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 LT |
28 | * preferred Try a specific node first before normal fallback. |
29 | * As a special case node -1 here means do the allocation | |
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 | ||
68 | #include <linux/mempolicy.h> | |
69 | #include <linux/mm.h> | |
70 | #include <linux/highmem.h> | |
71 | #include <linux/hugetlb.h> | |
72 | #include <linux/kernel.h> | |
73 | #include <linux/sched.h> | |
1da177e4 LT |
74 | #include <linux/nodemask.h> |
75 | #include <linux/cpuset.h> | |
1da177e4 LT |
76 | #include <linux/slab.h> |
77 | #include <linux/string.h> | |
b95f1b31 | 78 | #include <linux/export.h> |
b488893a | 79 | #include <linux/nsproxy.h> |
1da177e4 LT |
80 | #include <linux/interrupt.h> |
81 | #include <linux/init.h> | |
82 | #include <linux/compat.h> | |
dc9aa5b9 | 83 | #include <linux/swap.h> |
1a75a6c8 CL |
84 | #include <linux/seq_file.h> |
85 | #include <linux/proc_fs.h> | |
b20a3503 | 86 | #include <linux/migrate.h> |
62b61f61 | 87 | #include <linux/ksm.h> |
95a402c3 | 88 | #include <linux/rmap.h> |
86c3a764 | 89 | #include <linux/security.h> |
dbcb0f19 | 90 | #include <linux/syscalls.h> |
095f1fc4 | 91 | #include <linux/ctype.h> |
6d9c285a | 92 | #include <linux/mm_inline.h> |
b24f53a0 | 93 | #include <linux/mmu_notifier.h> |
dc9aa5b9 | 94 | |
1da177e4 LT |
95 | #include <asm/tlbflush.h> |
96 | #include <asm/uaccess.h> | |
778d3b0f | 97 | #include <linux/random.h> |
1da177e4 | 98 | |
62695a84 NP |
99 | #include "internal.h" |
100 | ||
38e35860 | 101 | /* Internal flags */ |
dc9aa5b9 | 102 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 103 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 104 | |
fcc234f8 PE |
105 | static struct kmem_cache *policy_cache; |
106 | static struct kmem_cache *sn_cache; | |
1da177e4 | 107 | |
1da177e4 LT |
108 | /* Highest zone. An specific allocation for a zone below that is not |
109 | policied. */ | |
6267276f | 110 | enum zone_type policy_zone = 0; |
1da177e4 | 111 | |
bea904d5 LS |
112 | /* |
113 | * run-time system-wide default policy => local allocation | |
114 | */ | |
e754d79d | 115 | static struct mempolicy default_policy = { |
1da177e4 | 116 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 117 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 118 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
119 | }; |
120 | ||
5606e387 MG |
121 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
122 | ||
123 | static struct mempolicy *get_task_policy(struct task_struct *p) | |
124 | { | |
125 | struct mempolicy *pol = p->mempolicy; | |
126 | int node; | |
127 | ||
128 | if (!pol) { | |
129 | node = numa_node_id(); | |
130 | if (node != -1) | |
131 | pol = &preferred_node_policy[node]; | |
132 | ||
133 | /* preferred_node_policy is not initialised early in boot */ | |
134 | if (!pol->mode) | |
135 | pol = NULL; | |
136 | } | |
137 | ||
138 | return pol; | |
139 | } | |
140 | ||
37012946 DR |
141 | static const struct mempolicy_operations { |
142 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
708c1bbc MX |
143 | /* |
144 | * If read-side task has no lock to protect task->mempolicy, write-side | |
145 | * task will rebind the task->mempolicy by two step. The first step is | |
146 | * setting all the newly nodes, and the second step is cleaning all the | |
147 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
148 | * page. | |
149 | * If we have a lock to protect task->mempolicy in read-side, we do | |
150 | * rebind directly. | |
151 | * | |
152 | * step: | |
153 | * MPOL_REBIND_ONCE - do rebind work at once | |
154 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
155 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
156 | */ | |
157 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes, | |
158 | enum mpol_rebind_step step); | |
37012946 DR |
159 | } mpol_ops[MPOL_MAX]; |
160 | ||
19770b32 | 161 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 162 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 163 | { |
19770b32 | 164 | int nd, k; |
1da177e4 | 165 | |
19770b32 MG |
166 | for_each_node_mask(nd, *nodemask) { |
167 | struct zone *z; | |
168 | ||
169 | for (k = 0; k <= policy_zone; k++) { | |
170 | z = &NODE_DATA(nd)->node_zones[k]; | |
171 | if (z->present_pages > 0) | |
172 | return 1; | |
dd942ae3 | 173 | } |
8af5e2eb | 174 | } |
19770b32 MG |
175 | |
176 | return 0; | |
1da177e4 LT |
177 | } |
178 | ||
f5b087b5 DR |
179 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
180 | { | |
6d556294 | 181 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
182 | } |
183 | ||
184 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
185 | const nodemask_t *rel) | |
186 | { | |
187 | nodemask_t tmp; | |
188 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
189 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
190 | } |
191 | ||
37012946 DR |
192 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
193 | { | |
194 | if (nodes_empty(*nodes)) | |
195 | return -EINVAL; | |
196 | pol->v.nodes = *nodes; | |
197 | return 0; | |
198 | } | |
199 | ||
200 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
201 | { | |
202 | if (!nodes) | |
fc36b8d3 | 203 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
204 | else if (nodes_empty(*nodes)) |
205 | return -EINVAL; /* no allowed nodes */ | |
206 | else | |
207 | pol->v.preferred_node = first_node(*nodes); | |
208 | return 0; | |
209 | } | |
210 | ||
211 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
212 | { | |
213 | if (!is_valid_nodemask(nodes)) | |
214 | return -EINVAL; | |
215 | pol->v.nodes = *nodes; | |
216 | return 0; | |
217 | } | |
218 | ||
58568d2a MX |
219 | /* |
220 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
221 | * any, for the new policy. mpol_new() has already validated the nodes | |
222 | * parameter with respect to the policy mode and flags. But, we need to | |
223 | * handle an empty nodemask with MPOL_PREFERRED here. | |
224 | * | |
225 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
226 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
227 | */ | |
4bfc4495 KH |
228 | static int mpol_set_nodemask(struct mempolicy *pol, |
229 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 230 | { |
58568d2a MX |
231 | int ret; |
232 | ||
233 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
234 | if (pol == NULL) | |
235 | return 0; | |
01f13bd6 | 236 | /* Check N_MEMORY */ |
4bfc4495 | 237 | nodes_and(nsc->mask1, |
01f13bd6 | 238 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
239 | |
240 | VM_BUG_ON(!nodes); | |
241 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
242 | nodes = NULL; /* explicit local allocation */ | |
243 | else { | |
244 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
4bfc4495 | 245 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
58568d2a | 246 | else |
4bfc4495 KH |
247 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
248 | ||
58568d2a MX |
249 | if (mpol_store_user_nodemask(pol)) |
250 | pol->w.user_nodemask = *nodes; | |
251 | else | |
252 | pol->w.cpuset_mems_allowed = | |
253 | cpuset_current_mems_allowed; | |
254 | } | |
255 | ||
4bfc4495 KH |
256 | if (nodes) |
257 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
258 | else | |
259 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
260 | return ret; |
261 | } | |
262 | ||
263 | /* | |
264 | * This function just creates a new policy, does some check and simple | |
265 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
266 | */ | |
028fec41 DR |
267 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
268 | nodemask_t *nodes) | |
1da177e4 LT |
269 | { |
270 | struct mempolicy *policy; | |
271 | ||
028fec41 DR |
272 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
273 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 274 | |
3e1f0645 DR |
275 | if (mode == MPOL_DEFAULT) { |
276 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 277 | return ERR_PTR(-EINVAL); |
d3a71033 | 278 | return NULL; |
37012946 | 279 | } |
3e1f0645 DR |
280 | VM_BUG_ON(!nodes); |
281 | ||
282 | /* | |
283 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
284 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
285 | * All other modes require a valid pointer to a non-empty nodemask. | |
286 | */ | |
287 | if (mode == MPOL_PREFERRED) { | |
288 | if (nodes_empty(*nodes)) { | |
289 | if (((flags & MPOL_F_STATIC_NODES) || | |
290 | (flags & MPOL_F_RELATIVE_NODES))) | |
291 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 292 | } |
479e2802 PZ |
293 | } else if (mode == MPOL_LOCAL) { |
294 | if (!nodes_empty(*nodes)) | |
295 | return ERR_PTR(-EINVAL); | |
296 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
297 | } else if (nodes_empty(*nodes)) |
298 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
299 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
300 | if (!policy) | |
301 | return ERR_PTR(-ENOMEM); | |
302 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 303 | policy->mode = mode; |
3e1f0645 | 304 | policy->flags = flags; |
37012946 | 305 | |
1da177e4 | 306 | return policy; |
37012946 DR |
307 | } |
308 | ||
52cd3b07 LS |
309 | /* Slow path of a mpol destructor. */ |
310 | void __mpol_put(struct mempolicy *p) | |
311 | { | |
312 | if (!atomic_dec_and_test(&p->refcnt)) | |
313 | return; | |
52cd3b07 LS |
314 | kmem_cache_free(policy_cache, p); |
315 | } | |
316 | ||
708c1bbc MX |
317 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
318 | enum mpol_rebind_step step) | |
37012946 DR |
319 | { |
320 | } | |
321 | ||
708c1bbc MX |
322 | /* |
323 | * step: | |
324 | * MPOL_REBIND_ONCE - do rebind work at once | |
325 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
326 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
327 | */ | |
328 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
329 | enum mpol_rebind_step step) | |
37012946 DR |
330 | { |
331 | nodemask_t tmp; | |
332 | ||
333 | if (pol->flags & MPOL_F_STATIC_NODES) | |
334 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
335 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
336 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
337 | else { | |
708c1bbc MX |
338 | /* |
339 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
340 | * result | |
341 | */ | |
342 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
343 | nodes_remap(tmp, pol->v.nodes, | |
344 | pol->w.cpuset_mems_allowed, *nodes); | |
345 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
346 | } else if (step == MPOL_REBIND_STEP2) { | |
347 | tmp = pol->w.cpuset_mems_allowed; | |
348 | pol->w.cpuset_mems_allowed = *nodes; | |
349 | } else | |
350 | BUG(); | |
37012946 | 351 | } |
f5b087b5 | 352 | |
708c1bbc MX |
353 | if (nodes_empty(tmp)) |
354 | tmp = *nodes; | |
355 | ||
356 | if (step == MPOL_REBIND_STEP1) | |
357 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
358 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
359 | pol->v.nodes = tmp; | |
360 | else | |
361 | BUG(); | |
362 | ||
37012946 DR |
363 | if (!node_isset(current->il_next, tmp)) { |
364 | current->il_next = next_node(current->il_next, tmp); | |
365 | if (current->il_next >= MAX_NUMNODES) | |
366 | current->il_next = first_node(tmp); | |
367 | if (current->il_next >= MAX_NUMNODES) | |
368 | current->il_next = numa_node_id(); | |
369 | } | |
370 | } | |
371 | ||
372 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
373 | const nodemask_t *nodes, |
374 | enum mpol_rebind_step step) | |
37012946 DR |
375 | { |
376 | nodemask_t tmp; | |
377 | ||
37012946 DR |
378 | if (pol->flags & MPOL_F_STATIC_NODES) { |
379 | int node = first_node(pol->w.user_nodemask); | |
380 | ||
fc36b8d3 | 381 | if (node_isset(node, *nodes)) { |
37012946 | 382 | pol->v.preferred_node = node; |
fc36b8d3 LS |
383 | pol->flags &= ~MPOL_F_LOCAL; |
384 | } else | |
385 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
386 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
387 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
388 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 389 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
390 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
391 | pol->w.cpuset_mems_allowed, | |
392 | *nodes); | |
393 | pol->w.cpuset_mems_allowed = *nodes; | |
394 | } | |
1da177e4 LT |
395 | } |
396 | ||
708c1bbc MX |
397 | /* |
398 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
399 | * | |
400 | * If read-side task has no lock to protect task->mempolicy, write-side | |
401 | * task will rebind the task->mempolicy by two step. The first step is | |
402 | * setting all the newly nodes, and the second step is cleaning all the | |
403 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
404 | * page. | |
405 | * If we have a lock to protect task->mempolicy in read-side, we do | |
406 | * rebind directly. | |
407 | * | |
408 | * step: | |
409 | * MPOL_REBIND_ONCE - do rebind work at once | |
410 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
411 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
412 | */ | |
413 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
414 | enum mpol_rebind_step step) | |
1d0d2680 | 415 | { |
1d0d2680 DR |
416 | if (!pol) |
417 | return; | |
89c522c7 | 418 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
1d0d2680 DR |
419 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
420 | return; | |
708c1bbc MX |
421 | |
422 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
423 | return; | |
424 | ||
425 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
426 | BUG(); | |
427 | ||
428 | if (step == MPOL_REBIND_STEP1) | |
429 | pol->flags |= MPOL_F_REBINDING; | |
430 | else if (step == MPOL_REBIND_STEP2) | |
431 | pol->flags &= ~MPOL_F_REBINDING; | |
432 | else if (step >= MPOL_REBIND_NSTEP) | |
433 | BUG(); | |
434 | ||
435 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
436 | } |
437 | ||
438 | /* | |
439 | * Wrapper for mpol_rebind_policy() that just requires task | |
440 | * pointer, and updates task mempolicy. | |
58568d2a MX |
441 | * |
442 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
443 | */ |
444 | ||
708c1bbc MX |
445 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
446 | enum mpol_rebind_step step) | |
1d0d2680 | 447 | { |
708c1bbc | 448 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
449 | } |
450 | ||
451 | /* | |
452 | * Rebind each vma in mm to new nodemask. | |
453 | * | |
454 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
455 | */ | |
456 | ||
457 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
458 | { | |
459 | struct vm_area_struct *vma; | |
460 | ||
461 | down_write(&mm->mmap_sem); | |
462 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 463 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
464 | up_write(&mm->mmap_sem); |
465 | } | |
466 | ||
37012946 DR |
467 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
468 | [MPOL_DEFAULT] = { | |
469 | .rebind = mpol_rebind_default, | |
470 | }, | |
471 | [MPOL_INTERLEAVE] = { | |
472 | .create = mpol_new_interleave, | |
473 | .rebind = mpol_rebind_nodemask, | |
474 | }, | |
475 | [MPOL_PREFERRED] = { | |
476 | .create = mpol_new_preferred, | |
477 | .rebind = mpol_rebind_preferred, | |
478 | }, | |
479 | [MPOL_BIND] = { | |
480 | .create = mpol_new_bind, | |
481 | .rebind = mpol_rebind_nodemask, | |
482 | }, | |
483 | }; | |
484 | ||
fc301289 CL |
485 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
486 | unsigned long flags); | |
1a75a6c8 | 487 | |
38e35860 | 488 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 489 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
490 | unsigned long addr, unsigned long end, |
491 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 492 | void *private) |
1da177e4 | 493 | { |
91612e0d HD |
494 | pte_t *orig_pte; |
495 | pte_t *pte; | |
705e87c0 | 496 | spinlock_t *ptl; |
941150a3 | 497 | |
705e87c0 | 498 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 499 | do { |
6aab341e | 500 | struct page *page; |
25ba77c1 | 501 | int nid; |
91612e0d HD |
502 | |
503 | if (!pte_present(*pte)) | |
1da177e4 | 504 | continue; |
6aab341e LT |
505 | page = vm_normal_page(vma, addr, *pte); |
506 | if (!page) | |
1da177e4 | 507 | continue; |
053837fc | 508 | /* |
62b61f61 HD |
509 | * vm_normal_page() filters out zero pages, but there might |
510 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
511 | * And we cannot move PageKsm pages sensibly or safely yet. | |
053837fc | 512 | */ |
62b61f61 | 513 | if (PageReserved(page) || PageKsm(page)) |
f4598c8b | 514 | continue; |
6aab341e | 515 | nid = page_to_nid(page); |
38e35860 CL |
516 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
517 | continue; | |
518 | ||
b1f72d18 | 519 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 520 | migrate_page_add(page, private, flags); |
38e35860 CL |
521 | else |
522 | break; | |
91612e0d | 523 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 524 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
525 | return addr != end; |
526 | } | |
527 | ||
b5810039 | 528 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
529 | unsigned long addr, unsigned long end, |
530 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 531 | void *private) |
91612e0d HD |
532 | { |
533 | pmd_t *pmd; | |
534 | unsigned long next; | |
535 | ||
536 | pmd = pmd_offset(pud, addr); | |
537 | do { | |
538 | next = pmd_addr_end(addr, end); | |
e180377f | 539 | split_huge_page_pmd(vma, addr, pmd); |
1a5a9906 | 540 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
91612e0d | 541 | continue; |
dc9aa5b9 | 542 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 543 | flags, private)) |
91612e0d HD |
544 | return -EIO; |
545 | } while (pmd++, addr = next, addr != end); | |
546 | return 0; | |
547 | } | |
548 | ||
b5810039 | 549 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
550 | unsigned long addr, unsigned long end, |
551 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 552 | void *private) |
91612e0d HD |
553 | { |
554 | pud_t *pud; | |
555 | unsigned long next; | |
556 | ||
557 | pud = pud_offset(pgd, addr); | |
558 | do { | |
559 | next = pud_addr_end(addr, end); | |
560 | if (pud_none_or_clear_bad(pud)) | |
561 | continue; | |
dc9aa5b9 | 562 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 563 | flags, private)) |
91612e0d HD |
564 | return -EIO; |
565 | } while (pud++, addr = next, addr != end); | |
566 | return 0; | |
567 | } | |
568 | ||
b5810039 | 569 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
570 | unsigned long addr, unsigned long end, |
571 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 572 | void *private) |
91612e0d HD |
573 | { |
574 | pgd_t *pgd; | |
575 | unsigned long next; | |
576 | ||
b5810039 | 577 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
578 | do { |
579 | next = pgd_addr_end(addr, end); | |
580 | if (pgd_none_or_clear_bad(pgd)) | |
581 | continue; | |
dc9aa5b9 | 582 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 583 | flags, private)) |
91612e0d HD |
584 | return -EIO; |
585 | } while (pgd++, addr = next, addr != end); | |
586 | return 0; | |
1da177e4 LT |
587 | } |
588 | ||
b24f53a0 LS |
589 | #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE |
590 | /* | |
4b10e7d5 MG |
591 | * This is used to mark a range of virtual addresses to be inaccessible. |
592 | * These are later cleared by a NUMA hinting fault. Depending on these | |
593 | * faults, pages may be migrated for better NUMA placement. | |
594 | * | |
595 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
596 | * an architecture makes a different choice, it will need further | |
597 | * changes to the core. | |
b24f53a0 | 598 | */ |
4b10e7d5 MG |
599 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
600 | unsigned long addr, unsigned long end) | |
b24f53a0 | 601 | { |
4b10e7d5 MG |
602 | int nr_updated; |
603 | BUILD_BUG_ON(_PAGE_NUMA != _PAGE_PROTNONE); | |
b24f53a0 | 604 | |
4b10e7d5 | 605 | nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1); |
03c5a6e1 MG |
606 | if (nr_updated) |
607 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 608 | |
4b10e7d5 | 609 | return nr_updated; |
b24f53a0 LS |
610 | } |
611 | #else | |
612 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
613 | unsigned long addr, unsigned long end) | |
614 | { | |
615 | return 0; | |
616 | } | |
617 | #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */ | |
618 | ||
dc9aa5b9 CL |
619 | /* |
620 | * Check if all pages in a range are on a set of nodes. | |
621 | * If pagelist != NULL then isolate pages from the LRU and | |
622 | * put them on the pagelist. | |
623 | */ | |
1da177e4 LT |
624 | static struct vm_area_struct * |
625 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 626 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
627 | { |
628 | int err; | |
629 | struct vm_area_struct *first, *vma, *prev; | |
630 | ||
053837fc | 631 | |
1da177e4 LT |
632 | first = find_vma(mm, start); |
633 | if (!first) | |
634 | return ERR_PTR(-EFAULT); | |
635 | prev = NULL; | |
636 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
b24f53a0 LS |
637 | unsigned long endvma = vma->vm_end; |
638 | ||
639 | if (endvma > end) | |
640 | endvma = end; | |
641 | if (vma->vm_start > start) | |
642 | start = vma->vm_start; | |
643 | ||
dc9aa5b9 CL |
644 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
645 | if (!vma->vm_next && vma->vm_end < end) | |
646 | return ERR_PTR(-EFAULT); | |
647 | if (prev && prev->vm_end < vma->vm_start) | |
648 | return ERR_PTR(-EFAULT); | |
649 | } | |
b24f53a0 LS |
650 | |
651 | if (is_vm_hugetlb_page(vma)) | |
652 | goto next; | |
653 | ||
654 | if (flags & MPOL_MF_LAZY) { | |
655 | change_prot_numa(vma, start, endvma); | |
656 | goto next; | |
657 | } | |
658 | ||
659 | if ((flags & MPOL_MF_STRICT) || | |
dc9aa5b9 | 660 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
b24f53a0 | 661 | vma_migratable(vma))) { |
dc9aa5b9 | 662 | |
dc9aa5b9 | 663 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 664 | flags, private); |
1da177e4 LT |
665 | if (err) { |
666 | first = ERR_PTR(err); | |
667 | break; | |
668 | } | |
669 | } | |
b24f53a0 | 670 | next: |
1da177e4 LT |
671 | prev = vma; |
672 | } | |
673 | return first; | |
674 | } | |
675 | ||
869833f2 KM |
676 | /* |
677 | * Apply policy to a single VMA | |
678 | * This must be called with the mmap_sem held for writing. | |
679 | */ | |
680 | static int vma_replace_policy(struct vm_area_struct *vma, | |
681 | struct mempolicy *pol) | |
8d34694c | 682 | { |
869833f2 KM |
683 | int err; |
684 | struct mempolicy *old; | |
685 | struct mempolicy *new; | |
8d34694c KM |
686 | |
687 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
688 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
689 | vma->vm_ops, vma->vm_file, | |
690 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
691 | ||
869833f2 KM |
692 | new = mpol_dup(pol); |
693 | if (IS_ERR(new)) | |
694 | return PTR_ERR(new); | |
695 | ||
696 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 697 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
698 | if (err) |
699 | goto err_out; | |
8d34694c | 700 | } |
869833f2 KM |
701 | |
702 | old = vma->vm_policy; | |
703 | vma->vm_policy = new; /* protected by mmap_sem */ | |
704 | mpol_put(old); | |
705 | ||
706 | return 0; | |
707 | err_out: | |
708 | mpol_put(new); | |
8d34694c KM |
709 | return err; |
710 | } | |
711 | ||
1da177e4 | 712 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
713 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
714 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
715 | { |
716 | struct vm_area_struct *next; | |
9d8cebd4 KM |
717 | struct vm_area_struct *prev; |
718 | struct vm_area_struct *vma; | |
719 | int err = 0; | |
e26a5114 | 720 | pgoff_t pgoff; |
9d8cebd4 KM |
721 | unsigned long vmstart; |
722 | unsigned long vmend; | |
1da177e4 | 723 | |
097d5910 | 724 | vma = find_vma(mm, start); |
9d8cebd4 KM |
725 | if (!vma || vma->vm_start > start) |
726 | return -EFAULT; | |
727 | ||
097d5910 | 728 | prev = vma->vm_prev; |
e26a5114 KM |
729 | if (start > vma->vm_start) |
730 | prev = vma; | |
731 | ||
9d8cebd4 | 732 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 733 | next = vma->vm_next; |
9d8cebd4 KM |
734 | vmstart = max(start, vma->vm_start); |
735 | vmend = min(end, vma->vm_end); | |
736 | ||
e26a5114 KM |
737 | if (mpol_equal(vma_policy(vma), new_pol)) |
738 | continue; | |
739 | ||
740 | pgoff = vma->vm_pgoff + | |
741 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 742 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 743 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 744 | new_pol); |
9d8cebd4 KM |
745 | if (prev) { |
746 | vma = prev; | |
747 | next = vma->vm_next; | |
748 | continue; | |
749 | } | |
750 | if (vma->vm_start != vmstart) { | |
751 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
752 | if (err) | |
753 | goto out; | |
754 | } | |
755 | if (vma->vm_end != vmend) { | |
756 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
757 | if (err) | |
758 | goto out; | |
759 | } | |
869833f2 | 760 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
761 | if (err) |
762 | goto out; | |
1da177e4 | 763 | } |
9d8cebd4 KM |
764 | |
765 | out: | |
1da177e4 LT |
766 | return err; |
767 | } | |
768 | ||
c61afb18 PJ |
769 | /* |
770 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
771 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
772 | * with other PF_* flag bits) on memory allocation hot code paths. | |
773 | * | |
774 | * If called from outside this file, the task 'p' should -only- be | |
775 | * a newly forked child not yet visible on the task list, because | |
776 | * manipulating the task flags of a visible task is not safe. | |
777 | * | |
778 | * The above limitation is why this routine has the funny name | |
779 | * mpol_fix_fork_child_flag(). | |
780 | * | |
781 | * It is also safe to call this with a task pointer of current, | |
782 | * which the static wrapper mpol_set_task_struct_flag() does, | |
783 | * for use within this file. | |
784 | */ | |
785 | ||
786 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
787 | { | |
788 | if (p->mempolicy) | |
789 | p->flags |= PF_MEMPOLICY; | |
790 | else | |
791 | p->flags &= ~PF_MEMPOLICY; | |
792 | } | |
793 | ||
794 | static void mpol_set_task_struct_flag(void) | |
795 | { | |
796 | mpol_fix_fork_child_flag(current); | |
797 | } | |
798 | ||
1da177e4 | 799 | /* Set the process memory policy */ |
028fec41 DR |
800 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
801 | nodemask_t *nodes) | |
1da177e4 | 802 | { |
58568d2a | 803 | struct mempolicy *new, *old; |
f4e53d91 | 804 | struct mm_struct *mm = current->mm; |
4bfc4495 | 805 | NODEMASK_SCRATCH(scratch); |
58568d2a | 806 | int ret; |
1da177e4 | 807 | |
4bfc4495 KH |
808 | if (!scratch) |
809 | return -ENOMEM; | |
f4e53d91 | 810 | |
4bfc4495 KH |
811 | new = mpol_new(mode, flags, nodes); |
812 | if (IS_ERR(new)) { | |
813 | ret = PTR_ERR(new); | |
814 | goto out; | |
815 | } | |
f4e53d91 LS |
816 | /* |
817 | * prevent changing our mempolicy while show_numa_maps() | |
818 | * is using it. | |
819 | * Note: do_set_mempolicy() can be called at init time | |
820 | * with no 'mm'. | |
821 | */ | |
822 | if (mm) | |
823 | down_write(&mm->mmap_sem); | |
58568d2a | 824 | task_lock(current); |
4bfc4495 | 825 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
826 | if (ret) { |
827 | task_unlock(current); | |
828 | if (mm) | |
829 | up_write(&mm->mmap_sem); | |
830 | mpol_put(new); | |
4bfc4495 | 831 | goto out; |
58568d2a MX |
832 | } |
833 | old = current->mempolicy; | |
1da177e4 | 834 | current->mempolicy = new; |
c61afb18 | 835 | mpol_set_task_struct_flag(); |
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 | { |
6484eb3e | 1002 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
1003 | } |
1004 | ||
7e2ab150 CL |
1005 | /* |
1006 | * Migrate pages from one node to a target node. | |
1007 | * Returns error or the number of pages not migrated. | |
1008 | */ | |
dbcb0f19 AB |
1009 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1010 | int flags) | |
7e2ab150 CL |
1011 | { |
1012 | nodemask_t nmask; | |
1013 | LIST_HEAD(pagelist); | |
1014 | int err = 0; | |
1015 | ||
1016 | nodes_clear(nmask); | |
1017 | node_set(source, nmask); | |
6ce3c4c0 | 1018 | |
08270807 MK |
1019 | /* |
1020 | * This does not "check" the range but isolates all pages that | |
1021 | * need migration. Between passing in the full user address | |
1022 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1023 | */ | |
1024 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
1025 | check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, | |
7e2ab150 CL |
1026 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1027 | ||
cf608ac1 | 1028 | if (!list_empty(&pagelist)) { |
7f0f2496 | 1029 | err = migrate_pages(&pagelist, new_node_page, dest, |
7b2a2d4a MG |
1030 | false, MIGRATE_SYNC, |
1031 | MR_SYSCALL); | |
cf608ac1 MK |
1032 | if (err) |
1033 | putback_lru_pages(&pagelist); | |
1034 | } | |
95a402c3 | 1035 | |
7e2ab150 | 1036 | return err; |
6ce3c4c0 CL |
1037 | } |
1038 | ||
39743889 | 1039 | /* |
7e2ab150 CL |
1040 | * Move pages between the two nodesets so as to preserve the physical |
1041 | * layout as much as possible. | |
39743889 CL |
1042 | * |
1043 | * Returns the number of page that could not be moved. | |
1044 | */ | |
0ce72d4f AM |
1045 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1046 | const nodemask_t *to, int flags) | |
39743889 | 1047 | { |
7e2ab150 | 1048 | int busy = 0; |
0aedadf9 | 1049 | int err; |
7e2ab150 | 1050 | nodemask_t tmp; |
39743889 | 1051 | |
0aedadf9 CL |
1052 | err = migrate_prep(); |
1053 | if (err) | |
1054 | return err; | |
1055 | ||
53f2556b | 1056 | down_read(&mm->mmap_sem); |
39743889 | 1057 | |
0ce72d4f | 1058 | err = migrate_vmas(mm, from, to, flags); |
7b2259b3 CL |
1059 | if (err) |
1060 | goto out; | |
1061 | ||
da0aa138 KM |
1062 | /* |
1063 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1064 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1065 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1066 | * The pair of nodemasks 'to' and 'from' define the map. | |
1067 | * | |
1068 | * If no pair of bits is found that way, fallback to picking some | |
1069 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1070 | * 'source' and 'dest' bits are the same, this represents a node | |
1071 | * that will be migrating to itself, so no pages need move. | |
1072 | * | |
1073 | * If no bits are left in 'tmp', or if all remaining bits left | |
1074 | * in 'tmp' correspond to the same bit in 'to', return false | |
1075 | * (nothing left to migrate). | |
1076 | * | |
1077 | * This lets us pick a pair of nodes to migrate between, such that | |
1078 | * if possible the dest node is not already occupied by some other | |
1079 | * source node, minimizing the risk of overloading the memory on a | |
1080 | * node that would happen if we migrated incoming memory to a node | |
1081 | * before migrating outgoing memory source that same node. | |
1082 | * | |
1083 | * A single scan of tmp is sufficient. As we go, we remember the | |
1084 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1085 | * that not only moved, but what's better, moved to an empty slot | |
1086 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1087 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1088 | * most recent <s, d> pair that moved. If we get all the way through |
1089 | * the scan of tmp without finding any node that moved, much less | |
1090 | * moved to an empty node, then there is nothing left worth migrating. | |
1091 | */ | |
d4984711 | 1092 | |
0ce72d4f | 1093 | tmp = *from; |
7e2ab150 CL |
1094 | while (!nodes_empty(tmp)) { |
1095 | int s,d; | |
1096 | int source = -1; | |
1097 | int dest = 0; | |
1098 | ||
1099 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1100 | |
1101 | /* | |
1102 | * do_migrate_pages() tries to maintain the relative | |
1103 | * node relationship of the pages established between | |
1104 | * threads and memory areas. | |
1105 | * | |
1106 | * However if the number of source nodes is not equal to | |
1107 | * the number of destination nodes we can not preserve | |
1108 | * this node relative relationship. In that case, skip | |
1109 | * copying memory from a node that is in the destination | |
1110 | * mask. | |
1111 | * | |
1112 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1113 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1114 | */ | |
1115 | ||
0ce72d4f AM |
1116 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1117 | (node_isset(s, *to))) | |
4a5b18cc LW |
1118 | continue; |
1119 | ||
0ce72d4f | 1120 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1121 | if (s == d) |
1122 | continue; | |
1123 | ||
1124 | source = s; /* Node moved. Memorize */ | |
1125 | dest = d; | |
1126 | ||
1127 | /* dest not in remaining from nodes? */ | |
1128 | if (!node_isset(dest, tmp)) | |
1129 | break; | |
1130 | } | |
1131 | if (source == -1) | |
1132 | break; | |
1133 | ||
1134 | node_clear(source, tmp); | |
1135 | err = migrate_to_node(mm, source, dest, flags); | |
1136 | if (err > 0) | |
1137 | busy += err; | |
1138 | if (err < 0) | |
1139 | break; | |
39743889 | 1140 | } |
7b2259b3 | 1141 | out: |
39743889 | 1142 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1143 | if (err < 0) |
1144 | return err; | |
1145 | return busy; | |
b20a3503 CL |
1146 | |
1147 | } | |
1148 | ||
3ad33b24 LS |
1149 | /* |
1150 | * Allocate a new page for page migration based on vma policy. | |
1151 | * Start assuming that page is mapped by vma pointed to by @private. | |
1152 | * Search forward from there, if not. N.B., this assumes that the | |
1153 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1154 | * is in virtual address order. | |
1155 | */ | |
742755a1 | 1156 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1157 | { |
1158 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 1159 | unsigned long uninitialized_var(address); |
95a402c3 | 1160 | |
3ad33b24 LS |
1161 | while (vma) { |
1162 | address = page_address_in_vma(page, vma); | |
1163 | if (address != -EFAULT) | |
1164 | break; | |
1165 | vma = vma->vm_next; | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * if !vma, alloc_page_vma() will use task or system default policy | |
1170 | */ | |
1171 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 1172 | } |
b20a3503 CL |
1173 | #else |
1174 | ||
1175 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1176 | unsigned long flags) | |
1177 | { | |
39743889 CL |
1178 | } |
1179 | ||
0ce72d4f AM |
1180 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1181 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1182 | { |
1183 | return -ENOSYS; | |
1184 | } | |
95a402c3 | 1185 | |
69939749 | 1186 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1187 | { |
1188 | return NULL; | |
1189 | } | |
b20a3503 CL |
1190 | #endif |
1191 | ||
dbcb0f19 | 1192 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1193 | unsigned short mode, unsigned short mode_flags, |
1194 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
1195 | { |
1196 | struct vm_area_struct *vma; | |
1197 | struct mm_struct *mm = current->mm; | |
1198 | struct mempolicy *new; | |
1199 | unsigned long end; | |
1200 | int err; | |
1201 | LIST_HEAD(pagelist); | |
1202 | ||
b24f53a0 | 1203 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1204 | return -EINVAL; |
74c00241 | 1205 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1206 | return -EPERM; |
1207 | ||
1208 | if (start & ~PAGE_MASK) | |
1209 | return -EINVAL; | |
1210 | ||
1211 | if (mode == MPOL_DEFAULT) | |
1212 | flags &= ~MPOL_MF_STRICT; | |
1213 | ||
1214 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1215 | end = start + len; | |
1216 | ||
1217 | if (end < start) | |
1218 | return -EINVAL; | |
1219 | if (end == start) | |
1220 | return 0; | |
1221 | ||
028fec41 | 1222 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1223 | if (IS_ERR(new)) |
1224 | return PTR_ERR(new); | |
1225 | ||
b24f53a0 LS |
1226 | if (flags & MPOL_MF_LAZY) |
1227 | new->flags |= MPOL_F_MOF; | |
1228 | ||
6ce3c4c0 CL |
1229 | /* |
1230 | * If we are using the default policy then operation | |
1231 | * on discontinuous address spaces is okay after all | |
1232 | */ | |
1233 | if (!new) | |
1234 | flags |= MPOL_MF_DISCONTIG_OK; | |
1235 | ||
028fec41 DR |
1236 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1237 | start, start + len, mode, mode_flags, | |
1238 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 | 1239 | |
0aedadf9 CL |
1240 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1241 | ||
1242 | err = migrate_prep(); | |
1243 | if (err) | |
b05ca738 | 1244 | goto mpol_out; |
0aedadf9 | 1245 | } |
4bfc4495 KH |
1246 | { |
1247 | NODEMASK_SCRATCH(scratch); | |
1248 | if (scratch) { | |
1249 | down_write(&mm->mmap_sem); | |
1250 | task_lock(current); | |
1251 | err = mpol_set_nodemask(new, nmask, scratch); | |
1252 | task_unlock(current); | |
1253 | if (err) | |
1254 | up_write(&mm->mmap_sem); | |
1255 | } else | |
1256 | err = -ENOMEM; | |
1257 | NODEMASK_SCRATCH_FREE(scratch); | |
1258 | } | |
b05ca738 KM |
1259 | if (err) |
1260 | goto mpol_out; | |
1261 | ||
6ce3c4c0 CL |
1262 | vma = check_range(mm, start, end, nmask, |
1263 | flags | MPOL_MF_INVERT, &pagelist); | |
1264 | ||
b24f53a0 | 1265 | err = PTR_ERR(vma); /* maybe ... */ |
a720094d | 1266 | if (!IS_ERR(vma)) |
9d8cebd4 | 1267 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1268 | |
b24f53a0 LS |
1269 | if (!err) { |
1270 | int nr_failed = 0; | |
1271 | ||
cf608ac1 | 1272 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1273 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
95a402c3 | 1274 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
7f0f2496 | 1275 | (unsigned long)vma, |
7b2a2d4a MG |
1276 | false, MIGRATE_SYNC, |
1277 | MR_MEMPOLICY_MBIND); | |
cf608ac1 MK |
1278 | if (nr_failed) |
1279 | putback_lru_pages(&pagelist); | |
1280 | } | |
6ce3c4c0 | 1281 | |
b24f53a0 | 1282 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1283 | err = -EIO; |
ab8a3e14 KM |
1284 | } else |
1285 | putback_lru_pages(&pagelist); | |
b20a3503 | 1286 | |
6ce3c4c0 | 1287 | up_write(&mm->mmap_sem); |
b05ca738 | 1288 | mpol_out: |
f0be3d32 | 1289 | mpol_put(new); |
6ce3c4c0 CL |
1290 | return err; |
1291 | } | |
1292 | ||
8bccd85f CL |
1293 | /* |
1294 | * User space interface with variable sized bitmaps for nodelists. | |
1295 | */ | |
1296 | ||
1297 | /* Copy a node mask from user space. */ | |
39743889 | 1298 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1299 | unsigned long maxnode) |
1300 | { | |
1301 | unsigned long k; | |
1302 | unsigned long nlongs; | |
1303 | unsigned long endmask; | |
1304 | ||
1305 | --maxnode; | |
1306 | nodes_clear(*nodes); | |
1307 | if (maxnode == 0 || !nmask) | |
1308 | return 0; | |
a9c930ba | 1309 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1310 | return -EINVAL; |
8bccd85f CL |
1311 | |
1312 | nlongs = BITS_TO_LONGS(maxnode); | |
1313 | if ((maxnode % BITS_PER_LONG) == 0) | |
1314 | endmask = ~0UL; | |
1315 | else | |
1316 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1317 | ||
1318 | /* When the user specified more nodes than supported just check | |
1319 | if the non supported part is all zero. */ | |
1320 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1321 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1322 | return -EINVAL; | |
1323 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1324 | unsigned long t; | |
1325 | if (get_user(t, nmask + k)) | |
1326 | return -EFAULT; | |
1327 | if (k == nlongs - 1) { | |
1328 | if (t & endmask) | |
1329 | return -EINVAL; | |
1330 | } else if (t) | |
1331 | return -EINVAL; | |
1332 | } | |
1333 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1334 | endmask = ~0UL; | |
1335 | } | |
1336 | ||
1337 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1338 | return -EFAULT; | |
1339 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | /* Copy a kernel node mask to user space */ | |
1344 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1345 | nodemask_t *nodes) | |
1346 | { | |
1347 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1348 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1349 | ||
1350 | if (copy > nbytes) { | |
1351 | if (copy > PAGE_SIZE) | |
1352 | return -EINVAL; | |
1353 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1354 | return -EFAULT; | |
1355 | copy = nbytes; | |
1356 | } | |
1357 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1358 | } | |
1359 | ||
938bb9f5 HC |
1360 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1361 | unsigned long, mode, unsigned long __user *, nmask, | |
1362 | unsigned long, maxnode, unsigned, flags) | |
8bccd85f CL |
1363 | { |
1364 | nodemask_t nodes; | |
1365 | int err; | |
028fec41 | 1366 | unsigned short mode_flags; |
8bccd85f | 1367 | |
028fec41 DR |
1368 | mode_flags = mode & MPOL_MODE_FLAGS; |
1369 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1370 | if (mode >= MPOL_MAX) |
1371 | return -EINVAL; | |
4c50bc01 DR |
1372 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1373 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1374 | return -EINVAL; | |
8bccd85f CL |
1375 | err = get_nodes(&nodes, nmask, maxnode); |
1376 | if (err) | |
1377 | return err; | |
028fec41 | 1378 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1379 | } |
1380 | ||
1381 | /* Set the process memory policy */ | |
938bb9f5 HC |
1382 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
1383 | unsigned long, maxnode) | |
8bccd85f CL |
1384 | { |
1385 | int err; | |
1386 | nodemask_t nodes; | |
028fec41 | 1387 | unsigned short flags; |
8bccd85f | 1388 | |
028fec41 DR |
1389 | flags = mode & MPOL_MODE_FLAGS; |
1390 | mode &= ~MPOL_MODE_FLAGS; | |
1391 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1392 | return -EINVAL; |
4c50bc01 DR |
1393 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1394 | return -EINVAL; | |
8bccd85f CL |
1395 | err = get_nodes(&nodes, nmask, maxnode); |
1396 | if (err) | |
1397 | return err; | |
028fec41 | 1398 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1399 | } |
1400 | ||
938bb9f5 HC |
1401 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1402 | const unsigned long __user *, old_nodes, | |
1403 | const unsigned long __user *, new_nodes) | |
39743889 | 1404 | { |
c69e8d9c | 1405 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1406 | struct mm_struct *mm = NULL; |
39743889 | 1407 | struct task_struct *task; |
39743889 CL |
1408 | nodemask_t task_nodes; |
1409 | int err; | |
596d7cfa KM |
1410 | nodemask_t *old; |
1411 | nodemask_t *new; | |
1412 | NODEMASK_SCRATCH(scratch); | |
1413 | ||
1414 | if (!scratch) | |
1415 | return -ENOMEM; | |
39743889 | 1416 | |
596d7cfa KM |
1417 | old = &scratch->mask1; |
1418 | new = &scratch->mask2; | |
1419 | ||
1420 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1421 | if (err) |
596d7cfa | 1422 | goto out; |
39743889 | 1423 | |
596d7cfa | 1424 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1425 | if (err) |
596d7cfa | 1426 | goto out; |
39743889 CL |
1427 | |
1428 | /* Find the mm_struct */ | |
55cfaa3c | 1429 | rcu_read_lock(); |
228ebcbe | 1430 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1431 | if (!task) { |
55cfaa3c | 1432 | rcu_read_unlock(); |
596d7cfa KM |
1433 | err = -ESRCH; |
1434 | goto out; | |
39743889 | 1435 | } |
3268c63e | 1436 | get_task_struct(task); |
39743889 | 1437 | |
596d7cfa | 1438 | err = -EINVAL; |
39743889 CL |
1439 | |
1440 | /* | |
1441 | * Check if this process has the right to modify the specified | |
1442 | * process. The right exists if the process has administrative | |
7f927fcc | 1443 | * capabilities, superuser privileges or the same |
39743889 CL |
1444 | * userid as the target process. |
1445 | */ | |
c69e8d9c | 1446 | tcred = __task_cred(task); |
b38a86eb EB |
1447 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1448 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1449 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1450 | rcu_read_unlock(); |
39743889 | 1451 | err = -EPERM; |
3268c63e | 1452 | goto out_put; |
39743889 | 1453 | } |
c69e8d9c | 1454 | rcu_read_unlock(); |
39743889 CL |
1455 | |
1456 | task_nodes = cpuset_mems_allowed(task); | |
1457 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1458 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1459 | err = -EPERM; |
3268c63e | 1460 | goto out_put; |
39743889 CL |
1461 | } |
1462 | ||
01f13bd6 | 1463 | if (!nodes_subset(*new, node_states[N_MEMORY])) { |
3b42d28b | 1464 | err = -EINVAL; |
3268c63e | 1465 | goto out_put; |
3b42d28b CL |
1466 | } |
1467 | ||
86c3a764 DQ |
1468 | err = security_task_movememory(task); |
1469 | if (err) | |
3268c63e | 1470 | goto out_put; |
86c3a764 | 1471 | |
3268c63e CL |
1472 | mm = get_task_mm(task); |
1473 | put_task_struct(task); | |
f2a9ef88 SL |
1474 | |
1475 | if (!mm) { | |
3268c63e | 1476 | err = -EINVAL; |
f2a9ef88 SL |
1477 | goto out; |
1478 | } | |
1479 | ||
1480 | err = do_migrate_pages(mm, old, new, | |
1481 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1482 | |
1483 | mmput(mm); | |
1484 | out: | |
596d7cfa KM |
1485 | NODEMASK_SCRATCH_FREE(scratch); |
1486 | ||
39743889 | 1487 | return err; |
3268c63e CL |
1488 | |
1489 | out_put: | |
1490 | put_task_struct(task); | |
1491 | goto out; | |
1492 | ||
39743889 CL |
1493 | } |
1494 | ||
1495 | ||
8bccd85f | 1496 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1497 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1498 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1499 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1500 | { |
dbcb0f19 AB |
1501 | int err; |
1502 | int uninitialized_var(pval); | |
8bccd85f CL |
1503 | nodemask_t nodes; |
1504 | ||
1505 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1506 | return -EINVAL; | |
1507 | ||
1508 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1509 | ||
1510 | if (err) | |
1511 | return err; | |
1512 | ||
1513 | if (policy && put_user(pval, policy)) | |
1514 | return -EFAULT; | |
1515 | ||
1516 | if (nmask) | |
1517 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1518 | ||
1519 | return err; | |
1520 | } | |
1521 | ||
1da177e4 LT |
1522 | #ifdef CONFIG_COMPAT |
1523 | ||
1524 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1525 | compat_ulong_t __user *nmask, | |
1526 | compat_ulong_t maxnode, | |
1527 | compat_ulong_t addr, compat_ulong_t flags) | |
1528 | { | |
1529 | long err; | |
1530 | unsigned long __user *nm = NULL; | |
1531 | unsigned long nr_bits, alloc_size; | |
1532 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1533 | ||
1534 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1535 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1536 | ||
1537 | if (nmask) | |
1538 | nm = compat_alloc_user_space(alloc_size); | |
1539 | ||
1540 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1541 | ||
1542 | if (!err && nmask) { | |
2bbff6c7 KH |
1543 | unsigned long copy_size; |
1544 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1545 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1546 | /* ensure entire bitmap is zeroed */ |
1547 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1548 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1549 | } | |
1550 | ||
1551 | return err; | |
1552 | } | |
1553 | ||
1554 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1555 | compat_ulong_t maxnode) | |
1556 | { | |
1557 | long err = 0; | |
1558 | unsigned long __user *nm = NULL; | |
1559 | unsigned long nr_bits, alloc_size; | |
1560 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1561 | ||
1562 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1563 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1564 | ||
1565 | if (nmask) { | |
1566 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1567 | nm = compat_alloc_user_space(alloc_size); | |
1568 | err |= copy_to_user(nm, bm, alloc_size); | |
1569 | } | |
1570 | ||
1571 | if (err) | |
1572 | return -EFAULT; | |
1573 | ||
1574 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1575 | } | |
1576 | ||
1577 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1578 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1579 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1580 | { | |
1581 | long err = 0; | |
1582 | unsigned long __user *nm = NULL; | |
1583 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1584 | nodemask_t bm; |
1da177e4 LT |
1585 | |
1586 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1587 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1588 | ||
1589 | if (nmask) { | |
dfcd3c0d | 1590 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1591 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1592 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | if (err) | |
1596 | return -EFAULT; | |
1597 | ||
1598 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1599 | } | |
1600 | ||
1601 | #endif | |
1602 | ||
480eccf9 LS |
1603 | /* |
1604 | * get_vma_policy(@task, @vma, @addr) | |
1605 | * @task - task for fallback if vma policy == default | |
1606 | * @vma - virtual memory area whose policy is sought | |
1607 | * @addr - address in @vma for shared policy lookup | |
1608 | * | |
1609 | * Returns effective policy for a VMA at specified address. | |
1610 | * Falls back to @task or system default policy, as necessary. | |
32f8516a DR |
1611 | * Current or other task's task mempolicy and non-shared vma policies must be |
1612 | * protected by task_lock(task) by the caller. | |
52cd3b07 LS |
1613 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1614 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1615 | * freeing by another task. It is the caller's responsibility to free the | |
1616 | * extra reference for shared policies. | |
480eccf9 | 1617 | */ |
d98f6cb6 | 1618 | struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1619 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1620 | { |
5606e387 | 1621 | struct mempolicy *pol = get_task_policy(task); |
1da177e4 LT |
1622 | |
1623 | if (vma) { | |
480eccf9 | 1624 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1625 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1626 | addr); | |
1627 | if (vpol) | |
1628 | pol = vpol; | |
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 LT |
1641 | } |
1642 | if (!pol) | |
1643 | pol = &default_policy; | |
1644 | return pol; | |
1645 | } | |
1646 | ||
52cd3b07 LS |
1647 | /* |
1648 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1649 | * page allocation | |
1650 | */ | |
1651 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1652 | { |
1653 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1654 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1655 | gfp_zone(gfp) >= policy_zone && |
1656 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1657 | return &policy->v.nodes; | |
1658 | ||
1659 | return NULL; | |
1660 | } | |
1661 | ||
52cd3b07 | 1662 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1663 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1664 | int nd) | |
1da177e4 | 1665 | { |
45c4745a | 1666 | switch (policy->mode) { |
1da177e4 | 1667 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1668 | if (!(policy->flags & MPOL_F_LOCAL)) |
1669 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1670 | break; |
1671 | case MPOL_BIND: | |
19770b32 | 1672 | /* |
52cd3b07 LS |
1673 | * Normally, MPOL_BIND allocations are node-local within the |
1674 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1675 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1676 | * the first node in the mask instead. |
19770b32 | 1677 | */ |
19770b32 MG |
1678 | if (unlikely(gfp & __GFP_THISNODE) && |
1679 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1680 | nd = first_node(policy->v.nodes); | |
1681 | break; | |
1da177e4 | 1682 | default: |
1da177e4 LT |
1683 | BUG(); |
1684 | } | |
0e88460d | 1685 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1686 | } |
1687 | ||
1688 | /* Do dynamic interleaving for a process */ | |
1689 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1690 | { | |
1691 | unsigned nid, next; | |
1692 | struct task_struct *me = current; | |
1693 | ||
1694 | nid = me->il_next; | |
dfcd3c0d | 1695 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1696 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1697 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1698 | if (next < MAX_NUMNODES) |
1699 | me->il_next = next; | |
1da177e4 LT |
1700 | return nid; |
1701 | } | |
1702 | ||
dc85da15 CL |
1703 | /* |
1704 | * Depending on the memory policy provide a node from which to allocate the | |
1705 | * next slab entry. | |
52cd3b07 LS |
1706 | * @policy must be protected by freeing by the caller. If @policy is |
1707 | * the current task's mempolicy, this protection is implicit, as only the | |
1708 | * task can change it's policy. The system default policy requires no | |
1709 | * such protection. | |
dc85da15 | 1710 | */ |
e7b691b0 | 1711 | unsigned slab_node(void) |
dc85da15 | 1712 | { |
e7b691b0 AK |
1713 | struct mempolicy *policy; |
1714 | ||
1715 | if (in_interrupt()) | |
1716 | return numa_node_id(); | |
1717 | ||
1718 | policy = current->mempolicy; | |
fc36b8d3 | 1719 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1720 | return numa_node_id(); |
1721 | ||
1722 | switch (policy->mode) { | |
1723 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1724 | /* |
1725 | * handled MPOL_F_LOCAL above | |
1726 | */ | |
1727 | return policy->v.preferred_node; | |
765c4507 | 1728 | |
dc85da15 CL |
1729 | case MPOL_INTERLEAVE: |
1730 | return interleave_nodes(policy); | |
1731 | ||
dd1a239f | 1732 | case MPOL_BIND: { |
dc85da15 CL |
1733 | /* |
1734 | * Follow bind policy behavior and start allocation at the | |
1735 | * first node. | |
1736 | */ | |
19770b32 MG |
1737 | struct zonelist *zonelist; |
1738 | struct zone *zone; | |
1739 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1740 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1741 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1742 | &policy->v.nodes, | |
1743 | &zone); | |
800416f7 | 1744 | return zone ? zone->node : numa_node_id(); |
dd1a239f | 1745 | } |
dc85da15 | 1746 | |
dc85da15 | 1747 | default: |
bea904d5 | 1748 | BUG(); |
dc85da15 CL |
1749 | } |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | /* Do static interleaving for a VMA with known offset. */ |
1753 | static unsigned offset_il_node(struct mempolicy *pol, | |
1754 | struct vm_area_struct *vma, unsigned long off) | |
1755 | { | |
dfcd3c0d | 1756 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1757 | unsigned target; |
1da177e4 LT |
1758 | int c; |
1759 | int nid = -1; | |
1760 | ||
f5b087b5 DR |
1761 | if (!nnodes) |
1762 | return numa_node_id(); | |
1763 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1764 | c = 0; |
1765 | do { | |
dfcd3c0d | 1766 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1767 | c++; |
1768 | } while (c <= target); | |
1da177e4 LT |
1769 | return nid; |
1770 | } | |
1771 | ||
5da7ca86 CL |
1772 | /* Determine a node number for interleave */ |
1773 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1774 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1775 | { | |
1776 | if (vma) { | |
1777 | unsigned long off; | |
1778 | ||
3b98b087 NA |
1779 | /* |
1780 | * for small pages, there is no difference between | |
1781 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1782 | * for huge pages, since vm_pgoff is in units of small | |
1783 | * pages, we need to shift off the always 0 bits to get | |
1784 | * a useful offset. | |
1785 | */ | |
1786 | BUG_ON(shift < PAGE_SHIFT); | |
1787 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1788 | off += (addr - vma->vm_start) >> shift; |
1789 | return offset_il_node(pol, vma, off); | |
1790 | } else | |
1791 | return interleave_nodes(pol); | |
1792 | } | |
1793 | ||
778d3b0f MH |
1794 | /* |
1795 | * Return the bit number of a random bit set in the nodemask. | |
1796 | * (returns -1 if nodemask is empty) | |
1797 | */ | |
1798 | int node_random(const nodemask_t *maskp) | |
1799 | { | |
1800 | int w, bit = -1; | |
1801 | ||
1802 | w = nodes_weight(*maskp); | |
1803 | if (w) | |
1804 | bit = bitmap_ord_to_pos(maskp->bits, | |
1805 | get_random_int() % w, MAX_NUMNODES); | |
1806 | return bit; | |
1807 | } | |
1808 | ||
00ac59ad | 1809 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1810 | /* |
1811 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1812 | * @vma = virtual memory area whose policy is sought | |
1813 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1814 | * @gfp_flags = for requested zone | |
19770b32 MG |
1815 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1816 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1817 | * |
52cd3b07 LS |
1818 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1819 | * to the struct mempolicy for conditional unref after allocation. | |
1820 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1821 | * @nodemask for filtering the zonelist. | |
c0ff7453 MX |
1822 | * |
1823 | * Must be protected by get_mems_allowed() | |
480eccf9 | 1824 | */ |
396faf03 | 1825 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1826 | gfp_t gfp_flags, struct mempolicy **mpol, |
1827 | nodemask_t **nodemask) | |
5da7ca86 | 1828 | { |
480eccf9 | 1829 | struct zonelist *zl; |
5da7ca86 | 1830 | |
52cd3b07 | 1831 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1832 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1833 | |
52cd3b07 LS |
1834 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1835 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1836 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1837 | } else { |
2f5f9486 | 1838 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1839 | if ((*mpol)->mode == MPOL_BIND) |
1840 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1841 | } |
1842 | return zl; | |
5da7ca86 | 1843 | } |
06808b08 LS |
1844 | |
1845 | /* | |
1846 | * init_nodemask_of_mempolicy | |
1847 | * | |
1848 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1849 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1850 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1851 | * initialize the argument nodemask to contain the single node for | |
1852 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1853 | * of non-default mempolicy. | |
1854 | * | |
1855 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1856 | * because the current task is examining it's own mempolicy and a task's | |
1857 | * mempolicy is only ever changed by the task itself. | |
1858 | * | |
1859 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1860 | */ | |
1861 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1862 | { | |
1863 | struct mempolicy *mempolicy; | |
1864 | int nid; | |
1865 | ||
1866 | if (!(mask && current->mempolicy)) | |
1867 | return false; | |
1868 | ||
c0ff7453 | 1869 | task_lock(current); |
06808b08 LS |
1870 | mempolicy = current->mempolicy; |
1871 | switch (mempolicy->mode) { | |
1872 | case MPOL_PREFERRED: | |
1873 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1874 | nid = numa_node_id(); | |
1875 | else | |
1876 | nid = mempolicy->v.preferred_node; | |
1877 | init_nodemask_of_node(mask, nid); | |
1878 | break; | |
1879 | ||
1880 | case MPOL_BIND: | |
1881 | /* Fall through */ | |
1882 | case MPOL_INTERLEAVE: | |
1883 | *mask = mempolicy->v.nodes; | |
1884 | break; | |
1885 | ||
1886 | default: | |
1887 | BUG(); | |
1888 | } | |
c0ff7453 | 1889 | task_unlock(current); |
06808b08 LS |
1890 | |
1891 | return true; | |
1892 | } | |
00ac59ad | 1893 | #endif |
5da7ca86 | 1894 | |
6f48d0eb DR |
1895 | /* |
1896 | * mempolicy_nodemask_intersects | |
1897 | * | |
1898 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1899 | * policy. Otherwise, check for intersection between mask and the policy | |
1900 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1901 | * policy, always return true since it may allocate elsewhere on fallback. | |
1902 | * | |
1903 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1904 | */ | |
1905 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1906 | const nodemask_t *mask) | |
1907 | { | |
1908 | struct mempolicy *mempolicy; | |
1909 | bool ret = true; | |
1910 | ||
1911 | if (!mask) | |
1912 | return ret; | |
1913 | task_lock(tsk); | |
1914 | mempolicy = tsk->mempolicy; | |
1915 | if (!mempolicy) | |
1916 | goto out; | |
1917 | ||
1918 | switch (mempolicy->mode) { | |
1919 | case MPOL_PREFERRED: | |
1920 | /* | |
1921 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1922 | * allocate from, they may fallback to other nodes when oom. | |
1923 | * Thus, it's possible for tsk to have allocated memory from | |
1924 | * nodes in mask. | |
1925 | */ | |
1926 | break; | |
1927 | case MPOL_BIND: | |
1928 | case MPOL_INTERLEAVE: | |
1929 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1930 | break; | |
1931 | default: | |
1932 | BUG(); | |
1933 | } | |
1934 | out: | |
1935 | task_unlock(tsk); | |
1936 | return ret; | |
1937 | } | |
1938 | ||
1da177e4 LT |
1939 | /* Allocate a page in interleaved policy. |
1940 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1941 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1942 | unsigned nid) | |
1da177e4 LT |
1943 | { |
1944 | struct zonelist *zl; | |
1945 | struct page *page; | |
1946 | ||
0e88460d | 1947 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1948 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1949 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1950 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1951 | return page; |
1952 | } | |
1953 | ||
1954 | /** | |
0bbbc0b3 | 1955 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1956 | * |
1957 | * @gfp: | |
1958 | * %GFP_USER user allocation. | |
1959 | * %GFP_KERNEL kernel allocations, | |
1960 | * %GFP_HIGHMEM highmem/user allocations, | |
1961 | * %GFP_FS allocation should not call back into a file system. | |
1962 | * %GFP_ATOMIC don't sleep. | |
1963 | * | |
0bbbc0b3 | 1964 | * @order:Order of the GFP allocation. |
1da177e4 LT |
1965 | * @vma: Pointer to VMA or NULL if not available. |
1966 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1967 | * | |
1968 | * This function allocates a page from the kernel page pool and applies | |
1969 | * a NUMA policy associated with the VMA or the current process. | |
1970 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1971 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1972 | * all allocations for pages that will be mapped into | |
1973 | * user space. Returns NULL when no page can be allocated. | |
1974 | * | |
1975 | * Should be called with the mm_sem of the vma hold. | |
1976 | */ | |
1977 | struct page * | |
0bbbc0b3 | 1978 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 1979 | unsigned long addr, int node) |
1da177e4 | 1980 | { |
cc9a6c87 | 1981 | struct mempolicy *pol; |
c0ff7453 | 1982 | struct page *page; |
cc9a6c87 MG |
1983 | unsigned int cpuset_mems_cookie; |
1984 | ||
1985 | retry_cpuset: | |
1986 | pol = get_vma_policy(current, vma, addr); | |
1987 | cpuset_mems_cookie = get_mems_allowed(); | |
1da177e4 | 1988 | |
45c4745a | 1989 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1990 | unsigned nid; |
5da7ca86 | 1991 | |
8eac563c | 1992 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 1993 | mpol_cond_put(pol); |
0bbbc0b3 | 1994 | page = alloc_page_interleave(gfp, order, nid); |
cc9a6c87 MG |
1995 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1996 | goto retry_cpuset; | |
1997 | ||
c0ff7453 | 1998 | return page; |
1da177e4 | 1999 | } |
212a0a6f DR |
2000 | page = __alloc_pages_nodemask(gfp, order, |
2001 | policy_zonelist(gfp, pol, node), | |
0bbbc0b3 | 2002 | policy_nodemask(gfp, pol)); |
212a0a6f DR |
2003 | if (unlikely(mpol_needs_cond_ref(pol))) |
2004 | __mpol_put(pol); | |
cc9a6c87 MG |
2005 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
2006 | goto retry_cpuset; | |
c0ff7453 | 2007 | return page; |
1da177e4 LT |
2008 | } |
2009 | ||
2010 | /** | |
2011 | * alloc_pages_current - Allocate pages. | |
2012 | * | |
2013 | * @gfp: | |
2014 | * %GFP_USER user allocation, | |
2015 | * %GFP_KERNEL kernel allocation, | |
2016 | * %GFP_HIGHMEM highmem allocation, | |
2017 | * %GFP_FS don't call back into a file system. | |
2018 | * %GFP_ATOMIC don't sleep. | |
2019 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2020 | * | |
2021 | * Allocate a page from the kernel page pool. When not in | |
2022 | * interrupt context and apply the current process NUMA policy. | |
2023 | * Returns NULL when no page can be allocated. | |
2024 | * | |
cf2a473c | 2025 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
2026 | * 1) it's ok to take cpuset_sem (can WAIT), and |
2027 | * 2) allocating for current task (not interrupt). | |
2028 | */ | |
dd0fc66f | 2029 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2030 | { |
5606e387 | 2031 | struct mempolicy *pol = get_task_policy(current); |
c0ff7453 | 2032 | struct page *page; |
cc9a6c87 | 2033 | unsigned int cpuset_mems_cookie; |
1da177e4 | 2034 | |
9b819d20 | 2035 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 2036 | pol = &default_policy; |
52cd3b07 | 2037 | |
cc9a6c87 MG |
2038 | retry_cpuset: |
2039 | cpuset_mems_cookie = get_mems_allowed(); | |
2040 | ||
52cd3b07 LS |
2041 | /* |
2042 | * No reference counting needed for current->mempolicy | |
2043 | * nor system default_policy | |
2044 | */ | |
45c4745a | 2045 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2046 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2047 | else | |
2048 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
2049 | policy_zonelist(gfp, pol, numa_node_id()), |
2050 | policy_nodemask(gfp, pol)); | |
cc9a6c87 MG |
2051 | |
2052 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2053 | goto retry_cpuset; | |
2054 | ||
c0ff7453 | 2055 | return page; |
1da177e4 LT |
2056 | } |
2057 | EXPORT_SYMBOL(alloc_pages_current); | |
2058 | ||
4225399a | 2059 | /* |
846a16bf | 2060 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2061 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2062 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2063 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2064 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2065 | * |
2066 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2067 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2068 | */ |
4225399a | 2069 | |
846a16bf LS |
2070 | /* Slow path of a mempolicy duplicate */ |
2071 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2072 | { |
2073 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2074 | ||
2075 | if (!new) | |
2076 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2077 | |
2078 | /* task's mempolicy is protected by alloc_lock */ | |
2079 | if (old == current->mempolicy) { | |
2080 | task_lock(current); | |
2081 | *new = *old; | |
2082 | task_unlock(current); | |
2083 | } else | |
2084 | *new = *old; | |
2085 | ||
99ee4ca7 | 2086 | rcu_read_lock(); |
4225399a PJ |
2087 | if (current_cpuset_is_being_rebound()) { |
2088 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2089 | if (new->flags & MPOL_F_REBINDING) |
2090 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2091 | else | |
2092 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2093 | } |
99ee4ca7 | 2094 | rcu_read_unlock(); |
1da177e4 | 2095 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2096 | return new; |
2097 | } | |
2098 | ||
2099 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2100 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2101 | { |
2102 | if (!a || !b) | |
fcfb4dcc | 2103 | return false; |
45c4745a | 2104 | if (a->mode != b->mode) |
fcfb4dcc | 2105 | return false; |
19800502 | 2106 | if (a->flags != b->flags) |
fcfb4dcc | 2107 | return false; |
19800502 BL |
2108 | if (mpol_store_user_nodemask(a)) |
2109 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2110 | return false; |
19800502 | 2111 | |
45c4745a | 2112 | switch (a->mode) { |
19770b32 MG |
2113 | case MPOL_BIND: |
2114 | /* Fall through */ | |
1da177e4 | 2115 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2116 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2117 | case MPOL_PREFERRED: |
75719661 | 2118 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2119 | default: |
2120 | BUG(); | |
fcfb4dcc | 2121 | return false; |
1da177e4 LT |
2122 | } |
2123 | } | |
2124 | ||
1da177e4 LT |
2125 | /* |
2126 | * Shared memory backing store policy support. | |
2127 | * | |
2128 | * Remember policies even when nobody has shared memory mapped. | |
2129 | * The policies are kept in Red-Black tree linked from the inode. | |
2130 | * They are protected by the sp->lock spinlock, which should be held | |
2131 | * for any accesses to the tree. | |
2132 | */ | |
2133 | ||
2134 | /* lookup first element intersecting start-end */ | |
42288fe3 | 2135 | /* Caller holds sp->lock */ |
1da177e4 LT |
2136 | static struct sp_node * |
2137 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2138 | { | |
2139 | struct rb_node *n = sp->root.rb_node; | |
2140 | ||
2141 | while (n) { | |
2142 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2143 | ||
2144 | if (start >= p->end) | |
2145 | n = n->rb_right; | |
2146 | else if (end <= p->start) | |
2147 | n = n->rb_left; | |
2148 | else | |
2149 | break; | |
2150 | } | |
2151 | if (!n) | |
2152 | return NULL; | |
2153 | for (;;) { | |
2154 | struct sp_node *w = NULL; | |
2155 | struct rb_node *prev = rb_prev(n); | |
2156 | if (!prev) | |
2157 | break; | |
2158 | w = rb_entry(prev, struct sp_node, nd); | |
2159 | if (w->end <= start) | |
2160 | break; | |
2161 | n = prev; | |
2162 | } | |
2163 | return rb_entry(n, struct sp_node, nd); | |
2164 | } | |
2165 | ||
2166 | /* Insert a new shared policy into the list. */ | |
2167 | /* Caller holds sp->lock */ | |
2168 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2169 | { | |
2170 | struct rb_node **p = &sp->root.rb_node; | |
2171 | struct rb_node *parent = NULL; | |
2172 | struct sp_node *nd; | |
2173 | ||
2174 | while (*p) { | |
2175 | parent = *p; | |
2176 | nd = rb_entry(parent, struct sp_node, nd); | |
2177 | if (new->start < nd->start) | |
2178 | p = &(*p)->rb_left; | |
2179 | else if (new->end > nd->end) | |
2180 | p = &(*p)->rb_right; | |
2181 | else | |
2182 | BUG(); | |
2183 | } | |
2184 | rb_link_node(&new->nd, parent, p); | |
2185 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2186 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2187 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2188 | } |
2189 | ||
2190 | /* Find shared policy intersecting idx */ | |
2191 | struct mempolicy * | |
2192 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2193 | { | |
2194 | struct mempolicy *pol = NULL; | |
2195 | struct sp_node *sn; | |
2196 | ||
2197 | if (!sp->root.rb_node) | |
2198 | return NULL; | |
42288fe3 | 2199 | spin_lock(&sp->lock); |
1da177e4 LT |
2200 | sn = sp_lookup(sp, idx, idx+1); |
2201 | if (sn) { | |
2202 | mpol_get(sn->policy); | |
2203 | pol = sn->policy; | |
2204 | } | |
42288fe3 | 2205 | spin_unlock(&sp->lock); |
1da177e4 LT |
2206 | return pol; |
2207 | } | |
2208 | ||
63f74ca2 KM |
2209 | static void sp_free(struct sp_node *n) |
2210 | { | |
2211 | mpol_put(n->policy); | |
2212 | kmem_cache_free(sn_cache, n); | |
2213 | } | |
2214 | ||
771fb4d8 LS |
2215 | /** |
2216 | * mpol_misplaced - check whether current page node is valid in policy | |
2217 | * | |
2218 | * @page - page to be checked | |
2219 | * @vma - vm area where page mapped | |
2220 | * @addr - virtual address where page mapped | |
2221 | * | |
2222 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2223 | * node id. | |
2224 | * | |
2225 | * Returns: | |
2226 | * -1 - not misplaced, page is in the right node | |
2227 | * node - node id where the page should be | |
2228 | * | |
2229 | * Policy determination "mimics" alloc_page_vma(). | |
2230 | * Called from fault path where we know the vma and faulting address. | |
2231 | */ | |
2232 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2233 | { | |
2234 | struct mempolicy *pol; | |
2235 | struct zone *zone; | |
2236 | int curnid = page_to_nid(page); | |
2237 | unsigned long pgoff; | |
2238 | int polnid = -1; | |
2239 | int ret = -1; | |
2240 | ||
2241 | BUG_ON(!vma); | |
2242 | ||
2243 | pol = get_vma_policy(current, vma, addr); | |
2244 | if (!(pol->flags & MPOL_F_MOF)) | |
2245 | goto out; | |
2246 | ||
2247 | switch (pol->mode) { | |
2248 | case MPOL_INTERLEAVE: | |
2249 | BUG_ON(addr >= vma->vm_end); | |
2250 | BUG_ON(addr < vma->vm_start); | |
2251 | ||
2252 | pgoff = vma->vm_pgoff; | |
2253 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2254 | polnid = offset_il_node(pol, vma, pgoff); | |
2255 | break; | |
2256 | ||
2257 | case MPOL_PREFERRED: | |
2258 | if (pol->flags & MPOL_F_LOCAL) | |
2259 | polnid = numa_node_id(); | |
2260 | else | |
2261 | polnid = pol->v.preferred_node; | |
2262 | break; | |
2263 | ||
2264 | case MPOL_BIND: | |
2265 | /* | |
2266 | * allows binding to multiple nodes. | |
2267 | * use current page if in policy nodemask, | |
2268 | * else select nearest allowed node, if any. | |
2269 | * If no allowed nodes, use current [!misplaced]. | |
2270 | */ | |
2271 | if (node_isset(curnid, pol->v.nodes)) | |
2272 | goto out; | |
2273 | (void)first_zones_zonelist( | |
2274 | node_zonelist(numa_node_id(), GFP_HIGHUSER), | |
2275 | gfp_zone(GFP_HIGHUSER), | |
2276 | &pol->v.nodes, &zone); | |
2277 | polnid = zone->node; | |
2278 | break; | |
2279 | ||
2280 | default: | |
2281 | BUG(); | |
2282 | } | |
5606e387 MG |
2283 | |
2284 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 MG |
2285 | if (pol->flags & MPOL_F_MORON) { |
2286 | int last_nid; | |
2287 | ||
5606e387 MG |
2288 | polnid = numa_node_id(); |
2289 | ||
e42c8ff2 MG |
2290 | /* |
2291 | * Multi-stage node selection is used in conjunction | |
2292 | * with a periodic migration fault to build a temporal | |
2293 | * task<->page relation. By using a two-stage filter we | |
2294 | * remove short/unlikely relations. | |
2295 | * | |
2296 | * Using P(p) ~ n_p / n_t as per frequentist | |
2297 | * probability, we can equate a task's usage of a | |
2298 | * particular page (n_p) per total usage of this | |
2299 | * page (n_t) (in a given time-span) to a probability. | |
2300 | * | |
2301 | * Our periodic faults will sample this probability and | |
2302 | * getting the same result twice in a row, given these | |
2303 | * samples are fully independent, is then given by | |
2304 | * P(n)^2, provided our sample period is sufficiently | |
2305 | * short compared to the usage pattern. | |
2306 | * | |
2307 | * This quadric squishes small probabilities, making | |
2308 | * it less likely we act on an unlikely task<->page | |
2309 | * relation. | |
2310 | */ | |
2311 | last_nid = page_xchg_last_nid(page, polnid); | |
2312 | if (last_nid != polnid) | |
2313 | goto out; | |
2314 | } | |
2315 | ||
771fb4d8 LS |
2316 | if (curnid != polnid) |
2317 | ret = polnid; | |
2318 | out: | |
2319 | mpol_cond_put(pol); | |
2320 | ||
2321 | return ret; | |
2322 | } | |
2323 | ||
1da177e4 LT |
2324 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2325 | { | |
140d5a49 | 2326 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2327 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2328 | sp_free(n); |
1da177e4 LT |
2329 | } |
2330 | ||
42288fe3 MG |
2331 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2332 | unsigned long end, struct mempolicy *pol) | |
2333 | { | |
2334 | node->start = start; | |
2335 | node->end = end; | |
2336 | node->policy = pol; | |
2337 | } | |
2338 | ||
dbcb0f19 AB |
2339 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2340 | struct mempolicy *pol) | |
1da177e4 | 2341 | { |
869833f2 KM |
2342 | struct sp_node *n; |
2343 | struct mempolicy *newpol; | |
1da177e4 | 2344 | |
869833f2 | 2345 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2346 | if (!n) |
2347 | return NULL; | |
869833f2 KM |
2348 | |
2349 | newpol = mpol_dup(pol); | |
2350 | if (IS_ERR(newpol)) { | |
2351 | kmem_cache_free(sn_cache, n); | |
2352 | return NULL; | |
2353 | } | |
2354 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2355 | sp_node_init(n, start, end, newpol); |
869833f2 | 2356 | |
1da177e4 LT |
2357 | return n; |
2358 | } | |
2359 | ||
2360 | /* Replace a policy range. */ | |
2361 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2362 | unsigned long end, struct sp_node *new) | |
2363 | { | |
b22d127a | 2364 | struct sp_node *n; |
42288fe3 MG |
2365 | struct sp_node *n_new = NULL; |
2366 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2367 | int ret = 0; |
1da177e4 | 2368 | |
42288fe3 MG |
2369 | restart: |
2370 | spin_lock(&sp->lock); | |
1da177e4 LT |
2371 | n = sp_lookup(sp, start, end); |
2372 | /* Take care of old policies in the same range. */ | |
2373 | while (n && n->start < end) { | |
2374 | struct rb_node *next = rb_next(&n->nd); | |
2375 | if (n->start >= start) { | |
2376 | if (n->end <= end) | |
2377 | sp_delete(sp, n); | |
2378 | else | |
2379 | n->start = end; | |
2380 | } else { | |
2381 | /* Old policy spanning whole new range. */ | |
2382 | if (n->end > end) { | |
42288fe3 MG |
2383 | if (!n_new) |
2384 | goto alloc_new; | |
2385 | ||
2386 | *mpol_new = *n->policy; | |
2387 | atomic_set(&mpol_new->refcnt, 1); | |
2388 | sp_node_init(n_new, n->end, end, mpol_new); | |
2389 | sp_insert(sp, n_new); | |
1da177e4 | 2390 | n->end = start; |
42288fe3 MG |
2391 | n_new = NULL; |
2392 | mpol_new = NULL; | |
1da177e4 LT |
2393 | break; |
2394 | } else | |
2395 | n->end = start; | |
2396 | } | |
2397 | if (!next) | |
2398 | break; | |
2399 | n = rb_entry(next, struct sp_node, nd); | |
2400 | } | |
2401 | if (new) | |
2402 | sp_insert(sp, new); | |
42288fe3 MG |
2403 | spin_unlock(&sp->lock); |
2404 | ret = 0; | |
2405 | ||
2406 | err_out: | |
2407 | if (mpol_new) | |
2408 | mpol_put(mpol_new); | |
2409 | if (n_new) | |
2410 | kmem_cache_free(sn_cache, n_new); | |
2411 | ||
b22d127a | 2412 | return ret; |
42288fe3 MG |
2413 | |
2414 | alloc_new: | |
2415 | spin_unlock(&sp->lock); | |
2416 | ret = -ENOMEM; | |
2417 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2418 | if (!n_new) | |
2419 | goto err_out; | |
2420 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2421 | if (!mpol_new) | |
2422 | goto err_out; | |
2423 | goto restart; | |
1da177e4 LT |
2424 | } |
2425 | ||
71fe804b LS |
2426 | /** |
2427 | * mpol_shared_policy_init - initialize shared policy for inode | |
2428 | * @sp: pointer to inode shared policy | |
2429 | * @mpol: struct mempolicy to install | |
2430 | * | |
2431 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2432 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2433 | * This must be released on exit. | |
4bfc4495 | 2434 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2435 | */ |
2436 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2437 | { | |
58568d2a MX |
2438 | int ret; |
2439 | ||
71fe804b | 2440 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
42288fe3 | 2441 | spin_lock_init(&sp->lock); |
71fe804b LS |
2442 | |
2443 | if (mpol) { | |
2444 | struct vm_area_struct pvma; | |
2445 | struct mempolicy *new; | |
4bfc4495 | 2446 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2447 | |
4bfc4495 | 2448 | if (!scratch) |
5c0c1654 | 2449 | goto put_mpol; |
71fe804b LS |
2450 | /* contextualize the tmpfs mount point mempolicy */ |
2451 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2452 | if (IS_ERR(new)) |
0cae3457 | 2453 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2454 | |
2455 | task_lock(current); | |
4bfc4495 | 2456 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2457 | task_unlock(current); |
15d77835 | 2458 | if (ret) |
5c0c1654 | 2459 | goto put_new; |
71fe804b LS |
2460 | |
2461 | /* Create pseudo-vma that contains just the policy */ | |
2462 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2463 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2464 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2465 | |
5c0c1654 | 2466 | put_new: |
71fe804b | 2467 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2468 | free_scratch: |
4bfc4495 | 2469 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2470 | put_mpol: |
2471 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2472 | } |
2473 | } | |
2474 | ||
1da177e4 LT |
2475 | int mpol_set_shared_policy(struct shared_policy *info, |
2476 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2477 | { | |
2478 | int err; | |
2479 | struct sp_node *new = NULL; | |
2480 | unsigned long sz = vma_pages(vma); | |
2481 | ||
028fec41 | 2482 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2483 | vma->vm_pgoff, |
45c4745a | 2484 | sz, npol ? npol->mode : -1, |
028fec41 | 2485 | npol ? npol->flags : -1, |
140d5a49 | 2486 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
2487 | |
2488 | if (npol) { | |
2489 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2490 | if (!new) | |
2491 | return -ENOMEM; | |
2492 | } | |
2493 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2494 | if (err && new) | |
63f74ca2 | 2495 | sp_free(new); |
1da177e4 LT |
2496 | return err; |
2497 | } | |
2498 | ||
2499 | /* Free a backing policy store on inode delete. */ | |
2500 | void mpol_free_shared_policy(struct shared_policy *p) | |
2501 | { | |
2502 | struct sp_node *n; | |
2503 | struct rb_node *next; | |
2504 | ||
2505 | if (!p->root.rb_node) | |
2506 | return; | |
42288fe3 | 2507 | spin_lock(&p->lock); |
1da177e4 LT |
2508 | next = rb_first(&p->root); |
2509 | while (next) { | |
2510 | n = rb_entry(next, struct sp_node, nd); | |
2511 | next = rb_next(&n->nd); | |
63f74ca2 | 2512 | sp_delete(p, n); |
1da177e4 | 2513 | } |
42288fe3 | 2514 | spin_unlock(&p->lock); |
1da177e4 LT |
2515 | } |
2516 | ||
1a687c2e MG |
2517 | #ifdef CONFIG_NUMA_BALANCING |
2518 | static bool __initdata numabalancing_override; | |
2519 | ||
2520 | static void __init check_numabalancing_enable(void) | |
2521 | { | |
2522 | bool numabalancing_default = false; | |
2523 | ||
2524 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2525 | numabalancing_default = true; | |
2526 | ||
2527 | if (nr_node_ids > 1 && !numabalancing_override) { | |
2528 | printk(KERN_INFO "Enabling automatic NUMA balancing. " | |
2529 | "Configure with numa_balancing= or sysctl"); | |
2530 | set_numabalancing_state(numabalancing_default); | |
2531 | } | |
2532 | } | |
2533 | ||
2534 | static int __init setup_numabalancing(char *str) | |
2535 | { | |
2536 | int ret = 0; | |
2537 | if (!str) | |
2538 | goto out; | |
2539 | numabalancing_override = true; | |
2540 | ||
2541 | if (!strcmp(str, "enable")) { | |
2542 | set_numabalancing_state(true); | |
2543 | ret = 1; | |
2544 | } else if (!strcmp(str, "disable")) { | |
2545 | set_numabalancing_state(false); | |
2546 | ret = 1; | |
2547 | } | |
2548 | out: | |
2549 | if (!ret) | |
2550 | printk(KERN_WARNING "Unable to parse numa_balancing=\n"); | |
2551 | ||
2552 | return ret; | |
2553 | } | |
2554 | __setup("numa_balancing=", setup_numabalancing); | |
2555 | #else | |
2556 | static inline void __init check_numabalancing_enable(void) | |
2557 | { | |
2558 | } | |
2559 | #endif /* CONFIG_NUMA_BALANCING */ | |
2560 | ||
1da177e4 LT |
2561 | /* assumes fs == KERNEL_DS */ |
2562 | void __init numa_policy_init(void) | |
2563 | { | |
b71636e2 PM |
2564 | nodemask_t interleave_nodes; |
2565 | unsigned long largest = 0; | |
2566 | int nid, prefer = 0; | |
2567 | ||
1da177e4 LT |
2568 | policy_cache = kmem_cache_create("numa_policy", |
2569 | sizeof(struct mempolicy), | |
20c2df83 | 2570 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2571 | |
2572 | sn_cache = kmem_cache_create("shared_policy_node", | |
2573 | sizeof(struct sp_node), | |
20c2df83 | 2574 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2575 | |
5606e387 MG |
2576 | for_each_node(nid) { |
2577 | preferred_node_policy[nid] = (struct mempolicy) { | |
2578 | .refcnt = ATOMIC_INIT(1), | |
2579 | .mode = MPOL_PREFERRED, | |
2580 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2581 | .v = { .preferred_node = nid, }, | |
2582 | }; | |
2583 | } | |
2584 | ||
b71636e2 PM |
2585 | /* |
2586 | * Set interleaving policy for system init. Interleaving is only | |
2587 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2588 | * fall back to the largest node if they're all smaller. | |
2589 | */ | |
2590 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2591 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2592 | unsigned long total_pages = node_present_pages(nid); |
2593 | ||
2594 | /* Preserve the largest node */ | |
2595 | if (largest < total_pages) { | |
2596 | largest = total_pages; | |
2597 | prefer = nid; | |
2598 | } | |
2599 | ||
2600 | /* Interleave this node? */ | |
2601 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2602 | node_set(nid, interleave_nodes); | |
2603 | } | |
2604 | ||
2605 | /* All too small, use the largest */ | |
2606 | if (unlikely(nodes_empty(interleave_nodes))) | |
2607 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2608 | |
028fec41 | 2609 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 | 2610 | printk("numa_policy_init: interleaving failed\n"); |
1a687c2e MG |
2611 | |
2612 | check_numabalancing_enable(); | |
1da177e4 LT |
2613 | } |
2614 | ||
8bccd85f | 2615 | /* Reset policy of current process to default */ |
1da177e4 LT |
2616 | void numa_default_policy(void) |
2617 | { | |
028fec41 | 2618 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2619 | } |
68860ec1 | 2620 | |
095f1fc4 LS |
2621 | /* |
2622 | * Parse and format mempolicy from/to strings | |
2623 | */ | |
2624 | ||
1a75a6c8 | 2625 | /* |
f2a07f40 | 2626 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2627 | */ |
345ace9c LS |
2628 | static const char * const policy_modes[] = |
2629 | { | |
2630 | [MPOL_DEFAULT] = "default", | |
2631 | [MPOL_PREFERRED] = "prefer", | |
2632 | [MPOL_BIND] = "bind", | |
2633 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2634 | [MPOL_LOCAL] = "local", |
345ace9c | 2635 | }; |
1a75a6c8 | 2636 | |
095f1fc4 LS |
2637 | |
2638 | #ifdef CONFIG_TMPFS | |
2639 | /** | |
f2a07f40 | 2640 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2641 | * @str: string containing mempolicy to parse |
71fe804b | 2642 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2643 | * |
2644 | * Format of input: | |
2645 | * <mode>[=<flags>][:<nodelist>] | |
2646 | * | |
71fe804b | 2647 | * On success, returns 0, else 1 |
095f1fc4 | 2648 | */ |
a7a88b23 | 2649 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2650 | { |
71fe804b | 2651 | struct mempolicy *new = NULL; |
b4652e84 | 2652 | unsigned short mode; |
f2a07f40 | 2653 | unsigned short mode_flags; |
71fe804b | 2654 | nodemask_t nodes; |
095f1fc4 LS |
2655 | char *nodelist = strchr(str, ':'); |
2656 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2657 | int err = 1; |
2658 | ||
2659 | if (nodelist) { | |
2660 | /* NUL-terminate mode or flags string */ | |
2661 | *nodelist++ = '\0'; | |
71fe804b | 2662 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2663 | goto out; |
01f13bd6 | 2664 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2665 | goto out; |
71fe804b LS |
2666 | } else |
2667 | nodes_clear(nodes); | |
2668 | ||
095f1fc4 LS |
2669 | if (flags) |
2670 | *flags++ = '\0'; /* terminate mode string */ | |
2671 | ||
479e2802 | 2672 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2673 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2674 | break; |
2675 | } | |
2676 | } | |
a720094d | 2677 | if (mode >= MPOL_MAX) |
095f1fc4 LS |
2678 | goto out; |
2679 | ||
71fe804b | 2680 | switch (mode) { |
095f1fc4 | 2681 | case MPOL_PREFERRED: |
71fe804b LS |
2682 | /* |
2683 | * Insist on a nodelist of one node only | |
2684 | */ | |
095f1fc4 LS |
2685 | if (nodelist) { |
2686 | char *rest = nodelist; | |
2687 | while (isdigit(*rest)) | |
2688 | rest++; | |
926f2ae0 KM |
2689 | if (*rest) |
2690 | goto out; | |
095f1fc4 LS |
2691 | } |
2692 | break; | |
095f1fc4 LS |
2693 | case MPOL_INTERLEAVE: |
2694 | /* | |
2695 | * Default to online nodes with memory if no nodelist | |
2696 | */ | |
2697 | if (!nodelist) | |
01f13bd6 | 2698 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2699 | break; |
71fe804b | 2700 | case MPOL_LOCAL: |
3f226aa1 | 2701 | /* |
71fe804b | 2702 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2703 | */ |
71fe804b | 2704 | if (nodelist) |
3f226aa1 | 2705 | goto out; |
71fe804b | 2706 | mode = MPOL_PREFERRED; |
3f226aa1 | 2707 | break; |
413b43de RT |
2708 | case MPOL_DEFAULT: |
2709 | /* | |
2710 | * Insist on a empty nodelist | |
2711 | */ | |
2712 | if (!nodelist) | |
2713 | err = 0; | |
2714 | goto out; | |
d69b2e63 KM |
2715 | case MPOL_BIND: |
2716 | /* | |
2717 | * Insist on a nodelist | |
2718 | */ | |
2719 | if (!nodelist) | |
2720 | goto out; | |
095f1fc4 LS |
2721 | } |
2722 | ||
71fe804b | 2723 | mode_flags = 0; |
095f1fc4 LS |
2724 | if (flags) { |
2725 | /* | |
2726 | * Currently, we only support two mutually exclusive | |
2727 | * mode flags. | |
2728 | */ | |
2729 | if (!strcmp(flags, "static")) | |
71fe804b | 2730 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2731 | else if (!strcmp(flags, "relative")) |
71fe804b | 2732 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2733 | else |
926f2ae0 | 2734 | goto out; |
095f1fc4 | 2735 | } |
71fe804b LS |
2736 | |
2737 | new = mpol_new(mode, mode_flags, &nodes); | |
2738 | if (IS_ERR(new)) | |
926f2ae0 KM |
2739 | goto out; |
2740 | ||
f2a07f40 HD |
2741 | /* |
2742 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2743 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2744 | */ | |
2745 | if (mode != MPOL_PREFERRED) | |
2746 | new->v.nodes = nodes; | |
2747 | else if (nodelist) | |
2748 | new->v.preferred_node = first_node(nodes); | |
2749 | else | |
2750 | new->flags |= MPOL_F_LOCAL; | |
2751 | ||
2752 | /* | |
2753 | * Save nodes for contextualization: this will be used to "clone" | |
2754 | * the mempolicy in a specific context [cpuset] at a later time. | |
2755 | */ | |
2756 | new->w.user_nodemask = nodes; | |
2757 | ||
926f2ae0 | 2758 | err = 0; |
71fe804b | 2759 | |
095f1fc4 LS |
2760 | out: |
2761 | /* Restore string for error message */ | |
2762 | if (nodelist) | |
2763 | *--nodelist = ':'; | |
2764 | if (flags) | |
2765 | *--flags = '='; | |
71fe804b LS |
2766 | if (!err) |
2767 | *mpol = new; | |
095f1fc4 LS |
2768 | return err; |
2769 | } | |
2770 | #endif /* CONFIG_TMPFS */ | |
2771 | ||
71fe804b LS |
2772 | /** |
2773 | * mpol_to_str - format a mempolicy structure for printing | |
2774 | * @buffer: to contain formatted mempolicy string | |
2775 | * @maxlen: length of @buffer | |
2776 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2777 | * |
1a75a6c8 CL |
2778 | * Convert a mempolicy into a string. |
2779 | * Returns the number of characters in buffer (if positive) | |
2780 | * or an error (negative) | |
2781 | */ | |
a7a88b23 | 2782 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2783 | { |
2784 | char *p = buffer; | |
2785 | int l; | |
2786 | nodemask_t nodes; | |
bea904d5 | 2787 | unsigned short mode; |
f5b087b5 | 2788 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 2789 | |
2291990a LS |
2790 | /* |
2791 | * Sanity check: room for longest mode, flag and some nodes | |
2792 | */ | |
2793 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); | |
2794 | ||
bea904d5 LS |
2795 | if (!pol || pol == &default_policy) |
2796 | mode = MPOL_DEFAULT; | |
2797 | else | |
2798 | mode = pol->mode; | |
2799 | ||
1a75a6c8 CL |
2800 | switch (mode) { |
2801 | case MPOL_DEFAULT: | |
2802 | nodes_clear(nodes); | |
2803 | break; | |
2804 | ||
2805 | case MPOL_PREFERRED: | |
2806 | nodes_clear(nodes); | |
fc36b8d3 | 2807 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 2808 | mode = MPOL_LOCAL; |
53f2556b | 2809 | else |
fc36b8d3 | 2810 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
2811 | break; |
2812 | ||
2813 | case MPOL_BIND: | |
19770b32 | 2814 | /* Fall through */ |
1a75a6c8 | 2815 | case MPOL_INTERLEAVE: |
f2a07f40 | 2816 | nodes = pol->v.nodes; |
1a75a6c8 CL |
2817 | break; |
2818 | ||
2819 | default: | |
80de7c31 | 2820 | return -EINVAL; |
1a75a6c8 CL |
2821 | } |
2822 | ||
345ace9c | 2823 | l = strlen(policy_modes[mode]); |
53f2556b LS |
2824 | if (buffer + maxlen < p + l + 1) |
2825 | return -ENOSPC; | |
1a75a6c8 | 2826 | |
345ace9c | 2827 | strcpy(p, policy_modes[mode]); |
1a75a6c8 CL |
2828 | p += l; |
2829 | ||
fc36b8d3 | 2830 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2831 | if (buffer + maxlen < p + 2) |
2832 | return -ENOSPC; | |
2833 | *p++ = '='; | |
2834 | ||
2291990a LS |
2835 | /* |
2836 | * Currently, the only defined flags are mutually exclusive | |
2837 | */ | |
f5b087b5 | 2838 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2839 | p += snprintf(p, buffer + maxlen - p, "static"); |
2840 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2841 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2842 | } |
2843 | ||
1a75a6c8 CL |
2844 | if (!nodes_empty(nodes)) { |
2845 | if (buffer + maxlen < p + 2) | |
2846 | return -ENOSPC; | |
095f1fc4 | 2847 | *p++ = ':'; |
1a75a6c8 CL |
2848 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2849 | } | |
2850 | return p - buffer; | |
2851 | } |