Commit | Line | Data |
---|---|---|
46aeb7e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Simple NUMA memory policy for the Linux kernel. | |
4 | * | |
5 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 6 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
7 | * |
8 | * NUMA policy allows the user to give hints in which node(s) memory should | |
9 | * be allocated. | |
10 | * | |
11 | * Support four policies per VMA and per process: | |
12 | * | |
13 | * The VMA policy has priority over the process policy for a page fault. | |
14 | * | |
15 | * interleave Allocate memory interleaved over a set of nodes, | |
16 | * with normal fallback if it fails. | |
17 | * For VMA based allocations this interleaves based on the | |
18 | * offset into the backing object or offset into the mapping | |
19 | * for anonymous memory. For process policy an process counter | |
20 | * is used. | |
8bccd85f | 21 | * |
1da177e4 LT |
22 | * bind Only allocate memory on a specific set of nodes, |
23 | * no fallback. | |
8bccd85f CL |
24 | * FIXME: memory is allocated starting with the first node |
25 | * to the last. It would be better if bind would truly restrict | |
26 | * the allocation to memory nodes instead | |
27 | * | |
1da177e4 | 28 | * preferred Try a specific node first before normal fallback. |
00ef2d2f | 29 | * As a special case NUMA_NO_NODE here means do the allocation |
1da177e4 LT |
30 | * on the local CPU. This is normally identical to default, |
31 | * but useful to set in a VMA when you have a non default | |
32 | * process policy. | |
8bccd85f | 33 | * |
b27abacc DH |
34 | * preferred many Try a set of nodes first before normal fallback. This is |
35 | * similar to preferred without the special case. | |
36 | * | |
1da177e4 LT |
37 | * default Allocate on the local node first, or when on a VMA |
38 | * use the process policy. This is what Linux always did | |
39 | * in a NUMA aware kernel and still does by, ahem, default. | |
40 | * | |
41 | * The process policy is applied for most non interrupt memory allocations | |
42 | * in that process' context. Interrupts ignore the policies and always | |
43 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
44 | * allocations for a VMA in the VM. | |
45 | * | |
46 | * Currently there are a few corner cases in swapping where the policy | |
47 | * is not applied, but the majority should be handled. When process policy | |
48 | * is used it is not remembered over swap outs/swap ins. | |
49 | * | |
50 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
51 | * requesting a lower zone just use default policy. This implies that | |
52 | * on systems with highmem kernel lowmem allocation don't get policied. | |
53 | * Same with GFP_DMA allocations. | |
54 | * | |
55 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between | |
56 | * all users and remembered even when nobody has memory mapped. | |
57 | */ | |
58 | ||
59 | /* Notebook: | |
60 | fix mmap readahead to honour policy and enable policy for any page cache | |
61 | object | |
62 | statistics for bigpages | |
63 | global policy for page cache? currently it uses process policy. Requires | |
64 | first item above. | |
65 | handle mremap for shared memory (currently ignored for the policy) | |
66 | grows down? | |
67 | make bind policy root only? It can trigger oom much faster and the | |
68 | kernel is not always grateful with that. | |
1da177e4 LT |
69 | */ |
70 | ||
b1de0d13 MH |
71 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
72 | ||
1da177e4 | 73 | #include <linux/mempolicy.h> |
a520110e | 74 | #include <linux/pagewalk.h> |
1da177e4 LT |
75 | #include <linux/highmem.h> |
76 | #include <linux/hugetlb.h> | |
77 | #include <linux/kernel.h> | |
78 | #include <linux/sched.h> | |
6e84f315 | 79 | #include <linux/sched/mm.h> |
6a3827d7 | 80 | #include <linux/sched/numa_balancing.h> |
f719ff9b | 81 | #include <linux/sched/task.h> |
1da177e4 LT |
82 | #include <linux/nodemask.h> |
83 | #include <linux/cpuset.h> | |
1da177e4 LT |
84 | #include <linux/slab.h> |
85 | #include <linux/string.h> | |
b95f1b31 | 86 | #include <linux/export.h> |
b488893a | 87 | #include <linux/nsproxy.h> |
1da177e4 LT |
88 | #include <linux/interrupt.h> |
89 | #include <linux/init.h> | |
90 | #include <linux/compat.h> | |
31367466 | 91 | #include <linux/ptrace.h> |
dc9aa5b9 | 92 | #include <linux/swap.h> |
1a75a6c8 CL |
93 | #include <linux/seq_file.h> |
94 | #include <linux/proc_fs.h> | |
b20a3503 | 95 | #include <linux/migrate.h> |
62b61f61 | 96 | #include <linux/ksm.h> |
95a402c3 | 97 | #include <linux/rmap.h> |
86c3a764 | 98 | #include <linux/security.h> |
dbcb0f19 | 99 | #include <linux/syscalls.h> |
095f1fc4 | 100 | #include <linux/ctype.h> |
6d9c285a | 101 | #include <linux/mm_inline.h> |
b24f53a0 | 102 | #include <linux/mmu_notifier.h> |
b1de0d13 | 103 | #include <linux/printk.h> |
c8633798 | 104 | #include <linux/swapops.h> |
dc9aa5b9 | 105 | |
1da177e4 | 106 | #include <asm/tlbflush.h> |
4a18419f | 107 | #include <asm/tlb.h> |
7c0f6ba6 | 108 | #include <linux/uaccess.h> |
1da177e4 | 109 | |
62695a84 NP |
110 | #include "internal.h" |
111 | ||
38e35860 | 112 | /* Internal flags */ |
dc9aa5b9 | 113 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 114 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 115 | |
fcc234f8 PE |
116 | static struct kmem_cache *policy_cache; |
117 | static struct kmem_cache *sn_cache; | |
1da177e4 | 118 | |
1da177e4 LT |
119 | /* Highest zone. An specific allocation for a zone below that is not |
120 | policied. */ | |
6267276f | 121 | enum zone_type policy_zone = 0; |
1da177e4 | 122 | |
bea904d5 LS |
123 | /* |
124 | * run-time system-wide default policy => local allocation | |
125 | */ | |
e754d79d | 126 | static struct mempolicy default_policy = { |
1da177e4 | 127 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
7858d7bc | 128 | .mode = MPOL_LOCAL, |
1da177e4 LT |
129 | }; |
130 | ||
5606e387 MG |
131 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
132 | ||
b2ca916c DW |
133 | /** |
134 | * numa_map_to_online_node - Find closest online node | |
f6e92f40 | 135 | * @node: Node id to start the search |
b2ca916c DW |
136 | * |
137 | * Lookup the next closest node by distance if @nid is not online. | |
dad5b023 RD |
138 | * |
139 | * Return: this @node if it is online, otherwise the closest node by distance | |
b2ca916c DW |
140 | */ |
141 | int numa_map_to_online_node(int node) | |
142 | { | |
4fcbe96e | 143 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 144 | |
4fcbe96e DW |
145 | if (node == NUMA_NO_NODE || node_online(node)) |
146 | return node; | |
b2ca916c DW |
147 | |
148 | min_node = node; | |
4fcbe96e DW |
149 | for_each_online_node(n) { |
150 | dist = node_distance(node, n); | |
151 | if (dist < min_dist) { | |
152 | min_dist = dist; | |
153 | min_node = n; | |
b2ca916c DW |
154 | } |
155 | } | |
156 | ||
157 | return min_node; | |
158 | } | |
159 | EXPORT_SYMBOL_GPL(numa_map_to_online_node); | |
160 | ||
74d2c3a0 | 161 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
162 | { |
163 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 164 | int node; |
5606e387 | 165 | |
f15ca78e ON |
166 | if (pol) |
167 | return pol; | |
5606e387 | 168 | |
f15ca78e ON |
169 | node = numa_node_id(); |
170 | if (node != NUMA_NO_NODE) { | |
171 | pol = &preferred_node_policy[node]; | |
172 | /* preferred_node_policy is not initialised early in boot */ | |
173 | if (pol->mode) | |
174 | return pol; | |
5606e387 MG |
175 | } |
176 | ||
f15ca78e | 177 | return &default_policy; |
5606e387 MG |
178 | } |
179 | ||
37012946 DR |
180 | static const struct mempolicy_operations { |
181 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 182 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
183 | } mpol_ops[MPOL_MAX]; |
184 | ||
f5b087b5 DR |
185 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
186 | { | |
6d556294 | 187 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
188 | } |
189 | ||
190 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
191 | const nodemask_t *rel) | |
192 | { | |
193 | nodemask_t tmp; | |
194 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
195 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
196 | } |
197 | ||
be897d48 | 198 | static int mpol_new_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
199 | { |
200 | if (nodes_empty(*nodes)) | |
201 | return -EINVAL; | |
269fbe72 | 202 | pol->nodes = *nodes; |
37012946 DR |
203 | return 0; |
204 | } | |
205 | ||
206 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
207 | { | |
7858d7bc FT |
208 | if (nodes_empty(*nodes)) |
209 | return -EINVAL; | |
269fbe72 BW |
210 | |
211 | nodes_clear(pol->nodes); | |
212 | node_set(first_node(*nodes), pol->nodes); | |
37012946 DR |
213 | return 0; |
214 | } | |
215 | ||
58568d2a MX |
216 | /* |
217 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
218 | * any, for the new policy. mpol_new() has already validated the nodes | |
7858d7bc | 219 | * parameter with respect to the policy mode and flags. |
58568d2a MX |
220 | * |
221 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
c1e8d7c6 | 222 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 223 | */ |
4bfc4495 KH |
224 | static int mpol_set_nodemask(struct mempolicy *pol, |
225 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 226 | { |
58568d2a MX |
227 | int ret; |
228 | ||
7858d7bc FT |
229 | /* |
230 | * Default (pol==NULL) resp. local memory policies are not a | |
231 | * subject of any remapping. They also do not need any special | |
232 | * constructor. | |
233 | */ | |
234 | if (!pol || pol->mode == MPOL_LOCAL) | |
58568d2a | 235 | return 0; |
7858d7bc | 236 | |
01f13bd6 | 237 | /* Check N_MEMORY */ |
4bfc4495 | 238 | nodes_and(nsc->mask1, |
01f13bd6 | 239 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
240 | |
241 | VM_BUG_ON(!nodes); | |
4bfc4495 | 242 | |
7858d7bc FT |
243 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
244 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); | |
245 | else | |
246 | nodes_and(nsc->mask2, *nodes, nsc->mask1); | |
58568d2a | 247 | |
7858d7bc FT |
248 | if (mpol_store_user_nodemask(pol)) |
249 | pol->w.user_nodemask = *nodes; | |
4bfc4495 | 250 | else |
7858d7bc FT |
251 | pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed; |
252 | ||
253 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
58568d2a MX |
254 | return ret; |
255 | } | |
256 | ||
257 | /* | |
258 | * This function just creates a new policy, does some check and simple | |
259 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
260 | */ | |
028fec41 DR |
261 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
262 | nodemask_t *nodes) | |
1da177e4 LT |
263 | { |
264 | struct mempolicy *policy; | |
265 | ||
028fec41 | 266 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 267 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 268 | |
3e1f0645 DR |
269 | if (mode == MPOL_DEFAULT) { |
270 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 271 | return ERR_PTR(-EINVAL); |
d3a71033 | 272 | return NULL; |
37012946 | 273 | } |
3e1f0645 DR |
274 | VM_BUG_ON(!nodes); |
275 | ||
276 | /* | |
277 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
278 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
279 | * All other modes require a valid pointer to a non-empty nodemask. | |
280 | */ | |
281 | if (mode == MPOL_PREFERRED) { | |
282 | if (nodes_empty(*nodes)) { | |
283 | if (((flags & MPOL_F_STATIC_NODES) || | |
284 | (flags & MPOL_F_RELATIVE_NODES))) | |
285 | return ERR_PTR(-EINVAL); | |
7858d7bc FT |
286 | |
287 | mode = MPOL_LOCAL; | |
3e1f0645 | 288 | } |
479e2802 | 289 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
290 | if (!nodes_empty(*nodes) || |
291 | (flags & MPOL_F_STATIC_NODES) || | |
292 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 | 293 | return ERR_PTR(-EINVAL); |
3e1f0645 DR |
294 | } else if (nodes_empty(*nodes)) |
295 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
296 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
297 | if (!policy) | |
298 | return ERR_PTR(-ENOMEM); | |
299 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 300 | policy->mode = mode; |
3e1f0645 | 301 | policy->flags = flags; |
c6018b4b | 302 | policy->home_node = NUMA_NO_NODE; |
37012946 | 303 | |
1da177e4 | 304 | return policy; |
37012946 DR |
305 | } |
306 | ||
52cd3b07 LS |
307 | /* Slow path of a mpol destructor. */ |
308 | void __mpol_put(struct mempolicy *p) | |
309 | { | |
310 | if (!atomic_dec_and_test(&p->refcnt)) | |
311 | return; | |
52cd3b07 LS |
312 | kmem_cache_free(policy_cache, p); |
313 | } | |
314 | ||
213980c0 | 315 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
316 | { |
317 | } | |
318 | ||
213980c0 | 319 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
320 | { |
321 | nodemask_t tmp; | |
322 | ||
323 | if (pol->flags & MPOL_F_STATIC_NODES) | |
324 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
325 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
326 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
327 | else { | |
269fbe72 | 328 | nodes_remap(tmp, pol->nodes, pol->w.cpuset_mems_allowed, |
213980c0 | 329 | *nodes); |
29b190fa | 330 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 331 | } |
f5b087b5 | 332 | |
708c1bbc MX |
333 | if (nodes_empty(tmp)) |
334 | tmp = *nodes; | |
335 | ||
269fbe72 | 336 | pol->nodes = tmp; |
37012946 DR |
337 | } |
338 | ||
339 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 340 | const nodemask_t *nodes) |
37012946 | 341 | { |
7858d7bc | 342 | pol->w.cpuset_mems_allowed = *nodes; |
1da177e4 LT |
343 | } |
344 | ||
708c1bbc MX |
345 | /* |
346 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
347 | * | |
c1e8d7c6 | 348 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
349 | * policies are protected by task->mems_allowed_seq to prevent a premature |
350 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 351 | */ |
213980c0 | 352 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 353 | { |
018160ad | 354 | if (!pol || pol->mode == MPOL_LOCAL) |
1d0d2680 | 355 | return; |
7858d7bc | 356 | if (!mpol_store_user_nodemask(pol) && |
1d0d2680 DR |
357 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
358 | return; | |
708c1bbc | 359 | |
213980c0 | 360 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
361 | } |
362 | ||
363 | /* | |
364 | * Wrapper for mpol_rebind_policy() that just requires task | |
365 | * pointer, and updates task mempolicy. | |
58568d2a MX |
366 | * |
367 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
368 | */ |
369 | ||
213980c0 | 370 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 371 | { |
213980c0 | 372 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
373 | } |
374 | ||
375 | /* | |
376 | * Rebind each vma in mm to new nodemask. | |
377 | * | |
c1e8d7c6 | 378 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 DR |
379 | */ |
380 | ||
381 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
382 | { | |
383 | struct vm_area_struct *vma; | |
66850be5 | 384 | VMA_ITERATOR(vmi, mm, 0); |
1d0d2680 | 385 | |
d8ed45c5 | 386 | mmap_write_lock(mm); |
66850be5 | 387 | for_each_vma(vmi, vma) |
213980c0 | 388 | mpol_rebind_policy(vma->vm_policy, new); |
d8ed45c5 | 389 | mmap_write_unlock(mm); |
1d0d2680 DR |
390 | } |
391 | ||
37012946 DR |
392 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
393 | [MPOL_DEFAULT] = { | |
394 | .rebind = mpol_rebind_default, | |
395 | }, | |
396 | [MPOL_INTERLEAVE] = { | |
be897d48 | 397 | .create = mpol_new_nodemask, |
37012946 DR |
398 | .rebind = mpol_rebind_nodemask, |
399 | }, | |
400 | [MPOL_PREFERRED] = { | |
401 | .create = mpol_new_preferred, | |
402 | .rebind = mpol_rebind_preferred, | |
403 | }, | |
404 | [MPOL_BIND] = { | |
be897d48 | 405 | .create = mpol_new_nodemask, |
37012946 DR |
406 | .rebind = mpol_rebind_nodemask, |
407 | }, | |
7858d7bc FT |
408 | [MPOL_LOCAL] = { |
409 | .rebind = mpol_rebind_default, | |
410 | }, | |
b27abacc | 411 | [MPOL_PREFERRED_MANY] = { |
be897d48 | 412 | .create = mpol_new_nodemask, |
b27abacc DH |
413 | .rebind = mpol_rebind_preferred, |
414 | }, | |
37012946 DR |
415 | }; |
416 | ||
a53190a4 | 417 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 418 | unsigned long flags); |
1a75a6c8 | 419 | |
6f4576e3 NH |
420 | struct queue_pages { |
421 | struct list_head *pagelist; | |
422 | unsigned long flags; | |
423 | nodemask_t *nmask; | |
f18da660 LX |
424 | unsigned long start; |
425 | unsigned long end; | |
426 | struct vm_area_struct *first; | |
6f4576e3 NH |
427 | }; |
428 | ||
88aaa2a1 NH |
429 | /* |
430 | * Check if the page's nid is in qp->nmask. | |
431 | * | |
432 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
433 | * in the invert of qp->nmask. | |
434 | */ | |
435 | static inline bool queue_pages_required(struct page *page, | |
436 | struct queue_pages *qp) | |
437 | { | |
438 | int nid = page_to_nid(page); | |
439 | unsigned long flags = qp->flags; | |
440 | ||
441 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
442 | } | |
443 | ||
a7f40cfe | 444 | /* |
bc78b5ed | 445 | * queue_pages_pmd() has three possible return values: |
e5947d23 YS |
446 | * 0 - pages are placed on the right node or queued successfully, or |
447 | * special page is met, i.e. huge zero page. | |
d8835445 YS |
448 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
449 | * specified. | |
d8835445 YS |
450 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an |
451 | * existing page was already on a node that does not follow the | |
452 | * policy. | |
a7f40cfe | 453 | */ |
c8633798 NH |
454 | static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
455 | unsigned long end, struct mm_walk *walk) | |
959a7e13 | 456 | __releases(ptl) |
c8633798 NH |
457 | { |
458 | int ret = 0; | |
459 | struct page *page; | |
460 | struct queue_pages *qp = walk->private; | |
461 | unsigned long flags; | |
462 | ||
463 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 464 | ret = -EIO; |
c8633798 NH |
465 | goto unlock; |
466 | } | |
467 | page = pmd_page(*pmd); | |
468 | if (is_huge_zero_page(page)) { | |
e5947d23 | 469 | walk->action = ACTION_CONTINUE; |
6d97cf88 | 470 | goto unlock; |
c8633798 | 471 | } |
d8835445 | 472 | if (!queue_pages_required(page, qp)) |
c8633798 | 473 | goto unlock; |
c8633798 | 474 | |
c8633798 NH |
475 | flags = qp->flags; |
476 | /* go to thp migration */ | |
a7f40cfe | 477 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 YS |
478 | if (!vma_migratable(walk->vma) || |
479 | migrate_page_add(page, qp->pagelist, flags)) { | |
d8835445 | 480 | ret = 1; |
a7f40cfe YS |
481 | goto unlock; |
482 | } | |
a7f40cfe YS |
483 | } else |
484 | ret = -EIO; | |
c8633798 NH |
485 | unlock: |
486 | spin_unlock(ptl); | |
c8633798 NH |
487 | return ret; |
488 | } | |
489 | ||
98094945 NH |
490 | /* |
491 | * Scan through pages checking if pages follow certain conditions, | |
492 | * and move them to the pagelist if they do. | |
d8835445 YS |
493 | * |
494 | * queue_pages_pte_range() has three possible return values: | |
e5947d23 YS |
495 | * 0 - pages are placed on the right node or queued successfully, or |
496 | * special page is met, i.e. zero page. | |
d8835445 YS |
497 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
498 | * specified. | |
499 | * -EIO - only MPOL_MF_STRICT was specified and an existing page was already | |
500 | * on a node that does not follow the policy. | |
98094945 | 501 | */ |
6f4576e3 NH |
502 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
503 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 504 | { |
6f4576e3 NH |
505 | struct vm_area_struct *vma = walk->vma; |
506 | struct page *page; | |
507 | struct queue_pages *qp = walk->private; | |
508 | unsigned long flags = qp->flags; | |
d8835445 | 509 | bool has_unmovable = false; |
3f088420 | 510 | pte_t *pte, *mapped_pte; |
705e87c0 | 511 | spinlock_t *ptl; |
941150a3 | 512 | |
c8633798 | 513 | ptl = pmd_trans_huge_lock(pmd, vma); |
bc78b5ed ML |
514 | if (ptl) |
515 | return queue_pages_pmd(pmd, ptl, addr, end, walk); | |
91612e0d | 516 | |
337d9abf NH |
517 | if (pmd_trans_unstable(pmd)) |
518 | return 0; | |
94723aaf | 519 | |
3f088420 | 520 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
6f4576e3 | 521 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
91612e0d | 522 | if (!pte_present(*pte)) |
1da177e4 | 523 | continue; |
6aab341e | 524 | page = vm_normal_page(vma, addr, *pte); |
3218f871 | 525 | if (!page || is_zone_device_page(page)) |
1da177e4 | 526 | continue; |
053837fc | 527 | /* |
62b61f61 HD |
528 | * vm_normal_page() filters out zero pages, but there might |
529 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 530 | */ |
b79bc0a0 | 531 | if (PageReserved(page)) |
f4598c8b | 532 | continue; |
88aaa2a1 | 533 | if (!queue_pages_required(page, qp)) |
38e35860 | 534 | continue; |
a7f40cfe | 535 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
d8835445 YS |
536 | /* MPOL_MF_STRICT must be specified if we get here */ |
537 | if (!vma_migratable(vma)) { | |
538 | has_unmovable = true; | |
a7f40cfe | 539 | break; |
d8835445 | 540 | } |
a53190a4 YS |
541 | |
542 | /* | |
543 | * Do not abort immediately since there may be | |
544 | * temporary off LRU pages in the range. Still | |
545 | * need migrate other LRU pages. | |
546 | */ | |
547 | if (migrate_page_add(page, qp->pagelist, flags)) | |
548 | has_unmovable = true; | |
a7f40cfe YS |
549 | } else |
550 | break; | |
6f4576e3 | 551 | } |
3f088420 | 552 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 553 | cond_resched(); |
d8835445 YS |
554 | |
555 | if (has_unmovable) | |
556 | return 1; | |
557 | ||
a7f40cfe | 558 | return addr != end ? -EIO : 0; |
91612e0d HD |
559 | } |
560 | ||
6f4576e3 NH |
561 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
562 | unsigned long addr, unsigned long end, | |
563 | struct mm_walk *walk) | |
e2d8cf40 | 564 | { |
dcf17635 | 565 | int ret = 0; |
e2d8cf40 | 566 | #ifdef CONFIG_HUGETLB_PAGE |
6f4576e3 | 567 | struct queue_pages *qp = walk->private; |
dcf17635 | 568 | unsigned long flags = (qp->flags & MPOL_MF_VALID); |
e2d8cf40 | 569 | struct page *page; |
cb900f41 | 570 | spinlock_t *ptl; |
d4c54919 | 571 | pte_t entry; |
e2d8cf40 | 572 | |
6f4576e3 NH |
573 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
574 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
575 | if (!pte_present(entry)) |
576 | goto unlock; | |
577 | page = pte_page(entry); | |
88aaa2a1 | 578 | if (!queue_pages_required(page, qp)) |
e2d8cf40 | 579 | goto unlock; |
dcf17635 LX |
580 | |
581 | if (flags == MPOL_MF_STRICT) { | |
582 | /* | |
583 | * STRICT alone means only detecting misplaced page and no | |
584 | * need to further check other vma. | |
585 | */ | |
586 | ret = -EIO; | |
587 | goto unlock; | |
588 | } | |
589 | ||
590 | if (!vma_migratable(walk->vma)) { | |
591 | /* | |
592 | * Must be STRICT with MOVE*, otherwise .test_walk() have | |
593 | * stopped walking current vma. | |
594 | * Detecting misplaced page but allow migrating pages which | |
595 | * have been queued. | |
596 | */ | |
597 | ret = 1; | |
598 | goto unlock; | |
599 | } | |
600 | ||
e2d8cf40 NH |
601 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ |
602 | if (flags & (MPOL_MF_MOVE_ALL) || | |
dcf17635 | 603 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) { |
7ce82f4c | 604 | if (isolate_hugetlb(page, qp->pagelist) && |
dcf17635 LX |
605 | (flags & MPOL_MF_STRICT)) |
606 | /* | |
607 | * Failed to isolate page but allow migrating pages | |
608 | * which have been queued. | |
609 | */ | |
610 | ret = 1; | |
611 | } | |
e2d8cf40 | 612 | unlock: |
cb900f41 | 613 | spin_unlock(ptl); |
e2d8cf40 NH |
614 | #else |
615 | BUG(); | |
616 | #endif | |
dcf17635 | 617 | return ret; |
1da177e4 LT |
618 | } |
619 | ||
5877231f | 620 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 621 | /* |
4b10e7d5 MG |
622 | * This is used to mark a range of virtual addresses to be inaccessible. |
623 | * These are later cleared by a NUMA hinting fault. Depending on these | |
624 | * faults, pages may be migrated for better NUMA placement. | |
625 | * | |
626 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
627 | * an architecture makes a different choice, it will need further | |
628 | * changes to the core. | |
b24f53a0 | 629 | */ |
4b10e7d5 MG |
630 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
631 | unsigned long addr, unsigned long end) | |
b24f53a0 | 632 | { |
4a18419f | 633 | struct mmu_gather tlb; |
4b10e7d5 | 634 | int nr_updated; |
b24f53a0 | 635 | |
4a18419f NA |
636 | tlb_gather_mmu(&tlb, vma->vm_mm); |
637 | ||
1ef488ed | 638 | nr_updated = change_protection(&tlb, vma, addr, end, MM_CP_PROT_NUMA); |
03c5a6e1 MG |
639 | if (nr_updated) |
640 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 641 | |
4a18419f NA |
642 | tlb_finish_mmu(&tlb); |
643 | ||
4b10e7d5 | 644 | return nr_updated; |
b24f53a0 LS |
645 | } |
646 | #else | |
647 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
648 | unsigned long addr, unsigned long end) | |
649 | { | |
650 | return 0; | |
651 | } | |
5877231f | 652 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 653 | |
6f4576e3 NH |
654 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
655 | struct mm_walk *walk) | |
656 | { | |
66850be5 | 657 | struct vm_area_struct *next, *vma = walk->vma; |
6f4576e3 NH |
658 | struct queue_pages *qp = walk->private; |
659 | unsigned long endvma = vma->vm_end; | |
660 | unsigned long flags = qp->flags; | |
661 | ||
a18b3ac2 | 662 | /* range check first */ |
ce33135c | 663 | VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma); |
f18da660 LX |
664 | |
665 | if (!qp->first) { | |
666 | qp->first = vma; | |
667 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
668 | (qp->start < vma->vm_start)) | |
669 | /* hole at head side of range */ | |
a18b3ac2 LX |
670 | return -EFAULT; |
671 | } | |
66850be5 | 672 | next = find_vma(vma->vm_mm, vma->vm_end); |
f18da660 LX |
673 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
674 | ((vma->vm_end < qp->end) && | |
66850be5 | 675 | (!next || vma->vm_end < next->vm_start))) |
f18da660 LX |
676 | /* hole at middle or tail of range */ |
677 | return -EFAULT; | |
a18b3ac2 | 678 | |
a7f40cfe YS |
679 | /* |
680 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
681 | * regardless of vma_migratable | |
682 | */ | |
683 | if (!vma_migratable(vma) && | |
684 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
685 | return 1; |
686 | ||
6f4576e3 NH |
687 | if (endvma > end) |
688 | endvma = end; | |
6f4576e3 | 689 | |
6f4576e3 NH |
690 | if (flags & MPOL_MF_LAZY) { |
691 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
3122e80e | 692 | if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) && |
4355c018 | 693 | !(vma->vm_flags & VM_MIXEDMAP)) |
6f4576e3 NH |
694 | change_prot_numa(vma, start, endvma); |
695 | return 1; | |
696 | } | |
697 | ||
77bf45e7 | 698 | /* queue pages from current vma */ |
a7f40cfe | 699 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
700 | return 0; |
701 | return 1; | |
702 | } | |
703 | ||
7b86ac33 CH |
704 | static const struct mm_walk_ops queue_pages_walk_ops = { |
705 | .hugetlb_entry = queue_pages_hugetlb, | |
706 | .pmd_entry = queue_pages_pte_range, | |
707 | .test_walk = queue_pages_test_walk, | |
708 | }; | |
709 | ||
dc9aa5b9 | 710 | /* |
98094945 NH |
711 | * Walk through page tables and collect pages to be migrated. |
712 | * | |
713 | * If pages found in a given range are on a set of nodes (determined by | |
714 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
715 | * passed via @private. |
716 | * | |
717 | * queue_pages_range() has three possible return values: | |
718 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
719 | * specified. | |
720 | * 0 - queue pages successfully or no misplaced page. | |
a85dfc30 YS |
721 | * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or |
722 | * memory range specified by nodemask and maxnode points outside | |
723 | * your accessible address space (-EFAULT) | |
dc9aa5b9 | 724 | */ |
d05f0cdc | 725 | static int |
98094945 | 726 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
727 | nodemask_t *nodes, unsigned long flags, |
728 | struct list_head *pagelist) | |
1da177e4 | 729 | { |
f18da660 | 730 | int err; |
6f4576e3 NH |
731 | struct queue_pages qp = { |
732 | .pagelist = pagelist, | |
733 | .flags = flags, | |
734 | .nmask = nodes, | |
f18da660 LX |
735 | .start = start, |
736 | .end = end, | |
737 | .first = NULL, | |
6f4576e3 | 738 | }; |
6f4576e3 | 739 | |
f18da660 LX |
740 | err = walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp); |
741 | ||
742 | if (!qp.first) | |
743 | /* whole range in hole */ | |
744 | err = -EFAULT; | |
745 | ||
746 | return err; | |
1da177e4 LT |
747 | } |
748 | ||
869833f2 KM |
749 | /* |
750 | * Apply policy to a single VMA | |
c1e8d7c6 | 751 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
752 | */ |
753 | static int vma_replace_policy(struct vm_area_struct *vma, | |
754 | struct mempolicy *pol) | |
8d34694c | 755 | { |
869833f2 KM |
756 | int err; |
757 | struct mempolicy *old; | |
758 | struct mempolicy *new; | |
8d34694c KM |
759 | |
760 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
761 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
762 | vma->vm_ops, vma->vm_file, | |
763 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
764 | ||
869833f2 KM |
765 | new = mpol_dup(pol); |
766 | if (IS_ERR(new)) | |
767 | return PTR_ERR(new); | |
768 | ||
769 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 770 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
771 | if (err) |
772 | goto err_out; | |
8d34694c | 773 | } |
869833f2 KM |
774 | |
775 | old = vma->vm_policy; | |
c1e8d7c6 | 776 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
777 | mpol_put(old); |
778 | ||
779 | return 0; | |
780 | err_out: | |
781 | mpol_put(new); | |
8d34694c KM |
782 | return err; |
783 | } | |
784 | ||
1da177e4 | 785 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
786 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
787 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 | 788 | { |
7329e3eb | 789 | MA_STATE(mas, &mm->mm_mt, start, start); |
9d8cebd4 KM |
790 | struct vm_area_struct *prev; |
791 | struct vm_area_struct *vma; | |
792 | int err = 0; | |
e26a5114 | 793 | pgoff_t pgoff; |
9d8cebd4 | 794 | |
7329e3eb LH |
795 | prev = mas_prev(&mas, 0); |
796 | if (unlikely(!prev)) | |
797 | mas_set(&mas, start); | |
798 | ||
799 | vma = mas_find(&mas, end - 1); | |
800 | if (WARN_ON(!vma)) | |
801 | return 0; | |
802 | ||
803 | if (start > vma->vm_start) | |
804 | prev = vma; | |
e26a5114 | 805 | |
66850be5 LH |
806 | for (; vma; vma = mas_next(&mas, end - 1)) { |
807 | unsigned long vmstart = max(start, vma->vm_start); | |
808 | unsigned long vmend = min(end, vma->vm_end); | |
9d8cebd4 | 809 | |
e26a5114 | 810 | if (mpol_equal(vma_policy(vma), new_pol)) |
66850be5 | 811 | goto next; |
e26a5114 KM |
812 | |
813 | pgoff = vma->vm_pgoff + | |
814 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 815 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af | 816 | vma->anon_vma, vma->vm_file, pgoff, |
9a10064f | 817 | new_pol, vma->vm_userfaultfd_ctx, |
5c26f6ac | 818 | anon_vma_name(vma)); |
9d8cebd4 | 819 | if (prev) { |
66850be5 LH |
820 | /* vma_merge() invalidated the mas */ |
821 | mas_pause(&mas); | |
9d8cebd4 | 822 | vma = prev; |
3964acd0 | 823 | goto replace; |
9d8cebd4 KM |
824 | } |
825 | if (vma->vm_start != vmstart) { | |
826 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
827 | if (err) | |
828 | goto out; | |
66850be5 LH |
829 | /* split_vma() invalidated the mas */ |
830 | mas_pause(&mas); | |
9d8cebd4 KM |
831 | } |
832 | if (vma->vm_end != vmend) { | |
833 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
834 | if (err) | |
835 | goto out; | |
66850be5 LH |
836 | /* split_vma() invalidated the mas */ |
837 | mas_pause(&mas); | |
9d8cebd4 | 838 | } |
66850be5 | 839 | replace: |
869833f2 | 840 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
841 | if (err) |
842 | goto out; | |
66850be5 LH |
843 | next: |
844 | prev = vma; | |
1da177e4 | 845 | } |
9d8cebd4 | 846 | |
66850be5 | 847 | out: |
1da177e4 LT |
848 | return err; |
849 | } | |
850 | ||
1da177e4 | 851 | /* Set the process memory policy */ |
028fec41 DR |
852 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
853 | nodemask_t *nodes) | |
1da177e4 | 854 | { |
58568d2a | 855 | struct mempolicy *new, *old; |
4bfc4495 | 856 | NODEMASK_SCRATCH(scratch); |
58568d2a | 857 | int ret; |
1da177e4 | 858 | |
4bfc4495 KH |
859 | if (!scratch) |
860 | return -ENOMEM; | |
f4e53d91 | 861 | |
4bfc4495 KH |
862 | new = mpol_new(mode, flags, nodes); |
863 | if (IS_ERR(new)) { | |
864 | ret = PTR_ERR(new); | |
865 | goto out; | |
866 | } | |
2c7c3a7d | 867 | |
12c1dc8e | 868 | task_lock(current); |
4bfc4495 | 869 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 870 | if (ret) { |
12c1dc8e | 871 | task_unlock(current); |
58568d2a | 872 | mpol_put(new); |
4bfc4495 | 873 | goto out; |
58568d2a | 874 | } |
12c1dc8e | 875 | |
58568d2a | 876 | old = current->mempolicy; |
1da177e4 | 877 | current->mempolicy = new; |
45816682 VB |
878 | if (new && new->mode == MPOL_INTERLEAVE) |
879 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 880 | task_unlock(current); |
58568d2a | 881 | mpol_put(old); |
4bfc4495 KH |
882 | ret = 0; |
883 | out: | |
884 | NODEMASK_SCRATCH_FREE(scratch); | |
885 | return ret; | |
1da177e4 LT |
886 | } |
887 | ||
bea904d5 LS |
888 | /* |
889 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
890 | * |
891 | * Called with task's alloc_lock held | |
bea904d5 LS |
892 | */ |
893 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 894 | { |
dfcd3c0d | 895 | nodes_clear(*nodes); |
bea904d5 LS |
896 | if (p == &default_policy) |
897 | return; | |
898 | ||
45c4745a | 899 | switch (p->mode) { |
19770b32 | 900 | case MPOL_BIND: |
1da177e4 | 901 | case MPOL_INTERLEAVE: |
269fbe72 | 902 | case MPOL_PREFERRED: |
b27abacc | 903 | case MPOL_PREFERRED_MANY: |
269fbe72 | 904 | *nodes = p->nodes; |
1da177e4 | 905 | break; |
7858d7bc FT |
906 | case MPOL_LOCAL: |
907 | /* return empty node mask for local allocation */ | |
908 | break; | |
1da177e4 LT |
909 | default: |
910 | BUG(); | |
911 | } | |
912 | } | |
913 | ||
3b9aadf7 | 914 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 915 | { |
ba841078 | 916 | struct page *p = NULL; |
f728b9c4 | 917 | int ret; |
1da177e4 | 918 | |
f728b9c4 JH |
919 | ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p); |
920 | if (ret > 0) { | |
921 | ret = page_to_nid(p); | |
1da177e4 LT |
922 | put_page(p); |
923 | } | |
f728b9c4 | 924 | return ret; |
1da177e4 LT |
925 | } |
926 | ||
1da177e4 | 927 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
928 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
929 | unsigned long addr, unsigned long flags) | |
1da177e4 | 930 | { |
8bccd85f | 931 | int err; |
1da177e4 LT |
932 | struct mm_struct *mm = current->mm; |
933 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 934 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 935 | |
754af6f5 LS |
936 | if (flags & |
937 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 938 | return -EINVAL; |
754af6f5 LS |
939 | |
940 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
941 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
942 | return -EINVAL; | |
943 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 944 | task_lock(current); |
754af6f5 | 945 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 946 | task_unlock(current); |
754af6f5 LS |
947 | return 0; |
948 | } | |
949 | ||
1da177e4 | 950 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
951 | /* |
952 | * Do NOT fall back to task policy if the | |
953 | * vma/shared policy at addr is NULL. We | |
954 | * want to return MPOL_DEFAULT in this case. | |
955 | */ | |
d8ed45c5 | 956 | mmap_read_lock(mm); |
33e3575c | 957 | vma = vma_lookup(mm, addr); |
1da177e4 | 958 | if (!vma) { |
d8ed45c5 | 959 | mmap_read_unlock(mm); |
1da177e4 LT |
960 | return -EFAULT; |
961 | } | |
962 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
963 | pol = vma->vm_ops->get_policy(vma, addr); | |
964 | else | |
965 | pol = vma->vm_policy; | |
966 | } else if (addr) | |
967 | return -EINVAL; | |
968 | ||
969 | if (!pol) | |
bea904d5 | 970 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
971 | |
972 | if (flags & MPOL_F_NODE) { | |
973 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 | 974 | /* |
f728b9c4 JH |
975 | * Take a refcount on the mpol, because we are about to |
976 | * drop the mmap_lock, after which only "pol" remains | |
977 | * valid, "vma" is stale. | |
3b9aadf7 AA |
978 | */ |
979 | pol_refcount = pol; | |
980 | vma = NULL; | |
981 | mpol_get(pol); | |
f728b9c4 | 982 | mmap_read_unlock(mm); |
3b9aadf7 | 983 | err = lookup_node(mm, addr); |
1da177e4 LT |
984 | if (err < 0) |
985 | goto out; | |
8bccd85f | 986 | *policy = err; |
1da177e4 | 987 | } else if (pol == current->mempolicy && |
45c4745a | 988 | pol->mode == MPOL_INTERLEAVE) { |
269fbe72 | 989 | *policy = next_node_in(current->il_prev, pol->nodes); |
1da177e4 LT |
990 | } else { |
991 | err = -EINVAL; | |
992 | goto out; | |
993 | } | |
bea904d5 LS |
994 | } else { |
995 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
996 | pol->mode; | |
d79df630 DR |
997 | /* |
998 | * Internal mempolicy flags must be masked off before exposing | |
999 | * the policy to userspace. | |
1000 | */ | |
1001 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1002 | } |
1da177e4 | 1003 | |
1da177e4 | 1004 | err = 0; |
58568d2a | 1005 | if (nmask) { |
c6b6ef8b LS |
1006 | if (mpol_store_user_nodemask(pol)) { |
1007 | *nmask = pol->w.user_nodemask; | |
1008 | } else { | |
1009 | task_lock(current); | |
1010 | get_policy_nodemask(pol, nmask); | |
1011 | task_unlock(current); | |
1012 | } | |
58568d2a | 1013 | } |
1da177e4 LT |
1014 | |
1015 | out: | |
52cd3b07 | 1016 | mpol_cond_put(pol); |
1da177e4 | 1017 | if (vma) |
d8ed45c5 | 1018 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1019 | if (pol_refcount) |
1020 | mpol_put(pol_refcount); | |
1da177e4 LT |
1021 | return err; |
1022 | } | |
1023 | ||
b20a3503 | 1024 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 1025 | /* |
c8633798 | 1026 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 1027 | */ |
a53190a4 | 1028 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 1029 | unsigned long flags) |
6ce3c4c0 | 1030 | { |
c8633798 | 1031 | struct page *head = compound_head(page); |
6ce3c4c0 | 1032 | /* |
fc301289 | 1033 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 1034 | */ |
c8633798 NH |
1035 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
1036 | if (!isolate_lru_page(head)) { | |
1037 | list_add_tail(&head->lru, pagelist); | |
1038 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1039 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1040 | thp_nr_pages(head)); |
a53190a4 YS |
1041 | } else if (flags & MPOL_MF_STRICT) { |
1042 | /* | |
1043 | * Non-movable page may reach here. And, there may be | |
1044 | * temporary off LRU pages or non-LRU movable pages. | |
1045 | * Treat them as unmovable pages since they can't be | |
1046 | * isolated, so they can't be moved at the moment. It | |
1047 | * should return -EIO for this case too. | |
1048 | */ | |
1049 | return -EIO; | |
62695a84 NP |
1050 | } |
1051 | } | |
a53190a4 YS |
1052 | |
1053 | return 0; | |
7e2ab150 | 1054 | } |
6ce3c4c0 | 1055 | |
7e2ab150 CL |
1056 | /* |
1057 | * Migrate pages from one node to a target node. | |
1058 | * Returns error or the number of pages not migrated. | |
1059 | */ | |
dbcb0f19 AB |
1060 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1061 | int flags) | |
7e2ab150 CL |
1062 | { |
1063 | nodemask_t nmask; | |
66850be5 | 1064 | struct vm_area_struct *vma; |
7e2ab150 CL |
1065 | LIST_HEAD(pagelist); |
1066 | int err = 0; | |
a0976311 JK |
1067 | struct migration_target_control mtc = { |
1068 | .nid = dest, | |
1069 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1070 | }; | |
7e2ab150 CL |
1071 | |
1072 | nodes_clear(nmask); | |
1073 | node_set(source, nmask); | |
6ce3c4c0 | 1074 | |
08270807 MK |
1075 | /* |
1076 | * This does not "check" the range but isolates all pages that | |
1077 | * need migration. Between passing in the full user address | |
1078 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1079 | */ | |
66850be5 | 1080 | vma = find_vma(mm, 0); |
08270807 | 1081 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); |
66850be5 | 1082 | queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1083 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1084 | ||
cf608ac1 | 1085 | if (!list_empty(&pagelist)) { |
a0976311 | 1086 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
5ac95884 | 1087 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
cf608ac1 | 1088 | if (err) |
e2d8cf40 | 1089 | putback_movable_pages(&pagelist); |
cf608ac1 | 1090 | } |
95a402c3 | 1091 | |
7e2ab150 | 1092 | return err; |
6ce3c4c0 CL |
1093 | } |
1094 | ||
39743889 | 1095 | /* |
7e2ab150 CL |
1096 | * Move pages between the two nodesets so as to preserve the physical |
1097 | * layout as much as possible. | |
39743889 CL |
1098 | * |
1099 | * Returns the number of page that could not be moved. | |
1100 | */ | |
0ce72d4f AM |
1101 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1102 | const nodemask_t *to, int flags) | |
39743889 | 1103 | { |
7e2ab150 | 1104 | int busy = 0; |
f555befd | 1105 | int err = 0; |
7e2ab150 | 1106 | nodemask_t tmp; |
39743889 | 1107 | |
361a2a22 | 1108 | lru_cache_disable(); |
0aedadf9 | 1109 | |
d8ed45c5 | 1110 | mmap_read_lock(mm); |
39743889 | 1111 | |
da0aa138 KM |
1112 | /* |
1113 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1114 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1115 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1116 | * The pair of nodemasks 'to' and 'from' define the map. | |
1117 | * | |
1118 | * If no pair of bits is found that way, fallback to picking some | |
1119 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1120 | * 'source' and 'dest' bits are the same, this represents a node | |
1121 | * that will be migrating to itself, so no pages need move. | |
1122 | * | |
1123 | * If no bits are left in 'tmp', or if all remaining bits left | |
1124 | * in 'tmp' correspond to the same bit in 'to', return false | |
1125 | * (nothing left to migrate). | |
1126 | * | |
1127 | * This lets us pick a pair of nodes to migrate between, such that | |
1128 | * if possible the dest node is not already occupied by some other | |
1129 | * source node, minimizing the risk of overloading the memory on a | |
1130 | * node that would happen if we migrated incoming memory to a node | |
1131 | * before migrating outgoing memory source that same node. | |
1132 | * | |
1133 | * A single scan of tmp is sufficient. As we go, we remember the | |
1134 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1135 | * that not only moved, but what's better, moved to an empty slot | |
1136 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1137 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1138 | * most recent <s, d> pair that moved. If we get all the way through |
1139 | * the scan of tmp without finding any node that moved, much less | |
1140 | * moved to an empty node, then there is nothing left worth migrating. | |
1141 | */ | |
d4984711 | 1142 | |
0ce72d4f | 1143 | tmp = *from; |
7e2ab150 | 1144 | while (!nodes_empty(tmp)) { |
68d68ff6 | 1145 | int s, d; |
b76ac7e7 | 1146 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1147 | int dest = 0; |
1148 | ||
1149 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1150 | |
1151 | /* | |
1152 | * do_migrate_pages() tries to maintain the relative | |
1153 | * node relationship of the pages established between | |
1154 | * threads and memory areas. | |
1155 | * | |
1156 | * However if the number of source nodes is not equal to | |
1157 | * the number of destination nodes we can not preserve | |
1158 | * this node relative relationship. In that case, skip | |
1159 | * copying memory from a node that is in the destination | |
1160 | * mask. | |
1161 | * | |
1162 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1163 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1164 | */ | |
1165 | ||
0ce72d4f AM |
1166 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1167 | (node_isset(s, *to))) | |
4a5b18cc LW |
1168 | continue; |
1169 | ||
0ce72d4f | 1170 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1171 | if (s == d) |
1172 | continue; | |
1173 | ||
1174 | source = s; /* Node moved. Memorize */ | |
1175 | dest = d; | |
1176 | ||
1177 | /* dest not in remaining from nodes? */ | |
1178 | if (!node_isset(dest, tmp)) | |
1179 | break; | |
1180 | } | |
b76ac7e7 | 1181 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1182 | break; |
1183 | ||
1184 | node_clear(source, tmp); | |
1185 | err = migrate_to_node(mm, source, dest, flags); | |
1186 | if (err > 0) | |
1187 | busy += err; | |
1188 | if (err < 0) | |
1189 | break; | |
39743889 | 1190 | } |
d8ed45c5 | 1191 | mmap_read_unlock(mm); |
d479960e | 1192 | |
361a2a22 | 1193 | lru_cache_enable(); |
7e2ab150 CL |
1194 | if (err < 0) |
1195 | return err; | |
1196 | return busy; | |
b20a3503 CL |
1197 | |
1198 | } | |
1199 | ||
3ad33b24 LS |
1200 | /* |
1201 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1202 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1203 | * Search forward from there, if not. N.B., this assumes that the |
1204 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1205 | * is in virtual address order. | |
1206 | */ | |
666feb21 | 1207 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1208 | { |
ec4858e0 | 1209 | struct folio *dst, *src = page_folio(page); |
d05f0cdc | 1210 | struct vm_area_struct *vma; |
3f649ab7 | 1211 | unsigned long address; |
66850be5 | 1212 | VMA_ITERATOR(vmi, current->mm, start); |
ec4858e0 | 1213 | gfp_t gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL; |
95a402c3 | 1214 | |
66850be5 | 1215 | for_each_vma(vmi, vma) { |
3ad33b24 LS |
1216 | address = page_address_in_vma(page, vma); |
1217 | if (address != -EFAULT) | |
1218 | break; | |
3ad33b24 | 1219 | } |
11c731e8 | 1220 | |
ec4858e0 MWO |
1221 | if (folio_test_hugetlb(src)) |
1222 | return alloc_huge_page_vma(page_hstate(&src->page), | |
389c8178 | 1223 | vma, address); |
ec4858e0 MWO |
1224 | |
1225 | if (folio_test_large(src)) | |
1226 | gfp = GFP_TRANSHUGE; | |
1227 | ||
0bf598d8 | 1228 | /* |
ec4858e0 | 1229 | * if !vma, vma_alloc_folio() will use task or system default policy |
0bf598d8 | 1230 | */ |
ec4858e0 MWO |
1231 | dst = vma_alloc_folio(gfp, folio_order(src), vma, address, |
1232 | folio_test_large(src)); | |
1233 | return &dst->page; | |
95a402c3 | 1234 | } |
b20a3503 CL |
1235 | #else |
1236 | ||
a53190a4 | 1237 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1238 | unsigned long flags) |
1239 | { | |
a53190a4 | 1240 | return -EIO; |
39743889 CL |
1241 | } |
1242 | ||
0ce72d4f AM |
1243 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1244 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1245 | { |
1246 | return -ENOSYS; | |
1247 | } | |
95a402c3 | 1248 | |
666feb21 | 1249 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1250 | { |
1251 | return NULL; | |
1252 | } | |
b20a3503 CL |
1253 | #endif |
1254 | ||
dbcb0f19 | 1255 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1256 | unsigned short mode, unsigned short mode_flags, |
1257 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1258 | { |
6ce3c4c0 CL |
1259 | struct mm_struct *mm = current->mm; |
1260 | struct mempolicy *new; | |
1261 | unsigned long end; | |
1262 | int err; | |
d8835445 | 1263 | int ret; |
6ce3c4c0 CL |
1264 | LIST_HEAD(pagelist); |
1265 | ||
b24f53a0 | 1266 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1267 | return -EINVAL; |
74c00241 | 1268 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1269 | return -EPERM; |
1270 | ||
1271 | if (start & ~PAGE_MASK) | |
1272 | return -EINVAL; | |
1273 | ||
1274 | if (mode == MPOL_DEFAULT) | |
1275 | flags &= ~MPOL_MF_STRICT; | |
1276 | ||
aaa31e05 | 1277 | len = PAGE_ALIGN(len); |
6ce3c4c0 CL |
1278 | end = start + len; |
1279 | ||
1280 | if (end < start) | |
1281 | return -EINVAL; | |
1282 | if (end == start) | |
1283 | return 0; | |
1284 | ||
028fec41 | 1285 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1286 | if (IS_ERR(new)) |
1287 | return PTR_ERR(new); | |
1288 | ||
b24f53a0 LS |
1289 | if (flags & MPOL_MF_LAZY) |
1290 | new->flags |= MPOL_F_MOF; | |
1291 | ||
6ce3c4c0 CL |
1292 | /* |
1293 | * If we are using the default policy then operation | |
1294 | * on discontinuous address spaces is okay after all | |
1295 | */ | |
1296 | if (!new) | |
1297 | flags |= MPOL_MF_DISCONTIG_OK; | |
1298 | ||
028fec41 DR |
1299 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1300 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1301 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1302 | |
0aedadf9 CL |
1303 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1304 | ||
361a2a22 | 1305 | lru_cache_disable(); |
0aedadf9 | 1306 | } |
4bfc4495 KH |
1307 | { |
1308 | NODEMASK_SCRATCH(scratch); | |
1309 | if (scratch) { | |
d8ed45c5 | 1310 | mmap_write_lock(mm); |
4bfc4495 | 1311 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1312 | if (err) |
d8ed45c5 | 1313 | mmap_write_unlock(mm); |
4bfc4495 KH |
1314 | } else |
1315 | err = -ENOMEM; | |
1316 | NODEMASK_SCRATCH_FREE(scratch); | |
1317 | } | |
b05ca738 KM |
1318 | if (err) |
1319 | goto mpol_out; | |
1320 | ||
d8835445 | 1321 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1322 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1323 | |
1324 | if (ret < 0) { | |
a85dfc30 | 1325 | err = ret; |
d8835445 YS |
1326 | goto up_out; |
1327 | } | |
1328 | ||
1329 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1330 | |
b24f53a0 LS |
1331 | if (!err) { |
1332 | int nr_failed = 0; | |
1333 | ||
cf608ac1 | 1334 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1335 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc | 1336 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
5ac95884 | 1337 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL); |
cf608ac1 | 1338 | if (nr_failed) |
74060e4d | 1339 | putback_movable_pages(&pagelist); |
cf608ac1 | 1340 | } |
6ce3c4c0 | 1341 | |
d8835445 | 1342 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1343 | err = -EIO; |
a85dfc30 | 1344 | } else { |
d8835445 | 1345 | up_out: |
a85dfc30 YS |
1346 | if (!list_empty(&pagelist)) |
1347 | putback_movable_pages(&pagelist); | |
1348 | } | |
1349 | ||
d8ed45c5 | 1350 | mmap_write_unlock(mm); |
d8835445 | 1351 | mpol_out: |
f0be3d32 | 1352 | mpol_put(new); |
d479960e | 1353 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1354 | lru_cache_enable(); |
6ce3c4c0 CL |
1355 | return err; |
1356 | } | |
1357 | ||
8bccd85f CL |
1358 | /* |
1359 | * User space interface with variable sized bitmaps for nodelists. | |
1360 | */ | |
e130242d AB |
1361 | static int get_bitmap(unsigned long *mask, const unsigned long __user *nmask, |
1362 | unsigned long maxnode) | |
1363 | { | |
1364 | unsigned long nlongs = BITS_TO_LONGS(maxnode); | |
1365 | int ret; | |
1366 | ||
1367 | if (in_compat_syscall()) | |
1368 | ret = compat_get_bitmap(mask, | |
1369 | (const compat_ulong_t __user *)nmask, | |
1370 | maxnode); | |
1371 | else | |
1372 | ret = copy_from_user(mask, nmask, | |
1373 | nlongs * sizeof(unsigned long)); | |
1374 | ||
1375 | if (ret) | |
1376 | return -EFAULT; | |
1377 | ||
1378 | if (maxnode % BITS_PER_LONG) | |
1379 | mask[nlongs - 1] &= (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1380 | ||
1381 | return 0; | |
1382 | } | |
8bccd85f CL |
1383 | |
1384 | /* Copy a node mask from user space. */ | |
39743889 | 1385 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1386 | unsigned long maxnode) |
1387 | { | |
8bccd85f CL |
1388 | --maxnode; |
1389 | nodes_clear(*nodes); | |
1390 | if (maxnode == 0 || !nmask) | |
1391 | return 0; | |
a9c930ba | 1392 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1393 | return -EINVAL; |
8bccd85f | 1394 | |
56521e7a YX |
1395 | /* |
1396 | * When the user specified more nodes than supported just check | |
e130242d AB |
1397 | * if the non supported part is all zero, one word at a time, |
1398 | * starting at the end. | |
56521e7a | 1399 | */ |
e130242d AB |
1400 | while (maxnode > MAX_NUMNODES) { |
1401 | unsigned long bits = min_t(unsigned long, maxnode, BITS_PER_LONG); | |
1402 | unsigned long t; | |
8bccd85f | 1403 | |
000eca5d | 1404 | if (get_bitmap(&t, &nmask[(maxnode - 1) / BITS_PER_LONG], bits)) |
56521e7a | 1405 | return -EFAULT; |
e130242d AB |
1406 | |
1407 | if (maxnode - bits >= MAX_NUMNODES) { | |
1408 | maxnode -= bits; | |
1409 | } else { | |
1410 | maxnode = MAX_NUMNODES; | |
1411 | t &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1412 | } | |
1413 | if (t) | |
56521e7a YX |
1414 | return -EINVAL; |
1415 | } | |
1416 | ||
e130242d | 1417 | return get_bitmap(nodes_addr(*nodes), nmask, maxnode); |
8bccd85f CL |
1418 | } |
1419 | ||
1420 | /* Copy a kernel node mask to user space */ | |
1421 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1422 | nodemask_t *nodes) | |
1423 | { | |
1424 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1425 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
e130242d AB |
1426 | bool compat = in_compat_syscall(); |
1427 | ||
1428 | if (compat) | |
1429 | nbytes = BITS_TO_COMPAT_LONGS(nr_node_ids) * sizeof(compat_long_t); | |
8bccd85f CL |
1430 | |
1431 | if (copy > nbytes) { | |
1432 | if (copy > PAGE_SIZE) | |
1433 | return -EINVAL; | |
1434 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1435 | return -EFAULT; | |
1436 | copy = nbytes; | |
e130242d | 1437 | maxnode = nr_node_ids; |
8bccd85f | 1438 | } |
e130242d AB |
1439 | |
1440 | if (compat) | |
1441 | return compat_put_bitmap((compat_ulong_t __user *)mask, | |
1442 | nodes_addr(*nodes), maxnode); | |
1443 | ||
8bccd85f CL |
1444 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; |
1445 | } | |
1446 | ||
95837924 FT |
1447 | /* Basic parameter sanity check used by both mbind() and set_mempolicy() */ |
1448 | static inline int sanitize_mpol_flags(int *mode, unsigned short *flags) | |
1449 | { | |
1450 | *flags = *mode & MPOL_MODE_FLAGS; | |
1451 | *mode &= ~MPOL_MODE_FLAGS; | |
b27abacc | 1452 | |
a38a59fd | 1453 | if ((unsigned int)(*mode) >= MPOL_MAX) |
95837924 FT |
1454 | return -EINVAL; |
1455 | if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES)) | |
1456 | return -EINVAL; | |
6d2aec9e ED |
1457 | if (*flags & MPOL_F_NUMA_BALANCING) { |
1458 | if (*mode != MPOL_BIND) | |
1459 | return -EINVAL; | |
1460 | *flags |= (MPOL_F_MOF | MPOL_F_MORON); | |
1461 | } | |
95837924 FT |
1462 | return 0; |
1463 | } | |
1464 | ||
e7dc9ad6 DB |
1465 | static long kernel_mbind(unsigned long start, unsigned long len, |
1466 | unsigned long mode, const unsigned long __user *nmask, | |
1467 | unsigned long maxnode, unsigned int flags) | |
8bccd85f | 1468 | { |
95837924 | 1469 | unsigned short mode_flags; |
8bccd85f | 1470 | nodemask_t nodes; |
95837924 | 1471 | int lmode = mode; |
8bccd85f CL |
1472 | int err; |
1473 | ||
057d3389 | 1474 | start = untagged_addr(start); |
95837924 FT |
1475 | err = sanitize_mpol_flags(&lmode, &mode_flags); |
1476 | if (err) | |
1477 | return err; | |
1478 | ||
8bccd85f CL |
1479 | err = get_nodes(&nodes, nmask, maxnode); |
1480 | if (err) | |
1481 | return err; | |
95837924 FT |
1482 | |
1483 | return do_mbind(start, len, lmode, mode_flags, &nodes, flags); | |
8bccd85f CL |
1484 | } |
1485 | ||
c6018b4b AK |
1486 | SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, len, |
1487 | unsigned long, home_node, unsigned long, flags) | |
1488 | { | |
1489 | struct mm_struct *mm = current->mm; | |
1490 | struct vm_area_struct *vma; | |
e976936c | 1491 | struct mempolicy *new, *old; |
c6018b4b AK |
1492 | unsigned long vmstart; |
1493 | unsigned long vmend; | |
1494 | unsigned long end; | |
1495 | int err = -ENOENT; | |
66850be5 | 1496 | VMA_ITERATOR(vmi, mm, start); |
c6018b4b AK |
1497 | |
1498 | start = untagged_addr(start); | |
1499 | if (start & ~PAGE_MASK) | |
1500 | return -EINVAL; | |
1501 | /* | |
1502 | * flags is used for future extension if any. | |
1503 | */ | |
1504 | if (flags != 0) | |
1505 | return -EINVAL; | |
1506 | ||
1507 | /* | |
1508 | * Check home_node is online to avoid accessing uninitialized | |
1509 | * NODE_DATA. | |
1510 | */ | |
1511 | if (home_node >= MAX_NUMNODES || !node_online(home_node)) | |
1512 | return -EINVAL; | |
1513 | ||
aaa31e05 | 1514 | len = PAGE_ALIGN(len); |
c6018b4b AK |
1515 | end = start + len; |
1516 | ||
1517 | if (end < start) | |
1518 | return -EINVAL; | |
1519 | if (end == start) | |
1520 | return 0; | |
1521 | mmap_write_lock(mm); | |
66850be5 | 1522 | for_each_vma_range(vmi, vma, end) { |
c6018b4b AK |
1523 | /* |
1524 | * If any vma in the range got policy other than MPOL_BIND | |
1525 | * or MPOL_PREFERRED_MANY we return error. We don't reset | |
1526 | * the home node for vmas we already updated before. | |
1527 | */ | |
e976936c MH |
1528 | old = vma_policy(vma); |
1529 | if (!old) | |
1530 | continue; | |
1531 | if (old->mode != MPOL_BIND && old->mode != MPOL_PREFERRED_MANY) { | |
c6018b4b AK |
1532 | err = -EOPNOTSUPP; |
1533 | break; | |
1534 | } | |
e976936c MH |
1535 | new = mpol_dup(old); |
1536 | if (IS_ERR(new)) { | |
1537 | err = PTR_ERR(new); | |
1538 | break; | |
1539 | } | |
c6018b4b AK |
1540 | |
1541 | new->home_node = home_node; | |
e976936c MH |
1542 | vmstart = max(start, vma->vm_start); |
1543 | vmend = min(end, vma->vm_end); | |
c6018b4b AK |
1544 | err = mbind_range(mm, vmstart, vmend, new); |
1545 | mpol_put(new); | |
1546 | if (err) | |
1547 | break; | |
1548 | } | |
1549 | mmap_write_unlock(mm); | |
1550 | return err; | |
1551 | } | |
1552 | ||
e7dc9ad6 DB |
1553 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1554 | unsigned long, mode, const unsigned long __user *, nmask, | |
1555 | unsigned long, maxnode, unsigned int, flags) | |
1556 | { | |
1557 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1558 | } | |
1559 | ||
8bccd85f | 1560 | /* Set the process memory policy */ |
af03c4ac DB |
1561 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1562 | unsigned long maxnode) | |
8bccd85f | 1563 | { |
95837924 | 1564 | unsigned short mode_flags; |
8bccd85f | 1565 | nodemask_t nodes; |
95837924 FT |
1566 | int lmode = mode; |
1567 | int err; | |
1568 | ||
1569 | err = sanitize_mpol_flags(&lmode, &mode_flags); | |
1570 | if (err) | |
1571 | return err; | |
8bccd85f | 1572 | |
8bccd85f CL |
1573 | err = get_nodes(&nodes, nmask, maxnode); |
1574 | if (err) | |
1575 | return err; | |
95837924 FT |
1576 | |
1577 | return do_set_mempolicy(lmode, mode_flags, &nodes); | |
8bccd85f CL |
1578 | } |
1579 | ||
af03c4ac DB |
1580 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1581 | unsigned long, maxnode) | |
1582 | { | |
1583 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1584 | } | |
1585 | ||
b6e9b0ba DB |
1586 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1587 | const unsigned long __user *old_nodes, | |
1588 | const unsigned long __user *new_nodes) | |
39743889 | 1589 | { |
596d7cfa | 1590 | struct mm_struct *mm = NULL; |
39743889 | 1591 | struct task_struct *task; |
39743889 CL |
1592 | nodemask_t task_nodes; |
1593 | int err; | |
596d7cfa KM |
1594 | nodemask_t *old; |
1595 | nodemask_t *new; | |
1596 | NODEMASK_SCRATCH(scratch); | |
1597 | ||
1598 | if (!scratch) | |
1599 | return -ENOMEM; | |
39743889 | 1600 | |
596d7cfa KM |
1601 | old = &scratch->mask1; |
1602 | new = &scratch->mask2; | |
1603 | ||
1604 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1605 | if (err) |
596d7cfa | 1606 | goto out; |
39743889 | 1607 | |
596d7cfa | 1608 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1609 | if (err) |
596d7cfa | 1610 | goto out; |
39743889 CL |
1611 | |
1612 | /* Find the mm_struct */ | |
55cfaa3c | 1613 | rcu_read_lock(); |
228ebcbe | 1614 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1615 | if (!task) { |
55cfaa3c | 1616 | rcu_read_unlock(); |
596d7cfa KM |
1617 | err = -ESRCH; |
1618 | goto out; | |
39743889 | 1619 | } |
3268c63e | 1620 | get_task_struct(task); |
39743889 | 1621 | |
596d7cfa | 1622 | err = -EINVAL; |
39743889 CL |
1623 | |
1624 | /* | |
31367466 OE |
1625 | * Check if this process has the right to modify the specified process. |
1626 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1627 | */ |
31367466 | 1628 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1629 | rcu_read_unlock(); |
39743889 | 1630 | err = -EPERM; |
3268c63e | 1631 | goto out_put; |
39743889 | 1632 | } |
c69e8d9c | 1633 | rcu_read_unlock(); |
39743889 CL |
1634 | |
1635 | task_nodes = cpuset_mems_allowed(task); | |
1636 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1637 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1638 | err = -EPERM; |
3268c63e | 1639 | goto out_put; |
39743889 CL |
1640 | } |
1641 | ||
0486a38b YX |
1642 | task_nodes = cpuset_mems_allowed(current); |
1643 | nodes_and(*new, *new, task_nodes); | |
1644 | if (nodes_empty(*new)) | |
1645 | goto out_put; | |
1646 | ||
86c3a764 DQ |
1647 | err = security_task_movememory(task); |
1648 | if (err) | |
3268c63e | 1649 | goto out_put; |
86c3a764 | 1650 | |
3268c63e CL |
1651 | mm = get_task_mm(task); |
1652 | put_task_struct(task); | |
f2a9ef88 SL |
1653 | |
1654 | if (!mm) { | |
3268c63e | 1655 | err = -EINVAL; |
f2a9ef88 SL |
1656 | goto out; |
1657 | } | |
1658 | ||
1659 | err = do_migrate_pages(mm, old, new, | |
1660 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1661 | |
1662 | mmput(mm); | |
1663 | out: | |
596d7cfa KM |
1664 | NODEMASK_SCRATCH_FREE(scratch); |
1665 | ||
39743889 | 1666 | return err; |
3268c63e CL |
1667 | |
1668 | out_put: | |
1669 | put_task_struct(task); | |
1670 | goto out; | |
1671 | ||
39743889 CL |
1672 | } |
1673 | ||
b6e9b0ba DB |
1674 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1675 | const unsigned long __user *, old_nodes, | |
1676 | const unsigned long __user *, new_nodes) | |
1677 | { | |
1678 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1679 | } | |
1680 | ||
39743889 | 1681 | |
8bccd85f | 1682 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1683 | static int kernel_get_mempolicy(int __user *policy, |
1684 | unsigned long __user *nmask, | |
1685 | unsigned long maxnode, | |
1686 | unsigned long addr, | |
1687 | unsigned long flags) | |
8bccd85f | 1688 | { |
dbcb0f19 | 1689 | int err; |
3f649ab7 | 1690 | int pval; |
8bccd85f CL |
1691 | nodemask_t nodes; |
1692 | ||
050c17f2 | 1693 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1694 | return -EINVAL; |
1695 | ||
4605f057 WH |
1696 | addr = untagged_addr(addr); |
1697 | ||
8bccd85f CL |
1698 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1699 | ||
1700 | if (err) | |
1701 | return err; | |
1702 | ||
1703 | if (policy && put_user(pval, policy)) | |
1704 | return -EFAULT; | |
1705 | ||
1706 | if (nmask) | |
1707 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1708 | ||
1709 | return err; | |
1710 | } | |
1711 | ||
af03c4ac DB |
1712 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1713 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1714 | unsigned long, addr, unsigned long, flags) | |
1715 | { | |
1716 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1717 | } | |
1718 | ||
20ca87f2 LX |
1719 | bool vma_migratable(struct vm_area_struct *vma) |
1720 | { | |
1721 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1722 | return false; | |
1723 | ||
1724 | /* | |
1725 | * DAX device mappings require predictable access latency, so avoid | |
1726 | * incurring periodic faults. | |
1727 | */ | |
1728 | if (vma_is_dax(vma)) | |
1729 | return false; | |
1730 | ||
1731 | if (is_vm_hugetlb_page(vma) && | |
1732 | !hugepage_migration_supported(hstate_vma(vma))) | |
1733 | return false; | |
1734 | ||
1735 | /* | |
1736 | * Migration allocates pages in the highest zone. If we cannot | |
1737 | * do so then migration (at least from node to node) is not | |
1738 | * possible. | |
1739 | */ | |
1740 | if (vma->vm_file && | |
1741 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1742 | < policy_zone) | |
1743 | return false; | |
1744 | return true; | |
1745 | } | |
1746 | ||
74d2c3a0 ON |
1747 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1748 | unsigned long addr) | |
1da177e4 | 1749 | { |
8d90274b | 1750 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1751 | |
1752 | if (vma) { | |
480eccf9 | 1753 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1754 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1755 | } else if (vma->vm_policy) { |
1da177e4 | 1756 | pol = vma->vm_policy; |
00442ad0 MG |
1757 | |
1758 | /* | |
1759 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1760 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1761 | * count on these policies which will be dropped by | |
1762 | * mpol_cond_put() later | |
1763 | */ | |
1764 | if (mpol_needs_cond_ref(pol)) | |
1765 | mpol_get(pol); | |
1766 | } | |
1da177e4 | 1767 | } |
f15ca78e | 1768 | |
74d2c3a0 ON |
1769 | return pol; |
1770 | } | |
1771 | ||
1772 | /* | |
dd6eecb9 | 1773 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1774 | * @vma: virtual memory area whose policy is sought |
1775 | * @addr: address in @vma for shared policy lookup | |
1776 | * | |
1777 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1778 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1779 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1780 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1781 | * freeing by another task. It is the caller's responsibility to free the | |
1782 | * extra reference for shared policies. | |
1783 | */ | |
ac79f78d | 1784 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1785 | unsigned long addr) |
74d2c3a0 ON |
1786 | { |
1787 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1788 | ||
8d90274b | 1789 | if (!pol) |
dd6eecb9 | 1790 | pol = get_task_policy(current); |
8d90274b | 1791 | |
1da177e4 LT |
1792 | return pol; |
1793 | } | |
1794 | ||
6b6482bb | 1795 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1796 | { |
6b6482bb | 1797 | struct mempolicy *pol; |
fc314724 | 1798 | |
6b6482bb ON |
1799 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1800 | bool ret = false; | |
fc314724 | 1801 | |
6b6482bb ON |
1802 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1803 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1804 | ret = true; | |
1805 | mpol_cond_put(pol); | |
8d90274b | 1806 | |
6b6482bb | 1807 | return ret; |
fc314724 MG |
1808 | } |
1809 | ||
6b6482bb | 1810 | pol = vma->vm_policy; |
8d90274b | 1811 | if (!pol) |
6b6482bb | 1812 | pol = get_task_policy(current); |
8d90274b | 1813 | |
fc314724 MG |
1814 | return pol->flags & MPOL_F_MOF; |
1815 | } | |
1816 | ||
d2226ebd | 1817 | bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
d3eb1570 LJ |
1818 | { |
1819 | enum zone_type dynamic_policy_zone = policy_zone; | |
1820 | ||
1821 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1822 | ||
1823 | /* | |
269fbe72 | 1824 | * if policy->nodes has movable memory only, |
d3eb1570 LJ |
1825 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. |
1826 | * | |
269fbe72 | 1827 | * policy->nodes is intersect with node_states[N_MEMORY]. |
f0953a1b | 1828 | * so if the following test fails, it implies |
269fbe72 | 1829 | * policy->nodes has movable memory only. |
d3eb1570 | 1830 | */ |
269fbe72 | 1831 | if (!nodes_intersects(policy->nodes, node_states[N_HIGH_MEMORY])) |
d3eb1570 LJ |
1832 | dynamic_policy_zone = ZONE_MOVABLE; |
1833 | ||
1834 | return zone >= dynamic_policy_zone; | |
1835 | } | |
1836 | ||
52cd3b07 LS |
1837 | /* |
1838 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1839 | * page allocation | |
1840 | */ | |
8ca39e68 | 1841 | nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
19770b32 | 1842 | { |
b27abacc DH |
1843 | int mode = policy->mode; |
1844 | ||
19770b32 | 1845 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ |
b27abacc DH |
1846 | if (unlikely(mode == MPOL_BIND) && |
1847 | apply_policy_zone(policy, gfp_zone(gfp)) && | |
1848 | cpuset_nodemask_valid_mems_allowed(&policy->nodes)) | |
1849 | return &policy->nodes; | |
1850 | ||
1851 | if (mode == MPOL_PREFERRED_MANY) | |
269fbe72 | 1852 | return &policy->nodes; |
19770b32 MG |
1853 | |
1854 | return NULL; | |
1855 | } | |
1856 | ||
b27abacc DH |
1857 | /* |
1858 | * Return the preferred node id for 'prefer' mempolicy, and return | |
1859 | * the given id for all other policies. | |
1860 | * | |
1861 | * policy_node() is always coupled with policy_nodemask(), which | |
1862 | * secures the nodemask limit for 'bind' and 'prefer-many' policy. | |
1863 | */ | |
f8fd5253 | 1864 | static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd) |
1da177e4 | 1865 | { |
7858d7bc | 1866 | if (policy->mode == MPOL_PREFERRED) { |
269fbe72 | 1867 | nd = first_node(policy->nodes); |
7858d7bc | 1868 | } else { |
19770b32 | 1869 | /* |
6d840958 MH |
1870 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1871 | * because we might easily break the expectation to stay on the | |
1872 | * requested node and not break the policy. | |
19770b32 | 1873 | */ |
6d840958 | 1874 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1875 | } |
6d840958 | 1876 | |
c6018b4b AK |
1877 | if ((policy->mode == MPOL_BIND || |
1878 | policy->mode == MPOL_PREFERRED_MANY) && | |
1879 | policy->home_node != NUMA_NO_NODE) | |
1880 | return policy->home_node; | |
1881 | ||
04ec6264 | 1882 | return nd; |
1da177e4 LT |
1883 | } |
1884 | ||
1885 | /* Do dynamic interleaving for a process */ | |
1886 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1887 | { | |
45816682 | 1888 | unsigned next; |
1da177e4 LT |
1889 | struct task_struct *me = current; |
1890 | ||
269fbe72 | 1891 | next = next_node_in(me->il_prev, policy->nodes); |
f5b087b5 | 1892 | if (next < MAX_NUMNODES) |
45816682 VB |
1893 | me->il_prev = next; |
1894 | return next; | |
1da177e4 LT |
1895 | } |
1896 | ||
dc85da15 CL |
1897 | /* |
1898 | * Depending on the memory policy provide a node from which to allocate the | |
1899 | * next slab entry. | |
1900 | */ | |
2a389610 | 1901 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1902 | { |
e7b691b0 | 1903 | struct mempolicy *policy; |
2a389610 | 1904 | int node = numa_mem_id(); |
e7b691b0 | 1905 | |
38b031dd | 1906 | if (!in_task()) |
2a389610 | 1907 | return node; |
e7b691b0 AK |
1908 | |
1909 | policy = current->mempolicy; | |
7858d7bc | 1910 | if (!policy) |
2a389610 | 1911 | return node; |
bea904d5 LS |
1912 | |
1913 | switch (policy->mode) { | |
1914 | case MPOL_PREFERRED: | |
269fbe72 | 1915 | return first_node(policy->nodes); |
765c4507 | 1916 | |
dc85da15 CL |
1917 | case MPOL_INTERLEAVE: |
1918 | return interleave_nodes(policy); | |
1919 | ||
b27abacc DH |
1920 | case MPOL_BIND: |
1921 | case MPOL_PREFERRED_MANY: | |
1922 | { | |
c33d6c06 MG |
1923 | struct zoneref *z; |
1924 | ||
dc85da15 CL |
1925 | /* |
1926 | * Follow bind policy behavior and start allocation at the | |
1927 | * first node. | |
1928 | */ | |
19770b32 | 1929 | struct zonelist *zonelist; |
19770b32 | 1930 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1931 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 | 1932 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
269fbe72 | 1933 | &policy->nodes); |
c1093b74 | 1934 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1935 | } |
7858d7bc FT |
1936 | case MPOL_LOCAL: |
1937 | return node; | |
dc85da15 | 1938 | |
dc85da15 | 1939 | default: |
bea904d5 | 1940 | BUG(); |
dc85da15 CL |
1941 | } |
1942 | } | |
1943 | ||
fee83b3a AM |
1944 | /* |
1945 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
269fbe72 | 1946 | * node in pol->nodes (starting from n=0), wrapping around if n exceeds the |
fee83b3a AM |
1947 | * number of present nodes. |
1948 | */ | |
98c70baa | 1949 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1950 | { |
276aeee1 | 1951 | nodemask_t nodemask = pol->nodes; |
1952 | unsigned int target, nnodes; | |
fee83b3a AM |
1953 | int i; |
1954 | int nid; | |
276aeee1 | 1955 | /* |
1956 | * The barrier will stabilize the nodemask in a register or on | |
1957 | * the stack so that it will stop changing under the code. | |
1958 | * | |
1959 | * Between first_node() and next_node(), pol->nodes could be changed | |
1960 | * by other threads. So we put pol->nodes in a local stack. | |
1961 | */ | |
1962 | barrier(); | |
1da177e4 | 1963 | |
276aeee1 | 1964 | nnodes = nodes_weight(nodemask); |
f5b087b5 DR |
1965 | if (!nnodes) |
1966 | return numa_node_id(); | |
fee83b3a | 1967 | target = (unsigned int)n % nnodes; |
276aeee1 | 1968 | nid = first_node(nodemask); |
fee83b3a | 1969 | for (i = 0; i < target; i++) |
276aeee1 | 1970 | nid = next_node(nid, nodemask); |
1da177e4 LT |
1971 | return nid; |
1972 | } | |
1973 | ||
5da7ca86 CL |
1974 | /* Determine a node number for interleave */ |
1975 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1976 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1977 | { | |
1978 | if (vma) { | |
1979 | unsigned long off; | |
1980 | ||
3b98b087 NA |
1981 | /* |
1982 | * for small pages, there is no difference between | |
1983 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1984 | * for huge pages, since vm_pgoff is in units of small | |
1985 | * pages, we need to shift off the always 0 bits to get | |
1986 | * a useful offset. | |
1987 | */ | |
1988 | BUG_ON(shift < PAGE_SHIFT); | |
1989 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 1990 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 1991 | return offset_il_node(pol, off); |
5da7ca86 CL |
1992 | } else |
1993 | return interleave_nodes(pol); | |
1994 | } | |
1995 | ||
00ac59ad | 1996 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 1997 | /* |
04ec6264 | 1998 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
1999 | * @vma: virtual memory area whose policy is sought |
2000 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2001 | * @gfp_flags: for requested zone | |
2002 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
b27abacc | 2003 | * @nodemask: pointer to nodemask pointer for 'bind' and 'prefer-many' policy |
480eccf9 | 2004 | * |
04ec6264 | 2005 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 | 2006 | * to the struct mempolicy for conditional unref after allocation. |
b27abacc DH |
2007 | * If the effective policy is 'bind' or 'prefer-many', returns a pointer |
2008 | * to the mempolicy's @nodemask for filtering the zonelist. | |
c0ff7453 | 2009 | * |
d26914d1 | 2010 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2011 | */ |
04ec6264 VB |
2012 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2013 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2014 | { |
04ec6264 | 2015 | int nid; |
b27abacc | 2016 | int mode; |
5da7ca86 | 2017 | |
dd6eecb9 | 2018 | *mpol = get_vma_policy(vma, addr); |
b27abacc DH |
2019 | *nodemask = NULL; |
2020 | mode = (*mpol)->mode; | |
5da7ca86 | 2021 | |
b27abacc | 2022 | if (unlikely(mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2023 | nid = interleave_nid(*mpol, vma, addr, |
2024 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2025 | } else { |
04ec6264 | 2026 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
b27abacc | 2027 | if (mode == MPOL_BIND || mode == MPOL_PREFERRED_MANY) |
269fbe72 | 2028 | *nodemask = &(*mpol)->nodes; |
480eccf9 | 2029 | } |
04ec6264 | 2030 | return nid; |
5da7ca86 | 2031 | } |
06808b08 LS |
2032 | |
2033 | /* | |
2034 | * init_nodemask_of_mempolicy | |
2035 | * | |
2036 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2037 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2038 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2039 | * initialize the argument nodemask to contain the single node for | |
2040 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2041 | * of non-default mempolicy. | |
2042 | * | |
2043 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2044 | * because the current task is examining it's own mempolicy and a task's | |
2045 | * mempolicy is only ever changed by the task itself. | |
2046 | * | |
2047 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2048 | */ | |
2049 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2050 | { | |
2051 | struct mempolicy *mempolicy; | |
06808b08 LS |
2052 | |
2053 | if (!(mask && current->mempolicy)) | |
2054 | return false; | |
2055 | ||
c0ff7453 | 2056 | task_lock(current); |
06808b08 LS |
2057 | mempolicy = current->mempolicy; |
2058 | switch (mempolicy->mode) { | |
2059 | case MPOL_PREFERRED: | |
b27abacc | 2060 | case MPOL_PREFERRED_MANY: |
06808b08 | 2061 | case MPOL_BIND: |
06808b08 | 2062 | case MPOL_INTERLEAVE: |
269fbe72 | 2063 | *mask = mempolicy->nodes; |
7858d7bc FT |
2064 | break; |
2065 | ||
2066 | case MPOL_LOCAL: | |
269fbe72 | 2067 | init_nodemask_of_node(mask, numa_node_id()); |
06808b08 LS |
2068 | break; |
2069 | ||
2070 | default: | |
2071 | BUG(); | |
2072 | } | |
c0ff7453 | 2073 | task_unlock(current); |
06808b08 LS |
2074 | |
2075 | return true; | |
2076 | } | |
00ac59ad | 2077 | #endif |
5da7ca86 | 2078 | |
6f48d0eb | 2079 | /* |
b26e517a | 2080 | * mempolicy_in_oom_domain |
6f48d0eb | 2081 | * |
b26e517a FT |
2082 | * If tsk's mempolicy is "bind", check for intersection between mask and |
2083 | * the policy nodemask. Otherwise, return true for all other policies | |
2084 | * including "interleave", as a tsk with "interleave" policy may have | |
2085 | * memory allocated from all nodes in system. | |
6f48d0eb DR |
2086 | * |
2087 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2088 | */ | |
b26e517a | 2089 | bool mempolicy_in_oom_domain(struct task_struct *tsk, |
6f48d0eb DR |
2090 | const nodemask_t *mask) |
2091 | { | |
2092 | struct mempolicy *mempolicy; | |
2093 | bool ret = true; | |
2094 | ||
2095 | if (!mask) | |
2096 | return ret; | |
b26e517a | 2097 | |
6f48d0eb DR |
2098 | task_lock(tsk); |
2099 | mempolicy = tsk->mempolicy; | |
b26e517a | 2100 | if (mempolicy && mempolicy->mode == MPOL_BIND) |
269fbe72 | 2101 | ret = nodes_intersects(mempolicy->nodes, *mask); |
6f48d0eb | 2102 | task_unlock(tsk); |
b26e517a | 2103 | |
6f48d0eb DR |
2104 | return ret; |
2105 | } | |
2106 | ||
1da177e4 LT |
2107 | /* Allocate a page in interleaved policy. |
2108 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2109 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2110 | unsigned nid) | |
1da177e4 | 2111 | { |
1da177e4 LT |
2112 | struct page *page; |
2113 | ||
84172f4b | 2114 | page = __alloc_pages(gfp, order, nid, NULL); |
4518085e KW |
2115 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2116 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2117 | return page; | |
de55c8b2 AR |
2118 | if (page && page_to_nid(page) == nid) { |
2119 | preempt_disable(); | |
f19298b9 | 2120 | __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT); |
de55c8b2 AR |
2121 | preempt_enable(); |
2122 | } | |
1da177e4 LT |
2123 | return page; |
2124 | } | |
2125 | ||
4c54d949 FT |
2126 | static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order, |
2127 | int nid, struct mempolicy *pol) | |
2128 | { | |
2129 | struct page *page; | |
2130 | gfp_t preferred_gfp; | |
2131 | ||
2132 | /* | |
2133 | * This is a two pass approach. The first pass will only try the | |
2134 | * preferred nodes but skip the direct reclaim and allow the | |
2135 | * allocation to fail, while the second pass will try all the | |
2136 | * nodes in system. | |
2137 | */ | |
2138 | preferred_gfp = gfp | __GFP_NOWARN; | |
2139 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2140 | page = __alloc_pages(preferred_gfp, order, nid, &pol->nodes); | |
2141 | if (!page) | |
c0455116 | 2142 | page = __alloc_pages(gfp, order, nid, NULL); |
4c54d949 FT |
2143 | |
2144 | return page; | |
2145 | } | |
2146 | ||
1da177e4 | 2147 | /** |
adf88aa8 | 2148 | * vma_alloc_folio - Allocate a folio for a VMA. |
eb350739 | 2149 | * @gfp: GFP flags. |
adf88aa8 | 2150 | * @order: Order of the folio. |
eb350739 MWO |
2151 | * @vma: Pointer to VMA or NULL if not available. |
2152 | * @addr: Virtual address of the allocation. Must be inside @vma. | |
eb350739 | 2153 | * @hugepage: For hugepages try only the preferred node if possible. |
1da177e4 | 2154 | * |
adf88aa8 | 2155 | * Allocate a folio for a specific address in @vma, using the appropriate |
eb350739 MWO |
2156 | * NUMA policy. When @vma is not NULL the caller must hold the mmap_lock |
2157 | * of the mm_struct of the VMA to prevent it from going away. Should be | |
adf88aa8 | 2158 | * used for all allocations for folios that will be mapped into user space. |
1da177e4 | 2159 | * |
adf88aa8 | 2160 | * Return: The folio on success or NULL if allocation fails. |
1da177e4 | 2161 | */ |
adf88aa8 | 2162 | struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma, |
be1a13eb | 2163 | unsigned long addr, bool hugepage) |
1da177e4 | 2164 | { |
cc9a6c87 | 2165 | struct mempolicy *pol; |
be1a13eb | 2166 | int node = numa_node_id(); |
adf88aa8 | 2167 | struct folio *folio; |
04ec6264 | 2168 | int preferred_nid; |
be97a41b | 2169 | nodemask_t *nmask; |
cc9a6c87 | 2170 | |
dd6eecb9 | 2171 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2172 | |
0867a57c | 2173 | if (pol->mode == MPOL_INTERLEAVE) { |
adf88aa8 | 2174 | struct page *page; |
0867a57c VB |
2175 | unsigned nid; |
2176 | ||
2177 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2178 | mpol_cond_put(pol); | |
adf88aa8 | 2179 | gfp |= __GFP_COMP; |
0867a57c | 2180 | page = alloc_page_interleave(gfp, order, nid); |
adf88aa8 MWO |
2181 | if (page && order > 1) |
2182 | prep_transhuge_page(page); | |
2183 | folio = (struct folio *)page; | |
0867a57c | 2184 | goto out; |
19deb769 DR |
2185 | } |
2186 | ||
4c54d949 | 2187 | if (pol->mode == MPOL_PREFERRED_MANY) { |
adf88aa8 MWO |
2188 | struct page *page; |
2189 | ||
c0455116 | 2190 | node = policy_node(gfp, pol, node); |
adf88aa8 | 2191 | gfp |= __GFP_COMP; |
4c54d949 FT |
2192 | page = alloc_pages_preferred_many(gfp, order, node, pol); |
2193 | mpol_cond_put(pol); | |
adf88aa8 MWO |
2194 | if (page && order > 1) |
2195 | prep_transhuge_page(page); | |
2196 | folio = (struct folio *)page; | |
4c54d949 FT |
2197 | goto out; |
2198 | } | |
2199 | ||
19deb769 DR |
2200 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { |
2201 | int hpage_node = node; | |
2202 | ||
2203 | /* | |
2204 | * For hugepage allocation and non-interleave policy which | |
2205 | * allows the current node (or other explicitly preferred | |
2206 | * node) we only try to allocate from the current/preferred | |
2207 | * node and don't fall back to other nodes, as the cost of | |
2208 | * remote accesses would likely offset THP benefits. | |
2209 | * | |
b27abacc | 2210 | * If the policy is interleave or does not allow the current |
19deb769 DR |
2211 | * node in its nodemask, we allocate the standard way. |
2212 | */ | |
7858d7bc | 2213 | if (pol->mode == MPOL_PREFERRED) |
269fbe72 | 2214 | hpage_node = first_node(pol->nodes); |
19deb769 DR |
2215 | |
2216 | nmask = policy_nodemask(gfp, pol); | |
2217 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2218 | mpol_cond_put(pol); | |
cc638f32 VB |
2219 | /* |
2220 | * First, try to allocate THP only on local node, but | |
2221 | * don't reclaim unnecessarily, just compact. | |
2222 | */ | |
adf88aa8 MWO |
2223 | folio = __folio_alloc_node(gfp | __GFP_THISNODE | |
2224 | __GFP_NORETRY, order, hpage_node); | |
76e654cc DR |
2225 | |
2226 | /* | |
2227 | * If hugepage allocations are configured to always | |
2228 | * synchronous compact or the vma has been madvised | |
2229 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2230 | * memory with both reclaim and compact as well. |
76e654cc | 2231 | */ |
adf88aa8 MWO |
2232 | if (!folio && (gfp & __GFP_DIRECT_RECLAIM)) |
2233 | folio = __folio_alloc(gfp, order, hpage_node, | |
2234 | nmask); | |
76e654cc | 2235 | |
19deb769 DR |
2236 | goto out; |
2237 | } | |
356ff8a9 DR |
2238 | } |
2239 | ||
be97a41b | 2240 | nmask = policy_nodemask(gfp, pol); |
04ec6264 | 2241 | preferred_nid = policy_node(gfp, pol, node); |
adf88aa8 | 2242 | folio = __folio_alloc(gfp, order, preferred_nid, nmask); |
d51e9894 | 2243 | mpol_cond_put(pol); |
be97a41b | 2244 | out: |
f584b680 MWO |
2245 | return folio; |
2246 | } | |
adf88aa8 | 2247 | EXPORT_SYMBOL(vma_alloc_folio); |
f584b680 | 2248 | |
1da177e4 | 2249 | /** |
6421ec76 MWO |
2250 | * alloc_pages - Allocate pages. |
2251 | * @gfp: GFP flags. | |
2252 | * @order: Power of two of number of pages to allocate. | |
1da177e4 | 2253 | * |
6421ec76 MWO |
2254 | * Allocate 1 << @order contiguous pages. The physical address of the |
2255 | * first page is naturally aligned (eg an order-3 allocation will be aligned | |
2256 | * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current | |
2257 | * process is honoured when in process context. | |
1da177e4 | 2258 | * |
6421ec76 MWO |
2259 | * Context: Can be called from any context, providing the appropriate GFP |
2260 | * flags are used. | |
2261 | * Return: The page on success or NULL if allocation fails. | |
1da177e4 | 2262 | */ |
d7f946d0 | 2263 | struct page *alloc_pages(gfp_t gfp, unsigned order) |
1da177e4 | 2264 | { |
8d90274b | 2265 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2266 | struct page *page; |
1da177e4 | 2267 | |
8d90274b ON |
2268 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2269 | pol = get_task_policy(current); | |
52cd3b07 LS |
2270 | |
2271 | /* | |
2272 | * No reference counting needed for current->mempolicy | |
2273 | * nor system default_policy | |
2274 | */ | |
45c4745a | 2275 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 | 2276 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
4c54d949 FT |
2277 | else if (pol->mode == MPOL_PREFERRED_MANY) |
2278 | page = alloc_pages_preferred_many(gfp, order, | |
c0455116 | 2279 | policy_node(gfp, pol, numa_node_id()), pol); |
c0ff7453 | 2280 | else |
84172f4b | 2281 | page = __alloc_pages(gfp, order, |
04ec6264 | 2282 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2283 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2284 | |
c0ff7453 | 2285 | return page; |
1da177e4 | 2286 | } |
d7f946d0 | 2287 | EXPORT_SYMBOL(alloc_pages); |
1da177e4 | 2288 | |
cc09cb13 MWO |
2289 | struct folio *folio_alloc(gfp_t gfp, unsigned order) |
2290 | { | |
2291 | struct page *page = alloc_pages(gfp | __GFP_COMP, order); | |
2292 | ||
2293 | if (page && order > 1) | |
2294 | prep_transhuge_page(page); | |
2295 | return (struct folio *)page; | |
2296 | } | |
2297 | EXPORT_SYMBOL(folio_alloc); | |
2298 | ||
c00b6b96 CW |
2299 | static unsigned long alloc_pages_bulk_array_interleave(gfp_t gfp, |
2300 | struct mempolicy *pol, unsigned long nr_pages, | |
2301 | struct page **page_array) | |
2302 | { | |
2303 | int nodes; | |
2304 | unsigned long nr_pages_per_node; | |
2305 | int delta; | |
2306 | int i; | |
2307 | unsigned long nr_allocated; | |
2308 | unsigned long total_allocated = 0; | |
2309 | ||
2310 | nodes = nodes_weight(pol->nodes); | |
2311 | nr_pages_per_node = nr_pages / nodes; | |
2312 | delta = nr_pages - nodes * nr_pages_per_node; | |
2313 | ||
2314 | for (i = 0; i < nodes; i++) { | |
2315 | if (delta) { | |
2316 | nr_allocated = __alloc_pages_bulk(gfp, | |
2317 | interleave_nodes(pol), NULL, | |
2318 | nr_pages_per_node + 1, NULL, | |
2319 | page_array); | |
2320 | delta--; | |
2321 | } else { | |
2322 | nr_allocated = __alloc_pages_bulk(gfp, | |
2323 | interleave_nodes(pol), NULL, | |
2324 | nr_pages_per_node, NULL, page_array); | |
2325 | } | |
2326 | ||
2327 | page_array += nr_allocated; | |
2328 | total_allocated += nr_allocated; | |
2329 | } | |
2330 | ||
2331 | return total_allocated; | |
2332 | } | |
2333 | ||
2334 | static unsigned long alloc_pages_bulk_array_preferred_many(gfp_t gfp, int nid, | |
2335 | struct mempolicy *pol, unsigned long nr_pages, | |
2336 | struct page **page_array) | |
2337 | { | |
2338 | gfp_t preferred_gfp; | |
2339 | unsigned long nr_allocated = 0; | |
2340 | ||
2341 | preferred_gfp = gfp | __GFP_NOWARN; | |
2342 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2343 | ||
2344 | nr_allocated = __alloc_pages_bulk(preferred_gfp, nid, &pol->nodes, | |
2345 | nr_pages, NULL, page_array); | |
2346 | ||
2347 | if (nr_allocated < nr_pages) | |
2348 | nr_allocated += __alloc_pages_bulk(gfp, numa_node_id(), NULL, | |
2349 | nr_pages - nr_allocated, NULL, | |
2350 | page_array + nr_allocated); | |
2351 | return nr_allocated; | |
2352 | } | |
2353 | ||
2354 | /* alloc pages bulk and mempolicy should be considered at the | |
2355 | * same time in some situation such as vmalloc. | |
2356 | * | |
2357 | * It can accelerate memory allocation especially interleaving | |
2358 | * allocate memory. | |
2359 | */ | |
2360 | unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, | |
2361 | unsigned long nr_pages, struct page **page_array) | |
2362 | { | |
2363 | struct mempolicy *pol = &default_policy; | |
2364 | ||
2365 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) | |
2366 | pol = get_task_policy(current); | |
2367 | ||
2368 | if (pol->mode == MPOL_INTERLEAVE) | |
2369 | return alloc_pages_bulk_array_interleave(gfp, pol, | |
2370 | nr_pages, page_array); | |
2371 | ||
2372 | if (pol->mode == MPOL_PREFERRED_MANY) | |
2373 | return alloc_pages_bulk_array_preferred_many(gfp, | |
2374 | numa_node_id(), pol, nr_pages, page_array); | |
2375 | ||
2376 | return __alloc_pages_bulk(gfp, policy_node(gfp, pol, numa_node_id()), | |
2377 | policy_nodemask(gfp, pol), nr_pages, NULL, | |
2378 | page_array); | |
2379 | } | |
2380 | ||
ef0855d3 ON |
2381 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2382 | { | |
2383 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2384 | ||
2385 | if (IS_ERR(pol)) | |
2386 | return PTR_ERR(pol); | |
2387 | dst->vm_policy = pol; | |
2388 | return 0; | |
2389 | } | |
2390 | ||
4225399a | 2391 | /* |
846a16bf | 2392 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2393 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2394 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2395 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2396 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2397 | * |
2398 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2399 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2400 | */ |
4225399a | 2401 | |
846a16bf LS |
2402 | /* Slow path of a mempolicy duplicate */ |
2403 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2404 | { |
2405 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2406 | ||
2407 | if (!new) | |
2408 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2409 | |
2410 | /* task's mempolicy is protected by alloc_lock */ | |
2411 | if (old == current->mempolicy) { | |
2412 | task_lock(current); | |
2413 | *new = *old; | |
2414 | task_unlock(current); | |
2415 | } else | |
2416 | *new = *old; | |
2417 | ||
4225399a PJ |
2418 | if (current_cpuset_is_being_rebound()) { |
2419 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2420 | mpol_rebind_policy(new, &mems); |
4225399a | 2421 | } |
1da177e4 | 2422 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2423 | return new; |
2424 | } | |
2425 | ||
2426 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2427 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2428 | { |
2429 | if (!a || !b) | |
fcfb4dcc | 2430 | return false; |
45c4745a | 2431 | if (a->mode != b->mode) |
fcfb4dcc | 2432 | return false; |
19800502 | 2433 | if (a->flags != b->flags) |
fcfb4dcc | 2434 | return false; |
c6018b4b AK |
2435 | if (a->home_node != b->home_node) |
2436 | return false; | |
19800502 BL |
2437 | if (mpol_store_user_nodemask(a)) |
2438 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2439 | return false; |
19800502 | 2440 | |
45c4745a | 2441 | switch (a->mode) { |
19770b32 | 2442 | case MPOL_BIND: |
1da177e4 | 2443 | case MPOL_INTERLEAVE: |
1da177e4 | 2444 | case MPOL_PREFERRED: |
b27abacc | 2445 | case MPOL_PREFERRED_MANY: |
269fbe72 | 2446 | return !!nodes_equal(a->nodes, b->nodes); |
7858d7bc FT |
2447 | case MPOL_LOCAL: |
2448 | return true; | |
1da177e4 LT |
2449 | default: |
2450 | BUG(); | |
fcfb4dcc | 2451 | return false; |
1da177e4 LT |
2452 | } |
2453 | } | |
2454 | ||
1da177e4 LT |
2455 | /* |
2456 | * Shared memory backing store policy support. | |
2457 | * | |
2458 | * Remember policies even when nobody has shared memory mapped. | |
2459 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2460 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2461 | * for any accesses to the tree. |
2462 | */ | |
2463 | ||
4a8c7bb5 NZ |
2464 | /* |
2465 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2466 | * reading or for writing | |
2467 | */ | |
1da177e4 LT |
2468 | static struct sp_node * |
2469 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2470 | { | |
2471 | struct rb_node *n = sp->root.rb_node; | |
2472 | ||
2473 | while (n) { | |
2474 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2475 | ||
2476 | if (start >= p->end) | |
2477 | n = n->rb_right; | |
2478 | else if (end <= p->start) | |
2479 | n = n->rb_left; | |
2480 | else | |
2481 | break; | |
2482 | } | |
2483 | if (!n) | |
2484 | return NULL; | |
2485 | for (;;) { | |
2486 | struct sp_node *w = NULL; | |
2487 | struct rb_node *prev = rb_prev(n); | |
2488 | if (!prev) | |
2489 | break; | |
2490 | w = rb_entry(prev, struct sp_node, nd); | |
2491 | if (w->end <= start) | |
2492 | break; | |
2493 | n = prev; | |
2494 | } | |
2495 | return rb_entry(n, struct sp_node, nd); | |
2496 | } | |
2497 | ||
4a8c7bb5 NZ |
2498 | /* |
2499 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2500 | * writing. | |
2501 | */ | |
1da177e4 LT |
2502 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2503 | { | |
2504 | struct rb_node **p = &sp->root.rb_node; | |
2505 | struct rb_node *parent = NULL; | |
2506 | struct sp_node *nd; | |
2507 | ||
2508 | while (*p) { | |
2509 | parent = *p; | |
2510 | nd = rb_entry(parent, struct sp_node, nd); | |
2511 | if (new->start < nd->start) | |
2512 | p = &(*p)->rb_left; | |
2513 | else if (new->end > nd->end) | |
2514 | p = &(*p)->rb_right; | |
2515 | else | |
2516 | BUG(); | |
2517 | } | |
2518 | rb_link_node(&new->nd, parent, p); | |
2519 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2520 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2521 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | /* Find shared policy intersecting idx */ | |
2525 | struct mempolicy * | |
2526 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2527 | { | |
2528 | struct mempolicy *pol = NULL; | |
2529 | struct sp_node *sn; | |
2530 | ||
2531 | if (!sp->root.rb_node) | |
2532 | return NULL; | |
4a8c7bb5 | 2533 | read_lock(&sp->lock); |
1da177e4 LT |
2534 | sn = sp_lookup(sp, idx, idx+1); |
2535 | if (sn) { | |
2536 | mpol_get(sn->policy); | |
2537 | pol = sn->policy; | |
2538 | } | |
4a8c7bb5 | 2539 | read_unlock(&sp->lock); |
1da177e4 LT |
2540 | return pol; |
2541 | } | |
2542 | ||
63f74ca2 KM |
2543 | static void sp_free(struct sp_node *n) |
2544 | { | |
2545 | mpol_put(n->policy); | |
2546 | kmem_cache_free(sn_cache, n); | |
2547 | } | |
2548 | ||
771fb4d8 LS |
2549 | /** |
2550 | * mpol_misplaced - check whether current page node is valid in policy | |
2551 | * | |
b46e14ac FF |
2552 | * @page: page to be checked |
2553 | * @vma: vm area where page mapped | |
2554 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2555 | * |
2556 | * Lookup current policy node id for vma,addr and "compare to" page's | |
5f076944 | 2557 | * node id. Policy determination "mimics" alloc_page_vma(). |
771fb4d8 | 2558 | * Called from fault path where we know the vma and faulting address. |
5f076944 | 2559 | * |
062db293 BW |
2560 | * Return: NUMA_NO_NODE if the page is in a node that is valid for this |
2561 | * policy, or a suitable node ID to allocate a replacement page from. | |
771fb4d8 LS |
2562 | */ |
2563 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2564 | { | |
2565 | struct mempolicy *pol; | |
c33d6c06 | 2566 | struct zoneref *z; |
771fb4d8 LS |
2567 | int curnid = page_to_nid(page); |
2568 | unsigned long pgoff; | |
90572890 PZ |
2569 | int thiscpu = raw_smp_processor_id(); |
2570 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2571 | int polnid = NUMA_NO_NODE; |
062db293 | 2572 | int ret = NUMA_NO_NODE; |
771fb4d8 | 2573 | |
dd6eecb9 | 2574 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2575 | if (!(pol->flags & MPOL_F_MOF)) |
2576 | goto out; | |
2577 | ||
2578 | switch (pol->mode) { | |
2579 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2580 | pgoff = vma->vm_pgoff; |
2581 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2582 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2583 | break; |
2584 | ||
2585 | case MPOL_PREFERRED: | |
b27abacc DH |
2586 | if (node_isset(curnid, pol->nodes)) |
2587 | goto out; | |
269fbe72 | 2588 | polnid = first_node(pol->nodes); |
7858d7bc FT |
2589 | break; |
2590 | ||
2591 | case MPOL_LOCAL: | |
2592 | polnid = numa_node_id(); | |
771fb4d8 LS |
2593 | break; |
2594 | ||
2595 | case MPOL_BIND: | |
bda420b9 HY |
2596 | /* Optimize placement among multiple nodes via NUMA balancing */ |
2597 | if (pol->flags & MPOL_F_MORON) { | |
269fbe72 | 2598 | if (node_isset(thisnid, pol->nodes)) |
bda420b9 HY |
2599 | break; |
2600 | goto out; | |
2601 | } | |
b27abacc | 2602 | fallthrough; |
c33d6c06 | 2603 | |
b27abacc | 2604 | case MPOL_PREFERRED_MANY: |
771fb4d8 | 2605 | /* |
771fb4d8 LS |
2606 | * use current page if in policy nodemask, |
2607 | * else select nearest allowed node, if any. | |
2608 | * If no allowed nodes, use current [!misplaced]. | |
2609 | */ | |
269fbe72 | 2610 | if (node_isset(curnid, pol->nodes)) |
771fb4d8 | 2611 | goto out; |
c33d6c06 | 2612 | z = first_zones_zonelist( |
771fb4d8 LS |
2613 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2614 | gfp_zone(GFP_HIGHUSER), | |
269fbe72 | 2615 | &pol->nodes); |
c1093b74 | 2616 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2617 | break; |
2618 | ||
2619 | default: | |
2620 | BUG(); | |
2621 | } | |
5606e387 MG |
2622 | |
2623 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2624 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2625 | polnid = thisnid; |
5606e387 | 2626 | |
10f39042 | 2627 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2628 | goto out; |
e42c8ff2 MG |
2629 | } |
2630 | ||
771fb4d8 LS |
2631 | if (curnid != polnid) |
2632 | ret = polnid; | |
2633 | out: | |
2634 | mpol_cond_put(pol); | |
2635 | ||
2636 | return ret; | |
2637 | } | |
2638 | ||
c11600e4 DR |
2639 | /* |
2640 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2641 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2642 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2643 | * policy. | |
2644 | */ | |
2645 | void mpol_put_task_policy(struct task_struct *task) | |
2646 | { | |
2647 | struct mempolicy *pol; | |
2648 | ||
2649 | task_lock(task); | |
2650 | pol = task->mempolicy; | |
2651 | task->mempolicy = NULL; | |
2652 | task_unlock(task); | |
2653 | mpol_put(pol); | |
2654 | } | |
2655 | ||
1da177e4 LT |
2656 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2657 | { | |
140d5a49 | 2658 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2659 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2660 | sp_free(n); |
1da177e4 LT |
2661 | } |
2662 | ||
42288fe3 MG |
2663 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2664 | unsigned long end, struct mempolicy *pol) | |
2665 | { | |
2666 | node->start = start; | |
2667 | node->end = end; | |
2668 | node->policy = pol; | |
2669 | } | |
2670 | ||
dbcb0f19 AB |
2671 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2672 | struct mempolicy *pol) | |
1da177e4 | 2673 | { |
869833f2 KM |
2674 | struct sp_node *n; |
2675 | struct mempolicy *newpol; | |
1da177e4 | 2676 | |
869833f2 | 2677 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2678 | if (!n) |
2679 | return NULL; | |
869833f2 KM |
2680 | |
2681 | newpol = mpol_dup(pol); | |
2682 | if (IS_ERR(newpol)) { | |
2683 | kmem_cache_free(sn_cache, n); | |
2684 | return NULL; | |
2685 | } | |
2686 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2687 | sp_node_init(n, start, end, newpol); |
869833f2 | 2688 | |
1da177e4 LT |
2689 | return n; |
2690 | } | |
2691 | ||
2692 | /* Replace a policy range. */ | |
2693 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2694 | unsigned long end, struct sp_node *new) | |
2695 | { | |
b22d127a | 2696 | struct sp_node *n; |
42288fe3 MG |
2697 | struct sp_node *n_new = NULL; |
2698 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2699 | int ret = 0; |
1da177e4 | 2700 | |
42288fe3 | 2701 | restart: |
4a8c7bb5 | 2702 | write_lock(&sp->lock); |
1da177e4 LT |
2703 | n = sp_lookup(sp, start, end); |
2704 | /* Take care of old policies in the same range. */ | |
2705 | while (n && n->start < end) { | |
2706 | struct rb_node *next = rb_next(&n->nd); | |
2707 | if (n->start >= start) { | |
2708 | if (n->end <= end) | |
2709 | sp_delete(sp, n); | |
2710 | else | |
2711 | n->start = end; | |
2712 | } else { | |
2713 | /* Old policy spanning whole new range. */ | |
2714 | if (n->end > end) { | |
42288fe3 MG |
2715 | if (!n_new) |
2716 | goto alloc_new; | |
2717 | ||
2718 | *mpol_new = *n->policy; | |
2719 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2720 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2721 | n->end = start; |
5ca39575 | 2722 | sp_insert(sp, n_new); |
42288fe3 MG |
2723 | n_new = NULL; |
2724 | mpol_new = NULL; | |
1da177e4 LT |
2725 | break; |
2726 | } else | |
2727 | n->end = start; | |
2728 | } | |
2729 | if (!next) | |
2730 | break; | |
2731 | n = rb_entry(next, struct sp_node, nd); | |
2732 | } | |
2733 | if (new) | |
2734 | sp_insert(sp, new); | |
4a8c7bb5 | 2735 | write_unlock(&sp->lock); |
42288fe3 MG |
2736 | ret = 0; |
2737 | ||
2738 | err_out: | |
2739 | if (mpol_new) | |
2740 | mpol_put(mpol_new); | |
2741 | if (n_new) | |
2742 | kmem_cache_free(sn_cache, n_new); | |
2743 | ||
b22d127a | 2744 | return ret; |
42288fe3 MG |
2745 | |
2746 | alloc_new: | |
4a8c7bb5 | 2747 | write_unlock(&sp->lock); |
42288fe3 MG |
2748 | ret = -ENOMEM; |
2749 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2750 | if (!n_new) | |
2751 | goto err_out; | |
2752 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2753 | if (!mpol_new) | |
2754 | goto err_out; | |
4ad09955 | 2755 | atomic_set(&mpol_new->refcnt, 1); |
42288fe3 | 2756 | goto restart; |
1da177e4 LT |
2757 | } |
2758 | ||
71fe804b LS |
2759 | /** |
2760 | * mpol_shared_policy_init - initialize shared policy for inode | |
2761 | * @sp: pointer to inode shared policy | |
2762 | * @mpol: struct mempolicy to install | |
2763 | * | |
2764 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2765 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2766 | * This must be released on exit. | |
4bfc4495 | 2767 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2768 | */ |
2769 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2770 | { | |
58568d2a MX |
2771 | int ret; |
2772 | ||
71fe804b | 2773 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2774 | rwlock_init(&sp->lock); |
71fe804b LS |
2775 | |
2776 | if (mpol) { | |
2777 | struct vm_area_struct pvma; | |
2778 | struct mempolicy *new; | |
4bfc4495 | 2779 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2780 | |
4bfc4495 | 2781 | if (!scratch) |
5c0c1654 | 2782 | goto put_mpol; |
71fe804b LS |
2783 | /* contextualize the tmpfs mount point mempolicy */ |
2784 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2785 | if (IS_ERR(new)) |
0cae3457 | 2786 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2787 | |
2788 | task_lock(current); | |
4bfc4495 | 2789 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2790 | task_unlock(current); |
15d77835 | 2791 | if (ret) |
5c0c1654 | 2792 | goto put_new; |
71fe804b LS |
2793 | |
2794 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2795 | vma_init(&pvma, NULL); |
71fe804b LS |
2796 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2797 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2798 | |
5c0c1654 | 2799 | put_new: |
71fe804b | 2800 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2801 | free_scratch: |
4bfc4495 | 2802 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2803 | put_mpol: |
2804 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2805 | } |
2806 | } | |
2807 | ||
1da177e4 LT |
2808 | int mpol_set_shared_policy(struct shared_policy *info, |
2809 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2810 | { | |
2811 | int err; | |
2812 | struct sp_node *new = NULL; | |
2813 | unsigned long sz = vma_pages(vma); | |
2814 | ||
028fec41 | 2815 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2816 | vma->vm_pgoff, |
45c4745a | 2817 | sz, npol ? npol->mode : -1, |
028fec41 | 2818 | npol ? npol->flags : -1, |
269fbe72 | 2819 | npol ? nodes_addr(npol->nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2820 | |
2821 | if (npol) { | |
2822 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2823 | if (!new) | |
2824 | return -ENOMEM; | |
2825 | } | |
2826 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2827 | if (err && new) | |
63f74ca2 | 2828 | sp_free(new); |
1da177e4 LT |
2829 | return err; |
2830 | } | |
2831 | ||
2832 | /* Free a backing policy store on inode delete. */ | |
2833 | void mpol_free_shared_policy(struct shared_policy *p) | |
2834 | { | |
2835 | struct sp_node *n; | |
2836 | struct rb_node *next; | |
2837 | ||
2838 | if (!p->root.rb_node) | |
2839 | return; | |
4a8c7bb5 | 2840 | write_lock(&p->lock); |
1da177e4 LT |
2841 | next = rb_first(&p->root); |
2842 | while (next) { | |
2843 | n = rb_entry(next, struct sp_node, nd); | |
2844 | next = rb_next(&n->nd); | |
63f74ca2 | 2845 | sp_delete(p, n); |
1da177e4 | 2846 | } |
4a8c7bb5 | 2847 | write_unlock(&p->lock); |
1da177e4 LT |
2848 | } |
2849 | ||
1a687c2e | 2850 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2851 | static int __initdata numabalancing_override; |
1a687c2e MG |
2852 | |
2853 | static void __init check_numabalancing_enable(void) | |
2854 | { | |
2855 | bool numabalancing_default = false; | |
2856 | ||
2857 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2858 | numabalancing_default = true; | |
2859 | ||
c297663c MG |
2860 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2861 | if (numabalancing_override) | |
2862 | set_numabalancing_state(numabalancing_override == 1); | |
2863 | ||
b0dc2b9b | 2864 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2865 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2866 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2867 | set_numabalancing_state(numabalancing_default); |
2868 | } | |
2869 | } | |
2870 | ||
2871 | static int __init setup_numabalancing(char *str) | |
2872 | { | |
2873 | int ret = 0; | |
2874 | if (!str) | |
2875 | goto out; | |
1a687c2e MG |
2876 | |
2877 | if (!strcmp(str, "enable")) { | |
c297663c | 2878 | numabalancing_override = 1; |
1a687c2e MG |
2879 | ret = 1; |
2880 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2881 | numabalancing_override = -1; |
1a687c2e MG |
2882 | ret = 1; |
2883 | } | |
2884 | out: | |
2885 | if (!ret) | |
4a404bea | 2886 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2887 | |
2888 | return ret; | |
2889 | } | |
2890 | __setup("numa_balancing=", setup_numabalancing); | |
2891 | #else | |
2892 | static inline void __init check_numabalancing_enable(void) | |
2893 | { | |
2894 | } | |
2895 | #endif /* CONFIG_NUMA_BALANCING */ | |
2896 | ||
1da177e4 LT |
2897 | /* assumes fs == KERNEL_DS */ |
2898 | void __init numa_policy_init(void) | |
2899 | { | |
b71636e2 PM |
2900 | nodemask_t interleave_nodes; |
2901 | unsigned long largest = 0; | |
2902 | int nid, prefer = 0; | |
2903 | ||
1da177e4 LT |
2904 | policy_cache = kmem_cache_create("numa_policy", |
2905 | sizeof(struct mempolicy), | |
20c2df83 | 2906 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2907 | |
2908 | sn_cache = kmem_cache_create("shared_policy_node", | |
2909 | sizeof(struct sp_node), | |
20c2df83 | 2910 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2911 | |
5606e387 MG |
2912 | for_each_node(nid) { |
2913 | preferred_node_policy[nid] = (struct mempolicy) { | |
2914 | .refcnt = ATOMIC_INIT(1), | |
2915 | .mode = MPOL_PREFERRED, | |
2916 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
269fbe72 | 2917 | .nodes = nodemask_of_node(nid), |
5606e387 MG |
2918 | }; |
2919 | } | |
2920 | ||
b71636e2 PM |
2921 | /* |
2922 | * Set interleaving policy for system init. Interleaving is only | |
2923 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2924 | * fall back to the largest node if they're all smaller. | |
2925 | */ | |
2926 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2927 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2928 | unsigned long total_pages = node_present_pages(nid); |
2929 | ||
2930 | /* Preserve the largest node */ | |
2931 | if (largest < total_pages) { | |
2932 | largest = total_pages; | |
2933 | prefer = nid; | |
2934 | } | |
2935 | ||
2936 | /* Interleave this node? */ | |
2937 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2938 | node_set(nid, interleave_nodes); | |
2939 | } | |
2940 | ||
2941 | /* All too small, use the largest */ | |
2942 | if (unlikely(nodes_empty(interleave_nodes))) | |
2943 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2944 | |
028fec41 | 2945 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2946 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2947 | |
2948 | check_numabalancing_enable(); | |
1da177e4 LT |
2949 | } |
2950 | ||
8bccd85f | 2951 | /* Reset policy of current process to default */ |
1da177e4 LT |
2952 | void numa_default_policy(void) |
2953 | { | |
028fec41 | 2954 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2955 | } |
68860ec1 | 2956 | |
095f1fc4 LS |
2957 | /* |
2958 | * Parse and format mempolicy from/to strings | |
2959 | */ | |
2960 | ||
345ace9c LS |
2961 | static const char * const policy_modes[] = |
2962 | { | |
2963 | [MPOL_DEFAULT] = "default", | |
2964 | [MPOL_PREFERRED] = "prefer", | |
2965 | [MPOL_BIND] = "bind", | |
2966 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2967 | [MPOL_LOCAL] = "local", |
b27abacc | 2968 | [MPOL_PREFERRED_MANY] = "prefer (many)", |
345ace9c | 2969 | }; |
1a75a6c8 | 2970 | |
095f1fc4 LS |
2971 | |
2972 | #ifdef CONFIG_TMPFS | |
2973 | /** | |
f2a07f40 | 2974 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2975 | * @str: string containing mempolicy to parse |
71fe804b | 2976 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2977 | * |
2978 | * Format of input: | |
2979 | * <mode>[=<flags>][:<nodelist>] | |
2980 | * | |
dad5b023 | 2981 | * Return: %0 on success, else %1 |
095f1fc4 | 2982 | */ |
a7a88b23 | 2983 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2984 | { |
71fe804b | 2985 | struct mempolicy *new = NULL; |
f2a07f40 | 2986 | unsigned short mode_flags; |
71fe804b | 2987 | nodemask_t nodes; |
095f1fc4 LS |
2988 | char *nodelist = strchr(str, ':'); |
2989 | char *flags = strchr(str, '='); | |
dedf2c73 | 2990 | int err = 1, mode; |
095f1fc4 | 2991 | |
c7a91bc7 DC |
2992 | if (flags) |
2993 | *flags++ = '\0'; /* terminate mode string */ | |
2994 | ||
095f1fc4 LS |
2995 | if (nodelist) { |
2996 | /* NUL-terminate mode or flags string */ | |
2997 | *nodelist++ = '\0'; | |
71fe804b | 2998 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2999 | goto out; |
01f13bd6 | 3000 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 3001 | goto out; |
71fe804b LS |
3002 | } else |
3003 | nodes_clear(nodes); | |
3004 | ||
dedf2c73 | 3005 | mode = match_string(policy_modes, MPOL_MAX, str); |
3006 | if (mode < 0) | |
095f1fc4 LS |
3007 | goto out; |
3008 | ||
71fe804b | 3009 | switch (mode) { |
095f1fc4 | 3010 | case MPOL_PREFERRED: |
71fe804b | 3011 | /* |
aa9f7d51 RD |
3012 | * Insist on a nodelist of one node only, although later |
3013 | * we use first_node(nodes) to grab a single node, so here | |
3014 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 3015 | */ |
095f1fc4 LS |
3016 | if (nodelist) { |
3017 | char *rest = nodelist; | |
3018 | while (isdigit(*rest)) | |
3019 | rest++; | |
926f2ae0 KM |
3020 | if (*rest) |
3021 | goto out; | |
aa9f7d51 RD |
3022 | if (nodes_empty(nodes)) |
3023 | goto out; | |
095f1fc4 LS |
3024 | } |
3025 | break; | |
095f1fc4 LS |
3026 | case MPOL_INTERLEAVE: |
3027 | /* | |
3028 | * Default to online nodes with memory if no nodelist | |
3029 | */ | |
3030 | if (!nodelist) | |
01f13bd6 | 3031 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 3032 | break; |
71fe804b | 3033 | case MPOL_LOCAL: |
3f226aa1 | 3034 | /* |
71fe804b | 3035 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 3036 | */ |
71fe804b | 3037 | if (nodelist) |
3f226aa1 | 3038 | goto out; |
3f226aa1 | 3039 | break; |
413b43de RT |
3040 | case MPOL_DEFAULT: |
3041 | /* | |
3042 | * Insist on a empty nodelist | |
3043 | */ | |
3044 | if (!nodelist) | |
3045 | err = 0; | |
3046 | goto out; | |
b27abacc | 3047 | case MPOL_PREFERRED_MANY: |
d69b2e63 KM |
3048 | case MPOL_BIND: |
3049 | /* | |
3050 | * Insist on a nodelist | |
3051 | */ | |
3052 | if (!nodelist) | |
3053 | goto out; | |
095f1fc4 LS |
3054 | } |
3055 | ||
71fe804b | 3056 | mode_flags = 0; |
095f1fc4 LS |
3057 | if (flags) { |
3058 | /* | |
3059 | * Currently, we only support two mutually exclusive | |
3060 | * mode flags. | |
3061 | */ | |
3062 | if (!strcmp(flags, "static")) | |
71fe804b | 3063 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 3064 | else if (!strcmp(flags, "relative")) |
71fe804b | 3065 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 3066 | else |
926f2ae0 | 3067 | goto out; |
095f1fc4 | 3068 | } |
71fe804b LS |
3069 | |
3070 | new = mpol_new(mode, mode_flags, &nodes); | |
3071 | if (IS_ERR(new)) | |
926f2ae0 KM |
3072 | goto out; |
3073 | ||
f2a07f40 HD |
3074 | /* |
3075 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
3076 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
3077 | */ | |
269fbe72 BW |
3078 | if (mode != MPOL_PREFERRED) { |
3079 | new->nodes = nodes; | |
3080 | } else if (nodelist) { | |
3081 | nodes_clear(new->nodes); | |
3082 | node_set(first_node(nodes), new->nodes); | |
3083 | } else { | |
7858d7bc | 3084 | new->mode = MPOL_LOCAL; |
269fbe72 | 3085 | } |
f2a07f40 HD |
3086 | |
3087 | /* | |
3088 | * Save nodes for contextualization: this will be used to "clone" | |
3089 | * the mempolicy in a specific context [cpuset] at a later time. | |
3090 | */ | |
3091 | new->w.user_nodemask = nodes; | |
3092 | ||
926f2ae0 | 3093 | err = 0; |
71fe804b | 3094 | |
095f1fc4 LS |
3095 | out: |
3096 | /* Restore string for error message */ | |
3097 | if (nodelist) | |
3098 | *--nodelist = ':'; | |
3099 | if (flags) | |
3100 | *--flags = '='; | |
71fe804b LS |
3101 | if (!err) |
3102 | *mpol = new; | |
095f1fc4 LS |
3103 | return err; |
3104 | } | |
3105 | #endif /* CONFIG_TMPFS */ | |
3106 | ||
71fe804b LS |
3107 | /** |
3108 | * mpol_to_str - format a mempolicy structure for printing | |
3109 | * @buffer: to contain formatted mempolicy string | |
3110 | * @maxlen: length of @buffer | |
3111 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3112 | * |
948927ee DR |
3113 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3114 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3115 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3116 | */ |
948927ee | 3117 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3118 | { |
3119 | char *p = buffer; | |
948927ee DR |
3120 | nodemask_t nodes = NODE_MASK_NONE; |
3121 | unsigned short mode = MPOL_DEFAULT; | |
3122 | unsigned short flags = 0; | |
2291990a | 3123 | |
8790c71a | 3124 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3125 | mode = pol->mode; |
948927ee DR |
3126 | flags = pol->flags; |
3127 | } | |
bea904d5 | 3128 | |
1a75a6c8 CL |
3129 | switch (mode) { |
3130 | case MPOL_DEFAULT: | |
7858d7bc | 3131 | case MPOL_LOCAL: |
1a75a6c8 | 3132 | break; |
1a75a6c8 | 3133 | case MPOL_PREFERRED: |
b27abacc | 3134 | case MPOL_PREFERRED_MANY: |
1a75a6c8 | 3135 | case MPOL_BIND: |
1a75a6c8 | 3136 | case MPOL_INTERLEAVE: |
269fbe72 | 3137 | nodes = pol->nodes; |
1a75a6c8 | 3138 | break; |
1a75a6c8 | 3139 | default: |
948927ee DR |
3140 | WARN_ON_ONCE(1); |
3141 | snprintf(p, maxlen, "unknown"); | |
3142 | return; | |
1a75a6c8 CL |
3143 | } |
3144 | ||
b7a9f420 | 3145 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3146 | |
fc36b8d3 | 3147 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3148 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3149 | |
2291990a LS |
3150 | /* |
3151 | * Currently, the only defined flags are mutually exclusive | |
3152 | */ | |
f5b087b5 | 3153 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3154 | p += snprintf(p, buffer + maxlen - p, "static"); |
3155 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3156 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3157 | } |
3158 | ||
9e763e0f TH |
3159 | if (!nodes_empty(nodes)) |
3160 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3161 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3162 | } |