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 | ||
126 | static struct mempolicy *get_task_policy(struct task_struct *p) | |
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 LS |
683 | if (flags & MPOL_MF_LAZY) { |
684 | change_prot_numa(vma, start, endvma); | |
685 | goto next; | |
686 | } | |
687 | ||
688 | if ((flags & MPOL_MF_STRICT) || | |
dc9aa5b9 | 689 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
b24f53a0 | 690 | vma_migratable(vma))) { |
dc9aa5b9 | 691 | |
98094945 | 692 | err = queue_pages_pgd_range(vma, start, endvma, nodes, |
38e35860 | 693 | flags, private); |
d05f0cdc | 694 | if (err) |
1da177e4 | 695 | break; |
1da177e4 | 696 | } |
b24f53a0 | 697 | next: |
1da177e4 LT |
698 | prev = vma; |
699 | } | |
d05f0cdc | 700 | return err; |
1da177e4 LT |
701 | } |
702 | ||
869833f2 KM |
703 | /* |
704 | * Apply policy to a single VMA | |
705 | * This must be called with the mmap_sem held for writing. | |
706 | */ | |
707 | static int vma_replace_policy(struct vm_area_struct *vma, | |
708 | struct mempolicy *pol) | |
8d34694c | 709 | { |
869833f2 KM |
710 | int err; |
711 | struct mempolicy *old; | |
712 | struct mempolicy *new; | |
8d34694c KM |
713 | |
714 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
715 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
716 | vma->vm_ops, vma->vm_file, | |
717 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
718 | ||
869833f2 KM |
719 | new = mpol_dup(pol); |
720 | if (IS_ERR(new)) | |
721 | return PTR_ERR(new); | |
722 | ||
723 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 724 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
725 | if (err) |
726 | goto err_out; | |
8d34694c | 727 | } |
869833f2 KM |
728 | |
729 | old = vma->vm_policy; | |
730 | vma->vm_policy = new; /* protected by mmap_sem */ | |
731 | mpol_put(old); | |
732 | ||
733 | return 0; | |
734 | err_out: | |
735 | mpol_put(new); | |
8d34694c KM |
736 | return err; |
737 | } | |
738 | ||
1da177e4 | 739 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
740 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
741 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
742 | { |
743 | struct vm_area_struct *next; | |
9d8cebd4 KM |
744 | struct vm_area_struct *prev; |
745 | struct vm_area_struct *vma; | |
746 | int err = 0; | |
e26a5114 | 747 | pgoff_t pgoff; |
9d8cebd4 KM |
748 | unsigned long vmstart; |
749 | unsigned long vmend; | |
1da177e4 | 750 | |
097d5910 | 751 | vma = find_vma(mm, start); |
9d8cebd4 KM |
752 | if (!vma || vma->vm_start > start) |
753 | return -EFAULT; | |
754 | ||
097d5910 | 755 | prev = vma->vm_prev; |
e26a5114 KM |
756 | if (start > vma->vm_start) |
757 | prev = vma; | |
758 | ||
9d8cebd4 | 759 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 760 | next = vma->vm_next; |
9d8cebd4 KM |
761 | vmstart = max(start, vma->vm_start); |
762 | vmend = min(end, vma->vm_end); | |
763 | ||
e26a5114 KM |
764 | if (mpol_equal(vma_policy(vma), new_pol)) |
765 | continue; | |
766 | ||
767 | pgoff = vma->vm_pgoff + | |
768 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 769 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 770 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 771 | new_pol); |
9d8cebd4 KM |
772 | if (prev) { |
773 | vma = prev; | |
774 | next = vma->vm_next; | |
3964acd0 ON |
775 | if (mpol_equal(vma_policy(vma), new_pol)) |
776 | continue; | |
777 | /* vma_merge() joined vma && vma->next, case 8 */ | |
778 | goto replace; | |
9d8cebd4 KM |
779 | } |
780 | if (vma->vm_start != vmstart) { | |
781 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
782 | if (err) | |
783 | goto out; | |
784 | } | |
785 | if (vma->vm_end != vmend) { | |
786 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
787 | if (err) | |
788 | goto out; | |
789 | } | |
3964acd0 | 790 | replace: |
869833f2 | 791 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
792 | if (err) |
793 | goto out; | |
1da177e4 | 794 | } |
9d8cebd4 KM |
795 | |
796 | out: | |
1da177e4 LT |
797 | return err; |
798 | } | |
799 | ||
1da177e4 | 800 | /* Set the process memory policy */ |
028fec41 DR |
801 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
802 | nodemask_t *nodes) | |
1da177e4 | 803 | { |
58568d2a | 804 | struct mempolicy *new, *old; |
f4e53d91 | 805 | struct mm_struct *mm = current->mm; |
4bfc4495 | 806 | NODEMASK_SCRATCH(scratch); |
58568d2a | 807 | int ret; |
1da177e4 | 808 | |
4bfc4495 KH |
809 | if (!scratch) |
810 | return -ENOMEM; | |
f4e53d91 | 811 | |
4bfc4495 KH |
812 | new = mpol_new(mode, flags, nodes); |
813 | if (IS_ERR(new)) { | |
814 | ret = PTR_ERR(new); | |
815 | goto out; | |
816 | } | |
f4e53d91 LS |
817 | /* |
818 | * prevent changing our mempolicy while show_numa_maps() | |
819 | * is using it. | |
820 | * Note: do_set_mempolicy() can be called at init time | |
821 | * with no 'mm'. | |
822 | */ | |
823 | if (mm) | |
824 | down_write(&mm->mmap_sem); | |
58568d2a | 825 | task_lock(current); |
4bfc4495 | 826 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
827 | if (ret) { |
828 | task_unlock(current); | |
829 | if (mm) | |
830 | up_write(&mm->mmap_sem); | |
831 | mpol_put(new); | |
4bfc4495 | 832 | goto out; |
58568d2a MX |
833 | } |
834 | old = current->mempolicy; | |
1da177e4 | 835 | current->mempolicy = new; |
45c4745a | 836 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 837 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 838 | current->il_next = first_node(new->v.nodes); |
58568d2a | 839 | task_unlock(current); |
f4e53d91 LS |
840 | if (mm) |
841 | up_write(&mm->mmap_sem); | |
842 | ||
58568d2a | 843 | mpol_put(old); |
4bfc4495 KH |
844 | ret = 0; |
845 | out: | |
846 | NODEMASK_SCRATCH_FREE(scratch); | |
847 | return ret; | |
1da177e4 LT |
848 | } |
849 | ||
bea904d5 LS |
850 | /* |
851 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
852 | * |
853 | * Called with task's alloc_lock held | |
bea904d5 LS |
854 | */ |
855 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 856 | { |
dfcd3c0d | 857 | nodes_clear(*nodes); |
bea904d5 LS |
858 | if (p == &default_policy) |
859 | return; | |
860 | ||
45c4745a | 861 | switch (p->mode) { |
19770b32 MG |
862 | case MPOL_BIND: |
863 | /* Fall through */ | |
1da177e4 | 864 | case MPOL_INTERLEAVE: |
dfcd3c0d | 865 | *nodes = p->v.nodes; |
1da177e4 LT |
866 | break; |
867 | case MPOL_PREFERRED: | |
fc36b8d3 | 868 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 869 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 870 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
871 | break; |
872 | default: | |
873 | BUG(); | |
874 | } | |
875 | } | |
876 | ||
877 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
878 | { | |
879 | struct page *p; | |
880 | int err; | |
881 | ||
882 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
883 | if (err >= 0) { | |
884 | err = page_to_nid(p); | |
885 | put_page(p); | |
886 | } | |
887 | return err; | |
888 | } | |
889 | ||
1da177e4 | 890 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
891 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
892 | unsigned long addr, unsigned long flags) | |
1da177e4 | 893 | { |
8bccd85f | 894 | int err; |
1da177e4 LT |
895 | struct mm_struct *mm = current->mm; |
896 | struct vm_area_struct *vma = NULL; | |
897 | struct mempolicy *pol = current->mempolicy; | |
898 | ||
754af6f5 LS |
899 | if (flags & |
900 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 901 | return -EINVAL; |
754af6f5 LS |
902 | |
903 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
904 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
905 | return -EINVAL; | |
906 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 907 | task_lock(current); |
754af6f5 | 908 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 909 | task_unlock(current); |
754af6f5 LS |
910 | return 0; |
911 | } | |
912 | ||
1da177e4 | 913 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
914 | /* |
915 | * Do NOT fall back to task policy if the | |
916 | * vma/shared policy at addr is NULL. We | |
917 | * want to return MPOL_DEFAULT in this case. | |
918 | */ | |
1da177e4 LT |
919 | down_read(&mm->mmap_sem); |
920 | vma = find_vma_intersection(mm, addr, addr+1); | |
921 | if (!vma) { | |
922 | up_read(&mm->mmap_sem); | |
923 | return -EFAULT; | |
924 | } | |
925 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
926 | pol = vma->vm_ops->get_policy(vma, addr); | |
927 | else | |
928 | pol = vma->vm_policy; | |
929 | } else if (addr) | |
930 | return -EINVAL; | |
931 | ||
932 | if (!pol) | |
bea904d5 | 933 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
934 | |
935 | if (flags & MPOL_F_NODE) { | |
936 | if (flags & MPOL_F_ADDR) { | |
937 | err = lookup_node(mm, addr); | |
938 | if (err < 0) | |
939 | goto out; | |
8bccd85f | 940 | *policy = err; |
1da177e4 | 941 | } else if (pol == current->mempolicy && |
45c4745a | 942 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 943 | *policy = current->il_next; |
1da177e4 LT |
944 | } else { |
945 | err = -EINVAL; | |
946 | goto out; | |
947 | } | |
bea904d5 LS |
948 | } else { |
949 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
950 | pol->mode; | |
d79df630 DR |
951 | /* |
952 | * Internal mempolicy flags must be masked off before exposing | |
953 | * the policy to userspace. | |
954 | */ | |
955 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 956 | } |
1da177e4 LT |
957 | |
958 | if (vma) { | |
959 | up_read(¤t->mm->mmap_sem); | |
960 | vma = NULL; | |
961 | } | |
962 | ||
1da177e4 | 963 | err = 0; |
58568d2a | 964 | if (nmask) { |
c6b6ef8b LS |
965 | if (mpol_store_user_nodemask(pol)) { |
966 | *nmask = pol->w.user_nodemask; | |
967 | } else { | |
968 | task_lock(current); | |
969 | get_policy_nodemask(pol, nmask); | |
970 | task_unlock(current); | |
971 | } | |
58568d2a | 972 | } |
1da177e4 LT |
973 | |
974 | out: | |
52cd3b07 | 975 | mpol_cond_put(pol); |
1da177e4 LT |
976 | if (vma) |
977 | up_read(¤t->mm->mmap_sem); | |
978 | return err; | |
979 | } | |
980 | ||
b20a3503 | 981 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
982 | /* |
983 | * page migration | |
984 | */ | |
fc301289 CL |
985 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
986 | unsigned long flags) | |
6ce3c4c0 CL |
987 | { |
988 | /* | |
fc301289 | 989 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 990 | */ |
62695a84 NP |
991 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
992 | if (!isolate_lru_page(page)) { | |
993 | list_add_tail(&page->lru, pagelist); | |
6d9c285a KM |
994 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
995 | page_is_file_cache(page)); | |
62695a84 NP |
996 | } |
997 | } | |
7e2ab150 | 998 | } |
6ce3c4c0 | 999 | |
742755a1 | 1000 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 1001 | { |
e2d8cf40 NH |
1002 | if (PageHuge(page)) |
1003 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1004 | node); | |
1005 | else | |
1006 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); | |
95a402c3 CL |
1007 | } |
1008 | ||
7e2ab150 CL |
1009 | /* |
1010 | * Migrate pages from one node to a target node. | |
1011 | * Returns error or the number of pages not migrated. | |
1012 | */ | |
dbcb0f19 AB |
1013 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1014 | int flags) | |
7e2ab150 CL |
1015 | { |
1016 | nodemask_t nmask; | |
1017 | LIST_HEAD(pagelist); | |
1018 | int err = 0; | |
1019 | ||
1020 | nodes_clear(nmask); | |
1021 | node_set(source, nmask); | |
6ce3c4c0 | 1022 | |
08270807 MK |
1023 | /* |
1024 | * This does not "check" the range but isolates all pages that | |
1025 | * need migration. Between passing in the full user address | |
1026 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1027 | */ | |
1028 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1029 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1030 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1031 | ||
cf608ac1 | 1032 | if (!list_empty(&pagelist)) { |
68711a74 | 1033 | err = migrate_pages(&pagelist, new_node_page, NULL, dest, |
9c620e2b | 1034 | MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1035 | if (err) |
e2d8cf40 | 1036 | putback_movable_pages(&pagelist); |
cf608ac1 | 1037 | } |
95a402c3 | 1038 | |
7e2ab150 | 1039 | return err; |
6ce3c4c0 CL |
1040 | } |
1041 | ||
39743889 | 1042 | /* |
7e2ab150 CL |
1043 | * Move pages between the two nodesets so as to preserve the physical |
1044 | * layout as much as possible. | |
39743889 CL |
1045 | * |
1046 | * Returns the number of page that could not be moved. | |
1047 | */ | |
0ce72d4f AM |
1048 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1049 | const nodemask_t *to, int flags) | |
39743889 | 1050 | { |
7e2ab150 | 1051 | int busy = 0; |
0aedadf9 | 1052 | int err; |
7e2ab150 | 1053 | nodemask_t tmp; |
39743889 | 1054 | |
0aedadf9 CL |
1055 | err = migrate_prep(); |
1056 | if (err) | |
1057 | return err; | |
1058 | ||
53f2556b | 1059 | down_read(&mm->mmap_sem); |
39743889 | 1060 | |
0ce72d4f | 1061 | err = migrate_vmas(mm, from, to, flags); |
7b2259b3 CL |
1062 | if (err) |
1063 | goto out; | |
1064 | ||
da0aa138 KM |
1065 | /* |
1066 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1067 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1068 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1069 | * The pair of nodemasks 'to' and 'from' define the map. | |
1070 | * | |
1071 | * If no pair of bits is found that way, fallback to picking some | |
1072 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1073 | * 'source' and 'dest' bits are the same, this represents a node | |
1074 | * that will be migrating to itself, so no pages need move. | |
1075 | * | |
1076 | * If no bits are left in 'tmp', or if all remaining bits left | |
1077 | * in 'tmp' correspond to the same bit in 'to', return false | |
1078 | * (nothing left to migrate). | |
1079 | * | |
1080 | * This lets us pick a pair of nodes to migrate between, such that | |
1081 | * if possible the dest node is not already occupied by some other | |
1082 | * source node, minimizing the risk of overloading the memory on a | |
1083 | * node that would happen if we migrated incoming memory to a node | |
1084 | * before migrating outgoing memory source that same node. | |
1085 | * | |
1086 | * A single scan of tmp is sufficient. As we go, we remember the | |
1087 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1088 | * that not only moved, but what's better, moved to an empty slot | |
1089 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1090 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1091 | * most recent <s, d> pair that moved. If we get all the way through |
1092 | * the scan of tmp without finding any node that moved, much less | |
1093 | * moved to an empty node, then there is nothing left worth migrating. | |
1094 | */ | |
d4984711 | 1095 | |
0ce72d4f | 1096 | tmp = *from; |
7e2ab150 CL |
1097 | while (!nodes_empty(tmp)) { |
1098 | int s,d; | |
b76ac7e7 | 1099 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1100 | int dest = 0; |
1101 | ||
1102 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1103 | |
1104 | /* | |
1105 | * do_migrate_pages() tries to maintain the relative | |
1106 | * node relationship of the pages established between | |
1107 | * threads and memory areas. | |
1108 | * | |
1109 | * However if the number of source nodes is not equal to | |
1110 | * the number of destination nodes we can not preserve | |
1111 | * this node relative relationship. In that case, skip | |
1112 | * copying memory from a node that is in the destination | |
1113 | * mask. | |
1114 | * | |
1115 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1116 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1117 | */ | |
1118 | ||
0ce72d4f AM |
1119 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1120 | (node_isset(s, *to))) | |
4a5b18cc LW |
1121 | continue; |
1122 | ||
0ce72d4f | 1123 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1124 | if (s == d) |
1125 | continue; | |
1126 | ||
1127 | source = s; /* Node moved. Memorize */ | |
1128 | dest = d; | |
1129 | ||
1130 | /* dest not in remaining from nodes? */ | |
1131 | if (!node_isset(dest, tmp)) | |
1132 | break; | |
1133 | } | |
b76ac7e7 | 1134 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1135 | break; |
1136 | ||
1137 | node_clear(source, tmp); | |
1138 | err = migrate_to_node(mm, source, dest, flags); | |
1139 | if (err > 0) | |
1140 | busy += err; | |
1141 | if (err < 0) | |
1142 | break; | |
39743889 | 1143 | } |
7b2259b3 | 1144 | out: |
39743889 | 1145 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1146 | if (err < 0) |
1147 | return err; | |
1148 | return busy; | |
b20a3503 CL |
1149 | |
1150 | } | |
1151 | ||
3ad33b24 LS |
1152 | /* |
1153 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1154 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1155 | * Search forward from there, if not. N.B., this assumes that the |
1156 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1157 | * is in virtual address order. | |
1158 | */ | |
d05f0cdc | 1159 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 | 1160 | { |
d05f0cdc | 1161 | struct vm_area_struct *vma; |
3ad33b24 | 1162 | unsigned long uninitialized_var(address); |
95a402c3 | 1163 | |
d05f0cdc | 1164 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1165 | while (vma) { |
1166 | address = page_address_in_vma(page, vma); | |
1167 | if (address != -EFAULT) | |
1168 | break; | |
1169 | vma = vma->vm_next; | |
1170 | } | |
11c731e8 WL |
1171 | |
1172 | if (PageHuge(page)) { | |
cc81717e MH |
1173 | BUG_ON(!vma); |
1174 | return alloc_huge_page_noerr(vma, address, 1); | |
11c731e8 | 1175 | } |
0bf598d8 | 1176 | /* |
11c731e8 | 1177 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1178 | */ |
3ad33b24 | 1179 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
95a402c3 | 1180 | } |
b20a3503 CL |
1181 | #else |
1182 | ||
1183 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1184 | unsigned long flags) | |
1185 | { | |
39743889 CL |
1186 | } |
1187 | ||
0ce72d4f AM |
1188 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1189 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1190 | { |
1191 | return -ENOSYS; | |
1192 | } | |
95a402c3 | 1193 | |
d05f0cdc | 1194 | static struct page *new_page(struct page *page, unsigned long start, int **x) |
95a402c3 CL |
1195 | { |
1196 | return NULL; | |
1197 | } | |
b20a3503 CL |
1198 | #endif |
1199 | ||
dbcb0f19 | 1200 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1201 | unsigned short mode, unsigned short mode_flags, |
1202 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1203 | { |
6ce3c4c0 CL |
1204 | struct mm_struct *mm = current->mm; |
1205 | struct mempolicy *new; | |
1206 | unsigned long end; | |
1207 | int err; | |
1208 | LIST_HEAD(pagelist); | |
1209 | ||
b24f53a0 | 1210 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1211 | return -EINVAL; |
74c00241 | 1212 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1213 | return -EPERM; |
1214 | ||
1215 | if (start & ~PAGE_MASK) | |
1216 | return -EINVAL; | |
1217 | ||
1218 | if (mode == MPOL_DEFAULT) | |
1219 | flags &= ~MPOL_MF_STRICT; | |
1220 | ||
1221 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1222 | end = start + len; | |
1223 | ||
1224 | if (end < start) | |
1225 | return -EINVAL; | |
1226 | if (end == start) | |
1227 | return 0; | |
1228 | ||
028fec41 | 1229 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1230 | if (IS_ERR(new)) |
1231 | return PTR_ERR(new); | |
1232 | ||
b24f53a0 LS |
1233 | if (flags & MPOL_MF_LAZY) |
1234 | new->flags |= MPOL_F_MOF; | |
1235 | ||
6ce3c4c0 CL |
1236 | /* |
1237 | * If we are using the default policy then operation | |
1238 | * on discontinuous address spaces is okay after all | |
1239 | */ | |
1240 | if (!new) | |
1241 | flags |= MPOL_MF_DISCONTIG_OK; | |
1242 | ||
028fec41 DR |
1243 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1244 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1245 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1246 | |
0aedadf9 CL |
1247 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1248 | ||
1249 | err = migrate_prep(); | |
1250 | if (err) | |
b05ca738 | 1251 | goto mpol_out; |
0aedadf9 | 1252 | } |
4bfc4495 KH |
1253 | { |
1254 | NODEMASK_SCRATCH(scratch); | |
1255 | if (scratch) { | |
1256 | down_write(&mm->mmap_sem); | |
1257 | task_lock(current); | |
1258 | err = mpol_set_nodemask(new, nmask, scratch); | |
1259 | task_unlock(current); | |
1260 | if (err) | |
1261 | up_write(&mm->mmap_sem); | |
1262 | } else | |
1263 | err = -ENOMEM; | |
1264 | NODEMASK_SCRATCH_FREE(scratch); | |
1265 | } | |
b05ca738 KM |
1266 | if (err) |
1267 | goto mpol_out; | |
1268 | ||
d05f0cdc | 1269 | err = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1270 | flags | MPOL_MF_INVERT, &pagelist); |
d05f0cdc | 1271 | if (!err) |
9d8cebd4 | 1272 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1273 | |
b24f53a0 LS |
1274 | if (!err) { |
1275 | int nr_failed = 0; | |
1276 | ||
cf608ac1 | 1277 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1278 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1279 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1280 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1281 | if (nr_failed) |
74060e4d | 1282 | putback_movable_pages(&pagelist); |
cf608ac1 | 1283 | } |
6ce3c4c0 | 1284 | |
b24f53a0 | 1285 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1286 | err = -EIO; |
ab8a3e14 | 1287 | } else |
b0e5fd73 | 1288 | putback_movable_pages(&pagelist); |
b20a3503 | 1289 | |
6ce3c4c0 | 1290 | up_write(&mm->mmap_sem); |
b05ca738 | 1291 | mpol_out: |
f0be3d32 | 1292 | mpol_put(new); |
6ce3c4c0 CL |
1293 | return err; |
1294 | } | |
1295 | ||
8bccd85f CL |
1296 | /* |
1297 | * User space interface with variable sized bitmaps for nodelists. | |
1298 | */ | |
1299 | ||
1300 | /* Copy a node mask from user space. */ | |
39743889 | 1301 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1302 | unsigned long maxnode) |
1303 | { | |
1304 | unsigned long k; | |
1305 | unsigned long nlongs; | |
1306 | unsigned long endmask; | |
1307 | ||
1308 | --maxnode; | |
1309 | nodes_clear(*nodes); | |
1310 | if (maxnode == 0 || !nmask) | |
1311 | return 0; | |
a9c930ba | 1312 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1313 | return -EINVAL; |
8bccd85f CL |
1314 | |
1315 | nlongs = BITS_TO_LONGS(maxnode); | |
1316 | if ((maxnode % BITS_PER_LONG) == 0) | |
1317 | endmask = ~0UL; | |
1318 | else | |
1319 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1320 | ||
1321 | /* When the user specified more nodes than supported just check | |
1322 | if the non supported part is all zero. */ | |
1323 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1324 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1325 | return -EINVAL; | |
1326 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1327 | unsigned long t; | |
1328 | if (get_user(t, nmask + k)) | |
1329 | return -EFAULT; | |
1330 | if (k == nlongs - 1) { | |
1331 | if (t & endmask) | |
1332 | return -EINVAL; | |
1333 | } else if (t) | |
1334 | return -EINVAL; | |
1335 | } | |
1336 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1337 | endmask = ~0UL; | |
1338 | } | |
1339 | ||
1340 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1341 | return -EFAULT; | |
1342 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | /* Copy a kernel node mask to user space */ | |
1347 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1348 | nodemask_t *nodes) | |
1349 | { | |
1350 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1351 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1352 | ||
1353 | if (copy > nbytes) { | |
1354 | if (copy > PAGE_SIZE) | |
1355 | return -EINVAL; | |
1356 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1357 | return -EFAULT; | |
1358 | copy = nbytes; | |
1359 | } | |
1360 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1361 | } | |
1362 | ||
938bb9f5 | 1363 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
f7f28ca9 | 1364 | unsigned long, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1365 | unsigned long, maxnode, unsigned, flags) |
8bccd85f CL |
1366 | { |
1367 | nodemask_t nodes; | |
1368 | int err; | |
028fec41 | 1369 | unsigned short mode_flags; |
8bccd85f | 1370 | |
028fec41 DR |
1371 | mode_flags = mode & MPOL_MODE_FLAGS; |
1372 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1373 | if (mode >= MPOL_MAX) |
1374 | return -EINVAL; | |
4c50bc01 DR |
1375 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1376 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1377 | return -EINVAL; | |
8bccd85f CL |
1378 | err = get_nodes(&nodes, nmask, maxnode); |
1379 | if (err) | |
1380 | return err; | |
028fec41 | 1381 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1382 | } |
1383 | ||
1384 | /* Set the process memory policy */ | |
23c8902d | 1385 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
938bb9f5 | 1386 | unsigned long, maxnode) |
8bccd85f CL |
1387 | { |
1388 | int err; | |
1389 | nodemask_t nodes; | |
028fec41 | 1390 | unsigned short flags; |
8bccd85f | 1391 | |
028fec41 DR |
1392 | flags = mode & MPOL_MODE_FLAGS; |
1393 | mode &= ~MPOL_MODE_FLAGS; | |
1394 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1395 | return -EINVAL; |
4c50bc01 DR |
1396 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1397 | return -EINVAL; | |
8bccd85f CL |
1398 | err = get_nodes(&nodes, nmask, maxnode); |
1399 | if (err) | |
1400 | return err; | |
028fec41 | 1401 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1402 | } |
1403 | ||
938bb9f5 HC |
1404 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1405 | const unsigned long __user *, old_nodes, | |
1406 | const unsigned long __user *, new_nodes) | |
39743889 | 1407 | { |
c69e8d9c | 1408 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1409 | struct mm_struct *mm = NULL; |
39743889 | 1410 | struct task_struct *task; |
39743889 CL |
1411 | nodemask_t task_nodes; |
1412 | int err; | |
596d7cfa KM |
1413 | nodemask_t *old; |
1414 | nodemask_t *new; | |
1415 | NODEMASK_SCRATCH(scratch); | |
1416 | ||
1417 | if (!scratch) | |
1418 | return -ENOMEM; | |
39743889 | 1419 | |
596d7cfa KM |
1420 | old = &scratch->mask1; |
1421 | new = &scratch->mask2; | |
1422 | ||
1423 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1424 | if (err) |
596d7cfa | 1425 | goto out; |
39743889 | 1426 | |
596d7cfa | 1427 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1428 | if (err) |
596d7cfa | 1429 | goto out; |
39743889 CL |
1430 | |
1431 | /* Find the mm_struct */ | |
55cfaa3c | 1432 | rcu_read_lock(); |
228ebcbe | 1433 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1434 | if (!task) { |
55cfaa3c | 1435 | rcu_read_unlock(); |
596d7cfa KM |
1436 | err = -ESRCH; |
1437 | goto out; | |
39743889 | 1438 | } |
3268c63e | 1439 | get_task_struct(task); |
39743889 | 1440 | |
596d7cfa | 1441 | err = -EINVAL; |
39743889 CL |
1442 | |
1443 | /* | |
1444 | * Check if this process has the right to modify the specified | |
1445 | * process. The right exists if the process has administrative | |
7f927fcc | 1446 | * capabilities, superuser privileges or the same |
39743889 CL |
1447 | * userid as the target process. |
1448 | */ | |
c69e8d9c | 1449 | tcred = __task_cred(task); |
b38a86eb EB |
1450 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1451 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1452 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1453 | rcu_read_unlock(); |
39743889 | 1454 | err = -EPERM; |
3268c63e | 1455 | goto out_put; |
39743889 | 1456 | } |
c69e8d9c | 1457 | rcu_read_unlock(); |
39743889 CL |
1458 | |
1459 | task_nodes = cpuset_mems_allowed(task); | |
1460 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1461 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1462 | err = -EPERM; |
3268c63e | 1463 | goto out_put; |
39743889 CL |
1464 | } |
1465 | ||
01f13bd6 | 1466 | if (!nodes_subset(*new, node_states[N_MEMORY])) { |
3b42d28b | 1467 | err = -EINVAL; |
3268c63e | 1468 | goto out_put; |
3b42d28b CL |
1469 | } |
1470 | ||
86c3a764 DQ |
1471 | err = security_task_movememory(task); |
1472 | if (err) | |
3268c63e | 1473 | goto out_put; |
86c3a764 | 1474 | |
3268c63e CL |
1475 | mm = get_task_mm(task); |
1476 | put_task_struct(task); | |
f2a9ef88 SL |
1477 | |
1478 | if (!mm) { | |
3268c63e | 1479 | err = -EINVAL; |
f2a9ef88 SL |
1480 | goto out; |
1481 | } | |
1482 | ||
1483 | err = do_migrate_pages(mm, old, new, | |
1484 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1485 | |
1486 | mmput(mm); | |
1487 | out: | |
596d7cfa KM |
1488 | NODEMASK_SCRATCH_FREE(scratch); |
1489 | ||
39743889 | 1490 | return err; |
3268c63e CL |
1491 | |
1492 | out_put: | |
1493 | put_task_struct(task); | |
1494 | goto out; | |
1495 | ||
39743889 CL |
1496 | } |
1497 | ||
1498 | ||
8bccd85f | 1499 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1500 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1501 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1502 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1503 | { |
dbcb0f19 AB |
1504 | int err; |
1505 | int uninitialized_var(pval); | |
8bccd85f CL |
1506 | nodemask_t nodes; |
1507 | ||
1508 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1509 | return -EINVAL; | |
1510 | ||
1511 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1512 | ||
1513 | if (err) | |
1514 | return err; | |
1515 | ||
1516 | if (policy && put_user(pval, policy)) | |
1517 | return -EFAULT; | |
1518 | ||
1519 | if (nmask) | |
1520 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1521 | ||
1522 | return err; | |
1523 | } | |
1524 | ||
1da177e4 LT |
1525 | #ifdef CONFIG_COMPAT |
1526 | ||
c93e0f6c HC |
1527 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1528 | compat_ulong_t __user *, nmask, | |
1529 | compat_ulong_t, maxnode, | |
1530 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1531 | { |
1532 | long err; | |
1533 | unsigned long __user *nm = NULL; | |
1534 | unsigned long nr_bits, alloc_size; | |
1535 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1536 | ||
1537 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1538 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1539 | ||
1540 | if (nmask) | |
1541 | nm = compat_alloc_user_space(alloc_size); | |
1542 | ||
1543 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1544 | ||
1545 | if (!err && nmask) { | |
2bbff6c7 KH |
1546 | unsigned long copy_size; |
1547 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1548 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1549 | /* ensure entire bitmap is zeroed */ |
1550 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1551 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1552 | } | |
1553 | ||
1554 | return err; | |
1555 | } | |
1556 | ||
c93e0f6c HC |
1557 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1558 | compat_ulong_t, maxnode) | |
1da177e4 LT |
1559 | { |
1560 | long err = 0; | |
1561 | unsigned long __user *nm = NULL; | |
1562 | unsigned long nr_bits, alloc_size; | |
1563 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1564 | ||
1565 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1566 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1567 | ||
1568 | if (nmask) { | |
1569 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1570 | nm = compat_alloc_user_space(alloc_size); | |
1571 | err |= copy_to_user(nm, bm, alloc_size); | |
1572 | } | |
1573 | ||
1574 | if (err) | |
1575 | return -EFAULT; | |
1576 | ||
1577 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1578 | } | |
1579 | ||
c93e0f6c HC |
1580 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1581 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1582 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 LT |
1583 | { |
1584 | long err = 0; | |
1585 | unsigned long __user *nm = NULL; | |
1586 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1587 | nodemask_t bm; |
1da177e4 LT |
1588 | |
1589 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1590 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1591 | ||
1592 | if (nmask) { | |
dfcd3c0d | 1593 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1594 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1595 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1596 | } |
1597 | ||
1598 | if (err) | |
1599 | return -EFAULT; | |
1600 | ||
1601 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1602 | } | |
1603 | ||
1604 | #endif | |
1605 | ||
480eccf9 LS |
1606 | /* |
1607 | * get_vma_policy(@task, @vma, @addr) | |
b46e14ac FF |
1608 | * @task: task for fallback if vma policy == default |
1609 | * @vma: virtual memory area whose policy is sought | |
1610 | * @addr: address in @vma for shared policy lookup | |
480eccf9 LS |
1611 | * |
1612 | * Returns effective policy for a VMA at specified address. | |
1613 | * Falls back to @task or system default policy, as necessary. | |
32f8516a DR |
1614 | * Current or other task's task mempolicy and non-shared vma policies must be |
1615 | * protected by task_lock(task) by the caller. | |
52cd3b07 LS |
1616 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1617 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1618 | * freeing by another task. It is the caller's responsibility to free the | |
1619 | * extra reference for shared policies. | |
480eccf9 | 1620 | */ |
d98f6cb6 | 1621 | struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1622 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1623 | { |
8d90274b | 1624 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1625 | |
1626 | if (vma) { | |
480eccf9 | 1627 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1628 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1629 | } else if (vma->vm_policy) { |
1da177e4 | 1630 | pol = vma->vm_policy; |
00442ad0 MG |
1631 | |
1632 | /* | |
1633 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1634 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1635 | * count on these policies which will be dropped by | |
1636 | * mpol_cond_put() later | |
1637 | */ | |
1638 | if (mpol_needs_cond_ref(pol)) | |
1639 | mpol_get(pol); | |
1640 | } | |
1da177e4 | 1641 | } |
f15ca78e | 1642 | |
8d90274b ON |
1643 | if (!pol) |
1644 | pol = get_task_policy(task); | |
1645 | ||
1da177e4 LT |
1646 | return pol; |
1647 | } | |
1648 | ||
6b6482bb | 1649 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1650 | { |
6b6482bb | 1651 | struct mempolicy *pol; |
fc314724 | 1652 | |
6b6482bb ON |
1653 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1654 | bool ret = false; | |
fc314724 | 1655 | |
6b6482bb ON |
1656 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1657 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1658 | ret = true; | |
1659 | mpol_cond_put(pol); | |
8d90274b | 1660 | |
6b6482bb | 1661 | return ret; |
fc314724 MG |
1662 | } |
1663 | ||
6b6482bb | 1664 | pol = vma->vm_policy; |
8d90274b | 1665 | if (!pol) |
6b6482bb | 1666 | pol = get_task_policy(current); |
8d90274b | 1667 | |
fc314724 MG |
1668 | return pol->flags & MPOL_F_MOF; |
1669 | } | |
1670 | ||
d3eb1570 LJ |
1671 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1672 | { | |
1673 | enum zone_type dynamic_policy_zone = policy_zone; | |
1674 | ||
1675 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1676 | ||
1677 | /* | |
1678 | * if policy->v.nodes has movable memory only, | |
1679 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1680 | * | |
1681 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1682 | * so if the following test faile, it implies | |
1683 | * policy->v.nodes has movable memory only. | |
1684 | */ | |
1685 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1686 | dynamic_policy_zone = ZONE_MOVABLE; | |
1687 | ||
1688 | return zone >= dynamic_policy_zone; | |
1689 | } | |
1690 | ||
52cd3b07 LS |
1691 | /* |
1692 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1693 | * page allocation | |
1694 | */ | |
1695 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1696 | { |
1697 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1698 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1699 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1700 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1701 | return &policy->v.nodes; | |
1702 | ||
1703 | return NULL; | |
1704 | } | |
1705 | ||
52cd3b07 | 1706 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1707 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1708 | int nd) | |
1da177e4 | 1709 | { |
45c4745a | 1710 | switch (policy->mode) { |
1da177e4 | 1711 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1712 | if (!(policy->flags & MPOL_F_LOCAL)) |
1713 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1714 | break; |
1715 | case MPOL_BIND: | |
19770b32 | 1716 | /* |
52cd3b07 LS |
1717 | * Normally, MPOL_BIND allocations are node-local within the |
1718 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1719 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1720 | * the first node in the mask instead. |
19770b32 | 1721 | */ |
19770b32 MG |
1722 | if (unlikely(gfp & __GFP_THISNODE) && |
1723 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1724 | nd = first_node(policy->v.nodes); | |
1725 | break; | |
1da177e4 | 1726 | default: |
1da177e4 LT |
1727 | BUG(); |
1728 | } | |
0e88460d | 1729 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1730 | } |
1731 | ||
1732 | /* Do dynamic interleaving for a process */ | |
1733 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1734 | { | |
1735 | unsigned nid, next; | |
1736 | struct task_struct *me = current; | |
1737 | ||
1738 | nid = me->il_next; | |
dfcd3c0d | 1739 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1740 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1741 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1742 | if (next < MAX_NUMNODES) |
1743 | me->il_next = next; | |
1da177e4 LT |
1744 | return nid; |
1745 | } | |
1746 | ||
dc85da15 CL |
1747 | /* |
1748 | * Depending on the memory policy provide a node from which to allocate the | |
1749 | * next slab entry. | |
1750 | */ | |
2a389610 | 1751 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1752 | { |
e7b691b0 | 1753 | struct mempolicy *policy; |
2a389610 | 1754 | int node = numa_mem_id(); |
e7b691b0 AK |
1755 | |
1756 | if (in_interrupt()) | |
2a389610 | 1757 | return node; |
e7b691b0 AK |
1758 | |
1759 | policy = current->mempolicy; | |
fc36b8d3 | 1760 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1761 | return node; |
bea904d5 LS |
1762 | |
1763 | switch (policy->mode) { | |
1764 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1765 | /* |
1766 | * handled MPOL_F_LOCAL above | |
1767 | */ | |
1768 | return policy->v.preferred_node; | |
765c4507 | 1769 | |
dc85da15 CL |
1770 | case MPOL_INTERLEAVE: |
1771 | return interleave_nodes(policy); | |
1772 | ||
dd1a239f | 1773 | case MPOL_BIND: { |
dc85da15 CL |
1774 | /* |
1775 | * Follow bind policy behavior and start allocation at the | |
1776 | * first node. | |
1777 | */ | |
19770b32 MG |
1778 | struct zonelist *zonelist; |
1779 | struct zone *zone; | |
1780 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
2a389610 | 1781 | zonelist = &NODE_DATA(node)->node_zonelists[0]; |
19770b32 MG |
1782 | (void)first_zones_zonelist(zonelist, highest_zoneidx, |
1783 | &policy->v.nodes, | |
1784 | &zone); | |
2a389610 | 1785 | return zone ? zone->node : node; |
dd1a239f | 1786 | } |
dc85da15 | 1787 | |
dc85da15 | 1788 | default: |
bea904d5 | 1789 | BUG(); |
dc85da15 CL |
1790 | } |
1791 | } | |
1792 | ||
1da177e4 LT |
1793 | /* Do static interleaving for a VMA with known offset. */ |
1794 | static unsigned offset_il_node(struct mempolicy *pol, | |
1795 | struct vm_area_struct *vma, unsigned long off) | |
1796 | { | |
dfcd3c0d | 1797 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1798 | unsigned target; |
1da177e4 | 1799 | int c; |
b76ac7e7 | 1800 | int nid = NUMA_NO_NODE; |
1da177e4 | 1801 | |
f5b087b5 DR |
1802 | if (!nnodes) |
1803 | return numa_node_id(); | |
1804 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1805 | c = 0; |
1806 | do { | |
dfcd3c0d | 1807 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1808 | c++; |
1809 | } while (c <= target); | |
1da177e4 LT |
1810 | return nid; |
1811 | } | |
1812 | ||
5da7ca86 CL |
1813 | /* Determine a node number for interleave */ |
1814 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1815 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1816 | { | |
1817 | if (vma) { | |
1818 | unsigned long off; | |
1819 | ||
3b98b087 NA |
1820 | /* |
1821 | * for small pages, there is no difference between | |
1822 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1823 | * for huge pages, since vm_pgoff is in units of small | |
1824 | * pages, we need to shift off the always 0 bits to get | |
1825 | * a useful offset. | |
1826 | */ | |
1827 | BUG_ON(shift < PAGE_SHIFT); | |
1828 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1829 | off += (addr - vma->vm_start) >> shift; |
1830 | return offset_il_node(pol, vma, off); | |
1831 | } else | |
1832 | return interleave_nodes(pol); | |
1833 | } | |
1834 | ||
778d3b0f MH |
1835 | /* |
1836 | * Return the bit number of a random bit set in the nodemask. | |
b76ac7e7 | 1837 | * (returns NUMA_NO_NODE if nodemask is empty) |
778d3b0f MH |
1838 | */ |
1839 | int node_random(const nodemask_t *maskp) | |
1840 | { | |
b76ac7e7 | 1841 | int w, bit = NUMA_NO_NODE; |
778d3b0f MH |
1842 | |
1843 | w = nodes_weight(*maskp); | |
1844 | if (w) | |
1845 | bit = bitmap_ord_to_pos(maskp->bits, | |
1846 | get_random_int() % w, MAX_NUMNODES); | |
1847 | return bit; | |
1848 | } | |
1849 | ||
00ac59ad | 1850 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1851 | /* |
1852 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
b46e14ac FF |
1853 | * @vma: virtual memory area whose policy is sought |
1854 | * @addr: address in @vma for shared policy lookup and interleave policy | |
1855 | * @gfp_flags: for requested zone | |
1856 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
1857 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1858 | * |
52cd3b07 LS |
1859 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1860 | * to the struct mempolicy for conditional unref after allocation. | |
1861 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1862 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 1863 | * |
d26914d1 | 1864 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 1865 | */ |
396faf03 | 1866 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1867 | gfp_t gfp_flags, struct mempolicy **mpol, |
1868 | nodemask_t **nodemask) | |
5da7ca86 | 1869 | { |
480eccf9 | 1870 | struct zonelist *zl; |
5da7ca86 | 1871 | |
52cd3b07 | 1872 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1873 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1874 | |
52cd3b07 LS |
1875 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1876 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1877 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1878 | } else { |
2f5f9486 | 1879 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1880 | if ((*mpol)->mode == MPOL_BIND) |
1881 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1882 | } |
1883 | return zl; | |
5da7ca86 | 1884 | } |
06808b08 LS |
1885 | |
1886 | /* | |
1887 | * init_nodemask_of_mempolicy | |
1888 | * | |
1889 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1890 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1891 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1892 | * initialize the argument nodemask to contain the single node for | |
1893 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1894 | * of non-default mempolicy. | |
1895 | * | |
1896 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1897 | * because the current task is examining it's own mempolicy and a task's | |
1898 | * mempolicy is only ever changed by the task itself. | |
1899 | * | |
1900 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1901 | */ | |
1902 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1903 | { | |
1904 | struct mempolicy *mempolicy; | |
1905 | int nid; | |
1906 | ||
1907 | if (!(mask && current->mempolicy)) | |
1908 | return false; | |
1909 | ||
c0ff7453 | 1910 | task_lock(current); |
06808b08 LS |
1911 | mempolicy = current->mempolicy; |
1912 | switch (mempolicy->mode) { | |
1913 | case MPOL_PREFERRED: | |
1914 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1915 | nid = numa_node_id(); | |
1916 | else | |
1917 | nid = mempolicy->v.preferred_node; | |
1918 | init_nodemask_of_node(mask, nid); | |
1919 | break; | |
1920 | ||
1921 | case MPOL_BIND: | |
1922 | /* Fall through */ | |
1923 | case MPOL_INTERLEAVE: | |
1924 | *mask = mempolicy->v.nodes; | |
1925 | break; | |
1926 | ||
1927 | default: | |
1928 | BUG(); | |
1929 | } | |
c0ff7453 | 1930 | task_unlock(current); |
06808b08 LS |
1931 | |
1932 | return true; | |
1933 | } | |
00ac59ad | 1934 | #endif |
5da7ca86 | 1935 | |
6f48d0eb DR |
1936 | /* |
1937 | * mempolicy_nodemask_intersects | |
1938 | * | |
1939 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1940 | * policy. Otherwise, check for intersection between mask and the policy | |
1941 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1942 | * policy, always return true since it may allocate elsewhere on fallback. | |
1943 | * | |
1944 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1945 | */ | |
1946 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1947 | const nodemask_t *mask) | |
1948 | { | |
1949 | struct mempolicy *mempolicy; | |
1950 | bool ret = true; | |
1951 | ||
1952 | if (!mask) | |
1953 | return ret; | |
1954 | task_lock(tsk); | |
1955 | mempolicy = tsk->mempolicy; | |
1956 | if (!mempolicy) | |
1957 | goto out; | |
1958 | ||
1959 | switch (mempolicy->mode) { | |
1960 | case MPOL_PREFERRED: | |
1961 | /* | |
1962 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1963 | * allocate from, they may fallback to other nodes when oom. | |
1964 | * Thus, it's possible for tsk to have allocated memory from | |
1965 | * nodes in mask. | |
1966 | */ | |
1967 | break; | |
1968 | case MPOL_BIND: | |
1969 | case MPOL_INTERLEAVE: | |
1970 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1971 | break; | |
1972 | default: | |
1973 | BUG(); | |
1974 | } | |
1975 | out: | |
1976 | task_unlock(tsk); | |
1977 | return ret; | |
1978 | } | |
1979 | ||
1da177e4 LT |
1980 | /* Allocate a page in interleaved policy. |
1981 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1982 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1983 | unsigned nid) | |
1da177e4 LT |
1984 | { |
1985 | struct zonelist *zl; | |
1986 | struct page *page; | |
1987 | ||
0e88460d | 1988 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1989 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1990 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1991 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1992 | return page; |
1993 | } | |
1994 | ||
1995 | /** | |
0bbbc0b3 | 1996 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1997 | * |
1998 | * @gfp: | |
1999 | * %GFP_USER user allocation. | |
2000 | * %GFP_KERNEL kernel allocations, | |
2001 | * %GFP_HIGHMEM highmem/user allocations, | |
2002 | * %GFP_FS allocation should not call back into a file system. | |
2003 | * %GFP_ATOMIC don't sleep. | |
2004 | * | |
0bbbc0b3 | 2005 | * @order:Order of the GFP allocation. |
1da177e4 LT |
2006 | * @vma: Pointer to VMA or NULL if not available. |
2007 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
2008 | * | |
2009 | * This function allocates a page from the kernel page pool and applies | |
2010 | * a NUMA policy associated with the VMA or the current process. | |
2011 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
2012 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
2013 | * all allocations for pages that will be mapped into | |
2014 | * user space. Returns NULL when no page can be allocated. | |
2015 | * | |
2016 | * Should be called with the mm_sem of the vma hold. | |
2017 | */ | |
2018 | struct page * | |
0bbbc0b3 | 2019 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 2020 | unsigned long addr, int node) |
1da177e4 | 2021 | { |
cc9a6c87 | 2022 | struct mempolicy *pol; |
c0ff7453 | 2023 | struct page *page; |
cc9a6c87 MG |
2024 | unsigned int cpuset_mems_cookie; |
2025 | ||
2026 | retry_cpuset: | |
2027 | pol = get_vma_policy(current, vma, addr); | |
d26914d1 | 2028 | cpuset_mems_cookie = read_mems_allowed_begin(); |
1da177e4 | 2029 | |
45c4745a | 2030 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 2031 | unsigned nid; |
5da7ca86 | 2032 | |
8eac563c | 2033 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 2034 | mpol_cond_put(pol); |
0bbbc0b3 | 2035 | page = alloc_page_interleave(gfp, order, nid); |
d26914d1 | 2036 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2037 | goto retry_cpuset; |
2038 | ||
c0ff7453 | 2039 | return page; |
1da177e4 | 2040 | } |
212a0a6f DR |
2041 | page = __alloc_pages_nodemask(gfp, order, |
2042 | policy_zonelist(gfp, pol, node), | |
0bbbc0b3 | 2043 | policy_nodemask(gfp, pol)); |
2386740d | 2044 | mpol_cond_put(pol); |
d26914d1 | 2045 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 2046 | goto retry_cpuset; |
c0ff7453 | 2047 | return page; |
1da177e4 LT |
2048 | } |
2049 | ||
2050 | /** | |
2051 | * alloc_pages_current - Allocate pages. | |
2052 | * | |
2053 | * @gfp: | |
2054 | * %GFP_USER user allocation, | |
2055 | * %GFP_KERNEL kernel allocation, | |
2056 | * %GFP_HIGHMEM highmem allocation, | |
2057 | * %GFP_FS don't call back into a file system. | |
2058 | * %GFP_ATOMIC don't sleep. | |
2059 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2060 | * | |
2061 | * Allocate a page from the kernel page pool. When not in | |
2062 | * interrupt context and apply the current process NUMA policy. | |
2063 | * Returns NULL when no page can be allocated. | |
2064 | * | |
cf2a473c | 2065 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
2066 | * 1) it's ok to take cpuset_sem (can WAIT), and |
2067 | * 2) allocating for current task (not interrupt). | |
2068 | */ | |
dd0fc66f | 2069 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2070 | { |
8d90274b | 2071 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2072 | struct page *page; |
cc9a6c87 | 2073 | unsigned int cpuset_mems_cookie; |
1da177e4 | 2074 | |
8d90274b ON |
2075 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2076 | pol = get_task_policy(current); | |
52cd3b07 | 2077 | |
cc9a6c87 | 2078 | retry_cpuset: |
d26914d1 | 2079 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2080 | |
52cd3b07 LS |
2081 | /* |
2082 | * No reference counting needed for current->mempolicy | |
2083 | * nor system default_policy | |
2084 | */ | |
45c4745a | 2085 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2086 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2087 | else | |
2088 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
2089 | policy_zonelist(gfp, pol, numa_node_id()), |
2090 | policy_nodemask(gfp, pol)); | |
cc9a6c87 | 2091 | |
d26914d1 | 2092 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2093 | goto retry_cpuset; |
2094 | ||
c0ff7453 | 2095 | return page; |
1da177e4 LT |
2096 | } |
2097 | EXPORT_SYMBOL(alloc_pages_current); | |
2098 | ||
ef0855d3 ON |
2099 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2100 | { | |
2101 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2102 | ||
2103 | if (IS_ERR(pol)) | |
2104 | return PTR_ERR(pol); | |
2105 | dst->vm_policy = pol; | |
2106 | return 0; | |
2107 | } | |
2108 | ||
4225399a | 2109 | /* |
846a16bf | 2110 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2111 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2112 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2113 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2114 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2115 | * |
2116 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2117 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2118 | */ |
4225399a | 2119 | |
846a16bf LS |
2120 | /* Slow path of a mempolicy duplicate */ |
2121 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2122 | { |
2123 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2124 | ||
2125 | if (!new) | |
2126 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2127 | |
2128 | /* task's mempolicy is protected by alloc_lock */ | |
2129 | if (old == current->mempolicy) { | |
2130 | task_lock(current); | |
2131 | *new = *old; | |
2132 | task_unlock(current); | |
2133 | } else | |
2134 | *new = *old; | |
2135 | ||
4225399a PJ |
2136 | if (current_cpuset_is_being_rebound()) { |
2137 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2138 | if (new->flags & MPOL_F_REBINDING) |
2139 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2140 | else | |
2141 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2142 | } |
1da177e4 | 2143 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2144 | return new; |
2145 | } | |
2146 | ||
2147 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2148 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2149 | { |
2150 | if (!a || !b) | |
fcfb4dcc | 2151 | return false; |
45c4745a | 2152 | if (a->mode != b->mode) |
fcfb4dcc | 2153 | return false; |
19800502 | 2154 | if (a->flags != b->flags) |
fcfb4dcc | 2155 | return false; |
19800502 BL |
2156 | if (mpol_store_user_nodemask(a)) |
2157 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2158 | return false; |
19800502 | 2159 | |
45c4745a | 2160 | switch (a->mode) { |
19770b32 MG |
2161 | case MPOL_BIND: |
2162 | /* Fall through */ | |
1da177e4 | 2163 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2164 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2165 | case MPOL_PREFERRED: |
75719661 | 2166 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2167 | default: |
2168 | BUG(); | |
fcfb4dcc | 2169 | return false; |
1da177e4 LT |
2170 | } |
2171 | } | |
2172 | ||
1da177e4 LT |
2173 | /* |
2174 | * Shared memory backing store policy support. | |
2175 | * | |
2176 | * Remember policies even when nobody has shared memory mapped. | |
2177 | * The policies are kept in Red-Black tree linked from the inode. | |
2178 | * They are protected by the sp->lock spinlock, which should be held | |
2179 | * for any accesses to the tree. | |
2180 | */ | |
2181 | ||
2182 | /* lookup first element intersecting start-end */ | |
42288fe3 | 2183 | /* Caller holds sp->lock */ |
1da177e4 LT |
2184 | static struct sp_node * |
2185 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2186 | { | |
2187 | struct rb_node *n = sp->root.rb_node; | |
2188 | ||
2189 | while (n) { | |
2190 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2191 | ||
2192 | if (start >= p->end) | |
2193 | n = n->rb_right; | |
2194 | else if (end <= p->start) | |
2195 | n = n->rb_left; | |
2196 | else | |
2197 | break; | |
2198 | } | |
2199 | if (!n) | |
2200 | return NULL; | |
2201 | for (;;) { | |
2202 | struct sp_node *w = NULL; | |
2203 | struct rb_node *prev = rb_prev(n); | |
2204 | if (!prev) | |
2205 | break; | |
2206 | w = rb_entry(prev, struct sp_node, nd); | |
2207 | if (w->end <= start) | |
2208 | break; | |
2209 | n = prev; | |
2210 | } | |
2211 | return rb_entry(n, struct sp_node, nd); | |
2212 | } | |
2213 | ||
2214 | /* Insert a new shared policy into the list. */ | |
2215 | /* Caller holds sp->lock */ | |
2216 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2217 | { | |
2218 | struct rb_node **p = &sp->root.rb_node; | |
2219 | struct rb_node *parent = NULL; | |
2220 | struct sp_node *nd; | |
2221 | ||
2222 | while (*p) { | |
2223 | parent = *p; | |
2224 | nd = rb_entry(parent, struct sp_node, nd); | |
2225 | if (new->start < nd->start) | |
2226 | p = &(*p)->rb_left; | |
2227 | else if (new->end > nd->end) | |
2228 | p = &(*p)->rb_right; | |
2229 | else | |
2230 | BUG(); | |
2231 | } | |
2232 | rb_link_node(&new->nd, parent, p); | |
2233 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2234 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2235 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2236 | } |
2237 | ||
2238 | /* Find shared policy intersecting idx */ | |
2239 | struct mempolicy * | |
2240 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2241 | { | |
2242 | struct mempolicy *pol = NULL; | |
2243 | struct sp_node *sn; | |
2244 | ||
2245 | if (!sp->root.rb_node) | |
2246 | return NULL; | |
42288fe3 | 2247 | spin_lock(&sp->lock); |
1da177e4 LT |
2248 | sn = sp_lookup(sp, idx, idx+1); |
2249 | if (sn) { | |
2250 | mpol_get(sn->policy); | |
2251 | pol = sn->policy; | |
2252 | } | |
42288fe3 | 2253 | spin_unlock(&sp->lock); |
1da177e4 LT |
2254 | return pol; |
2255 | } | |
2256 | ||
63f74ca2 KM |
2257 | static void sp_free(struct sp_node *n) |
2258 | { | |
2259 | mpol_put(n->policy); | |
2260 | kmem_cache_free(sn_cache, n); | |
2261 | } | |
2262 | ||
771fb4d8 LS |
2263 | /** |
2264 | * mpol_misplaced - check whether current page node is valid in policy | |
2265 | * | |
b46e14ac FF |
2266 | * @page: page to be checked |
2267 | * @vma: vm area where page mapped | |
2268 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2269 | * |
2270 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2271 | * node id. | |
2272 | * | |
2273 | * Returns: | |
2274 | * -1 - not misplaced, page is in the right node | |
2275 | * node - node id where the page should be | |
2276 | * | |
2277 | * Policy determination "mimics" alloc_page_vma(). | |
2278 | * Called from fault path where we know the vma and faulting address. | |
2279 | */ | |
2280 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2281 | { | |
2282 | struct mempolicy *pol; | |
2283 | struct zone *zone; | |
2284 | int curnid = page_to_nid(page); | |
2285 | unsigned long pgoff; | |
90572890 PZ |
2286 | int thiscpu = raw_smp_processor_id(); |
2287 | int thisnid = cpu_to_node(thiscpu); | |
771fb4d8 LS |
2288 | int polnid = -1; |
2289 | int ret = -1; | |
2290 | ||
2291 | BUG_ON(!vma); | |
2292 | ||
2293 | pol = get_vma_policy(current, vma, addr); | |
2294 | if (!(pol->flags & MPOL_F_MOF)) | |
2295 | goto out; | |
2296 | ||
2297 | switch (pol->mode) { | |
2298 | case MPOL_INTERLEAVE: | |
2299 | BUG_ON(addr >= vma->vm_end); | |
2300 | BUG_ON(addr < vma->vm_start); | |
2301 | ||
2302 | pgoff = vma->vm_pgoff; | |
2303 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2304 | polnid = offset_il_node(pol, vma, pgoff); | |
2305 | break; | |
2306 | ||
2307 | case MPOL_PREFERRED: | |
2308 | if (pol->flags & MPOL_F_LOCAL) | |
2309 | polnid = numa_node_id(); | |
2310 | else | |
2311 | polnid = pol->v.preferred_node; | |
2312 | break; | |
2313 | ||
2314 | case MPOL_BIND: | |
2315 | /* | |
2316 | * allows binding to multiple nodes. | |
2317 | * use current page if in policy nodemask, | |
2318 | * else select nearest allowed node, if any. | |
2319 | * If no allowed nodes, use current [!misplaced]. | |
2320 | */ | |
2321 | if (node_isset(curnid, pol->v.nodes)) | |
2322 | goto out; | |
2323 | (void)first_zones_zonelist( | |
2324 | node_zonelist(numa_node_id(), GFP_HIGHUSER), | |
2325 | gfp_zone(GFP_HIGHUSER), | |
2326 | &pol->v.nodes, &zone); | |
2327 | polnid = zone->node; | |
2328 | break; | |
2329 | ||
2330 | default: | |
2331 | BUG(); | |
2332 | } | |
5606e387 MG |
2333 | |
2334 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2335 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2336 | polnid = thisnid; |
5606e387 | 2337 | |
10f39042 | 2338 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2339 | goto out; |
e42c8ff2 MG |
2340 | } |
2341 | ||
771fb4d8 LS |
2342 | if (curnid != polnid) |
2343 | ret = polnid; | |
2344 | out: | |
2345 | mpol_cond_put(pol); | |
2346 | ||
2347 | return ret; | |
2348 | } | |
2349 | ||
1da177e4 LT |
2350 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2351 | { | |
140d5a49 | 2352 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2353 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2354 | sp_free(n); |
1da177e4 LT |
2355 | } |
2356 | ||
42288fe3 MG |
2357 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2358 | unsigned long end, struct mempolicy *pol) | |
2359 | { | |
2360 | node->start = start; | |
2361 | node->end = end; | |
2362 | node->policy = pol; | |
2363 | } | |
2364 | ||
dbcb0f19 AB |
2365 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2366 | struct mempolicy *pol) | |
1da177e4 | 2367 | { |
869833f2 KM |
2368 | struct sp_node *n; |
2369 | struct mempolicy *newpol; | |
1da177e4 | 2370 | |
869833f2 | 2371 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2372 | if (!n) |
2373 | return NULL; | |
869833f2 KM |
2374 | |
2375 | newpol = mpol_dup(pol); | |
2376 | if (IS_ERR(newpol)) { | |
2377 | kmem_cache_free(sn_cache, n); | |
2378 | return NULL; | |
2379 | } | |
2380 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2381 | sp_node_init(n, start, end, newpol); |
869833f2 | 2382 | |
1da177e4 LT |
2383 | return n; |
2384 | } | |
2385 | ||
2386 | /* Replace a policy range. */ | |
2387 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2388 | unsigned long end, struct sp_node *new) | |
2389 | { | |
b22d127a | 2390 | struct sp_node *n; |
42288fe3 MG |
2391 | struct sp_node *n_new = NULL; |
2392 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2393 | int ret = 0; |
1da177e4 | 2394 | |
42288fe3 MG |
2395 | restart: |
2396 | spin_lock(&sp->lock); | |
1da177e4 LT |
2397 | n = sp_lookup(sp, start, end); |
2398 | /* Take care of old policies in the same range. */ | |
2399 | while (n && n->start < end) { | |
2400 | struct rb_node *next = rb_next(&n->nd); | |
2401 | if (n->start >= start) { | |
2402 | if (n->end <= end) | |
2403 | sp_delete(sp, n); | |
2404 | else | |
2405 | n->start = end; | |
2406 | } else { | |
2407 | /* Old policy spanning whole new range. */ | |
2408 | if (n->end > end) { | |
42288fe3 MG |
2409 | if (!n_new) |
2410 | goto alloc_new; | |
2411 | ||
2412 | *mpol_new = *n->policy; | |
2413 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2414 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2415 | n->end = start; |
5ca39575 | 2416 | sp_insert(sp, n_new); |
42288fe3 MG |
2417 | n_new = NULL; |
2418 | mpol_new = NULL; | |
1da177e4 LT |
2419 | break; |
2420 | } else | |
2421 | n->end = start; | |
2422 | } | |
2423 | if (!next) | |
2424 | break; | |
2425 | n = rb_entry(next, struct sp_node, nd); | |
2426 | } | |
2427 | if (new) | |
2428 | sp_insert(sp, new); | |
42288fe3 MG |
2429 | spin_unlock(&sp->lock); |
2430 | ret = 0; | |
2431 | ||
2432 | err_out: | |
2433 | if (mpol_new) | |
2434 | mpol_put(mpol_new); | |
2435 | if (n_new) | |
2436 | kmem_cache_free(sn_cache, n_new); | |
2437 | ||
b22d127a | 2438 | return ret; |
42288fe3 MG |
2439 | |
2440 | alloc_new: | |
2441 | spin_unlock(&sp->lock); | |
2442 | ret = -ENOMEM; | |
2443 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2444 | if (!n_new) | |
2445 | goto err_out; | |
2446 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2447 | if (!mpol_new) | |
2448 | goto err_out; | |
2449 | goto restart; | |
1da177e4 LT |
2450 | } |
2451 | ||
71fe804b LS |
2452 | /** |
2453 | * mpol_shared_policy_init - initialize shared policy for inode | |
2454 | * @sp: pointer to inode shared policy | |
2455 | * @mpol: struct mempolicy to install | |
2456 | * | |
2457 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2458 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2459 | * This must be released on exit. | |
4bfc4495 | 2460 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2461 | */ |
2462 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2463 | { | |
58568d2a MX |
2464 | int ret; |
2465 | ||
71fe804b | 2466 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
42288fe3 | 2467 | spin_lock_init(&sp->lock); |
71fe804b LS |
2468 | |
2469 | if (mpol) { | |
2470 | struct vm_area_struct pvma; | |
2471 | struct mempolicy *new; | |
4bfc4495 | 2472 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2473 | |
4bfc4495 | 2474 | if (!scratch) |
5c0c1654 | 2475 | goto put_mpol; |
71fe804b LS |
2476 | /* contextualize the tmpfs mount point mempolicy */ |
2477 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2478 | if (IS_ERR(new)) |
0cae3457 | 2479 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2480 | |
2481 | task_lock(current); | |
4bfc4495 | 2482 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2483 | task_unlock(current); |
15d77835 | 2484 | if (ret) |
5c0c1654 | 2485 | goto put_new; |
71fe804b LS |
2486 | |
2487 | /* Create pseudo-vma that contains just the policy */ | |
2488 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2489 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2490 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2491 | |
5c0c1654 | 2492 | put_new: |
71fe804b | 2493 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2494 | free_scratch: |
4bfc4495 | 2495 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2496 | put_mpol: |
2497 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2498 | } |
2499 | } | |
2500 | ||
1da177e4 LT |
2501 | int mpol_set_shared_policy(struct shared_policy *info, |
2502 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2503 | { | |
2504 | int err; | |
2505 | struct sp_node *new = NULL; | |
2506 | unsigned long sz = vma_pages(vma); | |
2507 | ||
028fec41 | 2508 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2509 | vma->vm_pgoff, |
45c4745a | 2510 | sz, npol ? npol->mode : -1, |
028fec41 | 2511 | npol ? npol->flags : -1, |
00ef2d2f | 2512 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2513 | |
2514 | if (npol) { | |
2515 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2516 | if (!new) | |
2517 | return -ENOMEM; | |
2518 | } | |
2519 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2520 | if (err && new) | |
63f74ca2 | 2521 | sp_free(new); |
1da177e4 LT |
2522 | return err; |
2523 | } | |
2524 | ||
2525 | /* Free a backing policy store on inode delete. */ | |
2526 | void mpol_free_shared_policy(struct shared_policy *p) | |
2527 | { | |
2528 | struct sp_node *n; | |
2529 | struct rb_node *next; | |
2530 | ||
2531 | if (!p->root.rb_node) | |
2532 | return; | |
42288fe3 | 2533 | spin_lock(&p->lock); |
1da177e4 LT |
2534 | next = rb_first(&p->root); |
2535 | while (next) { | |
2536 | n = rb_entry(next, struct sp_node, nd); | |
2537 | next = rb_next(&n->nd); | |
63f74ca2 | 2538 | sp_delete(p, n); |
1da177e4 | 2539 | } |
42288fe3 | 2540 | spin_unlock(&p->lock); |
1da177e4 LT |
2541 | } |
2542 | ||
1a687c2e | 2543 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2544 | static int __initdata numabalancing_override; |
1a687c2e MG |
2545 | |
2546 | static void __init check_numabalancing_enable(void) | |
2547 | { | |
2548 | bool numabalancing_default = false; | |
2549 | ||
2550 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2551 | numabalancing_default = true; | |
2552 | ||
c297663c MG |
2553 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2554 | if (numabalancing_override) | |
2555 | set_numabalancing_state(numabalancing_override == 1); | |
2556 | ||
1a687c2e | 2557 | if (nr_node_ids > 1 && !numabalancing_override) { |
4a404bea | 2558 | pr_info("%s automatic NUMA balancing. " |
c297663c MG |
2559 | "Configure with numa_balancing= or the " |
2560 | "kernel.numa_balancing sysctl", | |
2561 | numabalancing_default ? "Enabling" : "Disabling"); | |
1a687c2e MG |
2562 | set_numabalancing_state(numabalancing_default); |
2563 | } | |
2564 | } | |
2565 | ||
2566 | static int __init setup_numabalancing(char *str) | |
2567 | { | |
2568 | int ret = 0; | |
2569 | if (!str) | |
2570 | goto out; | |
1a687c2e MG |
2571 | |
2572 | if (!strcmp(str, "enable")) { | |
c297663c | 2573 | numabalancing_override = 1; |
1a687c2e MG |
2574 | ret = 1; |
2575 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2576 | numabalancing_override = -1; |
1a687c2e MG |
2577 | ret = 1; |
2578 | } | |
2579 | out: | |
2580 | if (!ret) | |
4a404bea | 2581 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2582 | |
2583 | return ret; | |
2584 | } | |
2585 | __setup("numa_balancing=", setup_numabalancing); | |
2586 | #else | |
2587 | static inline void __init check_numabalancing_enable(void) | |
2588 | { | |
2589 | } | |
2590 | #endif /* CONFIG_NUMA_BALANCING */ | |
2591 | ||
1da177e4 LT |
2592 | /* assumes fs == KERNEL_DS */ |
2593 | void __init numa_policy_init(void) | |
2594 | { | |
b71636e2 PM |
2595 | nodemask_t interleave_nodes; |
2596 | unsigned long largest = 0; | |
2597 | int nid, prefer = 0; | |
2598 | ||
1da177e4 LT |
2599 | policy_cache = kmem_cache_create("numa_policy", |
2600 | sizeof(struct mempolicy), | |
20c2df83 | 2601 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2602 | |
2603 | sn_cache = kmem_cache_create("shared_policy_node", | |
2604 | sizeof(struct sp_node), | |
20c2df83 | 2605 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2606 | |
5606e387 MG |
2607 | for_each_node(nid) { |
2608 | preferred_node_policy[nid] = (struct mempolicy) { | |
2609 | .refcnt = ATOMIC_INIT(1), | |
2610 | .mode = MPOL_PREFERRED, | |
2611 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2612 | .v = { .preferred_node = nid, }, | |
2613 | }; | |
2614 | } | |
2615 | ||
b71636e2 PM |
2616 | /* |
2617 | * Set interleaving policy for system init. Interleaving is only | |
2618 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2619 | * fall back to the largest node if they're all smaller. | |
2620 | */ | |
2621 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2622 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2623 | unsigned long total_pages = node_present_pages(nid); |
2624 | ||
2625 | /* Preserve the largest node */ | |
2626 | if (largest < total_pages) { | |
2627 | largest = total_pages; | |
2628 | prefer = nid; | |
2629 | } | |
2630 | ||
2631 | /* Interleave this node? */ | |
2632 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2633 | node_set(nid, interleave_nodes); | |
2634 | } | |
2635 | ||
2636 | /* All too small, use the largest */ | |
2637 | if (unlikely(nodes_empty(interleave_nodes))) | |
2638 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2639 | |
028fec41 | 2640 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2641 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2642 | |
2643 | check_numabalancing_enable(); | |
1da177e4 LT |
2644 | } |
2645 | ||
8bccd85f | 2646 | /* Reset policy of current process to default */ |
1da177e4 LT |
2647 | void numa_default_policy(void) |
2648 | { | |
028fec41 | 2649 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2650 | } |
68860ec1 | 2651 | |
095f1fc4 LS |
2652 | /* |
2653 | * Parse and format mempolicy from/to strings | |
2654 | */ | |
2655 | ||
1a75a6c8 | 2656 | /* |
f2a07f40 | 2657 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2658 | */ |
345ace9c LS |
2659 | static const char * const policy_modes[] = |
2660 | { | |
2661 | [MPOL_DEFAULT] = "default", | |
2662 | [MPOL_PREFERRED] = "prefer", | |
2663 | [MPOL_BIND] = "bind", | |
2664 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2665 | [MPOL_LOCAL] = "local", |
345ace9c | 2666 | }; |
1a75a6c8 | 2667 | |
095f1fc4 LS |
2668 | |
2669 | #ifdef CONFIG_TMPFS | |
2670 | /** | |
f2a07f40 | 2671 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2672 | * @str: string containing mempolicy to parse |
71fe804b | 2673 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2674 | * |
2675 | * Format of input: | |
2676 | * <mode>[=<flags>][:<nodelist>] | |
2677 | * | |
71fe804b | 2678 | * On success, returns 0, else 1 |
095f1fc4 | 2679 | */ |
a7a88b23 | 2680 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2681 | { |
71fe804b | 2682 | struct mempolicy *new = NULL; |
b4652e84 | 2683 | unsigned short mode; |
f2a07f40 | 2684 | unsigned short mode_flags; |
71fe804b | 2685 | nodemask_t nodes; |
095f1fc4 LS |
2686 | char *nodelist = strchr(str, ':'); |
2687 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2688 | int err = 1; |
2689 | ||
2690 | if (nodelist) { | |
2691 | /* NUL-terminate mode or flags string */ | |
2692 | *nodelist++ = '\0'; | |
71fe804b | 2693 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2694 | goto out; |
01f13bd6 | 2695 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2696 | goto out; |
71fe804b LS |
2697 | } else |
2698 | nodes_clear(nodes); | |
2699 | ||
095f1fc4 LS |
2700 | if (flags) |
2701 | *flags++ = '\0'; /* terminate mode string */ | |
2702 | ||
479e2802 | 2703 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2704 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2705 | break; |
2706 | } | |
2707 | } | |
a720094d | 2708 | if (mode >= MPOL_MAX) |
095f1fc4 LS |
2709 | goto out; |
2710 | ||
71fe804b | 2711 | switch (mode) { |
095f1fc4 | 2712 | case MPOL_PREFERRED: |
71fe804b LS |
2713 | /* |
2714 | * Insist on a nodelist of one node only | |
2715 | */ | |
095f1fc4 LS |
2716 | if (nodelist) { |
2717 | char *rest = nodelist; | |
2718 | while (isdigit(*rest)) | |
2719 | rest++; | |
926f2ae0 KM |
2720 | if (*rest) |
2721 | goto out; | |
095f1fc4 LS |
2722 | } |
2723 | break; | |
095f1fc4 LS |
2724 | case MPOL_INTERLEAVE: |
2725 | /* | |
2726 | * Default to online nodes with memory if no nodelist | |
2727 | */ | |
2728 | if (!nodelist) | |
01f13bd6 | 2729 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2730 | break; |
71fe804b | 2731 | case MPOL_LOCAL: |
3f226aa1 | 2732 | /* |
71fe804b | 2733 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2734 | */ |
71fe804b | 2735 | if (nodelist) |
3f226aa1 | 2736 | goto out; |
71fe804b | 2737 | mode = MPOL_PREFERRED; |
3f226aa1 | 2738 | break; |
413b43de RT |
2739 | case MPOL_DEFAULT: |
2740 | /* | |
2741 | * Insist on a empty nodelist | |
2742 | */ | |
2743 | if (!nodelist) | |
2744 | err = 0; | |
2745 | goto out; | |
d69b2e63 KM |
2746 | case MPOL_BIND: |
2747 | /* | |
2748 | * Insist on a nodelist | |
2749 | */ | |
2750 | if (!nodelist) | |
2751 | goto out; | |
095f1fc4 LS |
2752 | } |
2753 | ||
71fe804b | 2754 | mode_flags = 0; |
095f1fc4 LS |
2755 | if (flags) { |
2756 | /* | |
2757 | * Currently, we only support two mutually exclusive | |
2758 | * mode flags. | |
2759 | */ | |
2760 | if (!strcmp(flags, "static")) | |
71fe804b | 2761 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2762 | else if (!strcmp(flags, "relative")) |
71fe804b | 2763 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2764 | else |
926f2ae0 | 2765 | goto out; |
095f1fc4 | 2766 | } |
71fe804b LS |
2767 | |
2768 | new = mpol_new(mode, mode_flags, &nodes); | |
2769 | if (IS_ERR(new)) | |
926f2ae0 KM |
2770 | goto out; |
2771 | ||
f2a07f40 HD |
2772 | /* |
2773 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2774 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2775 | */ | |
2776 | if (mode != MPOL_PREFERRED) | |
2777 | new->v.nodes = nodes; | |
2778 | else if (nodelist) | |
2779 | new->v.preferred_node = first_node(nodes); | |
2780 | else | |
2781 | new->flags |= MPOL_F_LOCAL; | |
2782 | ||
2783 | /* | |
2784 | * Save nodes for contextualization: this will be used to "clone" | |
2785 | * the mempolicy in a specific context [cpuset] at a later time. | |
2786 | */ | |
2787 | new->w.user_nodemask = nodes; | |
2788 | ||
926f2ae0 | 2789 | err = 0; |
71fe804b | 2790 | |
095f1fc4 LS |
2791 | out: |
2792 | /* Restore string for error message */ | |
2793 | if (nodelist) | |
2794 | *--nodelist = ':'; | |
2795 | if (flags) | |
2796 | *--flags = '='; | |
71fe804b LS |
2797 | if (!err) |
2798 | *mpol = new; | |
095f1fc4 LS |
2799 | return err; |
2800 | } | |
2801 | #endif /* CONFIG_TMPFS */ | |
2802 | ||
71fe804b LS |
2803 | /** |
2804 | * mpol_to_str - format a mempolicy structure for printing | |
2805 | * @buffer: to contain formatted mempolicy string | |
2806 | * @maxlen: length of @buffer | |
2807 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2808 | * |
948927ee DR |
2809 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
2810 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
2811 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 2812 | */ |
948927ee | 2813 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2814 | { |
2815 | char *p = buffer; | |
948927ee DR |
2816 | nodemask_t nodes = NODE_MASK_NONE; |
2817 | unsigned short mode = MPOL_DEFAULT; | |
2818 | unsigned short flags = 0; | |
2291990a | 2819 | |
8790c71a | 2820 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 2821 | mode = pol->mode; |
948927ee DR |
2822 | flags = pol->flags; |
2823 | } | |
bea904d5 | 2824 | |
1a75a6c8 CL |
2825 | switch (mode) { |
2826 | case MPOL_DEFAULT: | |
1a75a6c8 | 2827 | break; |
1a75a6c8 | 2828 | case MPOL_PREFERRED: |
fc36b8d3 | 2829 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 2830 | mode = MPOL_LOCAL; |
53f2556b | 2831 | else |
fc36b8d3 | 2832 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 2833 | break; |
1a75a6c8 | 2834 | case MPOL_BIND: |
1a75a6c8 | 2835 | case MPOL_INTERLEAVE: |
f2a07f40 | 2836 | nodes = pol->v.nodes; |
1a75a6c8 | 2837 | break; |
1a75a6c8 | 2838 | default: |
948927ee DR |
2839 | WARN_ON_ONCE(1); |
2840 | snprintf(p, maxlen, "unknown"); | |
2841 | return; | |
1a75a6c8 CL |
2842 | } |
2843 | ||
b7a9f420 | 2844 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 2845 | |
fc36b8d3 | 2846 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 2847 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 2848 | |
2291990a LS |
2849 | /* |
2850 | * Currently, the only defined flags are mutually exclusive | |
2851 | */ | |
f5b087b5 | 2852 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2853 | p += snprintf(p, buffer + maxlen - p, "static"); |
2854 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2855 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2856 | } |
2857 | ||
1a75a6c8 | 2858 | if (!nodes_empty(nodes)) { |
948927ee | 2859 | p += snprintf(p, buffer + maxlen - p, ":"); |
1a75a6c8 CL |
2860 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2861 | } | |
1a75a6c8 | 2862 | } |