]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Simple NUMA memory policy for the Linux kernel. | |
3 | * | |
4 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 5 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
6 | * Subject to the GNU Public License, version 2. |
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 | * |
1da177e4 LT |
34 | * default Allocate on the local node first, or when on a VMA |
35 | * use the process policy. This is what Linux always did | |
36 | * in a NUMA aware kernel and still does by, ahem, default. | |
37 | * | |
38 | * The process policy is applied for most non interrupt memory allocations | |
39 | * in that process' context. Interrupts ignore the policies and always | |
40 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
41 | * allocations for a VMA in the VM. | |
42 | * | |
43 | * Currently there are a few corner cases in swapping where the policy | |
44 | * is not applied, but the majority should be handled. When process policy | |
45 | * is used it is not remembered over swap outs/swap ins. | |
46 | * | |
47 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
48 | * requesting a lower zone just use default policy. This implies that | |
49 | * on systems with highmem kernel lowmem allocation don't get policied. | |
50 | * Same with GFP_DMA allocations. | |
51 | * | |
52 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between | |
53 | * all users and remembered even when nobody has memory mapped. | |
54 | */ | |
55 | ||
56 | /* Notebook: | |
57 | fix mmap readahead to honour policy and enable policy for any page cache | |
58 | object | |
59 | statistics for bigpages | |
60 | global policy for page cache? currently it uses process policy. Requires | |
61 | first item above. | |
62 | handle mremap for shared memory (currently ignored for the policy) | |
63 | grows down? | |
64 | make bind policy root only? It can trigger oom much faster and the | |
65 | kernel is not always grateful with that. | |
1da177e4 LT |
66 | */ |
67 | ||
b1de0d13 MH |
68 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
69 | ||
1da177e4 LT |
70 | #include <linux/mempolicy.h> |
71 | #include <linux/mm.h> | |
72 | #include <linux/highmem.h> | |
73 | #include <linux/hugetlb.h> | |
74 | #include <linux/kernel.h> | |
75 | #include <linux/sched.h> | |
6e84f315 | 76 | #include <linux/sched/mm.h> |
1da177e4 LT |
77 | #include <linux/nodemask.h> |
78 | #include <linux/cpuset.h> | |
1da177e4 LT |
79 | #include <linux/slab.h> |
80 | #include <linux/string.h> | |
b95f1b31 | 81 | #include <linux/export.h> |
b488893a | 82 | #include <linux/nsproxy.h> |
1da177e4 LT |
83 | #include <linux/interrupt.h> |
84 | #include <linux/init.h> | |
85 | #include <linux/compat.h> | |
dc9aa5b9 | 86 | #include <linux/swap.h> |
1a75a6c8 CL |
87 | #include <linux/seq_file.h> |
88 | #include <linux/proc_fs.h> | |
b20a3503 | 89 | #include <linux/migrate.h> |
62b61f61 | 90 | #include <linux/ksm.h> |
95a402c3 | 91 | #include <linux/rmap.h> |
86c3a764 | 92 | #include <linux/security.h> |
dbcb0f19 | 93 | #include <linux/syscalls.h> |
095f1fc4 | 94 | #include <linux/ctype.h> |
6d9c285a | 95 | #include <linux/mm_inline.h> |
b24f53a0 | 96 | #include <linux/mmu_notifier.h> |
b1de0d13 | 97 | #include <linux/printk.h> |
dc9aa5b9 | 98 | |
1da177e4 | 99 | #include <asm/tlbflush.h> |
7c0f6ba6 | 100 | #include <linux/uaccess.h> |
1da177e4 | 101 | |
62695a84 NP |
102 | #include "internal.h" |
103 | ||
38e35860 | 104 | /* Internal flags */ |
dc9aa5b9 | 105 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 106 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 107 | |
fcc234f8 PE |
108 | static struct kmem_cache *policy_cache; |
109 | static struct kmem_cache *sn_cache; | |
1da177e4 | 110 | |
1da177e4 LT |
111 | /* Highest zone. An specific allocation for a zone below that is not |
112 | policied. */ | |
6267276f | 113 | enum zone_type policy_zone = 0; |
1da177e4 | 114 | |
bea904d5 LS |
115 | /* |
116 | * run-time system-wide default policy => local allocation | |
117 | */ | |
e754d79d | 118 | static struct mempolicy default_policy = { |
1da177e4 | 119 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 120 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 121 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
122 | }; |
123 | ||
5606e387 MG |
124 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
125 | ||
74d2c3a0 | 126 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
127 | { |
128 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 129 | int node; |
5606e387 | 130 | |
f15ca78e ON |
131 | if (pol) |
132 | return pol; | |
5606e387 | 133 | |
f15ca78e ON |
134 | node = numa_node_id(); |
135 | if (node != NUMA_NO_NODE) { | |
136 | pol = &preferred_node_policy[node]; | |
137 | /* preferred_node_policy is not initialised early in boot */ | |
138 | if (pol->mode) | |
139 | return pol; | |
5606e387 MG |
140 | } |
141 | ||
f15ca78e | 142 | return &default_policy; |
5606e387 MG |
143 | } |
144 | ||
37012946 DR |
145 | static const struct mempolicy_operations { |
146 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
708c1bbc MX |
147 | /* |
148 | * If read-side task has no lock to protect task->mempolicy, write-side | |
149 | * task will rebind the task->mempolicy by two step. The first step is | |
150 | * setting all the newly nodes, and the second step is cleaning all the | |
151 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
152 | * page. | |
153 | * If we have a lock to protect task->mempolicy in read-side, we do | |
154 | * rebind directly. | |
155 | * | |
156 | * step: | |
157 | * MPOL_REBIND_ONCE - do rebind work at once | |
158 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
159 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
160 | */ | |
161 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes, | |
162 | enum mpol_rebind_step step); | |
37012946 DR |
163 | } mpol_ops[MPOL_MAX]; |
164 | ||
f5b087b5 DR |
165 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
166 | { | |
6d556294 | 167 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
168 | } |
169 | ||
170 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
171 | const nodemask_t *rel) | |
172 | { | |
173 | nodemask_t tmp; | |
174 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
175 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
176 | } |
177 | ||
37012946 DR |
178 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
179 | { | |
180 | if (nodes_empty(*nodes)) | |
181 | return -EINVAL; | |
182 | pol->v.nodes = *nodes; | |
183 | return 0; | |
184 | } | |
185 | ||
186 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
187 | { | |
188 | if (!nodes) | |
fc36b8d3 | 189 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
190 | else if (nodes_empty(*nodes)) |
191 | return -EINVAL; /* no allowed nodes */ | |
192 | else | |
193 | pol->v.preferred_node = first_node(*nodes); | |
194 | return 0; | |
195 | } | |
196 | ||
197 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
198 | { | |
859f7ef1 | 199 | if (nodes_empty(*nodes)) |
37012946 DR |
200 | return -EINVAL; |
201 | pol->v.nodes = *nodes; | |
202 | return 0; | |
203 | } | |
204 | ||
58568d2a MX |
205 | /* |
206 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
207 | * any, for the new policy. mpol_new() has already validated the nodes | |
208 | * parameter with respect to the policy mode and flags. But, we need to | |
209 | * handle an empty nodemask with MPOL_PREFERRED here. | |
210 | * | |
211 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
212 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
213 | */ | |
4bfc4495 KH |
214 | static int mpol_set_nodemask(struct mempolicy *pol, |
215 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 216 | { |
58568d2a MX |
217 | int ret; |
218 | ||
219 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
220 | if (pol == NULL) | |
221 | return 0; | |
01f13bd6 | 222 | /* Check N_MEMORY */ |
4bfc4495 | 223 | nodes_and(nsc->mask1, |
01f13bd6 | 224 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
225 | |
226 | VM_BUG_ON(!nodes); | |
227 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
228 | nodes = NULL; /* explicit local allocation */ | |
229 | else { | |
230 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
859f7ef1 | 231 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); |
58568d2a | 232 | else |
4bfc4495 KH |
233 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
234 | ||
58568d2a MX |
235 | if (mpol_store_user_nodemask(pol)) |
236 | pol->w.user_nodemask = *nodes; | |
237 | else | |
238 | pol->w.cpuset_mems_allowed = | |
239 | cpuset_current_mems_allowed; | |
240 | } | |
241 | ||
4bfc4495 KH |
242 | if (nodes) |
243 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
244 | else | |
245 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
246 | return ret; |
247 | } | |
248 | ||
249 | /* | |
250 | * This function just creates a new policy, does some check and simple | |
251 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
252 | */ | |
028fec41 DR |
253 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
254 | nodemask_t *nodes) | |
1da177e4 LT |
255 | { |
256 | struct mempolicy *policy; | |
257 | ||
028fec41 | 258 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 259 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 260 | |
3e1f0645 DR |
261 | if (mode == MPOL_DEFAULT) { |
262 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 263 | return ERR_PTR(-EINVAL); |
d3a71033 | 264 | return NULL; |
37012946 | 265 | } |
3e1f0645 DR |
266 | VM_BUG_ON(!nodes); |
267 | ||
268 | /* | |
269 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
270 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
271 | * All other modes require a valid pointer to a non-empty nodemask. | |
272 | */ | |
273 | if (mode == MPOL_PREFERRED) { | |
274 | if (nodes_empty(*nodes)) { | |
275 | if (((flags & MPOL_F_STATIC_NODES) || | |
276 | (flags & MPOL_F_RELATIVE_NODES))) | |
277 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 278 | } |
479e2802 | 279 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
280 | if (!nodes_empty(*nodes) || |
281 | (flags & MPOL_F_STATIC_NODES) || | |
282 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 PZ |
283 | return ERR_PTR(-EINVAL); |
284 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
285 | } else if (nodes_empty(*nodes)) |
286 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
287 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
288 | if (!policy) | |
289 | return ERR_PTR(-ENOMEM); | |
290 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 291 | policy->mode = mode; |
3e1f0645 | 292 | policy->flags = flags; |
37012946 | 293 | |
1da177e4 | 294 | return policy; |
37012946 DR |
295 | } |
296 | ||
52cd3b07 LS |
297 | /* Slow path of a mpol destructor. */ |
298 | void __mpol_put(struct mempolicy *p) | |
299 | { | |
300 | if (!atomic_dec_and_test(&p->refcnt)) | |
301 | return; | |
52cd3b07 LS |
302 | kmem_cache_free(policy_cache, p); |
303 | } | |
304 | ||
708c1bbc MX |
305 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
306 | enum mpol_rebind_step step) | |
37012946 DR |
307 | { |
308 | } | |
309 | ||
708c1bbc MX |
310 | /* |
311 | * step: | |
312 | * MPOL_REBIND_ONCE - do rebind work at once | |
313 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
314 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
315 | */ | |
316 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
317 | enum mpol_rebind_step step) | |
37012946 DR |
318 | { |
319 | nodemask_t tmp; | |
320 | ||
321 | if (pol->flags & MPOL_F_STATIC_NODES) | |
322 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
323 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
324 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
325 | else { | |
708c1bbc MX |
326 | /* |
327 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
328 | * result | |
329 | */ | |
330 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
331 | nodes_remap(tmp, pol->v.nodes, | |
332 | pol->w.cpuset_mems_allowed, *nodes); | |
333 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
334 | } else if (step == MPOL_REBIND_STEP2) { | |
335 | tmp = pol->w.cpuset_mems_allowed; | |
336 | pol->w.cpuset_mems_allowed = *nodes; | |
337 | } else | |
338 | BUG(); | |
37012946 | 339 | } |
f5b087b5 | 340 | |
708c1bbc MX |
341 | if (nodes_empty(tmp)) |
342 | tmp = *nodes; | |
343 | ||
344 | if (step == MPOL_REBIND_STEP1) | |
345 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
346 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
347 | pol->v.nodes = tmp; | |
348 | else | |
349 | BUG(); | |
350 | ||
37012946 | 351 | if (!node_isset(current->il_next, tmp)) { |
0edaf86c | 352 | current->il_next = next_node_in(current->il_next, tmp); |
37012946 DR |
353 | if (current->il_next >= MAX_NUMNODES) |
354 | current->il_next = numa_node_id(); | |
355 | } | |
356 | } | |
357 | ||
358 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
359 | const nodemask_t *nodes, |
360 | enum mpol_rebind_step step) | |
37012946 DR |
361 | { |
362 | nodemask_t tmp; | |
363 | ||
37012946 DR |
364 | if (pol->flags & MPOL_F_STATIC_NODES) { |
365 | int node = first_node(pol->w.user_nodemask); | |
366 | ||
fc36b8d3 | 367 | if (node_isset(node, *nodes)) { |
37012946 | 368 | pol->v.preferred_node = node; |
fc36b8d3 LS |
369 | pol->flags &= ~MPOL_F_LOCAL; |
370 | } else | |
371 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
372 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
373 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
374 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 375 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
376 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
377 | pol->w.cpuset_mems_allowed, | |
378 | *nodes); | |
379 | pol->w.cpuset_mems_allowed = *nodes; | |
380 | } | |
1da177e4 LT |
381 | } |
382 | ||
708c1bbc MX |
383 | /* |
384 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
385 | * | |
386 | * If read-side task has no lock to protect task->mempolicy, write-side | |
387 | * task will rebind the task->mempolicy by two step. The first step is | |
388 | * setting all the newly nodes, and the second step is cleaning all the | |
389 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
390 | * page. | |
391 | * If we have a lock to protect task->mempolicy in read-side, we do | |
392 | * rebind directly. | |
393 | * | |
394 | * step: | |
395 | * MPOL_REBIND_ONCE - do rebind work at once | |
396 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
397 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
398 | */ | |
399 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
400 | enum mpol_rebind_step step) | |
1d0d2680 | 401 | { |
1d0d2680 DR |
402 | if (!pol) |
403 | return; | |
89c522c7 | 404 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
1d0d2680 DR |
405 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
406 | return; | |
708c1bbc MX |
407 | |
408 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
409 | return; | |
410 | ||
411 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
412 | BUG(); | |
413 | ||
414 | if (step == MPOL_REBIND_STEP1) | |
415 | pol->flags |= MPOL_F_REBINDING; | |
416 | else if (step == MPOL_REBIND_STEP2) | |
417 | pol->flags &= ~MPOL_F_REBINDING; | |
418 | else if (step >= MPOL_REBIND_NSTEP) | |
419 | BUG(); | |
420 | ||
421 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
422 | } |
423 | ||
424 | /* | |
425 | * Wrapper for mpol_rebind_policy() that just requires task | |
426 | * pointer, and updates task mempolicy. | |
58568d2a MX |
427 | * |
428 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
429 | */ |
430 | ||
708c1bbc MX |
431 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
432 | enum mpol_rebind_step step) | |
1d0d2680 | 433 | { |
708c1bbc | 434 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
435 | } |
436 | ||
437 | /* | |
438 | * Rebind each vma in mm to new nodemask. | |
439 | * | |
440 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
441 | */ | |
442 | ||
443 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
444 | { | |
445 | struct vm_area_struct *vma; | |
446 | ||
447 | down_write(&mm->mmap_sem); | |
448 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 449 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
450 | up_write(&mm->mmap_sem); |
451 | } | |
452 | ||
37012946 DR |
453 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
454 | [MPOL_DEFAULT] = { | |
455 | .rebind = mpol_rebind_default, | |
456 | }, | |
457 | [MPOL_INTERLEAVE] = { | |
458 | .create = mpol_new_interleave, | |
459 | .rebind = mpol_rebind_nodemask, | |
460 | }, | |
461 | [MPOL_PREFERRED] = { | |
462 | .create = mpol_new_preferred, | |
463 | .rebind = mpol_rebind_preferred, | |
464 | }, | |
465 | [MPOL_BIND] = { | |
466 | .create = mpol_new_bind, | |
467 | .rebind = mpol_rebind_nodemask, | |
468 | }, | |
469 | }; | |
470 | ||
fc301289 CL |
471 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
472 | unsigned long flags); | |
1a75a6c8 | 473 | |
6f4576e3 NH |
474 | struct queue_pages { |
475 | struct list_head *pagelist; | |
476 | unsigned long flags; | |
477 | nodemask_t *nmask; | |
478 | struct vm_area_struct *prev; | |
479 | }; | |
480 | ||
98094945 NH |
481 | /* |
482 | * Scan through pages checking if pages follow certain conditions, | |
483 | * and move them to the pagelist if they do. | |
484 | */ | |
6f4576e3 NH |
485 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
486 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 487 | { |
6f4576e3 NH |
488 | struct vm_area_struct *vma = walk->vma; |
489 | struct page *page; | |
490 | struct queue_pages *qp = walk->private; | |
491 | unsigned long flags = qp->flags; | |
248db92d | 492 | int nid, ret; |
91612e0d | 493 | pte_t *pte; |
705e87c0 | 494 | spinlock_t *ptl; |
941150a3 | 495 | |
248db92d KS |
496 | if (pmd_trans_huge(*pmd)) { |
497 | ptl = pmd_lock(walk->mm, pmd); | |
498 | if (pmd_trans_huge(*pmd)) { | |
499 | page = pmd_page(*pmd); | |
500 | if (is_huge_zero_page(page)) { | |
501 | spin_unlock(ptl); | |
fd60775a | 502 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
248db92d KS |
503 | } else { |
504 | get_page(page); | |
505 | spin_unlock(ptl); | |
506 | lock_page(page); | |
507 | ret = split_huge_page(page); | |
508 | unlock_page(page); | |
509 | put_page(page); | |
510 | if (ret) | |
511 | return 0; | |
512 | } | |
513 | } else { | |
514 | spin_unlock(ptl); | |
515 | } | |
516 | } | |
91612e0d | 517 | |
337d9abf NH |
518 | if (pmd_trans_unstable(pmd)) |
519 | return 0; | |
248db92d | 520 | retry: |
6f4576e3 NH |
521 | pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
522 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
91612e0d | 523 | if (!pte_present(*pte)) |
1da177e4 | 524 | continue; |
6aab341e LT |
525 | page = vm_normal_page(vma, addr, *pte); |
526 | if (!page) | |
1da177e4 | 527 | continue; |
053837fc | 528 | /* |
62b61f61 HD |
529 | * vm_normal_page() filters out zero pages, but there might |
530 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 531 | */ |
b79bc0a0 | 532 | if (PageReserved(page)) |
f4598c8b | 533 | continue; |
6aab341e | 534 | nid = page_to_nid(page); |
6f4576e3 | 535 | if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT)) |
38e35860 | 536 | continue; |
800d8c63 | 537 | if (PageTransCompound(page)) { |
248db92d KS |
538 | get_page(page); |
539 | pte_unmap_unlock(pte, ptl); | |
540 | lock_page(page); | |
541 | ret = split_huge_page(page); | |
542 | unlock_page(page); | |
543 | put_page(page); | |
544 | /* Failed to split -- skip. */ | |
545 | if (ret) { | |
546 | pte = pte_offset_map_lock(walk->mm, pmd, | |
547 | addr, &ptl); | |
548 | continue; | |
549 | } | |
550 | goto retry; | |
551 | } | |
38e35860 | 552 | |
77bf45e7 | 553 | migrate_page_add(page, qp->pagelist, flags); |
6f4576e3 NH |
554 | } |
555 | pte_unmap_unlock(pte - 1, ptl); | |
556 | cond_resched(); | |
557 | return 0; | |
91612e0d HD |
558 | } |
559 | ||
6f4576e3 NH |
560 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
561 | unsigned long addr, unsigned long end, | |
562 | struct mm_walk *walk) | |
e2d8cf40 NH |
563 | { |
564 | #ifdef CONFIG_HUGETLB_PAGE | |
6f4576e3 NH |
565 | struct queue_pages *qp = walk->private; |
566 | unsigned long flags = qp->flags; | |
e2d8cf40 NH |
567 | int nid; |
568 | struct page *page; | |
cb900f41 | 569 | spinlock_t *ptl; |
d4c54919 | 570 | pte_t entry; |
e2d8cf40 | 571 | |
6f4576e3 NH |
572 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
573 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
574 | if (!pte_present(entry)) |
575 | goto unlock; | |
576 | page = pte_page(entry); | |
e2d8cf40 | 577 | nid = page_to_nid(page); |
6f4576e3 | 578 | if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT)) |
e2d8cf40 NH |
579 | goto unlock; |
580 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ | |
581 | if (flags & (MPOL_MF_MOVE_ALL) || | |
582 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) | |
6f4576e3 | 583 | isolate_huge_page(page, qp->pagelist); |
e2d8cf40 | 584 | unlock: |
cb900f41 | 585 | spin_unlock(ptl); |
e2d8cf40 NH |
586 | #else |
587 | BUG(); | |
588 | #endif | |
91612e0d | 589 | return 0; |
1da177e4 LT |
590 | } |
591 | ||
5877231f | 592 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 593 | /* |
4b10e7d5 MG |
594 | * This is used to mark a range of virtual addresses to be inaccessible. |
595 | * These are later cleared by a NUMA hinting fault. Depending on these | |
596 | * faults, pages may be migrated for better NUMA placement. | |
597 | * | |
598 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
599 | * an architecture makes a different choice, it will need further | |
600 | * changes to the core. | |
b24f53a0 | 601 | */ |
4b10e7d5 MG |
602 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
603 | unsigned long addr, unsigned long end) | |
b24f53a0 | 604 | { |
4b10e7d5 | 605 | int nr_updated; |
b24f53a0 | 606 | |
4d942466 | 607 | nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1); |
03c5a6e1 MG |
608 | if (nr_updated) |
609 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 610 | |
4b10e7d5 | 611 | return nr_updated; |
b24f53a0 LS |
612 | } |
613 | #else | |
614 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
615 | unsigned long addr, unsigned long end) | |
616 | { | |
617 | return 0; | |
618 | } | |
5877231f | 619 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 620 | |
6f4576e3 NH |
621 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
622 | struct mm_walk *walk) | |
623 | { | |
624 | struct vm_area_struct *vma = walk->vma; | |
625 | struct queue_pages *qp = walk->private; | |
626 | unsigned long endvma = vma->vm_end; | |
627 | unsigned long flags = qp->flags; | |
628 | ||
77bf45e7 | 629 | if (!vma_migratable(vma)) |
48684a65 NH |
630 | return 1; |
631 | ||
6f4576e3 NH |
632 | if (endvma > end) |
633 | endvma = end; | |
634 | if (vma->vm_start > start) | |
635 | start = vma->vm_start; | |
636 | ||
637 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { | |
638 | if (!vma->vm_next && vma->vm_end < end) | |
639 | return -EFAULT; | |
640 | if (qp->prev && qp->prev->vm_end < vma->vm_start) | |
641 | return -EFAULT; | |
642 | } | |
643 | ||
644 | qp->prev = vma; | |
645 | ||
6f4576e3 NH |
646 | if (flags & MPOL_MF_LAZY) { |
647 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
4355c018 LC |
648 | if (!is_vm_hugetlb_page(vma) && |
649 | (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) && | |
650 | !(vma->vm_flags & VM_MIXEDMAP)) | |
6f4576e3 NH |
651 | change_prot_numa(vma, start, endvma); |
652 | return 1; | |
653 | } | |
654 | ||
77bf45e7 KS |
655 | /* queue pages from current vma */ |
656 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6f4576e3 NH |
657 | return 0; |
658 | return 1; | |
659 | } | |
660 | ||
dc9aa5b9 | 661 | /* |
98094945 NH |
662 | * Walk through page tables and collect pages to be migrated. |
663 | * | |
664 | * If pages found in a given range are on a set of nodes (determined by | |
665 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
666 | * passed via @private.) | |
dc9aa5b9 | 667 | */ |
d05f0cdc | 668 | static int |
98094945 | 669 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
670 | nodemask_t *nodes, unsigned long flags, |
671 | struct list_head *pagelist) | |
1da177e4 | 672 | { |
6f4576e3 NH |
673 | struct queue_pages qp = { |
674 | .pagelist = pagelist, | |
675 | .flags = flags, | |
676 | .nmask = nodes, | |
677 | .prev = NULL, | |
678 | }; | |
679 | struct mm_walk queue_pages_walk = { | |
680 | .hugetlb_entry = queue_pages_hugetlb, | |
681 | .pmd_entry = queue_pages_pte_range, | |
682 | .test_walk = queue_pages_test_walk, | |
683 | .mm = mm, | |
684 | .private = &qp, | |
685 | }; | |
686 | ||
687 | return walk_page_range(start, end, &queue_pages_walk); | |
1da177e4 LT |
688 | } |
689 | ||
869833f2 KM |
690 | /* |
691 | * Apply policy to a single VMA | |
692 | * This must be called with the mmap_sem held for writing. | |
693 | */ | |
694 | static int vma_replace_policy(struct vm_area_struct *vma, | |
695 | struct mempolicy *pol) | |
8d34694c | 696 | { |
869833f2 KM |
697 | int err; |
698 | struct mempolicy *old; | |
699 | struct mempolicy *new; | |
8d34694c KM |
700 | |
701 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
702 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
703 | vma->vm_ops, vma->vm_file, | |
704 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
705 | ||
869833f2 KM |
706 | new = mpol_dup(pol); |
707 | if (IS_ERR(new)) | |
708 | return PTR_ERR(new); | |
709 | ||
710 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 711 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
712 | if (err) |
713 | goto err_out; | |
8d34694c | 714 | } |
869833f2 KM |
715 | |
716 | old = vma->vm_policy; | |
717 | vma->vm_policy = new; /* protected by mmap_sem */ | |
718 | mpol_put(old); | |
719 | ||
720 | return 0; | |
721 | err_out: | |
722 | mpol_put(new); | |
8d34694c KM |
723 | return err; |
724 | } | |
725 | ||
1da177e4 | 726 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
727 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
728 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
729 | { |
730 | struct vm_area_struct *next; | |
9d8cebd4 KM |
731 | struct vm_area_struct *prev; |
732 | struct vm_area_struct *vma; | |
733 | int err = 0; | |
e26a5114 | 734 | pgoff_t pgoff; |
9d8cebd4 KM |
735 | unsigned long vmstart; |
736 | unsigned long vmend; | |
1da177e4 | 737 | |
097d5910 | 738 | vma = find_vma(mm, start); |
9d8cebd4 KM |
739 | if (!vma || vma->vm_start > start) |
740 | return -EFAULT; | |
741 | ||
097d5910 | 742 | prev = vma->vm_prev; |
e26a5114 KM |
743 | if (start > vma->vm_start) |
744 | prev = vma; | |
745 | ||
9d8cebd4 | 746 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 747 | next = vma->vm_next; |
9d8cebd4 KM |
748 | vmstart = max(start, vma->vm_start); |
749 | vmend = min(end, vma->vm_end); | |
750 | ||
e26a5114 KM |
751 | if (mpol_equal(vma_policy(vma), new_pol)) |
752 | continue; | |
753 | ||
754 | pgoff = vma->vm_pgoff + | |
755 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 756 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af AA |
757 | vma->anon_vma, vma->vm_file, pgoff, |
758 | new_pol, vma->vm_userfaultfd_ctx); | |
9d8cebd4 KM |
759 | if (prev) { |
760 | vma = prev; | |
761 | next = vma->vm_next; | |
3964acd0 ON |
762 | if (mpol_equal(vma_policy(vma), new_pol)) |
763 | continue; | |
764 | /* vma_merge() joined vma && vma->next, case 8 */ | |
765 | goto replace; | |
9d8cebd4 KM |
766 | } |
767 | if (vma->vm_start != vmstart) { | |
768 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
769 | if (err) | |
770 | goto out; | |
771 | } | |
772 | if (vma->vm_end != vmend) { | |
773 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
774 | if (err) | |
775 | goto out; | |
776 | } | |
3964acd0 | 777 | replace: |
869833f2 | 778 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
779 | if (err) |
780 | goto out; | |
1da177e4 | 781 | } |
9d8cebd4 KM |
782 | |
783 | out: | |
1da177e4 LT |
784 | return err; |
785 | } | |
786 | ||
1da177e4 | 787 | /* Set the process memory policy */ |
028fec41 DR |
788 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
789 | nodemask_t *nodes) | |
1da177e4 | 790 | { |
58568d2a | 791 | struct mempolicy *new, *old; |
4bfc4495 | 792 | NODEMASK_SCRATCH(scratch); |
58568d2a | 793 | int ret; |
1da177e4 | 794 | |
4bfc4495 KH |
795 | if (!scratch) |
796 | return -ENOMEM; | |
f4e53d91 | 797 | |
4bfc4495 KH |
798 | new = mpol_new(mode, flags, nodes); |
799 | if (IS_ERR(new)) { | |
800 | ret = PTR_ERR(new); | |
801 | goto out; | |
802 | } | |
2c7c3a7d | 803 | |
58568d2a | 804 | task_lock(current); |
4bfc4495 | 805 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
806 | if (ret) { |
807 | task_unlock(current); | |
58568d2a | 808 | mpol_put(new); |
4bfc4495 | 809 | goto out; |
58568d2a MX |
810 | } |
811 | old = current->mempolicy; | |
1da177e4 | 812 | current->mempolicy = new; |
45c4745a | 813 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 814 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 815 | current->il_next = first_node(new->v.nodes); |
58568d2a | 816 | task_unlock(current); |
58568d2a | 817 | mpol_put(old); |
4bfc4495 KH |
818 | ret = 0; |
819 | out: | |
820 | NODEMASK_SCRATCH_FREE(scratch); | |
821 | return ret; | |
1da177e4 LT |
822 | } |
823 | ||
bea904d5 LS |
824 | /* |
825 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
826 | * |
827 | * Called with task's alloc_lock held | |
bea904d5 LS |
828 | */ |
829 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 830 | { |
dfcd3c0d | 831 | nodes_clear(*nodes); |
bea904d5 LS |
832 | if (p == &default_policy) |
833 | return; | |
834 | ||
45c4745a | 835 | switch (p->mode) { |
19770b32 MG |
836 | case MPOL_BIND: |
837 | /* Fall through */ | |
1da177e4 | 838 | case MPOL_INTERLEAVE: |
dfcd3c0d | 839 | *nodes = p->v.nodes; |
1da177e4 LT |
840 | break; |
841 | case MPOL_PREFERRED: | |
fc36b8d3 | 842 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 843 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 844 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
845 | break; |
846 | default: | |
847 | BUG(); | |
848 | } | |
849 | } | |
850 | ||
d4edcf0d | 851 | static int lookup_node(unsigned long addr) |
1da177e4 LT |
852 | { |
853 | struct page *p; | |
854 | int err; | |
855 | ||
768ae309 | 856 | err = get_user_pages(addr & PAGE_MASK, 1, 0, &p, NULL); |
1da177e4 LT |
857 | if (err >= 0) { |
858 | err = page_to_nid(p); | |
859 | put_page(p); | |
860 | } | |
861 | return err; | |
862 | } | |
863 | ||
1da177e4 | 864 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
865 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
866 | unsigned long addr, unsigned long flags) | |
1da177e4 | 867 | { |
8bccd85f | 868 | int err; |
1da177e4 LT |
869 | struct mm_struct *mm = current->mm; |
870 | struct vm_area_struct *vma = NULL; | |
871 | struct mempolicy *pol = current->mempolicy; | |
872 | ||
754af6f5 LS |
873 | if (flags & |
874 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 875 | return -EINVAL; |
754af6f5 LS |
876 | |
877 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
878 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
879 | return -EINVAL; | |
880 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 881 | task_lock(current); |
754af6f5 | 882 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 883 | task_unlock(current); |
754af6f5 LS |
884 | return 0; |
885 | } | |
886 | ||
1da177e4 | 887 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
888 | /* |
889 | * Do NOT fall back to task policy if the | |
890 | * vma/shared policy at addr is NULL. We | |
891 | * want to return MPOL_DEFAULT in this case. | |
892 | */ | |
1da177e4 LT |
893 | down_read(&mm->mmap_sem); |
894 | vma = find_vma_intersection(mm, addr, addr+1); | |
895 | if (!vma) { | |
896 | up_read(&mm->mmap_sem); | |
897 | return -EFAULT; | |
898 | } | |
899 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
900 | pol = vma->vm_ops->get_policy(vma, addr); | |
901 | else | |
902 | pol = vma->vm_policy; | |
903 | } else if (addr) | |
904 | return -EINVAL; | |
905 | ||
906 | if (!pol) | |
bea904d5 | 907 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
908 | |
909 | if (flags & MPOL_F_NODE) { | |
910 | if (flags & MPOL_F_ADDR) { | |
d4edcf0d | 911 | err = lookup_node(addr); |
1da177e4 LT |
912 | if (err < 0) |
913 | goto out; | |
8bccd85f | 914 | *policy = err; |
1da177e4 | 915 | } else if (pol == current->mempolicy && |
45c4745a | 916 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 917 | *policy = current->il_next; |
1da177e4 LT |
918 | } else { |
919 | err = -EINVAL; | |
920 | goto out; | |
921 | } | |
bea904d5 LS |
922 | } else { |
923 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
924 | pol->mode; | |
d79df630 DR |
925 | /* |
926 | * Internal mempolicy flags must be masked off before exposing | |
927 | * the policy to userspace. | |
928 | */ | |
929 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 930 | } |
1da177e4 LT |
931 | |
932 | if (vma) { | |
933 | up_read(¤t->mm->mmap_sem); | |
934 | vma = NULL; | |
935 | } | |
936 | ||
1da177e4 | 937 | err = 0; |
58568d2a | 938 | if (nmask) { |
c6b6ef8b LS |
939 | if (mpol_store_user_nodemask(pol)) { |
940 | *nmask = pol->w.user_nodemask; | |
941 | } else { | |
942 | task_lock(current); | |
943 | get_policy_nodemask(pol, nmask); | |
944 | task_unlock(current); | |
945 | } | |
58568d2a | 946 | } |
1da177e4 LT |
947 | |
948 | out: | |
52cd3b07 | 949 | mpol_cond_put(pol); |
1da177e4 LT |
950 | if (vma) |
951 | up_read(¤t->mm->mmap_sem); | |
952 | return err; | |
953 | } | |
954 | ||
b20a3503 | 955 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
956 | /* |
957 | * page migration | |
958 | */ | |
fc301289 CL |
959 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
960 | unsigned long flags) | |
6ce3c4c0 CL |
961 | { |
962 | /* | |
fc301289 | 963 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 964 | */ |
62695a84 NP |
965 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
966 | if (!isolate_lru_page(page)) { | |
967 | list_add_tail(&page->lru, pagelist); | |
599d0c95 | 968 | inc_node_page_state(page, NR_ISOLATED_ANON + |
6d9c285a | 969 | page_is_file_cache(page)); |
62695a84 NP |
970 | } |
971 | } | |
7e2ab150 | 972 | } |
6ce3c4c0 | 973 | |
742755a1 | 974 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 975 | { |
e2d8cf40 NH |
976 | if (PageHuge(page)) |
977 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
978 | node); | |
979 | else | |
96db800f | 980 | return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE | |
b360edb4 | 981 | __GFP_THISNODE, 0); |
95a402c3 CL |
982 | } |
983 | ||
7e2ab150 CL |
984 | /* |
985 | * Migrate pages from one node to a target node. | |
986 | * Returns error or the number of pages not migrated. | |
987 | */ | |
dbcb0f19 AB |
988 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
989 | int flags) | |
7e2ab150 CL |
990 | { |
991 | nodemask_t nmask; | |
992 | LIST_HEAD(pagelist); | |
993 | int err = 0; | |
994 | ||
995 | nodes_clear(nmask); | |
996 | node_set(source, nmask); | |
6ce3c4c0 | 997 | |
08270807 MK |
998 | /* |
999 | * This does not "check" the range but isolates all pages that | |
1000 | * need migration. Between passing in the full user address | |
1001 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1002 | */ | |
1003 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1004 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1005 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1006 | ||
cf608ac1 | 1007 | if (!list_empty(&pagelist)) { |
68711a74 | 1008 | err = migrate_pages(&pagelist, new_node_page, NULL, dest, |
9c620e2b | 1009 | MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1010 | if (err) |
e2d8cf40 | 1011 | putback_movable_pages(&pagelist); |
cf608ac1 | 1012 | } |
95a402c3 | 1013 | |
7e2ab150 | 1014 | return err; |
6ce3c4c0 CL |
1015 | } |
1016 | ||
39743889 | 1017 | /* |
7e2ab150 CL |
1018 | * Move pages between the two nodesets so as to preserve the physical |
1019 | * layout as much as possible. | |
39743889 CL |
1020 | * |
1021 | * Returns the number of page that could not be moved. | |
1022 | */ | |
0ce72d4f AM |
1023 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1024 | const nodemask_t *to, int flags) | |
39743889 | 1025 | { |
7e2ab150 | 1026 | int busy = 0; |
0aedadf9 | 1027 | int err; |
7e2ab150 | 1028 | nodemask_t tmp; |
39743889 | 1029 | |
0aedadf9 CL |
1030 | err = migrate_prep(); |
1031 | if (err) | |
1032 | return err; | |
1033 | ||
53f2556b | 1034 | down_read(&mm->mmap_sem); |
39743889 | 1035 | |
da0aa138 KM |
1036 | /* |
1037 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1038 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1039 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1040 | * The pair of nodemasks 'to' and 'from' define the map. | |
1041 | * | |
1042 | * If no pair of bits is found that way, fallback to picking some | |
1043 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1044 | * 'source' and 'dest' bits are the same, this represents a node | |
1045 | * that will be migrating to itself, so no pages need move. | |
1046 | * | |
1047 | * If no bits are left in 'tmp', or if all remaining bits left | |
1048 | * in 'tmp' correspond to the same bit in 'to', return false | |
1049 | * (nothing left to migrate). | |
1050 | * | |
1051 | * This lets us pick a pair of nodes to migrate between, such that | |
1052 | * if possible the dest node is not already occupied by some other | |
1053 | * source node, minimizing the risk of overloading the memory on a | |
1054 | * node that would happen if we migrated incoming memory to a node | |
1055 | * before migrating outgoing memory source that same node. | |
1056 | * | |
1057 | * A single scan of tmp is sufficient. As we go, we remember the | |
1058 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1059 | * that not only moved, but what's better, moved to an empty slot | |
1060 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1061 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1062 | * most recent <s, d> pair that moved. If we get all the way through |
1063 | * the scan of tmp without finding any node that moved, much less | |
1064 | * moved to an empty node, then there is nothing left worth migrating. | |
1065 | */ | |
d4984711 | 1066 | |
0ce72d4f | 1067 | tmp = *from; |
7e2ab150 CL |
1068 | while (!nodes_empty(tmp)) { |
1069 | int s,d; | |
b76ac7e7 | 1070 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1071 | int dest = 0; |
1072 | ||
1073 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1074 | |
1075 | /* | |
1076 | * do_migrate_pages() tries to maintain the relative | |
1077 | * node relationship of the pages established between | |
1078 | * threads and memory areas. | |
1079 | * | |
1080 | * However if the number of source nodes is not equal to | |
1081 | * the number of destination nodes we can not preserve | |
1082 | * this node relative relationship. In that case, skip | |
1083 | * copying memory from a node that is in the destination | |
1084 | * mask. | |
1085 | * | |
1086 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1087 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1088 | */ | |
1089 | ||
0ce72d4f AM |
1090 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1091 | (node_isset(s, *to))) | |
4a5b18cc LW |
1092 | continue; |
1093 | ||
0ce72d4f | 1094 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1095 | if (s == d) |
1096 | continue; | |
1097 | ||
1098 | source = s; /* Node moved. Memorize */ | |
1099 | dest = d; | |
1100 | ||
1101 | /* dest not in remaining from nodes? */ | |
1102 | if (!node_isset(dest, tmp)) | |
1103 | break; | |
1104 | } | |
b76ac7e7 | 1105 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1106 | break; |
1107 | ||
1108 | node_clear(source, tmp); | |
1109 | err = migrate_to_node(mm, source, dest, flags); | |
1110 | if (err > 0) | |
1111 | busy += err; | |
1112 | if (err < 0) | |
1113 | break; | |
39743889 CL |
1114 | } |
1115 | up_read(&mm->mmap_sem); | |
7e2ab150 CL |
1116 | if (err < 0) |
1117 | return err; | |
1118 | return busy; | |
b20a3503 CL |
1119 | |
1120 | } | |
1121 | ||
3ad33b24 LS |
1122 | /* |
1123 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1124 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1125 | * Search forward from there, if not. N.B., this assumes that the |
1126 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1127 | * is in virtual address order. | |
1128 | */ | |
d05f0cdc | 1129 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 | 1130 | { |
d05f0cdc | 1131 | struct vm_area_struct *vma; |
3ad33b24 | 1132 | unsigned long uninitialized_var(address); |
95a402c3 | 1133 | |
d05f0cdc | 1134 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1135 | while (vma) { |
1136 | address = page_address_in_vma(page, vma); | |
1137 | if (address != -EFAULT) | |
1138 | break; | |
1139 | vma = vma->vm_next; | |
1140 | } | |
11c731e8 WL |
1141 | |
1142 | if (PageHuge(page)) { | |
cc81717e MH |
1143 | BUG_ON(!vma); |
1144 | return alloc_huge_page_noerr(vma, address, 1); | |
11c731e8 | 1145 | } |
0bf598d8 | 1146 | /* |
11c731e8 | 1147 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1148 | */ |
3ad33b24 | 1149 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
95a402c3 | 1150 | } |
b20a3503 CL |
1151 | #else |
1152 | ||
1153 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1154 | unsigned long flags) | |
1155 | { | |
39743889 CL |
1156 | } |
1157 | ||
0ce72d4f AM |
1158 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1159 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1160 | { |
1161 | return -ENOSYS; | |
1162 | } | |
95a402c3 | 1163 | |
d05f0cdc | 1164 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 CL |
1165 | { |
1166 | return NULL; | |
1167 | } | |
b20a3503 CL |
1168 | #endif |
1169 | ||
dbcb0f19 | 1170 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1171 | unsigned short mode, unsigned short mode_flags, |
1172 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1173 | { |
6ce3c4c0 CL |
1174 | struct mm_struct *mm = current->mm; |
1175 | struct mempolicy *new; | |
1176 | unsigned long end; | |
1177 | int err; | |
1178 | LIST_HEAD(pagelist); | |
1179 | ||
b24f53a0 | 1180 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1181 | return -EINVAL; |
74c00241 | 1182 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1183 | return -EPERM; |
1184 | ||
1185 | if (start & ~PAGE_MASK) | |
1186 | return -EINVAL; | |
1187 | ||
1188 | if (mode == MPOL_DEFAULT) | |
1189 | flags &= ~MPOL_MF_STRICT; | |
1190 | ||
1191 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1192 | end = start + len; | |
1193 | ||
1194 | if (end < start) | |
1195 | return -EINVAL; | |
1196 | if (end == start) | |
1197 | return 0; | |
1198 | ||
028fec41 | 1199 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1200 | if (IS_ERR(new)) |
1201 | return PTR_ERR(new); | |
1202 | ||
b24f53a0 LS |
1203 | if (flags & MPOL_MF_LAZY) |
1204 | new->flags |= MPOL_F_MOF; | |
1205 | ||
6ce3c4c0 CL |
1206 | /* |
1207 | * If we are using the default policy then operation | |
1208 | * on discontinuous address spaces is okay after all | |
1209 | */ | |
1210 | if (!new) | |
1211 | flags |= MPOL_MF_DISCONTIG_OK; | |
1212 | ||
028fec41 DR |
1213 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1214 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1215 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1216 | |
0aedadf9 CL |
1217 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1218 | ||
1219 | err = migrate_prep(); | |
1220 | if (err) | |
b05ca738 | 1221 | goto mpol_out; |
0aedadf9 | 1222 | } |
4bfc4495 KH |
1223 | { |
1224 | NODEMASK_SCRATCH(scratch); | |
1225 | if (scratch) { | |
1226 | down_write(&mm->mmap_sem); | |
1227 | task_lock(current); | |
1228 | err = mpol_set_nodemask(new, nmask, scratch); | |
1229 | task_unlock(current); | |
1230 | if (err) | |
1231 | up_write(&mm->mmap_sem); | |
1232 | } else | |
1233 | err = -ENOMEM; | |
1234 | NODEMASK_SCRATCH_FREE(scratch); | |
1235 | } | |
b05ca738 KM |
1236 | if (err) |
1237 | goto mpol_out; | |
1238 | ||
d05f0cdc | 1239 | err = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1240 | flags | MPOL_MF_INVERT, &pagelist); |
d05f0cdc | 1241 | if (!err) |
9d8cebd4 | 1242 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1243 | |
b24f53a0 LS |
1244 | if (!err) { |
1245 | int nr_failed = 0; | |
1246 | ||
cf608ac1 | 1247 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1248 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1249 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1250 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1251 | if (nr_failed) |
74060e4d | 1252 | putback_movable_pages(&pagelist); |
cf608ac1 | 1253 | } |
6ce3c4c0 | 1254 | |
b24f53a0 | 1255 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1256 | err = -EIO; |
ab8a3e14 | 1257 | } else |
b0e5fd73 | 1258 | putback_movable_pages(&pagelist); |
b20a3503 | 1259 | |
6ce3c4c0 | 1260 | up_write(&mm->mmap_sem); |
b05ca738 | 1261 | mpol_out: |
f0be3d32 | 1262 | mpol_put(new); |
6ce3c4c0 CL |
1263 | return err; |
1264 | } | |
1265 | ||
8bccd85f CL |
1266 | /* |
1267 | * User space interface with variable sized bitmaps for nodelists. | |
1268 | */ | |
1269 | ||
1270 | /* Copy a node mask from user space. */ | |
39743889 | 1271 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1272 | unsigned long maxnode) |
1273 | { | |
1274 | unsigned long k; | |
1275 | unsigned long nlongs; | |
1276 | unsigned long endmask; | |
1277 | ||
1278 | --maxnode; | |
1279 | nodes_clear(*nodes); | |
1280 | if (maxnode == 0 || !nmask) | |
1281 | return 0; | |
a9c930ba | 1282 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1283 | return -EINVAL; |
8bccd85f CL |
1284 | |
1285 | nlongs = BITS_TO_LONGS(maxnode); | |
1286 | if ((maxnode % BITS_PER_LONG) == 0) | |
1287 | endmask = ~0UL; | |
1288 | else | |
1289 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1290 | ||
1291 | /* When the user specified more nodes than supported just check | |
1292 | if the non supported part is all zero. */ | |
1293 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1294 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1295 | return -EINVAL; | |
1296 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1297 | unsigned long t; | |
1298 | if (get_user(t, nmask + k)) | |
1299 | return -EFAULT; | |
1300 | if (k == nlongs - 1) { | |
1301 | if (t & endmask) | |
1302 | return -EINVAL; | |
1303 | } else if (t) | |
1304 | return -EINVAL; | |
1305 | } | |
1306 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1307 | endmask = ~0UL; | |
1308 | } | |
1309 | ||
1310 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1311 | return -EFAULT; | |
1312 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | /* Copy a kernel node mask to user space */ | |
1317 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1318 | nodemask_t *nodes) | |
1319 | { | |
1320 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1321 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1322 | ||
1323 | if (copy > nbytes) { | |
1324 | if (copy > PAGE_SIZE) | |
1325 | return -EINVAL; | |
1326 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1327 | return -EFAULT; | |
1328 | copy = nbytes; | |
1329 | } | |
1330 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1331 | } | |
1332 | ||
938bb9f5 | 1333 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
f7f28ca9 | 1334 | unsigned long, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1335 | unsigned long, maxnode, unsigned, flags) |
8bccd85f CL |
1336 | { |
1337 | nodemask_t nodes; | |
1338 | int err; | |
028fec41 | 1339 | unsigned short mode_flags; |
8bccd85f | 1340 | |
028fec41 DR |
1341 | mode_flags = mode & MPOL_MODE_FLAGS; |
1342 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1343 | if (mode >= MPOL_MAX) |
1344 | return -EINVAL; | |
4c50bc01 DR |
1345 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1346 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1347 | return -EINVAL; | |
8bccd85f CL |
1348 | err = get_nodes(&nodes, nmask, maxnode); |
1349 | if (err) | |
1350 | return err; | |
028fec41 | 1351 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1352 | } |
1353 | ||
1354 | /* Set the process memory policy */ | |
23c8902d | 1355 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1356 | unsigned long, maxnode) |
8bccd85f CL |
1357 | { |
1358 | int err; | |
1359 | nodemask_t nodes; | |
028fec41 | 1360 | unsigned short flags; |
8bccd85f | 1361 | |
028fec41 DR |
1362 | flags = mode & MPOL_MODE_FLAGS; |
1363 | mode &= ~MPOL_MODE_FLAGS; | |
1364 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1365 | return -EINVAL; |
4c50bc01 DR |
1366 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1367 | return -EINVAL; | |
8bccd85f CL |
1368 | err = get_nodes(&nodes, nmask, maxnode); |
1369 | if (err) | |
1370 | return err; | |
028fec41 | 1371 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1372 | } |
1373 | ||
938bb9f5 HC |
1374 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1375 | const unsigned long __user *, old_nodes, | |
1376 | const unsigned long __user *, new_nodes) | |
39743889 | 1377 | { |
c69e8d9c | 1378 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1379 | struct mm_struct *mm = NULL; |
39743889 | 1380 | struct task_struct *task; |
39743889 CL |
1381 | nodemask_t task_nodes; |
1382 | int err; | |
596d7cfa KM |
1383 | nodemask_t *old; |
1384 | nodemask_t *new; | |
1385 | NODEMASK_SCRATCH(scratch); | |
1386 | ||
1387 | if (!scratch) | |
1388 | return -ENOMEM; | |
39743889 | 1389 | |
596d7cfa KM |
1390 | old = &scratch->mask1; |
1391 | new = &scratch->mask2; | |
1392 | ||
1393 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1394 | if (err) |
596d7cfa | 1395 | goto out; |
39743889 | 1396 | |
596d7cfa | 1397 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1398 | if (err) |
596d7cfa | 1399 | goto out; |
39743889 CL |
1400 | |
1401 | /* Find the mm_struct */ | |
55cfaa3c | 1402 | rcu_read_lock(); |
228ebcbe | 1403 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1404 | if (!task) { |
55cfaa3c | 1405 | rcu_read_unlock(); |
596d7cfa KM |
1406 | err = -ESRCH; |
1407 | goto out; | |
39743889 | 1408 | } |
3268c63e | 1409 | get_task_struct(task); |
39743889 | 1410 | |
596d7cfa | 1411 | err = -EINVAL; |
39743889 CL |
1412 | |
1413 | /* | |
1414 | * Check if this process has the right to modify the specified | |
1415 | * process. The right exists if the process has administrative | |
7f927fcc | 1416 | * capabilities, superuser privileges or the same |
39743889 CL |
1417 | * userid as the target process. |
1418 | */ | |
c69e8d9c | 1419 | tcred = __task_cred(task); |
b38a86eb EB |
1420 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1421 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1422 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1423 | rcu_read_unlock(); |
39743889 | 1424 | err = -EPERM; |
3268c63e | 1425 | goto out_put; |
39743889 | 1426 | } |
c69e8d9c | 1427 | rcu_read_unlock(); |
39743889 CL |
1428 | |
1429 | task_nodes = cpuset_mems_allowed(task); | |
1430 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1431 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1432 | err = -EPERM; |
3268c63e | 1433 | goto out_put; |
39743889 CL |
1434 | } |
1435 | ||
01f13bd6 | 1436 | if (!nodes_subset(*new, node_states[N_MEMORY])) { |
3b42d28b | 1437 | err = -EINVAL; |
3268c63e | 1438 | goto out_put; |
3b42d28b CL |
1439 | } |
1440 | ||
86c3a764 DQ |
1441 | err = security_task_movememory(task); |
1442 | if (err) | |
3268c63e | 1443 | goto out_put; |
86c3a764 | 1444 | |
3268c63e CL |
1445 | mm = get_task_mm(task); |
1446 | put_task_struct(task); | |
f2a9ef88 SL |
1447 | |
1448 | if (!mm) { | |
3268c63e | 1449 | err = -EINVAL; |
f2a9ef88 SL |
1450 | goto out; |
1451 | } | |
1452 | ||
1453 | err = do_migrate_pages(mm, old, new, | |
1454 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1455 | |
1456 | mmput(mm); | |
1457 | out: | |
596d7cfa KM |
1458 | NODEMASK_SCRATCH_FREE(scratch); |
1459 | ||
39743889 | 1460 | return err; |
3268c63e CL |
1461 | |
1462 | out_put: | |
1463 | put_task_struct(task); | |
1464 | goto out; | |
1465 | ||
39743889 CL |
1466 | } |
1467 | ||
1468 | ||
8bccd85f | 1469 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1470 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1471 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1472 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1473 | { |
dbcb0f19 AB |
1474 | int err; |
1475 | int uninitialized_var(pval); | |
8bccd85f CL |
1476 | nodemask_t nodes; |
1477 | ||
1478 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1479 | return -EINVAL; | |
1480 | ||
1481 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1482 | ||
1483 | if (err) | |
1484 | return err; | |
1485 | ||
1486 | if (policy && put_user(pval, policy)) | |
1487 | return -EFAULT; | |
1488 | ||
1489 | if (nmask) | |
1490 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1491 | ||
1492 | return err; | |
1493 | } | |
1494 | ||
1da177e4 LT |
1495 | #ifdef CONFIG_COMPAT |
1496 | ||
c93e0f6c HC |
1497 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1498 | compat_ulong_t __user *, nmask, | |
1499 | compat_ulong_t, maxnode, | |
1500 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1501 | { |
1502 | long err; | |
1503 | unsigned long __user *nm = NULL; | |
1504 | unsigned long nr_bits, alloc_size; | |
1505 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1506 | ||
1507 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1508 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1509 | ||
1510 | if (nmask) | |
1511 | nm = compat_alloc_user_space(alloc_size); | |
1512 | ||
1513 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1514 | ||
1515 | if (!err && nmask) { | |
2bbff6c7 KH |
1516 | unsigned long copy_size; |
1517 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1518 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1519 | /* ensure entire bitmap is zeroed */ |
1520 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1521 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1522 | } | |
1523 | ||
1524 | return err; | |
1525 | } | |
1526 | ||
c93e0f6c HC |
1527 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1528 | compat_ulong_t, maxnode) | |
1da177e4 LT |
1529 | { |
1530 | long err = 0; | |
1531 | unsigned long __user *nm = NULL; | |
1532 | unsigned long nr_bits, alloc_size; | |
1533 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1534 | ||
1535 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1536 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1537 | ||
1538 | if (nmask) { | |
1539 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1540 | nm = compat_alloc_user_space(alloc_size); | |
1541 | err |= copy_to_user(nm, bm, alloc_size); | |
1542 | } | |
1543 | ||
1544 | if (err) | |
1545 | return -EFAULT; | |
1546 | ||
1547 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1548 | } | |
1549 | ||
c93e0f6c HC |
1550 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1551 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1552 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 LT |
1553 | { |
1554 | long err = 0; | |
1555 | unsigned long __user *nm = NULL; | |
1556 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1557 | nodemask_t bm; |
1da177e4 LT |
1558 | |
1559 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1560 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1561 | ||
1562 | if (nmask) { | |
dfcd3c0d | 1563 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1564 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1565 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1566 | } |
1567 | ||
1568 | if (err) | |
1569 | return -EFAULT; | |
1570 | ||
1571 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1572 | } | |
1573 | ||
1574 | #endif | |
1575 | ||
74d2c3a0 ON |
1576 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1577 | unsigned long addr) | |
1da177e4 | 1578 | { |
8d90274b | 1579 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1580 | |
1581 | if (vma) { | |
480eccf9 | 1582 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1583 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1584 | } else if (vma->vm_policy) { |
1da177e4 | 1585 | pol = vma->vm_policy; |
00442ad0 MG |
1586 | |
1587 | /* | |
1588 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1589 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1590 | * count on these policies which will be dropped by | |
1591 | * mpol_cond_put() later | |
1592 | */ | |
1593 | if (mpol_needs_cond_ref(pol)) | |
1594 | mpol_get(pol); | |
1595 | } | |
1da177e4 | 1596 | } |
f15ca78e | 1597 | |
74d2c3a0 ON |
1598 | return pol; |
1599 | } | |
1600 | ||
1601 | /* | |
dd6eecb9 | 1602 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1603 | * @vma: virtual memory area whose policy is sought |
1604 | * @addr: address in @vma for shared policy lookup | |
1605 | * | |
1606 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1607 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1608 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1609 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1610 | * freeing by another task. It is the caller's responsibility to free the | |
1611 | * extra reference for shared policies. | |
1612 | */ | |
dd6eecb9 ON |
1613 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
1614 | unsigned long addr) | |
74d2c3a0 ON |
1615 | { |
1616 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1617 | ||
8d90274b | 1618 | if (!pol) |
dd6eecb9 | 1619 | pol = get_task_policy(current); |
8d90274b | 1620 | |
1da177e4 LT |
1621 | return pol; |
1622 | } | |
1623 | ||
6b6482bb | 1624 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1625 | { |
6b6482bb | 1626 | struct mempolicy *pol; |
fc314724 | 1627 | |
6b6482bb ON |
1628 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1629 | bool ret = false; | |
fc314724 | 1630 | |
6b6482bb ON |
1631 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1632 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1633 | ret = true; | |
1634 | mpol_cond_put(pol); | |
8d90274b | 1635 | |
6b6482bb | 1636 | return ret; |
fc314724 MG |
1637 | } |
1638 | ||
6b6482bb | 1639 | pol = vma->vm_policy; |
8d90274b | 1640 | if (!pol) |
6b6482bb | 1641 | pol = get_task_policy(current); |
8d90274b | 1642 | |
fc314724 MG |
1643 | return pol->flags & MPOL_F_MOF; |
1644 | } | |
1645 | ||
d3eb1570 LJ |
1646 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1647 | { | |
1648 | enum zone_type dynamic_policy_zone = policy_zone; | |
1649 | ||
1650 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1651 | ||
1652 | /* | |
1653 | * if policy->v.nodes has movable memory only, | |
1654 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1655 | * | |
1656 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1657 | * so if the following test faile, it implies | |
1658 | * policy->v.nodes has movable memory only. | |
1659 | */ | |
1660 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1661 | dynamic_policy_zone = ZONE_MOVABLE; | |
1662 | ||
1663 | return zone >= dynamic_policy_zone; | |
1664 | } | |
1665 | ||
52cd3b07 LS |
1666 | /* |
1667 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1668 | * page allocation | |
1669 | */ | |
1670 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1671 | { |
1672 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1673 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1674 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1675 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1676 | return &policy->v.nodes; | |
1677 | ||
1678 | return NULL; | |
1679 | } | |
1680 | ||
52cd3b07 | 1681 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1682 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1683 | int nd) | |
1da177e4 | 1684 | { |
6d840958 MH |
1685 | if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL)) |
1686 | nd = policy->v.preferred_node; | |
1687 | else { | |
19770b32 | 1688 | /* |
6d840958 MH |
1689 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1690 | * because we might easily break the expectation to stay on the | |
1691 | * requested node and not break the policy. | |
19770b32 | 1692 | */ |
6d840958 | 1693 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1694 | } |
6d840958 | 1695 | |
0e88460d | 1696 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1697 | } |
1698 | ||
1699 | /* Do dynamic interleaving for a process */ | |
1700 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1701 | { | |
1702 | unsigned nid, next; | |
1703 | struct task_struct *me = current; | |
1704 | ||
1705 | nid = me->il_next; | |
0edaf86c | 1706 | next = next_node_in(nid, policy->v.nodes); |
f5b087b5 DR |
1707 | if (next < MAX_NUMNODES) |
1708 | me->il_next = next; | |
1da177e4 LT |
1709 | return nid; |
1710 | } | |
1711 | ||
dc85da15 CL |
1712 | /* |
1713 | * Depending on the memory policy provide a node from which to allocate the | |
1714 | * next slab entry. | |
1715 | */ | |
2a389610 | 1716 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1717 | { |
e7b691b0 | 1718 | struct mempolicy *policy; |
2a389610 | 1719 | int node = numa_mem_id(); |
e7b691b0 AK |
1720 | |
1721 | if (in_interrupt()) | |
2a389610 | 1722 | return node; |
e7b691b0 AK |
1723 | |
1724 | policy = current->mempolicy; | |
fc36b8d3 | 1725 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1726 | return node; |
bea904d5 LS |
1727 | |
1728 | switch (policy->mode) { | |
1729 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1730 | /* |
1731 | * handled MPOL_F_LOCAL above | |
1732 | */ | |
1733 | return policy->v.preferred_node; | |
765c4507 | 1734 | |
dc85da15 CL |
1735 | case MPOL_INTERLEAVE: |
1736 | return interleave_nodes(policy); | |
1737 | ||
dd1a239f | 1738 | case MPOL_BIND: { |
c33d6c06 MG |
1739 | struct zoneref *z; |
1740 | ||
dc85da15 CL |
1741 | /* |
1742 | * Follow bind policy behavior and start allocation at the | |
1743 | * first node. | |
1744 | */ | |
19770b32 | 1745 | struct zonelist *zonelist; |
19770b32 | 1746 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1747 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 MG |
1748 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
1749 | &policy->v.nodes); | |
1750 | return z->zone ? z->zone->node : node; | |
dd1a239f | 1751 | } |
dc85da15 | 1752 | |
dc85da15 | 1753 | default: |
bea904d5 | 1754 | BUG(); |
dc85da15 CL |
1755 | } |
1756 | } | |
1757 | ||
fee83b3a AM |
1758 | /* |
1759 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
1760 | * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the | |
1761 | * number of present nodes. | |
1762 | */ | |
1da177e4 | 1763 | static unsigned offset_il_node(struct mempolicy *pol, |
fee83b3a | 1764 | struct vm_area_struct *vma, unsigned long n) |
1da177e4 | 1765 | { |
dfcd3c0d | 1766 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1767 | unsigned target; |
fee83b3a AM |
1768 | int i; |
1769 | int nid; | |
1da177e4 | 1770 | |
f5b087b5 DR |
1771 | if (!nnodes) |
1772 | return numa_node_id(); | |
fee83b3a AM |
1773 | target = (unsigned int)n % nnodes; |
1774 | nid = first_node(pol->v.nodes); | |
1775 | for (i = 0; i < target; i++) | |
dfcd3c0d | 1776 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1777 | return nid; |
1778 | } | |
1779 | ||
5da7ca86 CL |
1780 | /* Determine a node number for interleave */ |
1781 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1782 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1783 | { | |
1784 | if (vma) { | |
1785 | unsigned long off; | |
1786 | ||
3b98b087 NA |
1787 | /* |
1788 | * for small pages, there is no difference between | |
1789 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1790 | * for huge pages, since vm_pgoff is in units of small | |
1791 | * pages, we need to shift off the always 0 bits to get | |
1792 | * a useful offset. | |
1793 | */ | |
1794 | BUG_ON(shift < PAGE_SHIFT); | |
1795 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1796 | off += (addr - vma->vm_start) >> shift; |
1797 | return offset_il_node(pol, vma, off); | |
1798 | } else | |
1799 | return interleave_nodes(pol); | |
1800 | } | |
1801 | ||
00ac59ad | 1802 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1803 | /* |
1804 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
b46e14ac FF |
1805 | * @vma: virtual memory area whose policy is sought |
1806 | * @addr: address in @vma for shared policy lookup and interleave policy | |
1807 | * @gfp_flags: for requested zone | |
1808 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
1809 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1810 | * |
52cd3b07 LS |
1811 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1812 | * to the struct mempolicy for conditional unref after allocation. | |
1813 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1814 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 1815 | * |
d26914d1 | 1816 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 1817 | */ |
396faf03 | 1818 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1819 | gfp_t gfp_flags, struct mempolicy **mpol, |
1820 | nodemask_t **nodemask) | |
5da7ca86 | 1821 | { |
480eccf9 | 1822 | struct zonelist *zl; |
5da7ca86 | 1823 | |
dd6eecb9 | 1824 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 1825 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1826 | |
52cd3b07 LS |
1827 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1828 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1829 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1830 | } else { |
2f5f9486 | 1831 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1832 | if ((*mpol)->mode == MPOL_BIND) |
1833 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1834 | } |
1835 | return zl; | |
5da7ca86 | 1836 | } |
06808b08 LS |
1837 | |
1838 | /* | |
1839 | * init_nodemask_of_mempolicy | |
1840 | * | |
1841 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1842 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1843 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1844 | * initialize the argument nodemask to contain the single node for | |
1845 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1846 | * of non-default mempolicy. | |
1847 | * | |
1848 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1849 | * because the current task is examining it's own mempolicy and a task's | |
1850 | * mempolicy is only ever changed by the task itself. | |
1851 | * | |
1852 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1853 | */ | |
1854 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1855 | { | |
1856 | struct mempolicy *mempolicy; | |
1857 | int nid; | |
1858 | ||
1859 | if (!(mask && current->mempolicy)) | |
1860 | return false; | |
1861 | ||
c0ff7453 | 1862 | task_lock(current); |
06808b08 LS |
1863 | mempolicy = current->mempolicy; |
1864 | switch (mempolicy->mode) { | |
1865 | case MPOL_PREFERRED: | |
1866 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1867 | nid = numa_node_id(); | |
1868 | else | |
1869 | nid = mempolicy->v.preferred_node; | |
1870 | init_nodemask_of_node(mask, nid); | |
1871 | break; | |
1872 | ||
1873 | case MPOL_BIND: | |
1874 | /* Fall through */ | |
1875 | case MPOL_INTERLEAVE: | |
1876 | *mask = mempolicy->v.nodes; | |
1877 | break; | |
1878 | ||
1879 | default: | |
1880 | BUG(); | |
1881 | } | |
c0ff7453 | 1882 | task_unlock(current); |
06808b08 LS |
1883 | |
1884 | return true; | |
1885 | } | |
00ac59ad | 1886 | #endif |
5da7ca86 | 1887 | |
6f48d0eb DR |
1888 | /* |
1889 | * mempolicy_nodemask_intersects | |
1890 | * | |
1891 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1892 | * policy. Otherwise, check for intersection between mask and the policy | |
1893 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1894 | * policy, always return true since it may allocate elsewhere on fallback. | |
1895 | * | |
1896 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1897 | */ | |
1898 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1899 | const nodemask_t *mask) | |
1900 | { | |
1901 | struct mempolicy *mempolicy; | |
1902 | bool ret = true; | |
1903 | ||
1904 | if (!mask) | |
1905 | return ret; | |
1906 | task_lock(tsk); | |
1907 | mempolicy = tsk->mempolicy; | |
1908 | if (!mempolicy) | |
1909 | goto out; | |
1910 | ||
1911 | switch (mempolicy->mode) { | |
1912 | case MPOL_PREFERRED: | |
1913 | /* | |
1914 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1915 | * allocate from, they may fallback to other nodes when oom. | |
1916 | * Thus, it's possible for tsk to have allocated memory from | |
1917 | * nodes in mask. | |
1918 | */ | |
1919 | break; | |
1920 | case MPOL_BIND: | |
1921 | case MPOL_INTERLEAVE: | |
1922 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1923 | break; | |
1924 | default: | |
1925 | BUG(); | |
1926 | } | |
1927 | out: | |
1928 | task_unlock(tsk); | |
1929 | return ret; | |
1930 | } | |
1931 | ||
1da177e4 LT |
1932 | /* Allocate a page in interleaved policy. |
1933 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1934 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1935 | unsigned nid) | |
1da177e4 LT |
1936 | { |
1937 | struct zonelist *zl; | |
1938 | struct page *page; | |
1939 | ||
0e88460d | 1940 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1941 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1942 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1943 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1944 | return page; |
1945 | } | |
1946 | ||
1947 | /** | |
0bbbc0b3 | 1948 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1949 | * |
1950 | * @gfp: | |
1951 | * %GFP_USER user allocation. | |
1952 | * %GFP_KERNEL kernel allocations, | |
1953 | * %GFP_HIGHMEM highmem/user allocations, | |
1954 | * %GFP_FS allocation should not call back into a file system. | |
1955 | * %GFP_ATOMIC don't sleep. | |
1956 | * | |
0bbbc0b3 | 1957 | * @order:Order of the GFP allocation. |
1da177e4 LT |
1958 | * @vma: Pointer to VMA or NULL if not available. |
1959 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
be97a41b VB |
1960 | * @node: Which node to prefer for allocation (modulo policy). |
1961 | * @hugepage: for hugepages try only the preferred node if possible | |
1da177e4 LT |
1962 | * |
1963 | * This function allocates a page from the kernel page pool and applies | |
1964 | * a NUMA policy associated with the VMA or the current process. | |
1965 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1966 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
be97a41b VB |
1967 | * all allocations for pages that will be mapped into user space. Returns |
1968 | * NULL when no page can be allocated. | |
1da177e4 LT |
1969 | */ |
1970 | struct page * | |
0bbbc0b3 | 1971 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
be97a41b | 1972 | unsigned long addr, int node, bool hugepage) |
1da177e4 | 1973 | { |
cc9a6c87 | 1974 | struct mempolicy *pol; |
c0ff7453 | 1975 | struct page *page; |
cc9a6c87 | 1976 | unsigned int cpuset_mems_cookie; |
be97a41b VB |
1977 | struct zonelist *zl; |
1978 | nodemask_t *nmask; | |
cc9a6c87 MG |
1979 | |
1980 | retry_cpuset: | |
dd6eecb9 | 1981 | pol = get_vma_policy(vma, addr); |
d26914d1 | 1982 | cpuset_mems_cookie = read_mems_allowed_begin(); |
1da177e4 | 1983 | |
0867a57c VB |
1984 | if (pol->mode == MPOL_INTERLEAVE) { |
1985 | unsigned nid; | |
1986 | ||
1987 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
1988 | mpol_cond_put(pol); | |
1989 | page = alloc_page_interleave(gfp, order, nid); | |
1990 | goto out; | |
1991 | } | |
1992 | ||
1993 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { | |
1994 | int hpage_node = node; | |
1995 | ||
be97a41b VB |
1996 | /* |
1997 | * For hugepage allocation and non-interleave policy which | |
0867a57c VB |
1998 | * allows the current node (or other explicitly preferred |
1999 | * node) we only try to allocate from the current/preferred | |
2000 | * node and don't fall back to other nodes, as the cost of | |
2001 | * remote accesses would likely offset THP benefits. | |
be97a41b VB |
2002 | * |
2003 | * If the policy is interleave, or does not allow the current | |
2004 | * node in its nodemask, we allocate the standard way. | |
2005 | */ | |
0867a57c VB |
2006 | if (pol->mode == MPOL_PREFERRED && |
2007 | !(pol->flags & MPOL_F_LOCAL)) | |
2008 | hpage_node = pol->v.preferred_node; | |
2009 | ||
be97a41b | 2010 | nmask = policy_nodemask(gfp, pol); |
0867a57c | 2011 | if (!nmask || node_isset(hpage_node, *nmask)) { |
be97a41b | 2012 | mpol_cond_put(pol); |
96db800f | 2013 | page = __alloc_pages_node(hpage_node, |
5265047a | 2014 | gfp | __GFP_THISNODE, order); |
be97a41b VB |
2015 | goto out; |
2016 | } | |
2017 | } | |
2018 | ||
be97a41b VB |
2019 | nmask = policy_nodemask(gfp, pol); |
2020 | zl = policy_zonelist(gfp, pol, node); | |
be97a41b | 2021 | page = __alloc_pages_nodemask(gfp, order, zl, nmask); |
d51e9894 | 2022 | mpol_cond_put(pol); |
be97a41b | 2023 | out: |
d26914d1 | 2024 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 2025 | goto retry_cpuset; |
c0ff7453 | 2026 | return page; |
1da177e4 LT |
2027 | } |
2028 | ||
2029 | /** | |
2030 | * alloc_pages_current - Allocate pages. | |
2031 | * | |
2032 | * @gfp: | |
2033 | * %GFP_USER user allocation, | |
2034 | * %GFP_KERNEL kernel allocation, | |
2035 | * %GFP_HIGHMEM highmem allocation, | |
2036 | * %GFP_FS don't call back into a file system. | |
2037 | * %GFP_ATOMIC don't sleep. | |
2038 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2039 | * | |
2040 | * Allocate a page from the kernel page pool. When not in | |
2041 | * interrupt context and apply the current process NUMA policy. | |
2042 | * Returns NULL when no page can be allocated. | |
2043 | * | |
cf2a473c | 2044 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
2045 | * 1) it's ok to take cpuset_sem (can WAIT), and |
2046 | * 2) allocating for current task (not interrupt). | |
2047 | */ | |
dd0fc66f | 2048 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2049 | { |
8d90274b | 2050 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2051 | struct page *page; |
cc9a6c87 | 2052 | unsigned int cpuset_mems_cookie; |
1da177e4 | 2053 | |
8d90274b ON |
2054 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2055 | pol = get_task_policy(current); | |
52cd3b07 | 2056 | |
cc9a6c87 | 2057 | retry_cpuset: |
d26914d1 | 2058 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2059 | |
52cd3b07 LS |
2060 | /* |
2061 | * No reference counting needed for current->mempolicy | |
2062 | * nor system default_policy | |
2063 | */ | |
45c4745a | 2064 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2065 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2066 | else | |
2067 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
2068 | policy_zonelist(gfp, pol, numa_node_id()), |
2069 | policy_nodemask(gfp, pol)); | |
cc9a6c87 | 2070 | |
d26914d1 | 2071 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2072 | goto retry_cpuset; |
2073 | ||
c0ff7453 | 2074 | return page; |
1da177e4 LT |
2075 | } |
2076 | EXPORT_SYMBOL(alloc_pages_current); | |
2077 | ||
ef0855d3 ON |
2078 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2079 | { | |
2080 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2081 | ||
2082 | if (IS_ERR(pol)) | |
2083 | return PTR_ERR(pol); | |
2084 | dst->vm_policy = pol; | |
2085 | return 0; | |
2086 | } | |
2087 | ||
4225399a | 2088 | /* |
846a16bf | 2089 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2090 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2091 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2092 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2093 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2094 | * |
2095 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2096 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2097 | */ |
4225399a | 2098 | |
846a16bf LS |
2099 | /* Slow path of a mempolicy duplicate */ |
2100 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2101 | { |
2102 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2103 | ||
2104 | if (!new) | |
2105 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2106 | |
2107 | /* task's mempolicy is protected by alloc_lock */ | |
2108 | if (old == current->mempolicy) { | |
2109 | task_lock(current); | |
2110 | *new = *old; | |
2111 | task_unlock(current); | |
2112 | } else | |
2113 | *new = *old; | |
2114 | ||
4225399a PJ |
2115 | if (current_cpuset_is_being_rebound()) { |
2116 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2117 | if (new->flags & MPOL_F_REBINDING) |
2118 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2119 | else | |
2120 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2121 | } |
1da177e4 | 2122 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2123 | return new; |
2124 | } | |
2125 | ||
2126 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2127 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2128 | { |
2129 | if (!a || !b) | |
fcfb4dcc | 2130 | return false; |
45c4745a | 2131 | if (a->mode != b->mode) |
fcfb4dcc | 2132 | return false; |
19800502 | 2133 | if (a->flags != b->flags) |
fcfb4dcc | 2134 | return false; |
19800502 BL |
2135 | if (mpol_store_user_nodemask(a)) |
2136 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2137 | return false; |
19800502 | 2138 | |
45c4745a | 2139 | switch (a->mode) { |
19770b32 MG |
2140 | case MPOL_BIND: |
2141 | /* Fall through */ | |
1da177e4 | 2142 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2143 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2144 | case MPOL_PREFERRED: |
75719661 | 2145 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2146 | default: |
2147 | BUG(); | |
fcfb4dcc | 2148 | return false; |
1da177e4 LT |
2149 | } |
2150 | } | |
2151 | ||
1da177e4 LT |
2152 | /* |
2153 | * Shared memory backing store policy support. | |
2154 | * | |
2155 | * Remember policies even when nobody has shared memory mapped. | |
2156 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2157 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2158 | * for any accesses to the tree. |
2159 | */ | |
2160 | ||
4a8c7bb5 NZ |
2161 | /* |
2162 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2163 | * reading or for writing | |
2164 | */ | |
1da177e4 LT |
2165 | static struct sp_node * |
2166 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2167 | { | |
2168 | struct rb_node *n = sp->root.rb_node; | |
2169 | ||
2170 | while (n) { | |
2171 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2172 | ||
2173 | if (start >= p->end) | |
2174 | n = n->rb_right; | |
2175 | else if (end <= p->start) | |
2176 | n = n->rb_left; | |
2177 | else | |
2178 | break; | |
2179 | } | |
2180 | if (!n) | |
2181 | return NULL; | |
2182 | for (;;) { | |
2183 | struct sp_node *w = NULL; | |
2184 | struct rb_node *prev = rb_prev(n); | |
2185 | if (!prev) | |
2186 | break; | |
2187 | w = rb_entry(prev, struct sp_node, nd); | |
2188 | if (w->end <= start) | |
2189 | break; | |
2190 | n = prev; | |
2191 | } | |
2192 | return rb_entry(n, struct sp_node, nd); | |
2193 | } | |
2194 | ||
4a8c7bb5 NZ |
2195 | /* |
2196 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2197 | * writing. | |
2198 | */ | |
1da177e4 LT |
2199 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2200 | { | |
2201 | struct rb_node **p = &sp->root.rb_node; | |
2202 | struct rb_node *parent = NULL; | |
2203 | struct sp_node *nd; | |
2204 | ||
2205 | while (*p) { | |
2206 | parent = *p; | |
2207 | nd = rb_entry(parent, struct sp_node, nd); | |
2208 | if (new->start < nd->start) | |
2209 | p = &(*p)->rb_left; | |
2210 | else if (new->end > nd->end) | |
2211 | p = &(*p)->rb_right; | |
2212 | else | |
2213 | BUG(); | |
2214 | } | |
2215 | rb_link_node(&new->nd, parent, p); | |
2216 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2217 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2218 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2219 | } |
2220 | ||
2221 | /* Find shared policy intersecting idx */ | |
2222 | struct mempolicy * | |
2223 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2224 | { | |
2225 | struct mempolicy *pol = NULL; | |
2226 | struct sp_node *sn; | |
2227 | ||
2228 | if (!sp->root.rb_node) | |
2229 | return NULL; | |
4a8c7bb5 | 2230 | read_lock(&sp->lock); |
1da177e4 LT |
2231 | sn = sp_lookup(sp, idx, idx+1); |
2232 | if (sn) { | |
2233 | mpol_get(sn->policy); | |
2234 | pol = sn->policy; | |
2235 | } | |
4a8c7bb5 | 2236 | read_unlock(&sp->lock); |
1da177e4 LT |
2237 | return pol; |
2238 | } | |
2239 | ||
63f74ca2 KM |
2240 | static void sp_free(struct sp_node *n) |
2241 | { | |
2242 | mpol_put(n->policy); | |
2243 | kmem_cache_free(sn_cache, n); | |
2244 | } | |
2245 | ||
771fb4d8 LS |
2246 | /** |
2247 | * mpol_misplaced - check whether current page node is valid in policy | |
2248 | * | |
b46e14ac FF |
2249 | * @page: page to be checked |
2250 | * @vma: vm area where page mapped | |
2251 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2252 | * |
2253 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2254 | * node id. | |
2255 | * | |
2256 | * Returns: | |
2257 | * -1 - not misplaced, page is in the right node | |
2258 | * node - node id where the page should be | |
2259 | * | |
2260 | * Policy determination "mimics" alloc_page_vma(). | |
2261 | * Called from fault path where we know the vma and faulting address. | |
2262 | */ | |
2263 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2264 | { | |
2265 | struct mempolicy *pol; | |
c33d6c06 | 2266 | struct zoneref *z; |
771fb4d8 LS |
2267 | int curnid = page_to_nid(page); |
2268 | unsigned long pgoff; | |
90572890 PZ |
2269 | int thiscpu = raw_smp_processor_id(); |
2270 | int thisnid = cpu_to_node(thiscpu); | |
771fb4d8 LS |
2271 | int polnid = -1; |
2272 | int ret = -1; | |
2273 | ||
2274 | BUG_ON(!vma); | |
2275 | ||
dd6eecb9 | 2276 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2277 | if (!(pol->flags & MPOL_F_MOF)) |
2278 | goto out; | |
2279 | ||
2280 | switch (pol->mode) { | |
2281 | case MPOL_INTERLEAVE: | |
2282 | BUG_ON(addr >= vma->vm_end); | |
2283 | BUG_ON(addr < vma->vm_start); | |
2284 | ||
2285 | pgoff = vma->vm_pgoff; | |
2286 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2287 | polnid = offset_il_node(pol, vma, pgoff); | |
2288 | break; | |
2289 | ||
2290 | case MPOL_PREFERRED: | |
2291 | if (pol->flags & MPOL_F_LOCAL) | |
2292 | polnid = numa_node_id(); | |
2293 | else | |
2294 | polnid = pol->v.preferred_node; | |
2295 | break; | |
2296 | ||
2297 | case MPOL_BIND: | |
c33d6c06 | 2298 | |
771fb4d8 LS |
2299 | /* |
2300 | * allows binding to multiple nodes. | |
2301 | * use current page if in policy nodemask, | |
2302 | * else select nearest allowed node, if any. | |
2303 | * If no allowed nodes, use current [!misplaced]. | |
2304 | */ | |
2305 | if (node_isset(curnid, pol->v.nodes)) | |
2306 | goto out; | |
c33d6c06 | 2307 | z = first_zones_zonelist( |
771fb4d8 LS |
2308 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2309 | gfp_zone(GFP_HIGHUSER), | |
c33d6c06 MG |
2310 | &pol->v.nodes); |
2311 | polnid = z->zone->node; | |
771fb4d8 LS |
2312 | break; |
2313 | ||
2314 | default: | |
2315 | BUG(); | |
2316 | } | |
5606e387 MG |
2317 | |
2318 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2319 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2320 | polnid = thisnid; |
5606e387 | 2321 | |
10f39042 | 2322 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2323 | goto out; |
e42c8ff2 MG |
2324 | } |
2325 | ||
771fb4d8 LS |
2326 | if (curnid != polnid) |
2327 | ret = polnid; | |
2328 | out: | |
2329 | mpol_cond_put(pol); | |
2330 | ||
2331 | return ret; | |
2332 | } | |
2333 | ||
c11600e4 DR |
2334 | /* |
2335 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2336 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2337 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2338 | * policy. | |
2339 | */ | |
2340 | void mpol_put_task_policy(struct task_struct *task) | |
2341 | { | |
2342 | struct mempolicy *pol; | |
2343 | ||
2344 | task_lock(task); | |
2345 | pol = task->mempolicy; | |
2346 | task->mempolicy = NULL; | |
2347 | task_unlock(task); | |
2348 | mpol_put(pol); | |
2349 | } | |
2350 | ||
1da177e4 LT |
2351 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2352 | { | |
140d5a49 | 2353 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2354 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2355 | sp_free(n); |
1da177e4 LT |
2356 | } |
2357 | ||
42288fe3 MG |
2358 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2359 | unsigned long end, struct mempolicy *pol) | |
2360 | { | |
2361 | node->start = start; | |
2362 | node->end = end; | |
2363 | node->policy = pol; | |
2364 | } | |
2365 | ||
dbcb0f19 AB |
2366 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2367 | struct mempolicy *pol) | |
1da177e4 | 2368 | { |
869833f2 KM |
2369 | struct sp_node *n; |
2370 | struct mempolicy *newpol; | |
1da177e4 | 2371 | |
869833f2 | 2372 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2373 | if (!n) |
2374 | return NULL; | |
869833f2 KM |
2375 | |
2376 | newpol = mpol_dup(pol); | |
2377 | if (IS_ERR(newpol)) { | |
2378 | kmem_cache_free(sn_cache, n); | |
2379 | return NULL; | |
2380 | } | |
2381 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2382 | sp_node_init(n, start, end, newpol); |
869833f2 | 2383 | |
1da177e4 LT |
2384 | return n; |
2385 | } | |
2386 | ||
2387 | /* Replace a policy range. */ | |
2388 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2389 | unsigned long end, struct sp_node *new) | |
2390 | { | |
b22d127a | 2391 | struct sp_node *n; |
42288fe3 MG |
2392 | struct sp_node *n_new = NULL; |
2393 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2394 | int ret = 0; |
1da177e4 | 2395 | |
42288fe3 | 2396 | restart: |
4a8c7bb5 | 2397 | write_lock(&sp->lock); |
1da177e4 LT |
2398 | n = sp_lookup(sp, start, end); |
2399 | /* Take care of old policies in the same range. */ | |
2400 | while (n && n->start < end) { | |
2401 | struct rb_node *next = rb_next(&n->nd); | |
2402 | if (n->start >= start) { | |
2403 | if (n->end <= end) | |
2404 | sp_delete(sp, n); | |
2405 | else | |
2406 | n->start = end; | |
2407 | } else { | |
2408 | /* Old policy spanning whole new range. */ | |
2409 | if (n->end > end) { | |
42288fe3 MG |
2410 | if (!n_new) |
2411 | goto alloc_new; | |
2412 | ||
2413 | *mpol_new = *n->policy; | |
2414 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2415 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2416 | n->end = start; |
5ca39575 | 2417 | sp_insert(sp, n_new); |
42288fe3 MG |
2418 | n_new = NULL; |
2419 | mpol_new = NULL; | |
1da177e4 LT |
2420 | break; |
2421 | } else | |
2422 | n->end = start; | |
2423 | } | |
2424 | if (!next) | |
2425 | break; | |
2426 | n = rb_entry(next, struct sp_node, nd); | |
2427 | } | |
2428 | if (new) | |
2429 | sp_insert(sp, new); | |
4a8c7bb5 | 2430 | write_unlock(&sp->lock); |
42288fe3 MG |
2431 | ret = 0; |
2432 | ||
2433 | err_out: | |
2434 | if (mpol_new) | |
2435 | mpol_put(mpol_new); | |
2436 | if (n_new) | |
2437 | kmem_cache_free(sn_cache, n_new); | |
2438 | ||
b22d127a | 2439 | return ret; |
42288fe3 MG |
2440 | |
2441 | alloc_new: | |
4a8c7bb5 | 2442 | write_unlock(&sp->lock); |
42288fe3 MG |
2443 | ret = -ENOMEM; |
2444 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2445 | if (!n_new) | |
2446 | goto err_out; | |
2447 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2448 | if (!mpol_new) | |
2449 | goto err_out; | |
2450 | goto restart; | |
1da177e4 LT |
2451 | } |
2452 | ||
71fe804b LS |
2453 | /** |
2454 | * mpol_shared_policy_init - initialize shared policy for inode | |
2455 | * @sp: pointer to inode shared policy | |
2456 | * @mpol: struct mempolicy to install | |
2457 | * | |
2458 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2459 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2460 | * This must be released on exit. | |
4bfc4495 | 2461 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2462 | */ |
2463 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2464 | { | |
58568d2a MX |
2465 | int ret; |
2466 | ||
71fe804b | 2467 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2468 | rwlock_init(&sp->lock); |
71fe804b LS |
2469 | |
2470 | if (mpol) { | |
2471 | struct vm_area_struct pvma; | |
2472 | struct mempolicy *new; | |
4bfc4495 | 2473 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2474 | |
4bfc4495 | 2475 | if (!scratch) |
5c0c1654 | 2476 | goto put_mpol; |
71fe804b LS |
2477 | /* contextualize the tmpfs mount point mempolicy */ |
2478 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2479 | if (IS_ERR(new)) |
0cae3457 | 2480 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2481 | |
2482 | task_lock(current); | |
4bfc4495 | 2483 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2484 | task_unlock(current); |
15d77835 | 2485 | if (ret) |
5c0c1654 | 2486 | goto put_new; |
71fe804b LS |
2487 | |
2488 | /* Create pseudo-vma that contains just the policy */ | |
2489 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2490 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2491 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2492 | |
5c0c1654 | 2493 | put_new: |
71fe804b | 2494 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2495 | free_scratch: |
4bfc4495 | 2496 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2497 | put_mpol: |
2498 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2499 | } |
2500 | } | |
2501 | ||
1da177e4 LT |
2502 | int mpol_set_shared_policy(struct shared_policy *info, |
2503 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2504 | { | |
2505 | int err; | |
2506 | struct sp_node *new = NULL; | |
2507 | unsigned long sz = vma_pages(vma); | |
2508 | ||
028fec41 | 2509 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2510 | vma->vm_pgoff, |
45c4745a | 2511 | sz, npol ? npol->mode : -1, |
028fec41 | 2512 | npol ? npol->flags : -1, |
00ef2d2f | 2513 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2514 | |
2515 | if (npol) { | |
2516 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2517 | if (!new) | |
2518 | return -ENOMEM; | |
2519 | } | |
2520 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2521 | if (err && new) | |
63f74ca2 | 2522 | sp_free(new); |
1da177e4 LT |
2523 | return err; |
2524 | } | |
2525 | ||
2526 | /* Free a backing policy store on inode delete. */ | |
2527 | void mpol_free_shared_policy(struct shared_policy *p) | |
2528 | { | |
2529 | struct sp_node *n; | |
2530 | struct rb_node *next; | |
2531 | ||
2532 | if (!p->root.rb_node) | |
2533 | return; | |
4a8c7bb5 | 2534 | write_lock(&p->lock); |
1da177e4 LT |
2535 | next = rb_first(&p->root); |
2536 | while (next) { | |
2537 | n = rb_entry(next, struct sp_node, nd); | |
2538 | next = rb_next(&n->nd); | |
63f74ca2 | 2539 | sp_delete(p, n); |
1da177e4 | 2540 | } |
4a8c7bb5 | 2541 | write_unlock(&p->lock); |
1da177e4 LT |
2542 | } |
2543 | ||
1a687c2e | 2544 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2545 | static int __initdata numabalancing_override; |
1a687c2e MG |
2546 | |
2547 | static void __init check_numabalancing_enable(void) | |
2548 | { | |
2549 | bool numabalancing_default = false; | |
2550 | ||
2551 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2552 | numabalancing_default = true; | |
2553 | ||
c297663c MG |
2554 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2555 | if (numabalancing_override) | |
2556 | set_numabalancing_state(numabalancing_override == 1); | |
2557 | ||
b0dc2b9b | 2558 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2559 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2560 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2561 | set_numabalancing_state(numabalancing_default); |
2562 | } | |
2563 | } | |
2564 | ||
2565 | static int __init setup_numabalancing(char *str) | |
2566 | { | |
2567 | int ret = 0; | |
2568 | if (!str) | |
2569 | goto out; | |
1a687c2e MG |
2570 | |
2571 | if (!strcmp(str, "enable")) { | |
c297663c | 2572 | numabalancing_override = 1; |
1a687c2e MG |
2573 | ret = 1; |
2574 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2575 | numabalancing_override = -1; |
1a687c2e MG |
2576 | ret = 1; |
2577 | } | |
2578 | out: | |
2579 | if (!ret) | |
4a404bea | 2580 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2581 | |
2582 | return ret; | |
2583 | } | |
2584 | __setup("numa_balancing=", setup_numabalancing); | |
2585 | #else | |
2586 | static inline void __init check_numabalancing_enable(void) | |
2587 | { | |
2588 | } | |
2589 | #endif /* CONFIG_NUMA_BALANCING */ | |
2590 | ||
1da177e4 LT |
2591 | /* assumes fs == KERNEL_DS */ |
2592 | void __init numa_policy_init(void) | |
2593 | { | |
b71636e2 PM |
2594 | nodemask_t interleave_nodes; |
2595 | unsigned long largest = 0; | |
2596 | int nid, prefer = 0; | |
2597 | ||
1da177e4 LT |
2598 | policy_cache = kmem_cache_create("numa_policy", |
2599 | sizeof(struct mempolicy), | |
20c2df83 | 2600 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2601 | |
2602 | sn_cache = kmem_cache_create("shared_policy_node", | |
2603 | sizeof(struct sp_node), | |
20c2df83 | 2604 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2605 | |
5606e387 MG |
2606 | for_each_node(nid) { |
2607 | preferred_node_policy[nid] = (struct mempolicy) { | |
2608 | .refcnt = ATOMIC_INIT(1), | |
2609 | .mode = MPOL_PREFERRED, | |
2610 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2611 | .v = { .preferred_node = nid, }, | |
2612 | }; | |
2613 | } | |
2614 | ||
b71636e2 PM |
2615 | /* |
2616 | * Set interleaving policy for system init. Interleaving is only | |
2617 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2618 | * fall back to the largest node if they're all smaller. | |
2619 | */ | |
2620 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2621 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2622 | unsigned long total_pages = node_present_pages(nid); |
2623 | ||
2624 | /* Preserve the largest node */ | |
2625 | if (largest < total_pages) { | |
2626 | largest = total_pages; | |
2627 | prefer = nid; | |
2628 | } | |
2629 | ||
2630 | /* Interleave this node? */ | |
2631 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2632 | node_set(nid, interleave_nodes); | |
2633 | } | |
2634 | ||
2635 | /* All too small, use the largest */ | |
2636 | if (unlikely(nodes_empty(interleave_nodes))) | |
2637 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2638 | |
028fec41 | 2639 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2640 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2641 | |
2642 | check_numabalancing_enable(); | |
1da177e4 LT |
2643 | } |
2644 | ||
8bccd85f | 2645 | /* Reset policy of current process to default */ |
1da177e4 LT |
2646 | void numa_default_policy(void) |
2647 | { | |
028fec41 | 2648 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2649 | } |
68860ec1 | 2650 | |
095f1fc4 LS |
2651 | /* |
2652 | * Parse and format mempolicy from/to strings | |
2653 | */ | |
2654 | ||
1a75a6c8 | 2655 | /* |
f2a07f40 | 2656 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2657 | */ |
345ace9c LS |
2658 | static const char * const policy_modes[] = |
2659 | { | |
2660 | [MPOL_DEFAULT] = "default", | |
2661 | [MPOL_PREFERRED] = "prefer", | |
2662 | [MPOL_BIND] = "bind", | |
2663 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2664 | [MPOL_LOCAL] = "local", |
345ace9c | 2665 | }; |
1a75a6c8 | 2666 | |
095f1fc4 LS |
2667 | |
2668 | #ifdef CONFIG_TMPFS | |
2669 | /** | |
f2a07f40 | 2670 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2671 | * @str: string containing mempolicy to parse |
71fe804b | 2672 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2673 | * |
2674 | * Format of input: | |
2675 | * <mode>[=<flags>][:<nodelist>] | |
2676 | * | |
71fe804b | 2677 | * On success, returns 0, else 1 |
095f1fc4 | 2678 | */ |
a7a88b23 | 2679 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2680 | { |
71fe804b | 2681 | struct mempolicy *new = NULL; |
b4652e84 | 2682 | unsigned short mode; |
f2a07f40 | 2683 | unsigned short mode_flags; |
71fe804b | 2684 | nodemask_t nodes; |
095f1fc4 LS |
2685 | char *nodelist = strchr(str, ':'); |
2686 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2687 | int err = 1; |
2688 | ||
2689 | if (nodelist) { | |
2690 | /* NUL-terminate mode or flags string */ | |
2691 | *nodelist++ = '\0'; | |
71fe804b | 2692 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2693 | goto out; |
01f13bd6 | 2694 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2695 | goto out; |
71fe804b LS |
2696 | } else |
2697 | nodes_clear(nodes); | |
2698 | ||
095f1fc4 LS |
2699 | if (flags) |
2700 | *flags++ = '\0'; /* terminate mode string */ | |
2701 | ||
479e2802 | 2702 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2703 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2704 | break; |
2705 | } | |
2706 | } | |
a720094d | 2707 | if (mode >= MPOL_MAX) |
095f1fc4 LS |
2708 | goto out; |
2709 | ||
71fe804b | 2710 | switch (mode) { |
095f1fc4 | 2711 | case MPOL_PREFERRED: |
71fe804b LS |
2712 | /* |
2713 | * Insist on a nodelist of one node only | |
2714 | */ | |
095f1fc4 LS |
2715 | if (nodelist) { |
2716 | char *rest = nodelist; | |
2717 | while (isdigit(*rest)) | |
2718 | rest++; | |
926f2ae0 KM |
2719 | if (*rest) |
2720 | goto out; | |
095f1fc4 LS |
2721 | } |
2722 | break; | |
095f1fc4 LS |
2723 | case MPOL_INTERLEAVE: |
2724 | /* | |
2725 | * Default to online nodes with memory if no nodelist | |
2726 | */ | |
2727 | if (!nodelist) | |
01f13bd6 | 2728 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2729 | break; |
71fe804b | 2730 | case MPOL_LOCAL: |
3f226aa1 | 2731 | /* |
71fe804b | 2732 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2733 | */ |
71fe804b | 2734 | if (nodelist) |
3f226aa1 | 2735 | goto out; |
71fe804b | 2736 | mode = MPOL_PREFERRED; |
3f226aa1 | 2737 | break; |
413b43de RT |
2738 | case MPOL_DEFAULT: |
2739 | /* | |
2740 | * Insist on a empty nodelist | |
2741 | */ | |
2742 | if (!nodelist) | |
2743 | err = 0; | |
2744 | goto out; | |
d69b2e63 KM |
2745 | case MPOL_BIND: |
2746 | /* | |
2747 | * Insist on a nodelist | |
2748 | */ | |
2749 | if (!nodelist) | |
2750 | goto out; | |
095f1fc4 LS |
2751 | } |
2752 | ||
71fe804b | 2753 | mode_flags = 0; |
095f1fc4 LS |
2754 | if (flags) { |
2755 | /* | |
2756 | * Currently, we only support two mutually exclusive | |
2757 | * mode flags. | |
2758 | */ | |
2759 | if (!strcmp(flags, "static")) | |
71fe804b | 2760 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2761 | else if (!strcmp(flags, "relative")) |
71fe804b | 2762 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2763 | else |
926f2ae0 | 2764 | goto out; |
095f1fc4 | 2765 | } |
71fe804b LS |
2766 | |
2767 | new = mpol_new(mode, mode_flags, &nodes); | |
2768 | if (IS_ERR(new)) | |
926f2ae0 KM |
2769 | goto out; |
2770 | ||
f2a07f40 HD |
2771 | /* |
2772 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2773 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2774 | */ | |
2775 | if (mode != MPOL_PREFERRED) | |
2776 | new->v.nodes = nodes; | |
2777 | else if (nodelist) | |
2778 | new->v.preferred_node = first_node(nodes); | |
2779 | else | |
2780 | new->flags |= MPOL_F_LOCAL; | |
2781 | ||
2782 | /* | |
2783 | * Save nodes for contextualization: this will be used to "clone" | |
2784 | * the mempolicy in a specific context [cpuset] at a later time. | |
2785 | */ | |
2786 | new->w.user_nodemask = nodes; | |
2787 | ||
926f2ae0 | 2788 | err = 0; |
71fe804b | 2789 | |
095f1fc4 LS |
2790 | out: |
2791 | /* Restore string for error message */ | |
2792 | if (nodelist) | |
2793 | *--nodelist = ':'; | |
2794 | if (flags) | |
2795 | *--flags = '='; | |
71fe804b LS |
2796 | if (!err) |
2797 | *mpol = new; | |
095f1fc4 LS |
2798 | return err; |
2799 | } | |
2800 | #endif /* CONFIG_TMPFS */ | |
2801 | ||
71fe804b LS |
2802 | /** |
2803 | * mpol_to_str - format a mempolicy structure for printing | |
2804 | * @buffer: to contain formatted mempolicy string | |
2805 | * @maxlen: length of @buffer | |
2806 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2807 | * |
948927ee DR |
2808 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
2809 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
2810 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 2811 | */ |
948927ee | 2812 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2813 | { |
2814 | char *p = buffer; | |
948927ee DR |
2815 | nodemask_t nodes = NODE_MASK_NONE; |
2816 | unsigned short mode = MPOL_DEFAULT; | |
2817 | unsigned short flags = 0; | |
2291990a | 2818 | |
8790c71a | 2819 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 2820 | mode = pol->mode; |
948927ee DR |
2821 | flags = pol->flags; |
2822 | } | |
bea904d5 | 2823 | |
1a75a6c8 CL |
2824 | switch (mode) { |
2825 | case MPOL_DEFAULT: | |
1a75a6c8 | 2826 | break; |
1a75a6c8 | 2827 | case MPOL_PREFERRED: |
fc36b8d3 | 2828 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 2829 | mode = MPOL_LOCAL; |
53f2556b | 2830 | else |
fc36b8d3 | 2831 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 2832 | break; |
1a75a6c8 | 2833 | case MPOL_BIND: |
1a75a6c8 | 2834 | case MPOL_INTERLEAVE: |
f2a07f40 | 2835 | nodes = pol->v.nodes; |
1a75a6c8 | 2836 | break; |
1a75a6c8 | 2837 | default: |
948927ee DR |
2838 | WARN_ON_ONCE(1); |
2839 | snprintf(p, maxlen, "unknown"); | |
2840 | return; | |
1a75a6c8 CL |
2841 | } |
2842 | ||
b7a9f420 | 2843 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 2844 | |
fc36b8d3 | 2845 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 2846 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 2847 | |
2291990a LS |
2848 | /* |
2849 | * Currently, the only defined flags are mutually exclusive | |
2850 | */ | |
f5b087b5 | 2851 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2852 | p += snprintf(p, buffer + maxlen - p, "static"); |
2853 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2854 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2855 | } |
2856 | ||
9e763e0f TH |
2857 | if (!nodes_empty(nodes)) |
2858 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
2859 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 2860 | } |