Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Simple NUMA memory policy for the Linux kernel. | |
3 | * | |
4 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 5 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
6 | * Subject to the GNU Public License, version 2. |
7 | * | |
8 | * NUMA policy allows the user to give hints in which node(s) memory should | |
9 | * be allocated. | |
10 | * | |
11 | * Support four policies per VMA and per process: | |
12 | * | |
13 | * The VMA policy has priority over the process policy for a page fault. | |
14 | * | |
15 | * interleave Allocate memory interleaved over a set of nodes, | |
16 | * with normal fallback if it fails. | |
17 | * For VMA based allocations this interleaves based on the | |
18 | * offset into the backing object or offset into the mapping | |
19 | * for anonymous memory. For process policy an process counter | |
20 | * is used. | |
8bccd85f | 21 | * |
1da177e4 LT |
22 | * bind Only allocate memory on a specific set of nodes, |
23 | * no fallback. | |
8bccd85f CL |
24 | * FIXME: memory is allocated starting with the first node |
25 | * to the last. It would be better if bind would truly restrict | |
26 | * the allocation to memory nodes instead | |
27 | * | |
1da177e4 LT |
28 | * preferred Try a specific node first before normal fallback. |
29 | * As a special case node -1 here means do the allocation | |
30 | * on the local CPU. This is normally identical to default, | |
31 | * but useful to set in a VMA when you have a non default | |
32 | * process policy. | |
8bccd85f | 33 | * |
1da177e4 LT |
34 | * default Allocate on the local node first, or when on a VMA |
35 | * use the process policy. This is what Linux always did | |
36 | * in a NUMA aware kernel and still does by, ahem, default. | |
37 | * | |
38 | * The process policy is applied for most non interrupt memory allocations | |
39 | * in that process' context. Interrupts ignore the policies and always | |
40 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
41 | * allocations for a VMA in the VM. | |
42 | * | |
43 | * Currently there are a few corner cases in swapping where the policy | |
44 | * is not applied, but the majority should be handled. When process policy | |
45 | * is used it is not remembered over swap outs/swap ins. | |
46 | * | |
47 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
48 | * requesting a lower zone just use default policy. This implies that | |
49 | * on systems with highmem kernel lowmem allocation don't get policied. | |
50 | * Same with GFP_DMA allocations. | |
51 | * | |
52 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between | |
53 | * all users and remembered even when nobody has memory mapped. | |
54 | */ | |
55 | ||
56 | /* Notebook: | |
57 | fix mmap readahead to honour policy and enable policy for any page cache | |
58 | object | |
59 | statistics for bigpages | |
60 | global policy for page cache? currently it uses process policy. Requires | |
61 | first item above. | |
62 | handle mremap for shared memory (currently ignored for the policy) | |
63 | grows down? | |
64 | make bind policy root only? It can trigger oom much faster and the | |
65 | kernel is not always grateful with that. | |
1da177e4 LT |
66 | */ |
67 | ||
68 | #include <linux/mempolicy.h> | |
69 | #include <linux/mm.h> | |
70 | #include <linux/highmem.h> | |
71 | #include <linux/hugetlb.h> | |
72 | #include <linux/kernel.h> | |
73 | #include <linux/sched.h> | |
1da177e4 LT |
74 | #include <linux/nodemask.h> |
75 | #include <linux/cpuset.h> | |
76 | #include <linux/gfp.h> | |
77 | #include <linux/slab.h> | |
78 | #include <linux/string.h> | |
79 | #include <linux/module.h> | |
b488893a | 80 | #include <linux/nsproxy.h> |
1da177e4 LT |
81 | #include <linux/interrupt.h> |
82 | #include <linux/init.h> | |
83 | #include <linux/compat.h> | |
dc9aa5b9 | 84 | #include <linux/swap.h> |
1a75a6c8 CL |
85 | #include <linux/seq_file.h> |
86 | #include <linux/proc_fs.h> | |
b20a3503 | 87 | #include <linux/migrate.h> |
95a402c3 | 88 | #include <linux/rmap.h> |
86c3a764 | 89 | #include <linux/security.h> |
dbcb0f19 | 90 | #include <linux/syscalls.h> |
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 | 108 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
45c4745a | 109 | .mode = MPOL_DEFAULT, |
1da177e4 LT |
110 | }; |
111 | ||
37012946 DR |
112 | static const struct mempolicy_operations { |
113 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
114 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); | |
115 | } mpol_ops[MPOL_MAX]; | |
116 | ||
19770b32 | 117 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 118 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 119 | { |
19770b32 | 120 | int nd, k; |
1da177e4 | 121 | |
19770b32 MG |
122 | /* Check that there is something useful in this mask */ |
123 | k = policy_zone; | |
124 | ||
125 | for_each_node_mask(nd, *nodemask) { | |
126 | struct zone *z; | |
127 | ||
128 | for (k = 0; k <= policy_zone; k++) { | |
129 | z = &NODE_DATA(nd)->node_zones[k]; | |
130 | if (z->present_pages > 0) | |
131 | return 1; | |
dd942ae3 | 132 | } |
8af5e2eb | 133 | } |
19770b32 MG |
134 | |
135 | return 0; | |
1da177e4 LT |
136 | } |
137 | ||
f5b087b5 DR |
138 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
139 | { | |
4c50bc01 DR |
140 | return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES); |
141 | } | |
142 | ||
143 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
144 | const nodemask_t *rel) | |
145 | { | |
146 | nodemask_t tmp; | |
147 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
148 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
149 | } |
150 | ||
37012946 DR |
151 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
152 | { | |
153 | if (nodes_empty(*nodes)) | |
154 | return -EINVAL; | |
155 | pol->v.nodes = *nodes; | |
156 | return 0; | |
157 | } | |
158 | ||
159 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
160 | { | |
161 | if (!nodes) | |
162 | pol->v.preferred_node = -1; /* local allocation */ | |
163 | else if (nodes_empty(*nodes)) | |
164 | return -EINVAL; /* no allowed nodes */ | |
165 | else | |
166 | pol->v.preferred_node = first_node(*nodes); | |
167 | return 0; | |
168 | } | |
169 | ||
170 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
171 | { | |
172 | if (!is_valid_nodemask(nodes)) | |
173 | return -EINVAL; | |
174 | pol->v.nodes = *nodes; | |
175 | return 0; | |
176 | } | |
177 | ||
1da177e4 | 178 | /* Create a new policy */ |
028fec41 DR |
179 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
180 | nodemask_t *nodes) | |
1da177e4 LT |
181 | { |
182 | struct mempolicy *policy; | |
f5b087b5 | 183 | nodemask_t cpuset_context_nmask; |
37012946 | 184 | int ret; |
1da177e4 | 185 | |
028fec41 DR |
186 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
187 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 188 | |
3e1f0645 DR |
189 | if (mode == MPOL_DEFAULT) { |
190 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 191 | return ERR_PTR(-EINVAL); |
3e1f0645 | 192 | return NULL; |
37012946 | 193 | } |
3e1f0645 DR |
194 | VM_BUG_ON(!nodes); |
195 | ||
196 | /* | |
197 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
198 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
199 | * All other modes require a valid pointer to a non-empty nodemask. | |
200 | */ | |
201 | if (mode == MPOL_PREFERRED) { | |
202 | if (nodes_empty(*nodes)) { | |
203 | if (((flags & MPOL_F_STATIC_NODES) || | |
204 | (flags & MPOL_F_RELATIVE_NODES))) | |
205 | return ERR_PTR(-EINVAL); | |
206 | nodes = NULL; /* flag local alloc */ | |
207 | } | |
208 | } else if (nodes_empty(*nodes)) | |
209 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
210 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
211 | if (!policy) | |
212 | return ERR_PTR(-ENOMEM); | |
213 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 214 | policy->mode = mode; |
3e1f0645 | 215 | policy->flags = flags; |
37012946 | 216 | |
3e1f0645 DR |
217 | if (nodes) { |
218 | /* | |
219 | * cpuset related setup doesn't apply to local allocation | |
220 | */ | |
37012946 DR |
221 | cpuset_update_task_memory_state(); |
222 | if (flags & MPOL_F_RELATIVE_NODES) | |
223 | mpol_relative_nodemask(&cpuset_context_nmask, nodes, | |
224 | &cpuset_current_mems_allowed); | |
225 | else | |
226 | nodes_and(cpuset_context_nmask, *nodes, | |
227 | cpuset_current_mems_allowed); | |
228 | if (mpol_store_user_nodemask(policy)) | |
229 | policy->w.user_nodemask = *nodes; | |
230 | else | |
231 | policy->w.cpuset_mems_allowed = | |
232 | cpuset_mems_allowed(current); | |
233 | } | |
234 | ||
235 | ret = mpol_ops[mode].create(policy, | |
3e1f0645 | 236 | nodes ? &cpuset_context_nmask : NULL); |
37012946 DR |
237 | if (ret < 0) { |
238 | kmem_cache_free(policy_cache, policy); | |
239 | return ERR_PTR(ret); | |
240 | } | |
1da177e4 | 241 | return policy; |
37012946 DR |
242 | } |
243 | ||
244 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) | |
245 | { | |
246 | } | |
247 | ||
248 | static void mpol_rebind_nodemask(struct mempolicy *pol, | |
249 | const nodemask_t *nodes) | |
250 | { | |
251 | nodemask_t tmp; | |
252 | ||
253 | if (pol->flags & MPOL_F_STATIC_NODES) | |
254 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
255 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
256 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
257 | else { | |
258 | nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed, | |
259 | *nodes); | |
260 | pol->w.cpuset_mems_allowed = *nodes; | |
261 | } | |
f5b087b5 | 262 | |
37012946 DR |
263 | pol->v.nodes = tmp; |
264 | if (!node_isset(current->il_next, tmp)) { | |
265 | current->il_next = next_node(current->il_next, tmp); | |
266 | if (current->il_next >= MAX_NUMNODES) | |
267 | current->il_next = first_node(tmp); | |
268 | if (current->il_next >= MAX_NUMNODES) | |
269 | current->il_next = numa_node_id(); | |
270 | } | |
271 | } | |
272 | ||
273 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
274 | const nodemask_t *nodes) | |
275 | { | |
276 | nodemask_t tmp; | |
277 | ||
37012946 DR |
278 | if (pol->flags & MPOL_F_STATIC_NODES) { |
279 | int node = first_node(pol->w.user_nodemask); | |
280 | ||
281 | if (node_isset(node, *nodes)) | |
282 | pol->v.preferred_node = node; | |
283 | else | |
284 | pol->v.preferred_node = -1; | |
37012946 DR |
285 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
286 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
287 | pol->v.preferred_node = first_node(tmp); | |
3e1f0645 | 288 | } else if (pol->v.preferred_node != -1) { |
37012946 DR |
289 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
290 | pol->w.cpuset_mems_allowed, | |
291 | *nodes); | |
292 | pol->w.cpuset_mems_allowed = *nodes; | |
293 | } | |
1da177e4 LT |
294 | } |
295 | ||
1d0d2680 DR |
296 | /* Migrate a policy to a different set of nodes */ |
297 | static void mpol_rebind_policy(struct mempolicy *pol, | |
298 | const nodemask_t *newmask) | |
299 | { | |
1d0d2680 DR |
300 | if (!pol) |
301 | return; | |
1d0d2680 DR |
302 | if (!mpol_store_user_nodemask(pol) && |
303 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) | |
304 | return; | |
45c4745a | 305 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
306 | } |
307 | ||
308 | /* | |
309 | * Wrapper for mpol_rebind_policy() that just requires task | |
310 | * pointer, and updates task mempolicy. | |
311 | */ | |
312 | ||
313 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
314 | { | |
315 | mpol_rebind_policy(tsk->mempolicy, new); | |
316 | } | |
317 | ||
318 | /* | |
319 | * Rebind each vma in mm to new nodemask. | |
320 | * | |
321 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
322 | */ | |
323 | ||
324 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
325 | { | |
326 | struct vm_area_struct *vma; | |
327 | ||
328 | down_write(&mm->mmap_sem); | |
329 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
330 | mpol_rebind_policy(vma->vm_policy, new); | |
331 | up_write(&mm->mmap_sem); | |
332 | } | |
333 | ||
37012946 DR |
334 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
335 | [MPOL_DEFAULT] = { | |
336 | .rebind = mpol_rebind_default, | |
337 | }, | |
338 | [MPOL_INTERLEAVE] = { | |
339 | .create = mpol_new_interleave, | |
340 | .rebind = mpol_rebind_nodemask, | |
341 | }, | |
342 | [MPOL_PREFERRED] = { | |
343 | .create = mpol_new_preferred, | |
344 | .rebind = mpol_rebind_preferred, | |
345 | }, | |
346 | [MPOL_BIND] = { | |
347 | .create = mpol_new_bind, | |
348 | .rebind = mpol_rebind_nodemask, | |
349 | }, | |
350 | }; | |
351 | ||
397874df | 352 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
353 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
354 | unsigned long flags); | |
1a75a6c8 | 355 | |
38e35860 | 356 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 357 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
358 | unsigned long addr, unsigned long end, |
359 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 360 | void *private) |
1da177e4 | 361 | { |
91612e0d HD |
362 | pte_t *orig_pte; |
363 | pte_t *pte; | |
705e87c0 | 364 | spinlock_t *ptl; |
941150a3 | 365 | |
705e87c0 | 366 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 367 | do { |
6aab341e | 368 | struct page *page; |
25ba77c1 | 369 | int nid; |
91612e0d HD |
370 | |
371 | if (!pte_present(*pte)) | |
1da177e4 | 372 | continue; |
6aab341e LT |
373 | page = vm_normal_page(vma, addr, *pte); |
374 | if (!page) | |
1da177e4 | 375 | continue; |
053837fc NP |
376 | /* |
377 | * The check for PageReserved here is important to avoid | |
378 | * handling zero pages and other pages that may have been | |
379 | * marked special by the system. | |
380 | * | |
381 | * If the PageReserved would not be checked here then f.e. | |
382 | * the location of the zero page could have an influence | |
383 | * on MPOL_MF_STRICT, zero pages would be counted for | |
384 | * the per node stats, and there would be useless attempts | |
385 | * to put zero pages on the migration list. | |
386 | */ | |
f4598c8b CL |
387 | if (PageReserved(page)) |
388 | continue; | |
6aab341e | 389 | nid = page_to_nid(page); |
38e35860 CL |
390 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
391 | continue; | |
392 | ||
1a75a6c8 | 393 | if (flags & MPOL_MF_STATS) |
397874df | 394 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 395 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 396 | migrate_page_add(page, private, flags); |
38e35860 CL |
397 | else |
398 | break; | |
91612e0d | 399 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 400 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
401 | return addr != end; |
402 | } | |
403 | ||
b5810039 | 404 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
405 | unsigned long addr, unsigned long end, |
406 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 407 | void *private) |
91612e0d HD |
408 | { |
409 | pmd_t *pmd; | |
410 | unsigned long next; | |
411 | ||
412 | pmd = pmd_offset(pud, addr); | |
413 | do { | |
414 | next = pmd_addr_end(addr, end); | |
415 | if (pmd_none_or_clear_bad(pmd)) | |
416 | continue; | |
dc9aa5b9 | 417 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 418 | flags, private)) |
91612e0d HD |
419 | return -EIO; |
420 | } while (pmd++, addr = next, addr != end); | |
421 | return 0; | |
422 | } | |
423 | ||
b5810039 | 424 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
425 | unsigned long addr, unsigned long end, |
426 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 427 | void *private) |
91612e0d HD |
428 | { |
429 | pud_t *pud; | |
430 | unsigned long next; | |
431 | ||
432 | pud = pud_offset(pgd, addr); | |
433 | do { | |
434 | next = pud_addr_end(addr, end); | |
435 | if (pud_none_or_clear_bad(pud)) | |
436 | continue; | |
dc9aa5b9 | 437 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 438 | flags, private)) |
91612e0d HD |
439 | return -EIO; |
440 | } while (pud++, addr = next, addr != end); | |
441 | return 0; | |
442 | } | |
443 | ||
b5810039 | 444 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
445 | unsigned long addr, unsigned long end, |
446 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 447 | void *private) |
91612e0d HD |
448 | { |
449 | pgd_t *pgd; | |
450 | unsigned long next; | |
451 | ||
b5810039 | 452 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
453 | do { |
454 | next = pgd_addr_end(addr, end); | |
455 | if (pgd_none_or_clear_bad(pgd)) | |
456 | continue; | |
dc9aa5b9 | 457 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 458 | flags, private)) |
91612e0d HD |
459 | return -EIO; |
460 | } while (pgd++, addr = next, addr != end); | |
461 | return 0; | |
1da177e4 LT |
462 | } |
463 | ||
dc9aa5b9 CL |
464 | /* |
465 | * Check if all pages in a range are on a set of nodes. | |
466 | * If pagelist != NULL then isolate pages from the LRU and | |
467 | * put them on the pagelist. | |
468 | */ | |
1da177e4 LT |
469 | static struct vm_area_struct * |
470 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 471 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
472 | { |
473 | int err; | |
474 | struct vm_area_struct *first, *vma, *prev; | |
475 | ||
90036ee5 | 476 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
90036ee5 | 477 | |
b20a3503 CL |
478 | err = migrate_prep(); |
479 | if (err) | |
480 | return ERR_PTR(err); | |
90036ee5 | 481 | } |
053837fc | 482 | |
1da177e4 LT |
483 | first = find_vma(mm, start); |
484 | if (!first) | |
485 | return ERR_PTR(-EFAULT); | |
486 | prev = NULL; | |
487 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
488 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
489 | if (!vma->vm_next && vma->vm_end < end) | |
490 | return ERR_PTR(-EFAULT); | |
491 | if (prev && prev->vm_end < vma->vm_start) | |
492 | return ERR_PTR(-EFAULT); | |
493 | } | |
494 | if (!is_vm_hugetlb_page(vma) && | |
495 | ((flags & MPOL_MF_STRICT) || | |
496 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
497 | vma_migratable(vma)))) { | |
5b952b3c | 498 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 499 | |
5b952b3c AK |
500 | if (endvma > end) |
501 | endvma = end; | |
502 | if (vma->vm_start > start) | |
503 | start = vma->vm_start; | |
dc9aa5b9 | 504 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 505 | flags, private); |
1da177e4 LT |
506 | if (err) { |
507 | first = ERR_PTR(err); | |
508 | break; | |
509 | } | |
510 | } | |
511 | prev = vma; | |
512 | } | |
513 | return first; | |
514 | } | |
515 | ||
516 | /* Apply policy to a single VMA */ | |
517 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
518 | { | |
519 | int err = 0; | |
520 | struct mempolicy *old = vma->vm_policy; | |
521 | ||
140d5a49 | 522 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
523 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
524 | vma->vm_ops, vma->vm_file, | |
525 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
526 | ||
527 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
528 | err = vma->vm_ops->set_policy(vma, new); | |
529 | if (!err) { | |
530 | mpol_get(new); | |
531 | vma->vm_policy = new; | |
f0be3d32 | 532 | mpol_put(old); |
1da177e4 LT |
533 | } |
534 | return err; | |
535 | } | |
536 | ||
537 | /* Step 2: apply policy to a range and do splits. */ | |
538 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
539 | unsigned long end, struct mempolicy *new) | |
540 | { | |
541 | struct vm_area_struct *next; | |
542 | int err; | |
543 | ||
544 | err = 0; | |
545 | for (; vma && vma->vm_start < end; vma = next) { | |
546 | next = vma->vm_next; | |
547 | if (vma->vm_start < start) | |
548 | err = split_vma(vma->vm_mm, vma, start, 1); | |
549 | if (!err && vma->vm_end > end) | |
550 | err = split_vma(vma->vm_mm, vma, end, 0); | |
551 | if (!err) | |
552 | err = policy_vma(vma, new); | |
553 | if (err) | |
554 | break; | |
555 | } | |
556 | return err; | |
557 | } | |
558 | ||
c61afb18 PJ |
559 | /* |
560 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
561 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
562 | * with other PF_* flag bits) on memory allocation hot code paths. | |
563 | * | |
564 | * If called from outside this file, the task 'p' should -only- be | |
565 | * a newly forked child not yet visible on the task list, because | |
566 | * manipulating the task flags of a visible task is not safe. | |
567 | * | |
568 | * The above limitation is why this routine has the funny name | |
569 | * mpol_fix_fork_child_flag(). | |
570 | * | |
571 | * It is also safe to call this with a task pointer of current, | |
572 | * which the static wrapper mpol_set_task_struct_flag() does, | |
573 | * for use within this file. | |
574 | */ | |
575 | ||
576 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
577 | { | |
578 | if (p->mempolicy) | |
579 | p->flags |= PF_MEMPOLICY; | |
580 | else | |
581 | p->flags &= ~PF_MEMPOLICY; | |
582 | } | |
583 | ||
584 | static void mpol_set_task_struct_flag(void) | |
585 | { | |
586 | mpol_fix_fork_child_flag(current); | |
587 | } | |
588 | ||
1da177e4 | 589 | /* Set the process memory policy */ |
028fec41 DR |
590 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
591 | nodemask_t *nodes) | |
1da177e4 | 592 | { |
1da177e4 | 593 | struct mempolicy *new; |
f4e53d91 | 594 | struct mm_struct *mm = current->mm; |
1da177e4 | 595 | |
028fec41 | 596 | new = mpol_new(mode, flags, nodes); |
1da177e4 LT |
597 | if (IS_ERR(new)) |
598 | return PTR_ERR(new); | |
f4e53d91 LS |
599 | |
600 | /* | |
601 | * prevent changing our mempolicy while show_numa_maps() | |
602 | * is using it. | |
603 | * Note: do_set_mempolicy() can be called at init time | |
604 | * with no 'mm'. | |
605 | */ | |
606 | if (mm) | |
607 | down_write(&mm->mmap_sem); | |
f0be3d32 | 608 | mpol_put(current->mempolicy); |
1da177e4 | 609 | current->mempolicy = new; |
c61afb18 | 610 | mpol_set_task_struct_flag(); |
45c4745a | 611 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 612 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 613 | current->il_next = first_node(new->v.nodes); |
f4e53d91 LS |
614 | if (mm) |
615 | up_write(&mm->mmap_sem); | |
616 | ||
1da177e4 LT |
617 | return 0; |
618 | } | |
619 | ||
620 | /* Fill a zone bitmap for a policy */ | |
dfcd3c0d | 621 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
1da177e4 | 622 | { |
dfcd3c0d | 623 | nodes_clear(*nodes); |
45c4745a | 624 | switch (p->mode) { |
1da177e4 LT |
625 | case MPOL_DEFAULT: |
626 | break; | |
19770b32 MG |
627 | case MPOL_BIND: |
628 | /* Fall through */ | |
1da177e4 | 629 | case MPOL_INTERLEAVE: |
dfcd3c0d | 630 | *nodes = p->v.nodes; |
1da177e4 LT |
631 | break; |
632 | case MPOL_PREFERRED: | |
56bbd65d | 633 | /* or use current node instead of memory_map? */ |
1da177e4 | 634 | if (p->v.preferred_node < 0) |
56bbd65d | 635 | *nodes = node_states[N_HIGH_MEMORY]; |
1da177e4 | 636 | else |
dfcd3c0d | 637 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
638 | break; |
639 | default: | |
640 | BUG(); | |
641 | } | |
642 | } | |
643 | ||
644 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
645 | { | |
646 | struct page *p; | |
647 | int err; | |
648 | ||
649 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
650 | if (err >= 0) { | |
651 | err = page_to_nid(p); | |
652 | put_page(p); | |
653 | } | |
654 | return err; | |
655 | } | |
656 | ||
1da177e4 | 657 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
658 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
659 | unsigned long addr, unsigned long flags) | |
1da177e4 | 660 | { |
8bccd85f | 661 | int err; |
1da177e4 LT |
662 | struct mm_struct *mm = current->mm; |
663 | struct vm_area_struct *vma = NULL; | |
664 | struct mempolicy *pol = current->mempolicy; | |
665 | ||
cf2a473c | 666 | cpuset_update_task_memory_state(); |
754af6f5 LS |
667 | if (flags & |
668 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 669 | return -EINVAL; |
754af6f5 LS |
670 | |
671 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
672 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
673 | return -EINVAL; | |
674 | *policy = 0; /* just so it's initialized */ | |
675 | *nmask = cpuset_current_mems_allowed; | |
676 | return 0; | |
677 | } | |
678 | ||
1da177e4 LT |
679 | if (flags & MPOL_F_ADDR) { |
680 | down_read(&mm->mmap_sem); | |
681 | vma = find_vma_intersection(mm, addr, addr+1); | |
682 | if (!vma) { | |
683 | up_read(&mm->mmap_sem); | |
684 | return -EFAULT; | |
685 | } | |
686 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
687 | pol = vma->vm_ops->get_policy(vma, addr); | |
688 | else | |
689 | pol = vma->vm_policy; | |
690 | } else if (addr) | |
691 | return -EINVAL; | |
692 | ||
693 | if (!pol) | |
694 | pol = &default_policy; | |
695 | ||
696 | if (flags & MPOL_F_NODE) { | |
697 | if (flags & MPOL_F_ADDR) { | |
698 | err = lookup_node(mm, addr); | |
699 | if (err < 0) | |
700 | goto out; | |
8bccd85f | 701 | *policy = err; |
1da177e4 | 702 | } else if (pol == current->mempolicy && |
45c4745a | 703 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 704 | *policy = current->il_next; |
1da177e4 LT |
705 | } else { |
706 | err = -EINVAL; | |
707 | goto out; | |
708 | } | |
709 | } else | |
45c4745a | 710 | *policy = pol->mode | pol->flags; |
1da177e4 LT |
711 | |
712 | if (vma) { | |
713 | up_read(¤t->mm->mmap_sem); | |
714 | vma = NULL; | |
715 | } | |
716 | ||
1da177e4 | 717 | err = 0; |
8bccd85f CL |
718 | if (nmask) |
719 | get_zonemask(pol, nmask); | |
1da177e4 LT |
720 | |
721 | out: | |
722 | if (vma) | |
723 | up_read(¤t->mm->mmap_sem); | |
724 | return err; | |
725 | } | |
726 | ||
b20a3503 | 727 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
728 | /* |
729 | * page migration | |
730 | */ | |
fc301289 CL |
731 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
732 | unsigned long flags) | |
6ce3c4c0 CL |
733 | { |
734 | /* | |
fc301289 | 735 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 736 | */ |
b20a3503 CL |
737 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
738 | isolate_lru_page(page, pagelist); | |
7e2ab150 | 739 | } |
6ce3c4c0 | 740 | |
742755a1 | 741 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 742 | { |
769848c0 | 743 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
744 | } |
745 | ||
7e2ab150 CL |
746 | /* |
747 | * Migrate pages from one node to a target node. | |
748 | * Returns error or the number of pages not migrated. | |
749 | */ | |
dbcb0f19 AB |
750 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
751 | int flags) | |
7e2ab150 CL |
752 | { |
753 | nodemask_t nmask; | |
754 | LIST_HEAD(pagelist); | |
755 | int err = 0; | |
756 | ||
757 | nodes_clear(nmask); | |
758 | node_set(source, nmask); | |
6ce3c4c0 | 759 | |
7e2ab150 CL |
760 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
761 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
762 | ||
aaa994b3 | 763 | if (!list_empty(&pagelist)) |
95a402c3 CL |
764 | err = migrate_pages(&pagelist, new_node_page, dest); |
765 | ||
7e2ab150 | 766 | return err; |
6ce3c4c0 CL |
767 | } |
768 | ||
39743889 | 769 | /* |
7e2ab150 CL |
770 | * Move pages between the two nodesets so as to preserve the physical |
771 | * layout as much as possible. | |
39743889 CL |
772 | * |
773 | * Returns the number of page that could not be moved. | |
774 | */ | |
775 | int do_migrate_pages(struct mm_struct *mm, | |
776 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
777 | { | |
778 | LIST_HEAD(pagelist); | |
7e2ab150 CL |
779 | int busy = 0; |
780 | int err = 0; | |
781 | nodemask_t tmp; | |
39743889 | 782 | |
7e2ab150 | 783 | down_read(&mm->mmap_sem); |
39743889 | 784 | |
7b2259b3 CL |
785 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
786 | if (err) | |
787 | goto out; | |
788 | ||
7e2ab150 CL |
789 | /* |
790 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
791 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
792 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
793 | * The pair of nodemasks 'to' and 'from' define the map. | |
794 | * | |
795 | * If no pair of bits is found that way, fallback to picking some | |
796 | * pair of 'source' and 'dest' bits that are not the same. If the | |
797 | * 'source' and 'dest' bits are the same, this represents a node | |
798 | * that will be migrating to itself, so no pages need move. | |
799 | * | |
800 | * If no bits are left in 'tmp', or if all remaining bits left | |
801 | * in 'tmp' correspond to the same bit in 'to', return false | |
802 | * (nothing left to migrate). | |
803 | * | |
804 | * This lets us pick a pair of nodes to migrate between, such that | |
805 | * if possible the dest node is not already occupied by some other | |
806 | * source node, minimizing the risk of overloading the memory on a | |
807 | * node that would happen if we migrated incoming memory to a node | |
808 | * before migrating outgoing memory source that same node. | |
809 | * | |
810 | * A single scan of tmp is sufficient. As we go, we remember the | |
811 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
812 | * that not only moved, but what's better, moved to an empty slot | |
813 | * (d is not set in tmp), then we break out then, with that pair. | |
814 | * Otherwise when we finish scannng from_tmp, we at least have the | |
815 | * most recent <s, d> pair that moved. If we get all the way through | |
816 | * the scan of tmp without finding any node that moved, much less | |
817 | * moved to an empty node, then there is nothing left worth migrating. | |
818 | */ | |
d4984711 | 819 | |
7e2ab150 CL |
820 | tmp = *from_nodes; |
821 | while (!nodes_empty(tmp)) { | |
822 | int s,d; | |
823 | int source = -1; | |
824 | int dest = 0; | |
825 | ||
826 | for_each_node_mask(s, tmp) { | |
827 | d = node_remap(s, *from_nodes, *to_nodes); | |
828 | if (s == d) | |
829 | continue; | |
830 | ||
831 | source = s; /* Node moved. Memorize */ | |
832 | dest = d; | |
833 | ||
834 | /* dest not in remaining from nodes? */ | |
835 | if (!node_isset(dest, tmp)) | |
836 | break; | |
837 | } | |
838 | if (source == -1) | |
839 | break; | |
840 | ||
841 | node_clear(source, tmp); | |
842 | err = migrate_to_node(mm, source, dest, flags); | |
843 | if (err > 0) | |
844 | busy += err; | |
845 | if (err < 0) | |
846 | break; | |
39743889 | 847 | } |
7b2259b3 | 848 | out: |
39743889 | 849 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
850 | if (err < 0) |
851 | return err; | |
852 | return busy; | |
b20a3503 CL |
853 | |
854 | } | |
855 | ||
3ad33b24 LS |
856 | /* |
857 | * Allocate a new page for page migration based on vma policy. | |
858 | * Start assuming that page is mapped by vma pointed to by @private. | |
859 | * Search forward from there, if not. N.B., this assumes that the | |
860 | * list of pages handed to migrate_pages()--which is how we get here-- | |
861 | * is in virtual address order. | |
862 | */ | |
742755a1 | 863 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
864 | { |
865 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 866 | unsigned long uninitialized_var(address); |
95a402c3 | 867 | |
3ad33b24 LS |
868 | while (vma) { |
869 | address = page_address_in_vma(page, vma); | |
870 | if (address != -EFAULT) | |
871 | break; | |
872 | vma = vma->vm_next; | |
873 | } | |
874 | ||
875 | /* | |
876 | * if !vma, alloc_page_vma() will use task or system default policy | |
877 | */ | |
878 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 879 | } |
b20a3503 CL |
880 | #else |
881 | ||
882 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
883 | unsigned long flags) | |
884 | { | |
39743889 CL |
885 | } |
886 | ||
b20a3503 CL |
887 | int do_migrate_pages(struct mm_struct *mm, |
888 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
889 | { | |
890 | return -ENOSYS; | |
891 | } | |
95a402c3 | 892 | |
69939749 | 893 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
894 | { |
895 | return NULL; | |
896 | } | |
b20a3503 CL |
897 | #endif |
898 | ||
dbcb0f19 | 899 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
900 | unsigned short mode, unsigned short mode_flags, |
901 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
902 | { |
903 | struct vm_area_struct *vma; | |
904 | struct mm_struct *mm = current->mm; | |
905 | struct mempolicy *new; | |
906 | unsigned long end; | |
907 | int err; | |
908 | LIST_HEAD(pagelist); | |
909 | ||
a3b51e01 DR |
910 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
911 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 912 | return -EINVAL; |
74c00241 | 913 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
914 | return -EPERM; |
915 | ||
916 | if (start & ~PAGE_MASK) | |
917 | return -EINVAL; | |
918 | ||
919 | if (mode == MPOL_DEFAULT) | |
920 | flags &= ~MPOL_MF_STRICT; | |
921 | ||
922 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
923 | end = start + len; | |
924 | ||
925 | if (end < start) | |
926 | return -EINVAL; | |
927 | if (end == start) | |
928 | return 0; | |
929 | ||
028fec41 | 930 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
931 | if (IS_ERR(new)) |
932 | return PTR_ERR(new); | |
933 | ||
934 | /* | |
935 | * If we are using the default policy then operation | |
936 | * on discontinuous address spaces is okay after all | |
937 | */ | |
938 | if (!new) | |
939 | flags |= MPOL_MF_DISCONTIG_OK; | |
940 | ||
028fec41 DR |
941 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
942 | start, start + len, mode, mode_flags, | |
943 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 CL |
944 | |
945 | down_write(&mm->mmap_sem); | |
946 | vma = check_range(mm, start, end, nmask, | |
947 | flags | MPOL_MF_INVERT, &pagelist); | |
948 | ||
949 | err = PTR_ERR(vma); | |
950 | if (!IS_ERR(vma)) { | |
951 | int nr_failed = 0; | |
952 | ||
953 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 954 | |
6ce3c4c0 | 955 | if (!list_empty(&pagelist)) |
95a402c3 CL |
956 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
957 | (unsigned long)vma); | |
6ce3c4c0 CL |
958 | |
959 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
960 | err = -EIO; | |
961 | } | |
b20a3503 | 962 | |
6ce3c4c0 | 963 | up_write(&mm->mmap_sem); |
f0be3d32 | 964 | mpol_put(new); |
6ce3c4c0 CL |
965 | return err; |
966 | } | |
967 | ||
8bccd85f CL |
968 | /* |
969 | * User space interface with variable sized bitmaps for nodelists. | |
970 | */ | |
971 | ||
972 | /* Copy a node mask from user space. */ | |
39743889 | 973 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
974 | unsigned long maxnode) |
975 | { | |
976 | unsigned long k; | |
977 | unsigned long nlongs; | |
978 | unsigned long endmask; | |
979 | ||
980 | --maxnode; | |
981 | nodes_clear(*nodes); | |
982 | if (maxnode == 0 || !nmask) | |
983 | return 0; | |
a9c930ba | 984 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 985 | return -EINVAL; |
8bccd85f CL |
986 | |
987 | nlongs = BITS_TO_LONGS(maxnode); | |
988 | if ((maxnode % BITS_PER_LONG) == 0) | |
989 | endmask = ~0UL; | |
990 | else | |
991 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
992 | ||
993 | /* When the user specified more nodes than supported just check | |
994 | if the non supported part is all zero. */ | |
995 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
996 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
997 | return -EINVAL; | |
998 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
999 | unsigned long t; | |
1000 | if (get_user(t, nmask + k)) | |
1001 | return -EFAULT; | |
1002 | if (k == nlongs - 1) { | |
1003 | if (t & endmask) | |
1004 | return -EINVAL; | |
1005 | } else if (t) | |
1006 | return -EINVAL; | |
1007 | } | |
1008 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1009 | endmask = ~0UL; | |
1010 | } | |
1011 | ||
1012 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1013 | return -EFAULT; | |
1014 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1015 | return 0; | |
1016 | } | |
1017 | ||
1018 | /* Copy a kernel node mask to user space */ | |
1019 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1020 | nodemask_t *nodes) | |
1021 | { | |
1022 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1023 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1024 | ||
1025 | if (copy > nbytes) { | |
1026 | if (copy > PAGE_SIZE) | |
1027 | return -EINVAL; | |
1028 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1029 | return -EFAULT; | |
1030 | copy = nbytes; | |
1031 | } | |
1032 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1033 | } | |
1034 | ||
1035 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
1036 | unsigned long mode, | |
1037 | unsigned long __user *nmask, unsigned long maxnode, | |
1038 | unsigned flags) | |
1039 | { | |
1040 | nodemask_t nodes; | |
1041 | int err; | |
028fec41 | 1042 | unsigned short mode_flags; |
8bccd85f | 1043 | |
028fec41 DR |
1044 | mode_flags = mode & MPOL_MODE_FLAGS; |
1045 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1046 | if (mode >= MPOL_MAX) |
1047 | return -EINVAL; | |
4c50bc01 DR |
1048 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1049 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1050 | return -EINVAL; | |
8bccd85f CL |
1051 | err = get_nodes(&nodes, nmask, maxnode); |
1052 | if (err) | |
1053 | return err; | |
028fec41 | 1054 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1055 | } |
1056 | ||
1057 | /* Set the process memory policy */ | |
1058 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
1059 | unsigned long maxnode) | |
1060 | { | |
1061 | int err; | |
1062 | nodemask_t nodes; | |
028fec41 | 1063 | unsigned short flags; |
8bccd85f | 1064 | |
028fec41 DR |
1065 | flags = mode & MPOL_MODE_FLAGS; |
1066 | mode &= ~MPOL_MODE_FLAGS; | |
1067 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1068 | return -EINVAL; |
4c50bc01 DR |
1069 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1070 | return -EINVAL; | |
8bccd85f CL |
1071 | err = get_nodes(&nodes, nmask, maxnode); |
1072 | if (err) | |
1073 | return err; | |
028fec41 | 1074 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1075 | } |
1076 | ||
39743889 CL |
1077 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
1078 | const unsigned long __user *old_nodes, | |
1079 | const unsigned long __user *new_nodes) | |
1080 | { | |
1081 | struct mm_struct *mm; | |
1082 | struct task_struct *task; | |
1083 | nodemask_t old; | |
1084 | nodemask_t new; | |
1085 | nodemask_t task_nodes; | |
1086 | int err; | |
1087 | ||
1088 | err = get_nodes(&old, old_nodes, maxnode); | |
1089 | if (err) | |
1090 | return err; | |
1091 | ||
1092 | err = get_nodes(&new, new_nodes, maxnode); | |
1093 | if (err) | |
1094 | return err; | |
1095 | ||
1096 | /* Find the mm_struct */ | |
1097 | read_lock(&tasklist_lock); | |
228ebcbe | 1098 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 CL |
1099 | if (!task) { |
1100 | read_unlock(&tasklist_lock); | |
1101 | return -ESRCH; | |
1102 | } | |
1103 | mm = get_task_mm(task); | |
1104 | read_unlock(&tasklist_lock); | |
1105 | ||
1106 | if (!mm) | |
1107 | return -EINVAL; | |
1108 | ||
1109 | /* | |
1110 | * Check if this process has the right to modify the specified | |
1111 | * process. The right exists if the process has administrative | |
7f927fcc | 1112 | * capabilities, superuser privileges or the same |
39743889 CL |
1113 | * userid as the target process. |
1114 | */ | |
1115 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
1116 | (current->uid != task->suid) && (current->uid != task->uid) && | |
74c00241 | 1117 | !capable(CAP_SYS_NICE)) { |
39743889 CL |
1118 | err = -EPERM; |
1119 | goto out; | |
1120 | } | |
1121 | ||
1122 | task_nodes = cpuset_mems_allowed(task); | |
1123 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 1124 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
1125 | err = -EPERM; |
1126 | goto out; | |
1127 | } | |
1128 | ||
37b07e41 | 1129 | if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
1130 | err = -EINVAL; |
1131 | goto out; | |
1132 | } | |
1133 | ||
86c3a764 DQ |
1134 | err = security_task_movememory(task); |
1135 | if (err) | |
1136 | goto out; | |
1137 | ||
511030bc | 1138 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 1139 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
1140 | out: |
1141 | mmput(mm); | |
1142 | return err; | |
1143 | } | |
1144 | ||
1145 | ||
8bccd85f CL |
1146 | /* Retrieve NUMA policy */ |
1147 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
1148 | unsigned long __user *nmask, | |
1149 | unsigned long maxnode, | |
1150 | unsigned long addr, unsigned long flags) | |
1151 | { | |
dbcb0f19 AB |
1152 | int err; |
1153 | int uninitialized_var(pval); | |
8bccd85f CL |
1154 | nodemask_t nodes; |
1155 | ||
1156 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1157 | return -EINVAL; | |
1158 | ||
1159 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1160 | ||
1161 | if (err) | |
1162 | return err; | |
1163 | ||
1164 | if (policy && put_user(pval, policy)) | |
1165 | return -EFAULT; | |
1166 | ||
1167 | if (nmask) | |
1168 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1169 | ||
1170 | return err; | |
1171 | } | |
1172 | ||
1da177e4 LT |
1173 | #ifdef CONFIG_COMPAT |
1174 | ||
1175 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1176 | compat_ulong_t __user *nmask, | |
1177 | compat_ulong_t maxnode, | |
1178 | compat_ulong_t addr, compat_ulong_t flags) | |
1179 | { | |
1180 | long err; | |
1181 | unsigned long __user *nm = NULL; | |
1182 | unsigned long nr_bits, alloc_size; | |
1183 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1184 | ||
1185 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1186 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1187 | ||
1188 | if (nmask) | |
1189 | nm = compat_alloc_user_space(alloc_size); | |
1190 | ||
1191 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1192 | ||
1193 | if (!err && nmask) { | |
1194 | err = copy_from_user(bm, nm, alloc_size); | |
1195 | /* ensure entire bitmap is zeroed */ | |
1196 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1197 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1198 | } | |
1199 | ||
1200 | return err; | |
1201 | } | |
1202 | ||
1203 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1204 | compat_ulong_t maxnode) | |
1205 | { | |
1206 | long err = 0; | |
1207 | unsigned long __user *nm = NULL; | |
1208 | unsigned long nr_bits, alloc_size; | |
1209 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1210 | ||
1211 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1212 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1213 | ||
1214 | if (nmask) { | |
1215 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1216 | nm = compat_alloc_user_space(alloc_size); | |
1217 | err |= copy_to_user(nm, bm, alloc_size); | |
1218 | } | |
1219 | ||
1220 | if (err) | |
1221 | return -EFAULT; | |
1222 | ||
1223 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1224 | } | |
1225 | ||
1226 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1227 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1228 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1229 | { | |
1230 | long err = 0; | |
1231 | unsigned long __user *nm = NULL; | |
1232 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1233 | nodemask_t bm; |
1da177e4 LT |
1234 | |
1235 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1236 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1237 | ||
1238 | if (nmask) { | |
dfcd3c0d | 1239 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1240 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1241 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1242 | } |
1243 | ||
1244 | if (err) | |
1245 | return -EFAULT; | |
1246 | ||
1247 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1248 | } | |
1249 | ||
1250 | #endif | |
1251 | ||
480eccf9 LS |
1252 | /* |
1253 | * get_vma_policy(@task, @vma, @addr) | |
1254 | * @task - task for fallback if vma policy == default | |
1255 | * @vma - virtual memory area whose policy is sought | |
1256 | * @addr - address in @vma for shared policy lookup | |
1257 | * | |
1258 | * Returns effective policy for a VMA at specified address. | |
1259 | * Falls back to @task or system default policy, as necessary. | |
1260 | * Returned policy has extra reference count if shared, vma, | |
1261 | * or some other task's policy [show_numa_maps() can pass | |
1262 | * @task != current]. It is the caller's responsibility to | |
1263 | * free the reference in these cases. | |
1264 | */ | |
ae4d8c16 | 1265 | static struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1266 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1267 | { |
6e21c8f1 | 1268 | struct mempolicy *pol = task->mempolicy; |
480eccf9 | 1269 | int shared_pol = 0; |
1da177e4 LT |
1270 | |
1271 | if (vma) { | |
480eccf9 | 1272 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1273 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1274 | addr); | |
1275 | if (vpol) | |
1276 | pol = vpol; | |
480eccf9 LS |
1277 | shared_pol = 1; /* if pol non-NULL, add ref below */ |
1278 | } else if (vma->vm_policy && | |
45c4745a | 1279 | vma->vm_policy->mode != MPOL_DEFAULT) |
1da177e4 LT |
1280 | pol = vma->vm_policy; |
1281 | } | |
1282 | if (!pol) | |
1283 | pol = &default_policy; | |
480eccf9 LS |
1284 | else if (!shared_pol && pol != current->mempolicy) |
1285 | mpol_get(pol); /* vma or other task's policy */ | |
1da177e4 LT |
1286 | return pol; |
1287 | } | |
1288 | ||
19770b32 MG |
1289 | /* Return a nodemask representing a mempolicy */ |
1290 | static nodemask_t *nodemask_policy(gfp_t gfp, struct mempolicy *policy) | |
1291 | { | |
1292 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1293 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1294 | gfp_zone(gfp) >= policy_zone && |
1295 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1296 | return &policy->v.nodes; | |
1297 | ||
1298 | return NULL; | |
1299 | } | |
1300 | ||
1da177e4 | 1301 | /* Return a zonelist representing a mempolicy */ |
dd0fc66f | 1302 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
1da177e4 LT |
1303 | { |
1304 | int nd; | |
1305 | ||
45c4745a | 1306 | switch (policy->mode) { |
1da177e4 LT |
1307 | case MPOL_PREFERRED: |
1308 | nd = policy->v.preferred_node; | |
1309 | if (nd < 0) | |
1310 | nd = numa_node_id(); | |
1311 | break; | |
1312 | case MPOL_BIND: | |
19770b32 MG |
1313 | /* |
1314 | * Normally, MPOL_BIND allocations node-local are node-local | |
1315 | * within the allowed nodemask. However, if __GFP_THISNODE is | |
1316 | * set and the current node is part of the mask, we use the | |
1317 | * the zonelist for the first node in the mask instead. | |
1318 | */ | |
1319 | nd = numa_node_id(); | |
1320 | if (unlikely(gfp & __GFP_THISNODE) && | |
1321 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1322 | nd = first_node(policy->v.nodes); | |
1323 | break; | |
1da177e4 LT |
1324 | case MPOL_INTERLEAVE: /* should not happen */ |
1325 | case MPOL_DEFAULT: | |
1326 | nd = numa_node_id(); | |
1327 | break; | |
1328 | default: | |
1329 | nd = 0; | |
1330 | BUG(); | |
1331 | } | |
0e88460d | 1332 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1333 | } |
1334 | ||
1335 | /* Do dynamic interleaving for a process */ | |
1336 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1337 | { | |
1338 | unsigned nid, next; | |
1339 | struct task_struct *me = current; | |
1340 | ||
1341 | nid = me->il_next; | |
dfcd3c0d | 1342 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1343 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1344 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1345 | if (next < MAX_NUMNODES) |
1346 | me->il_next = next; | |
1da177e4 LT |
1347 | return nid; |
1348 | } | |
1349 | ||
dc85da15 CL |
1350 | /* |
1351 | * Depending on the memory policy provide a node from which to allocate the | |
1352 | * next slab entry. | |
1353 | */ | |
1354 | unsigned slab_node(struct mempolicy *policy) | |
1355 | { | |
45c4745a | 1356 | unsigned short pol = policy ? policy->mode : MPOL_DEFAULT; |
765c4507 CL |
1357 | |
1358 | switch (pol) { | |
dc85da15 CL |
1359 | case MPOL_INTERLEAVE: |
1360 | return interleave_nodes(policy); | |
1361 | ||
dd1a239f | 1362 | case MPOL_BIND: { |
dc85da15 CL |
1363 | /* |
1364 | * Follow bind policy behavior and start allocation at the | |
1365 | * first node. | |
1366 | */ | |
19770b32 MG |
1367 | struct zonelist *zonelist; |
1368 | struct zone *zone; | |
1369 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1370 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1371 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1372 | &policy->v.nodes, | |
1373 | &zone); | |
1374 | return zone->node; | |
dd1a239f | 1375 | } |
dc85da15 CL |
1376 | |
1377 | case MPOL_PREFERRED: | |
1378 | if (policy->v.preferred_node >= 0) | |
1379 | return policy->v.preferred_node; | |
1380 | /* Fall through */ | |
1381 | ||
1382 | default: | |
1383 | return numa_node_id(); | |
1384 | } | |
1385 | } | |
1386 | ||
1da177e4 LT |
1387 | /* Do static interleaving for a VMA with known offset. */ |
1388 | static unsigned offset_il_node(struct mempolicy *pol, | |
1389 | struct vm_area_struct *vma, unsigned long off) | |
1390 | { | |
dfcd3c0d | 1391 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1392 | unsigned target; |
1da177e4 LT |
1393 | int c; |
1394 | int nid = -1; | |
1395 | ||
f5b087b5 DR |
1396 | if (!nnodes) |
1397 | return numa_node_id(); | |
1398 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1399 | c = 0; |
1400 | do { | |
dfcd3c0d | 1401 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1402 | c++; |
1403 | } while (c <= target); | |
1da177e4 LT |
1404 | return nid; |
1405 | } | |
1406 | ||
5da7ca86 CL |
1407 | /* Determine a node number for interleave */ |
1408 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1409 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1410 | { | |
1411 | if (vma) { | |
1412 | unsigned long off; | |
1413 | ||
3b98b087 NA |
1414 | /* |
1415 | * for small pages, there is no difference between | |
1416 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1417 | * for huge pages, since vm_pgoff is in units of small | |
1418 | * pages, we need to shift off the always 0 bits to get | |
1419 | * a useful offset. | |
1420 | */ | |
1421 | BUG_ON(shift < PAGE_SHIFT); | |
1422 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1423 | off += (addr - vma->vm_start) >> shift; |
1424 | return offset_il_node(pol, vma, off); | |
1425 | } else | |
1426 | return interleave_nodes(pol); | |
1427 | } | |
1428 | ||
00ac59ad | 1429 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1430 | /* |
1431 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1432 | * @vma = virtual memory area whose policy is sought | |
1433 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1434 | * @gfp_flags = for requested zone | |
19770b32 MG |
1435 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1436 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 LS |
1437 | * |
1438 | * Returns a zonelist suitable for a huge page allocation. | |
19770b32 MG |
1439 | * If the effective policy is 'BIND, returns pointer to local node's zonelist, |
1440 | * and a pointer to the mempolicy's @nodemask for filtering the zonelist. | |
480eccf9 | 1441 | * If it is also a policy for which get_vma_policy() returns an extra |
19770b32 | 1442 | * reference, we must hold that reference until after the allocation. |
480eccf9 | 1443 | * In that case, return policy via @mpol so hugetlb allocation can drop |
19770b32 | 1444 | * the reference. For non-'BIND referenced policies, we can/do drop the |
480eccf9 LS |
1445 | * reference here, so the caller doesn't need to know about the special case |
1446 | * for default and current task policy. | |
1447 | */ | |
396faf03 | 1448 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1449 | gfp_t gfp_flags, struct mempolicy **mpol, |
1450 | nodemask_t **nodemask) | |
5da7ca86 CL |
1451 | { |
1452 | struct mempolicy *pol = get_vma_policy(current, vma, addr); | |
480eccf9 | 1453 | struct zonelist *zl; |
5da7ca86 | 1454 | |
480eccf9 | 1455 | *mpol = NULL; /* probably no unref needed */ |
19770b32 | 1456 | *nodemask = NULL; /* assume !MPOL_BIND */ |
45c4745a | 1457 | if (pol->mode == MPOL_BIND) { |
19770b32 | 1458 | *nodemask = &pol->v.nodes; |
45c4745a | 1459 | } else if (pol->mode == MPOL_INTERLEAVE) { |
5da7ca86 CL |
1460 | unsigned nid; |
1461 | ||
1462 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); | |
69682d85 LS |
1463 | if (unlikely(pol != &default_policy && |
1464 | pol != current->mempolicy)) | |
f0be3d32 | 1465 | __mpol_put(pol); /* finished with pol */ |
0e88460d | 1466 | return node_zonelist(nid, gfp_flags); |
5da7ca86 | 1467 | } |
480eccf9 LS |
1468 | |
1469 | zl = zonelist_policy(GFP_HIGHUSER, pol); | |
1470 | if (unlikely(pol != &default_policy && pol != current->mempolicy)) { | |
45c4745a | 1471 | if (pol->mode != MPOL_BIND) |
f0be3d32 | 1472 | __mpol_put(pol); /* finished with pol */ |
480eccf9 LS |
1473 | else |
1474 | *mpol = pol; /* unref needed after allocation */ | |
1475 | } | |
1476 | return zl; | |
5da7ca86 | 1477 | } |
00ac59ad | 1478 | #endif |
5da7ca86 | 1479 | |
1da177e4 LT |
1480 | /* Allocate a page in interleaved policy. |
1481 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1482 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1483 | unsigned nid) | |
1da177e4 LT |
1484 | { |
1485 | struct zonelist *zl; | |
1486 | struct page *page; | |
1487 | ||
0e88460d | 1488 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1489 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1490 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1491 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1492 | return page; |
1493 | } | |
1494 | ||
1495 | /** | |
1496 | * alloc_page_vma - Allocate a page for a VMA. | |
1497 | * | |
1498 | * @gfp: | |
1499 | * %GFP_USER user allocation. | |
1500 | * %GFP_KERNEL kernel allocations, | |
1501 | * %GFP_HIGHMEM highmem/user allocations, | |
1502 | * %GFP_FS allocation should not call back into a file system. | |
1503 | * %GFP_ATOMIC don't sleep. | |
1504 | * | |
1505 | * @vma: Pointer to VMA or NULL if not available. | |
1506 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1507 | * | |
1508 | * This function allocates a page from the kernel page pool and applies | |
1509 | * a NUMA policy associated with the VMA or the current process. | |
1510 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1511 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1512 | * all allocations for pages that will be mapped into | |
1513 | * user space. Returns NULL when no page can be allocated. | |
1514 | * | |
1515 | * Should be called with the mm_sem of the vma hold. | |
1516 | */ | |
1517 | struct page * | |
dd0fc66f | 1518 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1519 | { |
6e21c8f1 | 1520 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1521 | struct zonelist *zl; |
1da177e4 | 1522 | |
cf2a473c | 1523 | cpuset_update_task_memory_state(); |
1da177e4 | 1524 | |
45c4745a | 1525 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1526 | unsigned nid; |
5da7ca86 CL |
1527 | |
1528 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
69682d85 LS |
1529 | if (unlikely(pol != &default_policy && |
1530 | pol != current->mempolicy)) | |
f0be3d32 | 1531 | __mpol_put(pol); /* finished with pol */ |
1da177e4 LT |
1532 | return alloc_page_interleave(gfp, 0, nid); |
1533 | } | |
480eccf9 LS |
1534 | zl = zonelist_policy(gfp, pol); |
1535 | if (pol != &default_policy && pol != current->mempolicy) { | |
1536 | /* | |
1537 | * slow path: ref counted policy -- shared or vma | |
1538 | */ | |
19770b32 MG |
1539 | struct page *page = __alloc_pages_nodemask(gfp, 0, |
1540 | zl, nodemask_policy(gfp, pol)); | |
f0be3d32 | 1541 | __mpol_put(pol); |
480eccf9 LS |
1542 | return page; |
1543 | } | |
1544 | /* | |
1545 | * fast path: default or task policy | |
1546 | */ | |
19770b32 | 1547 | return __alloc_pages_nodemask(gfp, 0, zl, nodemask_policy(gfp, pol)); |
1da177e4 LT |
1548 | } |
1549 | ||
1550 | /** | |
1551 | * alloc_pages_current - Allocate pages. | |
1552 | * | |
1553 | * @gfp: | |
1554 | * %GFP_USER user allocation, | |
1555 | * %GFP_KERNEL kernel allocation, | |
1556 | * %GFP_HIGHMEM highmem allocation, | |
1557 | * %GFP_FS don't call back into a file system. | |
1558 | * %GFP_ATOMIC don't sleep. | |
1559 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1560 | * | |
1561 | * Allocate a page from the kernel page pool. When not in | |
1562 | * interrupt context and apply the current process NUMA policy. | |
1563 | * Returns NULL when no page can be allocated. | |
1564 | * | |
cf2a473c | 1565 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1566 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1567 | * 2) allocating for current task (not interrupt). | |
1568 | */ | |
dd0fc66f | 1569 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1570 | { |
1571 | struct mempolicy *pol = current->mempolicy; | |
1572 | ||
1573 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1574 | cpuset_update_task_memory_state(); |
9b819d20 | 1575 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 1576 | pol = &default_policy; |
45c4745a | 1577 | if (pol->mode == MPOL_INTERLEAVE) |
1da177e4 | 1578 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
19770b32 MG |
1579 | return __alloc_pages_nodemask(gfp, order, |
1580 | zonelist_policy(gfp, pol), nodemask_policy(gfp, pol)); | |
1da177e4 LT |
1581 | } |
1582 | EXPORT_SYMBOL(alloc_pages_current); | |
1583 | ||
4225399a | 1584 | /* |
846a16bf | 1585 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
1586 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
1587 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1588 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1589 | * further kernel/cpuset.c update_nodemask(). | |
1590 | */ | |
4225399a | 1591 | |
846a16bf LS |
1592 | /* Slow path of a mempolicy duplicate */ |
1593 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
1594 | { |
1595 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1596 | ||
1597 | if (!new) | |
1598 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1599 | if (current_cpuset_is_being_rebound()) { |
1600 | nodemask_t mems = cpuset_mems_allowed(current); | |
1601 | mpol_rebind_policy(old, &mems); | |
1602 | } | |
1da177e4 LT |
1603 | *new = *old; |
1604 | atomic_set(&new->refcnt, 1); | |
1da177e4 LT |
1605 | return new; |
1606 | } | |
1607 | ||
f5b087b5 DR |
1608 | static int mpol_match_intent(const struct mempolicy *a, |
1609 | const struct mempolicy *b) | |
1610 | { | |
1611 | if (a->flags != b->flags) | |
1612 | return 0; | |
1613 | if (!mpol_store_user_nodemask(a)) | |
1614 | return 1; | |
1615 | return nodes_equal(a->w.user_nodemask, b->w.user_nodemask); | |
1616 | } | |
1617 | ||
1da177e4 LT |
1618 | /* Slow path of a mempolicy comparison */ |
1619 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1620 | { | |
1621 | if (!a || !b) | |
1622 | return 0; | |
45c4745a | 1623 | if (a->mode != b->mode) |
1da177e4 | 1624 | return 0; |
45c4745a | 1625 | if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b)) |
f5b087b5 | 1626 | return 0; |
45c4745a | 1627 | switch (a->mode) { |
1da177e4 LT |
1628 | case MPOL_DEFAULT: |
1629 | return 1; | |
19770b32 MG |
1630 | case MPOL_BIND: |
1631 | /* Fall through */ | |
1da177e4 | 1632 | case MPOL_INTERLEAVE: |
dfcd3c0d | 1633 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 LT |
1634 | case MPOL_PREFERRED: |
1635 | return a->v.preferred_node == b->v.preferred_node; | |
1da177e4 LT |
1636 | default: |
1637 | BUG(); | |
1638 | return 0; | |
1639 | } | |
1640 | } | |
1641 | ||
1642 | /* Slow path of a mpol destructor. */ | |
f0be3d32 | 1643 | void __mpol_put(struct mempolicy *p) |
1da177e4 LT |
1644 | { |
1645 | if (!atomic_dec_and_test(&p->refcnt)) | |
1646 | return; | |
45c4745a | 1647 | p->mode = MPOL_DEFAULT; |
1da177e4 LT |
1648 | kmem_cache_free(policy_cache, p); |
1649 | } | |
1650 | ||
1da177e4 LT |
1651 | /* |
1652 | * Shared memory backing store policy support. | |
1653 | * | |
1654 | * Remember policies even when nobody has shared memory mapped. | |
1655 | * The policies are kept in Red-Black tree linked from the inode. | |
1656 | * They are protected by the sp->lock spinlock, which should be held | |
1657 | * for any accesses to the tree. | |
1658 | */ | |
1659 | ||
1660 | /* lookup first element intersecting start-end */ | |
1661 | /* Caller holds sp->lock */ | |
1662 | static struct sp_node * | |
1663 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1664 | { | |
1665 | struct rb_node *n = sp->root.rb_node; | |
1666 | ||
1667 | while (n) { | |
1668 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1669 | ||
1670 | if (start >= p->end) | |
1671 | n = n->rb_right; | |
1672 | else if (end <= p->start) | |
1673 | n = n->rb_left; | |
1674 | else | |
1675 | break; | |
1676 | } | |
1677 | if (!n) | |
1678 | return NULL; | |
1679 | for (;;) { | |
1680 | struct sp_node *w = NULL; | |
1681 | struct rb_node *prev = rb_prev(n); | |
1682 | if (!prev) | |
1683 | break; | |
1684 | w = rb_entry(prev, struct sp_node, nd); | |
1685 | if (w->end <= start) | |
1686 | break; | |
1687 | n = prev; | |
1688 | } | |
1689 | return rb_entry(n, struct sp_node, nd); | |
1690 | } | |
1691 | ||
1692 | /* Insert a new shared policy into the list. */ | |
1693 | /* Caller holds sp->lock */ | |
1694 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1695 | { | |
1696 | struct rb_node **p = &sp->root.rb_node; | |
1697 | struct rb_node *parent = NULL; | |
1698 | struct sp_node *nd; | |
1699 | ||
1700 | while (*p) { | |
1701 | parent = *p; | |
1702 | nd = rb_entry(parent, struct sp_node, nd); | |
1703 | if (new->start < nd->start) | |
1704 | p = &(*p)->rb_left; | |
1705 | else if (new->end > nd->end) | |
1706 | p = &(*p)->rb_right; | |
1707 | else | |
1708 | BUG(); | |
1709 | } | |
1710 | rb_link_node(&new->nd, parent, p); | |
1711 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1712 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 1713 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
1714 | } |
1715 | ||
1716 | /* Find shared policy intersecting idx */ | |
1717 | struct mempolicy * | |
1718 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1719 | { | |
1720 | struct mempolicy *pol = NULL; | |
1721 | struct sp_node *sn; | |
1722 | ||
1723 | if (!sp->root.rb_node) | |
1724 | return NULL; | |
1725 | spin_lock(&sp->lock); | |
1726 | sn = sp_lookup(sp, idx, idx+1); | |
1727 | if (sn) { | |
1728 | mpol_get(sn->policy); | |
1729 | pol = sn->policy; | |
1730 | } | |
1731 | spin_unlock(&sp->lock); | |
1732 | return pol; | |
1733 | } | |
1734 | ||
1735 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1736 | { | |
140d5a49 | 1737 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 1738 | rb_erase(&n->nd, &sp->root); |
f0be3d32 | 1739 | mpol_put(n->policy); |
1da177e4 LT |
1740 | kmem_cache_free(sn_cache, n); |
1741 | } | |
1742 | ||
dbcb0f19 AB |
1743 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
1744 | struct mempolicy *pol) | |
1da177e4 LT |
1745 | { |
1746 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1747 | ||
1748 | if (!n) | |
1749 | return NULL; | |
1750 | n->start = start; | |
1751 | n->end = end; | |
1752 | mpol_get(pol); | |
aab0b102 | 1753 | pol->flags |= MPOL_F_SHARED; /* for unref */ |
1da177e4 LT |
1754 | n->policy = pol; |
1755 | return n; | |
1756 | } | |
1757 | ||
1758 | /* Replace a policy range. */ | |
1759 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1760 | unsigned long end, struct sp_node *new) | |
1761 | { | |
1762 | struct sp_node *n, *new2 = NULL; | |
1763 | ||
1764 | restart: | |
1765 | spin_lock(&sp->lock); | |
1766 | n = sp_lookup(sp, start, end); | |
1767 | /* Take care of old policies in the same range. */ | |
1768 | while (n && n->start < end) { | |
1769 | struct rb_node *next = rb_next(&n->nd); | |
1770 | if (n->start >= start) { | |
1771 | if (n->end <= end) | |
1772 | sp_delete(sp, n); | |
1773 | else | |
1774 | n->start = end; | |
1775 | } else { | |
1776 | /* Old policy spanning whole new range. */ | |
1777 | if (n->end > end) { | |
1778 | if (!new2) { | |
1779 | spin_unlock(&sp->lock); | |
1780 | new2 = sp_alloc(end, n->end, n->policy); | |
1781 | if (!new2) | |
1782 | return -ENOMEM; | |
1783 | goto restart; | |
1784 | } | |
1785 | n->end = start; | |
1786 | sp_insert(sp, new2); | |
1787 | new2 = NULL; | |
1788 | break; | |
1789 | } else | |
1790 | n->end = start; | |
1791 | } | |
1792 | if (!next) | |
1793 | break; | |
1794 | n = rb_entry(next, struct sp_node, nd); | |
1795 | } | |
1796 | if (new) | |
1797 | sp_insert(sp, new); | |
1798 | spin_unlock(&sp->lock); | |
1799 | if (new2) { | |
f0be3d32 | 1800 | mpol_put(new2->policy); |
1da177e4 LT |
1801 | kmem_cache_free(sn_cache, new2); |
1802 | } | |
1803 | return 0; | |
1804 | } | |
1805 | ||
a3b51e01 | 1806 | void mpol_shared_policy_init(struct shared_policy *info, unsigned short policy, |
028fec41 | 1807 | unsigned short flags, nodemask_t *policy_nodes) |
7339ff83 RH |
1808 | { |
1809 | info->root = RB_ROOT; | |
1810 | spin_lock_init(&info->lock); | |
1811 | ||
1812 | if (policy != MPOL_DEFAULT) { | |
1813 | struct mempolicy *newpol; | |
1814 | ||
1815 | /* Falls back to MPOL_DEFAULT on any error */ | |
028fec41 | 1816 | newpol = mpol_new(policy, flags, policy_nodes); |
7339ff83 RH |
1817 | if (!IS_ERR(newpol)) { |
1818 | /* Create pseudo-vma that contains just the policy */ | |
1819 | struct vm_area_struct pvma; | |
1820 | ||
1821 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1822 | /* Policy covers entire file */ | |
1823 | pvma.vm_end = TASK_SIZE; | |
1824 | mpol_set_shared_policy(info, &pvma, newpol); | |
f0be3d32 | 1825 | mpol_put(newpol); |
7339ff83 RH |
1826 | } |
1827 | } | |
1828 | } | |
1829 | ||
1da177e4 LT |
1830 | int mpol_set_shared_policy(struct shared_policy *info, |
1831 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1832 | { | |
1833 | int err; | |
1834 | struct sp_node *new = NULL; | |
1835 | unsigned long sz = vma_pages(vma); | |
1836 | ||
028fec41 | 1837 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 1838 | vma->vm_pgoff, |
45c4745a | 1839 | sz, npol ? npol->mode : -1, |
028fec41 | 1840 | npol ? npol->flags : -1, |
140d5a49 | 1841 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1842 | |
1843 | if (npol) { | |
1844 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1845 | if (!new) | |
1846 | return -ENOMEM; | |
1847 | } | |
1848 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1849 | if (err && new) | |
1850 | kmem_cache_free(sn_cache, new); | |
1851 | return err; | |
1852 | } | |
1853 | ||
1854 | /* Free a backing policy store on inode delete. */ | |
1855 | void mpol_free_shared_policy(struct shared_policy *p) | |
1856 | { | |
1857 | struct sp_node *n; | |
1858 | struct rb_node *next; | |
1859 | ||
1860 | if (!p->root.rb_node) | |
1861 | return; | |
1862 | spin_lock(&p->lock); | |
1863 | next = rb_first(&p->root); | |
1864 | while (next) { | |
1865 | n = rb_entry(next, struct sp_node, nd); | |
1866 | next = rb_next(&n->nd); | |
90c5029e | 1867 | rb_erase(&n->nd, &p->root); |
f0be3d32 | 1868 | mpol_put(n->policy); |
1da177e4 LT |
1869 | kmem_cache_free(sn_cache, n); |
1870 | } | |
1871 | spin_unlock(&p->lock); | |
1da177e4 LT |
1872 | } |
1873 | ||
1874 | /* assumes fs == KERNEL_DS */ | |
1875 | void __init numa_policy_init(void) | |
1876 | { | |
b71636e2 PM |
1877 | nodemask_t interleave_nodes; |
1878 | unsigned long largest = 0; | |
1879 | int nid, prefer = 0; | |
1880 | ||
1da177e4 LT |
1881 | policy_cache = kmem_cache_create("numa_policy", |
1882 | sizeof(struct mempolicy), | |
20c2df83 | 1883 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
1884 | |
1885 | sn_cache = kmem_cache_create("shared_policy_node", | |
1886 | sizeof(struct sp_node), | |
20c2df83 | 1887 | 0, SLAB_PANIC, NULL); |
1da177e4 | 1888 | |
b71636e2 PM |
1889 | /* |
1890 | * Set interleaving policy for system init. Interleaving is only | |
1891 | * enabled across suitably sized nodes (default is >= 16MB), or | |
1892 | * fall back to the largest node if they're all smaller. | |
1893 | */ | |
1894 | nodes_clear(interleave_nodes); | |
56bbd65d | 1895 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
1896 | unsigned long total_pages = node_present_pages(nid); |
1897 | ||
1898 | /* Preserve the largest node */ | |
1899 | if (largest < total_pages) { | |
1900 | largest = total_pages; | |
1901 | prefer = nid; | |
1902 | } | |
1903 | ||
1904 | /* Interleave this node? */ | |
1905 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
1906 | node_set(nid, interleave_nodes); | |
1907 | } | |
1908 | ||
1909 | /* All too small, use the largest */ | |
1910 | if (unlikely(nodes_empty(interleave_nodes))) | |
1911 | node_set(prefer, interleave_nodes); | |
1da177e4 | 1912 | |
028fec41 | 1913 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
1914 | printk("numa_policy_init: interleaving failed\n"); |
1915 | } | |
1916 | ||
8bccd85f | 1917 | /* Reset policy of current process to default */ |
1da177e4 LT |
1918 | void numa_default_policy(void) |
1919 | { | |
028fec41 | 1920 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 1921 | } |
68860ec1 | 1922 | |
1a75a6c8 CL |
1923 | /* |
1924 | * Display pages allocated per node and memory policy via /proc. | |
1925 | */ | |
15ad7cdc HD |
1926 | static const char * const policy_types[] = |
1927 | { "default", "prefer", "bind", "interleave" }; | |
1a75a6c8 CL |
1928 | |
1929 | /* | |
1930 | * Convert a mempolicy into a string. | |
1931 | * Returns the number of characters in buffer (if positive) | |
1932 | * or an error (negative) | |
1933 | */ | |
1934 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1935 | { | |
1936 | char *p = buffer; | |
1937 | int l; | |
1938 | nodemask_t nodes; | |
45c4745a | 1939 | unsigned short mode = pol ? pol->mode : MPOL_DEFAULT; |
f5b087b5 | 1940 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 CL |
1941 | |
1942 | switch (mode) { | |
1943 | case MPOL_DEFAULT: | |
1944 | nodes_clear(nodes); | |
1945 | break; | |
1946 | ||
1947 | case MPOL_PREFERRED: | |
1948 | nodes_clear(nodes); | |
1949 | node_set(pol->v.preferred_node, nodes); | |
1950 | break; | |
1951 | ||
1952 | case MPOL_BIND: | |
19770b32 | 1953 | /* Fall through */ |
1a75a6c8 CL |
1954 | case MPOL_INTERLEAVE: |
1955 | nodes = pol->v.nodes; | |
1956 | break; | |
1957 | ||
1958 | default: | |
1959 | BUG(); | |
1960 | return -EFAULT; | |
1961 | } | |
1962 | ||
1963 | l = strlen(policy_types[mode]); | |
1964 | if (buffer + maxlen < p + l + 1) | |
1965 | return -ENOSPC; | |
1966 | ||
1967 | strcpy(p, policy_types[mode]); | |
1968 | p += l; | |
1969 | ||
f5b087b5 DR |
1970 | if (flags) { |
1971 | int need_bar = 0; | |
1972 | ||
1973 | if (buffer + maxlen < p + 2) | |
1974 | return -ENOSPC; | |
1975 | *p++ = '='; | |
1976 | ||
1977 | if (flags & MPOL_F_STATIC_NODES) | |
1978 | p += sprintf(p, "%sstatic", need_bar++ ? "|" : ""); | |
4c50bc01 DR |
1979 | if (flags & MPOL_F_RELATIVE_NODES) |
1980 | p += sprintf(p, "%srelative", need_bar++ ? "|" : ""); | |
f5b087b5 DR |
1981 | } |
1982 | ||
1a75a6c8 CL |
1983 | if (!nodes_empty(nodes)) { |
1984 | if (buffer + maxlen < p + 2) | |
1985 | return -ENOSPC; | |
1986 | *p++ = '='; | |
1987 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
1988 | } | |
1989 | return p - buffer; | |
1990 | } | |
1991 | ||
1992 | struct numa_maps { | |
1993 | unsigned long pages; | |
1994 | unsigned long anon; | |
397874df CL |
1995 | unsigned long active; |
1996 | unsigned long writeback; | |
1a75a6c8 | 1997 | unsigned long mapcount_max; |
397874df CL |
1998 | unsigned long dirty; |
1999 | unsigned long swapcache; | |
1a75a6c8 CL |
2000 | unsigned long node[MAX_NUMNODES]; |
2001 | }; | |
2002 | ||
397874df | 2003 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
2004 | { |
2005 | struct numa_maps *md = private; | |
2006 | int count = page_mapcount(page); | |
2007 | ||
397874df CL |
2008 | md->pages++; |
2009 | if (pte_dirty || PageDirty(page)) | |
2010 | md->dirty++; | |
1a75a6c8 | 2011 | |
397874df CL |
2012 | if (PageSwapCache(page)) |
2013 | md->swapcache++; | |
1a75a6c8 | 2014 | |
397874df CL |
2015 | if (PageActive(page)) |
2016 | md->active++; | |
2017 | ||
2018 | if (PageWriteback(page)) | |
2019 | md->writeback++; | |
1a75a6c8 CL |
2020 | |
2021 | if (PageAnon(page)) | |
2022 | md->anon++; | |
2023 | ||
397874df CL |
2024 | if (count > md->mapcount_max) |
2025 | md->mapcount_max = count; | |
2026 | ||
1a75a6c8 | 2027 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
2028 | } |
2029 | ||
7f709ed0 | 2030 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
2031 | static void check_huge_range(struct vm_area_struct *vma, |
2032 | unsigned long start, unsigned long end, | |
2033 | struct numa_maps *md) | |
2034 | { | |
2035 | unsigned long addr; | |
2036 | struct page *page; | |
2037 | ||
2038 | for (addr = start; addr < end; addr += HPAGE_SIZE) { | |
2039 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); | |
2040 | pte_t pte; | |
2041 | ||
2042 | if (!ptep) | |
2043 | continue; | |
2044 | ||
2045 | pte = *ptep; | |
2046 | if (pte_none(pte)) | |
2047 | continue; | |
2048 | ||
2049 | page = pte_page(pte); | |
2050 | if (!page) | |
2051 | continue; | |
2052 | ||
2053 | gather_stats(page, md, pte_dirty(*ptep)); | |
2054 | } | |
2055 | } | |
7f709ed0 AM |
2056 | #else |
2057 | static inline void check_huge_range(struct vm_area_struct *vma, | |
2058 | unsigned long start, unsigned long end, | |
2059 | struct numa_maps *md) | |
2060 | { | |
2061 | } | |
2062 | #endif | |
397874df | 2063 | |
1a75a6c8 CL |
2064 | int show_numa_map(struct seq_file *m, void *v) |
2065 | { | |
99f89551 | 2066 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
2067 | struct vm_area_struct *vma = v; |
2068 | struct numa_maps *md; | |
397874df CL |
2069 | struct file *file = vma->vm_file; |
2070 | struct mm_struct *mm = vma->vm_mm; | |
480eccf9 | 2071 | struct mempolicy *pol; |
1a75a6c8 CL |
2072 | int n; |
2073 | char buffer[50]; | |
2074 | ||
397874df | 2075 | if (!mm) |
1a75a6c8 CL |
2076 | return 0; |
2077 | ||
2078 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
2079 | if (!md) | |
2080 | return 0; | |
2081 | ||
480eccf9 LS |
2082 | pol = get_vma_policy(priv->task, vma, vma->vm_start); |
2083 | mpol_to_str(buffer, sizeof(buffer), pol); | |
2084 | /* | |
2085 | * unref shared or other task's mempolicy | |
2086 | */ | |
2087 | if (pol != &default_policy && pol != current->mempolicy) | |
f0be3d32 | 2088 | __mpol_put(pol); |
397874df CL |
2089 | |
2090 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
2091 | ||
2092 | if (file) { | |
2093 | seq_printf(m, " file="); | |
c32c2f63 | 2094 | seq_path(m, &file->f_path, "\n\t= "); |
397874df CL |
2095 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
2096 | seq_printf(m, " heap"); | |
2097 | } else if (vma->vm_start <= mm->start_stack && | |
2098 | vma->vm_end >= mm->start_stack) { | |
2099 | seq_printf(m, " stack"); | |
2100 | } | |
2101 | ||
2102 | if (is_vm_hugetlb_page(vma)) { | |
2103 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
2104 | seq_printf(m, " huge"); | |
2105 | } else { | |
a57ebfdb | 2106 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
56bbd65d | 2107 | &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md); |
397874df CL |
2108 | } |
2109 | ||
2110 | if (!md->pages) | |
2111 | goto out; | |
1a75a6c8 | 2112 | |
397874df CL |
2113 | if (md->anon) |
2114 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 2115 | |
397874df CL |
2116 | if (md->dirty) |
2117 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 2118 | |
397874df CL |
2119 | if (md->pages != md->anon && md->pages != md->dirty) |
2120 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 2121 | |
397874df CL |
2122 | if (md->mapcount_max > 1) |
2123 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 2124 | |
397874df CL |
2125 | if (md->swapcache) |
2126 | seq_printf(m," swapcache=%lu", md->swapcache); | |
2127 | ||
2128 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
2129 | seq_printf(m," active=%lu", md->active); | |
2130 | ||
2131 | if (md->writeback) | |
2132 | seq_printf(m," writeback=%lu", md->writeback); | |
2133 | ||
56bbd65d | 2134 | for_each_node_state(n, N_HIGH_MEMORY) |
397874df CL |
2135 | if (md->node[n]) |
2136 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
2137 | out: | |
2138 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
2139 | kfree(md); |
2140 | ||
2141 | if (m->count < m->size) | |
99f89551 | 2142 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
2143 | return 0; |
2144 | } |