blob: 959a8b8c7350c28ffa6cacced61b4c24977479da [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.
29 * As a special case node -1 here means do the allocation
30 * on the local CPU. This is normally identical to default,
31 * but useful to set in a VMA when you have a non default
32 * process policy.
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
68#include <linux/mempolicy.h>
69#include <linux/mm.h>
70#include <linux/highmem.h>
71#include <linux/hugetlb.h>
72#include <linux/kernel.h>
73#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <linux/nodemask.h>
75#include <linux/cpuset.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/slab.h>
77#include <linux/string.h>
78#include <linux/module.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070079#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <linux/interrupt.h>
81#include <linux/init.h>
82#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080083#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080084#include <linux/seq_file.h>
85#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080086#include <linux/migrate.h>
Hugh Dickins62b61f62009-12-14 17:59:33 -080087#include <linux/ksm.h>
Christoph Lameter95a402c2006-06-23 02:03:53 -070088#include <linux/rmap.h>
David Quigley86c3a762006-06-23 02:04:02 -070089#include <linux/security.h>
Adrian Bunkdbcb0f12007-10-16 01:26:26 -070090#include <linux/syscalls.h>
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -070091#include <linux/ctype.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080092#include <linux/mm_inline.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080093
Linus Torvalds1da177e2005-04-16 15:20:36 -070094#include <asm/tlbflush.h>
95#include <asm/uaccess.h>
96
Nick Piggin62695a82008-10-18 20:26:09 -070097#include "internal.h"
98
Christoph Lameter38e35862006-01-08 01:01:01 -080099/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800100#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800101#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800102#define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800103
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800104static struct kmem_cache *policy_cache;
105static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/* Highest zone. An specific allocation for a zone below that is not
108 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800109enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700111/*
112 * run-time system-wide default policy => local allocation
113 */
Andi Kleend42c6992005-07-06 19:56:03 +0200114struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700116 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700117 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118};
119
David Rientjes37012942008-04-28 02:12:33 -0700120static const struct mempolicy_operations {
121 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700122 /*
123 * If read-side task has no lock to protect task->mempolicy, write-side
124 * task will rebind the task->mempolicy by two step. The first step is
125 * setting all the newly nodes, and the second step is cleaning all the
126 * disallowed nodes. In this way, we can avoid finding no node to alloc
127 * page.
128 * If we have a lock to protect task->mempolicy in read-side, we do
129 * rebind directly.
130 *
131 * step:
132 * MPOL_REBIND_ONCE - do rebind work at once
133 * MPOL_REBIND_STEP1 - set all the newly nodes
134 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
135 */
136 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
137 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700138} mpol_ops[MPOL_MAX];
139
Mel Gorman19770b32008-04-28 02:12:18 -0700140/* Check that the nodemask contains at least one populated zone */
David Rientjes37012942008-04-28 02:12:33 -0700141static int is_valid_nodemask(const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
Mel Gorman19770b32008-04-28 02:12:18 -0700143 int nd, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Mel Gorman19770b32008-04-28 02:12:18 -0700145 for_each_node_mask(nd, *nodemask) {
146 struct zone *z;
147
148 for (k = 0; k <= policy_zone; k++) {
149 z = &NODE_DATA(nd)->node_zones[k];
150 if (z->present_pages > 0)
151 return 1;
Andi Kleendd942ae2006-02-17 01:39:16 +0100152 }
153 }
Mel Gorman19770b32008-04-28 02:12:18 -0700154
155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
David Rientjesf5b087b2008-04-28 02:12:27 -0700158static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
159{
Bob Liu6d556292010-05-24 14:31:59 -0700160 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700161}
162
163static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
164 const nodemask_t *rel)
165{
166 nodemask_t tmp;
167 nodes_fold(tmp, *orig, nodes_weight(*rel));
168 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700169}
170
David Rientjes37012942008-04-28 02:12:33 -0700171static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
172{
173 if (nodes_empty(*nodes))
174 return -EINVAL;
175 pol->v.nodes = *nodes;
176 return 0;
177}
178
179static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
180{
181 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700182 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700183 else if (nodes_empty(*nodes))
184 return -EINVAL; /* no allowed nodes */
185 else
186 pol->v.preferred_node = first_node(*nodes);
187 return 0;
188}
189
190static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
191{
192 if (!is_valid_nodemask(nodes))
193 return -EINVAL;
194 pol->v.nodes = *nodes;
195 return 0;
196}
197
Miao Xie58568d22009-06-16 15:31:49 -0700198/*
199 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
200 * any, for the new policy. mpol_new() has already validated the nodes
201 * parameter with respect to the policy mode and flags. But, we need to
202 * handle an empty nodemask with MPOL_PREFERRED here.
203 *
204 * Must be called holding task's alloc_lock to protect task's mems_allowed
205 * and mempolicy. May also be called holding the mmap_semaphore for write.
206 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700207static int mpol_set_nodemask(struct mempolicy *pol,
208 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700209{
Miao Xie58568d22009-06-16 15:31:49 -0700210 int ret;
211
212 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
213 if (pol == NULL)
214 return 0;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700215 /* Check N_HIGH_MEMORY */
216 nodes_and(nsc->mask1,
217 cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700218
219 VM_BUG_ON(!nodes);
220 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
221 nodes = NULL; /* explicit local allocation */
222 else {
223 if (pol->flags & MPOL_F_RELATIVE_NODES)
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700224 mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700225 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700226 nodes_and(nsc->mask2, *nodes, nsc->mask1);
227
Miao Xie58568d22009-06-16 15:31:49 -0700228 if (mpol_store_user_nodemask(pol))
229 pol->w.user_nodemask = *nodes;
230 else
231 pol->w.cpuset_mems_allowed =
232 cpuset_current_mems_allowed;
233 }
234
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700235 if (nodes)
236 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
237 else
238 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700239 return ret;
240}
241
242/*
243 * This function just creates a new policy, does some check and simple
244 * initialization. You must invoke mpol_set_nodemask() to set nodes.
245 */
David Rientjes028fec42008-04-28 02:12:25 -0700246static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
247 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 struct mempolicy *policy;
250
David Rientjes028fec42008-04-28 02:12:25 -0700251 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
252 mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
Paul Mundt140d5a42007-07-15 23:38:16 -0700253
David Rientjes3e1f0642008-04-28 02:12:34 -0700254 if (mode == MPOL_DEFAULT) {
255 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700256 return ERR_PTR(-EINVAL);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700257 return NULL; /* simply delete any existing policy */
David Rientjes37012942008-04-28 02:12:33 -0700258 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700259 VM_BUG_ON(!nodes);
260
261 /*
262 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
263 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
264 * All other modes require a valid pointer to a non-empty nodemask.
265 */
266 if (mode == MPOL_PREFERRED) {
267 if (nodes_empty(*nodes)) {
268 if (((flags & MPOL_F_STATIC_NODES) ||
269 (flags & MPOL_F_RELATIVE_NODES)))
270 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700271 }
272 } else if (nodes_empty(*nodes))
273 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
275 if (!policy)
276 return ERR_PTR(-ENOMEM);
277 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700278 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700279 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700280
David Rientjes37012942008-04-28 02:12:33 -0700281 return policy;
282}
283
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700284/* Slow path of a mpol destructor. */
285void __mpol_put(struct mempolicy *p)
286{
287 if (!atomic_dec_and_test(&p->refcnt))
288 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700289 kmem_cache_free(policy_cache, p);
290}
291
Miao Xie708c1bb2010-05-24 14:32:07 -0700292static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
293 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700294{
295}
296
Miao Xie708c1bb2010-05-24 14:32:07 -0700297/*
298 * step:
299 * MPOL_REBIND_ONCE - do rebind work at once
300 * MPOL_REBIND_STEP1 - set all the newly nodes
301 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
302 */
303static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
304 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700305{
306 nodemask_t tmp;
307
308 if (pol->flags & MPOL_F_STATIC_NODES)
309 nodes_and(tmp, pol->w.user_nodemask, *nodes);
310 else if (pol->flags & MPOL_F_RELATIVE_NODES)
311 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
312 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700313 /*
314 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
315 * result
316 */
317 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
318 nodes_remap(tmp, pol->v.nodes,
319 pol->w.cpuset_mems_allowed, *nodes);
320 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
321 } else if (step == MPOL_REBIND_STEP2) {
322 tmp = pol->w.cpuset_mems_allowed;
323 pol->w.cpuset_mems_allowed = *nodes;
324 } else
325 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700326 }
327
Miao Xie708c1bb2010-05-24 14:32:07 -0700328 if (nodes_empty(tmp))
329 tmp = *nodes;
330
331 if (step == MPOL_REBIND_STEP1)
332 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
333 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
334 pol->v.nodes = tmp;
335 else
336 BUG();
337
David Rientjes37012942008-04-28 02:12:33 -0700338 if (!node_isset(current->il_next, tmp)) {
339 current->il_next = next_node(current->il_next, tmp);
340 if (current->il_next >= MAX_NUMNODES)
341 current->il_next = first_node(tmp);
342 if (current->il_next >= MAX_NUMNODES)
343 current->il_next = numa_node_id();
344 }
345}
346
347static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700348 const nodemask_t *nodes,
349 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700350{
351 nodemask_t tmp;
352
David Rientjes37012942008-04-28 02:12:33 -0700353 if (pol->flags & MPOL_F_STATIC_NODES) {
354 int node = first_node(pol->w.user_nodemask);
355
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700356 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700357 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700358 pol->flags &= ~MPOL_F_LOCAL;
359 } else
360 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700361 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
362 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
363 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700364 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700365 pol->v.preferred_node = node_remap(pol->v.preferred_node,
366 pol->w.cpuset_mems_allowed,
367 *nodes);
368 pol->w.cpuset_mems_allowed = *nodes;
369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
Miao Xie708c1bb2010-05-24 14:32:07 -0700372/*
373 * mpol_rebind_policy - Migrate a policy to a different set of nodes
374 *
375 * If read-side task has no lock to protect task->mempolicy, write-side
376 * task will rebind the task->mempolicy by two step. The first step is
377 * setting all the newly nodes, and the second step is cleaning all the
378 * disallowed nodes. In this way, we can avoid finding no node to alloc
379 * page.
380 * If we have a lock to protect task->mempolicy in read-side, we do
381 * rebind directly.
382 *
383 * step:
384 * MPOL_REBIND_ONCE - do rebind work at once
385 * MPOL_REBIND_STEP1 - set all the newly nodes
386 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
387 */
388static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
389 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700390{
David Rientjes1d0d2682008-04-28 02:12:32 -0700391 if (!pol)
392 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700393 if (!mpol_store_user_nodemask(pol) && step == 0 &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700394 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
395 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700396
397 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
398 return;
399
400 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
401 BUG();
402
403 if (step == MPOL_REBIND_STEP1)
404 pol->flags |= MPOL_F_REBINDING;
405 else if (step == MPOL_REBIND_STEP2)
406 pol->flags &= ~MPOL_F_REBINDING;
407 else if (step >= MPOL_REBIND_NSTEP)
408 BUG();
409
410 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700411}
412
413/*
414 * Wrapper for mpol_rebind_policy() that just requires task
415 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700416 *
417 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700418 */
419
Miao Xie708c1bb2010-05-24 14:32:07 -0700420void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
421 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700422{
Miao Xie708c1bb2010-05-24 14:32:07 -0700423 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700424}
425
426/*
427 * Rebind each vma in mm to new nodemask.
428 *
429 * Call holding a reference to mm. Takes mm->mmap_sem during call.
430 */
431
432void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
433{
434 struct vm_area_struct *vma;
435
436 down_write(&mm->mmap_sem);
437 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700438 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700439 up_write(&mm->mmap_sem);
440}
441
David Rientjes37012942008-04-28 02:12:33 -0700442static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
443 [MPOL_DEFAULT] = {
444 .rebind = mpol_rebind_default,
445 },
446 [MPOL_INTERLEAVE] = {
447 .create = mpol_new_interleave,
448 .rebind = mpol_rebind_nodemask,
449 },
450 [MPOL_PREFERRED] = {
451 .create = mpol_new_preferred,
452 .rebind = mpol_rebind_preferred,
453 },
454 [MPOL_BIND] = {
455 .create = mpol_new_bind,
456 .rebind = mpol_rebind_nodemask,
457 },
458};
459
Christoph Lameter397874d2006-03-06 15:42:53 -0800460static void gather_stats(struct page *, void *, int pte_dirty);
Christoph Lameterfc301282006-01-18 17:42:29 -0800461static void migrate_page_add(struct page *page, struct list_head *pagelist,
462 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800463
Christoph Lameter38e35862006-01-08 01:01:01 -0800464/* Scan through pages checking if pages follow certain conditions. */
Nick Pigginb5810032005-10-29 18:16:12 -0700465static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800466 unsigned long addr, unsigned long end,
467 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800468 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Hugh Dickins91612e02005-06-21 17:15:07 -0700470 pte_t *orig_pte;
471 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700472 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700473
Hugh Dickins705e87c2005-10-29 18:16:27 -0700474 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700475 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800476 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800477 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700478
479 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800481 page = vm_normal_page(vma, addr, *pte);
482 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800484 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800485 * vm_normal_page() filters out zero pages, but there might
486 * still be PageReserved pages to skip, perhaps in a VDSO.
487 * And we cannot move PageKsm pages sensibly or safely yet.
Nick Piggin053837f2006-01-18 17:42:27 -0800488 */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800489 if (PageReserved(page) || PageKsm(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800490 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800491 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800492 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
493 continue;
494
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800495 if (flags & MPOL_MF_STATS)
Christoph Lameter397874d2006-03-06 15:42:53 -0800496 gather_stats(page, private, pte_dirty(*pte));
Nick Piggin053837f2006-01-18 17:42:27 -0800497 else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800498 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800499 else
500 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700501 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700502 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700503 return addr != end;
504}
505
Nick Pigginb5810032005-10-29 18:16:12 -0700506static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800507 unsigned long addr, unsigned long end,
508 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800509 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700510{
511 pmd_t *pmd;
512 unsigned long next;
513
514 pmd = pmd_offset(pud, addr);
515 do {
516 next = pmd_addr_end(addr, end);
Andrea Arcangelibae9c192011-01-13 15:46:46 -0800517 split_huge_page_pmd(vma->vm_mm, pmd);
Hugh Dickins91612e02005-06-21 17:15:07 -0700518 if (pmd_none_or_clear_bad(pmd))
519 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800520 if (check_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800521 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700522 return -EIO;
523 } while (pmd++, addr = next, addr != end);
524 return 0;
525}
526
Nick Pigginb5810032005-10-29 18:16:12 -0700527static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800528 unsigned long addr, unsigned long end,
529 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800530 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700531{
532 pud_t *pud;
533 unsigned long next;
534
535 pud = pud_offset(pgd, addr);
536 do {
537 next = pud_addr_end(addr, end);
538 if (pud_none_or_clear_bad(pud))
539 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800540 if (check_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800541 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700542 return -EIO;
543 } while (pud++, addr = next, addr != end);
544 return 0;
545}
546
Nick Pigginb5810032005-10-29 18:16:12 -0700547static inline int check_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800548 unsigned long addr, unsigned long end,
549 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800550 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700551{
552 pgd_t *pgd;
553 unsigned long next;
554
Nick Pigginb5810032005-10-29 18:16:12 -0700555 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700556 do {
557 next = pgd_addr_end(addr, end);
558 if (pgd_none_or_clear_bad(pgd))
559 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800560 if (check_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800561 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700562 return -EIO;
563 } while (pgd++, addr = next, addr != end);
564 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800567/*
568 * Check if all pages in a range are on a set of nodes.
569 * If pagelist != NULL then isolate pages from the LRU and
570 * put them on the pagelist.
571 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572static struct vm_area_struct *
573check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800574 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
576 int err;
577 struct vm_area_struct *first, *vma, *prev;
578
Nick Piggin053837f2006-01-18 17:42:27 -0800579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 first = find_vma(mm, start);
581 if (!first)
582 return ERR_PTR(-EFAULT);
583 prev = NULL;
584 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800585 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
586 if (!vma->vm_next && vma->vm_end < end)
587 return ERR_PTR(-EFAULT);
588 if (prev && prev->vm_end < vma->vm_start)
589 return ERR_PTR(-EFAULT);
590 }
591 if (!is_vm_hugetlb_page(vma) &&
592 ((flags & MPOL_MF_STRICT) ||
593 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
594 vma_migratable(vma)))) {
Andi Kleen5b952b32005-09-13 01:25:08 -0700595 unsigned long endvma = vma->vm_end;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800596
Andi Kleen5b952b32005-09-13 01:25:08 -0700597 if (endvma > end)
598 endvma = end;
599 if (vma->vm_start > start)
600 start = vma->vm_start;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800601 err = check_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800602 flags, private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (err) {
604 first = ERR_PTR(err);
605 break;
606 }
607 }
608 prev = vma;
609 }
610 return first;
611}
612
613/* Apply policy to a single VMA */
614static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new)
615{
616 int err = 0;
617 struct mempolicy *old = vma->vm_policy;
618
Paul Mundt140d5a42007-07-15 23:38:16 -0700619 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 vma->vm_start, vma->vm_end, vma->vm_pgoff,
621 vma->vm_ops, vma->vm_file,
622 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
623
624 if (vma->vm_ops && vma->vm_ops->set_policy)
625 err = vma->vm_ops->set_policy(vma, new);
626 if (!err) {
627 mpol_get(new);
628 vma->vm_policy = new;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700629 mpol_put(old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
631 return err;
632}
633
634/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800635static int mbind_range(struct mm_struct *mm, unsigned long start,
636 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800639 struct vm_area_struct *prev;
640 struct vm_area_struct *vma;
641 int err = 0;
642 pgoff_t pgoff;
643 unsigned long vmstart;
644 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800646 vma = find_vma_prev(mm, start, &prev);
647 if (!vma || vma->vm_start > start)
648 return -EFAULT;
649
650 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800652 vmstart = max(start, vma->vm_start);
653 vmend = min(end, vma->vm_end);
654
655 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
656 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
657 vma->anon_vma, vma->vm_file, pgoff, new_pol);
658 if (prev) {
659 vma = prev;
660 next = vma->vm_next;
661 continue;
662 }
663 if (vma->vm_start != vmstart) {
664 err = split_vma(vma->vm_mm, vma, vmstart, 1);
665 if (err)
666 goto out;
667 }
668 if (vma->vm_end != vmend) {
669 err = split_vma(vma->vm_mm, vma, vmend, 0);
670 if (err)
671 goto out;
672 }
673 err = policy_vma(vma, new_pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800675 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800677
678 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return err;
680}
681
Paul Jacksonc61afb12006-03-24 03:16:08 -0800682/*
683 * Update task->flags PF_MEMPOLICY bit: set iff non-default
684 * mempolicy. Allows more rapid checking of this (combined perhaps
685 * with other PF_* flag bits) on memory allocation hot code paths.
686 *
687 * If called from outside this file, the task 'p' should -only- be
688 * a newly forked child not yet visible on the task list, because
689 * manipulating the task flags of a visible task is not safe.
690 *
691 * The above limitation is why this routine has the funny name
692 * mpol_fix_fork_child_flag().
693 *
694 * It is also safe to call this with a task pointer of current,
695 * which the static wrapper mpol_set_task_struct_flag() does,
696 * for use within this file.
697 */
698
699void mpol_fix_fork_child_flag(struct task_struct *p)
700{
701 if (p->mempolicy)
702 p->flags |= PF_MEMPOLICY;
703 else
704 p->flags &= ~PF_MEMPOLICY;
705}
706
707static void mpol_set_task_struct_flag(void)
708{
709 mpol_fix_fork_child_flag(current);
710}
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700713static long do_set_mempolicy(unsigned short mode, unsigned short flags,
714 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Miao Xie58568d22009-06-16 15:31:49 -0700716 struct mempolicy *new, *old;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700717 struct mm_struct *mm = current->mm;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700718 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700719 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700721 if (!scratch)
722 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700723
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700724 new = mpol_new(mode, flags, nodes);
725 if (IS_ERR(new)) {
726 ret = PTR_ERR(new);
727 goto out;
728 }
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700729 /*
730 * prevent changing our mempolicy while show_numa_maps()
731 * is using it.
732 * Note: do_set_mempolicy() can be called at init time
733 * with no 'mm'.
734 */
735 if (mm)
736 down_write(&mm->mmap_sem);
Miao Xie58568d22009-06-16 15:31:49 -0700737 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700738 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700739 if (ret) {
740 task_unlock(current);
741 if (mm)
742 up_write(&mm->mmap_sem);
743 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700744 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700745 }
746 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 current->mempolicy = new;
Paul Jacksonc61afb12006-03-24 03:16:08 -0800748 mpol_set_task_struct_flag();
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700749 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700750 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700751 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700752 task_unlock(current);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700753 if (mm)
754 up_write(&mm->mmap_sem);
755
Miao Xie58568d22009-06-16 15:31:49 -0700756 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700757 ret = 0;
758out:
759 NODEMASK_SCRATCH_FREE(scratch);
760 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700763/*
764 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700765 *
766 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700767 */
768static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700770 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700771 if (p == &default_policy)
772 return;
773
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700774 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700775 case MPOL_BIND:
776 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700778 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 break;
780 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700781 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700782 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700783 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 break;
785 default:
786 BUG();
787 }
788}
789
790static int lookup_node(struct mm_struct *mm, unsigned long addr)
791{
792 struct page *p;
793 int err;
794
795 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
796 if (err >= 0) {
797 err = page_to_nid(p);
798 put_page(p);
799 }
800 return err;
801}
802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700804static long do_get_mempolicy(int *policy, nodemask_t *nmask,
805 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700807 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct mm_struct *mm = current->mm;
809 struct vm_area_struct *vma = NULL;
810 struct mempolicy *pol = current->mempolicy;
811
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700812 if (flags &
813 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700815
816 if (flags & MPOL_F_MEMS_ALLOWED) {
817 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
818 return -EINVAL;
819 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700820 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700821 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700822 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700823 return 0;
824 }
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700827 /*
828 * Do NOT fall back to task policy if the
829 * vma/shared policy at addr is NULL. We
830 * want to return MPOL_DEFAULT in this case.
831 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 down_read(&mm->mmap_sem);
833 vma = find_vma_intersection(mm, addr, addr+1);
834 if (!vma) {
835 up_read(&mm->mmap_sem);
836 return -EFAULT;
837 }
838 if (vma->vm_ops && vma->vm_ops->get_policy)
839 pol = vma->vm_ops->get_policy(vma, addr);
840 else
841 pol = vma->vm_policy;
842 } else if (addr)
843 return -EINVAL;
844
845 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700846 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 if (flags & MPOL_F_NODE) {
849 if (flags & MPOL_F_ADDR) {
850 err = lookup_node(mm, addr);
851 if (err < 0)
852 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700853 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700855 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700856 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 } else {
858 err = -EINVAL;
859 goto out;
860 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700861 } else {
862 *policy = pol == &default_policy ? MPOL_DEFAULT :
863 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700864 /*
865 * Internal mempolicy flags must be masked off before exposing
866 * the policy to userspace.
867 */
868 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 if (vma) {
872 up_read(&current->mm->mmap_sem);
873 vma = NULL;
874 }
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700877 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700878 if (mpol_store_user_nodemask(pol)) {
879 *nmask = pol->w.user_nodemask;
880 } else {
881 task_lock(current);
882 get_policy_nodemask(pol, nmask);
883 task_unlock(current);
884 }
Miao Xie58568d22009-06-16 15:31:49 -0700885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700888 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if (vma)
890 up_read(&current->mm->mmap_sem);
891 return err;
892}
893
Christoph Lameterb20a3502006-03-22 00:09:12 -0800894#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700895/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800896 * page migration
897 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800898static void migrate_page_add(struct page *page, struct list_head *pagelist,
899 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800900{
901 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800902 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800903 */
Nick Piggin62695a82008-10-18 20:26:09 -0700904 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
905 if (!isolate_lru_page(page)) {
906 list_add_tail(&page->lru, pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800907 inc_zone_page_state(page, NR_ISOLATED_ANON +
908 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700909 }
910 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800911}
912
Christoph Lameter742755a2006-06-23 02:03:55 -0700913static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700914{
Mel Gorman6484eb32009-06-16 15:31:54 -0700915 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700916}
917
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800918/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800919 * Migrate pages from one node to a target node.
920 * Returns error or the number of pages not migrated.
921 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700922static int migrate_to_node(struct mm_struct *mm, int source, int dest,
923 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800924{
925 nodemask_t nmask;
926 LIST_HEAD(pagelist);
927 int err = 0;
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700928 struct vm_area_struct *vma;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800929
930 nodes_clear(nmask);
931 node_set(source, nmask);
932
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700933 vma = check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800934 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700935 if (IS_ERR(vma))
936 return PTR_ERR(vma);
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800937
Minchan Kimcf608ac2010-10-26 14:21:29 -0700938 if (!list_empty(&pagelist)) {
Mel Gorman7f0f2492011-01-13 15:45:58 -0800939 err = migrate_pages(&pagelist, new_node_page, dest,
940 false, true);
Minchan Kimcf608ac2010-10-26 14:21:29 -0700941 if (err)
942 putback_lru_pages(&pagelist);
943 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700944
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800945 return err;
946}
947
948/*
949 * Move pages between the two nodesets so as to preserve the physical
950 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -0800951 *
952 * Returns the number of page that could not be moved.
953 */
954int do_migrate_pages(struct mm_struct *mm,
955 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
956{
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800957 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800958 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800959 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -0800960
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800961 err = migrate_prep();
962 if (err)
963 return err;
964
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700965 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -0800966
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700967 err = migrate_vmas(mm, from_nodes, to_nodes, flags);
968 if (err)
969 goto out;
970
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -0800971 /*
972 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
973 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
974 * bit in 'tmp', and return that <source, dest> pair for migration.
975 * The pair of nodemasks 'to' and 'from' define the map.
976 *
977 * If no pair of bits is found that way, fallback to picking some
978 * pair of 'source' and 'dest' bits that are not the same. If the
979 * 'source' and 'dest' bits are the same, this represents a node
980 * that will be migrating to itself, so no pages need move.
981 *
982 * If no bits are left in 'tmp', or if all remaining bits left
983 * in 'tmp' correspond to the same bit in 'to', return false
984 * (nothing left to migrate).
985 *
986 * This lets us pick a pair of nodes to migrate between, such that
987 * if possible the dest node is not already occupied by some other
988 * source node, minimizing the risk of overloading the memory on a
989 * node that would happen if we migrated incoming memory to a node
990 * before migrating outgoing memory source that same node.
991 *
992 * A single scan of tmp is sufficient. As we go, we remember the
993 * most recent <s, d> pair that moved (s != d). If we find a pair
994 * that not only moved, but what's better, moved to an empty slot
995 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +0100996 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -0800997 * most recent <s, d> pair that moved. If we get all the way through
998 * the scan of tmp without finding any node that moved, much less
999 * moved to an empty node, then there is nothing left worth migrating.
1000 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001001
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001002 tmp = *from_nodes;
1003 while (!nodes_empty(tmp)) {
1004 int s,d;
1005 int source = -1;
1006 int dest = 0;
1007
1008 for_each_node_mask(s, tmp) {
1009 d = node_remap(s, *from_nodes, *to_nodes);
1010 if (s == d)
1011 continue;
1012
1013 source = s; /* Node moved. Memorize */
1014 dest = d;
1015
1016 /* dest not in remaining from nodes? */
1017 if (!node_isset(dest, tmp))
1018 break;
1019 }
1020 if (source == -1)
1021 break;
1022
1023 node_clear(source, tmp);
1024 err = migrate_to_node(mm, source, dest, flags);
1025 if (err > 0)
1026 busy += err;
1027 if (err < 0)
1028 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001029 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001030out:
Christoph Lameter39743882006-01-08 01:00:51 -08001031 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001032 if (err < 0)
1033 return err;
1034 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001035
Christoph Lameter39743882006-01-08 01:00:51 -08001036}
1037
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001038/*
1039 * Allocate a new page for page migration based on vma policy.
1040 * Start assuming that page is mapped by vma pointed to by @private.
1041 * Search forward from there, if not. N.B., this assumes that the
1042 * list of pages handed to migrate_pages()--which is how we get here--
1043 * is in virtual address order.
1044 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001045static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001046{
1047 struct vm_area_struct *vma = (struct vm_area_struct *)private;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001048 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001049
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001050 while (vma) {
1051 address = page_address_in_vma(page, vma);
1052 if (address != -EFAULT)
1053 break;
1054 vma = vma->vm_next;
1055 }
1056
1057 /*
1058 * if !vma, alloc_page_vma() will use task or system default policy
1059 */
1060 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001061}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001062#else
1063
1064static void migrate_page_add(struct page *page, struct list_head *pagelist,
1065 unsigned long flags)
1066{
1067}
1068
1069int do_migrate_pages(struct mm_struct *mm,
1070 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
1071{
1072 return -ENOSYS;
1073}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001074
Keith Owens69939742006-10-11 01:21:28 -07001075static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001076{
1077 return NULL;
1078}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001079#endif
1080
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001081static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001082 unsigned short mode, unsigned short mode_flags,
1083 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001084{
1085 struct vm_area_struct *vma;
1086 struct mm_struct *mm = current->mm;
1087 struct mempolicy *new;
1088 unsigned long end;
1089 int err;
1090 LIST_HEAD(pagelist);
1091
David Rientjesa3b51e02008-04-28 02:12:23 -07001092 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
1093 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001094 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001095 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001096 return -EPERM;
1097
1098 if (start & ~PAGE_MASK)
1099 return -EINVAL;
1100
1101 if (mode == MPOL_DEFAULT)
1102 flags &= ~MPOL_MF_STRICT;
1103
1104 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1105 end = start + len;
1106
1107 if (end < start)
1108 return -EINVAL;
1109 if (end == start)
1110 return 0;
1111
David Rientjes028fec42008-04-28 02:12:25 -07001112 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001113 if (IS_ERR(new))
1114 return PTR_ERR(new);
1115
1116 /*
1117 * If we are using the default policy then operation
1118 * on discontinuous address spaces is okay after all
1119 */
1120 if (!new)
1121 flags |= MPOL_MF_DISCONTIG_OK;
1122
David Rientjes028fec42008-04-28 02:12:25 -07001123 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1124 start, start + len, mode, mode_flags,
1125 nmask ? nodes_addr(*nmask)[0] : -1);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001126
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001127 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1128
1129 err = migrate_prep();
1130 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001131 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001132 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001133 {
1134 NODEMASK_SCRATCH(scratch);
1135 if (scratch) {
1136 down_write(&mm->mmap_sem);
1137 task_lock(current);
1138 err = mpol_set_nodemask(new, nmask, scratch);
1139 task_unlock(current);
1140 if (err)
1141 up_write(&mm->mmap_sem);
1142 } else
1143 err = -ENOMEM;
1144 NODEMASK_SCRATCH_FREE(scratch);
1145 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001146 if (err)
1147 goto mpol_out;
1148
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001149 vma = check_range(mm, start, end, nmask,
1150 flags | MPOL_MF_INVERT, &pagelist);
1151
1152 err = PTR_ERR(vma);
1153 if (!IS_ERR(vma)) {
1154 int nr_failed = 0;
1155
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001156 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001157
Minchan Kimcf608ac2010-10-26 14:21:29 -07001158 if (!list_empty(&pagelist)) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001159 nr_failed = migrate_pages(&pagelist, new_vma_page,
Mel Gorman7f0f2492011-01-13 15:45:58 -08001160 (unsigned long)vma,
1161 false, true);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001162 if (nr_failed)
1163 putback_lru_pages(&pagelist);
1164 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001165
1166 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
1167 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001168 } else
1169 putback_lru_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001170
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001171 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001172 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001173 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001174 return err;
1175}
1176
Christoph Lameter39743882006-01-08 01:00:51 -08001177/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001178 * User space interface with variable sized bitmaps for nodelists.
1179 */
1180
1181/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001182static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001183 unsigned long maxnode)
1184{
1185 unsigned long k;
1186 unsigned long nlongs;
1187 unsigned long endmask;
1188
1189 --maxnode;
1190 nodes_clear(*nodes);
1191 if (maxnode == 0 || !nmask)
1192 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001193 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001194 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001195
1196 nlongs = BITS_TO_LONGS(maxnode);
1197 if ((maxnode % BITS_PER_LONG) == 0)
1198 endmask = ~0UL;
1199 else
1200 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1201
1202 /* When the user specified more nodes than supported just check
1203 if the non supported part is all zero. */
1204 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1205 if (nlongs > PAGE_SIZE/sizeof(long))
1206 return -EINVAL;
1207 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1208 unsigned long t;
1209 if (get_user(t, nmask + k))
1210 return -EFAULT;
1211 if (k == nlongs - 1) {
1212 if (t & endmask)
1213 return -EINVAL;
1214 } else if (t)
1215 return -EINVAL;
1216 }
1217 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1218 endmask = ~0UL;
1219 }
1220
1221 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1222 return -EFAULT;
1223 nodes_addr(*nodes)[nlongs-1] &= endmask;
1224 return 0;
1225}
1226
1227/* Copy a kernel node mask to user space */
1228static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1229 nodemask_t *nodes)
1230{
1231 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1232 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1233
1234 if (copy > nbytes) {
1235 if (copy > PAGE_SIZE)
1236 return -EINVAL;
1237 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1238 return -EFAULT;
1239 copy = nbytes;
1240 }
1241 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1242}
1243
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001244SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1245 unsigned long, mode, unsigned long __user *, nmask,
1246 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001247{
1248 nodemask_t nodes;
1249 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001250 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001251
David Rientjes028fec42008-04-28 02:12:25 -07001252 mode_flags = mode & MPOL_MODE_FLAGS;
1253 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001254 if (mode >= MPOL_MAX)
1255 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001256 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1257 (mode_flags & MPOL_F_RELATIVE_NODES))
1258 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001259 err = get_nodes(&nodes, nmask, maxnode);
1260 if (err)
1261 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001262 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001263}
1264
1265/* Set the process memory policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001266SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
1267 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001268{
1269 int err;
1270 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001271 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001272
David Rientjes028fec42008-04-28 02:12:25 -07001273 flags = mode & MPOL_MODE_FLAGS;
1274 mode &= ~MPOL_MODE_FLAGS;
1275 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001276 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001277 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1278 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001279 err = get_nodes(&nodes, nmask, maxnode);
1280 if (err)
1281 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001282 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001283}
1284
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001285SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1286 const unsigned long __user *, old_nodes,
1287 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001288{
David Howellsc69e8d92008-11-14 10:39:19 +11001289 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001290 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001291 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001292 nodemask_t task_nodes;
1293 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001294 nodemask_t *old;
1295 nodemask_t *new;
1296 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001297
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001298 if (!scratch)
1299 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001300
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001301 old = &scratch->mask1;
1302 new = &scratch->mask2;
1303
1304 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001305 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001306 goto out;
1307
1308 err = get_nodes(new, new_nodes, maxnode);
1309 if (err)
1310 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001311
1312 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001313 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001314 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001315 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001316 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001317 err = -ESRCH;
1318 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001319 }
1320 mm = get_task_mm(task);
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001321 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001322
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001323 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001324 if (!mm)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001325 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001326
1327 /*
1328 * Check if this process has the right to modify the specified
1329 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001330 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001331 * userid as the target process.
1332 */
David Howellsc69e8d92008-11-14 10:39:19 +11001333 rcu_read_lock();
1334 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001335 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1336 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001337 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001338 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001339 err = -EPERM;
1340 goto out;
1341 }
David Howellsc69e8d92008-11-14 10:39:19 +11001342 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001343
1344 task_nodes = cpuset_mems_allowed(task);
1345 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001346 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001347 err = -EPERM;
1348 goto out;
1349 }
1350
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001351 if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001352 err = -EINVAL;
1353 goto out;
1354 }
1355
David Quigley86c3a762006-06-23 02:04:02 -07001356 err = security_task_movememory(task);
1357 if (err)
1358 goto out;
1359
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001360 err = do_migrate_pages(mm, old, new,
Christoph Lameter74c00242006-03-14 19:50:21 -08001361 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter39743882006-01-08 01:00:51 -08001362out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001363 if (mm)
1364 mmput(mm);
1365 NODEMASK_SCRATCH_FREE(scratch);
1366
Christoph Lameter39743882006-01-08 01:00:51 -08001367 return err;
1368}
1369
1370
Christoph Lameter8bccd852005-10-29 18:16:59 -07001371/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001372SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1373 unsigned long __user *, nmask, unsigned long, maxnode,
1374 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001375{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001376 int err;
1377 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001378 nodemask_t nodes;
1379
1380 if (nmask != NULL && maxnode < MAX_NUMNODES)
1381 return -EINVAL;
1382
1383 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1384
1385 if (err)
1386 return err;
1387
1388 if (policy && put_user(pval, policy))
1389 return -EFAULT;
1390
1391 if (nmask)
1392 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1393
1394 return err;
1395}
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397#ifdef CONFIG_COMPAT
1398
1399asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1400 compat_ulong_t __user *nmask,
1401 compat_ulong_t maxnode,
1402 compat_ulong_t addr, compat_ulong_t flags)
1403{
1404 long err;
1405 unsigned long __user *nm = NULL;
1406 unsigned long nr_bits, alloc_size;
1407 DECLARE_BITMAP(bm, MAX_NUMNODES);
1408
1409 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1410 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1411
1412 if (nmask)
1413 nm = compat_alloc_user_space(alloc_size);
1414
1415 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1416
1417 if (!err && nmask) {
1418 err = copy_from_user(bm, nm, alloc_size);
1419 /* ensure entire bitmap is zeroed */
1420 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1421 err |= compat_put_bitmap(nmask, bm, nr_bits);
1422 }
1423
1424 return err;
1425}
1426
1427asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1428 compat_ulong_t maxnode)
1429{
1430 long err = 0;
1431 unsigned long __user *nm = NULL;
1432 unsigned long nr_bits, alloc_size;
1433 DECLARE_BITMAP(bm, MAX_NUMNODES);
1434
1435 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1436 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1437
1438 if (nmask) {
1439 err = compat_get_bitmap(bm, nmask, nr_bits);
1440 nm = compat_alloc_user_space(alloc_size);
1441 err |= copy_to_user(nm, bm, alloc_size);
1442 }
1443
1444 if (err)
1445 return -EFAULT;
1446
1447 return sys_set_mempolicy(mode, nm, nr_bits+1);
1448}
1449
1450asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1451 compat_ulong_t mode, compat_ulong_t __user *nmask,
1452 compat_ulong_t maxnode, compat_ulong_t flags)
1453{
1454 long err = 0;
1455 unsigned long __user *nm = NULL;
1456 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001457 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1460 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1461
1462 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001463 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001465 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 }
1467
1468 if (err)
1469 return -EFAULT;
1470
1471 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1472}
1473
1474#endif
1475
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001476/*
1477 * get_vma_policy(@task, @vma, @addr)
1478 * @task - task for fallback if vma policy == default
1479 * @vma - virtual memory area whose policy is sought
1480 * @addr - address in @vma for shared policy lookup
1481 *
1482 * Returns effective policy for a VMA at specified address.
1483 * Falls back to @task or system default policy, as necessary.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001484 * Current or other task's task mempolicy and non-shared vma policies
1485 * are protected by the task's mmap_sem, which must be held for read by
1486 * the caller.
1487 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1488 * count--added by the get_policy() vm_op, as appropriate--to protect against
1489 * freeing by another task. It is the caller's responsibility to free the
1490 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001491 */
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001492static struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001493 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001495 struct mempolicy *pol = task->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001498 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001499 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1500 addr);
1501 if (vpol)
1502 pol = vpol;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001503 } else if (vma->vm_policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 pol = vma->vm_policy;
1505 }
1506 if (!pol)
1507 pol = &default_policy;
1508 return pol;
1509}
1510
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001511/*
1512 * Return a nodemask representing a mempolicy for filtering nodes for
1513 * page allocation
1514 */
1515static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001516{
1517 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001518 if (unlikely(policy->mode == MPOL_BIND) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001519 gfp_zone(gfp) >= policy_zone &&
1520 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1521 return &policy->v.nodes;
1522
1523 return NULL;
1524}
1525
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001526/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001527static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1528 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001530 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001532 if (!(policy->flags & MPOL_F_LOCAL))
1533 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 break;
1535 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001536 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001537 * Normally, MPOL_BIND allocations are node-local within the
1538 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001539 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001540 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001541 */
Mel Gorman19770b32008-04-28 02:12:18 -07001542 if (unlikely(gfp & __GFP_THISNODE) &&
1543 unlikely(!node_isset(nd, policy->v.nodes)))
1544 nd = first_node(policy->v.nodes);
1545 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 BUG();
1548 }
Mel Gorman0e884602008-04-28 02:12:14 -07001549 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550}
1551
1552/* Do dynamic interleaving for a process */
1553static unsigned interleave_nodes(struct mempolicy *policy)
1554{
1555 unsigned nid, next;
1556 struct task_struct *me = current;
1557
1558 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001559 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001561 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001562 if (next < MAX_NUMNODES)
1563 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 return nid;
1565}
1566
Christoph Lameterdc85da12006-01-18 17:42:36 -08001567/*
1568 * Depending on the memory policy provide a node from which to allocate the
1569 * next slab entry.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001570 * @policy must be protected by freeing by the caller. If @policy is
1571 * the current task's mempolicy, this protection is implicit, as only the
1572 * task can change it's policy. The system default policy requires no
1573 * such protection.
Christoph Lameterdc85da12006-01-18 17:42:36 -08001574 */
1575unsigned slab_node(struct mempolicy *policy)
1576{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001577 if (!policy || policy->flags & MPOL_F_LOCAL)
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001578 return numa_node_id();
Christoph Lameter765c4502006-09-27 01:50:08 -07001579
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001580 switch (policy->mode) {
1581 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001582 /*
1583 * handled MPOL_F_LOCAL above
1584 */
1585 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001586
Christoph Lameterdc85da12006-01-18 17:42:36 -08001587 case MPOL_INTERLEAVE:
1588 return interleave_nodes(policy);
1589
Mel Gormandd1a2392008-04-28 02:12:17 -07001590 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001591 /*
1592 * Follow bind policy behavior and start allocation at the
1593 * first node.
1594 */
Mel Gorman19770b32008-04-28 02:12:18 -07001595 struct zonelist *zonelist;
1596 struct zone *zone;
1597 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1598 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1599 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1600 &policy->v.nodes,
1601 &zone);
Eric Dumazet800416f2010-10-27 19:33:43 +02001602 return zone ? zone->node : numa_node_id();
Mel Gormandd1a2392008-04-28 02:12:17 -07001603 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001604
Christoph Lameterdc85da12006-01-18 17:42:36 -08001605 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001606 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001607 }
1608}
1609
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610/* Do static interleaving for a VMA with known offset. */
1611static unsigned offset_il_node(struct mempolicy *pol,
1612 struct vm_area_struct *vma, unsigned long off)
1613{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001614 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001615 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 int c;
1617 int nid = -1;
1618
David Rientjesf5b087b2008-04-28 02:12:27 -07001619 if (!nnodes)
1620 return numa_node_id();
1621 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 c = 0;
1623 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001624 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 c++;
1626 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 return nid;
1628}
1629
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001630/* Determine a node number for interleave */
1631static inline unsigned interleave_nid(struct mempolicy *pol,
1632 struct vm_area_struct *vma, unsigned long addr, int shift)
1633{
1634 if (vma) {
1635 unsigned long off;
1636
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001637 /*
1638 * for small pages, there is no difference between
1639 * shift and PAGE_SHIFT, so the bit-shift is safe.
1640 * for huge pages, since vm_pgoff is in units of small
1641 * pages, we need to shift off the always 0 bits to get
1642 * a useful offset.
1643 */
1644 BUG_ON(shift < PAGE_SHIFT);
1645 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001646 off += (addr - vma->vm_start) >> shift;
1647 return offset_il_node(pol, vma, off);
1648 } else
1649 return interleave_nodes(pol);
1650}
1651
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001652#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001653/*
1654 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1655 * @vma = virtual memory area whose policy is sought
1656 * @addr = address in @vma for shared policy lookup and interleave policy
1657 * @gfp_flags = for requested zone
Mel Gorman19770b32008-04-28 02:12:18 -07001658 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1659 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001660 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001661 * Returns a zonelist suitable for a huge page allocation and a pointer
1662 * to the struct mempolicy for conditional unref after allocation.
1663 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1664 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001665 *
1666 * Must be protected by get_mems_allowed()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001667 */
Mel Gorman396faf02007-07-17 04:03:13 -07001668struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001669 gfp_t gfp_flags, struct mempolicy **mpol,
1670 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001671{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001672 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001673
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001674 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001675 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001676
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001677 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1678 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001679 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001680 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001681 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001682 if ((*mpol)->mode == MPOL_BIND)
1683 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001684 }
1685 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001686}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001687
1688/*
1689 * init_nodemask_of_mempolicy
1690 *
1691 * If the current task's mempolicy is "default" [NULL], return 'false'
1692 * to indicate default policy. Otherwise, extract the policy nodemask
1693 * for 'bind' or 'interleave' policy into the argument nodemask, or
1694 * initialize the argument nodemask to contain the single node for
1695 * 'preferred' or 'local' policy and return 'true' to indicate presence
1696 * of non-default mempolicy.
1697 *
1698 * We don't bother with reference counting the mempolicy [mpol_get/put]
1699 * because the current task is examining it's own mempolicy and a task's
1700 * mempolicy is only ever changed by the task itself.
1701 *
1702 * N.B., it is the caller's responsibility to free a returned nodemask.
1703 */
1704bool init_nodemask_of_mempolicy(nodemask_t *mask)
1705{
1706 struct mempolicy *mempolicy;
1707 int nid;
1708
1709 if (!(mask && current->mempolicy))
1710 return false;
1711
Miao Xiec0ff7452010-05-24 14:32:08 -07001712 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001713 mempolicy = current->mempolicy;
1714 switch (mempolicy->mode) {
1715 case MPOL_PREFERRED:
1716 if (mempolicy->flags & MPOL_F_LOCAL)
1717 nid = numa_node_id();
1718 else
1719 nid = mempolicy->v.preferred_node;
1720 init_nodemask_of_node(mask, nid);
1721 break;
1722
1723 case MPOL_BIND:
1724 /* Fall through */
1725 case MPOL_INTERLEAVE:
1726 *mask = mempolicy->v.nodes;
1727 break;
1728
1729 default:
1730 BUG();
1731 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001732 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001733
1734 return true;
1735}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001736#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001737
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001738/*
1739 * mempolicy_nodemask_intersects
1740 *
1741 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1742 * policy. Otherwise, check for intersection between mask and the policy
1743 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1744 * policy, always return true since it may allocate elsewhere on fallback.
1745 *
1746 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1747 */
1748bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1749 const nodemask_t *mask)
1750{
1751 struct mempolicy *mempolicy;
1752 bool ret = true;
1753
1754 if (!mask)
1755 return ret;
1756 task_lock(tsk);
1757 mempolicy = tsk->mempolicy;
1758 if (!mempolicy)
1759 goto out;
1760
1761 switch (mempolicy->mode) {
1762 case MPOL_PREFERRED:
1763 /*
1764 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1765 * allocate from, they may fallback to other nodes when oom.
1766 * Thus, it's possible for tsk to have allocated memory from
1767 * nodes in mask.
1768 */
1769 break;
1770 case MPOL_BIND:
1771 case MPOL_INTERLEAVE:
1772 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1773 break;
1774 default:
1775 BUG();
1776 }
1777out:
1778 task_unlock(tsk);
1779 return ret;
1780}
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782/* Allocate a page in interleaved policy.
1783 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001784static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1785 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786{
1787 struct zonelist *zl;
1788 struct page *page;
1789
Mel Gorman0e884602008-04-28 02:12:14 -07001790 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001792 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001793 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 return page;
1795}
1796
1797/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001798 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 *
1800 * @gfp:
1801 * %GFP_USER user allocation.
1802 * %GFP_KERNEL kernel allocations,
1803 * %GFP_HIGHMEM highmem/user allocations,
1804 * %GFP_FS allocation should not call back into a file system.
1805 * %GFP_ATOMIC don't sleep.
1806 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001807 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 * @vma: Pointer to VMA or NULL if not available.
1809 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1810 *
1811 * This function allocates a page from the kernel page pool and applies
1812 * a NUMA policy associated with the VMA or the current process.
1813 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1814 * mm_struct of the VMA to prevent it from going away. Should be used for
1815 * all allocations for pages that will be mapped into
1816 * user space. Returns NULL when no page can be allocated.
1817 *
1818 * Should be called with the mm_sem of the vma hold.
1819 */
1820struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001821alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Andi Kleen2f5f9482011-03-04 17:36:29 -08001822 unsigned long addr, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001824 struct mempolicy *pol = get_vma_policy(current, vma, addr);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001825 struct zonelist *zl;
Miao Xiec0ff7452010-05-24 14:32:08 -07001826 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Miao Xiec0ff7452010-05-24 14:32:08 -07001828 get_mems_allowed();
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001829 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001831
Andi Kleen8eac5632011-02-25 14:44:28 -08001832 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001833 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001834 page = alloc_page_interleave(gfp, order, nid);
Miao Xiec0ff7452010-05-24 14:32:08 -07001835 put_mems_allowed();
1836 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 }
Andi Kleen2f5f9482011-03-04 17:36:29 -08001838 zl = policy_zonelist(gfp, pol, node);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001839 if (unlikely(mpol_needs_cond_ref(pol))) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001840 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001841 * slow path: ref counted shared policy
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001842 */
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001843 struct page *page = __alloc_pages_nodemask(gfp, order,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001844 zl, policy_nodemask(gfp, pol));
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001845 __mpol_put(pol);
Miao Xiec0ff7452010-05-24 14:32:08 -07001846 put_mems_allowed();
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001847 return page;
1848 }
1849 /*
1850 * fast path: default or task policy
1851 */
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001852 page = __alloc_pages_nodemask(gfp, order, zl,
1853 policy_nodemask(gfp, pol));
Miao Xiec0ff7452010-05-24 14:32:08 -07001854 put_mems_allowed();
1855 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
1858/**
1859 * alloc_pages_current - Allocate pages.
1860 *
1861 * @gfp:
1862 * %GFP_USER user allocation,
1863 * %GFP_KERNEL kernel allocation,
1864 * %GFP_HIGHMEM highmem allocation,
1865 * %GFP_FS don't call back into a file system.
1866 * %GFP_ATOMIC don't sleep.
1867 * @order: Power of two of allocation size in pages. 0 is a single page.
1868 *
1869 * Allocate a page from the kernel page pool. When not in
1870 * interrupt context and apply the current process NUMA policy.
1871 * Returns NULL when no page can be allocated.
1872 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001873 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 * 1) it's ok to take cpuset_sem (can WAIT), and
1875 * 2) allocating for current task (not interrupt).
1876 */
Al Virodd0fc662005-10-07 07:46:04 +01001877struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
1879 struct mempolicy *pol = current->mempolicy;
Miao Xiec0ff7452010-05-24 14:32:08 -07001880 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
Christoph Lameter9b819d22006-09-25 23:31:40 -07001882 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001884
Miao Xiec0ff7452010-05-24 14:32:08 -07001885 get_mems_allowed();
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001886 /*
1887 * No reference counting needed for current->mempolicy
1888 * nor system default_policy
1889 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001890 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07001891 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
1892 else
1893 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08001894 policy_zonelist(gfp, pol, numa_node_id()),
1895 policy_nodemask(gfp, pol));
Miao Xiec0ff7452010-05-24 14:32:08 -07001896 put_mems_allowed();
1897 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899EXPORT_SYMBOL(alloc_pages_current);
1900
Paul Jackson42253992006-01-08 01:01:59 -08001901/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001902 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08001903 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1904 * with the mems_allowed returned by cpuset_mems_allowed(). This
1905 * keeps mempolicies cpuset relative after its cpuset moves. See
1906 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07001907 *
1908 * current's mempolicy may be rebinded by the other task(the task that changes
1909 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08001910 */
Paul Jackson42253992006-01-08 01:01:59 -08001911
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001912/* Slow path of a mempolicy duplicate */
1913struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914{
1915 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1916
1917 if (!new)
1918 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07001919
1920 /* task's mempolicy is protected by alloc_lock */
1921 if (old == current->mempolicy) {
1922 task_lock(current);
1923 *new = *old;
1924 task_unlock(current);
1925 } else
1926 *new = *old;
1927
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001928 rcu_read_lock();
Paul Jackson42253992006-01-08 01:01:59 -08001929 if (current_cpuset_is_being_rebound()) {
1930 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07001931 if (new->flags & MPOL_F_REBINDING)
1932 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
1933 else
1934 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08001935 }
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001936 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return new;
1939}
1940
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001941/*
1942 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1943 * eliminate the * MPOL_F_* flags that require conditional ref and
1944 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1945 * after return. Use the returned value.
1946 *
1947 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1948 * policy lookup, even if the policy needs/has extra ref on lookup.
1949 * shmem_readahead needs this.
1950 */
1951struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
1952 struct mempolicy *frompol)
1953{
1954 if (!mpol_needs_cond_ref(frompol))
1955 return frompol;
1956
1957 *tompol = *frompol;
1958 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
1959 __mpol_put(frompol);
1960 return tompol;
1961}
1962
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963/* Slow path of a mempolicy comparison */
1964int __mpol_equal(struct mempolicy *a, struct mempolicy *b)
1965{
1966 if (!a || !b)
1967 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001968 if (a->mode != b->mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return 0;
Bob Liu19800502010-05-24 14:32:01 -07001970 if (a->flags != b->flags)
David Rientjesf5b087b2008-04-28 02:12:27 -07001971 return 0;
Bob Liu19800502010-05-24 14:32:01 -07001972 if (mpol_store_user_nodemask(a))
1973 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
1974 return 0;
1975
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001976 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07001977 case MPOL_BIND:
1978 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -07001980 return nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07001982 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 default:
1984 BUG();
1985 return 0;
1986 }
1987}
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 * Shared memory backing store policy support.
1991 *
1992 * Remember policies even when nobody has shared memory mapped.
1993 * The policies are kept in Red-Black tree linked from the inode.
1994 * They are protected by the sp->lock spinlock, which should be held
1995 * for any accesses to the tree.
1996 */
1997
1998/* lookup first element intersecting start-end */
1999/* Caller holds sp->lock */
2000static struct sp_node *
2001sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2002{
2003 struct rb_node *n = sp->root.rb_node;
2004
2005 while (n) {
2006 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2007
2008 if (start >= p->end)
2009 n = n->rb_right;
2010 else if (end <= p->start)
2011 n = n->rb_left;
2012 else
2013 break;
2014 }
2015 if (!n)
2016 return NULL;
2017 for (;;) {
2018 struct sp_node *w = NULL;
2019 struct rb_node *prev = rb_prev(n);
2020 if (!prev)
2021 break;
2022 w = rb_entry(prev, struct sp_node, nd);
2023 if (w->end <= start)
2024 break;
2025 n = prev;
2026 }
2027 return rb_entry(n, struct sp_node, nd);
2028}
2029
2030/* Insert a new shared policy into the list. */
2031/* Caller holds sp->lock */
2032static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2033{
2034 struct rb_node **p = &sp->root.rb_node;
2035 struct rb_node *parent = NULL;
2036 struct sp_node *nd;
2037
2038 while (*p) {
2039 parent = *p;
2040 nd = rb_entry(parent, struct sp_node, nd);
2041 if (new->start < nd->start)
2042 p = &(*p)->rb_left;
2043 else if (new->end > nd->end)
2044 p = &(*p)->rb_right;
2045 else
2046 BUG();
2047 }
2048 rb_link_node(&new->nd, parent, p);
2049 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002050 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002051 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
2053
2054/* Find shared policy intersecting idx */
2055struct mempolicy *
2056mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2057{
2058 struct mempolicy *pol = NULL;
2059 struct sp_node *sn;
2060
2061 if (!sp->root.rb_node)
2062 return NULL;
2063 spin_lock(&sp->lock);
2064 sn = sp_lookup(sp, idx, idx+1);
2065 if (sn) {
2066 mpol_get(sn->policy);
2067 pol = sn->policy;
2068 }
2069 spin_unlock(&sp->lock);
2070 return pol;
2071}
2072
2073static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2074{
Paul Mundt140d5a42007-07-15 23:38:16 -07002075 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 rb_erase(&n->nd, &sp->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002077 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 kmem_cache_free(sn_cache, n);
2079}
2080
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002081static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2082 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
2084 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2085
2086 if (!n)
2087 return NULL;
2088 n->start = start;
2089 n->end = end;
2090 mpol_get(pol);
Lee Schermerhornaab0b102008-04-28 02:13:13 -07002091 pol->flags |= MPOL_F_SHARED; /* for unref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 n->policy = pol;
2093 return n;
2094}
2095
2096/* Replace a policy range. */
2097static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2098 unsigned long end, struct sp_node *new)
2099{
2100 struct sp_node *n, *new2 = NULL;
2101
2102restart:
2103 spin_lock(&sp->lock);
2104 n = sp_lookup(sp, start, end);
2105 /* Take care of old policies in the same range. */
2106 while (n && n->start < end) {
2107 struct rb_node *next = rb_next(&n->nd);
2108 if (n->start >= start) {
2109 if (n->end <= end)
2110 sp_delete(sp, n);
2111 else
2112 n->start = end;
2113 } else {
2114 /* Old policy spanning whole new range. */
2115 if (n->end > end) {
2116 if (!new2) {
2117 spin_unlock(&sp->lock);
2118 new2 = sp_alloc(end, n->end, n->policy);
2119 if (!new2)
2120 return -ENOMEM;
2121 goto restart;
2122 }
2123 n->end = start;
2124 sp_insert(sp, new2);
2125 new2 = NULL;
2126 break;
2127 } else
2128 n->end = start;
2129 }
2130 if (!next)
2131 break;
2132 n = rb_entry(next, struct sp_node, nd);
2133 }
2134 if (new)
2135 sp_insert(sp, new);
2136 spin_unlock(&sp->lock);
2137 if (new2) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002138 mpol_put(new2->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 kmem_cache_free(sn_cache, new2);
2140 }
2141 return 0;
2142}
2143
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002144/**
2145 * mpol_shared_policy_init - initialize shared policy for inode
2146 * @sp: pointer to inode shared policy
2147 * @mpol: struct mempolicy to install
2148 *
2149 * Install non-NULL @mpol in inode's shared policy rb-tree.
2150 * On entry, the current task has a reference on a non-NULL @mpol.
2151 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002152 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002153 */
2154void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002155{
Miao Xie58568d22009-06-16 15:31:49 -07002156 int ret;
2157
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002158 sp->root = RB_ROOT; /* empty tree == default mempolicy */
2159 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002160
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002161 if (mpol) {
2162 struct vm_area_struct pvma;
2163 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002164 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002165
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002166 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002167 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002168 /* contextualize the tmpfs mount point mempolicy */
2169 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002170 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002171 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002172
2173 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002174 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002175 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002176 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002177 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002178
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002179 /* Create pseudo-vma that contains just the policy */
2180 memset(&pvma, 0, sizeof(struct vm_area_struct));
2181 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2182 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002183
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002184put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002185 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002186free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002187 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002188put_mpol:
2189 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002190 }
2191}
2192
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193int mpol_set_shared_policy(struct shared_policy *info,
2194 struct vm_area_struct *vma, struct mempolicy *npol)
2195{
2196 int err;
2197 struct sp_node *new = NULL;
2198 unsigned long sz = vma_pages(vma);
2199
David Rientjes028fec42008-04-28 02:12:25 -07002200 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002202 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002203 npol ? npol->flags : -1,
Paul Mundt140d5a42007-07-15 23:38:16 -07002204 npol ? nodes_addr(npol->v.nodes)[0] : -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 if (npol) {
2207 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2208 if (!new)
2209 return -ENOMEM;
2210 }
2211 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2212 if (err && new)
2213 kmem_cache_free(sn_cache, new);
2214 return err;
2215}
2216
2217/* Free a backing policy store on inode delete. */
2218void mpol_free_shared_policy(struct shared_policy *p)
2219{
2220 struct sp_node *n;
2221 struct rb_node *next;
2222
2223 if (!p->root.rb_node)
2224 return;
2225 spin_lock(&p->lock);
2226 next = rb_first(&p->root);
2227 while (next) {
2228 n = rb_entry(next, struct sp_node, nd);
2229 next = rb_next(&n->nd);
Andi Kleen90c50292005-07-27 11:43:50 -07002230 rb_erase(&n->nd, &p->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002231 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 kmem_cache_free(sn_cache, n);
2233 }
2234 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235}
2236
2237/* assumes fs == KERNEL_DS */
2238void __init numa_policy_init(void)
2239{
Paul Mundtb71636e2007-07-15 23:38:15 -07002240 nodemask_t interleave_nodes;
2241 unsigned long largest = 0;
2242 int nid, prefer = 0;
2243
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 policy_cache = kmem_cache_create("numa_policy",
2245 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002246 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 sn_cache = kmem_cache_create("shared_policy_node",
2249 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002250 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
Paul Mundtb71636e2007-07-15 23:38:15 -07002252 /*
2253 * Set interleaving policy for system init. Interleaving is only
2254 * enabled across suitably sized nodes (default is >= 16MB), or
2255 * fall back to the largest node if they're all smaller.
2256 */
2257 nodes_clear(interleave_nodes);
Christoph Lameter56bbd652007-10-16 01:25:35 -07002258 for_each_node_state(nid, N_HIGH_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002259 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
Paul Mundtb71636e2007-07-15 23:38:15 -07002261 /* Preserve the largest node */
2262 if (largest < total_pages) {
2263 largest = total_pages;
2264 prefer = nid;
2265 }
2266
2267 /* Interleave this node? */
2268 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2269 node_set(nid, interleave_nodes);
2270 }
2271
2272 /* All too small, use the largest */
2273 if (unlikely(nodes_empty(interleave_nodes)))
2274 node_set(prefer, interleave_nodes);
2275
David Rientjes028fec42008-04-28 02:12:25 -07002276 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 printk("numa_policy_init: interleaving failed\n");
2278}
2279
Christoph Lameter8bccd852005-10-29 18:16:59 -07002280/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281void numa_default_policy(void)
2282{
David Rientjes028fec42008-04-28 02:12:25 -07002283 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284}
Paul Jackson68860ec2005-10-30 15:02:36 -08002285
Paul Jackson42253992006-01-08 01:01:59 -08002286/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002287 * Parse and format mempolicy from/to strings
2288 */
2289
2290/*
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002291 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002292 * Used only for mpol_parse_str() and mpol_to_str()
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002293 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002294#define MPOL_LOCAL MPOL_MAX
2295static const char * const policy_modes[] =
2296{
2297 [MPOL_DEFAULT] = "default",
2298 [MPOL_PREFERRED] = "prefer",
2299 [MPOL_BIND] = "bind",
2300 [MPOL_INTERLEAVE] = "interleave",
2301 [MPOL_LOCAL] = "local"
2302};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002303
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002304
2305#ifdef CONFIG_TMPFS
2306/**
2307 * mpol_parse_str - parse string to mempolicy
2308 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002309 * @mpol: pointer to struct mempolicy pointer, returned on success.
2310 * @no_context: flag whether to "contextualize" the mempolicy
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002311 *
2312 * Format of input:
2313 * <mode>[=<flags>][:<nodelist>]
2314 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002315 * if @no_context is true, save the input nodemask in w.user_nodemask in
2316 * the returned mempolicy. This will be used to "clone" the mempolicy in
2317 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2318 * mount option. Note that if 'static' or 'relative' mode flags were
2319 * specified, the input nodemask will already have been saved. Saving
2320 * it again is redundant, but safe.
2321 *
2322 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002323 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002324int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002325{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002326 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002327 unsigned short mode;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002328 unsigned short uninitialized_var(mode_flags);
2329 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002330 char *nodelist = strchr(str, ':');
2331 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002332 int err = 1;
2333
2334 if (nodelist) {
2335 /* NUL-terminate mode or flags string */
2336 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002337 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002338 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002339 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002340 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002341 } else
2342 nodes_clear(nodes);
2343
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002344 if (flags)
2345 *flags++ = '\0'; /* terminate mode string */
2346
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002347 for (mode = 0; mode <= MPOL_LOCAL; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002348 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002349 break;
2350 }
2351 }
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002352 if (mode > MPOL_LOCAL)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002353 goto out;
2354
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002355 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002356 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002357 /*
2358 * Insist on a nodelist of one node only
2359 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002360 if (nodelist) {
2361 char *rest = nodelist;
2362 while (isdigit(*rest))
2363 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002364 if (*rest)
2365 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002366 }
2367 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002368 case MPOL_INTERLEAVE:
2369 /*
2370 * Default to online nodes with memory if no nodelist
2371 */
2372 if (!nodelist)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002373 nodes = node_states[N_HIGH_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002374 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002375 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002376 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002377 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002378 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002379 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002380 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002381 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002382 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002383 case MPOL_DEFAULT:
2384 /*
2385 * Insist on a empty nodelist
2386 */
2387 if (!nodelist)
2388 err = 0;
2389 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002390 case MPOL_BIND:
2391 /*
2392 * Insist on a nodelist
2393 */
2394 if (!nodelist)
2395 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002396 }
2397
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002398 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002399 if (flags) {
2400 /*
2401 * Currently, we only support two mutually exclusive
2402 * mode flags.
2403 */
2404 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002405 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002406 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002407 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002408 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002409 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002410 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002411
2412 new = mpol_new(mode, mode_flags, &nodes);
2413 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002414 goto out;
2415
Lee Schermerhorne17f74a2010-05-24 14:32:02 -07002416 if (no_context) {
2417 /* save for contextualization */
2418 new->w.user_nodemask = nodes;
2419 } else {
Miao Xie58568d22009-06-16 15:31:49 -07002420 int ret;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002421 NODEMASK_SCRATCH(scratch);
2422 if (scratch) {
2423 task_lock(current);
2424 ret = mpol_set_nodemask(new, &nodes, scratch);
2425 task_unlock(current);
2426 } else
2427 ret = -ENOMEM;
2428 NODEMASK_SCRATCH_FREE(scratch);
2429 if (ret) {
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002430 mpol_put(new);
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002431 goto out;
Miao Xie58568d22009-06-16 15:31:49 -07002432 }
2433 }
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002434 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002435
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002436out:
2437 /* Restore string for error message */
2438 if (nodelist)
2439 *--nodelist = ':';
2440 if (flags)
2441 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002442 if (!err)
2443 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002444 return err;
2445}
2446#endif /* CONFIG_TMPFS */
2447
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002448/**
2449 * mpol_to_str - format a mempolicy structure for printing
2450 * @buffer: to contain formatted mempolicy string
2451 * @maxlen: length of @buffer
2452 * @pol: pointer to mempolicy to be formatted
2453 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2454 *
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002455 * Convert a mempolicy into a string.
2456 * Returns the number of characters in buffer (if positive)
2457 * or an error (negative)
2458 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002459int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002460{
2461 char *p = buffer;
2462 int l;
2463 nodemask_t nodes;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002464 unsigned short mode;
David Rientjesf5b087b2008-04-28 02:12:27 -07002465 unsigned short flags = pol ? pol->flags : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002466
Lee Schermerhorn22919902008-04-28 02:13:22 -07002467 /*
2468 * Sanity check: room for longest mode, flag and some nodes
2469 */
2470 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2471
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002472 if (!pol || pol == &default_policy)
2473 mode = MPOL_DEFAULT;
2474 else
2475 mode = pol->mode;
2476
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002477 switch (mode) {
2478 case MPOL_DEFAULT:
2479 nodes_clear(nodes);
2480 break;
2481
2482 case MPOL_PREFERRED:
2483 nodes_clear(nodes);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002484 if (flags & MPOL_F_LOCAL)
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002485 mode = MPOL_LOCAL; /* pseudo-policy */
2486 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002487 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002488 break;
2489
2490 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07002491 /* Fall through */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002492 case MPOL_INTERLEAVE:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002493 if (no_context)
2494 nodes = pol->w.user_nodemask;
2495 else
2496 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002497 break;
2498
2499 default:
2500 BUG();
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002501 }
2502
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002503 l = strlen(policy_modes[mode]);
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002504 if (buffer + maxlen < p + l + 1)
2505 return -ENOSPC;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002506
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002507 strcpy(p, policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002508 p += l;
2509
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002510 if (flags & MPOL_MODE_FLAGS) {
David Rientjesf5b087b2008-04-28 02:12:27 -07002511 if (buffer + maxlen < p + 2)
2512 return -ENOSPC;
2513 *p++ = '=';
2514
Lee Schermerhorn22919902008-04-28 02:13:22 -07002515 /*
2516 * Currently, the only defined flags are mutually exclusive
2517 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002518 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002519 p += snprintf(p, buffer + maxlen - p, "static");
2520 else if (flags & MPOL_F_RELATIVE_NODES)
2521 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002522 }
2523
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002524 if (!nodes_empty(nodes)) {
2525 if (buffer + maxlen < p + 2)
2526 return -ENOSPC;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002527 *p++ = ':';
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002528 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2529 }
2530 return p - buffer;
2531}
2532
2533struct numa_maps {
2534 unsigned long pages;
2535 unsigned long anon;
Christoph Lameter397874d2006-03-06 15:42:53 -08002536 unsigned long active;
2537 unsigned long writeback;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002538 unsigned long mapcount_max;
Christoph Lameter397874d2006-03-06 15:42:53 -08002539 unsigned long dirty;
2540 unsigned long swapcache;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002541 unsigned long node[MAX_NUMNODES];
2542};
2543
Christoph Lameter397874d2006-03-06 15:42:53 -08002544static void gather_stats(struct page *page, void *private, int pte_dirty)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002545{
2546 struct numa_maps *md = private;
2547 int count = page_mapcount(page);
2548
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002549 md->pages++;
Christoph Lameter397874d2006-03-06 15:42:53 -08002550 if (pte_dirty || PageDirty(page))
2551 md->dirty++;
2552
2553 if (PageSwapCache(page))
2554 md->swapcache++;
2555
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002556 if (PageActive(page) || PageUnevictable(page))
Christoph Lameter397874d2006-03-06 15:42:53 -08002557 md->active++;
2558
2559 if (PageWriteback(page))
2560 md->writeback++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002561
2562 if (PageAnon(page))
2563 md->anon++;
2564
Christoph Lameter397874d2006-03-06 15:42:53 -08002565 if (count > md->mapcount_max)
2566 md->mapcount_max = count;
2567
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002568 md->node[page_to_nid(page)]++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002569}
2570
Andrew Morton7f709ed2006-03-07 21:55:22 -08002571#ifdef CONFIG_HUGETLB_PAGE
Christoph Lameter397874d2006-03-06 15:42:53 -08002572static void check_huge_range(struct vm_area_struct *vma,
2573 unsigned long start, unsigned long end,
2574 struct numa_maps *md)
2575{
2576 unsigned long addr;
2577 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -07002578 struct hstate *h = hstate_vma(vma);
2579 unsigned long sz = huge_page_size(h);
Christoph Lameter397874d2006-03-06 15:42:53 -08002580
Andi Kleena5516432008-07-23 21:27:41 -07002581 for (addr = start; addr < end; addr += sz) {
2582 pte_t *ptep = huge_pte_offset(vma->vm_mm,
2583 addr & huge_page_mask(h));
Christoph Lameter397874d2006-03-06 15:42:53 -08002584 pte_t pte;
2585
2586 if (!ptep)
2587 continue;
2588
2589 pte = *ptep;
2590 if (pte_none(pte))
2591 continue;
2592
2593 page = pte_page(pte);
2594 if (!page)
2595 continue;
2596
2597 gather_stats(page, md, pte_dirty(*ptep));
2598 }
2599}
Andrew Morton7f709ed2006-03-07 21:55:22 -08002600#else
2601static inline void check_huge_range(struct vm_area_struct *vma,
2602 unsigned long start, unsigned long end,
2603 struct numa_maps *md)
2604{
2605}
2606#endif
Christoph Lameter397874d2006-03-06 15:42:53 -08002607
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002608/*
2609 * Display pages allocated per node and memory policy via /proc.
2610 */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002611int show_numa_map(struct seq_file *m, void *v)
2612{
Eric W. Biederman99f89552006-06-26 00:25:55 -07002613 struct proc_maps_private *priv = m->private;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002614 struct vm_area_struct *vma = v;
2615 struct numa_maps *md;
Christoph Lameter397874d2006-03-06 15:42:53 -08002616 struct file *file = vma->vm_file;
2617 struct mm_struct *mm = vma->vm_mm;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002618 struct mempolicy *pol;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002619 int n;
2620 char buffer[50];
2621
Christoph Lameter397874d2006-03-06 15:42:53 -08002622 if (!mm)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002623 return 0;
2624
2625 md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL);
2626 if (!md)
2627 return 0;
2628
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002629 pol = get_vma_policy(priv->task, vma, vma->vm_start);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002630 mpol_to_str(buffer, sizeof(buffer), pol, 0);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002631 mpol_cond_put(pol);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002632
Christoph Lameter397874d2006-03-06 15:42:53 -08002633 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002634
Christoph Lameter397874d2006-03-06 15:42:53 -08002635 if (file) {
2636 seq_printf(m, " file=");
Jan Blunckc32c2f62008-02-14 19:38:43 -08002637 seq_path(m, &file->f_path, "\n\t= ");
Christoph Lameter397874d2006-03-06 15:42:53 -08002638 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
2639 seq_printf(m, " heap");
2640 } else if (vma->vm_start <= mm->start_stack &&
2641 vma->vm_end >= mm->start_stack) {
2642 seq_printf(m, " stack");
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002643 }
Christoph Lameter397874d2006-03-06 15:42:53 -08002644
2645 if (is_vm_hugetlb_page(vma)) {
2646 check_huge_range(vma, vma->vm_start, vma->vm_end, md);
2647 seq_printf(m, " huge");
2648 } else {
2649 check_pgd_range(vma, vma->vm_start, vma->vm_end,
Christoph Lameter56bbd652007-10-16 01:25:35 -07002650 &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
Christoph Lameter397874d2006-03-06 15:42:53 -08002651 }
2652
2653 if (!md->pages)
2654 goto out;
2655
2656 if (md->anon)
2657 seq_printf(m," anon=%lu",md->anon);
2658
2659 if (md->dirty)
2660 seq_printf(m," dirty=%lu",md->dirty);
2661
2662 if (md->pages != md->anon && md->pages != md->dirty)
2663 seq_printf(m, " mapped=%lu", md->pages);
2664
2665 if (md->mapcount_max > 1)
2666 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2667
2668 if (md->swapcache)
2669 seq_printf(m," swapcache=%lu", md->swapcache);
2670
2671 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2672 seq_printf(m," active=%lu", md->active);
2673
2674 if (md->writeback)
2675 seq_printf(m," writeback=%lu", md->writeback);
2676
Christoph Lameter56bbd652007-10-16 01:25:35 -07002677 for_each_node_state(n, N_HIGH_MEMORY)
Christoph Lameter397874d2006-03-06 15:42:53 -08002678 if (md->node[n])
2679 seq_printf(m, " N%d=%lu", n, md->node[n]);
2680out:
2681 seq_putc(m, '\n');
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002682 kfree(md);
2683
2684 if (m->count < m->size)
Eric W. Biederman99f89552006-06-26 00:25:55 -07002685 m->version = (vma != priv->tail_vma) ? vma->vm_start : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002686 return 0;
2687}