]>
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> | |
1da177e4 LT |
76 | #include <linux/nodemask.h> |
77 | #include <linux/cpuset.h> | |
1da177e4 LT |
78 | #include <linux/slab.h> |
79 | #include <linux/string.h> | |
b95f1b31 | 80 | #include <linux/export.h> |
b488893a | 81 | #include <linux/nsproxy.h> |
1da177e4 LT |
82 | #include <linux/interrupt.h> |
83 | #include <linux/init.h> | |
84 | #include <linux/compat.h> | |
dc9aa5b9 | 85 | #include <linux/swap.h> |
1a75a6c8 CL |
86 | #include <linux/seq_file.h> |
87 | #include <linux/proc_fs.h> | |
b20a3503 | 88 | #include <linux/migrate.h> |
62b61f61 | 89 | #include <linux/ksm.h> |
95a402c3 | 90 | #include <linux/rmap.h> |
86c3a764 | 91 | #include <linux/security.h> |
dbcb0f19 | 92 | #include <linux/syscalls.h> |
095f1fc4 | 93 | #include <linux/ctype.h> |
6d9c285a | 94 | #include <linux/mm_inline.h> |
b24f53a0 | 95 | #include <linux/mmu_notifier.h> |
b1de0d13 | 96 | #include <linux/printk.h> |
dc9aa5b9 | 97 | |
1da177e4 LT |
98 | #include <asm/tlbflush.h> |
99 | #include <asm/uaccess.h> | |
778d3b0f | 100 | #include <linux/random.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 | ||
19770b32 | 165 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 166 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 167 | { |
d3eb1570 | 168 | return nodes_intersects(*nodemask, node_states[N_MEMORY]); |
1da177e4 LT |
169 | } |
170 | ||
f5b087b5 DR |
171 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
172 | { | |
6d556294 | 173 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
174 | } |
175 | ||
176 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
177 | const nodemask_t *rel) | |
178 | { | |
179 | nodemask_t tmp; | |
180 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
181 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
182 | } |
183 | ||
37012946 DR |
184 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
185 | { | |
186 | if (nodes_empty(*nodes)) | |
187 | return -EINVAL; | |
188 | pol->v.nodes = *nodes; | |
189 | return 0; | |
190 | } | |
191 | ||
192 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
193 | { | |
194 | if (!nodes) | |
fc36b8d3 | 195 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
196 | else if (nodes_empty(*nodes)) |
197 | return -EINVAL; /* no allowed nodes */ | |
198 | else | |
199 | pol->v.preferred_node = first_node(*nodes); | |
200 | return 0; | |
201 | } | |
202 | ||
203 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
204 | { | |
205 | if (!is_valid_nodemask(nodes)) | |
206 | return -EINVAL; | |
207 | pol->v.nodes = *nodes; | |
208 | return 0; | |
209 | } | |
210 | ||
58568d2a MX |
211 | /* |
212 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
213 | * any, for the new policy. mpol_new() has already validated the nodes | |
214 | * parameter with respect to the policy mode and flags. But, we need to | |
215 | * handle an empty nodemask with MPOL_PREFERRED here. | |
216 | * | |
217 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
218 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
219 | */ | |
4bfc4495 KH |
220 | static int mpol_set_nodemask(struct mempolicy *pol, |
221 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 222 | { |
58568d2a MX |
223 | int ret; |
224 | ||
225 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
226 | if (pol == NULL) | |
227 | return 0; | |
01f13bd6 | 228 | /* Check N_MEMORY */ |
4bfc4495 | 229 | nodes_and(nsc->mask1, |
01f13bd6 | 230 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
231 | |
232 | VM_BUG_ON(!nodes); | |
233 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
234 | nodes = NULL; /* explicit local allocation */ | |
235 | else { | |
236 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
4bfc4495 | 237 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
58568d2a | 238 | else |
4bfc4495 KH |
239 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
240 | ||
58568d2a MX |
241 | if (mpol_store_user_nodemask(pol)) |
242 | pol->w.user_nodemask = *nodes; | |
243 | else | |
244 | pol->w.cpuset_mems_allowed = | |
245 | cpuset_current_mems_allowed; | |
246 | } | |
247 | ||
4bfc4495 KH |
248 | if (nodes) |
249 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
250 | else | |
251 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
252 | return ret; |
253 | } | |
254 | ||
255 | /* | |
256 | * This function just creates a new policy, does some check and simple | |
257 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
258 | */ | |
028fec41 DR |
259 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
260 | nodemask_t *nodes) | |
1da177e4 LT |
261 | { |
262 | struct mempolicy *policy; | |
263 | ||
028fec41 | 264 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 265 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 266 | |
3e1f0645 DR |
267 | if (mode == MPOL_DEFAULT) { |
268 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 269 | return ERR_PTR(-EINVAL); |
d3a71033 | 270 | return NULL; |
37012946 | 271 | } |
3e1f0645 DR |
272 | VM_BUG_ON(!nodes); |
273 | ||
274 | /* | |
275 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
276 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
277 | * All other modes require a valid pointer to a non-empty nodemask. | |
278 | */ | |
279 | if (mode == MPOL_PREFERRED) { | |
280 | if (nodes_empty(*nodes)) { | |
281 | if (((flags & MPOL_F_STATIC_NODES) || | |
282 | (flags & MPOL_F_RELATIVE_NODES))) | |
283 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 284 | } |
479e2802 PZ |
285 | } else if (mode == MPOL_LOCAL) { |
286 | if (!nodes_empty(*nodes)) | |
287 | return ERR_PTR(-EINVAL); | |
288 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
289 | } else if (nodes_empty(*nodes)) |
290 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
291 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
292 | if (!policy) | |
293 | return ERR_PTR(-ENOMEM); | |
294 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 295 | policy->mode = mode; |
3e1f0645 | 296 | policy->flags = flags; |
37012946 | 297 | |
1da177e4 | 298 | return policy; |
37012946 DR |
299 | } |
300 | ||
52cd3b07 LS |
301 | /* Slow path of a mpol destructor. */ |
302 | void __mpol_put(struct mempolicy *p) | |
303 | { | |
304 | if (!atomic_dec_and_test(&p->refcnt)) | |
305 | return; | |
52cd3b07 LS |
306 | kmem_cache_free(policy_cache, p); |
307 | } | |
308 | ||
708c1bbc MX |
309 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
310 | enum mpol_rebind_step step) | |
37012946 DR |
311 | { |
312 | } | |
313 | ||
708c1bbc MX |
314 | /* |
315 | * step: | |
316 | * MPOL_REBIND_ONCE - do rebind work at once | |
317 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
318 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
319 | */ | |
320 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
321 | enum mpol_rebind_step step) | |
37012946 DR |
322 | { |
323 | nodemask_t tmp; | |
324 | ||
325 | if (pol->flags & MPOL_F_STATIC_NODES) | |
326 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
327 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
328 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
329 | else { | |
708c1bbc MX |
330 | /* |
331 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
332 | * result | |
333 | */ | |
334 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
335 | nodes_remap(tmp, pol->v.nodes, | |
336 | pol->w.cpuset_mems_allowed, *nodes); | |
337 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
338 | } else if (step == MPOL_REBIND_STEP2) { | |
339 | tmp = pol->w.cpuset_mems_allowed; | |
340 | pol->w.cpuset_mems_allowed = *nodes; | |
341 | } else | |
342 | BUG(); | |
37012946 | 343 | } |
f5b087b5 | 344 | |
708c1bbc MX |
345 | if (nodes_empty(tmp)) |
346 | tmp = *nodes; | |
347 | ||
348 | if (step == MPOL_REBIND_STEP1) | |
349 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
350 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
351 | pol->v.nodes = tmp; | |
352 | else | |
353 | BUG(); | |
354 | ||
37012946 DR |
355 | if (!node_isset(current->il_next, tmp)) { |
356 | current->il_next = next_node(current->il_next, tmp); | |
357 | if (current->il_next >= MAX_NUMNODES) | |
358 | current->il_next = first_node(tmp); | |
359 | if (current->il_next >= MAX_NUMNODES) | |
360 | current->il_next = numa_node_id(); | |
361 | } | |
362 | } | |
363 | ||
364 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
365 | const nodemask_t *nodes, |
366 | enum mpol_rebind_step step) | |
37012946 DR |
367 | { |
368 | nodemask_t tmp; | |
369 | ||
37012946 DR |
370 | if (pol->flags & MPOL_F_STATIC_NODES) { |
371 | int node = first_node(pol->w.user_nodemask); | |
372 | ||
fc36b8d3 | 373 | if (node_isset(node, *nodes)) { |
37012946 | 374 | pol->v.preferred_node = node; |
fc36b8d3 LS |
375 | pol->flags &= ~MPOL_F_LOCAL; |
376 | } else | |
377 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
378 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
379 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
380 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 381 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
382 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
383 | pol->w.cpuset_mems_allowed, | |
384 | *nodes); | |
385 | pol->w.cpuset_mems_allowed = *nodes; | |
386 | } | |
1da177e4 LT |
387 | } |
388 | ||
708c1bbc MX |
389 | /* |
390 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
391 | * | |
392 | * If read-side task has no lock to protect task->mempolicy, write-side | |
393 | * task will rebind the task->mempolicy by two step. The first step is | |
394 | * setting all the newly nodes, and the second step is cleaning all the | |
395 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
396 | * page. | |
397 | * If we have a lock to protect task->mempolicy in read-side, we do | |
398 | * rebind directly. | |
399 | * | |
400 | * step: | |
401 | * MPOL_REBIND_ONCE - do rebind work at once | |
402 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
403 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
404 | */ | |
405 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
406 | enum mpol_rebind_step step) | |
1d0d2680 | 407 | { |
1d0d2680 DR |
408 | if (!pol) |
409 | return; | |
89c522c7 | 410 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
1d0d2680 DR |
411 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
412 | return; | |
708c1bbc MX |
413 | |
414 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
415 | return; | |
416 | ||
417 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
418 | BUG(); | |
419 | ||
420 | if (step == MPOL_REBIND_STEP1) | |
421 | pol->flags |= MPOL_F_REBINDING; | |
422 | else if (step == MPOL_REBIND_STEP2) | |
423 | pol->flags &= ~MPOL_F_REBINDING; | |
424 | else if (step >= MPOL_REBIND_NSTEP) | |
425 | BUG(); | |
426 | ||
427 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
428 | } |
429 | ||
430 | /* | |
431 | * Wrapper for mpol_rebind_policy() that just requires task | |
432 | * pointer, and updates task mempolicy. | |
58568d2a MX |
433 | * |
434 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
435 | */ |
436 | ||
708c1bbc MX |
437 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
438 | enum mpol_rebind_step step) | |
1d0d2680 | 439 | { |
708c1bbc | 440 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
441 | } |
442 | ||
443 | /* | |
444 | * Rebind each vma in mm to new nodemask. | |
445 | * | |
446 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
447 | */ | |
448 | ||
449 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
450 | { | |
451 | struct vm_area_struct *vma; | |
452 | ||
453 | down_write(&mm->mmap_sem); | |
454 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 455 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
456 | up_write(&mm->mmap_sem); |
457 | } | |
458 | ||
37012946 DR |
459 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
460 | [MPOL_DEFAULT] = { | |
461 | .rebind = mpol_rebind_default, | |
462 | }, | |
463 | [MPOL_INTERLEAVE] = { | |
464 | .create = mpol_new_interleave, | |
465 | .rebind = mpol_rebind_nodemask, | |
466 | }, | |
467 | [MPOL_PREFERRED] = { | |
468 | .create = mpol_new_preferred, | |
469 | .rebind = mpol_rebind_preferred, | |
470 | }, | |
471 | [MPOL_BIND] = { | |
472 | .create = mpol_new_bind, | |
473 | .rebind = mpol_rebind_nodemask, | |
474 | }, | |
475 | }; | |
476 | ||
fc301289 CL |
477 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
478 | unsigned long flags); | |
1a75a6c8 | 479 | |
98094945 NH |
480 | /* |
481 | * Scan through pages checking if pages follow certain conditions, | |
482 | * and move them to the pagelist if they do. | |
483 | */ | |
484 | static int queue_pages_pte_range(struct vm_area_struct *vma, pmd_t *pmd, | |
dc9aa5b9 CL |
485 | unsigned long addr, unsigned long end, |
486 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 487 | void *private) |
1da177e4 | 488 | { |
91612e0d HD |
489 | pte_t *orig_pte; |
490 | pte_t *pte; | |
705e87c0 | 491 | spinlock_t *ptl; |
941150a3 | 492 | |
705e87c0 | 493 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 494 | do { |
6aab341e | 495 | struct page *page; |
25ba77c1 | 496 | int nid; |
91612e0d HD |
497 | |
498 | if (!pte_present(*pte)) | |
1da177e4 | 499 | continue; |
6aab341e LT |
500 | page = vm_normal_page(vma, addr, *pte); |
501 | if (!page) | |
1da177e4 | 502 | continue; |
053837fc | 503 | /* |
62b61f61 HD |
504 | * vm_normal_page() filters out zero pages, but there might |
505 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 506 | */ |
b79bc0a0 | 507 | if (PageReserved(page)) |
f4598c8b | 508 | continue; |
6aab341e | 509 | nid = page_to_nid(page); |
38e35860 CL |
510 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
511 | continue; | |
512 | ||
b1f72d18 | 513 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 514 | migrate_page_add(page, private, flags); |
38e35860 CL |
515 | else |
516 | break; | |
91612e0d | 517 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 518 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
519 | return addr != end; |
520 | } | |
521 | ||
98094945 NH |
522 | static void queue_pages_hugetlb_pmd_range(struct vm_area_struct *vma, |
523 | pmd_t *pmd, const nodemask_t *nodes, unsigned long flags, | |
e2d8cf40 NH |
524 | void *private) |
525 | { | |
526 | #ifdef CONFIG_HUGETLB_PAGE | |
527 | int nid; | |
528 | struct page *page; | |
cb900f41 | 529 | spinlock_t *ptl; |
d4c54919 | 530 | pte_t entry; |
e2d8cf40 | 531 | |
cb900f41 | 532 | ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, (pte_t *)pmd); |
d4c54919 NH |
533 | entry = huge_ptep_get((pte_t *)pmd); |
534 | if (!pte_present(entry)) | |
535 | goto unlock; | |
536 | page = pte_page(entry); | |
e2d8cf40 NH |
537 | nid = page_to_nid(page); |
538 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) | |
539 | goto unlock; | |
540 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ | |
541 | if (flags & (MPOL_MF_MOVE_ALL) || | |
542 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) | |
543 | isolate_huge_page(page, private); | |
544 | unlock: | |
cb900f41 | 545 | spin_unlock(ptl); |
e2d8cf40 NH |
546 | #else |
547 | BUG(); | |
548 | #endif | |
549 | } | |
550 | ||
98094945 | 551 | static inline int queue_pages_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
552 | unsigned long addr, unsigned long end, |
553 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 554 | void *private) |
91612e0d HD |
555 | { |
556 | pmd_t *pmd; | |
557 | unsigned long next; | |
558 | ||
559 | pmd = pmd_offset(pud, addr); | |
560 | do { | |
561 | next = pmd_addr_end(addr, end); | |
e2d8cf40 NH |
562 | if (!pmd_present(*pmd)) |
563 | continue; | |
564 | if (pmd_huge(*pmd) && is_vm_hugetlb_page(vma)) { | |
98094945 | 565 | queue_pages_hugetlb_pmd_range(vma, pmd, nodes, |
e2d8cf40 NH |
566 | flags, private); |
567 | continue; | |
568 | } | |
e180377f | 569 | split_huge_page_pmd(vma, addr, pmd); |
1a5a9906 | 570 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
91612e0d | 571 | continue; |
98094945 | 572 | if (queue_pages_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 573 | flags, private)) |
91612e0d HD |
574 | return -EIO; |
575 | } while (pmd++, addr = next, addr != end); | |
576 | return 0; | |
577 | } | |
578 | ||
98094945 | 579 | static inline int queue_pages_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
580 | unsigned long addr, unsigned long end, |
581 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 582 | void *private) |
91612e0d HD |
583 | { |
584 | pud_t *pud; | |
585 | unsigned long next; | |
586 | ||
587 | pud = pud_offset(pgd, addr); | |
588 | do { | |
589 | next = pud_addr_end(addr, end); | |
e2d8cf40 NH |
590 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) |
591 | continue; | |
91612e0d HD |
592 | if (pud_none_or_clear_bad(pud)) |
593 | continue; | |
98094945 | 594 | if (queue_pages_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 595 | flags, private)) |
91612e0d HD |
596 | return -EIO; |
597 | } while (pud++, addr = next, addr != end); | |
598 | return 0; | |
599 | } | |
600 | ||
98094945 | 601 | static inline int queue_pages_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
602 | unsigned long addr, unsigned long end, |
603 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 604 | void *private) |
91612e0d HD |
605 | { |
606 | pgd_t *pgd; | |
607 | unsigned long next; | |
608 | ||
b5810039 | 609 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
610 | do { |
611 | next = pgd_addr_end(addr, end); | |
612 | if (pgd_none_or_clear_bad(pgd)) | |
613 | continue; | |
98094945 | 614 | if (queue_pages_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 615 | flags, private)) |
91612e0d HD |
616 | return -EIO; |
617 | } while (pgd++, addr = next, addr != end); | |
618 | return 0; | |
1da177e4 LT |
619 | } |
620 | ||
5877231f | 621 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 622 | /* |
4b10e7d5 MG |
623 | * This is used to mark a range of virtual addresses to be inaccessible. |
624 | * These are later cleared by a NUMA hinting fault. Depending on these | |
625 | * faults, pages may be migrated for better NUMA placement. | |
626 | * | |
627 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
628 | * an architecture makes a different choice, it will need further | |
629 | * changes to the core. | |
b24f53a0 | 630 | */ |
4b10e7d5 MG |
631 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
632 | unsigned long addr, unsigned long end) | |
b24f53a0 | 633 | { |
4b10e7d5 | 634 | int nr_updated; |
b24f53a0 | 635 | |
4b10e7d5 | 636 | nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1); |
03c5a6e1 MG |
637 | if (nr_updated) |
638 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 639 | |
4b10e7d5 | 640 | return nr_updated; |
b24f53a0 LS |
641 | } |
642 | #else | |
643 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
644 | unsigned long addr, unsigned long end) | |
645 | { | |
646 | return 0; | |
647 | } | |
5877231f | 648 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 649 | |
dc9aa5b9 | 650 | /* |
98094945 NH |
651 | * Walk through page tables and collect pages to be migrated. |
652 | * | |
653 | * If pages found in a given range are on a set of nodes (determined by | |
654 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
655 | * passed via @private.) | |
dc9aa5b9 | 656 | */ |
d05f0cdc | 657 | static int |
98094945 | 658 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
38e35860 | 659 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 | 660 | { |
d05f0cdc HD |
661 | int err = 0; |
662 | struct vm_area_struct *vma, *prev; | |
053837fc | 663 | |
d05f0cdc HD |
664 | vma = find_vma(mm, start); |
665 | if (!vma) | |
666 | return -EFAULT; | |
1da177e4 | 667 | prev = NULL; |
d05f0cdc | 668 | for (; vma && vma->vm_start < end; vma = vma->vm_next) { |
b24f53a0 LS |
669 | unsigned long endvma = vma->vm_end; |
670 | ||
671 | if (endvma > end) | |
672 | endvma = end; | |
673 | if (vma->vm_start > start) | |
674 | start = vma->vm_start; | |
675 | ||
dc9aa5b9 CL |
676 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
677 | if (!vma->vm_next && vma->vm_end < end) | |
d05f0cdc | 678 | return -EFAULT; |
dc9aa5b9 | 679 | if (prev && prev->vm_end < vma->vm_start) |
d05f0cdc | 680 | return -EFAULT; |
dc9aa5b9 | 681 | } |
b24f53a0 | 682 | |
b24f53a0 | 683 | if (flags & MPOL_MF_LAZY) { |
2c0346a3 MG |
684 | /* Similar to task_numa_work, skip inaccessible VMAs */ |
685 | if (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) | |
686 | change_prot_numa(vma, start, endvma); | |
b24f53a0 LS |
687 | goto next; |
688 | } | |
689 | ||
690 | if ((flags & MPOL_MF_STRICT) || | |
dc9aa5b9 | 691 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
b24f53a0 | 692 | vma_migratable(vma))) { |
dc9aa5b9 | 693 | |
98094945 | 694 | err = queue_pages_pgd_range(vma, start, endvma, nodes, |
38e35860 | 695 | flags, private); |
d05f0cdc | 696 | if (err) |
1da177e4 | 697 | break; |
1da177e4 | 698 | } |
b24f53a0 | 699 | next: |
1da177e4 LT |
700 | prev = vma; |
701 | } | |
d05f0cdc | 702 | return err; |
1da177e4 LT |
703 | } |
704 | ||
869833f2 KM |
705 | /* |
706 | * Apply policy to a single VMA | |
707 | * This must be called with the mmap_sem held for writing. | |
708 | */ | |
709 | static int vma_replace_policy(struct vm_area_struct *vma, | |
710 | struct mempolicy *pol) | |
8d34694c | 711 | { |
869833f2 KM |
712 | int err; |
713 | struct mempolicy *old; | |
714 | struct mempolicy *new; | |
8d34694c KM |
715 | |
716 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
717 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
718 | vma->vm_ops, vma->vm_file, | |
719 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
720 | ||
869833f2 KM |
721 | new = mpol_dup(pol); |
722 | if (IS_ERR(new)) | |
723 | return PTR_ERR(new); | |
724 | ||
725 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 726 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
727 | if (err) |
728 | goto err_out; | |
8d34694c | 729 | } |
869833f2 KM |
730 | |
731 | old = vma->vm_policy; | |
732 | vma->vm_policy = new; /* protected by mmap_sem */ | |
733 | mpol_put(old); | |
734 | ||
735 | return 0; | |
736 | err_out: | |
737 | mpol_put(new); | |
8d34694c KM |
738 | return err; |
739 | } | |
740 | ||
1da177e4 | 741 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
742 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
743 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
744 | { |
745 | struct vm_area_struct *next; | |
9d8cebd4 KM |
746 | struct vm_area_struct *prev; |
747 | struct vm_area_struct *vma; | |
748 | int err = 0; | |
e26a5114 | 749 | pgoff_t pgoff; |
9d8cebd4 KM |
750 | unsigned long vmstart; |
751 | unsigned long vmend; | |
1da177e4 | 752 | |
097d5910 | 753 | vma = find_vma(mm, start); |
9d8cebd4 KM |
754 | if (!vma || vma->vm_start > start) |
755 | return -EFAULT; | |
756 | ||
097d5910 | 757 | prev = vma->vm_prev; |
e26a5114 KM |
758 | if (start > vma->vm_start) |
759 | prev = vma; | |
760 | ||
9d8cebd4 | 761 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 762 | next = vma->vm_next; |
9d8cebd4 KM |
763 | vmstart = max(start, vma->vm_start); |
764 | vmend = min(end, vma->vm_end); | |
765 | ||
e26a5114 KM |
766 | if (mpol_equal(vma_policy(vma), new_pol)) |
767 | continue; | |
768 | ||
769 | pgoff = vma->vm_pgoff + | |
770 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 771 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 772 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 773 | new_pol); |
9d8cebd4 KM |
774 | if (prev) { |
775 | vma = prev; | |
776 | next = vma->vm_next; | |
3964acd0 ON |
777 | if (mpol_equal(vma_policy(vma), new_pol)) |
778 | continue; | |
779 | /* vma_merge() joined vma && vma->next, case 8 */ | |
780 | goto replace; | |
9d8cebd4 KM |
781 | } |
782 | if (vma->vm_start != vmstart) { | |
783 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
784 | if (err) | |
785 | goto out; | |
786 | } | |
787 | if (vma->vm_end != vmend) { | |
788 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
789 | if (err) | |
790 | goto out; | |
791 | } | |
3964acd0 | 792 | replace: |
869833f2 | 793 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
794 | if (err) |
795 | goto out; | |
1da177e4 | 796 | } |
9d8cebd4 KM |
797 | |
798 | out: | |
1da177e4 LT |
799 | return err; |
800 | } | |
801 | ||
1da177e4 | 802 | /* Set the process memory policy */ |
028fec41 DR |
803 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
804 | nodemask_t *nodes) | |
1da177e4 | 805 | { |
58568d2a | 806 | struct mempolicy *new, *old; |
4bfc4495 | 807 | NODEMASK_SCRATCH(scratch); |
58568d2a | 808 | int ret; |
1da177e4 | 809 | |
4bfc4495 KH |
810 | if (!scratch) |
811 | return -ENOMEM; | |
f4e53d91 | 812 | |
4bfc4495 KH |
813 | new = mpol_new(mode, flags, nodes); |
814 | if (IS_ERR(new)) { | |
815 | ret = PTR_ERR(new); | |
816 | goto out; | |
817 | } | |
2c7c3a7d | 818 | |
58568d2a | 819 | task_lock(current); |
4bfc4495 | 820 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
821 | if (ret) { |
822 | task_unlock(current); | |
58568d2a | 823 | mpol_put(new); |
4bfc4495 | 824 | goto out; |
58568d2a MX |
825 | } |
826 | old = current->mempolicy; | |
1da177e4 | 827 | current->mempolicy = new; |
45c4745a | 828 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 829 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 830 | current->il_next = first_node(new->v.nodes); |
58568d2a | 831 | task_unlock(current); |
58568d2a | 832 | mpol_put(old); |
4bfc4495 KH |
833 | ret = 0; |
834 | out: | |
835 | NODEMASK_SCRATCH_FREE(scratch); | |
836 | return ret; | |
1da177e4 LT |
837 | } |
838 | ||
bea904d5 LS |
839 | /* |
840 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
841 | * |
842 | * Called with task's alloc_lock held | |
bea904d5 LS |
843 | */ |
844 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 845 | { |
dfcd3c0d | 846 | nodes_clear(*nodes); |
bea904d5 LS |
847 | if (p == &default_policy) |
848 | return; | |
849 | ||
45c4745a | 850 | switch (p->mode) { |
19770b32 MG |
851 | case MPOL_BIND: |
852 | /* Fall through */ | |
1da177e4 | 853 | case MPOL_INTERLEAVE: |
dfcd3c0d | 854 | *nodes = p->v.nodes; |
1da177e4 LT |
855 | break; |
856 | case MPOL_PREFERRED: | |
fc36b8d3 | 857 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 858 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 859 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
860 | break; |
861 | default: | |
862 | BUG(); | |
863 | } | |
864 | } | |
865 | ||
866 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
867 | { | |
868 | struct page *p; | |
869 | int err; | |
870 | ||
871 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
872 | if (err >= 0) { | |
873 | err = page_to_nid(p); | |
874 | put_page(p); | |
875 | } | |
876 | return err; | |
877 | } | |
878 | ||
1da177e4 | 879 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
880 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
881 | unsigned long addr, unsigned long flags) | |
1da177e4 | 882 | { |
8bccd85f | 883 | int err; |
1da177e4 LT |
884 | struct mm_struct *mm = current->mm; |
885 | struct vm_area_struct *vma = NULL; | |
886 | struct mempolicy *pol = current->mempolicy; | |
887 | ||
754af6f5 LS |
888 | if (flags & |
889 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 890 | return -EINVAL; |
754af6f5 LS |
891 | |
892 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
893 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
894 | return -EINVAL; | |
895 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 896 | task_lock(current); |
754af6f5 | 897 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 898 | task_unlock(current); |
754af6f5 LS |
899 | return 0; |
900 | } | |
901 | ||
1da177e4 | 902 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
903 | /* |
904 | * Do NOT fall back to task policy if the | |
905 | * vma/shared policy at addr is NULL. We | |
906 | * want to return MPOL_DEFAULT in this case. | |
907 | */ | |
1da177e4 LT |
908 | down_read(&mm->mmap_sem); |
909 | vma = find_vma_intersection(mm, addr, addr+1); | |
910 | if (!vma) { | |
911 | up_read(&mm->mmap_sem); | |
912 | return -EFAULT; | |
913 | } | |
914 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
915 | pol = vma->vm_ops->get_policy(vma, addr); | |
916 | else | |
917 | pol = vma->vm_policy; | |
918 | } else if (addr) | |
919 | return -EINVAL; | |
920 | ||
921 | if (!pol) | |
bea904d5 | 922 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
923 | |
924 | if (flags & MPOL_F_NODE) { | |
925 | if (flags & MPOL_F_ADDR) { | |
926 | err = lookup_node(mm, addr); | |
927 | if (err < 0) | |
928 | goto out; | |
8bccd85f | 929 | *policy = err; |
1da177e4 | 930 | } else if (pol == current->mempolicy && |
45c4745a | 931 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 932 | *policy = current->il_next; |
1da177e4 LT |
933 | } else { |
934 | err = -EINVAL; | |
935 | goto out; | |
936 | } | |
bea904d5 LS |
937 | } else { |
938 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
939 | pol->mode; | |
d79df630 DR |
940 | /* |
941 | * Internal mempolicy flags must be masked off before exposing | |
942 | * the policy to userspace. | |
943 | */ | |
944 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 945 | } |
1da177e4 LT |
946 | |
947 | if (vma) { | |
948 | up_read(¤t->mm->mmap_sem); | |
949 | vma = NULL; | |
950 | } | |
951 | ||
1da177e4 | 952 | err = 0; |
58568d2a | 953 | if (nmask) { |
c6b6ef8b LS |
954 | if (mpol_store_user_nodemask(pol)) { |
955 | *nmask = pol->w.user_nodemask; | |
956 | } else { | |
957 | task_lock(current); | |
958 | get_policy_nodemask(pol, nmask); | |
959 | task_unlock(current); | |
960 | } | |
58568d2a | 961 | } |
1da177e4 LT |
962 | |
963 | out: | |
52cd3b07 | 964 | mpol_cond_put(pol); |
1da177e4 LT |
965 | if (vma) |
966 | up_read(¤t->mm->mmap_sem); | |
967 | return err; | |
968 | } | |
969 | ||
b20a3503 | 970 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
971 | /* |
972 | * page migration | |
973 | */ | |
fc301289 CL |
974 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
975 | unsigned long flags) | |
6ce3c4c0 CL |
976 | { |
977 | /* | |
fc301289 | 978 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 979 | */ |
62695a84 NP |
980 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
981 | if (!isolate_lru_page(page)) { | |
982 | list_add_tail(&page->lru, pagelist); | |
6d9c285a KM |
983 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
984 | page_is_file_cache(page)); | |
62695a84 NP |
985 | } |
986 | } | |
7e2ab150 | 987 | } |
6ce3c4c0 | 988 | |
742755a1 | 989 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 990 | { |
e2d8cf40 NH |
991 | if (PageHuge(page)) |
992 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
993 | node); | |
994 | else | |
995 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); | |
95a402c3 CL |
996 | } |
997 | ||
7e2ab150 CL |
998 | /* |
999 | * Migrate pages from one node to a target node. | |
1000 | * Returns error or the number of pages not migrated. | |
1001 | */ | |
dbcb0f19 AB |
1002 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1003 | int flags) | |
7e2ab150 CL |
1004 | { |
1005 | nodemask_t nmask; | |
1006 | LIST_HEAD(pagelist); | |
1007 | int err = 0; | |
1008 | ||
1009 | nodes_clear(nmask); | |
1010 | node_set(source, nmask); | |
6ce3c4c0 | 1011 | |
08270807 MK |
1012 | /* |
1013 | * This does not "check" the range but isolates all pages that | |
1014 | * need migration. Between passing in the full user address | |
1015 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1016 | */ | |
1017 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1018 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1019 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1020 | ||
cf608ac1 | 1021 | if (!list_empty(&pagelist)) { |
68711a74 | 1022 | err = migrate_pages(&pagelist, new_node_page, NULL, dest, |
9c620e2b | 1023 | MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1024 | if (err) |
e2d8cf40 | 1025 | putback_movable_pages(&pagelist); |
cf608ac1 | 1026 | } |
95a402c3 | 1027 | |
7e2ab150 | 1028 | return err; |
6ce3c4c0 CL |
1029 | } |
1030 | ||
39743889 | 1031 | /* |
7e2ab150 CL |
1032 | * Move pages between the two nodesets so as to preserve the physical |
1033 | * layout as much as possible. | |
39743889 CL |
1034 | * |
1035 | * Returns the number of page that could not be moved. | |
1036 | */ | |
0ce72d4f AM |
1037 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1038 | const nodemask_t *to, int flags) | |
39743889 | 1039 | { |
7e2ab150 | 1040 | int busy = 0; |
0aedadf9 | 1041 | int err; |
7e2ab150 | 1042 | nodemask_t tmp; |
39743889 | 1043 | |
0aedadf9 CL |
1044 | err = migrate_prep(); |
1045 | if (err) | |
1046 | return err; | |
1047 | ||
53f2556b | 1048 | down_read(&mm->mmap_sem); |
39743889 | 1049 | |
0ce72d4f | 1050 | err = migrate_vmas(mm, from, to, flags); |
7b2259b3 CL |
1051 | if (err) |
1052 | goto out; | |
1053 | ||
da0aa138 KM |
1054 | /* |
1055 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1056 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1057 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1058 | * The pair of nodemasks 'to' and 'from' define the map. | |
1059 | * | |
1060 | * If no pair of bits is found that way, fallback to picking some | |
1061 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1062 | * 'source' and 'dest' bits are the same, this represents a node | |
1063 | * that will be migrating to itself, so no pages need move. | |
1064 | * | |
1065 | * If no bits are left in 'tmp', or if all remaining bits left | |
1066 | * in 'tmp' correspond to the same bit in 'to', return false | |
1067 | * (nothing left to migrate). | |
1068 | * | |
1069 | * This lets us pick a pair of nodes to migrate between, such that | |
1070 | * if possible the dest node is not already occupied by some other | |
1071 | * source node, minimizing the risk of overloading the memory on a | |
1072 | * node that would happen if we migrated incoming memory to a node | |
1073 | * before migrating outgoing memory source that same node. | |
1074 | * | |
1075 | * A single scan of tmp is sufficient. As we go, we remember the | |
1076 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1077 | * that not only moved, but what's better, moved to an empty slot | |
1078 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1079 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1080 | * most recent <s, d> pair that moved. If we get all the way through |
1081 | * the scan of tmp without finding any node that moved, much less | |
1082 | * moved to an empty node, then there is nothing left worth migrating. | |
1083 | */ | |
d4984711 | 1084 | |
0ce72d4f | 1085 | tmp = *from; |
7e2ab150 CL |
1086 | while (!nodes_empty(tmp)) { |
1087 | int s,d; | |
b76ac7e7 | 1088 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1089 | int dest = 0; |
1090 | ||
1091 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1092 | |
1093 | /* | |
1094 | * do_migrate_pages() tries to maintain the relative | |
1095 | * node relationship of the pages established between | |
1096 | * threads and memory areas. | |
1097 | * | |
1098 | * However if the number of source nodes is not equal to | |
1099 | * the number of destination nodes we can not preserve | |
1100 | * this node relative relationship. In that case, skip | |
1101 | * copying memory from a node that is in the destination | |
1102 | * mask. | |
1103 | * | |
1104 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1105 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1106 | */ | |
1107 | ||
0ce72d4f AM |
1108 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1109 | (node_isset(s, *to))) | |
4a5b18cc LW |
1110 | continue; |
1111 | ||
0ce72d4f | 1112 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1113 | if (s == d) |
1114 | continue; | |
1115 | ||
1116 | source = s; /* Node moved. Memorize */ | |
1117 | dest = d; | |
1118 | ||
1119 | /* dest not in remaining from nodes? */ | |
1120 | if (!node_isset(dest, tmp)) | |
1121 | break; | |
1122 | } | |
b76ac7e7 | 1123 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1124 | break; |
1125 | ||
1126 | node_clear(source, tmp); | |
1127 | err = migrate_to_node(mm, source, dest, flags); | |
1128 | if (err > 0) | |
1129 | busy += err; | |
1130 | if (err < 0) | |
1131 | break; | |
39743889 | 1132 | } |
7b2259b3 | 1133 | out: |
39743889 | 1134 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1135 | if (err < 0) |
1136 | return err; | |
1137 | return busy; | |
b20a3503 CL |
1138 | |
1139 | } | |
1140 | ||
3ad33b24 LS |
1141 | /* |
1142 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1143 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1144 | * Search forward from there, if not. N.B., this assumes that the |
1145 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1146 | * is in virtual address order. | |
1147 | */ | |
d05f0cdc | 1148 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 | 1149 | { |
d05f0cdc | 1150 | struct vm_area_struct *vma; |
3ad33b24 | 1151 | unsigned long uninitialized_var(address); |
95a402c3 | 1152 | |
d05f0cdc | 1153 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1154 | while (vma) { |
1155 | address = page_address_in_vma(page, vma); | |
1156 | if (address != -EFAULT) | |
1157 | break; | |
1158 | vma = vma->vm_next; | |
1159 | } | |
11c731e8 WL |
1160 | |
1161 | if (PageHuge(page)) { | |
cc81717e MH |
1162 | BUG_ON(!vma); |
1163 | return alloc_huge_page_noerr(vma, address, 1); | |
11c731e8 | 1164 | } |
0bf598d8 | 1165 | /* |
11c731e8 | 1166 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1167 | */ |
3ad33b24 | 1168 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
95a402c3 | 1169 | } |
b20a3503 CL |
1170 | #else |
1171 | ||
1172 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1173 | unsigned long flags) | |
1174 | { | |
39743889 CL |
1175 | } |
1176 | ||
0ce72d4f AM |
1177 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1178 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1179 | { |
1180 | return -ENOSYS; | |
1181 | } | |
95a402c3 | 1182 | |
d05f0cdc | 1183 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 CL |
1184 | { |
1185 | return NULL; | |
1186 | } | |
b20a3503 CL |
1187 | #endif |
1188 | ||
dbcb0f19 | 1189 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1190 | unsigned short mode, unsigned short mode_flags, |
1191 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1192 | { |
6ce3c4c0 CL |
1193 | struct mm_struct *mm = current->mm; |
1194 | struct mempolicy *new; | |
1195 | unsigned long end; | |
1196 | int err; | |
1197 | LIST_HEAD(pagelist); | |
1198 | ||
b24f53a0 | 1199 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1200 | return -EINVAL; |
74c00241 | 1201 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1202 | return -EPERM; |
1203 | ||
1204 | if (start & ~PAGE_MASK) | |
1205 | return -EINVAL; | |
1206 | ||
1207 | if (mode == MPOL_DEFAULT) | |
1208 | flags &= ~MPOL_MF_STRICT; | |
1209 | ||
1210 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1211 | end = start + len; | |
1212 | ||
1213 | if (end < start) | |
1214 | return -EINVAL; | |
1215 | if (end == start) | |
1216 | return 0; | |
1217 | ||
028fec41 | 1218 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1219 | if (IS_ERR(new)) |
1220 | return PTR_ERR(new); | |
1221 | ||
b24f53a0 LS |
1222 | if (flags & MPOL_MF_LAZY) |
1223 | new->flags |= MPOL_F_MOF; | |
1224 | ||
6ce3c4c0 CL |
1225 | /* |
1226 | * If we are using the default policy then operation | |
1227 | * on discontinuous address spaces is okay after all | |
1228 | */ | |
1229 | if (!new) | |
1230 | flags |= MPOL_MF_DISCONTIG_OK; | |
1231 | ||
028fec41 DR |
1232 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1233 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1234 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1235 | |
0aedadf9 CL |
1236 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1237 | ||
1238 | err = migrate_prep(); | |
1239 | if (err) | |
b05ca738 | 1240 | goto mpol_out; |
0aedadf9 | 1241 | } |
4bfc4495 KH |
1242 | { |
1243 | NODEMASK_SCRATCH(scratch); | |
1244 | if (scratch) { | |
1245 | down_write(&mm->mmap_sem); | |
1246 | task_lock(current); | |
1247 | err = mpol_set_nodemask(new, nmask, scratch); | |
1248 | task_unlock(current); | |
1249 | if (err) | |
1250 | up_write(&mm->mmap_sem); | |
1251 | } else | |
1252 | err = -ENOMEM; | |
1253 | NODEMASK_SCRATCH_FREE(scratch); | |
1254 | } | |
b05ca738 KM |
1255 | if (err) |
1256 | goto mpol_out; | |
1257 | ||
d05f0cdc | 1258 | err = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1259 | flags | MPOL_MF_INVERT, &pagelist); |
d05f0cdc | 1260 | if (!err) |
9d8cebd4 | 1261 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1262 | |
b24f53a0 LS |
1263 | if (!err) { |
1264 | int nr_failed = 0; | |
1265 | ||
cf608ac1 | 1266 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1267 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1268 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1269 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1270 | if (nr_failed) |
74060e4d | 1271 | putback_movable_pages(&pagelist); |
cf608ac1 | 1272 | } |
6ce3c4c0 | 1273 | |
b24f53a0 | 1274 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1275 | err = -EIO; |
ab8a3e14 | 1276 | } else |
b0e5fd73 | 1277 | putback_movable_pages(&pagelist); |
b20a3503 | 1278 | |
6ce3c4c0 | 1279 | up_write(&mm->mmap_sem); |
b05ca738 | 1280 | mpol_out: |
f0be3d32 | 1281 | mpol_put(new); |
6ce3c4c0 CL |
1282 | return err; |
1283 | } | |
1284 | ||
8bccd85f CL |
1285 | /* |
1286 | * User space interface with variable sized bitmaps for nodelists. | |
1287 | */ | |
1288 | ||
1289 | /* Copy a node mask from user space. */ | |
39743889 | 1290 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1291 | unsigned long maxnode) |
1292 | { | |
1293 | unsigned long k; | |
1294 | unsigned long nlongs; | |
1295 | unsigned long endmask; | |
1296 | ||
1297 | --maxnode; | |
1298 | nodes_clear(*nodes); | |
1299 | if (maxnode == 0 || !nmask) | |
1300 | return 0; | |
a9c930ba | 1301 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1302 | return -EINVAL; |
8bccd85f CL |
1303 | |
1304 | nlongs = BITS_TO_LONGS(maxnode); | |
1305 | if ((maxnode % BITS_PER_LONG) == 0) | |
1306 | endmask = ~0UL; | |
1307 | else | |
1308 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1309 | ||
1310 | /* When the user specified more nodes than supported just check | |
1311 | if the non supported part is all zero. */ | |
1312 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1313 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1314 | return -EINVAL; | |
1315 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1316 | unsigned long t; | |
1317 | if (get_user(t, nmask + k)) | |
1318 | return -EFAULT; | |
1319 | if (k == nlongs - 1) { | |
1320 | if (t & endmask) | |
1321 | return -EINVAL; | |
1322 | } else if (t) | |
1323 | return -EINVAL; | |
1324 | } | |
1325 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1326 | endmask = ~0UL; | |
1327 | } | |
1328 | ||
1329 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1330 | return -EFAULT; | |
1331 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1332 | return 0; | |
1333 | } | |
1334 | ||
1335 | /* Copy a kernel node mask to user space */ | |
1336 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1337 | nodemask_t *nodes) | |
1338 | { | |
1339 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1340 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1341 | ||
1342 | if (copy > nbytes) { | |
1343 | if (copy > PAGE_SIZE) | |
1344 | return -EINVAL; | |
1345 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1346 | return -EFAULT; | |
1347 | copy = nbytes; | |
1348 | } | |
1349 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1350 | } | |
1351 | ||
938bb9f5 | 1352 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
f7f28ca9 | 1353 | unsigned long, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1354 | unsigned long, maxnode, unsigned, flags) |
8bccd85f CL |
1355 | { |
1356 | nodemask_t nodes; | |
1357 | int err; | |
028fec41 | 1358 | unsigned short mode_flags; |
8bccd85f | 1359 | |
028fec41 DR |
1360 | mode_flags = mode & MPOL_MODE_FLAGS; |
1361 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1362 | if (mode >= MPOL_MAX) |
1363 | return -EINVAL; | |
4c50bc01 DR |
1364 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1365 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1366 | return -EINVAL; | |
8bccd85f CL |
1367 | err = get_nodes(&nodes, nmask, maxnode); |
1368 | if (err) | |
1369 | return err; | |
028fec41 | 1370 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1371 | } |
1372 | ||
1373 | /* Set the process memory policy */ | |
23c8902d | 1374 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1375 | unsigned long, maxnode) |
8bccd85f CL |
1376 | { |
1377 | int err; | |
1378 | nodemask_t nodes; | |
028fec41 | 1379 | unsigned short flags; |
8bccd85f | 1380 | |
028fec41 DR |
1381 | flags = mode & MPOL_MODE_FLAGS; |
1382 | mode &= ~MPOL_MODE_FLAGS; | |
1383 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1384 | return -EINVAL; |
4c50bc01 DR |
1385 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1386 | return -EINVAL; | |
8bccd85f CL |
1387 | err = get_nodes(&nodes, nmask, maxnode); |
1388 | if (err) | |
1389 | return err; | |
028fec41 | 1390 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1391 | } |
1392 | ||
938bb9f5 HC |
1393 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1394 | const unsigned long __user *, old_nodes, | |
1395 | const unsigned long __user *, new_nodes) | |
39743889 | 1396 | { |
c69e8d9c | 1397 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1398 | struct mm_struct *mm = NULL; |
39743889 | 1399 | struct task_struct *task; |
39743889 CL |
1400 | nodemask_t task_nodes; |
1401 | int err; | |
596d7cfa KM |
1402 | nodemask_t *old; |
1403 | nodemask_t *new; | |
1404 | NODEMASK_SCRATCH(scratch); | |
1405 | ||
1406 | if (!scratch) | |
1407 | return -ENOMEM; | |
39743889 | 1408 | |
596d7cfa KM |
1409 | old = &scratch->mask1; |
1410 | new = &scratch->mask2; | |
1411 | ||
1412 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1413 | if (err) |
596d7cfa | 1414 | goto out; |
39743889 | 1415 | |
596d7cfa | 1416 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1417 | if (err) |
596d7cfa | 1418 | goto out; |
39743889 CL |
1419 | |
1420 | /* Find the mm_struct */ | |
55cfaa3c | 1421 | rcu_read_lock(); |
228ebcbe | 1422 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1423 | if (!task) { |
55cfaa3c | 1424 | rcu_read_unlock(); |
596d7cfa KM |
1425 | err = -ESRCH; |
1426 | goto out; | |
39743889 | 1427 | } |
3268c63e | 1428 | get_task_struct(task); |
39743889 | 1429 | |
596d7cfa | 1430 | err = -EINVAL; |
39743889 CL |
1431 | |
1432 | /* | |
1433 | * Check if this process has the right to modify the specified | |
1434 | * process. The right exists if the process has administrative | |
7f927fcc | 1435 | * capabilities, superuser privileges or the same |
39743889 CL |
1436 | * userid as the target process. |
1437 | */ | |
c69e8d9c | 1438 | tcred = __task_cred(task); |
b38a86eb EB |
1439 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1440 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1441 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1442 | rcu_read_unlock(); |
39743889 | 1443 | err = -EPERM; |
3268c63e | 1444 | goto out_put; |
39743889 | 1445 | } |
c69e8d9c | 1446 | rcu_read_unlock(); |
39743889 CL |
1447 | |
1448 | task_nodes = cpuset_mems_allowed(task); | |
1449 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1450 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1451 | err = -EPERM; |
3268c63e | 1452 | goto out_put; |
39743889 CL |
1453 | } |
1454 | ||
01f13bd6 | 1455 | if (!nodes_subset(*new, node_states[N_MEMORY])) { |
3b42d28b | 1456 | err = -EINVAL; |
3268c63e | 1457 | goto out_put; |
3b42d28b CL |
1458 | } |
1459 | ||
86c3a764 DQ |
1460 | err = security_task_movememory(task); |
1461 | if (err) | |
3268c63e | 1462 | goto out_put; |
86c3a764 | 1463 | |
3268c63e CL |
1464 | mm = get_task_mm(task); |
1465 | put_task_struct(task); | |
f2a9ef88 SL |
1466 | |
1467 | if (!mm) { | |
3268c63e | 1468 | err = -EINVAL; |
f2a9ef88 SL |
1469 | goto out; |
1470 | } | |
1471 | ||
1472 | err = do_migrate_pages(mm, old, new, | |
1473 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1474 | |
1475 | mmput(mm); | |
1476 | out: | |
596d7cfa KM |
1477 | NODEMASK_SCRATCH_FREE(scratch); |
1478 | ||
39743889 | 1479 | return err; |
3268c63e CL |
1480 | |
1481 | out_put: | |
1482 | put_task_struct(task); | |
1483 | goto out; | |
1484 | ||
39743889 CL |
1485 | } |
1486 | ||
1487 | ||
8bccd85f | 1488 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1489 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1490 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1491 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1492 | { |
dbcb0f19 AB |
1493 | int err; |
1494 | int uninitialized_var(pval); | |
8bccd85f CL |
1495 | nodemask_t nodes; |
1496 | ||
1497 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1498 | return -EINVAL; | |
1499 | ||
1500 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1501 | ||
1502 | if (err) | |
1503 | return err; | |
1504 | ||
1505 | if (policy && put_user(pval, policy)) | |
1506 | return -EFAULT; | |
1507 | ||
1508 | if (nmask) | |
1509 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1510 | ||
1511 | return err; | |
1512 | } | |
1513 | ||
1da177e4 LT |
1514 | #ifdef CONFIG_COMPAT |
1515 | ||
c93e0f6c HC |
1516 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1517 | compat_ulong_t __user *, nmask, | |
1518 | compat_ulong_t, maxnode, | |
1519 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1520 | { |
1521 | long err; | |
1522 | unsigned long __user *nm = NULL; | |
1523 | unsigned long nr_bits, alloc_size; | |
1524 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1525 | ||
1526 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1527 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1528 | ||
1529 | if (nmask) | |
1530 | nm = compat_alloc_user_space(alloc_size); | |
1531 | ||
1532 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1533 | ||
1534 | if (!err && nmask) { | |
2bbff6c7 KH |
1535 | unsigned long copy_size; |
1536 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1537 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1538 | /* ensure entire bitmap is zeroed */ |
1539 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1540 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1541 | } | |
1542 | ||
1543 | return err; | |
1544 | } | |
1545 | ||
c93e0f6c HC |
1546 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1547 | compat_ulong_t, maxnode) | |
1da177e4 LT |
1548 | { |
1549 | long err = 0; | |
1550 | unsigned long __user *nm = NULL; | |
1551 | unsigned long nr_bits, alloc_size; | |
1552 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1553 | ||
1554 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1555 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1556 | ||
1557 | if (nmask) { | |
1558 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1559 | nm = compat_alloc_user_space(alloc_size); | |
1560 | err |= copy_to_user(nm, bm, alloc_size); | |
1561 | } | |
1562 | ||
1563 | if (err) | |
1564 | return -EFAULT; | |
1565 | ||
1566 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1567 | } | |
1568 | ||
c93e0f6c HC |
1569 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1570 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1571 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 LT |
1572 | { |
1573 | long err = 0; | |
1574 | unsigned long __user *nm = NULL; | |
1575 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1576 | nodemask_t bm; |
1da177e4 LT |
1577 | |
1578 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1579 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1580 | ||
1581 | if (nmask) { | |
dfcd3c0d | 1582 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1583 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1584 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1585 | } |
1586 | ||
1587 | if (err) | |
1588 | return -EFAULT; | |
1589 | ||
1590 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1591 | } | |
1592 | ||
1593 | #endif | |
1594 | ||
74d2c3a0 ON |
1595 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1596 | unsigned long addr) | |
1da177e4 | 1597 | { |
8d90274b | 1598 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1599 | |
1600 | if (vma) { | |
480eccf9 | 1601 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1602 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1603 | } else if (vma->vm_policy) { |
1da177e4 | 1604 | pol = vma->vm_policy; |
00442ad0 MG |
1605 | |
1606 | /* | |
1607 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1608 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1609 | * count on these policies which will be dropped by | |
1610 | * mpol_cond_put() later | |
1611 | */ | |
1612 | if (mpol_needs_cond_ref(pol)) | |
1613 | mpol_get(pol); | |
1614 | } | |
1da177e4 | 1615 | } |
f15ca78e | 1616 | |
74d2c3a0 ON |
1617 | return pol; |
1618 | } | |
1619 | ||
1620 | /* | |
dd6eecb9 | 1621 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1622 | * @vma: virtual memory area whose policy is sought |
1623 | * @addr: address in @vma for shared policy lookup | |
1624 | * | |
1625 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1626 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1627 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1628 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1629 | * freeing by another task. It is the caller's responsibility to free the | |
1630 | * extra reference for shared policies. | |
1631 | */ | |
dd6eecb9 ON |
1632 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
1633 | unsigned long addr) | |
74d2c3a0 ON |
1634 | { |
1635 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1636 | ||
8d90274b | 1637 | if (!pol) |
dd6eecb9 | 1638 | pol = get_task_policy(current); |
8d90274b | 1639 | |
1da177e4 LT |
1640 | return pol; |
1641 | } | |
1642 | ||
6b6482bb | 1643 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1644 | { |
6b6482bb | 1645 | struct mempolicy *pol; |
fc314724 | 1646 | |
6b6482bb ON |
1647 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1648 | bool ret = false; | |
fc314724 | 1649 | |
6b6482bb ON |
1650 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1651 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1652 | ret = true; | |
1653 | mpol_cond_put(pol); | |
8d90274b | 1654 | |
6b6482bb | 1655 | return ret; |
fc314724 MG |
1656 | } |
1657 | ||
6b6482bb | 1658 | pol = vma->vm_policy; |
8d90274b | 1659 | if (!pol) |
6b6482bb | 1660 | pol = get_task_policy(current); |
8d90274b | 1661 | |
fc314724 MG |
1662 | return pol->flags & MPOL_F_MOF; |
1663 | } | |
1664 | ||
d3eb1570 LJ |
1665 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1666 | { | |
1667 | enum zone_type dynamic_policy_zone = policy_zone; | |
1668 | ||
1669 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1670 | ||
1671 | /* | |
1672 | * if policy->v.nodes has movable memory only, | |
1673 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1674 | * | |
1675 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1676 | * so if the following test faile, it implies | |
1677 | * policy->v.nodes has movable memory only. | |
1678 | */ | |
1679 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1680 | dynamic_policy_zone = ZONE_MOVABLE; | |
1681 | ||
1682 | return zone >= dynamic_policy_zone; | |
1683 | } | |
1684 | ||
52cd3b07 LS |
1685 | /* |
1686 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1687 | * page allocation | |
1688 | */ | |
1689 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1690 | { |
1691 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1692 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1693 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1694 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1695 | return &policy->v.nodes; | |
1696 | ||
1697 | return NULL; | |
1698 | } | |
1699 | ||
52cd3b07 | 1700 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1701 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1702 | int nd) | |
1da177e4 | 1703 | { |
45c4745a | 1704 | switch (policy->mode) { |
1da177e4 | 1705 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1706 | if (!(policy->flags & MPOL_F_LOCAL)) |
1707 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1708 | break; |
1709 | case MPOL_BIND: | |
19770b32 | 1710 | /* |
52cd3b07 LS |
1711 | * Normally, MPOL_BIND allocations are node-local within the |
1712 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1713 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1714 | * the first node in the mask instead. |
19770b32 | 1715 | */ |
19770b32 MG |
1716 | if (unlikely(gfp & __GFP_THISNODE) && |
1717 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1718 | nd = first_node(policy->v.nodes); | |
1719 | break; | |
1da177e4 | 1720 | default: |
1da177e4 LT |
1721 | BUG(); |
1722 | } | |
0e88460d | 1723 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1724 | } |
1725 | ||
1726 | /* Do dynamic interleaving for a process */ | |
1727 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1728 | { | |
1729 | unsigned nid, next; | |
1730 | struct task_struct *me = current; | |
1731 | ||
1732 | nid = me->il_next; | |
dfcd3c0d | 1733 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1734 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1735 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1736 | if (next < MAX_NUMNODES) |
1737 | me->il_next = next; | |
1da177e4 LT |
1738 | return nid; |
1739 | } | |
1740 | ||
dc85da15 CL |
1741 | /* |
1742 | * Depending on the memory policy provide a node from which to allocate the | |
1743 | * next slab entry. | |
1744 | */ | |
2a389610 | 1745 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1746 | { |
e7b691b0 | 1747 | struct mempolicy *policy; |
2a389610 | 1748 | int node = numa_mem_id(); |
e7b691b0 AK |
1749 | |
1750 | if (in_interrupt()) | |
2a389610 | 1751 | return node; |
e7b691b0 AK |
1752 | |
1753 | policy = current->mempolicy; | |
fc36b8d3 | 1754 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1755 | return node; |
bea904d5 LS |
1756 | |
1757 | switch (policy->mode) { | |
1758 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1759 | /* |
1760 | * handled MPOL_F_LOCAL above | |
1761 | */ | |
1762 | return policy->v.preferred_node; | |
765c4507 | 1763 | |
dc85da15 CL |
1764 | case MPOL_INTERLEAVE: |
1765 | return interleave_nodes(policy); | |
1766 | ||
dd1a239f | 1767 | case MPOL_BIND: { |
dc85da15 CL |
1768 | /* |
1769 | * Follow bind policy behavior and start allocation at the | |
1770 | * first node. | |
1771 | */ | |
19770b32 MG |
1772 | struct zonelist *zonelist; |
1773 | struct zone *zone; | |
1774 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
2a389610 | 1775 | zonelist = &NODE_DATA(node)->node_zonelists[0]; |
19770b32 MG |
1776 | (void)first_zones_zonelist(zonelist, highest_zoneidx, |
1777 | &policy->v.nodes, | |
1778 | &zone); | |
2a389610 | 1779 | return zone ? zone->node : node; |
dd1a239f | 1780 | } |
dc85da15 | 1781 | |
dc85da15 | 1782 | default: |
bea904d5 | 1783 | BUG(); |
dc85da15 CL |
1784 | } |
1785 | } | |
1786 | ||
1da177e4 LT |
1787 | /* Do static interleaving for a VMA with known offset. */ |
1788 | static unsigned offset_il_node(struct mempolicy *pol, | |
1789 | struct vm_area_struct *vma, unsigned long off) | |
1790 | { | |
dfcd3c0d | 1791 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1792 | unsigned target; |
1da177e4 | 1793 | int c; |
b76ac7e7 | 1794 | int nid = NUMA_NO_NODE; |
1da177e4 | 1795 | |
f5b087b5 DR |
1796 | if (!nnodes) |
1797 | return numa_node_id(); | |
1798 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1799 | c = 0; |
1800 | do { | |
dfcd3c0d | 1801 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1802 | c++; |
1803 | } while (c <= target); | |
1da177e4 LT |
1804 | return nid; |
1805 | } | |
1806 | ||
5da7ca86 CL |
1807 | /* Determine a node number for interleave */ |
1808 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1809 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1810 | { | |
1811 | if (vma) { | |
1812 | unsigned long off; | |
1813 | ||
3b98b087 NA |
1814 | /* |
1815 | * for small pages, there is no difference between | |
1816 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1817 | * for huge pages, since vm_pgoff is in units of small | |
1818 | * pages, we need to shift off the always 0 bits to get | |
1819 | * a useful offset. | |
1820 | */ | |
1821 | BUG_ON(shift < PAGE_SHIFT); | |
1822 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1823 | off += (addr - vma->vm_start) >> shift; |
1824 | return offset_il_node(pol, vma, off); | |
1825 | } else | |
1826 | return interleave_nodes(pol); | |
1827 | } | |
1828 | ||
778d3b0f MH |
1829 | /* |
1830 | * Return the bit number of a random bit set in the nodemask. | |
b76ac7e7 | 1831 | * (returns NUMA_NO_NODE if nodemask is empty) |
778d3b0f MH |
1832 | */ |
1833 | int node_random(const nodemask_t *maskp) | |
1834 | { | |
b76ac7e7 | 1835 | int w, bit = NUMA_NO_NODE; |
778d3b0f MH |
1836 | |
1837 | w = nodes_weight(*maskp); | |
1838 | if (w) | |
1839 | bit = bitmap_ord_to_pos(maskp->bits, | |
1840 | get_random_int() % w, MAX_NUMNODES); | |
1841 | return bit; | |
1842 | } | |
1843 | ||
00ac59ad | 1844 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1845 | /* |
1846 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
b46e14ac FF |
1847 | * @vma: virtual memory area whose policy is sought |
1848 | * @addr: address in @vma for shared policy lookup and interleave policy | |
1849 | * @gfp_flags: for requested zone | |
1850 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
1851 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1852 | * |
52cd3b07 LS |
1853 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1854 | * to the struct mempolicy for conditional unref after allocation. | |
1855 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1856 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 1857 | * |
d26914d1 | 1858 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 1859 | */ |
396faf03 | 1860 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1861 | gfp_t gfp_flags, struct mempolicy **mpol, |
1862 | nodemask_t **nodemask) | |
5da7ca86 | 1863 | { |
480eccf9 | 1864 | struct zonelist *zl; |
5da7ca86 | 1865 | |
dd6eecb9 | 1866 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 1867 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1868 | |
52cd3b07 LS |
1869 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1870 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1871 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1872 | } else { |
2f5f9486 | 1873 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1874 | if ((*mpol)->mode == MPOL_BIND) |
1875 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1876 | } |
1877 | return zl; | |
5da7ca86 | 1878 | } |
06808b08 LS |
1879 | |
1880 | /* | |
1881 | * init_nodemask_of_mempolicy | |
1882 | * | |
1883 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1884 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1885 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1886 | * initialize the argument nodemask to contain the single node for | |
1887 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1888 | * of non-default mempolicy. | |
1889 | * | |
1890 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1891 | * because the current task is examining it's own mempolicy and a task's | |
1892 | * mempolicy is only ever changed by the task itself. | |
1893 | * | |
1894 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1895 | */ | |
1896 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1897 | { | |
1898 | struct mempolicy *mempolicy; | |
1899 | int nid; | |
1900 | ||
1901 | if (!(mask && current->mempolicy)) | |
1902 | return false; | |
1903 | ||
c0ff7453 | 1904 | task_lock(current); |
06808b08 LS |
1905 | mempolicy = current->mempolicy; |
1906 | switch (mempolicy->mode) { | |
1907 | case MPOL_PREFERRED: | |
1908 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1909 | nid = numa_node_id(); | |
1910 | else | |
1911 | nid = mempolicy->v.preferred_node; | |
1912 | init_nodemask_of_node(mask, nid); | |
1913 | break; | |
1914 | ||
1915 | case MPOL_BIND: | |
1916 | /* Fall through */ | |
1917 | case MPOL_INTERLEAVE: | |
1918 | *mask = mempolicy->v.nodes; | |
1919 | break; | |
1920 | ||
1921 | default: | |
1922 | BUG(); | |
1923 | } | |
c0ff7453 | 1924 | task_unlock(current); |
06808b08 LS |
1925 | |
1926 | return true; | |
1927 | } | |
00ac59ad | 1928 | #endif |
5da7ca86 | 1929 | |
6f48d0eb DR |
1930 | /* |
1931 | * mempolicy_nodemask_intersects | |
1932 | * | |
1933 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1934 | * policy. Otherwise, check for intersection between mask and the policy | |
1935 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1936 | * policy, always return true since it may allocate elsewhere on fallback. | |
1937 | * | |
1938 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1939 | */ | |
1940 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1941 | const nodemask_t *mask) | |
1942 | { | |
1943 | struct mempolicy *mempolicy; | |
1944 | bool ret = true; | |
1945 | ||
1946 | if (!mask) | |
1947 | return ret; | |
1948 | task_lock(tsk); | |
1949 | mempolicy = tsk->mempolicy; | |
1950 | if (!mempolicy) | |
1951 | goto out; | |
1952 | ||
1953 | switch (mempolicy->mode) { | |
1954 | case MPOL_PREFERRED: | |
1955 | /* | |
1956 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1957 | * allocate from, they may fallback to other nodes when oom. | |
1958 | * Thus, it's possible for tsk to have allocated memory from | |
1959 | * nodes in mask. | |
1960 | */ | |
1961 | break; | |
1962 | case MPOL_BIND: | |
1963 | case MPOL_INTERLEAVE: | |
1964 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1965 | break; | |
1966 | default: | |
1967 | BUG(); | |
1968 | } | |
1969 | out: | |
1970 | task_unlock(tsk); | |
1971 | return ret; | |
1972 | } | |
1973 | ||
1da177e4 LT |
1974 | /* Allocate a page in interleaved policy. |
1975 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1976 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1977 | unsigned nid) | |
1da177e4 LT |
1978 | { |
1979 | struct zonelist *zl; | |
1980 | struct page *page; | |
1981 | ||
0e88460d | 1982 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1983 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1984 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1985 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1986 | return page; |
1987 | } | |
1988 | ||
1989 | /** | |
0bbbc0b3 | 1990 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1991 | * |
1992 | * @gfp: | |
1993 | * %GFP_USER user allocation. | |
1994 | * %GFP_KERNEL kernel allocations, | |
1995 | * %GFP_HIGHMEM highmem/user allocations, | |
1996 | * %GFP_FS allocation should not call back into a file system. | |
1997 | * %GFP_ATOMIC don't sleep. | |
1998 | * | |
0bbbc0b3 | 1999 | * @order:Order of the GFP allocation. |
1da177e4 LT |
2000 | * @vma: Pointer to VMA or NULL if not available. |
2001 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
2002 | * | |
2003 | * This function allocates a page from the kernel page pool and applies | |
2004 | * a NUMA policy associated with the VMA or the current process. | |
2005 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
2006 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
2007 | * all allocations for pages that will be mapped into | |
2008 | * user space. Returns NULL when no page can be allocated. | |
2009 | * | |
2010 | * Should be called with the mm_sem of the vma hold. | |
2011 | */ | |
2012 | struct page * | |
0bbbc0b3 | 2013 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 2014 | unsigned long addr, int node) |
1da177e4 | 2015 | { |
cc9a6c87 | 2016 | struct mempolicy *pol; |
c0ff7453 | 2017 | struct page *page; |
cc9a6c87 MG |
2018 | unsigned int cpuset_mems_cookie; |
2019 | ||
2020 | retry_cpuset: | |
dd6eecb9 | 2021 | pol = get_vma_policy(vma, addr); |
d26914d1 | 2022 | cpuset_mems_cookie = read_mems_allowed_begin(); |
1da177e4 | 2023 | |
45c4745a | 2024 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 2025 | unsigned nid; |
5da7ca86 | 2026 | |
8eac563c | 2027 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 2028 | mpol_cond_put(pol); |
0bbbc0b3 | 2029 | page = alloc_page_interleave(gfp, order, nid); |
d26914d1 | 2030 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2031 | goto retry_cpuset; |
2032 | ||
c0ff7453 | 2033 | return page; |
1da177e4 | 2034 | } |
212a0a6f DR |
2035 | page = __alloc_pages_nodemask(gfp, order, |
2036 | policy_zonelist(gfp, pol, node), | |
0bbbc0b3 | 2037 | policy_nodemask(gfp, pol)); |
2386740d | 2038 | mpol_cond_put(pol); |
d26914d1 | 2039 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 2040 | goto retry_cpuset; |
c0ff7453 | 2041 | return page; |
1da177e4 LT |
2042 | } |
2043 | ||
2044 | /** | |
2045 | * alloc_pages_current - Allocate pages. | |
2046 | * | |
2047 | * @gfp: | |
2048 | * %GFP_USER user allocation, | |
2049 | * %GFP_KERNEL kernel allocation, | |
2050 | * %GFP_HIGHMEM highmem allocation, | |
2051 | * %GFP_FS don't call back into a file system. | |
2052 | * %GFP_ATOMIC don't sleep. | |
2053 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2054 | * | |
2055 | * Allocate a page from the kernel page pool. When not in | |
2056 | * interrupt context and apply the current process NUMA policy. | |
2057 | * Returns NULL when no page can be allocated. | |
2058 | * | |
cf2a473c | 2059 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
2060 | * 1) it's ok to take cpuset_sem (can WAIT), and |
2061 | * 2) allocating for current task (not interrupt). | |
2062 | */ | |
dd0fc66f | 2063 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2064 | { |
8d90274b | 2065 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2066 | struct page *page; |
cc9a6c87 | 2067 | unsigned int cpuset_mems_cookie; |
1da177e4 | 2068 | |
8d90274b ON |
2069 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2070 | pol = get_task_policy(current); | |
52cd3b07 | 2071 | |
cc9a6c87 | 2072 | retry_cpuset: |
d26914d1 | 2073 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2074 | |
52cd3b07 LS |
2075 | /* |
2076 | * No reference counting needed for current->mempolicy | |
2077 | * nor system default_policy | |
2078 | */ | |
45c4745a | 2079 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2080 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2081 | else | |
2082 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
2083 | policy_zonelist(gfp, pol, numa_node_id()), |
2084 | policy_nodemask(gfp, pol)); | |
cc9a6c87 | 2085 | |
d26914d1 | 2086 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2087 | goto retry_cpuset; |
2088 | ||
c0ff7453 | 2089 | return page; |
1da177e4 LT |
2090 | } |
2091 | EXPORT_SYMBOL(alloc_pages_current); | |
2092 | ||
ef0855d3 ON |
2093 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2094 | { | |
2095 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2096 | ||
2097 | if (IS_ERR(pol)) | |
2098 | return PTR_ERR(pol); | |
2099 | dst->vm_policy = pol; | |
2100 | return 0; | |
2101 | } | |
2102 | ||
4225399a | 2103 | /* |
846a16bf | 2104 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2105 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2106 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2107 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2108 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2109 | * |
2110 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2111 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2112 | */ |
4225399a | 2113 | |
846a16bf LS |
2114 | /* Slow path of a mempolicy duplicate */ |
2115 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2116 | { |
2117 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2118 | ||
2119 | if (!new) | |
2120 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2121 | |
2122 | /* task's mempolicy is protected by alloc_lock */ | |
2123 | if (old == current->mempolicy) { | |
2124 | task_lock(current); | |
2125 | *new = *old; | |
2126 | task_unlock(current); | |
2127 | } else | |
2128 | *new = *old; | |
2129 | ||
4225399a PJ |
2130 | if (current_cpuset_is_being_rebound()) { |
2131 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2132 | if (new->flags & MPOL_F_REBINDING) |
2133 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2134 | else | |
2135 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2136 | } |
1da177e4 | 2137 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2138 | return new; |
2139 | } | |
2140 | ||
2141 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2142 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2143 | { |
2144 | if (!a || !b) | |
fcfb4dcc | 2145 | return false; |
45c4745a | 2146 | if (a->mode != b->mode) |
fcfb4dcc | 2147 | return false; |
19800502 | 2148 | if (a->flags != b->flags) |
fcfb4dcc | 2149 | return false; |
19800502 BL |
2150 | if (mpol_store_user_nodemask(a)) |
2151 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2152 | return false; |
19800502 | 2153 | |
45c4745a | 2154 | switch (a->mode) { |
19770b32 MG |
2155 | case MPOL_BIND: |
2156 | /* Fall through */ | |
1da177e4 | 2157 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2158 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2159 | case MPOL_PREFERRED: |
75719661 | 2160 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2161 | default: |
2162 | BUG(); | |
fcfb4dcc | 2163 | return false; |
1da177e4 LT |
2164 | } |
2165 | } | |
2166 | ||
1da177e4 LT |
2167 | /* |
2168 | * Shared memory backing store policy support. | |
2169 | * | |
2170 | * Remember policies even when nobody has shared memory mapped. | |
2171 | * The policies are kept in Red-Black tree linked from the inode. | |
2172 | * They are protected by the sp->lock spinlock, which should be held | |
2173 | * for any accesses to the tree. | |
2174 | */ | |
2175 | ||
2176 | /* lookup first element intersecting start-end */ | |
42288fe3 | 2177 | /* Caller holds sp->lock */ |
1da177e4 LT |
2178 | static struct sp_node * |
2179 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2180 | { | |
2181 | struct rb_node *n = sp->root.rb_node; | |
2182 | ||
2183 | while (n) { | |
2184 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2185 | ||
2186 | if (start >= p->end) | |
2187 | n = n->rb_right; | |
2188 | else if (end <= p->start) | |
2189 | n = n->rb_left; | |
2190 | else | |
2191 | break; | |
2192 | } | |
2193 | if (!n) | |
2194 | return NULL; | |
2195 | for (;;) { | |
2196 | struct sp_node *w = NULL; | |
2197 | struct rb_node *prev = rb_prev(n); | |
2198 | if (!prev) | |
2199 | break; | |
2200 | w = rb_entry(prev, struct sp_node, nd); | |
2201 | if (w->end <= start) | |
2202 | break; | |
2203 | n = prev; | |
2204 | } | |
2205 | return rb_entry(n, struct sp_node, nd); | |
2206 | } | |
2207 | ||
2208 | /* Insert a new shared policy into the list. */ | |
2209 | /* Caller holds sp->lock */ | |
2210 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2211 | { | |
2212 | struct rb_node **p = &sp->root.rb_node; | |
2213 | struct rb_node *parent = NULL; | |
2214 | struct sp_node *nd; | |
2215 | ||
2216 | while (*p) { | |
2217 | parent = *p; | |
2218 | nd = rb_entry(parent, struct sp_node, nd); | |
2219 | if (new->start < nd->start) | |
2220 | p = &(*p)->rb_left; | |
2221 | else if (new->end > nd->end) | |
2222 | p = &(*p)->rb_right; | |
2223 | else | |
2224 | BUG(); | |
2225 | } | |
2226 | rb_link_node(&new->nd, parent, p); | |
2227 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2228 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2229 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2230 | } |
2231 | ||
2232 | /* Find shared policy intersecting idx */ | |
2233 | struct mempolicy * | |
2234 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2235 | { | |
2236 | struct mempolicy *pol = NULL; | |
2237 | struct sp_node *sn; | |
2238 | ||
2239 | if (!sp->root.rb_node) | |
2240 | return NULL; | |
42288fe3 | 2241 | spin_lock(&sp->lock); |
1da177e4 LT |
2242 | sn = sp_lookup(sp, idx, idx+1); |
2243 | if (sn) { | |
2244 | mpol_get(sn->policy); | |
2245 | pol = sn->policy; | |
2246 | } | |
42288fe3 | 2247 | spin_unlock(&sp->lock); |
1da177e4 LT |
2248 | return pol; |
2249 | } | |
2250 | ||
63f74ca2 KM |
2251 | static void sp_free(struct sp_node *n) |
2252 | { | |
2253 | mpol_put(n->policy); | |
2254 | kmem_cache_free(sn_cache, n); | |
2255 | } | |
2256 | ||
771fb4d8 LS |
2257 | /** |
2258 | * mpol_misplaced - check whether current page node is valid in policy | |
2259 | * | |
b46e14ac FF |
2260 | * @page: page to be checked |
2261 | * @vma: vm area where page mapped | |
2262 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2263 | * |
2264 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2265 | * node id. | |
2266 | * | |
2267 | * Returns: | |
2268 | * -1 - not misplaced, page is in the right node | |
2269 | * node - node id where the page should be | |
2270 | * | |
2271 | * Policy determination "mimics" alloc_page_vma(). | |
2272 | * Called from fault path where we know the vma and faulting address. | |
2273 | */ | |
2274 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2275 | { | |
2276 | struct mempolicy *pol; | |
2277 | struct zone *zone; | |
2278 | int curnid = page_to_nid(page); | |
2279 | unsigned long pgoff; | |
90572890 PZ |
2280 | int thiscpu = raw_smp_processor_id(); |
2281 | int thisnid = cpu_to_node(thiscpu); | |
771fb4d8 LS |
2282 | int polnid = -1; |
2283 | int ret = -1; | |
2284 | ||
2285 | BUG_ON(!vma); | |
2286 | ||
dd6eecb9 | 2287 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2288 | if (!(pol->flags & MPOL_F_MOF)) |
2289 | goto out; | |
2290 | ||
2291 | switch (pol->mode) { | |
2292 | case MPOL_INTERLEAVE: | |
2293 | BUG_ON(addr >= vma->vm_end); | |
2294 | BUG_ON(addr < vma->vm_start); | |
2295 | ||
2296 | pgoff = vma->vm_pgoff; | |
2297 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2298 | polnid = offset_il_node(pol, vma, pgoff); | |
2299 | break; | |
2300 | ||
2301 | case MPOL_PREFERRED: | |
2302 | if (pol->flags & MPOL_F_LOCAL) | |
2303 | polnid = numa_node_id(); | |
2304 | else | |
2305 | polnid = pol->v.preferred_node; | |
2306 | break; | |
2307 | ||
2308 | case MPOL_BIND: | |
2309 | /* | |
2310 | * allows binding to multiple nodes. | |
2311 | * use current page if in policy nodemask, | |
2312 | * else select nearest allowed node, if any. | |
2313 | * If no allowed nodes, use current [!misplaced]. | |
2314 | */ | |
2315 | if (node_isset(curnid, pol->v.nodes)) | |
2316 | goto out; | |
2317 | (void)first_zones_zonelist( | |
2318 | node_zonelist(numa_node_id(), GFP_HIGHUSER), | |
2319 | gfp_zone(GFP_HIGHUSER), | |
2320 | &pol->v.nodes, &zone); | |
2321 | polnid = zone->node; | |
2322 | break; | |
2323 | ||
2324 | default: | |
2325 | BUG(); | |
2326 | } | |
5606e387 MG |
2327 | |
2328 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2329 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2330 | polnid = thisnid; |
5606e387 | 2331 | |
10f39042 | 2332 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2333 | goto out; |
e42c8ff2 MG |
2334 | } |
2335 | ||
771fb4d8 LS |
2336 | if (curnid != polnid) |
2337 | ret = polnid; | |
2338 | out: | |
2339 | mpol_cond_put(pol); | |
2340 | ||
2341 | return ret; | |
2342 | } | |
2343 | ||
1da177e4 LT |
2344 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2345 | { | |
140d5a49 | 2346 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2347 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2348 | sp_free(n); |
1da177e4 LT |
2349 | } |
2350 | ||
42288fe3 MG |
2351 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2352 | unsigned long end, struct mempolicy *pol) | |
2353 | { | |
2354 | node->start = start; | |
2355 | node->end = end; | |
2356 | node->policy = pol; | |
2357 | } | |
2358 | ||
dbcb0f19 AB |
2359 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2360 | struct mempolicy *pol) | |
1da177e4 | 2361 | { |
869833f2 KM |
2362 | struct sp_node *n; |
2363 | struct mempolicy *newpol; | |
1da177e4 | 2364 | |
869833f2 | 2365 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2366 | if (!n) |
2367 | return NULL; | |
869833f2 KM |
2368 | |
2369 | newpol = mpol_dup(pol); | |
2370 | if (IS_ERR(newpol)) { | |
2371 | kmem_cache_free(sn_cache, n); | |
2372 | return NULL; | |
2373 | } | |
2374 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2375 | sp_node_init(n, start, end, newpol); |
869833f2 | 2376 | |
1da177e4 LT |
2377 | return n; |
2378 | } | |
2379 | ||
2380 | /* Replace a policy range. */ | |
2381 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2382 | unsigned long end, struct sp_node *new) | |
2383 | { | |
b22d127a | 2384 | struct sp_node *n; |
42288fe3 MG |
2385 | struct sp_node *n_new = NULL; |
2386 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2387 | int ret = 0; |
1da177e4 | 2388 | |
42288fe3 MG |
2389 | restart: |
2390 | spin_lock(&sp->lock); | |
1da177e4 LT |
2391 | n = sp_lookup(sp, start, end); |
2392 | /* Take care of old policies in the same range. */ | |
2393 | while (n && n->start < end) { | |
2394 | struct rb_node *next = rb_next(&n->nd); | |
2395 | if (n->start >= start) { | |
2396 | if (n->end <= end) | |
2397 | sp_delete(sp, n); | |
2398 | else | |
2399 | n->start = end; | |
2400 | } else { | |
2401 | /* Old policy spanning whole new range. */ | |
2402 | if (n->end > end) { | |
42288fe3 MG |
2403 | if (!n_new) |
2404 | goto alloc_new; | |
2405 | ||
2406 | *mpol_new = *n->policy; | |
2407 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2408 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2409 | n->end = start; |
5ca39575 | 2410 | sp_insert(sp, n_new); |
42288fe3 MG |
2411 | n_new = NULL; |
2412 | mpol_new = NULL; | |
1da177e4 LT |
2413 | break; |
2414 | } else | |
2415 | n->end = start; | |
2416 | } | |
2417 | if (!next) | |
2418 | break; | |
2419 | n = rb_entry(next, struct sp_node, nd); | |
2420 | } | |
2421 | if (new) | |
2422 | sp_insert(sp, new); | |
42288fe3 MG |
2423 | spin_unlock(&sp->lock); |
2424 | ret = 0; | |
2425 | ||
2426 | err_out: | |
2427 | if (mpol_new) | |
2428 | mpol_put(mpol_new); | |
2429 | if (n_new) | |
2430 | kmem_cache_free(sn_cache, n_new); | |
2431 | ||
b22d127a | 2432 | return ret; |
42288fe3 MG |
2433 | |
2434 | alloc_new: | |
2435 | spin_unlock(&sp->lock); | |
2436 | ret = -ENOMEM; | |
2437 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2438 | if (!n_new) | |
2439 | goto err_out; | |
2440 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2441 | if (!mpol_new) | |
2442 | goto err_out; | |
2443 | goto restart; | |
1da177e4 LT |
2444 | } |
2445 | ||
71fe804b LS |
2446 | /** |
2447 | * mpol_shared_policy_init - initialize shared policy for inode | |
2448 | * @sp: pointer to inode shared policy | |
2449 | * @mpol: struct mempolicy to install | |
2450 | * | |
2451 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2452 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2453 | * This must be released on exit. | |
4bfc4495 | 2454 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2455 | */ |
2456 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2457 | { | |
58568d2a MX |
2458 | int ret; |
2459 | ||
71fe804b | 2460 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
42288fe3 | 2461 | spin_lock_init(&sp->lock); |
71fe804b LS |
2462 | |
2463 | if (mpol) { | |
2464 | struct vm_area_struct pvma; | |
2465 | struct mempolicy *new; | |
4bfc4495 | 2466 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2467 | |
4bfc4495 | 2468 | if (!scratch) |
5c0c1654 | 2469 | goto put_mpol; |
71fe804b LS |
2470 | /* contextualize the tmpfs mount point mempolicy */ |
2471 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2472 | if (IS_ERR(new)) |
0cae3457 | 2473 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2474 | |
2475 | task_lock(current); | |
4bfc4495 | 2476 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2477 | task_unlock(current); |
15d77835 | 2478 | if (ret) |
5c0c1654 | 2479 | goto put_new; |
71fe804b LS |
2480 | |
2481 | /* Create pseudo-vma that contains just the policy */ | |
2482 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2483 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2484 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2485 | |
5c0c1654 | 2486 | put_new: |
71fe804b | 2487 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2488 | free_scratch: |
4bfc4495 | 2489 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2490 | put_mpol: |
2491 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2492 | } |
2493 | } | |
2494 | ||
1da177e4 LT |
2495 | int mpol_set_shared_policy(struct shared_policy *info, |
2496 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2497 | { | |
2498 | int err; | |
2499 | struct sp_node *new = NULL; | |
2500 | unsigned long sz = vma_pages(vma); | |
2501 | ||
028fec41 | 2502 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2503 | vma->vm_pgoff, |
45c4745a | 2504 | sz, npol ? npol->mode : -1, |
028fec41 | 2505 | npol ? npol->flags : -1, |
00ef2d2f | 2506 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2507 | |
2508 | if (npol) { | |
2509 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2510 | if (!new) | |
2511 | return -ENOMEM; | |
2512 | } | |
2513 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2514 | if (err && new) | |
63f74ca2 | 2515 | sp_free(new); |
1da177e4 LT |
2516 | return err; |
2517 | } | |
2518 | ||
2519 | /* Free a backing policy store on inode delete. */ | |
2520 | void mpol_free_shared_policy(struct shared_policy *p) | |
2521 | { | |
2522 | struct sp_node *n; | |
2523 | struct rb_node *next; | |
2524 | ||
2525 | if (!p->root.rb_node) | |
2526 | return; | |
42288fe3 | 2527 | spin_lock(&p->lock); |
1da177e4 LT |
2528 | next = rb_first(&p->root); |
2529 | while (next) { | |
2530 | n = rb_entry(next, struct sp_node, nd); | |
2531 | next = rb_next(&n->nd); | |
63f74ca2 | 2532 | sp_delete(p, n); |
1da177e4 | 2533 | } |
42288fe3 | 2534 | spin_unlock(&p->lock); |
1da177e4 LT |
2535 | } |
2536 | ||
1a687c2e | 2537 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2538 | static int __initdata numabalancing_override; |
1a687c2e MG |
2539 | |
2540 | static void __init check_numabalancing_enable(void) | |
2541 | { | |
2542 | bool numabalancing_default = false; | |
2543 | ||
2544 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2545 | numabalancing_default = true; | |
2546 | ||
c297663c MG |
2547 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2548 | if (numabalancing_override) | |
2549 | set_numabalancing_state(numabalancing_override == 1); | |
2550 | ||
1a687c2e | 2551 | if (nr_node_ids > 1 && !numabalancing_override) { |
4a404bea | 2552 | pr_info("%s automatic NUMA balancing. " |
c297663c MG |
2553 | "Configure with numa_balancing= or the " |
2554 | "kernel.numa_balancing sysctl", | |
2555 | numabalancing_default ? "Enabling" : "Disabling"); | |
1a687c2e MG |
2556 | set_numabalancing_state(numabalancing_default); |
2557 | } | |
2558 | } | |
2559 | ||
2560 | static int __init setup_numabalancing(char *str) | |
2561 | { | |
2562 | int ret = 0; | |
2563 | if (!str) | |
2564 | goto out; | |
1a687c2e MG |
2565 | |
2566 | if (!strcmp(str, "enable")) { | |
c297663c | 2567 | numabalancing_override = 1; |
1a687c2e MG |
2568 | ret = 1; |
2569 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2570 | numabalancing_override = -1; |
1a687c2e MG |
2571 | ret = 1; |
2572 | } | |
2573 | out: | |
2574 | if (!ret) | |
4a404bea | 2575 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2576 | |
2577 | return ret; | |
2578 | } | |
2579 | __setup("numa_balancing=", setup_numabalancing); | |
2580 | #else | |
2581 | static inline void __init check_numabalancing_enable(void) | |
2582 | { | |
2583 | } | |
2584 | #endif /* CONFIG_NUMA_BALANCING */ | |
2585 | ||
1da177e4 LT |
2586 | /* assumes fs == KERNEL_DS */ |
2587 | void __init numa_policy_init(void) | |
2588 | { | |
b71636e2 PM |
2589 | nodemask_t interleave_nodes; |
2590 | unsigned long largest = 0; | |
2591 | int nid, prefer = 0; | |
2592 | ||
1da177e4 LT |
2593 | policy_cache = kmem_cache_create("numa_policy", |
2594 | sizeof(struct mempolicy), | |
20c2df83 | 2595 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2596 | |
2597 | sn_cache = kmem_cache_create("shared_policy_node", | |
2598 | sizeof(struct sp_node), | |
20c2df83 | 2599 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2600 | |
5606e387 MG |
2601 | for_each_node(nid) { |
2602 | preferred_node_policy[nid] = (struct mempolicy) { | |
2603 | .refcnt = ATOMIC_INIT(1), | |
2604 | .mode = MPOL_PREFERRED, | |
2605 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2606 | .v = { .preferred_node = nid, }, | |
2607 | }; | |
2608 | } | |
2609 | ||
b71636e2 PM |
2610 | /* |
2611 | * Set interleaving policy for system init. Interleaving is only | |
2612 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2613 | * fall back to the largest node if they're all smaller. | |
2614 | */ | |
2615 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2616 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2617 | unsigned long total_pages = node_present_pages(nid); |
2618 | ||
2619 | /* Preserve the largest node */ | |
2620 | if (largest < total_pages) { | |
2621 | largest = total_pages; | |
2622 | prefer = nid; | |
2623 | } | |
2624 | ||
2625 | /* Interleave this node? */ | |
2626 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2627 | node_set(nid, interleave_nodes); | |
2628 | } | |
2629 | ||
2630 | /* All too small, use the largest */ | |
2631 | if (unlikely(nodes_empty(interleave_nodes))) | |
2632 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2633 | |
028fec41 | 2634 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2635 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2636 | |
2637 | check_numabalancing_enable(); | |
1da177e4 LT |
2638 | } |
2639 | ||
8bccd85f | 2640 | /* Reset policy of current process to default */ |
1da177e4 LT |
2641 | void numa_default_policy(void) |
2642 | { | |
028fec41 | 2643 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2644 | } |
68860ec1 | 2645 | |
095f1fc4 LS |
2646 | /* |
2647 | * Parse and format mempolicy from/to strings | |
2648 | */ | |
2649 | ||
1a75a6c8 | 2650 | /* |
f2a07f40 | 2651 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2652 | */ |
345ace9c LS |
2653 | static const char * const policy_modes[] = |
2654 | { | |
2655 | [MPOL_DEFAULT] = "default", | |
2656 | [MPOL_PREFERRED] = "prefer", | |
2657 | [MPOL_BIND] = "bind", | |
2658 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2659 | [MPOL_LOCAL] = "local", |
345ace9c | 2660 | }; |
1a75a6c8 | 2661 | |
095f1fc4 LS |
2662 | |
2663 | #ifdef CONFIG_TMPFS | |
2664 | /** | |
f2a07f40 | 2665 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2666 | * @str: string containing mempolicy to parse |
71fe804b | 2667 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2668 | * |
2669 | * Format of input: | |
2670 | * <mode>[=<flags>][:<nodelist>] | |
2671 | * | |
71fe804b | 2672 | * On success, returns 0, else 1 |
095f1fc4 | 2673 | */ |
a7a88b23 | 2674 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2675 | { |
71fe804b | 2676 | struct mempolicy *new = NULL; |
b4652e84 | 2677 | unsigned short mode; |
f2a07f40 | 2678 | unsigned short mode_flags; |
71fe804b | 2679 | nodemask_t nodes; |
095f1fc4 LS |
2680 | char *nodelist = strchr(str, ':'); |
2681 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2682 | int err = 1; |
2683 | ||
2684 | if (nodelist) { | |
2685 | /* NUL-terminate mode or flags string */ | |
2686 | *nodelist++ = '\0'; | |
71fe804b | 2687 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2688 | goto out; |
01f13bd6 | 2689 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2690 | goto out; |
71fe804b LS |
2691 | } else |
2692 | nodes_clear(nodes); | |
2693 | ||
095f1fc4 LS |
2694 | if (flags) |
2695 | *flags++ = '\0'; /* terminate mode string */ | |
2696 | ||
479e2802 | 2697 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2698 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2699 | break; |
2700 | } | |
2701 | } | |
a720094d | 2702 | if (mode >= MPOL_MAX) |
095f1fc4 LS |
2703 | goto out; |
2704 | ||
71fe804b | 2705 | switch (mode) { |
095f1fc4 | 2706 | case MPOL_PREFERRED: |
71fe804b LS |
2707 | /* |
2708 | * Insist on a nodelist of one node only | |
2709 | */ | |
095f1fc4 LS |
2710 | if (nodelist) { |
2711 | char *rest = nodelist; | |
2712 | while (isdigit(*rest)) | |
2713 | rest++; | |
926f2ae0 KM |
2714 | if (*rest) |
2715 | goto out; | |
095f1fc4 LS |
2716 | } |
2717 | break; | |
095f1fc4 LS |
2718 | case MPOL_INTERLEAVE: |
2719 | /* | |
2720 | * Default to online nodes with memory if no nodelist | |
2721 | */ | |
2722 | if (!nodelist) | |
01f13bd6 | 2723 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2724 | break; |
71fe804b | 2725 | case MPOL_LOCAL: |
3f226aa1 | 2726 | /* |
71fe804b | 2727 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2728 | */ |
71fe804b | 2729 | if (nodelist) |
3f226aa1 | 2730 | goto out; |
71fe804b | 2731 | mode = MPOL_PREFERRED; |
3f226aa1 | 2732 | break; |
413b43de RT |
2733 | case MPOL_DEFAULT: |
2734 | /* | |
2735 | * Insist on a empty nodelist | |
2736 | */ | |
2737 | if (!nodelist) | |
2738 | err = 0; | |
2739 | goto out; | |
d69b2e63 KM |
2740 | case MPOL_BIND: |
2741 | /* | |
2742 | * Insist on a nodelist | |
2743 | */ | |
2744 | if (!nodelist) | |
2745 | goto out; | |
095f1fc4 LS |
2746 | } |
2747 | ||
71fe804b | 2748 | mode_flags = 0; |
095f1fc4 LS |
2749 | if (flags) { |
2750 | /* | |
2751 | * Currently, we only support two mutually exclusive | |
2752 | * mode flags. | |
2753 | */ | |
2754 | if (!strcmp(flags, "static")) | |
71fe804b | 2755 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2756 | else if (!strcmp(flags, "relative")) |
71fe804b | 2757 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2758 | else |
926f2ae0 | 2759 | goto out; |
095f1fc4 | 2760 | } |
71fe804b LS |
2761 | |
2762 | new = mpol_new(mode, mode_flags, &nodes); | |
2763 | if (IS_ERR(new)) | |
926f2ae0 KM |
2764 | goto out; |
2765 | ||
f2a07f40 HD |
2766 | /* |
2767 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2768 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2769 | */ | |
2770 | if (mode != MPOL_PREFERRED) | |
2771 | new->v.nodes = nodes; | |
2772 | else if (nodelist) | |
2773 | new->v.preferred_node = first_node(nodes); | |
2774 | else | |
2775 | new->flags |= MPOL_F_LOCAL; | |
2776 | ||
2777 | /* | |
2778 | * Save nodes for contextualization: this will be used to "clone" | |
2779 | * the mempolicy in a specific context [cpuset] at a later time. | |
2780 | */ | |
2781 | new->w.user_nodemask = nodes; | |
2782 | ||
926f2ae0 | 2783 | err = 0; |
71fe804b | 2784 | |
095f1fc4 LS |
2785 | out: |
2786 | /* Restore string for error message */ | |
2787 | if (nodelist) | |
2788 | *--nodelist = ':'; | |
2789 | if (flags) | |
2790 | *--flags = '='; | |
71fe804b LS |
2791 | if (!err) |
2792 | *mpol = new; | |
095f1fc4 LS |
2793 | return err; |
2794 | } | |
2795 | #endif /* CONFIG_TMPFS */ | |
2796 | ||
71fe804b LS |
2797 | /** |
2798 | * mpol_to_str - format a mempolicy structure for printing | |
2799 | * @buffer: to contain formatted mempolicy string | |
2800 | * @maxlen: length of @buffer | |
2801 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2802 | * |
948927ee DR |
2803 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
2804 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
2805 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 2806 | */ |
948927ee | 2807 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2808 | { |
2809 | char *p = buffer; | |
948927ee DR |
2810 | nodemask_t nodes = NODE_MASK_NONE; |
2811 | unsigned short mode = MPOL_DEFAULT; | |
2812 | unsigned short flags = 0; | |
2291990a | 2813 | |
8790c71a | 2814 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 2815 | mode = pol->mode; |
948927ee DR |
2816 | flags = pol->flags; |
2817 | } | |
bea904d5 | 2818 | |
1a75a6c8 CL |
2819 | switch (mode) { |
2820 | case MPOL_DEFAULT: | |
1a75a6c8 | 2821 | break; |
1a75a6c8 | 2822 | case MPOL_PREFERRED: |
fc36b8d3 | 2823 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 2824 | mode = MPOL_LOCAL; |
53f2556b | 2825 | else |
fc36b8d3 | 2826 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 2827 | break; |
1a75a6c8 | 2828 | case MPOL_BIND: |
1a75a6c8 | 2829 | case MPOL_INTERLEAVE: |
f2a07f40 | 2830 | nodes = pol->v.nodes; |
1a75a6c8 | 2831 | break; |
1a75a6c8 | 2832 | default: |
948927ee DR |
2833 | WARN_ON_ONCE(1); |
2834 | snprintf(p, maxlen, "unknown"); | |
2835 | return; | |
1a75a6c8 CL |
2836 | } |
2837 | ||
b7a9f420 | 2838 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 2839 | |
fc36b8d3 | 2840 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 2841 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 2842 | |
2291990a LS |
2843 | /* |
2844 | * Currently, the only defined flags are mutually exclusive | |
2845 | */ | |
f5b087b5 | 2846 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2847 | p += snprintf(p, buffer + maxlen - p, "static"); |
2848 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2849 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2850 | } |
2851 | ||
1a75a6c8 | 2852 | if (!nodes_empty(nodes)) { |
948927ee | 2853 | p += snprintf(p, buffer + maxlen - p, ":"); |
1a75a6c8 CL |
2854 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2855 | } | |
1a75a6c8 | 2856 | } |