blob: 9ff56571e6d567d974258f918fb92d1de47ae7fa [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>
100
Nick Piggin62695a82008-10-18 20:26:09 -0700101#include "internal.h"
102
Christoph Lameter38e35862006-01-08 01:01:01 -0800103/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800104#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800105#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800106
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800107static struct kmem_cache *policy_cache;
108static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/* Highest zone. An specific allocation for a zone below that is not
111 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800112enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700114/*
115 * run-time system-wide default policy => local allocation
116 */
H Hartley Sweetene754d792011-10-31 17:09:23 -0700117static struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700119 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700120 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121};
122
Mel Gorman5606e382012-11-02 18:19:13 +0000123static struct mempolicy preferred_node_policy[MAX_NUMNODES];
124
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -0700125struct mempolicy *get_task_policy(struct task_struct *p)
Mel Gorman5606e382012-11-02 18:19:13 +0000126{
127 struct mempolicy *pol = p->mempolicy;
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700128 int node;
Mel Gorman5606e382012-11-02 18:19:13 +0000129
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700130 if (pol)
131 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000132
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700133 node = numa_node_id();
134 if (node != NUMA_NO_NODE) {
135 pol = &preferred_node_policy[node];
136 /* preferred_node_policy is not initialised early in boot */
137 if (pol->mode)
138 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000139 }
140
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700141 return &default_policy;
Mel Gorman5606e382012-11-02 18:19:13 +0000142}
143
David Rientjes37012942008-04-28 02:12:33 -0700144static const struct mempolicy_operations {
145 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700146 /*
147 * If read-side task has no lock to protect task->mempolicy, write-side
148 * task will rebind the task->mempolicy by two step. The first step is
149 * setting all the newly nodes, and the second step is cleaning all the
150 * disallowed nodes. In this way, we can avoid finding no node to alloc
151 * page.
152 * If we have a lock to protect task->mempolicy in read-side, we do
153 * rebind directly.
154 *
155 * step:
156 * MPOL_REBIND_ONCE - do rebind work at once
157 * MPOL_REBIND_STEP1 - set all the newly nodes
158 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
159 */
160 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
161 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700162} mpol_ops[MPOL_MAX];
163
David Rientjesf5b087b2008-04-28 02:12:27 -0700164static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
165{
Bob Liu6d556292010-05-24 14:31:59 -0700166 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700167}
168
169static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
170 const nodemask_t *rel)
171{
172 nodemask_t tmp;
173 nodes_fold(tmp, *orig, nodes_weight(*rel));
174 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700175}
176
David Rientjes37012942008-04-28 02:12:33 -0700177static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
178{
179 if (nodes_empty(*nodes))
180 return -EINVAL;
181 pol->v.nodes = *nodes;
182 return 0;
183}
184
185static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
186{
187 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700188 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700189 else if (nodes_empty(*nodes))
190 return -EINVAL; /* no allowed nodes */
191 else
192 pol->v.preferred_node = first_node(*nodes);
193 return 0;
194}
195
196static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
197{
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800198 if (nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700199 return -EINVAL;
200 pol->v.nodes = *nodes;
201 return 0;
202}
203
Miao Xie58568d22009-06-16 15:31:49 -0700204/*
205 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
206 * any, for the new policy. mpol_new() has already validated the nodes
207 * parameter with respect to the policy mode and flags. But, we need to
208 * handle an empty nodemask with MPOL_PREFERRED here.
209 *
210 * Must be called holding task's alloc_lock to protect task's mems_allowed
211 * and mempolicy. May also be called holding the mmap_semaphore for write.
212 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700213static int mpol_set_nodemask(struct mempolicy *pol,
214 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700215{
Miao Xie58568d22009-06-16 15:31:49 -0700216 int ret;
217
218 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
219 if (pol == NULL)
220 return 0;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800221 /* Check N_MEMORY */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700222 nodes_and(nsc->mask1,
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800223 cpuset_current_mems_allowed, node_states[N_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700224
225 VM_BUG_ON(!nodes);
226 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
227 nodes = NULL; /* explicit local allocation */
228 else {
229 if (pol->flags & MPOL_F_RELATIVE_NODES)
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800230 mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700231 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700232 nodes_and(nsc->mask2, *nodes, nsc->mask1);
233
Miao Xie58568d22009-06-16 15:31:49 -0700234 if (mpol_store_user_nodemask(pol))
235 pol->w.user_nodemask = *nodes;
236 else
237 pol->w.cpuset_mems_allowed =
238 cpuset_current_mems_allowed;
239 }
240
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700241 if (nodes)
242 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
243 else
244 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700245 return ret;
246}
247
248/*
249 * This function just creates a new policy, does some check and simple
250 * initialization. You must invoke mpol_set_nodemask() to set nodes.
251 */
David Rientjes028fec42008-04-28 02:12:25 -0700252static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
253 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 struct mempolicy *policy;
256
David Rientjes028fec42008-04-28 02:12:25 -0700257 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
David Rientjes00ef2d22013-02-22 16:35:36 -0800258 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
Paul Mundt140d5a42007-07-15 23:38:16 -0700259
David Rientjes3e1f0642008-04-28 02:12:34 -0700260 if (mode == MPOL_DEFAULT) {
261 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700262 return ERR_PTR(-EINVAL);
Lee Schermerhornd3a71032012-10-25 14:16:29 +0200263 return NULL;
David Rientjes37012942008-04-28 02:12:33 -0700264 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700265 VM_BUG_ON(!nodes);
266
267 /*
268 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
269 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
270 * All other modes require a valid pointer to a non-empty nodemask.
271 */
272 if (mode == MPOL_PREFERRED) {
273 if (nodes_empty(*nodes)) {
274 if (((flags & MPOL_F_STATIC_NODES) ||
275 (flags & MPOL_F_RELATIVE_NODES)))
276 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700277 }
Peter Zijlstra479e2802012-10-25 14:16:28 +0200278 } else if (mode == MPOL_LOCAL) {
279 if (!nodes_empty(*nodes))
280 return ERR_PTR(-EINVAL);
281 mode = MPOL_PREFERRED;
David Rientjes3e1f0642008-04-28 02:12:34 -0700282 } else if (nodes_empty(*nodes))
283 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
285 if (!policy)
286 return ERR_PTR(-ENOMEM);
287 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700288 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700289 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700290
David Rientjes37012942008-04-28 02:12:33 -0700291 return policy;
292}
293
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700294/* Slow path of a mpol destructor. */
295void __mpol_put(struct mempolicy *p)
296{
297 if (!atomic_dec_and_test(&p->refcnt))
298 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700299 kmem_cache_free(policy_cache, p);
300}
301
Miao Xie708c1bb2010-05-24 14:32:07 -0700302static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
303 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700304{
305}
306
Miao Xie708c1bb2010-05-24 14:32:07 -0700307/*
308 * step:
309 * MPOL_REBIND_ONCE - do rebind work at once
310 * MPOL_REBIND_STEP1 - set all the newly nodes
311 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
312 */
313static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
314 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700315{
316 nodemask_t tmp;
317
318 if (pol->flags & MPOL_F_STATIC_NODES)
319 nodes_and(tmp, pol->w.user_nodemask, *nodes);
320 else if (pol->flags & MPOL_F_RELATIVE_NODES)
321 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
322 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700323 /*
324 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
325 * result
326 */
327 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
328 nodes_remap(tmp, pol->v.nodes,
329 pol->w.cpuset_mems_allowed, *nodes);
330 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
331 } else if (step == MPOL_REBIND_STEP2) {
332 tmp = pol->w.cpuset_mems_allowed;
333 pol->w.cpuset_mems_allowed = *nodes;
334 } else
335 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700336 }
337
Miao Xie708c1bb2010-05-24 14:32:07 -0700338 if (nodes_empty(tmp))
339 tmp = *nodes;
340
341 if (step == MPOL_REBIND_STEP1)
342 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
343 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
344 pol->v.nodes = tmp;
345 else
346 BUG();
347
David Rientjes37012942008-04-28 02:12:33 -0700348 if (!node_isset(current->il_next, tmp)) {
Andrew Morton0edaf862016-05-19 17:10:58 -0700349 current->il_next = next_node_in(current->il_next, tmp);
David Rientjes37012942008-04-28 02:12:33 -0700350 if (current->il_next >= MAX_NUMNODES)
351 current->il_next = numa_node_id();
352 }
353}
354
355static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700356 const nodemask_t *nodes,
357 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700358{
359 nodemask_t tmp;
360
David Rientjes37012942008-04-28 02:12:33 -0700361 if (pol->flags & MPOL_F_STATIC_NODES) {
362 int node = first_node(pol->w.user_nodemask);
363
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700364 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700365 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700366 pol->flags &= ~MPOL_F_LOCAL;
367 } else
368 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700369 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
370 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
371 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700372 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700373 pol->v.preferred_node = node_remap(pol->v.preferred_node,
374 pol->w.cpuset_mems_allowed,
375 *nodes);
376 pol->w.cpuset_mems_allowed = *nodes;
377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
Miao Xie708c1bb2010-05-24 14:32:07 -0700380/*
381 * mpol_rebind_policy - Migrate a policy to a different set of nodes
382 *
383 * If read-side task has no lock to protect task->mempolicy, write-side
384 * task will rebind the task->mempolicy by two step. The first step is
385 * setting all the newly nodes, and the second step is cleaning all the
386 * disallowed nodes. In this way, we can avoid finding no node to alloc
387 * page.
388 * If we have a lock to protect task->mempolicy in read-side, we do
389 * rebind directly.
390 *
391 * step:
392 * MPOL_REBIND_ONCE - do rebind work at once
393 * MPOL_REBIND_STEP1 - set all the newly nodes
394 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
395 */
396static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
397 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700398{
David Rientjes1d0d2682008-04-28 02:12:32 -0700399 if (!pol)
400 return;
Wang Sheng-Hui89c522c2012-05-29 15:06:16 -0700401 if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700402 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
403 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700404
405 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
406 return;
407
408 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
409 BUG();
410
411 if (step == MPOL_REBIND_STEP1)
412 pol->flags |= MPOL_F_REBINDING;
413 else if (step == MPOL_REBIND_STEP2)
414 pol->flags &= ~MPOL_F_REBINDING;
415 else if (step >= MPOL_REBIND_NSTEP)
416 BUG();
417
418 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700419}
420
421/*
422 * Wrapper for mpol_rebind_policy() that just requires task
423 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700424 *
425 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700426 */
427
Miao Xie708c1bb2010-05-24 14:32:07 -0700428void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
429 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700430{
Miao Xie708c1bb2010-05-24 14:32:07 -0700431 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700432}
433
434/*
435 * Rebind each vma in mm to new nodemask.
436 *
437 * Call holding a reference to mm. Takes mm->mmap_sem during call.
438 */
439
440void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
441{
442 struct vm_area_struct *vma;
443
444 down_write(&mm->mmap_sem);
445 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700446 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700447 up_write(&mm->mmap_sem);
448}
449
David Rientjes37012942008-04-28 02:12:33 -0700450static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
451 [MPOL_DEFAULT] = {
452 .rebind = mpol_rebind_default,
453 },
454 [MPOL_INTERLEAVE] = {
455 .create = mpol_new_interleave,
456 .rebind = mpol_rebind_nodemask,
457 },
458 [MPOL_PREFERRED] = {
459 .create = mpol_new_preferred,
460 .rebind = mpol_rebind_preferred,
461 },
462 [MPOL_BIND] = {
463 .create = mpol_new_bind,
464 .rebind = mpol_rebind_nodemask,
465 },
466};
467
Christoph Lameterfc301282006-01-18 17:42:29 -0800468static void migrate_page_add(struct page *page, struct list_head *pagelist,
469 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800470
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800471struct queue_pages {
472 struct list_head *pagelist;
473 unsigned long flags;
474 nodemask_t *nmask;
475 struct vm_area_struct *prev;
476};
477
Naoya Horiguchi98094942013-09-11 14:22:14 -0700478/*
479 * Scan through pages checking if pages follow certain conditions,
480 * and move them to the pagelist if they do.
481 */
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800482static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
483 unsigned long end, struct mm_walk *walk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800485 struct vm_area_struct *vma = walk->vma;
486 struct page *page;
487 struct queue_pages *qp = walk->private;
488 unsigned long flags = qp->flags;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800489 int nid, ret;
Hugh Dickins91612e02005-06-21 17:15:07 -0700490 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700491 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700492
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800493 if (pmd_trans_huge(*pmd)) {
494 ptl = pmd_lock(walk->mm, pmd);
495 if (pmd_trans_huge(*pmd)) {
496 page = pmd_page(*pmd);
497 if (is_huge_zero_page(page)) {
498 spin_unlock(ptl);
499 split_huge_pmd(vma, pmd, addr);
500 } else {
501 get_page(page);
502 spin_unlock(ptl);
503 lock_page(page);
504 ret = split_huge_page(page);
505 unlock_page(page);
506 put_page(page);
507 if (ret)
508 return 0;
509 }
510 } else {
511 spin_unlock(ptl);
512 }
513 }
Hugh Dickins91612e02005-06-21 17:15:07 -0700514
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700515 if (pmd_trans_unstable(pmd))
516 return 0;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800517retry:
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800518 pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
519 for (; addr != end; pte++, addr += PAGE_SIZE) {
Hugh Dickins91612e02005-06-21 17:15:07 -0700520 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800522 page = vm_normal_page(vma, addr, *pte);
523 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800525 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800526 * vm_normal_page() filters out zero pages, but there might
527 * still be PageReserved pages to skip, perhaps in a VDSO.
Nick Piggin053837f2006-01-18 17:42:27 -0800528 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800529 if (PageReserved(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800530 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800531 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800532 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Christoph Lameter38e35862006-01-08 01:01:01 -0800533 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700534 if (PageTransCompound(page)) {
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800535 get_page(page);
536 pte_unmap_unlock(pte, ptl);
537 lock_page(page);
538 ret = split_huge_page(page);
539 unlock_page(page);
540 put_page(page);
541 /* Failed to split -- skip. */
542 if (ret) {
543 pte = pte_offset_map_lock(walk->mm, pmd,
544 addr, &ptl);
545 continue;
546 }
547 goto retry;
548 }
Christoph Lameter38e35862006-01-08 01:01:01 -0800549
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800550 migrate_page_add(page, qp->pagelist, flags);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800551 }
552 pte_unmap_unlock(pte - 1, ptl);
553 cond_resched();
554 return 0;
Hugh Dickins91612e02005-06-21 17:15:07 -0700555}
556
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800557static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
558 unsigned long addr, unsigned long end,
559 struct mm_walk *walk)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700560{
561#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800562 struct queue_pages *qp = walk->private;
563 unsigned long flags = qp->flags;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700564 int nid;
565 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800566 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400567 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700568
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800569 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
570 entry = huge_ptep_get(pte);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400571 if (!pte_present(entry))
572 goto unlock;
573 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700574 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800575 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700576 goto unlock;
577 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
578 if (flags & (MPOL_MF_MOVE_ALL) ||
579 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800580 isolate_huge_page(page, qp->pagelist);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700581unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800582 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700583#else
584 BUG();
585#endif
Hugh Dickins91612e02005-06-21 17:15:07 -0700586 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530589#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200590/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200591 * This is used to mark a range of virtual addresses to be inaccessible.
592 * These are later cleared by a NUMA hinting fault. Depending on these
593 * faults, pages may be migrated for better NUMA placement.
594 *
595 * This is assuming that NUMA faults are handled using PROT_NONE. If
596 * an architecture makes a different choice, it will need further
597 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200598 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200599unsigned long change_prot_numa(struct vm_area_struct *vma,
600 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200601{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200602 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200603
Mel Gorman4d942462015-02-12 14:58:28 -0800604 nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000605 if (nr_updated)
606 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200607
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200608 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200609}
610#else
611static unsigned long change_prot_numa(struct vm_area_struct *vma,
612 unsigned long addr, unsigned long end)
613{
614 return 0;
615}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530616#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200617
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800618static int queue_pages_test_walk(unsigned long start, unsigned long end,
619 struct mm_walk *walk)
620{
621 struct vm_area_struct *vma = walk->vma;
622 struct queue_pages *qp = walk->private;
623 unsigned long endvma = vma->vm_end;
624 unsigned long flags = qp->flags;
625
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800626 if (!vma_migratable(vma))
Naoya Horiguchi48684a62015-02-11 15:28:06 -0800627 return 1;
628
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800629 if (endvma > end)
630 endvma = end;
631 if (vma->vm_start > start)
632 start = vma->vm_start;
633
634 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
635 if (!vma->vm_next && vma->vm_end < end)
636 return -EFAULT;
637 if (qp->prev && qp->prev->vm_end < vma->vm_start)
638 return -EFAULT;
639 }
640
641 qp->prev = vma;
642
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800643 if (flags & MPOL_MF_LAZY) {
644 /* Similar to task_numa_work, skip inaccessible VMAs */
Liang Chen4355c012016-03-15 14:56:42 -0700645 if (!is_vm_hugetlb_page(vma) &&
646 (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) &&
647 !(vma->vm_flags & VM_MIXEDMAP))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800648 change_prot_numa(vma, start, endvma);
649 return 1;
650 }
651
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800652 /* queue pages from current vma */
653 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800654 return 0;
655 return 1;
656}
657
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800658/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700659 * Walk through page tables and collect pages to be migrated.
660 *
661 * If pages found in a given range are on a set of nodes (determined by
662 * @nodes and @flags,) it's isolated and queued to the pagelist which is
663 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800664 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700665static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700666queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800667 nodemask_t *nodes, unsigned long flags,
668 struct list_head *pagelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800670 struct queue_pages qp = {
671 .pagelist = pagelist,
672 .flags = flags,
673 .nmask = nodes,
674 .prev = NULL,
675 };
676 struct mm_walk queue_pages_walk = {
677 .hugetlb_entry = queue_pages_hugetlb,
678 .pmd_entry = queue_pages_pte_range,
679 .test_walk = queue_pages_test_walk,
680 .mm = mm,
681 .private = &qp,
682 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800684 return walk_page_range(start, end, &queue_pages_walk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700687/*
688 * Apply policy to a single VMA
689 * This must be called with the mmap_sem held for writing.
690 */
691static int vma_replace_policy(struct vm_area_struct *vma,
692 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700693{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700694 int err;
695 struct mempolicy *old;
696 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700697
698 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
699 vma->vm_start, vma->vm_end, vma->vm_pgoff,
700 vma->vm_ops, vma->vm_file,
701 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
702
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700703 new = mpol_dup(pol);
704 if (IS_ERR(new))
705 return PTR_ERR(new);
706
707 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700708 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700709 if (err)
710 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700711 }
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700712
713 old = vma->vm_policy;
714 vma->vm_policy = new; /* protected by mmap_sem */
715 mpol_put(old);
716
717 return 0;
718 err_out:
719 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700720 return err;
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800724static int mbind_range(struct mm_struct *mm, unsigned long start,
725 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800728 struct vm_area_struct *prev;
729 struct vm_area_struct *vma;
730 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800731 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800732 unsigned long vmstart;
733 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Linus Torvalds097d5912012-03-06 18:23:36 -0800735 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800736 if (!vma || vma->vm_start > start)
737 return -EFAULT;
738
Linus Torvalds097d5912012-03-06 18:23:36 -0800739 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800740 if (start > vma->vm_start)
741 prev = vma;
742
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800743 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800745 vmstart = max(start, vma->vm_start);
746 vmend = min(end, vma->vm_end);
747
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800748 if (mpol_equal(vma_policy(vma), new_pol))
749 continue;
750
751 pgoff = vma->vm_pgoff +
752 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800753 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700754 vma->anon_vma, vma->vm_file, pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -0700755 new_pol, vma->vm_userfaultfd_ctx,
756 vma_get_anon_name(vma));
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800757 if (prev) {
758 vma = prev;
759 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700760 if (mpol_equal(vma_policy(vma), new_pol))
761 continue;
762 /* vma_merge() joined vma && vma->next, case 8 */
763 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800764 }
765 if (vma->vm_start != vmstart) {
766 err = split_vma(vma->vm_mm, vma, vmstart, 1);
767 if (err)
768 goto out;
769 }
770 if (vma->vm_end != vmend) {
771 err = split_vma(vma->vm_mm, vma, vmend, 0);
772 if (err)
773 goto out;
774 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700775 replace:
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700776 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700777 if (err)
778 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800780
781 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return err;
783}
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700786static long do_set_mempolicy(unsigned short mode, unsigned short flags,
787 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Miao Xie58568d22009-06-16 15:31:49 -0700789 struct mempolicy *new, *old;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700790 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700791 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700793 if (!scratch)
794 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700795
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700796 new = mpol_new(mode, flags, nodes);
797 if (IS_ERR(new)) {
798 ret = PTR_ERR(new);
799 goto out;
800 }
Oleg Nesterov2c7c3a72014-10-09 15:27:55 -0700801
Miao Xie58568d22009-06-16 15:31:49 -0700802 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700803 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700804 if (ret) {
805 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700806 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700807 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700808 }
809 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 current->mempolicy = new;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700811 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700812 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700813 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700814 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700815 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700816 ret = 0;
817out:
818 NODEMASK_SCRATCH_FREE(scratch);
819 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700822/*
823 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700824 *
825 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700826 */
827static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700829 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700830 if (p == &default_policy)
831 return;
832
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700833 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700834 case MPOL_BIND:
835 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700837 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 break;
839 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700840 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700841 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700842 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 break;
844 default:
845 BUG();
846 }
847}
848
Dave Hansend4edcf02016-02-12 13:01:56 -0800849static int lookup_node(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 struct page *p;
852 int err;
853
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100854 err = get_user_pages(addr & PAGE_MASK, 1, 0, &p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (err >= 0) {
856 err = page_to_nid(p);
857 put_page(p);
858 }
859 return err;
860}
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700863static long do_get_mempolicy(int *policy, nodemask_t *nmask,
864 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700866 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct mm_struct *mm = current->mm;
868 struct vm_area_struct *vma = NULL;
869 struct mempolicy *pol = current->mempolicy;
870
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700871 if (flags &
872 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700874
875 if (flags & MPOL_F_MEMS_ALLOWED) {
876 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
877 return -EINVAL;
878 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700879 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700880 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700881 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700882 return 0;
883 }
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700886 /*
887 * Do NOT fall back to task policy if the
888 * vma/shared policy at addr is NULL. We
889 * want to return MPOL_DEFAULT in this case.
890 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 down_read(&mm->mmap_sem);
892 vma = find_vma_intersection(mm, addr, addr+1);
893 if (!vma) {
894 up_read(&mm->mmap_sem);
895 return -EFAULT;
896 }
897 if (vma->vm_ops && vma->vm_ops->get_policy)
898 pol = vma->vm_ops->get_policy(vma, addr);
899 else
900 pol = vma->vm_policy;
901 } else if (addr)
902 return -EINVAL;
903
904 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700905 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 if (flags & MPOL_F_NODE) {
908 if (flags & MPOL_F_ADDR) {
Dave Hansend4edcf02016-02-12 13:01:56 -0800909 err = lookup_node(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (err < 0)
911 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700912 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700914 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700915 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 } else {
917 err = -EINVAL;
918 goto out;
919 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700920 } else {
921 *policy = pol == &default_policy ? MPOL_DEFAULT :
922 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700923 /*
924 * Internal mempolicy flags must be masked off before exposing
925 * the policy to userspace.
926 */
927 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 if (vma) {
931 up_read(&current->mm->mmap_sem);
932 vma = NULL;
933 }
934
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700936 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700937 if (mpol_store_user_nodemask(pol)) {
938 *nmask = pol->w.user_nodemask;
939 } else {
940 task_lock(current);
941 get_policy_nodemask(pol, nmask);
942 task_unlock(current);
943 }
Miao Xie58568d22009-06-16 15:31:49 -0700944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700947 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (vma)
949 up_read(&current->mm->mmap_sem);
950 return err;
951}
952
Christoph Lameterb20a3502006-03-22 00:09:12 -0800953#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700954/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800955 * page migration
956 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800957static void migrate_page_add(struct page *page, struct list_head *pagelist,
958 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800959{
960 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800961 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800962 */
Nick Piggin62695a82008-10-18 20:26:09 -0700963 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
964 if (!isolate_lru_page(page)) {
965 list_add_tail(&page->lru, pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -0700966 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800967 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700968 }
969 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800970}
971
Christoph Lameter742755a2006-06-23 02:03:55 -0700972static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700973{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700974 if (PageHuge(page))
975 return alloc_huge_page_node(page_hstate(compound_head(page)),
976 node);
977 else
Vlastimil Babka96db8002015-09-08 15:03:50 -0700978 return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE |
David Rientjesb360edb2015-04-14 15:46:52 -0700979 __GFP_THISNODE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700980}
981
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800982/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800983 * Migrate pages from one node to a target node.
984 * Returns error or the number of pages not migrated.
985 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700986static int migrate_to_node(struct mm_struct *mm, int source, int dest,
987 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800988{
989 nodemask_t nmask;
990 LIST_HEAD(pagelist);
991 int err = 0;
992
993 nodes_clear(nmask);
994 node_set(source, nmask);
995
Minchan Kim08270802012-10-08 16:33:38 -0700996 /*
997 * This does not "check" the range but isolates all pages that
998 * need migration. Between passing in the full user address
999 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
1000 */
1001 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -07001002 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001003 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1004
Minchan Kimcf608ac2010-10-26 14:21:29 -07001005 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001006 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001007 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001008 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001009 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001010 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001011
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001012 return err;
1013}
1014
1015/*
1016 * Move pages between the two nodesets so as to preserve the physical
1017 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001018 *
1019 * Returns the number of page that could not be moved.
1020 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001021int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1022 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001023{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001024 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001025 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001026 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001027
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001028 err = migrate_prep();
1029 if (err)
1030 return err;
1031
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001032 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001033
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001034 /*
1035 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1036 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1037 * bit in 'tmp', and return that <source, dest> pair for migration.
1038 * The pair of nodemasks 'to' and 'from' define the map.
1039 *
1040 * If no pair of bits is found that way, fallback to picking some
1041 * pair of 'source' and 'dest' bits that are not the same. If the
1042 * 'source' and 'dest' bits are the same, this represents a node
1043 * that will be migrating to itself, so no pages need move.
1044 *
1045 * If no bits are left in 'tmp', or if all remaining bits left
1046 * in 'tmp' correspond to the same bit in 'to', return false
1047 * (nothing left to migrate).
1048 *
1049 * This lets us pick a pair of nodes to migrate between, such that
1050 * if possible the dest node is not already occupied by some other
1051 * source node, minimizing the risk of overloading the memory on a
1052 * node that would happen if we migrated incoming memory to a node
1053 * before migrating outgoing memory source that same node.
1054 *
1055 * A single scan of tmp is sufficient. As we go, we remember the
1056 * most recent <s, d> pair that moved (s != d). If we find a pair
1057 * that not only moved, but what's better, moved to an empty slot
1058 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001059 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001060 * most recent <s, d> pair that moved. If we get all the way through
1061 * the scan of tmp without finding any node that moved, much less
1062 * moved to an empty node, then there is nothing left worth migrating.
1063 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001064
Andrew Morton0ce72d42012-05-29 15:06:24 -07001065 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001066 while (!nodes_empty(tmp)) {
1067 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001068 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001069 int dest = 0;
1070
1071 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001072
1073 /*
1074 * do_migrate_pages() tries to maintain the relative
1075 * node relationship of the pages established between
1076 * threads and memory areas.
1077 *
1078 * However if the number of source nodes is not equal to
1079 * the number of destination nodes we can not preserve
1080 * this node relative relationship. In that case, skip
1081 * copying memory from a node that is in the destination
1082 * mask.
1083 *
1084 * Example: [2,3,4] -> [3,4,5] moves everything.
1085 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1086 */
1087
Andrew Morton0ce72d42012-05-29 15:06:24 -07001088 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1089 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001090 continue;
1091
Andrew Morton0ce72d42012-05-29 15:06:24 -07001092 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001093 if (s == d)
1094 continue;
1095
1096 source = s; /* Node moved. Memorize */
1097 dest = d;
1098
1099 /* dest not in remaining from nodes? */
1100 if (!node_isset(dest, tmp))
1101 break;
1102 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001103 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001104 break;
1105
1106 node_clear(source, tmp);
1107 err = migrate_to_node(mm, source, dest, flags);
1108 if (err > 0)
1109 busy += err;
1110 if (err < 0)
1111 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001112 }
1113 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001114 if (err < 0)
1115 return err;
1116 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001117
Christoph Lameter39743882006-01-08 01:00:51 -08001118}
1119
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001120/*
1121 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001122 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001123 * Search forward from there, if not. N.B., this assumes that the
1124 * list of pages handed to migrate_pages()--which is how we get here--
1125 * is in virtual address order.
1126 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001127static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001128{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001129 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001130 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001131
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001132 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001133 while (vma) {
1134 address = page_address_in_vma(page, vma);
1135 if (address != -EFAULT)
1136 break;
1137 vma = vma->vm_next;
1138 }
1139
Wanpeng Li11c731e2013-12-18 17:08:56 -08001140 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001141 BUG_ON(!vma);
1142 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001143 }
1144 /*
1145 * if !vma, alloc_page_vma() will use task or system default policy
1146 */
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001147 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001148}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001149#else
1150
1151static void migrate_page_add(struct page *page, struct list_head *pagelist,
1152 unsigned long flags)
1153{
1154}
1155
Andrew Morton0ce72d42012-05-29 15:06:24 -07001156int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1157 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001158{
1159 return -ENOSYS;
1160}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001161
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001162static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001163{
1164 return NULL;
1165}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001166#endif
1167
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001168static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001169 unsigned short mode, unsigned short mode_flags,
1170 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001171{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001172 struct mm_struct *mm = current->mm;
1173 struct mempolicy *new;
1174 unsigned long end;
1175 int err;
1176 LIST_HEAD(pagelist);
1177
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001178 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001179 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001180 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001181 return -EPERM;
1182
1183 if (start & ~PAGE_MASK)
1184 return -EINVAL;
1185
1186 if (mode == MPOL_DEFAULT)
1187 flags &= ~MPOL_MF_STRICT;
1188
1189 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1190 end = start + len;
1191
1192 if (end < start)
1193 return -EINVAL;
1194 if (end == start)
1195 return 0;
1196
David Rientjes028fec42008-04-28 02:12:25 -07001197 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001198 if (IS_ERR(new))
1199 return PTR_ERR(new);
1200
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001201 if (flags & MPOL_MF_LAZY)
1202 new->flags |= MPOL_F_MOF;
1203
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001204 /*
1205 * If we are using the default policy then operation
1206 * on discontinuous address spaces is okay after all
1207 */
1208 if (!new)
1209 flags |= MPOL_MF_DISCONTIG_OK;
1210
David Rientjes028fec42008-04-28 02:12:25 -07001211 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1212 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001213 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001214
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001215 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1216
1217 err = migrate_prep();
1218 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001219 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001220 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001221 {
1222 NODEMASK_SCRATCH(scratch);
1223 if (scratch) {
1224 down_write(&mm->mmap_sem);
1225 task_lock(current);
1226 err = mpol_set_nodemask(new, nmask, scratch);
1227 task_unlock(current);
1228 if (err)
1229 up_write(&mm->mmap_sem);
1230 } else
1231 err = -ENOMEM;
1232 NODEMASK_SCRATCH_FREE(scratch);
1233 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001234 if (err)
1235 goto mpol_out;
1236
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001237 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001238 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001239 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001240 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001241
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001242 if (!err) {
1243 int nr_failed = 0;
1244
Minchan Kimcf608ac2010-10-26 14:21:29 -07001245 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001246 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001247 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1248 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001249 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001250 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001251 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001252
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001253 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001254 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001255 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001256 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001257
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001258 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001259 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001260 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001261 return err;
1262}
1263
Christoph Lameter39743882006-01-08 01:00:51 -08001264/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001265 * User space interface with variable sized bitmaps for nodelists.
1266 */
1267
1268/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001269static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001270 unsigned long maxnode)
1271{
1272 unsigned long k;
1273 unsigned long nlongs;
1274 unsigned long endmask;
1275
1276 --maxnode;
1277 nodes_clear(*nodes);
1278 if (maxnode == 0 || !nmask)
1279 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001280 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001281 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001282
1283 nlongs = BITS_TO_LONGS(maxnode);
1284 if ((maxnode % BITS_PER_LONG) == 0)
1285 endmask = ~0UL;
1286 else
1287 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1288
1289 /* When the user specified more nodes than supported just check
1290 if the non supported part is all zero. */
1291 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1292 if (nlongs > PAGE_SIZE/sizeof(long))
1293 return -EINVAL;
1294 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1295 unsigned long t;
1296 if (get_user(t, nmask + k))
1297 return -EFAULT;
1298 if (k == nlongs - 1) {
1299 if (t & endmask)
1300 return -EINVAL;
1301 } else if (t)
1302 return -EINVAL;
1303 }
1304 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1305 endmask = ~0UL;
1306 }
1307
1308 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1309 return -EFAULT;
1310 nodes_addr(*nodes)[nlongs-1] &= endmask;
1311 return 0;
1312}
1313
1314/* Copy a kernel node mask to user space */
1315static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1316 nodemask_t *nodes)
1317{
1318 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1319 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1320
1321 if (copy > nbytes) {
1322 if (copy > PAGE_SIZE)
1323 return -EINVAL;
1324 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1325 return -EFAULT;
1326 copy = nbytes;
1327 }
1328 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1329}
1330
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001331SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001332 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001333 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001334{
1335 nodemask_t nodes;
1336 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001337 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001338
David Rientjes028fec42008-04-28 02:12:25 -07001339 mode_flags = mode & MPOL_MODE_FLAGS;
1340 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001341 if (mode >= MPOL_MAX)
1342 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001343 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1344 (mode_flags & MPOL_F_RELATIVE_NODES))
1345 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001346 err = get_nodes(&nodes, nmask, maxnode);
1347 if (err)
1348 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001349 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001350}
1351
1352/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001353SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001354 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001355{
1356 int err;
1357 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001358 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001359
David Rientjes028fec42008-04-28 02:12:25 -07001360 flags = mode & MPOL_MODE_FLAGS;
1361 mode &= ~MPOL_MODE_FLAGS;
1362 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001363 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001364 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1365 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001366 err = get_nodes(&nodes, nmask, maxnode);
1367 if (err)
1368 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001369 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001370}
1371
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001372SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1373 const unsigned long __user *, old_nodes,
1374 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001375{
David Howellsc69e8d92008-11-14 10:39:19 +11001376 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001377 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001378 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001379 nodemask_t task_nodes;
1380 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001381 nodemask_t *old;
1382 nodemask_t *new;
1383 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001384
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001385 if (!scratch)
1386 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001387
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001388 old = &scratch->mask1;
1389 new = &scratch->mask2;
1390
1391 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001392 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001393 goto out;
1394
1395 err = get_nodes(new, new_nodes, maxnode);
1396 if (err)
1397 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001398
1399 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001400 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001401 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001402 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001403 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001404 err = -ESRCH;
1405 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001406 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001407 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001408
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001409 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001410
1411 /*
1412 * Check if this process has the right to modify the specified
1413 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001414 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001415 * userid as the target process.
1416 */
David Howellsc69e8d92008-11-14 10:39:19 +11001417 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001418 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1419 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001420 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001421 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001422 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001423 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001424 }
David Howellsc69e8d92008-11-14 10:39:19 +11001425 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001426
1427 task_nodes = cpuset_mems_allowed(task);
1428 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001429 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001430 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001431 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001432 }
1433
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08001434 if (!nodes_subset(*new, node_states[N_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001435 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001436 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001437 }
1438
David Quigley86c3a762006-06-23 02:04:02 -07001439 err = security_task_movememory(task);
1440 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001441 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001442
Christoph Lameter3268c632012-03-21 16:34:06 -07001443 mm = get_task_mm(task);
1444 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001445
1446 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001447 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001448 goto out;
1449 }
1450
1451 err = do_migrate_pages(mm, old, new,
1452 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001453
1454 mmput(mm);
1455out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001456 NODEMASK_SCRATCH_FREE(scratch);
1457
Christoph Lameter39743882006-01-08 01:00:51 -08001458 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001459
1460out_put:
1461 put_task_struct(task);
1462 goto out;
1463
Christoph Lameter39743882006-01-08 01:00:51 -08001464}
1465
1466
Christoph Lameter8bccd852005-10-29 18:16:59 -07001467/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001468SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1469 unsigned long __user *, nmask, unsigned long, maxnode,
1470 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001471{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001472 int err;
1473 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001474 nodemask_t nodes;
1475
1476 if (nmask != NULL && maxnode < MAX_NUMNODES)
1477 return -EINVAL;
1478
1479 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1480
1481 if (err)
1482 return err;
1483
1484 if (policy && put_user(pval, policy))
1485 return -EFAULT;
1486
1487 if (nmask)
1488 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1489
1490 return err;
1491}
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493#ifdef CONFIG_COMPAT
1494
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001495COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1496 compat_ulong_t __user *, nmask,
1497 compat_ulong_t, maxnode,
1498 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
1500 long err;
1501 unsigned long __user *nm = NULL;
1502 unsigned long nr_bits, alloc_size;
1503 DECLARE_BITMAP(bm, MAX_NUMNODES);
1504
1505 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1506 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1507
1508 if (nmask)
1509 nm = compat_alloc_user_space(alloc_size);
1510
1511 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1512
1513 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001514 unsigned long copy_size;
1515 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1516 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 /* ensure entire bitmap is zeroed */
1518 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1519 err |= compat_put_bitmap(nmask, bm, nr_bits);
1520 }
1521
1522 return err;
1523}
1524
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001525COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1526 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 unsigned long __user *nm = NULL;
1529 unsigned long nr_bits, alloc_size;
1530 DECLARE_BITMAP(bm, MAX_NUMNODES);
1531
1532 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1533 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1534
1535 if (nmask) {
Chris Sallscddab762017-04-07 23:48:11 -07001536 if (compat_get_bitmap(bm, nmask, nr_bits))
1537 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 nm = compat_alloc_user_space(alloc_size);
Chris Sallscddab762017-04-07 23:48:11 -07001539 if (copy_to_user(nm, bm, alloc_size))
1540 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 }
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 return sys_set_mempolicy(mode, nm, nr_bits+1);
1544}
1545
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001546COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1547 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1548 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 unsigned long __user *nm = NULL;
1551 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001552 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1555 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1556
1557 if (nmask) {
Chris Sallscddab762017-04-07 23:48:11 -07001558 if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits))
1559 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 nm = compat_alloc_user_space(alloc_size);
Chris Sallscddab762017-04-07 23:48:11 -07001561 if (copy_to_user(nm, nodes_addr(bm), alloc_size))
1562 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 }
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1566}
1567
1568#endif
1569
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001570struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1571 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572{
Oleg Nesterov8d902742014-10-09 15:27:45 -07001573 struct mempolicy *pol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
1575 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001576 if (vma->vm_ops && vma->vm_ops->get_policy) {
Oleg Nesterov8d902742014-10-09 15:27:45 -07001577 pol = vma->vm_ops->get_policy(vma, addr);
Mel Gorman00442ad2012-10-08 16:29:20 -07001578 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001580
1581 /*
1582 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1583 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1584 * count on these policies which will be dropped by
1585 * mpol_cond_put() later
1586 */
1587 if (mpol_needs_cond_ref(pol))
1588 mpol_get(pol);
1589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001591
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001592 return pol;
1593}
1594
1595/*
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001596 * get_vma_policy(@vma, @addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001597 * @vma: virtual memory area whose policy is sought
1598 * @addr: address in @vma for shared policy lookup
1599 *
1600 * Returns effective policy for a VMA at specified address.
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001601 * Falls back to current->mempolicy or system default policy, as necessary.
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001602 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1603 * count--added by the get_policy() vm_op, as appropriate--to protect against
1604 * freeing by another task. It is the caller's responsibility to free the
1605 * extra reference for shared policies.
1606 */
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001607static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1608 unsigned long addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001609{
1610 struct mempolicy *pol = __get_vma_policy(vma, addr);
1611
Oleg Nesterov8d902742014-10-09 15:27:45 -07001612 if (!pol)
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001613 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 return pol;
1616}
1617
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001618bool vma_policy_mof(struct vm_area_struct *vma)
Mel Gormanfc3147242013-10-07 11:29:09 +01001619{
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001620 struct mempolicy *pol;
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001621
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001622 if (vma->vm_ops && vma->vm_ops->get_policy) {
1623 bool ret = false;
Mel Gormanfc3147242013-10-07 11:29:09 +01001624
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001625 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1626 if (pol && (pol->flags & MPOL_F_MOF))
1627 ret = true;
1628 mpol_cond_put(pol);
Mel Gormanfc3147242013-10-07 11:29:09 +01001629
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001630 return ret;
Mel Gormanfc3147242013-10-07 11:29:09 +01001631 }
1632
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001633 pol = vma->vm_policy;
Oleg Nesterov8d902742014-10-09 15:27:45 -07001634 if (!pol)
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001635 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001636
Mel Gormanfc3147242013-10-07 11:29:09 +01001637 return pol->flags & MPOL_F_MOF;
1638}
1639
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001640static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1641{
1642 enum zone_type dynamic_policy_zone = policy_zone;
1643
1644 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1645
1646 /*
1647 * if policy->v.nodes has movable memory only,
1648 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1649 *
1650 * policy->v.nodes is intersect with node_states[N_MEMORY].
1651 * so if the following test faile, it implies
1652 * policy->v.nodes has movable memory only.
1653 */
1654 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1655 dynamic_policy_zone = ZONE_MOVABLE;
1656
1657 return zone >= dynamic_policy_zone;
1658}
1659
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001660/*
1661 * Return a nodemask representing a mempolicy for filtering nodes for
1662 * page allocation
1663 */
1664static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001665{
1666 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001667 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001668 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001669 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1670 return &policy->v.nodes;
1671
1672 return NULL;
1673}
1674
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001675/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001676static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1677 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001679 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001681 if (!(policy->flags & MPOL_F_LOCAL))
1682 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 break;
1684 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001685 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001686 * Normally, MPOL_BIND allocations are node-local within the
1687 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001688 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001689 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001690 */
Mel Gorman19770b32008-04-28 02:12:18 -07001691 if (unlikely(gfp & __GFP_THISNODE) &&
1692 unlikely(!node_isset(nd, policy->v.nodes)))
1693 nd = first_node(policy->v.nodes);
1694 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 BUG();
1697 }
Mel Gorman0e884602008-04-28 02:12:14 -07001698 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699}
1700
1701/* Do dynamic interleaving for a process */
1702static unsigned interleave_nodes(struct mempolicy *policy)
1703{
1704 unsigned nid, next;
1705 struct task_struct *me = current;
1706
1707 nid = me->il_next;
Andrew Morton0edaf862016-05-19 17:10:58 -07001708 next = next_node_in(nid, policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001709 if (next < MAX_NUMNODES)
1710 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return nid;
1712}
1713
Christoph Lameterdc85da12006-01-18 17:42:36 -08001714/*
1715 * Depending on the memory policy provide a node from which to allocate the
1716 * next slab entry.
1717 */
David Rientjes2a389612014-04-07 15:37:29 -07001718unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001719{
Andi Kleene7b691b2012-06-09 02:40:03 -07001720 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001721 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001722
1723 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001724 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001725
1726 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001727 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001728 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001729
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001730 switch (policy->mode) {
1731 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001732 /*
1733 * handled MPOL_F_LOCAL above
1734 */
1735 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001736
Christoph Lameterdc85da12006-01-18 17:42:36 -08001737 case MPOL_INTERLEAVE:
1738 return interleave_nodes(policy);
1739
Mel Gormandd1a2392008-04-28 02:12:17 -07001740 case MPOL_BIND: {
Mel Gormanc33d6c02016-05-19 17:14:10 -07001741 struct zoneref *z;
1742
Christoph Lameterdc85da12006-01-18 17:42:36 -08001743 /*
1744 * Follow bind policy behavior and start allocation at the
1745 * first node.
1746 */
Mel Gorman19770b32008-04-28 02:12:18 -07001747 struct zonelist *zonelist;
Mel Gorman19770b32008-04-28 02:12:18 -07001748 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07001749 zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK];
Mel Gormanc33d6c02016-05-19 17:14:10 -07001750 z = first_zones_zonelist(zonelist, highest_zoneidx,
1751 &policy->v.nodes);
1752 return z->zone ? z->zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001753 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001754
Christoph Lameterdc85da12006-01-18 17:42:36 -08001755 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001756 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001757 }
1758}
1759
Andrew Mortonfee83b32016-05-19 17:11:43 -07001760/*
1761 * Do static interleaving for a VMA with known offset @n. Returns the n'th
1762 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
1763 * number of present nodes.
1764 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765static unsigned offset_il_node(struct mempolicy *pol,
Andrew Mortonfee83b32016-05-19 17:11:43 -07001766 struct vm_area_struct *vma, unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001768 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001769 unsigned target;
Andrew Mortonfee83b32016-05-19 17:11:43 -07001770 int i;
1771 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
David Rientjesf5b087b2008-04-28 02:12:27 -07001773 if (!nnodes)
1774 return numa_node_id();
Andrew Mortonfee83b32016-05-19 17:11:43 -07001775 target = (unsigned int)n % nnodes;
1776 nid = first_node(pol->v.nodes);
1777 for (i = 0; i < target; i++)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001778 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return nid;
1780}
1781
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001782/* Determine a node number for interleave */
1783static inline unsigned interleave_nid(struct mempolicy *pol,
1784 struct vm_area_struct *vma, unsigned long addr, int shift)
1785{
1786 if (vma) {
1787 unsigned long off;
1788
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001789 /*
1790 * for small pages, there is no difference between
1791 * shift and PAGE_SHIFT, so the bit-shift is safe.
1792 * for huge pages, since vm_pgoff is in units of small
1793 * pages, we need to shift off the always 0 bits to get
1794 * a useful offset.
1795 */
1796 BUG_ON(shift < PAGE_SHIFT);
1797 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001798 off += (addr - vma->vm_start) >> shift;
1799 return offset_il_node(pol, vma, off);
1800 } else
1801 return interleave_nodes(pol);
1802}
1803
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001804#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001805/*
1806 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001807 * @vma: virtual memory area whose policy is sought
1808 * @addr: address in @vma for shared policy lookup and interleave policy
1809 * @gfp_flags: for requested zone
1810 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1811 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001812 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001813 * Returns a zonelist suitable for a huge page allocation and a pointer
1814 * to the struct mempolicy for conditional unref after allocation.
1815 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1816 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001817 *
Mel Gormand26914d2014-04-03 14:47:24 -07001818 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001819 */
Mel Gorman396faf02007-07-17 04:03:13 -07001820struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001821 gfp_t gfp_flags, struct mempolicy **mpol,
1822 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001823{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001824 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001825
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001826 *mpol = get_vma_policy(vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001827 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001828
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001829 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1830 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001831 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001832 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001833 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001834 if ((*mpol)->mode == MPOL_BIND)
1835 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001836 }
1837 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001838}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001839
1840/*
1841 * init_nodemask_of_mempolicy
1842 *
1843 * If the current task's mempolicy is "default" [NULL], return 'false'
1844 * to indicate default policy. Otherwise, extract the policy nodemask
1845 * for 'bind' or 'interleave' policy into the argument nodemask, or
1846 * initialize the argument nodemask to contain the single node for
1847 * 'preferred' or 'local' policy and return 'true' to indicate presence
1848 * of non-default mempolicy.
1849 *
1850 * We don't bother with reference counting the mempolicy [mpol_get/put]
1851 * because the current task is examining it's own mempolicy and a task's
1852 * mempolicy is only ever changed by the task itself.
1853 *
1854 * N.B., it is the caller's responsibility to free a returned nodemask.
1855 */
1856bool init_nodemask_of_mempolicy(nodemask_t *mask)
1857{
1858 struct mempolicy *mempolicy;
1859 int nid;
1860
1861 if (!(mask && current->mempolicy))
1862 return false;
1863
Miao Xiec0ff7452010-05-24 14:32:08 -07001864 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001865 mempolicy = current->mempolicy;
1866 switch (mempolicy->mode) {
1867 case MPOL_PREFERRED:
1868 if (mempolicy->flags & MPOL_F_LOCAL)
1869 nid = numa_node_id();
1870 else
1871 nid = mempolicy->v.preferred_node;
1872 init_nodemask_of_node(mask, nid);
1873 break;
1874
1875 case MPOL_BIND:
1876 /* Fall through */
1877 case MPOL_INTERLEAVE:
1878 *mask = mempolicy->v.nodes;
1879 break;
1880
1881 default:
1882 BUG();
1883 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001884 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001885
1886 return true;
1887}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001888#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001889
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001890/*
1891 * mempolicy_nodemask_intersects
1892 *
1893 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1894 * policy. Otherwise, check for intersection between mask and the policy
1895 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1896 * policy, always return true since it may allocate elsewhere on fallback.
1897 *
1898 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1899 */
1900bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1901 const nodemask_t *mask)
1902{
1903 struct mempolicy *mempolicy;
1904 bool ret = true;
1905
1906 if (!mask)
1907 return ret;
1908 task_lock(tsk);
1909 mempolicy = tsk->mempolicy;
1910 if (!mempolicy)
1911 goto out;
1912
1913 switch (mempolicy->mode) {
1914 case MPOL_PREFERRED:
1915 /*
1916 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1917 * allocate from, they may fallback to other nodes when oom.
1918 * Thus, it's possible for tsk to have allocated memory from
1919 * nodes in mask.
1920 */
1921 break;
1922 case MPOL_BIND:
1923 case MPOL_INTERLEAVE:
1924 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1925 break;
1926 default:
1927 BUG();
1928 }
1929out:
1930 task_unlock(tsk);
1931 return ret;
1932}
1933
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934/* Allocate a page in interleaved policy.
1935 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001936static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1937 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938{
1939 struct zonelist *zl;
1940 struct page *page;
1941
Mel Gorman0e884602008-04-28 02:12:14 -07001942 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001944 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001945 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return page;
1947}
1948
1949/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001950 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 *
1952 * @gfp:
1953 * %GFP_USER user allocation.
1954 * %GFP_KERNEL kernel allocations,
1955 * %GFP_HIGHMEM highmem/user allocations,
1956 * %GFP_FS allocation should not call back into a file system.
1957 * %GFP_ATOMIC don't sleep.
1958 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001959 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 * @vma: Pointer to VMA or NULL if not available.
1961 * @addr: Virtual Address of the allocation. Must be inside the VMA.
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001962 * @node: Which node to prefer for allocation (modulo policy).
1963 * @hugepage: for hugepages try only the preferred node if possible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 *
1965 * This function allocates a page from the kernel page pool and applies
1966 * a NUMA policy associated with the VMA or the current process.
1967 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1968 * mm_struct of the VMA to prevent it from going away. Should be used for
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001969 * all allocations for pages that will be mapped into user space. Returns
1970 * NULL when no page can be allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 */
1972struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001973alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001974 unsigned long addr, int node, bool hugepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975{
Mel Gormancc9a6c82012-03-21 16:34:11 -07001976 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07001977 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001978 unsigned int cpuset_mems_cookie;
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001979 struct zonelist *zl;
1980 nodemask_t *nmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
Mel Gormancc9a6c82012-03-21 16:34:11 -07001982retry_cpuset:
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001983 pol = get_vma_policy(vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07001984 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07001985
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001986 if (pol->mode == MPOL_INTERLEAVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001988
Andi Kleen8eac5632011-02-25 14:44:28 -08001989 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001990 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001991 page = alloc_page_interleave(gfp, order, nid);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001992 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 }
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001994
Vlastimil Babka0867a572015-06-24 16:58:48 -07001995 if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
1996 int hpage_node = node;
1997
1998 /*
1999 * For hugepage allocation and non-interleave policy which
2000 * allows the current node (or other explicitly preferred
2001 * node) we only try to allocate from the current/preferred
2002 * node and don't fall back to other nodes, as the cost of
2003 * remote accesses would likely offset THP benefits.
2004 *
2005 * If the policy is interleave, or does not allow the current
2006 * node in its nodemask, we allocate the standard way.
2007 */
2008 if (pol->mode == MPOL_PREFERRED &&
2009 !(pol->flags & MPOL_F_LOCAL))
2010 hpage_node = pol->v.preferred_node;
2011
2012 nmask = policy_nodemask(gfp, pol);
2013 if (!nmask || node_isset(hpage_node, *nmask)) {
2014 mpol_cond_put(pol);
Vlastimil Babka96db8002015-09-08 15:03:50 -07002015 page = __alloc_pages_node(hpage_node,
Vlastimil Babka0867a572015-06-24 16:58:48 -07002016 gfp | __GFP_THISNODE, order);
2017 goto out;
2018 }
2019 }
2020
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002021 nmask = policy_nodemask(gfp, pol);
2022 zl = policy_zonelist(gfp, pol, node);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002023 page = __alloc_pages_nodemask(gfp, order, zl, nmask);
Vlastimil Babka9b1a1ae2017-01-24 15:18:18 -08002024 mpol_cond_put(pol);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002025out:
Mel Gormand26914d2014-04-03 14:47:24 -07002026 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002027 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002028 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030
2031/**
2032 * alloc_pages_current - Allocate pages.
2033 *
2034 * @gfp:
2035 * %GFP_USER user allocation,
2036 * %GFP_KERNEL kernel allocation,
2037 * %GFP_HIGHMEM highmem allocation,
2038 * %GFP_FS don't call back into a file system.
2039 * %GFP_ATOMIC don't sleep.
2040 * @order: Power of two of allocation size in pages. 0 is a single page.
2041 *
2042 * Allocate a page from the kernel page pool. When not in
2043 * interrupt context and apply the current process NUMA policy.
2044 * Returns NULL when no page can be allocated.
2045 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002046 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 * 1) it's ok to take cpuset_sem (can WAIT), and
2048 * 2) allocating for current task (not interrupt).
2049 */
Al Virodd0fc662005-10-07 07:46:04 +01002050struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051{
Oleg Nesterov8d902742014-10-09 15:27:45 -07002052 struct mempolicy *pol = &default_policy;
Miao Xiec0ff7452010-05-24 14:32:08 -07002053 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002054 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Oleg Nesterov8d902742014-10-09 15:27:45 -07002056 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2057 pol = get_task_policy(current);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002058
Mel Gormancc9a6c82012-03-21 16:34:11 -07002059retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002060 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002061
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002062 /*
2063 * No reference counting needed for current->mempolicy
2064 * nor system default_policy
2065 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002066 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002067 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2068 else
2069 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002070 policy_zonelist(gfp, pol, numa_node_id()),
2071 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002072
Mel Gormand26914d2014-04-03 14:47:24 -07002073 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002074 goto retry_cpuset;
2075
Miao Xiec0ff7452010-05-24 14:32:08 -07002076 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077}
2078EXPORT_SYMBOL(alloc_pages_current);
2079
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002080int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2081{
2082 struct mempolicy *pol = mpol_dup(vma_policy(src));
2083
2084 if (IS_ERR(pol))
2085 return PTR_ERR(pol);
2086 dst->vm_policy = pol;
2087 return 0;
2088}
2089
Paul Jackson42253992006-01-08 01:01:59 -08002090/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002091 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002092 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2093 * with the mems_allowed returned by cpuset_mems_allowed(). This
2094 * keeps mempolicies cpuset relative after its cpuset moves. See
2095 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002096 *
2097 * current's mempolicy may be rebinded by the other task(the task that changes
2098 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002099 */
Paul Jackson42253992006-01-08 01:01:59 -08002100
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002101/* Slow path of a mempolicy duplicate */
2102struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2105
2106 if (!new)
2107 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002108
2109 /* task's mempolicy is protected by alloc_lock */
2110 if (old == current->mempolicy) {
2111 task_lock(current);
2112 *new = *old;
2113 task_unlock(current);
2114 } else
2115 *new = *old;
2116
Paul Jackson42253992006-01-08 01:01:59 -08002117 if (current_cpuset_is_being_rebound()) {
2118 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002119 if (new->flags & MPOL_F_REBINDING)
2120 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2121 else
2122 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 return new;
2126}
2127
2128/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002129bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
2131 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002132 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002133 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002134 return false;
Bob Liu19800502010-05-24 14:32:01 -07002135 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002136 return false;
Bob Liu19800502010-05-24 14:32:01 -07002137 if (mpol_store_user_nodemask(a))
2138 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002139 return false;
Bob Liu19800502010-05-24 14:32:01 -07002140
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002141 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002142 case MPOL_BIND:
2143 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002145 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002147 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 default:
2149 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002150 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 }
2152}
2153
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 * Shared memory backing store policy support.
2156 *
2157 * Remember policies even when nobody has shared memory mapped.
2158 * The policies are kept in Red-Black tree linked from the inode.
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002159 * They are protected by the sp->lock rwlock, which should be held
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 * for any accesses to the tree.
2161 */
2162
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002163/*
2164 * lookup first element intersecting start-end. Caller holds sp->lock for
2165 * reading or for writing
2166 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167static struct sp_node *
2168sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2169{
2170 struct rb_node *n = sp->root.rb_node;
2171
2172 while (n) {
2173 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2174
2175 if (start >= p->end)
2176 n = n->rb_right;
2177 else if (end <= p->start)
2178 n = n->rb_left;
2179 else
2180 break;
2181 }
2182 if (!n)
2183 return NULL;
2184 for (;;) {
2185 struct sp_node *w = NULL;
2186 struct rb_node *prev = rb_prev(n);
2187 if (!prev)
2188 break;
2189 w = rb_entry(prev, struct sp_node, nd);
2190 if (w->end <= start)
2191 break;
2192 n = prev;
2193 }
2194 return rb_entry(n, struct sp_node, nd);
2195}
2196
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002197/*
2198 * Insert a new shared policy into the list. Caller holds sp->lock for
2199 * writing.
2200 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2202{
2203 struct rb_node **p = &sp->root.rb_node;
2204 struct rb_node *parent = NULL;
2205 struct sp_node *nd;
2206
2207 while (*p) {
2208 parent = *p;
2209 nd = rb_entry(parent, struct sp_node, nd);
2210 if (new->start < nd->start)
2211 p = &(*p)->rb_left;
2212 else if (new->end > nd->end)
2213 p = &(*p)->rb_right;
2214 else
2215 BUG();
2216 }
2217 rb_link_node(&new->nd, parent, p);
2218 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002219 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002220 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221}
2222
2223/* Find shared policy intersecting idx */
2224struct mempolicy *
2225mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2226{
2227 struct mempolicy *pol = NULL;
2228 struct sp_node *sn;
2229
2230 if (!sp->root.rb_node)
2231 return NULL;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002232 read_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 sn = sp_lookup(sp, idx, idx+1);
2234 if (sn) {
2235 mpol_get(sn->policy);
2236 pol = sn->policy;
2237 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002238 read_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 return pol;
2240}
2241
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002242static void sp_free(struct sp_node *n)
2243{
2244 mpol_put(n->policy);
2245 kmem_cache_free(sn_cache, n);
2246}
2247
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002248/**
2249 * mpol_misplaced - check whether current page node is valid in policy
2250 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002251 * @page: page to be checked
2252 * @vma: vm area where page mapped
2253 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002254 *
2255 * Lookup current policy node id for vma,addr and "compare to" page's
2256 * node id.
2257 *
2258 * Returns:
2259 * -1 - not misplaced, page is in the right node
2260 * node - node id where the page should be
2261 *
2262 * Policy determination "mimics" alloc_page_vma().
2263 * Called from fault path where we know the vma and faulting address.
2264 */
2265int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2266{
2267 struct mempolicy *pol;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002268 struct zoneref *z;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002269 int curnid = page_to_nid(page);
2270 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002271 int thiscpu = raw_smp_processor_id();
2272 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002273 int polnid = -1;
2274 int ret = -1;
2275
2276 BUG_ON(!vma);
2277
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002278 pol = get_vma_policy(vma, addr);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002279 if (!(pol->flags & MPOL_F_MOF))
2280 goto out;
2281
2282 switch (pol->mode) {
2283 case MPOL_INTERLEAVE:
2284 BUG_ON(addr >= vma->vm_end);
2285 BUG_ON(addr < vma->vm_start);
2286
2287 pgoff = vma->vm_pgoff;
2288 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2289 polnid = offset_il_node(pol, vma, pgoff);
2290 break;
2291
2292 case MPOL_PREFERRED:
2293 if (pol->flags & MPOL_F_LOCAL)
2294 polnid = numa_node_id();
2295 else
2296 polnid = pol->v.preferred_node;
2297 break;
2298
2299 case MPOL_BIND:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002300
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002301 /*
2302 * allows binding to multiple nodes.
2303 * use current page if in policy nodemask,
2304 * else select nearest allowed node, if any.
2305 * If no allowed nodes, use current [!misplaced].
2306 */
2307 if (node_isset(curnid, pol->v.nodes))
2308 goto out;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002309 z = first_zones_zonelist(
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002310 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2311 gfp_zone(GFP_HIGHUSER),
Mel Gormanc33d6c02016-05-19 17:14:10 -07002312 &pol->v.nodes);
2313 polnid = z->zone->node;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002314 break;
2315
2316 default:
2317 BUG();
2318 }
Mel Gorman5606e382012-11-02 18:19:13 +00002319
2320 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002321 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002322 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002323
Rik van Riel10f39042014-01-27 17:03:44 -05002324 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002325 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002326 }
2327
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002328 if (curnid != polnid)
2329 ret = polnid;
2330out:
2331 mpol_cond_put(pol);
2332
2333 return ret;
2334}
2335
David Rientjesc11600e2016-09-01 16:15:07 -07002336/*
2337 * Drop the (possibly final) reference to task->mempolicy. It needs to be
2338 * dropped after task->mempolicy is set to NULL so that any allocation done as
2339 * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed
2340 * policy.
2341 */
2342void mpol_put_task_policy(struct task_struct *task)
2343{
2344 struct mempolicy *pol;
2345
2346 task_lock(task);
2347 pol = task->mempolicy;
2348 task->mempolicy = NULL;
2349 task_unlock(task);
2350 mpol_put(pol);
2351}
2352
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2354{
Paul Mundt140d5a42007-07-15 23:38:16 -07002355 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002357 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358}
2359
Mel Gorman42288fe2012-12-21 23:10:25 +00002360static void sp_node_init(struct sp_node *node, unsigned long start,
2361 unsigned long end, struct mempolicy *pol)
2362{
2363 node->start = start;
2364 node->end = end;
2365 node->policy = pol;
2366}
2367
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002368static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2369 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002371 struct sp_node *n;
2372 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002374 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 if (!n)
2376 return NULL;
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002377
2378 newpol = mpol_dup(pol);
2379 if (IS_ERR(newpol)) {
2380 kmem_cache_free(sn_cache, n);
2381 return NULL;
2382 }
2383 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002384 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002385
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 return n;
2387}
2388
2389/* Replace a policy range. */
2390static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2391 unsigned long end, struct sp_node *new)
2392{
Mel Gormanb22d1272012-10-08 16:29:17 -07002393 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002394 struct sp_node *n_new = NULL;
2395 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002396 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
Mel Gorman42288fe2012-12-21 23:10:25 +00002398restart:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002399 write_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 n = sp_lookup(sp, start, end);
2401 /* Take care of old policies in the same range. */
2402 while (n && n->start < end) {
2403 struct rb_node *next = rb_next(&n->nd);
2404 if (n->start >= start) {
2405 if (n->end <= end)
2406 sp_delete(sp, n);
2407 else
2408 n->start = end;
2409 } else {
2410 /* Old policy spanning whole new range. */
2411 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002412 if (!n_new)
2413 goto alloc_new;
2414
2415 *mpol_new = *n->policy;
2416 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002417 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002419 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002420 n_new = NULL;
2421 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 break;
2423 } else
2424 n->end = start;
2425 }
2426 if (!next)
2427 break;
2428 n = rb_entry(next, struct sp_node, nd);
2429 }
2430 if (new)
2431 sp_insert(sp, new);
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002432 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002433 ret = 0;
2434
2435err_out:
2436 if (mpol_new)
2437 mpol_put(mpol_new);
2438 if (n_new)
2439 kmem_cache_free(sn_cache, n_new);
2440
Mel Gormanb22d1272012-10-08 16:29:17 -07002441 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002442
2443alloc_new:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002444 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002445 ret = -ENOMEM;
2446 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2447 if (!n_new)
2448 goto err_out;
2449 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2450 if (!mpol_new)
2451 goto err_out;
2452 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453}
2454
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002455/**
2456 * mpol_shared_policy_init - initialize shared policy for inode
2457 * @sp: pointer to inode shared policy
2458 * @mpol: struct mempolicy to install
2459 *
2460 * Install non-NULL @mpol in inode's shared policy rb-tree.
2461 * On entry, the current task has a reference on a non-NULL @mpol.
2462 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002463 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002464 */
2465void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002466{
Miao Xie58568d22009-06-16 15:31:49 -07002467 int ret;
2468
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002469 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002470 rwlock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002471
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002472 if (mpol) {
2473 struct vm_area_struct pvma;
2474 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002475 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002476
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002477 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002478 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002479 /* contextualize the tmpfs mount point mempolicy */
2480 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002481 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002482 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002483
2484 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002485 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002486 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002487 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002488 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002489
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002490 /* Create pseudo-vma that contains just the policy */
2491 memset(&pvma, 0, sizeof(struct vm_area_struct));
2492 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2493 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002494
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002495put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002496 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002497free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002498 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002499put_mpol:
2500 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002501 }
2502}
2503
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504int mpol_set_shared_policy(struct shared_policy *info,
2505 struct vm_area_struct *vma, struct mempolicy *npol)
2506{
2507 int err;
2508 struct sp_node *new = NULL;
2509 unsigned long sz = vma_pages(vma);
2510
David Rientjes028fec42008-04-28 02:12:25 -07002511 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002513 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002514 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002515 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
2517 if (npol) {
2518 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2519 if (!new)
2520 return -ENOMEM;
2521 }
2522 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2523 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002524 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 return err;
2526}
2527
2528/* Free a backing policy store on inode delete. */
2529void mpol_free_shared_policy(struct shared_policy *p)
2530{
2531 struct sp_node *n;
2532 struct rb_node *next;
2533
2534 if (!p->root.rb_node)
2535 return;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002536 write_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 next = rb_first(&p->root);
2538 while (next) {
2539 n = rb_entry(next, struct sp_node, nd);
2540 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002541 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002543 write_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544}
2545
Mel Gorman1a687c22012-11-22 11:16:36 +00002546#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002547static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002548
2549static void __init check_numabalancing_enable(void)
2550{
2551 bool numabalancing_default = false;
2552
2553 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2554 numabalancing_default = true;
2555
Mel Gormanc2976632014-01-29 14:05:42 -08002556 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2557 if (numabalancing_override)
2558 set_numabalancing_state(numabalancing_override == 1);
2559
Mel Gormanb0dc2b92015-05-14 15:17:09 -07002560 if (num_online_nodes() > 1 && !numabalancing_override) {
Joe Perches756a025f02016-03-17 14:19:47 -07002561 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
Mel Gormanc2976632014-01-29 14:05:42 -08002562 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002563 set_numabalancing_state(numabalancing_default);
2564 }
2565}
2566
2567static int __init setup_numabalancing(char *str)
2568{
2569 int ret = 0;
2570 if (!str)
2571 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002572
2573 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002574 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002575 ret = 1;
2576 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002577 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002578 ret = 1;
2579 }
2580out:
2581 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002582 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002583
2584 return ret;
2585}
2586__setup("numa_balancing=", setup_numabalancing);
2587#else
2588static inline void __init check_numabalancing_enable(void)
2589{
2590}
2591#endif /* CONFIG_NUMA_BALANCING */
2592
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593/* assumes fs == KERNEL_DS */
2594void __init numa_policy_init(void)
2595{
Paul Mundtb71636e2007-07-15 23:38:15 -07002596 nodemask_t interleave_nodes;
2597 unsigned long largest = 0;
2598 int nid, prefer = 0;
2599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 policy_cache = kmem_cache_create("numa_policy",
2601 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002602 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
2604 sn_cache = kmem_cache_create("shared_policy_node",
2605 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002606 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Mel Gorman5606e382012-11-02 18:19:13 +00002608 for_each_node(nid) {
2609 preferred_node_policy[nid] = (struct mempolicy) {
2610 .refcnt = ATOMIC_INIT(1),
2611 .mode = MPOL_PREFERRED,
2612 .flags = MPOL_F_MOF | MPOL_F_MORON,
2613 .v = { .preferred_node = nid, },
2614 };
2615 }
2616
Paul Mundtb71636e2007-07-15 23:38:15 -07002617 /*
2618 * Set interleaving policy for system init. Interleaving is only
2619 * enabled across suitably sized nodes (default is >= 16MB), or
2620 * fall back to the largest node if they're all smaller.
2621 */
2622 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002623 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002624 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Paul Mundtb71636e2007-07-15 23:38:15 -07002626 /* Preserve the largest node */
2627 if (largest < total_pages) {
2628 largest = total_pages;
2629 prefer = nid;
2630 }
2631
2632 /* Interleave this node? */
2633 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2634 node_set(nid, interleave_nodes);
2635 }
2636
2637 /* All too small, use the largest */
2638 if (unlikely(nodes_empty(interleave_nodes)))
2639 node_set(prefer, interleave_nodes);
2640
David Rientjes028fec42008-04-28 02:12:25 -07002641 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002642 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002643
2644 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645}
2646
Christoph Lameter8bccd852005-10-29 18:16:59 -07002647/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648void numa_default_policy(void)
2649{
David Rientjes028fec42008-04-28 02:12:25 -07002650 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651}
Paul Jackson68860ec2005-10-30 15:02:36 -08002652
Paul Jackson42253992006-01-08 01:01:59 -08002653/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002654 * Parse and format mempolicy from/to strings
2655 */
2656
2657/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002658 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002659 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002660static const char * const policy_modes[] =
2661{
2662 [MPOL_DEFAULT] = "default",
2663 [MPOL_PREFERRED] = "prefer",
2664 [MPOL_BIND] = "bind",
2665 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002666 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002667};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002668
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002669
2670#ifdef CONFIG_TMPFS
2671/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002672 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002673 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002674 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002675 *
2676 * Format of input:
2677 * <mode>[=<flags>][:<nodelist>]
2678 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002679 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002680 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002681int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002682{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002683 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002684 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002685 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002686 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002687 char *nodelist = strchr(str, ':');
2688 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002689 int err = 1;
2690
2691 if (nodelist) {
2692 /* NUL-terminate mode or flags string */
2693 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002694 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002695 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002696 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002697 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002698 } else
2699 nodes_clear(nodes);
2700
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002701 if (flags)
2702 *flags++ = '\0'; /* terminate mode string */
2703
Peter Zijlstra479e2802012-10-25 14:16:28 +02002704 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002705 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002706 break;
2707 }
2708 }
Mel Gormana7200942012-11-16 09:37:58 +00002709 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002710 goto out;
2711
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002712 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002713 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002714 /*
2715 * Insist on a nodelist of one node only
2716 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002717 if (nodelist) {
2718 char *rest = nodelist;
2719 while (isdigit(*rest))
2720 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002721 if (*rest)
2722 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002723 }
2724 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002725 case MPOL_INTERLEAVE:
2726 /*
2727 * Default to online nodes with memory if no nodelist
2728 */
2729 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002730 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002731 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002732 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002733 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002734 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002735 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002736 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002737 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002738 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002739 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002740 case MPOL_DEFAULT:
2741 /*
2742 * Insist on a empty nodelist
2743 */
2744 if (!nodelist)
2745 err = 0;
2746 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002747 case MPOL_BIND:
2748 /*
2749 * Insist on a nodelist
2750 */
2751 if (!nodelist)
2752 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002753 }
2754
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002755 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002756 if (flags) {
2757 /*
2758 * Currently, we only support two mutually exclusive
2759 * mode flags.
2760 */
2761 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002762 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002763 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002764 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002765 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002766 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002767 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002768
2769 new = mpol_new(mode, mode_flags, &nodes);
2770 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002771 goto out;
2772
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002773 /*
2774 * Save nodes for mpol_to_str() to show the tmpfs mount options
2775 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2776 */
2777 if (mode != MPOL_PREFERRED)
2778 new->v.nodes = nodes;
2779 else if (nodelist)
2780 new->v.preferred_node = first_node(nodes);
2781 else
2782 new->flags |= MPOL_F_LOCAL;
2783
2784 /*
2785 * Save nodes for contextualization: this will be used to "clone"
2786 * the mempolicy in a specific context [cpuset] at a later time.
2787 */
2788 new->w.user_nodemask = nodes;
2789
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002790 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002791
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002792out:
2793 /* Restore string for error message */
2794 if (nodelist)
2795 *--nodelist = ':';
2796 if (flags)
2797 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002798 if (!err)
2799 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002800 return err;
2801}
2802#endif /* CONFIG_TMPFS */
2803
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002804/**
2805 * mpol_to_str - format a mempolicy structure for printing
2806 * @buffer: to contain formatted mempolicy string
2807 * @maxlen: length of @buffer
2808 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002809 *
David Rientjes948927e2013-11-12 15:07:28 -08002810 * Convert @pol into a string. If @buffer is too short, truncate the string.
2811 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2812 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002813 */
David Rientjes948927e2013-11-12 15:07:28 -08002814void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002815{
2816 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002817 nodemask_t nodes = NODE_MASK_NONE;
2818 unsigned short mode = MPOL_DEFAULT;
2819 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002820
David Rientjes8790c712014-01-30 15:46:08 -08002821 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002822 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002823 flags = pol->flags;
2824 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002825
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002826 switch (mode) {
2827 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002828 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002829 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002830 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002831 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002832 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002833 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002834 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002835 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002836 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002837 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002838 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002839 default:
David Rientjes948927e2013-11-12 15:07:28 -08002840 WARN_ON_ONCE(1);
2841 snprintf(p, maxlen, "unknown");
2842 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002843 }
2844
David Rientjesb7a9f422013-11-21 14:32:06 -08002845 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002846
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002847 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002848 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002849
Lee Schermerhorn22919902008-04-28 02:13:22 -07002850 /*
2851 * Currently, the only defined flags are mutually exclusive
2852 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002853 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002854 p += snprintf(p, buffer + maxlen - p, "static");
2855 else if (flags & MPOL_F_RELATIVE_NODES)
2856 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002857 }
2858
Tejun Heo9e763e02015-02-13 14:38:02 -08002859 if (!nodes_empty(nodes))
2860 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2861 nodemask_pr_args(&nodes));
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002862}