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> | |
b20a3503 | 89 | #include <linux/migrate.h> |
95a402c3 | 90 | #include <linux/rmap.h> |
86c3a764 | 91 | #include <linux/security.h> |
dc9aa5b9 | 92 | |
1da177e4 LT |
93 | #include <asm/tlbflush.h> |
94 | #include <asm/uaccess.h> | |
95 | ||
38e35860 | 96 | /* Internal flags */ |
dc9aa5b9 | 97 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 98 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
1a75a6c8 | 99 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
dc9aa5b9 | 100 | |
fcc234f8 PE |
101 | static struct kmem_cache *policy_cache; |
102 | static struct kmem_cache *sn_cache; | |
1da177e4 | 103 | |
1da177e4 LT |
104 | /* Highest zone. An specific allocation for a zone below that is not |
105 | policied. */ | |
6267276f | 106 | enum zone_type policy_zone = 0; |
1da177e4 | 107 | |
d42c6997 | 108 | struct mempolicy default_policy = { |
1da177e4 LT |
109 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
110 | .policy = MPOL_DEFAULT, | |
111 | }; | |
112 | ||
1da177e4 | 113 | /* Do sanity checking on a policy */ |
dfcd3c0d | 114 | static int mpol_check_policy(int mode, nodemask_t *nodes) |
1da177e4 | 115 | { |
dfcd3c0d | 116 | int empty = nodes_empty(*nodes); |
1da177e4 LT |
117 | |
118 | switch (mode) { | |
119 | case MPOL_DEFAULT: | |
120 | if (!empty) | |
121 | return -EINVAL; | |
122 | break; | |
123 | case MPOL_BIND: | |
124 | case MPOL_INTERLEAVE: | |
125 | /* Preferred will only use the first bit, but allow | |
126 | more for now. */ | |
127 | if (empty) | |
128 | return -EINVAL; | |
129 | break; | |
130 | } | |
dfcd3c0d | 131 | return nodes_subset(*nodes, node_online_map) ? 0 : -EINVAL; |
1da177e4 | 132 | } |
dd942ae3 | 133 | |
1da177e4 | 134 | /* Generate a custom zonelist for the BIND policy. */ |
dfcd3c0d | 135 | static struct zonelist *bind_zonelist(nodemask_t *nodes) |
1da177e4 LT |
136 | { |
137 | struct zonelist *zl; | |
2f6726e5 CL |
138 | int num, max, nd; |
139 | enum zone_type k; | |
1da177e4 | 140 | |
dfcd3c0d | 141 | max = 1 + MAX_NR_ZONES * nodes_weight(*nodes); |
9276b1bc | 142 | max++; /* space for zlcache_ptr (see mmzone.h) */ |
dd942ae3 | 143 | zl = kmalloc(sizeof(struct zone *) * max, GFP_KERNEL); |
1da177e4 | 144 | if (!zl) |
8af5e2eb | 145 | return ERR_PTR(-ENOMEM); |
9276b1bc | 146 | zl->zlcache_ptr = NULL; |
1da177e4 | 147 | num = 0; |
dd942ae3 AK |
148 | /* First put in the highest zones from all nodes, then all the next |
149 | lower zones etc. Avoid empty zones because the memory allocator | |
150 | doesn't like them. If you implement node hot removal you | |
151 | have to fix that. */ | |
b377fd39 | 152 | k = MAX_NR_ZONES - 1; |
2f6726e5 | 153 | while (1) { |
dd942ae3 AK |
154 | for_each_node_mask(nd, *nodes) { |
155 | struct zone *z = &NODE_DATA(nd)->node_zones[k]; | |
156 | if (z->present_pages > 0) | |
157 | zl->zones[num++] = z; | |
158 | } | |
2f6726e5 CL |
159 | if (k == 0) |
160 | break; | |
161 | k--; | |
dd942ae3 | 162 | } |
8af5e2eb KH |
163 | if (num == 0) { |
164 | kfree(zl); | |
165 | return ERR_PTR(-EINVAL); | |
166 | } | |
1da177e4 LT |
167 | zl->zones[num] = NULL; |
168 | return zl; | |
169 | } | |
170 | ||
171 | /* Create a new policy */ | |
dfcd3c0d | 172 | static struct mempolicy *mpol_new(int mode, nodemask_t *nodes) |
1da177e4 LT |
173 | { |
174 | struct mempolicy *policy; | |
175 | ||
140d5a49 PM |
176 | pr_debug("setting mode %d nodes[0] %lx\n", |
177 | mode, nodes ? nodes_addr(*nodes)[0] : -1); | |
178 | ||
1da177e4 LT |
179 | if (mode == MPOL_DEFAULT) |
180 | return NULL; | |
181 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
182 | if (!policy) | |
183 | return ERR_PTR(-ENOMEM); | |
184 | atomic_set(&policy->refcnt, 1); | |
185 | switch (mode) { | |
186 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 187 | policy->v.nodes = *nodes; |
8f493d79 AK |
188 | if (nodes_weight(*nodes) == 0) { |
189 | kmem_cache_free(policy_cache, policy); | |
190 | return ERR_PTR(-EINVAL); | |
191 | } | |
1da177e4 LT |
192 | break; |
193 | case MPOL_PREFERRED: | |
dfcd3c0d | 194 | policy->v.preferred_node = first_node(*nodes); |
1da177e4 LT |
195 | if (policy->v.preferred_node >= MAX_NUMNODES) |
196 | policy->v.preferred_node = -1; | |
197 | break; | |
198 | case MPOL_BIND: | |
199 | policy->v.zonelist = bind_zonelist(nodes); | |
8af5e2eb KH |
200 | if (IS_ERR(policy->v.zonelist)) { |
201 | void *error_code = policy->v.zonelist; | |
1da177e4 | 202 | kmem_cache_free(policy_cache, policy); |
8af5e2eb | 203 | return error_code; |
1da177e4 LT |
204 | } |
205 | break; | |
206 | } | |
207 | policy->policy = mode; | |
74cb2155 | 208 | policy->cpuset_mems_allowed = cpuset_mems_allowed(current); |
1da177e4 LT |
209 | return policy; |
210 | } | |
211 | ||
397874df | 212 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
213 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
214 | unsigned long flags); | |
1a75a6c8 | 215 | |
38e35860 | 216 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 217 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
218 | unsigned long addr, unsigned long end, |
219 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 220 | void *private) |
1da177e4 | 221 | { |
91612e0d HD |
222 | pte_t *orig_pte; |
223 | pte_t *pte; | |
705e87c0 | 224 | spinlock_t *ptl; |
941150a3 | 225 | |
705e87c0 | 226 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 227 | do { |
6aab341e | 228 | struct page *page; |
25ba77c1 | 229 | int nid; |
91612e0d HD |
230 | |
231 | if (!pte_present(*pte)) | |
1da177e4 | 232 | continue; |
6aab341e LT |
233 | page = vm_normal_page(vma, addr, *pte); |
234 | if (!page) | |
1da177e4 | 235 | continue; |
053837fc NP |
236 | /* |
237 | * The check for PageReserved here is important to avoid | |
238 | * handling zero pages and other pages that may have been | |
239 | * marked special by the system. | |
240 | * | |
241 | * If the PageReserved would not be checked here then f.e. | |
242 | * the location of the zero page could have an influence | |
243 | * on MPOL_MF_STRICT, zero pages would be counted for | |
244 | * the per node stats, and there would be useless attempts | |
245 | * to put zero pages on the migration list. | |
246 | */ | |
f4598c8b CL |
247 | if (PageReserved(page)) |
248 | continue; | |
6aab341e | 249 | nid = page_to_nid(page); |
38e35860 CL |
250 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
251 | continue; | |
252 | ||
1a75a6c8 | 253 | if (flags & MPOL_MF_STATS) |
397874df | 254 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 255 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 256 | migrate_page_add(page, private, flags); |
38e35860 CL |
257 | else |
258 | break; | |
91612e0d | 259 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 260 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
261 | return addr != end; |
262 | } | |
263 | ||
b5810039 | 264 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
265 | unsigned long addr, unsigned long end, |
266 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 267 | void *private) |
91612e0d HD |
268 | { |
269 | pmd_t *pmd; | |
270 | unsigned long next; | |
271 | ||
272 | pmd = pmd_offset(pud, addr); | |
273 | do { | |
274 | next = pmd_addr_end(addr, end); | |
275 | if (pmd_none_or_clear_bad(pmd)) | |
276 | continue; | |
dc9aa5b9 | 277 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 278 | flags, private)) |
91612e0d HD |
279 | return -EIO; |
280 | } while (pmd++, addr = next, addr != end); | |
281 | return 0; | |
282 | } | |
283 | ||
b5810039 | 284 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
285 | unsigned long addr, unsigned long end, |
286 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 287 | void *private) |
91612e0d HD |
288 | { |
289 | pud_t *pud; | |
290 | unsigned long next; | |
291 | ||
292 | pud = pud_offset(pgd, addr); | |
293 | do { | |
294 | next = pud_addr_end(addr, end); | |
295 | if (pud_none_or_clear_bad(pud)) | |
296 | continue; | |
dc9aa5b9 | 297 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 298 | flags, private)) |
91612e0d HD |
299 | return -EIO; |
300 | } while (pud++, addr = next, addr != end); | |
301 | return 0; | |
302 | } | |
303 | ||
b5810039 | 304 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
305 | unsigned long addr, unsigned long end, |
306 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 307 | void *private) |
91612e0d HD |
308 | { |
309 | pgd_t *pgd; | |
310 | unsigned long next; | |
311 | ||
b5810039 | 312 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
313 | do { |
314 | next = pgd_addr_end(addr, end); | |
315 | if (pgd_none_or_clear_bad(pgd)) | |
316 | continue; | |
dc9aa5b9 | 317 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 318 | flags, private)) |
91612e0d HD |
319 | return -EIO; |
320 | } while (pgd++, addr = next, addr != end); | |
321 | return 0; | |
1da177e4 LT |
322 | } |
323 | ||
dc9aa5b9 CL |
324 | /* |
325 | * Check if all pages in a range are on a set of nodes. | |
326 | * If pagelist != NULL then isolate pages from the LRU and | |
327 | * put them on the pagelist. | |
328 | */ | |
1da177e4 LT |
329 | static struct vm_area_struct * |
330 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 331 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
332 | { |
333 | int err; | |
334 | struct vm_area_struct *first, *vma, *prev; | |
335 | ||
90036ee5 | 336 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
90036ee5 | 337 | |
b20a3503 CL |
338 | err = migrate_prep(); |
339 | if (err) | |
340 | return ERR_PTR(err); | |
90036ee5 | 341 | } |
053837fc | 342 | |
1da177e4 LT |
343 | first = find_vma(mm, start); |
344 | if (!first) | |
345 | return ERR_PTR(-EFAULT); | |
346 | prev = NULL; | |
347 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
348 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
349 | if (!vma->vm_next && vma->vm_end < end) | |
350 | return ERR_PTR(-EFAULT); | |
351 | if (prev && prev->vm_end < vma->vm_start) | |
352 | return ERR_PTR(-EFAULT); | |
353 | } | |
354 | if (!is_vm_hugetlb_page(vma) && | |
355 | ((flags & MPOL_MF_STRICT) || | |
356 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
357 | vma_migratable(vma)))) { | |
5b952b3c | 358 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 359 | |
5b952b3c AK |
360 | if (endvma > end) |
361 | endvma = end; | |
362 | if (vma->vm_start > start) | |
363 | start = vma->vm_start; | |
dc9aa5b9 | 364 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 365 | flags, private); |
1da177e4 LT |
366 | if (err) { |
367 | first = ERR_PTR(err); | |
368 | break; | |
369 | } | |
370 | } | |
371 | prev = vma; | |
372 | } | |
373 | return first; | |
374 | } | |
375 | ||
376 | /* Apply policy to a single VMA */ | |
377 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
378 | { | |
379 | int err = 0; | |
380 | struct mempolicy *old = vma->vm_policy; | |
381 | ||
140d5a49 | 382 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
383 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
384 | vma->vm_ops, vma->vm_file, | |
385 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
386 | ||
387 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
388 | err = vma->vm_ops->set_policy(vma, new); | |
389 | if (!err) { | |
390 | mpol_get(new); | |
391 | vma->vm_policy = new; | |
392 | mpol_free(old); | |
393 | } | |
394 | return err; | |
395 | } | |
396 | ||
397 | /* Step 2: apply policy to a range and do splits. */ | |
398 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
399 | unsigned long end, struct mempolicy *new) | |
400 | { | |
401 | struct vm_area_struct *next; | |
402 | int err; | |
403 | ||
404 | err = 0; | |
405 | for (; vma && vma->vm_start < end; vma = next) { | |
406 | next = vma->vm_next; | |
407 | if (vma->vm_start < start) | |
408 | err = split_vma(vma->vm_mm, vma, start, 1); | |
409 | if (!err && vma->vm_end > end) | |
410 | err = split_vma(vma->vm_mm, vma, end, 0); | |
411 | if (!err) | |
412 | err = policy_vma(vma, new); | |
413 | if (err) | |
414 | break; | |
415 | } | |
416 | return err; | |
417 | } | |
418 | ||
8bccd85f CL |
419 | static int contextualize_policy(int mode, nodemask_t *nodes) |
420 | { | |
421 | if (!nodes) | |
422 | return 0; | |
423 | ||
cf2a473c | 424 | cpuset_update_task_memory_state(); |
5966514d PJ |
425 | if (!cpuset_nodes_subset_current_mems_allowed(*nodes)) |
426 | return -EINVAL; | |
8bccd85f CL |
427 | return mpol_check_policy(mode, nodes); |
428 | } | |
429 | ||
c61afb18 PJ |
430 | |
431 | /* | |
432 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
433 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
434 | * with other PF_* flag bits) on memory allocation hot code paths. | |
435 | * | |
436 | * If called from outside this file, the task 'p' should -only- be | |
437 | * a newly forked child not yet visible on the task list, because | |
438 | * manipulating the task flags of a visible task is not safe. | |
439 | * | |
440 | * The above limitation is why this routine has the funny name | |
441 | * mpol_fix_fork_child_flag(). | |
442 | * | |
443 | * It is also safe to call this with a task pointer of current, | |
444 | * which the static wrapper mpol_set_task_struct_flag() does, | |
445 | * for use within this file. | |
446 | */ | |
447 | ||
448 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
449 | { | |
450 | if (p->mempolicy) | |
451 | p->flags |= PF_MEMPOLICY; | |
452 | else | |
453 | p->flags &= ~PF_MEMPOLICY; | |
454 | } | |
455 | ||
456 | static void mpol_set_task_struct_flag(void) | |
457 | { | |
458 | mpol_fix_fork_child_flag(current); | |
459 | } | |
460 | ||
1da177e4 | 461 | /* Set the process memory policy */ |
8bccd85f | 462 | long do_set_mempolicy(int mode, nodemask_t *nodes) |
1da177e4 | 463 | { |
1da177e4 | 464 | struct mempolicy *new; |
1da177e4 | 465 | |
8bccd85f | 466 | if (contextualize_policy(mode, nodes)) |
1da177e4 | 467 | return -EINVAL; |
8bccd85f | 468 | new = mpol_new(mode, nodes); |
1da177e4 LT |
469 | if (IS_ERR(new)) |
470 | return PTR_ERR(new); | |
471 | mpol_free(current->mempolicy); | |
472 | current->mempolicy = new; | |
c61afb18 | 473 | mpol_set_task_struct_flag(); |
1da177e4 | 474 | if (new && new->policy == MPOL_INTERLEAVE) |
dfcd3c0d | 475 | current->il_next = first_node(new->v.nodes); |
1da177e4 LT |
476 | return 0; |
477 | } | |
478 | ||
479 | /* Fill a zone bitmap for a policy */ | |
dfcd3c0d | 480 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
1da177e4 LT |
481 | { |
482 | int i; | |
483 | ||
dfcd3c0d | 484 | nodes_clear(*nodes); |
1da177e4 LT |
485 | switch (p->policy) { |
486 | case MPOL_BIND: | |
487 | for (i = 0; p->v.zonelist->zones[i]; i++) | |
89fa3024 | 488 | node_set(zone_to_nid(p->v.zonelist->zones[i]), |
8bccd85f | 489 | *nodes); |
1da177e4 LT |
490 | break; |
491 | case MPOL_DEFAULT: | |
492 | break; | |
493 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 494 | *nodes = p->v.nodes; |
1da177e4 LT |
495 | break; |
496 | case MPOL_PREFERRED: | |
497 | /* or use current node instead of online map? */ | |
498 | if (p->v.preferred_node < 0) | |
dfcd3c0d | 499 | *nodes = node_online_map; |
1da177e4 | 500 | else |
dfcd3c0d | 501 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
502 | break; |
503 | default: | |
504 | BUG(); | |
505 | } | |
506 | } | |
507 | ||
508 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
509 | { | |
510 | struct page *p; | |
511 | int err; | |
512 | ||
513 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
514 | if (err >= 0) { | |
515 | err = page_to_nid(p); | |
516 | put_page(p); | |
517 | } | |
518 | return err; | |
519 | } | |
520 | ||
1da177e4 | 521 | /* Retrieve NUMA policy */ |
8bccd85f CL |
522 | long do_get_mempolicy(int *policy, nodemask_t *nmask, |
523 | unsigned long addr, unsigned long flags) | |
1da177e4 | 524 | { |
8bccd85f | 525 | int err; |
1da177e4 LT |
526 | struct mm_struct *mm = current->mm; |
527 | struct vm_area_struct *vma = NULL; | |
528 | struct mempolicy *pol = current->mempolicy; | |
529 | ||
cf2a473c | 530 | cpuset_update_task_memory_state(); |
1da177e4 LT |
531 | if (flags & ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR)) |
532 | return -EINVAL; | |
1da177e4 LT |
533 | if (flags & MPOL_F_ADDR) { |
534 | down_read(&mm->mmap_sem); | |
535 | vma = find_vma_intersection(mm, addr, addr+1); | |
536 | if (!vma) { | |
537 | up_read(&mm->mmap_sem); | |
538 | return -EFAULT; | |
539 | } | |
540 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
541 | pol = vma->vm_ops->get_policy(vma, addr); | |
542 | else | |
543 | pol = vma->vm_policy; | |
544 | } else if (addr) | |
545 | return -EINVAL; | |
546 | ||
547 | if (!pol) | |
548 | pol = &default_policy; | |
549 | ||
550 | if (flags & MPOL_F_NODE) { | |
551 | if (flags & MPOL_F_ADDR) { | |
552 | err = lookup_node(mm, addr); | |
553 | if (err < 0) | |
554 | goto out; | |
8bccd85f | 555 | *policy = err; |
1da177e4 LT |
556 | } else if (pol == current->mempolicy && |
557 | pol->policy == MPOL_INTERLEAVE) { | |
8bccd85f | 558 | *policy = current->il_next; |
1da177e4 LT |
559 | } else { |
560 | err = -EINVAL; | |
561 | goto out; | |
562 | } | |
563 | } else | |
8bccd85f | 564 | *policy = pol->policy; |
1da177e4 LT |
565 | |
566 | if (vma) { | |
567 | up_read(¤t->mm->mmap_sem); | |
568 | vma = NULL; | |
569 | } | |
570 | ||
1da177e4 | 571 | err = 0; |
8bccd85f CL |
572 | if (nmask) |
573 | get_zonemask(pol, nmask); | |
1da177e4 LT |
574 | |
575 | out: | |
576 | if (vma) | |
577 | up_read(¤t->mm->mmap_sem); | |
578 | return err; | |
579 | } | |
580 | ||
b20a3503 | 581 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
582 | /* |
583 | * page migration | |
584 | */ | |
fc301289 CL |
585 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
586 | unsigned long flags) | |
6ce3c4c0 CL |
587 | { |
588 | /* | |
fc301289 | 589 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 590 | */ |
b20a3503 CL |
591 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
592 | isolate_lru_page(page, pagelist); | |
7e2ab150 | 593 | } |
6ce3c4c0 | 594 | |
742755a1 | 595 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 596 | { |
769848c0 | 597 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
598 | } |
599 | ||
7e2ab150 CL |
600 | /* |
601 | * Migrate pages from one node to a target node. | |
602 | * Returns error or the number of pages not migrated. | |
603 | */ | |
604 | int migrate_to_node(struct mm_struct *mm, int source, int dest, int flags) | |
605 | { | |
606 | nodemask_t nmask; | |
607 | LIST_HEAD(pagelist); | |
608 | int err = 0; | |
609 | ||
610 | nodes_clear(nmask); | |
611 | node_set(source, nmask); | |
6ce3c4c0 | 612 | |
7e2ab150 CL |
613 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
614 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
615 | ||
aaa994b3 | 616 | if (!list_empty(&pagelist)) |
95a402c3 CL |
617 | err = migrate_pages(&pagelist, new_node_page, dest); |
618 | ||
7e2ab150 | 619 | return err; |
6ce3c4c0 CL |
620 | } |
621 | ||
39743889 | 622 | /* |
7e2ab150 CL |
623 | * Move pages between the two nodesets so as to preserve the physical |
624 | * layout as much as possible. | |
39743889 CL |
625 | * |
626 | * Returns the number of page that could not be moved. | |
627 | */ | |
628 | int do_migrate_pages(struct mm_struct *mm, | |
629 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
630 | { | |
631 | LIST_HEAD(pagelist); | |
7e2ab150 CL |
632 | int busy = 0; |
633 | int err = 0; | |
634 | nodemask_t tmp; | |
39743889 | 635 | |
7e2ab150 | 636 | down_read(&mm->mmap_sem); |
39743889 | 637 | |
7b2259b3 CL |
638 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
639 | if (err) | |
640 | goto out; | |
641 | ||
7e2ab150 CL |
642 | /* |
643 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
644 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
645 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
646 | * The pair of nodemasks 'to' and 'from' define the map. | |
647 | * | |
648 | * If no pair of bits is found that way, fallback to picking some | |
649 | * pair of 'source' and 'dest' bits that are not the same. If the | |
650 | * 'source' and 'dest' bits are the same, this represents a node | |
651 | * that will be migrating to itself, so no pages need move. | |
652 | * | |
653 | * If no bits are left in 'tmp', or if all remaining bits left | |
654 | * in 'tmp' correspond to the same bit in 'to', return false | |
655 | * (nothing left to migrate). | |
656 | * | |
657 | * This lets us pick a pair of nodes to migrate between, such that | |
658 | * if possible the dest node is not already occupied by some other | |
659 | * source node, minimizing the risk of overloading the memory on a | |
660 | * node that would happen if we migrated incoming memory to a node | |
661 | * before migrating outgoing memory source that same node. | |
662 | * | |
663 | * A single scan of tmp is sufficient. As we go, we remember the | |
664 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
665 | * that not only moved, but what's better, moved to an empty slot | |
666 | * (d is not set in tmp), then we break out then, with that pair. | |
667 | * Otherwise when we finish scannng from_tmp, we at least have the | |
668 | * most recent <s, d> pair that moved. If we get all the way through | |
669 | * the scan of tmp without finding any node that moved, much less | |
670 | * moved to an empty node, then there is nothing left worth migrating. | |
671 | */ | |
d4984711 | 672 | |
7e2ab150 CL |
673 | tmp = *from_nodes; |
674 | while (!nodes_empty(tmp)) { | |
675 | int s,d; | |
676 | int source = -1; | |
677 | int dest = 0; | |
678 | ||
679 | for_each_node_mask(s, tmp) { | |
680 | d = node_remap(s, *from_nodes, *to_nodes); | |
681 | if (s == d) | |
682 | continue; | |
683 | ||
684 | source = s; /* Node moved. Memorize */ | |
685 | dest = d; | |
686 | ||
687 | /* dest not in remaining from nodes? */ | |
688 | if (!node_isset(dest, tmp)) | |
689 | break; | |
690 | } | |
691 | if (source == -1) | |
692 | break; | |
693 | ||
694 | node_clear(source, tmp); | |
695 | err = migrate_to_node(mm, source, dest, flags); | |
696 | if (err > 0) | |
697 | busy += err; | |
698 | if (err < 0) | |
699 | break; | |
39743889 | 700 | } |
7b2259b3 | 701 | out: |
39743889 | 702 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
703 | if (err < 0) |
704 | return err; | |
705 | return busy; | |
b20a3503 CL |
706 | |
707 | } | |
708 | ||
742755a1 | 709 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
710 | { |
711 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
712 | ||
769848c0 MG |
713 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
714 | page_address_in_vma(page, vma)); | |
95a402c3 | 715 | } |
b20a3503 CL |
716 | #else |
717 | ||
718 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
719 | unsigned long flags) | |
720 | { | |
39743889 CL |
721 | } |
722 | ||
b20a3503 CL |
723 | int do_migrate_pages(struct mm_struct *mm, |
724 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
725 | { | |
726 | return -ENOSYS; | |
727 | } | |
95a402c3 | 728 | |
69939749 | 729 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
730 | { |
731 | return NULL; | |
732 | } | |
b20a3503 CL |
733 | #endif |
734 | ||
6ce3c4c0 CL |
735 | long do_mbind(unsigned long start, unsigned long len, |
736 | unsigned long mode, nodemask_t *nmask, unsigned long flags) | |
737 | { | |
738 | struct vm_area_struct *vma; | |
739 | struct mm_struct *mm = current->mm; | |
740 | struct mempolicy *new; | |
741 | unsigned long end; | |
742 | int err; | |
743 | LIST_HEAD(pagelist); | |
744 | ||
745 | if ((flags & ~(unsigned long)(MPOL_MF_STRICT | | |
746 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
747 | || mode > MPOL_MAX) | |
748 | return -EINVAL; | |
74c00241 | 749 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
750 | return -EPERM; |
751 | ||
752 | if (start & ~PAGE_MASK) | |
753 | return -EINVAL; | |
754 | ||
755 | if (mode == MPOL_DEFAULT) | |
756 | flags &= ~MPOL_MF_STRICT; | |
757 | ||
758 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
759 | end = start + len; | |
760 | ||
761 | if (end < start) | |
762 | return -EINVAL; | |
763 | if (end == start) | |
764 | return 0; | |
765 | ||
766 | if (mpol_check_policy(mode, nmask)) | |
767 | return -EINVAL; | |
768 | ||
769 | new = mpol_new(mode, nmask); | |
770 | if (IS_ERR(new)) | |
771 | return PTR_ERR(new); | |
772 | ||
773 | /* | |
774 | * If we are using the default policy then operation | |
775 | * on discontinuous address spaces is okay after all | |
776 | */ | |
777 | if (!new) | |
778 | flags |= MPOL_MF_DISCONTIG_OK; | |
779 | ||
140d5a49 PM |
780 | pr_debug("mbind %lx-%lx mode:%ld nodes:%lx\n",start,start+len, |
781 | mode, nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 CL |
782 | |
783 | down_write(&mm->mmap_sem); | |
784 | vma = check_range(mm, start, end, nmask, | |
785 | flags | MPOL_MF_INVERT, &pagelist); | |
786 | ||
787 | err = PTR_ERR(vma); | |
788 | if (!IS_ERR(vma)) { | |
789 | int nr_failed = 0; | |
790 | ||
791 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 792 | |
6ce3c4c0 | 793 | if (!list_empty(&pagelist)) |
95a402c3 CL |
794 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
795 | (unsigned long)vma); | |
6ce3c4c0 CL |
796 | |
797 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
798 | err = -EIO; | |
799 | } | |
b20a3503 | 800 | |
6ce3c4c0 CL |
801 | up_write(&mm->mmap_sem); |
802 | mpol_free(new); | |
803 | return err; | |
804 | } | |
805 | ||
8bccd85f CL |
806 | /* |
807 | * User space interface with variable sized bitmaps for nodelists. | |
808 | */ | |
809 | ||
810 | /* Copy a node mask from user space. */ | |
39743889 | 811 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
812 | unsigned long maxnode) |
813 | { | |
814 | unsigned long k; | |
815 | unsigned long nlongs; | |
816 | unsigned long endmask; | |
817 | ||
818 | --maxnode; | |
819 | nodes_clear(*nodes); | |
820 | if (maxnode == 0 || !nmask) | |
821 | return 0; | |
a9c930ba | 822 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 823 | return -EINVAL; |
8bccd85f CL |
824 | |
825 | nlongs = BITS_TO_LONGS(maxnode); | |
826 | if ((maxnode % BITS_PER_LONG) == 0) | |
827 | endmask = ~0UL; | |
828 | else | |
829 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
830 | ||
831 | /* When the user specified more nodes than supported just check | |
832 | if the non supported part is all zero. */ | |
833 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
834 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
835 | return -EINVAL; | |
836 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
837 | unsigned long t; | |
838 | if (get_user(t, nmask + k)) | |
839 | return -EFAULT; | |
840 | if (k == nlongs - 1) { | |
841 | if (t & endmask) | |
842 | return -EINVAL; | |
843 | } else if (t) | |
844 | return -EINVAL; | |
845 | } | |
846 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
847 | endmask = ~0UL; | |
848 | } | |
849 | ||
850 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
851 | return -EFAULT; | |
852 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
853 | return 0; | |
854 | } | |
855 | ||
856 | /* Copy a kernel node mask to user space */ | |
857 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
858 | nodemask_t *nodes) | |
859 | { | |
860 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
861 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
862 | ||
863 | if (copy > nbytes) { | |
864 | if (copy > PAGE_SIZE) | |
865 | return -EINVAL; | |
866 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
867 | return -EFAULT; | |
868 | copy = nbytes; | |
869 | } | |
870 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
871 | } | |
872 | ||
873 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
874 | unsigned long mode, | |
875 | unsigned long __user *nmask, unsigned long maxnode, | |
876 | unsigned flags) | |
877 | { | |
878 | nodemask_t nodes; | |
879 | int err; | |
880 | ||
881 | err = get_nodes(&nodes, nmask, maxnode); | |
882 | if (err) | |
883 | return err; | |
30150f8d CL |
884 | #ifdef CONFIG_CPUSETS |
885 | /* Restrict the nodes to the allowed nodes in the cpuset */ | |
886 | nodes_and(nodes, nodes, current->mems_allowed); | |
887 | #endif | |
8bccd85f CL |
888 | return do_mbind(start, len, mode, &nodes, flags); |
889 | } | |
890 | ||
891 | /* Set the process memory policy */ | |
892 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
893 | unsigned long maxnode) | |
894 | { | |
895 | int err; | |
896 | nodemask_t nodes; | |
897 | ||
898 | if (mode < 0 || mode > MPOL_MAX) | |
899 | return -EINVAL; | |
900 | err = get_nodes(&nodes, nmask, maxnode); | |
901 | if (err) | |
902 | return err; | |
903 | return do_set_mempolicy(mode, &nodes); | |
904 | } | |
905 | ||
39743889 CL |
906 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
907 | const unsigned long __user *old_nodes, | |
908 | const unsigned long __user *new_nodes) | |
909 | { | |
910 | struct mm_struct *mm; | |
911 | struct task_struct *task; | |
912 | nodemask_t old; | |
913 | nodemask_t new; | |
914 | nodemask_t task_nodes; | |
915 | int err; | |
916 | ||
917 | err = get_nodes(&old, old_nodes, maxnode); | |
918 | if (err) | |
919 | return err; | |
920 | ||
921 | err = get_nodes(&new, new_nodes, maxnode); | |
922 | if (err) | |
923 | return err; | |
924 | ||
925 | /* Find the mm_struct */ | |
926 | read_lock(&tasklist_lock); | |
927 | task = pid ? find_task_by_pid(pid) : current; | |
928 | if (!task) { | |
929 | read_unlock(&tasklist_lock); | |
930 | return -ESRCH; | |
931 | } | |
932 | mm = get_task_mm(task); | |
933 | read_unlock(&tasklist_lock); | |
934 | ||
935 | if (!mm) | |
936 | return -EINVAL; | |
937 | ||
938 | /* | |
939 | * Check if this process has the right to modify the specified | |
940 | * process. The right exists if the process has administrative | |
7f927fcc | 941 | * capabilities, superuser privileges or the same |
39743889 CL |
942 | * userid as the target process. |
943 | */ | |
944 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
945 | (current->uid != task->suid) && (current->uid != task->uid) && | |
74c00241 | 946 | !capable(CAP_SYS_NICE)) { |
39743889 CL |
947 | err = -EPERM; |
948 | goto out; | |
949 | } | |
950 | ||
951 | task_nodes = cpuset_mems_allowed(task); | |
952 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 953 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
954 | err = -EPERM; |
955 | goto out; | |
956 | } | |
957 | ||
86c3a764 DQ |
958 | err = security_task_movememory(task); |
959 | if (err) | |
960 | goto out; | |
961 | ||
511030bc | 962 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 963 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
964 | out: |
965 | mmput(mm); | |
966 | return err; | |
967 | } | |
968 | ||
969 | ||
8bccd85f CL |
970 | /* Retrieve NUMA policy */ |
971 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
972 | unsigned long __user *nmask, | |
973 | unsigned long maxnode, | |
974 | unsigned long addr, unsigned long flags) | |
975 | { | |
976 | int err, pval; | |
977 | nodemask_t nodes; | |
978 | ||
979 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
980 | return -EINVAL; | |
981 | ||
982 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
983 | ||
984 | if (err) | |
985 | return err; | |
986 | ||
987 | if (policy && put_user(pval, policy)) | |
988 | return -EFAULT; | |
989 | ||
990 | if (nmask) | |
991 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
992 | ||
993 | return err; | |
994 | } | |
995 | ||
1da177e4 LT |
996 | #ifdef CONFIG_COMPAT |
997 | ||
998 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
999 | compat_ulong_t __user *nmask, | |
1000 | compat_ulong_t maxnode, | |
1001 | compat_ulong_t addr, compat_ulong_t flags) | |
1002 | { | |
1003 | long err; | |
1004 | unsigned long __user *nm = NULL; | |
1005 | unsigned long nr_bits, alloc_size; | |
1006 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1007 | ||
1008 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1009 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1010 | ||
1011 | if (nmask) | |
1012 | nm = compat_alloc_user_space(alloc_size); | |
1013 | ||
1014 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1015 | ||
1016 | if (!err && nmask) { | |
1017 | err = copy_from_user(bm, nm, alloc_size); | |
1018 | /* ensure entire bitmap is zeroed */ | |
1019 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1020 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1021 | } | |
1022 | ||
1023 | return err; | |
1024 | } | |
1025 | ||
1026 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1027 | compat_ulong_t maxnode) | |
1028 | { | |
1029 | long err = 0; | |
1030 | unsigned long __user *nm = NULL; | |
1031 | unsigned long nr_bits, alloc_size; | |
1032 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1033 | ||
1034 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1035 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1036 | ||
1037 | if (nmask) { | |
1038 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1039 | nm = compat_alloc_user_space(alloc_size); | |
1040 | err |= copy_to_user(nm, bm, alloc_size); | |
1041 | } | |
1042 | ||
1043 | if (err) | |
1044 | return -EFAULT; | |
1045 | ||
1046 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1047 | } | |
1048 | ||
1049 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1050 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1051 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1052 | { | |
1053 | long err = 0; | |
1054 | unsigned long __user *nm = NULL; | |
1055 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1056 | nodemask_t bm; |
1da177e4 LT |
1057 | |
1058 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1059 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1060 | ||
1061 | if (nmask) { | |
dfcd3c0d | 1062 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1063 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1064 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1065 | } |
1066 | ||
1067 | if (err) | |
1068 | return -EFAULT; | |
1069 | ||
1070 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1071 | } | |
1072 | ||
1073 | #endif | |
1074 | ||
1075 | /* Return effective policy for a VMA */ | |
48fce342 CL |
1076 | static struct mempolicy * get_vma_policy(struct task_struct *task, |
1077 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 1078 | { |
6e21c8f1 | 1079 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1080 | |
1081 | if (vma) { | |
1082 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
8bccd85f | 1083 | pol = vma->vm_ops->get_policy(vma, addr); |
1da177e4 LT |
1084 | else if (vma->vm_policy && |
1085 | vma->vm_policy->policy != MPOL_DEFAULT) | |
1086 | pol = vma->vm_policy; | |
1087 | } | |
1088 | if (!pol) | |
1089 | pol = &default_policy; | |
1090 | return pol; | |
1091 | } | |
1092 | ||
1093 | /* Return a zonelist representing a mempolicy */ | |
dd0fc66f | 1094 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
1da177e4 LT |
1095 | { |
1096 | int nd; | |
1097 | ||
1098 | switch (policy->policy) { | |
1099 | case MPOL_PREFERRED: | |
1100 | nd = policy->v.preferred_node; | |
1101 | if (nd < 0) | |
1102 | nd = numa_node_id(); | |
1103 | break; | |
1104 | case MPOL_BIND: | |
1105 | /* Lower zones don't get a policy applied */ | |
1106 | /* Careful: current->mems_allowed might have moved */ | |
19655d34 | 1107 | if (gfp_zone(gfp) >= policy_zone) |
1da177e4 LT |
1108 | if (cpuset_zonelist_valid_mems_allowed(policy->v.zonelist)) |
1109 | return policy->v.zonelist; | |
1110 | /*FALL THROUGH*/ | |
1111 | case MPOL_INTERLEAVE: /* should not happen */ | |
1112 | case MPOL_DEFAULT: | |
1113 | nd = numa_node_id(); | |
1114 | break; | |
1115 | default: | |
1116 | nd = 0; | |
1117 | BUG(); | |
1118 | } | |
af4ca457 | 1119 | return NODE_DATA(nd)->node_zonelists + gfp_zone(gfp); |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | /* Do dynamic interleaving for a process */ | |
1123 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1124 | { | |
1125 | unsigned nid, next; | |
1126 | struct task_struct *me = current; | |
1127 | ||
1128 | nid = me->il_next; | |
dfcd3c0d | 1129 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1130 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1131 | next = first_node(policy->v.nodes); |
1da177e4 LT |
1132 | me->il_next = next; |
1133 | return nid; | |
1134 | } | |
1135 | ||
dc85da15 CL |
1136 | /* |
1137 | * Depending on the memory policy provide a node from which to allocate the | |
1138 | * next slab entry. | |
1139 | */ | |
1140 | unsigned slab_node(struct mempolicy *policy) | |
1141 | { | |
765c4507 CL |
1142 | int pol = policy ? policy->policy : MPOL_DEFAULT; |
1143 | ||
1144 | switch (pol) { | |
dc85da15 CL |
1145 | case MPOL_INTERLEAVE: |
1146 | return interleave_nodes(policy); | |
1147 | ||
1148 | case MPOL_BIND: | |
1149 | /* | |
1150 | * Follow bind policy behavior and start allocation at the | |
1151 | * first node. | |
1152 | */ | |
89fa3024 | 1153 | return zone_to_nid(policy->v.zonelist->zones[0]); |
dc85da15 CL |
1154 | |
1155 | case MPOL_PREFERRED: | |
1156 | if (policy->v.preferred_node >= 0) | |
1157 | return policy->v.preferred_node; | |
1158 | /* Fall through */ | |
1159 | ||
1160 | default: | |
1161 | return numa_node_id(); | |
1162 | } | |
1163 | } | |
1164 | ||
1da177e4 LT |
1165 | /* Do static interleaving for a VMA with known offset. */ |
1166 | static unsigned offset_il_node(struct mempolicy *pol, | |
1167 | struct vm_area_struct *vma, unsigned long off) | |
1168 | { | |
dfcd3c0d | 1169 | unsigned nnodes = nodes_weight(pol->v.nodes); |
1da177e4 LT |
1170 | unsigned target = (unsigned)off % nnodes; |
1171 | int c; | |
1172 | int nid = -1; | |
1173 | ||
1174 | c = 0; | |
1175 | do { | |
dfcd3c0d | 1176 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1177 | c++; |
1178 | } while (c <= target); | |
1da177e4 LT |
1179 | return nid; |
1180 | } | |
1181 | ||
5da7ca86 CL |
1182 | /* Determine a node number for interleave */ |
1183 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1184 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1185 | { | |
1186 | if (vma) { | |
1187 | unsigned long off; | |
1188 | ||
3b98b087 NA |
1189 | /* |
1190 | * for small pages, there is no difference between | |
1191 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1192 | * for huge pages, since vm_pgoff is in units of small | |
1193 | * pages, we need to shift off the always 0 bits to get | |
1194 | * a useful offset. | |
1195 | */ | |
1196 | BUG_ON(shift < PAGE_SHIFT); | |
1197 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1198 | off += (addr - vma->vm_start) >> shift; |
1199 | return offset_il_node(pol, vma, off); | |
1200 | } else | |
1201 | return interleave_nodes(pol); | |
1202 | } | |
1203 | ||
00ac59ad | 1204 | #ifdef CONFIG_HUGETLBFS |
5da7ca86 | 1205 | /* Return a zonelist suitable for a huge page allocation. */ |
396faf03 MG |
1206 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
1207 | gfp_t gfp_flags) | |
5da7ca86 CL |
1208 | { |
1209 | struct mempolicy *pol = get_vma_policy(current, vma, addr); | |
1210 | ||
1211 | if (pol->policy == MPOL_INTERLEAVE) { | |
1212 | unsigned nid; | |
1213 | ||
1214 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); | |
396faf03 | 1215 | return NODE_DATA(nid)->node_zonelists + gfp_zone(gfp_flags); |
5da7ca86 CL |
1216 | } |
1217 | return zonelist_policy(GFP_HIGHUSER, pol); | |
1218 | } | |
00ac59ad | 1219 | #endif |
5da7ca86 | 1220 | |
1da177e4 LT |
1221 | /* Allocate a page in interleaved policy. |
1222 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1223 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1224 | unsigned nid) | |
1da177e4 LT |
1225 | { |
1226 | struct zonelist *zl; | |
1227 | struct page *page; | |
1228 | ||
af4ca457 | 1229 | zl = NODE_DATA(nid)->node_zonelists + gfp_zone(gfp); |
1da177e4 | 1230 | page = __alloc_pages(gfp, order, zl); |
ca889e6c CL |
1231 | if (page && page_zone(page) == zl->zones[0]) |
1232 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); | |
1da177e4 LT |
1233 | return page; |
1234 | } | |
1235 | ||
1236 | /** | |
1237 | * alloc_page_vma - Allocate a page for a VMA. | |
1238 | * | |
1239 | * @gfp: | |
1240 | * %GFP_USER user allocation. | |
1241 | * %GFP_KERNEL kernel allocations, | |
1242 | * %GFP_HIGHMEM highmem/user allocations, | |
1243 | * %GFP_FS allocation should not call back into a file system. | |
1244 | * %GFP_ATOMIC don't sleep. | |
1245 | * | |
1246 | * @vma: Pointer to VMA or NULL if not available. | |
1247 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1248 | * | |
1249 | * This function allocates a page from the kernel page pool and applies | |
1250 | * a NUMA policy associated with the VMA or the current process. | |
1251 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1252 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1253 | * all allocations for pages that will be mapped into | |
1254 | * user space. Returns NULL when no page can be allocated. | |
1255 | * | |
1256 | * Should be called with the mm_sem of the vma hold. | |
1257 | */ | |
1258 | struct page * | |
dd0fc66f | 1259 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1260 | { |
6e21c8f1 | 1261 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
1da177e4 | 1262 | |
cf2a473c | 1263 | cpuset_update_task_memory_state(); |
1da177e4 LT |
1264 | |
1265 | if (unlikely(pol->policy == MPOL_INTERLEAVE)) { | |
1266 | unsigned nid; | |
5da7ca86 CL |
1267 | |
1268 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
1da177e4 LT |
1269 | return alloc_page_interleave(gfp, 0, nid); |
1270 | } | |
1271 | return __alloc_pages(gfp, 0, zonelist_policy(gfp, pol)); | |
1272 | } | |
1273 | ||
1274 | /** | |
1275 | * alloc_pages_current - Allocate pages. | |
1276 | * | |
1277 | * @gfp: | |
1278 | * %GFP_USER user allocation, | |
1279 | * %GFP_KERNEL kernel allocation, | |
1280 | * %GFP_HIGHMEM highmem allocation, | |
1281 | * %GFP_FS don't call back into a file system. | |
1282 | * %GFP_ATOMIC don't sleep. | |
1283 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1284 | * | |
1285 | * Allocate a page from the kernel page pool. When not in | |
1286 | * interrupt context and apply the current process NUMA policy. | |
1287 | * Returns NULL when no page can be allocated. | |
1288 | * | |
cf2a473c | 1289 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1290 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1291 | * 2) allocating for current task (not interrupt). | |
1292 | */ | |
dd0fc66f | 1293 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1294 | { |
1295 | struct mempolicy *pol = current->mempolicy; | |
1296 | ||
1297 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1298 | cpuset_update_task_memory_state(); |
9b819d20 | 1299 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 LT |
1300 | pol = &default_policy; |
1301 | if (pol->policy == MPOL_INTERLEAVE) | |
1302 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); | |
1303 | return __alloc_pages(gfp, order, zonelist_policy(gfp, pol)); | |
1304 | } | |
1305 | EXPORT_SYMBOL(alloc_pages_current); | |
1306 | ||
4225399a PJ |
1307 | /* |
1308 | * If mpol_copy() sees current->cpuset == cpuset_being_rebound, then it | |
1309 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() | |
1310 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1311 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1312 | * further kernel/cpuset.c update_nodemask(). | |
1313 | */ | |
1314 | void *cpuset_being_rebound; | |
1315 | ||
1da177e4 LT |
1316 | /* Slow path of a mempolicy copy */ |
1317 | struct mempolicy *__mpol_copy(struct mempolicy *old) | |
1318 | { | |
1319 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1320 | ||
1321 | if (!new) | |
1322 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1323 | if (current_cpuset_is_being_rebound()) { |
1324 | nodemask_t mems = cpuset_mems_allowed(current); | |
1325 | mpol_rebind_policy(old, &mems); | |
1326 | } | |
1da177e4 LT |
1327 | *new = *old; |
1328 | atomic_set(&new->refcnt, 1); | |
1329 | if (new->policy == MPOL_BIND) { | |
1330 | int sz = ksize(old->v.zonelist); | |
e94b1766 | 1331 | new->v.zonelist = kmemdup(old->v.zonelist, sz, GFP_KERNEL); |
1da177e4 LT |
1332 | if (!new->v.zonelist) { |
1333 | kmem_cache_free(policy_cache, new); | |
1334 | return ERR_PTR(-ENOMEM); | |
1335 | } | |
1da177e4 LT |
1336 | } |
1337 | return new; | |
1338 | } | |
1339 | ||
1340 | /* Slow path of a mempolicy comparison */ | |
1341 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1342 | { | |
1343 | if (!a || !b) | |
1344 | return 0; | |
1345 | if (a->policy != b->policy) | |
1346 | return 0; | |
1347 | switch (a->policy) { | |
1348 | case MPOL_DEFAULT: | |
1349 | return 1; | |
1350 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 1351 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 LT |
1352 | case MPOL_PREFERRED: |
1353 | return a->v.preferred_node == b->v.preferred_node; | |
1354 | case MPOL_BIND: { | |
1355 | int i; | |
1356 | for (i = 0; a->v.zonelist->zones[i]; i++) | |
1357 | if (a->v.zonelist->zones[i] != b->v.zonelist->zones[i]) | |
1358 | return 0; | |
1359 | return b->v.zonelist->zones[i] == NULL; | |
1360 | } | |
1361 | default: | |
1362 | BUG(); | |
1363 | return 0; | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | /* Slow path of a mpol destructor. */ | |
1368 | void __mpol_free(struct mempolicy *p) | |
1369 | { | |
1370 | if (!atomic_dec_and_test(&p->refcnt)) | |
1371 | return; | |
1372 | if (p->policy == MPOL_BIND) | |
1373 | kfree(p->v.zonelist); | |
1374 | p->policy = MPOL_DEFAULT; | |
1375 | kmem_cache_free(policy_cache, p); | |
1376 | } | |
1377 | ||
1da177e4 LT |
1378 | /* |
1379 | * Shared memory backing store policy support. | |
1380 | * | |
1381 | * Remember policies even when nobody has shared memory mapped. | |
1382 | * The policies are kept in Red-Black tree linked from the inode. | |
1383 | * They are protected by the sp->lock spinlock, which should be held | |
1384 | * for any accesses to the tree. | |
1385 | */ | |
1386 | ||
1387 | /* lookup first element intersecting start-end */ | |
1388 | /* Caller holds sp->lock */ | |
1389 | static struct sp_node * | |
1390 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1391 | { | |
1392 | struct rb_node *n = sp->root.rb_node; | |
1393 | ||
1394 | while (n) { | |
1395 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1396 | ||
1397 | if (start >= p->end) | |
1398 | n = n->rb_right; | |
1399 | else if (end <= p->start) | |
1400 | n = n->rb_left; | |
1401 | else | |
1402 | break; | |
1403 | } | |
1404 | if (!n) | |
1405 | return NULL; | |
1406 | for (;;) { | |
1407 | struct sp_node *w = NULL; | |
1408 | struct rb_node *prev = rb_prev(n); | |
1409 | if (!prev) | |
1410 | break; | |
1411 | w = rb_entry(prev, struct sp_node, nd); | |
1412 | if (w->end <= start) | |
1413 | break; | |
1414 | n = prev; | |
1415 | } | |
1416 | return rb_entry(n, struct sp_node, nd); | |
1417 | } | |
1418 | ||
1419 | /* Insert a new shared policy into the list. */ | |
1420 | /* Caller holds sp->lock */ | |
1421 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1422 | { | |
1423 | struct rb_node **p = &sp->root.rb_node; | |
1424 | struct rb_node *parent = NULL; | |
1425 | struct sp_node *nd; | |
1426 | ||
1427 | while (*p) { | |
1428 | parent = *p; | |
1429 | nd = rb_entry(parent, struct sp_node, nd); | |
1430 | if (new->start < nd->start) | |
1431 | p = &(*p)->rb_left; | |
1432 | else if (new->end > nd->end) | |
1433 | p = &(*p)->rb_right; | |
1434 | else | |
1435 | BUG(); | |
1436 | } | |
1437 | rb_link_node(&new->nd, parent, p); | |
1438 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1439 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
1da177e4 LT |
1440 | new->policy ? new->policy->policy : 0); |
1441 | } | |
1442 | ||
1443 | /* Find shared policy intersecting idx */ | |
1444 | struct mempolicy * | |
1445 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1446 | { | |
1447 | struct mempolicy *pol = NULL; | |
1448 | struct sp_node *sn; | |
1449 | ||
1450 | if (!sp->root.rb_node) | |
1451 | return NULL; | |
1452 | spin_lock(&sp->lock); | |
1453 | sn = sp_lookup(sp, idx, idx+1); | |
1454 | if (sn) { | |
1455 | mpol_get(sn->policy); | |
1456 | pol = sn->policy; | |
1457 | } | |
1458 | spin_unlock(&sp->lock); | |
1459 | return pol; | |
1460 | } | |
1461 | ||
1462 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1463 | { | |
140d5a49 | 1464 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 LT |
1465 | rb_erase(&n->nd, &sp->root); |
1466 | mpol_free(n->policy); | |
1467 | kmem_cache_free(sn_cache, n); | |
1468 | } | |
1469 | ||
1470 | struct sp_node * | |
1471 | sp_alloc(unsigned long start, unsigned long end, struct mempolicy *pol) | |
1472 | { | |
1473 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1474 | ||
1475 | if (!n) | |
1476 | return NULL; | |
1477 | n->start = start; | |
1478 | n->end = end; | |
1479 | mpol_get(pol); | |
1480 | n->policy = pol; | |
1481 | return n; | |
1482 | } | |
1483 | ||
1484 | /* Replace a policy range. */ | |
1485 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1486 | unsigned long end, struct sp_node *new) | |
1487 | { | |
1488 | struct sp_node *n, *new2 = NULL; | |
1489 | ||
1490 | restart: | |
1491 | spin_lock(&sp->lock); | |
1492 | n = sp_lookup(sp, start, end); | |
1493 | /* Take care of old policies in the same range. */ | |
1494 | while (n && n->start < end) { | |
1495 | struct rb_node *next = rb_next(&n->nd); | |
1496 | if (n->start >= start) { | |
1497 | if (n->end <= end) | |
1498 | sp_delete(sp, n); | |
1499 | else | |
1500 | n->start = end; | |
1501 | } else { | |
1502 | /* Old policy spanning whole new range. */ | |
1503 | if (n->end > end) { | |
1504 | if (!new2) { | |
1505 | spin_unlock(&sp->lock); | |
1506 | new2 = sp_alloc(end, n->end, n->policy); | |
1507 | if (!new2) | |
1508 | return -ENOMEM; | |
1509 | goto restart; | |
1510 | } | |
1511 | n->end = start; | |
1512 | sp_insert(sp, new2); | |
1513 | new2 = NULL; | |
1514 | break; | |
1515 | } else | |
1516 | n->end = start; | |
1517 | } | |
1518 | if (!next) | |
1519 | break; | |
1520 | n = rb_entry(next, struct sp_node, nd); | |
1521 | } | |
1522 | if (new) | |
1523 | sp_insert(sp, new); | |
1524 | spin_unlock(&sp->lock); | |
1525 | if (new2) { | |
1526 | mpol_free(new2->policy); | |
1527 | kmem_cache_free(sn_cache, new2); | |
1528 | } | |
1529 | return 0; | |
1530 | } | |
1531 | ||
7339ff83 RH |
1532 | void mpol_shared_policy_init(struct shared_policy *info, int policy, |
1533 | nodemask_t *policy_nodes) | |
1534 | { | |
1535 | info->root = RB_ROOT; | |
1536 | spin_lock_init(&info->lock); | |
1537 | ||
1538 | if (policy != MPOL_DEFAULT) { | |
1539 | struct mempolicy *newpol; | |
1540 | ||
1541 | /* Falls back to MPOL_DEFAULT on any error */ | |
1542 | newpol = mpol_new(policy, policy_nodes); | |
1543 | if (!IS_ERR(newpol)) { | |
1544 | /* Create pseudo-vma that contains just the policy */ | |
1545 | struct vm_area_struct pvma; | |
1546 | ||
1547 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1548 | /* Policy covers entire file */ | |
1549 | pvma.vm_end = TASK_SIZE; | |
1550 | mpol_set_shared_policy(info, &pvma, newpol); | |
1551 | mpol_free(newpol); | |
1552 | } | |
1553 | } | |
1554 | } | |
1555 | ||
1da177e4 LT |
1556 | int mpol_set_shared_policy(struct shared_policy *info, |
1557 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1558 | { | |
1559 | int err; | |
1560 | struct sp_node *new = NULL; | |
1561 | unsigned long sz = vma_pages(vma); | |
1562 | ||
140d5a49 | 1563 | pr_debug("set_shared_policy %lx sz %lu %d %lx\n", |
1da177e4 LT |
1564 | vma->vm_pgoff, |
1565 | sz, npol? npol->policy : -1, | |
140d5a49 | 1566 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1567 | |
1568 | if (npol) { | |
1569 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1570 | if (!new) | |
1571 | return -ENOMEM; | |
1572 | } | |
1573 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1574 | if (err && new) | |
1575 | kmem_cache_free(sn_cache, new); | |
1576 | return err; | |
1577 | } | |
1578 | ||
1579 | /* Free a backing policy store on inode delete. */ | |
1580 | void mpol_free_shared_policy(struct shared_policy *p) | |
1581 | { | |
1582 | struct sp_node *n; | |
1583 | struct rb_node *next; | |
1584 | ||
1585 | if (!p->root.rb_node) | |
1586 | return; | |
1587 | spin_lock(&p->lock); | |
1588 | next = rb_first(&p->root); | |
1589 | while (next) { | |
1590 | n = rb_entry(next, struct sp_node, nd); | |
1591 | next = rb_next(&n->nd); | |
90c5029e | 1592 | rb_erase(&n->nd, &p->root); |
1da177e4 LT |
1593 | mpol_free(n->policy); |
1594 | kmem_cache_free(sn_cache, n); | |
1595 | } | |
1596 | spin_unlock(&p->lock); | |
1da177e4 LT |
1597 | } |
1598 | ||
1599 | /* assumes fs == KERNEL_DS */ | |
1600 | void __init numa_policy_init(void) | |
1601 | { | |
b71636e2 PM |
1602 | nodemask_t interleave_nodes; |
1603 | unsigned long largest = 0; | |
1604 | int nid, prefer = 0; | |
1605 | ||
1da177e4 LT |
1606 | policy_cache = kmem_cache_create("numa_policy", |
1607 | sizeof(struct mempolicy), | |
20c2df83 | 1608 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
1609 | |
1610 | sn_cache = kmem_cache_create("shared_policy_node", | |
1611 | sizeof(struct sp_node), | |
20c2df83 | 1612 | 0, SLAB_PANIC, NULL); |
1da177e4 | 1613 | |
b71636e2 PM |
1614 | /* |
1615 | * Set interleaving policy for system init. Interleaving is only | |
1616 | * enabled across suitably sized nodes (default is >= 16MB), or | |
1617 | * fall back to the largest node if they're all smaller. | |
1618 | */ | |
1619 | nodes_clear(interleave_nodes); | |
1620 | for_each_online_node(nid) { | |
1621 | unsigned long total_pages = node_present_pages(nid); | |
1622 | ||
1623 | /* Preserve the largest node */ | |
1624 | if (largest < total_pages) { | |
1625 | largest = total_pages; | |
1626 | prefer = nid; | |
1627 | } | |
1628 | ||
1629 | /* Interleave this node? */ | |
1630 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
1631 | node_set(nid, interleave_nodes); | |
1632 | } | |
1633 | ||
1634 | /* All too small, use the largest */ | |
1635 | if (unlikely(nodes_empty(interleave_nodes))) | |
1636 | node_set(prefer, interleave_nodes); | |
1da177e4 | 1637 | |
b71636e2 | 1638 | if (do_set_mempolicy(MPOL_INTERLEAVE, &interleave_nodes)) |
1da177e4 LT |
1639 | printk("numa_policy_init: interleaving failed\n"); |
1640 | } | |
1641 | ||
8bccd85f | 1642 | /* Reset policy of current process to default */ |
1da177e4 LT |
1643 | void numa_default_policy(void) |
1644 | { | |
8bccd85f | 1645 | do_set_mempolicy(MPOL_DEFAULT, NULL); |
1da177e4 | 1646 | } |
68860ec1 PJ |
1647 | |
1648 | /* Migrate a policy to a different set of nodes */ | |
74cb2155 | 1649 | void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
68860ec1 | 1650 | { |
74cb2155 | 1651 | nodemask_t *mpolmask; |
68860ec1 PJ |
1652 | nodemask_t tmp; |
1653 | ||
1654 | if (!pol) | |
1655 | return; | |
74cb2155 PJ |
1656 | mpolmask = &pol->cpuset_mems_allowed; |
1657 | if (nodes_equal(*mpolmask, *newmask)) | |
1658 | return; | |
68860ec1 PJ |
1659 | |
1660 | switch (pol->policy) { | |
1661 | case MPOL_DEFAULT: | |
1662 | break; | |
1663 | case MPOL_INTERLEAVE: | |
74cb2155 | 1664 | nodes_remap(tmp, pol->v.nodes, *mpolmask, *newmask); |
68860ec1 | 1665 | pol->v.nodes = tmp; |
74cb2155 PJ |
1666 | *mpolmask = *newmask; |
1667 | current->il_next = node_remap(current->il_next, | |
1668 | *mpolmask, *newmask); | |
68860ec1 PJ |
1669 | break; |
1670 | case MPOL_PREFERRED: | |
1671 | pol->v.preferred_node = node_remap(pol->v.preferred_node, | |
74cb2155 PJ |
1672 | *mpolmask, *newmask); |
1673 | *mpolmask = *newmask; | |
68860ec1 PJ |
1674 | break; |
1675 | case MPOL_BIND: { | |
1676 | nodemask_t nodes; | |
1677 | struct zone **z; | |
1678 | struct zonelist *zonelist; | |
1679 | ||
1680 | nodes_clear(nodes); | |
1681 | for (z = pol->v.zonelist->zones; *z; z++) | |
89fa3024 | 1682 | node_set(zone_to_nid(*z), nodes); |
74cb2155 | 1683 | nodes_remap(tmp, nodes, *mpolmask, *newmask); |
68860ec1 PJ |
1684 | nodes = tmp; |
1685 | ||
1686 | zonelist = bind_zonelist(&nodes); | |
1687 | ||
1688 | /* If no mem, then zonelist is NULL and we keep old zonelist. | |
1689 | * If that old zonelist has no remaining mems_allowed nodes, | |
1690 | * then zonelist_policy() will "FALL THROUGH" to MPOL_DEFAULT. | |
1691 | */ | |
1692 | ||
8af5e2eb | 1693 | if (!IS_ERR(zonelist)) { |
68860ec1 PJ |
1694 | /* Good - got mem - substitute new zonelist */ |
1695 | kfree(pol->v.zonelist); | |
1696 | pol->v.zonelist = zonelist; | |
1697 | } | |
74cb2155 | 1698 | *mpolmask = *newmask; |
68860ec1 PJ |
1699 | break; |
1700 | } | |
1701 | default: | |
1702 | BUG(); | |
1703 | break; | |
1704 | } | |
1705 | } | |
1706 | ||
1707 | /* | |
74cb2155 PJ |
1708 | * Wrapper for mpol_rebind_policy() that just requires task |
1709 | * pointer, and updates task mempolicy. | |
68860ec1 | 1710 | */ |
74cb2155 PJ |
1711 | |
1712 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
68860ec1 | 1713 | { |
74cb2155 | 1714 | mpol_rebind_policy(tsk->mempolicy, new); |
68860ec1 | 1715 | } |
1a75a6c8 | 1716 | |
4225399a PJ |
1717 | /* |
1718 | * Rebind each vma in mm to new nodemask. | |
1719 | * | |
1720 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
1721 | */ | |
1722 | ||
1723 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
1724 | { | |
1725 | struct vm_area_struct *vma; | |
1726 | ||
1727 | down_write(&mm->mmap_sem); | |
1728 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
1729 | mpol_rebind_policy(vma->vm_policy, new); | |
1730 | up_write(&mm->mmap_sem); | |
1731 | } | |
1732 | ||
1a75a6c8 CL |
1733 | /* |
1734 | * Display pages allocated per node and memory policy via /proc. | |
1735 | */ | |
1736 | ||
15ad7cdc HD |
1737 | static const char * const policy_types[] = |
1738 | { "default", "prefer", "bind", "interleave" }; | |
1a75a6c8 CL |
1739 | |
1740 | /* | |
1741 | * Convert a mempolicy into a string. | |
1742 | * Returns the number of characters in buffer (if positive) | |
1743 | * or an error (negative) | |
1744 | */ | |
1745 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1746 | { | |
1747 | char *p = buffer; | |
1748 | int l; | |
1749 | nodemask_t nodes; | |
1750 | int mode = pol ? pol->policy : MPOL_DEFAULT; | |
1751 | ||
1752 | switch (mode) { | |
1753 | case MPOL_DEFAULT: | |
1754 | nodes_clear(nodes); | |
1755 | break; | |
1756 | ||
1757 | case MPOL_PREFERRED: | |
1758 | nodes_clear(nodes); | |
1759 | node_set(pol->v.preferred_node, nodes); | |
1760 | break; | |
1761 | ||
1762 | case MPOL_BIND: | |
1763 | get_zonemask(pol, &nodes); | |
1764 | break; | |
1765 | ||
1766 | case MPOL_INTERLEAVE: | |
1767 | nodes = pol->v.nodes; | |
1768 | break; | |
1769 | ||
1770 | default: | |
1771 | BUG(); | |
1772 | return -EFAULT; | |
1773 | } | |
1774 | ||
1775 | l = strlen(policy_types[mode]); | |
1776 | if (buffer + maxlen < p + l + 1) | |
1777 | return -ENOSPC; | |
1778 | ||
1779 | strcpy(p, policy_types[mode]); | |
1780 | p += l; | |
1781 | ||
1782 | if (!nodes_empty(nodes)) { | |
1783 | if (buffer + maxlen < p + 2) | |
1784 | return -ENOSPC; | |
1785 | *p++ = '='; | |
1786 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
1787 | } | |
1788 | return p - buffer; | |
1789 | } | |
1790 | ||
1791 | struct numa_maps { | |
1792 | unsigned long pages; | |
1793 | unsigned long anon; | |
397874df CL |
1794 | unsigned long active; |
1795 | unsigned long writeback; | |
1a75a6c8 | 1796 | unsigned long mapcount_max; |
397874df CL |
1797 | unsigned long dirty; |
1798 | unsigned long swapcache; | |
1a75a6c8 CL |
1799 | unsigned long node[MAX_NUMNODES]; |
1800 | }; | |
1801 | ||
397874df | 1802 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
1803 | { |
1804 | struct numa_maps *md = private; | |
1805 | int count = page_mapcount(page); | |
1806 | ||
397874df CL |
1807 | md->pages++; |
1808 | if (pte_dirty || PageDirty(page)) | |
1809 | md->dirty++; | |
1a75a6c8 | 1810 | |
397874df CL |
1811 | if (PageSwapCache(page)) |
1812 | md->swapcache++; | |
1a75a6c8 | 1813 | |
397874df CL |
1814 | if (PageActive(page)) |
1815 | md->active++; | |
1816 | ||
1817 | if (PageWriteback(page)) | |
1818 | md->writeback++; | |
1a75a6c8 CL |
1819 | |
1820 | if (PageAnon(page)) | |
1821 | md->anon++; | |
1822 | ||
397874df CL |
1823 | if (count > md->mapcount_max) |
1824 | md->mapcount_max = count; | |
1825 | ||
1a75a6c8 | 1826 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
1827 | } |
1828 | ||
7f709ed0 | 1829 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
1830 | static void check_huge_range(struct vm_area_struct *vma, |
1831 | unsigned long start, unsigned long end, | |
1832 | struct numa_maps *md) | |
1833 | { | |
1834 | unsigned long addr; | |
1835 | struct page *page; | |
1836 | ||
1837 | for (addr = start; addr < end; addr += HPAGE_SIZE) { | |
1838 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); | |
1839 | pte_t pte; | |
1840 | ||
1841 | if (!ptep) | |
1842 | continue; | |
1843 | ||
1844 | pte = *ptep; | |
1845 | if (pte_none(pte)) | |
1846 | continue; | |
1847 | ||
1848 | page = pte_page(pte); | |
1849 | if (!page) | |
1850 | continue; | |
1851 | ||
1852 | gather_stats(page, md, pte_dirty(*ptep)); | |
1853 | } | |
1854 | } | |
7f709ed0 AM |
1855 | #else |
1856 | static inline void check_huge_range(struct vm_area_struct *vma, | |
1857 | unsigned long start, unsigned long end, | |
1858 | struct numa_maps *md) | |
1859 | { | |
1860 | } | |
1861 | #endif | |
397874df | 1862 | |
1a75a6c8 CL |
1863 | int show_numa_map(struct seq_file *m, void *v) |
1864 | { | |
99f89551 | 1865 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
1866 | struct vm_area_struct *vma = v; |
1867 | struct numa_maps *md; | |
397874df CL |
1868 | struct file *file = vma->vm_file; |
1869 | struct mm_struct *mm = vma->vm_mm; | |
1a75a6c8 CL |
1870 | int n; |
1871 | char buffer[50]; | |
1872 | ||
397874df | 1873 | if (!mm) |
1a75a6c8 CL |
1874 | return 0; |
1875 | ||
1876 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
1877 | if (!md) | |
1878 | return 0; | |
1879 | ||
397874df | 1880 | mpol_to_str(buffer, sizeof(buffer), |
99f89551 | 1881 | get_vma_policy(priv->task, vma, vma->vm_start)); |
397874df CL |
1882 | |
1883 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
1884 | ||
1885 | if (file) { | |
1886 | seq_printf(m, " file="); | |
e9536ae7 | 1887 | seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n\t= "); |
397874df CL |
1888 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
1889 | seq_printf(m, " heap"); | |
1890 | } else if (vma->vm_start <= mm->start_stack && | |
1891 | vma->vm_end >= mm->start_stack) { | |
1892 | seq_printf(m, " stack"); | |
1893 | } | |
1894 | ||
1895 | if (is_vm_hugetlb_page(vma)) { | |
1896 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
1897 | seq_printf(m, " huge"); | |
1898 | } else { | |
a57ebfdb | 1899 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
397874df CL |
1900 | &node_online_map, MPOL_MF_STATS, md); |
1901 | } | |
1902 | ||
1903 | if (!md->pages) | |
1904 | goto out; | |
1a75a6c8 | 1905 | |
397874df CL |
1906 | if (md->anon) |
1907 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 1908 | |
397874df CL |
1909 | if (md->dirty) |
1910 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 1911 | |
397874df CL |
1912 | if (md->pages != md->anon && md->pages != md->dirty) |
1913 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 1914 | |
397874df CL |
1915 | if (md->mapcount_max > 1) |
1916 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 1917 | |
397874df CL |
1918 | if (md->swapcache) |
1919 | seq_printf(m," swapcache=%lu", md->swapcache); | |
1920 | ||
1921 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
1922 | seq_printf(m," active=%lu", md->active); | |
1923 | ||
1924 | if (md->writeback) | |
1925 | seq_printf(m," writeback=%lu", md->writeback); | |
1926 | ||
1927 | for_each_online_node(n) | |
1928 | if (md->node[n]) | |
1929 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
1930 | out: | |
1931 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
1932 | kfree(md); |
1933 | ||
1934 | if (m->count < m->size) | |
99f89551 | 1935 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
1936 | return 0; |
1937 | } | |
1938 |