | |
3498 | } | |
3499 | ||
3500 | node_attrs = kcalloc(nr_node_ids, sizeof(struct iw_node_attr *), | |
3501 | GFP_KERNEL); | |
3502 | if (!node_attrs) { | |
3503 | err = -ENOMEM; | |
3504 | goto mempol_out; | |
3505 | } | |
3506 | ||
3507 | err = kobject_init_and_add(mempolicy_kobj, &mempolicy_ktype, mm_kobj, | |
3508 | "mempolicy"); | |
3509 | if (err) | |
3510 | goto node_out; | |
3511 | ||
3512 | err = add_weighted_interleave_group(mempolicy_kobj); | |
3513 | if (err) { | |
3514 | pr_err("mempolicy sysfs structure failed to initialize\n"); | |
3515 | kobject_put(mempolicy_kobj); | |
3516 | return err; | |
3517 | } | |
3518 | ||
3519 | return err; | |
3520 | node_out: | |
3521 | kfree(node_attrs); | |
3522 | mempol_out: | |
3523 | kfree(mempolicy_kobj); | |
3524 | err_out: | |
3525 | pr_err("failed to add mempolicy kobject to the system\n"); | |
3526 | return err; | |
3527 | } | |
3528 | ||
3529 | late_initcall(mempolicy_sysfs_init); | |
3530 | #endif /* CONFIG_SYSFS */ |