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> | |
76 | #include <linux/gfp.h> | |
77 | #include <linux/slab.h> | |
78 | #include <linux/string.h> | |
79 | #include <linux/module.h> | |
b488893a | 80 | #include <linux/nsproxy.h> |
1da177e4 LT |
81 | #include <linux/interrupt.h> |
82 | #include <linux/init.h> | |
83 | #include <linux/compat.h> | |
dc9aa5b9 | 84 | #include <linux/swap.h> |
1a75a6c8 CL |
85 | #include <linux/seq_file.h> |
86 | #include <linux/proc_fs.h> | |
b20a3503 | 87 | #include <linux/migrate.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> |
dc9aa5b9 | 92 | |
1da177e4 LT |
93 | #include <asm/tlbflush.h> |
94 | #include <asm/uaccess.h> | |
95 | ||
62695a84 NP |
96 | #include "internal.h" |
97 | ||
38e35860 | 98 | /* Internal flags */ |
dc9aa5b9 | 99 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 100 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
1a75a6c8 | 101 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
dc9aa5b9 | 102 | |
fcc234f8 PE |
103 | static struct kmem_cache *policy_cache; |
104 | static struct kmem_cache *sn_cache; | |
1da177e4 | 105 | |
1da177e4 LT |
106 | /* Highest zone. An specific allocation for a zone below that is not |
107 | policied. */ | |
6267276f | 108 | enum zone_type policy_zone = 0; |
1da177e4 | 109 | |
bea904d5 LS |
110 | /* |
111 | * run-time system-wide default policy => local allocation | |
112 | */ | |
d42c6997 | 113 | struct mempolicy default_policy = { |
1da177e4 | 114 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 115 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 116 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
117 | }; |
118 | ||
37012946 DR |
119 | static const struct mempolicy_operations { |
120 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
121 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); | |
122 | } mpol_ops[MPOL_MAX]; | |
123 | ||
19770b32 | 124 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 125 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 126 | { |
19770b32 | 127 | int nd, k; |
1da177e4 | 128 | |
19770b32 MG |
129 | /* Check that there is something useful in this mask */ |
130 | k = policy_zone; | |
131 | ||
132 | for_each_node_mask(nd, *nodemask) { | |
133 | struct zone *z; | |
134 | ||
135 | for (k = 0; k <= policy_zone; k++) { | |
136 | z = &NODE_DATA(nd)->node_zones[k]; | |
137 | if (z->present_pages > 0) | |
138 | return 1; | |
dd942ae3 | 139 | } |
8af5e2eb | 140 | } |
19770b32 MG |
141 | |
142 | return 0; | |
1da177e4 LT |
143 | } |
144 | ||
f5b087b5 DR |
145 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
146 | { | |
4c50bc01 DR |
147 | return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES); |
148 | } | |
149 | ||
150 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
151 | const nodemask_t *rel) | |
152 | { | |
153 | nodemask_t tmp; | |
154 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
155 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
156 | } |
157 | ||
37012946 DR |
158 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
159 | { | |
160 | if (nodes_empty(*nodes)) | |
161 | return -EINVAL; | |
162 | pol->v.nodes = *nodes; | |
163 | return 0; | |
164 | } | |
165 | ||
166 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
167 | { | |
168 | if (!nodes) | |
fc36b8d3 | 169 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
170 | else if (nodes_empty(*nodes)) |
171 | return -EINVAL; /* no allowed nodes */ | |
172 | else | |
173 | pol->v.preferred_node = first_node(*nodes); | |
174 | return 0; | |
175 | } | |
176 | ||
177 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
178 | { | |
179 | if (!is_valid_nodemask(nodes)) | |
180 | return -EINVAL; | |
181 | pol->v.nodes = *nodes; | |
182 | return 0; | |
183 | } | |
184 | ||
58568d2a MX |
185 | /* |
186 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
187 | * any, for the new policy. mpol_new() has already validated the nodes | |
188 | * parameter with respect to the policy mode and flags. But, we need to | |
189 | * handle an empty nodemask with MPOL_PREFERRED here. | |
190 | * | |
191 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
192 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
193 | */ | |
4bfc4495 KH |
194 | static int mpol_set_nodemask(struct mempolicy *pol, |
195 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 196 | { |
58568d2a MX |
197 | int ret; |
198 | ||
199 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
200 | if (pol == NULL) | |
201 | return 0; | |
4bfc4495 KH |
202 | /* Check N_HIGH_MEMORY */ |
203 | nodes_and(nsc->mask1, | |
204 | cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]); | |
58568d2a MX |
205 | |
206 | VM_BUG_ON(!nodes); | |
207 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
208 | nodes = NULL; /* explicit local allocation */ | |
209 | else { | |
210 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
4bfc4495 | 211 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
58568d2a | 212 | else |
4bfc4495 KH |
213 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
214 | ||
58568d2a MX |
215 | if (mpol_store_user_nodemask(pol)) |
216 | pol->w.user_nodemask = *nodes; | |
217 | else | |
218 | pol->w.cpuset_mems_allowed = | |
219 | cpuset_current_mems_allowed; | |
220 | } | |
221 | ||
4bfc4495 KH |
222 | if (nodes) |
223 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
224 | else | |
225 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
226 | return ret; |
227 | } | |
228 | ||
229 | /* | |
230 | * This function just creates a new policy, does some check and simple | |
231 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
232 | */ | |
028fec41 DR |
233 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
234 | nodemask_t *nodes) | |
1da177e4 LT |
235 | { |
236 | struct mempolicy *policy; | |
237 | ||
028fec41 DR |
238 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
239 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 240 | |
3e1f0645 DR |
241 | if (mode == MPOL_DEFAULT) { |
242 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 243 | return ERR_PTR(-EINVAL); |
bea904d5 | 244 | return NULL; /* simply delete any existing policy */ |
37012946 | 245 | } |
3e1f0645 DR |
246 | VM_BUG_ON(!nodes); |
247 | ||
248 | /* | |
249 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
250 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
251 | * All other modes require a valid pointer to a non-empty nodemask. | |
252 | */ | |
253 | if (mode == MPOL_PREFERRED) { | |
254 | if (nodes_empty(*nodes)) { | |
255 | if (((flags & MPOL_F_STATIC_NODES) || | |
256 | (flags & MPOL_F_RELATIVE_NODES))) | |
257 | return ERR_PTR(-EINVAL); | |
3e1f0645 DR |
258 | } |
259 | } else if (nodes_empty(*nodes)) | |
260 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
261 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
262 | if (!policy) | |
263 | return ERR_PTR(-ENOMEM); | |
264 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 265 | policy->mode = mode; |
3e1f0645 | 266 | policy->flags = flags; |
37012946 | 267 | |
1da177e4 | 268 | return policy; |
37012946 DR |
269 | } |
270 | ||
52cd3b07 LS |
271 | /* Slow path of a mpol destructor. */ |
272 | void __mpol_put(struct mempolicy *p) | |
273 | { | |
274 | if (!atomic_dec_and_test(&p->refcnt)) | |
275 | return; | |
52cd3b07 LS |
276 | kmem_cache_free(policy_cache, p); |
277 | } | |
278 | ||
37012946 DR |
279 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
280 | { | |
281 | } | |
282 | ||
283 | static void mpol_rebind_nodemask(struct mempolicy *pol, | |
284 | const nodemask_t *nodes) | |
285 | { | |
286 | nodemask_t tmp; | |
287 | ||
288 | if (pol->flags & MPOL_F_STATIC_NODES) | |
289 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
290 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
291 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
292 | else { | |
293 | nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed, | |
294 | *nodes); | |
295 | pol->w.cpuset_mems_allowed = *nodes; | |
296 | } | |
f5b087b5 | 297 | |
37012946 DR |
298 | pol->v.nodes = tmp; |
299 | if (!node_isset(current->il_next, tmp)) { | |
300 | current->il_next = next_node(current->il_next, tmp); | |
301 | if (current->il_next >= MAX_NUMNODES) | |
302 | current->il_next = first_node(tmp); | |
303 | if (current->il_next >= MAX_NUMNODES) | |
304 | current->il_next = numa_node_id(); | |
305 | } | |
306 | } | |
307 | ||
308 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
309 | const nodemask_t *nodes) | |
310 | { | |
311 | nodemask_t tmp; | |
312 | ||
37012946 DR |
313 | if (pol->flags & MPOL_F_STATIC_NODES) { |
314 | int node = first_node(pol->w.user_nodemask); | |
315 | ||
fc36b8d3 | 316 | if (node_isset(node, *nodes)) { |
37012946 | 317 | pol->v.preferred_node = node; |
fc36b8d3 LS |
318 | pol->flags &= ~MPOL_F_LOCAL; |
319 | } else | |
320 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
321 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
322 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
323 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 324 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
325 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
326 | pol->w.cpuset_mems_allowed, | |
327 | *nodes); | |
328 | pol->w.cpuset_mems_allowed = *nodes; | |
329 | } | |
1da177e4 LT |
330 | } |
331 | ||
1d0d2680 DR |
332 | /* Migrate a policy to a different set of nodes */ |
333 | static void mpol_rebind_policy(struct mempolicy *pol, | |
334 | const nodemask_t *newmask) | |
335 | { | |
1d0d2680 DR |
336 | if (!pol) |
337 | return; | |
1d0d2680 DR |
338 | if (!mpol_store_user_nodemask(pol) && |
339 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) | |
340 | return; | |
45c4745a | 341 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
342 | } |
343 | ||
344 | /* | |
345 | * Wrapper for mpol_rebind_policy() that just requires task | |
346 | * pointer, and updates task mempolicy. | |
58568d2a MX |
347 | * |
348 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
349 | */ |
350 | ||
351 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
352 | { | |
353 | mpol_rebind_policy(tsk->mempolicy, new); | |
354 | } | |
355 | ||
356 | /* | |
357 | * Rebind each vma in mm to new nodemask. | |
358 | * | |
359 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
360 | */ | |
361 | ||
362 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
363 | { | |
364 | struct vm_area_struct *vma; | |
365 | ||
366 | down_write(&mm->mmap_sem); | |
367 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
368 | mpol_rebind_policy(vma->vm_policy, new); | |
369 | up_write(&mm->mmap_sem); | |
370 | } | |
371 | ||
37012946 DR |
372 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
373 | [MPOL_DEFAULT] = { | |
374 | .rebind = mpol_rebind_default, | |
375 | }, | |
376 | [MPOL_INTERLEAVE] = { | |
377 | .create = mpol_new_interleave, | |
378 | .rebind = mpol_rebind_nodemask, | |
379 | }, | |
380 | [MPOL_PREFERRED] = { | |
381 | .create = mpol_new_preferred, | |
382 | .rebind = mpol_rebind_preferred, | |
383 | }, | |
384 | [MPOL_BIND] = { | |
385 | .create = mpol_new_bind, | |
386 | .rebind = mpol_rebind_nodemask, | |
387 | }, | |
388 | }; | |
389 | ||
397874df | 390 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
391 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
392 | unsigned long flags); | |
1a75a6c8 | 393 | |
38e35860 | 394 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 395 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
396 | unsigned long addr, unsigned long end, |
397 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 398 | void *private) |
1da177e4 | 399 | { |
91612e0d HD |
400 | pte_t *orig_pte; |
401 | pte_t *pte; | |
705e87c0 | 402 | spinlock_t *ptl; |
941150a3 | 403 | |
705e87c0 | 404 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 405 | do { |
6aab341e | 406 | struct page *page; |
25ba77c1 | 407 | int nid; |
91612e0d HD |
408 | |
409 | if (!pte_present(*pte)) | |
1da177e4 | 410 | continue; |
6aab341e LT |
411 | page = vm_normal_page(vma, addr, *pte); |
412 | if (!page) | |
1da177e4 | 413 | continue; |
053837fc NP |
414 | /* |
415 | * The check for PageReserved here is important to avoid | |
416 | * handling zero pages and other pages that may have been | |
417 | * marked special by the system. | |
418 | * | |
419 | * If the PageReserved would not be checked here then f.e. | |
420 | * the location of the zero page could have an influence | |
421 | * on MPOL_MF_STRICT, zero pages would be counted for | |
422 | * the per node stats, and there would be useless attempts | |
423 | * to put zero pages on the migration list. | |
424 | */ | |
f4598c8b CL |
425 | if (PageReserved(page)) |
426 | continue; | |
6aab341e | 427 | nid = page_to_nid(page); |
38e35860 CL |
428 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
429 | continue; | |
430 | ||
1a75a6c8 | 431 | if (flags & MPOL_MF_STATS) |
397874df | 432 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 433 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 434 | migrate_page_add(page, private, flags); |
38e35860 CL |
435 | else |
436 | break; | |
91612e0d | 437 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 438 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
439 | return addr != end; |
440 | } | |
441 | ||
b5810039 | 442 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
443 | unsigned long addr, unsigned long end, |
444 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 445 | void *private) |
91612e0d HD |
446 | { |
447 | pmd_t *pmd; | |
448 | unsigned long next; | |
449 | ||
450 | pmd = pmd_offset(pud, addr); | |
451 | do { | |
452 | next = pmd_addr_end(addr, end); | |
453 | if (pmd_none_or_clear_bad(pmd)) | |
454 | continue; | |
dc9aa5b9 | 455 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 456 | flags, private)) |
91612e0d HD |
457 | return -EIO; |
458 | } while (pmd++, addr = next, addr != end); | |
459 | return 0; | |
460 | } | |
461 | ||
b5810039 | 462 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
463 | unsigned long addr, unsigned long end, |
464 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 465 | void *private) |
91612e0d HD |
466 | { |
467 | pud_t *pud; | |
468 | unsigned long next; | |
469 | ||
470 | pud = pud_offset(pgd, addr); | |
471 | do { | |
472 | next = pud_addr_end(addr, end); | |
473 | if (pud_none_or_clear_bad(pud)) | |
474 | continue; | |
dc9aa5b9 | 475 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 476 | flags, private)) |
91612e0d HD |
477 | return -EIO; |
478 | } while (pud++, addr = next, addr != end); | |
479 | return 0; | |
480 | } | |
481 | ||
b5810039 | 482 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
483 | unsigned long addr, unsigned long end, |
484 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 485 | void *private) |
91612e0d HD |
486 | { |
487 | pgd_t *pgd; | |
488 | unsigned long next; | |
489 | ||
b5810039 | 490 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
491 | do { |
492 | next = pgd_addr_end(addr, end); | |
493 | if (pgd_none_or_clear_bad(pgd)) | |
494 | continue; | |
dc9aa5b9 | 495 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 496 | flags, private)) |
91612e0d HD |
497 | return -EIO; |
498 | } while (pgd++, addr = next, addr != end); | |
499 | return 0; | |
1da177e4 LT |
500 | } |
501 | ||
dc9aa5b9 CL |
502 | /* |
503 | * Check if all pages in a range are on a set of nodes. | |
504 | * If pagelist != NULL then isolate pages from the LRU and | |
505 | * put them on the pagelist. | |
506 | */ | |
1da177e4 LT |
507 | static struct vm_area_struct * |
508 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 509 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
510 | { |
511 | int err; | |
512 | struct vm_area_struct *first, *vma, *prev; | |
513 | ||
053837fc | 514 | |
1da177e4 LT |
515 | first = find_vma(mm, start); |
516 | if (!first) | |
517 | return ERR_PTR(-EFAULT); | |
518 | prev = NULL; | |
519 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
520 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
521 | if (!vma->vm_next && vma->vm_end < end) | |
522 | return ERR_PTR(-EFAULT); | |
523 | if (prev && prev->vm_end < vma->vm_start) | |
524 | return ERR_PTR(-EFAULT); | |
525 | } | |
526 | if (!is_vm_hugetlb_page(vma) && | |
527 | ((flags & MPOL_MF_STRICT) || | |
528 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
529 | vma_migratable(vma)))) { | |
5b952b3c | 530 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 531 | |
5b952b3c AK |
532 | if (endvma > end) |
533 | endvma = end; | |
534 | if (vma->vm_start > start) | |
535 | start = vma->vm_start; | |
dc9aa5b9 | 536 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 537 | flags, private); |
1da177e4 LT |
538 | if (err) { |
539 | first = ERR_PTR(err); | |
540 | break; | |
541 | } | |
542 | } | |
543 | prev = vma; | |
544 | } | |
545 | return first; | |
546 | } | |
547 | ||
548 | /* Apply policy to a single VMA */ | |
549 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
550 | { | |
551 | int err = 0; | |
552 | struct mempolicy *old = vma->vm_policy; | |
553 | ||
140d5a49 | 554 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
555 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
556 | vma->vm_ops, vma->vm_file, | |
557 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
558 | ||
559 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
560 | err = vma->vm_ops->set_policy(vma, new); | |
561 | if (!err) { | |
562 | mpol_get(new); | |
563 | vma->vm_policy = new; | |
f0be3d32 | 564 | mpol_put(old); |
1da177e4 LT |
565 | } |
566 | return err; | |
567 | } | |
568 | ||
569 | /* Step 2: apply policy to a range and do splits. */ | |
570 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
571 | unsigned long end, struct mempolicy *new) | |
572 | { | |
573 | struct vm_area_struct *next; | |
574 | int err; | |
575 | ||
576 | err = 0; | |
577 | for (; vma && vma->vm_start < end; vma = next) { | |
578 | next = vma->vm_next; | |
579 | if (vma->vm_start < start) | |
580 | err = split_vma(vma->vm_mm, vma, start, 1); | |
581 | if (!err && vma->vm_end > end) | |
582 | err = split_vma(vma->vm_mm, vma, end, 0); | |
583 | if (!err) | |
584 | err = policy_vma(vma, new); | |
585 | if (err) | |
586 | break; | |
587 | } | |
588 | return err; | |
589 | } | |
590 | ||
c61afb18 PJ |
591 | /* |
592 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
593 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
594 | * with other PF_* flag bits) on memory allocation hot code paths. | |
595 | * | |
596 | * If called from outside this file, the task 'p' should -only- be | |
597 | * a newly forked child not yet visible on the task list, because | |
598 | * manipulating the task flags of a visible task is not safe. | |
599 | * | |
600 | * The above limitation is why this routine has the funny name | |
601 | * mpol_fix_fork_child_flag(). | |
602 | * | |
603 | * It is also safe to call this with a task pointer of current, | |
604 | * which the static wrapper mpol_set_task_struct_flag() does, | |
605 | * for use within this file. | |
606 | */ | |
607 | ||
608 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
609 | { | |
610 | if (p->mempolicy) | |
611 | p->flags |= PF_MEMPOLICY; | |
612 | else | |
613 | p->flags &= ~PF_MEMPOLICY; | |
614 | } | |
615 | ||
616 | static void mpol_set_task_struct_flag(void) | |
617 | { | |
618 | mpol_fix_fork_child_flag(current); | |
619 | } | |
620 | ||
1da177e4 | 621 | /* Set the process memory policy */ |
028fec41 DR |
622 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
623 | nodemask_t *nodes) | |
1da177e4 | 624 | { |
58568d2a | 625 | struct mempolicy *new, *old; |
f4e53d91 | 626 | struct mm_struct *mm = current->mm; |
4bfc4495 | 627 | NODEMASK_SCRATCH(scratch); |
58568d2a | 628 | int ret; |
1da177e4 | 629 | |
4bfc4495 KH |
630 | if (!scratch) |
631 | return -ENOMEM; | |
f4e53d91 | 632 | |
4bfc4495 KH |
633 | new = mpol_new(mode, flags, nodes); |
634 | if (IS_ERR(new)) { | |
635 | ret = PTR_ERR(new); | |
636 | goto out; | |
637 | } | |
f4e53d91 LS |
638 | /* |
639 | * prevent changing our mempolicy while show_numa_maps() | |
640 | * is using it. | |
641 | * Note: do_set_mempolicy() can be called at init time | |
642 | * with no 'mm'. | |
643 | */ | |
644 | if (mm) | |
645 | down_write(&mm->mmap_sem); | |
58568d2a | 646 | task_lock(current); |
4bfc4495 | 647 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
648 | if (ret) { |
649 | task_unlock(current); | |
650 | if (mm) | |
651 | up_write(&mm->mmap_sem); | |
652 | mpol_put(new); | |
4bfc4495 | 653 | goto out; |
58568d2a MX |
654 | } |
655 | old = current->mempolicy; | |
1da177e4 | 656 | current->mempolicy = new; |
c61afb18 | 657 | mpol_set_task_struct_flag(); |
45c4745a | 658 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 659 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 660 | current->il_next = first_node(new->v.nodes); |
58568d2a | 661 | task_unlock(current); |
f4e53d91 LS |
662 | if (mm) |
663 | up_write(&mm->mmap_sem); | |
664 | ||
58568d2a | 665 | mpol_put(old); |
4bfc4495 KH |
666 | ret = 0; |
667 | out: | |
668 | NODEMASK_SCRATCH_FREE(scratch); | |
669 | return ret; | |
1da177e4 LT |
670 | } |
671 | ||
bea904d5 LS |
672 | /* |
673 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
674 | * |
675 | * Called with task's alloc_lock held | |
bea904d5 LS |
676 | */ |
677 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 678 | { |
dfcd3c0d | 679 | nodes_clear(*nodes); |
bea904d5 LS |
680 | if (p == &default_policy) |
681 | return; | |
682 | ||
45c4745a | 683 | switch (p->mode) { |
19770b32 MG |
684 | case MPOL_BIND: |
685 | /* Fall through */ | |
1da177e4 | 686 | case MPOL_INTERLEAVE: |
dfcd3c0d | 687 | *nodes = p->v.nodes; |
1da177e4 LT |
688 | break; |
689 | case MPOL_PREFERRED: | |
fc36b8d3 | 690 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 691 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 692 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
693 | break; |
694 | default: | |
695 | BUG(); | |
696 | } | |
697 | } | |
698 | ||
699 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
700 | { | |
701 | struct page *p; | |
702 | int err; | |
703 | ||
704 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
705 | if (err >= 0) { | |
706 | err = page_to_nid(p); | |
707 | put_page(p); | |
708 | } | |
709 | return err; | |
710 | } | |
711 | ||
1da177e4 | 712 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
713 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
714 | unsigned long addr, unsigned long flags) | |
1da177e4 | 715 | { |
8bccd85f | 716 | int err; |
1da177e4 LT |
717 | struct mm_struct *mm = current->mm; |
718 | struct vm_area_struct *vma = NULL; | |
719 | struct mempolicy *pol = current->mempolicy; | |
720 | ||
754af6f5 LS |
721 | if (flags & |
722 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 723 | return -EINVAL; |
754af6f5 LS |
724 | |
725 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
726 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
727 | return -EINVAL; | |
728 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 729 | task_lock(current); |
754af6f5 | 730 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 731 | task_unlock(current); |
754af6f5 LS |
732 | return 0; |
733 | } | |
734 | ||
1da177e4 | 735 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
736 | /* |
737 | * Do NOT fall back to task policy if the | |
738 | * vma/shared policy at addr is NULL. We | |
739 | * want to return MPOL_DEFAULT in this case. | |
740 | */ | |
1da177e4 LT |
741 | down_read(&mm->mmap_sem); |
742 | vma = find_vma_intersection(mm, addr, addr+1); | |
743 | if (!vma) { | |
744 | up_read(&mm->mmap_sem); | |
745 | return -EFAULT; | |
746 | } | |
747 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
748 | pol = vma->vm_ops->get_policy(vma, addr); | |
749 | else | |
750 | pol = vma->vm_policy; | |
751 | } else if (addr) | |
752 | return -EINVAL; | |
753 | ||
754 | if (!pol) | |
bea904d5 | 755 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
756 | |
757 | if (flags & MPOL_F_NODE) { | |
758 | if (flags & MPOL_F_ADDR) { | |
759 | err = lookup_node(mm, addr); | |
760 | if (err < 0) | |
761 | goto out; | |
8bccd85f | 762 | *policy = err; |
1da177e4 | 763 | } else if (pol == current->mempolicy && |
45c4745a | 764 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 765 | *policy = current->il_next; |
1da177e4 LT |
766 | } else { |
767 | err = -EINVAL; | |
768 | goto out; | |
769 | } | |
bea904d5 LS |
770 | } else { |
771 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
772 | pol->mode; | |
d79df630 DR |
773 | /* |
774 | * Internal mempolicy flags must be masked off before exposing | |
775 | * the policy to userspace. | |
776 | */ | |
777 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 778 | } |
1da177e4 LT |
779 | |
780 | if (vma) { | |
781 | up_read(¤t->mm->mmap_sem); | |
782 | vma = NULL; | |
783 | } | |
784 | ||
1da177e4 | 785 | err = 0; |
58568d2a MX |
786 | if (nmask) { |
787 | task_lock(current); | |
bea904d5 | 788 | get_policy_nodemask(pol, nmask); |
58568d2a MX |
789 | task_unlock(current); |
790 | } | |
1da177e4 LT |
791 | |
792 | out: | |
52cd3b07 | 793 | mpol_cond_put(pol); |
1da177e4 LT |
794 | if (vma) |
795 | up_read(¤t->mm->mmap_sem); | |
796 | return err; | |
797 | } | |
798 | ||
b20a3503 | 799 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
800 | /* |
801 | * page migration | |
802 | */ | |
fc301289 CL |
803 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
804 | unsigned long flags) | |
6ce3c4c0 CL |
805 | { |
806 | /* | |
fc301289 | 807 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 808 | */ |
62695a84 NP |
809 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
810 | if (!isolate_lru_page(page)) { | |
811 | list_add_tail(&page->lru, pagelist); | |
812 | } | |
813 | } | |
7e2ab150 | 814 | } |
6ce3c4c0 | 815 | |
742755a1 | 816 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 817 | { |
6484eb3e | 818 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
819 | } |
820 | ||
7e2ab150 CL |
821 | /* |
822 | * Migrate pages from one node to a target node. | |
823 | * Returns error or the number of pages not migrated. | |
824 | */ | |
dbcb0f19 AB |
825 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
826 | int flags) | |
7e2ab150 CL |
827 | { |
828 | nodemask_t nmask; | |
829 | LIST_HEAD(pagelist); | |
830 | int err = 0; | |
831 | ||
832 | nodes_clear(nmask); | |
833 | node_set(source, nmask); | |
6ce3c4c0 | 834 | |
7e2ab150 CL |
835 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
836 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
837 | ||
aaa994b3 | 838 | if (!list_empty(&pagelist)) |
95a402c3 CL |
839 | err = migrate_pages(&pagelist, new_node_page, dest); |
840 | ||
7e2ab150 | 841 | return err; |
6ce3c4c0 CL |
842 | } |
843 | ||
39743889 | 844 | /* |
7e2ab150 CL |
845 | * Move pages between the two nodesets so as to preserve the physical |
846 | * layout as much as possible. | |
39743889 CL |
847 | * |
848 | * Returns the number of page that could not be moved. | |
849 | */ | |
850 | int do_migrate_pages(struct mm_struct *mm, | |
851 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
852 | { | |
7e2ab150 | 853 | int busy = 0; |
0aedadf9 | 854 | int err; |
7e2ab150 | 855 | nodemask_t tmp; |
39743889 | 856 | |
0aedadf9 CL |
857 | err = migrate_prep(); |
858 | if (err) | |
859 | return err; | |
860 | ||
53f2556b | 861 | down_read(&mm->mmap_sem); |
39743889 | 862 | |
7b2259b3 CL |
863 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
864 | if (err) | |
865 | goto out; | |
866 | ||
7e2ab150 CL |
867 | /* |
868 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
869 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
870 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
871 | * The pair of nodemasks 'to' and 'from' define the map. | |
872 | * | |
873 | * If no pair of bits is found that way, fallback to picking some | |
874 | * pair of 'source' and 'dest' bits that are not the same. If the | |
875 | * 'source' and 'dest' bits are the same, this represents a node | |
876 | * that will be migrating to itself, so no pages need move. | |
877 | * | |
878 | * If no bits are left in 'tmp', or if all remaining bits left | |
879 | * in 'tmp' correspond to the same bit in 'to', return false | |
880 | * (nothing left to migrate). | |
881 | * | |
882 | * This lets us pick a pair of nodes to migrate between, such that | |
883 | * if possible the dest node is not already occupied by some other | |
884 | * source node, minimizing the risk of overloading the memory on a | |
885 | * node that would happen if we migrated incoming memory to a node | |
886 | * before migrating outgoing memory source that same node. | |
887 | * | |
888 | * A single scan of tmp is sufficient. As we go, we remember the | |
889 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
890 | * that not only moved, but what's better, moved to an empty slot | |
891 | * (d is not set in tmp), then we break out then, with that pair. | |
892 | * Otherwise when we finish scannng from_tmp, we at least have the | |
893 | * most recent <s, d> pair that moved. If we get all the way through | |
894 | * the scan of tmp without finding any node that moved, much less | |
895 | * moved to an empty node, then there is nothing left worth migrating. | |
896 | */ | |
d4984711 | 897 | |
7e2ab150 CL |
898 | tmp = *from_nodes; |
899 | while (!nodes_empty(tmp)) { | |
900 | int s,d; | |
901 | int source = -1; | |
902 | int dest = 0; | |
903 | ||
904 | for_each_node_mask(s, tmp) { | |
905 | d = node_remap(s, *from_nodes, *to_nodes); | |
906 | if (s == d) | |
907 | continue; | |
908 | ||
909 | source = s; /* Node moved. Memorize */ | |
910 | dest = d; | |
911 | ||
912 | /* dest not in remaining from nodes? */ | |
913 | if (!node_isset(dest, tmp)) | |
914 | break; | |
915 | } | |
916 | if (source == -1) | |
917 | break; | |
918 | ||
919 | node_clear(source, tmp); | |
920 | err = migrate_to_node(mm, source, dest, flags); | |
921 | if (err > 0) | |
922 | busy += err; | |
923 | if (err < 0) | |
924 | break; | |
39743889 | 925 | } |
7b2259b3 | 926 | out: |
39743889 | 927 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
928 | if (err < 0) |
929 | return err; | |
930 | return busy; | |
b20a3503 CL |
931 | |
932 | } | |
933 | ||
3ad33b24 LS |
934 | /* |
935 | * Allocate a new page for page migration based on vma policy. | |
936 | * Start assuming that page is mapped by vma pointed to by @private. | |
937 | * Search forward from there, if not. N.B., this assumes that the | |
938 | * list of pages handed to migrate_pages()--which is how we get here-- | |
939 | * is in virtual address order. | |
940 | */ | |
742755a1 | 941 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
942 | { |
943 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 944 | unsigned long uninitialized_var(address); |
95a402c3 | 945 | |
3ad33b24 LS |
946 | while (vma) { |
947 | address = page_address_in_vma(page, vma); | |
948 | if (address != -EFAULT) | |
949 | break; | |
950 | vma = vma->vm_next; | |
951 | } | |
952 | ||
953 | /* | |
954 | * if !vma, alloc_page_vma() will use task or system default policy | |
955 | */ | |
956 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 957 | } |
b20a3503 CL |
958 | #else |
959 | ||
960 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
961 | unsigned long flags) | |
962 | { | |
39743889 CL |
963 | } |
964 | ||
b20a3503 CL |
965 | int do_migrate_pages(struct mm_struct *mm, |
966 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
967 | { | |
968 | return -ENOSYS; | |
969 | } | |
95a402c3 | 970 | |
69939749 | 971 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
972 | { |
973 | return NULL; | |
974 | } | |
b20a3503 CL |
975 | #endif |
976 | ||
dbcb0f19 | 977 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
978 | unsigned short mode, unsigned short mode_flags, |
979 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
980 | { |
981 | struct vm_area_struct *vma; | |
982 | struct mm_struct *mm = current->mm; | |
983 | struct mempolicy *new; | |
984 | unsigned long end; | |
985 | int err; | |
986 | LIST_HEAD(pagelist); | |
987 | ||
a3b51e01 DR |
988 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
989 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 990 | return -EINVAL; |
74c00241 | 991 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
992 | return -EPERM; |
993 | ||
994 | if (start & ~PAGE_MASK) | |
995 | return -EINVAL; | |
996 | ||
997 | if (mode == MPOL_DEFAULT) | |
998 | flags &= ~MPOL_MF_STRICT; | |
999 | ||
1000 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1001 | end = start + len; | |
1002 | ||
1003 | if (end < start) | |
1004 | return -EINVAL; | |
1005 | if (end == start) | |
1006 | return 0; | |
1007 | ||
028fec41 | 1008 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1009 | if (IS_ERR(new)) |
1010 | return PTR_ERR(new); | |
1011 | ||
1012 | /* | |
1013 | * If we are using the default policy then operation | |
1014 | * on discontinuous address spaces is okay after all | |
1015 | */ | |
1016 | if (!new) | |
1017 | flags |= MPOL_MF_DISCONTIG_OK; | |
1018 | ||
028fec41 DR |
1019 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1020 | start, start + len, mode, mode_flags, | |
1021 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 | 1022 | |
0aedadf9 CL |
1023 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1024 | ||
1025 | err = migrate_prep(); | |
1026 | if (err) | |
b05ca738 | 1027 | goto mpol_out; |
0aedadf9 | 1028 | } |
4bfc4495 KH |
1029 | { |
1030 | NODEMASK_SCRATCH(scratch); | |
1031 | if (scratch) { | |
1032 | down_write(&mm->mmap_sem); | |
1033 | task_lock(current); | |
1034 | err = mpol_set_nodemask(new, nmask, scratch); | |
1035 | task_unlock(current); | |
1036 | if (err) | |
1037 | up_write(&mm->mmap_sem); | |
1038 | } else | |
1039 | err = -ENOMEM; | |
1040 | NODEMASK_SCRATCH_FREE(scratch); | |
1041 | } | |
b05ca738 KM |
1042 | if (err) |
1043 | goto mpol_out; | |
1044 | ||
6ce3c4c0 CL |
1045 | vma = check_range(mm, start, end, nmask, |
1046 | flags | MPOL_MF_INVERT, &pagelist); | |
1047 | ||
1048 | err = PTR_ERR(vma); | |
1049 | if (!IS_ERR(vma)) { | |
1050 | int nr_failed = 0; | |
1051 | ||
1052 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 1053 | |
6ce3c4c0 | 1054 | if (!list_empty(&pagelist)) |
95a402c3 CL |
1055 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
1056 | (unsigned long)vma); | |
6ce3c4c0 CL |
1057 | |
1058 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
1059 | err = -EIO; | |
ab8a3e14 KM |
1060 | } else |
1061 | putback_lru_pages(&pagelist); | |
b20a3503 | 1062 | |
6ce3c4c0 | 1063 | up_write(&mm->mmap_sem); |
b05ca738 | 1064 | mpol_out: |
f0be3d32 | 1065 | mpol_put(new); |
6ce3c4c0 CL |
1066 | return err; |
1067 | } | |
1068 | ||
8bccd85f CL |
1069 | /* |
1070 | * User space interface with variable sized bitmaps for nodelists. | |
1071 | */ | |
1072 | ||
1073 | /* Copy a node mask from user space. */ | |
39743889 | 1074 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1075 | unsigned long maxnode) |
1076 | { | |
1077 | unsigned long k; | |
1078 | unsigned long nlongs; | |
1079 | unsigned long endmask; | |
1080 | ||
1081 | --maxnode; | |
1082 | nodes_clear(*nodes); | |
1083 | if (maxnode == 0 || !nmask) | |
1084 | return 0; | |
a9c930ba | 1085 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1086 | return -EINVAL; |
8bccd85f CL |
1087 | |
1088 | nlongs = BITS_TO_LONGS(maxnode); | |
1089 | if ((maxnode % BITS_PER_LONG) == 0) | |
1090 | endmask = ~0UL; | |
1091 | else | |
1092 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1093 | ||
1094 | /* When the user specified more nodes than supported just check | |
1095 | if the non supported part is all zero. */ | |
1096 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1097 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1098 | return -EINVAL; | |
1099 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1100 | unsigned long t; | |
1101 | if (get_user(t, nmask + k)) | |
1102 | return -EFAULT; | |
1103 | if (k == nlongs - 1) { | |
1104 | if (t & endmask) | |
1105 | return -EINVAL; | |
1106 | } else if (t) | |
1107 | return -EINVAL; | |
1108 | } | |
1109 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1110 | endmask = ~0UL; | |
1111 | } | |
1112 | ||
1113 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1114 | return -EFAULT; | |
1115 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | /* Copy a kernel node mask to user space */ | |
1120 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1121 | nodemask_t *nodes) | |
1122 | { | |
1123 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1124 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1125 | ||
1126 | if (copy > nbytes) { | |
1127 | if (copy > PAGE_SIZE) | |
1128 | return -EINVAL; | |
1129 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1130 | return -EFAULT; | |
1131 | copy = nbytes; | |
1132 | } | |
1133 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1134 | } | |
1135 | ||
938bb9f5 HC |
1136 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1137 | unsigned long, mode, unsigned long __user *, nmask, | |
1138 | unsigned long, maxnode, unsigned, flags) | |
8bccd85f CL |
1139 | { |
1140 | nodemask_t nodes; | |
1141 | int err; | |
028fec41 | 1142 | unsigned short mode_flags; |
8bccd85f | 1143 | |
028fec41 DR |
1144 | mode_flags = mode & MPOL_MODE_FLAGS; |
1145 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1146 | if (mode >= MPOL_MAX) |
1147 | return -EINVAL; | |
4c50bc01 DR |
1148 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1149 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1150 | return -EINVAL; | |
8bccd85f CL |
1151 | err = get_nodes(&nodes, nmask, maxnode); |
1152 | if (err) | |
1153 | return err; | |
028fec41 | 1154 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1155 | } |
1156 | ||
1157 | /* Set the process memory policy */ | |
938bb9f5 HC |
1158 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
1159 | unsigned long, maxnode) | |
8bccd85f CL |
1160 | { |
1161 | int err; | |
1162 | nodemask_t nodes; | |
028fec41 | 1163 | unsigned short flags; |
8bccd85f | 1164 | |
028fec41 DR |
1165 | flags = mode & MPOL_MODE_FLAGS; |
1166 | mode &= ~MPOL_MODE_FLAGS; | |
1167 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1168 | return -EINVAL; |
4c50bc01 DR |
1169 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1170 | return -EINVAL; | |
8bccd85f CL |
1171 | err = get_nodes(&nodes, nmask, maxnode); |
1172 | if (err) | |
1173 | return err; | |
028fec41 | 1174 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1175 | } |
1176 | ||
938bb9f5 HC |
1177 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1178 | const unsigned long __user *, old_nodes, | |
1179 | const unsigned long __user *, new_nodes) | |
39743889 | 1180 | { |
c69e8d9c | 1181 | const struct cred *cred = current_cred(), *tcred; |
39743889 CL |
1182 | struct mm_struct *mm; |
1183 | struct task_struct *task; | |
1184 | nodemask_t old; | |
1185 | nodemask_t new; | |
1186 | nodemask_t task_nodes; | |
1187 | int err; | |
1188 | ||
1189 | err = get_nodes(&old, old_nodes, maxnode); | |
1190 | if (err) | |
1191 | return err; | |
1192 | ||
1193 | err = get_nodes(&new, new_nodes, maxnode); | |
1194 | if (err) | |
1195 | return err; | |
1196 | ||
1197 | /* Find the mm_struct */ | |
1198 | read_lock(&tasklist_lock); | |
228ebcbe | 1199 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 CL |
1200 | if (!task) { |
1201 | read_unlock(&tasklist_lock); | |
1202 | return -ESRCH; | |
1203 | } | |
1204 | mm = get_task_mm(task); | |
1205 | read_unlock(&tasklist_lock); | |
1206 | ||
1207 | if (!mm) | |
1208 | return -EINVAL; | |
1209 | ||
1210 | /* | |
1211 | * Check if this process has the right to modify the specified | |
1212 | * process. The right exists if the process has administrative | |
7f927fcc | 1213 | * capabilities, superuser privileges or the same |
39743889 CL |
1214 | * userid as the target process. |
1215 | */ | |
c69e8d9c DH |
1216 | rcu_read_lock(); |
1217 | tcred = __task_cred(task); | |
b6dff3ec DH |
1218 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1219 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
74c00241 | 1220 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1221 | rcu_read_unlock(); |
39743889 CL |
1222 | err = -EPERM; |
1223 | goto out; | |
1224 | } | |
c69e8d9c | 1225 | rcu_read_unlock(); |
39743889 CL |
1226 | |
1227 | task_nodes = cpuset_mems_allowed(task); | |
1228 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 1229 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
1230 | err = -EPERM; |
1231 | goto out; | |
1232 | } | |
1233 | ||
37b07e41 | 1234 | if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
1235 | err = -EINVAL; |
1236 | goto out; | |
1237 | } | |
1238 | ||
86c3a764 DQ |
1239 | err = security_task_movememory(task); |
1240 | if (err) | |
1241 | goto out; | |
1242 | ||
511030bc | 1243 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 1244 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
1245 | out: |
1246 | mmput(mm); | |
1247 | return err; | |
1248 | } | |
1249 | ||
1250 | ||
8bccd85f | 1251 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1252 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1253 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1254 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1255 | { |
dbcb0f19 AB |
1256 | int err; |
1257 | int uninitialized_var(pval); | |
8bccd85f CL |
1258 | nodemask_t nodes; |
1259 | ||
1260 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1261 | return -EINVAL; | |
1262 | ||
1263 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1264 | ||
1265 | if (err) | |
1266 | return err; | |
1267 | ||
1268 | if (policy && put_user(pval, policy)) | |
1269 | return -EFAULT; | |
1270 | ||
1271 | if (nmask) | |
1272 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1273 | ||
1274 | return err; | |
1275 | } | |
1276 | ||
1da177e4 LT |
1277 | #ifdef CONFIG_COMPAT |
1278 | ||
1279 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1280 | compat_ulong_t __user *nmask, | |
1281 | compat_ulong_t maxnode, | |
1282 | compat_ulong_t addr, compat_ulong_t flags) | |
1283 | { | |
1284 | long err; | |
1285 | unsigned long __user *nm = NULL; | |
1286 | unsigned long nr_bits, alloc_size; | |
1287 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1288 | ||
1289 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1290 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1291 | ||
1292 | if (nmask) | |
1293 | nm = compat_alloc_user_space(alloc_size); | |
1294 | ||
1295 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1296 | ||
1297 | if (!err && nmask) { | |
1298 | err = copy_from_user(bm, nm, alloc_size); | |
1299 | /* ensure entire bitmap is zeroed */ | |
1300 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1301 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1302 | } | |
1303 | ||
1304 | return err; | |
1305 | } | |
1306 | ||
1307 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1308 | compat_ulong_t maxnode) | |
1309 | { | |
1310 | long err = 0; | |
1311 | unsigned long __user *nm = NULL; | |
1312 | unsigned long nr_bits, alloc_size; | |
1313 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1314 | ||
1315 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1316 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1317 | ||
1318 | if (nmask) { | |
1319 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1320 | nm = compat_alloc_user_space(alloc_size); | |
1321 | err |= copy_to_user(nm, bm, alloc_size); | |
1322 | } | |
1323 | ||
1324 | if (err) | |
1325 | return -EFAULT; | |
1326 | ||
1327 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1328 | } | |
1329 | ||
1330 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1331 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1332 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1333 | { | |
1334 | long err = 0; | |
1335 | unsigned long __user *nm = NULL; | |
1336 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1337 | nodemask_t bm; |
1da177e4 LT |
1338 | |
1339 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1340 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1341 | ||
1342 | if (nmask) { | |
dfcd3c0d | 1343 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1344 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1345 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1346 | } |
1347 | ||
1348 | if (err) | |
1349 | return -EFAULT; | |
1350 | ||
1351 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1352 | } | |
1353 | ||
1354 | #endif | |
1355 | ||
480eccf9 LS |
1356 | /* |
1357 | * get_vma_policy(@task, @vma, @addr) | |
1358 | * @task - task for fallback if vma policy == default | |
1359 | * @vma - virtual memory area whose policy is sought | |
1360 | * @addr - address in @vma for shared policy lookup | |
1361 | * | |
1362 | * Returns effective policy for a VMA at specified address. | |
1363 | * Falls back to @task or system default policy, as necessary. | |
52cd3b07 LS |
1364 | * Current or other task's task mempolicy and non-shared vma policies |
1365 | * are protected by the task's mmap_sem, which must be held for read by | |
1366 | * the caller. | |
1367 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference | |
1368 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1369 | * freeing by another task. It is the caller's responsibility to free the | |
1370 | * extra reference for shared policies. | |
480eccf9 | 1371 | */ |
ae4d8c16 | 1372 | static struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1373 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1374 | { |
6e21c8f1 | 1375 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1376 | |
1377 | if (vma) { | |
480eccf9 | 1378 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1379 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1380 | addr); | |
1381 | if (vpol) | |
1382 | pol = vpol; | |
bea904d5 | 1383 | } else if (vma->vm_policy) |
1da177e4 LT |
1384 | pol = vma->vm_policy; |
1385 | } | |
1386 | if (!pol) | |
1387 | pol = &default_policy; | |
1388 | return pol; | |
1389 | } | |
1390 | ||
52cd3b07 LS |
1391 | /* |
1392 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1393 | * page allocation | |
1394 | */ | |
1395 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1396 | { |
1397 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1398 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1399 | gfp_zone(gfp) >= policy_zone && |
1400 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1401 | return &policy->v.nodes; | |
1402 | ||
1403 | return NULL; | |
1404 | } | |
1405 | ||
52cd3b07 LS |
1406 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
1407 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy) | |
1da177e4 | 1408 | { |
fc36b8d3 | 1409 | int nd = numa_node_id(); |
1da177e4 | 1410 | |
45c4745a | 1411 | switch (policy->mode) { |
1da177e4 | 1412 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1413 | if (!(policy->flags & MPOL_F_LOCAL)) |
1414 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1415 | break; |
1416 | case MPOL_BIND: | |
19770b32 | 1417 | /* |
52cd3b07 LS |
1418 | * Normally, MPOL_BIND allocations are node-local within the |
1419 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
1420 | * current node is part of the mask, we use the zonelist for | |
1421 | * the first node in the mask instead. | |
19770b32 | 1422 | */ |
19770b32 MG |
1423 | if (unlikely(gfp & __GFP_THISNODE) && |
1424 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1425 | nd = first_node(policy->v.nodes); | |
1426 | break; | |
1da177e4 | 1427 | case MPOL_INTERLEAVE: /* should not happen */ |
1da177e4 LT |
1428 | break; |
1429 | default: | |
1da177e4 LT |
1430 | BUG(); |
1431 | } | |
0e88460d | 1432 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | /* Do dynamic interleaving for a process */ | |
1436 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1437 | { | |
1438 | unsigned nid, next; | |
1439 | struct task_struct *me = current; | |
1440 | ||
1441 | nid = me->il_next; | |
dfcd3c0d | 1442 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1443 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1444 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1445 | if (next < MAX_NUMNODES) |
1446 | me->il_next = next; | |
1da177e4 LT |
1447 | return nid; |
1448 | } | |
1449 | ||
dc85da15 CL |
1450 | /* |
1451 | * Depending on the memory policy provide a node from which to allocate the | |
1452 | * next slab entry. | |
52cd3b07 LS |
1453 | * @policy must be protected by freeing by the caller. If @policy is |
1454 | * the current task's mempolicy, this protection is implicit, as only the | |
1455 | * task can change it's policy. The system default policy requires no | |
1456 | * such protection. | |
dc85da15 CL |
1457 | */ |
1458 | unsigned slab_node(struct mempolicy *policy) | |
1459 | { | |
fc36b8d3 | 1460 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1461 | return numa_node_id(); |
1462 | ||
1463 | switch (policy->mode) { | |
1464 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1465 | /* |
1466 | * handled MPOL_F_LOCAL above | |
1467 | */ | |
1468 | return policy->v.preferred_node; | |
765c4507 | 1469 | |
dc85da15 CL |
1470 | case MPOL_INTERLEAVE: |
1471 | return interleave_nodes(policy); | |
1472 | ||
dd1a239f | 1473 | case MPOL_BIND: { |
dc85da15 CL |
1474 | /* |
1475 | * Follow bind policy behavior and start allocation at the | |
1476 | * first node. | |
1477 | */ | |
19770b32 MG |
1478 | struct zonelist *zonelist; |
1479 | struct zone *zone; | |
1480 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1481 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1482 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1483 | &policy->v.nodes, | |
1484 | &zone); | |
1485 | return zone->node; | |
dd1a239f | 1486 | } |
dc85da15 | 1487 | |
dc85da15 | 1488 | default: |
bea904d5 | 1489 | BUG(); |
dc85da15 CL |
1490 | } |
1491 | } | |
1492 | ||
1da177e4 LT |
1493 | /* Do static interleaving for a VMA with known offset. */ |
1494 | static unsigned offset_il_node(struct mempolicy *pol, | |
1495 | struct vm_area_struct *vma, unsigned long off) | |
1496 | { | |
dfcd3c0d | 1497 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1498 | unsigned target; |
1da177e4 LT |
1499 | int c; |
1500 | int nid = -1; | |
1501 | ||
f5b087b5 DR |
1502 | if (!nnodes) |
1503 | return numa_node_id(); | |
1504 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1505 | c = 0; |
1506 | do { | |
dfcd3c0d | 1507 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1508 | c++; |
1509 | } while (c <= target); | |
1da177e4 LT |
1510 | return nid; |
1511 | } | |
1512 | ||
5da7ca86 CL |
1513 | /* Determine a node number for interleave */ |
1514 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1515 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1516 | { | |
1517 | if (vma) { | |
1518 | unsigned long off; | |
1519 | ||
3b98b087 NA |
1520 | /* |
1521 | * for small pages, there is no difference between | |
1522 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1523 | * for huge pages, since vm_pgoff is in units of small | |
1524 | * pages, we need to shift off the always 0 bits to get | |
1525 | * a useful offset. | |
1526 | */ | |
1527 | BUG_ON(shift < PAGE_SHIFT); | |
1528 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1529 | off += (addr - vma->vm_start) >> shift; |
1530 | return offset_il_node(pol, vma, off); | |
1531 | } else | |
1532 | return interleave_nodes(pol); | |
1533 | } | |
1534 | ||
00ac59ad | 1535 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1536 | /* |
1537 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1538 | * @vma = virtual memory area whose policy is sought | |
1539 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1540 | * @gfp_flags = for requested zone | |
19770b32 MG |
1541 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1542 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1543 | * |
52cd3b07 LS |
1544 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1545 | * to the struct mempolicy for conditional unref after allocation. | |
1546 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1547 | * @nodemask for filtering the zonelist. | |
480eccf9 | 1548 | */ |
396faf03 | 1549 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1550 | gfp_t gfp_flags, struct mempolicy **mpol, |
1551 | nodemask_t **nodemask) | |
5da7ca86 | 1552 | { |
480eccf9 | 1553 | struct zonelist *zl; |
5da7ca86 | 1554 | |
52cd3b07 | 1555 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1556 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1557 | |
52cd3b07 LS |
1558 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1559 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1560 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 LS |
1561 | } else { |
1562 | zl = policy_zonelist(gfp_flags, *mpol); | |
1563 | if ((*mpol)->mode == MPOL_BIND) | |
1564 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1565 | } |
1566 | return zl; | |
5da7ca86 | 1567 | } |
00ac59ad | 1568 | #endif |
5da7ca86 | 1569 | |
1da177e4 LT |
1570 | /* Allocate a page in interleaved policy. |
1571 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1572 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1573 | unsigned nid) | |
1da177e4 LT |
1574 | { |
1575 | struct zonelist *zl; | |
1576 | struct page *page; | |
1577 | ||
0e88460d | 1578 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1579 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1580 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1581 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1582 | return page; |
1583 | } | |
1584 | ||
1585 | /** | |
1586 | * alloc_page_vma - Allocate a page for a VMA. | |
1587 | * | |
1588 | * @gfp: | |
1589 | * %GFP_USER user allocation. | |
1590 | * %GFP_KERNEL kernel allocations, | |
1591 | * %GFP_HIGHMEM highmem/user allocations, | |
1592 | * %GFP_FS allocation should not call back into a file system. | |
1593 | * %GFP_ATOMIC don't sleep. | |
1594 | * | |
1595 | * @vma: Pointer to VMA or NULL if not available. | |
1596 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1597 | * | |
1598 | * This function allocates a page from the kernel page pool and applies | |
1599 | * a NUMA policy associated with the VMA or the current process. | |
1600 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1601 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1602 | * all allocations for pages that will be mapped into | |
1603 | * user space. Returns NULL when no page can be allocated. | |
1604 | * | |
1605 | * Should be called with the mm_sem of the vma hold. | |
1606 | */ | |
1607 | struct page * | |
dd0fc66f | 1608 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1609 | { |
6e21c8f1 | 1610 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1611 | struct zonelist *zl; |
1da177e4 | 1612 | |
45c4745a | 1613 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1614 | unsigned nid; |
5da7ca86 CL |
1615 | |
1616 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
52cd3b07 | 1617 | mpol_cond_put(pol); |
1da177e4 LT |
1618 | return alloc_page_interleave(gfp, 0, nid); |
1619 | } | |
52cd3b07 LS |
1620 | zl = policy_zonelist(gfp, pol); |
1621 | if (unlikely(mpol_needs_cond_ref(pol))) { | |
480eccf9 | 1622 | /* |
52cd3b07 | 1623 | * slow path: ref counted shared policy |
480eccf9 | 1624 | */ |
19770b32 | 1625 | struct page *page = __alloc_pages_nodemask(gfp, 0, |
52cd3b07 | 1626 | zl, policy_nodemask(gfp, pol)); |
f0be3d32 | 1627 | __mpol_put(pol); |
480eccf9 LS |
1628 | return page; |
1629 | } | |
1630 | /* | |
1631 | * fast path: default or task policy | |
1632 | */ | |
52cd3b07 | 1633 | return __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol)); |
1da177e4 LT |
1634 | } |
1635 | ||
1636 | /** | |
1637 | * alloc_pages_current - Allocate pages. | |
1638 | * | |
1639 | * @gfp: | |
1640 | * %GFP_USER user allocation, | |
1641 | * %GFP_KERNEL kernel allocation, | |
1642 | * %GFP_HIGHMEM highmem allocation, | |
1643 | * %GFP_FS don't call back into a file system. | |
1644 | * %GFP_ATOMIC don't sleep. | |
1645 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1646 | * | |
1647 | * Allocate a page from the kernel page pool. When not in | |
1648 | * interrupt context and apply the current process NUMA policy. | |
1649 | * Returns NULL when no page can be allocated. | |
1650 | * | |
cf2a473c | 1651 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1652 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1653 | * 2) allocating for current task (not interrupt). | |
1654 | */ | |
dd0fc66f | 1655 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1656 | { |
1657 | struct mempolicy *pol = current->mempolicy; | |
1658 | ||
9b819d20 | 1659 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 1660 | pol = &default_policy; |
52cd3b07 LS |
1661 | |
1662 | /* | |
1663 | * No reference counting needed for current->mempolicy | |
1664 | * nor system default_policy | |
1665 | */ | |
45c4745a | 1666 | if (pol->mode == MPOL_INTERLEAVE) |
1da177e4 | 1667 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
19770b32 | 1668 | return __alloc_pages_nodemask(gfp, order, |
52cd3b07 | 1669 | policy_zonelist(gfp, pol), policy_nodemask(gfp, pol)); |
1da177e4 LT |
1670 | } |
1671 | EXPORT_SYMBOL(alloc_pages_current); | |
1672 | ||
4225399a | 1673 | /* |
846a16bf | 1674 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
1675 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
1676 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1677 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1678 | * further kernel/cpuset.c update_nodemask(). | |
1679 | */ | |
4225399a | 1680 | |
846a16bf LS |
1681 | /* Slow path of a mempolicy duplicate */ |
1682 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
1683 | { |
1684 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1685 | ||
1686 | if (!new) | |
1687 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1688 | if (current_cpuset_is_being_rebound()) { |
1689 | nodemask_t mems = cpuset_mems_allowed(current); | |
1690 | mpol_rebind_policy(old, &mems); | |
1691 | } | |
1da177e4 LT |
1692 | *new = *old; |
1693 | atomic_set(&new->refcnt, 1); | |
1da177e4 LT |
1694 | return new; |
1695 | } | |
1696 | ||
52cd3b07 LS |
1697 | /* |
1698 | * If *frompol needs [has] an extra ref, copy *frompol to *tompol , | |
1699 | * eliminate the * MPOL_F_* flags that require conditional ref and | |
1700 | * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly | |
1701 | * after return. Use the returned value. | |
1702 | * | |
1703 | * Allows use of a mempolicy for, e.g., multiple allocations with a single | |
1704 | * policy lookup, even if the policy needs/has extra ref on lookup. | |
1705 | * shmem_readahead needs this. | |
1706 | */ | |
1707 | struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
1708 | struct mempolicy *frompol) | |
1709 | { | |
1710 | if (!mpol_needs_cond_ref(frompol)) | |
1711 | return frompol; | |
1712 | ||
1713 | *tompol = *frompol; | |
1714 | tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */ | |
1715 | __mpol_put(frompol); | |
1716 | return tompol; | |
1717 | } | |
1718 | ||
f5b087b5 DR |
1719 | static int mpol_match_intent(const struct mempolicy *a, |
1720 | const struct mempolicy *b) | |
1721 | { | |
1722 | if (a->flags != b->flags) | |
1723 | return 0; | |
1724 | if (!mpol_store_user_nodemask(a)) | |
1725 | return 1; | |
1726 | return nodes_equal(a->w.user_nodemask, b->w.user_nodemask); | |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | /* Slow path of a mempolicy comparison */ |
1730 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1731 | { | |
1732 | if (!a || !b) | |
1733 | return 0; | |
45c4745a | 1734 | if (a->mode != b->mode) |
1da177e4 | 1735 | return 0; |
45c4745a | 1736 | if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b)) |
f5b087b5 | 1737 | return 0; |
45c4745a | 1738 | switch (a->mode) { |
19770b32 MG |
1739 | case MPOL_BIND: |
1740 | /* Fall through */ | |
1da177e4 | 1741 | case MPOL_INTERLEAVE: |
dfcd3c0d | 1742 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 1743 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1744 | return a->v.preferred_node == b->v.preferred_node && |
1745 | a->flags == b->flags; | |
1da177e4 LT |
1746 | default: |
1747 | BUG(); | |
1748 | return 0; | |
1749 | } | |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | /* |
1753 | * Shared memory backing store policy support. | |
1754 | * | |
1755 | * Remember policies even when nobody has shared memory mapped. | |
1756 | * The policies are kept in Red-Black tree linked from the inode. | |
1757 | * They are protected by the sp->lock spinlock, which should be held | |
1758 | * for any accesses to the tree. | |
1759 | */ | |
1760 | ||
1761 | /* lookup first element intersecting start-end */ | |
1762 | /* Caller holds sp->lock */ | |
1763 | static struct sp_node * | |
1764 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1765 | { | |
1766 | struct rb_node *n = sp->root.rb_node; | |
1767 | ||
1768 | while (n) { | |
1769 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1770 | ||
1771 | if (start >= p->end) | |
1772 | n = n->rb_right; | |
1773 | else if (end <= p->start) | |
1774 | n = n->rb_left; | |
1775 | else | |
1776 | break; | |
1777 | } | |
1778 | if (!n) | |
1779 | return NULL; | |
1780 | for (;;) { | |
1781 | struct sp_node *w = NULL; | |
1782 | struct rb_node *prev = rb_prev(n); | |
1783 | if (!prev) | |
1784 | break; | |
1785 | w = rb_entry(prev, struct sp_node, nd); | |
1786 | if (w->end <= start) | |
1787 | break; | |
1788 | n = prev; | |
1789 | } | |
1790 | return rb_entry(n, struct sp_node, nd); | |
1791 | } | |
1792 | ||
1793 | /* Insert a new shared policy into the list. */ | |
1794 | /* Caller holds sp->lock */ | |
1795 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1796 | { | |
1797 | struct rb_node **p = &sp->root.rb_node; | |
1798 | struct rb_node *parent = NULL; | |
1799 | struct sp_node *nd; | |
1800 | ||
1801 | while (*p) { | |
1802 | parent = *p; | |
1803 | nd = rb_entry(parent, struct sp_node, nd); | |
1804 | if (new->start < nd->start) | |
1805 | p = &(*p)->rb_left; | |
1806 | else if (new->end > nd->end) | |
1807 | p = &(*p)->rb_right; | |
1808 | else | |
1809 | BUG(); | |
1810 | } | |
1811 | rb_link_node(&new->nd, parent, p); | |
1812 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1813 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 1814 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
1815 | } |
1816 | ||
1817 | /* Find shared policy intersecting idx */ | |
1818 | struct mempolicy * | |
1819 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1820 | { | |
1821 | struct mempolicy *pol = NULL; | |
1822 | struct sp_node *sn; | |
1823 | ||
1824 | if (!sp->root.rb_node) | |
1825 | return NULL; | |
1826 | spin_lock(&sp->lock); | |
1827 | sn = sp_lookup(sp, idx, idx+1); | |
1828 | if (sn) { | |
1829 | mpol_get(sn->policy); | |
1830 | pol = sn->policy; | |
1831 | } | |
1832 | spin_unlock(&sp->lock); | |
1833 | return pol; | |
1834 | } | |
1835 | ||
1836 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1837 | { | |
140d5a49 | 1838 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 1839 | rb_erase(&n->nd, &sp->root); |
f0be3d32 | 1840 | mpol_put(n->policy); |
1da177e4 LT |
1841 | kmem_cache_free(sn_cache, n); |
1842 | } | |
1843 | ||
dbcb0f19 AB |
1844 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
1845 | struct mempolicy *pol) | |
1da177e4 LT |
1846 | { |
1847 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1848 | ||
1849 | if (!n) | |
1850 | return NULL; | |
1851 | n->start = start; | |
1852 | n->end = end; | |
1853 | mpol_get(pol); | |
aab0b102 | 1854 | pol->flags |= MPOL_F_SHARED; /* for unref */ |
1da177e4 LT |
1855 | n->policy = pol; |
1856 | return n; | |
1857 | } | |
1858 | ||
1859 | /* Replace a policy range. */ | |
1860 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1861 | unsigned long end, struct sp_node *new) | |
1862 | { | |
1863 | struct sp_node *n, *new2 = NULL; | |
1864 | ||
1865 | restart: | |
1866 | spin_lock(&sp->lock); | |
1867 | n = sp_lookup(sp, start, end); | |
1868 | /* Take care of old policies in the same range. */ | |
1869 | while (n && n->start < end) { | |
1870 | struct rb_node *next = rb_next(&n->nd); | |
1871 | if (n->start >= start) { | |
1872 | if (n->end <= end) | |
1873 | sp_delete(sp, n); | |
1874 | else | |
1875 | n->start = end; | |
1876 | } else { | |
1877 | /* Old policy spanning whole new range. */ | |
1878 | if (n->end > end) { | |
1879 | if (!new2) { | |
1880 | spin_unlock(&sp->lock); | |
1881 | new2 = sp_alloc(end, n->end, n->policy); | |
1882 | if (!new2) | |
1883 | return -ENOMEM; | |
1884 | goto restart; | |
1885 | } | |
1886 | n->end = start; | |
1887 | sp_insert(sp, new2); | |
1888 | new2 = NULL; | |
1889 | break; | |
1890 | } else | |
1891 | n->end = start; | |
1892 | } | |
1893 | if (!next) | |
1894 | break; | |
1895 | n = rb_entry(next, struct sp_node, nd); | |
1896 | } | |
1897 | if (new) | |
1898 | sp_insert(sp, new); | |
1899 | spin_unlock(&sp->lock); | |
1900 | if (new2) { | |
f0be3d32 | 1901 | mpol_put(new2->policy); |
1da177e4 LT |
1902 | kmem_cache_free(sn_cache, new2); |
1903 | } | |
1904 | return 0; | |
1905 | } | |
1906 | ||
71fe804b LS |
1907 | /** |
1908 | * mpol_shared_policy_init - initialize shared policy for inode | |
1909 | * @sp: pointer to inode shared policy | |
1910 | * @mpol: struct mempolicy to install | |
1911 | * | |
1912 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
1913 | * On entry, the current task has a reference on a non-NULL @mpol. | |
1914 | * This must be released on exit. | |
4bfc4495 | 1915 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
1916 | */ |
1917 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
1918 | { | |
58568d2a MX |
1919 | int ret; |
1920 | ||
71fe804b LS |
1921 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
1922 | spin_lock_init(&sp->lock); | |
1923 | ||
1924 | if (mpol) { | |
1925 | struct vm_area_struct pvma; | |
1926 | struct mempolicy *new; | |
4bfc4495 | 1927 | NODEMASK_SCRATCH(scratch); |
71fe804b | 1928 | |
4bfc4495 KH |
1929 | if (!scratch) |
1930 | return; | |
71fe804b LS |
1931 | /* contextualize the tmpfs mount point mempolicy */ |
1932 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
58568d2a MX |
1933 | if (IS_ERR(new)) { |
1934 | mpol_put(mpol); /* drop our ref on sb mpol */ | |
4bfc4495 | 1935 | NODEMASK_SCRATCH_FREE(scratch); |
71fe804b | 1936 | return; /* no valid nodemask intersection */ |
58568d2a MX |
1937 | } |
1938 | ||
1939 | task_lock(current); | |
4bfc4495 | 1940 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a MX |
1941 | task_unlock(current); |
1942 | mpol_put(mpol); /* drop our ref on sb mpol */ | |
1943 | if (ret) { | |
4bfc4495 | 1944 | NODEMASK_SCRATCH_FREE(scratch); |
58568d2a MX |
1945 | mpol_put(new); |
1946 | return; | |
1947 | } | |
71fe804b LS |
1948 | |
1949 | /* Create pseudo-vma that contains just the policy */ | |
1950 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1951 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
1952 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
1953 | mpol_put(new); /* drop initial ref */ | |
4bfc4495 | 1954 | NODEMASK_SCRATCH_FREE(scratch); |
7339ff83 RH |
1955 | } |
1956 | } | |
1957 | ||
1da177e4 LT |
1958 | int mpol_set_shared_policy(struct shared_policy *info, |
1959 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1960 | { | |
1961 | int err; | |
1962 | struct sp_node *new = NULL; | |
1963 | unsigned long sz = vma_pages(vma); | |
1964 | ||
028fec41 | 1965 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 1966 | vma->vm_pgoff, |
45c4745a | 1967 | sz, npol ? npol->mode : -1, |
028fec41 | 1968 | npol ? npol->flags : -1, |
140d5a49 | 1969 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1970 | |
1971 | if (npol) { | |
1972 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1973 | if (!new) | |
1974 | return -ENOMEM; | |
1975 | } | |
1976 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1977 | if (err && new) | |
1978 | kmem_cache_free(sn_cache, new); | |
1979 | return err; | |
1980 | } | |
1981 | ||
1982 | /* Free a backing policy store on inode delete. */ | |
1983 | void mpol_free_shared_policy(struct shared_policy *p) | |
1984 | { | |
1985 | struct sp_node *n; | |
1986 | struct rb_node *next; | |
1987 | ||
1988 | if (!p->root.rb_node) | |
1989 | return; | |
1990 | spin_lock(&p->lock); | |
1991 | next = rb_first(&p->root); | |
1992 | while (next) { | |
1993 | n = rb_entry(next, struct sp_node, nd); | |
1994 | next = rb_next(&n->nd); | |
90c5029e | 1995 | rb_erase(&n->nd, &p->root); |
f0be3d32 | 1996 | mpol_put(n->policy); |
1da177e4 LT |
1997 | kmem_cache_free(sn_cache, n); |
1998 | } | |
1999 | spin_unlock(&p->lock); | |
1da177e4 LT |
2000 | } |
2001 | ||
2002 | /* assumes fs == KERNEL_DS */ | |
2003 | void __init numa_policy_init(void) | |
2004 | { | |
b71636e2 PM |
2005 | nodemask_t interleave_nodes; |
2006 | unsigned long largest = 0; | |
2007 | int nid, prefer = 0; | |
2008 | ||
1da177e4 LT |
2009 | policy_cache = kmem_cache_create("numa_policy", |
2010 | sizeof(struct mempolicy), | |
20c2df83 | 2011 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2012 | |
2013 | sn_cache = kmem_cache_create("shared_policy_node", | |
2014 | sizeof(struct sp_node), | |
20c2df83 | 2015 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2016 | |
b71636e2 PM |
2017 | /* |
2018 | * Set interleaving policy for system init. Interleaving is only | |
2019 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2020 | * fall back to the largest node if they're all smaller. | |
2021 | */ | |
2022 | nodes_clear(interleave_nodes); | |
56bbd65d | 2023 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
2024 | unsigned long total_pages = node_present_pages(nid); |
2025 | ||
2026 | /* Preserve the largest node */ | |
2027 | if (largest < total_pages) { | |
2028 | largest = total_pages; | |
2029 | prefer = nid; | |
2030 | } | |
2031 | ||
2032 | /* Interleave this node? */ | |
2033 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2034 | node_set(nid, interleave_nodes); | |
2035 | } | |
2036 | ||
2037 | /* All too small, use the largest */ | |
2038 | if (unlikely(nodes_empty(interleave_nodes))) | |
2039 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2040 | |
028fec41 | 2041 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
2042 | printk("numa_policy_init: interleaving failed\n"); |
2043 | } | |
2044 | ||
8bccd85f | 2045 | /* Reset policy of current process to default */ |
1da177e4 LT |
2046 | void numa_default_policy(void) |
2047 | { | |
028fec41 | 2048 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2049 | } |
68860ec1 | 2050 | |
095f1fc4 LS |
2051 | /* |
2052 | * Parse and format mempolicy from/to strings | |
2053 | */ | |
2054 | ||
1a75a6c8 | 2055 | /* |
fc36b8d3 | 2056 | * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag |
3f226aa1 | 2057 | * Used only for mpol_parse_str() and mpol_to_str() |
1a75a6c8 | 2058 | */ |
53f2556b | 2059 | #define MPOL_LOCAL (MPOL_INTERLEAVE + 1) |
15ad7cdc | 2060 | static const char * const policy_types[] = |
53f2556b | 2061 | { "default", "prefer", "bind", "interleave", "local" }; |
1a75a6c8 | 2062 | |
095f1fc4 LS |
2063 | |
2064 | #ifdef CONFIG_TMPFS | |
2065 | /** | |
2066 | * mpol_parse_str - parse string to mempolicy | |
2067 | * @str: string containing mempolicy to parse | |
71fe804b LS |
2068 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
2069 | * @no_context: flag whether to "contextualize" the mempolicy | |
095f1fc4 LS |
2070 | * |
2071 | * Format of input: | |
2072 | * <mode>[=<flags>][:<nodelist>] | |
2073 | * | |
71fe804b LS |
2074 | * if @no_context is true, save the input nodemask in w.user_nodemask in |
2075 | * the returned mempolicy. This will be used to "clone" the mempolicy in | |
2076 | * a specific context [cpuset] at a later time. Used to parse tmpfs mpol | |
2077 | * mount option. Note that if 'static' or 'relative' mode flags were | |
2078 | * specified, the input nodemask will already have been saved. Saving | |
2079 | * it again is redundant, but safe. | |
2080 | * | |
2081 | * On success, returns 0, else 1 | |
095f1fc4 | 2082 | */ |
71fe804b | 2083 | int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context) |
095f1fc4 | 2084 | { |
71fe804b LS |
2085 | struct mempolicy *new = NULL; |
2086 | unsigned short uninitialized_var(mode); | |
2087 | unsigned short uninitialized_var(mode_flags); | |
2088 | nodemask_t nodes; | |
095f1fc4 LS |
2089 | char *nodelist = strchr(str, ':'); |
2090 | char *flags = strchr(str, '='); | |
2091 | int i; | |
2092 | int err = 1; | |
2093 | ||
2094 | if (nodelist) { | |
2095 | /* NUL-terminate mode or flags string */ | |
2096 | *nodelist++ = '\0'; | |
71fe804b | 2097 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2098 | goto out; |
71fe804b | 2099 | if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY])) |
095f1fc4 | 2100 | goto out; |
71fe804b LS |
2101 | } else |
2102 | nodes_clear(nodes); | |
2103 | ||
095f1fc4 LS |
2104 | if (flags) |
2105 | *flags++ = '\0'; /* terminate mode string */ | |
2106 | ||
3f226aa1 | 2107 | for (i = 0; i <= MPOL_LOCAL; i++) { |
095f1fc4 | 2108 | if (!strcmp(str, policy_types[i])) { |
71fe804b | 2109 | mode = i; |
095f1fc4 LS |
2110 | break; |
2111 | } | |
2112 | } | |
3f226aa1 | 2113 | if (i > MPOL_LOCAL) |
095f1fc4 LS |
2114 | goto out; |
2115 | ||
71fe804b | 2116 | switch (mode) { |
095f1fc4 | 2117 | case MPOL_PREFERRED: |
71fe804b LS |
2118 | /* |
2119 | * Insist on a nodelist of one node only | |
2120 | */ | |
095f1fc4 LS |
2121 | if (nodelist) { |
2122 | char *rest = nodelist; | |
2123 | while (isdigit(*rest)) | |
2124 | rest++; | |
2125 | if (!*rest) | |
2126 | err = 0; | |
2127 | } | |
2128 | break; | |
095f1fc4 LS |
2129 | case MPOL_INTERLEAVE: |
2130 | /* | |
2131 | * Default to online nodes with memory if no nodelist | |
2132 | */ | |
2133 | if (!nodelist) | |
71fe804b | 2134 | nodes = node_states[N_HIGH_MEMORY]; |
095f1fc4 | 2135 | err = 0; |
3f226aa1 | 2136 | break; |
71fe804b | 2137 | case MPOL_LOCAL: |
3f226aa1 | 2138 | /* |
71fe804b | 2139 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2140 | */ |
71fe804b | 2141 | if (nodelist) |
3f226aa1 | 2142 | goto out; |
71fe804b | 2143 | mode = MPOL_PREFERRED; |
3f226aa1 | 2144 | break; |
71fe804b LS |
2145 | |
2146 | /* | |
2147 | * case MPOL_BIND: mpol_new() enforces non-empty nodemask. | |
2148 | * case MPOL_DEFAULT: mpol_new() enforces empty nodemask, ignores flags. | |
2149 | */ | |
095f1fc4 LS |
2150 | } |
2151 | ||
71fe804b | 2152 | mode_flags = 0; |
095f1fc4 LS |
2153 | if (flags) { |
2154 | /* | |
2155 | * Currently, we only support two mutually exclusive | |
2156 | * mode flags. | |
2157 | */ | |
2158 | if (!strcmp(flags, "static")) | |
71fe804b | 2159 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2160 | else if (!strcmp(flags, "relative")) |
71fe804b | 2161 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 LS |
2162 | else |
2163 | err = 1; | |
2164 | } | |
71fe804b LS |
2165 | |
2166 | new = mpol_new(mode, mode_flags, &nodes); | |
2167 | if (IS_ERR(new)) | |
2168 | err = 1; | |
58568d2a MX |
2169 | else { |
2170 | int ret; | |
4bfc4495 KH |
2171 | NODEMASK_SCRATCH(scratch); |
2172 | if (scratch) { | |
2173 | task_lock(current); | |
2174 | ret = mpol_set_nodemask(new, &nodes, scratch); | |
2175 | task_unlock(current); | |
2176 | } else | |
2177 | ret = -ENOMEM; | |
2178 | NODEMASK_SCRATCH_FREE(scratch); | |
2179 | if (ret) { | |
58568d2a | 2180 | err = 1; |
4bfc4495 KH |
2181 | mpol_put(new); |
2182 | } else if (no_context) { | |
58568d2a MX |
2183 | /* save for contextualization */ |
2184 | new->w.user_nodemask = nodes; | |
2185 | } | |
2186 | } | |
71fe804b | 2187 | |
095f1fc4 LS |
2188 | out: |
2189 | /* Restore string for error message */ | |
2190 | if (nodelist) | |
2191 | *--nodelist = ':'; | |
2192 | if (flags) | |
2193 | *--flags = '='; | |
71fe804b LS |
2194 | if (!err) |
2195 | *mpol = new; | |
095f1fc4 LS |
2196 | return err; |
2197 | } | |
2198 | #endif /* CONFIG_TMPFS */ | |
2199 | ||
71fe804b LS |
2200 | /** |
2201 | * mpol_to_str - format a mempolicy structure for printing | |
2202 | * @buffer: to contain formatted mempolicy string | |
2203 | * @maxlen: length of @buffer | |
2204 | * @pol: pointer to mempolicy to be formatted | |
2205 | * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask | |
2206 | * | |
1a75a6c8 CL |
2207 | * Convert a mempolicy into a string. |
2208 | * Returns the number of characters in buffer (if positive) | |
2209 | * or an error (negative) | |
2210 | */ | |
71fe804b | 2211 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context) |
1a75a6c8 CL |
2212 | { |
2213 | char *p = buffer; | |
2214 | int l; | |
2215 | nodemask_t nodes; | |
bea904d5 | 2216 | unsigned short mode; |
f5b087b5 | 2217 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 2218 | |
2291990a LS |
2219 | /* |
2220 | * Sanity check: room for longest mode, flag and some nodes | |
2221 | */ | |
2222 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); | |
2223 | ||
bea904d5 LS |
2224 | if (!pol || pol == &default_policy) |
2225 | mode = MPOL_DEFAULT; | |
2226 | else | |
2227 | mode = pol->mode; | |
2228 | ||
1a75a6c8 CL |
2229 | switch (mode) { |
2230 | case MPOL_DEFAULT: | |
2231 | nodes_clear(nodes); | |
2232 | break; | |
2233 | ||
2234 | case MPOL_PREFERRED: | |
2235 | nodes_clear(nodes); | |
fc36b8d3 | 2236 | if (flags & MPOL_F_LOCAL) |
53f2556b LS |
2237 | mode = MPOL_LOCAL; /* pseudo-policy */ |
2238 | else | |
fc36b8d3 | 2239 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
2240 | break; |
2241 | ||
2242 | case MPOL_BIND: | |
19770b32 | 2243 | /* Fall through */ |
1a75a6c8 | 2244 | case MPOL_INTERLEAVE: |
71fe804b LS |
2245 | if (no_context) |
2246 | nodes = pol->w.user_nodemask; | |
2247 | else | |
2248 | nodes = pol->v.nodes; | |
1a75a6c8 CL |
2249 | break; |
2250 | ||
2251 | default: | |
2252 | BUG(); | |
1a75a6c8 CL |
2253 | } |
2254 | ||
2255 | l = strlen(policy_types[mode]); | |
53f2556b LS |
2256 | if (buffer + maxlen < p + l + 1) |
2257 | return -ENOSPC; | |
1a75a6c8 CL |
2258 | |
2259 | strcpy(p, policy_types[mode]); | |
2260 | p += l; | |
2261 | ||
fc36b8d3 | 2262 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2263 | if (buffer + maxlen < p + 2) |
2264 | return -ENOSPC; | |
2265 | *p++ = '='; | |
2266 | ||
2291990a LS |
2267 | /* |
2268 | * Currently, the only defined flags are mutually exclusive | |
2269 | */ | |
f5b087b5 | 2270 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2271 | p += snprintf(p, buffer + maxlen - p, "static"); |
2272 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2273 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2274 | } |
2275 | ||
1a75a6c8 CL |
2276 | if (!nodes_empty(nodes)) { |
2277 | if (buffer + maxlen < p + 2) | |
2278 | return -ENOSPC; | |
095f1fc4 | 2279 | *p++ = ':'; |
1a75a6c8 CL |
2280 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2281 | } | |
2282 | return p - buffer; | |
2283 | } | |
2284 | ||
2285 | struct numa_maps { | |
2286 | unsigned long pages; | |
2287 | unsigned long anon; | |
397874df CL |
2288 | unsigned long active; |
2289 | unsigned long writeback; | |
1a75a6c8 | 2290 | unsigned long mapcount_max; |
397874df CL |
2291 | unsigned long dirty; |
2292 | unsigned long swapcache; | |
1a75a6c8 CL |
2293 | unsigned long node[MAX_NUMNODES]; |
2294 | }; | |
2295 | ||
397874df | 2296 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
2297 | { |
2298 | struct numa_maps *md = private; | |
2299 | int count = page_mapcount(page); | |
2300 | ||
397874df CL |
2301 | md->pages++; |
2302 | if (pte_dirty || PageDirty(page)) | |
2303 | md->dirty++; | |
1a75a6c8 | 2304 | |
397874df CL |
2305 | if (PageSwapCache(page)) |
2306 | md->swapcache++; | |
1a75a6c8 | 2307 | |
894bc310 | 2308 | if (PageActive(page) || PageUnevictable(page)) |
397874df CL |
2309 | md->active++; |
2310 | ||
2311 | if (PageWriteback(page)) | |
2312 | md->writeback++; | |
1a75a6c8 CL |
2313 | |
2314 | if (PageAnon(page)) | |
2315 | md->anon++; | |
2316 | ||
397874df CL |
2317 | if (count > md->mapcount_max) |
2318 | md->mapcount_max = count; | |
2319 | ||
1a75a6c8 | 2320 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
2321 | } |
2322 | ||
7f709ed0 | 2323 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
2324 | static void check_huge_range(struct vm_area_struct *vma, |
2325 | unsigned long start, unsigned long end, | |
2326 | struct numa_maps *md) | |
2327 | { | |
2328 | unsigned long addr; | |
2329 | struct page *page; | |
a5516438 AK |
2330 | struct hstate *h = hstate_vma(vma); |
2331 | unsigned long sz = huge_page_size(h); | |
397874df | 2332 | |
a5516438 AK |
2333 | for (addr = start; addr < end; addr += sz) { |
2334 | pte_t *ptep = huge_pte_offset(vma->vm_mm, | |
2335 | addr & huge_page_mask(h)); | |
397874df CL |
2336 | pte_t pte; |
2337 | ||
2338 | if (!ptep) | |
2339 | continue; | |
2340 | ||
2341 | pte = *ptep; | |
2342 | if (pte_none(pte)) | |
2343 | continue; | |
2344 | ||
2345 | page = pte_page(pte); | |
2346 | if (!page) | |
2347 | continue; | |
2348 | ||
2349 | gather_stats(page, md, pte_dirty(*ptep)); | |
2350 | } | |
2351 | } | |
7f709ed0 AM |
2352 | #else |
2353 | static inline void check_huge_range(struct vm_area_struct *vma, | |
2354 | unsigned long start, unsigned long end, | |
2355 | struct numa_maps *md) | |
2356 | { | |
2357 | } | |
2358 | #endif | |
397874df | 2359 | |
53f2556b LS |
2360 | /* |
2361 | * Display pages allocated per node and memory policy via /proc. | |
2362 | */ | |
1a75a6c8 CL |
2363 | int show_numa_map(struct seq_file *m, void *v) |
2364 | { | |
99f89551 | 2365 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
2366 | struct vm_area_struct *vma = v; |
2367 | struct numa_maps *md; | |
397874df CL |
2368 | struct file *file = vma->vm_file; |
2369 | struct mm_struct *mm = vma->vm_mm; | |
480eccf9 | 2370 | struct mempolicy *pol; |
1a75a6c8 CL |
2371 | int n; |
2372 | char buffer[50]; | |
2373 | ||
397874df | 2374 | if (!mm) |
1a75a6c8 CL |
2375 | return 0; |
2376 | ||
2377 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
2378 | if (!md) | |
2379 | return 0; | |
2380 | ||
480eccf9 | 2381 | pol = get_vma_policy(priv->task, vma, vma->vm_start); |
71fe804b | 2382 | mpol_to_str(buffer, sizeof(buffer), pol, 0); |
52cd3b07 | 2383 | mpol_cond_put(pol); |
397874df CL |
2384 | |
2385 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
2386 | ||
2387 | if (file) { | |
2388 | seq_printf(m, " file="); | |
c32c2f63 | 2389 | seq_path(m, &file->f_path, "\n\t= "); |
397874df CL |
2390 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
2391 | seq_printf(m, " heap"); | |
2392 | } else if (vma->vm_start <= mm->start_stack && | |
2393 | vma->vm_end >= mm->start_stack) { | |
2394 | seq_printf(m, " stack"); | |
2395 | } | |
2396 | ||
2397 | if (is_vm_hugetlb_page(vma)) { | |
2398 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
2399 | seq_printf(m, " huge"); | |
2400 | } else { | |
a57ebfdb | 2401 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
56bbd65d | 2402 | &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md); |
397874df CL |
2403 | } |
2404 | ||
2405 | if (!md->pages) | |
2406 | goto out; | |
1a75a6c8 | 2407 | |
397874df CL |
2408 | if (md->anon) |
2409 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 2410 | |
397874df CL |
2411 | if (md->dirty) |
2412 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 2413 | |
397874df CL |
2414 | if (md->pages != md->anon && md->pages != md->dirty) |
2415 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 2416 | |
397874df CL |
2417 | if (md->mapcount_max > 1) |
2418 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 2419 | |
397874df CL |
2420 | if (md->swapcache) |
2421 | seq_printf(m," swapcache=%lu", md->swapcache); | |
2422 | ||
2423 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
2424 | seq_printf(m," active=%lu", md->active); | |
2425 | ||
2426 | if (md->writeback) | |
2427 | seq_printf(m," writeback=%lu", md->writeback); | |
2428 | ||
56bbd65d | 2429 | for_each_node_state(n, N_HIGH_MEMORY) |
397874df CL |
2430 | if (md->node[n]) |
2431 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
2432 | out: | |
2433 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
2434 | kfree(md); |
2435 | ||
2436 | if (m->count < m->size) | |
99f89551 | 2437 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
2438 | return 0; |
2439 | } |