blob: e08e2c4da63a5dc45bfcfb9f56cd440d557d0772 [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>
76#include <linux/gfp.h>
77#include <linux/slab.h>
78#include <linux/string.h>
79#include <linux/module.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070080#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <linux/interrupt.h>
82#include <linux/init.h>
83#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080084#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080085#include <linux/seq_file.h>
86#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080087#include <linux/migrate.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>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093#include <asm/tlbflush.h>
94#include <asm/uaccess.h>
95
Nick Piggin62695a82008-10-18 20:26:09 -070096#include "internal.h"
97
Christoph Lameter38e35862006-01-08 01:01:01 -080098/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080099#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800100#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800101#define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800102
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800103static struct kmem_cache *policy_cache;
104static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/* Highest zone. An specific allocation for a zone below that is not
107 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800108enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700110/*
111 * run-time system-wide default policy => local allocation
112 */
Andi Kleend42c6992005-07-06 19:56:03 +0200113struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700115 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700116 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117};
118
David Rientjes37012942008-04-28 02:12:33 -0700119static const struct mempolicy_operations {
120 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
121 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes);
122} mpol_ops[MPOL_MAX];
123
Mel Gorman19770b32008-04-28 02:12:18 -0700124/* Check that the nodemask contains at least one populated zone */
David Rientjes37012942008-04-28 02:12:33 -0700125static int is_valid_nodemask(const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Mel Gorman19770b32008-04-28 02:12:18 -0700127 int nd, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Mel Gorman19770b32008-04-28 02:12:18 -0700129 /* Check that there is something useful in this mask */
130 k = policy_zone;
131
132 for_each_node_mask(nd, *nodemask) {
133 struct zone *z;
134
135 for (k = 0; k <= policy_zone; k++) {
136 z = &NODE_DATA(nd)->node_zones[k];
137 if (z->present_pages > 0)
138 return 1;
Andi Kleendd942ae2006-02-17 01:39:16 +0100139 }
140 }
Mel Gorman19770b32008-04-28 02:12:18 -0700141
142 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
David Rientjesf5b087b2008-04-28 02:12:27 -0700145static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
146{
David Rientjes4c50bc02008-04-28 02:12:30 -0700147 return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES);
148}
149
150static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
151 const nodemask_t *rel)
152{
153 nodemask_t tmp;
154 nodes_fold(tmp, *orig, nodes_weight(*rel));
155 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700156}
157
David Rientjes37012942008-04-28 02:12:33 -0700158static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
159{
160 if (nodes_empty(*nodes))
161 return -EINVAL;
162 pol->v.nodes = *nodes;
163 return 0;
164}
165
166static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
167{
168 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700169 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700170 else if (nodes_empty(*nodes))
171 return -EINVAL; /* no allowed nodes */
172 else
173 pol->v.preferred_node = first_node(*nodes);
174 return 0;
175}
176
177static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
178{
179 if (!is_valid_nodemask(nodes))
180 return -EINVAL;
181 pol->v.nodes = *nodes;
182 return 0;
183}
184
Miao Xie58568d22009-06-16 15:31:49 -0700185/*
186 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
187 * any, for the new policy. mpol_new() has already validated the nodes
188 * parameter with respect to the policy mode and flags. But, we need to
189 * handle an empty nodemask with MPOL_PREFERRED here.
190 *
191 * Must be called holding task's alloc_lock to protect task's mems_allowed
192 * and mempolicy. May also be called holding the mmap_semaphore for write.
193 */
194static int mpol_set_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
195{
196 nodemask_t cpuset_context_nmask;
197 int ret;
198
199 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
200 if (pol == NULL)
201 return 0;
202
203 VM_BUG_ON(!nodes);
204 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
205 nodes = NULL; /* explicit local allocation */
206 else {
207 if (pol->flags & MPOL_F_RELATIVE_NODES)
208 mpol_relative_nodemask(&cpuset_context_nmask, nodes,
209 &cpuset_current_mems_allowed);
210 else
211 nodes_and(cpuset_context_nmask, *nodes,
212 cpuset_current_mems_allowed);
213 if (mpol_store_user_nodemask(pol))
214 pol->w.user_nodemask = *nodes;
215 else
216 pol->w.cpuset_mems_allowed =
217 cpuset_current_mems_allowed;
218 }
219
220 ret = mpol_ops[pol->mode].create(pol,
221 nodes ? &cpuset_context_nmask : NULL);
222 return ret;
223}
224
225/*
226 * This function just creates a new policy, does some check and simple
227 * initialization. You must invoke mpol_set_nodemask() to set nodes.
228 */
David Rientjes028fec42008-04-28 02:12:25 -0700229static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
230 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 struct mempolicy *policy;
233
David Rientjes028fec42008-04-28 02:12:25 -0700234 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
235 mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
Paul Mundt140d5a42007-07-15 23:38:16 -0700236
David Rientjes3e1f0642008-04-28 02:12:34 -0700237 if (mode == MPOL_DEFAULT) {
238 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700239 return ERR_PTR(-EINVAL);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700240 return NULL; /* simply delete any existing policy */
David Rientjes37012942008-04-28 02:12:33 -0700241 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700242 VM_BUG_ON(!nodes);
243
244 /*
245 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
246 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
247 * All other modes require a valid pointer to a non-empty nodemask.
248 */
249 if (mode == MPOL_PREFERRED) {
250 if (nodes_empty(*nodes)) {
251 if (((flags & MPOL_F_STATIC_NODES) ||
252 (flags & MPOL_F_RELATIVE_NODES)))
253 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700254 }
255 } else if (nodes_empty(*nodes))
256 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
258 if (!policy)
259 return ERR_PTR(-ENOMEM);
260 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700261 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700262 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700263
David Rientjes37012942008-04-28 02:12:33 -0700264 return policy;
265}
266
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700267/* Slow path of a mpol destructor. */
268void __mpol_put(struct mempolicy *p)
269{
270 if (!atomic_dec_and_test(&p->refcnt))
271 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700272 kmem_cache_free(policy_cache, p);
273}
274
David Rientjes37012942008-04-28 02:12:33 -0700275static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
276{
277}
278
279static void mpol_rebind_nodemask(struct mempolicy *pol,
280 const nodemask_t *nodes)
281{
282 nodemask_t tmp;
283
284 if (pol->flags & MPOL_F_STATIC_NODES)
285 nodes_and(tmp, pol->w.user_nodemask, *nodes);
286 else if (pol->flags & MPOL_F_RELATIVE_NODES)
287 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
288 else {
289 nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed,
290 *nodes);
291 pol->w.cpuset_mems_allowed = *nodes;
292 }
293
294 pol->v.nodes = tmp;
295 if (!node_isset(current->il_next, tmp)) {
296 current->il_next = next_node(current->il_next, tmp);
297 if (current->il_next >= MAX_NUMNODES)
298 current->il_next = first_node(tmp);
299 if (current->il_next >= MAX_NUMNODES)
300 current->il_next = numa_node_id();
301 }
302}
303
304static void mpol_rebind_preferred(struct mempolicy *pol,
305 const nodemask_t *nodes)
306{
307 nodemask_t tmp;
308
David Rientjes37012942008-04-28 02:12:33 -0700309 if (pol->flags & MPOL_F_STATIC_NODES) {
310 int node = first_node(pol->w.user_nodemask);
311
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700312 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700313 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700314 pol->flags &= ~MPOL_F_LOCAL;
315 } else
316 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700317 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
318 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
319 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700320 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700321 pol->v.preferred_node = node_remap(pol->v.preferred_node,
322 pol->w.cpuset_mems_allowed,
323 *nodes);
324 pol->w.cpuset_mems_allowed = *nodes;
325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
David Rientjes1d0d2682008-04-28 02:12:32 -0700328/* Migrate a policy to a different set of nodes */
329static void mpol_rebind_policy(struct mempolicy *pol,
330 const nodemask_t *newmask)
331{
David Rientjes1d0d2682008-04-28 02:12:32 -0700332 if (!pol)
333 return;
David Rientjes1d0d2682008-04-28 02:12:32 -0700334 if (!mpol_store_user_nodemask(pol) &&
335 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
336 return;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700337 mpol_ops[pol->mode].rebind(pol, newmask);
David Rientjes1d0d2682008-04-28 02:12:32 -0700338}
339
340/*
341 * Wrapper for mpol_rebind_policy() that just requires task
342 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700343 *
344 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700345 */
346
347void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
348{
349 mpol_rebind_policy(tsk->mempolicy, new);
350}
351
352/*
353 * Rebind each vma in mm to new nodemask.
354 *
355 * Call holding a reference to mm. Takes mm->mmap_sem during call.
356 */
357
358void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
359{
360 struct vm_area_struct *vma;
361
362 down_write(&mm->mmap_sem);
363 for (vma = mm->mmap; vma; vma = vma->vm_next)
364 mpol_rebind_policy(vma->vm_policy, new);
365 up_write(&mm->mmap_sem);
366}
367
David Rientjes37012942008-04-28 02:12:33 -0700368static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
369 [MPOL_DEFAULT] = {
370 .rebind = mpol_rebind_default,
371 },
372 [MPOL_INTERLEAVE] = {
373 .create = mpol_new_interleave,
374 .rebind = mpol_rebind_nodemask,
375 },
376 [MPOL_PREFERRED] = {
377 .create = mpol_new_preferred,
378 .rebind = mpol_rebind_preferred,
379 },
380 [MPOL_BIND] = {
381 .create = mpol_new_bind,
382 .rebind = mpol_rebind_nodemask,
383 },
384};
385
Christoph Lameter397874d2006-03-06 15:42:53 -0800386static void gather_stats(struct page *, void *, int pte_dirty);
Christoph Lameterfc301282006-01-18 17:42:29 -0800387static void migrate_page_add(struct page *page, struct list_head *pagelist,
388 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800389
Christoph Lameter38e35862006-01-08 01:01:01 -0800390/* Scan through pages checking if pages follow certain conditions. */
Nick Pigginb5810032005-10-29 18:16:12 -0700391static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800392 unsigned long addr, unsigned long end,
393 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800394 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Hugh Dickins91612e02005-06-21 17:15:07 -0700396 pte_t *orig_pte;
397 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700398 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700399
Hugh Dickins705e87c2005-10-29 18:16:27 -0700400 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700401 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800402 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800403 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700404
405 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800407 page = vm_normal_page(vma, addr, *pte);
408 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800410 /*
411 * The check for PageReserved here is important to avoid
412 * handling zero pages and other pages that may have been
413 * marked special by the system.
414 *
415 * If the PageReserved would not be checked here then f.e.
416 * the location of the zero page could have an influence
417 * on MPOL_MF_STRICT, zero pages would be counted for
418 * the per node stats, and there would be useless attempts
419 * to put zero pages on the migration list.
420 */
Christoph Lameterf4598c82006-01-12 01:05:20 -0800421 if (PageReserved(page))
422 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800423 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800424 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
425 continue;
426
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800427 if (flags & MPOL_MF_STATS)
Christoph Lameter397874d2006-03-06 15:42:53 -0800428 gather_stats(page, private, pte_dirty(*pte));
Nick Piggin053837f2006-01-18 17:42:27 -0800429 else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800430 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800431 else
432 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700433 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700434 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700435 return addr != end;
436}
437
Nick Pigginb5810032005-10-29 18:16:12 -0700438static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800439 unsigned long addr, unsigned long end,
440 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800441 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700442{
443 pmd_t *pmd;
444 unsigned long next;
445
446 pmd = pmd_offset(pud, addr);
447 do {
448 next = pmd_addr_end(addr, end);
449 if (pmd_none_or_clear_bad(pmd))
450 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800451 if (check_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800452 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700453 return -EIO;
454 } while (pmd++, addr = next, addr != end);
455 return 0;
456}
457
Nick Pigginb5810032005-10-29 18:16:12 -0700458static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800459 unsigned long addr, unsigned long end,
460 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800461 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700462{
463 pud_t *pud;
464 unsigned long next;
465
466 pud = pud_offset(pgd, addr);
467 do {
468 next = pud_addr_end(addr, end);
469 if (pud_none_or_clear_bad(pud))
470 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800471 if (check_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800472 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700473 return -EIO;
474 } while (pud++, addr = next, addr != end);
475 return 0;
476}
477
Nick Pigginb5810032005-10-29 18:16:12 -0700478static inline int check_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800479 unsigned long addr, unsigned long end,
480 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800481 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700482{
483 pgd_t *pgd;
484 unsigned long next;
485
Nick Pigginb5810032005-10-29 18:16:12 -0700486 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700487 do {
488 next = pgd_addr_end(addr, end);
489 if (pgd_none_or_clear_bad(pgd))
490 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800491 if (check_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800492 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700493 return -EIO;
494 } while (pgd++, addr = next, addr != end);
495 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
497
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800498/*
499 * Check if all pages in a range are on a set of nodes.
500 * If pagelist != NULL then isolate pages from the LRU and
501 * put them on the pagelist.
502 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503static struct vm_area_struct *
504check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800505 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
507 int err;
508 struct vm_area_struct *first, *vma, *prev;
509
Nick Piggin053837f2006-01-18 17:42:27 -0800510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 first = find_vma(mm, start);
512 if (!first)
513 return ERR_PTR(-EFAULT);
514 prev = NULL;
515 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800516 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
517 if (!vma->vm_next && vma->vm_end < end)
518 return ERR_PTR(-EFAULT);
519 if (prev && prev->vm_end < vma->vm_start)
520 return ERR_PTR(-EFAULT);
521 }
522 if (!is_vm_hugetlb_page(vma) &&
523 ((flags & MPOL_MF_STRICT) ||
524 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
525 vma_migratable(vma)))) {
Andi Kleen5b952b32005-09-13 01:25:08 -0700526 unsigned long endvma = vma->vm_end;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800527
Andi Kleen5b952b32005-09-13 01:25:08 -0700528 if (endvma > end)
529 endvma = end;
530 if (vma->vm_start > start)
531 start = vma->vm_start;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800532 err = check_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800533 flags, private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (err) {
535 first = ERR_PTR(err);
536 break;
537 }
538 }
539 prev = vma;
540 }
541 return first;
542}
543
544/* Apply policy to a single VMA */
545static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new)
546{
547 int err = 0;
548 struct mempolicy *old = vma->vm_policy;
549
Paul Mundt140d5a42007-07-15 23:38:16 -0700550 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 vma->vm_start, vma->vm_end, vma->vm_pgoff,
552 vma->vm_ops, vma->vm_file,
553 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
554
555 if (vma->vm_ops && vma->vm_ops->set_policy)
556 err = vma->vm_ops->set_policy(vma, new);
557 if (!err) {
558 mpol_get(new);
559 vma->vm_policy = new;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700560 mpol_put(old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 }
562 return err;
563}
564
565/* Step 2: apply policy to a range and do splits. */
566static int mbind_range(struct vm_area_struct *vma, unsigned long start,
567 unsigned long end, struct mempolicy *new)
568{
569 struct vm_area_struct *next;
570 int err;
571
572 err = 0;
573 for (; vma && vma->vm_start < end; vma = next) {
574 next = vma->vm_next;
575 if (vma->vm_start < start)
576 err = split_vma(vma->vm_mm, vma, start, 1);
577 if (!err && vma->vm_end > end)
578 err = split_vma(vma->vm_mm, vma, end, 0);
579 if (!err)
580 err = policy_vma(vma, new);
581 if (err)
582 break;
583 }
584 return err;
585}
586
Paul Jacksonc61afb12006-03-24 03:16:08 -0800587/*
588 * Update task->flags PF_MEMPOLICY bit: set iff non-default
589 * mempolicy. Allows more rapid checking of this (combined perhaps
590 * with other PF_* flag bits) on memory allocation hot code paths.
591 *
592 * If called from outside this file, the task 'p' should -only- be
593 * a newly forked child not yet visible on the task list, because
594 * manipulating the task flags of a visible task is not safe.
595 *
596 * The above limitation is why this routine has the funny name
597 * mpol_fix_fork_child_flag().
598 *
599 * It is also safe to call this with a task pointer of current,
600 * which the static wrapper mpol_set_task_struct_flag() does,
601 * for use within this file.
602 */
603
604void mpol_fix_fork_child_flag(struct task_struct *p)
605{
606 if (p->mempolicy)
607 p->flags |= PF_MEMPOLICY;
608 else
609 p->flags &= ~PF_MEMPOLICY;
610}
611
612static void mpol_set_task_struct_flag(void)
613{
614 mpol_fix_fork_child_flag(current);
615}
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700618static long do_set_mempolicy(unsigned short mode, unsigned short flags,
619 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
Miao Xie58568d22009-06-16 15:31:49 -0700621 struct mempolicy *new, *old;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700622 struct mm_struct *mm = current->mm;
Miao Xie58568d22009-06-16 15:31:49 -0700623 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
David Rientjes028fec42008-04-28 02:12:25 -0700625 new = mpol_new(mode, flags, nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (IS_ERR(new))
627 return PTR_ERR(new);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700628
629 /*
630 * prevent changing our mempolicy while show_numa_maps()
631 * is using it.
632 * Note: do_set_mempolicy() can be called at init time
633 * with no 'mm'.
634 */
635 if (mm)
636 down_write(&mm->mmap_sem);
Miao Xie58568d22009-06-16 15:31:49 -0700637 task_lock(current);
638 ret = mpol_set_nodemask(new, nodes);
639 if (ret) {
640 task_unlock(current);
641 if (mm)
642 up_write(&mm->mmap_sem);
643 mpol_put(new);
644 return ret;
645 }
646 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 current->mempolicy = new;
Paul Jacksonc61afb12006-03-24 03:16:08 -0800648 mpol_set_task_struct_flag();
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700649 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700650 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700651 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700652 task_unlock(current);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700653 if (mm)
654 up_write(&mm->mmap_sem);
655
Miao Xie58568d22009-06-16 15:31:49 -0700656 mpol_put(old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 return 0;
658}
659
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700660/*
661 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700662 *
663 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700664 */
665static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700667 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700668 if (p == &default_policy)
669 return;
670
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700671 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700672 case MPOL_BIND:
673 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700675 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 break;
677 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700678 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700679 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700680 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 break;
682 default:
683 BUG();
684 }
685}
686
687static int lookup_node(struct mm_struct *mm, unsigned long addr)
688{
689 struct page *p;
690 int err;
691
692 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
693 if (err >= 0) {
694 err = page_to_nid(p);
695 put_page(p);
696 }
697 return err;
698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700701static long do_get_mempolicy(int *policy, nodemask_t *nmask,
702 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700704 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 struct mm_struct *mm = current->mm;
706 struct vm_area_struct *vma = NULL;
707 struct mempolicy *pol = current->mempolicy;
708
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700709 if (flags &
710 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700712
713 if (flags & MPOL_F_MEMS_ALLOWED) {
714 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
715 return -EINVAL;
716 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700717 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700718 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700719 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700720 return 0;
721 }
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700724 /*
725 * Do NOT fall back to task policy if the
726 * vma/shared policy at addr is NULL. We
727 * want to return MPOL_DEFAULT in this case.
728 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 down_read(&mm->mmap_sem);
730 vma = find_vma_intersection(mm, addr, addr+1);
731 if (!vma) {
732 up_read(&mm->mmap_sem);
733 return -EFAULT;
734 }
735 if (vma->vm_ops && vma->vm_ops->get_policy)
736 pol = vma->vm_ops->get_policy(vma, addr);
737 else
738 pol = vma->vm_policy;
739 } else if (addr)
740 return -EINVAL;
741
742 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700743 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
745 if (flags & MPOL_F_NODE) {
746 if (flags & MPOL_F_ADDR) {
747 err = lookup_node(mm, addr);
748 if (err < 0)
749 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700750 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700752 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700753 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 } else {
755 err = -EINVAL;
756 goto out;
757 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700758 } else {
759 *policy = pol == &default_policy ? MPOL_DEFAULT :
760 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700761 /*
762 * Internal mempolicy flags must be masked off before exposing
763 * the policy to userspace.
764 */
765 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 if (vma) {
769 up_read(&current->mm->mmap_sem);
770 vma = NULL;
771 }
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700774 if (nmask) {
775 task_lock(current);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700776 get_policy_nodemask(pol, nmask);
Miao Xie58568d22009-06-16 15:31:49 -0700777 task_unlock(current);
778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700781 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (vma)
783 up_read(&current->mm->mmap_sem);
784 return err;
785}
786
Christoph Lameterb20a3502006-03-22 00:09:12 -0800787#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700788/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800789 * page migration
790 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800791static void migrate_page_add(struct page *page, struct list_head *pagelist,
792 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800793{
794 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800795 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800796 */
Nick Piggin62695a82008-10-18 20:26:09 -0700797 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
798 if (!isolate_lru_page(page)) {
799 list_add_tail(&page->lru, pagelist);
800 }
801 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800802}
803
Christoph Lameter742755a2006-06-23 02:03:55 -0700804static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700805{
Mel Gorman6484eb32009-06-16 15:31:54 -0700806 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700807}
808
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800809/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800810 * Migrate pages from one node to a target node.
811 * Returns error or the number of pages not migrated.
812 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700813static int migrate_to_node(struct mm_struct *mm, int source, int dest,
814 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800815{
816 nodemask_t nmask;
817 LIST_HEAD(pagelist);
818 int err = 0;
819
820 nodes_clear(nmask);
821 node_set(source, nmask);
822
823 check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask,
824 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
825
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700826 if (!list_empty(&pagelist))
Christoph Lameter95a402c2006-06-23 02:03:53 -0700827 err = migrate_pages(&pagelist, new_node_page, dest);
828
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800829 return err;
830}
831
832/*
833 * Move pages between the two nodesets so as to preserve the physical
834 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -0800835 *
836 * Returns the number of page that could not be moved.
837 */
838int do_migrate_pages(struct mm_struct *mm,
839 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
840{
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800841 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800842 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800843 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -0800844
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800845 err = migrate_prep();
846 if (err)
847 return err;
848
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700849 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -0800850
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700851 err = migrate_vmas(mm, from_nodes, to_nodes, flags);
852 if (err)
853 goto out;
854
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800855/*
856 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
857 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
858 * bit in 'tmp', and return that <source, dest> pair for migration.
859 * The pair of nodemasks 'to' and 'from' define the map.
860 *
861 * If no pair of bits is found that way, fallback to picking some
862 * pair of 'source' and 'dest' bits that are not the same. If the
863 * 'source' and 'dest' bits are the same, this represents a node
864 * that will be migrating to itself, so no pages need move.
865 *
866 * If no bits are left in 'tmp', or if all remaining bits left
867 * in 'tmp' correspond to the same bit in 'to', return false
868 * (nothing left to migrate).
869 *
870 * This lets us pick a pair of nodes to migrate between, such that
871 * if possible the dest node is not already occupied by some other
872 * source node, minimizing the risk of overloading the memory on a
873 * node that would happen if we migrated incoming memory to a node
874 * before migrating outgoing memory source that same node.
875 *
876 * A single scan of tmp is sufficient. As we go, we remember the
877 * most recent <s, d> pair that moved (s != d). If we find a pair
878 * that not only moved, but what's better, moved to an empty slot
879 * (d is not set in tmp), then we break out then, with that pair.
880 * Otherwise when we finish scannng from_tmp, we at least have the
881 * most recent <s, d> pair that moved. If we get all the way through
882 * the scan of tmp without finding any node that moved, much less
883 * moved to an empty node, then there is nothing left worth migrating.
884 */
Christoph Lameterd4984712006-01-08 01:00:55 -0800885
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800886 tmp = *from_nodes;
887 while (!nodes_empty(tmp)) {
888 int s,d;
889 int source = -1;
890 int dest = 0;
891
892 for_each_node_mask(s, tmp) {
893 d = node_remap(s, *from_nodes, *to_nodes);
894 if (s == d)
895 continue;
896
897 source = s; /* Node moved. Memorize */
898 dest = d;
899
900 /* dest not in remaining from nodes? */
901 if (!node_isset(dest, tmp))
902 break;
903 }
904 if (source == -1)
905 break;
906
907 node_clear(source, tmp);
908 err = migrate_to_node(mm, source, dest, flags);
909 if (err > 0)
910 busy += err;
911 if (err < 0)
912 break;
Christoph Lameter39743882006-01-08 01:00:51 -0800913 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700914out:
Christoph Lameter39743882006-01-08 01:00:51 -0800915 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800916 if (err < 0)
917 return err;
918 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800919
Christoph Lameter39743882006-01-08 01:00:51 -0800920}
921
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800922/*
923 * Allocate a new page for page migration based on vma policy.
924 * Start assuming that page is mapped by vma pointed to by @private.
925 * Search forward from there, if not. N.B., this assumes that the
926 * list of pages handed to migrate_pages()--which is how we get here--
927 * is in virtual address order.
928 */
Christoph Lameter742755a2006-06-23 02:03:55 -0700929static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700930{
931 struct vm_area_struct *vma = (struct vm_area_struct *)private;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800932 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700933
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800934 while (vma) {
935 address = page_address_in_vma(page, vma);
936 if (address != -EFAULT)
937 break;
938 vma = vma->vm_next;
939 }
940
941 /*
942 * if !vma, alloc_page_vma() will use task or system default policy
943 */
944 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700945}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800946#else
947
948static void migrate_page_add(struct page *page, struct list_head *pagelist,
949 unsigned long flags)
950{
951}
952
953int do_migrate_pages(struct mm_struct *mm,
954 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
955{
956 return -ENOSYS;
957}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700958
Keith Owens69939742006-10-11 01:21:28 -0700959static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700960{
961 return NULL;
962}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800963#endif
964
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700965static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -0700966 unsigned short mode, unsigned short mode_flags,
967 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800968{
969 struct vm_area_struct *vma;
970 struct mm_struct *mm = current->mm;
971 struct mempolicy *new;
972 unsigned long end;
973 int err;
974 LIST_HEAD(pagelist);
975
David Rientjesa3b51e02008-04-28 02:12:23 -0700976 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
977 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800978 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -0800979 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800980 return -EPERM;
981
982 if (start & ~PAGE_MASK)
983 return -EINVAL;
984
985 if (mode == MPOL_DEFAULT)
986 flags &= ~MPOL_MF_STRICT;
987
988 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
989 end = start + len;
990
991 if (end < start)
992 return -EINVAL;
993 if (end == start)
994 return 0;
995
David Rientjes028fec42008-04-28 02:12:25 -0700996 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800997 if (IS_ERR(new))
998 return PTR_ERR(new);
999
1000 /*
1001 * If we are using the default policy then operation
1002 * on discontinuous address spaces is okay after all
1003 */
1004 if (!new)
1005 flags |= MPOL_MF_DISCONTIG_OK;
1006
David Rientjes028fec42008-04-28 02:12:25 -07001007 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1008 start, start + len, mode, mode_flags,
1009 nmask ? nodes_addr(*nmask)[0] : -1);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001010
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001011 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1012
1013 err = migrate_prep();
1014 if (err)
1015 return err;
1016 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001017 down_write(&mm->mmap_sem);
Miao Xie58568d22009-06-16 15:31:49 -07001018 task_lock(current);
1019 err = mpol_set_nodemask(new, nmask);
1020 task_unlock(current);
1021 if (err) {
1022 up_write(&mm->mmap_sem);
1023 mpol_put(new);
1024 return err;
1025 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001026 vma = check_range(mm, start, end, nmask,
1027 flags | MPOL_MF_INVERT, &pagelist);
1028
1029 err = PTR_ERR(vma);
1030 if (!IS_ERR(vma)) {
1031 int nr_failed = 0;
1032
1033 err = mbind_range(vma, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001034
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001035 if (!list_empty(&pagelist))
Christoph Lameter95a402c2006-06-23 02:03:53 -07001036 nr_failed = migrate_pages(&pagelist, new_vma_page,
1037 (unsigned long)vma);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001038
1039 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
1040 err = -EIO;
1041 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001042
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001043 up_write(&mm->mmap_sem);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001044 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001045 return err;
1046}
1047
Christoph Lameter39743882006-01-08 01:00:51 -08001048/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001049 * User space interface with variable sized bitmaps for nodelists.
1050 */
1051
1052/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001053static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001054 unsigned long maxnode)
1055{
1056 unsigned long k;
1057 unsigned long nlongs;
1058 unsigned long endmask;
1059
1060 --maxnode;
1061 nodes_clear(*nodes);
1062 if (maxnode == 0 || !nmask)
1063 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001064 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001065 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001066
1067 nlongs = BITS_TO_LONGS(maxnode);
1068 if ((maxnode % BITS_PER_LONG) == 0)
1069 endmask = ~0UL;
1070 else
1071 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1072
1073 /* When the user specified more nodes than supported just check
1074 if the non supported part is all zero. */
1075 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1076 if (nlongs > PAGE_SIZE/sizeof(long))
1077 return -EINVAL;
1078 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1079 unsigned long t;
1080 if (get_user(t, nmask + k))
1081 return -EFAULT;
1082 if (k == nlongs - 1) {
1083 if (t & endmask)
1084 return -EINVAL;
1085 } else if (t)
1086 return -EINVAL;
1087 }
1088 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1089 endmask = ~0UL;
1090 }
1091
1092 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1093 return -EFAULT;
1094 nodes_addr(*nodes)[nlongs-1] &= endmask;
1095 return 0;
1096}
1097
1098/* Copy a kernel node mask to user space */
1099static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1100 nodemask_t *nodes)
1101{
1102 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1103 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1104
1105 if (copy > nbytes) {
1106 if (copy > PAGE_SIZE)
1107 return -EINVAL;
1108 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1109 return -EFAULT;
1110 copy = nbytes;
1111 }
1112 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1113}
1114
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001115SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1116 unsigned long, mode, unsigned long __user *, nmask,
1117 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001118{
1119 nodemask_t nodes;
1120 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001121 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001122
David Rientjes028fec42008-04-28 02:12:25 -07001123 mode_flags = mode & MPOL_MODE_FLAGS;
1124 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001125 if (mode >= MPOL_MAX)
1126 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001127 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1128 (mode_flags & MPOL_F_RELATIVE_NODES))
1129 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001130 err = get_nodes(&nodes, nmask, maxnode);
1131 if (err)
1132 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001133 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001134}
1135
1136/* Set the process memory policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001137SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
1138 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001139{
1140 int err;
1141 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001142 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001143
David Rientjes028fec42008-04-28 02:12:25 -07001144 flags = mode & MPOL_MODE_FLAGS;
1145 mode &= ~MPOL_MODE_FLAGS;
1146 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001147 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001148 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1149 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001150 err = get_nodes(&nodes, nmask, maxnode);
1151 if (err)
1152 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001153 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001154}
1155
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001156SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1157 const unsigned long __user *, old_nodes,
1158 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001159{
David Howellsc69e8d92008-11-14 10:39:19 +11001160 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter39743882006-01-08 01:00:51 -08001161 struct mm_struct *mm;
1162 struct task_struct *task;
1163 nodemask_t old;
1164 nodemask_t new;
1165 nodemask_t task_nodes;
1166 int err;
1167
1168 err = get_nodes(&old, old_nodes, maxnode);
1169 if (err)
1170 return err;
1171
1172 err = get_nodes(&new, new_nodes, maxnode);
1173 if (err)
1174 return err;
1175
1176 /* Find the mm_struct */
1177 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001178 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001179 if (!task) {
1180 read_unlock(&tasklist_lock);
1181 return -ESRCH;
1182 }
1183 mm = get_task_mm(task);
1184 read_unlock(&tasklist_lock);
1185
1186 if (!mm)
1187 return -EINVAL;
1188
1189 /*
1190 * Check if this process has the right to modify the specified
1191 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001192 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001193 * userid as the target process.
1194 */
David Howellsc69e8d92008-11-14 10:39:19 +11001195 rcu_read_lock();
1196 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001197 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1198 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001199 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001200 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001201 err = -EPERM;
1202 goto out;
1203 }
David Howellsc69e8d92008-11-14 10:39:19 +11001204 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001205
1206 task_nodes = cpuset_mems_allowed(task);
1207 /* Is the user allowed to access the target nodes? */
Christoph Lameter74c00242006-03-14 19:50:21 -08001208 if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001209 err = -EPERM;
1210 goto out;
1211 }
1212
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001213 if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001214 err = -EINVAL;
1215 goto out;
1216 }
1217
David Quigley86c3a762006-06-23 02:04:02 -07001218 err = security_task_movememory(task);
1219 if (err)
1220 goto out;
1221
Christoph Lameter511030b2006-02-28 16:58:57 -08001222 err = do_migrate_pages(mm, &old, &new,
Christoph Lameter74c00242006-03-14 19:50:21 -08001223 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter39743882006-01-08 01:00:51 -08001224out:
1225 mmput(mm);
1226 return err;
1227}
1228
1229
Christoph Lameter8bccd852005-10-29 18:16:59 -07001230/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001231SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1232 unsigned long __user *, nmask, unsigned long, maxnode,
1233 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001234{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001235 int err;
1236 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001237 nodemask_t nodes;
1238
1239 if (nmask != NULL && maxnode < MAX_NUMNODES)
1240 return -EINVAL;
1241
1242 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1243
1244 if (err)
1245 return err;
1246
1247 if (policy && put_user(pval, policy))
1248 return -EFAULT;
1249
1250 if (nmask)
1251 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1252
1253 return err;
1254}
1255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256#ifdef CONFIG_COMPAT
1257
1258asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1259 compat_ulong_t __user *nmask,
1260 compat_ulong_t maxnode,
1261 compat_ulong_t addr, compat_ulong_t flags)
1262{
1263 long err;
1264 unsigned long __user *nm = NULL;
1265 unsigned long nr_bits, alloc_size;
1266 DECLARE_BITMAP(bm, MAX_NUMNODES);
1267
1268 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1269 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1270
1271 if (nmask)
1272 nm = compat_alloc_user_space(alloc_size);
1273
1274 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1275
1276 if (!err && nmask) {
1277 err = copy_from_user(bm, nm, alloc_size);
1278 /* ensure entire bitmap is zeroed */
1279 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1280 err |= compat_put_bitmap(nmask, bm, nr_bits);
1281 }
1282
1283 return err;
1284}
1285
1286asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1287 compat_ulong_t maxnode)
1288{
1289 long err = 0;
1290 unsigned long __user *nm = NULL;
1291 unsigned long nr_bits, alloc_size;
1292 DECLARE_BITMAP(bm, MAX_NUMNODES);
1293
1294 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1295 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1296
1297 if (nmask) {
1298 err = compat_get_bitmap(bm, nmask, nr_bits);
1299 nm = compat_alloc_user_space(alloc_size);
1300 err |= copy_to_user(nm, bm, alloc_size);
1301 }
1302
1303 if (err)
1304 return -EFAULT;
1305
1306 return sys_set_mempolicy(mode, nm, nr_bits+1);
1307}
1308
1309asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1310 compat_ulong_t mode, compat_ulong_t __user *nmask,
1311 compat_ulong_t maxnode, compat_ulong_t flags)
1312{
1313 long err = 0;
1314 unsigned long __user *nm = NULL;
1315 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001316 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1319 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1320
1321 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001322 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001324 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 }
1326
1327 if (err)
1328 return -EFAULT;
1329
1330 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1331}
1332
1333#endif
1334
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001335/*
1336 * get_vma_policy(@task, @vma, @addr)
1337 * @task - task for fallback if vma policy == default
1338 * @vma - virtual memory area whose policy is sought
1339 * @addr - address in @vma for shared policy lookup
1340 *
1341 * Returns effective policy for a VMA at specified address.
1342 * Falls back to @task or system default policy, as necessary.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001343 * Current or other task's task mempolicy and non-shared vma policies
1344 * are protected by the task's mmap_sem, which must be held for read by
1345 * the caller.
1346 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1347 * count--added by the get_policy() vm_op, as appropriate--to protect against
1348 * freeing by another task. It is the caller's responsibility to free the
1349 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001350 */
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001351static struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001352 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001354 struct mempolicy *pol = task->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001357 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001358 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1359 addr);
1360 if (vpol)
1361 pol = vpol;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001362 } else if (vma->vm_policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 pol = vma->vm_policy;
1364 }
1365 if (!pol)
1366 pol = &default_policy;
1367 return pol;
1368}
1369
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001370/*
1371 * Return a nodemask representing a mempolicy for filtering nodes for
1372 * page allocation
1373 */
1374static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001375{
1376 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001377 if (unlikely(policy->mode == MPOL_BIND) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001378 gfp_zone(gfp) >= policy_zone &&
1379 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1380 return &policy->v.nodes;
1381
1382 return NULL;
1383}
1384
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001385/* Return a zonelist indicated by gfp for node representing a mempolicy */
1386static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001388 int nd = numa_node_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001390 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001392 if (!(policy->flags & MPOL_F_LOCAL))
1393 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 break;
1395 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001396 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001397 * Normally, MPOL_BIND allocations are node-local within the
1398 * allowed nodemask. However, if __GFP_THISNODE is set and the
1399 * current node is part of the mask, we use the zonelist for
1400 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001401 */
Mel Gorman19770b32008-04-28 02:12:18 -07001402 if (unlikely(gfp & __GFP_THISNODE) &&
1403 unlikely(!node_isset(nd, policy->v.nodes)))
1404 nd = first_node(policy->v.nodes);
1405 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 case MPOL_INTERLEAVE: /* should not happen */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 break;
1408 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 BUG();
1410 }
Mel Gorman0e884602008-04-28 02:12:14 -07001411 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
1414/* Do dynamic interleaving for a process */
1415static unsigned interleave_nodes(struct mempolicy *policy)
1416{
1417 unsigned nid, next;
1418 struct task_struct *me = current;
1419
1420 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001421 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001423 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001424 if (next < MAX_NUMNODES)
1425 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return nid;
1427}
1428
Christoph Lameterdc85da12006-01-18 17:42:36 -08001429/*
1430 * Depending on the memory policy provide a node from which to allocate the
1431 * next slab entry.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001432 * @policy must be protected by freeing by the caller. If @policy is
1433 * the current task's mempolicy, this protection is implicit, as only the
1434 * task can change it's policy. The system default policy requires no
1435 * such protection.
Christoph Lameterdc85da12006-01-18 17:42:36 -08001436 */
1437unsigned slab_node(struct mempolicy *policy)
1438{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001439 if (!policy || policy->flags & MPOL_F_LOCAL)
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001440 return numa_node_id();
Christoph Lameter765c4502006-09-27 01:50:08 -07001441
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001442 switch (policy->mode) {
1443 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001444 /*
1445 * handled MPOL_F_LOCAL above
1446 */
1447 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001448
Christoph Lameterdc85da12006-01-18 17:42:36 -08001449 case MPOL_INTERLEAVE:
1450 return interleave_nodes(policy);
1451
Mel Gormandd1a2392008-04-28 02:12:17 -07001452 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001453 /*
1454 * Follow bind policy behavior and start allocation at the
1455 * first node.
1456 */
Mel Gorman19770b32008-04-28 02:12:18 -07001457 struct zonelist *zonelist;
1458 struct zone *zone;
1459 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1460 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1461 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1462 &policy->v.nodes,
1463 &zone);
1464 return zone->node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001465 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001466
Christoph Lameterdc85da12006-01-18 17:42:36 -08001467 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001468 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001469 }
1470}
1471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472/* Do static interleaving for a VMA with known offset. */
1473static unsigned offset_il_node(struct mempolicy *pol,
1474 struct vm_area_struct *vma, unsigned long off)
1475{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001476 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001477 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 int c;
1479 int nid = -1;
1480
David Rientjesf5b087b2008-04-28 02:12:27 -07001481 if (!nnodes)
1482 return numa_node_id();
1483 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 c = 0;
1485 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001486 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 c++;
1488 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return nid;
1490}
1491
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001492/* Determine a node number for interleave */
1493static inline unsigned interleave_nid(struct mempolicy *pol,
1494 struct vm_area_struct *vma, unsigned long addr, int shift)
1495{
1496 if (vma) {
1497 unsigned long off;
1498
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001499 /*
1500 * for small pages, there is no difference between
1501 * shift and PAGE_SHIFT, so the bit-shift is safe.
1502 * for huge pages, since vm_pgoff is in units of small
1503 * pages, we need to shift off the always 0 bits to get
1504 * a useful offset.
1505 */
1506 BUG_ON(shift < PAGE_SHIFT);
1507 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001508 off += (addr - vma->vm_start) >> shift;
1509 return offset_il_node(pol, vma, off);
1510 } else
1511 return interleave_nodes(pol);
1512}
1513
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001514#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001515/*
1516 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1517 * @vma = virtual memory area whose policy is sought
1518 * @addr = address in @vma for shared policy lookup and interleave policy
1519 * @gfp_flags = for requested zone
Mel Gorman19770b32008-04-28 02:12:18 -07001520 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1521 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001522 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001523 * Returns a zonelist suitable for a huge page allocation and a pointer
1524 * to the struct mempolicy for conditional unref after allocation.
1525 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1526 * @nodemask for filtering the zonelist.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001527 */
Mel Gorman396faf02007-07-17 04:03:13 -07001528struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001529 gfp_t gfp_flags, struct mempolicy **mpol,
1530 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001531{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001532 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001533
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001534 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001535 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001536
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001537 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1538 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001539 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001540 } else {
1541 zl = policy_zonelist(gfp_flags, *mpol);
1542 if ((*mpol)->mode == MPOL_BIND)
1543 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001544 }
1545 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001546}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001547#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/* Allocate a page in interleaved policy.
1550 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001551static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1552 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553{
1554 struct zonelist *zl;
1555 struct page *page;
1556
Mel Gorman0e884602008-04-28 02:12:14 -07001557 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001559 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001560 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 return page;
1562}
1563
1564/**
1565 * alloc_page_vma - Allocate a page for a VMA.
1566 *
1567 * @gfp:
1568 * %GFP_USER user allocation.
1569 * %GFP_KERNEL kernel allocations,
1570 * %GFP_HIGHMEM highmem/user allocations,
1571 * %GFP_FS allocation should not call back into a file system.
1572 * %GFP_ATOMIC don't sleep.
1573 *
1574 * @vma: Pointer to VMA or NULL if not available.
1575 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1576 *
1577 * This function allocates a page from the kernel page pool and applies
1578 * a NUMA policy associated with the VMA or the current process.
1579 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1580 * mm_struct of the VMA to prevent it from going away. Should be used for
1581 * all allocations for pages that will be mapped into
1582 * user space. Returns NULL when no page can be allocated.
1583 *
1584 * Should be called with the mm_sem of the vma hold.
1585 */
1586struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001587alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001589 struct mempolicy *pol = get_vma_policy(current, vma, addr);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001590 struct zonelist *zl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001592 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001594
1595 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001596 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 return alloc_page_interleave(gfp, 0, nid);
1598 }
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001599 zl = policy_zonelist(gfp, pol);
1600 if (unlikely(mpol_needs_cond_ref(pol))) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001601 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001602 * slow path: ref counted shared policy
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001603 */
Mel Gorman19770b32008-04-28 02:12:18 -07001604 struct page *page = __alloc_pages_nodemask(gfp, 0,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001605 zl, policy_nodemask(gfp, pol));
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001606 __mpol_put(pol);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001607 return page;
1608 }
1609 /*
1610 * fast path: default or task policy
1611 */
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001612 return __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613}
1614
1615/**
1616 * alloc_pages_current - Allocate pages.
1617 *
1618 * @gfp:
1619 * %GFP_USER user allocation,
1620 * %GFP_KERNEL kernel allocation,
1621 * %GFP_HIGHMEM highmem allocation,
1622 * %GFP_FS don't call back into a file system.
1623 * %GFP_ATOMIC don't sleep.
1624 * @order: Power of two of allocation size in pages. 0 is a single page.
1625 *
1626 * Allocate a page from the kernel page pool. When not in
1627 * interrupt context and apply the current process NUMA policy.
1628 * Returns NULL when no page can be allocated.
1629 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001630 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 * 1) it's ok to take cpuset_sem (can WAIT), and
1632 * 2) allocating for current task (not interrupt).
1633 */
Al Virodd0fc662005-10-07 07:46:04 +01001634struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635{
1636 struct mempolicy *pol = current->mempolicy;
1637
Christoph Lameter9b819d22006-09-25 23:31:40 -07001638 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001640
1641 /*
1642 * No reference counting needed for current->mempolicy
1643 * nor system default_policy
1644 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001645 if (pol->mode == MPOL_INTERLEAVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return alloc_page_interleave(gfp, order, interleave_nodes(pol));
Mel Gorman19770b32008-04-28 02:12:18 -07001647 return __alloc_pages_nodemask(gfp, order,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001648 policy_zonelist(gfp, pol), policy_nodemask(gfp, pol));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649}
1650EXPORT_SYMBOL(alloc_pages_current);
1651
Paul Jackson42253992006-01-08 01:01:59 -08001652/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001653 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08001654 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1655 * with the mems_allowed returned by cpuset_mems_allowed(). This
1656 * keeps mempolicies cpuset relative after its cpuset moves. See
1657 * further kernel/cpuset.c update_nodemask().
1658 */
Paul Jackson42253992006-01-08 01:01:59 -08001659
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001660/* Slow path of a mempolicy duplicate */
1661struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
1663 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1664
1665 if (!new)
1666 return ERR_PTR(-ENOMEM);
Paul Jackson42253992006-01-08 01:01:59 -08001667 if (current_cpuset_is_being_rebound()) {
1668 nodemask_t mems = cpuset_mems_allowed(current);
1669 mpol_rebind_policy(old, &mems);
1670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 *new = *old;
1672 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return new;
1674}
1675
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001676/*
1677 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1678 * eliminate the * MPOL_F_* flags that require conditional ref and
1679 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1680 * after return. Use the returned value.
1681 *
1682 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1683 * policy lookup, even if the policy needs/has extra ref on lookup.
1684 * shmem_readahead needs this.
1685 */
1686struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
1687 struct mempolicy *frompol)
1688{
1689 if (!mpol_needs_cond_ref(frompol))
1690 return frompol;
1691
1692 *tompol = *frompol;
1693 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
1694 __mpol_put(frompol);
1695 return tompol;
1696}
1697
David Rientjesf5b087b2008-04-28 02:12:27 -07001698static int mpol_match_intent(const struct mempolicy *a,
1699 const struct mempolicy *b)
1700{
1701 if (a->flags != b->flags)
1702 return 0;
1703 if (!mpol_store_user_nodemask(a))
1704 return 1;
1705 return nodes_equal(a->w.user_nodemask, b->w.user_nodemask);
1706}
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708/* Slow path of a mempolicy comparison */
1709int __mpol_equal(struct mempolicy *a, struct mempolicy *b)
1710{
1711 if (!a || !b)
1712 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001713 if (a->mode != b->mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001715 if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b))
David Rientjesf5b087b2008-04-28 02:12:27 -07001716 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001717 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07001718 case MPOL_BIND:
1719 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -07001721 return nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001723 return a->v.preferred_node == b->v.preferred_node &&
1724 a->flags == b->flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 default:
1726 BUG();
1727 return 0;
1728 }
1729}
1730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 * Shared memory backing store policy support.
1733 *
1734 * Remember policies even when nobody has shared memory mapped.
1735 * The policies are kept in Red-Black tree linked from the inode.
1736 * They are protected by the sp->lock spinlock, which should be held
1737 * for any accesses to the tree.
1738 */
1739
1740/* lookup first element intersecting start-end */
1741/* Caller holds sp->lock */
1742static struct sp_node *
1743sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
1744{
1745 struct rb_node *n = sp->root.rb_node;
1746
1747 while (n) {
1748 struct sp_node *p = rb_entry(n, struct sp_node, nd);
1749
1750 if (start >= p->end)
1751 n = n->rb_right;
1752 else if (end <= p->start)
1753 n = n->rb_left;
1754 else
1755 break;
1756 }
1757 if (!n)
1758 return NULL;
1759 for (;;) {
1760 struct sp_node *w = NULL;
1761 struct rb_node *prev = rb_prev(n);
1762 if (!prev)
1763 break;
1764 w = rb_entry(prev, struct sp_node, nd);
1765 if (w->end <= start)
1766 break;
1767 n = prev;
1768 }
1769 return rb_entry(n, struct sp_node, nd);
1770}
1771
1772/* Insert a new shared policy into the list. */
1773/* Caller holds sp->lock */
1774static void sp_insert(struct shared_policy *sp, struct sp_node *new)
1775{
1776 struct rb_node **p = &sp->root.rb_node;
1777 struct rb_node *parent = NULL;
1778 struct sp_node *nd;
1779
1780 while (*p) {
1781 parent = *p;
1782 nd = rb_entry(parent, struct sp_node, nd);
1783 if (new->start < nd->start)
1784 p = &(*p)->rb_left;
1785 else if (new->end > nd->end)
1786 p = &(*p)->rb_right;
1787 else
1788 BUG();
1789 }
1790 rb_link_node(&new->nd, parent, p);
1791 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07001792 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001793 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794}
1795
1796/* Find shared policy intersecting idx */
1797struct mempolicy *
1798mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
1799{
1800 struct mempolicy *pol = NULL;
1801 struct sp_node *sn;
1802
1803 if (!sp->root.rb_node)
1804 return NULL;
1805 spin_lock(&sp->lock);
1806 sn = sp_lookup(sp, idx, idx+1);
1807 if (sn) {
1808 mpol_get(sn->policy);
1809 pol = sn->policy;
1810 }
1811 spin_unlock(&sp->lock);
1812 return pol;
1813}
1814
1815static void sp_delete(struct shared_policy *sp, struct sp_node *n)
1816{
Paul Mundt140d5a42007-07-15 23:38:16 -07001817 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 rb_erase(&n->nd, &sp->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001819 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 kmem_cache_free(sn_cache, n);
1821}
1822
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001823static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
1824 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825{
1826 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
1827
1828 if (!n)
1829 return NULL;
1830 n->start = start;
1831 n->end = end;
1832 mpol_get(pol);
Lee Schermerhornaab0b102008-04-28 02:13:13 -07001833 pol->flags |= MPOL_F_SHARED; /* for unref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 n->policy = pol;
1835 return n;
1836}
1837
1838/* Replace a policy range. */
1839static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
1840 unsigned long end, struct sp_node *new)
1841{
1842 struct sp_node *n, *new2 = NULL;
1843
1844restart:
1845 spin_lock(&sp->lock);
1846 n = sp_lookup(sp, start, end);
1847 /* Take care of old policies in the same range. */
1848 while (n && n->start < end) {
1849 struct rb_node *next = rb_next(&n->nd);
1850 if (n->start >= start) {
1851 if (n->end <= end)
1852 sp_delete(sp, n);
1853 else
1854 n->start = end;
1855 } else {
1856 /* Old policy spanning whole new range. */
1857 if (n->end > end) {
1858 if (!new2) {
1859 spin_unlock(&sp->lock);
1860 new2 = sp_alloc(end, n->end, n->policy);
1861 if (!new2)
1862 return -ENOMEM;
1863 goto restart;
1864 }
1865 n->end = start;
1866 sp_insert(sp, new2);
1867 new2 = NULL;
1868 break;
1869 } else
1870 n->end = start;
1871 }
1872 if (!next)
1873 break;
1874 n = rb_entry(next, struct sp_node, nd);
1875 }
1876 if (new)
1877 sp_insert(sp, new);
1878 spin_unlock(&sp->lock);
1879 if (new2) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001880 mpol_put(new2->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 kmem_cache_free(sn_cache, new2);
1882 }
1883 return 0;
1884}
1885
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001886/**
1887 * mpol_shared_policy_init - initialize shared policy for inode
1888 * @sp: pointer to inode shared policy
1889 * @mpol: struct mempolicy to install
1890 *
1891 * Install non-NULL @mpol in inode's shared policy rb-tree.
1892 * On entry, the current task has a reference on a non-NULL @mpol.
1893 * This must be released on exit.
1894 */
1895void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08001896{
Miao Xie58568d22009-06-16 15:31:49 -07001897 int ret;
1898
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001899 sp->root = RB_ROOT; /* empty tree == default mempolicy */
1900 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08001901
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001902 if (mpol) {
1903 struct vm_area_struct pvma;
1904 struct mempolicy *new;
Robin Holt7339ff82006-01-14 13:20:48 -08001905
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001906 /* contextualize the tmpfs mount point mempolicy */
1907 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Miao Xie58568d22009-06-16 15:31:49 -07001908 if (IS_ERR(new)) {
1909 mpol_put(mpol); /* drop our ref on sb mpol */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001910 return; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07001911 }
1912
1913 task_lock(current);
1914 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask);
1915 task_unlock(current);
1916 mpol_put(mpol); /* drop our ref on sb mpol */
1917 if (ret) {
1918 mpol_put(new);
1919 return;
1920 }
Robin Holt7339ff82006-01-14 13:20:48 -08001921
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001922 /* Create pseudo-vma that contains just the policy */
1923 memset(&pvma, 0, sizeof(struct vm_area_struct));
1924 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
1925 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
1926 mpol_put(new); /* drop initial ref */
Robin Holt7339ff82006-01-14 13:20:48 -08001927 }
1928}
1929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930int mpol_set_shared_policy(struct shared_policy *info,
1931 struct vm_area_struct *vma, struct mempolicy *npol)
1932{
1933 int err;
1934 struct sp_node *new = NULL;
1935 unsigned long sz = vma_pages(vma);
1936
David Rientjes028fec42008-04-28 02:12:25 -07001937 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001939 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07001940 npol ? npol->flags : -1,
Paul Mundt140d5a42007-07-15 23:38:16 -07001941 npol ? nodes_addr(npol->v.nodes)[0] : -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
1943 if (npol) {
1944 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
1945 if (!new)
1946 return -ENOMEM;
1947 }
1948 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
1949 if (err && new)
1950 kmem_cache_free(sn_cache, new);
1951 return err;
1952}
1953
1954/* Free a backing policy store on inode delete. */
1955void mpol_free_shared_policy(struct shared_policy *p)
1956{
1957 struct sp_node *n;
1958 struct rb_node *next;
1959
1960 if (!p->root.rb_node)
1961 return;
1962 spin_lock(&p->lock);
1963 next = rb_first(&p->root);
1964 while (next) {
1965 n = rb_entry(next, struct sp_node, nd);
1966 next = rb_next(&n->nd);
Andi Kleen90c50292005-07-27 11:43:50 -07001967 rb_erase(&n->nd, &p->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001968 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 kmem_cache_free(sn_cache, n);
1970 }
1971 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972}
1973
1974/* assumes fs == KERNEL_DS */
1975void __init numa_policy_init(void)
1976{
Paul Mundtb71636e2007-07-15 23:38:15 -07001977 nodemask_t interleave_nodes;
1978 unsigned long largest = 0;
1979 int nid, prefer = 0;
1980
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 policy_cache = kmem_cache_create("numa_policy",
1982 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09001983 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985 sn_cache = kmem_cache_create("shared_policy_node",
1986 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09001987 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Paul Mundtb71636e2007-07-15 23:38:15 -07001989 /*
1990 * Set interleaving policy for system init. Interleaving is only
1991 * enabled across suitably sized nodes (default is >= 16MB), or
1992 * fall back to the largest node if they're all smaller.
1993 */
1994 nodes_clear(interleave_nodes);
Christoph Lameter56bbd652007-10-16 01:25:35 -07001995 for_each_node_state(nid, N_HIGH_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07001996 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Paul Mundtb71636e2007-07-15 23:38:15 -07001998 /* Preserve the largest node */
1999 if (largest < total_pages) {
2000 largest = total_pages;
2001 prefer = nid;
2002 }
2003
2004 /* Interleave this node? */
2005 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2006 node_set(nid, interleave_nodes);
2007 }
2008
2009 /* All too small, use the largest */
2010 if (unlikely(nodes_empty(interleave_nodes)))
2011 node_set(prefer, interleave_nodes);
2012
David Rientjes028fec42008-04-28 02:12:25 -07002013 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 printk("numa_policy_init: interleaving failed\n");
2015}
2016
Christoph Lameter8bccd852005-10-29 18:16:59 -07002017/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018void numa_default_policy(void)
2019{
David Rientjes028fec42008-04-28 02:12:25 -07002020 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021}
Paul Jackson68860ec2005-10-30 15:02:36 -08002022
Paul Jackson42253992006-01-08 01:01:59 -08002023/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002024 * Parse and format mempolicy from/to strings
2025 */
2026
2027/*
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002028 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002029 * Used only for mpol_parse_str() and mpol_to_str()
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002030 */
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002031#define MPOL_LOCAL (MPOL_INTERLEAVE + 1)
Helge Deller15ad7cd2006-12-06 20:40:36 -08002032static const char * const policy_types[] =
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002033 { "default", "prefer", "bind", "interleave", "local" };
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002034
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002035
2036#ifdef CONFIG_TMPFS
2037/**
2038 * mpol_parse_str - parse string to mempolicy
2039 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002040 * @mpol: pointer to struct mempolicy pointer, returned on success.
2041 * @no_context: flag whether to "contextualize" the mempolicy
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002042 *
2043 * Format of input:
2044 * <mode>[=<flags>][:<nodelist>]
2045 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002046 * if @no_context is true, save the input nodemask in w.user_nodemask in
2047 * the returned mempolicy. This will be used to "clone" the mempolicy in
2048 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2049 * mount option. Note that if 'static' or 'relative' mode flags were
2050 * specified, the input nodemask will already have been saved. Saving
2051 * it again is redundant, but safe.
2052 *
2053 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002054 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002055int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002056{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002057 struct mempolicy *new = NULL;
2058 unsigned short uninitialized_var(mode);
2059 unsigned short uninitialized_var(mode_flags);
2060 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002061 char *nodelist = strchr(str, ':');
2062 char *flags = strchr(str, '=');
2063 int i;
2064 int err = 1;
2065
2066 if (nodelist) {
2067 /* NUL-terminate mode or flags string */
2068 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002069 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002070 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002071 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002072 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002073 } else
2074 nodes_clear(nodes);
2075
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002076 if (flags)
2077 *flags++ = '\0'; /* terminate mode string */
2078
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002079 for (i = 0; i <= MPOL_LOCAL; i++) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002080 if (!strcmp(str, policy_types[i])) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002081 mode = i;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002082 break;
2083 }
2084 }
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002085 if (i > MPOL_LOCAL)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002086 goto out;
2087
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002088 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002089 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002090 /*
2091 * Insist on a nodelist of one node only
2092 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002093 if (nodelist) {
2094 char *rest = nodelist;
2095 while (isdigit(*rest))
2096 rest++;
2097 if (!*rest)
2098 err = 0;
2099 }
2100 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002101 case MPOL_INTERLEAVE:
2102 /*
2103 * Default to online nodes with memory if no nodelist
2104 */
2105 if (!nodelist)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002106 nodes = node_states[N_HIGH_MEMORY];
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002107 err = 0;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002108 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002109 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002110 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002111 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002112 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002113 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002114 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002115 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002116 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002117
2118 /*
2119 * case MPOL_BIND: mpol_new() enforces non-empty nodemask.
2120 * case MPOL_DEFAULT: mpol_new() enforces empty nodemask, ignores flags.
2121 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002122 }
2123
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002124 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002125 if (flags) {
2126 /*
2127 * Currently, we only support two mutually exclusive
2128 * mode flags.
2129 */
2130 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002131 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002132 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002133 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002134 else
2135 err = 1;
2136 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002137
2138 new = mpol_new(mode, mode_flags, &nodes);
2139 if (IS_ERR(new))
2140 err = 1;
Miao Xie58568d22009-06-16 15:31:49 -07002141 else {
2142 int ret;
2143
2144 task_lock(current);
2145 ret = mpol_set_nodemask(new, &nodes);
2146 task_unlock(current);
2147 if (ret)
2148 err = 1;
2149 else if (no_context) {
2150 /* save for contextualization */
2151 new->w.user_nodemask = nodes;
2152 }
2153 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002154
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002155out:
2156 /* Restore string for error message */
2157 if (nodelist)
2158 *--nodelist = ':';
2159 if (flags)
2160 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002161 if (!err)
2162 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002163 return err;
2164}
2165#endif /* CONFIG_TMPFS */
2166
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002167/**
2168 * mpol_to_str - format a mempolicy structure for printing
2169 * @buffer: to contain formatted mempolicy string
2170 * @maxlen: length of @buffer
2171 * @pol: pointer to mempolicy to be formatted
2172 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2173 *
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002174 * Convert a mempolicy into a string.
2175 * Returns the number of characters in buffer (if positive)
2176 * or an error (negative)
2177 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002178int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002179{
2180 char *p = buffer;
2181 int l;
2182 nodemask_t nodes;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002183 unsigned short mode;
David Rientjesf5b087b2008-04-28 02:12:27 -07002184 unsigned short flags = pol ? pol->flags : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002185
Lee Schermerhorn22919902008-04-28 02:13:22 -07002186 /*
2187 * Sanity check: room for longest mode, flag and some nodes
2188 */
2189 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2190
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002191 if (!pol || pol == &default_policy)
2192 mode = MPOL_DEFAULT;
2193 else
2194 mode = pol->mode;
2195
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002196 switch (mode) {
2197 case MPOL_DEFAULT:
2198 nodes_clear(nodes);
2199 break;
2200
2201 case MPOL_PREFERRED:
2202 nodes_clear(nodes);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002203 if (flags & MPOL_F_LOCAL)
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002204 mode = MPOL_LOCAL; /* pseudo-policy */
2205 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002206 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002207 break;
2208
2209 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07002210 /* Fall through */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002211 case MPOL_INTERLEAVE:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002212 if (no_context)
2213 nodes = pol->w.user_nodemask;
2214 else
2215 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002216 break;
2217
2218 default:
2219 BUG();
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002220 }
2221
2222 l = strlen(policy_types[mode]);
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002223 if (buffer + maxlen < p + l + 1)
2224 return -ENOSPC;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002225
2226 strcpy(p, policy_types[mode]);
2227 p += l;
2228
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002229 if (flags & MPOL_MODE_FLAGS) {
David Rientjesf5b087b2008-04-28 02:12:27 -07002230 if (buffer + maxlen < p + 2)
2231 return -ENOSPC;
2232 *p++ = '=';
2233
Lee Schermerhorn22919902008-04-28 02:13:22 -07002234 /*
2235 * Currently, the only defined flags are mutually exclusive
2236 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002237 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002238 p += snprintf(p, buffer + maxlen - p, "static");
2239 else if (flags & MPOL_F_RELATIVE_NODES)
2240 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002241 }
2242
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002243 if (!nodes_empty(nodes)) {
2244 if (buffer + maxlen < p + 2)
2245 return -ENOSPC;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002246 *p++ = ':';
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002247 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2248 }
2249 return p - buffer;
2250}
2251
2252struct numa_maps {
2253 unsigned long pages;
2254 unsigned long anon;
Christoph Lameter397874d2006-03-06 15:42:53 -08002255 unsigned long active;
2256 unsigned long writeback;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002257 unsigned long mapcount_max;
Christoph Lameter397874d2006-03-06 15:42:53 -08002258 unsigned long dirty;
2259 unsigned long swapcache;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002260 unsigned long node[MAX_NUMNODES];
2261};
2262
Christoph Lameter397874d2006-03-06 15:42:53 -08002263static void gather_stats(struct page *page, void *private, int pte_dirty)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002264{
2265 struct numa_maps *md = private;
2266 int count = page_mapcount(page);
2267
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002268 md->pages++;
Christoph Lameter397874d2006-03-06 15:42:53 -08002269 if (pte_dirty || PageDirty(page))
2270 md->dirty++;
2271
2272 if (PageSwapCache(page))
2273 md->swapcache++;
2274
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002275 if (PageActive(page) || PageUnevictable(page))
Christoph Lameter397874d2006-03-06 15:42:53 -08002276 md->active++;
2277
2278 if (PageWriteback(page))
2279 md->writeback++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002280
2281 if (PageAnon(page))
2282 md->anon++;
2283
Christoph Lameter397874d2006-03-06 15:42:53 -08002284 if (count > md->mapcount_max)
2285 md->mapcount_max = count;
2286
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002287 md->node[page_to_nid(page)]++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002288}
2289
Andrew Morton7f709ed2006-03-07 21:55:22 -08002290#ifdef CONFIG_HUGETLB_PAGE
Christoph Lameter397874d2006-03-06 15:42:53 -08002291static void check_huge_range(struct vm_area_struct *vma,
2292 unsigned long start, unsigned long end,
2293 struct numa_maps *md)
2294{
2295 unsigned long addr;
2296 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -07002297 struct hstate *h = hstate_vma(vma);
2298 unsigned long sz = huge_page_size(h);
Christoph Lameter397874d2006-03-06 15:42:53 -08002299
Andi Kleena5516432008-07-23 21:27:41 -07002300 for (addr = start; addr < end; addr += sz) {
2301 pte_t *ptep = huge_pte_offset(vma->vm_mm,
2302 addr & huge_page_mask(h));
Christoph Lameter397874d2006-03-06 15:42:53 -08002303 pte_t pte;
2304
2305 if (!ptep)
2306 continue;
2307
2308 pte = *ptep;
2309 if (pte_none(pte))
2310 continue;
2311
2312 page = pte_page(pte);
2313 if (!page)
2314 continue;
2315
2316 gather_stats(page, md, pte_dirty(*ptep));
2317 }
2318}
Andrew Morton7f709ed2006-03-07 21:55:22 -08002319#else
2320static inline void check_huge_range(struct vm_area_struct *vma,
2321 unsigned long start, unsigned long end,
2322 struct numa_maps *md)
2323{
2324}
2325#endif
Christoph Lameter397874d2006-03-06 15:42:53 -08002326
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002327/*
2328 * Display pages allocated per node and memory policy via /proc.
2329 */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002330int show_numa_map(struct seq_file *m, void *v)
2331{
Eric W. Biederman99f89552006-06-26 00:25:55 -07002332 struct proc_maps_private *priv = m->private;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002333 struct vm_area_struct *vma = v;
2334 struct numa_maps *md;
Christoph Lameter397874d2006-03-06 15:42:53 -08002335 struct file *file = vma->vm_file;
2336 struct mm_struct *mm = vma->vm_mm;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002337 struct mempolicy *pol;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002338 int n;
2339 char buffer[50];
2340
Christoph Lameter397874d2006-03-06 15:42:53 -08002341 if (!mm)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002342 return 0;
2343
2344 md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL);
2345 if (!md)
2346 return 0;
2347
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002348 pol = get_vma_policy(priv->task, vma, vma->vm_start);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002349 mpol_to_str(buffer, sizeof(buffer), pol, 0);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002350 mpol_cond_put(pol);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002351
Christoph Lameter397874d2006-03-06 15:42:53 -08002352 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002353
Christoph Lameter397874d2006-03-06 15:42:53 -08002354 if (file) {
2355 seq_printf(m, " file=");
Jan Blunckc32c2f62008-02-14 19:38:43 -08002356 seq_path(m, &file->f_path, "\n\t= ");
Christoph Lameter397874d2006-03-06 15:42:53 -08002357 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
2358 seq_printf(m, " heap");
2359 } else if (vma->vm_start <= mm->start_stack &&
2360 vma->vm_end >= mm->start_stack) {
2361 seq_printf(m, " stack");
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002362 }
Christoph Lameter397874d2006-03-06 15:42:53 -08002363
2364 if (is_vm_hugetlb_page(vma)) {
2365 check_huge_range(vma, vma->vm_start, vma->vm_end, md);
2366 seq_printf(m, " huge");
2367 } else {
2368 check_pgd_range(vma, vma->vm_start, vma->vm_end,
Christoph Lameter56bbd652007-10-16 01:25:35 -07002369 &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
Christoph Lameter397874d2006-03-06 15:42:53 -08002370 }
2371
2372 if (!md->pages)
2373 goto out;
2374
2375 if (md->anon)
2376 seq_printf(m," anon=%lu",md->anon);
2377
2378 if (md->dirty)
2379 seq_printf(m," dirty=%lu",md->dirty);
2380
2381 if (md->pages != md->anon && md->pages != md->dirty)
2382 seq_printf(m, " mapped=%lu", md->pages);
2383
2384 if (md->mapcount_max > 1)
2385 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2386
2387 if (md->swapcache)
2388 seq_printf(m," swapcache=%lu", md->swapcache);
2389
2390 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2391 seq_printf(m," active=%lu", md->active);
2392
2393 if (md->writeback)
2394 seq_printf(m," writeback=%lu", md->writeback);
2395
Christoph Lameter56bbd652007-10-16 01:25:35 -07002396 for_each_node_state(n, N_HIGH_MEMORY)
Christoph Lameter397874d2006-03-06 15:42:53 -08002397 if (md->node[n])
2398 seq_printf(m, " N%d=%lu", n, md->node[n]);
2399out:
2400 seq_putc(m, '\n');
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002401 kfree(md);
2402
2403 if (m->count < m->size)
Eric W. Biederman99f89552006-06-26 00:25:55 -07002404 m->version = (vma != priv->tail_vma) ? vma->vm_start : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002405 return 0;
2406}