blob: 36f42573a3352fa6ba452eea2d76ed2021c8605c [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/* Create a new policy */
David Rientjes028fec42008-04-28 02:12:25 -0700186static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
187 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 struct mempolicy *policy;
David Rientjesf5b087b2008-04-28 02:12:27 -0700190 nodemask_t cpuset_context_nmask;
David Rientjes37012942008-04-28 02:12:33 -0700191 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
David Rientjes028fec42008-04-28 02:12:25 -0700193 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
194 mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
Paul Mundt140d5a42007-07-15 23:38:16 -0700195
David Rientjes3e1f0642008-04-28 02:12:34 -0700196 if (mode == MPOL_DEFAULT) {
197 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700198 return ERR_PTR(-EINVAL);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700199 return NULL; /* simply delete any existing policy */
David Rientjes37012942008-04-28 02:12:33 -0700200 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700201 VM_BUG_ON(!nodes);
202
203 /*
204 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
205 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
206 * All other modes require a valid pointer to a non-empty nodemask.
207 */
208 if (mode == MPOL_PREFERRED) {
209 if (nodes_empty(*nodes)) {
210 if (((flags & MPOL_F_STATIC_NODES) ||
211 (flags & MPOL_F_RELATIVE_NODES)))
212 return ERR_PTR(-EINVAL);
213 nodes = NULL; /* flag local alloc */
214 }
215 } else if (nodes_empty(*nodes))
216 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
218 if (!policy)
219 return ERR_PTR(-ENOMEM);
220 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700221 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700222 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700223
David Rientjes3e1f0642008-04-28 02:12:34 -0700224 if (nodes) {
225 /*
226 * cpuset related setup doesn't apply to local allocation
227 */
David Rientjes37012942008-04-28 02:12:33 -0700228 cpuset_update_task_memory_state();
229 if (flags & MPOL_F_RELATIVE_NODES)
230 mpol_relative_nodemask(&cpuset_context_nmask, nodes,
231 &cpuset_current_mems_allowed);
232 else
233 nodes_and(cpuset_context_nmask, *nodes,
234 cpuset_current_mems_allowed);
235 if (mpol_store_user_nodemask(policy))
236 policy->w.user_nodemask = *nodes;
237 else
238 policy->w.cpuset_mems_allowed =
239 cpuset_mems_allowed(current);
240 }
241
242 ret = mpol_ops[mode].create(policy,
David Rientjes3e1f0642008-04-28 02:12:34 -0700243 nodes ? &cpuset_context_nmask : NULL);
David Rientjes37012942008-04-28 02:12:33 -0700244 if (ret < 0) {
245 kmem_cache_free(policy_cache, policy);
246 return ERR_PTR(ret);
247 }
248 return policy;
249}
250
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700251/* Slow path of a mpol destructor. */
252void __mpol_put(struct mempolicy *p)
253{
254 if (!atomic_dec_and_test(&p->refcnt))
255 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700256 kmem_cache_free(policy_cache, p);
257}
258
David Rientjes37012942008-04-28 02:12:33 -0700259static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
260{
261}
262
263static void mpol_rebind_nodemask(struct mempolicy *pol,
264 const nodemask_t *nodes)
265{
266 nodemask_t tmp;
267
268 if (pol->flags & MPOL_F_STATIC_NODES)
269 nodes_and(tmp, pol->w.user_nodemask, *nodes);
270 else if (pol->flags & MPOL_F_RELATIVE_NODES)
271 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
272 else {
273 nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed,
274 *nodes);
275 pol->w.cpuset_mems_allowed = *nodes;
276 }
277
278 pol->v.nodes = tmp;
279 if (!node_isset(current->il_next, tmp)) {
280 current->il_next = next_node(current->il_next, tmp);
281 if (current->il_next >= MAX_NUMNODES)
282 current->il_next = first_node(tmp);
283 if (current->il_next >= MAX_NUMNODES)
284 current->il_next = numa_node_id();
285 }
286}
287
288static void mpol_rebind_preferred(struct mempolicy *pol,
289 const nodemask_t *nodes)
290{
291 nodemask_t tmp;
292
David Rientjes37012942008-04-28 02:12:33 -0700293 if (pol->flags & MPOL_F_STATIC_NODES) {
294 int node = first_node(pol->w.user_nodemask);
295
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700296 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700297 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700298 pol->flags &= ~MPOL_F_LOCAL;
299 } else
300 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700301 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
302 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
303 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700304 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700305 pol->v.preferred_node = node_remap(pol->v.preferred_node,
306 pol->w.cpuset_mems_allowed,
307 *nodes);
308 pol->w.cpuset_mems_allowed = *nodes;
309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
David Rientjes1d0d2682008-04-28 02:12:32 -0700312/* Migrate a policy to a different set of nodes */
313static void mpol_rebind_policy(struct mempolicy *pol,
314 const nodemask_t *newmask)
315{
David Rientjes1d0d2682008-04-28 02:12:32 -0700316 if (!pol)
317 return;
David Rientjes1d0d2682008-04-28 02:12:32 -0700318 if (!mpol_store_user_nodemask(pol) &&
319 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
320 return;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700321 mpol_ops[pol->mode].rebind(pol, newmask);
David Rientjes1d0d2682008-04-28 02:12:32 -0700322}
323
324/*
325 * Wrapper for mpol_rebind_policy() that just requires task
326 * pointer, and updates task mempolicy.
327 */
328
329void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
330{
331 mpol_rebind_policy(tsk->mempolicy, new);
332}
333
334/*
335 * Rebind each vma in mm to new nodemask.
336 *
337 * Call holding a reference to mm. Takes mm->mmap_sem during call.
338 */
339
340void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
341{
342 struct vm_area_struct *vma;
343
344 down_write(&mm->mmap_sem);
345 for (vma = mm->mmap; vma; vma = vma->vm_next)
346 mpol_rebind_policy(vma->vm_policy, new);
347 up_write(&mm->mmap_sem);
348}
349
David Rientjes37012942008-04-28 02:12:33 -0700350static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
351 [MPOL_DEFAULT] = {
352 .rebind = mpol_rebind_default,
353 },
354 [MPOL_INTERLEAVE] = {
355 .create = mpol_new_interleave,
356 .rebind = mpol_rebind_nodemask,
357 },
358 [MPOL_PREFERRED] = {
359 .create = mpol_new_preferred,
360 .rebind = mpol_rebind_preferred,
361 },
362 [MPOL_BIND] = {
363 .create = mpol_new_bind,
364 .rebind = mpol_rebind_nodemask,
365 },
366};
367
Christoph Lameter397874d2006-03-06 15:42:53 -0800368static void gather_stats(struct page *, void *, int pte_dirty);
Christoph Lameterfc301282006-01-18 17:42:29 -0800369static void migrate_page_add(struct page *page, struct list_head *pagelist,
370 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800371
Christoph Lameter38e35862006-01-08 01:01:01 -0800372/* Scan through pages checking if pages follow certain conditions. */
Nick Pigginb5810032005-10-29 18:16:12 -0700373static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800374 unsigned long addr, unsigned long end,
375 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800376 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Hugh Dickins91612e02005-06-21 17:15:07 -0700378 pte_t *orig_pte;
379 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700380 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700381
Hugh Dickins705e87c2005-10-29 18:16:27 -0700382 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700383 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800384 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800385 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700386
387 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800389 page = vm_normal_page(vma, addr, *pte);
390 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800392 /*
393 * The check for PageReserved here is important to avoid
394 * handling zero pages and other pages that may have been
395 * marked special by the system.
396 *
397 * If the PageReserved would not be checked here then f.e.
398 * the location of the zero page could have an influence
399 * on MPOL_MF_STRICT, zero pages would be counted for
400 * the per node stats, and there would be useless attempts
401 * to put zero pages on the migration list.
402 */
Christoph Lameterf4598c82006-01-12 01:05:20 -0800403 if (PageReserved(page))
404 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800405 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800406 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
407 continue;
408
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800409 if (flags & MPOL_MF_STATS)
Christoph Lameter397874d2006-03-06 15:42:53 -0800410 gather_stats(page, private, pte_dirty(*pte));
Nick Piggin053837f2006-01-18 17:42:27 -0800411 else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800412 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800413 else
414 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700415 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700416 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700417 return addr != end;
418}
419
Nick Pigginb5810032005-10-29 18:16:12 -0700420static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800421 unsigned long addr, unsigned long end,
422 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800423 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700424{
425 pmd_t *pmd;
426 unsigned long next;
427
428 pmd = pmd_offset(pud, addr);
429 do {
430 next = pmd_addr_end(addr, end);
431 if (pmd_none_or_clear_bad(pmd))
432 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800433 if (check_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800434 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700435 return -EIO;
436 } while (pmd++, addr = next, addr != end);
437 return 0;
438}
439
Nick Pigginb5810032005-10-29 18:16:12 -0700440static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800441 unsigned long addr, unsigned long end,
442 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800443 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700444{
445 pud_t *pud;
446 unsigned long next;
447
448 pud = pud_offset(pgd, addr);
449 do {
450 next = pud_addr_end(addr, end);
451 if (pud_none_or_clear_bad(pud))
452 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800453 if (check_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800454 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700455 return -EIO;
456 } while (pud++, addr = next, addr != end);
457 return 0;
458}
459
Nick Pigginb5810032005-10-29 18:16:12 -0700460static inline int check_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800461 unsigned long addr, unsigned long end,
462 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800463 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700464{
465 pgd_t *pgd;
466 unsigned long next;
467
Nick Pigginb5810032005-10-29 18:16:12 -0700468 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700469 do {
470 next = pgd_addr_end(addr, end);
471 if (pgd_none_or_clear_bad(pgd))
472 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800473 if (check_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800474 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700475 return -EIO;
476 } while (pgd++, addr = next, addr != end);
477 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478}
479
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800480/*
481 * Check if all pages in a range are on a set of nodes.
482 * If pagelist != NULL then isolate pages from the LRU and
483 * put them on the pagelist.
484 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485static struct vm_area_struct *
486check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800487 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
489 int err;
490 struct vm_area_struct *first, *vma, *prev;
491
Christoph Lameter90036ee2006-03-16 23:03:59 -0800492 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
Christoph Lameter90036ee2006-03-16 23:03:59 -0800493
Christoph Lameterb20a3502006-03-22 00:09:12 -0800494 err = migrate_prep();
495 if (err)
496 return ERR_PTR(err);
Christoph Lameter90036ee2006-03-16 23:03:59 -0800497 }
Nick Piggin053837f2006-01-18 17:42:27 -0800498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 first = find_vma(mm, start);
500 if (!first)
501 return ERR_PTR(-EFAULT);
502 prev = NULL;
503 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800504 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
505 if (!vma->vm_next && vma->vm_end < end)
506 return ERR_PTR(-EFAULT);
507 if (prev && prev->vm_end < vma->vm_start)
508 return ERR_PTR(-EFAULT);
509 }
510 if (!is_vm_hugetlb_page(vma) &&
511 ((flags & MPOL_MF_STRICT) ||
512 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
513 vma_migratable(vma)))) {
Andi Kleen5b952b32005-09-13 01:25:08 -0700514 unsigned long endvma = vma->vm_end;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800515
Andi Kleen5b952b32005-09-13 01:25:08 -0700516 if (endvma > end)
517 endvma = end;
518 if (vma->vm_start > start)
519 start = vma->vm_start;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800520 err = check_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800521 flags, private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (err) {
523 first = ERR_PTR(err);
524 break;
525 }
526 }
527 prev = vma;
528 }
529 return first;
530}
531
532/* Apply policy to a single VMA */
533static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new)
534{
535 int err = 0;
536 struct mempolicy *old = vma->vm_policy;
537
Paul Mundt140d5a42007-07-15 23:38:16 -0700538 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 vma->vm_start, vma->vm_end, vma->vm_pgoff,
540 vma->vm_ops, vma->vm_file,
541 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
542
543 if (vma->vm_ops && vma->vm_ops->set_policy)
544 err = vma->vm_ops->set_policy(vma, new);
545 if (!err) {
546 mpol_get(new);
547 vma->vm_policy = new;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700548 mpol_put(old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550 return err;
551}
552
553/* Step 2: apply policy to a range and do splits. */
554static int mbind_range(struct vm_area_struct *vma, unsigned long start,
555 unsigned long end, struct mempolicy *new)
556{
557 struct vm_area_struct *next;
558 int err;
559
560 err = 0;
561 for (; vma && vma->vm_start < end; vma = next) {
562 next = vma->vm_next;
563 if (vma->vm_start < start)
564 err = split_vma(vma->vm_mm, vma, start, 1);
565 if (!err && vma->vm_end > end)
566 err = split_vma(vma->vm_mm, vma, end, 0);
567 if (!err)
568 err = policy_vma(vma, new);
569 if (err)
570 break;
571 }
572 return err;
573}
574
Paul Jacksonc61afb12006-03-24 03:16:08 -0800575/*
576 * Update task->flags PF_MEMPOLICY bit: set iff non-default
577 * mempolicy. Allows more rapid checking of this (combined perhaps
578 * with other PF_* flag bits) on memory allocation hot code paths.
579 *
580 * If called from outside this file, the task 'p' should -only- be
581 * a newly forked child not yet visible on the task list, because
582 * manipulating the task flags of a visible task is not safe.
583 *
584 * The above limitation is why this routine has the funny name
585 * mpol_fix_fork_child_flag().
586 *
587 * It is also safe to call this with a task pointer of current,
588 * which the static wrapper mpol_set_task_struct_flag() does,
589 * for use within this file.
590 */
591
592void mpol_fix_fork_child_flag(struct task_struct *p)
593{
594 if (p->mempolicy)
595 p->flags |= PF_MEMPOLICY;
596 else
597 p->flags &= ~PF_MEMPOLICY;
598}
599
600static void mpol_set_task_struct_flag(void)
601{
602 mpol_fix_fork_child_flag(current);
603}
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700606static long do_set_mempolicy(unsigned short mode, unsigned short flags,
607 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 struct mempolicy *new;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700610 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
David Rientjes028fec42008-04-28 02:12:25 -0700612 new = mpol_new(mode, flags, nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (IS_ERR(new))
614 return PTR_ERR(new);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700615
616 /*
617 * prevent changing our mempolicy while show_numa_maps()
618 * is using it.
619 * Note: do_set_mempolicy() can be called at init time
620 * with no 'mm'.
621 */
622 if (mm)
623 down_write(&mm->mmap_sem);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700624 mpol_put(current->mempolicy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 current->mempolicy = new;
Paul Jacksonc61afb12006-03-24 03:16:08 -0800626 mpol_set_task_struct_flag();
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700627 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700628 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700629 current->il_next = first_node(new->v.nodes);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700630 if (mm)
631 up_write(&mm->mmap_sem);
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return 0;
634}
635
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700636/*
637 * Return nodemask for policy for get_mempolicy() query
638 */
639static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700641 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700642 if (p == &default_policy)
643 return;
644
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700645 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700646 case MPOL_BIND:
647 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700649 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 break;
651 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700652 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700653 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700654 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 break;
656 default:
657 BUG();
658 }
659}
660
661static int lookup_node(struct mm_struct *mm, unsigned long addr)
662{
663 struct page *p;
664 int err;
665
666 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
667 if (err >= 0) {
668 err = page_to_nid(p);
669 put_page(p);
670 }
671 return err;
672}
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700675static long do_get_mempolicy(int *policy, nodemask_t *nmask,
676 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700678 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 struct mm_struct *mm = current->mm;
680 struct vm_area_struct *vma = NULL;
681 struct mempolicy *pol = current->mempolicy;
682
Paul Jacksoncf2a473c2006-01-08 01:01:54 -0800683 cpuset_update_task_memory_state();
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700684 if (flags &
685 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700687
688 if (flags & MPOL_F_MEMS_ALLOWED) {
689 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
690 return -EINVAL;
691 *policy = 0; /* just so it's initialized */
692 *nmask = cpuset_current_mems_allowed;
693 return 0;
694 }
695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700697 /*
698 * Do NOT fall back to task policy if the
699 * vma/shared policy at addr is NULL. We
700 * want to return MPOL_DEFAULT in this case.
701 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 down_read(&mm->mmap_sem);
703 vma = find_vma_intersection(mm, addr, addr+1);
704 if (!vma) {
705 up_read(&mm->mmap_sem);
706 return -EFAULT;
707 }
708 if (vma->vm_ops && vma->vm_ops->get_policy)
709 pol = vma->vm_ops->get_policy(vma, addr);
710 else
711 pol = vma->vm_policy;
712 } else if (addr)
713 return -EINVAL;
714
715 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700716 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 if (flags & MPOL_F_NODE) {
719 if (flags & MPOL_F_ADDR) {
720 err = lookup_node(mm, addr);
721 if (err < 0)
722 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700723 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700725 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700726 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 } else {
728 err = -EINVAL;
729 goto out;
730 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700731 } else {
732 *policy = pol == &default_policy ? MPOL_DEFAULT :
733 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700734 /*
735 * Internal mempolicy flags must be masked off before exposing
736 * the policy to userspace.
737 */
738 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741 if (vma) {
742 up_read(&current->mm->mmap_sem);
743 vma = NULL;
744 }
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 err = 0;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700747 if (nmask)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700748 get_policy_nodemask(pol, nmask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700751 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 if (vma)
753 up_read(&current->mm->mmap_sem);
754 return err;
755}
756
Christoph Lameterb20a3502006-03-22 00:09:12 -0800757#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700758/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800759 * page migration
760 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800761static void migrate_page_add(struct page *page, struct list_head *pagelist,
762 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800763{
764 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800765 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800766 */
Nick Piggin62695a82008-10-18 20:26:09 -0700767 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
768 if (!isolate_lru_page(page)) {
769 list_add_tail(&page->lru, pagelist);
770 }
771 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800772}
773
Christoph Lameter742755a2006-06-23 02:03:55 -0700774static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700775{
Mel Gorman769848c2007-07-17 04:03:05 -0700776 return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700777}
778
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800779/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800780 * Migrate pages from one node to a target node.
781 * Returns error or the number of pages not migrated.
782 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700783static int migrate_to_node(struct mm_struct *mm, int source, int dest,
784 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800785{
786 nodemask_t nmask;
787 LIST_HEAD(pagelist);
788 int err = 0;
789
790 nodes_clear(nmask);
791 node_set(source, nmask);
792
793 check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask,
794 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
795
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700796 if (!list_empty(&pagelist))
Christoph Lameter95a402c2006-06-23 02:03:53 -0700797 err = migrate_pages(&pagelist, new_node_page, dest);
798
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800799 return err;
800}
801
802/*
803 * Move pages between the two nodesets so as to preserve the physical
804 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -0800805 *
806 * Returns the number of page that could not be moved.
807 */
808int do_migrate_pages(struct mm_struct *mm,
809 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
810{
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800811 int busy = 0;
812 int err = 0;
813 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -0800814
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700815 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -0800816
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700817 err = migrate_vmas(mm, from_nodes, to_nodes, flags);
818 if (err)
819 goto out;
820
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800821/*
822 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
823 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
824 * bit in 'tmp', and return that <source, dest> pair for migration.
825 * The pair of nodemasks 'to' and 'from' define the map.
826 *
827 * If no pair of bits is found that way, fallback to picking some
828 * pair of 'source' and 'dest' bits that are not the same. If the
829 * 'source' and 'dest' bits are the same, this represents a node
830 * that will be migrating to itself, so no pages need move.
831 *
832 * If no bits are left in 'tmp', or if all remaining bits left
833 * in 'tmp' correspond to the same bit in 'to', return false
834 * (nothing left to migrate).
835 *
836 * This lets us pick a pair of nodes to migrate between, such that
837 * if possible the dest node is not already occupied by some other
838 * source node, minimizing the risk of overloading the memory on a
839 * node that would happen if we migrated incoming memory to a node
840 * before migrating outgoing memory source that same node.
841 *
842 * A single scan of tmp is sufficient. As we go, we remember the
843 * most recent <s, d> pair that moved (s != d). If we find a pair
844 * that not only moved, but what's better, moved to an empty slot
845 * (d is not set in tmp), then we break out then, with that pair.
846 * Otherwise when we finish scannng from_tmp, we at least have the
847 * most recent <s, d> pair that moved. If we get all the way through
848 * the scan of tmp without finding any node that moved, much less
849 * moved to an empty node, then there is nothing left worth migrating.
850 */
Christoph Lameterd4984712006-01-08 01:00:55 -0800851
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800852 tmp = *from_nodes;
853 while (!nodes_empty(tmp)) {
854 int s,d;
855 int source = -1;
856 int dest = 0;
857
858 for_each_node_mask(s, tmp) {
859 d = node_remap(s, *from_nodes, *to_nodes);
860 if (s == d)
861 continue;
862
863 source = s; /* Node moved. Memorize */
864 dest = d;
865
866 /* dest not in remaining from nodes? */
867 if (!node_isset(dest, tmp))
868 break;
869 }
870 if (source == -1)
871 break;
872
873 node_clear(source, tmp);
874 err = migrate_to_node(mm, source, dest, flags);
875 if (err > 0)
876 busy += err;
877 if (err < 0)
878 break;
Christoph Lameter39743882006-01-08 01:00:51 -0800879 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700880out:
Christoph Lameter39743882006-01-08 01:00:51 -0800881 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800882 if (err < 0)
883 return err;
884 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800885
Christoph Lameter39743882006-01-08 01:00:51 -0800886}
887
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800888/*
889 * Allocate a new page for page migration based on vma policy.
890 * Start assuming that page is mapped by vma pointed to by @private.
891 * Search forward from there, if not. N.B., this assumes that the
892 * list of pages handed to migrate_pages()--which is how we get here--
893 * is in virtual address order.
894 */
Christoph Lameter742755a2006-06-23 02:03:55 -0700895static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700896{
897 struct vm_area_struct *vma = (struct vm_area_struct *)private;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800898 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700899
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800900 while (vma) {
901 address = page_address_in_vma(page, vma);
902 if (address != -EFAULT)
903 break;
904 vma = vma->vm_next;
905 }
906
907 /*
908 * if !vma, alloc_page_vma() will use task or system default policy
909 */
910 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700911}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800912#else
913
914static void migrate_page_add(struct page *page, struct list_head *pagelist,
915 unsigned long flags)
916{
917}
918
919int do_migrate_pages(struct mm_struct *mm,
920 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
921{
922 return -ENOSYS;
923}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700924
Keith Owens69939742006-10-11 01:21:28 -0700925static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700926{
927 return NULL;
928}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800929#endif
930
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700931static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -0700932 unsigned short mode, unsigned short mode_flags,
933 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800934{
935 struct vm_area_struct *vma;
936 struct mm_struct *mm = current->mm;
937 struct mempolicy *new;
938 unsigned long end;
939 int err;
940 LIST_HEAD(pagelist);
941
David Rientjesa3b51e02008-04-28 02:12:23 -0700942 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
943 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800944 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -0800945 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800946 return -EPERM;
947
948 if (start & ~PAGE_MASK)
949 return -EINVAL;
950
951 if (mode == MPOL_DEFAULT)
952 flags &= ~MPOL_MF_STRICT;
953
954 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
955 end = start + len;
956
957 if (end < start)
958 return -EINVAL;
959 if (end == start)
960 return 0;
961
David Rientjes028fec42008-04-28 02:12:25 -0700962 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800963 if (IS_ERR(new))
964 return PTR_ERR(new);
965
966 /*
967 * If we are using the default policy then operation
968 * on discontinuous address spaces is okay after all
969 */
970 if (!new)
971 flags |= MPOL_MF_DISCONTIG_OK;
972
David Rientjes028fec42008-04-28 02:12:25 -0700973 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
974 start, start + len, mode, mode_flags,
975 nmask ? nodes_addr(*nmask)[0] : -1);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800976
977 down_write(&mm->mmap_sem);
978 vma = check_range(mm, start, end, nmask,
979 flags | MPOL_MF_INVERT, &pagelist);
980
981 err = PTR_ERR(vma);
982 if (!IS_ERR(vma)) {
983 int nr_failed = 0;
984
985 err = mbind_range(vma, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800986
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800987 if (!list_empty(&pagelist))
Christoph Lameter95a402c2006-06-23 02:03:53 -0700988 nr_failed = migrate_pages(&pagelist, new_vma_page,
989 (unsigned long)vma);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800990
991 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
992 err = -EIO;
993 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800994
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800995 up_write(&mm->mmap_sem);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700996 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800997 return err;
998}
999
Christoph Lameter39743882006-01-08 01:00:51 -08001000/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001001 * User space interface with variable sized bitmaps for nodelists.
1002 */
1003
1004/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001005static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001006 unsigned long maxnode)
1007{
1008 unsigned long k;
1009 unsigned long nlongs;
1010 unsigned long endmask;
1011
1012 --maxnode;
1013 nodes_clear(*nodes);
1014 if (maxnode == 0 || !nmask)
1015 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001016 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001017 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001018
1019 nlongs = BITS_TO_LONGS(maxnode);
1020 if ((maxnode % BITS_PER_LONG) == 0)
1021 endmask = ~0UL;
1022 else
1023 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1024
1025 /* When the user specified more nodes than supported just check
1026 if the non supported part is all zero. */
1027 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1028 if (nlongs > PAGE_SIZE/sizeof(long))
1029 return -EINVAL;
1030 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1031 unsigned long t;
1032 if (get_user(t, nmask + k))
1033 return -EFAULT;
1034 if (k == nlongs - 1) {
1035 if (t & endmask)
1036 return -EINVAL;
1037 } else if (t)
1038 return -EINVAL;
1039 }
1040 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1041 endmask = ~0UL;
1042 }
1043
1044 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1045 return -EFAULT;
1046 nodes_addr(*nodes)[nlongs-1] &= endmask;
1047 return 0;
1048}
1049
1050/* Copy a kernel node mask to user space */
1051static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1052 nodemask_t *nodes)
1053{
1054 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1055 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1056
1057 if (copy > nbytes) {
1058 if (copy > PAGE_SIZE)
1059 return -EINVAL;
1060 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1061 return -EFAULT;
1062 copy = nbytes;
1063 }
1064 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1065}
1066
1067asmlinkage long sys_mbind(unsigned long start, unsigned long len,
1068 unsigned long mode,
1069 unsigned long __user *nmask, unsigned long maxnode,
1070 unsigned flags)
1071{
1072 nodemask_t nodes;
1073 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001074 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001075
David Rientjes028fec42008-04-28 02:12:25 -07001076 mode_flags = mode & MPOL_MODE_FLAGS;
1077 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001078 if (mode >= MPOL_MAX)
1079 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001080 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1081 (mode_flags & MPOL_F_RELATIVE_NODES))
1082 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001083 err = get_nodes(&nodes, nmask, maxnode);
1084 if (err)
1085 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001086 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001087}
1088
1089/* Set the process memory policy */
1090asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask,
1091 unsigned long maxnode)
1092{
1093 int err;
1094 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001095 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001096
David Rientjes028fec42008-04-28 02:12:25 -07001097 flags = mode & MPOL_MODE_FLAGS;
1098 mode &= ~MPOL_MODE_FLAGS;
1099 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001100 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001101 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1102 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001103 err = get_nodes(&nodes, nmask, maxnode);
1104 if (err)
1105 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001106 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001107}
1108
Christoph Lameter39743882006-01-08 01:00:51 -08001109asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode,
1110 const unsigned long __user *old_nodes,
1111 const unsigned long __user *new_nodes)
1112{
1113 struct mm_struct *mm;
1114 struct task_struct *task;
1115 nodemask_t old;
1116 nodemask_t new;
1117 nodemask_t task_nodes;
1118 int err;
1119
1120 err = get_nodes(&old, old_nodes, maxnode);
1121 if (err)
1122 return err;
1123
1124 err = get_nodes(&new, new_nodes, maxnode);
1125 if (err)
1126 return err;
1127
1128 /* Find the mm_struct */
1129 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001130 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001131 if (!task) {
1132 read_unlock(&tasklist_lock);
1133 return -ESRCH;
1134 }
1135 mm = get_task_mm(task);
1136 read_unlock(&tasklist_lock);
1137
1138 if (!mm)
1139 return -EINVAL;
1140
1141 /*
1142 * Check if this process has the right to modify the specified
1143 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001144 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001145 * userid as the target process.
1146 */
1147 if ((current->euid != task->suid) && (current->euid != task->uid) &&
1148 (current->uid != task->suid) && (current->uid != task->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001149 !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001150 err = -EPERM;
1151 goto out;
1152 }
1153
1154 task_nodes = cpuset_mems_allowed(task);
1155 /* Is the user allowed to access the target nodes? */
Christoph Lameter74c00242006-03-14 19:50:21 -08001156 if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001157 err = -EPERM;
1158 goto out;
1159 }
1160
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001161 if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001162 err = -EINVAL;
1163 goto out;
1164 }
1165
David Quigley86c3a762006-06-23 02:04:02 -07001166 err = security_task_movememory(task);
1167 if (err)
1168 goto out;
1169
Christoph Lameter511030b2006-02-28 16:58:57 -08001170 err = do_migrate_pages(mm, &old, &new,
Christoph Lameter74c00242006-03-14 19:50:21 -08001171 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter39743882006-01-08 01:00:51 -08001172out:
1173 mmput(mm);
1174 return err;
1175}
1176
1177
Christoph Lameter8bccd852005-10-29 18:16:59 -07001178/* Retrieve NUMA policy */
1179asmlinkage long sys_get_mempolicy(int __user *policy,
1180 unsigned long __user *nmask,
1181 unsigned long maxnode,
1182 unsigned long addr, unsigned long flags)
1183{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001184 int err;
1185 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001186 nodemask_t nodes;
1187
1188 if (nmask != NULL && maxnode < MAX_NUMNODES)
1189 return -EINVAL;
1190
1191 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1192
1193 if (err)
1194 return err;
1195
1196 if (policy && put_user(pval, policy))
1197 return -EFAULT;
1198
1199 if (nmask)
1200 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1201
1202 return err;
1203}
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205#ifdef CONFIG_COMPAT
1206
1207asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1208 compat_ulong_t __user *nmask,
1209 compat_ulong_t maxnode,
1210 compat_ulong_t addr, compat_ulong_t flags)
1211{
1212 long err;
1213 unsigned long __user *nm = NULL;
1214 unsigned long nr_bits, alloc_size;
1215 DECLARE_BITMAP(bm, MAX_NUMNODES);
1216
1217 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1218 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1219
1220 if (nmask)
1221 nm = compat_alloc_user_space(alloc_size);
1222
1223 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1224
1225 if (!err && nmask) {
1226 err = copy_from_user(bm, nm, alloc_size);
1227 /* ensure entire bitmap is zeroed */
1228 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1229 err |= compat_put_bitmap(nmask, bm, nr_bits);
1230 }
1231
1232 return err;
1233}
1234
1235asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1236 compat_ulong_t maxnode)
1237{
1238 long err = 0;
1239 unsigned long __user *nm = NULL;
1240 unsigned long nr_bits, alloc_size;
1241 DECLARE_BITMAP(bm, MAX_NUMNODES);
1242
1243 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1244 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1245
1246 if (nmask) {
1247 err = compat_get_bitmap(bm, nmask, nr_bits);
1248 nm = compat_alloc_user_space(alloc_size);
1249 err |= copy_to_user(nm, bm, alloc_size);
1250 }
1251
1252 if (err)
1253 return -EFAULT;
1254
1255 return sys_set_mempolicy(mode, nm, nr_bits+1);
1256}
1257
1258asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1259 compat_ulong_t mode, compat_ulong_t __user *nmask,
1260 compat_ulong_t maxnode, compat_ulong_t flags)
1261{
1262 long err = 0;
1263 unsigned long __user *nm = NULL;
1264 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001265 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1268 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1269
1270 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001271 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001273 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
1275
1276 if (err)
1277 return -EFAULT;
1278
1279 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1280}
1281
1282#endif
1283
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001284/*
1285 * get_vma_policy(@task, @vma, @addr)
1286 * @task - task for fallback if vma policy == default
1287 * @vma - virtual memory area whose policy is sought
1288 * @addr - address in @vma for shared policy lookup
1289 *
1290 * Returns effective policy for a VMA at specified address.
1291 * Falls back to @task or system default policy, as necessary.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001292 * Current or other task's task mempolicy and non-shared vma policies
1293 * are protected by the task's mmap_sem, which must be held for read by
1294 * the caller.
1295 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1296 * count--added by the get_policy() vm_op, as appropriate--to protect against
1297 * freeing by another task. It is the caller's responsibility to free the
1298 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001299 */
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001300static struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001301 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001303 struct mempolicy *pol = task->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
1305 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001306 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001307 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1308 addr);
1309 if (vpol)
1310 pol = vpol;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001311 } else if (vma->vm_policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 pol = vma->vm_policy;
1313 }
1314 if (!pol)
1315 pol = &default_policy;
1316 return pol;
1317}
1318
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001319/*
1320 * Return a nodemask representing a mempolicy for filtering nodes for
1321 * page allocation
1322 */
1323static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001324{
1325 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001326 if (unlikely(policy->mode == MPOL_BIND) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001327 gfp_zone(gfp) >= policy_zone &&
1328 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1329 return &policy->v.nodes;
1330
1331 return NULL;
1332}
1333
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001334/* Return a zonelist indicated by gfp for node representing a mempolicy */
1335static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001337 int nd = numa_node_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001339 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001341 if (!(policy->flags & MPOL_F_LOCAL))
1342 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 break;
1344 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001345 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001346 * Normally, MPOL_BIND allocations are node-local within the
1347 * allowed nodemask. However, if __GFP_THISNODE is set and the
1348 * current node is part of the mask, we use the zonelist for
1349 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001350 */
Mel Gorman19770b32008-04-28 02:12:18 -07001351 if (unlikely(gfp & __GFP_THISNODE) &&
1352 unlikely(!node_isset(nd, policy->v.nodes)))
1353 nd = first_node(policy->v.nodes);
1354 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 case MPOL_INTERLEAVE: /* should not happen */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 break;
1357 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 BUG();
1359 }
Mel Gorman0e884602008-04-28 02:12:14 -07001360 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361}
1362
1363/* Do dynamic interleaving for a process */
1364static unsigned interleave_nodes(struct mempolicy *policy)
1365{
1366 unsigned nid, next;
1367 struct task_struct *me = current;
1368
1369 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001370 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001372 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001373 if (next < MAX_NUMNODES)
1374 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 return nid;
1376}
1377
Christoph Lameterdc85da12006-01-18 17:42:36 -08001378/*
1379 * Depending on the memory policy provide a node from which to allocate the
1380 * next slab entry.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001381 * @policy must be protected by freeing by the caller. If @policy is
1382 * the current task's mempolicy, this protection is implicit, as only the
1383 * task can change it's policy. The system default policy requires no
1384 * such protection.
Christoph Lameterdc85da12006-01-18 17:42:36 -08001385 */
1386unsigned slab_node(struct mempolicy *policy)
1387{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001388 if (!policy || policy->flags & MPOL_F_LOCAL)
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001389 return numa_node_id();
Christoph Lameter765c4502006-09-27 01:50:08 -07001390
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001391 switch (policy->mode) {
1392 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001393 /*
1394 * handled MPOL_F_LOCAL above
1395 */
1396 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001397
Christoph Lameterdc85da12006-01-18 17:42:36 -08001398 case MPOL_INTERLEAVE:
1399 return interleave_nodes(policy);
1400
Mel Gormandd1a2392008-04-28 02:12:17 -07001401 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001402 /*
1403 * Follow bind policy behavior and start allocation at the
1404 * first node.
1405 */
Mel Gorman19770b32008-04-28 02:12:18 -07001406 struct zonelist *zonelist;
1407 struct zone *zone;
1408 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1409 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1410 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1411 &policy->v.nodes,
1412 &zone);
1413 return zone->node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001414 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001415
Christoph Lameterdc85da12006-01-18 17:42:36 -08001416 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001417 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001418 }
1419}
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421/* Do static interleaving for a VMA with known offset. */
1422static unsigned offset_il_node(struct mempolicy *pol,
1423 struct vm_area_struct *vma, unsigned long off)
1424{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001425 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001426 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 int c;
1428 int nid = -1;
1429
David Rientjesf5b087b2008-04-28 02:12:27 -07001430 if (!nnodes)
1431 return numa_node_id();
1432 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 c = 0;
1434 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001435 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 c++;
1437 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 return nid;
1439}
1440
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001441/* Determine a node number for interleave */
1442static inline unsigned interleave_nid(struct mempolicy *pol,
1443 struct vm_area_struct *vma, unsigned long addr, int shift)
1444{
1445 if (vma) {
1446 unsigned long off;
1447
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001448 /*
1449 * for small pages, there is no difference between
1450 * shift and PAGE_SHIFT, so the bit-shift is safe.
1451 * for huge pages, since vm_pgoff is in units of small
1452 * pages, we need to shift off the always 0 bits to get
1453 * a useful offset.
1454 */
1455 BUG_ON(shift < PAGE_SHIFT);
1456 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001457 off += (addr - vma->vm_start) >> shift;
1458 return offset_il_node(pol, vma, off);
1459 } else
1460 return interleave_nodes(pol);
1461}
1462
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001463#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001464/*
1465 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1466 * @vma = virtual memory area whose policy is sought
1467 * @addr = address in @vma for shared policy lookup and interleave policy
1468 * @gfp_flags = for requested zone
Mel Gorman19770b32008-04-28 02:12:18 -07001469 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1470 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001471 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001472 * Returns a zonelist suitable for a huge page allocation and a pointer
1473 * to the struct mempolicy for conditional unref after allocation.
1474 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1475 * @nodemask for filtering the zonelist.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001476 */
Mel Gorman396faf02007-07-17 04:03:13 -07001477struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001478 gfp_t gfp_flags, struct mempolicy **mpol,
1479 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001480{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001481 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001482
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001483 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001484 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001485
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001486 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1487 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001488 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001489 } else {
1490 zl = policy_zonelist(gfp_flags, *mpol);
1491 if ((*mpol)->mode == MPOL_BIND)
1492 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001493 }
1494 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001495}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001496#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498/* Allocate a page in interleaved policy.
1499 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001500static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1501 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
1503 struct zonelist *zl;
1504 struct page *page;
1505
Mel Gorman0e884602008-04-28 02:12:14 -07001506 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001508 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001509 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 return page;
1511}
1512
1513/**
1514 * alloc_page_vma - Allocate a page for a VMA.
1515 *
1516 * @gfp:
1517 * %GFP_USER user allocation.
1518 * %GFP_KERNEL kernel allocations,
1519 * %GFP_HIGHMEM highmem/user allocations,
1520 * %GFP_FS allocation should not call back into a file system.
1521 * %GFP_ATOMIC don't sleep.
1522 *
1523 * @vma: Pointer to VMA or NULL if not available.
1524 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1525 *
1526 * This function allocates a page from the kernel page pool and applies
1527 * a NUMA policy associated with the VMA or the current process.
1528 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1529 * mm_struct of the VMA to prevent it from going away. Should be used for
1530 * all allocations for pages that will be mapped into
1531 * user space. Returns NULL when no page can be allocated.
1532 *
1533 * Should be called with the mm_sem of the vma hold.
1534 */
1535struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001536alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001538 struct mempolicy *pol = get_vma_policy(current, vma, addr);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001539 struct zonelist *zl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001541 cpuset_update_task_memory_state();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001543 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001545
1546 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001547 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 return alloc_page_interleave(gfp, 0, nid);
1549 }
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001550 zl = policy_zonelist(gfp, pol);
1551 if (unlikely(mpol_needs_cond_ref(pol))) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001552 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001553 * slow path: ref counted shared policy
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001554 */
Mel Gorman19770b32008-04-28 02:12:18 -07001555 struct page *page = __alloc_pages_nodemask(gfp, 0,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001556 zl, policy_nodemask(gfp, pol));
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001557 __mpol_put(pol);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001558 return page;
1559 }
1560 /*
1561 * fast path: default or task policy
1562 */
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001563 return __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
1566/**
1567 * alloc_pages_current - Allocate pages.
1568 *
1569 * @gfp:
1570 * %GFP_USER user allocation,
1571 * %GFP_KERNEL kernel allocation,
1572 * %GFP_HIGHMEM highmem allocation,
1573 * %GFP_FS don't call back into a file system.
1574 * %GFP_ATOMIC don't sleep.
1575 * @order: Power of two of allocation size in pages. 0 is a single page.
1576 *
1577 * Allocate a page from the kernel page pool. When not in
1578 * interrupt context and apply the current process NUMA policy.
1579 * Returns NULL when no page can be allocated.
1580 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001581 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 * 1) it's ok to take cpuset_sem (can WAIT), and
1583 * 2) allocating for current task (not interrupt).
1584 */
Al Virodd0fc662005-10-07 07:46:04 +01001585struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
1587 struct mempolicy *pol = current->mempolicy;
1588
1589 if ((gfp & __GFP_WAIT) && !in_interrupt())
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001590 cpuset_update_task_memory_state();
Christoph Lameter9b819d22006-09-25 23:31:40 -07001591 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001593
1594 /*
1595 * No reference counting needed for current->mempolicy
1596 * nor system default_policy
1597 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001598 if (pol->mode == MPOL_INTERLEAVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 return alloc_page_interleave(gfp, order, interleave_nodes(pol));
Mel Gorman19770b32008-04-28 02:12:18 -07001600 return __alloc_pages_nodemask(gfp, order,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001601 policy_zonelist(gfp, pol), policy_nodemask(gfp, pol));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603EXPORT_SYMBOL(alloc_pages_current);
1604
Paul Jackson42253992006-01-08 01:01:59 -08001605/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001606 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08001607 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1608 * with the mems_allowed returned by cpuset_mems_allowed(). This
1609 * keeps mempolicies cpuset relative after its cpuset moves. See
1610 * further kernel/cpuset.c update_nodemask().
1611 */
Paul Jackson42253992006-01-08 01:01:59 -08001612
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001613/* Slow path of a mempolicy duplicate */
1614struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
1616 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1617
1618 if (!new)
1619 return ERR_PTR(-ENOMEM);
Paul Jackson42253992006-01-08 01:01:59 -08001620 if (current_cpuset_is_being_rebound()) {
1621 nodemask_t mems = cpuset_mems_allowed(current);
1622 mpol_rebind_policy(old, &mems);
1623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 *new = *old;
1625 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return new;
1627}
1628
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001629/*
1630 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1631 * eliminate the * MPOL_F_* flags that require conditional ref and
1632 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1633 * after return. Use the returned value.
1634 *
1635 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1636 * policy lookup, even if the policy needs/has extra ref on lookup.
1637 * shmem_readahead needs this.
1638 */
1639struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
1640 struct mempolicy *frompol)
1641{
1642 if (!mpol_needs_cond_ref(frompol))
1643 return frompol;
1644
1645 *tompol = *frompol;
1646 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
1647 __mpol_put(frompol);
1648 return tompol;
1649}
1650
David Rientjesf5b087b2008-04-28 02:12:27 -07001651static int mpol_match_intent(const struct mempolicy *a,
1652 const struct mempolicy *b)
1653{
1654 if (a->flags != b->flags)
1655 return 0;
1656 if (!mpol_store_user_nodemask(a))
1657 return 1;
1658 return nodes_equal(a->w.user_nodemask, b->w.user_nodemask);
1659}
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661/* Slow path of a mempolicy comparison */
1662int __mpol_equal(struct mempolicy *a, struct mempolicy *b)
1663{
1664 if (!a || !b)
1665 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001666 if (a->mode != b->mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001668 if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b))
David Rientjesf5b087b2008-04-28 02:12:27 -07001669 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001670 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07001671 case MPOL_BIND:
1672 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -07001674 return nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001676 return a->v.preferred_node == b->v.preferred_node &&
1677 a->flags == b->flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 default:
1679 BUG();
1680 return 0;
1681 }
1682}
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 * Shared memory backing store policy support.
1686 *
1687 * Remember policies even when nobody has shared memory mapped.
1688 * The policies are kept in Red-Black tree linked from the inode.
1689 * They are protected by the sp->lock spinlock, which should be held
1690 * for any accesses to the tree.
1691 */
1692
1693/* lookup first element intersecting start-end */
1694/* Caller holds sp->lock */
1695static struct sp_node *
1696sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
1697{
1698 struct rb_node *n = sp->root.rb_node;
1699
1700 while (n) {
1701 struct sp_node *p = rb_entry(n, struct sp_node, nd);
1702
1703 if (start >= p->end)
1704 n = n->rb_right;
1705 else if (end <= p->start)
1706 n = n->rb_left;
1707 else
1708 break;
1709 }
1710 if (!n)
1711 return NULL;
1712 for (;;) {
1713 struct sp_node *w = NULL;
1714 struct rb_node *prev = rb_prev(n);
1715 if (!prev)
1716 break;
1717 w = rb_entry(prev, struct sp_node, nd);
1718 if (w->end <= start)
1719 break;
1720 n = prev;
1721 }
1722 return rb_entry(n, struct sp_node, nd);
1723}
1724
1725/* Insert a new shared policy into the list. */
1726/* Caller holds sp->lock */
1727static void sp_insert(struct shared_policy *sp, struct sp_node *new)
1728{
1729 struct rb_node **p = &sp->root.rb_node;
1730 struct rb_node *parent = NULL;
1731 struct sp_node *nd;
1732
1733 while (*p) {
1734 parent = *p;
1735 nd = rb_entry(parent, struct sp_node, nd);
1736 if (new->start < nd->start)
1737 p = &(*p)->rb_left;
1738 else if (new->end > nd->end)
1739 p = &(*p)->rb_right;
1740 else
1741 BUG();
1742 }
1743 rb_link_node(&new->nd, parent, p);
1744 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07001745 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001746 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
1749/* Find shared policy intersecting idx */
1750struct mempolicy *
1751mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
1752{
1753 struct mempolicy *pol = NULL;
1754 struct sp_node *sn;
1755
1756 if (!sp->root.rb_node)
1757 return NULL;
1758 spin_lock(&sp->lock);
1759 sn = sp_lookup(sp, idx, idx+1);
1760 if (sn) {
1761 mpol_get(sn->policy);
1762 pol = sn->policy;
1763 }
1764 spin_unlock(&sp->lock);
1765 return pol;
1766}
1767
1768static void sp_delete(struct shared_policy *sp, struct sp_node *n)
1769{
Paul Mundt140d5a42007-07-15 23:38:16 -07001770 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 rb_erase(&n->nd, &sp->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001772 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 kmem_cache_free(sn_cache, n);
1774}
1775
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001776static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
1777 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778{
1779 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
1780
1781 if (!n)
1782 return NULL;
1783 n->start = start;
1784 n->end = end;
1785 mpol_get(pol);
Lee Schermerhornaab0b102008-04-28 02:13:13 -07001786 pol->flags |= MPOL_F_SHARED; /* for unref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 n->policy = pol;
1788 return n;
1789}
1790
1791/* Replace a policy range. */
1792static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
1793 unsigned long end, struct sp_node *new)
1794{
1795 struct sp_node *n, *new2 = NULL;
1796
1797restart:
1798 spin_lock(&sp->lock);
1799 n = sp_lookup(sp, start, end);
1800 /* Take care of old policies in the same range. */
1801 while (n && n->start < end) {
1802 struct rb_node *next = rb_next(&n->nd);
1803 if (n->start >= start) {
1804 if (n->end <= end)
1805 sp_delete(sp, n);
1806 else
1807 n->start = end;
1808 } else {
1809 /* Old policy spanning whole new range. */
1810 if (n->end > end) {
1811 if (!new2) {
1812 spin_unlock(&sp->lock);
1813 new2 = sp_alloc(end, n->end, n->policy);
1814 if (!new2)
1815 return -ENOMEM;
1816 goto restart;
1817 }
1818 n->end = start;
1819 sp_insert(sp, new2);
1820 new2 = NULL;
1821 break;
1822 } else
1823 n->end = start;
1824 }
1825 if (!next)
1826 break;
1827 n = rb_entry(next, struct sp_node, nd);
1828 }
1829 if (new)
1830 sp_insert(sp, new);
1831 spin_unlock(&sp->lock);
1832 if (new2) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001833 mpol_put(new2->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 kmem_cache_free(sn_cache, new2);
1835 }
1836 return 0;
1837}
1838
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001839/**
1840 * mpol_shared_policy_init - initialize shared policy for inode
1841 * @sp: pointer to inode shared policy
1842 * @mpol: struct mempolicy to install
1843 *
1844 * Install non-NULL @mpol in inode's shared policy rb-tree.
1845 * On entry, the current task has a reference on a non-NULL @mpol.
1846 * This must be released on exit.
1847 */
1848void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08001849{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001850 sp->root = RB_ROOT; /* empty tree == default mempolicy */
1851 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08001852
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001853 if (mpol) {
1854 struct vm_area_struct pvma;
1855 struct mempolicy *new;
Robin Holt7339ff82006-01-14 13:20:48 -08001856
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001857 /* contextualize the tmpfs mount point mempolicy */
1858 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
1859 mpol_put(mpol); /* drop our ref on sb mpol */
1860 if (IS_ERR(new))
1861 return; /* no valid nodemask intersection */
Robin Holt7339ff82006-01-14 13:20:48 -08001862
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001863 /* Create pseudo-vma that contains just the policy */
1864 memset(&pvma, 0, sizeof(struct vm_area_struct));
1865 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
1866 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
1867 mpol_put(new); /* drop initial ref */
Robin Holt7339ff82006-01-14 13:20:48 -08001868 }
1869}
1870
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871int mpol_set_shared_policy(struct shared_policy *info,
1872 struct vm_area_struct *vma, struct mempolicy *npol)
1873{
1874 int err;
1875 struct sp_node *new = NULL;
1876 unsigned long sz = vma_pages(vma);
1877
David Rientjes028fec42008-04-28 02:12:25 -07001878 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001880 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07001881 npol ? npol->flags : -1,
Paul Mundt140d5a42007-07-15 23:38:16 -07001882 npol ? nodes_addr(npol->v.nodes)[0] : -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 if (npol) {
1885 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
1886 if (!new)
1887 return -ENOMEM;
1888 }
1889 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
1890 if (err && new)
1891 kmem_cache_free(sn_cache, new);
1892 return err;
1893}
1894
1895/* Free a backing policy store on inode delete. */
1896void mpol_free_shared_policy(struct shared_policy *p)
1897{
1898 struct sp_node *n;
1899 struct rb_node *next;
1900
1901 if (!p->root.rb_node)
1902 return;
1903 spin_lock(&p->lock);
1904 next = rb_first(&p->root);
1905 while (next) {
1906 n = rb_entry(next, struct sp_node, nd);
1907 next = rb_next(&n->nd);
Andi Kleen90c50292005-07-27 11:43:50 -07001908 rb_erase(&n->nd, &p->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001909 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 kmem_cache_free(sn_cache, n);
1911 }
1912 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
1915/* assumes fs == KERNEL_DS */
1916void __init numa_policy_init(void)
1917{
Paul Mundtb71636e2007-07-15 23:38:15 -07001918 nodemask_t interleave_nodes;
1919 unsigned long largest = 0;
1920 int nid, prefer = 0;
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 policy_cache = kmem_cache_create("numa_policy",
1923 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09001924 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
1926 sn_cache = kmem_cache_create("shared_policy_node",
1927 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09001928 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Paul Mundtb71636e2007-07-15 23:38:15 -07001930 /*
1931 * Set interleaving policy for system init. Interleaving is only
1932 * enabled across suitably sized nodes (default is >= 16MB), or
1933 * fall back to the largest node if they're all smaller.
1934 */
1935 nodes_clear(interleave_nodes);
Christoph Lameter56bbd652007-10-16 01:25:35 -07001936 for_each_node_state(nid, N_HIGH_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07001937 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Paul Mundtb71636e2007-07-15 23:38:15 -07001939 /* Preserve the largest node */
1940 if (largest < total_pages) {
1941 largest = total_pages;
1942 prefer = nid;
1943 }
1944
1945 /* Interleave this node? */
1946 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
1947 node_set(nid, interleave_nodes);
1948 }
1949
1950 /* All too small, use the largest */
1951 if (unlikely(nodes_empty(interleave_nodes)))
1952 node_set(prefer, interleave_nodes);
1953
David Rientjes028fec42008-04-28 02:12:25 -07001954 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 printk("numa_policy_init: interleaving failed\n");
1956}
1957
Christoph Lameter8bccd852005-10-29 18:16:59 -07001958/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959void numa_default_policy(void)
1960{
David Rientjes028fec42008-04-28 02:12:25 -07001961 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962}
Paul Jackson68860ec2005-10-30 15:02:36 -08001963
Paul Jackson42253992006-01-08 01:01:59 -08001964/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001965 * Parse and format mempolicy from/to strings
1966 */
1967
1968/*
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001969 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07001970 * Used only for mpol_parse_str() and mpol_to_str()
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08001971 */
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001972#define MPOL_LOCAL (MPOL_INTERLEAVE + 1)
Helge Deller15ad7cd2006-12-06 20:40:36 -08001973static const char * const policy_types[] =
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001974 { "default", "prefer", "bind", "interleave", "local" };
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08001975
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001976
1977#ifdef CONFIG_TMPFS
1978/**
1979 * mpol_parse_str - parse string to mempolicy
1980 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001981 * @mpol: pointer to struct mempolicy pointer, returned on success.
1982 * @no_context: flag whether to "contextualize" the mempolicy
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001983 *
1984 * Format of input:
1985 * <mode>[=<flags>][:<nodelist>]
1986 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001987 * if @no_context is true, save the input nodemask in w.user_nodemask in
1988 * the returned mempolicy. This will be used to "clone" the mempolicy in
1989 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
1990 * mount option. Note that if 'static' or 'relative' mode flags were
1991 * specified, the input nodemask will already have been saved. Saving
1992 * it again is redundant, but safe.
1993 *
1994 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001995 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001996int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001997{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001998 struct mempolicy *new = NULL;
1999 unsigned short uninitialized_var(mode);
2000 unsigned short uninitialized_var(mode_flags);
2001 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002002 char *nodelist = strchr(str, ':');
2003 char *flags = strchr(str, '=');
2004 int i;
2005 int err = 1;
2006
2007 if (nodelist) {
2008 /* NUL-terminate mode or flags string */
2009 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002010 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002011 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002012 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002013 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002014 } else
2015 nodes_clear(nodes);
2016
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002017 if (flags)
2018 *flags++ = '\0'; /* terminate mode string */
2019
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002020 for (i = 0; i <= MPOL_LOCAL; i++) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002021 if (!strcmp(str, policy_types[i])) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002022 mode = i;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002023 break;
2024 }
2025 }
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002026 if (i > MPOL_LOCAL)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002027 goto out;
2028
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002029 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002030 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002031 /*
2032 * Insist on a nodelist of one node only
2033 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002034 if (nodelist) {
2035 char *rest = nodelist;
2036 while (isdigit(*rest))
2037 rest++;
2038 if (!*rest)
2039 err = 0;
2040 }
2041 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002042 case MPOL_INTERLEAVE:
2043 /*
2044 * Default to online nodes with memory if no nodelist
2045 */
2046 if (!nodelist)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002047 nodes = node_states[N_HIGH_MEMORY];
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002048 err = 0;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002049 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002050 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002051 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002052 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002053 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002054 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002055 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002056 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002057 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002058
2059 /*
2060 * case MPOL_BIND: mpol_new() enforces non-empty nodemask.
2061 * case MPOL_DEFAULT: mpol_new() enforces empty nodemask, ignores flags.
2062 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002063 }
2064
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002065 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002066 if (flags) {
2067 /*
2068 * Currently, we only support two mutually exclusive
2069 * mode flags.
2070 */
2071 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002072 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002073 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002074 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002075 else
2076 err = 1;
2077 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002078
2079 new = mpol_new(mode, mode_flags, &nodes);
2080 if (IS_ERR(new))
2081 err = 1;
2082 else if (no_context)
2083 new->w.user_nodemask = nodes; /* save for contextualization */
2084
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002085out:
2086 /* Restore string for error message */
2087 if (nodelist)
2088 *--nodelist = ':';
2089 if (flags)
2090 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002091 if (!err)
2092 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002093 return err;
2094}
2095#endif /* CONFIG_TMPFS */
2096
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002097/**
2098 * mpol_to_str - format a mempolicy structure for printing
2099 * @buffer: to contain formatted mempolicy string
2100 * @maxlen: length of @buffer
2101 * @pol: pointer to mempolicy to be formatted
2102 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2103 *
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002104 * Convert a mempolicy into a string.
2105 * Returns the number of characters in buffer (if positive)
2106 * or an error (negative)
2107 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002108int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002109{
2110 char *p = buffer;
2111 int l;
2112 nodemask_t nodes;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002113 unsigned short mode;
David Rientjesf5b087b2008-04-28 02:12:27 -07002114 unsigned short flags = pol ? pol->flags : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002115
Lee Schermerhorn22919902008-04-28 02:13:22 -07002116 /*
2117 * Sanity check: room for longest mode, flag and some nodes
2118 */
2119 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2120
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002121 if (!pol || pol == &default_policy)
2122 mode = MPOL_DEFAULT;
2123 else
2124 mode = pol->mode;
2125
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002126 switch (mode) {
2127 case MPOL_DEFAULT:
2128 nodes_clear(nodes);
2129 break;
2130
2131 case MPOL_PREFERRED:
2132 nodes_clear(nodes);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002133 if (flags & MPOL_F_LOCAL)
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002134 mode = MPOL_LOCAL; /* pseudo-policy */
2135 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002136 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002137 break;
2138
2139 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07002140 /* Fall through */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002141 case MPOL_INTERLEAVE:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002142 if (no_context)
2143 nodes = pol->w.user_nodemask;
2144 else
2145 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002146 break;
2147
2148 default:
2149 BUG();
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002150 }
2151
2152 l = strlen(policy_types[mode]);
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002153 if (buffer + maxlen < p + l + 1)
2154 return -ENOSPC;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002155
2156 strcpy(p, policy_types[mode]);
2157 p += l;
2158
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002159 if (flags & MPOL_MODE_FLAGS) {
David Rientjesf5b087b2008-04-28 02:12:27 -07002160 if (buffer + maxlen < p + 2)
2161 return -ENOSPC;
2162 *p++ = '=';
2163
Lee Schermerhorn22919902008-04-28 02:13:22 -07002164 /*
2165 * Currently, the only defined flags are mutually exclusive
2166 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002167 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002168 p += snprintf(p, buffer + maxlen - p, "static");
2169 else if (flags & MPOL_F_RELATIVE_NODES)
2170 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002171 }
2172
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002173 if (!nodes_empty(nodes)) {
2174 if (buffer + maxlen < p + 2)
2175 return -ENOSPC;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002176 *p++ = ':';
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002177 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2178 }
2179 return p - buffer;
2180}
2181
2182struct numa_maps {
2183 unsigned long pages;
2184 unsigned long anon;
Christoph Lameter397874d2006-03-06 15:42:53 -08002185 unsigned long active;
2186 unsigned long writeback;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002187 unsigned long mapcount_max;
Christoph Lameter397874d2006-03-06 15:42:53 -08002188 unsigned long dirty;
2189 unsigned long swapcache;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002190 unsigned long node[MAX_NUMNODES];
2191};
2192
Christoph Lameter397874d2006-03-06 15:42:53 -08002193static void gather_stats(struct page *page, void *private, int pte_dirty)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002194{
2195 struct numa_maps *md = private;
2196 int count = page_mapcount(page);
2197
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002198 md->pages++;
Christoph Lameter397874d2006-03-06 15:42:53 -08002199 if (pte_dirty || PageDirty(page))
2200 md->dirty++;
2201
2202 if (PageSwapCache(page))
2203 md->swapcache++;
2204
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002205 if (PageActive(page) || PageUnevictable(page))
Christoph Lameter397874d2006-03-06 15:42:53 -08002206 md->active++;
2207
2208 if (PageWriteback(page))
2209 md->writeback++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002210
2211 if (PageAnon(page))
2212 md->anon++;
2213
Christoph Lameter397874d2006-03-06 15:42:53 -08002214 if (count > md->mapcount_max)
2215 md->mapcount_max = count;
2216
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002217 md->node[page_to_nid(page)]++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002218}
2219
Andrew Morton7f709ed2006-03-07 21:55:22 -08002220#ifdef CONFIG_HUGETLB_PAGE
Christoph Lameter397874d2006-03-06 15:42:53 -08002221static void check_huge_range(struct vm_area_struct *vma,
2222 unsigned long start, unsigned long end,
2223 struct numa_maps *md)
2224{
2225 unsigned long addr;
2226 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -07002227 struct hstate *h = hstate_vma(vma);
2228 unsigned long sz = huge_page_size(h);
Christoph Lameter397874d2006-03-06 15:42:53 -08002229
Andi Kleena5516432008-07-23 21:27:41 -07002230 for (addr = start; addr < end; addr += sz) {
2231 pte_t *ptep = huge_pte_offset(vma->vm_mm,
2232 addr & huge_page_mask(h));
Christoph Lameter397874d2006-03-06 15:42:53 -08002233 pte_t pte;
2234
2235 if (!ptep)
2236 continue;
2237
2238 pte = *ptep;
2239 if (pte_none(pte))
2240 continue;
2241
2242 page = pte_page(pte);
2243 if (!page)
2244 continue;
2245
2246 gather_stats(page, md, pte_dirty(*ptep));
2247 }
2248}
Andrew Morton7f709ed2006-03-07 21:55:22 -08002249#else
2250static inline void check_huge_range(struct vm_area_struct *vma,
2251 unsigned long start, unsigned long end,
2252 struct numa_maps *md)
2253{
2254}
2255#endif
Christoph Lameter397874d2006-03-06 15:42:53 -08002256
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002257/*
2258 * Display pages allocated per node and memory policy via /proc.
2259 */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002260int show_numa_map(struct seq_file *m, void *v)
2261{
Eric W. Biederman99f89552006-06-26 00:25:55 -07002262 struct proc_maps_private *priv = m->private;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002263 struct vm_area_struct *vma = v;
2264 struct numa_maps *md;
Christoph Lameter397874d2006-03-06 15:42:53 -08002265 struct file *file = vma->vm_file;
2266 struct mm_struct *mm = vma->vm_mm;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002267 struct mempolicy *pol;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002268 int n;
2269 char buffer[50];
2270
Christoph Lameter397874d2006-03-06 15:42:53 -08002271 if (!mm)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002272 return 0;
2273
2274 md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL);
2275 if (!md)
2276 return 0;
2277
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002278 pol = get_vma_policy(priv->task, vma, vma->vm_start);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002279 mpol_to_str(buffer, sizeof(buffer), pol, 0);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002280 mpol_cond_put(pol);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002281
Christoph Lameter397874d2006-03-06 15:42:53 -08002282 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002283
Christoph Lameter397874d2006-03-06 15:42:53 -08002284 if (file) {
2285 seq_printf(m, " file=");
Jan Blunckc32c2f62008-02-14 19:38:43 -08002286 seq_path(m, &file->f_path, "\n\t= ");
Christoph Lameter397874d2006-03-06 15:42:53 -08002287 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
2288 seq_printf(m, " heap");
2289 } else if (vma->vm_start <= mm->start_stack &&
2290 vma->vm_end >= mm->start_stack) {
2291 seq_printf(m, " stack");
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002292 }
Christoph Lameter397874d2006-03-06 15:42:53 -08002293
2294 if (is_vm_hugetlb_page(vma)) {
2295 check_huge_range(vma, vma->vm_start, vma->vm_end, md);
2296 seq_printf(m, " huge");
2297 } else {
2298 check_pgd_range(vma, vma->vm_start, vma->vm_end,
Christoph Lameter56bbd652007-10-16 01:25:35 -07002299 &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
Christoph Lameter397874d2006-03-06 15:42:53 -08002300 }
2301
2302 if (!md->pages)
2303 goto out;
2304
2305 if (md->anon)
2306 seq_printf(m," anon=%lu",md->anon);
2307
2308 if (md->dirty)
2309 seq_printf(m," dirty=%lu",md->dirty);
2310
2311 if (md->pages != md->anon && md->pages != md->dirty)
2312 seq_printf(m, " mapped=%lu", md->pages);
2313
2314 if (md->mapcount_max > 1)
2315 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2316
2317 if (md->swapcache)
2318 seq_printf(m," swapcache=%lu", md->swapcache);
2319
2320 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2321 seq_printf(m," active=%lu", md->active);
2322
2323 if (md->writeback)
2324 seq_printf(m," writeback=%lu", md->writeback);
2325
Christoph Lameter56bbd652007-10-16 01:25:35 -07002326 for_each_node_state(n, N_HIGH_MEMORY)
Christoph Lameter397874d2006-03-06 15:42:53 -08002327 if (md->node[n])
2328 seq_printf(m, " N%d=%lu", n, md->node[n]);
2329out:
2330 seq_putc(m, '\n');
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002331 kfree(md);
2332
2333 if (m->count < m->size)
Eric W. Biederman99f89552006-06-26 00:25:55 -07002334 m->version = (vma != priv->tail_vma) ? vma->vm_start : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002335 return 0;
2336}