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 | |
bea904d5 LS |
107 | /* |
108 | * run-time system-wide default policy => local allocation | |
109 | */ | |
d42c6997 | 110 | struct mempolicy default_policy = { |
1da177e4 | 111 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 112 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 113 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
114 | }; |
115 | ||
37012946 DR |
116 | static const struct mempolicy_operations { |
117 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
118 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); | |
119 | } mpol_ops[MPOL_MAX]; | |
120 | ||
19770b32 | 121 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 122 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 123 | { |
19770b32 | 124 | int nd, k; |
1da177e4 | 125 | |
19770b32 MG |
126 | /* Check that there is something useful in this mask */ |
127 | k = policy_zone; | |
128 | ||
129 | for_each_node_mask(nd, *nodemask) { | |
130 | struct zone *z; | |
131 | ||
132 | for (k = 0; k <= policy_zone; k++) { | |
133 | z = &NODE_DATA(nd)->node_zones[k]; | |
134 | if (z->present_pages > 0) | |
135 | return 1; | |
dd942ae3 | 136 | } |
8af5e2eb | 137 | } |
19770b32 MG |
138 | |
139 | return 0; | |
1da177e4 LT |
140 | } |
141 | ||
f5b087b5 DR |
142 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
143 | { | |
4c50bc01 DR |
144 | return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES); |
145 | } | |
146 | ||
147 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
148 | const nodemask_t *rel) | |
149 | { | |
150 | nodemask_t tmp; | |
151 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
152 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
153 | } |
154 | ||
37012946 DR |
155 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
156 | { | |
157 | if (nodes_empty(*nodes)) | |
158 | return -EINVAL; | |
159 | pol->v.nodes = *nodes; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
164 | { | |
165 | if (!nodes) | |
fc36b8d3 | 166 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
167 | else if (nodes_empty(*nodes)) |
168 | return -EINVAL; /* no allowed nodes */ | |
169 | else | |
170 | pol->v.preferred_node = first_node(*nodes); | |
171 | return 0; | |
172 | } | |
173 | ||
174 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
175 | { | |
176 | if (!is_valid_nodemask(nodes)) | |
177 | return -EINVAL; | |
178 | pol->v.nodes = *nodes; | |
179 | return 0; | |
180 | } | |
181 | ||
1da177e4 | 182 | /* Create a new policy */ |
028fec41 DR |
183 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
184 | nodemask_t *nodes) | |
1da177e4 LT |
185 | { |
186 | struct mempolicy *policy; | |
f5b087b5 | 187 | nodemask_t cpuset_context_nmask; |
37012946 | 188 | int ret; |
1da177e4 | 189 | |
028fec41 DR |
190 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
191 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 192 | |
3e1f0645 DR |
193 | if (mode == MPOL_DEFAULT) { |
194 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 195 | return ERR_PTR(-EINVAL); |
bea904d5 | 196 | return NULL; /* simply delete any existing policy */ |
37012946 | 197 | } |
3e1f0645 DR |
198 | VM_BUG_ON(!nodes); |
199 | ||
200 | /* | |
201 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
202 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
203 | * All other modes require a valid pointer to a non-empty nodemask. | |
204 | */ | |
205 | if (mode == MPOL_PREFERRED) { | |
206 | if (nodes_empty(*nodes)) { | |
207 | if (((flags & MPOL_F_STATIC_NODES) || | |
208 | (flags & MPOL_F_RELATIVE_NODES))) | |
209 | return ERR_PTR(-EINVAL); | |
210 | nodes = NULL; /* flag local alloc */ | |
211 | } | |
212 | } else if (nodes_empty(*nodes)) | |
213 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
214 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
215 | if (!policy) | |
216 | return ERR_PTR(-ENOMEM); | |
217 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 218 | policy->mode = mode; |
3e1f0645 | 219 | policy->flags = flags; |
37012946 | 220 | |
3e1f0645 DR |
221 | if (nodes) { |
222 | /* | |
223 | * cpuset related setup doesn't apply to local allocation | |
224 | */ | |
37012946 DR |
225 | cpuset_update_task_memory_state(); |
226 | if (flags & MPOL_F_RELATIVE_NODES) | |
227 | mpol_relative_nodemask(&cpuset_context_nmask, nodes, | |
228 | &cpuset_current_mems_allowed); | |
229 | else | |
230 | nodes_and(cpuset_context_nmask, *nodes, | |
231 | cpuset_current_mems_allowed); | |
232 | if (mpol_store_user_nodemask(policy)) | |
233 | policy->w.user_nodemask = *nodes; | |
234 | else | |
235 | policy->w.cpuset_mems_allowed = | |
236 | cpuset_mems_allowed(current); | |
237 | } | |
238 | ||
239 | ret = mpol_ops[mode].create(policy, | |
3e1f0645 | 240 | nodes ? &cpuset_context_nmask : NULL); |
37012946 DR |
241 | if (ret < 0) { |
242 | kmem_cache_free(policy_cache, policy); | |
243 | return ERR_PTR(ret); | |
244 | } | |
1da177e4 | 245 | return policy; |
37012946 DR |
246 | } |
247 | ||
52cd3b07 LS |
248 | /* Slow path of a mpol destructor. */ |
249 | void __mpol_put(struct mempolicy *p) | |
250 | { | |
251 | if (!atomic_dec_and_test(&p->refcnt)) | |
252 | return; | |
52cd3b07 LS |
253 | kmem_cache_free(policy_cache, p); |
254 | } | |
255 | ||
37012946 DR |
256 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
257 | { | |
258 | } | |
259 | ||
260 | static void mpol_rebind_nodemask(struct mempolicy *pol, | |
261 | const nodemask_t *nodes) | |
262 | { | |
263 | nodemask_t tmp; | |
264 | ||
265 | if (pol->flags & MPOL_F_STATIC_NODES) | |
266 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
267 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
268 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
269 | else { | |
270 | nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed, | |
271 | *nodes); | |
272 | pol->w.cpuset_mems_allowed = *nodes; | |
273 | } | |
f5b087b5 | 274 | |
37012946 DR |
275 | pol->v.nodes = tmp; |
276 | if (!node_isset(current->il_next, tmp)) { | |
277 | current->il_next = next_node(current->il_next, tmp); | |
278 | if (current->il_next >= MAX_NUMNODES) | |
279 | current->il_next = first_node(tmp); | |
280 | if (current->il_next >= MAX_NUMNODES) | |
281 | current->il_next = numa_node_id(); | |
282 | } | |
283 | } | |
284 | ||
285 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
286 | const nodemask_t *nodes) | |
287 | { | |
288 | nodemask_t tmp; | |
289 | ||
37012946 DR |
290 | if (pol->flags & MPOL_F_STATIC_NODES) { |
291 | int node = first_node(pol->w.user_nodemask); | |
292 | ||
fc36b8d3 | 293 | if (node_isset(node, *nodes)) { |
37012946 | 294 | pol->v.preferred_node = node; |
fc36b8d3 LS |
295 | pol->flags &= ~MPOL_F_LOCAL; |
296 | } else | |
297 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
298 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
299 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
300 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 301 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
302 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
303 | pol->w.cpuset_mems_allowed, | |
304 | *nodes); | |
305 | pol->w.cpuset_mems_allowed = *nodes; | |
306 | } | |
1da177e4 LT |
307 | } |
308 | ||
1d0d2680 DR |
309 | /* Migrate a policy to a different set of nodes */ |
310 | static void mpol_rebind_policy(struct mempolicy *pol, | |
311 | const nodemask_t *newmask) | |
312 | { | |
1d0d2680 DR |
313 | if (!pol) |
314 | return; | |
1d0d2680 DR |
315 | if (!mpol_store_user_nodemask(pol) && |
316 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) | |
317 | return; | |
45c4745a | 318 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
319 | } |
320 | ||
321 | /* | |
322 | * Wrapper for mpol_rebind_policy() that just requires task | |
323 | * pointer, and updates task mempolicy. | |
324 | */ | |
325 | ||
326 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
327 | { | |
328 | mpol_rebind_policy(tsk->mempolicy, new); | |
329 | } | |
330 | ||
331 | /* | |
332 | * Rebind each vma in mm to new nodemask. | |
333 | * | |
334 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
335 | */ | |
336 | ||
337 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
338 | { | |
339 | struct vm_area_struct *vma; | |
340 | ||
341 | down_write(&mm->mmap_sem); | |
342 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
343 | mpol_rebind_policy(vma->vm_policy, new); | |
344 | up_write(&mm->mmap_sem); | |
345 | } | |
346 | ||
37012946 DR |
347 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
348 | [MPOL_DEFAULT] = { | |
349 | .rebind = mpol_rebind_default, | |
350 | }, | |
351 | [MPOL_INTERLEAVE] = { | |
352 | .create = mpol_new_interleave, | |
353 | .rebind = mpol_rebind_nodemask, | |
354 | }, | |
355 | [MPOL_PREFERRED] = { | |
356 | .create = mpol_new_preferred, | |
357 | .rebind = mpol_rebind_preferred, | |
358 | }, | |
359 | [MPOL_BIND] = { | |
360 | .create = mpol_new_bind, | |
361 | .rebind = mpol_rebind_nodemask, | |
362 | }, | |
363 | }; | |
364 | ||
397874df | 365 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
366 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
367 | unsigned long flags); | |
1a75a6c8 | 368 | |
38e35860 | 369 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 370 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
371 | unsigned long addr, unsigned long end, |
372 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 373 | void *private) |
1da177e4 | 374 | { |
91612e0d HD |
375 | pte_t *orig_pte; |
376 | pte_t *pte; | |
705e87c0 | 377 | spinlock_t *ptl; |
941150a3 | 378 | |
705e87c0 | 379 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 380 | do { |
6aab341e | 381 | struct page *page; |
25ba77c1 | 382 | int nid; |
91612e0d HD |
383 | |
384 | if (!pte_present(*pte)) | |
1da177e4 | 385 | continue; |
6aab341e LT |
386 | page = vm_normal_page(vma, addr, *pte); |
387 | if (!page) | |
1da177e4 | 388 | continue; |
053837fc NP |
389 | /* |
390 | * The check for PageReserved here is important to avoid | |
391 | * handling zero pages and other pages that may have been | |
392 | * marked special by the system. | |
393 | * | |
394 | * If the PageReserved would not be checked here then f.e. | |
395 | * the location of the zero page could have an influence | |
396 | * on MPOL_MF_STRICT, zero pages would be counted for | |
397 | * the per node stats, and there would be useless attempts | |
398 | * to put zero pages on the migration list. | |
399 | */ | |
f4598c8b CL |
400 | if (PageReserved(page)) |
401 | continue; | |
6aab341e | 402 | nid = page_to_nid(page); |
38e35860 CL |
403 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
404 | continue; | |
405 | ||
1a75a6c8 | 406 | if (flags & MPOL_MF_STATS) |
397874df | 407 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 408 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 409 | migrate_page_add(page, private, flags); |
38e35860 CL |
410 | else |
411 | break; | |
91612e0d | 412 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 413 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
414 | return addr != end; |
415 | } | |
416 | ||
b5810039 | 417 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
418 | unsigned long addr, unsigned long end, |
419 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 420 | void *private) |
91612e0d HD |
421 | { |
422 | pmd_t *pmd; | |
423 | unsigned long next; | |
424 | ||
425 | pmd = pmd_offset(pud, addr); | |
426 | do { | |
427 | next = pmd_addr_end(addr, end); | |
428 | if (pmd_none_or_clear_bad(pmd)) | |
429 | continue; | |
dc9aa5b9 | 430 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 431 | flags, private)) |
91612e0d HD |
432 | return -EIO; |
433 | } while (pmd++, addr = next, addr != end); | |
434 | return 0; | |
435 | } | |
436 | ||
b5810039 | 437 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
438 | unsigned long addr, unsigned long end, |
439 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 440 | void *private) |
91612e0d HD |
441 | { |
442 | pud_t *pud; | |
443 | unsigned long next; | |
444 | ||
445 | pud = pud_offset(pgd, addr); | |
446 | do { | |
447 | next = pud_addr_end(addr, end); | |
448 | if (pud_none_or_clear_bad(pud)) | |
449 | continue; | |
dc9aa5b9 | 450 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 451 | flags, private)) |
91612e0d HD |
452 | return -EIO; |
453 | } while (pud++, addr = next, addr != end); | |
454 | return 0; | |
455 | } | |
456 | ||
b5810039 | 457 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
458 | unsigned long addr, unsigned long end, |
459 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 460 | void *private) |
91612e0d HD |
461 | { |
462 | pgd_t *pgd; | |
463 | unsigned long next; | |
464 | ||
b5810039 | 465 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
466 | do { |
467 | next = pgd_addr_end(addr, end); | |
468 | if (pgd_none_or_clear_bad(pgd)) | |
469 | continue; | |
dc9aa5b9 | 470 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 471 | flags, private)) |
91612e0d HD |
472 | return -EIO; |
473 | } while (pgd++, addr = next, addr != end); | |
474 | return 0; | |
1da177e4 LT |
475 | } |
476 | ||
dc9aa5b9 CL |
477 | /* |
478 | * Check if all pages in a range are on a set of nodes. | |
479 | * If pagelist != NULL then isolate pages from the LRU and | |
480 | * put them on the pagelist. | |
481 | */ | |
1da177e4 LT |
482 | static struct vm_area_struct * |
483 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 484 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
485 | { |
486 | int err; | |
487 | struct vm_area_struct *first, *vma, *prev; | |
488 | ||
90036ee5 | 489 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
90036ee5 | 490 | |
b20a3503 CL |
491 | err = migrate_prep(); |
492 | if (err) | |
493 | return ERR_PTR(err); | |
90036ee5 | 494 | } |
053837fc | 495 | |
1da177e4 LT |
496 | first = find_vma(mm, start); |
497 | if (!first) | |
498 | return ERR_PTR(-EFAULT); | |
499 | prev = NULL; | |
500 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
501 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
502 | if (!vma->vm_next && vma->vm_end < end) | |
503 | return ERR_PTR(-EFAULT); | |
504 | if (prev && prev->vm_end < vma->vm_start) | |
505 | return ERR_PTR(-EFAULT); | |
506 | } | |
507 | if (!is_vm_hugetlb_page(vma) && | |
508 | ((flags & MPOL_MF_STRICT) || | |
509 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
510 | vma_migratable(vma)))) { | |
5b952b3c | 511 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 512 | |
5b952b3c AK |
513 | if (endvma > end) |
514 | endvma = end; | |
515 | if (vma->vm_start > start) | |
516 | start = vma->vm_start; | |
dc9aa5b9 | 517 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 518 | flags, private); |
1da177e4 LT |
519 | if (err) { |
520 | first = ERR_PTR(err); | |
521 | break; | |
522 | } | |
523 | } | |
524 | prev = vma; | |
525 | } | |
526 | return first; | |
527 | } | |
528 | ||
529 | /* Apply policy to a single VMA */ | |
530 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
531 | { | |
532 | int err = 0; | |
533 | struct mempolicy *old = vma->vm_policy; | |
534 | ||
140d5a49 | 535 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
536 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
537 | vma->vm_ops, vma->vm_file, | |
538 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
539 | ||
540 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
541 | err = vma->vm_ops->set_policy(vma, new); | |
542 | if (!err) { | |
543 | mpol_get(new); | |
544 | vma->vm_policy = new; | |
f0be3d32 | 545 | mpol_put(old); |
1da177e4 LT |
546 | } |
547 | return err; | |
548 | } | |
549 | ||
550 | /* Step 2: apply policy to a range and do splits. */ | |
551 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
552 | unsigned long end, struct mempolicy *new) | |
553 | { | |
554 | struct vm_area_struct *next; | |
555 | int err; | |
556 | ||
557 | err = 0; | |
558 | for (; vma && vma->vm_start < end; vma = next) { | |
559 | next = vma->vm_next; | |
560 | if (vma->vm_start < start) | |
561 | err = split_vma(vma->vm_mm, vma, start, 1); | |
562 | if (!err && vma->vm_end > end) | |
563 | err = split_vma(vma->vm_mm, vma, end, 0); | |
564 | if (!err) | |
565 | err = policy_vma(vma, new); | |
566 | if (err) | |
567 | break; | |
568 | } | |
569 | return err; | |
570 | } | |
571 | ||
c61afb18 PJ |
572 | /* |
573 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
574 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
575 | * with other PF_* flag bits) on memory allocation hot code paths. | |
576 | * | |
577 | * If called from outside this file, the task 'p' should -only- be | |
578 | * a newly forked child not yet visible on the task list, because | |
579 | * manipulating the task flags of a visible task is not safe. | |
580 | * | |
581 | * The above limitation is why this routine has the funny name | |
582 | * mpol_fix_fork_child_flag(). | |
583 | * | |
584 | * It is also safe to call this with a task pointer of current, | |
585 | * which the static wrapper mpol_set_task_struct_flag() does, | |
586 | * for use within this file. | |
587 | */ | |
588 | ||
589 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
590 | { | |
591 | if (p->mempolicy) | |
592 | p->flags |= PF_MEMPOLICY; | |
593 | else | |
594 | p->flags &= ~PF_MEMPOLICY; | |
595 | } | |
596 | ||
597 | static void mpol_set_task_struct_flag(void) | |
598 | { | |
599 | mpol_fix_fork_child_flag(current); | |
600 | } | |
601 | ||
1da177e4 | 602 | /* Set the process memory policy */ |
028fec41 DR |
603 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
604 | nodemask_t *nodes) | |
1da177e4 | 605 | { |
1da177e4 | 606 | struct mempolicy *new; |
f4e53d91 | 607 | struct mm_struct *mm = current->mm; |
1da177e4 | 608 | |
028fec41 | 609 | new = mpol_new(mode, flags, nodes); |
1da177e4 LT |
610 | if (IS_ERR(new)) |
611 | return PTR_ERR(new); | |
f4e53d91 LS |
612 | |
613 | /* | |
614 | * prevent changing our mempolicy while show_numa_maps() | |
615 | * is using it. | |
616 | * Note: do_set_mempolicy() can be called at init time | |
617 | * with no 'mm'. | |
618 | */ | |
619 | if (mm) | |
620 | down_write(&mm->mmap_sem); | |
f0be3d32 | 621 | mpol_put(current->mempolicy); |
1da177e4 | 622 | current->mempolicy = new; |
c61afb18 | 623 | mpol_set_task_struct_flag(); |
45c4745a | 624 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 625 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 626 | current->il_next = first_node(new->v.nodes); |
f4e53d91 LS |
627 | if (mm) |
628 | up_write(&mm->mmap_sem); | |
629 | ||
1da177e4 LT |
630 | return 0; |
631 | } | |
632 | ||
bea904d5 LS |
633 | /* |
634 | * Return nodemask for policy for get_mempolicy() query | |
635 | */ | |
636 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 637 | { |
dfcd3c0d | 638 | nodes_clear(*nodes); |
bea904d5 LS |
639 | if (p == &default_policy) |
640 | return; | |
641 | ||
45c4745a | 642 | switch (p->mode) { |
19770b32 MG |
643 | case MPOL_BIND: |
644 | /* Fall through */ | |
1da177e4 | 645 | case MPOL_INTERLEAVE: |
dfcd3c0d | 646 | *nodes = p->v.nodes; |
1da177e4 LT |
647 | break; |
648 | case MPOL_PREFERRED: | |
fc36b8d3 | 649 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 650 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 651 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
652 | break; |
653 | default: | |
654 | BUG(); | |
655 | } | |
656 | } | |
657 | ||
658 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
659 | { | |
660 | struct page *p; | |
661 | int err; | |
662 | ||
663 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
664 | if (err >= 0) { | |
665 | err = page_to_nid(p); | |
666 | put_page(p); | |
667 | } | |
668 | return err; | |
669 | } | |
670 | ||
1da177e4 | 671 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
672 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
673 | unsigned long addr, unsigned long flags) | |
1da177e4 | 674 | { |
8bccd85f | 675 | int err; |
1da177e4 LT |
676 | struct mm_struct *mm = current->mm; |
677 | struct vm_area_struct *vma = NULL; | |
678 | struct mempolicy *pol = current->mempolicy; | |
679 | ||
cf2a473c | 680 | cpuset_update_task_memory_state(); |
754af6f5 LS |
681 | if (flags & |
682 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 683 | return -EINVAL; |
754af6f5 LS |
684 | |
685 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
686 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
687 | return -EINVAL; | |
688 | *policy = 0; /* just so it's initialized */ | |
689 | *nmask = cpuset_current_mems_allowed; | |
690 | return 0; | |
691 | } | |
692 | ||
1da177e4 | 693 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
694 | /* |
695 | * Do NOT fall back to task policy if the | |
696 | * vma/shared policy at addr is NULL. We | |
697 | * want to return MPOL_DEFAULT in this case. | |
698 | */ | |
1da177e4 LT |
699 | down_read(&mm->mmap_sem); |
700 | vma = find_vma_intersection(mm, addr, addr+1); | |
701 | if (!vma) { | |
702 | up_read(&mm->mmap_sem); | |
703 | return -EFAULT; | |
704 | } | |
705 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
706 | pol = vma->vm_ops->get_policy(vma, addr); | |
707 | else | |
708 | pol = vma->vm_policy; | |
709 | } else if (addr) | |
710 | return -EINVAL; | |
711 | ||
712 | if (!pol) | |
bea904d5 | 713 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
714 | |
715 | if (flags & MPOL_F_NODE) { | |
716 | if (flags & MPOL_F_ADDR) { | |
717 | err = lookup_node(mm, addr); | |
718 | if (err < 0) | |
719 | goto out; | |
8bccd85f | 720 | *policy = err; |
1da177e4 | 721 | } else if (pol == current->mempolicy && |
45c4745a | 722 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 723 | *policy = current->il_next; |
1da177e4 LT |
724 | } else { |
725 | err = -EINVAL; | |
726 | goto out; | |
727 | } | |
bea904d5 LS |
728 | } else { |
729 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
730 | pol->mode; | |
731 | *policy |= pol->flags; | |
732 | } | |
1da177e4 LT |
733 | |
734 | if (vma) { | |
735 | up_read(¤t->mm->mmap_sem); | |
736 | vma = NULL; | |
737 | } | |
738 | ||
1da177e4 | 739 | err = 0; |
8bccd85f | 740 | if (nmask) |
bea904d5 | 741 | get_policy_nodemask(pol, nmask); |
1da177e4 LT |
742 | |
743 | out: | |
52cd3b07 | 744 | mpol_cond_put(pol); |
1da177e4 LT |
745 | if (vma) |
746 | up_read(¤t->mm->mmap_sem); | |
747 | return err; | |
748 | } | |
749 | ||
b20a3503 | 750 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
751 | /* |
752 | * page migration | |
753 | */ | |
fc301289 CL |
754 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
755 | unsigned long flags) | |
6ce3c4c0 CL |
756 | { |
757 | /* | |
fc301289 | 758 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 759 | */ |
b20a3503 CL |
760 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
761 | isolate_lru_page(page, pagelist); | |
7e2ab150 | 762 | } |
6ce3c4c0 | 763 | |
742755a1 | 764 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 765 | { |
769848c0 | 766 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
767 | } |
768 | ||
7e2ab150 CL |
769 | /* |
770 | * Migrate pages from one node to a target node. | |
771 | * Returns error or the number of pages not migrated. | |
772 | */ | |
dbcb0f19 AB |
773 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
774 | int flags) | |
7e2ab150 CL |
775 | { |
776 | nodemask_t nmask; | |
777 | LIST_HEAD(pagelist); | |
778 | int err = 0; | |
779 | ||
780 | nodes_clear(nmask); | |
781 | node_set(source, nmask); | |
6ce3c4c0 | 782 | |
7e2ab150 CL |
783 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
784 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
785 | ||
aaa994b3 | 786 | if (!list_empty(&pagelist)) |
95a402c3 CL |
787 | err = migrate_pages(&pagelist, new_node_page, dest); |
788 | ||
7e2ab150 | 789 | return err; |
6ce3c4c0 CL |
790 | } |
791 | ||
39743889 | 792 | /* |
7e2ab150 CL |
793 | * Move pages between the two nodesets so as to preserve the physical |
794 | * layout as much as possible. | |
39743889 CL |
795 | * |
796 | * Returns the number of page that could not be moved. | |
797 | */ | |
798 | int do_migrate_pages(struct mm_struct *mm, | |
799 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
800 | { | |
801 | LIST_HEAD(pagelist); | |
7e2ab150 CL |
802 | int busy = 0; |
803 | int err = 0; | |
804 | nodemask_t tmp; | |
39743889 | 805 | |
53f2556b | 806 | down_read(&mm->mmap_sem); |
39743889 | 807 | |
7b2259b3 CL |
808 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
809 | if (err) | |
810 | goto out; | |
811 | ||
7e2ab150 CL |
812 | /* |
813 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
814 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
815 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
816 | * The pair of nodemasks 'to' and 'from' define the map. | |
817 | * | |
818 | * If no pair of bits is found that way, fallback to picking some | |
819 | * pair of 'source' and 'dest' bits that are not the same. If the | |
820 | * 'source' and 'dest' bits are the same, this represents a node | |
821 | * that will be migrating to itself, so no pages need move. | |
822 | * | |
823 | * If no bits are left in 'tmp', or if all remaining bits left | |
824 | * in 'tmp' correspond to the same bit in 'to', return false | |
825 | * (nothing left to migrate). | |
826 | * | |
827 | * This lets us pick a pair of nodes to migrate between, such that | |
828 | * if possible the dest node is not already occupied by some other | |
829 | * source node, minimizing the risk of overloading the memory on a | |
830 | * node that would happen if we migrated incoming memory to a node | |
831 | * before migrating outgoing memory source that same node. | |
832 | * | |
833 | * A single scan of tmp is sufficient. As we go, we remember the | |
834 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
835 | * that not only moved, but what's better, moved to an empty slot | |
836 | * (d is not set in tmp), then we break out then, with that pair. | |
837 | * Otherwise when we finish scannng from_tmp, we at least have the | |
838 | * most recent <s, d> pair that moved. If we get all the way through | |
839 | * the scan of tmp without finding any node that moved, much less | |
840 | * moved to an empty node, then there is nothing left worth migrating. | |
841 | */ | |
d4984711 | 842 | |
7e2ab150 CL |
843 | tmp = *from_nodes; |
844 | while (!nodes_empty(tmp)) { | |
845 | int s,d; | |
846 | int source = -1; | |
847 | int dest = 0; | |
848 | ||
849 | for_each_node_mask(s, tmp) { | |
850 | d = node_remap(s, *from_nodes, *to_nodes); | |
851 | if (s == d) | |
852 | continue; | |
853 | ||
854 | source = s; /* Node moved. Memorize */ | |
855 | dest = d; | |
856 | ||
857 | /* dest not in remaining from nodes? */ | |
858 | if (!node_isset(dest, tmp)) | |
859 | break; | |
860 | } | |
861 | if (source == -1) | |
862 | break; | |
863 | ||
864 | node_clear(source, tmp); | |
865 | err = migrate_to_node(mm, source, dest, flags); | |
866 | if (err > 0) | |
867 | busy += err; | |
868 | if (err < 0) | |
869 | break; | |
39743889 | 870 | } |
7b2259b3 | 871 | out: |
39743889 | 872 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
873 | if (err < 0) |
874 | return err; | |
875 | return busy; | |
b20a3503 CL |
876 | |
877 | } | |
878 | ||
3ad33b24 LS |
879 | /* |
880 | * Allocate a new page for page migration based on vma policy. | |
881 | * Start assuming that page is mapped by vma pointed to by @private. | |
882 | * Search forward from there, if not. N.B., this assumes that the | |
883 | * list of pages handed to migrate_pages()--which is how we get here-- | |
884 | * is in virtual address order. | |
885 | */ | |
742755a1 | 886 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
887 | { |
888 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 889 | unsigned long uninitialized_var(address); |
95a402c3 | 890 | |
3ad33b24 LS |
891 | while (vma) { |
892 | address = page_address_in_vma(page, vma); | |
893 | if (address != -EFAULT) | |
894 | break; | |
895 | vma = vma->vm_next; | |
896 | } | |
897 | ||
898 | /* | |
899 | * if !vma, alloc_page_vma() will use task or system default policy | |
900 | */ | |
901 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 902 | } |
b20a3503 CL |
903 | #else |
904 | ||
905 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
906 | unsigned long flags) | |
907 | { | |
39743889 CL |
908 | } |
909 | ||
b20a3503 CL |
910 | int do_migrate_pages(struct mm_struct *mm, |
911 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
912 | { | |
913 | return -ENOSYS; | |
914 | } | |
95a402c3 | 915 | |
69939749 | 916 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
917 | { |
918 | return NULL; | |
919 | } | |
b20a3503 CL |
920 | #endif |
921 | ||
dbcb0f19 | 922 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
923 | unsigned short mode, unsigned short mode_flags, |
924 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
925 | { |
926 | struct vm_area_struct *vma; | |
927 | struct mm_struct *mm = current->mm; | |
928 | struct mempolicy *new; | |
929 | unsigned long end; | |
930 | int err; | |
931 | LIST_HEAD(pagelist); | |
932 | ||
a3b51e01 DR |
933 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
934 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 935 | return -EINVAL; |
74c00241 | 936 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
937 | return -EPERM; |
938 | ||
939 | if (start & ~PAGE_MASK) | |
940 | return -EINVAL; | |
941 | ||
942 | if (mode == MPOL_DEFAULT) | |
943 | flags &= ~MPOL_MF_STRICT; | |
944 | ||
945 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
946 | end = start + len; | |
947 | ||
948 | if (end < start) | |
949 | return -EINVAL; | |
950 | if (end == start) | |
951 | return 0; | |
952 | ||
028fec41 | 953 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
954 | if (IS_ERR(new)) |
955 | return PTR_ERR(new); | |
956 | ||
957 | /* | |
958 | * If we are using the default policy then operation | |
959 | * on discontinuous address spaces is okay after all | |
960 | */ | |
961 | if (!new) | |
962 | flags |= MPOL_MF_DISCONTIG_OK; | |
963 | ||
028fec41 DR |
964 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
965 | start, start + len, mode, mode_flags, | |
966 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 CL |
967 | |
968 | down_write(&mm->mmap_sem); | |
969 | vma = check_range(mm, start, end, nmask, | |
970 | flags | MPOL_MF_INVERT, &pagelist); | |
971 | ||
972 | err = PTR_ERR(vma); | |
973 | if (!IS_ERR(vma)) { | |
974 | int nr_failed = 0; | |
975 | ||
976 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 977 | |
6ce3c4c0 | 978 | if (!list_empty(&pagelist)) |
95a402c3 CL |
979 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
980 | (unsigned long)vma); | |
6ce3c4c0 CL |
981 | |
982 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
983 | err = -EIO; | |
984 | } | |
b20a3503 | 985 | |
6ce3c4c0 | 986 | up_write(&mm->mmap_sem); |
f0be3d32 | 987 | mpol_put(new); |
6ce3c4c0 CL |
988 | return err; |
989 | } | |
990 | ||
8bccd85f CL |
991 | /* |
992 | * User space interface with variable sized bitmaps for nodelists. | |
993 | */ | |
994 | ||
995 | /* Copy a node mask from user space. */ | |
39743889 | 996 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
997 | unsigned long maxnode) |
998 | { | |
999 | unsigned long k; | |
1000 | unsigned long nlongs; | |
1001 | unsigned long endmask; | |
1002 | ||
1003 | --maxnode; | |
1004 | nodes_clear(*nodes); | |
1005 | if (maxnode == 0 || !nmask) | |
1006 | return 0; | |
a9c930ba | 1007 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1008 | return -EINVAL; |
8bccd85f CL |
1009 | |
1010 | nlongs = BITS_TO_LONGS(maxnode); | |
1011 | if ((maxnode % BITS_PER_LONG) == 0) | |
1012 | endmask = ~0UL; | |
1013 | else | |
1014 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1015 | ||
1016 | /* When the user specified more nodes than supported just check | |
1017 | if the non supported part is all zero. */ | |
1018 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1019 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1020 | return -EINVAL; | |
1021 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1022 | unsigned long t; | |
1023 | if (get_user(t, nmask + k)) | |
1024 | return -EFAULT; | |
1025 | if (k == nlongs - 1) { | |
1026 | if (t & endmask) | |
1027 | return -EINVAL; | |
1028 | } else if (t) | |
1029 | return -EINVAL; | |
1030 | } | |
1031 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1032 | endmask = ~0UL; | |
1033 | } | |
1034 | ||
1035 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1036 | return -EFAULT; | |
1037 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | /* Copy a kernel node mask to user space */ | |
1042 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1043 | nodemask_t *nodes) | |
1044 | { | |
1045 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1046 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1047 | ||
1048 | if (copy > nbytes) { | |
1049 | if (copy > PAGE_SIZE) | |
1050 | return -EINVAL; | |
1051 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1052 | return -EFAULT; | |
1053 | copy = nbytes; | |
1054 | } | |
1055 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1056 | } | |
1057 | ||
1058 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
1059 | unsigned long mode, | |
1060 | unsigned long __user *nmask, unsigned long maxnode, | |
1061 | unsigned flags) | |
1062 | { | |
1063 | nodemask_t nodes; | |
1064 | int err; | |
028fec41 | 1065 | unsigned short mode_flags; |
8bccd85f | 1066 | |
028fec41 DR |
1067 | mode_flags = mode & MPOL_MODE_FLAGS; |
1068 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1069 | if (mode >= MPOL_MAX) |
1070 | return -EINVAL; | |
4c50bc01 DR |
1071 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1072 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1073 | return -EINVAL; | |
8bccd85f CL |
1074 | err = get_nodes(&nodes, nmask, maxnode); |
1075 | if (err) | |
1076 | return err; | |
028fec41 | 1077 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1078 | } |
1079 | ||
1080 | /* Set the process memory policy */ | |
1081 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
1082 | unsigned long maxnode) | |
1083 | { | |
1084 | int err; | |
1085 | nodemask_t nodes; | |
028fec41 | 1086 | unsigned short flags; |
8bccd85f | 1087 | |
028fec41 DR |
1088 | flags = mode & MPOL_MODE_FLAGS; |
1089 | mode &= ~MPOL_MODE_FLAGS; | |
1090 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1091 | return -EINVAL; |
4c50bc01 DR |
1092 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1093 | return -EINVAL; | |
8bccd85f CL |
1094 | err = get_nodes(&nodes, nmask, maxnode); |
1095 | if (err) | |
1096 | return err; | |
028fec41 | 1097 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1098 | } |
1099 | ||
39743889 CL |
1100 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
1101 | const unsigned long __user *old_nodes, | |
1102 | const unsigned long __user *new_nodes) | |
1103 | { | |
1104 | struct mm_struct *mm; | |
1105 | struct task_struct *task; | |
1106 | nodemask_t old; | |
1107 | nodemask_t new; | |
1108 | nodemask_t task_nodes; | |
1109 | int err; | |
1110 | ||
1111 | err = get_nodes(&old, old_nodes, maxnode); | |
1112 | if (err) | |
1113 | return err; | |
1114 | ||
1115 | err = get_nodes(&new, new_nodes, maxnode); | |
1116 | if (err) | |
1117 | return err; | |
1118 | ||
1119 | /* Find the mm_struct */ | |
1120 | read_lock(&tasklist_lock); | |
228ebcbe | 1121 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 CL |
1122 | if (!task) { |
1123 | read_unlock(&tasklist_lock); | |
1124 | return -ESRCH; | |
1125 | } | |
1126 | mm = get_task_mm(task); | |
1127 | read_unlock(&tasklist_lock); | |
1128 | ||
1129 | if (!mm) | |
1130 | return -EINVAL; | |
1131 | ||
1132 | /* | |
1133 | * Check if this process has the right to modify the specified | |
1134 | * process. The right exists if the process has administrative | |
7f927fcc | 1135 | * capabilities, superuser privileges or the same |
39743889 CL |
1136 | * userid as the target process. |
1137 | */ | |
1138 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
1139 | (current->uid != task->suid) && (current->uid != task->uid) && | |
74c00241 | 1140 | !capable(CAP_SYS_NICE)) { |
39743889 CL |
1141 | err = -EPERM; |
1142 | goto out; | |
1143 | } | |
1144 | ||
1145 | task_nodes = cpuset_mems_allowed(task); | |
1146 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 1147 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
1148 | err = -EPERM; |
1149 | goto out; | |
1150 | } | |
1151 | ||
37b07e41 | 1152 | if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
1153 | err = -EINVAL; |
1154 | goto out; | |
1155 | } | |
1156 | ||
86c3a764 DQ |
1157 | err = security_task_movememory(task); |
1158 | if (err) | |
1159 | goto out; | |
1160 | ||
511030bc | 1161 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 1162 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
1163 | out: |
1164 | mmput(mm); | |
1165 | return err; | |
1166 | } | |
1167 | ||
1168 | ||
8bccd85f CL |
1169 | /* Retrieve NUMA policy */ |
1170 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
1171 | unsigned long __user *nmask, | |
1172 | unsigned long maxnode, | |
1173 | unsigned long addr, unsigned long flags) | |
1174 | { | |
dbcb0f19 AB |
1175 | int err; |
1176 | int uninitialized_var(pval); | |
8bccd85f CL |
1177 | nodemask_t nodes; |
1178 | ||
1179 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1180 | return -EINVAL; | |
1181 | ||
1182 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1183 | ||
1184 | if (err) | |
1185 | return err; | |
1186 | ||
1187 | if (policy && put_user(pval, policy)) | |
1188 | return -EFAULT; | |
1189 | ||
1190 | if (nmask) | |
1191 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1192 | ||
1193 | return err; | |
1194 | } | |
1195 | ||
1da177e4 LT |
1196 | #ifdef CONFIG_COMPAT |
1197 | ||
1198 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1199 | compat_ulong_t __user *nmask, | |
1200 | compat_ulong_t maxnode, | |
1201 | compat_ulong_t addr, compat_ulong_t flags) | |
1202 | { | |
1203 | long err; | |
1204 | unsigned long __user *nm = NULL; | |
1205 | unsigned long nr_bits, alloc_size; | |
1206 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1207 | ||
1208 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1209 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1210 | ||
1211 | if (nmask) | |
1212 | nm = compat_alloc_user_space(alloc_size); | |
1213 | ||
1214 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1215 | ||
1216 | if (!err && nmask) { | |
1217 | err = copy_from_user(bm, nm, alloc_size); | |
1218 | /* ensure entire bitmap is zeroed */ | |
1219 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1220 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1221 | } | |
1222 | ||
1223 | return err; | |
1224 | } | |
1225 | ||
1226 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1227 | compat_ulong_t maxnode) | |
1228 | { | |
1229 | long err = 0; | |
1230 | unsigned long __user *nm = NULL; | |
1231 | unsigned long nr_bits, alloc_size; | |
1232 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1233 | ||
1234 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1235 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1236 | ||
1237 | if (nmask) { | |
1238 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1239 | nm = compat_alloc_user_space(alloc_size); | |
1240 | err |= copy_to_user(nm, bm, alloc_size); | |
1241 | } | |
1242 | ||
1243 | if (err) | |
1244 | return -EFAULT; | |
1245 | ||
1246 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1247 | } | |
1248 | ||
1249 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1250 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1251 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1252 | { | |
1253 | long err = 0; | |
1254 | unsigned long __user *nm = NULL; | |
1255 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1256 | nodemask_t bm; |
1da177e4 LT |
1257 | |
1258 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1259 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1260 | ||
1261 | if (nmask) { | |
dfcd3c0d | 1262 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1263 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1264 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | if (err) | |
1268 | return -EFAULT; | |
1269 | ||
1270 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1271 | } | |
1272 | ||
1273 | #endif | |
1274 | ||
480eccf9 LS |
1275 | /* |
1276 | * get_vma_policy(@task, @vma, @addr) | |
1277 | * @task - task for fallback if vma policy == default | |
1278 | * @vma - virtual memory area whose policy is sought | |
1279 | * @addr - address in @vma for shared policy lookup | |
1280 | * | |
1281 | * Returns effective policy for a VMA at specified address. | |
1282 | * Falls back to @task or system default policy, as necessary. | |
52cd3b07 LS |
1283 | * Current or other task's task mempolicy and non-shared vma policies |
1284 | * are protected by the task's mmap_sem, which must be held for read by | |
1285 | * the caller. | |
1286 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference | |
1287 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1288 | * freeing by another task. It is the caller's responsibility to free the | |
1289 | * extra reference for shared policies. | |
480eccf9 | 1290 | */ |
ae4d8c16 | 1291 | static struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1292 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1293 | { |
6e21c8f1 | 1294 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1295 | |
1296 | if (vma) { | |
480eccf9 | 1297 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1298 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1299 | addr); | |
1300 | if (vpol) | |
1301 | pol = vpol; | |
bea904d5 | 1302 | } else if (vma->vm_policy) |
1da177e4 LT |
1303 | pol = vma->vm_policy; |
1304 | } | |
1305 | if (!pol) | |
1306 | pol = &default_policy; | |
1307 | return pol; | |
1308 | } | |
1309 | ||
52cd3b07 LS |
1310 | /* |
1311 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1312 | * page allocation | |
1313 | */ | |
1314 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1315 | { |
1316 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1317 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1318 | gfp_zone(gfp) >= policy_zone && |
1319 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1320 | return &policy->v.nodes; | |
1321 | ||
1322 | return NULL; | |
1323 | } | |
1324 | ||
52cd3b07 LS |
1325 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
1326 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy) | |
1da177e4 | 1327 | { |
fc36b8d3 | 1328 | int nd = numa_node_id(); |
1da177e4 | 1329 | |
45c4745a | 1330 | switch (policy->mode) { |
1da177e4 | 1331 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1332 | if (!(policy->flags & MPOL_F_LOCAL)) |
1333 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1334 | break; |
1335 | case MPOL_BIND: | |
19770b32 | 1336 | /* |
52cd3b07 LS |
1337 | * Normally, MPOL_BIND allocations are node-local within the |
1338 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
1339 | * current node is part of the mask, we use the zonelist for | |
1340 | * the first node in the mask instead. | |
19770b32 | 1341 | */ |
19770b32 MG |
1342 | if (unlikely(gfp & __GFP_THISNODE) && |
1343 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1344 | nd = first_node(policy->v.nodes); | |
1345 | break; | |
1da177e4 | 1346 | case MPOL_INTERLEAVE: /* should not happen */ |
1da177e4 LT |
1347 | break; |
1348 | default: | |
1da177e4 LT |
1349 | BUG(); |
1350 | } | |
0e88460d | 1351 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1352 | } |
1353 | ||
1354 | /* Do dynamic interleaving for a process */ | |
1355 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1356 | { | |
1357 | unsigned nid, next; | |
1358 | struct task_struct *me = current; | |
1359 | ||
1360 | nid = me->il_next; | |
dfcd3c0d | 1361 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1362 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1363 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1364 | if (next < MAX_NUMNODES) |
1365 | me->il_next = next; | |
1da177e4 LT |
1366 | return nid; |
1367 | } | |
1368 | ||
dc85da15 CL |
1369 | /* |
1370 | * Depending on the memory policy provide a node from which to allocate the | |
1371 | * next slab entry. | |
52cd3b07 LS |
1372 | * @policy must be protected by freeing by the caller. If @policy is |
1373 | * the current task's mempolicy, this protection is implicit, as only the | |
1374 | * task can change it's policy. The system default policy requires no | |
1375 | * such protection. | |
dc85da15 CL |
1376 | */ |
1377 | unsigned slab_node(struct mempolicy *policy) | |
1378 | { | |
fc36b8d3 | 1379 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1380 | return numa_node_id(); |
1381 | ||
1382 | switch (policy->mode) { | |
1383 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1384 | /* |
1385 | * handled MPOL_F_LOCAL above | |
1386 | */ | |
1387 | return policy->v.preferred_node; | |
765c4507 | 1388 | |
dc85da15 CL |
1389 | case MPOL_INTERLEAVE: |
1390 | return interleave_nodes(policy); | |
1391 | ||
dd1a239f | 1392 | case MPOL_BIND: { |
dc85da15 CL |
1393 | /* |
1394 | * Follow bind policy behavior and start allocation at the | |
1395 | * first node. | |
1396 | */ | |
19770b32 MG |
1397 | struct zonelist *zonelist; |
1398 | struct zone *zone; | |
1399 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1400 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1401 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1402 | &policy->v.nodes, | |
1403 | &zone); | |
1404 | return zone->node; | |
dd1a239f | 1405 | } |
dc85da15 | 1406 | |
dc85da15 | 1407 | default: |
bea904d5 | 1408 | BUG(); |
dc85da15 CL |
1409 | } |
1410 | } | |
1411 | ||
1da177e4 LT |
1412 | /* Do static interleaving for a VMA with known offset. */ |
1413 | static unsigned offset_il_node(struct mempolicy *pol, | |
1414 | struct vm_area_struct *vma, unsigned long off) | |
1415 | { | |
dfcd3c0d | 1416 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1417 | unsigned target; |
1da177e4 LT |
1418 | int c; |
1419 | int nid = -1; | |
1420 | ||
f5b087b5 DR |
1421 | if (!nnodes) |
1422 | return numa_node_id(); | |
1423 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1424 | c = 0; |
1425 | do { | |
dfcd3c0d | 1426 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1427 | c++; |
1428 | } while (c <= target); | |
1da177e4 LT |
1429 | return nid; |
1430 | } | |
1431 | ||
5da7ca86 CL |
1432 | /* Determine a node number for interleave */ |
1433 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1434 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1435 | { | |
1436 | if (vma) { | |
1437 | unsigned long off; | |
1438 | ||
3b98b087 NA |
1439 | /* |
1440 | * for small pages, there is no difference between | |
1441 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1442 | * for huge pages, since vm_pgoff is in units of small | |
1443 | * pages, we need to shift off the always 0 bits to get | |
1444 | * a useful offset. | |
1445 | */ | |
1446 | BUG_ON(shift < PAGE_SHIFT); | |
1447 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1448 | off += (addr - vma->vm_start) >> shift; |
1449 | return offset_il_node(pol, vma, off); | |
1450 | } else | |
1451 | return interleave_nodes(pol); | |
1452 | } | |
1453 | ||
00ac59ad | 1454 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1455 | /* |
1456 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1457 | * @vma = virtual memory area whose policy is sought | |
1458 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1459 | * @gfp_flags = for requested zone | |
19770b32 MG |
1460 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1461 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1462 | * |
52cd3b07 LS |
1463 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1464 | * to the struct mempolicy for conditional unref after allocation. | |
1465 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1466 | * @nodemask for filtering the zonelist. | |
480eccf9 | 1467 | */ |
396faf03 | 1468 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1469 | gfp_t gfp_flags, struct mempolicy **mpol, |
1470 | nodemask_t **nodemask) | |
5da7ca86 | 1471 | { |
480eccf9 | 1472 | struct zonelist *zl; |
5da7ca86 | 1473 | |
52cd3b07 | 1474 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1475 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1476 | |
52cd3b07 LS |
1477 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1478 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
1479 | HPAGE_SHIFT), gfp_flags); | |
1480 | } else { | |
1481 | zl = policy_zonelist(gfp_flags, *mpol); | |
1482 | if ((*mpol)->mode == MPOL_BIND) | |
1483 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1484 | } |
1485 | return zl; | |
5da7ca86 | 1486 | } |
00ac59ad | 1487 | #endif |
5da7ca86 | 1488 | |
1da177e4 LT |
1489 | /* Allocate a page in interleaved policy. |
1490 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1491 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1492 | unsigned nid) | |
1da177e4 LT |
1493 | { |
1494 | struct zonelist *zl; | |
1495 | struct page *page; | |
1496 | ||
0e88460d | 1497 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1498 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1499 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1500 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1501 | return page; |
1502 | } | |
1503 | ||
1504 | /** | |
1505 | * alloc_page_vma - Allocate a page for a VMA. | |
1506 | * | |
1507 | * @gfp: | |
1508 | * %GFP_USER user allocation. | |
1509 | * %GFP_KERNEL kernel allocations, | |
1510 | * %GFP_HIGHMEM highmem/user allocations, | |
1511 | * %GFP_FS allocation should not call back into a file system. | |
1512 | * %GFP_ATOMIC don't sleep. | |
1513 | * | |
1514 | * @vma: Pointer to VMA or NULL if not available. | |
1515 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1516 | * | |
1517 | * This function allocates a page from the kernel page pool and applies | |
1518 | * a NUMA policy associated with the VMA or the current process. | |
1519 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1520 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1521 | * all allocations for pages that will be mapped into | |
1522 | * user space. Returns NULL when no page can be allocated. | |
1523 | * | |
1524 | * Should be called with the mm_sem of the vma hold. | |
1525 | */ | |
1526 | struct page * | |
dd0fc66f | 1527 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1528 | { |
6e21c8f1 | 1529 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1530 | struct zonelist *zl; |
1da177e4 | 1531 | |
cf2a473c | 1532 | cpuset_update_task_memory_state(); |
1da177e4 | 1533 | |
45c4745a | 1534 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1535 | unsigned nid; |
5da7ca86 CL |
1536 | |
1537 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
52cd3b07 | 1538 | mpol_cond_put(pol); |
1da177e4 LT |
1539 | return alloc_page_interleave(gfp, 0, nid); |
1540 | } | |
52cd3b07 LS |
1541 | zl = policy_zonelist(gfp, pol); |
1542 | if (unlikely(mpol_needs_cond_ref(pol))) { | |
480eccf9 | 1543 | /* |
52cd3b07 | 1544 | * slow path: ref counted shared policy |
480eccf9 | 1545 | */ |
19770b32 | 1546 | struct page *page = __alloc_pages_nodemask(gfp, 0, |
52cd3b07 | 1547 | zl, policy_nodemask(gfp, pol)); |
f0be3d32 | 1548 | __mpol_put(pol); |
480eccf9 LS |
1549 | return page; |
1550 | } | |
1551 | /* | |
1552 | * fast path: default or task policy | |
1553 | */ | |
52cd3b07 | 1554 | return __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol)); |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | /** | |
1558 | * alloc_pages_current - Allocate pages. | |
1559 | * | |
1560 | * @gfp: | |
1561 | * %GFP_USER user allocation, | |
1562 | * %GFP_KERNEL kernel allocation, | |
1563 | * %GFP_HIGHMEM highmem allocation, | |
1564 | * %GFP_FS don't call back into a file system. | |
1565 | * %GFP_ATOMIC don't sleep. | |
1566 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1567 | * | |
1568 | * Allocate a page from the kernel page pool. When not in | |
1569 | * interrupt context and apply the current process NUMA policy. | |
1570 | * Returns NULL when no page can be allocated. | |
1571 | * | |
cf2a473c | 1572 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1573 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1574 | * 2) allocating for current task (not interrupt). | |
1575 | */ | |
dd0fc66f | 1576 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1577 | { |
1578 | struct mempolicy *pol = current->mempolicy; | |
1579 | ||
1580 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1581 | cpuset_update_task_memory_state(); |
9b819d20 | 1582 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 1583 | pol = &default_policy; |
52cd3b07 LS |
1584 | |
1585 | /* | |
1586 | * No reference counting needed for current->mempolicy | |
1587 | * nor system default_policy | |
1588 | */ | |
45c4745a | 1589 | if (pol->mode == MPOL_INTERLEAVE) |
1da177e4 | 1590 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
19770b32 | 1591 | return __alloc_pages_nodemask(gfp, order, |
52cd3b07 | 1592 | policy_zonelist(gfp, pol), policy_nodemask(gfp, pol)); |
1da177e4 LT |
1593 | } |
1594 | EXPORT_SYMBOL(alloc_pages_current); | |
1595 | ||
4225399a | 1596 | /* |
846a16bf | 1597 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
1598 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
1599 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1600 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1601 | * further kernel/cpuset.c update_nodemask(). | |
1602 | */ | |
4225399a | 1603 | |
846a16bf LS |
1604 | /* Slow path of a mempolicy duplicate */ |
1605 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
1606 | { |
1607 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1608 | ||
1609 | if (!new) | |
1610 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1611 | if (current_cpuset_is_being_rebound()) { |
1612 | nodemask_t mems = cpuset_mems_allowed(current); | |
1613 | mpol_rebind_policy(old, &mems); | |
1614 | } | |
1da177e4 LT |
1615 | *new = *old; |
1616 | atomic_set(&new->refcnt, 1); | |
1da177e4 LT |
1617 | return new; |
1618 | } | |
1619 | ||
52cd3b07 LS |
1620 | /* |
1621 | * If *frompol needs [has] an extra ref, copy *frompol to *tompol , | |
1622 | * eliminate the * MPOL_F_* flags that require conditional ref and | |
1623 | * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly | |
1624 | * after return. Use the returned value. | |
1625 | * | |
1626 | * Allows use of a mempolicy for, e.g., multiple allocations with a single | |
1627 | * policy lookup, even if the policy needs/has extra ref on lookup. | |
1628 | * shmem_readahead needs this. | |
1629 | */ | |
1630 | struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
1631 | struct mempolicy *frompol) | |
1632 | { | |
1633 | if (!mpol_needs_cond_ref(frompol)) | |
1634 | return frompol; | |
1635 | ||
1636 | *tompol = *frompol; | |
1637 | tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */ | |
1638 | __mpol_put(frompol); | |
1639 | return tompol; | |
1640 | } | |
1641 | ||
f5b087b5 DR |
1642 | static int mpol_match_intent(const struct mempolicy *a, |
1643 | const struct mempolicy *b) | |
1644 | { | |
1645 | if (a->flags != b->flags) | |
1646 | return 0; | |
1647 | if (!mpol_store_user_nodemask(a)) | |
1648 | return 1; | |
1649 | return nodes_equal(a->w.user_nodemask, b->w.user_nodemask); | |
1650 | } | |
1651 | ||
1da177e4 LT |
1652 | /* Slow path of a mempolicy comparison */ |
1653 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1654 | { | |
1655 | if (!a || !b) | |
1656 | return 0; | |
45c4745a | 1657 | if (a->mode != b->mode) |
1da177e4 | 1658 | return 0; |
45c4745a | 1659 | if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b)) |
f5b087b5 | 1660 | return 0; |
45c4745a | 1661 | switch (a->mode) { |
19770b32 MG |
1662 | case MPOL_BIND: |
1663 | /* Fall through */ | |
1da177e4 | 1664 | case MPOL_INTERLEAVE: |
dfcd3c0d | 1665 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 1666 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1667 | return a->v.preferred_node == b->v.preferred_node && |
1668 | a->flags == b->flags; | |
1da177e4 LT |
1669 | default: |
1670 | BUG(); | |
1671 | return 0; | |
1672 | } | |
1673 | } | |
1674 | ||
1da177e4 LT |
1675 | /* |
1676 | * Shared memory backing store policy support. | |
1677 | * | |
1678 | * Remember policies even when nobody has shared memory mapped. | |
1679 | * The policies are kept in Red-Black tree linked from the inode. | |
1680 | * They are protected by the sp->lock spinlock, which should be held | |
1681 | * for any accesses to the tree. | |
1682 | */ | |
1683 | ||
1684 | /* lookup first element intersecting start-end */ | |
1685 | /* Caller holds sp->lock */ | |
1686 | static struct sp_node * | |
1687 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1688 | { | |
1689 | struct rb_node *n = sp->root.rb_node; | |
1690 | ||
1691 | while (n) { | |
1692 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1693 | ||
1694 | if (start >= p->end) | |
1695 | n = n->rb_right; | |
1696 | else if (end <= p->start) | |
1697 | n = n->rb_left; | |
1698 | else | |
1699 | break; | |
1700 | } | |
1701 | if (!n) | |
1702 | return NULL; | |
1703 | for (;;) { | |
1704 | struct sp_node *w = NULL; | |
1705 | struct rb_node *prev = rb_prev(n); | |
1706 | if (!prev) | |
1707 | break; | |
1708 | w = rb_entry(prev, struct sp_node, nd); | |
1709 | if (w->end <= start) | |
1710 | break; | |
1711 | n = prev; | |
1712 | } | |
1713 | return rb_entry(n, struct sp_node, nd); | |
1714 | } | |
1715 | ||
1716 | /* Insert a new shared policy into the list. */ | |
1717 | /* Caller holds sp->lock */ | |
1718 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1719 | { | |
1720 | struct rb_node **p = &sp->root.rb_node; | |
1721 | struct rb_node *parent = NULL; | |
1722 | struct sp_node *nd; | |
1723 | ||
1724 | while (*p) { | |
1725 | parent = *p; | |
1726 | nd = rb_entry(parent, struct sp_node, nd); | |
1727 | if (new->start < nd->start) | |
1728 | p = &(*p)->rb_left; | |
1729 | else if (new->end > nd->end) | |
1730 | p = &(*p)->rb_right; | |
1731 | else | |
1732 | BUG(); | |
1733 | } | |
1734 | rb_link_node(&new->nd, parent, p); | |
1735 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1736 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 1737 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
1738 | } |
1739 | ||
1740 | /* Find shared policy intersecting idx */ | |
1741 | struct mempolicy * | |
1742 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1743 | { | |
1744 | struct mempolicy *pol = NULL; | |
1745 | struct sp_node *sn; | |
1746 | ||
1747 | if (!sp->root.rb_node) | |
1748 | return NULL; | |
1749 | spin_lock(&sp->lock); | |
1750 | sn = sp_lookup(sp, idx, idx+1); | |
1751 | if (sn) { | |
1752 | mpol_get(sn->policy); | |
1753 | pol = sn->policy; | |
1754 | } | |
1755 | spin_unlock(&sp->lock); | |
1756 | return pol; | |
1757 | } | |
1758 | ||
1759 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1760 | { | |
140d5a49 | 1761 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 1762 | rb_erase(&n->nd, &sp->root); |
f0be3d32 | 1763 | mpol_put(n->policy); |
1da177e4 LT |
1764 | kmem_cache_free(sn_cache, n); |
1765 | } | |
1766 | ||
dbcb0f19 AB |
1767 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
1768 | struct mempolicy *pol) | |
1da177e4 LT |
1769 | { |
1770 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1771 | ||
1772 | if (!n) | |
1773 | return NULL; | |
1774 | n->start = start; | |
1775 | n->end = end; | |
1776 | mpol_get(pol); | |
aab0b102 | 1777 | pol->flags |= MPOL_F_SHARED; /* for unref */ |
1da177e4 LT |
1778 | n->policy = pol; |
1779 | return n; | |
1780 | } | |
1781 | ||
1782 | /* Replace a policy range. */ | |
1783 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1784 | unsigned long end, struct sp_node *new) | |
1785 | { | |
1786 | struct sp_node *n, *new2 = NULL; | |
1787 | ||
1788 | restart: | |
1789 | spin_lock(&sp->lock); | |
1790 | n = sp_lookup(sp, start, end); | |
1791 | /* Take care of old policies in the same range. */ | |
1792 | while (n && n->start < end) { | |
1793 | struct rb_node *next = rb_next(&n->nd); | |
1794 | if (n->start >= start) { | |
1795 | if (n->end <= end) | |
1796 | sp_delete(sp, n); | |
1797 | else | |
1798 | n->start = end; | |
1799 | } else { | |
1800 | /* Old policy spanning whole new range. */ | |
1801 | if (n->end > end) { | |
1802 | if (!new2) { | |
1803 | spin_unlock(&sp->lock); | |
1804 | new2 = sp_alloc(end, n->end, n->policy); | |
1805 | if (!new2) | |
1806 | return -ENOMEM; | |
1807 | goto restart; | |
1808 | } | |
1809 | n->end = start; | |
1810 | sp_insert(sp, new2); | |
1811 | new2 = NULL; | |
1812 | break; | |
1813 | } else | |
1814 | n->end = start; | |
1815 | } | |
1816 | if (!next) | |
1817 | break; | |
1818 | n = rb_entry(next, struct sp_node, nd); | |
1819 | } | |
1820 | if (new) | |
1821 | sp_insert(sp, new); | |
1822 | spin_unlock(&sp->lock); | |
1823 | if (new2) { | |
f0be3d32 | 1824 | mpol_put(new2->policy); |
1da177e4 LT |
1825 | kmem_cache_free(sn_cache, new2); |
1826 | } | |
1827 | return 0; | |
1828 | } | |
1829 | ||
a3b51e01 | 1830 | void mpol_shared_policy_init(struct shared_policy *info, unsigned short policy, |
028fec41 | 1831 | unsigned short flags, nodemask_t *policy_nodes) |
7339ff83 RH |
1832 | { |
1833 | info->root = RB_ROOT; | |
1834 | spin_lock_init(&info->lock); | |
1835 | ||
1836 | if (policy != MPOL_DEFAULT) { | |
1837 | struct mempolicy *newpol; | |
1838 | ||
bea904d5 | 1839 | /* Falls back to NULL policy [MPOL_DEFAULT] on any error */ |
028fec41 | 1840 | newpol = mpol_new(policy, flags, policy_nodes); |
7339ff83 RH |
1841 | if (!IS_ERR(newpol)) { |
1842 | /* Create pseudo-vma that contains just the policy */ | |
1843 | struct vm_area_struct pvma; | |
1844 | ||
1845 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1846 | /* Policy covers entire file */ | |
1847 | pvma.vm_end = TASK_SIZE; | |
1848 | mpol_set_shared_policy(info, &pvma, newpol); | |
f0be3d32 | 1849 | mpol_put(newpol); |
7339ff83 RH |
1850 | } |
1851 | } | |
1852 | } | |
1853 | ||
1da177e4 LT |
1854 | int mpol_set_shared_policy(struct shared_policy *info, |
1855 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1856 | { | |
1857 | int err; | |
1858 | struct sp_node *new = NULL; | |
1859 | unsigned long sz = vma_pages(vma); | |
1860 | ||
028fec41 | 1861 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 1862 | vma->vm_pgoff, |
45c4745a | 1863 | sz, npol ? npol->mode : -1, |
028fec41 | 1864 | npol ? npol->flags : -1, |
140d5a49 | 1865 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1866 | |
1867 | if (npol) { | |
1868 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1869 | if (!new) | |
1870 | return -ENOMEM; | |
1871 | } | |
1872 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1873 | if (err && new) | |
1874 | kmem_cache_free(sn_cache, new); | |
1875 | return err; | |
1876 | } | |
1877 | ||
1878 | /* Free a backing policy store on inode delete. */ | |
1879 | void mpol_free_shared_policy(struct shared_policy *p) | |
1880 | { | |
1881 | struct sp_node *n; | |
1882 | struct rb_node *next; | |
1883 | ||
1884 | if (!p->root.rb_node) | |
1885 | return; | |
1886 | spin_lock(&p->lock); | |
1887 | next = rb_first(&p->root); | |
1888 | while (next) { | |
1889 | n = rb_entry(next, struct sp_node, nd); | |
1890 | next = rb_next(&n->nd); | |
90c5029e | 1891 | rb_erase(&n->nd, &p->root); |
f0be3d32 | 1892 | mpol_put(n->policy); |
1da177e4 LT |
1893 | kmem_cache_free(sn_cache, n); |
1894 | } | |
1895 | spin_unlock(&p->lock); | |
1da177e4 LT |
1896 | } |
1897 | ||
1898 | /* assumes fs == KERNEL_DS */ | |
1899 | void __init numa_policy_init(void) | |
1900 | { | |
b71636e2 PM |
1901 | nodemask_t interleave_nodes; |
1902 | unsigned long largest = 0; | |
1903 | int nid, prefer = 0; | |
1904 | ||
1da177e4 LT |
1905 | policy_cache = kmem_cache_create("numa_policy", |
1906 | sizeof(struct mempolicy), | |
20c2df83 | 1907 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
1908 | |
1909 | sn_cache = kmem_cache_create("shared_policy_node", | |
1910 | sizeof(struct sp_node), | |
20c2df83 | 1911 | 0, SLAB_PANIC, NULL); |
1da177e4 | 1912 | |
b71636e2 PM |
1913 | /* |
1914 | * Set interleaving policy for system init. Interleaving is only | |
1915 | * enabled across suitably sized nodes (default is >= 16MB), or | |
1916 | * fall back to the largest node if they're all smaller. | |
1917 | */ | |
1918 | nodes_clear(interleave_nodes); | |
56bbd65d | 1919 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
1920 | unsigned long total_pages = node_present_pages(nid); |
1921 | ||
1922 | /* Preserve the largest node */ | |
1923 | if (largest < total_pages) { | |
1924 | largest = total_pages; | |
1925 | prefer = nid; | |
1926 | } | |
1927 | ||
1928 | /* Interleave this node? */ | |
1929 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
1930 | node_set(nid, interleave_nodes); | |
1931 | } | |
1932 | ||
1933 | /* All too small, use the largest */ | |
1934 | if (unlikely(nodes_empty(interleave_nodes))) | |
1935 | node_set(prefer, interleave_nodes); | |
1da177e4 | 1936 | |
028fec41 | 1937 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
1938 | printk("numa_policy_init: interleaving failed\n"); |
1939 | } | |
1940 | ||
8bccd85f | 1941 | /* Reset policy of current process to default */ |
1da177e4 LT |
1942 | void numa_default_policy(void) |
1943 | { | |
028fec41 | 1944 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 1945 | } |
68860ec1 | 1946 | |
1a75a6c8 | 1947 | /* |
fc36b8d3 | 1948 | * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag |
53f2556b | 1949 | * Used only for mpol_to_str() |
1a75a6c8 | 1950 | */ |
53f2556b | 1951 | #define MPOL_LOCAL (MPOL_INTERLEAVE + 1) |
15ad7cdc | 1952 | static const char * const policy_types[] = |
53f2556b | 1953 | { "default", "prefer", "bind", "interleave", "local" }; |
1a75a6c8 CL |
1954 | |
1955 | /* | |
1956 | * Convert a mempolicy into a string. | |
1957 | * Returns the number of characters in buffer (if positive) | |
1958 | * or an error (negative) | |
1959 | */ | |
1960 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1961 | { | |
1962 | char *p = buffer; | |
1963 | int l; | |
1964 | nodemask_t nodes; | |
bea904d5 | 1965 | unsigned short mode; |
f5b087b5 | 1966 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 1967 | |
bea904d5 LS |
1968 | if (!pol || pol == &default_policy) |
1969 | mode = MPOL_DEFAULT; | |
1970 | else | |
1971 | mode = pol->mode; | |
1972 | ||
1a75a6c8 CL |
1973 | switch (mode) { |
1974 | case MPOL_DEFAULT: | |
1975 | nodes_clear(nodes); | |
1976 | break; | |
1977 | ||
1978 | case MPOL_PREFERRED: | |
1979 | nodes_clear(nodes); | |
fc36b8d3 | 1980 | if (flags & MPOL_F_LOCAL) |
53f2556b LS |
1981 | mode = MPOL_LOCAL; /* pseudo-policy */ |
1982 | else | |
fc36b8d3 | 1983 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
1984 | break; |
1985 | ||
1986 | case MPOL_BIND: | |
19770b32 | 1987 | /* Fall through */ |
1a75a6c8 CL |
1988 | case MPOL_INTERLEAVE: |
1989 | nodes = pol->v.nodes; | |
1990 | break; | |
1991 | ||
1992 | default: | |
1993 | BUG(); | |
1994 | return -EFAULT; | |
1995 | } | |
1996 | ||
1997 | l = strlen(policy_types[mode]); | |
53f2556b LS |
1998 | if (buffer + maxlen < p + l + 1) |
1999 | return -ENOSPC; | |
1a75a6c8 CL |
2000 | |
2001 | strcpy(p, policy_types[mode]); | |
2002 | p += l; | |
2003 | ||
fc36b8d3 | 2004 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2005 | int need_bar = 0; |
2006 | ||
2007 | if (buffer + maxlen < p + 2) | |
2008 | return -ENOSPC; | |
2009 | *p++ = '='; | |
2010 | ||
2011 | if (flags & MPOL_F_STATIC_NODES) | |
2012 | p += sprintf(p, "%sstatic", need_bar++ ? "|" : ""); | |
4c50bc01 DR |
2013 | if (flags & MPOL_F_RELATIVE_NODES) |
2014 | p += sprintf(p, "%srelative", need_bar++ ? "|" : ""); | |
f5b087b5 DR |
2015 | } |
2016 | ||
1a75a6c8 CL |
2017 | if (!nodes_empty(nodes)) { |
2018 | if (buffer + maxlen < p + 2) | |
2019 | return -ENOSPC; | |
2020 | *p++ = '='; | |
2021 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
2022 | } | |
2023 | return p - buffer; | |
2024 | } | |
2025 | ||
2026 | struct numa_maps { | |
2027 | unsigned long pages; | |
2028 | unsigned long anon; | |
397874df CL |
2029 | unsigned long active; |
2030 | unsigned long writeback; | |
1a75a6c8 | 2031 | unsigned long mapcount_max; |
397874df CL |
2032 | unsigned long dirty; |
2033 | unsigned long swapcache; | |
1a75a6c8 CL |
2034 | unsigned long node[MAX_NUMNODES]; |
2035 | }; | |
2036 | ||
397874df | 2037 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
2038 | { |
2039 | struct numa_maps *md = private; | |
2040 | int count = page_mapcount(page); | |
2041 | ||
397874df CL |
2042 | md->pages++; |
2043 | if (pte_dirty || PageDirty(page)) | |
2044 | md->dirty++; | |
1a75a6c8 | 2045 | |
397874df CL |
2046 | if (PageSwapCache(page)) |
2047 | md->swapcache++; | |
1a75a6c8 | 2048 | |
397874df CL |
2049 | if (PageActive(page)) |
2050 | md->active++; | |
2051 | ||
2052 | if (PageWriteback(page)) | |
2053 | md->writeback++; | |
1a75a6c8 CL |
2054 | |
2055 | if (PageAnon(page)) | |
2056 | md->anon++; | |
2057 | ||
397874df CL |
2058 | if (count > md->mapcount_max) |
2059 | md->mapcount_max = count; | |
2060 | ||
1a75a6c8 | 2061 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
2062 | } |
2063 | ||
7f709ed0 | 2064 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
2065 | static void check_huge_range(struct vm_area_struct *vma, |
2066 | unsigned long start, unsigned long end, | |
2067 | struct numa_maps *md) | |
2068 | { | |
2069 | unsigned long addr; | |
2070 | struct page *page; | |
2071 | ||
2072 | for (addr = start; addr < end; addr += HPAGE_SIZE) { | |
2073 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); | |
2074 | pte_t pte; | |
2075 | ||
2076 | if (!ptep) | |
2077 | continue; | |
2078 | ||
2079 | pte = *ptep; | |
2080 | if (pte_none(pte)) | |
2081 | continue; | |
2082 | ||
2083 | page = pte_page(pte); | |
2084 | if (!page) | |
2085 | continue; | |
2086 | ||
2087 | gather_stats(page, md, pte_dirty(*ptep)); | |
2088 | } | |
2089 | } | |
7f709ed0 AM |
2090 | #else |
2091 | static inline void check_huge_range(struct vm_area_struct *vma, | |
2092 | unsigned long start, unsigned long end, | |
2093 | struct numa_maps *md) | |
2094 | { | |
2095 | } | |
2096 | #endif | |
397874df | 2097 | |
53f2556b LS |
2098 | /* |
2099 | * Display pages allocated per node and memory policy via /proc. | |
2100 | */ | |
1a75a6c8 CL |
2101 | int show_numa_map(struct seq_file *m, void *v) |
2102 | { | |
99f89551 | 2103 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
2104 | struct vm_area_struct *vma = v; |
2105 | struct numa_maps *md; | |
397874df CL |
2106 | struct file *file = vma->vm_file; |
2107 | struct mm_struct *mm = vma->vm_mm; | |
480eccf9 | 2108 | struct mempolicy *pol; |
1a75a6c8 CL |
2109 | int n; |
2110 | char buffer[50]; | |
2111 | ||
397874df | 2112 | if (!mm) |
1a75a6c8 CL |
2113 | return 0; |
2114 | ||
2115 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
2116 | if (!md) | |
2117 | return 0; | |
2118 | ||
480eccf9 LS |
2119 | pol = get_vma_policy(priv->task, vma, vma->vm_start); |
2120 | mpol_to_str(buffer, sizeof(buffer), pol); | |
52cd3b07 | 2121 | mpol_cond_put(pol); |
397874df CL |
2122 | |
2123 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
2124 | ||
2125 | if (file) { | |
2126 | seq_printf(m, " file="); | |
c32c2f63 | 2127 | seq_path(m, &file->f_path, "\n\t= "); |
397874df CL |
2128 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
2129 | seq_printf(m, " heap"); | |
2130 | } else if (vma->vm_start <= mm->start_stack && | |
2131 | vma->vm_end >= mm->start_stack) { | |
2132 | seq_printf(m, " stack"); | |
2133 | } | |
2134 | ||
2135 | if (is_vm_hugetlb_page(vma)) { | |
2136 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
2137 | seq_printf(m, " huge"); | |
2138 | } else { | |
a57ebfdb | 2139 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
56bbd65d | 2140 | &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md); |
397874df CL |
2141 | } |
2142 | ||
2143 | if (!md->pages) | |
2144 | goto out; | |
1a75a6c8 | 2145 | |
397874df CL |
2146 | if (md->anon) |
2147 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 2148 | |
397874df CL |
2149 | if (md->dirty) |
2150 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 2151 | |
397874df CL |
2152 | if (md->pages != md->anon && md->pages != md->dirty) |
2153 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 2154 | |
397874df CL |
2155 | if (md->mapcount_max > 1) |
2156 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 2157 | |
397874df CL |
2158 | if (md->swapcache) |
2159 | seq_printf(m," swapcache=%lu", md->swapcache); | |
2160 | ||
2161 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
2162 | seq_printf(m," active=%lu", md->active); | |
2163 | ||
2164 | if (md->writeback) | |
2165 | seq_printf(m," writeback=%lu", md->writeback); | |
2166 | ||
56bbd65d | 2167 | for_each_node_state(n, N_HIGH_MEMORY) |
397874df CL |
2168 | if (md->node[n]) |
2169 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
2170 | out: | |
2171 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
2172 | kfree(md); |
2173 | ||
2174 | if (m->count < m->size) | |
99f89551 | 2175 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
2176 | return 0; |
2177 | } |