]>
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. | |
66 | could replace all the switch()es with a mempolicy_ops structure. | |
67 | */ | |
68 | ||
69 | #include <linux/mempolicy.h> | |
70 | #include <linux/mm.h> | |
71 | #include <linux/highmem.h> | |
72 | #include <linux/hugetlb.h> | |
73 | #include <linux/kernel.h> | |
74 | #include <linux/sched.h> | |
75 | #include <linux/mm.h> | |
76 | #include <linux/nodemask.h> | |
77 | #include <linux/cpuset.h> | |
78 | #include <linux/gfp.h> | |
79 | #include <linux/slab.h> | |
80 | #include <linux/string.h> | |
81 | #include <linux/module.h> | |
82 | #include <linux/interrupt.h> | |
83 | #include <linux/init.h> | |
84 | #include <linux/compat.h> | |
85 | #include <linux/mempolicy.h> | |
dc9aa5b9 | 86 | #include <linux/swap.h> |
1a75a6c8 CL |
87 | #include <linux/seq_file.h> |
88 | #include <linux/proc_fs.h> | |
dc9aa5b9 | 89 | |
1da177e4 LT |
90 | #include <asm/tlbflush.h> |
91 | #include <asm/uaccess.h> | |
92 | ||
38e35860 | 93 | /* Internal flags */ |
dc9aa5b9 | 94 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 95 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
1a75a6c8 | 96 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
dc9aa5b9 | 97 | |
1da177e4 LT |
98 | static kmem_cache_t *policy_cache; |
99 | static kmem_cache_t *sn_cache; | |
100 | ||
101 | #define PDprintk(fmt...) | |
102 | ||
103 | /* Highest zone. An specific allocation for a zone below that is not | |
104 | policied. */ | |
4be38e35 | 105 | int policy_zone = ZONE_DMA; |
1da177e4 | 106 | |
d42c6997 | 107 | struct mempolicy default_policy = { |
1da177e4 LT |
108 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
109 | .policy = MPOL_DEFAULT, | |
110 | }; | |
111 | ||
1da177e4 | 112 | /* Do sanity checking on a policy */ |
dfcd3c0d | 113 | static int mpol_check_policy(int mode, nodemask_t *nodes) |
1da177e4 | 114 | { |
dfcd3c0d | 115 | int empty = nodes_empty(*nodes); |
1da177e4 LT |
116 | |
117 | switch (mode) { | |
118 | case MPOL_DEFAULT: | |
119 | if (!empty) | |
120 | return -EINVAL; | |
121 | break; | |
122 | case MPOL_BIND: | |
123 | case MPOL_INTERLEAVE: | |
124 | /* Preferred will only use the first bit, but allow | |
125 | more for now. */ | |
126 | if (empty) | |
127 | return -EINVAL; | |
128 | break; | |
129 | } | |
dfcd3c0d | 130 | return nodes_subset(*nodes, node_online_map) ? 0 : -EINVAL; |
1da177e4 | 131 | } |
1da177e4 | 132 | /* Generate a custom zonelist for the BIND policy. */ |
dfcd3c0d | 133 | static struct zonelist *bind_zonelist(nodemask_t *nodes) |
1da177e4 LT |
134 | { |
135 | struct zonelist *zl; | |
136 | int num, max, nd; | |
137 | ||
dfcd3c0d | 138 | max = 1 + MAX_NR_ZONES * nodes_weight(*nodes); |
1da177e4 LT |
139 | zl = kmalloc(sizeof(void *) * max, GFP_KERNEL); |
140 | if (!zl) | |
141 | return NULL; | |
142 | num = 0; | |
4be38e35 CL |
143 | for_each_node_mask(nd, *nodes) |
144 | zl->zones[num++] = &NODE_DATA(nd)->node_zones[policy_zone]; | |
1da177e4 LT |
145 | zl->zones[num] = NULL; |
146 | return zl; | |
147 | } | |
148 | ||
149 | /* Create a new policy */ | |
dfcd3c0d | 150 | static struct mempolicy *mpol_new(int mode, nodemask_t *nodes) |
1da177e4 LT |
151 | { |
152 | struct mempolicy *policy; | |
153 | ||
dfcd3c0d | 154 | PDprintk("setting mode %d nodes[0] %lx\n", mode, nodes_addr(*nodes)[0]); |
1da177e4 LT |
155 | if (mode == MPOL_DEFAULT) |
156 | return NULL; | |
157 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
158 | if (!policy) | |
159 | return ERR_PTR(-ENOMEM); | |
160 | atomic_set(&policy->refcnt, 1); | |
161 | switch (mode) { | |
162 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 163 | policy->v.nodes = *nodes; |
8f493d79 AK |
164 | if (nodes_weight(*nodes) == 0) { |
165 | kmem_cache_free(policy_cache, policy); | |
166 | return ERR_PTR(-EINVAL); | |
167 | } | |
1da177e4 LT |
168 | break; |
169 | case MPOL_PREFERRED: | |
dfcd3c0d | 170 | policy->v.preferred_node = first_node(*nodes); |
1da177e4 LT |
171 | if (policy->v.preferred_node >= MAX_NUMNODES) |
172 | policy->v.preferred_node = -1; | |
173 | break; | |
174 | case MPOL_BIND: | |
175 | policy->v.zonelist = bind_zonelist(nodes); | |
176 | if (policy->v.zonelist == NULL) { | |
177 | kmem_cache_free(policy_cache, policy); | |
178 | return ERR_PTR(-ENOMEM); | |
179 | } | |
180 | break; | |
181 | } | |
182 | policy->policy = mode; | |
74cb2155 | 183 | policy->cpuset_mems_allowed = cpuset_mems_allowed(current); |
1da177e4 LT |
184 | return policy; |
185 | } | |
186 | ||
1a75a6c8 | 187 | static void gather_stats(struct page *, void *); |
fc301289 CL |
188 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
189 | unsigned long flags); | |
1a75a6c8 | 190 | |
38e35860 | 191 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 192 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
193 | unsigned long addr, unsigned long end, |
194 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 195 | void *private) |
1da177e4 | 196 | { |
91612e0d HD |
197 | pte_t *orig_pte; |
198 | pte_t *pte; | |
705e87c0 | 199 | spinlock_t *ptl; |
941150a3 | 200 | |
705e87c0 | 201 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 202 | do { |
6aab341e | 203 | struct page *page; |
91612e0d HD |
204 | unsigned int nid; |
205 | ||
206 | if (!pte_present(*pte)) | |
1da177e4 | 207 | continue; |
6aab341e LT |
208 | page = vm_normal_page(vma, addr, *pte); |
209 | if (!page) | |
1da177e4 | 210 | continue; |
053837fc NP |
211 | /* |
212 | * The check for PageReserved here is important to avoid | |
213 | * handling zero pages and other pages that may have been | |
214 | * marked special by the system. | |
215 | * | |
216 | * If the PageReserved would not be checked here then f.e. | |
217 | * the location of the zero page could have an influence | |
218 | * on MPOL_MF_STRICT, zero pages would be counted for | |
219 | * the per node stats, and there would be useless attempts | |
220 | * to put zero pages on the migration list. | |
221 | */ | |
f4598c8b CL |
222 | if (PageReserved(page)) |
223 | continue; | |
6aab341e | 224 | nid = page_to_nid(page); |
38e35860 CL |
225 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
226 | continue; | |
227 | ||
1a75a6c8 CL |
228 | if (flags & MPOL_MF_STATS) |
229 | gather_stats(page, private); | |
053837fc | 230 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 231 | migrate_page_add(page, private, flags); |
38e35860 CL |
232 | else |
233 | break; | |
91612e0d | 234 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 235 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
236 | return addr != end; |
237 | } | |
238 | ||
b5810039 | 239 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
240 | unsigned long addr, unsigned long end, |
241 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 242 | void *private) |
91612e0d HD |
243 | { |
244 | pmd_t *pmd; | |
245 | unsigned long next; | |
246 | ||
247 | pmd = pmd_offset(pud, addr); | |
248 | do { | |
249 | next = pmd_addr_end(addr, end); | |
250 | if (pmd_none_or_clear_bad(pmd)) | |
251 | continue; | |
dc9aa5b9 | 252 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 253 | flags, private)) |
91612e0d HD |
254 | return -EIO; |
255 | } while (pmd++, addr = next, addr != end); | |
256 | return 0; | |
257 | } | |
258 | ||
b5810039 | 259 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
260 | unsigned long addr, unsigned long end, |
261 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 262 | void *private) |
91612e0d HD |
263 | { |
264 | pud_t *pud; | |
265 | unsigned long next; | |
266 | ||
267 | pud = pud_offset(pgd, addr); | |
268 | do { | |
269 | next = pud_addr_end(addr, end); | |
270 | if (pud_none_or_clear_bad(pud)) | |
271 | continue; | |
dc9aa5b9 | 272 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 273 | flags, private)) |
91612e0d HD |
274 | return -EIO; |
275 | } while (pud++, addr = next, addr != end); | |
276 | return 0; | |
277 | } | |
278 | ||
b5810039 | 279 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
280 | unsigned long addr, unsigned long end, |
281 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 282 | void *private) |
91612e0d HD |
283 | { |
284 | pgd_t *pgd; | |
285 | unsigned long next; | |
286 | ||
b5810039 | 287 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
288 | do { |
289 | next = pgd_addr_end(addr, end); | |
290 | if (pgd_none_or_clear_bad(pgd)) | |
291 | continue; | |
dc9aa5b9 | 292 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 293 | flags, private)) |
91612e0d HD |
294 | return -EIO; |
295 | } while (pgd++, addr = next, addr != end); | |
296 | return 0; | |
1da177e4 LT |
297 | } |
298 | ||
dc9aa5b9 CL |
299 | /* Check if a vma is migratable */ |
300 | static inline int vma_migratable(struct vm_area_struct *vma) | |
301 | { | |
302 | if (vma->vm_flags & ( | |
f4598c8b | 303 | VM_LOCKED|VM_IO|VM_HUGETLB|VM_PFNMAP|VM_RESERVED)) |
dc9aa5b9 CL |
304 | return 0; |
305 | return 1; | |
306 | } | |
307 | ||
308 | /* | |
309 | * Check if all pages in a range are on a set of nodes. | |
310 | * If pagelist != NULL then isolate pages from the LRU and | |
311 | * put them on the pagelist. | |
312 | */ | |
1da177e4 LT |
313 | static struct vm_area_struct * |
314 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 315 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
316 | { |
317 | int err; | |
318 | struct vm_area_struct *first, *vma, *prev; | |
319 | ||
053837fc NP |
320 | /* Clear the LRU lists so pages can be isolated */ |
321 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
322 | lru_add_drain_all(); | |
323 | ||
1da177e4 LT |
324 | first = find_vma(mm, start); |
325 | if (!first) | |
326 | return ERR_PTR(-EFAULT); | |
327 | prev = NULL; | |
328 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
329 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
330 | if (!vma->vm_next && vma->vm_end < end) | |
331 | return ERR_PTR(-EFAULT); | |
332 | if (prev && prev->vm_end < vma->vm_start) | |
333 | return ERR_PTR(-EFAULT); | |
334 | } | |
335 | if (!is_vm_hugetlb_page(vma) && | |
336 | ((flags & MPOL_MF_STRICT) || | |
337 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
338 | vma_migratable(vma)))) { | |
5b952b3c | 339 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 340 | |
5b952b3c AK |
341 | if (endvma > end) |
342 | endvma = end; | |
343 | if (vma->vm_start > start) | |
344 | start = vma->vm_start; | |
dc9aa5b9 | 345 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 346 | flags, private); |
1da177e4 LT |
347 | if (err) { |
348 | first = ERR_PTR(err); | |
349 | break; | |
350 | } | |
351 | } | |
352 | prev = vma; | |
353 | } | |
354 | return first; | |
355 | } | |
356 | ||
357 | /* Apply policy to a single VMA */ | |
358 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
359 | { | |
360 | int err = 0; | |
361 | struct mempolicy *old = vma->vm_policy; | |
362 | ||
363 | PDprintk("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
364 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
365 | vma->vm_ops, vma->vm_file, | |
366 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
367 | ||
368 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
369 | err = vma->vm_ops->set_policy(vma, new); | |
370 | if (!err) { | |
371 | mpol_get(new); | |
372 | vma->vm_policy = new; | |
373 | mpol_free(old); | |
374 | } | |
375 | return err; | |
376 | } | |
377 | ||
378 | /* Step 2: apply policy to a range and do splits. */ | |
379 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
380 | unsigned long end, struct mempolicy *new) | |
381 | { | |
382 | struct vm_area_struct *next; | |
383 | int err; | |
384 | ||
385 | err = 0; | |
386 | for (; vma && vma->vm_start < end; vma = next) { | |
387 | next = vma->vm_next; | |
388 | if (vma->vm_start < start) | |
389 | err = split_vma(vma->vm_mm, vma, start, 1); | |
390 | if (!err && vma->vm_end > end) | |
391 | err = split_vma(vma->vm_mm, vma, end, 0); | |
392 | if (!err) | |
393 | err = policy_vma(vma, new); | |
394 | if (err) | |
395 | break; | |
396 | } | |
397 | return err; | |
398 | } | |
399 | ||
8bccd85f CL |
400 | static int contextualize_policy(int mode, nodemask_t *nodes) |
401 | { | |
402 | if (!nodes) | |
403 | return 0; | |
404 | ||
cf2a473c | 405 | cpuset_update_task_memory_state(); |
5966514d PJ |
406 | if (!cpuset_nodes_subset_current_mems_allowed(*nodes)) |
407 | return -EINVAL; | |
8bccd85f CL |
408 | return mpol_check_policy(mode, nodes); |
409 | } | |
410 | ||
1da177e4 | 411 | /* Set the process memory policy */ |
8bccd85f | 412 | long do_set_mempolicy(int mode, nodemask_t *nodes) |
1da177e4 | 413 | { |
1da177e4 | 414 | struct mempolicy *new; |
1da177e4 | 415 | |
8bccd85f | 416 | if (contextualize_policy(mode, nodes)) |
1da177e4 | 417 | return -EINVAL; |
8bccd85f | 418 | new = mpol_new(mode, nodes); |
1da177e4 LT |
419 | if (IS_ERR(new)) |
420 | return PTR_ERR(new); | |
421 | mpol_free(current->mempolicy); | |
422 | current->mempolicy = new; | |
423 | if (new && new->policy == MPOL_INTERLEAVE) | |
dfcd3c0d | 424 | current->il_next = first_node(new->v.nodes); |
1da177e4 LT |
425 | return 0; |
426 | } | |
427 | ||
428 | /* Fill a zone bitmap for a policy */ | |
dfcd3c0d | 429 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
1da177e4 LT |
430 | { |
431 | int i; | |
432 | ||
dfcd3c0d | 433 | nodes_clear(*nodes); |
1da177e4 LT |
434 | switch (p->policy) { |
435 | case MPOL_BIND: | |
436 | for (i = 0; p->v.zonelist->zones[i]; i++) | |
8bccd85f CL |
437 | node_set(p->v.zonelist->zones[i]->zone_pgdat->node_id, |
438 | *nodes); | |
1da177e4 LT |
439 | break; |
440 | case MPOL_DEFAULT: | |
441 | break; | |
442 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 443 | *nodes = p->v.nodes; |
1da177e4 LT |
444 | break; |
445 | case MPOL_PREFERRED: | |
446 | /* or use current node instead of online map? */ | |
447 | if (p->v.preferred_node < 0) | |
dfcd3c0d | 448 | *nodes = node_online_map; |
1da177e4 | 449 | else |
dfcd3c0d | 450 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
451 | break; |
452 | default: | |
453 | BUG(); | |
454 | } | |
455 | } | |
456 | ||
457 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
458 | { | |
459 | struct page *p; | |
460 | int err; | |
461 | ||
462 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
463 | if (err >= 0) { | |
464 | err = page_to_nid(p); | |
465 | put_page(p); | |
466 | } | |
467 | return err; | |
468 | } | |
469 | ||
1da177e4 | 470 | /* Retrieve NUMA policy */ |
8bccd85f CL |
471 | long do_get_mempolicy(int *policy, nodemask_t *nmask, |
472 | unsigned long addr, unsigned long flags) | |
1da177e4 | 473 | { |
8bccd85f | 474 | int err; |
1da177e4 LT |
475 | struct mm_struct *mm = current->mm; |
476 | struct vm_area_struct *vma = NULL; | |
477 | struct mempolicy *pol = current->mempolicy; | |
478 | ||
cf2a473c | 479 | cpuset_update_task_memory_state(); |
1da177e4 LT |
480 | if (flags & ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR)) |
481 | return -EINVAL; | |
1da177e4 LT |
482 | if (flags & MPOL_F_ADDR) { |
483 | down_read(&mm->mmap_sem); | |
484 | vma = find_vma_intersection(mm, addr, addr+1); | |
485 | if (!vma) { | |
486 | up_read(&mm->mmap_sem); | |
487 | return -EFAULT; | |
488 | } | |
489 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
490 | pol = vma->vm_ops->get_policy(vma, addr); | |
491 | else | |
492 | pol = vma->vm_policy; | |
493 | } else if (addr) | |
494 | return -EINVAL; | |
495 | ||
496 | if (!pol) | |
497 | pol = &default_policy; | |
498 | ||
499 | if (flags & MPOL_F_NODE) { | |
500 | if (flags & MPOL_F_ADDR) { | |
501 | err = lookup_node(mm, addr); | |
502 | if (err < 0) | |
503 | goto out; | |
8bccd85f | 504 | *policy = err; |
1da177e4 LT |
505 | } else if (pol == current->mempolicy && |
506 | pol->policy == MPOL_INTERLEAVE) { | |
8bccd85f | 507 | *policy = current->il_next; |
1da177e4 LT |
508 | } else { |
509 | err = -EINVAL; | |
510 | goto out; | |
511 | } | |
512 | } else | |
8bccd85f | 513 | *policy = pol->policy; |
1da177e4 LT |
514 | |
515 | if (vma) { | |
516 | up_read(¤t->mm->mmap_sem); | |
517 | vma = NULL; | |
518 | } | |
519 | ||
1da177e4 | 520 | err = 0; |
8bccd85f CL |
521 | if (nmask) |
522 | get_zonemask(pol, nmask); | |
1da177e4 LT |
523 | |
524 | out: | |
525 | if (vma) | |
526 | up_read(¤t->mm->mmap_sem); | |
527 | return err; | |
528 | } | |
529 | ||
6ce3c4c0 CL |
530 | /* |
531 | * page migration | |
532 | */ | |
533 | ||
fc301289 CL |
534 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
535 | unsigned long flags) | |
6ce3c4c0 CL |
536 | { |
537 | /* | |
fc301289 | 538 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 539 | */ |
fc301289 | 540 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
053837fc | 541 | if (isolate_lru_page(page)) |
6ce3c4c0 | 542 | list_add(&page->lru, pagelist); |
6ce3c4c0 CL |
543 | } |
544 | } | |
545 | ||
546 | static int swap_pages(struct list_head *pagelist) | |
547 | { | |
548 | LIST_HEAD(moved); | |
549 | LIST_HEAD(failed); | |
550 | int n; | |
551 | ||
552 | n = migrate_pages(pagelist, NULL, &moved, &failed); | |
553 | putback_lru_pages(&failed); | |
554 | putback_lru_pages(&moved); | |
555 | ||
556 | return n; | |
557 | } | |
558 | ||
39743889 CL |
559 | /* |
560 | * For now migrate_pages simply swaps out the pages from nodes that are in | |
561 | * the source set but not in the target set. In the future, we would | |
562 | * want a function that moves pages between the two nodesets in such | |
563 | * a way as to preserve the physical layout as much as possible. | |
564 | * | |
565 | * Returns the number of page that could not be moved. | |
566 | */ | |
567 | int do_migrate_pages(struct mm_struct *mm, | |
568 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
569 | { | |
570 | LIST_HEAD(pagelist); | |
571 | int count = 0; | |
572 | nodemask_t nodes; | |
573 | ||
574 | nodes_andnot(nodes, *from_nodes, *to_nodes); | |
39743889 CL |
575 | |
576 | down_read(&mm->mmap_sem); | |
577 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nodes, | |
578 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
d4984711 | 579 | |
39743889 | 580 | if (!list_empty(&pagelist)) { |
d4984711 CL |
581 | count = swap_pages(&pagelist); |
582 | putback_lru_pages(&pagelist); | |
39743889 | 583 | } |
d4984711 | 584 | |
39743889 CL |
585 | up_read(&mm->mmap_sem); |
586 | return count; | |
587 | } | |
588 | ||
6ce3c4c0 CL |
589 | long do_mbind(unsigned long start, unsigned long len, |
590 | unsigned long mode, nodemask_t *nmask, unsigned long flags) | |
591 | { | |
592 | struct vm_area_struct *vma; | |
593 | struct mm_struct *mm = current->mm; | |
594 | struct mempolicy *new; | |
595 | unsigned long end; | |
596 | int err; | |
597 | LIST_HEAD(pagelist); | |
598 | ||
599 | if ((flags & ~(unsigned long)(MPOL_MF_STRICT | | |
600 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
601 | || mode > MPOL_MAX) | |
602 | return -EINVAL; | |
603 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_RESOURCE)) | |
604 | return -EPERM; | |
605 | ||
606 | if (start & ~PAGE_MASK) | |
607 | return -EINVAL; | |
608 | ||
609 | if (mode == MPOL_DEFAULT) | |
610 | flags &= ~MPOL_MF_STRICT; | |
611 | ||
612 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
613 | end = start + len; | |
614 | ||
615 | if (end < start) | |
616 | return -EINVAL; | |
617 | if (end == start) | |
618 | return 0; | |
619 | ||
620 | if (mpol_check_policy(mode, nmask)) | |
621 | return -EINVAL; | |
622 | ||
623 | new = mpol_new(mode, nmask); | |
624 | if (IS_ERR(new)) | |
625 | return PTR_ERR(new); | |
626 | ||
627 | /* | |
628 | * If we are using the default policy then operation | |
629 | * on discontinuous address spaces is okay after all | |
630 | */ | |
631 | if (!new) | |
632 | flags |= MPOL_MF_DISCONTIG_OK; | |
633 | ||
634 | PDprintk("mbind %lx-%lx mode:%ld nodes:%lx\n",start,start+len, | |
635 | mode,nodes_addr(nodes)[0]); | |
636 | ||
637 | down_write(&mm->mmap_sem); | |
638 | vma = check_range(mm, start, end, nmask, | |
639 | flags | MPOL_MF_INVERT, &pagelist); | |
640 | ||
641 | err = PTR_ERR(vma); | |
642 | if (!IS_ERR(vma)) { | |
643 | int nr_failed = 0; | |
644 | ||
645 | err = mbind_range(vma, start, end, new); | |
646 | if (!list_empty(&pagelist)) | |
647 | nr_failed = swap_pages(&pagelist); | |
648 | ||
649 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
650 | err = -EIO; | |
651 | } | |
652 | if (!list_empty(&pagelist)) | |
653 | putback_lru_pages(&pagelist); | |
654 | ||
655 | up_write(&mm->mmap_sem); | |
656 | mpol_free(new); | |
657 | return err; | |
658 | } | |
659 | ||
8bccd85f CL |
660 | /* |
661 | * User space interface with variable sized bitmaps for nodelists. | |
662 | */ | |
663 | ||
664 | /* Copy a node mask from user space. */ | |
39743889 | 665 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
666 | unsigned long maxnode) |
667 | { | |
668 | unsigned long k; | |
669 | unsigned long nlongs; | |
670 | unsigned long endmask; | |
671 | ||
672 | --maxnode; | |
673 | nodes_clear(*nodes); | |
674 | if (maxnode == 0 || !nmask) | |
675 | return 0; | |
676 | ||
677 | nlongs = BITS_TO_LONGS(maxnode); | |
678 | if ((maxnode % BITS_PER_LONG) == 0) | |
679 | endmask = ~0UL; | |
680 | else | |
681 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
682 | ||
683 | /* When the user specified more nodes than supported just check | |
684 | if the non supported part is all zero. */ | |
685 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
686 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
687 | return -EINVAL; | |
688 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
689 | unsigned long t; | |
690 | if (get_user(t, nmask + k)) | |
691 | return -EFAULT; | |
692 | if (k == nlongs - 1) { | |
693 | if (t & endmask) | |
694 | return -EINVAL; | |
695 | } else if (t) | |
696 | return -EINVAL; | |
697 | } | |
698 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
699 | endmask = ~0UL; | |
700 | } | |
701 | ||
702 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
703 | return -EFAULT; | |
704 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
705 | return 0; | |
706 | } | |
707 | ||
708 | /* Copy a kernel node mask to user space */ | |
709 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
710 | nodemask_t *nodes) | |
711 | { | |
712 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
713 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
714 | ||
715 | if (copy > nbytes) { | |
716 | if (copy > PAGE_SIZE) | |
717 | return -EINVAL; | |
718 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
719 | return -EFAULT; | |
720 | copy = nbytes; | |
721 | } | |
722 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
723 | } | |
724 | ||
725 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
726 | unsigned long mode, | |
727 | unsigned long __user *nmask, unsigned long maxnode, | |
728 | unsigned flags) | |
729 | { | |
730 | nodemask_t nodes; | |
731 | int err; | |
732 | ||
733 | err = get_nodes(&nodes, nmask, maxnode); | |
734 | if (err) | |
735 | return err; | |
736 | return do_mbind(start, len, mode, &nodes, flags); | |
737 | } | |
738 | ||
739 | /* Set the process memory policy */ | |
740 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
741 | unsigned long maxnode) | |
742 | { | |
743 | int err; | |
744 | nodemask_t nodes; | |
745 | ||
746 | if (mode < 0 || mode > MPOL_MAX) | |
747 | return -EINVAL; | |
748 | err = get_nodes(&nodes, nmask, maxnode); | |
749 | if (err) | |
750 | return err; | |
751 | return do_set_mempolicy(mode, &nodes); | |
752 | } | |
753 | ||
39743889 CL |
754 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
755 | const unsigned long __user *old_nodes, | |
756 | const unsigned long __user *new_nodes) | |
757 | { | |
758 | struct mm_struct *mm; | |
759 | struct task_struct *task; | |
760 | nodemask_t old; | |
761 | nodemask_t new; | |
762 | nodemask_t task_nodes; | |
763 | int err; | |
764 | ||
765 | err = get_nodes(&old, old_nodes, maxnode); | |
766 | if (err) | |
767 | return err; | |
768 | ||
769 | err = get_nodes(&new, new_nodes, maxnode); | |
770 | if (err) | |
771 | return err; | |
772 | ||
773 | /* Find the mm_struct */ | |
774 | read_lock(&tasklist_lock); | |
775 | task = pid ? find_task_by_pid(pid) : current; | |
776 | if (!task) { | |
777 | read_unlock(&tasklist_lock); | |
778 | return -ESRCH; | |
779 | } | |
780 | mm = get_task_mm(task); | |
781 | read_unlock(&tasklist_lock); | |
782 | ||
783 | if (!mm) | |
784 | return -EINVAL; | |
785 | ||
786 | /* | |
787 | * Check if this process has the right to modify the specified | |
788 | * process. The right exists if the process has administrative | |
789 | * capabilities, superuser priviledges or the same | |
790 | * userid as the target process. | |
791 | */ | |
792 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
793 | (current->uid != task->suid) && (current->uid != task->uid) && | |
794 | !capable(CAP_SYS_ADMIN)) { | |
795 | err = -EPERM; | |
796 | goto out; | |
797 | } | |
798 | ||
799 | task_nodes = cpuset_mems_allowed(task); | |
800 | /* Is the user allowed to access the target nodes? */ | |
801 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_ADMIN)) { | |
802 | err = -EPERM; | |
803 | goto out; | |
804 | } | |
805 | ||
806 | err = do_migrate_pages(mm, &old, &new, MPOL_MF_MOVE); | |
807 | out: | |
808 | mmput(mm); | |
809 | return err; | |
810 | } | |
811 | ||
812 | ||
8bccd85f CL |
813 | /* Retrieve NUMA policy */ |
814 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
815 | unsigned long __user *nmask, | |
816 | unsigned long maxnode, | |
817 | unsigned long addr, unsigned long flags) | |
818 | { | |
819 | int err, pval; | |
820 | nodemask_t nodes; | |
821 | ||
822 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
823 | return -EINVAL; | |
824 | ||
825 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
826 | ||
827 | if (err) | |
828 | return err; | |
829 | ||
830 | if (policy && put_user(pval, policy)) | |
831 | return -EFAULT; | |
832 | ||
833 | if (nmask) | |
834 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
835 | ||
836 | return err; | |
837 | } | |
838 | ||
1da177e4 LT |
839 | #ifdef CONFIG_COMPAT |
840 | ||
841 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
842 | compat_ulong_t __user *nmask, | |
843 | compat_ulong_t maxnode, | |
844 | compat_ulong_t addr, compat_ulong_t flags) | |
845 | { | |
846 | long err; | |
847 | unsigned long __user *nm = NULL; | |
848 | unsigned long nr_bits, alloc_size; | |
849 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
850 | ||
851 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
852 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
853 | ||
854 | if (nmask) | |
855 | nm = compat_alloc_user_space(alloc_size); | |
856 | ||
857 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
858 | ||
859 | if (!err && nmask) { | |
860 | err = copy_from_user(bm, nm, alloc_size); | |
861 | /* ensure entire bitmap is zeroed */ | |
862 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
863 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
864 | } | |
865 | ||
866 | return err; | |
867 | } | |
868 | ||
869 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
870 | compat_ulong_t maxnode) | |
871 | { | |
872 | long err = 0; | |
873 | unsigned long __user *nm = NULL; | |
874 | unsigned long nr_bits, alloc_size; | |
875 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
876 | ||
877 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
878 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
879 | ||
880 | if (nmask) { | |
881 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
882 | nm = compat_alloc_user_space(alloc_size); | |
883 | err |= copy_to_user(nm, bm, alloc_size); | |
884 | } | |
885 | ||
886 | if (err) | |
887 | return -EFAULT; | |
888 | ||
889 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
890 | } | |
891 | ||
892 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
893 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
894 | compat_ulong_t maxnode, compat_ulong_t flags) | |
895 | { | |
896 | long err = 0; | |
897 | unsigned long __user *nm = NULL; | |
898 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 899 | nodemask_t bm; |
1da177e4 LT |
900 | |
901 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
902 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
903 | ||
904 | if (nmask) { | |
dfcd3c0d | 905 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 906 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 907 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
908 | } |
909 | ||
910 | if (err) | |
911 | return -EFAULT; | |
912 | ||
913 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
914 | } | |
915 | ||
916 | #endif | |
917 | ||
918 | /* Return effective policy for a VMA */ | |
48fce342 CL |
919 | static struct mempolicy * get_vma_policy(struct task_struct *task, |
920 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 921 | { |
6e21c8f1 | 922 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
923 | |
924 | if (vma) { | |
925 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
8bccd85f | 926 | pol = vma->vm_ops->get_policy(vma, addr); |
1da177e4 LT |
927 | else if (vma->vm_policy && |
928 | vma->vm_policy->policy != MPOL_DEFAULT) | |
929 | pol = vma->vm_policy; | |
930 | } | |
931 | if (!pol) | |
932 | pol = &default_policy; | |
933 | return pol; | |
934 | } | |
935 | ||
936 | /* Return a zonelist representing a mempolicy */ | |
dd0fc66f | 937 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
1da177e4 LT |
938 | { |
939 | int nd; | |
940 | ||
941 | switch (policy->policy) { | |
942 | case MPOL_PREFERRED: | |
943 | nd = policy->v.preferred_node; | |
944 | if (nd < 0) | |
945 | nd = numa_node_id(); | |
946 | break; | |
947 | case MPOL_BIND: | |
948 | /* Lower zones don't get a policy applied */ | |
949 | /* Careful: current->mems_allowed might have moved */ | |
af4ca457 | 950 | if (gfp_zone(gfp) >= policy_zone) |
1da177e4 LT |
951 | if (cpuset_zonelist_valid_mems_allowed(policy->v.zonelist)) |
952 | return policy->v.zonelist; | |
953 | /*FALL THROUGH*/ | |
954 | case MPOL_INTERLEAVE: /* should not happen */ | |
955 | case MPOL_DEFAULT: | |
956 | nd = numa_node_id(); | |
957 | break; | |
958 | default: | |
959 | nd = 0; | |
960 | BUG(); | |
961 | } | |
af4ca457 | 962 | return NODE_DATA(nd)->node_zonelists + gfp_zone(gfp); |
1da177e4 LT |
963 | } |
964 | ||
965 | /* Do dynamic interleaving for a process */ | |
966 | static unsigned interleave_nodes(struct mempolicy *policy) | |
967 | { | |
968 | unsigned nid, next; | |
969 | struct task_struct *me = current; | |
970 | ||
971 | nid = me->il_next; | |
dfcd3c0d | 972 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 973 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 974 | next = first_node(policy->v.nodes); |
1da177e4 LT |
975 | me->il_next = next; |
976 | return nid; | |
977 | } | |
978 | ||
979 | /* Do static interleaving for a VMA with known offset. */ | |
980 | static unsigned offset_il_node(struct mempolicy *pol, | |
981 | struct vm_area_struct *vma, unsigned long off) | |
982 | { | |
dfcd3c0d | 983 | unsigned nnodes = nodes_weight(pol->v.nodes); |
1da177e4 LT |
984 | unsigned target = (unsigned)off % nnodes; |
985 | int c; | |
986 | int nid = -1; | |
987 | ||
988 | c = 0; | |
989 | do { | |
dfcd3c0d | 990 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
991 | c++; |
992 | } while (c <= target); | |
1da177e4 LT |
993 | return nid; |
994 | } | |
995 | ||
5da7ca86 CL |
996 | /* Determine a node number for interleave */ |
997 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
998 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
999 | { | |
1000 | if (vma) { | |
1001 | unsigned long off; | |
1002 | ||
1003 | off = vma->vm_pgoff; | |
1004 | off += (addr - vma->vm_start) >> shift; | |
1005 | return offset_il_node(pol, vma, off); | |
1006 | } else | |
1007 | return interleave_nodes(pol); | |
1008 | } | |
1009 | ||
1010 | /* Return a zonelist suitable for a huge page allocation. */ | |
1011 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr) | |
1012 | { | |
1013 | struct mempolicy *pol = get_vma_policy(current, vma, addr); | |
1014 | ||
1015 | if (pol->policy == MPOL_INTERLEAVE) { | |
1016 | unsigned nid; | |
1017 | ||
1018 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); | |
1019 | return NODE_DATA(nid)->node_zonelists + gfp_zone(GFP_HIGHUSER); | |
1020 | } | |
1021 | return zonelist_policy(GFP_HIGHUSER, pol); | |
1022 | } | |
1023 | ||
1da177e4 LT |
1024 | /* Allocate a page in interleaved policy. |
1025 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1026 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1027 | unsigned nid) | |
1da177e4 LT |
1028 | { |
1029 | struct zonelist *zl; | |
1030 | struct page *page; | |
1031 | ||
af4ca457 | 1032 | zl = NODE_DATA(nid)->node_zonelists + gfp_zone(gfp); |
1da177e4 LT |
1033 | page = __alloc_pages(gfp, order, zl); |
1034 | if (page && page_zone(page) == zl->zones[0]) { | |
e7c8d5c9 | 1035 | zone_pcp(zl->zones[0],get_cpu())->interleave_hit++; |
1da177e4 LT |
1036 | put_cpu(); |
1037 | } | |
1038 | return page; | |
1039 | } | |
1040 | ||
1041 | /** | |
1042 | * alloc_page_vma - Allocate a page for a VMA. | |
1043 | * | |
1044 | * @gfp: | |
1045 | * %GFP_USER user allocation. | |
1046 | * %GFP_KERNEL kernel allocations, | |
1047 | * %GFP_HIGHMEM highmem/user allocations, | |
1048 | * %GFP_FS allocation should not call back into a file system. | |
1049 | * %GFP_ATOMIC don't sleep. | |
1050 | * | |
1051 | * @vma: Pointer to VMA or NULL if not available. | |
1052 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1053 | * | |
1054 | * This function allocates a page from the kernel page pool and applies | |
1055 | * a NUMA policy associated with the VMA or the current process. | |
1056 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1057 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1058 | * all allocations for pages that will be mapped into | |
1059 | * user space. Returns NULL when no page can be allocated. | |
1060 | * | |
1061 | * Should be called with the mm_sem of the vma hold. | |
1062 | */ | |
1063 | struct page * | |
dd0fc66f | 1064 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1065 | { |
6e21c8f1 | 1066 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
1da177e4 | 1067 | |
cf2a473c | 1068 | cpuset_update_task_memory_state(); |
1da177e4 LT |
1069 | |
1070 | if (unlikely(pol->policy == MPOL_INTERLEAVE)) { | |
1071 | unsigned nid; | |
5da7ca86 CL |
1072 | |
1073 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
1da177e4 LT |
1074 | return alloc_page_interleave(gfp, 0, nid); |
1075 | } | |
1076 | return __alloc_pages(gfp, 0, zonelist_policy(gfp, pol)); | |
1077 | } | |
1078 | ||
1079 | /** | |
1080 | * alloc_pages_current - Allocate pages. | |
1081 | * | |
1082 | * @gfp: | |
1083 | * %GFP_USER user allocation, | |
1084 | * %GFP_KERNEL kernel allocation, | |
1085 | * %GFP_HIGHMEM highmem allocation, | |
1086 | * %GFP_FS don't call back into a file system. | |
1087 | * %GFP_ATOMIC don't sleep. | |
1088 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1089 | * | |
1090 | * Allocate a page from the kernel page pool. When not in | |
1091 | * interrupt context and apply the current process NUMA policy. | |
1092 | * Returns NULL when no page can be allocated. | |
1093 | * | |
cf2a473c | 1094 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1095 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1096 | * 2) allocating for current task (not interrupt). | |
1097 | */ | |
dd0fc66f | 1098 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1099 | { |
1100 | struct mempolicy *pol = current->mempolicy; | |
1101 | ||
1102 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1103 | cpuset_update_task_memory_state(); |
1da177e4 LT |
1104 | if (!pol || in_interrupt()) |
1105 | pol = &default_policy; | |
1106 | if (pol->policy == MPOL_INTERLEAVE) | |
1107 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); | |
1108 | return __alloc_pages(gfp, order, zonelist_policy(gfp, pol)); | |
1109 | } | |
1110 | EXPORT_SYMBOL(alloc_pages_current); | |
1111 | ||
4225399a PJ |
1112 | /* |
1113 | * If mpol_copy() sees current->cpuset == cpuset_being_rebound, then it | |
1114 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() | |
1115 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1116 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1117 | * further kernel/cpuset.c update_nodemask(). | |
1118 | */ | |
1119 | void *cpuset_being_rebound; | |
1120 | ||
1da177e4 LT |
1121 | /* Slow path of a mempolicy copy */ |
1122 | struct mempolicy *__mpol_copy(struct mempolicy *old) | |
1123 | { | |
1124 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1125 | ||
1126 | if (!new) | |
1127 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1128 | if (current_cpuset_is_being_rebound()) { |
1129 | nodemask_t mems = cpuset_mems_allowed(current); | |
1130 | mpol_rebind_policy(old, &mems); | |
1131 | } | |
1da177e4 LT |
1132 | *new = *old; |
1133 | atomic_set(&new->refcnt, 1); | |
1134 | if (new->policy == MPOL_BIND) { | |
1135 | int sz = ksize(old->v.zonelist); | |
1136 | new->v.zonelist = kmalloc(sz, SLAB_KERNEL); | |
1137 | if (!new->v.zonelist) { | |
1138 | kmem_cache_free(policy_cache, new); | |
1139 | return ERR_PTR(-ENOMEM); | |
1140 | } | |
1141 | memcpy(new->v.zonelist, old->v.zonelist, sz); | |
1142 | } | |
1143 | return new; | |
1144 | } | |
1145 | ||
1146 | /* Slow path of a mempolicy comparison */ | |
1147 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1148 | { | |
1149 | if (!a || !b) | |
1150 | return 0; | |
1151 | if (a->policy != b->policy) | |
1152 | return 0; | |
1153 | switch (a->policy) { | |
1154 | case MPOL_DEFAULT: | |
1155 | return 1; | |
1156 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 1157 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 LT |
1158 | case MPOL_PREFERRED: |
1159 | return a->v.preferred_node == b->v.preferred_node; | |
1160 | case MPOL_BIND: { | |
1161 | int i; | |
1162 | for (i = 0; a->v.zonelist->zones[i]; i++) | |
1163 | if (a->v.zonelist->zones[i] != b->v.zonelist->zones[i]) | |
1164 | return 0; | |
1165 | return b->v.zonelist->zones[i] == NULL; | |
1166 | } | |
1167 | default: | |
1168 | BUG(); | |
1169 | return 0; | |
1170 | } | |
1171 | } | |
1172 | ||
1173 | /* Slow path of a mpol destructor. */ | |
1174 | void __mpol_free(struct mempolicy *p) | |
1175 | { | |
1176 | if (!atomic_dec_and_test(&p->refcnt)) | |
1177 | return; | |
1178 | if (p->policy == MPOL_BIND) | |
1179 | kfree(p->v.zonelist); | |
1180 | p->policy = MPOL_DEFAULT; | |
1181 | kmem_cache_free(policy_cache, p); | |
1182 | } | |
1183 | ||
1da177e4 LT |
1184 | /* |
1185 | * Shared memory backing store policy support. | |
1186 | * | |
1187 | * Remember policies even when nobody has shared memory mapped. | |
1188 | * The policies are kept in Red-Black tree linked from the inode. | |
1189 | * They are protected by the sp->lock spinlock, which should be held | |
1190 | * for any accesses to the tree. | |
1191 | */ | |
1192 | ||
1193 | /* lookup first element intersecting start-end */ | |
1194 | /* Caller holds sp->lock */ | |
1195 | static struct sp_node * | |
1196 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1197 | { | |
1198 | struct rb_node *n = sp->root.rb_node; | |
1199 | ||
1200 | while (n) { | |
1201 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1202 | ||
1203 | if (start >= p->end) | |
1204 | n = n->rb_right; | |
1205 | else if (end <= p->start) | |
1206 | n = n->rb_left; | |
1207 | else | |
1208 | break; | |
1209 | } | |
1210 | if (!n) | |
1211 | return NULL; | |
1212 | for (;;) { | |
1213 | struct sp_node *w = NULL; | |
1214 | struct rb_node *prev = rb_prev(n); | |
1215 | if (!prev) | |
1216 | break; | |
1217 | w = rb_entry(prev, struct sp_node, nd); | |
1218 | if (w->end <= start) | |
1219 | break; | |
1220 | n = prev; | |
1221 | } | |
1222 | return rb_entry(n, struct sp_node, nd); | |
1223 | } | |
1224 | ||
1225 | /* Insert a new shared policy into the list. */ | |
1226 | /* Caller holds sp->lock */ | |
1227 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1228 | { | |
1229 | struct rb_node **p = &sp->root.rb_node; | |
1230 | struct rb_node *parent = NULL; | |
1231 | struct sp_node *nd; | |
1232 | ||
1233 | while (*p) { | |
1234 | parent = *p; | |
1235 | nd = rb_entry(parent, struct sp_node, nd); | |
1236 | if (new->start < nd->start) | |
1237 | p = &(*p)->rb_left; | |
1238 | else if (new->end > nd->end) | |
1239 | p = &(*p)->rb_right; | |
1240 | else | |
1241 | BUG(); | |
1242 | } | |
1243 | rb_link_node(&new->nd, parent, p); | |
1244 | rb_insert_color(&new->nd, &sp->root); | |
1245 | PDprintk("inserting %lx-%lx: %d\n", new->start, new->end, | |
1246 | new->policy ? new->policy->policy : 0); | |
1247 | } | |
1248 | ||
1249 | /* Find shared policy intersecting idx */ | |
1250 | struct mempolicy * | |
1251 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1252 | { | |
1253 | struct mempolicy *pol = NULL; | |
1254 | struct sp_node *sn; | |
1255 | ||
1256 | if (!sp->root.rb_node) | |
1257 | return NULL; | |
1258 | spin_lock(&sp->lock); | |
1259 | sn = sp_lookup(sp, idx, idx+1); | |
1260 | if (sn) { | |
1261 | mpol_get(sn->policy); | |
1262 | pol = sn->policy; | |
1263 | } | |
1264 | spin_unlock(&sp->lock); | |
1265 | return pol; | |
1266 | } | |
1267 | ||
1268 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1269 | { | |
1270 | PDprintk("deleting %lx-l%x\n", n->start, n->end); | |
1271 | rb_erase(&n->nd, &sp->root); | |
1272 | mpol_free(n->policy); | |
1273 | kmem_cache_free(sn_cache, n); | |
1274 | } | |
1275 | ||
1276 | struct sp_node * | |
1277 | sp_alloc(unsigned long start, unsigned long end, struct mempolicy *pol) | |
1278 | { | |
1279 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1280 | ||
1281 | if (!n) | |
1282 | return NULL; | |
1283 | n->start = start; | |
1284 | n->end = end; | |
1285 | mpol_get(pol); | |
1286 | n->policy = pol; | |
1287 | return n; | |
1288 | } | |
1289 | ||
1290 | /* Replace a policy range. */ | |
1291 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1292 | unsigned long end, struct sp_node *new) | |
1293 | { | |
1294 | struct sp_node *n, *new2 = NULL; | |
1295 | ||
1296 | restart: | |
1297 | spin_lock(&sp->lock); | |
1298 | n = sp_lookup(sp, start, end); | |
1299 | /* Take care of old policies in the same range. */ | |
1300 | while (n && n->start < end) { | |
1301 | struct rb_node *next = rb_next(&n->nd); | |
1302 | if (n->start >= start) { | |
1303 | if (n->end <= end) | |
1304 | sp_delete(sp, n); | |
1305 | else | |
1306 | n->start = end; | |
1307 | } else { | |
1308 | /* Old policy spanning whole new range. */ | |
1309 | if (n->end > end) { | |
1310 | if (!new2) { | |
1311 | spin_unlock(&sp->lock); | |
1312 | new2 = sp_alloc(end, n->end, n->policy); | |
1313 | if (!new2) | |
1314 | return -ENOMEM; | |
1315 | goto restart; | |
1316 | } | |
1317 | n->end = start; | |
1318 | sp_insert(sp, new2); | |
1319 | new2 = NULL; | |
1320 | break; | |
1321 | } else | |
1322 | n->end = start; | |
1323 | } | |
1324 | if (!next) | |
1325 | break; | |
1326 | n = rb_entry(next, struct sp_node, nd); | |
1327 | } | |
1328 | if (new) | |
1329 | sp_insert(sp, new); | |
1330 | spin_unlock(&sp->lock); | |
1331 | if (new2) { | |
1332 | mpol_free(new2->policy); | |
1333 | kmem_cache_free(sn_cache, new2); | |
1334 | } | |
1335 | return 0; | |
1336 | } | |
1337 | ||
7339ff83 RH |
1338 | void mpol_shared_policy_init(struct shared_policy *info, int policy, |
1339 | nodemask_t *policy_nodes) | |
1340 | { | |
1341 | info->root = RB_ROOT; | |
1342 | spin_lock_init(&info->lock); | |
1343 | ||
1344 | if (policy != MPOL_DEFAULT) { | |
1345 | struct mempolicy *newpol; | |
1346 | ||
1347 | /* Falls back to MPOL_DEFAULT on any error */ | |
1348 | newpol = mpol_new(policy, policy_nodes); | |
1349 | if (!IS_ERR(newpol)) { | |
1350 | /* Create pseudo-vma that contains just the policy */ | |
1351 | struct vm_area_struct pvma; | |
1352 | ||
1353 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1354 | /* Policy covers entire file */ | |
1355 | pvma.vm_end = TASK_SIZE; | |
1356 | mpol_set_shared_policy(info, &pvma, newpol); | |
1357 | mpol_free(newpol); | |
1358 | } | |
1359 | } | |
1360 | } | |
1361 | ||
1da177e4 LT |
1362 | int mpol_set_shared_policy(struct shared_policy *info, |
1363 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1364 | { | |
1365 | int err; | |
1366 | struct sp_node *new = NULL; | |
1367 | unsigned long sz = vma_pages(vma); | |
1368 | ||
1369 | PDprintk("set_shared_policy %lx sz %lu %d %lx\n", | |
1370 | vma->vm_pgoff, | |
1371 | sz, npol? npol->policy : -1, | |
dfcd3c0d | 1372 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1373 | |
1374 | if (npol) { | |
1375 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1376 | if (!new) | |
1377 | return -ENOMEM; | |
1378 | } | |
1379 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1380 | if (err && new) | |
1381 | kmem_cache_free(sn_cache, new); | |
1382 | return err; | |
1383 | } | |
1384 | ||
1385 | /* Free a backing policy store on inode delete. */ | |
1386 | void mpol_free_shared_policy(struct shared_policy *p) | |
1387 | { | |
1388 | struct sp_node *n; | |
1389 | struct rb_node *next; | |
1390 | ||
1391 | if (!p->root.rb_node) | |
1392 | return; | |
1393 | spin_lock(&p->lock); | |
1394 | next = rb_first(&p->root); | |
1395 | while (next) { | |
1396 | n = rb_entry(next, struct sp_node, nd); | |
1397 | next = rb_next(&n->nd); | |
90c5029e | 1398 | rb_erase(&n->nd, &p->root); |
1da177e4 LT |
1399 | mpol_free(n->policy); |
1400 | kmem_cache_free(sn_cache, n); | |
1401 | } | |
1402 | spin_unlock(&p->lock); | |
1da177e4 LT |
1403 | } |
1404 | ||
1405 | /* assumes fs == KERNEL_DS */ | |
1406 | void __init numa_policy_init(void) | |
1407 | { | |
1408 | policy_cache = kmem_cache_create("numa_policy", | |
1409 | sizeof(struct mempolicy), | |
1410 | 0, SLAB_PANIC, NULL, NULL); | |
1411 | ||
1412 | sn_cache = kmem_cache_create("shared_policy_node", | |
1413 | sizeof(struct sp_node), | |
1414 | 0, SLAB_PANIC, NULL, NULL); | |
1415 | ||
1416 | /* Set interleaving policy for system init. This way not all | |
1417 | the data structures allocated at system boot end up in node zero. */ | |
1418 | ||
8bccd85f | 1419 | if (do_set_mempolicy(MPOL_INTERLEAVE, &node_online_map)) |
1da177e4 LT |
1420 | printk("numa_policy_init: interleaving failed\n"); |
1421 | } | |
1422 | ||
8bccd85f | 1423 | /* Reset policy of current process to default */ |
1da177e4 LT |
1424 | void numa_default_policy(void) |
1425 | { | |
8bccd85f | 1426 | do_set_mempolicy(MPOL_DEFAULT, NULL); |
1da177e4 | 1427 | } |
68860ec1 PJ |
1428 | |
1429 | /* Migrate a policy to a different set of nodes */ | |
74cb2155 | 1430 | void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
68860ec1 | 1431 | { |
74cb2155 | 1432 | nodemask_t *mpolmask; |
68860ec1 PJ |
1433 | nodemask_t tmp; |
1434 | ||
1435 | if (!pol) | |
1436 | return; | |
74cb2155 PJ |
1437 | mpolmask = &pol->cpuset_mems_allowed; |
1438 | if (nodes_equal(*mpolmask, *newmask)) | |
1439 | return; | |
68860ec1 PJ |
1440 | |
1441 | switch (pol->policy) { | |
1442 | case MPOL_DEFAULT: | |
1443 | break; | |
1444 | case MPOL_INTERLEAVE: | |
74cb2155 | 1445 | nodes_remap(tmp, pol->v.nodes, *mpolmask, *newmask); |
68860ec1 | 1446 | pol->v.nodes = tmp; |
74cb2155 PJ |
1447 | *mpolmask = *newmask; |
1448 | current->il_next = node_remap(current->il_next, | |
1449 | *mpolmask, *newmask); | |
68860ec1 PJ |
1450 | break; |
1451 | case MPOL_PREFERRED: | |
1452 | pol->v.preferred_node = node_remap(pol->v.preferred_node, | |
74cb2155 PJ |
1453 | *mpolmask, *newmask); |
1454 | *mpolmask = *newmask; | |
68860ec1 PJ |
1455 | break; |
1456 | case MPOL_BIND: { | |
1457 | nodemask_t nodes; | |
1458 | struct zone **z; | |
1459 | struct zonelist *zonelist; | |
1460 | ||
1461 | nodes_clear(nodes); | |
1462 | for (z = pol->v.zonelist->zones; *z; z++) | |
1463 | node_set((*z)->zone_pgdat->node_id, nodes); | |
74cb2155 | 1464 | nodes_remap(tmp, nodes, *mpolmask, *newmask); |
68860ec1 PJ |
1465 | nodes = tmp; |
1466 | ||
1467 | zonelist = bind_zonelist(&nodes); | |
1468 | ||
1469 | /* If no mem, then zonelist is NULL and we keep old zonelist. | |
1470 | * If that old zonelist has no remaining mems_allowed nodes, | |
1471 | * then zonelist_policy() will "FALL THROUGH" to MPOL_DEFAULT. | |
1472 | */ | |
1473 | ||
1474 | if (zonelist) { | |
1475 | /* Good - got mem - substitute new zonelist */ | |
1476 | kfree(pol->v.zonelist); | |
1477 | pol->v.zonelist = zonelist; | |
1478 | } | |
74cb2155 | 1479 | *mpolmask = *newmask; |
68860ec1 PJ |
1480 | break; |
1481 | } | |
1482 | default: | |
1483 | BUG(); | |
1484 | break; | |
1485 | } | |
1486 | } | |
1487 | ||
1488 | /* | |
74cb2155 PJ |
1489 | * Wrapper for mpol_rebind_policy() that just requires task |
1490 | * pointer, and updates task mempolicy. | |
68860ec1 | 1491 | */ |
74cb2155 PJ |
1492 | |
1493 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
68860ec1 | 1494 | { |
74cb2155 | 1495 | mpol_rebind_policy(tsk->mempolicy, new); |
68860ec1 | 1496 | } |
1a75a6c8 | 1497 | |
4225399a PJ |
1498 | /* |
1499 | * Rebind each vma in mm to new nodemask. | |
1500 | * | |
1501 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
1502 | */ | |
1503 | ||
1504 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
1505 | { | |
1506 | struct vm_area_struct *vma; | |
1507 | ||
1508 | down_write(&mm->mmap_sem); | |
1509 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
1510 | mpol_rebind_policy(vma->vm_policy, new); | |
1511 | up_write(&mm->mmap_sem); | |
1512 | } | |
1513 | ||
1a75a6c8 CL |
1514 | /* |
1515 | * Display pages allocated per node and memory policy via /proc. | |
1516 | */ | |
1517 | ||
1518 | static const char *policy_types[] = { "default", "prefer", "bind", | |
1519 | "interleave" }; | |
1520 | ||
1521 | /* | |
1522 | * Convert a mempolicy into a string. | |
1523 | * Returns the number of characters in buffer (if positive) | |
1524 | * or an error (negative) | |
1525 | */ | |
1526 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1527 | { | |
1528 | char *p = buffer; | |
1529 | int l; | |
1530 | nodemask_t nodes; | |
1531 | int mode = pol ? pol->policy : MPOL_DEFAULT; | |
1532 | ||
1533 | switch (mode) { | |
1534 | case MPOL_DEFAULT: | |
1535 | nodes_clear(nodes); | |
1536 | break; | |
1537 | ||
1538 | case MPOL_PREFERRED: | |
1539 | nodes_clear(nodes); | |
1540 | node_set(pol->v.preferred_node, nodes); | |
1541 | break; | |
1542 | ||
1543 | case MPOL_BIND: | |
1544 | get_zonemask(pol, &nodes); | |
1545 | break; | |
1546 | ||
1547 | case MPOL_INTERLEAVE: | |
1548 | nodes = pol->v.nodes; | |
1549 | break; | |
1550 | ||
1551 | default: | |
1552 | BUG(); | |
1553 | return -EFAULT; | |
1554 | } | |
1555 | ||
1556 | l = strlen(policy_types[mode]); | |
1557 | if (buffer + maxlen < p + l + 1) | |
1558 | return -ENOSPC; | |
1559 | ||
1560 | strcpy(p, policy_types[mode]); | |
1561 | p += l; | |
1562 | ||
1563 | if (!nodes_empty(nodes)) { | |
1564 | if (buffer + maxlen < p + 2) | |
1565 | return -ENOSPC; | |
1566 | *p++ = '='; | |
1567 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
1568 | } | |
1569 | return p - buffer; | |
1570 | } | |
1571 | ||
1572 | struct numa_maps { | |
1573 | unsigned long pages; | |
1574 | unsigned long anon; | |
1575 | unsigned long mapped; | |
1576 | unsigned long mapcount_max; | |
1577 | unsigned long node[MAX_NUMNODES]; | |
1578 | }; | |
1579 | ||
1580 | static void gather_stats(struct page *page, void *private) | |
1581 | { | |
1582 | struct numa_maps *md = private; | |
1583 | int count = page_mapcount(page); | |
1584 | ||
1585 | if (count) | |
1586 | md->mapped++; | |
1587 | ||
1588 | if (count > md->mapcount_max) | |
1589 | md->mapcount_max = count; | |
1590 | ||
1591 | md->pages++; | |
1592 | ||
1593 | if (PageAnon(page)) | |
1594 | md->anon++; | |
1595 | ||
1596 | md->node[page_to_nid(page)]++; | |
1597 | cond_resched(); | |
1598 | } | |
1599 | ||
1600 | int show_numa_map(struct seq_file *m, void *v) | |
1601 | { | |
1602 | struct task_struct *task = m->private; | |
1603 | struct vm_area_struct *vma = v; | |
1604 | struct numa_maps *md; | |
1605 | int n; | |
1606 | char buffer[50]; | |
1607 | ||
1608 | if (!vma->vm_mm) | |
1609 | return 0; | |
1610 | ||
1611 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
1612 | if (!md) | |
1613 | return 0; | |
1614 | ||
1615 | check_pgd_range(vma, vma->vm_start, vma->vm_end, | |
1616 | &node_online_map, MPOL_MF_STATS, md); | |
1617 | ||
1618 | if (md->pages) { | |
1619 | mpol_to_str(buffer, sizeof(buffer), | |
1620 | get_vma_policy(task, vma, vma->vm_start)); | |
1621 | ||
1622 | seq_printf(m, "%08lx %s pages=%lu mapped=%lu maxref=%lu", | |
1623 | vma->vm_start, buffer, md->pages, | |
1624 | md->mapped, md->mapcount_max); | |
1625 | ||
1626 | if (md->anon) | |
1627 | seq_printf(m," anon=%lu",md->anon); | |
1628 | ||
1629 | for_each_online_node(n) | |
1630 | if (md->node[n]) | |
1631 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
1632 | ||
1633 | seq_putc(m, '\n'); | |
1634 | } | |
1635 | kfree(md); | |
1636 | ||
1637 | if (m->count < m->size) | |
1638 | m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; | |
1639 | return 0; | |
1640 | } | |
1641 |