]>
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> |
7c0f6ba6 | 107 | #include <linux/uaccess.h> |
1da177e4 | 108 | |
62695a84 NP |
109 | #include "internal.h" |
110 | ||
38e35860 | 111 | /* Internal flags */ |
dc9aa5b9 | 112 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 113 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 114 | |
fcc234f8 PE |
115 | static struct kmem_cache *policy_cache; |
116 | static struct kmem_cache *sn_cache; | |
1da177e4 | 117 | |
1da177e4 LT |
118 | /* Highest zone. An specific allocation for a zone below that is not |
119 | policied. */ | |
6267276f | 120 | enum zone_type policy_zone = 0; |
1da177e4 | 121 | |
bea904d5 LS |
122 | /* |
123 | * run-time system-wide default policy => local allocation | |
124 | */ | |
e754d79d | 125 | static struct mempolicy default_policy = { |
1da177e4 | 126 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
7858d7bc | 127 | .mode = MPOL_LOCAL, |
1da177e4 LT |
128 | }; |
129 | ||
5606e387 MG |
130 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
131 | ||
b2ca916c DW |
132 | /** |
133 | * numa_map_to_online_node - Find closest online node | |
f6e92f40 | 134 | * @node: Node id to start the search |
b2ca916c DW |
135 | * |
136 | * Lookup the next closest node by distance if @nid is not online. | |
137 | */ | |
138 | int numa_map_to_online_node(int node) | |
139 | { | |
4fcbe96e | 140 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 141 | |
4fcbe96e DW |
142 | if (node == NUMA_NO_NODE || node_online(node)) |
143 | return node; | |
b2ca916c DW |
144 | |
145 | min_node = node; | |
4fcbe96e DW |
146 | for_each_online_node(n) { |
147 | dist = node_distance(node, n); | |
148 | if (dist < min_dist) { | |
149 | min_dist = dist; | |
150 | min_node = n; | |
b2ca916c DW |
151 | } |
152 | } | |
153 | ||
154 | return min_node; | |
155 | } | |
156 | EXPORT_SYMBOL_GPL(numa_map_to_online_node); | |
157 | ||
74d2c3a0 | 158 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
159 | { |
160 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 161 | int node; |
5606e387 | 162 | |
f15ca78e ON |
163 | if (pol) |
164 | return pol; | |
5606e387 | 165 | |
f15ca78e ON |
166 | node = numa_node_id(); |
167 | if (node != NUMA_NO_NODE) { | |
168 | pol = &preferred_node_policy[node]; | |
169 | /* preferred_node_policy is not initialised early in boot */ | |
170 | if (pol->mode) | |
171 | return pol; | |
5606e387 MG |
172 | } |
173 | ||
f15ca78e | 174 | return &default_policy; |
5606e387 MG |
175 | } |
176 | ||
37012946 DR |
177 | static const struct mempolicy_operations { |
178 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 179 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
180 | } mpol_ops[MPOL_MAX]; |
181 | ||
f5b087b5 DR |
182 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
183 | { | |
6d556294 | 184 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
185 | } |
186 | ||
187 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
188 | const nodemask_t *rel) | |
189 | { | |
190 | nodemask_t tmp; | |
191 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
192 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
193 | } |
194 | ||
37012946 DR |
195 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
196 | { | |
197 | if (nodes_empty(*nodes)) | |
198 | return -EINVAL; | |
269fbe72 | 199 | pol->nodes = *nodes; |
37012946 DR |
200 | return 0; |
201 | } | |
202 | ||
203 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
204 | { | |
7858d7bc FT |
205 | if (nodes_empty(*nodes)) |
206 | return -EINVAL; | |
269fbe72 BW |
207 | |
208 | nodes_clear(pol->nodes); | |
209 | node_set(first_node(*nodes), pol->nodes); | |
37012946 DR |
210 | return 0; |
211 | } | |
212 | ||
b27abacc DH |
213 | static int mpol_new_preferred_many(struct mempolicy *pol, const nodemask_t *nodes) |
214 | { | |
215 | if (nodes_empty(*nodes)) | |
216 | return -EINVAL; | |
217 | pol->nodes = *nodes; | |
218 | return 0; | |
219 | } | |
220 | ||
37012946 DR |
221 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) |
222 | { | |
859f7ef1 | 223 | if (nodes_empty(*nodes)) |
37012946 | 224 | return -EINVAL; |
269fbe72 | 225 | pol->nodes = *nodes; |
37012946 DR |
226 | return 0; |
227 | } | |
228 | ||
58568d2a MX |
229 | /* |
230 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
231 | * any, for the new policy. mpol_new() has already validated the nodes | |
7858d7bc | 232 | * parameter with respect to the policy mode and flags. |
58568d2a MX |
233 | * |
234 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
c1e8d7c6 | 235 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 236 | */ |
4bfc4495 KH |
237 | static int mpol_set_nodemask(struct mempolicy *pol, |
238 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 239 | { |
58568d2a MX |
240 | int ret; |
241 | ||
7858d7bc FT |
242 | /* |
243 | * Default (pol==NULL) resp. local memory policies are not a | |
244 | * subject of any remapping. They also do not need any special | |
245 | * constructor. | |
246 | */ | |
247 | if (!pol || pol->mode == MPOL_LOCAL) | |
58568d2a | 248 | return 0; |
7858d7bc | 249 | |
01f13bd6 | 250 | /* Check N_MEMORY */ |
4bfc4495 | 251 | nodes_and(nsc->mask1, |
01f13bd6 | 252 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
253 | |
254 | VM_BUG_ON(!nodes); | |
4bfc4495 | 255 | |
7858d7bc FT |
256 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
257 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); | |
258 | else | |
259 | nodes_and(nsc->mask2, *nodes, nsc->mask1); | |
58568d2a | 260 | |
7858d7bc FT |
261 | if (mpol_store_user_nodemask(pol)) |
262 | pol->w.user_nodemask = *nodes; | |
4bfc4495 | 263 | else |
7858d7bc FT |
264 | pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed; |
265 | ||
266 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
58568d2a MX |
267 | return ret; |
268 | } | |
269 | ||
270 | /* | |
271 | * This function just creates a new policy, does some check and simple | |
272 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
273 | */ | |
028fec41 DR |
274 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
275 | nodemask_t *nodes) | |
1da177e4 LT |
276 | { |
277 | struct mempolicy *policy; | |
278 | ||
028fec41 | 279 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 280 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 281 | |
3e1f0645 DR |
282 | if (mode == MPOL_DEFAULT) { |
283 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 284 | return ERR_PTR(-EINVAL); |
d3a71033 | 285 | return NULL; |
37012946 | 286 | } |
3e1f0645 DR |
287 | VM_BUG_ON(!nodes); |
288 | ||
289 | /* | |
290 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
291 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
292 | * All other modes require a valid pointer to a non-empty nodemask. | |
293 | */ | |
294 | if (mode == MPOL_PREFERRED) { | |
295 | if (nodes_empty(*nodes)) { | |
296 | if (((flags & MPOL_F_STATIC_NODES) || | |
297 | (flags & MPOL_F_RELATIVE_NODES))) | |
298 | return ERR_PTR(-EINVAL); | |
7858d7bc FT |
299 | |
300 | mode = MPOL_LOCAL; | |
3e1f0645 | 301 | } |
479e2802 | 302 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
303 | if (!nodes_empty(*nodes) || |
304 | (flags & MPOL_F_STATIC_NODES) || | |
305 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 | 306 | return ERR_PTR(-EINVAL); |
3e1f0645 DR |
307 | } else if (nodes_empty(*nodes)) |
308 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
309 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
310 | if (!policy) | |
311 | return ERR_PTR(-ENOMEM); | |
312 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 313 | policy->mode = mode; |
3e1f0645 | 314 | policy->flags = flags; |
37012946 | 315 | |
1da177e4 | 316 | return policy; |
37012946 DR |
317 | } |
318 | ||
52cd3b07 LS |
319 | /* Slow path of a mpol destructor. */ |
320 | void __mpol_put(struct mempolicy *p) | |
321 | { | |
322 | if (!atomic_dec_and_test(&p->refcnt)) | |
323 | return; | |
52cd3b07 LS |
324 | kmem_cache_free(policy_cache, p); |
325 | } | |
326 | ||
213980c0 | 327 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
328 | { |
329 | } | |
330 | ||
213980c0 | 331 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
332 | { |
333 | nodemask_t tmp; | |
334 | ||
335 | if (pol->flags & MPOL_F_STATIC_NODES) | |
336 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
337 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
338 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
339 | else { | |
269fbe72 | 340 | nodes_remap(tmp, pol->nodes, pol->w.cpuset_mems_allowed, |
213980c0 | 341 | *nodes); |
29b190fa | 342 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 343 | } |
f5b087b5 | 344 | |
708c1bbc MX |
345 | if (nodes_empty(tmp)) |
346 | tmp = *nodes; | |
347 | ||
269fbe72 | 348 | pol->nodes = tmp; |
37012946 DR |
349 | } |
350 | ||
351 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 352 | const nodemask_t *nodes) |
37012946 | 353 | { |
7858d7bc | 354 | pol->w.cpuset_mems_allowed = *nodes; |
1da177e4 LT |
355 | } |
356 | ||
708c1bbc MX |
357 | /* |
358 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
359 | * | |
c1e8d7c6 | 360 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
361 | * policies are protected by task->mems_allowed_seq to prevent a premature |
362 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 363 | */ |
213980c0 | 364 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 365 | { |
1d0d2680 DR |
366 | if (!pol) |
367 | return; | |
7858d7bc | 368 | if (!mpol_store_user_nodemask(pol) && |
1d0d2680 DR |
369 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
370 | return; | |
708c1bbc | 371 | |
213980c0 | 372 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
373 | } |
374 | ||
375 | /* | |
376 | * Wrapper for mpol_rebind_policy() that just requires task | |
377 | * pointer, and updates task mempolicy. | |
58568d2a MX |
378 | * |
379 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
380 | */ |
381 | ||
213980c0 | 382 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 383 | { |
213980c0 | 384 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
385 | } |
386 | ||
387 | /* | |
388 | * Rebind each vma in mm to new nodemask. | |
389 | * | |
c1e8d7c6 | 390 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 DR |
391 | */ |
392 | ||
393 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
394 | { | |
395 | struct vm_area_struct *vma; | |
396 | ||
d8ed45c5 | 397 | mmap_write_lock(mm); |
1d0d2680 | 398 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
213980c0 | 399 | mpol_rebind_policy(vma->vm_policy, new); |
d8ed45c5 | 400 | mmap_write_unlock(mm); |
1d0d2680 DR |
401 | } |
402 | ||
37012946 DR |
403 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
404 | [MPOL_DEFAULT] = { | |
405 | .rebind = mpol_rebind_default, | |
406 | }, | |
407 | [MPOL_INTERLEAVE] = { | |
408 | .create = mpol_new_interleave, | |
409 | .rebind = mpol_rebind_nodemask, | |
410 | }, | |
411 | [MPOL_PREFERRED] = { | |
412 | .create = mpol_new_preferred, | |
413 | .rebind = mpol_rebind_preferred, | |
414 | }, | |
415 | [MPOL_BIND] = { | |
416 | .create = mpol_new_bind, | |
417 | .rebind = mpol_rebind_nodemask, | |
418 | }, | |
7858d7bc FT |
419 | [MPOL_LOCAL] = { |
420 | .rebind = mpol_rebind_default, | |
421 | }, | |
b27abacc DH |
422 | [MPOL_PREFERRED_MANY] = { |
423 | .create = mpol_new_preferred_many, | |
424 | .rebind = mpol_rebind_preferred, | |
425 | }, | |
37012946 DR |
426 | }; |
427 | ||
a53190a4 | 428 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 429 | unsigned long flags); |
1a75a6c8 | 430 | |
6f4576e3 NH |
431 | struct queue_pages { |
432 | struct list_head *pagelist; | |
433 | unsigned long flags; | |
434 | nodemask_t *nmask; | |
f18da660 LX |
435 | unsigned long start; |
436 | unsigned long end; | |
437 | struct vm_area_struct *first; | |
6f4576e3 NH |
438 | }; |
439 | ||
88aaa2a1 NH |
440 | /* |
441 | * Check if the page's nid is in qp->nmask. | |
442 | * | |
443 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
444 | * in the invert of qp->nmask. | |
445 | */ | |
446 | static inline bool queue_pages_required(struct page *page, | |
447 | struct queue_pages *qp) | |
448 | { | |
449 | int nid = page_to_nid(page); | |
450 | unsigned long flags = qp->flags; | |
451 | ||
452 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
453 | } | |
454 | ||
a7f40cfe | 455 | /* |
d8835445 | 456 | * queue_pages_pmd() has four possible return values: |
e5947d23 YS |
457 | * 0 - pages are placed on the right node or queued successfully, or |
458 | * special page is met, i.e. huge zero page. | |
d8835445 YS |
459 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
460 | * specified. | |
461 | * 2 - THP was split. | |
462 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an | |
463 | * existing page was already on a node that does not follow the | |
464 | * policy. | |
a7f40cfe | 465 | */ |
c8633798 NH |
466 | static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
467 | unsigned long end, struct mm_walk *walk) | |
959a7e13 | 468 | __releases(ptl) |
c8633798 NH |
469 | { |
470 | int ret = 0; | |
471 | struct page *page; | |
472 | struct queue_pages *qp = walk->private; | |
473 | unsigned long flags; | |
474 | ||
475 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 476 | ret = -EIO; |
c8633798 NH |
477 | goto unlock; |
478 | } | |
479 | page = pmd_page(*pmd); | |
480 | if (is_huge_zero_page(page)) { | |
481 | spin_unlock(ptl); | |
e5947d23 | 482 | walk->action = ACTION_CONTINUE; |
c8633798 NH |
483 | goto out; |
484 | } | |
d8835445 | 485 | if (!queue_pages_required(page, qp)) |
c8633798 | 486 | goto unlock; |
c8633798 | 487 | |
c8633798 NH |
488 | flags = qp->flags; |
489 | /* go to thp migration */ | |
a7f40cfe | 490 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 YS |
491 | if (!vma_migratable(walk->vma) || |
492 | migrate_page_add(page, qp->pagelist, flags)) { | |
d8835445 | 493 | ret = 1; |
a7f40cfe YS |
494 | goto unlock; |
495 | } | |
a7f40cfe YS |
496 | } else |
497 | ret = -EIO; | |
c8633798 NH |
498 | unlock: |
499 | spin_unlock(ptl); | |
500 | out: | |
501 | return ret; | |
502 | } | |
503 | ||
98094945 NH |
504 | /* |
505 | * Scan through pages checking if pages follow certain conditions, | |
506 | * and move them to the pagelist if they do. | |
d8835445 YS |
507 | * |
508 | * queue_pages_pte_range() has three possible return values: | |
e5947d23 YS |
509 | * 0 - pages are placed on the right node or queued successfully, or |
510 | * special page is met, i.e. zero page. | |
d8835445 YS |
511 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
512 | * specified. | |
513 | * -EIO - only MPOL_MF_STRICT was specified and an existing page was already | |
514 | * on a node that does not follow the policy. | |
98094945 | 515 | */ |
6f4576e3 NH |
516 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
517 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 518 | { |
6f4576e3 NH |
519 | struct vm_area_struct *vma = walk->vma; |
520 | struct page *page; | |
521 | struct queue_pages *qp = walk->private; | |
522 | unsigned long flags = qp->flags; | |
c8633798 | 523 | int ret; |
d8835445 | 524 | bool has_unmovable = false; |
3f088420 | 525 | pte_t *pte, *mapped_pte; |
705e87c0 | 526 | spinlock_t *ptl; |
941150a3 | 527 | |
c8633798 NH |
528 | ptl = pmd_trans_huge_lock(pmd, vma); |
529 | if (ptl) { | |
530 | ret = queue_pages_pmd(pmd, ptl, addr, end, walk); | |
d8835445 | 531 | if (ret != 2) |
a7f40cfe | 532 | return ret; |
248db92d | 533 | } |
d8835445 | 534 | /* THP was split, fall through to pte walk */ |
91612e0d | 535 | |
337d9abf NH |
536 | if (pmd_trans_unstable(pmd)) |
537 | return 0; | |
94723aaf | 538 | |
3f088420 | 539 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
6f4576e3 | 540 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
91612e0d | 541 | if (!pte_present(*pte)) |
1da177e4 | 542 | continue; |
6aab341e LT |
543 | page = vm_normal_page(vma, addr, *pte); |
544 | if (!page) | |
1da177e4 | 545 | continue; |
053837fc | 546 | /* |
62b61f61 HD |
547 | * vm_normal_page() filters out zero pages, but there might |
548 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 549 | */ |
b79bc0a0 | 550 | if (PageReserved(page)) |
f4598c8b | 551 | continue; |
88aaa2a1 | 552 | if (!queue_pages_required(page, qp)) |
38e35860 | 553 | continue; |
a7f40cfe | 554 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
d8835445 YS |
555 | /* MPOL_MF_STRICT must be specified if we get here */ |
556 | if (!vma_migratable(vma)) { | |
557 | has_unmovable = true; | |
a7f40cfe | 558 | break; |
d8835445 | 559 | } |
a53190a4 YS |
560 | |
561 | /* | |
562 | * Do not abort immediately since there may be | |
563 | * temporary off LRU pages in the range. Still | |
564 | * need migrate other LRU pages. | |
565 | */ | |
566 | if (migrate_page_add(page, qp->pagelist, flags)) | |
567 | has_unmovable = true; | |
a7f40cfe YS |
568 | } else |
569 | break; | |
6f4576e3 | 570 | } |
3f088420 | 571 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 572 | cond_resched(); |
d8835445 YS |
573 | |
574 | if (has_unmovable) | |
575 | return 1; | |
576 | ||
a7f40cfe | 577 | return addr != end ? -EIO : 0; |
91612e0d HD |
578 | } |
579 | ||
6f4576e3 NH |
580 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
581 | unsigned long addr, unsigned long end, | |
582 | struct mm_walk *walk) | |
e2d8cf40 | 583 | { |
dcf17635 | 584 | int ret = 0; |
e2d8cf40 | 585 | #ifdef CONFIG_HUGETLB_PAGE |
6f4576e3 | 586 | struct queue_pages *qp = walk->private; |
dcf17635 | 587 | unsigned long flags = (qp->flags & MPOL_MF_VALID); |
e2d8cf40 | 588 | struct page *page; |
cb900f41 | 589 | spinlock_t *ptl; |
d4c54919 | 590 | pte_t entry; |
e2d8cf40 | 591 | |
6f4576e3 NH |
592 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
593 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
594 | if (!pte_present(entry)) |
595 | goto unlock; | |
596 | page = pte_page(entry); | |
88aaa2a1 | 597 | if (!queue_pages_required(page, qp)) |
e2d8cf40 | 598 | goto unlock; |
dcf17635 LX |
599 | |
600 | if (flags == MPOL_MF_STRICT) { | |
601 | /* | |
602 | * STRICT alone means only detecting misplaced page and no | |
603 | * need to further check other vma. | |
604 | */ | |
605 | ret = -EIO; | |
606 | goto unlock; | |
607 | } | |
608 | ||
609 | if (!vma_migratable(walk->vma)) { | |
610 | /* | |
611 | * Must be STRICT with MOVE*, otherwise .test_walk() have | |
612 | * stopped walking current vma. | |
613 | * Detecting misplaced page but allow migrating pages which | |
614 | * have been queued. | |
615 | */ | |
616 | ret = 1; | |
617 | goto unlock; | |
618 | } | |
619 | ||
e2d8cf40 NH |
620 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ |
621 | if (flags & (MPOL_MF_MOVE_ALL) || | |
dcf17635 LX |
622 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) { |
623 | if (!isolate_huge_page(page, qp->pagelist) && | |
624 | (flags & MPOL_MF_STRICT)) | |
625 | /* | |
626 | * Failed to isolate page but allow migrating pages | |
627 | * which have been queued. | |
628 | */ | |
629 | ret = 1; | |
630 | } | |
e2d8cf40 | 631 | unlock: |
cb900f41 | 632 | spin_unlock(ptl); |
e2d8cf40 NH |
633 | #else |
634 | BUG(); | |
635 | #endif | |
dcf17635 | 636 | return ret; |
1da177e4 LT |
637 | } |
638 | ||
5877231f | 639 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 640 | /* |
4b10e7d5 MG |
641 | * This is used to mark a range of virtual addresses to be inaccessible. |
642 | * These are later cleared by a NUMA hinting fault. Depending on these | |
643 | * faults, pages may be migrated for better NUMA placement. | |
644 | * | |
645 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
646 | * an architecture makes a different choice, it will need further | |
647 | * changes to the core. | |
b24f53a0 | 648 | */ |
4b10e7d5 MG |
649 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
650 | unsigned long addr, unsigned long end) | |
b24f53a0 | 651 | { |
4b10e7d5 | 652 | int nr_updated; |
b24f53a0 | 653 | |
58705444 | 654 | nr_updated = change_protection(vma, addr, end, PAGE_NONE, MM_CP_PROT_NUMA); |
03c5a6e1 MG |
655 | if (nr_updated) |
656 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 657 | |
4b10e7d5 | 658 | return nr_updated; |
b24f53a0 LS |
659 | } |
660 | #else | |
661 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
662 | unsigned long addr, unsigned long end) | |
663 | { | |
664 | return 0; | |
665 | } | |
5877231f | 666 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 667 | |
6f4576e3 NH |
668 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
669 | struct mm_walk *walk) | |
670 | { | |
671 | struct vm_area_struct *vma = walk->vma; | |
672 | struct queue_pages *qp = walk->private; | |
673 | unsigned long endvma = vma->vm_end; | |
674 | unsigned long flags = qp->flags; | |
675 | ||
a18b3ac2 | 676 | /* range check first */ |
ce33135c | 677 | VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma); |
f18da660 LX |
678 | |
679 | if (!qp->first) { | |
680 | qp->first = vma; | |
681 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
682 | (qp->start < vma->vm_start)) | |
683 | /* hole at head side of range */ | |
a18b3ac2 LX |
684 | return -EFAULT; |
685 | } | |
f18da660 LX |
686 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
687 | ((vma->vm_end < qp->end) && | |
688 | (!vma->vm_next || vma->vm_end < vma->vm_next->vm_start))) | |
689 | /* hole at middle or tail of range */ | |
690 | return -EFAULT; | |
a18b3ac2 | 691 | |
a7f40cfe YS |
692 | /* |
693 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
694 | * regardless of vma_migratable | |
695 | */ | |
696 | if (!vma_migratable(vma) && | |
697 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
698 | return 1; |
699 | ||
6f4576e3 NH |
700 | if (endvma > end) |
701 | endvma = end; | |
6f4576e3 | 702 | |
6f4576e3 NH |
703 | if (flags & MPOL_MF_LAZY) { |
704 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
3122e80e | 705 | if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) && |
4355c018 | 706 | !(vma->vm_flags & VM_MIXEDMAP)) |
6f4576e3 NH |
707 | change_prot_numa(vma, start, endvma); |
708 | return 1; | |
709 | } | |
710 | ||
77bf45e7 | 711 | /* queue pages from current vma */ |
a7f40cfe | 712 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
713 | return 0; |
714 | return 1; | |
715 | } | |
716 | ||
7b86ac33 CH |
717 | static const struct mm_walk_ops queue_pages_walk_ops = { |
718 | .hugetlb_entry = queue_pages_hugetlb, | |
719 | .pmd_entry = queue_pages_pte_range, | |
720 | .test_walk = queue_pages_test_walk, | |
721 | }; | |
722 | ||
dc9aa5b9 | 723 | /* |
98094945 NH |
724 | * Walk through page tables and collect pages to be migrated. |
725 | * | |
726 | * If pages found in a given range are on a set of nodes (determined by | |
727 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
728 | * passed via @private. |
729 | * | |
730 | * queue_pages_range() has three possible return values: | |
731 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
732 | * specified. | |
733 | * 0 - queue pages successfully or no misplaced page. | |
a85dfc30 YS |
734 | * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or |
735 | * memory range specified by nodemask and maxnode points outside | |
736 | * your accessible address space (-EFAULT) | |
dc9aa5b9 | 737 | */ |
d05f0cdc | 738 | static int |
98094945 | 739 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
740 | nodemask_t *nodes, unsigned long flags, |
741 | struct list_head *pagelist) | |
1da177e4 | 742 | { |
f18da660 | 743 | int err; |
6f4576e3 NH |
744 | struct queue_pages qp = { |
745 | .pagelist = pagelist, | |
746 | .flags = flags, | |
747 | .nmask = nodes, | |
f18da660 LX |
748 | .start = start, |
749 | .end = end, | |
750 | .first = NULL, | |
6f4576e3 | 751 | }; |
6f4576e3 | 752 | |
f18da660 LX |
753 | err = walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp); |
754 | ||
755 | if (!qp.first) | |
756 | /* whole range in hole */ | |
757 | err = -EFAULT; | |
758 | ||
759 | return err; | |
1da177e4 LT |
760 | } |
761 | ||
869833f2 KM |
762 | /* |
763 | * Apply policy to a single VMA | |
c1e8d7c6 | 764 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
765 | */ |
766 | static int vma_replace_policy(struct vm_area_struct *vma, | |
767 | struct mempolicy *pol) | |
8d34694c | 768 | { |
869833f2 KM |
769 | int err; |
770 | struct mempolicy *old; | |
771 | struct mempolicy *new; | |
8d34694c KM |
772 | |
773 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
774 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
775 | vma->vm_ops, vma->vm_file, | |
776 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
777 | ||
869833f2 KM |
778 | new = mpol_dup(pol); |
779 | if (IS_ERR(new)) | |
780 | return PTR_ERR(new); | |
781 | ||
782 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 783 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
784 | if (err) |
785 | goto err_out; | |
8d34694c | 786 | } |
869833f2 KM |
787 | |
788 | old = vma->vm_policy; | |
c1e8d7c6 | 789 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
790 | mpol_put(old); |
791 | ||
792 | return 0; | |
793 | err_out: | |
794 | mpol_put(new); | |
8d34694c KM |
795 | return err; |
796 | } | |
797 | ||
1da177e4 | 798 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
799 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
800 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
801 | { |
802 | struct vm_area_struct *next; | |
9d8cebd4 KM |
803 | struct vm_area_struct *prev; |
804 | struct vm_area_struct *vma; | |
805 | int err = 0; | |
e26a5114 | 806 | pgoff_t pgoff; |
9d8cebd4 KM |
807 | unsigned long vmstart; |
808 | unsigned long vmend; | |
1da177e4 | 809 | |
097d5910 | 810 | vma = find_vma(mm, start); |
f18da660 | 811 | VM_BUG_ON(!vma); |
9d8cebd4 | 812 | |
097d5910 | 813 | prev = vma->vm_prev; |
e26a5114 KM |
814 | if (start > vma->vm_start) |
815 | prev = vma; | |
816 | ||
9d8cebd4 | 817 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 818 | next = vma->vm_next; |
9d8cebd4 KM |
819 | vmstart = max(start, vma->vm_start); |
820 | vmend = min(end, vma->vm_end); | |
821 | ||
e26a5114 KM |
822 | if (mpol_equal(vma_policy(vma), new_pol)) |
823 | continue; | |
824 | ||
825 | pgoff = vma->vm_pgoff + | |
826 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 827 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af AA |
828 | vma->anon_vma, vma->vm_file, pgoff, |
829 | new_pol, vma->vm_userfaultfd_ctx); | |
9d8cebd4 KM |
830 | if (prev) { |
831 | vma = prev; | |
832 | next = vma->vm_next; | |
3964acd0 ON |
833 | if (mpol_equal(vma_policy(vma), new_pol)) |
834 | continue; | |
835 | /* vma_merge() joined vma && vma->next, case 8 */ | |
836 | goto replace; | |
9d8cebd4 KM |
837 | } |
838 | if (vma->vm_start != vmstart) { | |
839 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
840 | if (err) | |
841 | goto out; | |
842 | } | |
843 | if (vma->vm_end != vmend) { | |
844 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
845 | if (err) | |
846 | goto out; | |
847 | } | |
3964acd0 | 848 | replace: |
869833f2 | 849 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
850 | if (err) |
851 | goto out; | |
1da177e4 | 852 | } |
9d8cebd4 KM |
853 | |
854 | out: | |
1da177e4 LT |
855 | return err; |
856 | } | |
857 | ||
1da177e4 | 858 | /* Set the process memory policy */ |
028fec41 DR |
859 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
860 | nodemask_t *nodes) | |
1da177e4 | 861 | { |
58568d2a | 862 | struct mempolicy *new, *old; |
4bfc4495 | 863 | NODEMASK_SCRATCH(scratch); |
58568d2a | 864 | int ret; |
1da177e4 | 865 | |
4bfc4495 KH |
866 | if (!scratch) |
867 | return -ENOMEM; | |
f4e53d91 | 868 | |
4bfc4495 KH |
869 | new = mpol_new(mode, flags, nodes); |
870 | if (IS_ERR(new)) { | |
871 | ret = PTR_ERR(new); | |
872 | goto out; | |
873 | } | |
2c7c3a7d | 874 | |
bda420b9 HY |
875 | if (flags & MPOL_F_NUMA_BALANCING) { |
876 | if (new && new->mode == MPOL_BIND) { | |
877 | new->flags |= (MPOL_F_MOF | MPOL_F_MORON); | |
878 | } else { | |
879 | ret = -EINVAL; | |
880 | mpol_put(new); | |
881 | goto out; | |
882 | } | |
883 | } | |
884 | ||
4bfc4495 | 885 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 886 | if (ret) { |
58568d2a | 887 | mpol_put(new); |
4bfc4495 | 888 | goto out; |
58568d2a | 889 | } |
78b132e9 | 890 | task_lock(current); |
58568d2a | 891 | old = current->mempolicy; |
1da177e4 | 892 | current->mempolicy = new; |
45816682 VB |
893 | if (new && new->mode == MPOL_INTERLEAVE) |
894 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 895 | task_unlock(current); |
58568d2a | 896 | mpol_put(old); |
4bfc4495 KH |
897 | ret = 0; |
898 | out: | |
899 | NODEMASK_SCRATCH_FREE(scratch); | |
900 | return ret; | |
1da177e4 LT |
901 | } |
902 | ||
bea904d5 LS |
903 | /* |
904 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
905 | * |
906 | * Called with task's alloc_lock held | |
bea904d5 LS |
907 | */ |
908 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 909 | { |
dfcd3c0d | 910 | nodes_clear(*nodes); |
bea904d5 LS |
911 | if (p == &default_policy) |
912 | return; | |
913 | ||
45c4745a | 914 | switch (p->mode) { |
19770b32 | 915 | case MPOL_BIND: |
1da177e4 | 916 | case MPOL_INTERLEAVE: |
269fbe72 | 917 | case MPOL_PREFERRED: |
b27abacc | 918 | case MPOL_PREFERRED_MANY: |
269fbe72 | 919 | *nodes = p->nodes; |
1da177e4 | 920 | break; |
7858d7bc FT |
921 | case MPOL_LOCAL: |
922 | /* return empty node mask for local allocation */ | |
923 | break; | |
1da177e4 LT |
924 | default: |
925 | BUG(); | |
926 | } | |
927 | } | |
928 | ||
3b9aadf7 | 929 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 930 | { |
ba841078 | 931 | struct page *p = NULL; |
1da177e4 LT |
932 | int err; |
933 | ||
3b9aadf7 AA |
934 | int locked = 1; |
935 | err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked); | |
2d3a36a4 | 936 | if (err > 0) { |
1da177e4 LT |
937 | err = page_to_nid(p); |
938 | put_page(p); | |
939 | } | |
3b9aadf7 | 940 | if (locked) |
d8ed45c5 | 941 | mmap_read_unlock(mm); |
1da177e4 LT |
942 | return err; |
943 | } | |
944 | ||
1da177e4 | 945 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
946 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
947 | unsigned long addr, unsigned long flags) | |
1da177e4 | 948 | { |
8bccd85f | 949 | int err; |
1da177e4 LT |
950 | struct mm_struct *mm = current->mm; |
951 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 952 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 953 | |
754af6f5 LS |
954 | if (flags & |
955 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 956 | return -EINVAL; |
754af6f5 LS |
957 | |
958 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
959 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
960 | return -EINVAL; | |
961 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 962 | task_lock(current); |
754af6f5 | 963 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 964 | task_unlock(current); |
754af6f5 LS |
965 | return 0; |
966 | } | |
967 | ||
1da177e4 | 968 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
969 | /* |
970 | * Do NOT fall back to task policy if the | |
971 | * vma/shared policy at addr is NULL. We | |
972 | * want to return MPOL_DEFAULT in this case. | |
973 | */ | |
d8ed45c5 | 974 | mmap_read_lock(mm); |
33e3575c | 975 | vma = vma_lookup(mm, addr); |
1da177e4 | 976 | if (!vma) { |
d8ed45c5 | 977 | mmap_read_unlock(mm); |
1da177e4 LT |
978 | return -EFAULT; |
979 | } | |
980 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
981 | pol = vma->vm_ops->get_policy(vma, addr); | |
982 | else | |
983 | pol = vma->vm_policy; | |
984 | } else if (addr) | |
985 | return -EINVAL; | |
986 | ||
987 | if (!pol) | |
bea904d5 | 988 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
989 | |
990 | if (flags & MPOL_F_NODE) { | |
991 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 AA |
992 | /* |
993 | * Take a refcount on the mpol, lookup_node() | |
baf2f90b | 994 | * will drop the mmap_lock, so after calling |
3b9aadf7 AA |
995 | * lookup_node() only "pol" remains valid, "vma" |
996 | * is stale. | |
997 | */ | |
998 | pol_refcount = pol; | |
999 | vma = NULL; | |
1000 | mpol_get(pol); | |
1001 | err = lookup_node(mm, addr); | |
1da177e4 LT |
1002 | if (err < 0) |
1003 | goto out; | |
8bccd85f | 1004 | *policy = err; |
1da177e4 | 1005 | } else if (pol == current->mempolicy && |
45c4745a | 1006 | pol->mode == MPOL_INTERLEAVE) { |
269fbe72 | 1007 | *policy = next_node_in(current->il_prev, pol->nodes); |
1da177e4 LT |
1008 | } else { |
1009 | err = -EINVAL; | |
1010 | goto out; | |
1011 | } | |
bea904d5 LS |
1012 | } else { |
1013 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
1014 | pol->mode; | |
d79df630 DR |
1015 | /* |
1016 | * Internal mempolicy flags must be masked off before exposing | |
1017 | * the policy to userspace. | |
1018 | */ | |
1019 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1020 | } |
1da177e4 | 1021 | |
1da177e4 | 1022 | err = 0; |
58568d2a | 1023 | if (nmask) { |
c6b6ef8b LS |
1024 | if (mpol_store_user_nodemask(pol)) { |
1025 | *nmask = pol->w.user_nodemask; | |
1026 | } else { | |
1027 | task_lock(current); | |
1028 | get_policy_nodemask(pol, nmask); | |
1029 | task_unlock(current); | |
1030 | } | |
58568d2a | 1031 | } |
1da177e4 LT |
1032 | |
1033 | out: | |
52cd3b07 | 1034 | mpol_cond_put(pol); |
1da177e4 | 1035 | if (vma) |
d8ed45c5 | 1036 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1037 | if (pol_refcount) |
1038 | mpol_put(pol_refcount); | |
1da177e4 LT |
1039 | return err; |
1040 | } | |
1041 | ||
b20a3503 | 1042 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 1043 | /* |
c8633798 | 1044 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 1045 | */ |
a53190a4 | 1046 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 1047 | unsigned long flags) |
6ce3c4c0 | 1048 | { |
c8633798 | 1049 | struct page *head = compound_head(page); |
6ce3c4c0 | 1050 | /* |
fc301289 | 1051 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 1052 | */ |
c8633798 NH |
1053 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
1054 | if (!isolate_lru_page(head)) { | |
1055 | list_add_tail(&head->lru, pagelist); | |
1056 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1057 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1058 | thp_nr_pages(head)); |
a53190a4 YS |
1059 | } else if (flags & MPOL_MF_STRICT) { |
1060 | /* | |
1061 | * Non-movable page may reach here. And, there may be | |
1062 | * temporary off LRU pages or non-LRU movable pages. | |
1063 | * Treat them as unmovable pages since they can't be | |
1064 | * isolated, so they can't be moved at the moment. It | |
1065 | * should return -EIO for this case too. | |
1066 | */ | |
1067 | return -EIO; | |
62695a84 NP |
1068 | } |
1069 | } | |
a53190a4 YS |
1070 | |
1071 | return 0; | |
7e2ab150 | 1072 | } |
6ce3c4c0 | 1073 | |
7e2ab150 CL |
1074 | /* |
1075 | * Migrate pages from one node to a target node. | |
1076 | * Returns error or the number of pages not migrated. | |
1077 | */ | |
dbcb0f19 AB |
1078 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1079 | int flags) | |
7e2ab150 CL |
1080 | { |
1081 | nodemask_t nmask; | |
1082 | LIST_HEAD(pagelist); | |
1083 | int err = 0; | |
a0976311 JK |
1084 | struct migration_target_control mtc = { |
1085 | .nid = dest, | |
1086 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1087 | }; | |
7e2ab150 CL |
1088 | |
1089 | nodes_clear(nmask); | |
1090 | node_set(source, nmask); | |
6ce3c4c0 | 1091 | |
08270807 MK |
1092 | /* |
1093 | * This does not "check" the range but isolates all pages that | |
1094 | * need migration. Between passing in the full user address | |
1095 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1096 | */ | |
1097 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1098 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1099 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1100 | ||
cf608ac1 | 1101 | if (!list_empty(&pagelist)) { |
a0976311 | 1102 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
5ac95884 | 1103 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
cf608ac1 | 1104 | if (err) |
e2d8cf40 | 1105 | putback_movable_pages(&pagelist); |
cf608ac1 | 1106 | } |
95a402c3 | 1107 | |
7e2ab150 | 1108 | return err; |
6ce3c4c0 CL |
1109 | } |
1110 | ||
39743889 | 1111 | /* |
7e2ab150 CL |
1112 | * Move pages between the two nodesets so as to preserve the physical |
1113 | * layout as much as possible. | |
39743889 CL |
1114 | * |
1115 | * Returns the number of page that could not be moved. | |
1116 | */ | |
0ce72d4f AM |
1117 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1118 | const nodemask_t *to, int flags) | |
39743889 | 1119 | { |
7e2ab150 | 1120 | int busy = 0; |
f555befd | 1121 | int err = 0; |
7e2ab150 | 1122 | nodemask_t tmp; |
39743889 | 1123 | |
361a2a22 | 1124 | lru_cache_disable(); |
0aedadf9 | 1125 | |
d8ed45c5 | 1126 | mmap_read_lock(mm); |
39743889 | 1127 | |
da0aa138 KM |
1128 | /* |
1129 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1130 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1131 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1132 | * The pair of nodemasks 'to' and 'from' define the map. | |
1133 | * | |
1134 | * If no pair of bits is found that way, fallback to picking some | |
1135 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1136 | * 'source' and 'dest' bits are the same, this represents a node | |
1137 | * that will be migrating to itself, so no pages need move. | |
1138 | * | |
1139 | * If no bits are left in 'tmp', or if all remaining bits left | |
1140 | * in 'tmp' correspond to the same bit in 'to', return false | |
1141 | * (nothing left to migrate). | |
1142 | * | |
1143 | * This lets us pick a pair of nodes to migrate between, such that | |
1144 | * if possible the dest node is not already occupied by some other | |
1145 | * source node, minimizing the risk of overloading the memory on a | |
1146 | * node that would happen if we migrated incoming memory to a node | |
1147 | * before migrating outgoing memory source that same node. | |
1148 | * | |
1149 | * A single scan of tmp is sufficient. As we go, we remember the | |
1150 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1151 | * that not only moved, but what's better, moved to an empty slot | |
1152 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1153 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1154 | * most recent <s, d> pair that moved. If we get all the way through |
1155 | * the scan of tmp without finding any node that moved, much less | |
1156 | * moved to an empty node, then there is nothing left worth migrating. | |
1157 | */ | |
d4984711 | 1158 | |
0ce72d4f | 1159 | tmp = *from; |
7e2ab150 | 1160 | while (!nodes_empty(tmp)) { |
68d68ff6 | 1161 | int s, d; |
b76ac7e7 | 1162 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1163 | int dest = 0; |
1164 | ||
1165 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1166 | |
1167 | /* | |
1168 | * do_migrate_pages() tries to maintain the relative | |
1169 | * node relationship of the pages established between | |
1170 | * threads and memory areas. | |
1171 | * | |
1172 | * However if the number of source nodes is not equal to | |
1173 | * the number of destination nodes we can not preserve | |
1174 | * this node relative relationship. In that case, skip | |
1175 | * copying memory from a node that is in the destination | |
1176 | * mask. | |
1177 | * | |
1178 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1179 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1180 | */ | |
1181 | ||
0ce72d4f AM |
1182 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1183 | (node_isset(s, *to))) | |
4a5b18cc LW |
1184 | continue; |
1185 | ||
0ce72d4f | 1186 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1187 | if (s == d) |
1188 | continue; | |
1189 | ||
1190 | source = s; /* Node moved. Memorize */ | |
1191 | dest = d; | |
1192 | ||
1193 | /* dest not in remaining from nodes? */ | |
1194 | if (!node_isset(dest, tmp)) | |
1195 | break; | |
1196 | } | |
b76ac7e7 | 1197 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1198 | break; |
1199 | ||
1200 | node_clear(source, tmp); | |
1201 | err = migrate_to_node(mm, source, dest, flags); | |
1202 | if (err > 0) | |
1203 | busy += err; | |
1204 | if (err < 0) | |
1205 | break; | |
39743889 | 1206 | } |
d8ed45c5 | 1207 | mmap_read_unlock(mm); |
d479960e | 1208 | |
361a2a22 | 1209 | lru_cache_enable(); |
7e2ab150 CL |
1210 | if (err < 0) |
1211 | return err; | |
1212 | return busy; | |
b20a3503 CL |
1213 | |
1214 | } | |
1215 | ||
3ad33b24 LS |
1216 | /* |
1217 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1218 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1219 | * Search forward from there, if not. N.B., this assumes that the |
1220 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1221 | * is in virtual address order. | |
1222 | */ | |
666feb21 | 1223 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1224 | { |
d05f0cdc | 1225 | struct vm_area_struct *vma; |
3f649ab7 | 1226 | unsigned long address; |
95a402c3 | 1227 | |
d05f0cdc | 1228 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1229 | while (vma) { |
1230 | address = page_address_in_vma(page, vma); | |
1231 | if (address != -EFAULT) | |
1232 | break; | |
1233 | vma = vma->vm_next; | |
1234 | } | |
11c731e8 WL |
1235 | |
1236 | if (PageHuge(page)) { | |
389c8178 MH |
1237 | return alloc_huge_page_vma(page_hstate(compound_head(page)), |
1238 | vma, address); | |
94723aaf | 1239 | } else if (PageTransHuge(page)) { |
c8633798 NH |
1240 | struct page *thp; |
1241 | ||
19deb769 DR |
1242 | thp = alloc_hugepage_vma(GFP_TRANSHUGE, vma, address, |
1243 | HPAGE_PMD_ORDER); | |
c8633798 NH |
1244 | if (!thp) |
1245 | return NULL; | |
1246 | prep_transhuge_page(thp); | |
1247 | return thp; | |
11c731e8 | 1248 | } |
0bf598d8 | 1249 | /* |
11c731e8 | 1250 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1251 | */ |
0f556856 MH |
1252 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL, |
1253 | vma, address); | |
95a402c3 | 1254 | } |
b20a3503 CL |
1255 | #else |
1256 | ||
a53190a4 | 1257 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1258 | unsigned long flags) |
1259 | { | |
a53190a4 | 1260 | return -EIO; |
39743889 CL |
1261 | } |
1262 | ||
0ce72d4f AM |
1263 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1264 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1265 | { |
1266 | return -ENOSYS; | |
1267 | } | |
95a402c3 | 1268 | |
666feb21 | 1269 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1270 | { |
1271 | return NULL; | |
1272 | } | |
b20a3503 CL |
1273 | #endif |
1274 | ||
dbcb0f19 | 1275 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1276 | unsigned short mode, unsigned short mode_flags, |
1277 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1278 | { |
6ce3c4c0 CL |
1279 | struct mm_struct *mm = current->mm; |
1280 | struct mempolicy *new; | |
1281 | unsigned long end; | |
1282 | int err; | |
d8835445 | 1283 | int ret; |
6ce3c4c0 CL |
1284 | LIST_HEAD(pagelist); |
1285 | ||
b24f53a0 | 1286 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1287 | return -EINVAL; |
74c00241 | 1288 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1289 | return -EPERM; |
1290 | ||
1291 | if (start & ~PAGE_MASK) | |
1292 | return -EINVAL; | |
1293 | ||
1294 | if (mode == MPOL_DEFAULT) | |
1295 | flags &= ~MPOL_MF_STRICT; | |
1296 | ||
1297 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1298 | end = start + len; | |
1299 | ||
1300 | if (end < start) | |
1301 | return -EINVAL; | |
1302 | if (end == start) | |
1303 | return 0; | |
1304 | ||
028fec41 | 1305 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1306 | if (IS_ERR(new)) |
1307 | return PTR_ERR(new); | |
1308 | ||
b24f53a0 LS |
1309 | if (flags & MPOL_MF_LAZY) |
1310 | new->flags |= MPOL_F_MOF; | |
1311 | ||
6ce3c4c0 CL |
1312 | /* |
1313 | * If we are using the default policy then operation | |
1314 | * on discontinuous address spaces is okay after all | |
1315 | */ | |
1316 | if (!new) | |
1317 | flags |= MPOL_MF_DISCONTIG_OK; | |
1318 | ||
028fec41 DR |
1319 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1320 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1321 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1322 | |
0aedadf9 CL |
1323 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1324 | ||
361a2a22 | 1325 | lru_cache_disable(); |
0aedadf9 | 1326 | } |
4bfc4495 KH |
1327 | { |
1328 | NODEMASK_SCRATCH(scratch); | |
1329 | if (scratch) { | |
d8ed45c5 | 1330 | mmap_write_lock(mm); |
4bfc4495 | 1331 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1332 | if (err) |
d8ed45c5 | 1333 | mmap_write_unlock(mm); |
4bfc4495 KH |
1334 | } else |
1335 | err = -ENOMEM; | |
1336 | NODEMASK_SCRATCH_FREE(scratch); | |
1337 | } | |
b05ca738 KM |
1338 | if (err) |
1339 | goto mpol_out; | |
1340 | ||
d8835445 | 1341 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1342 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1343 | |
1344 | if (ret < 0) { | |
a85dfc30 | 1345 | err = ret; |
d8835445 YS |
1346 | goto up_out; |
1347 | } | |
1348 | ||
1349 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1350 | |
b24f53a0 LS |
1351 | if (!err) { |
1352 | int nr_failed = 0; | |
1353 | ||
cf608ac1 | 1354 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1355 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc | 1356 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
5ac95884 | 1357 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL); |
cf608ac1 | 1358 | if (nr_failed) |
74060e4d | 1359 | putback_movable_pages(&pagelist); |
cf608ac1 | 1360 | } |
6ce3c4c0 | 1361 | |
d8835445 | 1362 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1363 | err = -EIO; |
a85dfc30 | 1364 | } else { |
d8835445 | 1365 | up_out: |
a85dfc30 YS |
1366 | if (!list_empty(&pagelist)) |
1367 | putback_movable_pages(&pagelist); | |
1368 | } | |
1369 | ||
d8ed45c5 | 1370 | mmap_write_unlock(mm); |
d8835445 | 1371 | mpol_out: |
f0be3d32 | 1372 | mpol_put(new); |
d479960e | 1373 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1374 | lru_cache_enable(); |
6ce3c4c0 CL |
1375 | return err; |
1376 | } | |
1377 | ||
8bccd85f CL |
1378 | /* |
1379 | * User space interface with variable sized bitmaps for nodelists. | |
1380 | */ | |
1381 | ||
1382 | /* Copy a node mask from user space. */ | |
39743889 | 1383 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1384 | unsigned long maxnode) |
1385 | { | |
1386 | unsigned long k; | |
56521e7a | 1387 | unsigned long t; |
8bccd85f CL |
1388 | unsigned long nlongs; |
1389 | unsigned long endmask; | |
1390 | ||
1391 | --maxnode; | |
1392 | nodes_clear(*nodes); | |
1393 | if (maxnode == 0 || !nmask) | |
1394 | return 0; | |
a9c930ba | 1395 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1396 | return -EINVAL; |
8bccd85f CL |
1397 | |
1398 | nlongs = BITS_TO_LONGS(maxnode); | |
1399 | if ((maxnode % BITS_PER_LONG) == 0) | |
1400 | endmask = ~0UL; | |
1401 | else | |
1402 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1403 | ||
56521e7a YX |
1404 | /* |
1405 | * When the user specified more nodes than supported just check | |
1406 | * if the non supported part is all zero. | |
1407 | * | |
1408 | * If maxnode have more longs than MAX_NUMNODES, check | |
1409 | * the bits in that area first. And then go through to | |
1410 | * check the rest bits which equal or bigger than MAX_NUMNODES. | |
1411 | * Otherwise, just check bits [MAX_NUMNODES, maxnode). | |
1412 | */ | |
8bccd85f | 1413 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
8bccd85f | 1414 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
8bccd85f CL |
1415 | if (get_user(t, nmask + k)) |
1416 | return -EFAULT; | |
1417 | if (k == nlongs - 1) { | |
1418 | if (t & endmask) | |
1419 | return -EINVAL; | |
1420 | } else if (t) | |
1421 | return -EINVAL; | |
1422 | } | |
1423 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1424 | endmask = ~0UL; | |
1425 | } | |
1426 | ||
56521e7a YX |
1427 | if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) { |
1428 | unsigned long valid_mask = endmask; | |
1429 | ||
1430 | valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1431 | if (get_user(t, nmask + nlongs - 1)) | |
1432 | return -EFAULT; | |
1433 | if (t & valid_mask) | |
1434 | return -EINVAL; | |
1435 | } | |
1436 | ||
8bccd85f CL |
1437 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
1438 | return -EFAULT; | |
1439 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1440 | return 0; | |
1441 | } | |
1442 | ||
1443 | /* Copy a kernel node mask to user space */ | |
1444 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1445 | nodemask_t *nodes) | |
1446 | { | |
1447 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1448 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
8bccd85f CL |
1449 | |
1450 | if (copy > nbytes) { | |
1451 | if (copy > PAGE_SIZE) | |
1452 | return -EINVAL; | |
1453 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1454 | return -EFAULT; | |
1455 | copy = nbytes; | |
1456 | } | |
1457 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1458 | } | |
1459 | ||
95837924 FT |
1460 | /* Basic parameter sanity check used by both mbind() and set_mempolicy() */ |
1461 | static inline int sanitize_mpol_flags(int *mode, unsigned short *flags) | |
1462 | { | |
1463 | *flags = *mode & MPOL_MODE_FLAGS; | |
1464 | *mode &= ~MPOL_MODE_FLAGS; | |
b27abacc DH |
1465 | |
1466 | /* | |
1467 | * The check should be 'mode >= MPOL_MAX', but as 'prefer_many' | |
1468 | * is not fully implemented, don't permit it to be used for now, | |
1469 | * and the logic will be restored in following patch | |
1470 | */ | |
1471 | if ((unsigned int)(*mode) >= MPOL_PREFERRED_MANY) | |
95837924 FT |
1472 | return -EINVAL; |
1473 | if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES)) | |
1474 | return -EINVAL; | |
1475 | ||
1476 | return 0; | |
1477 | } | |
1478 | ||
e7dc9ad6 DB |
1479 | static long kernel_mbind(unsigned long start, unsigned long len, |
1480 | unsigned long mode, const unsigned long __user *nmask, | |
1481 | unsigned long maxnode, unsigned int flags) | |
8bccd85f | 1482 | { |
95837924 | 1483 | unsigned short mode_flags; |
8bccd85f | 1484 | nodemask_t nodes; |
95837924 | 1485 | int lmode = mode; |
8bccd85f CL |
1486 | int err; |
1487 | ||
057d3389 | 1488 | start = untagged_addr(start); |
95837924 FT |
1489 | err = sanitize_mpol_flags(&lmode, &mode_flags); |
1490 | if (err) | |
1491 | return err; | |
1492 | ||
8bccd85f CL |
1493 | err = get_nodes(&nodes, nmask, maxnode); |
1494 | if (err) | |
1495 | return err; | |
95837924 FT |
1496 | |
1497 | return do_mbind(start, len, lmode, mode_flags, &nodes, flags); | |
8bccd85f CL |
1498 | } |
1499 | ||
e7dc9ad6 DB |
1500 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1501 | unsigned long, mode, const unsigned long __user *, nmask, | |
1502 | unsigned long, maxnode, unsigned int, flags) | |
1503 | { | |
1504 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1505 | } | |
1506 | ||
8bccd85f | 1507 | /* Set the process memory policy */ |
af03c4ac DB |
1508 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1509 | unsigned long maxnode) | |
8bccd85f | 1510 | { |
95837924 | 1511 | unsigned short mode_flags; |
8bccd85f | 1512 | nodemask_t nodes; |
95837924 FT |
1513 | int lmode = mode; |
1514 | int err; | |
1515 | ||
1516 | err = sanitize_mpol_flags(&lmode, &mode_flags); | |
1517 | if (err) | |
1518 | return err; | |
8bccd85f | 1519 | |
8bccd85f CL |
1520 | err = get_nodes(&nodes, nmask, maxnode); |
1521 | if (err) | |
1522 | return err; | |
95837924 FT |
1523 | |
1524 | return do_set_mempolicy(lmode, mode_flags, &nodes); | |
8bccd85f CL |
1525 | } |
1526 | ||
af03c4ac DB |
1527 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1528 | unsigned long, maxnode) | |
1529 | { | |
1530 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1531 | } | |
1532 | ||
b6e9b0ba DB |
1533 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1534 | const unsigned long __user *old_nodes, | |
1535 | const unsigned long __user *new_nodes) | |
39743889 | 1536 | { |
596d7cfa | 1537 | struct mm_struct *mm = NULL; |
39743889 | 1538 | struct task_struct *task; |
39743889 CL |
1539 | nodemask_t task_nodes; |
1540 | int err; | |
596d7cfa KM |
1541 | nodemask_t *old; |
1542 | nodemask_t *new; | |
1543 | NODEMASK_SCRATCH(scratch); | |
1544 | ||
1545 | if (!scratch) | |
1546 | return -ENOMEM; | |
39743889 | 1547 | |
596d7cfa KM |
1548 | old = &scratch->mask1; |
1549 | new = &scratch->mask2; | |
1550 | ||
1551 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1552 | if (err) |
596d7cfa | 1553 | goto out; |
39743889 | 1554 | |
596d7cfa | 1555 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1556 | if (err) |
596d7cfa | 1557 | goto out; |
39743889 CL |
1558 | |
1559 | /* Find the mm_struct */ | |
55cfaa3c | 1560 | rcu_read_lock(); |
228ebcbe | 1561 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1562 | if (!task) { |
55cfaa3c | 1563 | rcu_read_unlock(); |
596d7cfa KM |
1564 | err = -ESRCH; |
1565 | goto out; | |
39743889 | 1566 | } |
3268c63e | 1567 | get_task_struct(task); |
39743889 | 1568 | |
596d7cfa | 1569 | err = -EINVAL; |
39743889 CL |
1570 | |
1571 | /* | |
31367466 OE |
1572 | * Check if this process has the right to modify the specified process. |
1573 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1574 | */ |
31367466 | 1575 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1576 | rcu_read_unlock(); |
39743889 | 1577 | err = -EPERM; |
3268c63e | 1578 | goto out_put; |
39743889 | 1579 | } |
c69e8d9c | 1580 | rcu_read_unlock(); |
39743889 CL |
1581 | |
1582 | task_nodes = cpuset_mems_allowed(task); | |
1583 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1584 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1585 | err = -EPERM; |
3268c63e | 1586 | goto out_put; |
39743889 CL |
1587 | } |
1588 | ||
0486a38b YX |
1589 | task_nodes = cpuset_mems_allowed(current); |
1590 | nodes_and(*new, *new, task_nodes); | |
1591 | if (nodes_empty(*new)) | |
1592 | goto out_put; | |
1593 | ||
86c3a764 DQ |
1594 | err = security_task_movememory(task); |
1595 | if (err) | |
3268c63e | 1596 | goto out_put; |
86c3a764 | 1597 | |
3268c63e CL |
1598 | mm = get_task_mm(task); |
1599 | put_task_struct(task); | |
f2a9ef88 SL |
1600 | |
1601 | if (!mm) { | |
3268c63e | 1602 | err = -EINVAL; |
f2a9ef88 SL |
1603 | goto out; |
1604 | } | |
1605 | ||
1606 | err = do_migrate_pages(mm, old, new, | |
1607 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1608 | |
1609 | mmput(mm); | |
1610 | out: | |
596d7cfa KM |
1611 | NODEMASK_SCRATCH_FREE(scratch); |
1612 | ||
39743889 | 1613 | return err; |
3268c63e CL |
1614 | |
1615 | out_put: | |
1616 | put_task_struct(task); | |
1617 | goto out; | |
1618 | ||
39743889 CL |
1619 | } |
1620 | ||
b6e9b0ba DB |
1621 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1622 | const unsigned long __user *, old_nodes, | |
1623 | const unsigned long __user *, new_nodes) | |
1624 | { | |
1625 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1626 | } | |
1627 | ||
39743889 | 1628 | |
8bccd85f | 1629 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1630 | static int kernel_get_mempolicy(int __user *policy, |
1631 | unsigned long __user *nmask, | |
1632 | unsigned long maxnode, | |
1633 | unsigned long addr, | |
1634 | unsigned long flags) | |
8bccd85f | 1635 | { |
dbcb0f19 | 1636 | int err; |
3f649ab7 | 1637 | int pval; |
8bccd85f CL |
1638 | nodemask_t nodes; |
1639 | ||
050c17f2 | 1640 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1641 | return -EINVAL; |
1642 | ||
4605f057 WH |
1643 | addr = untagged_addr(addr); |
1644 | ||
8bccd85f CL |
1645 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1646 | ||
1647 | if (err) | |
1648 | return err; | |
1649 | ||
1650 | if (policy && put_user(pval, policy)) | |
1651 | return -EFAULT; | |
1652 | ||
1653 | if (nmask) | |
1654 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1655 | ||
1656 | return err; | |
1657 | } | |
1658 | ||
af03c4ac DB |
1659 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1660 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1661 | unsigned long, addr, unsigned long, flags) | |
1662 | { | |
1663 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1664 | } | |
1665 | ||
1da177e4 LT |
1666 | #ifdef CONFIG_COMPAT |
1667 | ||
c93e0f6c HC |
1668 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1669 | compat_ulong_t __user *, nmask, | |
1670 | compat_ulong_t, maxnode, | |
1671 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1672 | { |
1673 | long err; | |
1674 | unsigned long __user *nm = NULL; | |
1675 | unsigned long nr_bits, alloc_size; | |
1676 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1677 | ||
050c17f2 | 1678 | nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids); |
1da177e4 LT |
1679 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
1680 | ||
1681 | if (nmask) | |
1682 | nm = compat_alloc_user_space(alloc_size); | |
1683 | ||
af03c4ac | 1684 | err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
1da177e4 LT |
1685 | |
1686 | if (!err && nmask) { | |
2bbff6c7 KH |
1687 | unsigned long copy_size; |
1688 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1689 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1690 | /* ensure entire bitmap is zeroed */ |
1691 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1692 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1693 | } | |
1694 | ||
1695 | return err; | |
1696 | } | |
1697 | ||
c93e0f6c HC |
1698 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1699 | compat_ulong_t, maxnode) | |
1da177e4 | 1700 | { |
1da177e4 LT |
1701 | unsigned long __user *nm = NULL; |
1702 | unsigned long nr_bits, alloc_size; | |
1703 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1704 | ||
1705 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1706 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1707 | ||
1708 | if (nmask) { | |
cf01fb99 CS |
1709 | if (compat_get_bitmap(bm, nmask, nr_bits)) |
1710 | return -EFAULT; | |
1da177e4 | 1711 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1712 | if (copy_to_user(nm, bm, alloc_size)) |
1713 | return -EFAULT; | |
1da177e4 LT |
1714 | } |
1715 | ||
af03c4ac | 1716 | return kernel_set_mempolicy(mode, nm, nr_bits+1); |
1da177e4 LT |
1717 | } |
1718 | ||
c93e0f6c HC |
1719 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1720 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1721 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 | 1722 | { |
1da177e4 LT |
1723 | unsigned long __user *nm = NULL; |
1724 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1725 | nodemask_t bm; |
1da177e4 LT |
1726 | |
1727 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1728 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1729 | ||
1730 | if (nmask) { | |
cf01fb99 CS |
1731 | if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits)) |
1732 | return -EFAULT; | |
1da177e4 | 1733 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1734 | if (copy_to_user(nm, nodes_addr(bm), alloc_size)) |
1735 | return -EFAULT; | |
1da177e4 LT |
1736 | } |
1737 | ||
e7dc9ad6 | 1738 | return kernel_mbind(start, len, mode, nm, nr_bits+1, flags); |
1da177e4 LT |
1739 | } |
1740 | ||
b6e9b0ba DB |
1741 | COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid, |
1742 | compat_ulong_t, maxnode, | |
1743 | const compat_ulong_t __user *, old_nodes, | |
1744 | const compat_ulong_t __user *, new_nodes) | |
1745 | { | |
1746 | unsigned long __user *old = NULL; | |
1747 | unsigned long __user *new = NULL; | |
1748 | nodemask_t tmp_mask; | |
1749 | unsigned long nr_bits; | |
1750 | unsigned long size; | |
1751 | ||
1752 | nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES); | |
1753 | size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1754 | if (old_nodes) { | |
1755 | if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits)) | |
1756 | return -EFAULT; | |
1757 | old = compat_alloc_user_space(new_nodes ? size * 2 : size); | |
1758 | if (new_nodes) | |
1759 | new = old + size / sizeof(unsigned long); | |
1760 | if (copy_to_user(old, nodes_addr(tmp_mask), size)) | |
1761 | return -EFAULT; | |
1762 | } | |
1763 | if (new_nodes) { | |
1764 | if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits)) | |
1765 | return -EFAULT; | |
1766 | if (new == NULL) | |
1767 | new = compat_alloc_user_space(size); | |
1768 | if (copy_to_user(new, nodes_addr(tmp_mask), size)) | |
1769 | return -EFAULT; | |
1770 | } | |
1771 | return kernel_migrate_pages(pid, nr_bits + 1, old, new); | |
1772 | } | |
1773 | ||
1774 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 1775 | |
20ca87f2 LX |
1776 | bool vma_migratable(struct vm_area_struct *vma) |
1777 | { | |
1778 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1779 | return false; | |
1780 | ||
1781 | /* | |
1782 | * DAX device mappings require predictable access latency, so avoid | |
1783 | * incurring periodic faults. | |
1784 | */ | |
1785 | if (vma_is_dax(vma)) | |
1786 | return false; | |
1787 | ||
1788 | if (is_vm_hugetlb_page(vma) && | |
1789 | !hugepage_migration_supported(hstate_vma(vma))) | |
1790 | return false; | |
1791 | ||
1792 | /* | |
1793 | * Migration allocates pages in the highest zone. If we cannot | |
1794 | * do so then migration (at least from node to node) is not | |
1795 | * possible. | |
1796 | */ | |
1797 | if (vma->vm_file && | |
1798 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1799 | < policy_zone) | |
1800 | return false; | |
1801 | return true; | |
1802 | } | |
1803 | ||
74d2c3a0 ON |
1804 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1805 | unsigned long addr) | |
1da177e4 | 1806 | { |
8d90274b | 1807 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1808 | |
1809 | if (vma) { | |
480eccf9 | 1810 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1811 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1812 | } else if (vma->vm_policy) { |
1da177e4 | 1813 | pol = vma->vm_policy; |
00442ad0 MG |
1814 | |
1815 | /* | |
1816 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1817 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1818 | * count on these policies which will be dropped by | |
1819 | * mpol_cond_put() later | |
1820 | */ | |
1821 | if (mpol_needs_cond_ref(pol)) | |
1822 | mpol_get(pol); | |
1823 | } | |
1da177e4 | 1824 | } |
f15ca78e | 1825 | |
74d2c3a0 ON |
1826 | return pol; |
1827 | } | |
1828 | ||
1829 | /* | |
dd6eecb9 | 1830 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1831 | * @vma: virtual memory area whose policy is sought |
1832 | * @addr: address in @vma for shared policy lookup | |
1833 | * | |
1834 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1835 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1836 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1837 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1838 | * freeing by another task. It is the caller's responsibility to free the | |
1839 | * extra reference for shared policies. | |
1840 | */ | |
ac79f78d | 1841 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1842 | unsigned long addr) |
74d2c3a0 ON |
1843 | { |
1844 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1845 | ||
8d90274b | 1846 | if (!pol) |
dd6eecb9 | 1847 | pol = get_task_policy(current); |
8d90274b | 1848 | |
1da177e4 LT |
1849 | return pol; |
1850 | } | |
1851 | ||
6b6482bb | 1852 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1853 | { |
6b6482bb | 1854 | struct mempolicy *pol; |
fc314724 | 1855 | |
6b6482bb ON |
1856 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1857 | bool ret = false; | |
fc314724 | 1858 | |
6b6482bb ON |
1859 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1860 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1861 | ret = true; | |
1862 | mpol_cond_put(pol); | |
8d90274b | 1863 | |
6b6482bb | 1864 | return ret; |
fc314724 MG |
1865 | } |
1866 | ||
6b6482bb | 1867 | pol = vma->vm_policy; |
8d90274b | 1868 | if (!pol) |
6b6482bb | 1869 | pol = get_task_policy(current); |
8d90274b | 1870 | |
fc314724 MG |
1871 | return pol->flags & MPOL_F_MOF; |
1872 | } | |
1873 | ||
d3eb1570 LJ |
1874 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1875 | { | |
1876 | enum zone_type dynamic_policy_zone = policy_zone; | |
1877 | ||
1878 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1879 | ||
1880 | /* | |
269fbe72 | 1881 | * if policy->nodes has movable memory only, |
d3eb1570 LJ |
1882 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. |
1883 | * | |
269fbe72 | 1884 | * policy->nodes is intersect with node_states[N_MEMORY]. |
f0953a1b | 1885 | * so if the following test fails, it implies |
269fbe72 | 1886 | * policy->nodes has movable memory only. |
d3eb1570 | 1887 | */ |
269fbe72 | 1888 | if (!nodes_intersects(policy->nodes, node_states[N_HIGH_MEMORY])) |
d3eb1570 LJ |
1889 | dynamic_policy_zone = ZONE_MOVABLE; |
1890 | ||
1891 | return zone >= dynamic_policy_zone; | |
1892 | } | |
1893 | ||
52cd3b07 LS |
1894 | /* |
1895 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1896 | * page allocation | |
1897 | */ | |
8ca39e68 | 1898 | nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
19770b32 | 1899 | { |
b27abacc DH |
1900 | int mode = policy->mode; |
1901 | ||
19770b32 | 1902 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ |
b27abacc DH |
1903 | if (unlikely(mode == MPOL_BIND) && |
1904 | apply_policy_zone(policy, gfp_zone(gfp)) && | |
1905 | cpuset_nodemask_valid_mems_allowed(&policy->nodes)) | |
1906 | return &policy->nodes; | |
1907 | ||
1908 | if (mode == MPOL_PREFERRED_MANY) | |
269fbe72 | 1909 | return &policy->nodes; |
19770b32 MG |
1910 | |
1911 | return NULL; | |
1912 | } | |
1913 | ||
b27abacc DH |
1914 | /* |
1915 | * Return the preferred node id for 'prefer' mempolicy, and return | |
1916 | * the given id for all other policies. | |
1917 | * | |
1918 | * policy_node() is always coupled with policy_nodemask(), which | |
1919 | * secures the nodemask limit for 'bind' and 'prefer-many' policy. | |
1920 | */ | |
f8fd5253 | 1921 | static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd) |
1da177e4 | 1922 | { |
7858d7bc | 1923 | if (policy->mode == MPOL_PREFERRED) { |
269fbe72 | 1924 | nd = first_node(policy->nodes); |
7858d7bc | 1925 | } else { |
19770b32 | 1926 | /* |
6d840958 MH |
1927 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1928 | * because we might easily break the expectation to stay on the | |
1929 | * requested node and not break the policy. | |
19770b32 | 1930 | */ |
6d840958 | 1931 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1932 | } |
6d840958 | 1933 | |
04ec6264 | 1934 | return nd; |
1da177e4 LT |
1935 | } |
1936 | ||
1937 | /* Do dynamic interleaving for a process */ | |
1938 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1939 | { | |
45816682 | 1940 | unsigned next; |
1da177e4 LT |
1941 | struct task_struct *me = current; |
1942 | ||
269fbe72 | 1943 | next = next_node_in(me->il_prev, policy->nodes); |
f5b087b5 | 1944 | if (next < MAX_NUMNODES) |
45816682 VB |
1945 | me->il_prev = next; |
1946 | return next; | |
1da177e4 LT |
1947 | } |
1948 | ||
dc85da15 CL |
1949 | /* |
1950 | * Depending on the memory policy provide a node from which to allocate the | |
1951 | * next slab entry. | |
1952 | */ | |
2a389610 | 1953 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1954 | { |
e7b691b0 | 1955 | struct mempolicy *policy; |
2a389610 | 1956 | int node = numa_mem_id(); |
e7b691b0 AK |
1957 | |
1958 | if (in_interrupt()) | |
2a389610 | 1959 | return node; |
e7b691b0 AK |
1960 | |
1961 | policy = current->mempolicy; | |
7858d7bc | 1962 | if (!policy) |
2a389610 | 1963 | return node; |
bea904d5 LS |
1964 | |
1965 | switch (policy->mode) { | |
1966 | case MPOL_PREFERRED: | |
269fbe72 | 1967 | return first_node(policy->nodes); |
765c4507 | 1968 | |
dc85da15 CL |
1969 | case MPOL_INTERLEAVE: |
1970 | return interleave_nodes(policy); | |
1971 | ||
b27abacc DH |
1972 | case MPOL_BIND: |
1973 | case MPOL_PREFERRED_MANY: | |
1974 | { | |
c33d6c06 MG |
1975 | struct zoneref *z; |
1976 | ||
dc85da15 CL |
1977 | /* |
1978 | * Follow bind policy behavior and start allocation at the | |
1979 | * first node. | |
1980 | */ | |
19770b32 | 1981 | struct zonelist *zonelist; |
19770b32 | 1982 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1983 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 | 1984 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
269fbe72 | 1985 | &policy->nodes); |
c1093b74 | 1986 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1987 | } |
7858d7bc FT |
1988 | case MPOL_LOCAL: |
1989 | return node; | |
dc85da15 | 1990 | |
dc85da15 | 1991 | default: |
bea904d5 | 1992 | BUG(); |
dc85da15 CL |
1993 | } |
1994 | } | |
1995 | ||
fee83b3a AM |
1996 | /* |
1997 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
269fbe72 | 1998 | * node in pol->nodes (starting from n=0), wrapping around if n exceeds the |
fee83b3a AM |
1999 | * number of present nodes. |
2000 | */ | |
98c70baa | 2001 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 2002 | { |
269fbe72 | 2003 | unsigned nnodes = nodes_weight(pol->nodes); |
f5b087b5 | 2004 | unsigned target; |
fee83b3a AM |
2005 | int i; |
2006 | int nid; | |
1da177e4 | 2007 | |
f5b087b5 DR |
2008 | if (!nnodes) |
2009 | return numa_node_id(); | |
fee83b3a | 2010 | target = (unsigned int)n % nnodes; |
269fbe72 | 2011 | nid = first_node(pol->nodes); |
fee83b3a | 2012 | for (i = 0; i < target; i++) |
269fbe72 | 2013 | nid = next_node(nid, pol->nodes); |
1da177e4 LT |
2014 | return nid; |
2015 | } | |
2016 | ||
5da7ca86 CL |
2017 | /* Determine a node number for interleave */ |
2018 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
2019 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
2020 | { | |
2021 | if (vma) { | |
2022 | unsigned long off; | |
2023 | ||
3b98b087 NA |
2024 | /* |
2025 | * for small pages, there is no difference between | |
2026 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
2027 | * for huge pages, since vm_pgoff is in units of small | |
2028 | * pages, we need to shift off the always 0 bits to get | |
2029 | * a useful offset. | |
2030 | */ | |
2031 | BUG_ON(shift < PAGE_SHIFT); | |
2032 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 2033 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 2034 | return offset_il_node(pol, off); |
5da7ca86 CL |
2035 | } else |
2036 | return interleave_nodes(pol); | |
2037 | } | |
2038 | ||
00ac59ad | 2039 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2040 | /* |
04ec6264 | 2041 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2042 | * @vma: virtual memory area whose policy is sought |
2043 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2044 | * @gfp_flags: for requested zone | |
2045 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
b27abacc | 2046 | * @nodemask: pointer to nodemask pointer for 'bind' and 'prefer-many' policy |
480eccf9 | 2047 | * |
04ec6264 | 2048 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 | 2049 | * to the struct mempolicy for conditional unref after allocation. |
b27abacc DH |
2050 | * If the effective policy is 'bind' or 'prefer-many', returns a pointer |
2051 | * to the mempolicy's @nodemask for filtering the zonelist. | |
c0ff7453 | 2052 | * |
d26914d1 | 2053 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2054 | */ |
04ec6264 VB |
2055 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2056 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2057 | { |
04ec6264 | 2058 | int nid; |
b27abacc | 2059 | int mode; |
5da7ca86 | 2060 | |
dd6eecb9 | 2061 | *mpol = get_vma_policy(vma, addr); |
b27abacc DH |
2062 | *nodemask = NULL; |
2063 | mode = (*mpol)->mode; | |
5da7ca86 | 2064 | |
b27abacc | 2065 | if (unlikely(mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2066 | nid = interleave_nid(*mpol, vma, addr, |
2067 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2068 | } else { |
04ec6264 | 2069 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
b27abacc | 2070 | if (mode == MPOL_BIND || mode == MPOL_PREFERRED_MANY) |
269fbe72 | 2071 | *nodemask = &(*mpol)->nodes; |
480eccf9 | 2072 | } |
04ec6264 | 2073 | return nid; |
5da7ca86 | 2074 | } |
06808b08 LS |
2075 | |
2076 | /* | |
2077 | * init_nodemask_of_mempolicy | |
2078 | * | |
2079 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2080 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2081 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2082 | * initialize the argument nodemask to contain the single node for | |
2083 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2084 | * of non-default mempolicy. | |
2085 | * | |
2086 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2087 | * because the current task is examining it's own mempolicy and a task's | |
2088 | * mempolicy is only ever changed by the task itself. | |
2089 | * | |
2090 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2091 | */ | |
2092 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2093 | { | |
2094 | struct mempolicy *mempolicy; | |
06808b08 LS |
2095 | |
2096 | if (!(mask && current->mempolicy)) | |
2097 | return false; | |
2098 | ||
c0ff7453 | 2099 | task_lock(current); |
06808b08 LS |
2100 | mempolicy = current->mempolicy; |
2101 | switch (mempolicy->mode) { | |
2102 | case MPOL_PREFERRED: | |
b27abacc | 2103 | case MPOL_PREFERRED_MANY: |
06808b08 | 2104 | case MPOL_BIND: |
06808b08 | 2105 | case MPOL_INTERLEAVE: |
269fbe72 | 2106 | *mask = mempolicy->nodes; |
7858d7bc FT |
2107 | break; |
2108 | ||
2109 | case MPOL_LOCAL: | |
269fbe72 | 2110 | init_nodemask_of_node(mask, numa_node_id()); |
06808b08 LS |
2111 | break; |
2112 | ||
2113 | default: | |
2114 | BUG(); | |
2115 | } | |
c0ff7453 | 2116 | task_unlock(current); |
06808b08 LS |
2117 | |
2118 | return true; | |
2119 | } | |
00ac59ad | 2120 | #endif |
5da7ca86 | 2121 | |
6f48d0eb | 2122 | /* |
b26e517a | 2123 | * mempolicy_in_oom_domain |
6f48d0eb | 2124 | * |
b26e517a FT |
2125 | * If tsk's mempolicy is "bind", check for intersection between mask and |
2126 | * the policy nodemask. Otherwise, return true for all other policies | |
2127 | * including "interleave", as a tsk with "interleave" policy may have | |
2128 | * memory allocated from all nodes in system. | |
6f48d0eb DR |
2129 | * |
2130 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2131 | */ | |
b26e517a | 2132 | bool mempolicy_in_oom_domain(struct task_struct *tsk, |
6f48d0eb DR |
2133 | const nodemask_t *mask) |
2134 | { | |
2135 | struct mempolicy *mempolicy; | |
2136 | bool ret = true; | |
2137 | ||
2138 | if (!mask) | |
2139 | return ret; | |
b26e517a | 2140 | |
6f48d0eb DR |
2141 | task_lock(tsk); |
2142 | mempolicy = tsk->mempolicy; | |
b26e517a | 2143 | if (mempolicy && mempolicy->mode == MPOL_BIND) |
269fbe72 | 2144 | ret = nodes_intersects(mempolicy->nodes, *mask); |
6f48d0eb | 2145 | task_unlock(tsk); |
b26e517a | 2146 | |
6f48d0eb DR |
2147 | return ret; |
2148 | } | |
2149 | ||
1da177e4 LT |
2150 | /* Allocate a page in interleaved policy. |
2151 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2152 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2153 | unsigned nid) | |
1da177e4 | 2154 | { |
1da177e4 LT |
2155 | struct page *page; |
2156 | ||
84172f4b | 2157 | page = __alloc_pages(gfp, order, nid, NULL); |
4518085e KW |
2158 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2159 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2160 | return page; | |
de55c8b2 AR |
2161 | if (page && page_to_nid(page) == nid) { |
2162 | preempt_disable(); | |
f19298b9 | 2163 | __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT); |
de55c8b2 AR |
2164 | preempt_enable(); |
2165 | } | |
1da177e4 LT |
2166 | return page; |
2167 | } | |
2168 | ||
4c54d949 FT |
2169 | static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order, |
2170 | int nid, struct mempolicy *pol) | |
2171 | { | |
2172 | struct page *page; | |
2173 | gfp_t preferred_gfp; | |
2174 | ||
2175 | /* | |
2176 | * This is a two pass approach. The first pass will only try the | |
2177 | * preferred nodes but skip the direct reclaim and allow the | |
2178 | * allocation to fail, while the second pass will try all the | |
2179 | * nodes in system. | |
2180 | */ | |
2181 | preferred_gfp = gfp | __GFP_NOWARN; | |
2182 | preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); | |
2183 | page = __alloc_pages(preferred_gfp, order, nid, &pol->nodes); | |
2184 | if (!page) | |
2185 | page = __alloc_pages(gfp, order, numa_node_id(), NULL); | |
2186 | ||
2187 | return page; | |
2188 | } | |
2189 | ||
1da177e4 | 2190 | /** |
eb350739 MWO |
2191 | * alloc_pages_vma - Allocate a page for a VMA. |
2192 | * @gfp: GFP flags. | |
2193 | * @order: Order of the GFP allocation. | |
2194 | * @vma: Pointer to VMA or NULL if not available. | |
2195 | * @addr: Virtual address of the allocation. Must be inside @vma. | |
2196 | * @node: Which node to prefer for allocation (modulo policy). | |
2197 | * @hugepage: For hugepages try only the preferred node if possible. | |
1da177e4 | 2198 | * |
eb350739 MWO |
2199 | * Allocate a page for a specific address in @vma, using the appropriate |
2200 | * NUMA policy. When @vma is not NULL the caller must hold the mmap_lock | |
2201 | * of the mm_struct of the VMA to prevent it from going away. Should be | |
2202 | * used for all allocations for pages that will be mapped into user space. | |
1da177e4 | 2203 | * |
eb350739 | 2204 | * Return: The page on success or NULL if allocation fails. |
1da177e4 | 2205 | */ |
eb350739 | 2206 | struct page *alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
19deb769 | 2207 | unsigned long addr, int node, bool hugepage) |
1da177e4 | 2208 | { |
cc9a6c87 | 2209 | struct mempolicy *pol; |
c0ff7453 | 2210 | struct page *page; |
04ec6264 | 2211 | int preferred_nid; |
be97a41b | 2212 | nodemask_t *nmask; |
cc9a6c87 | 2213 | |
dd6eecb9 | 2214 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2215 | |
0867a57c VB |
2216 | if (pol->mode == MPOL_INTERLEAVE) { |
2217 | unsigned nid; | |
2218 | ||
2219 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2220 | mpol_cond_put(pol); | |
2221 | page = alloc_page_interleave(gfp, order, nid); | |
2222 | goto out; | |
19deb769 DR |
2223 | } |
2224 | ||
4c54d949 FT |
2225 | if (pol->mode == MPOL_PREFERRED_MANY) { |
2226 | page = alloc_pages_preferred_many(gfp, order, node, pol); | |
2227 | mpol_cond_put(pol); | |
2228 | goto out; | |
2229 | } | |
2230 | ||
19deb769 DR |
2231 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { |
2232 | int hpage_node = node; | |
2233 | ||
2234 | /* | |
2235 | * For hugepage allocation and non-interleave policy which | |
2236 | * allows the current node (or other explicitly preferred | |
2237 | * node) we only try to allocate from the current/preferred | |
2238 | * node and don't fall back to other nodes, as the cost of | |
2239 | * remote accesses would likely offset THP benefits. | |
2240 | * | |
b27abacc | 2241 | * If the policy is interleave or does not allow the current |
19deb769 DR |
2242 | * node in its nodemask, we allocate the standard way. |
2243 | */ | |
7858d7bc | 2244 | if (pol->mode == MPOL_PREFERRED) |
269fbe72 | 2245 | hpage_node = first_node(pol->nodes); |
19deb769 DR |
2246 | |
2247 | nmask = policy_nodemask(gfp, pol); | |
2248 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2249 | mpol_cond_put(pol); | |
cc638f32 VB |
2250 | /* |
2251 | * First, try to allocate THP only on local node, but | |
2252 | * don't reclaim unnecessarily, just compact. | |
2253 | */ | |
19deb769 | 2254 | page = __alloc_pages_node(hpage_node, |
cc638f32 | 2255 | gfp | __GFP_THISNODE | __GFP_NORETRY, order); |
76e654cc DR |
2256 | |
2257 | /* | |
2258 | * If hugepage allocations are configured to always | |
2259 | * synchronous compact or the vma has been madvised | |
2260 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2261 | * memory with both reclaim and compact as well. |
76e654cc DR |
2262 | */ |
2263 | if (!page && (gfp & __GFP_DIRECT_RECLAIM)) | |
2264 | page = __alloc_pages_node(hpage_node, | |
cc638f32 | 2265 | gfp, order); |
76e654cc | 2266 | |
19deb769 DR |
2267 | goto out; |
2268 | } | |
356ff8a9 DR |
2269 | } |
2270 | ||
be97a41b | 2271 | nmask = policy_nodemask(gfp, pol); |
04ec6264 | 2272 | preferred_nid = policy_node(gfp, pol, node); |
84172f4b | 2273 | page = __alloc_pages(gfp, order, preferred_nid, nmask); |
d51e9894 | 2274 | mpol_cond_put(pol); |
be97a41b | 2275 | out: |
c0ff7453 | 2276 | return page; |
1da177e4 | 2277 | } |
69262215 | 2278 | EXPORT_SYMBOL(alloc_pages_vma); |
1da177e4 LT |
2279 | |
2280 | /** | |
6421ec76 MWO |
2281 | * alloc_pages - Allocate pages. |
2282 | * @gfp: GFP flags. | |
2283 | * @order: Power of two of number of pages to allocate. | |
1da177e4 | 2284 | * |
6421ec76 MWO |
2285 | * Allocate 1 << @order contiguous pages. The physical address of the |
2286 | * first page is naturally aligned (eg an order-3 allocation will be aligned | |
2287 | * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current | |
2288 | * process is honoured when in process context. | |
1da177e4 | 2289 | * |
6421ec76 MWO |
2290 | * Context: Can be called from any context, providing the appropriate GFP |
2291 | * flags are used. | |
2292 | * Return: The page on success or NULL if allocation fails. | |
1da177e4 | 2293 | */ |
d7f946d0 | 2294 | struct page *alloc_pages(gfp_t gfp, unsigned order) |
1da177e4 | 2295 | { |
8d90274b | 2296 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2297 | struct page *page; |
1da177e4 | 2298 | |
8d90274b ON |
2299 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2300 | pol = get_task_policy(current); | |
52cd3b07 LS |
2301 | |
2302 | /* | |
2303 | * No reference counting needed for current->mempolicy | |
2304 | * nor system default_policy | |
2305 | */ | |
45c4745a | 2306 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 | 2307 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
4c54d949 FT |
2308 | else if (pol->mode == MPOL_PREFERRED_MANY) |
2309 | page = alloc_pages_preferred_many(gfp, order, | |
2310 | numa_node_id(), pol); | |
c0ff7453 | 2311 | else |
84172f4b | 2312 | page = __alloc_pages(gfp, order, |
04ec6264 | 2313 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2314 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2315 | |
c0ff7453 | 2316 | return page; |
1da177e4 | 2317 | } |
d7f946d0 | 2318 | EXPORT_SYMBOL(alloc_pages); |
1da177e4 | 2319 | |
ef0855d3 ON |
2320 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2321 | { | |
2322 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2323 | ||
2324 | if (IS_ERR(pol)) | |
2325 | return PTR_ERR(pol); | |
2326 | dst->vm_policy = pol; | |
2327 | return 0; | |
2328 | } | |
2329 | ||
4225399a | 2330 | /* |
846a16bf | 2331 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2332 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2333 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2334 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2335 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2336 | * |
2337 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2338 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2339 | */ |
4225399a | 2340 | |
846a16bf LS |
2341 | /* Slow path of a mempolicy duplicate */ |
2342 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2343 | { |
2344 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2345 | ||
2346 | if (!new) | |
2347 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2348 | |
2349 | /* task's mempolicy is protected by alloc_lock */ | |
2350 | if (old == current->mempolicy) { | |
2351 | task_lock(current); | |
2352 | *new = *old; | |
2353 | task_unlock(current); | |
2354 | } else | |
2355 | *new = *old; | |
2356 | ||
4225399a PJ |
2357 | if (current_cpuset_is_being_rebound()) { |
2358 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2359 | mpol_rebind_policy(new, &mems); |
4225399a | 2360 | } |
1da177e4 | 2361 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2362 | return new; |
2363 | } | |
2364 | ||
2365 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2366 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2367 | { |
2368 | if (!a || !b) | |
fcfb4dcc | 2369 | return false; |
45c4745a | 2370 | if (a->mode != b->mode) |
fcfb4dcc | 2371 | return false; |
19800502 | 2372 | if (a->flags != b->flags) |
fcfb4dcc | 2373 | return false; |
19800502 BL |
2374 | if (mpol_store_user_nodemask(a)) |
2375 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2376 | return false; |
19800502 | 2377 | |
45c4745a | 2378 | switch (a->mode) { |
19770b32 | 2379 | case MPOL_BIND: |
1da177e4 | 2380 | case MPOL_INTERLEAVE: |
1da177e4 | 2381 | case MPOL_PREFERRED: |
b27abacc | 2382 | case MPOL_PREFERRED_MANY: |
269fbe72 | 2383 | return !!nodes_equal(a->nodes, b->nodes); |
7858d7bc FT |
2384 | case MPOL_LOCAL: |
2385 | return true; | |
1da177e4 LT |
2386 | default: |
2387 | BUG(); | |
fcfb4dcc | 2388 | return false; |
1da177e4 LT |
2389 | } |
2390 | } | |
2391 | ||
1da177e4 LT |
2392 | /* |
2393 | * Shared memory backing store policy support. | |
2394 | * | |
2395 | * Remember policies even when nobody has shared memory mapped. | |
2396 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2397 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2398 | * for any accesses to the tree. |
2399 | */ | |
2400 | ||
4a8c7bb5 NZ |
2401 | /* |
2402 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2403 | * reading or for writing | |
2404 | */ | |
1da177e4 LT |
2405 | static struct sp_node * |
2406 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2407 | { | |
2408 | struct rb_node *n = sp->root.rb_node; | |
2409 | ||
2410 | while (n) { | |
2411 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2412 | ||
2413 | if (start >= p->end) | |
2414 | n = n->rb_right; | |
2415 | else if (end <= p->start) | |
2416 | n = n->rb_left; | |
2417 | else | |
2418 | break; | |
2419 | } | |
2420 | if (!n) | |
2421 | return NULL; | |
2422 | for (;;) { | |
2423 | struct sp_node *w = NULL; | |
2424 | struct rb_node *prev = rb_prev(n); | |
2425 | if (!prev) | |
2426 | break; | |
2427 | w = rb_entry(prev, struct sp_node, nd); | |
2428 | if (w->end <= start) | |
2429 | break; | |
2430 | n = prev; | |
2431 | } | |
2432 | return rb_entry(n, struct sp_node, nd); | |
2433 | } | |
2434 | ||
4a8c7bb5 NZ |
2435 | /* |
2436 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2437 | * writing. | |
2438 | */ | |
1da177e4 LT |
2439 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2440 | { | |
2441 | struct rb_node **p = &sp->root.rb_node; | |
2442 | struct rb_node *parent = NULL; | |
2443 | struct sp_node *nd; | |
2444 | ||
2445 | while (*p) { | |
2446 | parent = *p; | |
2447 | nd = rb_entry(parent, struct sp_node, nd); | |
2448 | if (new->start < nd->start) | |
2449 | p = &(*p)->rb_left; | |
2450 | else if (new->end > nd->end) | |
2451 | p = &(*p)->rb_right; | |
2452 | else | |
2453 | BUG(); | |
2454 | } | |
2455 | rb_link_node(&new->nd, parent, p); | |
2456 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2457 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2458 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2459 | } |
2460 | ||
2461 | /* Find shared policy intersecting idx */ | |
2462 | struct mempolicy * | |
2463 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2464 | { | |
2465 | struct mempolicy *pol = NULL; | |
2466 | struct sp_node *sn; | |
2467 | ||
2468 | if (!sp->root.rb_node) | |
2469 | return NULL; | |
4a8c7bb5 | 2470 | read_lock(&sp->lock); |
1da177e4 LT |
2471 | sn = sp_lookup(sp, idx, idx+1); |
2472 | if (sn) { | |
2473 | mpol_get(sn->policy); | |
2474 | pol = sn->policy; | |
2475 | } | |
4a8c7bb5 | 2476 | read_unlock(&sp->lock); |
1da177e4 LT |
2477 | return pol; |
2478 | } | |
2479 | ||
63f74ca2 KM |
2480 | static void sp_free(struct sp_node *n) |
2481 | { | |
2482 | mpol_put(n->policy); | |
2483 | kmem_cache_free(sn_cache, n); | |
2484 | } | |
2485 | ||
771fb4d8 LS |
2486 | /** |
2487 | * mpol_misplaced - check whether current page node is valid in policy | |
2488 | * | |
b46e14ac FF |
2489 | * @page: page to be checked |
2490 | * @vma: vm area where page mapped | |
2491 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2492 | * |
2493 | * Lookup current policy node id for vma,addr and "compare to" page's | |
5f076944 | 2494 | * node id. Policy determination "mimics" alloc_page_vma(). |
771fb4d8 | 2495 | * Called from fault path where we know the vma and faulting address. |
5f076944 | 2496 | * |
062db293 BW |
2497 | * Return: NUMA_NO_NODE if the page is in a node that is valid for this |
2498 | * policy, or a suitable node ID to allocate a replacement page from. | |
771fb4d8 LS |
2499 | */ |
2500 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2501 | { | |
2502 | struct mempolicy *pol; | |
c33d6c06 | 2503 | struct zoneref *z; |
771fb4d8 LS |
2504 | int curnid = page_to_nid(page); |
2505 | unsigned long pgoff; | |
90572890 PZ |
2506 | int thiscpu = raw_smp_processor_id(); |
2507 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2508 | int polnid = NUMA_NO_NODE; |
062db293 | 2509 | int ret = NUMA_NO_NODE; |
771fb4d8 | 2510 | |
dd6eecb9 | 2511 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2512 | if (!(pol->flags & MPOL_F_MOF)) |
2513 | goto out; | |
2514 | ||
2515 | switch (pol->mode) { | |
2516 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2517 | pgoff = vma->vm_pgoff; |
2518 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2519 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2520 | break; |
2521 | ||
2522 | case MPOL_PREFERRED: | |
b27abacc DH |
2523 | if (node_isset(curnid, pol->nodes)) |
2524 | goto out; | |
269fbe72 | 2525 | polnid = first_node(pol->nodes); |
7858d7bc FT |
2526 | break; |
2527 | ||
2528 | case MPOL_LOCAL: | |
2529 | polnid = numa_node_id(); | |
771fb4d8 LS |
2530 | break; |
2531 | ||
2532 | case MPOL_BIND: | |
bda420b9 HY |
2533 | /* Optimize placement among multiple nodes via NUMA balancing */ |
2534 | if (pol->flags & MPOL_F_MORON) { | |
269fbe72 | 2535 | if (node_isset(thisnid, pol->nodes)) |
bda420b9 HY |
2536 | break; |
2537 | goto out; | |
2538 | } | |
b27abacc | 2539 | fallthrough; |
c33d6c06 | 2540 | |
b27abacc | 2541 | case MPOL_PREFERRED_MANY: |
771fb4d8 | 2542 | /* |
771fb4d8 LS |
2543 | * use current page if in policy nodemask, |
2544 | * else select nearest allowed node, if any. | |
2545 | * If no allowed nodes, use current [!misplaced]. | |
2546 | */ | |
269fbe72 | 2547 | if (node_isset(curnid, pol->nodes)) |
771fb4d8 | 2548 | goto out; |
c33d6c06 | 2549 | z = first_zones_zonelist( |
771fb4d8 LS |
2550 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2551 | gfp_zone(GFP_HIGHUSER), | |
269fbe72 | 2552 | &pol->nodes); |
c1093b74 | 2553 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2554 | break; |
2555 | ||
2556 | default: | |
2557 | BUG(); | |
2558 | } | |
5606e387 MG |
2559 | |
2560 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2561 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2562 | polnid = thisnid; |
5606e387 | 2563 | |
10f39042 | 2564 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2565 | goto out; |
e42c8ff2 MG |
2566 | } |
2567 | ||
771fb4d8 LS |
2568 | if (curnid != polnid) |
2569 | ret = polnid; | |
2570 | out: | |
2571 | mpol_cond_put(pol); | |
2572 | ||
2573 | return ret; | |
2574 | } | |
2575 | ||
c11600e4 DR |
2576 | /* |
2577 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2578 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2579 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2580 | * policy. | |
2581 | */ | |
2582 | void mpol_put_task_policy(struct task_struct *task) | |
2583 | { | |
2584 | struct mempolicy *pol; | |
2585 | ||
2586 | task_lock(task); | |
2587 | pol = task->mempolicy; | |
2588 | task->mempolicy = NULL; | |
2589 | task_unlock(task); | |
2590 | mpol_put(pol); | |
2591 | } | |
2592 | ||
1da177e4 LT |
2593 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2594 | { | |
140d5a49 | 2595 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2596 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2597 | sp_free(n); |
1da177e4 LT |
2598 | } |
2599 | ||
42288fe3 MG |
2600 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2601 | unsigned long end, struct mempolicy *pol) | |
2602 | { | |
2603 | node->start = start; | |
2604 | node->end = end; | |
2605 | node->policy = pol; | |
2606 | } | |
2607 | ||
dbcb0f19 AB |
2608 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2609 | struct mempolicy *pol) | |
1da177e4 | 2610 | { |
869833f2 KM |
2611 | struct sp_node *n; |
2612 | struct mempolicy *newpol; | |
1da177e4 | 2613 | |
869833f2 | 2614 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2615 | if (!n) |
2616 | return NULL; | |
869833f2 KM |
2617 | |
2618 | newpol = mpol_dup(pol); | |
2619 | if (IS_ERR(newpol)) { | |
2620 | kmem_cache_free(sn_cache, n); | |
2621 | return NULL; | |
2622 | } | |
2623 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2624 | sp_node_init(n, start, end, newpol); |
869833f2 | 2625 | |
1da177e4 LT |
2626 | return n; |
2627 | } | |
2628 | ||
2629 | /* Replace a policy range. */ | |
2630 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2631 | unsigned long end, struct sp_node *new) | |
2632 | { | |
b22d127a | 2633 | struct sp_node *n; |
42288fe3 MG |
2634 | struct sp_node *n_new = NULL; |
2635 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2636 | int ret = 0; |
1da177e4 | 2637 | |
42288fe3 | 2638 | restart: |
4a8c7bb5 | 2639 | write_lock(&sp->lock); |
1da177e4 LT |
2640 | n = sp_lookup(sp, start, end); |
2641 | /* Take care of old policies in the same range. */ | |
2642 | while (n && n->start < end) { | |
2643 | struct rb_node *next = rb_next(&n->nd); | |
2644 | if (n->start >= start) { | |
2645 | if (n->end <= end) | |
2646 | sp_delete(sp, n); | |
2647 | else | |
2648 | n->start = end; | |
2649 | } else { | |
2650 | /* Old policy spanning whole new range. */ | |
2651 | if (n->end > end) { | |
42288fe3 MG |
2652 | if (!n_new) |
2653 | goto alloc_new; | |
2654 | ||
2655 | *mpol_new = *n->policy; | |
2656 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2657 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2658 | n->end = start; |
5ca39575 | 2659 | sp_insert(sp, n_new); |
42288fe3 MG |
2660 | n_new = NULL; |
2661 | mpol_new = NULL; | |
1da177e4 LT |
2662 | break; |
2663 | } else | |
2664 | n->end = start; | |
2665 | } | |
2666 | if (!next) | |
2667 | break; | |
2668 | n = rb_entry(next, struct sp_node, nd); | |
2669 | } | |
2670 | if (new) | |
2671 | sp_insert(sp, new); | |
4a8c7bb5 | 2672 | write_unlock(&sp->lock); |
42288fe3 MG |
2673 | ret = 0; |
2674 | ||
2675 | err_out: | |
2676 | if (mpol_new) | |
2677 | mpol_put(mpol_new); | |
2678 | if (n_new) | |
2679 | kmem_cache_free(sn_cache, n_new); | |
2680 | ||
b22d127a | 2681 | return ret; |
42288fe3 MG |
2682 | |
2683 | alloc_new: | |
4a8c7bb5 | 2684 | write_unlock(&sp->lock); |
42288fe3 MG |
2685 | ret = -ENOMEM; |
2686 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2687 | if (!n_new) | |
2688 | goto err_out; | |
2689 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2690 | if (!mpol_new) | |
2691 | goto err_out; | |
2692 | goto restart; | |
1da177e4 LT |
2693 | } |
2694 | ||
71fe804b LS |
2695 | /** |
2696 | * mpol_shared_policy_init - initialize shared policy for inode | |
2697 | * @sp: pointer to inode shared policy | |
2698 | * @mpol: struct mempolicy to install | |
2699 | * | |
2700 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2701 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2702 | * This must be released on exit. | |
4bfc4495 | 2703 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2704 | */ |
2705 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2706 | { | |
58568d2a MX |
2707 | int ret; |
2708 | ||
71fe804b | 2709 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2710 | rwlock_init(&sp->lock); |
71fe804b LS |
2711 | |
2712 | if (mpol) { | |
2713 | struct vm_area_struct pvma; | |
2714 | struct mempolicy *new; | |
4bfc4495 | 2715 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2716 | |
4bfc4495 | 2717 | if (!scratch) |
5c0c1654 | 2718 | goto put_mpol; |
71fe804b LS |
2719 | /* contextualize the tmpfs mount point mempolicy */ |
2720 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2721 | if (IS_ERR(new)) |
0cae3457 | 2722 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2723 | |
2724 | task_lock(current); | |
4bfc4495 | 2725 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2726 | task_unlock(current); |
15d77835 | 2727 | if (ret) |
5c0c1654 | 2728 | goto put_new; |
71fe804b LS |
2729 | |
2730 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2731 | vma_init(&pvma, NULL); |
71fe804b LS |
2732 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2733 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2734 | |
5c0c1654 | 2735 | put_new: |
71fe804b | 2736 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2737 | free_scratch: |
4bfc4495 | 2738 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2739 | put_mpol: |
2740 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2741 | } |
2742 | } | |
2743 | ||
1da177e4 LT |
2744 | int mpol_set_shared_policy(struct shared_policy *info, |
2745 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2746 | { | |
2747 | int err; | |
2748 | struct sp_node *new = NULL; | |
2749 | unsigned long sz = vma_pages(vma); | |
2750 | ||
028fec41 | 2751 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2752 | vma->vm_pgoff, |
45c4745a | 2753 | sz, npol ? npol->mode : -1, |
028fec41 | 2754 | npol ? npol->flags : -1, |
269fbe72 | 2755 | npol ? nodes_addr(npol->nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2756 | |
2757 | if (npol) { | |
2758 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2759 | if (!new) | |
2760 | return -ENOMEM; | |
2761 | } | |
2762 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2763 | if (err && new) | |
63f74ca2 | 2764 | sp_free(new); |
1da177e4 LT |
2765 | return err; |
2766 | } | |
2767 | ||
2768 | /* Free a backing policy store on inode delete. */ | |
2769 | void mpol_free_shared_policy(struct shared_policy *p) | |
2770 | { | |
2771 | struct sp_node *n; | |
2772 | struct rb_node *next; | |
2773 | ||
2774 | if (!p->root.rb_node) | |
2775 | return; | |
4a8c7bb5 | 2776 | write_lock(&p->lock); |
1da177e4 LT |
2777 | next = rb_first(&p->root); |
2778 | while (next) { | |
2779 | n = rb_entry(next, struct sp_node, nd); | |
2780 | next = rb_next(&n->nd); | |
63f74ca2 | 2781 | sp_delete(p, n); |
1da177e4 | 2782 | } |
4a8c7bb5 | 2783 | write_unlock(&p->lock); |
1da177e4 LT |
2784 | } |
2785 | ||
1a687c2e | 2786 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2787 | static int __initdata numabalancing_override; |
1a687c2e MG |
2788 | |
2789 | static void __init check_numabalancing_enable(void) | |
2790 | { | |
2791 | bool numabalancing_default = false; | |
2792 | ||
2793 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2794 | numabalancing_default = true; | |
2795 | ||
c297663c MG |
2796 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2797 | if (numabalancing_override) | |
2798 | set_numabalancing_state(numabalancing_override == 1); | |
2799 | ||
b0dc2b9b | 2800 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2801 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2802 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2803 | set_numabalancing_state(numabalancing_default); |
2804 | } | |
2805 | } | |
2806 | ||
2807 | static int __init setup_numabalancing(char *str) | |
2808 | { | |
2809 | int ret = 0; | |
2810 | if (!str) | |
2811 | goto out; | |
1a687c2e MG |
2812 | |
2813 | if (!strcmp(str, "enable")) { | |
c297663c | 2814 | numabalancing_override = 1; |
1a687c2e MG |
2815 | ret = 1; |
2816 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2817 | numabalancing_override = -1; |
1a687c2e MG |
2818 | ret = 1; |
2819 | } | |
2820 | out: | |
2821 | if (!ret) | |
4a404bea | 2822 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2823 | |
2824 | return ret; | |
2825 | } | |
2826 | __setup("numa_balancing=", setup_numabalancing); | |
2827 | #else | |
2828 | static inline void __init check_numabalancing_enable(void) | |
2829 | { | |
2830 | } | |
2831 | #endif /* CONFIG_NUMA_BALANCING */ | |
2832 | ||
1da177e4 LT |
2833 | /* assumes fs == KERNEL_DS */ |
2834 | void __init numa_policy_init(void) | |
2835 | { | |
b71636e2 PM |
2836 | nodemask_t interleave_nodes; |
2837 | unsigned long largest = 0; | |
2838 | int nid, prefer = 0; | |
2839 | ||
1da177e4 LT |
2840 | policy_cache = kmem_cache_create("numa_policy", |
2841 | sizeof(struct mempolicy), | |
20c2df83 | 2842 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2843 | |
2844 | sn_cache = kmem_cache_create("shared_policy_node", | |
2845 | sizeof(struct sp_node), | |
20c2df83 | 2846 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2847 | |
5606e387 MG |
2848 | for_each_node(nid) { |
2849 | preferred_node_policy[nid] = (struct mempolicy) { | |
2850 | .refcnt = ATOMIC_INIT(1), | |
2851 | .mode = MPOL_PREFERRED, | |
2852 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
269fbe72 | 2853 | .nodes = nodemask_of_node(nid), |
5606e387 MG |
2854 | }; |
2855 | } | |
2856 | ||
b71636e2 PM |
2857 | /* |
2858 | * Set interleaving policy for system init. Interleaving is only | |
2859 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2860 | * fall back to the largest node if they're all smaller. | |
2861 | */ | |
2862 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2863 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2864 | unsigned long total_pages = node_present_pages(nid); |
2865 | ||
2866 | /* Preserve the largest node */ | |
2867 | if (largest < total_pages) { | |
2868 | largest = total_pages; | |
2869 | prefer = nid; | |
2870 | } | |
2871 | ||
2872 | /* Interleave this node? */ | |
2873 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2874 | node_set(nid, interleave_nodes); | |
2875 | } | |
2876 | ||
2877 | /* All too small, use the largest */ | |
2878 | if (unlikely(nodes_empty(interleave_nodes))) | |
2879 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2880 | |
028fec41 | 2881 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2882 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2883 | |
2884 | check_numabalancing_enable(); | |
1da177e4 LT |
2885 | } |
2886 | ||
8bccd85f | 2887 | /* Reset policy of current process to default */ |
1da177e4 LT |
2888 | void numa_default_policy(void) |
2889 | { | |
028fec41 | 2890 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2891 | } |
68860ec1 | 2892 | |
095f1fc4 LS |
2893 | /* |
2894 | * Parse and format mempolicy from/to strings | |
2895 | */ | |
2896 | ||
345ace9c LS |
2897 | static const char * const policy_modes[] = |
2898 | { | |
2899 | [MPOL_DEFAULT] = "default", | |
2900 | [MPOL_PREFERRED] = "prefer", | |
2901 | [MPOL_BIND] = "bind", | |
2902 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2903 | [MPOL_LOCAL] = "local", |
b27abacc | 2904 | [MPOL_PREFERRED_MANY] = "prefer (many)", |
345ace9c | 2905 | }; |
1a75a6c8 | 2906 | |
095f1fc4 LS |
2907 | |
2908 | #ifdef CONFIG_TMPFS | |
2909 | /** | |
f2a07f40 | 2910 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2911 | * @str: string containing mempolicy to parse |
71fe804b | 2912 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2913 | * |
2914 | * Format of input: | |
2915 | * <mode>[=<flags>][:<nodelist>] | |
2916 | * | |
71fe804b | 2917 | * On success, returns 0, else 1 |
095f1fc4 | 2918 | */ |
a7a88b23 | 2919 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2920 | { |
71fe804b | 2921 | struct mempolicy *new = NULL; |
f2a07f40 | 2922 | unsigned short mode_flags; |
71fe804b | 2923 | nodemask_t nodes; |
095f1fc4 LS |
2924 | char *nodelist = strchr(str, ':'); |
2925 | char *flags = strchr(str, '='); | |
dedf2c73 | 2926 | int err = 1, mode; |
095f1fc4 | 2927 | |
c7a91bc7 DC |
2928 | if (flags) |
2929 | *flags++ = '\0'; /* terminate mode string */ | |
2930 | ||
095f1fc4 LS |
2931 | if (nodelist) { |
2932 | /* NUL-terminate mode or flags string */ | |
2933 | *nodelist++ = '\0'; | |
71fe804b | 2934 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2935 | goto out; |
01f13bd6 | 2936 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2937 | goto out; |
71fe804b LS |
2938 | } else |
2939 | nodes_clear(nodes); | |
2940 | ||
dedf2c73 | 2941 | mode = match_string(policy_modes, MPOL_MAX, str); |
2942 | if (mode < 0) | |
095f1fc4 LS |
2943 | goto out; |
2944 | ||
71fe804b | 2945 | switch (mode) { |
095f1fc4 | 2946 | case MPOL_PREFERRED: |
71fe804b | 2947 | /* |
aa9f7d51 RD |
2948 | * Insist on a nodelist of one node only, although later |
2949 | * we use first_node(nodes) to grab a single node, so here | |
2950 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 2951 | */ |
095f1fc4 LS |
2952 | if (nodelist) { |
2953 | char *rest = nodelist; | |
2954 | while (isdigit(*rest)) | |
2955 | rest++; | |
926f2ae0 KM |
2956 | if (*rest) |
2957 | goto out; | |
aa9f7d51 RD |
2958 | if (nodes_empty(nodes)) |
2959 | goto out; | |
095f1fc4 LS |
2960 | } |
2961 | break; | |
095f1fc4 LS |
2962 | case MPOL_INTERLEAVE: |
2963 | /* | |
2964 | * Default to online nodes with memory if no nodelist | |
2965 | */ | |
2966 | if (!nodelist) | |
01f13bd6 | 2967 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2968 | break; |
71fe804b | 2969 | case MPOL_LOCAL: |
3f226aa1 | 2970 | /* |
71fe804b | 2971 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2972 | */ |
71fe804b | 2973 | if (nodelist) |
3f226aa1 | 2974 | goto out; |
3f226aa1 | 2975 | break; |
413b43de RT |
2976 | case MPOL_DEFAULT: |
2977 | /* | |
2978 | * Insist on a empty nodelist | |
2979 | */ | |
2980 | if (!nodelist) | |
2981 | err = 0; | |
2982 | goto out; | |
b27abacc | 2983 | case MPOL_PREFERRED_MANY: |
d69b2e63 KM |
2984 | case MPOL_BIND: |
2985 | /* | |
2986 | * Insist on a nodelist | |
2987 | */ | |
2988 | if (!nodelist) | |
2989 | goto out; | |
095f1fc4 LS |
2990 | } |
2991 | ||
71fe804b | 2992 | mode_flags = 0; |
095f1fc4 LS |
2993 | if (flags) { |
2994 | /* | |
2995 | * Currently, we only support two mutually exclusive | |
2996 | * mode flags. | |
2997 | */ | |
2998 | if (!strcmp(flags, "static")) | |
71fe804b | 2999 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 3000 | else if (!strcmp(flags, "relative")) |
71fe804b | 3001 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 3002 | else |
926f2ae0 | 3003 | goto out; |
095f1fc4 | 3004 | } |
71fe804b LS |
3005 | |
3006 | new = mpol_new(mode, mode_flags, &nodes); | |
3007 | if (IS_ERR(new)) | |
926f2ae0 KM |
3008 | goto out; |
3009 | ||
f2a07f40 HD |
3010 | /* |
3011 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
3012 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
3013 | */ | |
269fbe72 BW |
3014 | if (mode != MPOL_PREFERRED) { |
3015 | new->nodes = nodes; | |
3016 | } else if (nodelist) { | |
3017 | nodes_clear(new->nodes); | |
3018 | node_set(first_node(nodes), new->nodes); | |
3019 | } else { | |
7858d7bc | 3020 | new->mode = MPOL_LOCAL; |
269fbe72 | 3021 | } |
f2a07f40 HD |
3022 | |
3023 | /* | |
3024 | * Save nodes for contextualization: this will be used to "clone" | |
3025 | * the mempolicy in a specific context [cpuset] at a later time. | |
3026 | */ | |
3027 | new->w.user_nodemask = nodes; | |
3028 | ||
926f2ae0 | 3029 | err = 0; |
71fe804b | 3030 | |
095f1fc4 LS |
3031 | out: |
3032 | /* Restore string for error message */ | |
3033 | if (nodelist) | |
3034 | *--nodelist = ':'; | |
3035 | if (flags) | |
3036 | *--flags = '='; | |
71fe804b LS |
3037 | if (!err) |
3038 | *mpol = new; | |
095f1fc4 LS |
3039 | return err; |
3040 | } | |
3041 | #endif /* CONFIG_TMPFS */ | |
3042 | ||
71fe804b LS |
3043 | /** |
3044 | * mpol_to_str - format a mempolicy structure for printing | |
3045 | * @buffer: to contain formatted mempolicy string | |
3046 | * @maxlen: length of @buffer | |
3047 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3048 | * |
948927ee DR |
3049 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3050 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3051 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3052 | */ |
948927ee | 3053 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3054 | { |
3055 | char *p = buffer; | |
948927ee DR |
3056 | nodemask_t nodes = NODE_MASK_NONE; |
3057 | unsigned short mode = MPOL_DEFAULT; | |
3058 | unsigned short flags = 0; | |
2291990a | 3059 | |
8790c71a | 3060 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3061 | mode = pol->mode; |
948927ee DR |
3062 | flags = pol->flags; |
3063 | } | |
bea904d5 | 3064 | |
1a75a6c8 CL |
3065 | switch (mode) { |
3066 | case MPOL_DEFAULT: | |
7858d7bc | 3067 | case MPOL_LOCAL: |
1a75a6c8 | 3068 | break; |
1a75a6c8 | 3069 | case MPOL_PREFERRED: |
b27abacc | 3070 | case MPOL_PREFERRED_MANY: |
1a75a6c8 | 3071 | case MPOL_BIND: |
1a75a6c8 | 3072 | case MPOL_INTERLEAVE: |
269fbe72 | 3073 | nodes = pol->nodes; |
1a75a6c8 | 3074 | break; |
1a75a6c8 | 3075 | default: |
948927ee DR |
3076 | WARN_ON_ONCE(1); |
3077 | snprintf(p, maxlen, "unknown"); | |
3078 | return; | |
1a75a6c8 CL |
3079 | } |
3080 | ||
b7a9f420 | 3081 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3082 | |
fc36b8d3 | 3083 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3084 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3085 | |
2291990a LS |
3086 | /* |
3087 | * Currently, the only defined flags are mutually exclusive | |
3088 | */ | |
f5b087b5 | 3089 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3090 | p += snprintf(p, buffer + maxlen - p, "static"); |
3091 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3092 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3093 | } |
3094 | ||
9e763e0f TH |
3095 | if (!nodes_empty(nodes)) |
3096 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3097 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3098 | } |
20b51af1 HY |
3099 | |
3100 | bool numa_demotion_enabled = false; | |
3101 | ||
3102 | #ifdef CONFIG_SYSFS | |
3103 | static ssize_t numa_demotion_enabled_show(struct kobject *kobj, | |
3104 | struct kobj_attribute *attr, char *buf) | |
3105 | { | |
3106 | return sysfs_emit(buf, "%s\n", | |
3107 | numa_demotion_enabled? "true" : "false"); | |
3108 | } | |
3109 | ||
3110 | static ssize_t numa_demotion_enabled_store(struct kobject *kobj, | |
3111 | struct kobj_attribute *attr, | |
3112 | const char *buf, size_t count) | |
3113 | { | |
3114 | if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1)) | |
3115 | numa_demotion_enabled = true; | |
3116 | else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1)) | |
3117 | numa_demotion_enabled = false; | |
3118 | else | |
3119 | return -EINVAL; | |
3120 | ||
3121 | return count; | |
3122 | } | |
3123 | ||
3124 | static struct kobj_attribute numa_demotion_enabled_attr = | |
3125 | __ATTR(demotion_enabled, 0644, numa_demotion_enabled_show, | |
3126 | numa_demotion_enabled_store); | |
3127 | ||
3128 | static struct attribute *numa_attrs[] = { | |
3129 | &numa_demotion_enabled_attr.attr, | |
3130 | NULL, | |
3131 | }; | |
3132 | ||
3133 | static const struct attribute_group numa_attr_group = { | |
3134 | .attrs = numa_attrs, | |
3135 | }; | |
3136 | ||
3137 | static int __init numa_init_sysfs(void) | |
3138 | { | |
3139 | int err; | |
3140 | struct kobject *numa_kobj; | |
3141 | ||
3142 | numa_kobj = kobject_create_and_add("numa", mm_kobj); | |
3143 | if (!numa_kobj) { | |
3144 | pr_err("failed to create numa kobject\n"); | |
3145 | return -ENOMEM; | |
3146 | } | |
3147 | err = sysfs_create_group(numa_kobj, &numa_attr_group); | |
3148 | if (err) { | |
3149 | pr_err("failed to register numa group\n"); | |
3150 | goto delete_obj; | |
3151 | } | |
3152 | return 0; | |
3153 | ||
3154 | delete_obj: | |
3155 | kobject_put(numa_kobj); | |
3156 | return err; | |
3157 | } | |
3158 | subsys_initcall(numa_init_sysfs); | |
3159 | #endif |