]>
Commit | Line | Data |
---|---|---|
46aeb7e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Simple NUMA memory policy for the Linux kernel. | |
4 | * | |
5 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 6 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
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 | 28 | * preferred Try a specific node first before normal fallback. |
00ef2d2f | 29 | * As a special case NUMA_NO_NODE here means do the allocation |
1da177e4 LT |
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 | ||
b1de0d13 MH |
68 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
69 | ||
1da177e4 | 70 | #include <linux/mempolicy.h> |
a520110e | 71 | #include <linux/pagewalk.h> |
1da177e4 LT |
72 | #include <linux/highmem.h> |
73 | #include <linux/hugetlb.h> | |
74 | #include <linux/kernel.h> | |
75 | #include <linux/sched.h> | |
6e84f315 | 76 | #include <linux/sched/mm.h> |
6a3827d7 | 77 | #include <linux/sched/numa_balancing.h> |
f719ff9b | 78 | #include <linux/sched/task.h> |
1da177e4 LT |
79 | #include <linux/nodemask.h> |
80 | #include <linux/cpuset.h> | |
1da177e4 LT |
81 | #include <linux/slab.h> |
82 | #include <linux/string.h> | |
b95f1b31 | 83 | #include <linux/export.h> |
b488893a | 84 | #include <linux/nsproxy.h> |
1da177e4 LT |
85 | #include <linux/interrupt.h> |
86 | #include <linux/init.h> | |
87 | #include <linux/compat.h> | |
31367466 | 88 | #include <linux/ptrace.h> |
dc9aa5b9 | 89 | #include <linux/swap.h> |
1a75a6c8 CL |
90 | #include <linux/seq_file.h> |
91 | #include <linux/proc_fs.h> | |
b20a3503 | 92 | #include <linux/migrate.h> |
62b61f61 | 93 | #include <linux/ksm.h> |
95a402c3 | 94 | #include <linux/rmap.h> |
86c3a764 | 95 | #include <linux/security.h> |
dbcb0f19 | 96 | #include <linux/syscalls.h> |
095f1fc4 | 97 | #include <linux/ctype.h> |
6d9c285a | 98 | #include <linux/mm_inline.h> |
b24f53a0 | 99 | #include <linux/mmu_notifier.h> |
b1de0d13 | 100 | #include <linux/printk.h> |
c8633798 | 101 | #include <linux/swapops.h> |
dc9aa5b9 | 102 | |
1da177e4 | 103 | #include <asm/tlbflush.h> |
7c0f6ba6 | 104 | #include <linux/uaccess.h> |
1da177e4 | 105 | |
62695a84 NP |
106 | #include "internal.h" |
107 | ||
38e35860 | 108 | /* Internal flags */ |
dc9aa5b9 | 109 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 110 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 111 | |
fcc234f8 PE |
112 | static struct kmem_cache *policy_cache; |
113 | static struct kmem_cache *sn_cache; | |
1da177e4 | 114 | |
1da177e4 LT |
115 | /* Highest zone. An specific allocation for a zone below that is not |
116 | policied. */ | |
6267276f | 117 | enum zone_type policy_zone = 0; |
1da177e4 | 118 | |
bea904d5 LS |
119 | /* |
120 | * run-time system-wide default policy => local allocation | |
121 | */ | |
e754d79d | 122 | static struct mempolicy default_policy = { |
1da177e4 | 123 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 124 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 125 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
126 | }; |
127 | ||
5606e387 MG |
128 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
129 | ||
74d2c3a0 | 130 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
131 | { |
132 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 133 | int node; |
5606e387 | 134 | |
f15ca78e ON |
135 | if (pol) |
136 | return pol; | |
5606e387 | 137 | |
f15ca78e ON |
138 | node = numa_node_id(); |
139 | if (node != NUMA_NO_NODE) { | |
140 | pol = &preferred_node_policy[node]; | |
141 | /* preferred_node_policy is not initialised early in boot */ | |
142 | if (pol->mode) | |
143 | return pol; | |
5606e387 MG |
144 | } |
145 | ||
f15ca78e | 146 | return &default_policy; |
5606e387 MG |
147 | } |
148 | ||
37012946 DR |
149 | static const struct mempolicy_operations { |
150 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 151 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
152 | } mpol_ops[MPOL_MAX]; |
153 | ||
f5b087b5 DR |
154 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
155 | { | |
6d556294 | 156 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
157 | } |
158 | ||
159 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
160 | const nodemask_t *rel) | |
161 | { | |
162 | nodemask_t tmp; | |
163 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
164 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
165 | } |
166 | ||
37012946 DR |
167 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
168 | { | |
169 | if (nodes_empty(*nodes)) | |
170 | return -EINVAL; | |
171 | pol->v.nodes = *nodes; | |
172 | return 0; | |
173 | } | |
174 | ||
175 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
176 | { | |
177 | if (!nodes) | |
fc36b8d3 | 178 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
179 | else if (nodes_empty(*nodes)) |
180 | return -EINVAL; /* no allowed nodes */ | |
181 | else | |
182 | pol->v.preferred_node = first_node(*nodes); | |
183 | return 0; | |
184 | } | |
185 | ||
186 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
187 | { | |
859f7ef1 | 188 | if (nodes_empty(*nodes)) |
37012946 DR |
189 | return -EINVAL; |
190 | pol->v.nodes = *nodes; | |
191 | return 0; | |
192 | } | |
193 | ||
58568d2a MX |
194 | /* |
195 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
196 | * any, for the new policy. mpol_new() has already validated the nodes | |
197 | * parameter with respect to the policy mode and flags. But, we need to | |
198 | * handle an empty nodemask with MPOL_PREFERRED here. | |
199 | * | |
200 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
201 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
202 | */ | |
4bfc4495 KH |
203 | static int mpol_set_nodemask(struct mempolicy *pol, |
204 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 205 | { |
58568d2a MX |
206 | int ret; |
207 | ||
208 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
209 | if (pol == NULL) | |
210 | return 0; | |
01f13bd6 | 211 | /* Check N_MEMORY */ |
4bfc4495 | 212 | nodes_and(nsc->mask1, |
01f13bd6 | 213 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
214 | |
215 | VM_BUG_ON(!nodes); | |
216 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
217 | nodes = NULL; /* explicit local allocation */ | |
218 | else { | |
219 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
859f7ef1 | 220 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); |
58568d2a | 221 | else |
4bfc4495 KH |
222 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
223 | ||
58568d2a MX |
224 | if (mpol_store_user_nodemask(pol)) |
225 | pol->w.user_nodemask = *nodes; | |
226 | else | |
227 | pol->w.cpuset_mems_allowed = | |
228 | cpuset_current_mems_allowed; | |
229 | } | |
230 | ||
4bfc4495 KH |
231 | if (nodes) |
232 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
233 | else | |
234 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
235 | return ret; |
236 | } | |
237 | ||
238 | /* | |
239 | * This function just creates a new policy, does some check and simple | |
240 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
241 | */ | |
028fec41 DR |
242 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
243 | nodemask_t *nodes) | |
1da177e4 LT |
244 | { |
245 | struct mempolicy *policy; | |
246 | ||
028fec41 | 247 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 248 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 249 | |
3e1f0645 DR |
250 | if (mode == MPOL_DEFAULT) { |
251 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 252 | return ERR_PTR(-EINVAL); |
d3a71033 | 253 | return NULL; |
37012946 | 254 | } |
3e1f0645 DR |
255 | VM_BUG_ON(!nodes); |
256 | ||
257 | /* | |
258 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
259 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
260 | * All other modes require a valid pointer to a non-empty nodemask. | |
261 | */ | |
262 | if (mode == MPOL_PREFERRED) { | |
263 | if (nodes_empty(*nodes)) { | |
264 | if (((flags & MPOL_F_STATIC_NODES) || | |
265 | (flags & MPOL_F_RELATIVE_NODES))) | |
266 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 267 | } |
479e2802 | 268 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
269 | if (!nodes_empty(*nodes) || |
270 | (flags & MPOL_F_STATIC_NODES) || | |
271 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 PZ |
272 | return ERR_PTR(-EINVAL); |
273 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
274 | } else if (nodes_empty(*nodes)) |
275 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
276 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
277 | if (!policy) | |
278 | return ERR_PTR(-ENOMEM); | |
279 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 280 | policy->mode = mode; |
3e1f0645 | 281 | policy->flags = flags; |
37012946 | 282 | |
1da177e4 | 283 | return policy; |
37012946 DR |
284 | } |
285 | ||
52cd3b07 LS |
286 | /* Slow path of a mpol destructor. */ |
287 | void __mpol_put(struct mempolicy *p) | |
288 | { | |
289 | if (!atomic_dec_and_test(&p->refcnt)) | |
290 | return; | |
52cd3b07 LS |
291 | kmem_cache_free(policy_cache, p); |
292 | } | |
293 | ||
213980c0 | 294 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
295 | { |
296 | } | |
297 | ||
213980c0 | 298 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
299 | { |
300 | nodemask_t tmp; | |
301 | ||
302 | if (pol->flags & MPOL_F_STATIC_NODES) | |
303 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
304 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
305 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
306 | else { | |
213980c0 VB |
307 | nodes_remap(tmp, pol->v.nodes,pol->w.cpuset_mems_allowed, |
308 | *nodes); | |
29b190fa | 309 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 310 | } |
f5b087b5 | 311 | |
708c1bbc MX |
312 | if (nodes_empty(tmp)) |
313 | tmp = *nodes; | |
314 | ||
213980c0 | 315 | pol->v.nodes = tmp; |
37012946 DR |
316 | } |
317 | ||
318 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 319 | const nodemask_t *nodes) |
37012946 DR |
320 | { |
321 | nodemask_t tmp; | |
322 | ||
37012946 DR |
323 | if (pol->flags & MPOL_F_STATIC_NODES) { |
324 | int node = first_node(pol->w.user_nodemask); | |
325 | ||
fc36b8d3 | 326 | if (node_isset(node, *nodes)) { |
37012946 | 327 | pol->v.preferred_node = node; |
fc36b8d3 LS |
328 | pol->flags &= ~MPOL_F_LOCAL; |
329 | } else | |
330 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
331 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
332 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
333 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 334 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
335 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
336 | pol->w.cpuset_mems_allowed, | |
337 | *nodes); | |
338 | pol->w.cpuset_mems_allowed = *nodes; | |
339 | } | |
1da177e4 LT |
340 | } |
341 | ||
708c1bbc MX |
342 | /* |
343 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
344 | * | |
213980c0 VB |
345 | * Per-vma policies are protected by mmap_sem. Allocations using per-task |
346 | * policies are protected by task->mems_allowed_seq to prevent a premature | |
347 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 348 | */ |
213980c0 | 349 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 350 | { |
1d0d2680 DR |
351 | if (!pol) |
352 | return; | |
2e25644e | 353 | if (!mpol_store_user_nodemask(pol) && !(pol->flags & MPOL_F_LOCAL) && |
1d0d2680 DR |
354 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
355 | return; | |
708c1bbc | 356 | |
213980c0 | 357 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
358 | } |
359 | ||
360 | /* | |
361 | * Wrapper for mpol_rebind_policy() that just requires task | |
362 | * pointer, and updates task mempolicy. | |
58568d2a MX |
363 | * |
364 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
365 | */ |
366 | ||
213980c0 | 367 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 368 | { |
213980c0 | 369 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
370 | } |
371 | ||
372 | /* | |
373 | * Rebind each vma in mm to new nodemask. | |
374 | * | |
375 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
376 | */ | |
377 | ||
378 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
379 | { | |
380 | struct vm_area_struct *vma; | |
381 | ||
382 | down_write(&mm->mmap_sem); | |
383 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
213980c0 | 384 | mpol_rebind_policy(vma->vm_policy, new); |
1d0d2680 DR |
385 | up_write(&mm->mmap_sem); |
386 | } | |
387 | ||
37012946 DR |
388 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
389 | [MPOL_DEFAULT] = { | |
390 | .rebind = mpol_rebind_default, | |
391 | }, | |
392 | [MPOL_INTERLEAVE] = { | |
393 | .create = mpol_new_interleave, | |
394 | .rebind = mpol_rebind_nodemask, | |
395 | }, | |
396 | [MPOL_PREFERRED] = { | |
397 | .create = mpol_new_preferred, | |
398 | .rebind = mpol_rebind_preferred, | |
399 | }, | |
400 | [MPOL_BIND] = { | |
401 | .create = mpol_new_bind, | |
402 | .rebind = mpol_rebind_nodemask, | |
403 | }, | |
404 | }; | |
405 | ||
a53190a4 | 406 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 407 | unsigned long flags); |
1a75a6c8 | 408 | |
6f4576e3 NH |
409 | struct queue_pages { |
410 | struct list_head *pagelist; | |
411 | unsigned long flags; | |
412 | nodemask_t *nmask; | |
413 | struct vm_area_struct *prev; | |
414 | }; | |
415 | ||
88aaa2a1 NH |
416 | /* |
417 | * Check if the page's nid is in qp->nmask. | |
418 | * | |
419 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
420 | * in the invert of qp->nmask. | |
421 | */ | |
422 | static inline bool queue_pages_required(struct page *page, | |
423 | struct queue_pages *qp) | |
424 | { | |
425 | int nid = page_to_nid(page); | |
426 | unsigned long flags = qp->flags; | |
427 | ||
428 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
429 | } | |
430 | ||
a7f40cfe | 431 | /* |
d8835445 YS |
432 | * queue_pages_pmd() has four possible return values: |
433 | * 0 - pages are placed on the right node or queued successfully. | |
434 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
435 | * specified. | |
436 | * 2 - THP was split. | |
437 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an | |
438 | * existing page was already on a node that does not follow the | |
439 | * policy. | |
a7f40cfe | 440 | */ |
c8633798 NH |
441 | static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
442 | unsigned long end, struct mm_walk *walk) | |
443 | { | |
444 | int ret = 0; | |
445 | struct page *page; | |
446 | struct queue_pages *qp = walk->private; | |
447 | unsigned long flags; | |
448 | ||
449 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 450 | ret = -EIO; |
c8633798 NH |
451 | goto unlock; |
452 | } | |
453 | page = pmd_page(*pmd); | |
454 | if (is_huge_zero_page(page)) { | |
455 | spin_unlock(ptl); | |
456 | __split_huge_pmd(walk->vma, pmd, addr, false, NULL); | |
d8835445 | 457 | ret = 2; |
c8633798 NH |
458 | goto out; |
459 | } | |
d8835445 | 460 | if (!queue_pages_required(page, qp)) |
c8633798 | 461 | goto unlock; |
c8633798 | 462 | |
c8633798 NH |
463 | flags = qp->flags; |
464 | /* go to thp migration */ | |
a7f40cfe | 465 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 YS |
466 | if (!vma_migratable(walk->vma) || |
467 | migrate_page_add(page, qp->pagelist, flags)) { | |
d8835445 | 468 | ret = 1; |
a7f40cfe YS |
469 | goto unlock; |
470 | } | |
a7f40cfe YS |
471 | } else |
472 | ret = -EIO; | |
c8633798 NH |
473 | unlock: |
474 | spin_unlock(ptl); | |
475 | out: | |
476 | return ret; | |
477 | } | |
478 | ||
98094945 NH |
479 | /* |
480 | * Scan through pages checking if pages follow certain conditions, | |
481 | * and move them to the pagelist if they do. | |
d8835445 YS |
482 | * |
483 | * queue_pages_pte_range() has three possible return values: | |
484 | * 0 - pages are placed on the right node or queued successfully. | |
485 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
486 | * specified. | |
487 | * -EIO - only MPOL_MF_STRICT was specified and an existing page was already | |
488 | * on a node that does not follow the policy. | |
98094945 | 489 | */ |
6f4576e3 NH |
490 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
491 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 492 | { |
6f4576e3 NH |
493 | struct vm_area_struct *vma = walk->vma; |
494 | struct page *page; | |
495 | struct queue_pages *qp = walk->private; | |
496 | unsigned long flags = qp->flags; | |
c8633798 | 497 | int ret; |
d8835445 | 498 | bool has_unmovable = false; |
91612e0d | 499 | pte_t *pte; |
705e87c0 | 500 | spinlock_t *ptl; |
941150a3 | 501 | |
c8633798 NH |
502 | ptl = pmd_trans_huge_lock(pmd, vma); |
503 | if (ptl) { | |
504 | ret = queue_pages_pmd(pmd, ptl, addr, end, walk); | |
d8835445 | 505 | if (ret != 2) |
a7f40cfe | 506 | return ret; |
248db92d | 507 | } |
d8835445 | 508 | /* THP was split, fall through to pte walk */ |
91612e0d | 509 | |
337d9abf NH |
510 | if (pmd_trans_unstable(pmd)) |
511 | return 0; | |
94723aaf | 512 | |
6f4576e3 NH |
513 | pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
514 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
91612e0d | 515 | if (!pte_present(*pte)) |
1da177e4 | 516 | continue; |
6aab341e LT |
517 | page = vm_normal_page(vma, addr, *pte); |
518 | if (!page) | |
1da177e4 | 519 | continue; |
053837fc | 520 | /* |
62b61f61 HD |
521 | * vm_normal_page() filters out zero pages, but there might |
522 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 523 | */ |
b79bc0a0 | 524 | if (PageReserved(page)) |
f4598c8b | 525 | continue; |
88aaa2a1 | 526 | if (!queue_pages_required(page, qp)) |
38e35860 | 527 | continue; |
a7f40cfe | 528 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
d8835445 YS |
529 | /* MPOL_MF_STRICT must be specified if we get here */ |
530 | if (!vma_migratable(vma)) { | |
531 | has_unmovable = true; | |
a7f40cfe | 532 | break; |
d8835445 | 533 | } |
a53190a4 YS |
534 | |
535 | /* | |
536 | * Do not abort immediately since there may be | |
537 | * temporary off LRU pages in the range. Still | |
538 | * need migrate other LRU pages. | |
539 | */ | |
540 | if (migrate_page_add(page, qp->pagelist, flags)) | |
541 | has_unmovable = true; | |
a7f40cfe YS |
542 | } else |
543 | break; | |
6f4576e3 NH |
544 | } |
545 | pte_unmap_unlock(pte - 1, ptl); | |
546 | cond_resched(); | |
d8835445 YS |
547 | |
548 | if (has_unmovable) | |
549 | return 1; | |
550 | ||
a7f40cfe | 551 | return addr != end ? -EIO : 0; |
91612e0d HD |
552 | } |
553 | ||
6f4576e3 NH |
554 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
555 | unsigned long addr, unsigned long end, | |
556 | struct mm_walk *walk) | |
e2d8cf40 NH |
557 | { |
558 | #ifdef CONFIG_HUGETLB_PAGE | |
6f4576e3 NH |
559 | struct queue_pages *qp = walk->private; |
560 | unsigned long flags = qp->flags; | |
e2d8cf40 | 561 | struct page *page; |
cb900f41 | 562 | spinlock_t *ptl; |
d4c54919 | 563 | pte_t entry; |
e2d8cf40 | 564 | |
6f4576e3 NH |
565 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
566 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
567 | if (!pte_present(entry)) |
568 | goto unlock; | |
569 | page = pte_page(entry); | |
88aaa2a1 | 570 | if (!queue_pages_required(page, qp)) |
e2d8cf40 NH |
571 | goto unlock; |
572 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ | |
573 | if (flags & (MPOL_MF_MOVE_ALL) || | |
574 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) | |
6f4576e3 | 575 | isolate_huge_page(page, qp->pagelist); |
e2d8cf40 | 576 | unlock: |
cb900f41 | 577 | spin_unlock(ptl); |
e2d8cf40 NH |
578 | #else |
579 | BUG(); | |
580 | #endif | |
91612e0d | 581 | return 0; |
1da177e4 LT |
582 | } |
583 | ||
5877231f | 584 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 585 | /* |
4b10e7d5 MG |
586 | * This is used to mark a range of virtual addresses to be inaccessible. |
587 | * These are later cleared by a NUMA hinting fault. Depending on these | |
588 | * faults, pages may be migrated for better NUMA placement. | |
589 | * | |
590 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
591 | * an architecture makes a different choice, it will need further | |
592 | * changes to the core. | |
b24f53a0 | 593 | */ |
4b10e7d5 MG |
594 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
595 | unsigned long addr, unsigned long end) | |
b24f53a0 | 596 | { |
4b10e7d5 | 597 | int nr_updated; |
b24f53a0 | 598 | |
4d942466 | 599 | nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1); |
03c5a6e1 MG |
600 | if (nr_updated) |
601 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 602 | |
4b10e7d5 | 603 | return nr_updated; |
b24f53a0 LS |
604 | } |
605 | #else | |
606 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
607 | unsigned long addr, unsigned long end) | |
608 | { | |
609 | return 0; | |
610 | } | |
5877231f | 611 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 612 | |
6f4576e3 NH |
613 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
614 | struct mm_walk *walk) | |
615 | { | |
616 | struct vm_area_struct *vma = walk->vma; | |
617 | struct queue_pages *qp = walk->private; | |
618 | unsigned long endvma = vma->vm_end; | |
619 | unsigned long flags = qp->flags; | |
620 | ||
a7f40cfe YS |
621 | /* |
622 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
623 | * regardless of vma_migratable | |
624 | */ | |
625 | if (!vma_migratable(vma) && | |
626 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
627 | return 1; |
628 | ||
6f4576e3 NH |
629 | if (endvma > end) |
630 | endvma = end; | |
631 | if (vma->vm_start > start) | |
632 | start = vma->vm_start; | |
633 | ||
634 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { | |
635 | if (!vma->vm_next && vma->vm_end < end) | |
636 | return -EFAULT; | |
637 | if (qp->prev && qp->prev->vm_end < vma->vm_start) | |
638 | return -EFAULT; | |
639 | } | |
640 | ||
641 | qp->prev = vma; | |
642 | ||
6f4576e3 NH |
643 | if (flags & MPOL_MF_LAZY) { |
644 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
4355c018 LC |
645 | if (!is_vm_hugetlb_page(vma) && |
646 | (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) && | |
647 | !(vma->vm_flags & VM_MIXEDMAP)) | |
6f4576e3 NH |
648 | change_prot_numa(vma, start, endvma); |
649 | return 1; | |
650 | } | |
651 | ||
77bf45e7 | 652 | /* queue pages from current vma */ |
a7f40cfe | 653 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
654 | return 0; |
655 | return 1; | |
656 | } | |
657 | ||
7b86ac33 CH |
658 | static const struct mm_walk_ops queue_pages_walk_ops = { |
659 | .hugetlb_entry = queue_pages_hugetlb, | |
660 | .pmd_entry = queue_pages_pte_range, | |
661 | .test_walk = queue_pages_test_walk, | |
662 | }; | |
663 | ||
dc9aa5b9 | 664 | /* |
98094945 NH |
665 | * Walk through page tables and collect pages to be migrated. |
666 | * | |
667 | * If pages found in a given range are on a set of nodes (determined by | |
668 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
669 | * passed via @private. |
670 | * | |
671 | * queue_pages_range() has three possible return values: | |
672 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
673 | * specified. | |
674 | * 0 - queue pages successfully or no misplaced page. | |
675 | * -EIO - there is misplaced page and only MPOL_MF_STRICT was specified. | |
dc9aa5b9 | 676 | */ |
d05f0cdc | 677 | static int |
98094945 | 678 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
679 | nodemask_t *nodes, unsigned long flags, |
680 | struct list_head *pagelist) | |
1da177e4 | 681 | { |
6f4576e3 NH |
682 | struct queue_pages qp = { |
683 | .pagelist = pagelist, | |
684 | .flags = flags, | |
685 | .nmask = nodes, | |
686 | .prev = NULL, | |
687 | }; | |
6f4576e3 | 688 | |
7b86ac33 | 689 | return walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp); |
1da177e4 LT |
690 | } |
691 | ||
869833f2 KM |
692 | /* |
693 | * Apply policy to a single VMA | |
694 | * This must be called with the mmap_sem held for writing. | |
695 | */ | |
696 | static int vma_replace_policy(struct vm_area_struct *vma, | |
697 | struct mempolicy *pol) | |
8d34694c | 698 | { |
869833f2 KM |
699 | int err; |
700 | struct mempolicy *old; | |
701 | struct mempolicy *new; | |
8d34694c KM |
702 | |
703 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
704 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
705 | vma->vm_ops, vma->vm_file, | |
706 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
707 | ||
869833f2 KM |
708 | new = mpol_dup(pol); |
709 | if (IS_ERR(new)) | |
710 | return PTR_ERR(new); | |
711 | ||
712 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 713 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
714 | if (err) |
715 | goto err_out; | |
8d34694c | 716 | } |
869833f2 KM |
717 | |
718 | old = vma->vm_policy; | |
719 | vma->vm_policy = new; /* protected by mmap_sem */ | |
720 | mpol_put(old); | |
721 | ||
722 | return 0; | |
723 | err_out: | |
724 | mpol_put(new); | |
8d34694c KM |
725 | return err; |
726 | } | |
727 | ||
1da177e4 | 728 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
729 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
730 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
731 | { |
732 | struct vm_area_struct *next; | |
9d8cebd4 KM |
733 | struct vm_area_struct *prev; |
734 | struct vm_area_struct *vma; | |
735 | int err = 0; | |
e26a5114 | 736 | pgoff_t pgoff; |
9d8cebd4 KM |
737 | unsigned long vmstart; |
738 | unsigned long vmend; | |
1da177e4 | 739 | |
097d5910 | 740 | vma = find_vma(mm, start); |
9d8cebd4 KM |
741 | if (!vma || vma->vm_start > start) |
742 | return -EFAULT; | |
743 | ||
097d5910 | 744 | prev = vma->vm_prev; |
e26a5114 KM |
745 | if (start > vma->vm_start) |
746 | prev = vma; | |
747 | ||
9d8cebd4 | 748 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 749 | next = vma->vm_next; |
9d8cebd4 KM |
750 | vmstart = max(start, vma->vm_start); |
751 | vmend = min(end, vma->vm_end); | |
752 | ||
e26a5114 KM |
753 | if (mpol_equal(vma_policy(vma), new_pol)) |
754 | continue; | |
755 | ||
756 | pgoff = vma->vm_pgoff + | |
757 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 758 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af AA |
759 | vma->anon_vma, vma->vm_file, pgoff, |
760 | new_pol, vma->vm_userfaultfd_ctx); | |
9d8cebd4 KM |
761 | if (prev) { |
762 | vma = prev; | |
763 | next = vma->vm_next; | |
3964acd0 ON |
764 | if (mpol_equal(vma_policy(vma), new_pol)) |
765 | continue; | |
766 | /* vma_merge() joined vma && vma->next, case 8 */ | |
767 | goto replace; | |
9d8cebd4 KM |
768 | } |
769 | if (vma->vm_start != vmstart) { | |
770 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
771 | if (err) | |
772 | goto out; | |
773 | } | |
774 | if (vma->vm_end != vmend) { | |
775 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
776 | if (err) | |
777 | goto out; | |
778 | } | |
3964acd0 | 779 | replace: |
869833f2 | 780 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
781 | if (err) |
782 | goto out; | |
1da177e4 | 783 | } |
9d8cebd4 KM |
784 | |
785 | out: | |
1da177e4 LT |
786 | return err; |
787 | } | |
788 | ||
1da177e4 | 789 | /* Set the process memory policy */ |
028fec41 DR |
790 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
791 | nodemask_t *nodes) | |
1da177e4 | 792 | { |
58568d2a | 793 | struct mempolicy *new, *old; |
4bfc4495 | 794 | NODEMASK_SCRATCH(scratch); |
58568d2a | 795 | int ret; |
1da177e4 | 796 | |
4bfc4495 KH |
797 | if (!scratch) |
798 | return -ENOMEM; | |
f4e53d91 | 799 | |
4bfc4495 KH |
800 | new = mpol_new(mode, flags, nodes); |
801 | if (IS_ERR(new)) { | |
802 | ret = PTR_ERR(new); | |
803 | goto out; | |
804 | } | |
2c7c3a7d | 805 | |
58568d2a | 806 | task_lock(current); |
4bfc4495 | 807 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
808 | if (ret) { |
809 | task_unlock(current); | |
58568d2a | 810 | mpol_put(new); |
4bfc4495 | 811 | goto out; |
58568d2a MX |
812 | } |
813 | old = current->mempolicy; | |
1da177e4 | 814 | current->mempolicy = new; |
45816682 VB |
815 | if (new && new->mode == MPOL_INTERLEAVE) |
816 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 817 | task_unlock(current); |
58568d2a | 818 | mpol_put(old); |
4bfc4495 KH |
819 | ret = 0; |
820 | out: | |
821 | NODEMASK_SCRATCH_FREE(scratch); | |
822 | return ret; | |
1da177e4 LT |
823 | } |
824 | ||
bea904d5 LS |
825 | /* |
826 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
827 | * |
828 | * Called with task's alloc_lock held | |
bea904d5 LS |
829 | */ |
830 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 831 | { |
dfcd3c0d | 832 | nodes_clear(*nodes); |
bea904d5 LS |
833 | if (p == &default_policy) |
834 | return; | |
835 | ||
45c4745a | 836 | switch (p->mode) { |
19770b32 MG |
837 | case MPOL_BIND: |
838 | /* Fall through */ | |
1da177e4 | 839 | case MPOL_INTERLEAVE: |
dfcd3c0d | 840 | *nodes = p->v.nodes; |
1da177e4 LT |
841 | break; |
842 | case MPOL_PREFERRED: | |
fc36b8d3 | 843 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 844 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 845 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
846 | break; |
847 | default: | |
848 | BUG(); | |
849 | } | |
850 | } | |
851 | ||
3b9aadf7 | 852 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
853 | { |
854 | struct page *p; | |
855 | int err; | |
856 | ||
3b9aadf7 AA |
857 | int locked = 1; |
858 | err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked); | |
1da177e4 LT |
859 | if (err >= 0) { |
860 | err = page_to_nid(p); | |
861 | put_page(p); | |
862 | } | |
3b9aadf7 AA |
863 | if (locked) |
864 | up_read(&mm->mmap_sem); | |
1da177e4 LT |
865 | return err; |
866 | } | |
867 | ||
1da177e4 | 868 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
869 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
870 | unsigned long addr, unsigned long flags) | |
1da177e4 | 871 | { |
8bccd85f | 872 | int err; |
1da177e4 LT |
873 | struct mm_struct *mm = current->mm; |
874 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 875 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 876 | |
754af6f5 LS |
877 | if (flags & |
878 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 879 | return -EINVAL; |
754af6f5 LS |
880 | |
881 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
882 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
883 | return -EINVAL; | |
884 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 885 | task_lock(current); |
754af6f5 | 886 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 887 | task_unlock(current); |
754af6f5 LS |
888 | return 0; |
889 | } | |
890 | ||
1da177e4 | 891 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
892 | /* |
893 | * Do NOT fall back to task policy if the | |
894 | * vma/shared policy at addr is NULL. We | |
895 | * want to return MPOL_DEFAULT in this case. | |
896 | */ | |
1da177e4 LT |
897 | down_read(&mm->mmap_sem); |
898 | vma = find_vma_intersection(mm, addr, addr+1); | |
899 | if (!vma) { | |
900 | up_read(&mm->mmap_sem); | |
901 | return -EFAULT; | |
902 | } | |
903 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
904 | pol = vma->vm_ops->get_policy(vma, addr); | |
905 | else | |
906 | pol = vma->vm_policy; | |
907 | } else if (addr) | |
908 | return -EINVAL; | |
909 | ||
910 | if (!pol) | |
bea904d5 | 911 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
912 | |
913 | if (flags & MPOL_F_NODE) { | |
914 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 AA |
915 | /* |
916 | * Take a refcount on the mpol, lookup_node() | |
917 | * wil drop the mmap_sem, so after calling | |
918 | * lookup_node() only "pol" remains valid, "vma" | |
919 | * is stale. | |
920 | */ | |
921 | pol_refcount = pol; | |
922 | vma = NULL; | |
923 | mpol_get(pol); | |
924 | err = lookup_node(mm, addr); | |
1da177e4 LT |
925 | if (err < 0) |
926 | goto out; | |
8bccd85f | 927 | *policy = err; |
1da177e4 | 928 | } else if (pol == current->mempolicy && |
45c4745a | 929 | pol->mode == MPOL_INTERLEAVE) { |
45816682 | 930 | *policy = next_node_in(current->il_prev, pol->v.nodes); |
1da177e4 LT |
931 | } else { |
932 | err = -EINVAL; | |
933 | goto out; | |
934 | } | |
bea904d5 LS |
935 | } else { |
936 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
937 | pol->mode; | |
d79df630 DR |
938 | /* |
939 | * Internal mempolicy flags must be masked off before exposing | |
940 | * the policy to userspace. | |
941 | */ | |
942 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 943 | } |
1da177e4 | 944 | |
1da177e4 | 945 | err = 0; |
58568d2a | 946 | if (nmask) { |
c6b6ef8b LS |
947 | if (mpol_store_user_nodemask(pol)) { |
948 | *nmask = pol->w.user_nodemask; | |
949 | } else { | |
950 | task_lock(current); | |
951 | get_policy_nodemask(pol, nmask); | |
952 | task_unlock(current); | |
953 | } | |
58568d2a | 954 | } |
1da177e4 LT |
955 | |
956 | out: | |
52cd3b07 | 957 | mpol_cond_put(pol); |
1da177e4 | 958 | if (vma) |
3b9aadf7 AA |
959 | up_read(&mm->mmap_sem); |
960 | if (pol_refcount) | |
961 | mpol_put(pol_refcount); | |
1da177e4 LT |
962 | return err; |
963 | } | |
964 | ||
b20a3503 | 965 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 966 | /* |
c8633798 | 967 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 968 | */ |
a53190a4 | 969 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 970 | unsigned long flags) |
6ce3c4c0 | 971 | { |
c8633798 | 972 | struct page *head = compound_head(page); |
6ce3c4c0 | 973 | /* |
fc301289 | 974 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 975 | */ |
c8633798 NH |
976 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
977 | if (!isolate_lru_page(head)) { | |
978 | list_add_tail(&head->lru, pagelist); | |
979 | mod_node_page_state(page_pgdat(head), | |
980 | NR_ISOLATED_ANON + page_is_file_cache(head), | |
981 | hpage_nr_pages(head)); | |
a53190a4 YS |
982 | } else if (flags & MPOL_MF_STRICT) { |
983 | /* | |
984 | * Non-movable page may reach here. And, there may be | |
985 | * temporary off LRU pages or non-LRU movable pages. | |
986 | * Treat them as unmovable pages since they can't be | |
987 | * isolated, so they can't be moved at the moment. It | |
988 | * should return -EIO for this case too. | |
989 | */ | |
990 | return -EIO; | |
62695a84 NP |
991 | } |
992 | } | |
a53190a4 YS |
993 | |
994 | return 0; | |
7e2ab150 | 995 | } |
6ce3c4c0 | 996 | |
a49bd4d7 | 997 | /* page allocation callback for NUMA node migration */ |
666feb21 | 998 | struct page *alloc_new_node_page(struct page *page, unsigned long node) |
95a402c3 | 999 | { |
e2d8cf40 NH |
1000 | if (PageHuge(page)) |
1001 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1002 | node); | |
94723aaf | 1003 | else if (PageTransHuge(page)) { |
c8633798 NH |
1004 | struct page *thp; |
1005 | ||
1006 | thp = alloc_pages_node(node, | |
1007 | (GFP_TRANSHUGE | __GFP_THISNODE), | |
1008 | HPAGE_PMD_ORDER); | |
1009 | if (!thp) | |
1010 | return NULL; | |
1011 | prep_transhuge_page(thp); | |
1012 | return thp; | |
1013 | } else | |
96db800f | 1014 | return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE | |
b360edb4 | 1015 | __GFP_THISNODE, 0); |
95a402c3 CL |
1016 | } |
1017 | ||
7e2ab150 CL |
1018 | /* |
1019 | * Migrate pages from one node to a target node. | |
1020 | * Returns error or the number of pages not migrated. | |
1021 | */ | |
dbcb0f19 AB |
1022 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1023 | int flags) | |
7e2ab150 CL |
1024 | { |
1025 | nodemask_t nmask; | |
1026 | LIST_HEAD(pagelist); | |
1027 | int err = 0; | |
1028 | ||
1029 | nodes_clear(nmask); | |
1030 | node_set(source, nmask); | |
6ce3c4c0 | 1031 | |
08270807 MK |
1032 | /* |
1033 | * This does not "check" the range but isolates all pages that | |
1034 | * need migration. Between passing in the full user address | |
1035 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1036 | */ | |
1037 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1038 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1039 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1040 | ||
cf608ac1 | 1041 | if (!list_empty(&pagelist)) { |
a49bd4d7 | 1042 | err = migrate_pages(&pagelist, alloc_new_node_page, NULL, dest, |
9c620e2b | 1043 | MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1044 | if (err) |
e2d8cf40 | 1045 | putback_movable_pages(&pagelist); |
cf608ac1 | 1046 | } |
95a402c3 | 1047 | |
7e2ab150 | 1048 | return err; |
6ce3c4c0 CL |
1049 | } |
1050 | ||
39743889 | 1051 | /* |
7e2ab150 CL |
1052 | * Move pages between the two nodesets so as to preserve the physical |
1053 | * layout as much as possible. | |
39743889 CL |
1054 | * |
1055 | * Returns the number of page that could not be moved. | |
1056 | */ | |
0ce72d4f AM |
1057 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1058 | const nodemask_t *to, int flags) | |
39743889 | 1059 | { |
7e2ab150 | 1060 | int busy = 0; |
0aedadf9 | 1061 | int err; |
7e2ab150 | 1062 | nodemask_t tmp; |
39743889 | 1063 | |
0aedadf9 CL |
1064 | err = migrate_prep(); |
1065 | if (err) | |
1066 | return err; | |
1067 | ||
53f2556b | 1068 | down_read(&mm->mmap_sem); |
39743889 | 1069 | |
da0aa138 KM |
1070 | /* |
1071 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1072 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1073 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1074 | * The pair of nodemasks 'to' and 'from' define the map. | |
1075 | * | |
1076 | * If no pair of bits is found that way, fallback to picking some | |
1077 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1078 | * 'source' and 'dest' bits are the same, this represents a node | |
1079 | * that will be migrating to itself, so no pages need move. | |
1080 | * | |
1081 | * If no bits are left in 'tmp', or if all remaining bits left | |
1082 | * in 'tmp' correspond to the same bit in 'to', return false | |
1083 | * (nothing left to migrate). | |
1084 | * | |
1085 | * This lets us pick a pair of nodes to migrate between, such that | |
1086 | * if possible the dest node is not already occupied by some other | |
1087 | * source node, minimizing the risk of overloading the memory on a | |
1088 | * node that would happen if we migrated incoming memory to a node | |
1089 | * before migrating outgoing memory source that same node. | |
1090 | * | |
1091 | * A single scan of tmp is sufficient. As we go, we remember the | |
1092 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1093 | * that not only moved, but what's better, moved to an empty slot | |
1094 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1095 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1096 | * most recent <s, d> pair that moved. If we get all the way through |
1097 | * the scan of tmp without finding any node that moved, much less | |
1098 | * moved to an empty node, then there is nothing left worth migrating. | |
1099 | */ | |
d4984711 | 1100 | |
0ce72d4f | 1101 | tmp = *from; |
7e2ab150 CL |
1102 | while (!nodes_empty(tmp)) { |
1103 | int s,d; | |
b76ac7e7 | 1104 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1105 | int dest = 0; |
1106 | ||
1107 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1108 | |
1109 | /* | |
1110 | * do_migrate_pages() tries to maintain the relative | |
1111 | * node relationship of the pages established between | |
1112 | * threads and memory areas. | |
1113 | * | |
1114 | * However if the number of source nodes is not equal to | |
1115 | * the number of destination nodes we can not preserve | |
1116 | * this node relative relationship. In that case, skip | |
1117 | * copying memory from a node that is in the destination | |
1118 | * mask. | |
1119 | * | |
1120 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1121 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1122 | */ | |
1123 | ||
0ce72d4f AM |
1124 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1125 | (node_isset(s, *to))) | |
4a5b18cc LW |
1126 | continue; |
1127 | ||
0ce72d4f | 1128 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1129 | if (s == d) |
1130 | continue; | |
1131 | ||
1132 | source = s; /* Node moved. Memorize */ | |
1133 | dest = d; | |
1134 | ||
1135 | /* dest not in remaining from nodes? */ | |
1136 | if (!node_isset(dest, tmp)) | |
1137 | break; | |
1138 | } | |
b76ac7e7 | 1139 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1140 | break; |
1141 | ||
1142 | node_clear(source, tmp); | |
1143 | err = migrate_to_node(mm, source, dest, flags); | |
1144 | if (err > 0) | |
1145 | busy += err; | |
1146 | if (err < 0) | |
1147 | break; | |
39743889 CL |
1148 | } |
1149 | up_read(&mm->mmap_sem); | |
7e2ab150 CL |
1150 | if (err < 0) |
1151 | return err; | |
1152 | return busy; | |
b20a3503 CL |
1153 | |
1154 | } | |
1155 | ||
3ad33b24 LS |
1156 | /* |
1157 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1158 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1159 | * Search forward from there, if not. N.B., this assumes that the |
1160 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1161 | * is in virtual address order. | |
1162 | */ | |
666feb21 | 1163 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1164 | { |
d05f0cdc | 1165 | struct vm_area_struct *vma; |
3ad33b24 | 1166 | unsigned long uninitialized_var(address); |
95a402c3 | 1167 | |
d05f0cdc | 1168 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1169 | while (vma) { |
1170 | address = page_address_in_vma(page, vma); | |
1171 | if (address != -EFAULT) | |
1172 | break; | |
1173 | vma = vma->vm_next; | |
1174 | } | |
11c731e8 WL |
1175 | |
1176 | if (PageHuge(page)) { | |
389c8178 MH |
1177 | return alloc_huge_page_vma(page_hstate(compound_head(page)), |
1178 | vma, address); | |
94723aaf | 1179 | } else if (PageTransHuge(page)) { |
c8633798 NH |
1180 | struct page *thp; |
1181 | ||
92717d42 AA |
1182 | thp = alloc_pages_vma(GFP_TRANSHUGE, HPAGE_PMD_ORDER, vma, |
1183 | address, numa_node_id()); | |
c8633798 NH |
1184 | if (!thp) |
1185 | return NULL; | |
1186 | prep_transhuge_page(thp); | |
1187 | return thp; | |
11c731e8 | 1188 | } |
0bf598d8 | 1189 | /* |
11c731e8 | 1190 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1191 | */ |
0f556856 MH |
1192 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL, |
1193 | vma, address); | |
95a402c3 | 1194 | } |
b20a3503 CL |
1195 | #else |
1196 | ||
a53190a4 | 1197 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1198 | unsigned long flags) |
1199 | { | |
a53190a4 | 1200 | return -EIO; |
39743889 CL |
1201 | } |
1202 | ||
0ce72d4f AM |
1203 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1204 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1205 | { |
1206 | return -ENOSYS; | |
1207 | } | |
95a402c3 | 1208 | |
666feb21 | 1209 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1210 | { |
1211 | return NULL; | |
1212 | } | |
b20a3503 CL |
1213 | #endif |
1214 | ||
dbcb0f19 | 1215 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1216 | unsigned short mode, unsigned short mode_flags, |
1217 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1218 | { |
6ce3c4c0 CL |
1219 | struct mm_struct *mm = current->mm; |
1220 | struct mempolicy *new; | |
1221 | unsigned long end; | |
1222 | int err; | |
d8835445 | 1223 | int ret; |
6ce3c4c0 CL |
1224 | LIST_HEAD(pagelist); |
1225 | ||
b24f53a0 | 1226 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1227 | return -EINVAL; |
74c00241 | 1228 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1229 | return -EPERM; |
1230 | ||
1231 | if (start & ~PAGE_MASK) | |
1232 | return -EINVAL; | |
1233 | ||
1234 | if (mode == MPOL_DEFAULT) | |
1235 | flags &= ~MPOL_MF_STRICT; | |
1236 | ||
1237 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1238 | end = start + len; | |
1239 | ||
1240 | if (end < start) | |
1241 | return -EINVAL; | |
1242 | if (end == start) | |
1243 | return 0; | |
1244 | ||
028fec41 | 1245 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1246 | if (IS_ERR(new)) |
1247 | return PTR_ERR(new); | |
1248 | ||
b24f53a0 LS |
1249 | if (flags & MPOL_MF_LAZY) |
1250 | new->flags |= MPOL_F_MOF; | |
1251 | ||
6ce3c4c0 CL |
1252 | /* |
1253 | * If we are using the default policy then operation | |
1254 | * on discontinuous address spaces is okay after all | |
1255 | */ | |
1256 | if (!new) | |
1257 | flags |= MPOL_MF_DISCONTIG_OK; | |
1258 | ||
028fec41 DR |
1259 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1260 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1261 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1262 | |
0aedadf9 CL |
1263 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1264 | ||
1265 | err = migrate_prep(); | |
1266 | if (err) | |
b05ca738 | 1267 | goto mpol_out; |
0aedadf9 | 1268 | } |
4bfc4495 KH |
1269 | { |
1270 | NODEMASK_SCRATCH(scratch); | |
1271 | if (scratch) { | |
1272 | down_write(&mm->mmap_sem); | |
1273 | task_lock(current); | |
1274 | err = mpol_set_nodemask(new, nmask, scratch); | |
1275 | task_unlock(current); | |
1276 | if (err) | |
1277 | up_write(&mm->mmap_sem); | |
1278 | } else | |
1279 | err = -ENOMEM; | |
1280 | NODEMASK_SCRATCH_FREE(scratch); | |
1281 | } | |
b05ca738 KM |
1282 | if (err) |
1283 | goto mpol_out; | |
1284 | ||
d8835445 | 1285 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1286 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1287 | |
1288 | if (ret < 0) { | |
1289 | err = -EIO; | |
1290 | goto up_out; | |
1291 | } | |
1292 | ||
1293 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1294 | |
b24f53a0 LS |
1295 | if (!err) { |
1296 | int nr_failed = 0; | |
1297 | ||
cf608ac1 | 1298 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1299 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1300 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1301 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1302 | if (nr_failed) |
74060e4d | 1303 | putback_movable_pages(&pagelist); |
cf608ac1 | 1304 | } |
6ce3c4c0 | 1305 | |
d8835445 | 1306 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1307 | err = -EIO; |
ab8a3e14 | 1308 | } else |
b0e5fd73 | 1309 | putback_movable_pages(&pagelist); |
b20a3503 | 1310 | |
d8835445 | 1311 | up_out: |
6ce3c4c0 | 1312 | up_write(&mm->mmap_sem); |
d8835445 | 1313 | mpol_out: |
f0be3d32 | 1314 | mpol_put(new); |
6ce3c4c0 CL |
1315 | return err; |
1316 | } | |
1317 | ||
8bccd85f CL |
1318 | /* |
1319 | * User space interface with variable sized bitmaps for nodelists. | |
1320 | */ | |
1321 | ||
1322 | /* Copy a node mask from user space. */ | |
39743889 | 1323 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1324 | unsigned long maxnode) |
1325 | { | |
1326 | unsigned long k; | |
56521e7a | 1327 | unsigned long t; |
8bccd85f CL |
1328 | unsigned long nlongs; |
1329 | unsigned long endmask; | |
1330 | ||
1331 | --maxnode; | |
1332 | nodes_clear(*nodes); | |
1333 | if (maxnode == 0 || !nmask) | |
1334 | return 0; | |
a9c930ba | 1335 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1336 | return -EINVAL; |
8bccd85f CL |
1337 | |
1338 | nlongs = BITS_TO_LONGS(maxnode); | |
1339 | if ((maxnode % BITS_PER_LONG) == 0) | |
1340 | endmask = ~0UL; | |
1341 | else | |
1342 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1343 | ||
56521e7a YX |
1344 | /* |
1345 | * When the user specified more nodes than supported just check | |
1346 | * if the non supported part is all zero. | |
1347 | * | |
1348 | * If maxnode have more longs than MAX_NUMNODES, check | |
1349 | * the bits in that area first. And then go through to | |
1350 | * check the rest bits which equal or bigger than MAX_NUMNODES. | |
1351 | * Otherwise, just check bits [MAX_NUMNODES, maxnode). | |
1352 | */ | |
8bccd85f | 1353 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
8bccd85f | 1354 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
8bccd85f CL |
1355 | if (get_user(t, nmask + k)) |
1356 | return -EFAULT; | |
1357 | if (k == nlongs - 1) { | |
1358 | if (t & endmask) | |
1359 | return -EINVAL; | |
1360 | } else if (t) | |
1361 | return -EINVAL; | |
1362 | } | |
1363 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1364 | endmask = ~0UL; | |
1365 | } | |
1366 | ||
56521e7a YX |
1367 | if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) { |
1368 | unsigned long valid_mask = endmask; | |
1369 | ||
1370 | valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1371 | if (get_user(t, nmask + nlongs - 1)) | |
1372 | return -EFAULT; | |
1373 | if (t & valid_mask) | |
1374 | return -EINVAL; | |
1375 | } | |
1376 | ||
8bccd85f CL |
1377 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
1378 | return -EFAULT; | |
1379 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1380 | return 0; | |
1381 | } | |
1382 | ||
1383 | /* Copy a kernel node mask to user space */ | |
1384 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1385 | nodemask_t *nodes) | |
1386 | { | |
1387 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1388 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
8bccd85f CL |
1389 | |
1390 | if (copy > nbytes) { | |
1391 | if (copy > PAGE_SIZE) | |
1392 | return -EINVAL; | |
1393 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1394 | return -EFAULT; | |
1395 | copy = nbytes; | |
1396 | } | |
1397 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1398 | } | |
1399 | ||
e7dc9ad6 DB |
1400 | static long kernel_mbind(unsigned long start, unsigned long len, |
1401 | unsigned long mode, const unsigned long __user *nmask, | |
1402 | unsigned long maxnode, unsigned int flags) | |
8bccd85f CL |
1403 | { |
1404 | nodemask_t nodes; | |
1405 | int err; | |
028fec41 | 1406 | unsigned short mode_flags; |
8bccd85f | 1407 | |
028fec41 DR |
1408 | mode_flags = mode & MPOL_MODE_FLAGS; |
1409 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1410 | if (mode >= MPOL_MAX) |
1411 | return -EINVAL; | |
4c50bc01 DR |
1412 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1413 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1414 | return -EINVAL; | |
8bccd85f CL |
1415 | err = get_nodes(&nodes, nmask, maxnode); |
1416 | if (err) | |
1417 | return err; | |
028fec41 | 1418 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1419 | } |
1420 | ||
e7dc9ad6 DB |
1421 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1422 | unsigned long, mode, const unsigned long __user *, nmask, | |
1423 | unsigned long, maxnode, unsigned int, flags) | |
1424 | { | |
1425 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1426 | } | |
1427 | ||
8bccd85f | 1428 | /* Set the process memory policy */ |
af03c4ac DB |
1429 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1430 | unsigned long maxnode) | |
8bccd85f CL |
1431 | { |
1432 | int err; | |
1433 | nodemask_t nodes; | |
028fec41 | 1434 | unsigned short flags; |
8bccd85f | 1435 | |
028fec41 DR |
1436 | flags = mode & MPOL_MODE_FLAGS; |
1437 | mode &= ~MPOL_MODE_FLAGS; | |
1438 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1439 | return -EINVAL; |
4c50bc01 DR |
1440 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1441 | return -EINVAL; | |
8bccd85f CL |
1442 | err = get_nodes(&nodes, nmask, maxnode); |
1443 | if (err) | |
1444 | return err; | |
028fec41 | 1445 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1446 | } |
1447 | ||
af03c4ac DB |
1448 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1449 | unsigned long, maxnode) | |
1450 | { | |
1451 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1452 | } | |
1453 | ||
b6e9b0ba DB |
1454 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1455 | const unsigned long __user *old_nodes, | |
1456 | const unsigned long __user *new_nodes) | |
39743889 | 1457 | { |
596d7cfa | 1458 | struct mm_struct *mm = NULL; |
39743889 | 1459 | struct task_struct *task; |
39743889 CL |
1460 | nodemask_t task_nodes; |
1461 | int err; | |
596d7cfa KM |
1462 | nodemask_t *old; |
1463 | nodemask_t *new; | |
1464 | NODEMASK_SCRATCH(scratch); | |
1465 | ||
1466 | if (!scratch) | |
1467 | return -ENOMEM; | |
39743889 | 1468 | |
596d7cfa KM |
1469 | old = &scratch->mask1; |
1470 | new = &scratch->mask2; | |
1471 | ||
1472 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1473 | if (err) |
596d7cfa | 1474 | goto out; |
39743889 | 1475 | |
596d7cfa | 1476 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1477 | if (err) |
596d7cfa | 1478 | goto out; |
39743889 CL |
1479 | |
1480 | /* Find the mm_struct */ | |
55cfaa3c | 1481 | rcu_read_lock(); |
228ebcbe | 1482 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1483 | if (!task) { |
55cfaa3c | 1484 | rcu_read_unlock(); |
596d7cfa KM |
1485 | err = -ESRCH; |
1486 | goto out; | |
39743889 | 1487 | } |
3268c63e | 1488 | get_task_struct(task); |
39743889 | 1489 | |
596d7cfa | 1490 | err = -EINVAL; |
39743889 CL |
1491 | |
1492 | /* | |
31367466 OE |
1493 | * Check if this process has the right to modify the specified process. |
1494 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1495 | */ |
31367466 | 1496 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1497 | rcu_read_unlock(); |
39743889 | 1498 | err = -EPERM; |
3268c63e | 1499 | goto out_put; |
39743889 | 1500 | } |
c69e8d9c | 1501 | rcu_read_unlock(); |
39743889 CL |
1502 | |
1503 | task_nodes = cpuset_mems_allowed(task); | |
1504 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1505 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1506 | err = -EPERM; |
3268c63e | 1507 | goto out_put; |
39743889 CL |
1508 | } |
1509 | ||
0486a38b YX |
1510 | task_nodes = cpuset_mems_allowed(current); |
1511 | nodes_and(*new, *new, task_nodes); | |
1512 | if (nodes_empty(*new)) | |
1513 | goto out_put; | |
1514 | ||
86c3a764 DQ |
1515 | err = security_task_movememory(task); |
1516 | if (err) | |
3268c63e | 1517 | goto out_put; |
86c3a764 | 1518 | |
3268c63e CL |
1519 | mm = get_task_mm(task); |
1520 | put_task_struct(task); | |
f2a9ef88 SL |
1521 | |
1522 | if (!mm) { | |
3268c63e | 1523 | err = -EINVAL; |
f2a9ef88 SL |
1524 | goto out; |
1525 | } | |
1526 | ||
1527 | err = do_migrate_pages(mm, old, new, | |
1528 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1529 | |
1530 | mmput(mm); | |
1531 | out: | |
596d7cfa KM |
1532 | NODEMASK_SCRATCH_FREE(scratch); |
1533 | ||
39743889 | 1534 | return err; |
3268c63e CL |
1535 | |
1536 | out_put: | |
1537 | put_task_struct(task); | |
1538 | goto out; | |
1539 | ||
39743889 CL |
1540 | } |
1541 | ||
b6e9b0ba DB |
1542 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1543 | const unsigned long __user *, old_nodes, | |
1544 | const unsigned long __user *, new_nodes) | |
1545 | { | |
1546 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1547 | } | |
1548 | ||
39743889 | 1549 | |
8bccd85f | 1550 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1551 | static int kernel_get_mempolicy(int __user *policy, |
1552 | unsigned long __user *nmask, | |
1553 | unsigned long maxnode, | |
1554 | unsigned long addr, | |
1555 | unsigned long flags) | |
8bccd85f | 1556 | { |
dbcb0f19 AB |
1557 | int err; |
1558 | int uninitialized_var(pval); | |
8bccd85f CL |
1559 | nodemask_t nodes; |
1560 | ||
050c17f2 | 1561 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1562 | return -EINVAL; |
1563 | ||
1564 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1565 | ||
1566 | if (err) | |
1567 | return err; | |
1568 | ||
1569 | if (policy && put_user(pval, policy)) | |
1570 | return -EFAULT; | |
1571 | ||
1572 | if (nmask) | |
1573 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1574 | ||
1575 | return err; | |
1576 | } | |
1577 | ||
af03c4ac DB |
1578 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1579 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1580 | unsigned long, addr, unsigned long, flags) | |
1581 | { | |
1582 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1583 | } | |
1584 | ||
1da177e4 LT |
1585 | #ifdef CONFIG_COMPAT |
1586 | ||
c93e0f6c HC |
1587 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1588 | compat_ulong_t __user *, nmask, | |
1589 | compat_ulong_t, maxnode, | |
1590 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1591 | { |
1592 | long err; | |
1593 | unsigned long __user *nm = NULL; | |
1594 | unsigned long nr_bits, alloc_size; | |
1595 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1596 | ||
050c17f2 | 1597 | nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids); |
1da177e4 LT |
1598 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
1599 | ||
1600 | if (nmask) | |
1601 | nm = compat_alloc_user_space(alloc_size); | |
1602 | ||
af03c4ac | 1603 | err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
1da177e4 LT |
1604 | |
1605 | if (!err && nmask) { | |
2bbff6c7 KH |
1606 | unsigned long copy_size; |
1607 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1608 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1609 | /* ensure entire bitmap is zeroed */ |
1610 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1611 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1612 | } | |
1613 | ||
1614 | return err; | |
1615 | } | |
1616 | ||
c93e0f6c HC |
1617 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1618 | compat_ulong_t, maxnode) | |
1da177e4 | 1619 | { |
1da177e4 LT |
1620 | unsigned long __user *nm = NULL; |
1621 | unsigned long nr_bits, alloc_size; | |
1622 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1623 | ||
1624 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1625 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1626 | ||
1627 | if (nmask) { | |
cf01fb99 CS |
1628 | if (compat_get_bitmap(bm, nmask, nr_bits)) |
1629 | return -EFAULT; | |
1da177e4 | 1630 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1631 | if (copy_to_user(nm, bm, alloc_size)) |
1632 | return -EFAULT; | |
1da177e4 LT |
1633 | } |
1634 | ||
af03c4ac | 1635 | return kernel_set_mempolicy(mode, nm, nr_bits+1); |
1da177e4 LT |
1636 | } |
1637 | ||
c93e0f6c HC |
1638 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1639 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1640 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 | 1641 | { |
1da177e4 LT |
1642 | unsigned long __user *nm = NULL; |
1643 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1644 | nodemask_t bm; |
1da177e4 LT |
1645 | |
1646 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1647 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1648 | ||
1649 | if (nmask) { | |
cf01fb99 CS |
1650 | if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits)) |
1651 | return -EFAULT; | |
1da177e4 | 1652 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1653 | if (copy_to_user(nm, nodes_addr(bm), alloc_size)) |
1654 | return -EFAULT; | |
1da177e4 LT |
1655 | } |
1656 | ||
e7dc9ad6 | 1657 | return kernel_mbind(start, len, mode, nm, nr_bits+1, flags); |
1da177e4 LT |
1658 | } |
1659 | ||
b6e9b0ba DB |
1660 | COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid, |
1661 | compat_ulong_t, maxnode, | |
1662 | const compat_ulong_t __user *, old_nodes, | |
1663 | const compat_ulong_t __user *, new_nodes) | |
1664 | { | |
1665 | unsigned long __user *old = NULL; | |
1666 | unsigned long __user *new = NULL; | |
1667 | nodemask_t tmp_mask; | |
1668 | unsigned long nr_bits; | |
1669 | unsigned long size; | |
1670 | ||
1671 | nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES); | |
1672 | size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1673 | if (old_nodes) { | |
1674 | if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits)) | |
1675 | return -EFAULT; | |
1676 | old = compat_alloc_user_space(new_nodes ? size * 2 : size); | |
1677 | if (new_nodes) | |
1678 | new = old + size / sizeof(unsigned long); | |
1679 | if (copy_to_user(old, nodes_addr(tmp_mask), size)) | |
1680 | return -EFAULT; | |
1681 | } | |
1682 | if (new_nodes) { | |
1683 | if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits)) | |
1684 | return -EFAULT; | |
1685 | if (new == NULL) | |
1686 | new = compat_alloc_user_space(size); | |
1687 | if (copy_to_user(new, nodes_addr(tmp_mask), size)) | |
1688 | return -EFAULT; | |
1689 | } | |
1690 | return kernel_migrate_pages(pid, nr_bits + 1, old, new); | |
1691 | } | |
1692 | ||
1693 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 1694 | |
74d2c3a0 ON |
1695 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1696 | unsigned long addr) | |
1da177e4 | 1697 | { |
8d90274b | 1698 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1699 | |
1700 | if (vma) { | |
480eccf9 | 1701 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1702 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1703 | } else if (vma->vm_policy) { |
1da177e4 | 1704 | pol = vma->vm_policy; |
00442ad0 MG |
1705 | |
1706 | /* | |
1707 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1708 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1709 | * count on these policies which will be dropped by | |
1710 | * mpol_cond_put() later | |
1711 | */ | |
1712 | if (mpol_needs_cond_ref(pol)) | |
1713 | mpol_get(pol); | |
1714 | } | |
1da177e4 | 1715 | } |
f15ca78e | 1716 | |
74d2c3a0 ON |
1717 | return pol; |
1718 | } | |
1719 | ||
1720 | /* | |
dd6eecb9 | 1721 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1722 | * @vma: virtual memory area whose policy is sought |
1723 | * @addr: address in @vma for shared policy lookup | |
1724 | * | |
1725 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1726 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1727 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1728 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1729 | * freeing by another task. It is the caller's responsibility to free the | |
1730 | * extra reference for shared policies. | |
1731 | */ | |
a8282608 | 1732 | struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1733 | unsigned long addr) |
74d2c3a0 ON |
1734 | { |
1735 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1736 | ||
8d90274b | 1737 | if (!pol) |
dd6eecb9 | 1738 | pol = get_task_policy(current); |
8d90274b | 1739 | |
1da177e4 LT |
1740 | return pol; |
1741 | } | |
1742 | ||
6b6482bb | 1743 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1744 | { |
6b6482bb | 1745 | struct mempolicy *pol; |
fc314724 | 1746 | |
6b6482bb ON |
1747 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1748 | bool ret = false; | |
fc314724 | 1749 | |
6b6482bb ON |
1750 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1751 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1752 | ret = true; | |
1753 | mpol_cond_put(pol); | |
8d90274b | 1754 | |
6b6482bb | 1755 | return ret; |
fc314724 MG |
1756 | } |
1757 | ||
6b6482bb | 1758 | pol = vma->vm_policy; |
8d90274b | 1759 | if (!pol) |
6b6482bb | 1760 | pol = get_task_policy(current); |
8d90274b | 1761 | |
fc314724 MG |
1762 | return pol->flags & MPOL_F_MOF; |
1763 | } | |
1764 | ||
d3eb1570 LJ |
1765 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1766 | { | |
1767 | enum zone_type dynamic_policy_zone = policy_zone; | |
1768 | ||
1769 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1770 | ||
1771 | /* | |
1772 | * if policy->v.nodes has movable memory only, | |
1773 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1774 | * | |
1775 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1776 | * so if the following test faile, it implies | |
1777 | * policy->v.nodes has movable memory only. | |
1778 | */ | |
1779 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1780 | dynamic_policy_zone = ZONE_MOVABLE; | |
1781 | ||
1782 | return zone >= dynamic_policy_zone; | |
1783 | } | |
1784 | ||
52cd3b07 LS |
1785 | /* |
1786 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1787 | * page allocation | |
1788 | */ | |
1789 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1790 | { |
1791 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1792 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1793 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1794 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1795 | return &policy->v.nodes; | |
1796 | ||
1797 | return NULL; | |
1798 | } | |
1799 | ||
04ec6264 VB |
1800 | /* Return the node id preferred by the given mempolicy, or the given id */ |
1801 | static int policy_node(gfp_t gfp, struct mempolicy *policy, | |
1802 | int nd) | |
1da177e4 | 1803 | { |
6d840958 MH |
1804 | if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL)) |
1805 | nd = policy->v.preferred_node; | |
1806 | else { | |
19770b32 | 1807 | /* |
6d840958 MH |
1808 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1809 | * because we might easily break the expectation to stay on the | |
1810 | * requested node and not break the policy. | |
19770b32 | 1811 | */ |
6d840958 | 1812 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1813 | } |
6d840958 | 1814 | |
04ec6264 | 1815 | return nd; |
1da177e4 LT |
1816 | } |
1817 | ||
1818 | /* Do dynamic interleaving for a process */ | |
1819 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1820 | { | |
45816682 | 1821 | unsigned next; |
1da177e4 LT |
1822 | struct task_struct *me = current; |
1823 | ||
45816682 | 1824 | next = next_node_in(me->il_prev, policy->v.nodes); |
f5b087b5 | 1825 | if (next < MAX_NUMNODES) |
45816682 VB |
1826 | me->il_prev = next; |
1827 | return next; | |
1da177e4 LT |
1828 | } |
1829 | ||
dc85da15 CL |
1830 | /* |
1831 | * Depending on the memory policy provide a node from which to allocate the | |
1832 | * next slab entry. | |
1833 | */ | |
2a389610 | 1834 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1835 | { |
e7b691b0 | 1836 | struct mempolicy *policy; |
2a389610 | 1837 | int node = numa_mem_id(); |
e7b691b0 AK |
1838 | |
1839 | if (in_interrupt()) | |
2a389610 | 1840 | return node; |
e7b691b0 AK |
1841 | |
1842 | policy = current->mempolicy; | |
fc36b8d3 | 1843 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1844 | return node; |
bea904d5 LS |
1845 | |
1846 | switch (policy->mode) { | |
1847 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1848 | /* |
1849 | * handled MPOL_F_LOCAL above | |
1850 | */ | |
1851 | return policy->v.preferred_node; | |
765c4507 | 1852 | |
dc85da15 CL |
1853 | case MPOL_INTERLEAVE: |
1854 | return interleave_nodes(policy); | |
1855 | ||
dd1a239f | 1856 | case MPOL_BIND: { |
c33d6c06 MG |
1857 | struct zoneref *z; |
1858 | ||
dc85da15 CL |
1859 | /* |
1860 | * Follow bind policy behavior and start allocation at the | |
1861 | * first node. | |
1862 | */ | |
19770b32 | 1863 | struct zonelist *zonelist; |
19770b32 | 1864 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1865 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 MG |
1866 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
1867 | &policy->v.nodes); | |
c1093b74 | 1868 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1869 | } |
dc85da15 | 1870 | |
dc85da15 | 1871 | default: |
bea904d5 | 1872 | BUG(); |
dc85da15 CL |
1873 | } |
1874 | } | |
1875 | ||
fee83b3a AM |
1876 | /* |
1877 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
1878 | * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the | |
1879 | * number of present nodes. | |
1880 | */ | |
98c70baa | 1881 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1882 | { |
dfcd3c0d | 1883 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1884 | unsigned target; |
fee83b3a AM |
1885 | int i; |
1886 | int nid; | |
1da177e4 | 1887 | |
f5b087b5 DR |
1888 | if (!nnodes) |
1889 | return numa_node_id(); | |
fee83b3a AM |
1890 | target = (unsigned int)n % nnodes; |
1891 | nid = first_node(pol->v.nodes); | |
1892 | for (i = 0; i < target; i++) | |
dfcd3c0d | 1893 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1894 | return nid; |
1895 | } | |
1896 | ||
5da7ca86 CL |
1897 | /* Determine a node number for interleave */ |
1898 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1899 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1900 | { | |
1901 | if (vma) { | |
1902 | unsigned long off; | |
1903 | ||
3b98b087 NA |
1904 | /* |
1905 | * for small pages, there is no difference between | |
1906 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1907 | * for huge pages, since vm_pgoff is in units of small | |
1908 | * pages, we need to shift off the always 0 bits to get | |
1909 | * a useful offset. | |
1910 | */ | |
1911 | BUG_ON(shift < PAGE_SHIFT); | |
1912 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 1913 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 1914 | return offset_il_node(pol, off); |
5da7ca86 CL |
1915 | } else |
1916 | return interleave_nodes(pol); | |
1917 | } | |
1918 | ||
00ac59ad | 1919 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 1920 | /* |
04ec6264 | 1921 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
1922 | * @vma: virtual memory area whose policy is sought |
1923 | * @addr: address in @vma for shared policy lookup and interleave policy | |
1924 | * @gfp_flags: for requested zone | |
1925 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
1926 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1927 | * |
04ec6264 | 1928 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 LS |
1929 | * to the struct mempolicy for conditional unref after allocation. |
1930 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1931 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 1932 | * |
d26914d1 | 1933 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 1934 | */ |
04ec6264 VB |
1935 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
1936 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 1937 | { |
04ec6264 | 1938 | int nid; |
5da7ca86 | 1939 | |
dd6eecb9 | 1940 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 1941 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1942 | |
52cd3b07 | 1943 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
1944 | nid = interleave_nid(*mpol, vma, addr, |
1945 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 1946 | } else { |
04ec6264 | 1947 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1948 | if ((*mpol)->mode == MPOL_BIND) |
1949 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 | 1950 | } |
04ec6264 | 1951 | return nid; |
5da7ca86 | 1952 | } |
06808b08 LS |
1953 | |
1954 | /* | |
1955 | * init_nodemask_of_mempolicy | |
1956 | * | |
1957 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1958 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1959 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1960 | * initialize the argument nodemask to contain the single node for | |
1961 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1962 | * of non-default mempolicy. | |
1963 | * | |
1964 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1965 | * because the current task is examining it's own mempolicy and a task's | |
1966 | * mempolicy is only ever changed by the task itself. | |
1967 | * | |
1968 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1969 | */ | |
1970 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1971 | { | |
1972 | struct mempolicy *mempolicy; | |
1973 | int nid; | |
1974 | ||
1975 | if (!(mask && current->mempolicy)) | |
1976 | return false; | |
1977 | ||
c0ff7453 | 1978 | task_lock(current); |
06808b08 LS |
1979 | mempolicy = current->mempolicy; |
1980 | switch (mempolicy->mode) { | |
1981 | case MPOL_PREFERRED: | |
1982 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1983 | nid = numa_node_id(); | |
1984 | else | |
1985 | nid = mempolicy->v.preferred_node; | |
1986 | init_nodemask_of_node(mask, nid); | |
1987 | break; | |
1988 | ||
1989 | case MPOL_BIND: | |
1990 | /* Fall through */ | |
1991 | case MPOL_INTERLEAVE: | |
1992 | *mask = mempolicy->v.nodes; | |
1993 | break; | |
1994 | ||
1995 | default: | |
1996 | BUG(); | |
1997 | } | |
c0ff7453 | 1998 | task_unlock(current); |
06808b08 LS |
1999 | |
2000 | return true; | |
2001 | } | |
00ac59ad | 2002 | #endif |
5da7ca86 | 2003 | |
6f48d0eb DR |
2004 | /* |
2005 | * mempolicy_nodemask_intersects | |
2006 | * | |
2007 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
2008 | * policy. Otherwise, check for intersection between mask and the policy | |
2009 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
2010 | * policy, always return true since it may allocate elsewhere on fallback. | |
2011 | * | |
2012 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2013 | */ | |
2014 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
2015 | const nodemask_t *mask) | |
2016 | { | |
2017 | struct mempolicy *mempolicy; | |
2018 | bool ret = true; | |
2019 | ||
2020 | if (!mask) | |
2021 | return ret; | |
2022 | task_lock(tsk); | |
2023 | mempolicy = tsk->mempolicy; | |
2024 | if (!mempolicy) | |
2025 | goto out; | |
2026 | ||
2027 | switch (mempolicy->mode) { | |
2028 | case MPOL_PREFERRED: | |
2029 | /* | |
2030 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
2031 | * allocate from, they may fallback to other nodes when oom. | |
2032 | * Thus, it's possible for tsk to have allocated memory from | |
2033 | * nodes in mask. | |
2034 | */ | |
2035 | break; | |
2036 | case MPOL_BIND: | |
2037 | case MPOL_INTERLEAVE: | |
2038 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
2039 | break; | |
2040 | default: | |
2041 | BUG(); | |
2042 | } | |
2043 | out: | |
2044 | task_unlock(tsk); | |
2045 | return ret; | |
2046 | } | |
2047 | ||
1da177e4 LT |
2048 | /* Allocate a page in interleaved policy. |
2049 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2050 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2051 | unsigned nid) | |
1da177e4 | 2052 | { |
1da177e4 LT |
2053 | struct page *page; |
2054 | ||
04ec6264 | 2055 | page = __alloc_pages(gfp, order, nid); |
4518085e KW |
2056 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2057 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2058 | return page; | |
de55c8b2 AR |
2059 | if (page && page_to_nid(page) == nid) { |
2060 | preempt_disable(); | |
2061 | __inc_numa_state(page_zone(page), NUMA_INTERLEAVE_HIT); | |
2062 | preempt_enable(); | |
2063 | } | |
1da177e4 LT |
2064 | return page; |
2065 | } | |
2066 | ||
2067 | /** | |
0bbbc0b3 | 2068 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
2069 | * |
2070 | * @gfp: | |
2071 | * %GFP_USER user allocation. | |
2072 | * %GFP_KERNEL kernel allocations, | |
2073 | * %GFP_HIGHMEM highmem/user allocations, | |
2074 | * %GFP_FS allocation should not call back into a file system. | |
2075 | * %GFP_ATOMIC don't sleep. | |
2076 | * | |
0bbbc0b3 | 2077 | * @order:Order of the GFP allocation. |
1da177e4 LT |
2078 | * @vma: Pointer to VMA or NULL if not available. |
2079 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
be97a41b | 2080 | * @node: Which node to prefer for allocation (modulo policy). |
1da177e4 LT |
2081 | * |
2082 | * This function allocates a page from the kernel page pool and applies | |
2083 | * a NUMA policy associated with the VMA or the current process. | |
2084 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
2085 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
be97a41b VB |
2086 | * all allocations for pages that will be mapped into user space. Returns |
2087 | * NULL when no page can be allocated. | |
1da177e4 LT |
2088 | */ |
2089 | struct page * | |
0bbbc0b3 | 2090 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
92717d42 | 2091 | unsigned long addr, int node) |
1da177e4 | 2092 | { |
cc9a6c87 | 2093 | struct mempolicy *pol; |
c0ff7453 | 2094 | struct page *page; |
04ec6264 | 2095 | int preferred_nid; |
be97a41b | 2096 | nodemask_t *nmask; |
cc9a6c87 | 2097 | |
dd6eecb9 | 2098 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2099 | |
0867a57c VB |
2100 | if (pol->mode == MPOL_INTERLEAVE) { |
2101 | unsigned nid; | |
2102 | ||
2103 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2104 | mpol_cond_put(pol); | |
2105 | page = alloc_page_interleave(gfp, order, nid); | |
2106 | goto out; | |
356ff8a9 DR |
2107 | } |
2108 | ||
be97a41b | 2109 | nmask = policy_nodemask(gfp, pol); |
04ec6264 VB |
2110 | preferred_nid = policy_node(gfp, pol, node); |
2111 | page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask); | |
d51e9894 | 2112 | mpol_cond_put(pol); |
be97a41b | 2113 | out: |
c0ff7453 | 2114 | return page; |
1da177e4 | 2115 | } |
69262215 | 2116 | EXPORT_SYMBOL(alloc_pages_vma); |
1da177e4 LT |
2117 | |
2118 | /** | |
2119 | * alloc_pages_current - Allocate pages. | |
2120 | * | |
2121 | * @gfp: | |
2122 | * %GFP_USER user allocation, | |
2123 | * %GFP_KERNEL kernel allocation, | |
2124 | * %GFP_HIGHMEM highmem allocation, | |
2125 | * %GFP_FS don't call back into a file system. | |
2126 | * %GFP_ATOMIC don't sleep. | |
2127 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2128 | * | |
2129 | * Allocate a page from the kernel page pool. When not in | |
2130 | * interrupt context and apply the current process NUMA policy. | |
2131 | * Returns NULL when no page can be allocated. | |
1da177e4 | 2132 | */ |
dd0fc66f | 2133 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2134 | { |
8d90274b | 2135 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2136 | struct page *page; |
1da177e4 | 2137 | |
8d90274b ON |
2138 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2139 | pol = get_task_policy(current); | |
52cd3b07 LS |
2140 | |
2141 | /* | |
2142 | * No reference counting needed for current->mempolicy | |
2143 | * nor system default_policy | |
2144 | */ | |
45c4745a | 2145 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2146 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2147 | else | |
2148 | page = __alloc_pages_nodemask(gfp, order, | |
04ec6264 | 2149 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2150 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2151 | |
c0ff7453 | 2152 | return page; |
1da177e4 LT |
2153 | } |
2154 | EXPORT_SYMBOL(alloc_pages_current); | |
2155 | ||
ef0855d3 ON |
2156 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2157 | { | |
2158 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2159 | ||
2160 | if (IS_ERR(pol)) | |
2161 | return PTR_ERR(pol); | |
2162 | dst->vm_policy = pol; | |
2163 | return 0; | |
2164 | } | |
2165 | ||
4225399a | 2166 | /* |
846a16bf | 2167 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2168 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2169 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2170 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2171 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2172 | * |
2173 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2174 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2175 | */ |
4225399a | 2176 | |
846a16bf LS |
2177 | /* Slow path of a mempolicy duplicate */ |
2178 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2179 | { |
2180 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2181 | ||
2182 | if (!new) | |
2183 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2184 | |
2185 | /* task's mempolicy is protected by alloc_lock */ | |
2186 | if (old == current->mempolicy) { | |
2187 | task_lock(current); | |
2188 | *new = *old; | |
2189 | task_unlock(current); | |
2190 | } else | |
2191 | *new = *old; | |
2192 | ||
4225399a PJ |
2193 | if (current_cpuset_is_being_rebound()) { |
2194 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2195 | mpol_rebind_policy(new, &mems); |
4225399a | 2196 | } |
1da177e4 | 2197 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2198 | return new; |
2199 | } | |
2200 | ||
2201 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2202 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2203 | { |
2204 | if (!a || !b) | |
fcfb4dcc | 2205 | return false; |
45c4745a | 2206 | if (a->mode != b->mode) |
fcfb4dcc | 2207 | return false; |
19800502 | 2208 | if (a->flags != b->flags) |
fcfb4dcc | 2209 | return false; |
19800502 BL |
2210 | if (mpol_store_user_nodemask(a)) |
2211 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2212 | return false; |
19800502 | 2213 | |
45c4745a | 2214 | switch (a->mode) { |
19770b32 MG |
2215 | case MPOL_BIND: |
2216 | /* Fall through */ | |
1da177e4 | 2217 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2218 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2219 | case MPOL_PREFERRED: |
8970a63e YX |
2220 | /* a's ->flags is the same as b's */ |
2221 | if (a->flags & MPOL_F_LOCAL) | |
2222 | return true; | |
75719661 | 2223 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2224 | default: |
2225 | BUG(); | |
fcfb4dcc | 2226 | return false; |
1da177e4 LT |
2227 | } |
2228 | } | |
2229 | ||
1da177e4 LT |
2230 | /* |
2231 | * Shared memory backing store policy support. | |
2232 | * | |
2233 | * Remember policies even when nobody has shared memory mapped. | |
2234 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2235 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2236 | * for any accesses to the tree. |
2237 | */ | |
2238 | ||
4a8c7bb5 NZ |
2239 | /* |
2240 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2241 | * reading or for writing | |
2242 | */ | |
1da177e4 LT |
2243 | static struct sp_node * |
2244 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2245 | { | |
2246 | struct rb_node *n = sp->root.rb_node; | |
2247 | ||
2248 | while (n) { | |
2249 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2250 | ||
2251 | if (start >= p->end) | |
2252 | n = n->rb_right; | |
2253 | else if (end <= p->start) | |
2254 | n = n->rb_left; | |
2255 | else | |
2256 | break; | |
2257 | } | |
2258 | if (!n) | |
2259 | return NULL; | |
2260 | for (;;) { | |
2261 | struct sp_node *w = NULL; | |
2262 | struct rb_node *prev = rb_prev(n); | |
2263 | if (!prev) | |
2264 | break; | |
2265 | w = rb_entry(prev, struct sp_node, nd); | |
2266 | if (w->end <= start) | |
2267 | break; | |
2268 | n = prev; | |
2269 | } | |
2270 | return rb_entry(n, struct sp_node, nd); | |
2271 | } | |
2272 | ||
4a8c7bb5 NZ |
2273 | /* |
2274 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2275 | * writing. | |
2276 | */ | |
1da177e4 LT |
2277 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2278 | { | |
2279 | struct rb_node **p = &sp->root.rb_node; | |
2280 | struct rb_node *parent = NULL; | |
2281 | struct sp_node *nd; | |
2282 | ||
2283 | while (*p) { | |
2284 | parent = *p; | |
2285 | nd = rb_entry(parent, struct sp_node, nd); | |
2286 | if (new->start < nd->start) | |
2287 | p = &(*p)->rb_left; | |
2288 | else if (new->end > nd->end) | |
2289 | p = &(*p)->rb_right; | |
2290 | else | |
2291 | BUG(); | |
2292 | } | |
2293 | rb_link_node(&new->nd, parent, p); | |
2294 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2295 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2296 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2297 | } |
2298 | ||
2299 | /* Find shared policy intersecting idx */ | |
2300 | struct mempolicy * | |
2301 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2302 | { | |
2303 | struct mempolicy *pol = NULL; | |
2304 | struct sp_node *sn; | |
2305 | ||
2306 | if (!sp->root.rb_node) | |
2307 | return NULL; | |
4a8c7bb5 | 2308 | read_lock(&sp->lock); |
1da177e4 LT |
2309 | sn = sp_lookup(sp, idx, idx+1); |
2310 | if (sn) { | |
2311 | mpol_get(sn->policy); | |
2312 | pol = sn->policy; | |
2313 | } | |
4a8c7bb5 | 2314 | read_unlock(&sp->lock); |
1da177e4 LT |
2315 | return pol; |
2316 | } | |
2317 | ||
63f74ca2 KM |
2318 | static void sp_free(struct sp_node *n) |
2319 | { | |
2320 | mpol_put(n->policy); | |
2321 | kmem_cache_free(sn_cache, n); | |
2322 | } | |
2323 | ||
771fb4d8 LS |
2324 | /** |
2325 | * mpol_misplaced - check whether current page node is valid in policy | |
2326 | * | |
b46e14ac FF |
2327 | * @page: page to be checked |
2328 | * @vma: vm area where page mapped | |
2329 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2330 | * |
2331 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2332 | * node id. | |
2333 | * | |
2334 | * Returns: | |
2335 | * -1 - not misplaced, page is in the right node | |
2336 | * node - node id where the page should be | |
2337 | * | |
2338 | * Policy determination "mimics" alloc_page_vma(). | |
2339 | * Called from fault path where we know the vma and faulting address. | |
2340 | */ | |
2341 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2342 | { | |
2343 | struct mempolicy *pol; | |
c33d6c06 | 2344 | struct zoneref *z; |
771fb4d8 LS |
2345 | int curnid = page_to_nid(page); |
2346 | unsigned long pgoff; | |
90572890 PZ |
2347 | int thiscpu = raw_smp_processor_id(); |
2348 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2349 | int polnid = NUMA_NO_NODE; |
771fb4d8 LS |
2350 | int ret = -1; |
2351 | ||
dd6eecb9 | 2352 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2353 | if (!(pol->flags & MPOL_F_MOF)) |
2354 | goto out; | |
2355 | ||
2356 | switch (pol->mode) { | |
2357 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2358 | pgoff = vma->vm_pgoff; |
2359 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2360 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2361 | break; |
2362 | ||
2363 | case MPOL_PREFERRED: | |
2364 | if (pol->flags & MPOL_F_LOCAL) | |
2365 | polnid = numa_node_id(); | |
2366 | else | |
2367 | polnid = pol->v.preferred_node; | |
2368 | break; | |
2369 | ||
2370 | case MPOL_BIND: | |
c33d6c06 | 2371 | |
771fb4d8 LS |
2372 | /* |
2373 | * allows binding to multiple nodes. | |
2374 | * use current page if in policy nodemask, | |
2375 | * else select nearest allowed node, if any. | |
2376 | * If no allowed nodes, use current [!misplaced]. | |
2377 | */ | |
2378 | if (node_isset(curnid, pol->v.nodes)) | |
2379 | goto out; | |
c33d6c06 | 2380 | z = first_zones_zonelist( |
771fb4d8 LS |
2381 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2382 | gfp_zone(GFP_HIGHUSER), | |
c33d6c06 | 2383 | &pol->v.nodes); |
c1093b74 | 2384 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2385 | break; |
2386 | ||
2387 | default: | |
2388 | BUG(); | |
2389 | } | |
5606e387 MG |
2390 | |
2391 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2392 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2393 | polnid = thisnid; |
5606e387 | 2394 | |
10f39042 | 2395 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2396 | goto out; |
e42c8ff2 MG |
2397 | } |
2398 | ||
771fb4d8 LS |
2399 | if (curnid != polnid) |
2400 | ret = polnid; | |
2401 | out: | |
2402 | mpol_cond_put(pol); | |
2403 | ||
2404 | return ret; | |
2405 | } | |
2406 | ||
c11600e4 DR |
2407 | /* |
2408 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2409 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2410 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2411 | * policy. | |
2412 | */ | |
2413 | void mpol_put_task_policy(struct task_struct *task) | |
2414 | { | |
2415 | struct mempolicy *pol; | |
2416 | ||
2417 | task_lock(task); | |
2418 | pol = task->mempolicy; | |
2419 | task->mempolicy = NULL; | |
2420 | task_unlock(task); | |
2421 | mpol_put(pol); | |
2422 | } | |
2423 | ||
1da177e4 LT |
2424 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2425 | { | |
140d5a49 | 2426 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2427 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2428 | sp_free(n); |
1da177e4 LT |
2429 | } |
2430 | ||
42288fe3 MG |
2431 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2432 | unsigned long end, struct mempolicy *pol) | |
2433 | { | |
2434 | node->start = start; | |
2435 | node->end = end; | |
2436 | node->policy = pol; | |
2437 | } | |
2438 | ||
dbcb0f19 AB |
2439 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2440 | struct mempolicy *pol) | |
1da177e4 | 2441 | { |
869833f2 KM |
2442 | struct sp_node *n; |
2443 | struct mempolicy *newpol; | |
1da177e4 | 2444 | |
869833f2 | 2445 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2446 | if (!n) |
2447 | return NULL; | |
869833f2 KM |
2448 | |
2449 | newpol = mpol_dup(pol); | |
2450 | if (IS_ERR(newpol)) { | |
2451 | kmem_cache_free(sn_cache, n); | |
2452 | return NULL; | |
2453 | } | |
2454 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2455 | sp_node_init(n, start, end, newpol); |
869833f2 | 2456 | |
1da177e4 LT |
2457 | return n; |
2458 | } | |
2459 | ||
2460 | /* Replace a policy range. */ | |
2461 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2462 | unsigned long end, struct sp_node *new) | |
2463 | { | |
b22d127a | 2464 | struct sp_node *n; |
42288fe3 MG |
2465 | struct sp_node *n_new = NULL; |
2466 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2467 | int ret = 0; |
1da177e4 | 2468 | |
42288fe3 | 2469 | restart: |
4a8c7bb5 | 2470 | write_lock(&sp->lock); |
1da177e4 LT |
2471 | n = sp_lookup(sp, start, end); |
2472 | /* Take care of old policies in the same range. */ | |
2473 | while (n && n->start < end) { | |
2474 | struct rb_node *next = rb_next(&n->nd); | |
2475 | if (n->start >= start) { | |
2476 | if (n->end <= end) | |
2477 | sp_delete(sp, n); | |
2478 | else | |
2479 | n->start = end; | |
2480 | } else { | |
2481 | /* Old policy spanning whole new range. */ | |
2482 | if (n->end > end) { | |
42288fe3 MG |
2483 | if (!n_new) |
2484 | goto alloc_new; | |
2485 | ||
2486 | *mpol_new = *n->policy; | |
2487 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2488 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2489 | n->end = start; |
5ca39575 | 2490 | sp_insert(sp, n_new); |
42288fe3 MG |
2491 | n_new = NULL; |
2492 | mpol_new = NULL; | |
1da177e4 LT |
2493 | break; |
2494 | } else | |
2495 | n->end = start; | |
2496 | } | |
2497 | if (!next) | |
2498 | break; | |
2499 | n = rb_entry(next, struct sp_node, nd); | |
2500 | } | |
2501 | if (new) | |
2502 | sp_insert(sp, new); | |
4a8c7bb5 | 2503 | write_unlock(&sp->lock); |
42288fe3 MG |
2504 | ret = 0; |
2505 | ||
2506 | err_out: | |
2507 | if (mpol_new) | |
2508 | mpol_put(mpol_new); | |
2509 | if (n_new) | |
2510 | kmem_cache_free(sn_cache, n_new); | |
2511 | ||
b22d127a | 2512 | return ret; |
42288fe3 MG |
2513 | |
2514 | alloc_new: | |
4a8c7bb5 | 2515 | write_unlock(&sp->lock); |
42288fe3 MG |
2516 | ret = -ENOMEM; |
2517 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2518 | if (!n_new) | |
2519 | goto err_out; | |
2520 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2521 | if (!mpol_new) | |
2522 | goto err_out; | |
2523 | goto restart; | |
1da177e4 LT |
2524 | } |
2525 | ||
71fe804b LS |
2526 | /** |
2527 | * mpol_shared_policy_init - initialize shared policy for inode | |
2528 | * @sp: pointer to inode shared policy | |
2529 | * @mpol: struct mempolicy to install | |
2530 | * | |
2531 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2532 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2533 | * This must be released on exit. | |
4bfc4495 | 2534 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2535 | */ |
2536 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2537 | { | |
58568d2a MX |
2538 | int ret; |
2539 | ||
71fe804b | 2540 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2541 | rwlock_init(&sp->lock); |
71fe804b LS |
2542 | |
2543 | if (mpol) { | |
2544 | struct vm_area_struct pvma; | |
2545 | struct mempolicy *new; | |
4bfc4495 | 2546 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2547 | |
4bfc4495 | 2548 | if (!scratch) |
5c0c1654 | 2549 | goto put_mpol; |
71fe804b LS |
2550 | /* contextualize the tmpfs mount point mempolicy */ |
2551 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2552 | if (IS_ERR(new)) |
0cae3457 | 2553 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2554 | |
2555 | task_lock(current); | |
4bfc4495 | 2556 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2557 | task_unlock(current); |
15d77835 | 2558 | if (ret) |
5c0c1654 | 2559 | goto put_new; |
71fe804b LS |
2560 | |
2561 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2562 | vma_init(&pvma, NULL); |
71fe804b LS |
2563 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2564 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2565 | |
5c0c1654 | 2566 | put_new: |
71fe804b | 2567 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2568 | free_scratch: |
4bfc4495 | 2569 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2570 | put_mpol: |
2571 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2572 | } |
2573 | } | |
2574 | ||
1da177e4 LT |
2575 | int mpol_set_shared_policy(struct shared_policy *info, |
2576 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2577 | { | |
2578 | int err; | |
2579 | struct sp_node *new = NULL; | |
2580 | unsigned long sz = vma_pages(vma); | |
2581 | ||
028fec41 | 2582 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2583 | vma->vm_pgoff, |
45c4745a | 2584 | sz, npol ? npol->mode : -1, |
028fec41 | 2585 | npol ? npol->flags : -1, |
00ef2d2f | 2586 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2587 | |
2588 | if (npol) { | |
2589 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2590 | if (!new) | |
2591 | return -ENOMEM; | |
2592 | } | |
2593 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2594 | if (err && new) | |
63f74ca2 | 2595 | sp_free(new); |
1da177e4 LT |
2596 | return err; |
2597 | } | |
2598 | ||
2599 | /* Free a backing policy store on inode delete. */ | |
2600 | void mpol_free_shared_policy(struct shared_policy *p) | |
2601 | { | |
2602 | struct sp_node *n; | |
2603 | struct rb_node *next; | |
2604 | ||
2605 | if (!p->root.rb_node) | |
2606 | return; | |
4a8c7bb5 | 2607 | write_lock(&p->lock); |
1da177e4 LT |
2608 | next = rb_first(&p->root); |
2609 | while (next) { | |
2610 | n = rb_entry(next, struct sp_node, nd); | |
2611 | next = rb_next(&n->nd); | |
63f74ca2 | 2612 | sp_delete(p, n); |
1da177e4 | 2613 | } |
4a8c7bb5 | 2614 | write_unlock(&p->lock); |
1da177e4 LT |
2615 | } |
2616 | ||
1a687c2e | 2617 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2618 | static int __initdata numabalancing_override; |
1a687c2e MG |
2619 | |
2620 | static void __init check_numabalancing_enable(void) | |
2621 | { | |
2622 | bool numabalancing_default = false; | |
2623 | ||
2624 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2625 | numabalancing_default = true; | |
2626 | ||
c297663c MG |
2627 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2628 | if (numabalancing_override) | |
2629 | set_numabalancing_state(numabalancing_override == 1); | |
2630 | ||
b0dc2b9b | 2631 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2632 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2633 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2634 | set_numabalancing_state(numabalancing_default); |
2635 | } | |
2636 | } | |
2637 | ||
2638 | static int __init setup_numabalancing(char *str) | |
2639 | { | |
2640 | int ret = 0; | |
2641 | if (!str) | |
2642 | goto out; | |
1a687c2e MG |
2643 | |
2644 | if (!strcmp(str, "enable")) { | |
c297663c | 2645 | numabalancing_override = 1; |
1a687c2e MG |
2646 | ret = 1; |
2647 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2648 | numabalancing_override = -1; |
1a687c2e MG |
2649 | ret = 1; |
2650 | } | |
2651 | out: | |
2652 | if (!ret) | |
4a404bea | 2653 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2654 | |
2655 | return ret; | |
2656 | } | |
2657 | __setup("numa_balancing=", setup_numabalancing); | |
2658 | #else | |
2659 | static inline void __init check_numabalancing_enable(void) | |
2660 | { | |
2661 | } | |
2662 | #endif /* CONFIG_NUMA_BALANCING */ | |
2663 | ||
1da177e4 LT |
2664 | /* assumes fs == KERNEL_DS */ |
2665 | void __init numa_policy_init(void) | |
2666 | { | |
b71636e2 PM |
2667 | nodemask_t interleave_nodes; |
2668 | unsigned long largest = 0; | |
2669 | int nid, prefer = 0; | |
2670 | ||
1da177e4 LT |
2671 | policy_cache = kmem_cache_create("numa_policy", |
2672 | sizeof(struct mempolicy), | |
20c2df83 | 2673 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2674 | |
2675 | sn_cache = kmem_cache_create("shared_policy_node", | |
2676 | sizeof(struct sp_node), | |
20c2df83 | 2677 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2678 | |
5606e387 MG |
2679 | for_each_node(nid) { |
2680 | preferred_node_policy[nid] = (struct mempolicy) { | |
2681 | .refcnt = ATOMIC_INIT(1), | |
2682 | .mode = MPOL_PREFERRED, | |
2683 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2684 | .v = { .preferred_node = nid, }, | |
2685 | }; | |
2686 | } | |
2687 | ||
b71636e2 PM |
2688 | /* |
2689 | * Set interleaving policy for system init. Interleaving is only | |
2690 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2691 | * fall back to the largest node if they're all smaller. | |
2692 | */ | |
2693 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2694 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2695 | unsigned long total_pages = node_present_pages(nid); |
2696 | ||
2697 | /* Preserve the largest node */ | |
2698 | if (largest < total_pages) { | |
2699 | largest = total_pages; | |
2700 | prefer = nid; | |
2701 | } | |
2702 | ||
2703 | /* Interleave this node? */ | |
2704 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2705 | node_set(nid, interleave_nodes); | |
2706 | } | |
2707 | ||
2708 | /* All too small, use the largest */ | |
2709 | if (unlikely(nodes_empty(interleave_nodes))) | |
2710 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2711 | |
028fec41 | 2712 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2713 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2714 | |
2715 | check_numabalancing_enable(); | |
1da177e4 LT |
2716 | } |
2717 | ||
8bccd85f | 2718 | /* Reset policy of current process to default */ |
1da177e4 LT |
2719 | void numa_default_policy(void) |
2720 | { | |
028fec41 | 2721 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2722 | } |
68860ec1 | 2723 | |
095f1fc4 LS |
2724 | /* |
2725 | * Parse and format mempolicy from/to strings | |
2726 | */ | |
2727 | ||
1a75a6c8 | 2728 | /* |
f2a07f40 | 2729 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2730 | */ |
345ace9c LS |
2731 | static const char * const policy_modes[] = |
2732 | { | |
2733 | [MPOL_DEFAULT] = "default", | |
2734 | [MPOL_PREFERRED] = "prefer", | |
2735 | [MPOL_BIND] = "bind", | |
2736 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2737 | [MPOL_LOCAL] = "local", |
345ace9c | 2738 | }; |
1a75a6c8 | 2739 | |
095f1fc4 LS |
2740 | |
2741 | #ifdef CONFIG_TMPFS | |
2742 | /** | |
f2a07f40 | 2743 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2744 | * @str: string containing mempolicy to parse |
71fe804b | 2745 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2746 | * |
2747 | * Format of input: | |
2748 | * <mode>[=<flags>][:<nodelist>] | |
2749 | * | |
71fe804b | 2750 | * On success, returns 0, else 1 |
095f1fc4 | 2751 | */ |
a7a88b23 | 2752 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2753 | { |
71fe804b | 2754 | struct mempolicy *new = NULL; |
f2a07f40 | 2755 | unsigned short mode_flags; |
71fe804b | 2756 | nodemask_t nodes; |
095f1fc4 LS |
2757 | char *nodelist = strchr(str, ':'); |
2758 | char *flags = strchr(str, '='); | |
dedf2c73 | 2759 | int err = 1, mode; |
095f1fc4 LS |
2760 | |
2761 | if (nodelist) { | |
2762 | /* NUL-terminate mode or flags string */ | |
2763 | *nodelist++ = '\0'; | |
71fe804b | 2764 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2765 | goto out; |
01f13bd6 | 2766 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2767 | goto out; |
71fe804b LS |
2768 | } else |
2769 | nodes_clear(nodes); | |
2770 | ||
095f1fc4 LS |
2771 | if (flags) |
2772 | *flags++ = '\0'; /* terminate mode string */ | |
2773 | ||
dedf2c73 | 2774 | mode = match_string(policy_modes, MPOL_MAX, str); |
2775 | if (mode < 0) | |
095f1fc4 LS |
2776 | goto out; |
2777 | ||
71fe804b | 2778 | switch (mode) { |
095f1fc4 | 2779 | case MPOL_PREFERRED: |
71fe804b LS |
2780 | /* |
2781 | * Insist on a nodelist of one node only | |
2782 | */ | |
095f1fc4 LS |
2783 | if (nodelist) { |
2784 | char *rest = nodelist; | |
2785 | while (isdigit(*rest)) | |
2786 | rest++; | |
926f2ae0 KM |
2787 | if (*rest) |
2788 | goto out; | |
095f1fc4 LS |
2789 | } |
2790 | break; | |
095f1fc4 LS |
2791 | case MPOL_INTERLEAVE: |
2792 | /* | |
2793 | * Default to online nodes with memory if no nodelist | |
2794 | */ | |
2795 | if (!nodelist) | |
01f13bd6 | 2796 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2797 | break; |
71fe804b | 2798 | case MPOL_LOCAL: |
3f226aa1 | 2799 | /* |
71fe804b | 2800 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2801 | */ |
71fe804b | 2802 | if (nodelist) |
3f226aa1 | 2803 | goto out; |
71fe804b | 2804 | mode = MPOL_PREFERRED; |
3f226aa1 | 2805 | break; |
413b43de RT |
2806 | case MPOL_DEFAULT: |
2807 | /* | |
2808 | * Insist on a empty nodelist | |
2809 | */ | |
2810 | if (!nodelist) | |
2811 | err = 0; | |
2812 | goto out; | |
d69b2e63 KM |
2813 | case MPOL_BIND: |
2814 | /* | |
2815 | * Insist on a nodelist | |
2816 | */ | |
2817 | if (!nodelist) | |
2818 | goto out; | |
095f1fc4 LS |
2819 | } |
2820 | ||
71fe804b | 2821 | mode_flags = 0; |
095f1fc4 LS |
2822 | if (flags) { |
2823 | /* | |
2824 | * Currently, we only support two mutually exclusive | |
2825 | * mode flags. | |
2826 | */ | |
2827 | if (!strcmp(flags, "static")) | |
71fe804b | 2828 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2829 | else if (!strcmp(flags, "relative")) |
71fe804b | 2830 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2831 | else |
926f2ae0 | 2832 | goto out; |
095f1fc4 | 2833 | } |
71fe804b LS |
2834 | |
2835 | new = mpol_new(mode, mode_flags, &nodes); | |
2836 | if (IS_ERR(new)) | |
926f2ae0 KM |
2837 | goto out; |
2838 | ||
f2a07f40 HD |
2839 | /* |
2840 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2841 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2842 | */ | |
2843 | if (mode != MPOL_PREFERRED) | |
2844 | new->v.nodes = nodes; | |
2845 | else if (nodelist) | |
2846 | new->v.preferred_node = first_node(nodes); | |
2847 | else | |
2848 | new->flags |= MPOL_F_LOCAL; | |
2849 | ||
2850 | /* | |
2851 | * Save nodes for contextualization: this will be used to "clone" | |
2852 | * the mempolicy in a specific context [cpuset] at a later time. | |
2853 | */ | |
2854 | new->w.user_nodemask = nodes; | |
2855 | ||
926f2ae0 | 2856 | err = 0; |
71fe804b | 2857 | |
095f1fc4 LS |
2858 | out: |
2859 | /* Restore string for error message */ | |
2860 | if (nodelist) | |
2861 | *--nodelist = ':'; | |
2862 | if (flags) | |
2863 | *--flags = '='; | |
71fe804b LS |
2864 | if (!err) |
2865 | *mpol = new; | |
095f1fc4 LS |
2866 | return err; |
2867 | } | |
2868 | #endif /* CONFIG_TMPFS */ | |
2869 | ||
71fe804b LS |
2870 | /** |
2871 | * mpol_to_str - format a mempolicy structure for printing | |
2872 | * @buffer: to contain formatted mempolicy string | |
2873 | * @maxlen: length of @buffer | |
2874 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2875 | * |
948927ee DR |
2876 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
2877 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
2878 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 2879 | */ |
948927ee | 2880 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2881 | { |
2882 | char *p = buffer; | |
948927ee DR |
2883 | nodemask_t nodes = NODE_MASK_NONE; |
2884 | unsigned short mode = MPOL_DEFAULT; | |
2885 | unsigned short flags = 0; | |
2291990a | 2886 | |
8790c71a | 2887 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 2888 | mode = pol->mode; |
948927ee DR |
2889 | flags = pol->flags; |
2890 | } | |
bea904d5 | 2891 | |
1a75a6c8 CL |
2892 | switch (mode) { |
2893 | case MPOL_DEFAULT: | |
1a75a6c8 | 2894 | break; |
1a75a6c8 | 2895 | case MPOL_PREFERRED: |
fc36b8d3 | 2896 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 2897 | mode = MPOL_LOCAL; |
53f2556b | 2898 | else |
fc36b8d3 | 2899 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 2900 | break; |
1a75a6c8 | 2901 | case MPOL_BIND: |
1a75a6c8 | 2902 | case MPOL_INTERLEAVE: |
f2a07f40 | 2903 | nodes = pol->v.nodes; |
1a75a6c8 | 2904 | break; |
1a75a6c8 | 2905 | default: |
948927ee DR |
2906 | WARN_ON_ONCE(1); |
2907 | snprintf(p, maxlen, "unknown"); | |
2908 | return; | |
1a75a6c8 CL |
2909 | } |
2910 | ||
b7a9f420 | 2911 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 2912 | |
fc36b8d3 | 2913 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 2914 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 2915 | |
2291990a LS |
2916 | /* |
2917 | * Currently, the only defined flags are mutually exclusive | |
2918 | */ | |
f5b087b5 | 2919 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2920 | p += snprintf(p, buffer + maxlen - p, "static"); |
2921 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2922 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2923 | } |
2924 | ||
9e763e0f TH |
2925 | if (!nodes_empty(nodes)) |
2926 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
2927 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 2928 | } |