blob: eb58de19f815d07adaa0a0485308dd0095d39423 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Simple NUMA memory policy for the Linux kernel.
3 *
4 * Copyright 2003,2004 Andi Kleen, SuSE Labs.
Christoph Lameter8bccd852005-10-29 18:16:59 -07005 * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * Subject to the GNU Public License, version 2.
7 *
8 * NUMA policy allows the user to give hints in which node(s) memory should
9 * be allocated.
10 *
11 * Support four policies per VMA and per process:
12 *
13 * The VMA policy has priority over the process policy for a page fault.
14 *
15 * interleave Allocate memory interleaved over a set of nodes,
16 * with normal fallback if it fails.
17 * For VMA based allocations this interleaves based on the
18 * offset into the backing object or offset into the mapping
19 * for anonymous memory. For process policy an process counter
20 * is used.
Christoph Lameter8bccd852005-10-29 18:16:59 -070021 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 * bind Only allocate memory on a specific set of nodes,
23 * no fallback.
Christoph Lameter8bccd852005-10-29 18:16:59 -070024 * 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 * preferred Try a specific node first before normal fallback.
David Rientjes00ef2d22013-02-22 16:35:36 -080029 * As a special case NUMA_NO_NODE here means do the allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * 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.
Christoph Lameter8bccd852005-10-29 18:16:59 -070033 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070066*/
67
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070068#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/mempolicy.h>
71#include <linux/mm.h>
72#include <linux/highmem.h>
73#include <linux/hugetlb.h>
74#include <linux/kernel.h>
75#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/nodemask.h>
77#include <linux/cpuset.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#include <linux/slab.h>
79#include <linux/string.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040080#include <linux/export.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070081#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include <linux/interrupt.h>
83#include <linux/init.h>
84#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080085#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080086#include <linux/seq_file.h>
87#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080088#include <linux/migrate.h>
Hugh Dickins62b61f62009-12-14 17:59:33 -080089#include <linux/ksm.h>
Christoph Lameter95a402c2006-06-23 02:03:53 -070090#include <linux/rmap.h>
David Quigley86c3a762006-06-23 02:04:02 -070091#include <linux/security.h>
Adrian Bunkdbcb0f12007-10-16 01:26:26 -070092#include <linux/syscalls.h>
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -070093#include <linux/ctype.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080094#include <linux/mm_inline.h>
Lee Schermerhornb24f53a2012-10-25 14:16:32 +020095#include <linux/mmu_notifier.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070096#include <linux/printk.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080097
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#include <asm/tlbflush.h>
99#include <asm/uaccess.h>
Michal Hocko778d3b02011-07-26 16:08:30 -0700100#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Nick Piggin62695a82008-10-18 20:26:09 -0700102#include "internal.h"
103
Christoph Lameter38e35862006-01-08 01:01:01 -0800104/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800105#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800106#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800107
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800108static struct kmem_cache *policy_cache;
109static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/* Highest zone. An specific allocation for a zone below that is not
112 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800113enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700115/*
116 * run-time system-wide default policy => local allocation
117 */
H Hartley Sweetene754d792011-10-31 17:09:23 -0700118static struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700120 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700121 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122};
123
Mel Gorman5606e382012-11-02 18:19:13 +0000124static struct mempolicy preferred_node_policy[MAX_NUMNODES];
125
126static struct mempolicy *get_task_policy(struct task_struct *p)
127{
128 struct mempolicy *pol = p->mempolicy;
Mel Gorman5606e382012-11-02 18:19:13 +0000129
130 if (!pol) {
Jianguo Wu1da6f0e2013-09-11 14:21:25 -0700131 int node = numa_node_id();
Mel Gorman5606e382012-11-02 18:19:13 +0000132
Jianguo Wu1da6f0e2013-09-11 14:21:25 -0700133 if (node != NUMA_NO_NODE) {
134 pol = &preferred_node_policy[node];
135 /*
136 * preferred_node_policy is not initialised early in
137 * boot
138 */
139 if (!pol->mode)
140 pol = NULL;
141 }
Mel Gorman5606e382012-11-02 18:19:13 +0000142 }
143
144 return pol;
145}
146
David Rientjes37012942008-04-28 02:12:33 -0700147static const struct mempolicy_operations {
148 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700149 /*
150 * If read-side task has no lock to protect task->mempolicy, write-side
151 * task will rebind the task->mempolicy by two step. The first step is
152 * setting all the newly nodes, and the second step is cleaning all the
153 * disallowed nodes. In this way, we can avoid finding no node to alloc
154 * page.
155 * If we have a lock to protect task->mempolicy in read-side, we do
156 * rebind directly.
157 *
158 * step:
159 * MPOL_REBIND_ONCE - do rebind work at once
160 * MPOL_REBIND_STEP1 - set all the newly nodes
161 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
162 */
163 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
164 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700165} mpol_ops[MPOL_MAX];
166
Mel Gorman19770b32008-04-28 02:12:18 -0700167/* Check that the nodemask contains at least one populated zone */
David Rientjes37012942008-04-28 02:12:33 -0700168static int is_valid_nodemask(const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Lai Jiangshand3eb1572013-02-22 16:33:22 -0800170 return nodes_intersects(*nodemask, node_states[N_MEMORY]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
David Rientjesf5b087b2008-04-28 02:12:27 -0700173static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
174{
Bob Liu6d556292010-05-24 14:31:59 -0700175 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700176}
177
178static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
179 const nodemask_t *rel)
180{
181 nodemask_t tmp;
182 nodes_fold(tmp, *orig, nodes_weight(*rel));
183 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700184}
185
David Rientjes37012942008-04-28 02:12:33 -0700186static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
187{
188 if (nodes_empty(*nodes))
189 return -EINVAL;
190 pol->v.nodes = *nodes;
191 return 0;
192}
193
194static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
195{
196 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700197 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700198 else if (nodes_empty(*nodes))
199 return -EINVAL; /* no allowed nodes */
200 else
201 pol->v.preferred_node = first_node(*nodes);
202 return 0;
203}
204
205static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
206{
207 if (!is_valid_nodemask(nodes))
208 return -EINVAL;
209 pol->v.nodes = *nodes;
210 return 0;
211}
212
Miao Xie58568d22009-06-16 15:31:49 -0700213/*
214 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
215 * any, for the new policy. mpol_new() has already validated the nodes
216 * parameter with respect to the policy mode and flags. But, we need to
217 * handle an empty nodemask with MPOL_PREFERRED here.
218 *
219 * Must be called holding task's alloc_lock to protect task's mems_allowed
220 * and mempolicy. May also be called holding the mmap_semaphore for write.
221 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700222static int mpol_set_nodemask(struct mempolicy *pol,
223 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700224{
Miao Xie58568d22009-06-16 15:31:49 -0700225 int ret;
226
227 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
228 if (pol == NULL)
229 return 0;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800230 /* Check N_MEMORY */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700231 nodes_and(nsc->mask1,
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800232 cpuset_current_mems_allowed, node_states[N_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700233
234 VM_BUG_ON(!nodes);
235 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
236 nodes = NULL; /* explicit local allocation */
237 else {
238 if (pol->flags & MPOL_F_RELATIVE_NODES)
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700239 mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700240 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700241 nodes_and(nsc->mask2, *nodes, nsc->mask1);
242
Miao Xie58568d22009-06-16 15:31:49 -0700243 if (mpol_store_user_nodemask(pol))
244 pol->w.user_nodemask = *nodes;
245 else
246 pol->w.cpuset_mems_allowed =
247 cpuset_current_mems_allowed;
248 }
249
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700250 if (nodes)
251 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
252 else
253 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700254 return ret;
255}
256
257/*
258 * This function just creates a new policy, does some check and simple
259 * initialization. You must invoke mpol_set_nodemask() to set nodes.
260 */
David Rientjes028fec42008-04-28 02:12:25 -0700261static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
262 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
264 struct mempolicy *policy;
265
David Rientjes028fec42008-04-28 02:12:25 -0700266 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
David Rientjes00ef2d22013-02-22 16:35:36 -0800267 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
Paul Mundt140d5a42007-07-15 23:38:16 -0700268
David Rientjes3e1f0642008-04-28 02:12:34 -0700269 if (mode == MPOL_DEFAULT) {
270 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700271 return ERR_PTR(-EINVAL);
Lee Schermerhornd3a71032012-10-25 14:16:29 +0200272 return NULL;
David Rientjes37012942008-04-28 02:12:33 -0700273 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700274 VM_BUG_ON(!nodes);
275
276 /*
277 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
278 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
279 * All other modes require a valid pointer to a non-empty nodemask.
280 */
281 if (mode == MPOL_PREFERRED) {
282 if (nodes_empty(*nodes)) {
283 if (((flags & MPOL_F_STATIC_NODES) ||
284 (flags & MPOL_F_RELATIVE_NODES)))
285 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700286 }
Peter Zijlstra479e2802012-10-25 14:16:28 +0200287 } else if (mode == MPOL_LOCAL) {
288 if (!nodes_empty(*nodes))
289 return ERR_PTR(-EINVAL);
290 mode = MPOL_PREFERRED;
David Rientjes3e1f0642008-04-28 02:12:34 -0700291 } else if (nodes_empty(*nodes))
292 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
294 if (!policy)
295 return ERR_PTR(-ENOMEM);
296 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700297 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700298 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700299
David Rientjes37012942008-04-28 02:12:33 -0700300 return policy;
301}
302
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700303/* Slow path of a mpol destructor. */
304void __mpol_put(struct mempolicy *p)
305{
306 if (!atomic_dec_and_test(&p->refcnt))
307 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700308 kmem_cache_free(policy_cache, p);
309}
310
Miao Xie708c1bb2010-05-24 14:32:07 -0700311static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
312 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700313{
314}
315
Miao Xie708c1bb2010-05-24 14:32:07 -0700316/*
317 * step:
318 * MPOL_REBIND_ONCE - do rebind work at once
319 * MPOL_REBIND_STEP1 - set all the newly nodes
320 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
321 */
322static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
323 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700324{
325 nodemask_t tmp;
326
327 if (pol->flags & MPOL_F_STATIC_NODES)
328 nodes_and(tmp, pol->w.user_nodemask, *nodes);
329 else if (pol->flags & MPOL_F_RELATIVE_NODES)
330 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
331 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700332 /*
333 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
334 * result
335 */
336 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
337 nodes_remap(tmp, pol->v.nodes,
338 pol->w.cpuset_mems_allowed, *nodes);
339 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
340 } else if (step == MPOL_REBIND_STEP2) {
341 tmp = pol->w.cpuset_mems_allowed;
342 pol->w.cpuset_mems_allowed = *nodes;
343 } else
344 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700345 }
346
Miao Xie708c1bb2010-05-24 14:32:07 -0700347 if (nodes_empty(tmp))
348 tmp = *nodes;
349
350 if (step == MPOL_REBIND_STEP1)
351 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
352 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
353 pol->v.nodes = tmp;
354 else
355 BUG();
356
David Rientjes37012942008-04-28 02:12:33 -0700357 if (!node_isset(current->il_next, tmp)) {
358 current->il_next = next_node(current->il_next, tmp);
359 if (current->il_next >= MAX_NUMNODES)
360 current->il_next = first_node(tmp);
361 if (current->il_next >= MAX_NUMNODES)
362 current->il_next = numa_node_id();
363 }
364}
365
366static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700367 const nodemask_t *nodes,
368 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700369{
370 nodemask_t tmp;
371
David Rientjes37012942008-04-28 02:12:33 -0700372 if (pol->flags & MPOL_F_STATIC_NODES) {
373 int node = first_node(pol->w.user_nodemask);
374
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700375 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700376 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700377 pol->flags &= ~MPOL_F_LOCAL;
378 } else
379 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700380 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
381 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
382 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700383 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700384 pol->v.preferred_node = node_remap(pol->v.preferred_node,
385 pol->w.cpuset_mems_allowed,
386 *nodes);
387 pol->w.cpuset_mems_allowed = *nodes;
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390
Miao Xie708c1bb2010-05-24 14:32:07 -0700391/*
392 * mpol_rebind_policy - Migrate a policy to a different set of nodes
393 *
394 * If read-side task has no lock to protect task->mempolicy, write-side
395 * task will rebind the task->mempolicy by two step. The first step is
396 * setting all the newly nodes, and the second step is cleaning all the
397 * disallowed nodes. In this way, we can avoid finding no node to alloc
398 * page.
399 * If we have a lock to protect task->mempolicy in read-side, we do
400 * rebind directly.
401 *
402 * step:
403 * MPOL_REBIND_ONCE - do rebind work at once
404 * MPOL_REBIND_STEP1 - set all the newly nodes
405 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
406 */
407static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
408 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700409{
David Rientjes1d0d2682008-04-28 02:12:32 -0700410 if (!pol)
411 return;
Wang Sheng-Hui89c522c2012-05-29 15:06:16 -0700412 if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700413 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
414 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700415
416 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
417 return;
418
419 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
420 BUG();
421
422 if (step == MPOL_REBIND_STEP1)
423 pol->flags |= MPOL_F_REBINDING;
424 else if (step == MPOL_REBIND_STEP2)
425 pol->flags &= ~MPOL_F_REBINDING;
426 else if (step >= MPOL_REBIND_NSTEP)
427 BUG();
428
429 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700430}
431
432/*
433 * Wrapper for mpol_rebind_policy() that just requires task
434 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700435 *
436 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700437 */
438
Miao Xie708c1bb2010-05-24 14:32:07 -0700439void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
440 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700441{
Miao Xie708c1bb2010-05-24 14:32:07 -0700442 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700443}
444
445/*
446 * Rebind each vma in mm to new nodemask.
447 *
448 * Call holding a reference to mm. Takes mm->mmap_sem during call.
449 */
450
451void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
452{
453 struct vm_area_struct *vma;
454
455 down_write(&mm->mmap_sem);
456 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700457 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700458 up_write(&mm->mmap_sem);
459}
460
David Rientjes37012942008-04-28 02:12:33 -0700461static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
462 [MPOL_DEFAULT] = {
463 .rebind = mpol_rebind_default,
464 },
465 [MPOL_INTERLEAVE] = {
466 .create = mpol_new_interleave,
467 .rebind = mpol_rebind_nodemask,
468 },
469 [MPOL_PREFERRED] = {
470 .create = mpol_new_preferred,
471 .rebind = mpol_rebind_preferred,
472 },
473 [MPOL_BIND] = {
474 .create = mpol_new_bind,
475 .rebind = mpol_rebind_nodemask,
476 },
477};
478
Christoph Lameterfc301282006-01-18 17:42:29 -0800479static void migrate_page_add(struct page *page, struct list_head *pagelist,
480 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800481
Naoya Horiguchi98094942013-09-11 14:22:14 -0700482/*
483 * Scan through pages checking if pages follow certain conditions,
484 * and move them to the pagelist if they do.
485 */
486static int queue_pages_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800487 unsigned long addr, unsigned long end,
488 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800489 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Hugh Dickins91612e02005-06-21 17:15:07 -0700491 pte_t *orig_pte;
492 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700493 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700494
Hugh Dickins705e87c2005-10-29 18:16:27 -0700495 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700496 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800497 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800498 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700499
500 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800502 page = vm_normal_page(vma, addr, *pte);
503 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800505 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800506 * vm_normal_page() filters out zero pages, but there might
507 * still be PageReserved pages to skip, perhaps in a VDSO.
Nick Piggin053837f2006-01-18 17:42:27 -0800508 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800509 if (PageReserved(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800510 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800511 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800512 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
513 continue;
514
Stephen Wilsonb1f72d12011-05-24 17:12:43 -0700515 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800516 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800517 else
518 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700519 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700520 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700521 return addr != end;
522}
523
Naoya Horiguchi98094942013-09-11 14:22:14 -0700524static void queue_pages_hugetlb_pmd_range(struct vm_area_struct *vma,
525 pmd_t *pmd, const nodemask_t *nodes, unsigned long flags,
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700526 void *private)
527{
528#ifdef CONFIG_HUGETLB_PAGE
529 int nid;
530 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800531 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400532 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700533
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800534 ptl = huge_pte_lock(hstate_vma(vma), vma->vm_mm, (pte_t *)pmd);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400535 entry = huge_ptep_get((pte_t *)pmd);
536 if (!pte_present(entry))
537 goto unlock;
538 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700539 nid = page_to_nid(page);
540 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
541 goto unlock;
542 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
543 if (flags & (MPOL_MF_MOVE_ALL) ||
544 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
545 isolate_huge_page(page, private);
546unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800547 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700548#else
549 BUG();
550#endif
551}
552
Naoya Horiguchi98094942013-09-11 14:22:14 -0700553static inline int queue_pages_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800554 unsigned long addr, unsigned long end,
555 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800556 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700557{
558 pmd_t *pmd;
559 unsigned long next;
560
561 pmd = pmd_offset(pud, addr);
562 do {
563 next = pmd_addr_end(addr, end);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700564 if (!pmd_present(*pmd))
565 continue;
566 if (pmd_huge(*pmd) && is_vm_hugetlb_page(vma)) {
Naoya Horiguchi98094942013-09-11 14:22:14 -0700567 queue_pages_hugetlb_pmd_range(vma, pmd, nodes,
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700568 flags, private);
569 continue;
570 }
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800571 split_huge_page_pmd(vma, addr, pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700572 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Hugh Dickins91612e02005-06-21 17:15:07 -0700573 continue;
Naoya Horiguchi98094942013-09-11 14:22:14 -0700574 if (queue_pages_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800575 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700576 return -EIO;
577 } while (pmd++, addr = next, addr != end);
578 return 0;
579}
580
Naoya Horiguchi98094942013-09-11 14:22:14 -0700581static inline int queue_pages_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800582 unsigned long addr, unsigned long end,
583 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800584 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700585{
586 pud_t *pud;
587 unsigned long next;
588
589 pud = pud_offset(pgd, addr);
590 do {
591 next = pud_addr_end(addr, end);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700592 if (pud_huge(*pud) && is_vm_hugetlb_page(vma))
593 continue;
Hugh Dickins91612e02005-06-21 17:15:07 -0700594 if (pud_none_or_clear_bad(pud))
595 continue;
Naoya Horiguchi98094942013-09-11 14:22:14 -0700596 if (queue_pages_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800597 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700598 return -EIO;
599 } while (pud++, addr = next, addr != end);
600 return 0;
601}
602
Naoya Horiguchi98094942013-09-11 14:22:14 -0700603static inline int queue_pages_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800604 unsigned long addr, unsigned long end,
605 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800606 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700607{
608 pgd_t *pgd;
609 unsigned long next;
610
Nick Pigginb5810032005-10-29 18:16:12 -0700611 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700612 do {
613 next = pgd_addr_end(addr, end);
614 if (pgd_none_or_clear_bad(pgd))
615 continue;
Naoya Horiguchi98094942013-09-11 14:22:14 -0700616 if (queue_pages_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800617 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700618 return -EIO;
619 } while (pgd++, addr = next, addr != end);
620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530623#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200624/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200625 * This is used to mark a range of virtual addresses to be inaccessible.
626 * These are later cleared by a NUMA hinting fault. Depending on these
627 * faults, pages may be migrated for better NUMA placement.
628 *
629 * This is assuming that NUMA faults are handled using PROT_NONE. If
630 * an architecture makes a different choice, it will need further
631 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200632 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200633unsigned long change_prot_numa(struct vm_area_struct *vma,
634 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200635{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200636 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200637
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200638 nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000639 if (nr_updated)
640 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200641
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200642 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200643}
644#else
645static unsigned long change_prot_numa(struct vm_area_struct *vma,
646 unsigned long addr, unsigned long end)
647{
648 return 0;
649}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530650#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200651
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800652/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700653 * Walk through page tables and collect pages to be migrated.
654 *
655 * If pages found in a given range are on a set of nodes (determined by
656 * @nodes and @flags,) it's isolated and queued to the pagelist which is
657 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800658 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700659static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700660queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800661 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700663 int err = 0;
664 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700666 vma = find_vma(mm, start);
667 if (!vma)
668 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 prev = NULL;
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700670 for (; vma && vma->vm_start < end; vma = vma->vm_next) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200671 unsigned long endvma = vma->vm_end;
672
673 if (endvma > end)
674 endvma = end;
675 if (vma->vm_start > start)
676 start = vma->vm_start;
677
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800678 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
679 if (!vma->vm_next && vma->vm_end < end)
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700680 return -EFAULT;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800681 if (prev && prev->vm_end < vma->vm_start)
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700682 return -EFAULT;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800683 }
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800684
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200685 if (flags & MPOL_MF_LAZY) {
686 change_prot_numa(vma, start, endvma);
687 goto next;
688 }
689
690 if ((flags & MPOL_MF_STRICT) ||
691 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
692 vma_migratable(vma))) {
693
Naoya Horiguchi98094942013-09-11 14:22:14 -0700694 err = queue_pages_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800695 flags, private);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700696 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200699next:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 prev = vma;
701 }
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700702 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700705/*
706 * Apply policy to a single VMA
707 * This must be called with the mmap_sem held for writing.
708 */
709static int vma_replace_policy(struct vm_area_struct *vma,
710 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700711{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700712 int err;
713 struct mempolicy *old;
714 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700715
716 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
717 vma->vm_start, vma->vm_end, vma->vm_pgoff,
718 vma->vm_ops, vma->vm_file,
719 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
720
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700721 new = mpol_dup(pol);
722 if (IS_ERR(new))
723 return PTR_ERR(new);
724
725 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700726 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700727 if (err)
728 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700729 }
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700730
731 old = vma->vm_policy;
732 vma->vm_policy = new; /* protected by mmap_sem */
733 mpol_put(old);
734
735 return 0;
736 err_out:
737 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700738 return err;
739}
740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800742static int mbind_range(struct mm_struct *mm, unsigned long start,
743 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800746 struct vm_area_struct *prev;
747 struct vm_area_struct *vma;
748 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800749 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800750 unsigned long vmstart;
751 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Linus Torvalds097d5912012-03-06 18:23:36 -0800753 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800754 if (!vma || vma->vm_start > start)
755 return -EFAULT;
756
Linus Torvalds097d5912012-03-06 18:23:36 -0800757 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800758 if (start > vma->vm_start)
759 prev = vma;
760
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800761 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800763 vmstart = max(start, vma->vm_start);
764 vmend = min(end, vma->vm_end);
765
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800766 if (mpol_equal(vma_policy(vma), new_pol))
767 continue;
768
769 pgoff = vma->vm_pgoff +
770 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800771 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800772 vma->anon_vma, vma->vm_file, pgoff,
Caspar Zhang8aacc9f2011-09-14 16:20:58 -0700773 new_pol);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800774 if (prev) {
775 vma = prev;
776 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700777 if (mpol_equal(vma_policy(vma), new_pol))
778 continue;
779 /* vma_merge() joined vma && vma->next, case 8 */
780 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800781 }
782 if (vma->vm_start != vmstart) {
783 err = split_vma(vma->vm_mm, vma, vmstart, 1);
784 if (err)
785 goto out;
786 }
787 if (vma->vm_end != vmend) {
788 err = split_vma(vma->vm_mm, vma, vmend, 0);
789 if (err)
790 goto out;
791 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700792 replace:
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700793 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700794 if (err)
795 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800797
798 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return err;
800}
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700803static long do_set_mempolicy(unsigned short mode, unsigned short flags,
804 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Miao Xie58568d22009-06-16 15:31:49 -0700806 struct mempolicy *new, *old;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700807 struct mm_struct *mm = current->mm;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700808 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700809 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700811 if (!scratch)
812 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700813
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700814 new = mpol_new(mode, flags, nodes);
815 if (IS_ERR(new)) {
816 ret = PTR_ERR(new);
817 goto out;
818 }
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700819 /*
820 * prevent changing our mempolicy while show_numa_maps()
821 * is using it.
822 * Note: do_set_mempolicy() can be called at init time
823 * with no 'mm'.
824 */
825 if (mm)
826 down_write(&mm->mmap_sem);
Miao Xie58568d22009-06-16 15:31:49 -0700827 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700828 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700829 if (ret) {
830 task_unlock(current);
831 if (mm)
832 up_write(&mm->mmap_sem);
833 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700834 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700835 }
836 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 current->mempolicy = new;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700838 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700839 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700840 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700841 task_unlock(current);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700842 if (mm)
843 up_write(&mm->mmap_sem);
844
Miao Xie58568d22009-06-16 15:31:49 -0700845 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700846 ret = 0;
847out:
848 NODEMASK_SCRATCH_FREE(scratch);
849 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850}
851
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700852/*
853 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700854 *
855 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700856 */
857static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700859 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700860 if (p == &default_policy)
861 return;
862
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700863 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700864 case MPOL_BIND:
865 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700867 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 break;
869 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700870 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700871 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700872 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 break;
874 default:
875 BUG();
876 }
877}
878
879static int lookup_node(struct mm_struct *mm, unsigned long addr)
880{
881 struct page *p;
882 int err;
883
884 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
885 if (err >= 0) {
886 err = page_to_nid(p);
887 put_page(p);
888 }
889 return err;
890}
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700893static long do_get_mempolicy(int *policy, nodemask_t *nmask,
894 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700896 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct mm_struct *mm = current->mm;
898 struct vm_area_struct *vma = NULL;
899 struct mempolicy *pol = current->mempolicy;
900
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700901 if (flags &
902 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700904
905 if (flags & MPOL_F_MEMS_ALLOWED) {
906 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
907 return -EINVAL;
908 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700909 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700910 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700911 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700912 return 0;
913 }
914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700916 /*
917 * Do NOT fall back to task policy if the
918 * vma/shared policy at addr is NULL. We
919 * want to return MPOL_DEFAULT in this case.
920 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 down_read(&mm->mmap_sem);
922 vma = find_vma_intersection(mm, addr, addr+1);
923 if (!vma) {
924 up_read(&mm->mmap_sem);
925 return -EFAULT;
926 }
927 if (vma->vm_ops && vma->vm_ops->get_policy)
928 pol = vma->vm_ops->get_policy(vma, addr);
929 else
930 pol = vma->vm_policy;
931 } else if (addr)
932 return -EINVAL;
933
934 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700935 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 if (flags & MPOL_F_NODE) {
938 if (flags & MPOL_F_ADDR) {
939 err = lookup_node(mm, addr);
940 if (err < 0)
941 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700942 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700944 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700945 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 } else {
947 err = -EINVAL;
948 goto out;
949 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700950 } else {
951 *policy = pol == &default_policy ? MPOL_DEFAULT :
952 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700953 /*
954 * Internal mempolicy flags must be masked off before exposing
955 * the policy to userspace.
956 */
957 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 if (vma) {
961 up_read(&current->mm->mmap_sem);
962 vma = NULL;
963 }
964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700966 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700967 if (mpol_store_user_nodemask(pol)) {
968 *nmask = pol->w.user_nodemask;
969 } else {
970 task_lock(current);
971 get_policy_nodemask(pol, nmask);
972 task_unlock(current);
973 }
Miao Xie58568d22009-06-16 15:31:49 -0700974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700977 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (vma)
979 up_read(&current->mm->mmap_sem);
980 return err;
981}
982
Christoph Lameterb20a3502006-03-22 00:09:12 -0800983#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700984/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800985 * page migration
986 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800987static void migrate_page_add(struct page *page, struct list_head *pagelist,
988 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800989{
990 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800991 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800992 */
Nick Piggin62695a82008-10-18 20:26:09 -0700993 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
994 if (!isolate_lru_page(page)) {
995 list_add_tail(&page->lru, pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800996 inc_zone_page_state(page, NR_ISOLATED_ANON +
997 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700998 }
999 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001000}
1001
Christoph Lameter742755a2006-06-23 02:03:55 -07001002static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001003{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001004 if (PageHuge(page))
1005 return alloc_huge_page_node(page_hstate(compound_head(page)),
1006 node);
1007 else
1008 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001009}
1010
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001011/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001012 * Migrate pages from one node to a target node.
1013 * Returns error or the number of pages not migrated.
1014 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001015static int migrate_to_node(struct mm_struct *mm, int source, int dest,
1016 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001017{
1018 nodemask_t nmask;
1019 LIST_HEAD(pagelist);
1020 int err = 0;
1021
1022 nodes_clear(nmask);
1023 node_set(source, nmask);
1024
Minchan Kim08270802012-10-08 16:33:38 -07001025 /*
1026 * This does not "check" the range but isolates all pages that
1027 * need migration. Between passing in the full user address
1028 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
1029 */
1030 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -07001031 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001032 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1033
Minchan Kimcf608ac2010-10-26 14:21:29 -07001034 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001035 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001036 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001037 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001038 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001039 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001040
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001041 return err;
1042}
1043
1044/*
1045 * Move pages between the two nodesets so as to preserve the physical
1046 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001047 *
1048 * Returns the number of page that could not be moved.
1049 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001050int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1051 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001052{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001053 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001054 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001055 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001056
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001057 err = migrate_prep();
1058 if (err)
1059 return err;
1060
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001061 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001062
Andrew Morton0ce72d42012-05-29 15:06:24 -07001063 err = migrate_vmas(mm, from, to, flags);
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001064 if (err)
1065 goto out;
1066
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001067 /*
1068 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1069 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1070 * bit in 'tmp', and return that <source, dest> pair for migration.
1071 * The pair of nodemasks 'to' and 'from' define the map.
1072 *
1073 * If no pair of bits is found that way, fallback to picking some
1074 * pair of 'source' and 'dest' bits that are not the same. If the
1075 * 'source' and 'dest' bits are the same, this represents a node
1076 * that will be migrating to itself, so no pages need move.
1077 *
1078 * If no bits are left in 'tmp', or if all remaining bits left
1079 * in 'tmp' correspond to the same bit in 'to', return false
1080 * (nothing left to migrate).
1081 *
1082 * This lets us pick a pair of nodes to migrate between, such that
1083 * if possible the dest node is not already occupied by some other
1084 * source node, minimizing the risk of overloading the memory on a
1085 * node that would happen if we migrated incoming memory to a node
1086 * before migrating outgoing memory source that same node.
1087 *
1088 * A single scan of tmp is sufficient. As we go, we remember the
1089 * most recent <s, d> pair that moved (s != d). If we find a pair
1090 * that not only moved, but what's better, moved to an empty slot
1091 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001092 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001093 * most recent <s, d> pair that moved. If we get all the way through
1094 * the scan of tmp without finding any node that moved, much less
1095 * moved to an empty node, then there is nothing left worth migrating.
1096 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001097
Andrew Morton0ce72d42012-05-29 15:06:24 -07001098 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001099 while (!nodes_empty(tmp)) {
1100 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001101 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001102 int dest = 0;
1103
1104 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001105
1106 /*
1107 * do_migrate_pages() tries to maintain the relative
1108 * node relationship of the pages established between
1109 * threads and memory areas.
1110 *
1111 * However if the number of source nodes is not equal to
1112 * the number of destination nodes we can not preserve
1113 * this node relative relationship. In that case, skip
1114 * copying memory from a node that is in the destination
1115 * mask.
1116 *
1117 * Example: [2,3,4] -> [3,4,5] moves everything.
1118 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1119 */
1120
Andrew Morton0ce72d42012-05-29 15:06:24 -07001121 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1122 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001123 continue;
1124
Andrew Morton0ce72d42012-05-29 15:06:24 -07001125 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001126 if (s == d)
1127 continue;
1128
1129 source = s; /* Node moved. Memorize */
1130 dest = d;
1131
1132 /* dest not in remaining from nodes? */
1133 if (!node_isset(dest, tmp))
1134 break;
1135 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001136 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001137 break;
1138
1139 node_clear(source, tmp);
1140 err = migrate_to_node(mm, source, dest, flags);
1141 if (err > 0)
1142 busy += err;
1143 if (err < 0)
1144 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001145 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001146out:
Christoph Lameter39743882006-01-08 01:00:51 -08001147 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001148 if (err < 0)
1149 return err;
1150 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001151
Christoph Lameter39743882006-01-08 01:00:51 -08001152}
1153
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001154/*
1155 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001156 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001157 * Search forward from there, if not. N.B., this assumes that the
1158 * list of pages handed to migrate_pages()--which is how we get here--
1159 * is in virtual address order.
1160 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001161static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001162{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001163 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001164 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001165
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001166 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001167 while (vma) {
1168 address = page_address_in_vma(page, vma);
1169 if (address != -EFAULT)
1170 break;
1171 vma = vma->vm_next;
1172 }
1173
Wanpeng Li11c731e2013-12-18 17:08:56 -08001174 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001175 BUG_ON(!vma);
1176 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001177 }
1178 /*
1179 * if !vma, alloc_page_vma() will use task or system default policy
1180 */
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001181 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001182}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001183#else
1184
1185static void migrate_page_add(struct page *page, struct list_head *pagelist,
1186 unsigned long flags)
1187{
1188}
1189
Andrew Morton0ce72d42012-05-29 15:06:24 -07001190int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1191 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001192{
1193 return -ENOSYS;
1194}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001195
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001196static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001197{
1198 return NULL;
1199}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001200#endif
1201
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001202static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001203 unsigned short mode, unsigned short mode_flags,
1204 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001205{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001206 struct mm_struct *mm = current->mm;
1207 struct mempolicy *new;
1208 unsigned long end;
1209 int err;
1210 LIST_HEAD(pagelist);
1211
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001212 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001213 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001214 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001215 return -EPERM;
1216
1217 if (start & ~PAGE_MASK)
1218 return -EINVAL;
1219
1220 if (mode == MPOL_DEFAULT)
1221 flags &= ~MPOL_MF_STRICT;
1222
1223 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1224 end = start + len;
1225
1226 if (end < start)
1227 return -EINVAL;
1228 if (end == start)
1229 return 0;
1230
David Rientjes028fec42008-04-28 02:12:25 -07001231 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001232 if (IS_ERR(new))
1233 return PTR_ERR(new);
1234
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001235 if (flags & MPOL_MF_LAZY)
1236 new->flags |= MPOL_F_MOF;
1237
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001238 /*
1239 * If we are using the default policy then operation
1240 * on discontinuous address spaces is okay after all
1241 */
1242 if (!new)
1243 flags |= MPOL_MF_DISCONTIG_OK;
1244
David Rientjes028fec42008-04-28 02:12:25 -07001245 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1246 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001247 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001248
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001249 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1250
1251 err = migrate_prep();
1252 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001253 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001254 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001255 {
1256 NODEMASK_SCRATCH(scratch);
1257 if (scratch) {
1258 down_write(&mm->mmap_sem);
1259 task_lock(current);
1260 err = mpol_set_nodemask(new, nmask, scratch);
1261 task_unlock(current);
1262 if (err)
1263 up_write(&mm->mmap_sem);
1264 } else
1265 err = -ENOMEM;
1266 NODEMASK_SCRATCH_FREE(scratch);
1267 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001268 if (err)
1269 goto mpol_out;
1270
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001271 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001272 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001273 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001274 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001275
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001276 if (!err) {
1277 int nr_failed = 0;
1278
Minchan Kimcf608ac2010-10-26 14:21:29 -07001279 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001280 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001281 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1282 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001283 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001284 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001285 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001286
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001287 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001288 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001289 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001290 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001291
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001292 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001293 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001294 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001295 return err;
1296}
1297
Christoph Lameter39743882006-01-08 01:00:51 -08001298/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001299 * User space interface with variable sized bitmaps for nodelists.
1300 */
1301
1302/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001303static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001304 unsigned long maxnode)
1305{
1306 unsigned long k;
1307 unsigned long nlongs;
1308 unsigned long endmask;
1309
1310 --maxnode;
1311 nodes_clear(*nodes);
1312 if (maxnode == 0 || !nmask)
1313 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001314 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001315 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001316
1317 nlongs = BITS_TO_LONGS(maxnode);
1318 if ((maxnode % BITS_PER_LONG) == 0)
1319 endmask = ~0UL;
1320 else
1321 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1322
1323 /* When the user specified more nodes than supported just check
1324 if the non supported part is all zero. */
1325 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1326 if (nlongs > PAGE_SIZE/sizeof(long))
1327 return -EINVAL;
1328 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1329 unsigned long t;
1330 if (get_user(t, nmask + k))
1331 return -EFAULT;
1332 if (k == nlongs - 1) {
1333 if (t & endmask)
1334 return -EINVAL;
1335 } else if (t)
1336 return -EINVAL;
1337 }
1338 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1339 endmask = ~0UL;
1340 }
1341
1342 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1343 return -EFAULT;
1344 nodes_addr(*nodes)[nlongs-1] &= endmask;
1345 return 0;
1346}
1347
1348/* Copy a kernel node mask to user space */
1349static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1350 nodemask_t *nodes)
1351{
1352 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1353 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1354
1355 if (copy > nbytes) {
1356 if (copy > PAGE_SIZE)
1357 return -EINVAL;
1358 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1359 return -EFAULT;
1360 copy = nbytes;
1361 }
1362 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1363}
1364
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001365SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001366 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001367 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001368{
1369 nodemask_t nodes;
1370 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001371 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001372
David Rientjes028fec42008-04-28 02:12:25 -07001373 mode_flags = mode & MPOL_MODE_FLAGS;
1374 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001375 if (mode >= MPOL_MAX)
1376 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001377 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1378 (mode_flags & MPOL_F_RELATIVE_NODES))
1379 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001380 err = get_nodes(&nodes, nmask, maxnode);
1381 if (err)
1382 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001383 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001384}
1385
1386/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001387SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001388 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001389{
1390 int err;
1391 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001392 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001393
David Rientjes028fec42008-04-28 02:12:25 -07001394 flags = mode & MPOL_MODE_FLAGS;
1395 mode &= ~MPOL_MODE_FLAGS;
1396 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001397 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001398 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1399 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001400 err = get_nodes(&nodes, nmask, maxnode);
1401 if (err)
1402 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001403 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001404}
1405
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001406SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1407 const unsigned long __user *, old_nodes,
1408 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001409{
David Howellsc69e8d92008-11-14 10:39:19 +11001410 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001411 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001412 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001413 nodemask_t task_nodes;
1414 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001415 nodemask_t *old;
1416 nodemask_t *new;
1417 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001418
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001419 if (!scratch)
1420 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001421
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001422 old = &scratch->mask1;
1423 new = &scratch->mask2;
1424
1425 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001426 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001427 goto out;
1428
1429 err = get_nodes(new, new_nodes, maxnode);
1430 if (err)
1431 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001432
1433 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001434 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001435 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001436 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001437 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001438 err = -ESRCH;
1439 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001440 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001441 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001442
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001443 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001444
1445 /*
1446 * Check if this process has the right to modify the specified
1447 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001448 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001449 * userid as the target process.
1450 */
David Howellsc69e8d92008-11-14 10:39:19 +11001451 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001452 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1453 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001454 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001455 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001456 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001457 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001458 }
David Howellsc69e8d92008-11-14 10:39:19 +11001459 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001460
1461 task_nodes = cpuset_mems_allowed(task);
1462 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001463 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001464 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001465 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001466 }
1467
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08001468 if (!nodes_subset(*new, node_states[N_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001469 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001470 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001471 }
1472
David Quigley86c3a762006-06-23 02:04:02 -07001473 err = security_task_movememory(task);
1474 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001475 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001476
Christoph Lameter3268c632012-03-21 16:34:06 -07001477 mm = get_task_mm(task);
1478 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001479
1480 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001481 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001482 goto out;
1483 }
1484
1485 err = do_migrate_pages(mm, old, new,
1486 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001487
1488 mmput(mm);
1489out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001490 NODEMASK_SCRATCH_FREE(scratch);
1491
Christoph Lameter39743882006-01-08 01:00:51 -08001492 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001493
1494out_put:
1495 put_task_struct(task);
1496 goto out;
1497
Christoph Lameter39743882006-01-08 01:00:51 -08001498}
1499
1500
Christoph Lameter8bccd852005-10-29 18:16:59 -07001501/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001502SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1503 unsigned long __user *, nmask, unsigned long, maxnode,
1504 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001505{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001506 int err;
1507 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001508 nodemask_t nodes;
1509
1510 if (nmask != NULL && maxnode < MAX_NUMNODES)
1511 return -EINVAL;
1512
1513 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1514
1515 if (err)
1516 return err;
1517
1518 if (policy && put_user(pval, policy))
1519 return -EFAULT;
1520
1521 if (nmask)
1522 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1523
1524 return err;
1525}
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527#ifdef CONFIG_COMPAT
1528
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001529COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1530 compat_ulong_t __user *, nmask,
1531 compat_ulong_t, maxnode,
1532 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
1534 long err;
1535 unsigned long __user *nm = NULL;
1536 unsigned long nr_bits, alloc_size;
1537 DECLARE_BITMAP(bm, MAX_NUMNODES);
1538
1539 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1540 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1541
1542 if (nmask)
1543 nm = compat_alloc_user_space(alloc_size);
1544
1545 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1546
1547 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001548 unsigned long copy_size;
1549 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1550 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 /* ensure entire bitmap is zeroed */
1552 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1553 err |= compat_put_bitmap(nmask, bm, nr_bits);
1554 }
1555
1556 return err;
1557}
1558
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001559COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1560 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
1562 long err = 0;
1563 unsigned long __user *nm = NULL;
1564 unsigned long nr_bits, alloc_size;
1565 DECLARE_BITMAP(bm, MAX_NUMNODES);
1566
1567 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1568 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1569
1570 if (nmask) {
1571 err = compat_get_bitmap(bm, nmask, nr_bits);
1572 nm = compat_alloc_user_space(alloc_size);
1573 err |= copy_to_user(nm, bm, alloc_size);
1574 }
1575
1576 if (err)
1577 return -EFAULT;
1578
1579 return sys_set_mempolicy(mode, nm, nr_bits+1);
1580}
1581
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001582COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1583 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1584 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585{
1586 long err = 0;
1587 unsigned long __user *nm = NULL;
1588 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001589 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1592 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1593
1594 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001595 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001597 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599
1600 if (err)
1601 return -EFAULT;
1602
1603 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1604}
1605
1606#endif
1607
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001608/*
1609 * get_vma_policy(@task, @vma, @addr)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001610 * @task: task for fallback if vma policy == default
1611 * @vma: virtual memory area whose policy is sought
1612 * @addr: address in @vma for shared policy lookup
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001613 *
1614 * Returns effective policy for a VMA at specified address.
1615 * Falls back to @task or system default policy, as necessary.
David Rientjes32f85162012-10-16 17:31:23 -07001616 * Current or other task's task mempolicy and non-shared vma policies must be
1617 * protected by task_lock(task) by the caller.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001618 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1619 * count--added by the get_policy() vm_op, as appropriate--to protect against
1620 * freeing by another task. It is the caller's responsibility to free the
1621 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001622 */
Stephen Wilsond98f6cb2011-05-24 17:12:41 -07001623struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001624 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625{
Mel Gorman5606e382012-11-02 18:19:13 +00001626 struct mempolicy *pol = get_task_policy(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001629 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001630 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1631 addr);
1632 if (vpol)
1633 pol = vpol;
Mel Gorman00442ad2012-10-08 16:29:20 -07001634 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001636
1637 /*
1638 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1639 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1640 * count on these policies which will be dropped by
1641 * mpol_cond_put() later
1642 */
1643 if (mpol_needs_cond_ref(pol))
1644 mpol_get(pol);
1645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 }
1647 if (!pol)
1648 pol = &default_policy;
1649 return pol;
1650}
1651
Mel Gormanfc3147242013-10-07 11:29:09 +01001652bool vma_policy_mof(struct task_struct *task, struct vm_area_struct *vma)
1653{
1654 struct mempolicy *pol = get_task_policy(task);
1655 if (vma) {
1656 if (vma->vm_ops && vma->vm_ops->get_policy) {
1657 bool ret = false;
1658
1659 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1660 if (pol && (pol->flags & MPOL_F_MOF))
1661 ret = true;
1662 mpol_cond_put(pol);
1663
1664 return ret;
1665 } else if (vma->vm_policy) {
1666 pol = vma->vm_policy;
1667 }
1668 }
1669
1670 if (!pol)
1671 return default_policy.flags & MPOL_F_MOF;
1672
1673 return pol->flags & MPOL_F_MOF;
1674}
1675
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001676static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1677{
1678 enum zone_type dynamic_policy_zone = policy_zone;
1679
1680 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1681
1682 /*
1683 * if policy->v.nodes has movable memory only,
1684 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1685 *
1686 * policy->v.nodes is intersect with node_states[N_MEMORY].
1687 * so if the following test faile, it implies
1688 * policy->v.nodes has movable memory only.
1689 */
1690 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1691 dynamic_policy_zone = ZONE_MOVABLE;
1692
1693 return zone >= dynamic_policy_zone;
1694}
1695
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001696/*
1697 * Return a nodemask representing a mempolicy for filtering nodes for
1698 * page allocation
1699 */
1700static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001701{
1702 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001703 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001704 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001705 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1706 return &policy->v.nodes;
1707
1708 return NULL;
1709}
1710
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001711/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001712static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1713 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001715 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001717 if (!(policy->flags & MPOL_F_LOCAL))
1718 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 break;
1720 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001721 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001722 * Normally, MPOL_BIND allocations are node-local within the
1723 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001724 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001725 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001726 */
Mel Gorman19770b32008-04-28 02:12:18 -07001727 if (unlikely(gfp & __GFP_THISNODE) &&
1728 unlikely(!node_isset(nd, policy->v.nodes)))
1729 nd = first_node(policy->v.nodes);
1730 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 BUG();
1733 }
Mel Gorman0e884602008-04-28 02:12:14 -07001734 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735}
1736
1737/* Do dynamic interleaving for a process */
1738static unsigned interleave_nodes(struct mempolicy *policy)
1739{
1740 unsigned nid, next;
1741 struct task_struct *me = current;
1742
1743 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001744 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001746 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001747 if (next < MAX_NUMNODES)
1748 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 return nid;
1750}
1751
Christoph Lameterdc85da12006-01-18 17:42:36 -08001752/*
1753 * Depending on the memory policy provide a node from which to allocate the
1754 * next slab entry.
1755 */
David Rientjes2a389612014-04-07 15:37:29 -07001756unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001757{
Andi Kleene7b691b2012-06-09 02:40:03 -07001758 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001759 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001760
1761 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001762 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001763
1764 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001765 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001766 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001767
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001768 switch (policy->mode) {
1769 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001770 /*
1771 * handled MPOL_F_LOCAL above
1772 */
1773 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001774
Christoph Lameterdc85da12006-01-18 17:42:36 -08001775 case MPOL_INTERLEAVE:
1776 return interleave_nodes(policy);
1777
Mel Gormandd1a2392008-04-28 02:12:17 -07001778 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001779 /*
1780 * Follow bind policy behavior and start allocation at the
1781 * first node.
1782 */
Mel Gorman19770b32008-04-28 02:12:18 -07001783 struct zonelist *zonelist;
1784 struct zone *zone;
1785 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
David Rientjes2a389612014-04-07 15:37:29 -07001786 zonelist = &NODE_DATA(node)->node_zonelists[0];
Mel Gorman19770b32008-04-28 02:12:18 -07001787 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1788 &policy->v.nodes,
1789 &zone);
David Rientjes2a389612014-04-07 15:37:29 -07001790 return zone ? zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001791 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001792
Christoph Lameterdc85da12006-01-18 17:42:36 -08001793 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001794 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001795 }
1796}
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798/* Do static interleaving for a VMA with known offset. */
1799static unsigned offset_il_node(struct mempolicy *pol,
1800 struct vm_area_struct *vma, unsigned long off)
1801{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001802 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001803 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 int c;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001805 int nid = NUMA_NO_NODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
David Rientjesf5b087b2008-04-28 02:12:27 -07001807 if (!nnodes)
1808 return numa_node_id();
1809 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 c = 0;
1811 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001812 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 c++;
1814 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 return nid;
1816}
1817
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001818/* Determine a node number for interleave */
1819static inline unsigned interleave_nid(struct mempolicy *pol,
1820 struct vm_area_struct *vma, unsigned long addr, int shift)
1821{
1822 if (vma) {
1823 unsigned long off;
1824
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001825 /*
1826 * for small pages, there is no difference between
1827 * shift and PAGE_SHIFT, so the bit-shift is safe.
1828 * for huge pages, since vm_pgoff is in units of small
1829 * pages, we need to shift off the always 0 bits to get
1830 * a useful offset.
1831 */
1832 BUG_ON(shift < PAGE_SHIFT);
1833 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001834 off += (addr - vma->vm_start) >> shift;
1835 return offset_il_node(pol, vma, off);
1836 } else
1837 return interleave_nodes(pol);
1838}
1839
Michal Hocko778d3b02011-07-26 16:08:30 -07001840/*
1841 * Return the bit number of a random bit set in the nodemask.
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001842 * (returns NUMA_NO_NODE if nodemask is empty)
Michal Hocko778d3b02011-07-26 16:08:30 -07001843 */
1844int node_random(const nodemask_t *maskp)
1845{
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001846 int w, bit = NUMA_NO_NODE;
Michal Hocko778d3b02011-07-26 16:08:30 -07001847
1848 w = nodes_weight(*maskp);
1849 if (w)
1850 bit = bitmap_ord_to_pos(maskp->bits,
1851 get_random_int() % w, MAX_NUMNODES);
1852 return bit;
1853}
1854
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001855#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001856/*
1857 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001858 * @vma: virtual memory area whose policy is sought
1859 * @addr: address in @vma for shared policy lookup and interleave policy
1860 * @gfp_flags: for requested zone
1861 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1862 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001863 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001864 * Returns a zonelist suitable for a huge page allocation and a pointer
1865 * to the struct mempolicy for conditional unref after allocation.
1866 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1867 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001868 *
Mel Gormand26914d2014-04-03 14:47:24 -07001869 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001870 */
Mel Gorman396faf02007-07-17 04:03:13 -07001871struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001872 gfp_t gfp_flags, struct mempolicy **mpol,
1873 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001874{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001875 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001876
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001877 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001878 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001879
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001880 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1881 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001882 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001883 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001884 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001885 if ((*mpol)->mode == MPOL_BIND)
1886 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001887 }
1888 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001889}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001890
1891/*
1892 * init_nodemask_of_mempolicy
1893 *
1894 * If the current task's mempolicy is "default" [NULL], return 'false'
1895 * to indicate default policy. Otherwise, extract the policy nodemask
1896 * for 'bind' or 'interleave' policy into the argument nodemask, or
1897 * initialize the argument nodemask to contain the single node for
1898 * 'preferred' or 'local' policy and return 'true' to indicate presence
1899 * of non-default mempolicy.
1900 *
1901 * We don't bother with reference counting the mempolicy [mpol_get/put]
1902 * because the current task is examining it's own mempolicy and a task's
1903 * mempolicy is only ever changed by the task itself.
1904 *
1905 * N.B., it is the caller's responsibility to free a returned nodemask.
1906 */
1907bool init_nodemask_of_mempolicy(nodemask_t *mask)
1908{
1909 struct mempolicy *mempolicy;
1910 int nid;
1911
1912 if (!(mask && current->mempolicy))
1913 return false;
1914
Miao Xiec0ff7452010-05-24 14:32:08 -07001915 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001916 mempolicy = current->mempolicy;
1917 switch (mempolicy->mode) {
1918 case MPOL_PREFERRED:
1919 if (mempolicy->flags & MPOL_F_LOCAL)
1920 nid = numa_node_id();
1921 else
1922 nid = mempolicy->v.preferred_node;
1923 init_nodemask_of_node(mask, nid);
1924 break;
1925
1926 case MPOL_BIND:
1927 /* Fall through */
1928 case MPOL_INTERLEAVE:
1929 *mask = mempolicy->v.nodes;
1930 break;
1931
1932 default:
1933 BUG();
1934 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001935 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001936
1937 return true;
1938}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001939#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001940
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001941/*
1942 * mempolicy_nodemask_intersects
1943 *
1944 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1945 * policy. Otherwise, check for intersection between mask and the policy
1946 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1947 * policy, always return true since it may allocate elsewhere on fallback.
1948 *
1949 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1950 */
1951bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1952 const nodemask_t *mask)
1953{
1954 struct mempolicy *mempolicy;
1955 bool ret = true;
1956
1957 if (!mask)
1958 return ret;
1959 task_lock(tsk);
1960 mempolicy = tsk->mempolicy;
1961 if (!mempolicy)
1962 goto out;
1963
1964 switch (mempolicy->mode) {
1965 case MPOL_PREFERRED:
1966 /*
1967 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1968 * allocate from, they may fallback to other nodes when oom.
1969 * Thus, it's possible for tsk to have allocated memory from
1970 * nodes in mask.
1971 */
1972 break;
1973 case MPOL_BIND:
1974 case MPOL_INTERLEAVE:
1975 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1976 break;
1977 default:
1978 BUG();
1979 }
1980out:
1981 task_unlock(tsk);
1982 return ret;
1983}
1984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985/* Allocate a page in interleaved policy.
1986 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001987static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1988 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
1990 struct zonelist *zl;
1991 struct page *page;
1992
Mel Gorman0e884602008-04-28 02:12:14 -07001993 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001995 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001996 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return page;
1998}
1999
2000/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002001 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 *
2003 * @gfp:
2004 * %GFP_USER user allocation.
2005 * %GFP_KERNEL kernel allocations,
2006 * %GFP_HIGHMEM highmem/user allocations,
2007 * %GFP_FS allocation should not call back into a file system.
2008 * %GFP_ATOMIC don't sleep.
2009 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002010 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 * @vma: Pointer to VMA or NULL if not available.
2012 * @addr: Virtual Address of the allocation. Must be inside the VMA.
2013 *
2014 * This function allocates a page from the kernel page pool and applies
2015 * a NUMA policy associated with the VMA or the current process.
2016 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
2017 * mm_struct of the VMA to prevent it from going away. Should be used for
2018 * all allocations for pages that will be mapped into
2019 * user space. Returns NULL when no page can be allocated.
2020 *
2021 * Should be called with the mm_sem of the vma hold.
2022 */
2023struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002024alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Andi Kleen2f5f9482011-03-04 17:36:29 -08002025 unsigned long addr, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026{
Mel Gormancc9a6c82012-03-21 16:34:11 -07002027 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07002028 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002029 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
Mel Gormancc9a6c82012-03-21 16:34:11 -07002031retry_cpuset:
2032 pol = get_vma_policy(current, vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07002033 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002034
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002035 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08002037
Andi Kleen8eac5632011-02-25 14:44:28 -08002038 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002039 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002040 page = alloc_page_interleave(gfp, order, nid);
Mel Gormand26914d2014-04-03 14:47:24 -07002041 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002042 goto retry_cpuset;
2043
Miao Xiec0ff7452010-05-24 14:32:08 -07002044 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 }
David Rientjes212a0a62012-12-11 16:02:51 -08002046 page = __alloc_pages_nodemask(gfp, order,
2047 policy_zonelist(gfp, pol, node),
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002048 policy_nodemask(gfp, pol));
David Rientjes212a0a62012-12-11 16:02:51 -08002049 if (unlikely(mpol_needs_cond_ref(pol)))
2050 __mpol_put(pol);
Mel Gormand26914d2014-04-03 14:47:24 -07002051 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002052 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002053 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054}
2055
2056/**
2057 * alloc_pages_current - Allocate pages.
2058 *
2059 * @gfp:
2060 * %GFP_USER user allocation,
2061 * %GFP_KERNEL kernel allocation,
2062 * %GFP_HIGHMEM highmem allocation,
2063 * %GFP_FS don't call back into a file system.
2064 * %GFP_ATOMIC don't sleep.
2065 * @order: Power of two of allocation size in pages. 0 is a single page.
2066 *
2067 * Allocate a page from the kernel page pool. When not in
2068 * interrupt context and apply the current process NUMA policy.
2069 * Returns NULL when no page can be allocated.
2070 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002071 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 * 1) it's ok to take cpuset_sem (can WAIT), and
2073 * 2) allocating for current task (not interrupt).
2074 */
Al Virodd0fc662005-10-07 07:46:04 +01002075struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
Mel Gorman5606e382012-11-02 18:19:13 +00002077 struct mempolicy *pol = get_task_policy(current);
Miao Xiec0ff7452010-05-24 14:32:08 -07002078 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002079 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Christoph Lameter9b819d22006-09-25 23:31:40 -07002081 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002083
Mel Gormancc9a6c82012-03-21 16:34:11 -07002084retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002085 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002086
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002087 /*
2088 * No reference counting needed for current->mempolicy
2089 * nor system default_policy
2090 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002091 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002092 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2093 else
2094 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002095 policy_zonelist(gfp, pol, numa_node_id()),
2096 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002097
Mel Gormand26914d2014-04-03 14:47:24 -07002098 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002099 goto retry_cpuset;
2100
Miao Xiec0ff7452010-05-24 14:32:08 -07002101 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102}
2103EXPORT_SYMBOL(alloc_pages_current);
2104
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002105int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2106{
2107 struct mempolicy *pol = mpol_dup(vma_policy(src));
2108
2109 if (IS_ERR(pol))
2110 return PTR_ERR(pol);
2111 dst->vm_policy = pol;
2112 return 0;
2113}
2114
Paul Jackson42253992006-01-08 01:01:59 -08002115/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002116 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002117 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2118 * with the mems_allowed returned by cpuset_mems_allowed(). This
2119 * keeps mempolicies cpuset relative after its cpuset moves. See
2120 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002121 *
2122 * current's mempolicy may be rebinded by the other task(the task that changes
2123 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002124 */
Paul Jackson42253992006-01-08 01:01:59 -08002125
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002126/* Slow path of a mempolicy duplicate */
2127struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128{
2129 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2130
2131 if (!new)
2132 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002133
2134 /* task's mempolicy is protected by alloc_lock */
2135 if (old == current->mempolicy) {
2136 task_lock(current);
2137 *new = *old;
2138 task_unlock(current);
2139 } else
2140 *new = *old;
2141
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08002142 rcu_read_lock();
Paul Jackson42253992006-01-08 01:01:59 -08002143 if (current_cpuset_is_being_rebound()) {
2144 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002145 if (new->flags & MPOL_F_REBINDING)
2146 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2147 else
2148 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002149 }
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08002150 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 return new;
2153}
2154
2155/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002156bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157{
2158 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002159 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002160 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002161 return false;
Bob Liu19800502010-05-24 14:32:01 -07002162 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002163 return false;
Bob Liu19800502010-05-24 14:32:01 -07002164 if (mpol_store_user_nodemask(a))
2165 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002166 return false;
Bob Liu19800502010-05-24 14:32:01 -07002167
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002168 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002169 case MPOL_BIND:
2170 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002172 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002174 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 default:
2176 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002177 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 }
2179}
2180
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 * Shared memory backing store policy support.
2183 *
2184 * Remember policies even when nobody has shared memory mapped.
2185 * The policies are kept in Red-Black tree linked from the inode.
2186 * They are protected by the sp->lock spinlock, which should be held
2187 * for any accesses to the tree.
2188 */
2189
2190/* lookup first element intersecting start-end */
Mel Gorman42288fe2012-12-21 23:10:25 +00002191/* Caller holds sp->lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192static struct sp_node *
2193sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2194{
2195 struct rb_node *n = sp->root.rb_node;
2196
2197 while (n) {
2198 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2199
2200 if (start >= p->end)
2201 n = n->rb_right;
2202 else if (end <= p->start)
2203 n = n->rb_left;
2204 else
2205 break;
2206 }
2207 if (!n)
2208 return NULL;
2209 for (;;) {
2210 struct sp_node *w = NULL;
2211 struct rb_node *prev = rb_prev(n);
2212 if (!prev)
2213 break;
2214 w = rb_entry(prev, struct sp_node, nd);
2215 if (w->end <= start)
2216 break;
2217 n = prev;
2218 }
2219 return rb_entry(n, struct sp_node, nd);
2220}
2221
2222/* Insert a new shared policy into the list. */
2223/* Caller holds sp->lock */
2224static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2225{
2226 struct rb_node **p = &sp->root.rb_node;
2227 struct rb_node *parent = NULL;
2228 struct sp_node *nd;
2229
2230 while (*p) {
2231 parent = *p;
2232 nd = rb_entry(parent, struct sp_node, nd);
2233 if (new->start < nd->start)
2234 p = &(*p)->rb_left;
2235 else if (new->end > nd->end)
2236 p = &(*p)->rb_right;
2237 else
2238 BUG();
2239 }
2240 rb_link_node(&new->nd, parent, p);
2241 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002242 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002243 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244}
2245
2246/* Find shared policy intersecting idx */
2247struct mempolicy *
2248mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2249{
2250 struct mempolicy *pol = NULL;
2251 struct sp_node *sn;
2252
2253 if (!sp->root.rb_node)
2254 return NULL;
Mel Gorman42288fe2012-12-21 23:10:25 +00002255 spin_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 sn = sp_lookup(sp, idx, idx+1);
2257 if (sn) {
2258 mpol_get(sn->policy);
2259 pol = sn->policy;
2260 }
Mel Gorman42288fe2012-12-21 23:10:25 +00002261 spin_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return pol;
2263}
2264
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002265static void sp_free(struct sp_node *n)
2266{
2267 mpol_put(n->policy);
2268 kmem_cache_free(sn_cache, n);
2269}
2270
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002271/**
2272 * mpol_misplaced - check whether current page node is valid in policy
2273 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002274 * @page: page to be checked
2275 * @vma: vm area where page mapped
2276 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002277 *
2278 * Lookup current policy node id for vma,addr and "compare to" page's
2279 * node id.
2280 *
2281 * Returns:
2282 * -1 - not misplaced, page is in the right node
2283 * node - node id where the page should be
2284 *
2285 * Policy determination "mimics" alloc_page_vma().
2286 * Called from fault path where we know the vma and faulting address.
2287 */
2288int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2289{
2290 struct mempolicy *pol;
2291 struct zone *zone;
2292 int curnid = page_to_nid(page);
2293 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002294 int thiscpu = raw_smp_processor_id();
2295 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002296 int polnid = -1;
2297 int ret = -1;
2298
2299 BUG_ON(!vma);
2300
2301 pol = get_vma_policy(current, vma, addr);
2302 if (!(pol->flags & MPOL_F_MOF))
2303 goto out;
2304
2305 switch (pol->mode) {
2306 case MPOL_INTERLEAVE:
2307 BUG_ON(addr >= vma->vm_end);
2308 BUG_ON(addr < vma->vm_start);
2309
2310 pgoff = vma->vm_pgoff;
2311 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2312 polnid = offset_il_node(pol, vma, pgoff);
2313 break;
2314
2315 case MPOL_PREFERRED:
2316 if (pol->flags & MPOL_F_LOCAL)
2317 polnid = numa_node_id();
2318 else
2319 polnid = pol->v.preferred_node;
2320 break;
2321
2322 case MPOL_BIND:
2323 /*
2324 * allows binding to multiple nodes.
2325 * use current page if in policy nodemask,
2326 * else select nearest allowed node, if any.
2327 * If no allowed nodes, use current [!misplaced].
2328 */
2329 if (node_isset(curnid, pol->v.nodes))
2330 goto out;
2331 (void)first_zones_zonelist(
2332 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2333 gfp_zone(GFP_HIGHUSER),
2334 &pol->v.nodes, &zone);
2335 polnid = zone->node;
2336 break;
2337
2338 default:
2339 BUG();
2340 }
Mel Gorman5606e382012-11-02 18:19:13 +00002341
2342 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002343 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002344 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002345
Rik van Riel10f39042014-01-27 17:03:44 -05002346 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002347 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002348 }
2349
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002350 if (curnid != polnid)
2351 ret = polnid;
2352out:
2353 mpol_cond_put(pol);
2354
2355 return ret;
2356}
2357
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2359{
Paul Mundt140d5a42007-07-15 23:38:16 -07002360 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002362 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363}
2364
Mel Gorman42288fe2012-12-21 23:10:25 +00002365static void sp_node_init(struct sp_node *node, unsigned long start,
2366 unsigned long end, struct mempolicy *pol)
2367{
2368 node->start = start;
2369 node->end = end;
2370 node->policy = pol;
2371}
2372
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002373static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2374 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002376 struct sp_node *n;
2377 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002379 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 if (!n)
2381 return NULL;
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002382
2383 newpol = mpol_dup(pol);
2384 if (IS_ERR(newpol)) {
2385 kmem_cache_free(sn_cache, n);
2386 return NULL;
2387 }
2388 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002389 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002390
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 return n;
2392}
2393
2394/* Replace a policy range. */
2395static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2396 unsigned long end, struct sp_node *new)
2397{
Mel Gormanb22d1272012-10-08 16:29:17 -07002398 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002399 struct sp_node *n_new = NULL;
2400 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002401 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Mel Gorman42288fe2012-12-21 23:10:25 +00002403restart:
2404 spin_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 n = sp_lookup(sp, start, end);
2406 /* Take care of old policies in the same range. */
2407 while (n && n->start < end) {
2408 struct rb_node *next = rb_next(&n->nd);
2409 if (n->start >= start) {
2410 if (n->end <= end)
2411 sp_delete(sp, n);
2412 else
2413 n->start = end;
2414 } else {
2415 /* Old policy spanning whole new range. */
2416 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002417 if (!n_new)
2418 goto alloc_new;
2419
2420 *mpol_new = *n->policy;
2421 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002422 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002424 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002425 n_new = NULL;
2426 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 break;
2428 } else
2429 n->end = start;
2430 }
2431 if (!next)
2432 break;
2433 n = rb_entry(next, struct sp_node, nd);
2434 }
2435 if (new)
2436 sp_insert(sp, new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002437 spin_unlock(&sp->lock);
2438 ret = 0;
2439
2440err_out:
2441 if (mpol_new)
2442 mpol_put(mpol_new);
2443 if (n_new)
2444 kmem_cache_free(sn_cache, n_new);
2445
Mel Gormanb22d1272012-10-08 16:29:17 -07002446 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002447
2448alloc_new:
2449 spin_unlock(&sp->lock);
2450 ret = -ENOMEM;
2451 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2452 if (!n_new)
2453 goto err_out;
2454 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2455 if (!mpol_new)
2456 goto err_out;
2457 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458}
2459
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002460/**
2461 * mpol_shared_policy_init - initialize shared policy for inode
2462 * @sp: pointer to inode shared policy
2463 * @mpol: struct mempolicy to install
2464 *
2465 * Install non-NULL @mpol in inode's shared policy rb-tree.
2466 * On entry, the current task has a reference on a non-NULL @mpol.
2467 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002468 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002469 */
2470void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002471{
Miao Xie58568d22009-06-16 15:31:49 -07002472 int ret;
2473
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002474 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Mel Gorman42288fe2012-12-21 23:10:25 +00002475 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002476
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002477 if (mpol) {
2478 struct vm_area_struct pvma;
2479 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002480 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002481
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002482 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002483 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002484 /* contextualize the tmpfs mount point mempolicy */
2485 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002486 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002487 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002488
2489 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002490 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002491 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002492 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002493 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002494
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002495 /* Create pseudo-vma that contains just the policy */
2496 memset(&pvma, 0, sizeof(struct vm_area_struct));
2497 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2498 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002499
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002500put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002501 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002502free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002503 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002504put_mpol:
2505 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002506 }
2507}
2508
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509int mpol_set_shared_policy(struct shared_policy *info,
2510 struct vm_area_struct *vma, struct mempolicy *npol)
2511{
2512 int err;
2513 struct sp_node *new = NULL;
2514 unsigned long sz = vma_pages(vma);
2515
David Rientjes028fec42008-04-28 02:12:25 -07002516 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002518 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002519 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002520 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
2522 if (npol) {
2523 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2524 if (!new)
2525 return -ENOMEM;
2526 }
2527 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2528 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002529 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 return err;
2531}
2532
2533/* Free a backing policy store on inode delete. */
2534void mpol_free_shared_policy(struct shared_policy *p)
2535{
2536 struct sp_node *n;
2537 struct rb_node *next;
2538
2539 if (!p->root.rb_node)
2540 return;
Mel Gorman42288fe2012-12-21 23:10:25 +00002541 spin_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 next = rb_first(&p->root);
2543 while (next) {
2544 n = rb_entry(next, struct sp_node, nd);
2545 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002546 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 }
Mel Gorman42288fe2012-12-21 23:10:25 +00002548 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549}
2550
Mel Gorman1a687c22012-11-22 11:16:36 +00002551#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002552static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002553
2554static void __init check_numabalancing_enable(void)
2555{
2556 bool numabalancing_default = false;
2557
2558 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2559 numabalancing_default = true;
2560
Mel Gormanc2976632014-01-29 14:05:42 -08002561 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2562 if (numabalancing_override)
2563 set_numabalancing_state(numabalancing_override == 1);
2564
Mel Gorman1a687c22012-11-22 11:16:36 +00002565 if (nr_node_ids > 1 && !numabalancing_override) {
Andrew Morton4a404be2014-01-29 14:05:43 -08002566 pr_info("%s automatic NUMA balancing. "
Mel Gormanc2976632014-01-29 14:05:42 -08002567 "Configure with numa_balancing= or the "
2568 "kernel.numa_balancing sysctl",
2569 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002570 set_numabalancing_state(numabalancing_default);
2571 }
2572}
2573
2574static int __init setup_numabalancing(char *str)
2575{
2576 int ret = 0;
2577 if (!str)
2578 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002579
2580 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002581 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002582 ret = 1;
2583 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002584 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002585 ret = 1;
2586 }
2587out:
2588 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002589 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002590
2591 return ret;
2592}
2593__setup("numa_balancing=", setup_numabalancing);
2594#else
2595static inline void __init check_numabalancing_enable(void)
2596{
2597}
2598#endif /* CONFIG_NUMA_BALANCING */
2599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600/* assumes fs == KERNEL_DS */
2601void __init numa_policy_init(void)
2602{
Paul Mundtb71636e2007-07-15 23:38:15 -07002603 nodemask_t interleave_nodes;
2604 unsigned long largest = 0;
2605 int nid, prefer = 0;
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 policy_cache = kmem_cache_create("numa_policy",
2608 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002609 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
2611 sn_cache = kmem_cache_create("shared_policy_node",
2612 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002613 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
Mel Gorman5606e382012-11-02 18:19:13 +00002615 for_each_node(nid) {
2616 preferred_node_policy[nid] = (struct mempolicy) {
2617 .refcnt = ATOMIC_INIT(1),
2618 .mode = MPOL_PREFERRED,
2619 .flags = MPOL_F_MOF | MPOL_F_MORON,
2620 .v = { .preferred_node = nid, },
2621 };
2622 }
2623
Paul Mundtb71636e2007-07-15 23:38:15 -07002624 /*
2625 * Set interleaving policy for system init. Interleaving is only
2626 * enabled across suitably sized nodes (default is >= 16MB), or
2627 * fall back to the largest node if they're all smaller.
2628 */
2629 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002630 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002631 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
Paul Mundtb71636e2007-07-15 23:38:15 -07002633 /* Preserve the largest node */
2634 if (largest < total_pages) {
2635 largest = total_pages;
2636 prefer = nid;
2637 }
2638
2639 /* Interleave this node? */
2640 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2641 node_set(nid, interleave_nodes);
2642 }
2643
2644 /* All too small, use the largest */
2645 if (unlikely(nodes_empty(interleave_nodes)))
2646 node_set(prefer, interleave_nodes);
2647
David Rientjes028fec42008-04-28 02:12:25 -07002648 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002649 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002650
2651 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652}
2653
Christoph Lameter8bccd852005-10-29 18:16:59 -07002654/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655void numa_default_policy(void)
2656{
David Rientjes028fec42008-04-28 02:12:25 -07002657 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658}
Paul Jackson68860ec2005-10-30 15:02:36 -08002659
Paul Jackson42253992006-01-08 01:01:59 -08002660/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002661 * Parse and format mempolicy from/to strings
2662 */
2663
2664/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002665 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002666 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002667static const char * const policy_modes[] =
2668{
2669 [MPOL_DEFAULT] = "default",
2670 [MPOL_PREFERRED] = "prefer",
2671 [MPOL_BIND] = "bind",
2672 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002673 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002674};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002675
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002676
2677#ifdef CONFIG_TMPFS
2678/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002679 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002680 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002681 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002682 *
2683 * Format of input:
2684 * <mode>[=<flags>][:<nodelist>]
2685 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002686 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002687 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002688int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002689{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002690 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002691 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002692 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002693 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002694 char *nodelist = strchr(str, ':');
2695 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002696 int err = 1;
2697
2698 if (nodelist) {
2699 /* NUL-terminate mode or flags string */
2700 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002701 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002702 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002703 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002704 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002705 } else
2706 nodes_clear(nodes);
2707
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002708 if (flags)
2709 *flags++ = '\0'; /* terminate mode string */
2710
Peter Zijlstra479e2802012-10-25 14:16:28 +02002711 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002712 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002713 break;
2714 }
2715 }
Mel Gormana7200942012-11-16 09:37:58 +00002716 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002717 goto out;
2718
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002719 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002720 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002721 /*
2722 * Insist on a nodelist of one node only
2723 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002724 if (nodelist) {
2725 char *rest = nodelist;
2726 while (isdigit(*rest))
2727 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002728 if (*rest)
2729 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002730 }
2731 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002732 case MPOL_INTERLEAVE:
2733 /*
2734 * Default to online nodes with memory if no nodelist
2735 */
2736 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002737 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002738 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002739 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002740 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002741 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002742 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002743 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002744 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002745 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002746 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002747 case MPOL_DEFAULT:
2748 /*
2749 * Insist on a empty nodelist
2750 */
2751 if (!nodelist)
2752 err = 0;
2753 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002754 case MPOL_BIND:
2755 /*
2756 * Insist on a nodelist
2757 */
2758 if (!nodelist)
2759 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002760 }
2761
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002762 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002763 if (flags) {
2764 /*
2765 * Currently, we only support two mutually exclusive
2766 * mode flags.
2767 */
2768 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002769 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002770 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002771 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002772 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002773 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002774 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002775
2776 new = mpol_new(mode, mode_flags, &nodes);
2777 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002778 goto out;
2779
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002780 /*
2781 * Save nodes for mpol_to_str() to show the tmpfs mount options
2782 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2783 */
2784 if (mode != MPOL_PREFERRED)
2785 new->v.nodes = nodes;
2786 else if (nodelist)
2787 new->v.preferred_node = first_node(nodes);
2788 else
2789 new->flags |= MPOL_F_LOCAL;
2790
2791 /*
2792 * Save nodes for contextualization: this will be used to "clone"
2793 * the mempolicy in a specific context [cpuset] at a later time.
2794 */
2795 new->w.user_nodemask = nodes;
2796
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002797 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002798
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002799out:
2800 /* Restore string for error message */
2801 if (nodelist)
2802 *--nodelist = ':';
2803 if (flags)
2804 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002805 if (!err)
2806 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002807 return err;
2808}
2809#endif /* CONFIG_TMPFS */
2810
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002811/**
2812 * mpol_to_str - format a mempolicy structure for printing
2813 * @buffer: to contain formatted mempolicy string
2814 * @maxlen: length of @buffer
2815 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002816 *
David Rientjes948927e2013-11-12 15:07:28 -08002817 * Convert @pol into a string. If @buffer is too short, truncate the string.
2818 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2819 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002820 */
David Rientjes948927e2013-11-12 15:07:28 -08002821void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002822{
2823 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002824 nodemask_t nodes = NODE_MASK_NONE;
2825 unsigned short mode = MPOL_DEFAULT;
2826 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002827
David Rientjes8790c712014-01-30 15:46:08 -08002828 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002829 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002830 flags = pol->flags;
2831 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002832
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002833 switch (mode) {
2834 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002835 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002836 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002837 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002838 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002839 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002840 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002841 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002842 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002843 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002844 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002845 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002846 default:
David Rientjes948927e2013-11-12 15:07:28 -08002847 WARN_ON_ONCE(1);
2848 snprintf(p, maxlen, "unknown");
2849 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002850 }
2851
David Rientjesb7a9f422013-11-21 14:32:06 -08002852 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002853
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002854 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002855 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002856
Lee Schermerhorn22919902008-04-28 02:13:22 -07002857 /*
2858 * Currently, the only defined flags are mutually exclusive
2859 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002860 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002861 p += snprintf(p, buffer + maxlen - p, "static");
2862 else if (flags & MPOL_F_RELATIVE_NODES)
2863 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002864 }
2865
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002866 if (!nodes_empty(nodes)) {
David Rientjes948927e2013-11-12 15:07:28 -08002867 p += snprintf(p, buffer + maxlen - p, ":");
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002868 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2869 }
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002870}