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> | |
1da177e4 LT |
75 | #include <linux/nodemask.h> |
76 | #include <linux/cpuset.h> | |
77 | #include <linux/gfp.h> | |
78 | #include <linux/slab.h> | |
79 | #include <linux/string.h> | |
80 | #include <linux/module.h> | |
81 | #include <linux/interrupt.h> | |
82 | #include <linux/init.h> | |
83 | #include <linux/compat.h> | |
dc9aa5b9 | 84 | #include <linux/swap.h> |
1a75a6c8 CL |
85 | #include <linux/seq_file.h> |
86 | #include <linux/proc_fs.h> | |
b20a3503 | 87 | #include <linux/migrate.h> |
95a402c3 | 88 | #include <linux/rmap.h> |
86c3a764 | 89 | #include <linux/security.h> |
dbcb0f19 | 90 | #include <linux/syscalls.h> |
dc9aa5b9 | 91 | |
1da177e4 LT |
92 | #include <asm/tlbflush.h> |
93 | #include <asm/uaccess.h> | |
94 | ||
38e35860 | 95 | /* Internal flags */ |
dc9aa5b9 | 96 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 97 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
1a75a6c8 | 98 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
dc9aa5b9 | 99 | |
fcc234f8 PE |
100 | static struct kmem_cache *policy_cache; |
101 | static struct kmem_cache *sn_cache; | |
1da177e4 | 102 | |
1da177e4 LT |
103 | /* Highest zone. An specific allocation for a zone below that is not |
104 | policied. */ | |
6267276f | 105 | enum zone_type policy_zone = 0; |
1da177e4 | 106 | |
d42c6997 | 107 | struct mempolicy default_policy = { |
1da177e4 LT |
108 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
109 | .policy = MPOL_DEFAULT, | |
110 | }; | |
111 | ||
dbcb0f19 AB |
112 | static void mpol_rebind_policy(struct mempolicy *pol, |
113 | const nodemask_t *newmask); | |
114 | ||
1da177e4 | 115 | /* Do sanity checking on a policy */ |
dfcd3c0d | 116 | static int mpol_check_policy(int mode, nodemask_t *nodes) |
1da177e4 | 117 | { |
dfcd3c0d | 118 | int empty = nodes_empty(*nodes); |
1da177e4 LT |
119 | |
120 | switch (mode) { | |
121 | case MPOL_DEFAULT: | |
122 | if (!empty) | |
123 | return -EINVAL; | |
124 | break; | |
125 | case MPOL_BIND: | |
126 | case MPOL_INTERLEAVE: | |
127 | /* Preferred will only use the first bit, but allow | |
128 | more for now. */ | |
129 | if (empty) | |
130 | return -EINVAL; | |
131 | break; | |
132 | } | |
37b07e41 | 133 | return nodes_subset(*nodes, node_states[N_HIGH_MEMORY]) ? 0 : -EINVAL; |
1da177e4 | 134 | } |
dd942ae3 | 135 | |
1da177e4 | 136 | /* Generate a custom zonelist for the BIND policy. */ |
dfcd3c0d | 137 | static struct zonelist *bind_zonelist(nodemask_t *nodes) |
1da177e4 LT |
138 | { |
139 | struct zonelist *zl; | |
2f6726e5 CL |
140 | int num, max, nd; |
141 | enum zone_type k; | |
1da177e4 | 142 | |
dfcd3c0d | 143 | max = 1 + MAX_NR_ZONES * nodes_weight(*nodes); |
9276b1bc | 144 | max++; /* space for zlcache_ptr (see mmzone.h) */ |
dd942ae3 | 145 | zl = kmalloc(sizeof(struct zone *) * max, GFP_KERNEL); |
1da177e4 | 146 | if (!zl) |
8af5e2eb | 147 | return ERR_PTR(-ENOMEM); |
9276b1bc | 148 | zl->zlcache_ptr = NULL; |
1da177e4 | 149 | num = 0; |
dd942ae3 AK |
150 | /* First put in the highest zones from all nodes, then all the next |
151 | lower zones etc. Avoid empty zones because the memory allocator | |
152 | doesn't like them. If you implement node hot removal you | |
153 | have to fix that. */ | |
b377fd39 | 154 | k = MAX_NR_ZONES - 1; |
2f6726e5 | 155 | while (1) { |
dd942ae3 AK |
156 | for_each_node_mask(nd, *nodes) { |
157 | struct zone *z = &NODE_DATA(nd)->node_zones[k]; | |
158 | if (z->present_pages > 0) | |
159 | zl->zones[num++] = z; | |
160 | } | |
2f6726e5 CL |
161 | if (k == 0) |
162 | break; | |
163 | k--; | |
dd942ae3 | 164 | } |
8af5e2eb KH |
165 | if (num == 0) { |
166 | kfree(zl); | |
167 | return ERR_PTR(-EINVAL); | |
168 | } | |
1da177e4 LT |
169 | zl->zones[num] = NULL; |
170 | return zl; | |
171 | } | |
172 | ||
173 | /* Create a new policy */ | |
dfcd3c0d | 174 | static struct mempolicy *mpol_new(int mode, nodemask_t *nodes) |
1da177e4 LT |
175 | { |
176 | struct mempolicy *policy; | |
177 | ||
140d5a49 PM |
178 | pr_debug("setting mode %d nodes[0] %lx\n", |
179 | mode, nodes ? nodes_addr(*nodes)[0] : -1); | |
180 | ||
1da177e4 LT |
181 | if (mode == MPOL_DEFAULT) |
182 | return NULL; | |
183 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
184 | if (!policy) | |
185 | return ERR_PTR(-ENOMEM); | |
186 | atomic_set(&policy->refcnt, 1); | |
187 | switch (mode) { | |
188 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 189 | policy->v.nodes = *nodes; |
6eaf806a CL |
190 | nodes_and(policy->v.nodes, policy->v.nodes, |
191 | node_states[N_HIGH_MEMORY]); | |
192 | if (nodes_weight(policy->v.nodes) == 0) { | |
8f493d79 AK |
193 | kmem_cache_free(policy_cache, policy); |
194 | return ERR_PTR(-EINVAL); | |
195 | } | |
1da177e4 LT |
196 | break; |
197 | case MPOL_PREFERRED: | |
dfcd3c0d | 198 | policy->v.preferred_node = first_node(*nodes); |
1da177e4 LT |
199 | if (policy->v.preferred_node >= MAX_NUMNODES) |
200 | policy->v.preferred_node = -1; | |
201 | break; | |
202 | case MPOL_BIND: | |
203 | policy->v.zonelist = bind_zonelist(nodes); | |
8af5e2eb KH |
204 | if (IS_ERR(policy->v.zonelist)) { |
205 | void *error_code = policy->v.zonelist; | |
1da177e4 | 206 | kmem_cache_free(policy_cache, policy); |
8af5e2eb | 207 | return error_code; |
1da177e4 LT |
208 | } |
209 | break; | |
210 | } | |
211 | policy->policy = mode; | |
74cb2155 | 212 | policy->cpuset_mems_allowed = cpuset_mems_allowed(current); |
1da177e4 LT |
213 | return policy; |
214 | } | |
215 | ||
397874df | 216 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
217 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
218 | unsigned long flags); | |
1a75a6c8 | 219 | |
38e35860 | 220 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 221 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
222 | unsigned long addr, unsigned long end, |
223 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 224 | void *private) |
1da177e4 | 225 | { |
91612e0d HD |
226 | pte_t *orig_pte; |
227 | pte_t *pte; | |
705e87c0 | 228 | spinlock_t *ptl; |
941150a3 | 229 | |
705e87c0 | 230 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 231 | do { |
6aab341e | 232 | struct page *page; |
25ba77c1 | 233 | int nid; |
91612e0d HD |
234 | |
235 | if (!pte_present(*pte)) | |
1da177e4 | 236 | continue; |
6aab341e LT |
237 | page = vm_normal_page(vma, addr, *pte); |
238 | if (!page) | |
1da177e4 | 239 | continue; |
053837fc NP |
240 | /* |
241 | * The check for PageReserved here is important to avoid | |
242 | * handling zero pages and other pages that may have been | |
243 | * marked special by the system. | |
244 | * | |
245 | * If the PageReserved would not be checked here then f.e. | |
246 | * the location of the zero page could have an influence | |
247 | * on MPOL_MF_STRICT, zero pages would be counted for | |
248 | * the per node stats, and there would be useless attempts | |
249 | * to put zero pages on the migration list. | |
250 | */ | |
f4598c8b CL |
251 | if (PageReserved(page)) |
252 | continue; | |
6aab341e | 253 | nid = page_to_nid(page); |
38e35860 CL |
254 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
255 | continue; | |
256 | ||
1a75a6c8 | 257 | if (flags & MPOL_MF_STATS) |
397874df | 258 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 259 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 260 | migrate_page_add(page, private, flags); |
38e35860 CL |
261 | else |
262 | break; | |
91612e0d | 263 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 264 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
265 | return addr != end; |
266 | } | |
267 | ||
b5810039 | 268 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
269 | unsigned long addr, unsigned long end, |
270 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 271 | void *private) |
91612e0d HD |
272 | { |
273 | pmd_t *pmd; | |
274 | unsigned long next; | |
275 | ||
276 | pmd = pmd_offset(pud, addr); | |
277 | do { | |
278 | next = pmd_addr_end(addr, end); | |
279 | if (pmd_none_or_clear_bad(pmd)) | |
280 | continue; | |
dc9aa5b9 | 281 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 282 | flags, private)) |
91612e0d HD |
283 | return -EIO; |
284 | } while (pmd++, addr = next, addr != end); | |
285 | return 0; | |
286 | } | |
287 | ||
b5810039 | 288 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
289 | unsigned long addr, unsigned long end, |
290 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 291 | void *private) |
91612e0d HD |
292 | { |
293 | pud_t *pud; | |
294 | unsigned long next; | |
295 | ||
296 | pud = pud_offset(pgd, addr); | |
297 | do { | |
298 | next = pud_addr_end(addr, end); | |
299 | if (pud_none_or_clear_bad(pud)) | |
300 | continue; | |
dc9aa5b9 | 301 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 302 | flags, private)) |
91612e0d HD |
303 | return -EIO; |
304 | } while (pud++, addr = next, addr != end); | |
305 | return 0; | |
306 | } | |
307 | ||
b5810039 | 308 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
309 | unsigned long addr, unsigned long end, |
310 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 311 | void *private) |
91612e0d HD |
312 | { |
313 | pgd_t *pgd; | |
314 | unsigned long next; | |
315 | ||
b5810039 | 316 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
317 | do { |
318 | next = pgd_addr_end(addr, end); | |
319 | if (pgd_none_or_clear_bad(pgd)) | |
320 | continue; | |
dc9aa5b9 | 321 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 322 | flags, private)) |
91612e0d HD |
323 | return -EIO; |
324 | } while (pgd++, addr = next, addr != end); | |
325 | return 0; | |
1da177e4 LT |
326 | } |
327 | ||
dc9aa5b9 CL |
328 | /* |
329 | * Check if all pages in a range are on a set of nodes. | |
330 | * If pagelist != NULL then isolate pages from the LRU and | |
331 | * put them on the pagelist. | |
332 | */ | |
1da177e4 LT |
333 | static struct vm_area_struct * |
334 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 335 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
336 | { |
337 | int err; | |
338 | struct vm_area_struct *first, *vma, *prev; | |
339 | ||
90036ee5 | 340 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
90036ee5 | 341 | |
b20a3503 CL |
342 | err = migrate_prep(); |
343 | if (err) | |
344 | return ERR_PTR(err); | |
90036ee5 | 345 | } |
053837fc | 346 | |
1da177e4 LT |
347 | first = find_vma(mm, start); |
348 | if (!first) | |
349 | return ERR_PTR(-EFAULT); | |
350 | prev = NULL; | |
351 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
352 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
353 | if (!vma->vm_next && vma->vm_end < end) | |
354 | return ERR_PTR(-EFAULT); | |
355 | if (prev && prev->vm_end < vma->vm_start) | |
356 | return ERR_PTR(-EFAULT); | |
357 | } | |
358 | if (!is_vm_hugetlb_page(vma) && | |
359 | ((flags & MPOL_MF_STRICT) || | |
360 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
361 | vma_migratable(vma)))) { | |
5b952b3c | 362 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 363 | |
5b952b3c AK |
364 | if (endvma > end) |
365 | endvma = end; | |
366 | if (vma->vm_start > start) | |
367 | start = vma->vm_start; | |
dc9aa5b9 | 368 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 369 | flags, private); |
1da177e4 LT |
370 | if (err) { |
371 | first = ERR_PTR(err); | |
372 | break; | |
373 | } | |
374 | } | |
375 | prev = vma; | |
376 | } | |
377 | return first; | |
378 | } | |
379 | ||
380 | /* Apply policy to a single VMA */ | |
381 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
382 | { | |
383 | int err = 0; | |
384 | struct mempolicy *old = vma->vm_policy; | |
385 | ||
140d5a49 | 386 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
387 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
388 | vma->vm_ops, vma->vm_file, | |
389 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
390 | ||
391 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
392 | err = vma->vm_ops->set_policy(vma, new); | |
393 | if (!err) { | |
394 | mpol_get(new); | |
395 | vma->vm_policy = new; | |
396 | mpol_free(old); | |
397 | } | |
398 | return err; | |
399 | } | |
400 | ||
401 | /* Step 2: apply policy to a range and do splits. */ | |
402 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
403 | unsigned long end, struct mempolicy *new) | |
404 | { | |
405 | struct vm_area_struct *next; | |
406 | int err; | |
407 | ||
408 | err = 0; | |
409 | for (; vma && vma->vm_start < end; vma = next) { | |
410 | next = vma->vm_next; | |
411 | if (vma->vm_start < start) | |
412 | err = split_vma(vma->vm_mm, vma, start, 1); | |
413 | if (!err && vma->vm_end > end) | |
414 | err = split_vma(vma->vm_mm, vma, end, 0); | |
415 | if (!err) | |
416 | err = policy_vma(vma, new); | |
417 | if (err) | |
418 | break; | |
419 | } | |
420 | return err; | |
421 | } | |
422 | ||
8bccd85f CL |
423 | static int contextualize_policy(int mode, nodemask_t *nodes) |
424 | { | |
425 | if (!nodes) | |
426 | return 0; | |
427 | ||
cf2a473c | 428 | cpuset_update_task_memory_state(); |
5966514d PJ |
429 | if (!cpuset_nodes_subset_current_mems_allowed(*nodes)) |
430 | return -EINVAL; | |
8bccd85f CL |
431 | return mpol_check_policy(mode, nodes); |
432 | } | |
433 | ||
c61afb18 PJ |
434 | |
435 | /* | |
436 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
437 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
438 | * with other PF_* flag bits) on memory allocation hot code paths. | |
439 | * | |
440 | * If called from outside this file, the task 'p' should -only- be | |
441 | * a newly forked child not yet visible on the task list, because | |
442 | * manipulating the task flags of a visible task is not safe. | |
443 | * | |
444 | * The above limitation is why this routine has the funny name | |
445 | * mpol_fix_fork_child_flag(). | |
446 | * | |
447 | * It is also safe to call this with a task pointer of current, | |
448 | * which the static wrapper mpol_set_task_struct_flag() does, | |
449 | * for use within this file. | |
450 | */ | |
451 | ||
452 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
453 | { | |
454 | if (p->mempolicy) | |
455 | p->flags |= PF_MEMPOLICY; | |
456 | else | |
457 | p->flags &= ~PF_MEMPOLICY; | |
458 | } | |
459 | ||
460 | static void mpol_set_task_struct_flag(void) | |
461 | { | |
462 | mpol_fix_fork_child_flag(current); | |
463 | } | |
464 | ||
1da177e4 | 465 | /* Set the process memory policy */ |
dbcb0f19 | 466 | static long do_set_mempolicy(int mode, nodemask_t *nodes) |
1da177e4 | 467 | { |
1da177e4 | 468 | struct mempolicy *new; |
1da177e4 | 469 | |
8bccd85f | 470 | if (contextualize_policy(mode, nodes)) |
1da177e4 | 471 | return -EINVAL; |
8bccd85f | 472 | new = mpol_new(mode, nodes); |
1da177e4 LT |
473 | if (IS_ERR(new)) |
474 | return PTR_ERR(new); | |
475 | mpol_free(current->mempolicy); | |
476 | current->mempolicy = new; | |
c61afb18 | 477 | mpol_set_task_struct_flag(); |
1da177e4 | 478 | if (new && new->policy == MPOL_INTERLEAVE) |
dfcd3c0d | 479 | current->il_next = first_node(new->v.nodes); |
1da177e4 LT |
480 | return 0; |
481 | } | |
482 | ||
483 | /* Fill a zone bitmap for a policy */ | |
dfcd3c0d | 484 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
1da177e4 LT |
485 | { |
486 | int i; | |
487 | ||
dfcd3c0d | 488 | nodes_clear(*nodes); |
1da177e4 LT |
489 | switch (p->policy) { |
490 | case MPOL_BIND: | |
491 | for (i = 0; p->v.zonelist->zones[i]; i++) | |
89fa3024 | 492 | node_set(zone_to_nid(p->v.zonelist->zones[i]), |
8bccd85f | 493 | *nodes); |
1da177e4 LT |
494 | break; |
495 | case MPOL_DEFAULT: | |
496 | break; | |
497 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 498 | *nodes = p->v.nodes; |
1da177e4 LT |
499 | break; |
500 | case MPOL_PREFERRED: | |
56bbd65d | 501 | /* or use current node instead of memory_map? */ |
1da177e4 | 502 | if (p->v.preferred_node < 0) |
56bbd65d | 503 | *nodes = node_states[N_HIGH_MEMORY]; |
1da177e4 | 504 | else |
dfcd3c0d | 505 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
506 | break; |
507 | default: | |
508 | BUG(); | |
509 | } | |
510 | } | |
511 | ||
512 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
513 | { | |
514 | struct page *p; | |
515 | int err; | |
516 | ||
517 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
518 | if (err >= 0) { | |
519 | err = page_to_nid(p); | |
520 | put_page(p); | |
521 | } | |
522 | return err; | |
523 | } | |
524 | ||
1da177e4 | 525 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
526 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
527 | unsigned long addr, unsigned long flags) | |
1da177e4 | 528 | { |
8bccd85f | 529 | int err; |
1da177e4 LT |
530 | struct mm_struct *mm = current->mm; |
531 | struct vm_area_struct *vma = NULL; | |
532 | struct mempolicy *pol = current->mempolicy; | |
533 | ||
cf2a473c | 534 | cpuset_update_task_memory_state(); |
754af6f5 LS |
535 | if (flags & |
536 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 537 | return -EINVAL; |
754af6f5 LS |
538 | |
539 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
540 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
541 | return -EINVAL; | |
542 | *policy = 0; /* just so it's initialized */ | |
543 | *nmask = cpuset_current_mems_allowed; | |
544 | return 0; | |
545 | } | |
546 | ||
1da177e4 LT |
547 | if (flags & MPOL_F_ADDR) { |
548 | down_read(&mm->mmap_sem); | |
549 | vma = find_vma_intersection(mm, addr, addr+1); | |
550 | if (!vma) { | |
551 | up_read(&mm->mmap_sem); | |
552 | return -EFAULT; | |
553 | } | |
554 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
555 | pol = vma->vm_ops->get_policy(vma, addr); | |
556 | else | |
557 | pol = vma->vm_policy; | |
558 | } else if (addr) | |
559 | return -EINVAL; | |
560 | ||
561 | if (!pol) | |
562 | pol = &default_policy; | |
563 | ||
564 | if (flags & MPOL_F_NODE) { | |
565 | if (flags & MPOL_F_ADDR) { | |
566 | err = lookup_node(mm, addr); | |
567 | if (err < 0) | |
568 | goto out; | |
8bccd85f | 569 | *policy = err; |
1da177e4 LT |
570 | } else if (pol == current->mempolicy && |
571 | pol->policy == MPOL_INTERLEAVE) { | |
8bccd85f | 572 | *policy = current->il_next; |
1da177e4 LT |
573 | } else { |
574 | err = -EINVAL; | |
575 | goto out; | |
576 | } | |
577 | } else | |
8bccd85f | 578 | *policy = pol->policy; |
1da177e4 LT |
579 | |
580 | if (vma) { | |
581 | up_read(¤t->mm->mmap_sem); | |
582 | vma = NULL; | |
583 | } | |
584 | ||
1da177e4 | 585 | err = 0; |
8bccd85f CL |
586 | if (nmask) |
587 | get_zonemask(pol, nmask); | |
1da177e4 LT |
588 | |
589 | out: | |
590 | if (vma) | |
591 | up_read(¤t->mm->mmap_sem); | |
592 | return err; | |
593 | } | |
594 | ||
b20a3503 | 595 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
596 | /* |
597 | * page migration | |
598 | */ | |
fc301289 CL |
599 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
600 | unsigned long flags) | |
6ce3c4c0 CL |
601 | { |
602 | /* | |
fc301289 | 603 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 604 | */ |
b20a3503 CL |
605 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
606 | isolate_lru_page(page, pagelist); | |
7e2ab150 | 607 | } |
6ce3c4c0 | 608 | |
742755a1 | 609 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 610 | { |
769848c0 | 611 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
612 | } |
613 | ||
7e2ab150 CL |
614 | /* |
615 | * Migrate pages from one node to a target node. | |
616 | * Returns error or the number of pages not migrated. | |
617 | */ | |
dbcb0f19 AB |
618 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
619 | int flags) | |
7e2ab150 CL |
620 | { |
621 | nodemask_t nmask; | |
622 | LIST_HEAD(pagelist); | |
623 | int err = 0; | |
624 | ||
625 | nodes_clear(nmask); | |
626 | node_set(source, nmask); | |
6ce3c4c0 | 627 | |
7e2ab150 CL |
628 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
629 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
630 | ||
aaa994b3 | 631 | if (!list_empty(&pagelist)) |
95a402c3 CL |
632 | err = migrate_pages(&pagelist, new_node_page, dest); |
633 | ||
7e2ab150 | 634 | return err; |
6ce3c4c0 CL |
635 | } |
636 | ||
39743889 | 637 | /* |
7e2ab150 CL |
638 | * Move pages between the two nodesets so as to preserve the physical |
639 | * layout as much as possible. | |
39743889 CL |
640 | * |
641 | * Returns the number of page that could not be moved. | |
642 | */ | |
643 | int do_migrate_pages(struct mm_struct *mm, | |
644 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
645 | { | |
646 | LIST_HEAD(pagelist); | |
7e2ab150 CL |
647 | int busy = 0; |
648 | int err = 0; | |
649 | nodemask_t tmp; | |
39743889 | 650 | |
7e2ab150 | 651 | down_read(&mm->mmap_sem); |
39743889 | 652 | |
7b2259b3 CL |
653 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
654 | if (err) | |
655 | goto out; | |
656 | ||
7e2ab150 CL |
657 | /* |
658 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
659 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
660 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
661 | * The pair of nodemasks 'to' and 'from' define the map. | |
662 | * | |
663 | * If no pair of bits is found that way, fallback to picking some | |
664 | * pair of 'source' and 'dest' bits that are not the same. If the | |
665 | * 'source' and 'dest' bits are the same, this represents a node | |
666 | * that will be migrating to itself, so no pages need move. | |
667 | * | |
668 | * If no bits are left in 'tmp', or if all remaining bits left | |
669 | * in 'tmp' correspond to the same bit in 'to', return false | |
670 | * (nothing left to migrate). | |
671 | * | |
672 | * This lets us pick a pair of nodes to migrate between, such that | |
673 | * if possible the dest node is not already occupied by some other | |
674 | * source node, minimizing the risk of overloading the memory on a | |
675 | * node that would happen if we migrated incoming memory to a node | |
676 | * before migrating outgoing memory source that same node. | |
677 | * | |
678 | * A single scan of tmp is sufficient. As we go, we remember the | |
679 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
680 | * that not only moved, but what's better, moved to an empty slot | |
681 | * (d is not set in tmp), then we break out then, with that pair. | |
682 | * Otherwise when we finish scannng from_tmp, we at least have the | |
683 | * most recent <s, d> pair that moved. If we get all the way through | |
684 | * the scan of tmp without finding any node that moved, much less | |
685 | * moved to an empty node, then there is nothing left worth migrating. | |
686 | */ | |
d4984711 | 687 | |
7e2ab150 CL |
688 | tmp = *from_nodes; |
689 | while (!nodes_empty(tmp)) { | |
690 | int s,d; | |
691 | int source = -1; | |
692 | int dest = 0; | |
693 | ||
694 | for_each_node_mask(s, tmp) { | |
695 | d = node_remap(s, *from_nodes, *to_nodes); | |
696 | if (s == d) | |
697 | continue; | |
698 | ||
699 | source = s; /* Node moved. Memorize */ | |
700 | dest = d; | |
701 | ||
702 | /* dest not in remaining from nodes? */ | |
703 | if (!node_isset(dest, tmp)) | |
704 | break; | |
705 | } | |
706 | if (source == -1) | |
707 | break; | |
708 | ||
709 | node_clear(source, tmp); | |
710 | err = migrate_to_node(mm, source, dest, flags); | |
711 | if (err > 0) | |
712 | busy += err; | |
713 | if (err < 0) | |
714 | break; | |
39743889 | 715 | } |
7b2259b3 | 716 | out: |
39743889 | 717 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
718 | if (err < 0) |
719 | return err; | |
720 | return busy; | |
b20a3503 CL |
721 | |
722 | } | |
723 | ||
742755a1 | 724 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
725 | { |
726 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
727 | ||
769848c0 MG |
728 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
729 | page_address_in_vma(page, vma)); | |
95a402c3 | 730 | } |
b20a3503 CL |
731 | #else |
732 | ||
733 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
734 | unsigned long flags) | |
735 | { | |
39743889 CL |
736 | } |
737 | ||
b20a3503 CL |
738 | int do_migrate_pages(struct mm_struct *mm, |
739 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
740 | { | |
741 | return -ENOSYS; | |
742 | } | |
95a402c3 | 743 | |
69939749 | 744 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
745 | { |
746 | return NULL; | |
747 | } | |
b20a3503 CL |
748 | #endif |
749 | ||
dbcb0f19 AB |
750 | static long do_mbind(unsigned long start, unsigned long len, |
751 | unsigned long mode, nodemask_t *nmask, | |
752 | unsigned long flags) | |
6ce3c4c0 CL |
753 | { |
754 | struct vm_area_struct *vma; | |
755 | struct mm_struct *mm = current->mm; | |
756 | struct mempolicy *new; | |
757 | unsigned long end; | |
758 | int err; | |
759 | LIST_HEAD(pagelist); | |
760 | ||
761 | if ((flags & ~(unsigned long)(MPOL_MF_STRICT | | |
762 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
763 | || mode > MPOL_MAX) | |
764 | return -EINVAL; | |
74c00241 | 765 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
766 | return -EPERM; |
767 | ||
768 | if (start & ~PAGE_MASK) | |
769 | return -EINVAL; | |
770 | ||
771 | if (mode == MPOL_DEFAULT) | |
772 | flags &= ~MPOL_MF_STRICT; | |
773 | ||
774 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
775 | end = start + len; | |
776 | ||
777 | if (end < start) | |
778 | return -EINVAL; | |
779 | if (end == start) | |
780 | return 0; | |
781 | ||
782 | if (mpol_check_policy(mode, nmask)) | |
783 | return -EINVAL; | |
784 | ||
785 | new = mpol_new(mode, nmask); | |
786 | if (IS_ERR(new)) | |
787 | return PTR_ERR(new); | |
788 | ||
789 | /* | |
790 | * If we are using the default policy then operation | |
791 | * on discontinuous address spaces is okay after all | |
792 | */ | |
793 | if (!new) | |
794 | flags |= MPOL_MF_DISCONTIG_OK; | |
795 | ||
140d5a49 PM |
796 | pr_debug("mbind %lx-%lx mode:%ld nodes:%lx\n",start,start+len, |
797 | mode, nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 CL |
798 | |
799 | down_write(&mm->mmap_sem); | |
800 | vma = check_range(mm, start, end, nmask, | |
801 | flags | MPOL_MF_INVERT, &pagelist); | |
802 | ||
803 | err = PTR_ERR(vma); | |
804 | if (!IS_ERR(vma)) { | |
805 | int nr_failed = 0; | |
806 | ||
807 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 808 | |
6ce3c4c0 | 809 | if (!list_empty(&pagelist)) |
95a402c3 CL |
810 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
811 | (unsigned long)vma); | |
6ce3c4c0 CL |
812 | |
813 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
814 | err = -EIO; | |
815 | } | |
b20a3503 | 816 | |
6ce3c4c0 CL |
817 | up_write(&mm->mmap_sem); |
818 | mpol_free(new); | |
819 | return err; | |
820 | } | |
821 | ||
8bccd85f CL |
822 | /* |
823 | * User space interface with variable sized bitmaps for nodelists. | |
824 | */ | |
825 | ||
826 | /* Copy a node mask from user space. */ | |
39743889 | 827 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
828 | unsigned long maxnode) |
829 | { | |
830 | unsigned long k; | |
831 | unsigned long nlongs; | |
832 | unsigned long endmask; | |
833 | ||
834 | --maxnode; | |
835 | nodes_clear(*nodes); | |
836 | if (maxnode == 0 || !nmask) | |
837 | return 0; | |
a9c930ba | 838 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 839 | return -EINVAL; |
8bccd85f CL |
840 | |
841 | nlongs = BITS_TO_LONGS(maxnode); | |
842 | if ((maxnode % BITS_PER_LONG) == 0) | |
843 | endmask = ~0UL; | |
844 | else | |
845 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
846 | ||
847 | /* When the user specified more nodes than supported just check | |
848 | if the non supported part is all zero. */ | |
849 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
850 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
851 | return -EINVAL; | |
852 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
853 | unsigned long t; | |
854 | if (get_user(t, nmask + k)) | |
855 | return -EFAULT; | |
856 | if (k == nlongs - 1) { | |
857 | if (t & endmask) | |
858 | return -EINVAL; | |
859 | } else if (t) | |
860 | return -EINVAL; | |
861 | } | |
862 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
863 | endmask = ~0UL; | |
864 | } | |
865 | ||
866 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
867 | return -EFAULT; | |
868 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
869 | return 0; | |
870 | } | |
871 | ||
872 | /* Copy a kernel node mask to user space */ | |
873 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
874 | nodemask_t *nodes) | |
875 | { | |
876 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
877 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
878 | ||
879 | if (copy > nbytes) { | |
880 | if (copy > PAGE_SIZE) | |
881 | return -EINVAL; | |
882 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
883 | return -EFAULT; | |
884 | copy = nbytes; | |
885 | } | |
886 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
887 | } | |
888 | ||
889 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
890 | unsigned long mode, | |
891 | unsigned long __user *nmask, unsigned long maxnode, | |
892 | unsigned flags) | |
893 | { | |
894 | nodemask_t nodes; | |
895 | int err; | |
896 | ||
897 | err = get_nodes(&nodes, nmask, maxnode); | |
898 | if (err) | |
899 | return err; | |
30150f8d CL |
900 | #ifdef CONFIG_CPUSETS |
901 | /* Restrict the nodes to the allowed nodes in the cpuset */ | |
902 | nodes_and(nodes, nodes, current->mems_allowed); | |
903 | #endif | |
8bccd85f CL |
904 | return do_mbind(start, len, mode, &nodes, flags); |
905 | } | |
906 | ||
907 | /* Set the process memory policy */ | |
908 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
909 | unsigned long maxnode) | |
910 | { | |
911 | int err; | |
912 | nodemask_t nodes; | |
913 | ||
914 | if (mode < 0 || mode > MPOL_MAX) | |
915 | return -EINVAL; | |
916 | err = get_nodes(&nodes, nmask, maxnode); | |
917 | if (err) | |
918 | return err; | |
919 | return do_set_mempolicy(mode, &nodes); | |
920 | } | |
921 | ||
39743889 CL |
922 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
923 | const unsigned long __user *old_nodes, | |
924 | const unsigned long __user *new_nodes) | |
925 | { | |
926 | struct mm_struct *mm; | |
927 | struct task_struct *task; | |
928 | nodemask_t old; | |
929 | nodemask_t new; | |
930 | nodemask_t task_nodes; | |
931 | int err; | |
932 | ||
933 | err = get_nodes(&old, old_nodes, maxnode); | |
934 | if (err) | |
935 | return err; | |
936 | ||
937 | err = get_nodes(&new, new_nodes, maxnode); | |
938 | if (err) | |
939 | return err; | |
940 | ||
941 | /* Find the mm_struct */ | |
942 | read_lock(&tasklist_lock); | |
943 | task = pid ? find_task_by_pid(pid) : current; | |
944 | if (!task) { | |
945 | read_unlock(&tasklist_lock); | |
946 | return -ESRCH; | |
947 | } | |
948 | mm = get_task_mm(task); | |
949 | read_unlock(&tasklist_lock); | |
950 | ||
951 | if (!mm) | |
952 | return -EINVAL; | |
953 | ||
954 | /* | |
955 | * Check if this process has the right to modify the specified | |
956 | * process. The right exists if the process has administrative | |
7f927fcc | 957 | * capabilities, superuser privileges or the same |
39743889 CL |
958 | * userid as the target process. |
959 | */ | |
960 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
961 | (current->uid != task->suid) && (current->uid != task->uid) && | |
74c00241 | 962 | !capable(CAP_SYS_NICE)) { |
39743889 CL |
963 | err = -EPERM; |
964 | goto out; | |
965 | } | |
966 | ||
967 | task_nodes = cpuset_mems_allowed(task); | |
968 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 969 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
970 | err = -EPERM; |
971 | goto out; | |
972 | } | |
973 | ||
37b07e41 | 974 | if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
975 | err = -EINVAL; |
976 | goto out; | |
977 | } | |
978 | ||
86c3a764 DQ |
979 | err = security_task_movememory(task); |
980 | if (err) | |
981 | goto out; | |
982 | ||
511030bc | 983 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 984 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
985 | out: |
986 | mmput(mm); | |
987 | return err; | |
988 | } | |
989 | ||
990 | ||
8bccd85f CL |
991 | /* Retrieve NUMA policy */ |
992 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
993 | unsigned long __user *nmask, | |
994 | unsigned long maxnode, | |
995 | unsigned long addr, unsigned long flags) | |
996 | { | |
dbcb0f19 AB |
997 | int err; |
998 | int uninitialized_var(pval); | |
8bccd85f CL |
999 | nodemask_t nodes; |
1000 | ||
1001 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1002 | return -EINVAL; | |
1003 | ||
1004 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1005 | ||
1006 | if (err) | |
1007 | return err; | |
1008 | ||
1009 | if (policy && put_user(pval, policy)) | |
1010 | return -EFAULT; | |
1011 | ||
1012 | if (nmask) | |
1013 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1014 | ||
1015 | return err; | |
1016 | } | |
1017 | ||
1da177e4 LT |
1018 | #ifdef CONFIG_COMPAT |
1019 | ||
1020 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1021 | compat_ulong_t __user *nmask, | |
1022 | compat_ulong_t maxnode, | |
1023 | compat_ulong_t addr, compat_ulong_t flags) | |
1024 | { | |
1025 | long err; | |
1026 | unsigned long __user *nm = NULL; | |
1027 | unsigned long nr_bits, alloc_size; | |
1028 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1029 | ||
1030 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1031 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1032 | ||
1033 | if (nmask) | |
1034 | nm = compat_alloc_user_space(alloc_size); | |
1035 | ||
1036 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1037 | ||
1038 | if (!err && nmask) { | |
1039 | err = copy_from_user(bm, nm, alloc_size); | |
1040 | /* ensure entire bitmap is zeroed */ | |
1041 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1042 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1043 | } | |
1044 | ||
1045 | return err; | |
1046 | } | |
1047 | ||
1048 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1049 | compat_ulong_t maxnode) | |
1050 | { | |
1051 | long err = 0; | |
1052 | unsigned long __user *nm = NULL; | |
1053 | unsigned long nr_bits, alloc_size; | |
1054 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1055 | ||
1056 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1057 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1058 | ||
1059 | if (nmask) { | |
1060 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1061 | nm = compat_alloc_user_space(alloc_size); | |
1062 | err |= copy_to_user(nm, bm, alloc_size); | |
1063 | } | |
1064 | ||
1065 | if (err) | |
1066 | return -EFAULT; | |
1067 | ||
1068 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1069 | } | |
1070 | ||
1071 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1072 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1073 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1074 | { | |
1075 | long err = 0; | |
1076 | unsigned long __user *nm = NULL; | |
1077 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1078 | nodemask_t bm; |
1da177e4 LT |
1079 | |
1080 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1081 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1082 | ||
1083 | if (nmask) { | |
dfcd3c0d | 1084 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1085 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1086 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1087 | } |
1088 | ||
1089 | if (err) | |
1090 | return -EFAULT; | |
1091 | ||
1092 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1093 | } | |
1094 | ||
1095 | #endif | |
1096 | ||
480eccf9 LS |
1097 | /* |
1098 | * get_vma_policy(@task, @vma, @addr) | |
1099 | * @task - task for fallback if vma policy == default | |
1100 | * @vma - virtual memory area whose policy is sought | |
1101 | * @addr - address in @vma for shared policy lookup | |
1102 | * | |
1103 | * Returns effective policy for a VMA at specified address. | |
1104 | * Falls back to @task or system default policy, as necessary. | |
1105 | * Returned policy has extra reference count if shared, vma, | |
1106 | * or some other task's policy [show_numa_maps() can pass | |
1107 | * @task != current]. It is the caller's responsibility to | |
1108 | * free the reference in these cases. | |
1109 | */ | |
48fce342 CL |
1110 | static struct mempolicy * get_vma_policy(struct task_struct *task, |
1111 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 1112 | { |
6e21c8f1 | 1113 | struct mempolicy *pol = task->mempolicy; |
480eccf9 | 1114 | int shared_pol = 0; |
1da177e4 LT |
1115 | |
1116 | if (vma) { | |
480eccf9 | 1117 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8bccd85f | 1118 | pol = vma->vm_ops->get_policy(vma, addr); |
480eccf9 LS |
1119 | shared_pol = 1; /* if pol non-NULL, add ref below */ |
1120 | } else if (vma->vm_policy && | |
1da177e4 LT |
1121 | vma->vm_policy->policy != MPOL_DEFAULT) |
1122 | pol = vma->vm_policy; | |
1123 | } | |
1124 | if (!pol) | |
1125 | pol = &default_policy; | |
480eccf9 LS |
1126 | else if (!shared_pol && pol != current->mempolicy) |
1127 | mpol_get(pol); /* vma or other task's policy */ | |
1da177e4 LT |
1128 | return pol; |
1129 | } | |
1130 | ||
1131 | /* Return a zonelist representing a mempolicy */ | |
dd0fc66f | 1132 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
1da177e4 LT |
1133 | { |
1134 | int nd; | |
1135 | ||
1136 | switch (policy->policy) { | |
1137 | case MPOL_PREFERRED: | |
1138 | nd = policy->v.preferred_node; | |
1139 | if (nd < 0) | |
1140 | nd = numa_node_id(); | |
1141 | break; | |
1142 | case MPOL_BIND: | |
1143 | /* Lower zones don't get a policy applied */ | |
1144 | /* Careful: current->mems_allowed might have moved */ | |
19655d34 | 1145 | if (gfp_zone(gfp) >= policy_zone) |
1da177e4 LT |
1146 | if (cpuset_zonelist_valid_mems_allowed(policy->v.zonelist)) |
1147 | return policy->v.zonelist; | |
1148 | /*FALL THROUGH*/ | |
1149 | case MPOL_INTERLEAVE: /* should not happen */ | |
1150 | case MPOL_DEFAULT: | |
1151 | nd = numa_node_id(); | |
1152 | break; | |
1153 | default: | |
1154 | nd = 0; | |
1155 | BUG(); | |
1156 | } | |
af4ca457 | 1157 | return NODE_DATA(nd)->node_zonelists + gfp_zone(gfp); |
1da177e4 LT |
1158 | } |
1159 | ||
1160 | /* Do dynamic interleaving for a process */ | |
1161 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1162 | { | |
1163 | unsigned nid, next; | |
1164 | struct task_struct *me = current; | |
1165 | ||
1166 | nid = me->il_next; | |
dfcd3c0d | 1167 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1168 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1169 | next = first_node(policy->v.nodes); |
1da177e4 LT |
1170 | me->il_next = next; |
1171 | return nid; | |
1172 | } | |
1173 | ||
dc85da15 CL |
1174 | /* |
1175 | * Depending on the memory policy provide a node from which to allocate the | |
1176 | * next slab entry. | |
1177 | */ | |
1178 | unsigned slab_node(struct mempolicy *policy) | |
1179 | { | |
765c4507 CL |
1180 | int pol = policy ? policy->policy : MPOL_DEFAULT; |
1181 | ||
1182 | switch (pol) { | |
dc85da15 CL |
1183 | case MPOL_INTERLEAVE: |
1184 | return interleave_nodes(policy); | |
1185 | ||
1186 | case MPOL_BIND: | |
1187 | /* | |
1188 | * Follow bind policy behavior and start allocation at the | |
1189 | * first node. | |
1190 | */ | |
89fa3024 | 1191 | return zone_to_nid(policy->v.zonelist->zones[0]); |
dc85da15 CL |
1192 | |
1193 | case MPOL_PREFERRED: | |
1194 | if (policy->v.preferred_node >= 0) | |
1195 | return policy->v.preferred_node; | |
1196 | /* Fall through */ | |
1197 | ||
1198 | default: | |
1199 | return numa_node_id(); | |
1200 | } | |
1201 | } | |
1202 | ||
1da177e4 LT |
1203 | /* Do static interleaving for a VMA with known offset. */ |
1204 | static unsigned offset_il_node(struct mempolicy *pol, | |
1205 | struct vm_area_struct *vma, unsigned long off) | |
1206 | { | |
dfcd3c0d | 1207 | unsigned nnodes = nodes_weight(pol->v.nodes); |
1da177e4 LT |
1208 | unsigned target = (unsigned)off % nnodes; |
1209 | int c; | |
1210 | int nid = -1; | |
1211 | ||
1212 | c = 0; | |
1213 | do { | |
dfcd3c0d | 1214 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1215 | c++; |
1216 | } while (c <= target); | |
1da177e4 LT |
1217 | return nid; |
1218 | } | |
1219 | ||
5da7ca86 CL |
1220 | /* Determine a node number for interleave */ |
1221 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1222 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1223 | { | |
1224 | if (vma) { | |
1225 | unsigned long off; | |
1226 | ||
3b98b087 NA |
1227 | /* |
1228 | * for small pages, there is no difference between | |
1229 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1230 | * for huge pages, since vm_pgoff is in units of small | |
1231 | * pages, we need to shift off the always 0 bits to get | |
1232 | * a useful offset. | |
1233 | */ | |
1234 | BUG_ON(shift < PAGE_SHIFT); | |
1235 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1236 | off += (addr - vma->vm_start) >> shift; |
1237 | return offset_il_node(pol, vma, off); | |
1238 | } else | |
1239 | return interleave_nodes(pol); | |
1240 | } | |
1241 | ||
00ac59ad | 1242 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1243 | /* |
1244 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1245 | * @vma = virtual memory area whose policy is sought | |
1246 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1247 | * @gfp_flags = for requested zone | |
1248 | * @mpol = pointer to mempolicy pointer for reference counted 'BIND policy | |
1249 | * | |
1250 | * Returns a zonelist suitable for a huge page allocation. | |
1251 | * If the effective policy is 'BIND, returns pointer to policy's zonelist. | |
1252 | * If it is also a policy for which get_vma_policy() returns an extra | |
1253 | * reference, we must hold that reference until after allocation. | |
1254 | * In that case, return policy via @mpol so hugetlb allocation can drop | |
1255 | * the reference. For non-'BIND referenced policies, we can/do drop the | |
1256 | * reference here, so the caller doesn't need to know about the special case | |
1257 | * for default and current task policy. | |
1258 | */ | |
396faf03 | 1259 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
480eccf9 | 1260 | gfp_t gfp_flags, struct mempolicy **mpol) |
5da7ca86 CL |
1261 | { |
1262 | struct mempolicy *pol = get_vma_policy(current, vma, addr); | |
480eccf9 | 1263 | struct zonelist *zl; |
5da7ca86 | 1264 | |
480eccf9 | 1265 | *mpol = NULL; /* probably no unref needed */ |
5da7ca86 CL |
1266 | if (pol->policy == MPOL_INTERLEAVE) { |
1267 | unsigned nid; | |
1268 | ||
1269 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); | |
480eccf9 | 1270 | __mpol_free(pol); /* finished with pol */ |
396faf03 | 1271 | return NODE_DATA(nid)->node_zonelists + gfp_zone(gfp_flags); |
5da7ca86 | 1272 | } |
480eccf9 LS |
1273 | |
1274 | zl = zonelist_policy(GFP_HIGHUSER, pol); | |
1275 | if (unlikely(pol != &default_policy && pol != current->mempolicy)) { | |
1276 | if (pol->policy != MPOL_BIND) | |
1277 | __mpol_free(pol); /* finished with pol */ | |
1278 | else | |
1279 | *mpol = pol; /* unref needed after allocation */ | |
1280 | } | |
1281 | return zl; | |
5da7ca86 | 1282 | } |
00ac59ad | 1283 | #endif |
5da7ca86 | 1284 | |
1da177e4 LT |
1285 | /* Allocate a page in interleaved policy. |
1286 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1287 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1288 | unsigned nid) | |
1da177e4 LT |
1289 | { |
1290 | struct zonelist *zl; | |
1291 | struct page *page; | |
1292 | ||
af4ca457 | 1293 | zl = NODE_DATA(nid)->node_zonelists + gfp_zone(gfp); |
1da177e4 | 1294 | page = __alloc_pages(gfp, order, zl); |
ca889e6c CL |
1295 | if (page && page_zone(page) == zl->zones[0]) |
1296 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); | |
1da177e4 LT |
1297 | return page; |
1298 | } | |
1299 | ||
1300 | /** | |
1301 | * alloc_page_vma - Allocate a page for a VMA. | |
1302 | * | |
1303 | * @gfp: | |
1304 | * %GFP_USER user allocation. | |
1305 | * %GFP_KERNEL kernel allocations, | |
1306 | * %GFP_HIGHMEM highmem/user allocations, | |
1307 | * %GFP_FS allocation should not call back into a file system. | |
1308 | * %GFP_ATOMIC don't sleep. | |
1309 | * | |
1310 | * @vma: Pointer to VMA or NULL if not available. | |
1311 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1312 | * | |
1313 | * This function allocates a page from the kernel page pool and applies | |
1314 | * a NUMA policy associated with the VMA or the current process. | |
1315 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1316 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1317 | * all allocations for pages that will be mapped into | |
1318 | * user space. Returns NULL when no page can be allocated. | |
1319 | * | |
1320 | * Should be called with the mm_sem of the vma hold. | |
1321 | */ | |
1322 | struct page * | |
dd0fc66f | 1323 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1324 | { |
6e21c8f1 | 1325 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1326 | struct zonelist *zl; |
1da177e4 | 1327 | |
cf2a473c | 1328 | cpuset_update_task_memory_state(); |
1da177e4 LT |
1329 | |
1330 | if (unlikely(pol->policy == MPOL_INTERLEAVE)) { | |
1331 | unsigned nid; | |
5da7ca86 CL |
1332 | |
1333 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
1da177e4 LT |
1334 | return alloc_page_interleave(gfp, 0, nid); |
1335 | } | |
480eccf9 LS |
1336 | zl = zonelist_policy(gfp, pol); |
1337 | if (pol != &default_policy && pol != current->mempolicy) { | |
1338 | /* | |
1339 | * slow path: ref counted policy -- shared or vma | |
1340 | */ | |
1341 | struct page *page = __alloc_pages(gfp, 0, zl); | |
1342 | __mpol_free(pol); | |
1343 | return page; | |
1344 | } | |
1345 | /* | |
1346 | * fast path: default or task policy | |
1347 | */ | |
1348 | return __alloc_pages(gfp, 0, zl); | |
1da177e4 LT |
1349 | } |
1350 | ||
1351 | /** | |
1352 | * alloc_pages_current - Allocate pages. | |
1353 | * | |
1354 | * @gfp: | |
1355 | * %GFP_USER user allocation, | |
1356 | * %GFP_KERNEL kernel allocation, | |
1357 | * %GFP_HIGHMEM highmem allocation, | |
1358 | * %GFP_FS don't call back into a file system. | |
1359 | * %GFP_ATOMIC don't sleep. | |
1360 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1361 | * | |
1362 | * Allocate a page from the kernel page pool. When not in | |
1363 | * interrupt context and apply the current process NUMA policy. | |
1364 | * Returns NULL when no page can be allocated. | |
1365 | * | |
cf2a473c | 1366 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1367 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1368 | * 2) allocating for current task (not interrupt). | |
1369 | */ | |
dd0fc66f | 1370 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1371 | { |
1372 | struct mempolicy *pol = current->mempolicy; | |
1373 | ||
1374 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1375 | cpuset_update_task_memory_state(); |
9b819d20 | 1376 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 LT |
1377 | pol = &default_policy; |
1378 | if (pol->policy == MPOL_INTERLEAVE) | |
1379 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); | |
1380 | return __alloc_pages(gfp, order, zonelist_policy(gfp, pol)); | |
1381 | } | |
1382 | EXPORT_SYMBOL(alloc_pages_current); | |
1383 | ||
4225399a PJ |
1384 | /* |
1385 | * If mpol_copy() sees current->cpuset == cpuset_being_rebound, then it | |
1386 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() | |
1387 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1388 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1389 | * further kernel/cpuset.c update_nodemask(). | |
1390 | */ | |
1391 | void *cpuset_being_rebound; | |
1392 | ||
1da177e4 LT |
1393 | /* Slow path of a mempolicy copy */ |
1394 | struct mempolicy *__mpol_copy(struct mempolicy *old) | |
1395 | { | |
1396 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1397 | ||
1398 | if (!new) | |
1399 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1400 | if (current_cpuset_is_being_rebound()) { |
1401 | nodemask_t mems = cpuset_mems_allowed(current); | |
1402 | mpol_rebind_policy(old, &mems); | |
1403 | } | |
1da177e4 LT |
1404 | *new = *old; |
1405 | atomic_set(&new->refcnt, 1); | |
1406 | if (new->policy == MPOL_BIND) { | |
1407 | int sz = ksize(old->v.zonelist); | |
e94b1766 | 1408 | new->v.zonelist = kmemdup(old->v.zonelist, sz, GFP_KERNEL); |
1da177e4 LT |
1409 | if (!new->v.zonelist) { |
1410 | kmem_cache_free(policy_cache, new); | |
1411 | return ERR_PTR(-ENOMEM); | |
1412 | } | |
1da177e4 LT |
1413 | } |
1414 | return new; | |
1415 | } | |
1416 | ||
1417 | /* Slow path of a mempolicy comparison */ | |
1418 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1419 | { | |
1420 | if (!a || !b) | |
1421 | return 0; | |
1422 | if (a->policy != b->policy) | |
1423 | return 0; | |
1424 | switch (a->policy) { | |
1425 | case MPOL_DEFAULT: | |
1426 | return 1; | |
1427 | case MPOL_INTERLEAVE: | |
dfcd3c0d | 1428 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 LT |
1429 | case MPOL_PREFERRED: |
1430 | return a->v.preferred_node == b->v.preferred_node; | |
1431 | case MPOL_BIND: { | |
1432 | int i; | |
1433 | for (i = 0; a->v.zonelist->zones[i]; i++) | |
1434 | if (a->v.zonelist->zones[i] != b->v.zonelist->zones[i]) | |
1435 | return 0; | |
1436 | return b->v.zonelist->zones[i] == NULL; | |
1437 | } | |
1438 | default: | |
1439 | BUG(); | |
1440 | return 0; | |
1441 | } | |
1442 | } | |
1443 | ||
1444 | /* Slow path of a mpol destructor. */ | |
1445 | void __mpol_free(struct mempolicy *p) | |
1446 | { | |
1447 | if (!atomic_dec_and_test(&p->refcnt)) | |
1448 | return; | |
1449 | if (p->policy == MPOL_BIND) | |
1450 | kfree(p->v.zonelist); | |
1451 | p->policy = MPOL_DEFAULT; | |
1452 | kmem_cache_free(policy_cache, p); | |
1453 | } | |
1454 | ||
1da177e4 LT |
1455 | /* |
1456 | * Shared memory backing store policy support. | |
1457 | * | |
1458 | * Remember policies even when nobody has shared memory mapped. | |
1459 | * The policies are kept in Red-Black tree linked from the inode. | |
1460 | * They are protected by the sp->lock spinlock, which should be held | |
1461 | * for any accesses to the tree. | |
1462 | */ | |
1463 | ||
1464 | /* lookup first element intersecting start-end */ | |
1465 | /* Caller holds sp->lock */ | |
1466 | static struct sp_node * | |
1467 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1468 | { | |
1469 | struct rb_node *n = sp->root.rb_node; | |
1470 | ||
1471 | while (n) { | |
1472 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1473 | ||
1474 | if (start >= p->end) | |
1475 | n = n->rb_right; | |
1476 | else if (end <= p->start) | |
1477 | n = n->rb_left; | |
1478 | else | |
1479 | break; | |
1480 | } | |
1481 | if (!n) | |
1482 | return NULL; | |
1483 | for (;;) { | |
1484 | struct sp_node *w = NULL; | |
1485 | struct rb_node *prev = rb_prev(n); | |
1486 | if (!prev) | |
1487 | break; | |
1488 | w = rb_entry(prev, struct sp_node, nd); | |
1489 | if (w->end <= start) | |
1490 | break; | |
1491 | n = prev; | |
1492 | } | |
1493 | return rb_entry(n, struct sp_node, nd); | |
1494 | } | |
1495 | ||
1496 | /* Insert a new shared policy into the list. */ | |
1497 | /* Caller holds sp->lock */ | |
1498 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1499 | { | |
1500 | struct rb_node **p = &sp->root.rb_node; | |
1501 | struct rb_node *parent = NULL; | |
1502 | struct sp_node *nd; | |
1503 | ||
1504 | while (*p) { | |
1505 | parent = *p; | |
1506 | nd = rb_entry(parent, struct sp_node, nd); | |
1507 | if (new->start < nd->start) | |
1508 | p = &(*p)->rb_left; | |
1509 | else if (new->end > nd->end) | |
1510 | p = &(*p)->rb_right; | |
1511 | else | |
1512 | BUG(); | |
1513 | } | |
1514 | rb_link_node(&new->nd, parent, p); | |
1515 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1516 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
1da177e4 LT |
1517 | new->policy ? new->policy->policy : 0); |
1518 | } | |
1519 | ||
1520 | /* Find shared policy intersecting idx */ | |
1521 | struct mempolicy * | |
1522 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1523 | { | |
1524 | struct mempolicy *pol = NULL; | |
1525 | struct sp_node *sn; | |
1526 | ||
1527 | if (!sp->root.rb_node) | |
1528 | return NULL; | |
1529 | spin_lock(&sp->lock); | |
1530 | sn = sp_lookup(sp, idx, idx+1); | |
1531 | if (sn) { | |
1532 | mpol_get(sn->policy); | |
1533 | pol = sn->policy; | |
1534 | } | |
1535 | spin_unlock(&sp->lock); | |
1536 | return pol; | |
1537 | } | |
1538 | ||
1539 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1540 | { | |
140d5a49 | 1541 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 LT |
1542 | rb_erase(&n->nd, &sp->root); |
1543 | mpol_free(n->policy); | |
1544 | kmem_cache_free(sn_cache, n); | |
1545 | } | |
1546 | ||
dbcb0f19 AB |
1547 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
1548 | struct mempolicy *pol) | |
1da177e4 LT |
1549 | { |
1550 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1551 | ||
1552 | if (!n) | |
1553 | return NULL; | |
1554 | n->start = start; | |
1555 | n->end = end; | |
1556 | mpol_get(pol); | |
1557 | n->policy = pol; | |
1558 | return n; | |
1559 | } | |
1560 | ||
1561 | /* Replace a policy range. */ | |
1562 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1563 | unsigned long end, struct sp_node *new) | |
1564 | { | |
1565 | struct sp_node *n, *new2 = NULL; | |
1566 | ||
1567 | restart: | |
1568 | spin_lock(&sp->lock); | |
1569 | n = sp_lookup(sp, start, end); | |
1570 | /* Take care of old policies in the same range. */ | |
1571 | while (n && n->start < end) { | |
1572 | struct rb_node *next = rb_next(&n->nd); | |
1573 | if (n->start >= start) { | |
1574 | if (n->end <= end) | |
1575 | sp_delete(sp, n); | |
1576 | else | |
1577 | n->start = end; | |
1578 | } else { | |
1579 | /* Old policy spanning whole new range. */ | |
1580 | if (n->end > end) { | |
1581 | if (!new2) { | |
1582 | spin_unlock(&sp->lock); | |
1583 | new2 = sp_alloc(end, n->end, n->policy); | |
1584 | if (!new2) | |
1585 | return -ENOMEM; | |
1586 | goto restart; | |
1587 | } | |
1588 | n->end = start; | |
1589 | sp_insert(sp, new2); | |
1590 | new2 = NULL; | |
1591 | break; | |
1592 | } else | |
1593 | n->end = start; | |
1594 | } | |
1595 | if (!next) | |
1596 | break; | |
1597 | n = rb_entry(next, struct sp_node, nd); | |
1598 | } | |
1599 | if (new) | |
1600 | sp_insert(sp, new); | |
1601 | spin_unlock(&sp->lock); | |
1602 | if (new2) { | |
1603 | mpol_free(new2->policy); | |
1604 | kmem_cache_free(sn_cache, new2); | |
1605 | } | |
1606 | return 0; | |
1607 | } | |
1608 | ||
7339ff83 RH |
1609 | void mpol_shared_policy_init(struct shared_policy *info, int policy, |
1610 | nodemask_t *policy_nodes) | |
1611 | { | |
1612 | info->root = RB_ROOT; | |
1613 | spin_lock_init(&info->lock); | |
1614 | ||
1615 | if (policy != MPOL_DEFAULT) { | |
1616 | struct mempolicy *newpol; | |
1617 | ||
1618 | /* Falls back to MPOL_DEFAULT on any error */ | |
1619 | newpol = mpol_new(policy, policy_nodes); | |
1620 | if (!IS_ERR(newpol)) { | |
1621 | /* Create pseudo-vma that contains just the policy */ | |
1622 | struct vm_area_struct pvma; | |
1623 | ||
1624 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1625 | /* Policy covers entire file */ | |
1626 | pvma.vm_end = TASK_SIZE; | |
1627 | mpol_set_shared_policy(info, &pvma, newpol); | |
1628 | mpol_free(newpol); | |
1629 | } | |
1630 | } | |
1631 | } | |
1632 | ||
1da177e4 LT |
1633 | int mpol_set_shared_policy(struct shared_policy *info, |
1634 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1635 | { | |
1636 | int err; | |
1637 | struct sp_node *new = NULL; | |
1638 | unsigned long sz = vma_pages(vma); | |
1639 | ||
140d5a49 | 1640 | pr_debug("set_shared_policy %lx sz %lu %d %lx\n", |
1da177e4 LT |
1641 | vma->vm_pgoff, |
1642 | sz, npol? npol->policy : -1, | |
140d5a49 | 1643 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1644 | |
1645 | if (npol) { | |
1646 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1647 | if (!new) | |
1648 | return -ENOMEM; | |
1649 | } | |
1650 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1651 | if (err && new) | |
1652 | kmem_cache_free(sn_cache, new); | |
1653 | return err; | |
1654 | } | |
1655 | ||
1656 | /* Free a backing policy store on inode delete. */ | |
1657 | void mpol_free_shared_policy(struct shared_policy *p) | |
1658 | { | |
1659 | struct sp_node *n; | |
1660 | struct rb_node *next; | |
1661 | ||
1662 | if (!p->root.rb_node) | |
1663 | return; | |
1664 | spin_lock(&p->lock); | |
1665 | next = rb_first(&p->root); | |
1666 | while (next) { | |
1667 | n = rb_entry(next, struct sp_node, nd); | |
1668 | next = rb_next(&n->nd); | |
90c5029e | 1669 | rb_erase(&n->nd, &p->root); |
1da177e4 LT |
1670 | mpol_free(n->policy); |
1671 | kmem_cache_free(sn_cache, n); | |
1672 | } | |
1673 | spin_unlock(&p->lock); | |
1da177e4 LT |
1674 | } |
1675 | ||
1676 | /* assumes fs == KERNEL_DS */ | |
1677 | void __init numa_policy_init(void) | |
1678 | { | |
b71636e2 PM |
1679 | nodemask_t interleave_nodes; |
1680 | unsigned long largest = 0; | |
1681 | int nid, prefer = 0; | |
1682 | ||
1da177e4 LT |
1683 | policy_cache = kmem_cache_create("numa_policy", |
1684 | sizeof(struct mempolicy), | |
20c2df83 | 1685 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
1686 | |
1687 | sn_cache = kmem_cache_create("shared_policy_node", | |
1688 | sizeof(struct sp_node), | |
20c2df83 | 1689 | 0, SLAB_PANIC, NULL); |
1da177e4 | 1690 | |
b71636e2 PM |
1691 | /* |
1692 | * Set interleaving policy for system init. Interleaving is only | |
1693 | * enabled across suitably sized nodes (default is >= 16MB), or | |
1694 | * fall back to the largest node if they're all smaller. | |
1695 | */ | |
1696 | nodes_clear(interleave_nodes); | |
56bbd65d | 1697 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
1698 | unsigned long total_pages = node_present_pages(nid); |
1699 | ||
1700 | /* Preserve the largest node */ | |
1701 | if (largest < total_pages) { | |
1702 | largest = total_pages; | |
1703 | prefer = nid; | |
1704 | } | |
1705 | ||
1706 | /* Interleave this node? */ | |
1707 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
1708 | node_set(nid, interleave_nodes); | |
1709 | } | |
1710 | ||
1711 | /* All too small, use the largest */ | |
1712 | if (unlikely(nodes_empty(interleave_nodes))) | |
1713 | node_set(prefer, interleave_nodes); | |
1da177e4 | 1714 | |
b71636e2 | 1715 | if (do_set_mempolicy(MPOL_INTERLEAVE, &interleave_nodes)) |
1da177e4 LT |
1716 | printk("numa_policy_init: interleaving failed\n"); |
1717 | } | |
1718 | ||
8bccd85f | 1719 | /* Reset policy of current process to default */ |
1da177e4 LT |
1720 | void numa_default_policy(void) |
1721 | { | |
8bccd85f | 1722 | do_set_mempolicy(MPOL_DEFAULT, NULL); |
1da177e4 | 1723 | } |
68860ec1 PJ |
1724 | |
1725 | /* Migrate a policy to a different set of nodes */ | |
dbcb0f19 AB |
1726 | static void mpol_rebind_policy(struct mempolicy *pol, |
1727 | const nodemask_t *newmask) | |
68860ec1 | 1728 | { |
74cb2155 | 1729 | nodemask_t *mpolmask; |
68860ec1 PJ |
1730 | nodemask_t tmp; |
1731 | ||
1732 | if (!pol) | |
1733 | return; | |
74cb2155 PJ |
1734 | mpolmask = &pol->cpuset_mems_allowed; |
1735 | if (nodes_equal(*mpolmask, *newmask)) | |
1736 | return; | |
68860ec1 PJ |
1737 | |
1738 | switch (pol->policy) { | |
1739 | case MPOL_DEFAULT: | |
1740 | break; | |
1741 | case MPOL_INTERLEAVE: | |
74cb2155 | 1742 | nodes_remap(tmp, pol->v.nodes, *mpolmask, *newmask); |
68860ec1 | 1743 | pol->v.nodes = tmp; |
74cb2155 PJ |
1744 | *mpolmask = *newmask; |
1745 | current->il_next = node_remap(current->il_next, | |
1746 | *mpolmask, *newmask); | |
68860ec1 PJ |
1747 | break; |
1748 | case MPOL_PREFERRED: | |
1749 | pol->v.preferred_node = node_remap(pol->v.preferred_node, | |
74cb2155 PJ |
1750 | *mpolmask, *newmask); |
1751 | *mpolmask = *newmask; | |
68860ec1 PJ |
1752 | break; |
1753 | case MPOL_BIND: { | |
1754 | nodemask_t nodes; | |
1755 | struct zone **z; | |
1756 | struct zonelist *zonelist; | |
1757 | ||
1758 | nodes_clear(nodes); | |
1759 | for (z = pol->v.zonelist->zones; *z; z++) | |
89fa3024 | 1760 | node_set(zone_to_nid(*z), nodes); |
74cb2155 | 1761 | nodes_remap(tmp, nodes, *mpolmask, *newmask); |
68860ec1 PJ |
1762 | nodes = tmp; |
1763 | ||
1764 | zonelist = bind_zonelist(&nodes); | |
1765 | ||
1766 | /* If no mem, then zonelist is NULL and we keep old zonelist. | |
1767 | * If that old zonelist has no remaining mems_allowed nodes, | |
1768 | * then zonelist_policy() will "FALL THROUGH" to MPOL_DEFAULT. | |
1769 | */ | |
1770 | ||
8af5e2eb | 1771 | if (!IS_ERR(zonelist)) { |
68860ec1 PJ |
1772 | /* Good - got mem - substitute new zonelist */ |
1773 | kfree(pol->v.zonelist); | |
1774 | pol->v.zonelist = zonelist; | |
1775 | } | |
74cb2155 | 1776 | *mpolmask = *newmask; |
68860ec1 PJ |
1777 | break; |
1778 | } | |
1779 | default: | |
1780 | BUG(); | |
1781 | break; | |
1782 | } | |
1783 | } | |
1784 | ||
1785 | /* | |
74cb2155 PJ |
1786 | * Wrapper for mpol_rebind_policy() that just requires task |
1787 | * pointer, and updates task mempolicy. | |
68860ec1 | 1788 | */ |
74cb2155 PJ |
1789 | |
1790 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
68860ec1 | 1791 | { |
74cb2155 | 1792 | mpol_rebind_policy(tsk->mempolicy, new); |
68860ec1 | 1793 | } |
1a75a6c8 | 1794 | |
4225399a PJ |
1795 | /* |
1796 | * Rebind each vma in mm to new nodemask. | |
1797 | * | |
1798 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
1799 | */ | |
1800 | ||
1801 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
1802 | { | |
1803 | struct vm_area_struct *vma; | |
1804 | ||
1805 | down_write(&mm->mmap_sem); | |
1806 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
1807 | mpol_rebind_policy(vma->vm_policy, new); | |
1808 | up_write(&mm->mmap_sem); | |
1809 | } | |
1810 | ||
1a75a6c8 CL |
1811 | /* |
1812 | * Display pages allocated per node and memory policy via /proc. | |
1813 | */ | |
1814 | ||
15ad7cdc HD |
1815 | static const char * const policy_types[] = |
1816 | { "default", "prefer", "bind", "interleave" }; | |
1a75a6c8 CL |
1817 | |
1818 | /* | |
1819 | * Convert a mempolicy into a string. | |
1820 | * Returns the number of characters in buffer (if positive) | |
1821 | * or an error (negative) | |
1822 | */ | |
1823 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1824 | { | |
1825 | char *p = buffer; | |
1826 | int l; | |
1827 | nodemask_t nodes; | |
1828 | int mode = pol ? pol->policy : MPOL_DEFAULT; | |
1829 | ||
1830 | switch (mode) { | |
1831 | case MPOL_DEFAULT: | |
1832 | nodes_clear(nodes); | |
1833 | break; | |
1834 | ||
1835 | case MPOL_PREFERRED: | |
1836 | nodes_clear(nodes); | |
1837 | node_set(pol->v.preferred_node, nodes); | |
1838 | break; | |
1839 | ||
1840 | case MPOL_BIND: | |
1841 | get_zonemask(pol, &nodes); | |
1842 | break; | |
1843 | ||
1844 | case MPOL_INTERLEAVE: | |
1845 | nodes = pol->v.nodes; | |
1846 | break; | |
1847 | ||
1848 | default: | |
1849 | BUG(); | |
1850 | return -EFAULT; | |
1851 | } | |
1852 | ||
1853 | l = strlen(policy_types[mode]); | |
1854 | if (buffer + maxlen < p + l + 1) | |
1855 | return -ENOSPC; | |
1856 | ||
1857 | strcpy(p, policy_types[mode]); | |
1858 | p += l; | |
1859 | ||
1860 | if (!nodes_empty(nodes)) { | |
1861 | if (buffer + maxlen < p + 2) | |
1862 | return -ENOSPC; | |
1863 | *p++ = '='; | |
1864 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
1865 | } | |
1866 | return p - buffer; | |
1867 | } | |
1868 | ||
1869 | struct numa_maps { | |
1870 | unsigned long pages; | |
1871 | unsigned long anon; | |
397874df CL |
1872 | unsigned long active; |
1873 | unsigned long writeback; | |
1a75a6c8 | 1874 | unsigned long mapcount_max; |
397874df CL |
1875 | unsigned long dirty; |
1876 | unsigned long swapcache; | |
1a75a6c8 CL |
1877 | unsigned long node[MAX_NUMNODES]; |
1878 | }; | |
1879 | ||
397874df | 1880 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
1881 | { |
1882 | struct numa_maps *md = private; | |
1883 | int count = page_mapcount(page); | |
1884 | ||
397874df CL |
1885 | md->pages++; |
1886 | if (pte_dirty || PageDirty(page)) | |
1887 | md->dirty++; | |
1a75a6c8 | 1888 | |
397874df CL |
1889 | if (PageSwapCache(page)) |
1890 | md->swapcache++; | |
1a75a6c8 | 1891 | |
397874df CL |
1892 | if (PageActive(page)) |
1893 | md->active++; | |
1894 | ||
1895 | if (PageWriteback(page)) | |
1896 | md->writeback++; | |
1a75a6c8 CL |
1897 | |
1898 | if (PageAnon(page)) | |
1899 | md->anon++; | |
1900 | ||
397874df CL |
1901 | if (count > md->mapcount_max) |
1902 | md->mapcount_max = count; | |
1903 | ||
1a75a6c8 | 1904 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
1905 | } |
1906 | ||
7f709ed0 | 1907 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
1908 | static void check_huge_range(struct vm_area_struct *vma, |
1909 | unsigned long start, unsigned long end, | |
1910 | struct numa_maps *md) | |
1911 | { | |
1912 | unsigned long addr; | |
1913 | struct page *page; | |
1914 | ||
1915 | for (addr = start; addr < end; addr += HPAGE_SIZE) { | |
1916 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); | |
1917 | pte_t pte; | |
1918 | ||
1919 | if (!ptep) | |
1920 | continue; | |
1921 | ||
1922 | pte = *ptep; | |
1923 | if (pte_none(pte)) | |
1924 | continue; | |
1925 | ||
1926 | page = pte_page(pte); | |
1927 | if (!page) | |
1928 | continue; | |
1929 | ||
1930 | gather_stats(page, md, pte_dirty(*ptep)); | |
1931 | } | |
1932 | } | |
7f709ed0 AM |
1933 | #else |
1934 | static inline void check_huge_range(struct vm_area_struct *vma, | |
1935 | unsigned long start, unsigned long end, | |
1936 | struct numa_maps *md) | |
1937 | { | |
1938 | } | |
1939 | #endif | |
397874df | 1940 | |
1a75a6c8 CL |
1941 | int show_numa_map(struct seq_file *m, void *v) |
1942 | { | |
99f89551 | 1943 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
1944 | struct vm_area_struct *vma = v; |
1945 | struct numa_maps *md; | |
397874df CL |
1946 | struct file *file = vma->vm_file; |
1947 | struct mm_struct *mm = vma->vm_mm; | |
480eccf9 | 1948 | struct mempolicy *pol; |
1a75a6c8 CL |
1949 | int n; |
1950 | char buffer[50]; | |
1951 | ||
397874df | 1952 | if (!mm) |
1a75a6c8 CL |
1953 | return 0; |
1954 | ||
1955 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
1956 | if (!md) | |
1957 | return 0; | |
1958 | ||
480eccf9 LS |
1959 | pol = get_vma_policy(priv->task, vma, vma->vm_start); |
1960 | mpol_to_str(buffer, sizeof(buffer), pol); | |
1961 | /* | |
1962 | * unref shared or other task's mempolicy | |
1963 | */ | |
1964 | if (pol != &default_policy && pol != current->mempolicy) | |
1965 | __mpol_free(pol); | |
397874df CL |
1966 | |
1967 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
1968 | ||
1969 | if (file) { | |
1970 | seq_printf(m, " file="); | |
e9536ae7 | 1971 | seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n\t= "); |
397874df CL |
1972 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
1973 | seq_printf(m, " heap"); | |
1974 | } else if (vma->vm_start <= mm->start_stack && | |
1975 | vma->vm_end >= mm->start_stack) { | |
1976 | seq_printf(m, " stack"); | |
1977 | } | |
1978 | ||
1979 | if (is_vm_hugetlb_page(vma)) { | |
1980 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
1981 | seq_printf(m, " huge"); | |
1982 | } else { | |
a57ebfdb | 1983 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
56bbd65d | 1984 | &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md); |
397874df CL |
1985 | } |
1986 | ||
1987 | if (!md->pages) | |
1988 | goto out; | |
1a75a6c8 | 1989 | |
397874df CL |
1990 | if (md->anon) |
1991 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 1992 | |
397874df CL |
1993 | if (md->dirty) |
1994 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 1995 | |
397874df CL |
1996 | if (md->pages != md->anon && md->pages != md->dirty) |
1997 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 1998 | |
397874df CL |
1999 | if (md->mapcount_max > 1) |
2000 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 2001 | |
397874df CL |
2002 | if (md->swapcache) |
2003 | seq_printf(m," swapcache=%lu", md->swapcache); | |
2004 | ||
2005 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
2006 | seq_printf(m," active=%lu", md->active); | |
2007 | ||
2008 | if (md->writeback) | |
2009 | seq_printf(m," writeback=%lu", md->writeback); | |
2010 | ||
56bbd65d | 2011 | for_each_node_state(n, N_HIGH_MEMORY) |
397874df CL |
2012 | if (md->node[n]) |
2013 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
2014 | out: | |
2015 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
2016 | kfree(md); |
2017 | ||
2018 | if (m->count < m->size) | |
99f89551 | 2019 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
2020 | return 0; |
2021 | } | |
2022 |