]>
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 | ||
37012946 DR |
121 | static const struct mempolicy_operations { |
122 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
708c1bbc MX |
123 | /* |
124 | * If read-side task has no lock to protect task->mempolicy, write-side | |
125 | * task will rebind the task->mempolicy by two step. The first step is | |
126 | * setting all the newly nodes, and the second step is cleaning all the | |
127 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
128 | * page. | |
129 | * If we have a lock to protect task->mempolicy in read-side, we do | |
130 | * rebind directly. | |
131 | * | |
132 | * step: | |
133 | * MPOL_REBIND_ONCE - do rebind work at once | |
134 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
135 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
136 | */ | |
137 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes, | |
138 | enum mpol_rebind_step step); | |
37012946 DR |
139 | } mpol_ops[MPOL_MAX]; |
140 | ||
19770b32 | 141 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 142 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 143 | { |
19770b32 | 144 | int nd, k; |
1da177e4 | 145 | |
19770b32 MG |
146 | for_each_node_mask(nd, *nodemask) { |
147 | struct zone *z; | |
148 | ||
149 | for (k = 0; k <= policy_zone; k++) { | |
150 | z = &NODE_DATA(nd)->node_zones[k]; | |
151 | if (z->present_pages > 0) | |
152 | return 1; | |
dd942ae3 | 153 | } |
8af5e2eb | 154 | } |
19770b32 MG |
155 | |
156 | return 0; | |
1da177e4 LT |
157 | } |
158 | ||
f5b087b5 DR |
159 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
160 | { | |
6d556294 | 161 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
162 | } |
163 | ||
164 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
165 | const nodemask_t *rel) | |
166 | { | |
167 | nodemask_t tmp; | |
168 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
169 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
170 | } |
171 | ||
37012946 DR |
172 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
173 | { | |
174 | if (nodes_empty(*nodes)) | |
175 | return -EINVAL; | |
176 | pol->v.nodes = *nodes; | |
177 | return 0; | |
178 | } | |
179 | ||
180 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
181 | { | |
182 | if (!nodes) | |
fc36b8d3 | 183 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
184 | else if (nodes_empty(*nodes)) |
185 | return -EINVAL; /* no allowed nodes */ | |
186 | else | |
187 | pol->v.preferred_node = first_node(*nodes); | |
188 | return 0; | |
189 | } | |
190 | ||
191 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
192 | { | |
193 | if (!is_valid_nodemask(nodes)) | |
194 | return -EINVAL; | |
195 | pol->v.nodes = *nodes; | |
196 | return 0; | |
197 | } | |
198 | ||
58568d2a MX |
199 | /* |
200 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
201 | * any, for the new policy. mpol_new() has already validated the nodes | |
202 | * parameter with respect to the policy mode and flags. But, we need to | |
203 | * handle an empty nodemask with MPOL_PREFERRED here. | |
204 | * | |
205 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
206 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
207 | */ | |
4bfc4495 KH |
208 | static int mpol_set_nodemask(struct mempolicy *pol, |
209 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 210 | { |
58568d2a MX |
211 | int ret; |
212 | ||
213 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
214 | if (pol == NULL) | |
215 | return 0; | |
4bfc4495 KH |
216 | /* Check N_HIGH_MEMORY */ |
217 | nodes_and(nsc->mask1, | |
218 | cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]); | |
58568d2a MX |
219 | |
220 | VM_BUG_ON(!nodes); | |
221 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
222 | nodes = NULL; /* explicit local allocation */ | |
223 | else { | |
224 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
4bfc4495 | 225 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
58568d2a | 226 | else |
4bfc4495 KH |
227 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
228 | ||
58568d2a MX |
229 | if (mpol_store_user_nodemask(pol)) |
230 | pol->w.user_nodemask = *nodes; | |
231 | else | |
232 | pol->w.cpuset_mems_allowed = | |
233 | cpuset_current_mems_allowed; | |
234 | } | |
235 | ||
4bfc4495 KH |
236 | if (nodes) |
237 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
238 | else | |
239 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
240 | return ret; |
241 | } | |
242 | ||
243 | /* | |
244 | * This function just creates a new policy, does some check and simple | |
245 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
246 | */ | |
028fec41 DR |
247 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
248 | nodemask_t *nodes) | |
1da177e4 LT |
249 | { |
250 | struct mempolicy *policy; | |
251 | ||
028fec41 DR |
252 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
253 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 254 | |
a720094d | 255 | if (mode == MPOL_DEFAULT) { |
3e1f0645 | 256 | if (nodes && !nodes_empty(*nodes)) |
37012946 | 257 | return ERR_PTR(-EINVAL); |
d3a71033 | 258 | return NULL; |
37012946 | 259 | } |
3e1f0645 DR |
260 | VM_BUG_ON(!nodes); |
261 | ||
262 | /* | |
263 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
264 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
265 | * All other modes require a valid pointer to a non-empty nodemask. | |
266 | */ | |
267 | if (mode == MPOL_PREFERRED) { | |
268 | if (nodes_empty(*nodes)) { | |
269 | if (((flags & MPOL_F_STATIC_NODES) || | |
270 | (flags & MPOL_F_RELATIVE_NODES))) | |
271 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 272 | } |
479e2802 PZ |
273 | } else if (mode == MPOL_LOCAL) { |
274 | if (!nodes_empty(*nodes)) | |
275 | return ERR_PTR(-EINVAL); | |
276 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
277 | } else if (nodes_empty(*nodes)) |
278 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
279 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
280 | if (!policy) | |
281 | return ERR_PTR(-ENOMEM); | |
282 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 283 | policy->mode = mode; |
3e1f0645 | 284 | policy->flags = flags; |
37012946 | 285 | |
1da177e4 | 286 | return policy; |
37012946 DR |
287 | } |
288 | ||
52cd3b07 LS |
289 | /* Slow path of a mpol destructor. */ |
290 | void __mpol_put(struct mempolicy *p) | |
291 | { | |
292 | if (!atomic_dec_and_test(&p->refcnt)) | |
293 | return; | |
52cd3b07 LS |
294 | kmem_cache_free(policy_cache, p); |
295 | } | |
296 | ||
708c1bbc MX |
297 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
298 | enum mpol_rebind_step step) | |
37012946 DR |
299 | { |
300 | } | |
301 | ||
708c1bbc MX |
302 | /* |
303 | * step: | |
304 | * MPOL_REBIND_ONCE - do rebind work at once | |
305 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
306 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
307 | */ | |
308 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
309 | enum mpol_rebind_step step) | |
37012946 DR |
310 | { |
311 | nodemask_t tmp; | |
312 | ||
313 | if (pol->flags & MPOL_F_STATIC_NODES) | |
314 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
315 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
316 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
317 | else { | |
708c1bbc MX |
318 | /* |
319 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
320 | * result | |
321 | */ | |
322 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
323 | nodes_remap(tmp, pol->v.nodes, | |
324 | pol->w.cpuset_mems_allowed, *nodes); | |
325 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
326 | } else if (step == MPOL_REBIND_STEP2) { | |
327 | tmp = pol->w.cpuset_mems_allowed; | |
328 | pol->w.cpuset_mems_allowed = *nodes; | |
329 | } else | |
330 | BUG(); | |
37012946 | 331 | } |
f5b087b5 | 332 | |
708c1bbc MX |
333 | if (nodes_empty(tmp)) |
334 | tmp = *nodes; | |
335 | ||
336 | if (step == MPOL_REBIND_STEP1) | |
337 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
338 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
339 | pol->v.nodes = tmp; | |
340 | else | |
341 | BUG(); | |
342 | ||
37012946 DR |
343 | if (!node_isset(current->il_next, tmp)) { |
344 | current->il_next = next_node(current->il_next, tmp); | |
345 | if (current->il_next >= MAX_NUMNODES) | |
346 | current->il_next = first_node(tmp); | |
347 | if (current->il_next >= MAX_NUMNODES) | |
348 | current->il_next = numa_node_id(); | |
349 | } | |
350 | } | |
351 | ||
352 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
353 | const nodemask_t *nodes, |
354 | enum mpol_rebind_step step) | |
37012946 DR |
355 | { |
356 | nodemask_t tmp; | |
357 | ||
37012946 DR |
358 | if (pol->flags & MPOL_F_STATIC_NODES) { |
359 | int node = first_node(pol->w.user_nodemask); | |
360 | ||
fc36b8d3 | 361 | if (node_isset(node, *nodes)) { |
37012946 | 362 | pol->v.preferred_node = node; |
fc36b8d3 LS |
363 | pol->flags &= ~MPOL_F_LOCAL; |
364 | } else | |
365 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
366 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
367 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
368 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 369 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
370 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
371 | pol->w.cpuset_mems_allowed, | |
372 | *nodes); | |
373 | pol->w.cpuset_mems_allowed = *nodes; | |
374 | } | |
1da177e4 LT |
375 | } |
376 | ||
708c1bbc MX |
377 | /* |
378 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
379 | * | |
380 | * If read-side task has no lock to protect task->mempolicy, write-side | |
381 | * task will rebind the task->mempolicy by two step. The first step is | |
382 | * setting all the newly nodes, and the second step is cleaning all the | |
383 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
384 | * page. | |
385 | * If we have a lock to protect task->mempolicy in read-side, we do | |
386 | * rebind directly. | |
387 | * | |
388 | * step: | |
389 | * MPOL_REBIND_ONCE - do rebind work at once | |
390 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
391 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
392 | */ | |
393 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
394 | enum mpol_rebind_step step) | |
1d0d2680 | 395 | { |
1d0d2680 DR |
396 | if (!pol) |
397 | return; | |
89c522c7 | 398 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
1d0d2680 DR |
399 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
400 | return; | |
708c1bbc MX |
401 | |
402 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
403 | return; | |
404 | ||
405 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
406 | BUG(); | |
407 | ||
408 | if (step == MPOL_REBIND_STEP1) | |
409 | pol->flags |= MPOL_F_REBINDING; | |
410 | else if (step == MPOL_REBIND_STEP2) | |
411 | pol->flags &= ~MPOL_F_REBINDING; | |
412 | else if (step >= MPOL_REBIND_NSTEP) | |
413 | BUG(); | |
414 | ||
415 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
416 | } |
417 | ||
418 | /* | |
419 | * Wrapper for mpol_rebind_policy() that just requires task | |
420 | * pointer, and updates task mempolicy. | |
58568d2a MX |
421 | * |
422 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
423 | */ |
424 | ||
708c1bbc MX |
425 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
426 | enum mpol_rebind_step step) | |
1d0d2680 | 427 | { |
708c1bbc | 428 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
429 | } |
430 | ||
431 | /* | |
432 | * Rebind each vma in mm to new nodemask. | |
433 | * | |
434 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
435 | */ | |
436 | ||
437 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
438 | { | |
439 | struct vm_area_struct *vma; | |
440 | ||
441 | down_write(&mm->mmap_sem); | |
442 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 443 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
444 | up_write(&mm->mmap_sem); |
445 | } | |
446 | ||
37012946 DR |
447 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
448 | [MPOL_DEFAULT] = { | |
449 | .rebind = mpol_rebind_default, | |
450 | }, | |
451 | [MPOL_INTERLEAVE] = { | |
452 | .create = mpol_new_interleave, | |
453 | .rebind = mpol_rebind_nodemask, | |
454 | }, | |
455 | [MPOL_PREFERRED] = { | |
456 | .create = mpol_new_preferred, | |
457 | .rebind = mpol_rebind_preferred, | |
458 | }, | |
459 | [MPOL_BIND] = { | |
460 | .create = mpol_new_bind, | |
461 | .rebind = mpol_rebind_nodemask, | |
462 | }, | |
463 | }; | |
464 | ||
fc301289 CL |
465 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
466 | unsigned long flags); | |
1a75a6c8 | 467 | |
38e35860 | 468 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 469 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
470 | unsigned long addr, unsigned long end, |
471 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 472 | void *private) |
1da177e4 | 473 | { |
91612e0d HD |
474 | pte_t *orig_pte; |
475 | pte_t *pte; | |
705e87c0 | 476 | spinlock_t *ptl; |
941150a3 | 477 | |
705e87c0 | 478 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 479 | do { |
6aab341e | 480 | struct page *page; |
25ba77c1 | 481 | int nid; |
91612e0d HD |
482 | |
483 | if (!pte_present(*pte)) | |
1da177e4 | 484 | continue; |
6aab341e LT |
485 | page = vm_normal_page(vma, addr, *pte); |
486 | if (!page) | |
1da177e4 | 487 | continue; |
053837fc | 488 | /* |
62b61f61 HD |
489 | * vm_normal_page() filters out zero pages, but there might |
490 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
491 | * And we cannot move PageKsm pages sensibly or safely yet. | |
053837fc | 492 | */ |
62b61f61 | 493 | if (PageReserved(page) || PageKsm(page)) |
f4598c8b | 494 | continue; |
6aab341e | 495 | nid = page_to_nid(page); |
38e35860 CL |
496 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
497 | continue; | |
498 | ||
b1f72d18 | 499 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 500 | migrate_page_add(page, private, flags); |
38e35860 CL |
501 | else |
502 | break; | |
91612e0d | 503 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 504 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
505 | return addr != end; |
506 | } | |
507 | ||
b5810039 | 508 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
509 | unsigned long addr, unsigned long end, |
510 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 511 | void *private) |
91612e0d HD |
512 | { |
513 | pmd_t *pmd; | |
514 | unsigned long next; | |
515 | ||
516 | pmd = pmd_offset(pud, addr); | |
517 | do { | |
518 | next = pmd_addr_end(addr, end); | |
bae9c19b | 519 | split_huge_page_pmd(vma->vm_mm, pmd); |
1a5a9906 | 520 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
91612e0d | 521 | continue; |
dc9aa5b9 | 522 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 523 | flags, private)) |
91612e0d HD |
524 | return -EIO; |
525 | } while (pmd++, addr = next, addr != end); | |
526 | return 0; | |
527 | } | |
528 | ||
b5810039 | 529 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
530 | unsigned long addr, unsigned long end, |
531 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 532 | void *private) |
91612e0d HD |
533 | { |
534 | pud_t *pud; | |
535 | unsigned long next; | |
536 | ||
537 | pud = pud_offset(pgd, addr); | |
538 | do { | |
539 | next = pud_addr_end(addr, end); | |
540 | if (pud_none_or_clear_bad(pud)) | |
541 | continue; | |
dc9aa5b9 | 542 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 543 | flags, private)) |
91612e0d HD |
544 | return -EIO; |
545 | } while (pud++, addr = next, addr != end); | |
546 | return 0; | |
547 | } | |
548 | ||
b5810039 | 549 | static inline int check_pgd_range(struct vm_area_struct *vma, |
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 | pgd_t *pgd; | |
555 | unsigned long next; | |
556 | ||
b5810039 | 557 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
558 | do { |
559 | next = pgd_addr_end(addr, end); | |
560 | if (pgd_none_or_clear_bad(pgd)) | |
561 | continue; | |
dc9aa5b9 | 562 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 563 | flags, private)) |
91612e0d HD |
564 | return -EIO; |
565 | } while (pgd++, addr = next, addr != end); | |
566 | return 0; | |
1da177e4 LT |
567 | } |
568 | ||
b24f53a0 LS |
569 | #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE |
570 | /* | |
4b10e7d5 MG |
571 | * This is used to mark a range of virtual addresses to be inaccessible. |
572 | * These are later cleared by a NUMA hinting fault. Depending on these | |
573 | * faults, pages may be migrated for better NUMA placement. | |
574 | * | |
575 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
576 | * an architecture makes a different choice, it will need further | |
577 | * changes to the core. | |
b24f53a0 | 578 | */ |
4b10e7d5 MG |
579 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
580 | unsigned long addr, unsigned long end) | |
b24f53a0 | 581 | { |
4b10e7d5 MG |
582 | int nr_updated; |
583 | BUILD_BUG_ON(_PAGE_NUMA != _PAGE_PROTNONE); | |
b24f53a0 | 584 | |
4b10e7d5 | 585 | nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1); |
b24f53a0 | 586 | |
4b10e7d5 | 587 | return nr_updated; |
b24f53a0 LS |
588 | } |
589 | #else | |
590 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
591 | unsigned long addr, unsigned long end) | |
592 | { | |
593 | return 0; | |
594 | } | |
595 | #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */ | |
596 | ||
dc9aa5b9 CL |
597 | /* |
598 | * Check if all pages in a range are on a set of nodes. | |
599 | * If pagelist != NULL then isolate pages from the LRU and | |
600 | * put them on the pagelist. | |
601 | */ | |
1da177e4 LT |
602 | static struct vm_area_struct * |
603 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 604 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
605 | { |
606 | int err; | |
607 | struct vm_area_struct *first, *vma, *prev; | |
608 | ||
053837fc | 609 | |
1da177e4 LT |
610 | first = find_vma(mm, start); |
611 | if (!first) | |
612 | return ERR_PTR(-EFAULT); | |
613 | prev = NULL; | |
614 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
b24f53a0 LS |
615 | unsigned long endvma = vma->vm_end; |
616 | ||
617 | if (endvma > end) | |
618 | endvma = end; | |
619 | if (vma->vm_start > start) | |
620 | start = vma->vm_start; | |
621 | ||
dc9aa5b9 CL |
622 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
623 | if (!vma->vm_next && vma->vm_end < end) | |
624 | return ERR_PTR(-EFAULT); | |
625 | if (prev && prev->vm_end < vma->vm_start) | |
626 | return ERR_PTR(-EFAULT); | |
627 | } | |
b24f53a0 LS |
628 | |
629 | if (is_vm_hugetlb_page(vma)) | |
630 | goto next; | |
631 | ||
632 | if (flags & MPOL_MF_LAZY) { | |
633 | change_prot_numa(vma, start, endvma); | |
634 | goto next; | |
635 | } | |
636 | ||
637 | if ((flags & MPOL_MF_STRICT) || | |
dc9aa5b9 | 638 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
b24f53a0 | 639 | vma_migratable(vma))) { |
dc9aa5b9 | 640 | |
dc9aa5b9 | 641 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 642 | flags, private); |
1da177e4 LT |
643 | if (err) { |
644 | first = ERR_PTR(err); | |
645 | break; | |
646 | } | |
647 | } | |
b24f53a0 | 648 | next: |
1da177e4 LT |
649 | prev = vma; |
650 | } | |
651 | return first; | |
652 | } | |
653 | ||
869833f2 KM |
654 | /* |
655 | * Apply policy to a single VMA | |
656 | * This must be called with the mmap_sem held for writing. | |
657 | */ | |
658 | static int vma_replace_policy(struct vm_area_struct *vma, | |
659 | struct mempolicy *pol) | |
8d34694c | 660 | { |
869833f2 KM |
661 | int err; |
662 | struct mempolicy *old; | |
663 | struct mempolicy *new; | |
8d34694c KM |
664 | |
665 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
666 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
667 | vma->vm_ops, vma->vm_file, | |
668 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
669 | ||
869833f2 KM |
670 | new = mpol_dup(pol); |
671 | if (IS_ERR(new)) | |
672 | return PTR_ERR(new); | |
673 | ||
674 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 675 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
676 | if (err) |
677 | goto err_out; | |
8d34694c | 678 | } |
869833f2 KM |
679 | |
680 | old = vma->vm_policy; | |
681 | vma->vm_policy = new; /* protected by mmap_sem */ | |
682 | mpol_put(old); | |
683 | ||
684 | return 0; | |
685 | err_out: | |
686 | mpol_put(new); | |
8d34694c KM |
687 | return err; |
688 | } | |
689 | ||
1da177e4 | 690 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
691 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
692 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
693 | { |
694 | struct vm_area_struct *next; | |
9d8cebd4 KM |
695 | struct vm_area_struct *prev; |
696 | struct vm_area_struct *vma; | |
697 | int err = 0; | |
e26a5114 | 698 | pgoff_t pgoff; |
9d8cebd4 KM |
699 | unsigned long vmstart; |
700 | unsigned long vmend; | |
1da177e4 | 701 | |
097d5910 | 702 | vma = find_vma(mm, start); |
9d8cebd4 KM |
703 | if (!vma || vma->vm_start > start) |
704 | return -EFAULT; | |
705 | ||
097d5910 | 706 | prev = vma->vm_prev; |
e26a5114 KM |
707 | if (start > vma->vm_start) |
708 | prev = vma; | |
709 | ||
9d8cebd4 | 710 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 711 | next = vma->vm_next; |
9d8cebd4 KM |
712 | vmstart = max(start, vma->vm_start); |
713 | vmend = min(end, vma->vm_end); | |
714 | ||
e26a5114 KM |
715 | if (mpol_equal(vma_policy(vma), new_pol)) |
716 | continue; | |
717 | ||
718 | pgoff = vma->vm_pgoff + | |
719 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 720 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 721 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 722 | new_pol); |
9d8cebd4 KM |
723 | if (prev) { |
724 | vma = prev; | |
725 | next = vma->vm_next; | |
726 | continue; | |
727 | } | |
728 | if (vma->vm_start != vmstart) { | |
729 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
730 | if (err) | |
731 | goto out; | |
732 | } | |
733 | if (vma->vm_end != vmend) { | |
734 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
735 | if (err) | |
736 | goto out; | |
737 | } | |
869833f2 | 738 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
739 | if (err) |
740 | goto out; | |
1da177e4 | 741 | } |
9d8cebd4 KM |
742 | |
743 | out: | |
1da177e4 LT |
744 | return err; |
745 | } | |
746 | ||
c61afb18 PJ |
747 | /* |
748 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
749 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
750 | * with other PF_* flag bits) on memory allocation hot code paths. | |
751 | * | |
752 | * If called from outside this file, the task 'p' should -only- be | |
753 | * a newly forked child not yet visible on the task list, because | |
754 | * manipulating the task flags of a visible task is not safe. | |
755 | * | |
756 | * The above limitation is why this routine has the funny name | |
757 | * mpol_fix_fork_child_flag(). | |
758 | * | |
759 | * It is also safe to call this with a task pointer of current, | |
760 | * which the static wrapper mpol_set_task_struct_flag() does, | |
761 | * for use within this file. | |
762 | */ | |
763 | ||
764 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
765 | { | |
766 | if (p->mempolicy) | |
767 | p->flags |= PF_MEMPOLICY; | |
768 | else | |
769 | p->flags &= ~PF_MEMPOLICY; | |
770 | } | |
771 | ||
772 | static void mpol_set_task_struct_flag(void) | |
773 | { | |
774 | mpol_fix_fork_child_flag(current); | |
775 | } | |
776 | ||
1da177e4 | 777 | /* Set the process memory policy */ |
028fec41 DR |
778 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
779 | nodemask_t *nodes) | |
1da177e4 | 780 | { |
58568d2a | 781 | struct mempolicy *new, *old; |
f4e53d91 | 782 | struct mm_struct *mm = current->mm; |
4bfc4495 | 783 | NODEMASK_SCRATCH(scratch); |
58568d2a | 784 | int ret; |
1da177e4 | 785 | |
4bfc4495 KH |
786 | if (!scratch) |
787 | return -ENOMEM; | |
f4e53d91 | 788 | |
4bfc4495 KH |
789 | new = mpol_new(mode, flags, nodes); |
790 | if (IS_ERR(new)) { | |
791 | ret = PTR_ERR(new); | |
792 | goto out; | |
793 | } | |
f4e53d91 LS |
794 | /* |
795 | * prevent changing our mempolicy while show_numa_maps() | |
796 | * is using it. | |
797 | * Note: do_set_mempolicy() can be called at init time | |
798 | * with no 'mm'. | |
799 | */ | |
800 | if (mm) | |
801 | down_write(&mm->mmap_sem); | |
58568d2a | 802 | task_lock(current); |
4bfc4495 | 803 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
804 | if (ret) { |
805 | task_unlock(current); | |
806 | if (mm) | |
807 | up_write(&mm->mmap_sem); | |
808 | mpol_put(new); | |
4bfc4495 | 809 | goto out; |
58568d2a MX |
810 | } |
811 | old = current->mempolicy; | |
1da177e4 | 812 | current->mempolicy = new; |
c61afb18 | 813 | mpol_set_task_struct_flag(); |
45c4745a | 814 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 815 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 816 | current->il_next = first_node(new->v.nodes); |
58568d2a | 817 | task_unlock(current); |
f4e53d91 LS |
818 | if (mm) |
819 | up_write(&mm->mmap_sem); | |
820 | ||
58568d2a | 821 | mpol_put(old); |
4bfc4495 KH |
822 | ret = 0; |
823 | out: | |
824 | NODEMASK_SCRATCH_FREE(scratch); | |
825 | return ret; | |
1da177e4 LT |
826 | } |
827 | ||
bea904d5 LS |
828 | /* |
829 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
830 | * |
831 | * Called with task's alloc_lock held | |
bea904d5 LS |
832 | */ |
833 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 834 | { |
dfcd3c0d | 835 | nodes_clear(*nodes); |
bea904d5 LS |
836 | if (p == &default_policy) |
837 | return; | |
838 | ||
45c4745a | 839 | switch (p->mode) { |
19770b32 MG |
840 | case MPOL_BIND: |
841 | /* Fall through */ | |
1da177e4 | 842 | case MPOL_INTERLEAVE: |
dfcd3c0d | 843 | *nodes = p->v.nodes; |
1da177e4 LT |
844 | break; |
845 | case MPOL_PREFERRED: | |
fc36b8d3 | 846 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 847 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 848 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
849 | break; |
850 | default: | |
851 | BUG(); | |
852 | } | |
853 | } | |
854 | ||
855 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
856 | { | |
857 | struct page *p; | |
858 | int err; | |
859 | ||
860 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
861 | if (err >= 0) { | |
862 | err = page_to_nid(p); | |
863 | put_page(p); | |
864 | } | |
865 | return err; | |
866 | } | |
867 | ||
1da177e4 | 868 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
869 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
870 | unsigned long addr, unsigned long flags) | |
1da177e4 | 871 | { |
8bccd85f | 872 | int err; |
1da177e4 LT |
873 | struct mm_struct *mm = current->mm; |
874 | struct vm_area_struct *vma = NULL; | |
875 | struct mempolicy *pol = current->mempolicy; | |
876 | ||
754af6f5 LS |
877 | if (flags & |
878 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 879 | return -EINVAL; |
754af6f5 LS |
880 | |
881 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
882 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
883 | return -EINVAL; | |
884 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 885 | task_lock(current); |
754af6f5 | 886 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 887 | task_unlock(current); |
754af6f5 LS |
888 | return 0; |
889 | } | |
890 | ||
1da177e4 | 891 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
892 | /* |
893 | * Do NOT fall back to task policy if the | |
894 | * vma/shared policy at addr is NULL. We | |
895 | * want to return MPOL_DEFAULT in this case. | |
896 | */ | |
1da177e4 LT |
897 | down_read(&mm->mmap_sem); |
898 | vma = find_vma_intersection(mm, addr, addr+1); | |
899 | if (!vma) { | |
900 | up_read(&mm->mmap_sem); | |
901 | return -EFAULT; | |
902 | } | |
903 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
904 | pol = vma->vm_ops->get_policy(vma, addr); | |
905 | else | |
906 | pol = vma->vm_policy; | |
907 | } else if (addr) | |
908 | return -EINVAL; | |
909 | ||
910 | if (!pol) | |
bea904d5 | 911 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
912 | |
913 | if (flags & MPOL_F_NODE) { | |
914 | if (flags & MPOL_F_ADDR) { | |
915 | err = lookup_node(mm, addr); | |
916 | if (err < 0) | |
917 | goto out; | |
8bccd85f | 918 | *policy = err; |
1da177e4 | 919 | } else if (pol == current->mempolicy && |
45c4745a | 920 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 921 | *policy = current->il_next; |
1da177e4 LT |
922 | } else { |
923 | err = -EINVAL; | |
924 | goto out; | |
925 | } | |
bea904d5 LS |
926 | } else { |
927 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
928 | pol->mode; | |
d79df630 DR |
929 | /* |
930 | * Internal mempolicy flags must be masked off before exposing | |
931 | * the policy to userspace. | |
932 | */ | |
933 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 934 | } |
1da177e4 LT |
935 | |
936 | if (vma) { | |
937 | up_read(¤t->mm->mmap_sem); | |
938 | vma = NULL; | |
939 | } | |
940 | ||
1da177e4 | 941 | err = 0; |
58568d2a | 942 | if (nmask) { |
c6b6ef8b LS |
943 | if (mpol_store_user_nodemask(pol)) { |
944 | *nmask = pol->w.user_nodemask; | |
945 | } else { | |
946 | task_lock(current); | |
947 | get_policy_nodemask(pol, nmask); | |
948 | task_unlock(current); | |
949 | } | |
58568d2a | 950 | } |
1da177e4 LT |
951 | |
952 | out: | |
52cd3b07 | 953 | mpol_cond_put(pol); |
1da177e4 LT |
954 | if (vma) |
955 | up_read(¤t->mm->mmap_sem); | |
956 | return err; | |
957 | } | |
958 | ||
b20a3503 | 959 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
960 | /* |
961 | * page migration | |
962 | */ | |
fc301289 CL |
963 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
964 | unsigned long flags) | |
6ce3c4c0 CL |
965 | { |
966 | /* | |
fc301289 | 967 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 968 | */ |
62695a84 NP |
969 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
970 | if (!isolate_lru_page(page)) { | |
971 | list_add_tail(&page->lru, pagelist); | |
6d9c285a KM |
972 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
973 | page_is_file_cache(page)); | |
62695a84 NP |
974 | } |
975 | } | |
7e2ab150 | 976 | } |
6ce3c4c0 | 977 | |
742755a1 | 978 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 979 | { |
6484eb3e | 980 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
981 | } |
982 | ||
7e2ab150 CL |
983 | /* |
984 | * Migrate pages from one node to a target node. | |
985 | * Returns error or the number of pages not migrated. | |
986 | */ | |
dbcb0f19 AB |
987 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
988 | int flags) | |
7e2ab150 CL |
989 | { |
990 | nodemask_t nmask; | |
991 | LIST_HEAD(pagelist); | |
992 | int err = 0; | |
993 | ||
994 | nodes_clear(nmask); | |
995 | node_set(source, nmask); | |
6ce3c4c0 | 996 | |
08270807 MK |
997 | /* |
998 | * This does not "check" the range but isolates all pages that | |
999 | * need migration. Between passing in the full user address | |
1000 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1001 | */ | |
1002 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
1003 | check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, | |
7e2ab150 CL |
1004 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1005 | ||
cf608ac1 | 1006 | if (!list_empty(&pagelist)) { |
7f0f2496 | 1007 | err = migrate_pages(&pagelist, new_node_page, dest, |
7b2a2d4a MG |
1008 | false, MIGRATE_SYNC, |
1009 | MR_SYSCALL); | |
cf608ac1 MK |
1010 | if (err) |
1011 | putback_lru_pages(&pagelist); | |
1012 | } | |
95a402c3 | 1013 | |
7e2ab150 | 1014 | return err; |
6ce3c4c0 CL |
1015 | } |
1016 | ||
39743889 | 1017 | /* |
7e2ab150 CL |
1018 | * Move pages between the two nodesets so as to preserve the physical |
1019 | * layout as much as possible. | |
39743889 CL |
1020 | * |
1021 | * Returns the number of page that could not be moved. | |
1022 | */ | |
0ce72d4f AM |
1023 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1024 | const nodemask_t *to, int flags) | |
39743889 | 1025 | { |
7e2ab150 | 1026 | int busy = 0; |
0aedadf9 | 1027 | int err; |
7e2ab150 | 1028 | nodemask_t tmp; |
39743889 | 1029 | |
0aedadf9 CL |
1030 | err = migrate_prep(); |
1031 | if (err) | |
1032 | return err; | |
1033 | ||
53f2556b | 1034 | down_read(&mm->mmap_sem); |
39743889 | 1035 | |
0ce72d4f | 1036 | err = migrate_vmas(mm, from, to, flags); |
7b2259b3 CL |
1037 | if (err) |
1038 | goto out; | |
1039 | ||
da0aa138 KM |
1040 | /* |
1041 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1042 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1043 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1044 | * The pair of nodemasks 'to' and 'from' define the map. | |
1045 | * | |
1046 | * If no pair of bits is found that way, fallback to picking some | |
1047 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1048 | * 'source' and 'dest' bits are the same, this represents a node | |
1049 | * that will be migrating to itself, so no pages need move. | |
1050 | * | |
1051 | * If no bits are left in 'tmp', or if all remaining bits left | |
1052 | * in 'tmp' correspond to the same bit in 'to', return false | |
1053 | * (nothing left to migrate). | |
1054 | * | |
1055 | * This lets us pick a pair of nodes to migrate between, such that | |
1056 | * if possible the dest node is not already occupied by some other | |
1057 | * source node, minimizing the risk of overloading the memory on a | |
1058 | * node that would happen if we migrated incoming memory to a node | |
1059 | * before migrating outgoing memory source that same node. | |
1060 | * | |
1061 | * A single scan of tmp is sufficient. As we go, we remember the | |
1062 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1063 | * that not only moved, but what's better, moved to an empty slot | |
1064 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1065 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1066 | * most recent <s, d> pair that moved. If we get all the way through |
1067 | * the scan of tmp without finding any node that moved, much less | |
1068 | * moved to an empty node, then there is nothing left worth migrating. | |
1069 | */ | |
d4984711 | 1070 | |
0ce72d4f | 1071 | tmp = *from; |
7e2ab150 CL |
1072 | while (!nodes_empty(tmp)) { |
1073 | int s,d; | |
1074 | int source = -1; | |
1075 | int dest = 0; | |
1076 | ||
1077 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1078 | |
1079 | /* | |
1080 | * do_migrate_pages() tries to maintain the relative | |
1081 | * node relationship of the pages established between | |
1082 | * threads and memory areas. | |
1083 | * | |
1084 | * However if the number of source nodes is not equal to | |
1085 | * the number of destination nodes we can not preserve | |
1086 | * this node relative relationship. In that case, skip | |
1087 | * copying memory from a node that is in the destination | |
1088 | * mask. | |
1089 | * | |
1090 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1091 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1092 | */ | |
1093 | ||
0ce72d4f AM |
1094 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1095 | (node_isset(s, *to))) | |
4a5b18cc LW |
1096 | continue; |
1097 | ||
0ce72d4f | 1098 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1099 | if (s == d) |
1100 | continue; | |
1101 | ||
1102 | source = s; /* Node moved. Memorize */ | |
1103 | dest = d; | |
1104 | ||
1105 | /* dest not in remaining from nodes? */ | |
1106 | if (!node_isset(dest, tmp)) | |
1107 | break; | |
1108 | } | |
1109 | if (source == -1) | |
1110 | break; | |
1111 | ||
1112 | node_clear(source, tmp); | |
1113 | err = migrate_to_node(mm, source, dest, flags); | |
1114 | if (err > 0) | |
1115 | busy += err; | |
1116 | if (err < 0) | |
1117 | break; | |
39743889 | 1118 | } |
7b2259b3 | 1119 | out: |
39743889 | 1120 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1121 | if (err < 0) |
1122 | return err; | |
1123 | return busy; | |
b20a3503 CL |
1124 | |
1125 | } | |
1126 | ||
3ad33b24 LS |
1127 | /* |
1128 | * Allocate a new page for page migration based on vma policy. | |
1129 | * Start assuming that page is mapped by vma pointed to by @private. | |
1130 | * Search forward from there, if not. N.B., this assumes that the | |
1131 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1132 | * is in virtual address order. | |
1133 | */ | |
742755a1 | 1134 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1135 | { |
1136 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 1137 | unsigned long uninitialized_var(address); |
95a402c3 | 1138 | |
3ad33b24 LS |
1139 | while (vma) { |
1140 | address = page_address_in_vma(page, vma); | |
1141 | if (address != -EFAULT) | |
1142 | break; | |
1143 | vma = vma->vm_next; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * if !vma, alloc_page_vma() will use task or system default policy | |
1148 | */ | |
1149 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 1150 | } |
b20a3503 CL |
1151 | #else |
1152 | ||
1153 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1154 | unsigned long flags) | |
1155 | { | |
39743889 CL |
1156 | } |
1157 | ||
0ce72d4f AM |
1158 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1159 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1160 | { |
1161 | return -ENOSYS; | |
1162 | } | |
95a402c3 | 1163 | |
69939749 | 1164 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1165 | { |
1166 | return NULL; | |
1167 | } | |
b20a3503 CL |
1168 | #endif |
1169 | ||
dbcb0f19 | 1170 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1171 | unsigned short mode, unsigned short mode_flags, |
1172 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
1173 | { |
1174 | struct vm_area_struct *vma; | |
1175 | struct mm_struct *mm = current->mm; | |
1176 | struct mempolicy *new; | |
1177 | unsigned long end; | |
1178 | int err; | |
1179 | LIST_HEAD(pagelist); | |
1180 | ||
b24f53a0 | 1181 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1182 | return -EINVAL; |
74c00241 | 1183 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1184 | return -EPERM; |
1185 | ||
1186 | if (start & ~PAGE_MASK) | |
1187 | return -EINVAL; | |
1188 | ||
a720094d | 1189 | if (mode == MPOL_DEFAULT) |
6ce3c4c0 CL |
1190 | flags &= ~MPOL_MF_STRICT; |
1191 | ||
1192 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1193 | end = start + len; | |
1194 | ||
1195 | if (end < start) | |
1196 | return -EINVAL; | |
1197 | if (end == start) | |
1198 | return 0; | |
1199 | ||
028fec41 | 1200 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1201 | if (IS_ERR(new)) |
1202 | return PTR_ERR(new); | |
1203 | ||
b24f53a0 LS |
1204 | if (flags & MPOL_MF_LAZY) |
1205 | new->flags |= MPOL_F_MOF; | |
1206 | ||
6ce3c4c0 CL |
1207 | /* |
1208 | * If we are using the default policy then operation | |
1209 | * on discontinuous address spaces is okay after all | |
1210 | */ | |
1211 | if (!new) | |
1212 | flags |= MPOL_MF_DISCONTIG_OK; | |
1213 | ||
028fec41 DR |
1214 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1215 | start, start + len, mode, mode_flags, | |
1216 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 | 1217 | |
0aedadf9 CL |
1218 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1219 | ||
1220 | err = migrate_prep(); | |
1221 | if (err) | |
b05ca738 | 1222 | goto mpol_out; |
0aedadf9 | 1223 | } |
4bfc4495 KH |
1224 | { |
1225 | NODEMASK_SCRATCH(scratch); | |
1226 | if (scratch) { | |
1227 | down_write(&mm->mmap_sem); | |
1228 | task_lock(current); | |
1229 | err = mpol_set_nodemask(new, nmask, scratch); | |
1230 | task_unlock(current); | |
1231 | if (err) | |
1232 | up_write(&mm->mmap_sem); | |
1233 | } else | |
1234 | err = -ENOMEM; | |
1235 | NODEMASK_SCRATCH_FREE(scratch); | |
1236 | } | |
b05ca738 KM |
1237 | if (err) |
1238 | goto mpol_out; | |
1239 | ||
6ce3c4c0 CL |
1240 | vma = check_range(mm, start, end, nmask, |
1241 | flags | MPOL_MF_INVERT, &pagelist); | |
1242 | ||
b24f53a0 | 1243 | err = PTR_ERR(vma); /* maybe ... */ |
a720094d | 1244 | if (!IS_ERR(vma)) |
9d8cebd4 | 1245 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1246 | |
b24f53a0 LS |
1247 | if (!err) { |
1248 | int nr_failed = 0; | |
1249 | ||
cf608ac1 | 1250 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1251 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
95a402c3 | 1252 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
7f0f2496 | 1253 | (unsigned long)vma, |
7b2a2d4a MG |
1254 | false, MIGRATE_SYNC, |
1255 | MR_MEMPOLICY_MBIND); | |
cf608ac1 MK |
1256 | if (nr_failed) |
1257 | putback_lru_pages(&pagelist); | |
1258 | } | |
6ce3c4c0 | 1259 | |
b24f53a0 | 1260 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
6ce3c4c0 | 1261 | err = -EIO; |
ab8a3e14 KM |
1262 | } else |
1263 | putback_lru_pages(&pagelist); | |
b20a3503 | 1264 | |
6ce3c4c0 | 1265 | up_write(&mm->mmap_sem); |
b05ca738 | 1266 | mpol_out: |
f0be3d32 | 1267 | mpol_put(new); |
6ce3c4c0 CL |
1268 | return err; |
1269 | } | |
1270 | ||
8bccd85f CL |
1271 | /* |
1272 | * User space interface with variable sized bitmaps for nodelists. | |
1273 | */ | |
1274 | ||
1275 | /* Copy a node mask from user space. */ | |
39743889 | 1276 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1277 | unsigned long maxnode) |
1278 | { | |
1279 | unsigned long k; | |
1280 | unsigned long nlongs; | |
1281 | unsigned long endmask; | |
1282 | ||
1283 | --maxnode; | |
1284 | nodes_clear(*nodes); | |
1285 | if (maxnode == 0 || !nmask) | |
1286 | return 0; | |
a9c930ba | 1287 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1288 | return -EINVAL; |
8bccd85f CL |
1289 | |
1290 | nlongs = BITS_TO_LONGS(maxnode); | |
1291 | if ((maxnode % BITS_PER_LONG) == 0) | |
1292 | endmask = ~0UL; | |
1293 | else | |
1294 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1295 | ||
1296 | /* When the user specified more nodes than supported just check | |
1297 | if the non supported part is all zero. */ | |
1298 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1299 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1300 | return -EINVAL; | |
1301 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1302 | unsigned long t; | |
1303 | if (get_user(t, nmask + k)) | |
1304 | return -EFAULT; | |
1305 | if (k == nlongs - 1) { | |
1306 | if (t & endmask) | |
1307 | return -EINVAL; | |
1308 | } else if (t) | |
1309 | return -EINVAL; | |
1310 | } | |
1311 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1312 | endmask = ~0UL; | |
1313 | } | |
1314 | ||
1315 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1316 | return -EFAULT; | |
1317 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1318 | return 0; | |
1319 | } | |
1320 | ||
1321 | /* Copy a kernel node mask to user space */ | |
1322 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1323 | nodemask_t *nodes) | |
1324 | { | |
1325 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1326 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1327 | ||
1328 | if (copy > nbytes) { | |
1329 | if (copy > PAGE_SIZE) | |
1330 | return -EINVAL; | |
1331 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1332 | return -EFAULT; | |
1333 | copy = nbytes; | |
1334 | } | |
1335 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1336 | } | |
1337 | ||
938bb9f5 HC |
1338 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1339 | unsigned long, mode, unsigned long __user *, nmask, | |
1340 | unsigned long, maxnode, unsigned, flags) | |
8bccd85f CL |
1341 | { |
1342 | nodemask_t nodes; | |
1343 | int err; | |
028fec41 | 1344 | unsigned short mode_flags; |
8bccd85f | 1345 | |
028fec41 DR |
1346 | mode_flags = mode & MPOL_MODE_FLAGS; |
1347 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1348 | if (mode >= MPOL_MAX) |
1349 | return -EINVAL; | |
4c50bc01 DR |
1350 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1351 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1352 | return -EINVAL; | |
8bccd85f CL |
1353 | err = get_nodes(&nodes, nmask, maxnode); |
1354 | if (err) | |
1355 | return err; | |
028fec41 | 1356 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1357 | } |
1358 | ||
1359 | /* Set the process memory policy */ | |
938bb9f5 HC |
1360 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
1361 | unsigned long, maxnode) | |
8bccd85f CL |
1362 | { |
1363 | int err; | |
1364 | nodemask_t nodes; | |
028fec41 | 1365 | unsigned short flags; |
8bccd85f | 1366 | |
028fec41 DR |
1367 | flags = mode & MPOL_MODE_FLAGS; |
1368 | mode &= ~MPOL_MODE_FLAGS; | |
1369 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1370 | return -EINVAL; |
4c50bc01 DR |
1371 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1372 | return -EINVAL; | |
8bccd85f CL |
1373 | err = get_nodes(&nodes, nmask, maxnode); |
1374 | if (err) | |
1375 | return err; | |
028fec41 | 1376 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1377 | } |
1378 | ||
938bb9f5 HC |
1379 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1380 | const unsigned long __user *, old_nodes, | |
1381 | const unsigned long __user *, new_nodes) | |
39743889 | 1382 | { |
c69e8d9c | 1383 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1384 | struct mm_struct *mm = NULL; |
39743889 | 1385 | struct task_struct *task; |
39743889 CL |
1386 | nodemask_t task_nodes; |
1387 | int err; | |
596d7cfa KM |
1388 | nodemask_t *old; |
1389 | nodemask_t *new; | |
1390 | NODEMASK_SCRATCH(scratch); | |
1391 | ||
1392 | if (!scratch) | |
1393 | return -ENOMEM; | |
39743889 | 1394 | |
596d7cfa KM |
1395 | old = &scratch->mask1; |
1396 | new = &scratch->mask2; | |
1397 | ||
1398 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1399 | if (err) |
596d7cfa | 1400 | goto out; |
39743889 | 1401 | |
596d7cfa | 1402 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1403 | if (err) |
596d7cfa | 1404 | goto out; |
39743889 CL |
1405 | |
1406 | /* Find the mm_struct */ | |
55cfaa3c | 1407 | rcu_read_lock(); |
228ebcbe | 1408 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1409 | if (!task) { |
55cfaa3c | 1410 | rcu_read_unlock(); |
596d7cfa KM |
1411 | err = -ESRCH; |
1412 | goto out; | |
39743889 | 1413 | } |
3268c63e | 1414 | get_task_struct(task); |
39743889 | 1415 | |
596d7cfa | 1416 | err = -EINVAL; |
39743889 CL |
1417 | |
1418 | /* | |
1419 | * Check if this process has the right to modify the specified | |
1420 | * process. The right exists if the process has administrative | |
7f927fcc | 1421 | * capabilities, superuser privileges or the same |
39743889 CL |
1422 | * userid as the target process. |
1423 | */ | |
c69e8d9c | 1424 | tcred = __task_cred(task); |
b38a86eb EB |
1425 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1426 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1427 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1428 | rcu_read_unlock(); |
39743889 | 1429 | err = -EPERM; |
3268c63e | 1430 | goto out_put; |
39743889 | 1431 | } |
c69e8d9c | 1432 | rcu_read_unlock(); |
39743889 CL |
1433 | |
1434 | task_nodes = cpuset_mems_allowed(task); | |
1435 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1436 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1437 | err = -EPERM; |
3268c63e | 1438 | goto out_put; |
39743889 CL |
1439 | } |
1440 | ||
596d7cfa | 1441 | if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) { |
3b42d28b | 1442 | err = -EINVAL; |
3268c63e | 1443 | goto out_put; |
3b42d28b CL |
1444 | } |
1445 | ||
86c3a764 DQ |
1446 | err = security_task_movememory(task); |
1447 | if (err) | |
3268c63e | 1448 | goto out_put; |
86c3a764 | 1449 | |
3268c63e CL |
1450 | mm = get_task_mm(task); |
1451 | put_task_struct(task); | |
f2a9ef88 SL |
1452 | |
1453 | if (!mm) { | |
3268c63e | 1454 | err = -EINVAL; |
f2a9ef88 SL |
1455 | goto out; |
1456 | } | |
1457 | ||
1458 | err = do_migrate_pages(mm, old, new, | |
1459 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1460 | |
1461 | mmput(mm); | |
1462 | out: | |
596d7cfa KM |
1463 | NODEMASK_SCRATCH_FREE(scratch); |
1464 | ||
39743889 | 1465 | return err; |
3268c63e CL |
1466 | |
1467 | out_put: | |
1468 | put_task_struct(task); | |
1469 | goto out; | |
1470 | ||
39743889 CL |
1471 | } |
1472 | ||
1473 | ||
8bccd85f | 1474 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1475 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1476 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1477 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1478 | { |
dbcb0f19 AB |
1479 | int err; |
1480 | int uninitialized_var(pval); | |
8bccd85f CL |
1481 | nodemask_t nodes; |
1482 | ||
1483 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1484 | return -EINVAL; | |
1485 | ||
1486 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1487 | ||
1488 | if (err) | |
1489 | return err; | |
1490 | ||
1491 | if (policy && put_user(pval, policy)) | |
1492 | return -EFAULT; | |
1493 | ||
1494 | if (nmask) | |
1495 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1496 | ||
1497 | return err; | |
1498 | } | |
1499 | ||
1da177e4 LT |
1500 | #ifdef CONFIG_COMPAT |
1501 | ||
1502 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1503 | compat_ulong_t __user *nmask, | |
1504 | compat_ulong_t maxnode, | |
1505 | compat_ulong_t addr, compat_ulong_t flags) | |
1506 | { | |
1507 | long err; | |
1508 | unsigned long __user *nm = NULL; | |
1509 | unsigned long nr_bits, alloc_size; | |
1510 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1511 | ||
1512 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1513 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1514 | ||
1515 | if (nmask) | |
1516 | nm = compat_alloc_user_space(alloc_size); | |
1517 | ||
1518 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1519 | ||
1520 | if (!err && nmask) { | |
2bbff6c7 KH |
1521 | unsigned long copy_size; |
1522 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1523 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1524 | /* ensure entire bitmap is zeroed */ |
1525 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1526 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1527 | } | |
1528 | ||
1529 | return err; | |
1530 | } | |
1531 | ||
1532 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1533 | compat_ulong_t maxnode) | |
1534 | { | |
1535 | long err = 0; | |
1536 | unsigned long __user *nm = NULL; | |
1537 | unsigned long nr_bits, alloc_size; | |
1538 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1539 | ||
1540 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1541 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1542 | ||
1543 | if (nmask) { | |
1544 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1545 | nm = compat_alloc_user_space(alloc_size); | |
1546 | err |= copy_to_user(nm, bm, alloc_size); | |
1547 | } | |
1548 | ||
1549 | if (err) | |
1550 | return -EFAULT; | |
1551 | ||
1552 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1553 | } | |
1554 | ||
1555 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1556 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1557 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1558 | { | |
1559 | long err = 0; | |
1560 | unsigned long __user *nm = NULL; | |
1561 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1562 | nodemask_t bm; |
1da177e4 LT |
1563 | |
1564 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1565 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1566 | ||
1567 | if (nmask) { | |
dfcd3c0d | 1568 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1569 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1570 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1571 | } |
1572 | ||
1573 | if (err) | |
1574 | return -EFAULT; | |
1575 | ||
1576 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1577 | } | |
1578 | ||
1579 | #endif | |
1580 | ||
480eccf9 LS |
1581 | /* |
1582 | * get_vma_policy(@task, @vma, @addr) | |
1583 | * @task - task for fallback if vma policy == default | |
1584 | * @vma - virtual memory area whose policy is sought | |
1585 | * @addr - address in @vma for shared policy lookup | |
1586 | * | |
1587 | * Returns effective policy for a VMA at specified address. | |
1588 | * Falls back to @task or system default policy, as necessary. | |
32f8516a DR |
1589 | * Current or other task's task mempolicy and non-shared vma policies must be |
1590 | * protected by task_lock(task) by the caller. | |
52cd3b07 LS |
1591 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1592 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1593 | * freeing by another task. It is the caller's responsibility to free the | |
1594 | * extra reference for shared policies. | |
480eccf9 | 1595 | */ |
d98f6cb6 | 1596 | struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1597 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1598 | { |
6e21c8f1 | 1599 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1600 | |
1601 | if (vma) { | |
480eccf9 | 1602 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1603 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1604 | addr); | |
1605 | if (vpol) | |
1606 | pol = vpol; | |
00442ad0 | 1607 | } else if (vma->vm_policy) { |
1da177e4 | 1608 | pol = vma->vm_policy; |
00442ad0 MG |
1609 | |
1610 | /* | |
1611 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1612 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1613 | * count on these policies which will be dropped by | |
1614 | * mpol_cond_put() later | |
1615 | */ | |
1616 | if (mpol_needs_cond_ref(pol)) | |
1617 | mpol_get(pol); | |
1618 | } | |
1da177e4 LT |
1619 | } |
1620 | if (!pol) | |
1621 | pol = &default_policy; | |
1622 | return pol; | |
1623 | } | |
1624 | ||
52cd3b07 LS |
1625 | /* |
1626 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1627 | * page allocation | |
1628 | */ | |
1629 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1630 | { |
1631 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1632 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1633 | gfp_zone(gfp) >= policy_zone && |
1634 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1635 | return &policy->v.nodes; | |
1636 | ||
1637 | return NULL; | |
1638 | } | |
1639 | ||
52cd3b07 | 1640 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1641 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1642 | int nd) | |
1da177e4 | 1643 | { |
45c4745a | 1644 | switch (policy->mode) { |
1da177e4 | 1645 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1646 | if (!(policy->flags & MPOL_F_LOCAL)) |
1647 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1648 | break; |
1649 | case MPOL_BIND: | |
19770b32 | 1650 | /* |
52cd3b07 LS |
1651 | * Normally, MPOL_BIND allocations are node-local within the |
1652 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1653 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1654 | * the first node in the mask instead. |
19770b32 | 1655 | */ |
19770b32 MG |
1656 | if (unlikely(gfp & __GFP_THISNODE) && |
1657 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1658 | nd = first_node(policy->v.nodes); | |
1659 | break; | |
1da177e4 | 1660 | default: |
1da177e4 LT |
1661 | BUG(); |
1662 | } | |
0e88460d | 1663 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1664 | } |
1665 | ||
1666 | /* Do dynamic interleaving for a process */ | |
1667 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1668 | { | |
1669 | unsigned nid, next; | |
1670 | struct task_struct *me = current; | |
1671 | ||
1672 | nid = me->il_next; | |
dfcd3c0d | 1673 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1674 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1675 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1676 | if (next < MAX_NUMNODES) |
1677 | me->il_next = next; | |
1da177e4 LT |
1678 | return nid; |
1679 | } | |
1680 | ||
dc85da15 CL |
1681 | /* |
1682 | * Depending on the memory policy provide a node from which to allocate the | |
1683 | * next slab entry. | |
52cd3b07 LS |
1684 | * @policy must be protected by freeing by the caller. If @policy is |
1685 | * the current task's mempolicy, this protection is implicit, as only the | |
1686 | * task can change it's policy. The system default policy requires no | |
1687 | * such protection. | |
dc85da15 | 1688 | */ |
e7b691b0 | 1689 | unsigned slab_node(void) |
dc85da15 | 1690 | { |
e7b691b0 AK |
1691 | struct mempolicy *policy; |
1692 | ||
1693 | if (in_interrupt()) | |
1694 | return numa_node_id(); | |
1695 | ||
1696 | policy = current->mempolicy; | |
fc36b8d3 | 1697 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1698 | return numa_node_id(); |
1699 | ||
1700 | switch (policy->mode) { | |
1701 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1702 | /* |
1703 | * handled MPOL_F_LOCAL above | |
1704 | */ | |
1705 | return policy->v.preferred_node; | |
765c4507 | 1706 | |
dc85da15 CL |
1707 | case MPOL_INTERLEAVE: |
1708 | return interleave_nodes(policy); | |
1709 | ||
dd1a239f | 1710 | case MPOL_BIND: { |
dc85da15 CL |
1711 | /* |
1712 | * Follow bind policy behavior and start allocation at the | |
1713 | * first node. | |
1714 | */ | |
19770b32 MG |
1715 | struct zonelist *zonelist; |
1716 | struct zone *zone; | |
1717 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1718 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1719 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1720 | &policy->v.nodes, | |
1721 | &zone); | |
800416f7 | 1722 | return zone ? zone->node : numa_node_id(); |
dd1a239f | 1723 | } |
dc85da15 | 1724 | |
dc85da15 | 1725 | default: |
bea904d5 | 1726 | BUG(); |
dc85da15 CL |
1727 | } |
1728 | } | |
1729 | ||
1da177e4 LT |
1730 | /* Do static interleaving for a VMA with known offset. */ |
1731 | static unsigned offset_il_node(struct mempolicy *pol, | |
1732 | struct vm_area_struct *vma, unsigned long off) | |
1733 | { | |
dfcd3c0d | 1734 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1735 | unsigned target; |
1da177e4 LT |
1736 | int c; |
1737 | int nid = -1; | |
1738 | ||
f5b087b5 DR |
1739 | if (!nnodes) |
1740 | return numa_node_id(); | |
1741 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1742 | c = 0; |
1743 | do { | |
dfcd3c0d | 1744 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1745 | c++; |
1746 | } while (c <= target); | |
1da177e4 LT |
1747 | return nid; |
1748 | } | |
1749 | ||
5da7ca86 CL |
1750 | /* Determine a node number for interleave */ |
1751 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1752 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1753 | { | |
1754 | if (vma) { | |
1755 | unsigned long off; | |
1756 | ||
3b98b087 NA |
1757 | /* |
1758 | * for small pages, there is no difference between | |
1759 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1760 | * for huge pages, since vm_pgoff is in units of small | |
1761 | * pages, we need to shift off the always 0 bits to get | |
1762 | * a useful offset. | |
1763 | */ | |
1764 | BUG_ON(shift < PAGE_SHIFT); | |
1765 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1766 | off += (addr - vma->vm_start) >> shift; |
1767 | return offset_il_node(pol, vma, off); | |
1768 | } else | |
1769 | return interleave_nodes(pol); | |
1770 | } | |
1771 | ||
778d3b0f MH |
1772 | /* |
1773 | * Return the bit number of a random bit set in the nodemask. | |
1774 | * (returns -1 if nodemask is empty) | |
1775 | */ | |
1776 | int node_random(const nodemask_t *maskp) | |
1777 | { | |
1778 | int w, bit = -1; | |
1779 | ||
1780 | w = nodes_weight(*maskp); | |
1781 | if (w) | |
1782 | bit = bitmap_ord_to_pos(maskp->bits, | |
1783 | get_random_int() % w, MAX_NUMNODES); | |
1784 | return bit; | |
1785 | } | |
1786 | ||
00ac59ad | 1787 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1788 | /* |
1789 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1790 | * @vma = virtual memory area whose policy is sought | |
1791 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1792 | * @gfp_flags = for requested zone | |
19770b32 MG |
1793 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1794 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1795 | * |
52cd3b07 LS |
1796 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1797 | * to the struct mempolicy for conditional unref after allocation. | |
1798 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1799 | * @nodemask for filtering the zonelist. | |
c0ff7453 MX |
1800 | * |
1801 | * Must be protected by get_mems_allowed() | |
480eccf9 | 1802 | */ |
396faf03 | 1803 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1804 | gfp_t gfp_flags, struct mempolicy **mpol, |
1805 | nodemask_t **nodemask) | |
5da7ca86 | 1806 | { |
480eccf9 | 1807 | struct zonelist *zl; |
5da7ca86 | 1808 | |
52cd3b07 | 1809 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1810 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1811 | |
52cd3b07 LS |
1812 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1813 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1814 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1815 | } else { |
2f5f9486 | 1816 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1817 | if ((*mpol)->mode == MPOL_BIND) |
1818 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1819 | } |
1820 | return zl; | |
5da7ca86 | 1821 | } |
06808b08 LS |
1822 | |
1823 | /* | |
1824 | * init_nodemask_of_mempolicy | |
1825 | * | |
1826 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1827 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1828 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1829 | * initialize the argument nodemask to contain the single node for | |
1830 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1831 | * of non-default mempolicy. | |
1832 | * | |
1833 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1834 | * because the current task is examining it's own mempolicy and a task's | |
1835 | * mempolicy is only ever changed by the task itself. | |
1836 | * | |
1837 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1838 | */ | |
1839 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1840 | { | |
1841 | struct mempolicy *mempolicy; | |
1842 | int nid; | |
1843 | ||
1844 | if (!(mask && current->mempolicy)) | |
1845 | return false; | |
1846 | ||
c0ff7453 | 1847 | task_lock(current); |
06808b08 LS |
1848 | mempolicy = current->mempolicy; |
1849 | switch (mempolicy->mode) { | |
1850 | case MPOL_PREFERRED: | |
1851 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1852 | nid = numa_node_id(); | |
1853 | else | |
1854 | nid = mempolicy->v.preferred_node; | |
1855 | init_nodemask_of_node(mask, nid); | |
1856 | break; | |
1857 | ||
1858 | case MPOL_BIND: | |
1859 | /* Fall through */ | |
1860 | case MPOL_INTERLEAVE: | |
1861 | *mask = mempolicy->v.nodes; | |
1862 | break; | |
1863 | ||
1864 | default: | |
1865 | BUG(); | |
1866 | } | |
c0ff7453 | 1867 | task_unlock(current); |
06808b08 LS |
1868 | |
1869 | return true; | |
1870 | } | |
00ac59ad | 1871 | #endif |
5da7ca86 | 1872 | |
6f48d0eb DR |
1873 | /* |
1874 | * mempolicy_nodemask_intersects | |
1875 | * | |
1876 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1877 | * policy. Otherwise, check for intersection between mask and the policy | |
1878 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1879 | * policy, always return true since it may allocate elsewhere on fallback. | |
1880 | * | |
1881 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1882 | */ | |
1883 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1884 | const nodemask_t *mask) | |
1885 | { | |
1886 | struct mempolicy *mempolicy; | |
1887 | bool ret = true; | |
1888 | ||
1889 | if (!mask) | |
1890 | return ret; | |
1891 | task_lock(tsk); | |
1892 | mempolicy = tsk->mempolicy; | |
1893 | if (!mempolicy) | |
1894 | goto out; | |
1895 | ||
1896 | switch (mempolicy->mode) { | |
1897 | case MPOL_PREFERRED: | |
1898 | /* | |
1899 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1900 | * allocate from, they may fallback to other nodes when oom. | |
1901 | * Thus, it's possible for tsk to have allocated memory from | |
1902 | * nodes in mask. | |
1903 | */ | |
1904 | break; | |
1905 | case MPOL_BIND: | |
1906 | case MPOL_INTERLEAVE: | |
1907 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1908 | break; | |
1909 | default: | |
1910 | BUG(); | |
1911 | } | |
1912 | out: | |
1913 | task_unlock(tsk); | |
1914 | return ret; | |
1915 | } | |
1916 | ||
1da177e4 LT |
1917 | /* Allocate a page in interleaved policy. |
1918 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1919 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1920 | unsigned nid) | |
1da177e4 LT |
1921 | { |
1922 | struct zonelist *zl; | |
1923 | struct page *page; | |
1924 | ||
0e88460d | 1925 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1926 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1927 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1928 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1929 | return page; |
1930 | } | |
1931 | ||
1932 | /** | |
0bbbc0b3 | 1933 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1934 | * |
1935 | * @gfp: | |
1936 | * %GFP_USER user allocation. | |
1937 | * %GFP_KERNEL kernel allocations, | |
1938 | * %GFP_HIGHMEM highmem/user allocations, | |
1939 | * %GFP_FS allocation should not call back into a file system. | |
1940 | * %GFP_ATOMIC don't sleep. | |
1941 | * | |
0bbbc0b3 | 1942 | * @order:Order of the GFP allocation. |
1da177e4 LT |
1943 | * @vma: Pointer to VMA or NULL if not available. |
1944 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1945 | * | |
1946 | * This function allocates a page from the kernel page pool and applies | |
1947 | * a NUMA policy associated with the VMA or the current process. | |
1948 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1949 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1950 | * all allocations for pages that will be mapped into | |
1951 | * user space. Returns NULL when no page can be allocated. | |
1952 | * | |
1953 | * Should be called with the mm_sem of the vma hold. | |
1954 | */ | |
1955 | struct page * | |
0bbbc0b3 | 1956 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 1957 | unsigned long addr, int node) |
1da177e4 | 1958 | { |
cc9a6c87 | 1959 | struct mempolicy *pol; |
480eccf9 | 1960 | struct zonelist *zl; |
c0ff7453 | 1961 | struct page *page; |
cc9a6c87 MG |
1962 | unsigned int cpuset_mems_cookie; |
1963 | ||
1964 | retry_cpuset: | |
1965 | pol = get_vma_policy(current, vma, addr); | |
1966 | cpuset_mems_cookie = get_mems_allowed(); | |
1da177e4 | 1967 | |
45c4745a | 1968 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1969 | unsigned nid; |
5da7ca86 | 1970 | |
8eac563c | 1971 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 1972 | mpol_cond_put(pol); |
0bbbc0b3 | 1973 | page = alloc_page_interleave(gfp, order, nid); |
cc9a6c87 MG |
1974 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1975 | goto retry_cpuset; | |
1976 | ||
c0ff7453 | 1977 | return page; |
1da177e4 | 1978 | } |
2f5f9486 | 1979 | zl = policy_zonelist(gfp, pol, node); |
52cd3b07 | 1980 | if (unlikely(mpol_needs_cond_ref(pol))) { |
480eccf9 | 1981 | /* |
52cd3b07 | 1982 | * slow path: ref counted shared policy |
480eccf9 | 1983 | */ |
0bbbc0b3 | 1984 | struct page *page = __alloc_pages_nodemask(gfp, order, |
52cd3b07 | 1985 | zl, policy_nodemask(gfp, pol)); |
f0be3d32 | 1986 | __mpol_put(pol); |
cc9a6c87 MG |
1987 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1988 | goto retry_cpuset; | |
480eccf9 LS |
1989 | return page; |
1990 | } | |
1991 | /* | |
1992 | * fast path: default or task policy | |
1993 | */ | |
0bbbc0b3 AA |
1994 | page = __alloc_pages_nodemask(gfp, order, zl, |
1995 | policy_nodemask(gfp, pol)); | |
cc9a6c87 MG |
1996 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1997 | goto retry_cpuset; | |
c0ff7453 | 1998 | return page; |
1da177e4 LT |
1999 | } |
2000 | ||
2001 | /** | |
2002 | * alloc_pages_current - Allocate pages. | |
2003 | * | |
2004 | * @gfp: | |
2005 | * %GFP_USER user allocation, | |
2006 | * %GFP_KERNEL kernel allocation, | |
2007 | * %GFP_HIGHMEM highmem allocation, | |
2008 | * %GFP_FS don't call back into a file system. | |
2009 | * %GFP_ATOMIC don't sleep. | |
2010 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2011 | * | |
2012 | * Allocate a page from the kernel page pool. When not in | |
2013 | * interrupt context and apply the current process NUMA policy. | |
2014 | * Returns NULL when no page can be allocated. | |
2015 | * | |
cf2a473c | 2016 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
2017 | * 1) it's ok to take cpuset_sem (can WAIT), and |
2018 | * 2) allocating for current task (not interrupt). | |
2019 | */ | |
dd0fc66f | 2020 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
2021 | { |
2022 | struct mempolicy *pol = current->mempolicy; | |
c0ff7453 | 2023 | struct page *page; |
cc9a6c87 | 2024 | unsigned int cpuset_mems_cookie; |
1da177e4 | 2025 | |
9b819d20 | 2026 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 2027 | pol = &default_policy; |
52cd3b07 | 2028 | |
cc9a6c87 MG |
2029 | retry_cpuset: |
2030 | cpuset_mems_cookie = get_mems_allowed(); | |
2031 | ||
52cd3b07 LS |
2032 | /* |
2033 | * No reference counting needed for current->mempolicy | |
2034 | * nor system default_policy | |
2035 | */ | |
45c4745a | 2036 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2037 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2038 | else | |
2039 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
2040 | policy_zonelist(gfp, pol, numa_node_id()), |
2041 | policy_nodemask(gfp, pol)); | |
cc9a6c87 MG |
2042 | |
2043 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2044 | goto retry_cpuset; | |
2045 | ||
c0ff7453 | 2046 | return page; |
1da177e4 LT |
2047 | } |
2048 | EXPORT_SYMBOL(alloc_pages_current); | |
2049 | ||
4225399a | 2050 | /* |
846a16bf | 2051 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2052 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2053 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2054 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2055 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2056 | * |
2057 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2058 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2059 | */ |
4225399a | 2060 | |
846a16bf LS |
2061 | /* Slow path of a mempolicy duplicate */ |
2062 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2063 | { |
2064 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2065 | ||
2066 | if (!new) | |
2067 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2068 | |
2069 | /* task's mempolicy is protected by alloc_lock */ | |
2070 | if (old == current->mempolicy) { | |
2071 | task_lock(current); | |
2072 | *new = *old; | |
2073 | task_unlock(current); | |
2074 | } else | |
2075 | *new = *old; | |
2076 | ||
99ee4ca7 | 2077 | rcu_read_lock(); |
4225399a PJ |
2078 | if (current_cpuset_is_being_rebound()) { |
2079 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2080 | if (new->flags & MPOL_F_REBINDING) |
2081 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2082 | else | |
2083 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2084 | } |
99ee4ca7 | 2085 | rcu_read_unlock(); |
1da177e4 | 2086 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2087 | return new; |
2088 | } | |
2089 | ||
52cd3b07 LS |
2090 | /* |
2091 | * If *frompol needs [has] an extra ref, copy *frompol to *tompol , | |
2092 | * eliminate the * MPOL_F_* flags that require conditional ref and | |
2093 | * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly | |
2094 | * after return. Use the returned value. | |
2095 | * | |
2096 | * Allows use of a mempolicy for, e.g., multiple allocations with a single | |
2097 | * policy lookup, even if the policy needs/has extra ref on lookup. | |
2098 | * shmem_readahead needs this. | |
2099 | */ | |
2100 | struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
2101 | struct mempolicy *frompol) | |
2102 | { | |
2103 | if (!mpol_needs_cond_ref(frompol)) | |
2104 | return frompol; | |
2105 | ||
2106 | *tompol = *frompol; | |
2107 | tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */ | |
2108 | __mpol_put(frompol); | |
2109 | return tompol; | |
2110 | } | |
2111 | ||
1da177e4 | 2112 | /* Slow path of a mempolicy comparison */ |
fcfb4dcc | 2113 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2114 | { |
2115 | if (!a || !b) | |
fcfb4dcc | 2116 | return false; |
45c4745a | 2117 | if (a->mode != b->mode) |
fcfb4dcc | 2118 | return false; |
19800502 | 2119 | if (a->flags != b->flags) |
fcfb4dcc | 2120 | return false; |
19800502 BL |
2121 | if (mpol_store_user_nodemask(a)) |
2122 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2123 | return false; |
19800502 | 2124 | |
45c4745a | 2125 | switch (a->mode) { |
19770b32 MG |
2126 | case MPOL_BIND: |
2127 | /* Fall through */ | |
1da177e4 | 2128 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2129 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2130 | case MPOL_PREFERRED: |
75719661 | 2131 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2132 | default: |
2133 | BUG(); | |
fcfb4dcc | 2134 | return false; |
1da177e4 LT |
2135 | } |
2136 | } | |
2137 | ||
1da177e4 LT |
2138 | /* |
2139 | * Shared memory backing store policy support. | |
2140 | * | |
2141 | * Remember policies even when nobody has shared memory mapped. | |
2142 | * The policies are kept in Red-Black tree linked from the inode. | |
2143 | * They are protected by the sp->lock spinlock, which should be held | |
2144 | * for any accesses to the tree. | |
2145 | */ | |
2146 | ||
2147 | /* lookup first element intersecting start-end */ | |
b22d127a | 2148 | /* Caller holds sp->mutex */ |
1da177e4 LT |
2149 | static struct sp_node * |
2150 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2151 | { | |
2152 | struct rb_node *n = sp->root.rb_node; | |
2153 | ||
2154 | while (n) { | |
2155 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2156 | ||
2157 | if (start >= p->end) | |
2158 | n = n->rb_right; | |
2159 | else if (end <= p->start) | |
2160 | n = n->rb_left; | |
2161 | else | |
2162 | break; | |
2163 | } | |
2164 | if (!n) | |
2165 | return NULL; | |
2166 | for (;;) { | |
2167 | struct sp_node *w = NULL; | |
2168 | struct rb_node *prev = rb_prev(n); | |
2169 | if (!prev) | |
2170 | break; | |
2171 | w = rb_entry(prev, struct sp_node, nd); | |
2172 | if (w->end <= start) | |
2173 | break; | |
2174 | n = prev; | |
2175 | } | |
2176 | return rb_entry(n, struct sp_node, nd); | |
2177 | } | |
2178 | ||
2179 | /* Insert a new shared policy into the list. */ | |
2180 | /* Caller holds sp->lock */ | |
2181 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2182 | { | |
2183 | struct rb_node **p = &sp->root.rb_node; | |
2184 | struct rb_node *parent = NULL; | |
2185 | struct sp_node *nd; | |
2186 | ||
2187 | while (*p) { | |
2188 | parent = *p; | |
2189 | nd = rb_entry(parent, struct sp_node, nd); | |
2190 | if (new->start < nd->start) | |
2191 | p = &(*p)->rb_left; | |
2192 | else if (new->end > nd->end) | |
2193 | p = &(*p)->rb_right; | |
2194 | else | |
2195 | BUG(); | |
2196 | } | |
2197 | rb_link_node(&new->nd, parent, p); | |
2198 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2199 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2200 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2201 | } |
2202 | ||
2203 | /* Find shared policy intersecting idx */ | |
2204 | struct mempolicy * | |
2205 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2206 | { | |
2207 | struct mempolicy *pol = NULL; | |
2208 | struct sp_node *sn; | |
2209 | ||
2210 | if (!sp->root.rb_node) | |
2211 | return NULL; | |
b22d127a | 2212 | mutex_lock(&sp->mutex); |
1da177e4 LT |
2213 | sn = sp_lookup(sp, idx, idx+1); |
2214 | if (sn) { | |
2215 | mpol_get(sn->policy); | |
2216 | pol = sn->policy; | |
2217 | } | |
b22d127a | 2218 | mutex_unlock(&sp->mutex); |
1da177e4 LT |
2219 | return pol; |
2220 | } | |
2221 | ||
63f74ca2 KM |
2222 | static void sp_free(struct sp_node *n) |
2223 | { | |
2224 | mpol_put(n->policy); | |
2225 | kmem_cache_free(sn_cache, n); | |
2226 | } | |
2227 | ||
771fb4d8 LS |
2228 | /** |
2229 | * mpol_misplaced - check whether current page node is valid in policy | |
2230 | * | |
2231 | * @page - page to be checked | |
2232 | * @vma - vm area where page mapped | |
2233 | * @addr - virtual address where page mapped | |
2234 | * | |
2235 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2236 | * node id. | |
2237 | * | |
2238 | * Returns: | |
2239 | * -1 - not misplaced, page is in the right node | |
2240 | * node - node id where the page should be | |
2241 | * | |
2242 | * Policy determination "mimics" alloc_page_vma(). | |
2243 | * Called from fault path where we know the vma and faulting address. | |
2244 | */ | |
2245 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2246 | { | |
2247 | struct mempolicy *pol; | |
2248 | struct zone *zone; | |
2249 | int curnid = page_to_nid(page); | |
2250 | unsigned long pgoff; | |
2251 | int polnid = -1; | |
2252 | int ret = -1; | |
2253 | ||
2254 | BUG_ON(!vma); | |
2255 | ||
2256 | pol = get_vma_policy(current, vma, addr); | |
2257 | if (!(pol->flags & MPOL_F_MOF)) | |
2258 | goto out; | |
2259 | ||
2260 | switch (pol->mode) { | |
2261 | case MPOL_INTERLEAVE: | |
2262 | BUG_ON(addr >= vma->vm_end); | |
2263 | BUG_ON(addr < vma->vm_start); | |
2264 | ||
2265 | pgoff = vma->vm_pgoff; | |
2266 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2267 | polnid = offset_il_node(pol, vma, pgoff); | |
2268 | break; | |
2269 | ||
2270 | case MPOL_PREFERRED: | |
2271 | if (pol->flags & MPOL_F_LOCAL) | |
2272 | polnid = numa_node_id(); | |
2273 | else | |
2274 | polnid = pol->v.preferred_node; | |
2275 | break; | |
2276 | ||
2277 | case MPOL_BIND: | |
2278 | /* | |
2279 | * allows binding to multiple nodes. | |
2280 | * use current page if in policy nodemask, | |
2281 | * else select nearest allowed node, if any. | |
2282 | * If no allowed nodes, use current [!misplaced]. | |
2283 | */ | |
2284 | if (node_isset(curnid, pol->v.nodes)) | |
2285 | goto out; | |
2286 | (void)first_zones_zonelist( | |
2287 | node_zonelist(numa_node_id(), GFP_HIGHUSER), | |
2288 | gfp_zone(GFP_HIGHUSER), | |
2289 | &pol->v.nodes, &zone); | |
2290 | polnid = zone->node; | |
2291 | break; | |
2292 | ||
2293 | default: | |
2294 | BUG(); | |
2295 | } | |
2296 | if (curnid != polnid) | |
2297 | ret = polnid; | |
2298 | out: | |
2299 | mpol_cond_put(pol); | |
2300 | ||
2301 | return ret; | |
2302 | } | |
2303 | ||
1da177e4 LT |
2304 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2305 | { | |
140d5a49 | 2306 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2307 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2308 | sp_free(n); |
1da177e4 LT |
2309 | } |
2310 | ||
dbcb0f19 AB |
2311 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2312 | struct mempolicy *pol) | |
1da177e4 | 2313 | { |
869833f2 KM |
2314 | struct sp_node *n; |
2315 | struct mempolicy *newpol; | |
1da177e4 | 2316 | |
869833f2 | 2317 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2318 | if (!n) |
2319 | return NULL; | |
869833f2 KM |
2320 | |
2321 | newpol = mpol_dup(pol); | |
2322 | if (IS_ERR(newpol)) { | |
2323 | kmem_cache_free(sn_cache, n); | |
2324 | return NULL; | |
2325 | } | |
2326 | newpol->flags |= MPOL_F_SHARED; | |
2327 | ||
1da177e4 LT |
2328 | n->start = start; |
2329 | n->end = end; | |
869833f2 KM |
2330 | n->policy = newpol; |
2331 | ||
1da177e4 LT |
2332 | return n; |
2333 | } | |
2334 | ||
2335 | /* Replace a policy range. */ | |
2336 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2337 | unsigned long end, struct sp_node *new) | |
2338 | { | |
b22d127a MG |
2339 | struct sp_node *n; |
2340 | int ret = 0; | |
1da177e4 | 2341 | |
b22d127a | 2342 | mutex_lock(&sp->mutex); |
1da177e4 LT |
2343 | n = sp_lookup(sp, start, end); |
2344 | /* Take care of old policies in the same range. */ | |
2345 | while (n && n->start < end) { | |
2346 | struct rb_node *next = rb_next(&n->nd); | |
2347 | if (n->start >= start) { | |
2348 | if (n->end <= end) | |
2349 | sp_delete(sp, n); | |
2350 | else | |
2351 | n->start = end; | |
2352 | } else { | |
2353 | /* Old policy spanning whole new range. */ | |
2354 | if (n->end > end) { | |
b22d127a MG |
2355 | struct sp_node *new2; |
2356 | new2 = sp_alloc(end, n->end, n->policy); | |
1da177e4 | 2357 | if (!new2) { |
b22d127a MG |
2358 | ret = -ENOMEM; |
2359 | goto out; | |
1da177e4 LT |
2360 | } |
2361 | n->end = start; | |
2362 | sp_insert(sp, new2); | |
1da177e4 LT |
2363 | break; |
2364 | } else | |
2365 | n->end = start; | |
2366 | } | |
2367 | if (!next) | |
2368 | break; | |
2369 | n = rb_entry(next, struct sp_node, nd); | |
2370 | } | |
2371 | if (new) | |
2372 | sp_insert(sp, new); | |
b22d127a MG |
2373 | out: |
2374 | mutex_unlock(&sp->mutex); | |
2375 | return ret; | |
1da177e4 LT |
2376 | } |
2377 | ||
71fe804b LS |
2378 | /** |
2379 | * mpol_shared_policy_init - initialize shared policy for inode | |
2380 | * @sp: pointer to inode shared policy | |
2381 | * @mpol: struct mempolicy to install | |
2382 | * | |
2383 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2384 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2385 | * This must be released on exit. | |
4bfc4495 | 2386 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2387 | */ |
2388 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2389 | { | |
58568d2a MX |
2390 | int ret; |
2391 | ||
71fe804b | 2392 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
b22d127a | 2393 | mutex_init(&sp->mutex); |
71fe804b LS |
2394 | |
2395 | if (mpol) { | |
2396 | struct vm_area_struct pvma; | |
2397 | struct mempolicy *new; | |
4bfc4495 | 2398 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2399 | |
4bfc4495 | 2400 | if (!scratch) |
5c0c1654 | 2401 | goto put_mpol; |
71fe804b LS |
2402 | /* contextualize the tmpfs mount point mempolicy */ |
2403 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2404 | if (IS_ERR(new)) |
0cae3457 | 2405 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2406 | |
2407 | task_lock(current); | |
4bfc4495 | 2408 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2409 | task_unlock(current); |
15d77835 | 2410 | if (ret) |
5c0c1654 | 2411 | goto put_new; |
71fe804b LS |
2412 | |
2413 | /* Create pseudo-vma that contains just the policy */ | |
2414 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2415 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2416 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2417 | |
5c0c1654 | 2418 | put_new: |
71fe804b | 2419 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2420 | free_scratch: |
4bfc4495 | 2421 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2422 | put_mpol: |
2423 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2424 | } |
2425 | } | |
2426 | ||
1da177e4 LT |
2427 | int mpol_set_shared_policy(struct shared_policy *info, |
2428 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2429 | { | |
2430 | int err; | |
2431 | struct sp_node *new = NULL; | |
2432 | unsigned long sz = vma_pages(vma); | |
2433 | ||
028fec41 | 2434 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2435 | vma->vm_pgoff, |
45c4745a | 2436 | sz, npol ? npol->mode : -1, |
028fec41 | 2437 | npol ? npol->flags : -1, |
140d5a49 | 2438 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
2439 | |
2440 | if (npol) { | |
2441 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2442 | if (!new) | |
2443 | return -ENOMEM; | |
2444 | } | |
2445 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2446 | if (err && new) | |
63f74ca2 | 2447 | sp_free(new); |
1da177e4 LT |
2448 | return err; |
2449 | } | |
2450 | ||
2451 | /* Free a backing policy store on inode delete. */ | |
2452 | void mpol_free_shared_policy(struct shared_policy *p) | |
2453 | { | |
2454 | struct sp_node *n; | |
2455 | struct rb_node *next; | |
2456 | ||
2457 | if (!p->root.rb_node) | |
2458 | return; | |
b22d127a | 2459 | mutex_lock(&p->mutex); |
1da177e4 LT |
2460 | next = rb_first(&p->root); |
2461 | while (next) { | |
2462 | n = rb_entry(next, struct sp_node, nd); | |
2463 | next = rb_next(&n->nd); | |
63f74ca2 | 2464 | sp_delete(p, n); |
1da177e4 | 2465 | } |
b22d127a | 2466 | mutex_unlock(&p->mutex); |
1da177e4 LT |
2467 | } |
2468 | ||
2469 | /* assumes fs == KERNEL_DS */ | |
2470 | void __init numa_policy_init(void) | |
2471 | { | |
b71636e2 PM |
2472 | nodemask_t interleave_nodes; |
2473 | unsigned long largest = 0; | |
2474 | int nid, prefer = 0; | |
2475 | ||
1da177e4 LT |
2476 | policy_cache = kmem_cache_create("numa_policy", |
2477 | sizeof(struct mempolicy), | |
20c2df83 | 2478 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2479 | |
2480 | sn_cache = kmem_cache_create("shared_policy_node", | |
2481 | sizeof(struct sp_node), | |
20c2df83 | 2482 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2483 | |
b71636e2 PM |
2484 | /* |
2485 | * Set interleaving policy for system init. Interleaving is only | |
2486 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2487 | * fall back to the largest node if they're all smaller. | |
2488 | */ | |
2489 | nodes_clear(interleave_nodes); | |
56bbd65d | 2490 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
2491 | unsigned long total_pages = node_present_pages(nid); |
2492 | ||
2493 | /* Preserve the largest node */ | |
2494 | if (largest < total_pages) { | |
2495 | largest = total_pages; | |
2496 | prefer = nid; | |
2497 | } | |
2498 | ||
2499 | /* Interleave this node? */ | |
2500 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2501 | node_set(nid, interleave_nodes); | |
2502 | } | |
2503 | ||
2504 | /* All too small, use the largest */ | |
2505 | if (unlikely(nodes_empty(interleave_nodes))) | |
2506 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2507 | |
028fec41 | 2508 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
2509 | printk("numa_policy_init: interleaving failed\n"); |
2510 | } | |
2511 | ||
8bccd85f | 2512 | /* Reset policy of current process to default */ |
1da177e4 LT |
2513 | void numa_default_policy(void) |
2514 | { | |
028fec41 | 2515 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2516 | } |
68860ec1 | 2517 | |
095f1fc4 LS |
2518 | /* |
2519 | * Parse and format mempolicy from/to strings | |
2520 | */ | |
2521 | ||
1a75a6c8 | 2522 | /* |
fc36b8d3 | 2523 | * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag |
3f226aa1 | 2524 | * Used only for mpol_parse_str() and mpol_to_str() |
1a75a6c8 | 2525 | */ |
345ace9c LS |
2526 | static const char * const policy_modes[] = |
2527 | { | |
2528 | [MPOL_DEFAULT] = "default", | |
2529 | [MPOL_PREFERRED] = "prefer", | |
2530 | [MPOL_BIND] = "bind", | |
2531 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2532 | [MPOL_LOCAL] = "local", |
345ace9c | 2533 | }; |
1a75a6c8 | 2534 | |
095f1fc4 LS |
2535 | |
2536 | #ifdef CONFIG_TMPFS | |
2537 | /** | |
2538 | * mpol_parse_str - parse string to mempolicy | |
2539 | * @str: string containing mempolicy to parse | |
71fe804b LS |
2540 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
2541 | * @no_context: flag whether to "contextualize" the mempolicy | |
095f1fc4 LS |
2542 | * |
2543 | * Format of input: | |
2544 | * <mode>[=<flags>][:<nodelist>] | |
2545 | * | |
71fe804b LS |
2546 | * if @no_context is true, save the input nodemask in w.user_nodemask in |
2547 | * the returned mempolicy. This will be used to "clone" the mempolicy in | |
2548 | * a specific context [cpuset] at a later time. Used to parse tmpfs mpol | |
2549 | * mount option. Note that if 'static' or 'relative' mode flags were | |
2550 | * specified, the input nodemask will already have been saved. Saving | |
2551 | * it again is redundant, but safe. | |
2552 | * | |
2553 | * On success, returns 0, else 1 | |
095f1fc4 | 2554 | */ |
71fe804b | 2555 | int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context) |
095f1fc4 | 2556 | { |
71fe804b | 2557 | struct mempolicy *new = NULL; |
b4652e84 | 2558 | unsigned short mode; |
71fe804b LS |
2559 | unsigned short uninitialized_var(mode_flags); |
2560 | nodemask_t nodes; | |
095f1fc4 LS |
2561 | char *nodelist = strchr(str, ':'); |
2562 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2563 | int err = 1; |
2564 | ||
2565 | if (nodelist) { | |
2566 | /* NUL-terminate mode or flags string */ | |
2567 | *nodelist++ = '\0'; | |
71fe804b | 2568 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2569 | goto out; |
71fe804b | 2570 | if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY])) |
095f1fc4 | 2571 | goto out; |
71fe804b LS |
2572 | } else |
2573 | nodes_clear(nodes); | |
2574 | ||
095f1fc4 LS |
2575 | if (flags) |
2576 | *flags++ = '\0'; /* terminate mode string */ | |
2577 | ||
479e2802 | 2578 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2579 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2580 | break; |
2581 | } | |
2582 | } | |
a720094d | 2583 | if (mode >= MPOL_MAX) |
095f1fc4 LS |
2584 | goto out; |
2585 | ||
71fe804b | 2586 | switch (mode) { |
095f1fc4 | 2587 | case MPOL_PREFERRED: |
71fe804b LS |
2588 | /* |
2589 | * Insist on a nodelist of one node only | |
2590 | */ | |
095f1fc4 LS |
2591 | if (nodelist) { |
2592 | char *rest = nodelist; | |
2593 | while (isdigit(*rest)) | |
2594 | rest++; | |
926f2ae0 KM |
2595 | if (*rest) |
2596 | goto out; | |
095f1fc4 LS |
2597 | } |
2598 | break; | |
095f1fc4 LS |
2599 | case MPOL_INTERLEAVE: |
2600 | /* | |
2601 | * Default to online nodes with memory if no nodelist | |
2602 | */ | |
2603 | if (!nodelist) | |
71fe804b | 2604 | nodes = node_states[N_HIGH_MEMORY]; |
3f226aa1 | 2605 | break; |
71fe804b | 2606 | case MPOL_LOCAL: |
3f226aa1 | 2607 | /* |
71fe804b | 2608 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2609 | */ |
71fe804b | 2610 | if (nodelist) |
3f226aa1 | 2611 | goto out; |
71fe804b | 2612 | mode = MPOL_PREFERRED; |
3f226aa1 | 2613 | break; |
413b43de RT |
2614 | case MPOL_DEFAULT: |
2615 | /* | |
2616 | * Insist on a empty nodelist | |
2617 | */ | |
2618 | if (!nodelist) | |
2619 | err = 0; | |
2620 | goto out; | |
d69b2e63 KM |
2621 | case MPOL_BIND: |
2622 | /* | |
2623 | * Insist on a nodelist | |
2624 | */ | |
2625 | if (!nodelist) | |
2626 | goto out; | |
095f1fc4 LS |
2627 | } |
2628 | ||
71fe804b | 2629 | mode_flags = 0; |
095f1fc4 LS |
2630 | if (flags) { |
2631 | /* | |
2632 | * Currently, we only support two mutually exclusive | |
2633 | * mode flags. | |
2634 | */ | |
2635 | if (!strcmp(flags, "static")) | |
71fe804b | 2636 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2637 | else if (!strcmp(flags, "relative")) |
71fe804b | 2638 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2639 | else |
926f2ae0 | 2640 | goto out; |
095f1fc4 | 2641 | } |
71fe804b LS |
2642 | |
2643 | new = mpol_new(mode, mode_flags, &nodes); | |
2644 | if (IS_ERR(new)) | |
926f2ae0 KM |
2645 | goto out; |
2646 | ||
e17f74af LS |
2647 | if (no_context) { |
2648 | /* save for contextualization */ | |
2649 | new->w.user_nodemask = nodes; | |
2650 | } else { | |
58568d2a | 2651 | int ret; |
4bfc4495 KH |
2652 | NODEMASK_SCRATCH(scratch); |
2653 | if (scratch) { | |
2654 | task_lock(current); | |
2655 | ret = mpol_set_nodemask(new, &nodes, scratch); | |
2656 | task_unlock(current); | |
2657 | } else | |
2658 | ret = -ENOMEM; | |
2659 | NODEMASK_SCRATCH_FREE(scratch); | |
2660 | if (ret) { | |
4bfc4495 | 2661 | mpol_put(new); |
926f2ae0 | 2662 | goto out; |
58568d2a MX |
2663 | } |
2664 | } | |
926f2ae0 | 2665 | err = 0; |
71fe804b | 2666 | |
095f1fc4 LS |
2667 | out: |
2668 | /* Restore string for error message */ | |
2669 | if (nodelist) | |
2670 | *--nodelist = ':'; | |
2671 | if (flags) | |
2672 | *--flags = '='; | |
71fe804b LS |
2673 | if (!err) |
2674 | *mpol = new; | |
095f1fc4 LS |
2675 | return err; |
2676 | } | |
2677 | #endif /* CONFIG_TMPFS */ | |
2678 | ||
71fe804b LS |
2679 | /** |
2680 | * mpol_to_str - format a mempolicy structure for printing | |
2681 | * @buffer: to contain formatted mempolicy string | |
2682 | * @maxlen: length of @buffer | |
2683 | * @pol: pointer to mempolicy to be formatted | |
2684 | * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask | |
2685 | * | |
1a75a6c8 CL |
2686 | * Convert a mempolicy into a string. |
2687 | * Returns the number of characters in buffer (if positive) | |
2688 | * or an error (negative) | |
2689 | */ | |
71fe804b | 2690 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context) |
1a75a6c8 CL |
2691 | { |
2692 | char *p = buffer; | |
2693 | int l; | |
2694 | nodemask_t nodes; | |
bea904d5 | 2695 | unsigned short mode; |
f5b087b5 | 2696 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 2697 | |
2291990a LS |
2698 | /* |
2699 | * Sanity check: room for longest mode, flag and some nodes | |
2700 | */ | |
2701 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); | |
2702 | ||
bea904d5 LS |
2703 | if (!pol || pol == &default_policy) |
2704 | mode = MPOL_DEFAULT; | |
2705 | else | |
2706 | mode = pol->mode; | |
2707 | ||
1a75a6c8 CL |
2708 | switch (mode) { |
2709 | case MPOL_DEFAULT: | |
2710 | nodes_clear(nodes); | |
2711 | break; | |
2712 | ||
2713 | case MPOL_PREFERRED: | |
2714 | nodes_clear(nodes); | |
fc36b8d3 | 2715 | if (flags & MPOL_F_LOCAL) |
53f2556b LS |
2716 | mode = MPOL_LOCAL; /* pseudo-policy */ |
2717 | else | |
fc36b8d3 | 2718 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
2719 | break; |
2720 | ||
2721 | case MPOL_BIND: | |
19770b32 | 2722 | /* Fall through */ |
1a75a6c8 | 2723 | case MPOL_INTERLEAVE: |
71fe804b LS |
2724 | if (no_context) |
2725 | nodes = pol->w.user_nodemask; | |
2726 | else | |
2727 | nodes = pol->v.nodes; | |
1a75a6c8 CL |
2728 | break; |
2729 | ||
2730 | default: | |
80de7c31 | 2731 | return -EINVAL; |
1a75a6c8 CL |
2732 | } |
2733 | ||
345ace9c | 2734 | l = strlen(policy_modes[mode]); |
53f2556b LS |
2735 | if (buffer + maxlen < p + l + 1) |
2736 | return -ENOSPC; | |
1a75a6c8 | 2737 | |
345ace9c | 2738 | strcpy(p, policy_modes[mode]); |
1a75a6c8 CL |
2739 | p += l; |
2740 | ||
fc36b8d3 | 2741 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2742 | if (buffer + maxlen < p + 2) |
2743 | return -ENOSPC; | |
2744 | *p++ = '='; | |
2745 | ||
2291990a LS |
2746 | /* |
2747 | * Currently, the only defined flags are mutually exclusive | |
2748 | */ | |
f5b087b5 | 2749 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2750 | p += snprintf(p, buffer + maxlen - p, "static"); |
2751 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2752 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2753 | } |
2754 | ||
1a75a6c8 CL |
2755 | if (!nodes_empty(nodes)) { |
2756 | if (buffer + maxlen < p + 2) | |
2757 | return -ENOSPC; | |
095f1fc4 | 2758 | *p++ = ':'; |
1a75a6c8 CL |
2759 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2760 | } | |
2761 | return p - buffer; | |
2762 | } |