4 fix mmap readahead to honour policy and enable policy for any page cache
6 statistics for bigpages
7 global policy for page cache? currently it uses process policy. Requires
9 handle mremap for shared memory (currently ignored for the policy)
11 make bind policy root only? It can trigger oom much faster and the
12 kernel is not always grateful with that.
15 #include <linux/mempolicy.h>
17 #include <linux/highmem.h>
18 #include <linux/hugetlb.h>
19 #include <linux/kernel.h>
20 #include <linux/sched.h>
21 #include <linux/nodemask.h>
22 #include <linux/cpuset.h>
23 #include <linux/slab.h>
24 #include <linux/string.h>
25 #include <linux/module.h>
26 #include <linux/nsproxy.h>
27 #include <linux/interrupt.h>
28 #include <linux/init.h>
29 #include <linux/compat.h>
30 #include <linux/swap.h>
31 #include <linux/seq_file.h>
32 #include <linux/proc_fs.h>
33 #include <linux/migrate.h>
34 #include <linux/ksm.h>
35 #include <linux/rmap.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/ctype.h>
39 #include <linux/mm_inline.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
47 #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
48 #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
49 #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */
51 static struct kmem_cache
*policy_cache
;
52 static struct kmem_cache
*sn_cache
;
54 /* Highest zone. An specific allocation for a zone below that is not
56 enum zone_type policy_zone
= 0;
59 * run-time system-wide default policy => local allocation
61 struct mempolicy default_policy
= {
62 .refcnt
= ATOMIC_INIT(1), /* never free it */
63 .mode
= MPOL_PREFERRED
,
64 .flags
= MPOL_F_LOCAL
,
67 static const struct mempolicy_operations
{
68 int (*create
)(struct mempolicy
*pol
, const nodemask_t
*nodes
);
70 * If read-side task has no lock to protect task->mempolicy, write-side
71 * task will rebind the task->mempolicy by two step. The first step is
72 * setting all the newly nodes, and the second step is cleaning all the
73 * disallowed nodes. In this way, we can avoid finding no node to alloc
75 * If we have a lock to protect task->mempolicy in read-side, we do
79 * MPOL_REBIND_ONCE - do rebind work at once
80 * MPOL_REBIND_STEP1 - set all the newly nodes
81 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
83 void (*rebind
)(struct mempolicy
*pol
, const nodemask_t
*nodes
,
84 enum mpol_rebind_step step
);
87 /* Check that the nodemask contains at least one populated zone */
88 static int is_valid_nodemask(const nodemask_t
*nodemask
)
92 for_each_node_mask(nd
, *nodemask
) {
95 for (k
= 0; k
<= policy_zone
; k
++) {
96 z
= &NODE_DATA(nd
)->node_zones
[k
];
97 if (z
->present_pages
> 0)
105 static inline int mpol_store_user_nodemask(const struct mempolicy
*pol
)
107 return pol
->flags
& MPOL_MODE_FLAGS
;
110 static void mpol_relative_nodemask(nodemask_t
*ret
, const nodemask_t
*orig
,
111 const nodemask_t
*rel
)
114 nodes_fold(tmp
, *orig
, nodes_weight(*rel
));
115 nodes_onto(*ret
, tmp
, *rel
);
118 static int mpol_new_interleave(struct mempolicy
*pol
, const nodemask_t
*nodes
)
120 if (nodes_empty(*nodes
))
122 pol
->v
.nodes
= *nodes
;
126 static int mpol_new_preferred(struct mempolicy
*pol
, const nodemask_t
*nodes
)
129 pol
->flags
|= MPOL_F_LOCAL
; /* local allocation */
130 else if (nodes_empty(*nodes
))
131 return -EINVAL
; /* no allowed nodes */
133 pol
->v
.preferred_node
= first_node(*nodes
);
137 static int mpol_new_bind(struct mempolicy
*pol
, const nodemask_t
*nodes
)
139 if (!is_valid_nodemask(nodes
))
141 pol
->v
.nodes
= *nodes
;
146 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
147 * any, for the new policy. mpol_new() has already validated the nodes
148 * parameter with respect to the policy mode and flags. But, we need to
149 * handle an empty nodemask with MPOL_PREFERRED here.
151 * Must be called holding task's alloc_lock to protect task's mems_allowed
152 * and mempolicy. May also be called holding the mmap_semaphore for write.
154 static int mpol_set_nodemask(struct mempolicy
*pol
,
155 const nodemask_t
*nodes
, struct nodemask_scratch
*nsc
)
159 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
162 /* Check N_HIGH_MEMORY */
163 nodes_and(nsc
->mask1
,
164 cpuset_current_mems_allowed
, node_states
[N_HIGH_MEMORY
]);
167 if (pol
->mode
== MPOL_PREFERRED
&& nodes_empty(*nodes
))
168 nodes
= NULL
; /* explicit local allocation */
170 if (pol
->flags
& MPOL_F_RELATIVE_NODES
)
171 mpol_relative_nodemask(&nsc
->mask2
, nodes
,&nsc
->mask1
);
173 nodes_and(nsc
->mask2
, *nodes
, nsc
->mask1
);
175 if (mpol_store_user_nodemask(pol
))
176 pol
->w
.user_nodemask
= *nodes
;
178 pol
->w
.cpuset_mems_allowed
=
179 cpuset_current_mems_allowed
;
183 ret
= mpol_ops
[pol
->mode
].create(pol
, &nsc
->mask2
);
185 ret
= mpol_ops
[pol
->mode
].create(pol
, NULL
);
190 * This function just creates a new policy, does some check and simple
191 * initialization. You must invoke mpol_set_nodemask() to set nodes.
193 static struct mempolicy
*mpol_new(unsigned short mode
, unsigned short flags
,
196 struct mempolicy
*policy
;
198 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
199 mode
, flags
, nodes
? nodes_addr(*nodes
)[0] : -1);
201 if (mode
== MPOL_DEFAULT
) {
202 if (nodes
&& !nodes_empty(*nodes
))
203 return ERR_PTR(-EINVAL
);
204 return NULL
; /* simply delete any existing policy */
209 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
210 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
211 * All other modes require a valid pointer to a non-empty nodemask.
213 if (mode
== MPOL_PREFERRED
) {
214 if (nodes_empty(*nodes
)) {
215 if (((flags
& MPOL_F_STATIC_NODES
) ||
216 (flags
& MPOL_F_RELATIVE_NODES
)))
217 return ERR_PTR(-EINVAL
);
219 } else if (nodes_empty(*nodes
))
220 return ERR_PTR(-EINVAL
);
221 policy
= kmem_cache_alloc(policy_cache
, GFP_KERNEL
);
223 return ERR_PTR(-ENOMEM
);
224 atomic_set(&policy
->refcnt
, 1);
226 policy
->flags
= flags
;
231 /* Slow path of a mpol destructor. */
232 void __mpol_put(struct mempolicy
*p
)
234 if (!atomic_dec_and_test(&p
->refcnt
))
236 kmem_cache_free(policy_cache
, p
);
239 static void mpol_rebind_default(struct mempolicy
*pol
, const nodemask_t
*nodes
,
240 enum mpol_rebind_step step
)
246 * MPOL_REBIND_ONCE - do rebind work at once
247 * MPOL_REBIND_STEP1 - set all the newly nodes
248 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
250 static void mpol_rebind_nodemask(struct mempolicy
*pol
, const nodemask_t
*nodes
,
251 enum mpol_rebind_step step
)
255 if (pol
->flags
& MPOL_F_STATIC_NODES
)
256 nodes_and(tmp
, pol
->w
.user_nodemask
, *nodes
);
257 else if (pol
->flags
& MPOL_F_RELATIVE_NODES
)
258 mpol_relative_nodemask(&tmp
, &pol
->w
.user_nodemask
, nodes
);
261 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
264 if (step
== MPOL_REBIND_ONCE
|| step
== MPOL_REBIND_STEP1
) {
265 nodes_remap(tmp
, pol
->v
.nodes
,
266 pol
->w
.cpuset_mems_allowed
, *nodes
);
267 pol
->w
.cpuset_mems_allowed
= step
? tmp
: *nodes
;
268 } else if (step
== MPOL_REBIND_STEP2
) {
269 tmp
= pol
->w
.cpuset_mems_allowed
;
270 pol
->w
.cpuset_mems_allowed
= *nodes
;
275 if (nodes_empty(tmp
))
278 if (step
== MPOL_REBIND_STEP1
)
279 nodes_or(pol
->v
.nodes
, pol
->v
.nodes
, tmp
);
280 else if (step
== MPOL_REBIND_ONCE
|| step
== MPOL_REBIND_STEP2
)
285 if (!node_isset(current
->il_next
, tmp
)) {
286 current
->il_next
= next_node(current
->il_next
, tmp
);
287 if (current
->il_next
>= MAX_NUMNODES
)
288 current
->il_next
= first_node(tmp
);
289 if (current
->il_next
>= MAX_NUMNODES
)
290 current
->il_next
= numa_node_id();
294 static void mpol_rebind_preferred(struct mempolicy
*pol
,
295 const nodemask_t
*nodes
,
296 enum mpol_rebind_step step
)
300 if (pol
->flags
& MPOL_F_STATIC_NODES
) {
301 int node
= first_node(pol
->w
.user_nodemask
);
303 if (node_isset(node
, *nodes
)) {
304 pol
->v
.preferred_node
= node
;
305 pol
->flags
&= ~MPOL_F_LOCAL
;
307 pol
->flags
|= MPOL_F_LOCAL
;
308 } else if (pol
->flags
& MPOL_F_RELATIVE_NODES
) {
309 mpol_relative_nodemask(&tmp
, &pol
->w
.user_nodemask
, nodes
);
310 pol
->v
.preferred_node
= first_node(tmp
);
311 } else if (!(pol
->flags
& MPOL_F_LOCAL
)) {
312 pol
->v
.preferred_node
= node_remap(pol
->v
.preferred_node
,
313 pol
->w
.cpuset_mems_allowed
,
315 pol
->w
.cpuset_mems_allowed
= *nodes
;
320 * mpol_rebind_policy - Migrate a policy to a different set of nodes
322 * If read-side task has no lock to protect task->mempolicy, write-side
323 * task will rebind the task->mempolicy by two step. The first step is
324 * setting all the newly nodes, and the second step is cleaning all the
325 * disallowed nodes. In this way, we can avoid finding no node to alloc
327 * If we have a lock to protect task->mempolicy in read-side, we do
331 * MPOL_REBIND_ONCE - do rebind work at once
332 * MPOL_REBIND_STEP1 - set all the newly nodes
333 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
335 static void mpol_rebind_policy(struct mempolicy
*pol
, const nodemask_t
*newmask
,
336 enum mpol_rebind_step step
)
340 if (!mpol_store_user_nodemask(pol
) && step
== 0 &&
341 nodes_equal(pol
->w
.cpuset_mems_allowed
, *newmask
))
344 if (step
== MPOL_REBIND_STEP1
&& (pol
->flags
& MPOL_F_REBINDING
))
347 if (step
== MPOL_REBIND_STEP2
&& !(pol
->flags
& MPOL_F_REBINDING
))
350 if (step
== MPOL_REBIND_STEP1
)
351 pol
->flags
|= MPOL_F_REBINDING
;
352 else if (step
== MPOL_REBIND_STEP2
)
353 pol
->flags
&= ~MPOL_F_REBINDING
;
354 else if (step
>= MPOL_REBIND_NSTEP
)
357 mpol_ops
[pol
->mode
].rebind(pol
, newmask
, step
);
361 * Wrapper for mpol_rebind_policy() that just requires task
362 * pointer, and updates task mempolicy.
364 * Called with task's alloc_lock held.
367 void mpol_rebind_task(struct task_struct
*tsk
, const nodemask_t
*new,
368 enum mpol_rebind_step step
)
370 mpol_rebind_policy(tsk
->mempolicy
, new, step
);
374 * Rebind each vma in mm to new nodemask.
376 * Call holding a reference to mm. Takes mm->mmap_sem during call.
379 void mpol_rebind_mm(struct mm_struct
*mm
, nodemask_t
*new)
381 struct vm_area_struct
*vma
;
383 down_write(&mm
->mmap_sem
);
384 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
)
385 mpol_rebind_policy(vma
->vm_policy
, new, MPOL_REBIND_ONCE
);
386 up_write(&mm
->mmap_sem
);
389 static const struct mempolicy_operations mpol_ops
[MPOL_MAX
] = {
391 .rebind
= mpol_rebind_default
,
393 [MPOL_INTERLEAVE
] = {
394 .create
= mpol_new_interleave
,
395 .rebind
= mpol_rebind_nodemask
,
398 .create
= mpol_new_preferred
,
399 .rebind
= mpol_rebind_preferred
,
402 .create
= mpol_new_bind
,
403 .rebind
= mpol_rebind_nodemask
,
407 static void gather_stats(struct page
*, void *, int pte_dirty
);
408 static void migrate_page_add(struct page
*page
, struct list_head
*pagelist
,
409 unsigned long flags
);
411 /* Scan through pages checking if pages follow certain conditions. */
412 static int check_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
413 unsigned long addr
, unsigned long end
,
414 const nodemask_t
*nodes
, unsigned long flags
,
421 orig_pte
= pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
426 if (!pte_present(*pte
))
428 page
= vm_normal_page(vma
, addr
, *pte
);
432 * vm_normal_page() filters out zero pages, but there might
433 * still be PageReserved pages to skip, perhaps in a VDSO.
434 * And we cannot move PageKsm pages sensibly or safely yet.
436 if (PageReserved(page
) || PageKsm(page
))
438 nid
= page_to_nid(page
);
439 if (node_isset(nid
, *nodes
) == !!(flags
& MPOL_MF_INVERT
))
442 if (flags
& MPOL_MF_STATS
)
443 gather_stats(page
, private, pte_dirty(*pte
));
444 else if (flags
& (MPOL_MF_MOVE
| MPOL_MF_MOVE_ALL
))
445 migrate_page_add(page
, private, flags
);
448 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
449 pte_unmap_unlock(orig_pte
, ptl
);
453 static inline int check_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
454 unsigned long addr
, unsigned long end
,
455 const nodemask_t
*nodes
, unsigned long flags
,
461 pmd
= pmd_offset(pud
, addr
);
463 next
= pmd_addr_end(addr
, end
);
464 if (pmd_none_or_clear_bad(pmd
))
466 if (check_pte_range(vma
, pmd
, addr
, next
, nodes
,
469 } while (pmd
++, addr
= next
, addr
!= end
);
473 static inline int check_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
474 unsigned long addr
, unsigned long end
,
475 const nodemask_t
*nodes
, unsigned long flags
,
481 pud
= pud_offset(pgd
, addr
);
483 next
= pud_addr_end(addr
, end
);
484 if (pud_none_or_clear_bad(pud
))
486 if (check_pmd_range(vma
, pud
, addr
, next
, nodes
,
489 } while (pud
++, addr
= next
, addr
!= end
);
493 static inline int check_pgd_range(struct vm_area_struct
*vma
,
494 unsigned long addr
, unsigned long end
,
495 const nodemask_t
*nodes
, unsigned long flags
,
501 pgd
= pgd_offset(vma
->vm_mm
, addr
);
503 next
= pgd_addr_end(addr
, end
);
504 if (pgd_none_or_clear_bad(pgd
))
506 if (check_pud_range(vma
, pgd
, addr
, next
, nodes
,
509 } while (pgd
++, addr
= next
, addr
!= end
);
514 * Check if all pages in a range are on a set of nodes.
515 * If pagelist != NULL then isolate pages from the LRU and
516 * put them on the pagelist.
518 static struct vm_area_struct
*
519 check_range(struct mm_struct
*mm
, unsigned long start
, unsigned long end
,
520 const nodemask_t
*nodes
, unsigned long flags
, void *private)
523 struct vm_area_struct
*first
, *vma
, *prev
;
526 first
= find_vma(mm
, start
);
528 return ERR_PTR(-EFAULT
);
530 for (vma
= first
; vma
&& vma
->vm_start
< end
; vma
= vma
->vm_next
) {
531 if (!(flags
& MPOL_MF_DISCONTIG_OK
)) {
532 if (!vma
->vm_next
&& vma
->vm_end
< end
)
533 return ERR_PTR(-EFAULT
);
534 if (prev
&& prev
->vm_end
< vma
->vm_start
)
535 return ERR_PTR(-EFAULT
);
537 if (!is_vm_hugetlb_page(vma
) &&
538 ((flags
& MPOL_MF_STRICT
) ||
539 ((flags
& (MPOL_MF_MOVE
| MPOL_MF_MOVE_ALL
)) &&
540 vma_migratable(vma
)))) {
541 unsigned long endvma
= vma
->vm_end
;
545 if (vma
->vm_start
> start
)
546 start
= vma
->vm_start
;
547 err
= check_pgd_range(vma
, start
, endvma
, nodes
,
550 first
= ERR_PTR(err
);
559 /* Apply policy to a single VMA */
560 static int policy_vma(struct vm_area_struct
*vma
, struct mempolicy
*new)
563 struct mempolicy
*old
= vma
->vm_policy
;
565 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
566 vma
->vm_start
, vma
->vm_end
, vma
->vm_pgoff
,
567 vma
->vm_ops
, vma
->vm_file
,
568 vma
->vm_ops
? vma
->vm_ops
->set_policy
: NULL
);
570 if (vma
->vm_ops
&& vma
->vm_ops
->set_policy
)
571 err
= vma
->vm_ops
->set_policy(vma
, new);
574 vma
->vm_policy
= new;
580 /* Step 2: apply policy to a range and do splits. */
581 static int mbind_range(struct mm_struct
*mm
, unsigned long start
,
582 unsigned long end
, struct mempolicy
*new_pol
)
584 struct vm_area_struct
*next
;
585 struct vm_area_struct
*prev
;
586 struct vm_area_struct
*vma
;
589 unsigned long vmstart
;
592 vma
= find_vma_prev(mm
, start
, &prev
);
593 if (!vma
|| vma
->vm_start
> start
)
596 for (; vma
&& vma
->vm_start
< end
; prev
= vma
, vma
= next
) {
598 vmstart
= max(start
, vma
->vm_start
);
599 vmend
= min(end
, vma
->vm_end
);
601 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
602 prev
= vma_merge(mm
, prev
, vmstart
, vmend
, vma
->vm_flags
,
603 vma
->anon_vma
, vma
->vm_file
, pgoff
, new_pol
);
609 if (vma
->vm_start
!= vmstart
) {
610 err
= split_vma(vma
->vm_mm
, vma
, vmstart
, 1);
614 if (vma
->vm_end
!= vmend
) {
615 err
= split_vma(vma
->vm_mm
, vma
, vmend
, 0);
619 err
= policy_vma(vma
, new_pol
);
629 * Update task->flags PF_MEMPOLICY bit: set iff non-default
630 * mempolicy. Allows more rapid checking of this (combined perhaps
631 * with other PF_* flag bits) on memory allocation hot code paths.
633 * If called from outside this file, the task 'p' should -only- be
634 * a newly forked child not yet visible on the task list, because
635 * manipulating the task flags of a visible task is not safe.
637 * The above limitation is why this routine has the funny name
638 * mpol_fix_fork_child_flag().
640 * It is also safe to call this with a task pointer of current,
641 * which the static wrapper mpol_set_task_struct_flag() does,
642 * for use within this file.
645 void mpol_fix_fork_child_flag(struct task_struct
*p
)
648 p
->flags
|= PF_MEMPOLICY
;
650 p
->flags
&= ~PF_MEMPOLICY
;
653 static void mpol_set_task_struct_flag(void)
655 mpol_fix_fork_child_flag(current
);
658 /* Set the process memory policy */
659 static long do_set_mempolicy(unsigned short mode
, unsigned short flags
,
662 struct mempolicy
*new, *old
;
663 struct mm_struct
*mm
= current
->mm
;
664 NODEMASK_SCRATCH(scratch
);
670 new = mpol_new(mode
, flags
, nodes
);
676 * prevent changing our mempolicy while show_numa_maps()
678 * Note: do_set_mempolicy() can be called at init time
682 down_write(&mm
->mmap_sem
);
684 ret
= mpol_set_nodemask(new, nodes
, scratch
);
686 task_unlock(current
);
688 up_write(&mm
->mmap_sem
);
692 old
= current
->mempolicy
;
693 current
->mempolicy
= new;
694 mpol_set_task_struct_flag();
695 if (new && new->mode
== MPOL_INTERLEAVE
&&
696 nodes_weight(new->v
.nodes
))
697 current
->il_next
= first_node(new->v
.nodes
);
698 task_unlock(current
);
700 up_write(&mm
->mmap_sem
);
705 NODEMASK_SCRATCH_FREE(scratch
);
710 * Return nodemask for policy for get_mempolicy() query
712 * Called with task's alloc_lock held
714 static void get_policy_nodemask(struct mempolicy
*p
, nodemask_t
*nodes
)
717 if (p
== &default_policy
)
723 case MPOL_INTERLEAVE
:
727 if (!(p
->flags
& MPOL_F_LOCAL
))
728 node_set(p
->v
.preferred_node
, *nodes
);
729 /* else return empty node mask for local allocation */
736 static int lookup_node(struct mm_struct
*mm
, unsigned long addr
)
741 err
= get_user_pages(current
, mm
, addr
& PAGE_MASK
, 1, 0, 0, &p
, NULL
);
743 err
= page_to_nid(p
);
749 /* Retrieve NUMA policy */
750 static long do_get_mempolicy(int *policy
, nodemask_t
*nmask
,
751 unsigned long addr
, unsigned long flags
)
754 struct mm_struct
*mm
= current
->mm
;
755 struct vm_area_struct
*vma
= NULL
;
756 struct mempolicy
*pol
= current
->mempolicy
;
759 ~(unsigned long)(MPOL_F_NODE
|MPOL_F_ADDR
|MPOL_F_MEMS_ALLOWED
))
762 if (flags
& MPOL_F_MEMS_ALLOWED
) {
763 if (flags
& (MPOL_F_NODE
|MPOL_F_ADDR
))
765 *policy
= 0; /* just so it's initialized */
767 *nmask
= cpuset_current_mems_allowed
;
768 task_unlock(current
);
772 if (flags
& MPOL_F_ADDR
) {
774 * Do NOT fall back to task policy if the
775 * vma/shared policy at addr is NULL. We
776 * want to return MPOL_DEFAULT in this case.
778 down_read(&mm
->mmap_sem
);
779 vma
= find_vma_intersection(mm
, addr
, addr
+1);
781 up_read(&mm
->mmap_sem
);
784 if (vma
->vm_ops
&& vma
->vm_ops
->get_policy
)
785 pol
= vma
->vm_ops
->get_policy(vma
, addr
);
787 pol
= vma
->vm_policy
;
792 pol
= &default_policy
; /* indicates default behavior */
794 if (flags
& MPOL_F_NODE
) {
795 if (flags
& MPOL_F_ADDR
) {
796 err
= lookup_node(mm
, addr
);
800 } else if (pol
== current
->mempolicy
&&
801 pol
->mode
== MPOL_INTERLEAVE
) {
802 *policy
= current
->il_next
;
808 *policy
= pol
== &default_policy
? MPOL_DEFAULT
:
811 * Internal mempolicy flags must be masked off before exposing
812 * the policy to userspace.
814 *policy
|= (pol
->flags
& MPOL_MODE_FLAGS
);
818 up_read(¤t
->mm
->mmap_sem
);
824 if (mpol_store_user_nodemask(pol
)) {
825 *nmask
= pol
->w
.user_nodemask
;
828 get_policy_nodemask(pol
, nmask
);
829 task_unlock(current
);
836 up_read(¤t
->mm
->mmap_sem
);
840 #ifdef CONFIG_MIGRATION
844 static void migrate_page_add(struct page
*page
, struct list_head
*pagelist
,
848 * Avoid migrating a page that is shared with others.
850 if ((flags
& MPOL_MF_MOVE_ALL
) || page_mapcount(page
) == 1) {
851 if (!isolate_lru_page(page
)) {
852 list_add_tail(&page
->lru
, pagelist
);
853 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
854 page_is_file_cache(page
));
859 static struct page
*new_node_page(struct page
*page
, unsigned long node
, int **x
)
861 return alloc_pages_exact_node(node
, GFP_HIGHUSER_MOVABLE
, 0);
865 * Migrate pages from one node to a target node.
866 * Returns error or the number of pages not migrated.
868 static int migrate_to_node(struct mm_struct
*mm
, int source
, int dest
,
876 node_set(source
, nmask
);
878 check_range(mm
, mm
->mmap
->vm_start
, mm
->task_size
, &nmask
,
879 flags
| MPOL_MF_DISCONTIG_OK
, &pagelist
);
881 if (!list_empty(&pagelist
))
882 err
= migrate_pages(&pagelist
, new_node_page
, dest
, 0);
888 * Move pages between the two nodesets so as to preserve the physical
889 * layout as much as possible.
891 * Returns the number of page that could not be moved.
893 int do_migrate_pages(struct mm_struct
*mm
,
894 const nodemask_t
*from_nodes
, const nodemask_t
*to_nodes
, int flags
)
900 err
= migrate_prep();
904 down_read(&mm
->mmap_sem
);
906 err
= migrate_vmas(mm
, from_nodes
, to_nodes
, flags
);
911 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
912 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
913 * bit in 'tmp', and return that <source, dest> pair for migration.
914 * The pair of nodemasks 'to' and 'from' define the map.
916 * If no pair of bits is found that way, fallback to picking some
917 * pair of 'source' and 'dest' bits that are not the same. If the
918 * 'source' and 'dest' bits are the same, this represents a node
919 * that will be migrating to itself, so no pages need move.
921 * If no bits are left in 'tmp', or if all remaining bits left
922 * in 'tmp' correspond to the same bit in 'to', return false
923 * (nothing left to migrate).
925 * This lets us pick a pair of nodes to migrate between, such that
926 * if possible the dest node is not already occupied by some other
927 * source node, minimizing the risk of overloading the memory on a
928 * node that would happen if we migrated incoming memory to a node
929 * before migrating outgoing memory source that same node.
931 * A single scan of tmp is sufficient. As we go, we remember the
932 * most recent <s, d> pair that moved (s != d). If we find a pair
933 * that not only moved, but what's better, moved to an empty slot
934 * (d is not set in tmp), then we break out then, with that pair.
935 * Otherwise when we finish scannng from_tmp, we at least have the
936 * most recent <s, d> pair that moved. If we get all the way through
937 * the scan of tmp without finding any node that moved, much less
938 * moved to an empty node, then there is nothing left worth migrating.
942 while (!nodes_empty(tmp
)) {
947 for_each_node_mask(s
, tmp
) {
948 d
= node_remap(s
, *from_nodes
, *to_nodes
);
952 source
= s
; /* Node moved. Memorize */
955 /* dest not in remaining from nodes? */
956 if (!node_isset(dest
, tmp
))
962 node_clear(source
, tmp
);
963 err
= migrate_to_node(mm
, source
, dest
, flags
);
970 up_read(&mm
->mmap_sem
);
978 * Allocate a new page for page migration based on vma policy.
979 * Start assuming that page is mapped by vma pointed to by @private.
980 * Search forward from there, if not. N.B., this assumes that the
981 * list of pages handed to migrate_pages()--which is how we get here--
982 * is in virtual address order.
984 static struct page
*new_vma_page(struct page
*page
, unsigned long private, int **x
)
986 struct vm_area_struct
*vma
= (struct vm_area_struct
*)private;
987 unsigned long uninitialized_var(address
);
990 address
= page_address_in_vma(page
, vma
);
991 if (address
!= -EFAULT
)
997 * if !vma, alloc_page_vma() will use task or system default policy
999 return alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
1003 static void migrate_page_add(struct page
*page
, struct list_head
*pagelist
,
1004 unsigned long flags
)
1008 int do_migrate_pages(struct mm_struct
*mm
,
1009 const nodemask_t
*from_nodes
, const nodemask_t
*to_nodes
, int flags
)
1014 static struct page
*new_vma_page(struct page
*page
, unsigned long private, int **x
)
1020 static long do_mbind(unsigned long start
, unsigned long len
,
1021 unsigned short mode
, unsigned short mode_flags
,
1022 nodemask_t
*nmask
, unsigned long flags
)
1024 struct vm_area_struct
*vma
;
1025 struct mm_struct
*mm
= current
->mm
;
1026 struct mempolicy
*new;
1029 LIST_HEAD(pagelist
);
1031 if (flags
& ~(unsigned long)(MPOL_MF_STRICT
|
1032 MPOL_MF_MOVE
| MPOL_MF_MOVE_ALL
))
1034 if ((flags
& MPOL_MF_MOVE_ALL
) && !capable(CAP_SYS_NICE
))
1037 if (start
& ~PAGE_MASK
)
1040 if (mode
== MPOL_DEFAULT
)
1041 flags
&= ~MPOL_MF_STRICT
;
1043 len
= (len
+ PAGE_SIZE
- 1) & PAGE_MASK
;
1051 new = mpol_new(mode
, mode_flags
, nmask
);
1053 return PTR_ERR(new);
1056 * If we are using the default policy then operation
1057 * on discontinuous address spaces is okay after all
1060 flags
|= MPOL_MF_DISCONTIG_OK
;
1062 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1063 start
, start
+ len
, mode
, mode_flags
,
1064 nmask
? nodes_addr(*nmask
)[0] : -1);
1066 if (flags
& (MPOL_MF_MOVE
| MPOL_MF_MOVE_ALL
)) {
1068 err
= migrate_prep();
1073 NODEMASK_SCRATCH(scratch
);
1075 down_write(&mm
->mmap_sem
);
1077 err
= mpol_set_nodemask(new, nmask
, scratch
);
1078 task_unlock(current
);
1080 up_write(&mm
->mmap_sem
);
1083 NODEMASK_SCRATCH_FREE(scratch
);
1088 vma
= check_range(mm
, start
, end
, nmask
,
1089 flags
| MPOL_MF_INVERT
, &pagelist
);
1095 err
= mbind_range(mm
, start
, end
, new);
1097 if (!list_empty(&pagelist
))
1098 nr_failed
= migrate_pages(&pagelist
, new_vma_page
,
1099 (unsigned long)vma
, 0);
1101 if (!err
&& nr_failed
&& (flags
& MPOL_MF_STRICT
))
1104 putback_lru_pages(&pagelist
);
1106 up_write(&mm
->mmap_sem
);
1113 * User space interface with variable sized bitmaps for nodelists.
1116 /* Copy a node mask from user space. */
1117 static int get_nodes(nodemask_t
*nodes
, const unsigned long __user
*nmask
,
1118 unsigned long maxnode
)
1121 unsigned long nlongs
;
1122 unsigned long endmask
;
1125 nodes_clear(*nodes
);
1126 if (maxnode
== 0 || !nmask
)
1128 if (maxnode
> PAGE_SIZE
*BITS_PER_BYTE
)
1131 nlongs
= BITS_TO_LONGS(maxnode
);
1132 if ((maxnode
% BITS_PER_LONG
) == 0)
1135 endmask
= (1UL << (maxnode
% BITS_PER_LONG
)) - 1;
1137 /* When the user specified more nodes than supported just check
1138 if the non supported part is all zero. */
1139 if (nlongs
> BITS_TO_LONGS(MAX_NUMNODES
)) {
1140 if (nlongs
> PAGE_SIZE
/sizeof(long))
1142 for (k
= BITS_TO_LONGS(MAX_NUMNODES
); k
< nlongs
; k
++) {
1144 if (get_user(t
, nmask
+ k
))
1146 if (k
== nlongs
- 1) {
1152 nlongs
= BITS_TO_LONGS(MAX_NUMNODES
);
1156 if (copy_from_user(nodes_addr(*nodes
), nmask
, nlongs
*sizeof(unsigned long)))
1158 nodes_addr(*nodes
)[nlongs
-1] &= endmask
;
1162 /* Copy a kernel node mask to user space */
1163 static int copy_nodes_to_user(unsigned long __user
*mask
, unsigned long maxnode
,
1166 unsigned long copy
= ALIGN(maxnode
-1, 64) / 8;
1167 const int nbytes
= BITS_TO_LONGS(MAX_NUMNODES
) * sizeof(long);
1169 if (copy
> nbytes
) {
1170 if (copy
> PAGE_SIZE
)
1172 if (clear_user((char __user
*)mask
+ nbytes
, copy
- nbytes
))
1176 return copy_to_user(mask
, nodes_addr(*nodes
), copy
) ? -EFAULT
: 0;
1179 SYSCALL_DEFINE6(mbind
, unsigned long, start
, unsigned long, len
,
1180 unsigned long, mode
, unsigned long __user
*, nmask
,
1181 unsigned long, maxnode
, unsigned, flags
)
1185 unsigned short mode_flags
;
1187 mode_flags
= mode
& MPOL_MODE_FLAGS
;
1188 mode
&= ~MPOL_MODE_FLAGS
;
1189 if (mode
>= MPOL_MAX
)
1191 if ((mode_flags
& MPOL_F_STATIC_NODES
) &&
1192 (mode_flags
& MPOL_F_RELATIVE_NODES
))
1194 err
= get_nodes(&nodes
, nmask
, maxnode
);
1197 return do_mbind(start
, len
, mode
, mode_flags
, &nodes
, flags
);
1200 /* Set the process memory policy */
1201 SYSCALL_DEFINE3(set_mempolicy
, int, mode
, unsigned long __user
*, nmask
,
1202 unsigned long, maxnode
)
1206 unsigned short flags
;
1208 flags
= mode
& MPOL_MODE_FLAGS
;
1209 mode
&= ~MPOL_MODE_FLAGS
;
1210 if ((unsigned int)mode
>= MPOL_MAX
)
1212 if ((flags
& MPOL_F_STATIC_NODES
) && (flags
& MPOL_F_RELATIVE_NODES
))
1214 err
= get_nodes(&nodes
, nmask
, maxnode
);
1217 return do_set_mempolicy(mode
, flags
, &nodes
);
1220 SYSCALL_DEFINE4(migrate_pages
, pid_t
, pid
, unsigned long, maxnode
,
1221 const unsigned long __user
*, old_nodes
,
1222 const unsigned long __user
*, new_nodes
)
1224 const struct cred
*cred
= current_cred(), *tcred
;
1225 struct mm_struct
*mm
= NULL
;
1226 struct task_struct
*task
;
1227 nodemask_t task_nodes
;
1231 NODEMASK_SCRATCH(scratch
);
1236 old
= &scratch
->mask1
;
1237 new = &scratch
->mask2
;
1239 err
= get_nodes(old
, old_nodes
, maxnode
);
1243 err
= get_nodes(new, new_nodes
, maxnode
);
1247 /* Find the mm_struct */
1248 read_lock(&tasklist_lock
);
1249 task
= pid
? find_task_by_vpid(pid
) : current
;
1251 read_unlock(&tasklist_lock
);
1255 mm
= get_task_mm(task
);
1256 read_unlock(&tasklist_lock
);
1263 * Check if this process has the right to modify the specified
1264 * process. The right exists if the process has administrative
1265 * capabilities, superuser privileges or the same
1266 * userid as the target process.
1269 tcred
= __task_cred(task
);
1270 if (cred
->euid
!= tcred
->suid
&& cred
->euid
!= tcred
->uid
&&
1271 cred
->uid
!= tcred
->suid
&& cred
->uid
!= tcred
->uid
&&
1272 !capable(CAP_SYS_NICE
)) {
1279 task_nodes
= cpuset_mems_allowed(task
);
1280 /* Is the user allowed to access the target nodes? */
1281 if (!nodes_subset(*new, task_nodes
) && !capable(CAP_SYS_NICE
)) {
1286 if (!nodes_subset(*new, node_states
[N_HIGH_MEMORY
])) {
1291 err
= security_task_movememory(task
);
1295 err
= do_migrate_pages(mm
, old
, new,
1296 capable(CAP_SYS_NICE
) ? MPOL_MF_MOVE_ALL
: MPOL_MF_MOVE
);
1300 NODEMASK_SCRATCH_FREE(scratch
);
1306 /* Retrieve NUMA policy */
1307 SYSCALL_DEFINE5(get_mempolicy
, int __user
*, policy
,
1308 unsigned long __user
*, nmask
, unsigned long, maxnode
,
1309 unsigned long, addr
, unsigned long, flags
)
1312 int uninitialized_var(pval
);
1315 if (nmask
!= NULL
&& maxnode
< MAX_NUMNODES
)
1318 err
= do_get_mempolicy(&pval
, &nodes
, addr
, flags
);
1323 if (policy
&& put_user(pval
, policy
))
1327 err
= copy_nodes_to_user(nmask
, maxnode
, &nodes
);
1332 #ifdef CONFIG_COMPAT
1334 asmlinkage
long compat_sys_get_mempolicy(int __user
*policy
,
1335 compat_ulong_t __user
*nmask
,
1336 compat_ulong_t maxnode
,
1337 compat_ulong_t addr
, compat_ulong_t flags
)
1340 unsigned long __user
*nm
= NULL
;
1341 unsigned long nr_bits
, alloc_size
;
1342 DECLARE_BITMAP(bm
, MAX_NUMNODES
);
1344 nr_bits
= min_t(unsigned long, maxnode
-1, MAX_NUMNODES
);
1345 alloc_size
= ALIGN(nr_bits
, BITS_PER_LONG
) / 8;
1348 nm
= compat_alloc_user_space(alloc_size
);
1350 err
= sys_get_mempolicy(policy
, nm
, nr_bits
+1, addr
, flags
);
1352 if (!err
&& nmask
) {
1353 err
= copy_from_user(bm
, nm
, alloc_size
);
1354 /* ensure entire bitmap is zeroed */
1355 err
|= clear_user(nmask
, ALIGN(maxnode
-1, 8) / 8);
1356 err
|= compat_put_bitmap(nmask
, bm
, nr_bits
);
1362 asmlinkage
long compat_sys_set_mempolicy(int mode
, compat_ulong_t __user
*nmask
,
1363 compat_ulong_t maxnode
)
1366 unsigned long __user
*nm
= NULL
;
1367 unsigned long nr_bits
, alloc_size
;
1368 DECLARE_BITMAP(bm
, MAX_NUMNODES
);
1370 nr_bits
= min_t(unsigned long, maxnode
-1, MAX_NUMNODES
);
1371 alloc_size
= ALIGN(nr_bits
, BITS_PER_LONG
) / 8;
1374 err
= compat_get_bitmap(bm
, nmask
, nr_bits
);
1375 nm
= compat_alloc_user_space(alloc_size
);
1376 err
|= copy_to_user(nm
, bm
, alloc_size
);
1382 return sys_set_mempolicy(mode
, nm
, nr_bits
+1);
1385 asmlinkage
long compat_sys_mbind(compat_ulong_t start
, compat_ulong_t len
,
1386 compat_ulong_t mode
, compat_ulong_t __user
*nmask
,
1387 compat_ulong_t maxnode
, compat_ulong_t flags
)
1390 unsigned long __user
*nm
= NULL
;
1391 unsigned long nr_bits
, alloc_size
;
1394 nr_bits
= min_t(unsigned long, maxnode
-1, MAX_NUMNODES
);
1395 alloc_size
= ALIGN(nr_bits
, BITS_PER_LONG
) / 8;
1398 err
= compat_get_bitmap(nodes_addr(bm
), nmask
, nr_bits
);
1399 nm
= compat_alloc_user_space(alloc_size
);
1400 err
|= copy_to_user(nm
, nodes_addr(bm
), alloc_size
);
1406 return sys_mbind(start
, len
, mode
, nm
, nr_bits
+1, flags
);
1412 * get_vma_policy(@task, @vma, @addr)
1413 * @task - task for fallback if vma policy == default
1414 * @vma - virtual memory area whose policy is sought
1415 * @addr - address in @vma for shared policy lookup
1417 * Returns effective policy for a VMA at specified address.
1418 * Falls back to @task or system default policy, as necessary.
1419 * Current or other task's task mempolicy and non-shared vma policies
1420 * are protected by the task's mmap_sem, which must be held for read by
1422 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1423 * count--added by the get_policy() vm_op, as appropriate--to protect against
1424 * freeing by another task. It is the caller's responsibility to free the
1425 * extra reference for shared policies.
1427 static struct mempolicy
*get_vma_policy(struct task_struct
*task
,
1428 struct vm_area_struct
*vma
, unsigned long addr
)
1430 struct mempolicy
*pol
= task
->mempolicy
;
1433 if (vma
->vm_ops
&& vma
->vm_ops
->get_policy
) {
1434 struct mempolicy
*vpol
= vma
->vm_ops
->get_policy(vma
,
1438 } else if (vma
->vm_policy
)
1439 pol
= vma
->vm_policy
;
1442 pol
= &default_policy
;
1447 * Return a nodemask representing a mempolicy for filtering nodes for
1450 static nodemask_t
*policy_nodemask(gfp_t gfp
, struct mempolicy
*policy
)
1452 /* Lower zones don't get a nodemask applied for MPOL_BIND */
1453 if (unlikely(policy
->mode
== MPOL_BIND
) &&
1454 gfp_zone(gfp
) >= policy_zone
&&
1455 cpuset_nodemask_valid_mems_allowed(&policy
->v
.nodes
))
1456 return &policy
->v
.nodes
;
1461 /* Return a zonelist indicated by gfp for node representing a mempolicy */
1462 static struct zonelist
*policy_zonelist(gfp_t gfp
, struct mempolicy
*policy
)
1464 int nd
= numa_node_id();
1466 switch (policy
->mode
) {
1467 case MPOL_PREFERRED
:
1468 if (!(policy
->flags
& MPOL_F_LOCAL
))
1469 nd
= policy
->v
.preferred_node
;
1473 * Normally, MPOL_BIND allocations are node-local within the
1474 * allowed nodemask. However, if __GFP_THISNODE is set and the
1475 * current node isn't part of the mask, we use the zonelist for
1476 * the first node in the mask instead.
1478 if (unlikely(gfp
& __GFP_THISNODE
) &&
1479 unlikely(!node_isset(nd
, policy
->v
.nodes
)))
1480 nd
= first_node(policy
->v
.nodes
);
1485 return node_zonelist(nd
, gfp
);
1488 /* Do dynamic interleaving for a process */
1489 static unsigned interleave_nodes(struct mempolicy
*policy
)
1492 struct task_struct
*me
= current
;
1495 next
= next_node(nid
, policy
->v
.nodes
);
1496 if (next
>= MAX_NUMNODES
)
1497 next
= first_node(policy
->v
.nodes
);
1498 if (next
< MAX_NUMNODES
)
1504 * Depending on the memory policy provide a node from which to allocate the
1506 * @policy must be protected by freeing by the caller. If @policy is
1507 * the current task's mempolicy, this protection is implicit, as only the
1508 * task can change it's policy. The system default policy requires no
1511 unsigned slab_node(struct mempolicy
*policy
)
1513 if (!policy
|| policy
->flags
& MPOL_F_LOCAL
)
1514 return numa_node_id();
1516 switch (policy
->mode
) {
1517 case MPOL_PREFERRED
:
1519 * handled MPOL_F_LOCAL above
1521 return policy
->v
.preferred_node
;
1523 case MPOL_INTERLEAVE
:
1524 return interleave_nodes(policy
);
1528 * Follow bind policy behavior and start allocation at the
1531 struct zonelist
*zonelist
;
1533 enum zone_type highest_zoneidx
= gfp_zone(GFP_KERNEL
);
1534 zonelist
= &NODE_DATA(numa_node_id())->node_zonelists
[0];
1535 (void)first_zones_zonelist(zonelist
, highest_zoneidx
,
1538 return zone
? zone
->node
: numa_node_id();
1546 /* Do static interleaving for a VMA with known offset. */
1547 static unsigned offset_il_node(struct mempolicy
*pol
,
1548 struct vm_area_struct
*vma
, unsigned long off
)
1550 unsigned nnodes
= nodes_weight(pol
->v
.nodes
);
1556 return numa_node_id();
1557 target
= (unsigned int)off
% nnodes
;
1560 nid
= next_node(nid
, pol
->v
.nodes
);
1562 } while (c
<= target
);
1566 /* Determine a node number for interleave */
1567 static inline unsigned interleave_nid(struct mempolicy
*pol
,
1568 struct vm_area_struct
*vma
, unsigned long addr
, int shift
)
1574 * for small pages, there is no difference between
1575 * shift and PAGE_SHIFT, so the bit-shift is safe.
1576 * for huge pages, since vm_pgoff is in units of small
1577 * pages, we need to shift off the always 0 bits to get
1580 BUG_ON(shift
< PAGE_SHIFT
);
1581 off
= vma
->vm_pgoff
>> (shift
- PAGE_SHIFT
);
1582 off
+= (addr
- vma
->vm_start
) >> shift
;
1583 return offset_il_node(pol
, vma
, off
);
1585 return interleave_nodes(pol
);
1588 #ifdef CONFIG_HUGETLBFS
1590 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1591 * @vma = virtual memory area whose policy is sought
1592 * @addr = address in @vma for shared policy lookup and interleave policy
1593 * @gfp_flags = for requested zone
1594 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1595 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
1597 * Returns a zonelist suitable for a huge page allocation and a pointer
1598 * to the struct mempolicy for conditional unref after allocation.
1599 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1600 * @nodemask for filtering the zonelist.
1602 * Must be protected by get_mems_allowed()
1604 struct zonelist
*huge_zonelist(struct vm_area_struct
*vma
, unsigned long addr
,
1605 gfp_t gfp_flags
, struct mempolicy
**mpol
,
1606 nodemask_t
**nodemask
)
1608 struct zonelist
*zl
;
1610 *mpol
= get_vma_policy(current
, vma
, addr
);
1611 *nodemask
= NULL
; /* assume !MPOL_BIND */
1613 if (unlikely((*mpol
)->mode
== MPOL_INTERLEAVE
)) {
1614 zl
= node_zonelist(interleave_nid(*mpol
, vma
, addr
,
1615 huge_page_shift(hstate_vma(vma
))), gfp_flags
);
1617 zl
= policy_zonelist(gfp_flags
, *mpol
);
1618 if ((*mpol
)->mode
== MPOL_BIND
)
1619 *nodemask
= &(*mpol
)->v
.nodes
;
1625 * init_nodemask_of_mempolicy
1627 * If the current task's mempolicy is "default" [NULL], return 'false'
1628 * to indicate default policy. Otherwise, extract the policy nodemask
1629 * for 'bind' or 'interleave' policy into the argument nodemask, or
1630 * initialize the argument nodemask to contain the single node for
1631 * 'preferred' or 'local' policy and return 'true' to indicate presence
1632 * of non-default mempolicy.
1634 * We don't bother with reference counting the mempolicy [mpol_get/put]
1635 * because the current task is examining it's own mempolicy and a task's
1636 * mempolicy is only ever changed by the task itself.
1638 * N.B., it is the caller's responsibility to free a returned nodemask.
1640 bool init_nodemask_of_mempolicy(nodemask_t
*mask
)
1642 struct mempolicy
*mempolicy
;
1645 if (!(mask
&& current
->mempolicy
))
1649 mempolicy
= current
->mempolicy
;
1650 switch (mempolicy
->mode
) {
1651 case MPOL_PREFERRED
:
1652 if (mempolicy
->flags
& MPOL_F_LOCAL
)
1653 nid
= numa_node_id();
1655 nid
= mempolicy
->v
.preferred_node
;
1656 init_nodemask_of_node(mask
, nid
);
1661 case MPOL_INTERLEAVE
:
1662 *mask
= mempolicy
->v
.nodes
;
1668 task_unlock(current
);
1675 * mempolicy_nodemask_intersects
1677 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1678 * policy. Otherwise, check for intersection between mask and the policy
1679 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1680 * policy, always return true since it may allocate elsewhere on fallback.
1682 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1684 bool mempolicy_nodemask_intersects(struct task_struct
*tsk
,
1685 const nodemask_t
*mask
)
1687 struct mempolicy
*mempolicy
;
1693 mempolicy
= tsk
->mempolicy
;
1697 switch (mempolicy
->mode
) {
1698 case MPOL_PREFERRED
:
1700 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1701 * allocate from, they may fallback to other nodes when oom.
1702 * Thus, it's possible for tsk to have allocated memory from
1707 case MPOL_INTERLEAVE
:
1708 ret
= nodes_intersects(mempolicy
->v
.nodes
, *mask
);
1718 /* Allocate a page in interleaved policy.
1719 Own path because it needs to do special accounting. */
1720 static struct page
*alloc_page_interleave(gfp_t gfp
, unsigned order
,
1723 struct zonelist
*zl
;
1726 zl
= node_zonelist(nid
, gfp
);
1727 page
= __alloc_pages(gfp
, order
, zl
);
1728 if (page
&& page_zone(page
) == zonelist_zone(&zl
->_zonerefs
[0]))
1729 inc_zone_page_state(page
, NUMA_INTERLEAVE_HIT
);
1734 * alloc_page_vma - Allocate a page for a VMA.
1737 * %GFP_USER user allocation.
1738 * %GFP_KERNEL kernel allocations,
1739 * %GFP_HIGHMEM highmem/user allocations,
1740 * %GFP_FS allocation should not call back into a file system.
1741 * %GFP_ATOMIC don't sleep.
1743 * @vma: Pointer to VMA or NULL if not available.
1744 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1746 * This function allocates a page from the kernel page pool and applies
1747 * a NUMA policy associated with the VMA or the current process.
1748 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1749 * mm_struct of the VMA to prevent it from going away. Should be used for
1750 * all allocations for pages that will be mapped into
1751 * user space. Returns NULL when no page can be allocated.
1753 * Should be called with the mm_sem of the vma hold.
1756 alloc_page_vma(gfp_t gfp
, struct vm_area_struct
*vma
, unsigned long addr
)
1758 struct mempolicy
*pol
= get_vma_policy(current
, vma
, addr
);
1759 struct zonelist
*zl
;
1763 if (unlikely(pol
->mode
== MPOL_INTERLEAVE
)) {
1766 nid
= interleave_nid(pol
, vma
, addr
, PAGE_SHIFT
);
1768 page
= alloc_page_interleave(gfp
, 0, nid
);
1772 zl
= policy_zonelist(gfp
, pol
);
1773 if (unlikely(mpol_needs_cond_ref(pol
))) {
1775 * slow path: ref counted shared policy
1777 struct page
*page
= __alloc_pages_nodemask(gfp
, 0,
1778 zl
, policy_nodemask(gfp
, pol
));
1784 * fast path: default or task policy
1786 page
= __alloc_pages_nodemask(gfp
, 0, zl
, policy_nodemask(gfp
, pol
));
1792 * alloc_pages_current - Allocate pages.
1795 * %GFP_USER user allocation,
1796 * %GFP_KERNEL kernel allocation,
1797 * %GFP_HIGHMEM highmem allocation,
1798 * %GFP_FS don't call back into a file system.
1799 * %GFP_ATOMIC don't sleep.
1800 * @order: Power of two of allocation size in pages. 0 is a single page.
1802 * Allocate a page from the kernel page pool. When not in
1803 * interrupt context and apply the current process NUMA policy.
1804 * Returns NULL when no page can be allocated.
1806 * Don't call cpuset_update_task_memory_state() unless
1807 * 1) it's ok to take cpuset_sem (can WAIT), and
1808 * 2) allocating for current task (not interrupt).
1810 struct page
*alloc_pages_current(gfp_t gfp
, unsigned order
)
1812 struct mempolicy
*pol
= current
->mempolicy
;
1815 if (!pol
|| in_interrupt() || (gfp
& __GFP_THISNODE
))
1816 pol
= &default_policy
;
1820 * No reference counting needed for current->mempolicy
1821 * nor system default_policy
1823 if (pol
->mode
== MPOL_INTERLEAVE
)
1824 page
= alloc_page_interleave(gfp
, order
, interleave_nodes(pol
));
1826 page
= __alloc_pages_nodemask(gfp
, order
,
1827 policy_zonelist(gfp
, pol
), policy_nodemask(gfp
, pol
));
1831 EXPORT_SYMBOL(alloc_pages_current
);
1834 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
1835 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1836 * with the mems_allowed returned by cpuset_mems_allowed(). This
1837 * keeps mempolicies cpuset relative after its cpuset moves. See
1838 * further kernel/cpuset.c update_nodemask().
1840 * current's mempolicy may be rebinded by the other task(the task that changes
1841 * cpuset's mems), so we needn't do rebind work for current task.
1844 /* Slow path of a mempolicy duplicate */
1845 struct mempolicy
*__mpol_dup(struct mempolicy
*old
)
1847 struct mempolicy
*new = kmem_cache_alloc(policy_cache
, GFP_KERNEL
);
1850 return ERR_PTR(-ENOMEM
);
1852 /* task's mempolicy is protected by alloc_lock */
1853 if (old
== current
->mempolicy
) {
1856 task_unlock(current
);
1861 if (current_cpuset_is_being_rebound()) {
1862 nodemask_t mems
= cpuset_mems_allowed(current
);
1863 if (new->flags
& MPOL_F_REBINDING
)
1864 mpol_rebind_policy(new, &mems
, MPOL_REBIND_STEP2
);
1866 mpol_rebind_policy(new, &mems
, MPOL_REBIND_ONCE
);
1869 atomic_set(&new->refcnt
, 1);
1874 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1875 * eliminate the * MPOL_F_* flags that require conditional ref and
1876 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1877 * after return. Use the returned value.
1879 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1880 * policy lookup, even if the policy needs/has extra ref on lookup.
1881 * shmem_readahead needs this.
1883 struct mempolicy
*__mpol_cond_copy(struct mempolicy
*tompol
,
1884 struct mempolicy
*frompol
)
1886 if (!mpol_needs_cond_ref(frompol
))
1890 tompol
->flags
&= ~MPOL_F_SHARED
; /* copy doesn't need unref */
1891 __mpol_put(frompol
);
1895 /* Slow path of a mempolicy comparison */
1896 int __mpol_equal(struct mempolicy
*a
, struct mempolicy
*b
)
1900 if (a
->mode
!= b
->mode
)
1902 if (a
->flags
!= b
->flags
)
1904 if (mpol_store_user_nodemask(a
))
1905 if (!nodes_equal(a
->w
.user_nodemask
, b
->w
.user_nodemask
))
1911 case MPOL_INTERLEAVE
:
1912 return nodes_equal(a
->v
.nodes
, b
->v
.nodes
);
1913 case MPOL_PREFERRED
:
1914 return a
->v
.preferred_node
== b
->v
.preferred_node
&&
1915 a
->flags
== b
->flags
;
1923 * Shared memory backing store policy support.
1925 * Remember policies even when nobody has shared memory mapped.
1926 * The policies are kept in Red-Black tree linked from the inode.
1927 * They are protected by the sp->lock spinlock, which should be held
1928 * for any accesses to the tree.
1931 /* lookup first element intersecting start-end */
1932 /* Caller holds sp->lock */
1933 static struct sp_node
*
1934 sp_lookup(struct shared_policy
*sp
, unsigned long start
, unsigned long end
)
1936 struct rb_node
*n
= sp
->root
.rb_node
;
1939 struct sp_node
*p
= rb_entry(n
, struct sp_node
, nd
);
1941 if (start
>= p
->end
)
1943 else if (end
<= p
->start
)
1951 struct sp_node
*w
= NULL
;
1952 struct rb_node
*prev
= rb_prev(n
);
1955 w
= rb_entry(prev
, struct sp_node
, nd
);
1956 if (w
->end
<= start
)
1960 return rb_entry(n
, struct sp_node
, nd
);
1963 /* Insert a new shared policy into the list. */
1964 /* Caller holds sp->lock */
1965 static void sp_insert(struct shared_policy
*sp
, struct sp_node
*new)
1967 struct rb_node
**p
= &sp
->root
.rb_node
;
1968 struct rb_node
*parent
= NULL
;
1973 nd
= rb_entry(parent
, struct sp_node
, nd
);
1974 if (new->start
< nd
->start
)
1976 else if (new->end
> nd
->end
)
1977 p
= &(*p
)->rb_right
;
1981 rb_link_node(&new->nd
, parent
, p
);
1982 rb_insert_color(&new->nd
, &sp
->root
);
1983 pr_debug("inserting %lx-%lx: %d\n", new->start
, new->end
,
1984 new->policy
? new->policy
->mode
: 0);
1987 /* Find shared policy intersecting idx */
1989 mpol_shared_policy_lookup(struct shared_policy
*sp
, unsigned long idx
)
1991 struct mempolicy
*pol
= NULL
;
1994 if (!sp
->root
.rb_node
)
1996 spin_lock(&sp
->lock
);
1997 sn
= sp_lookup(sp
, idx
, idx
+1);
1999 mpol_get(sn
->policy
);
2002 spin_unlock(&sp
->lock
);
2006 static void sp_delete(struct shared_policy
*sp
, struct sp_node
*n
)
2008 pr_debug("deleting %lx-l%lx\n", n
->start
, n
->end
);
2009 rb_erase(&n
->nd
, &sp
->root
);
2010 mpol_put(n
->policy
);
2011 kmem_cache_free(sn_cache
, n
);
2014 static struct sp_node
*sp_alloc(unsigned long start
, unsigned long end
,
2015 struct mempolicy
*pol
)
2017 struct sp_node
*n
= kmem_cache_alloc(sn_cache
, GFP_KERNEL
);
2024 pol
->flags
|= MPOL_F_SHARED
; /* for unref */
2029 /* Replace a policy range. */
2030 static int shared_policy_replace(struct shared_policy
*sp
, unsigned long start
,
2031 unsigned long end
, struct sp_node
*new)
2033 struct sp_node
*n
, *new2
= NULL
;
2036 spin_lock(&sp
->lock
);
2037 n
= sp_lookup(sp
, start
, end
);
2038 /* Take care of old policies in the same range. */
2039 while (n
&& n
->start
< end
) {
2040 struct rb_node
*next
= rb_next(&n
->nd
);
2041 if (n
->start
>= start
) {
2047 /* Old policy spanning whole new range. */
2050 spin_unlock(&sp
->lock
);
2051 new2
= sp_alloc(end
, n
->end
, n
->policy
);
2057 sp_insert(sp
, new2
);
2065 n
= rb_entry(next
, struct sp_node
, nd
);
2069 spin_unlock(&sp
->lock
);
2071 mpol_put(new2
->policy
);
2072 kmem_cache_free(sn_cache
, new2
);
2078 * mpol_shared_policy_init - initialize shared policy for inode
2079 * @sp: pointer to inode shared policy
2080 * @mpol: struct mempolicy to install
2082 * Install non-NULL @mpol in inode's shared policy rb-tree.
2083 * On entry, the current task has a reference on a non-NULL @mpol.
2084 * This must be released on exit.
2085 * This is called at get_inode() calls and we can use GFP_KERNEL.
2087 void mpol_shared_policy_init(struct shared_policy
*sp
, struct mempolicy
*mpol
)
2091 sp
->root
= RB_ROOT
; /* empty tree == default mempolicy */
2092 spin_lock_init(&sp
->lock
);
2095 struct vm_area_struct pvma
;
2096 struct mempolicy
*new;
2097 NODEMASK_SCRATCH(scratch
);
2101 /* contextualize the tmpfs mount point mempolicy */
2102 new = mpol_new(mpol
->mode
, mpol
->flags
, &mpol
->w
.user_nodemask
);
2104 goto free_scratch
; /* no valid nodemask intersection */
2107 ret
= mpol_set_nodemask(new, &mpol
->w
.user_nodemask
, scratch
);
2108 task_unlock(current
);
2112 /* Create pseudo-vma that contains just the policy */
2113 memset(&pvma
, 0, sizeof(struct vm_area_struct
));
2114 pvma
.vm_end
= TASK_SIZE
; /* policy covers entire file */
2115 mpol_set_shared_policy(sp
, &pvma
, new); /* adds ref */
2118 mpol_put(new); /* drop initial ref */
2120 NODEMASK_SCRATCH_FREE(scratch
);
2122 mpol_put(mpol
); /* drop our incoming ref on sb mpol */
2126 int mpol_set_shared_policy(struct shared_policy
*info
,
2127 struct vm_area_struct
*vma
, struct mempolicy
*npol
)
2130 struct sp_node
*new = NULL
;
2131 unsigned long sz
= vma_pages(vma
);
2133 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
2135 sz
, npol
? npol
->mode
: -1,
2136 npol
? npol
->flags
: -1,
2137 npol
? nodes_addr(npol
->v
.nodes
)[0] : -1);
2140 new = sp_alloc(vma
->vm_pgoff
, vma
->vm_pgoff
+ sz
, npol
);
2144 err
= shared_policy_replace(info
, vma
->vm_pgoff
, vma
->vm_pgoff
+sz
, new);
2146 kmem_cache_free(sn_cache
, new);
2150 /* Free a backing policy store on inode delete. */
2151 void mpol_free_shared_policy(struct shared_policy
*p
)
2154 struct rb_node
*next
;
2156 if (!p
->root
.rb_node
)
2158 spin_lock(&p
->lock
);
2159 next
= rb_first(&p
->root
);
2161 n
= rb_entry(next
, struct sp_node
, nd
);
2162 next
= rb_next(&n
->nd
);
2163 rb_erase(&n
->nd
, &p
->root
);
2164 mpol_put(n
->policy
);
2165 kmem_cache_free(sn_cache
, n
);
2167 spin_unlock(&p
->lock
);
2170 /* assumes fs == KERNEL_DS */
2171 void __init
numa_policy_init(void)
2173 nodemask_t interleave_nodes
;
2174 unsigned long largest
= 0;
2175 int nid
, prefer
= 0;
2177 policy_cache
= kmem_cache_create("numa_policy",
2178 sizeof(struct mempolicy
),
2179 0, SLAB_PANIC
, NULL
);
2181 sn_cache
= kmem_cache_create("shared_policy_node",
2182 sizeof(struct sp_node
),
2183 0, SLAB_PANIC
, NULL
);
2186 * Set interleaving policy for system init. Interleaving is only
2187 * enabled across suitably sized nodes (default is >= 16MB), or
2188 * fall back to the largest node if they're all smaller.
2190 nodes_clear(interleave_nodes
);
2191 for_each_node_state(nid
, N_HIGH_MEMORY
) {
2192 unsigned long total_pages
= node_present_pages(nid
);
2194 /* Preserve the largest node */
2195 if (largest
< total_pages
) {
2196 largest
= total_pages
;
2200 /* Interleave this node? */
2201 if ((total_pages
<< PAGE_SHIFT
) >= (16 << 20))
2202 node_set(nid
, interleave_nodes
);
2205 /* All too small, use the largest */
2206 if (unlikely(nodes_empty(interleave_nodes
)))
2207 node_set(prefer
, interleave_nodes
);
2209 if (do_set_mempolicy(MPOL_INTERLEAVE
, 0, &interleave_nodes
))
2210 printk("numa_policy_init: interleaving failed\n");
2213 /* Reset policy of current process to default */
2214 void numa_default_policy(void)
2216 do_set_mempolicy(MPOL_DEFAULT
, 0, NULL
);
2220 * Parse and format mempolicy from/to strings
2224 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
2225 * Used only for mpol_parse_str() and mpol_to_str()
2227 #define MPOL_LOCAL MPOL_MAX
2228 static const char * const policy_modes
[] =
2230 [MPOL_DEFAULT
] = "default",
2231 [MPOL_PREFERRED
] = "prefer",
2232 [MPOL_BIND
] = "bind",
2233 [MPOL_INTERLEAVE
] = "interleave",
2234 [MPOL_LOCAL
] = "local"
2240 * mpol_parse_str - parse string to mempolicy
2241 * @str: string containing mempolicy to parse
2242 * @mpol: pointer to struct mempolicy pointer, returned on success.
2243 * @no_context: flag whether to "contextualize" the mempolicy
2246 * <mode>[=<flags>][:<nodelist>]
2248 * if @no_context is true, save the input nodemask in w.user_nodemask in
2249 * the returned mempolicy. This will be used to "clone" the mempolicy in
2250 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2251 * mount option. Note that if 'static' or 'relative' mode flags were
2252 * specified, the input nodemask will already have been saved. Saving
2253 * it again is redundant, but safe.
2255 * On success, returns 0, else 1
2257 int mpol_parse_str(char *str
, struct mempolicy
**mpol
, int no_context
)
2259 struct mempolicy
*new = NULL
;
2260 unsigned short mode
;
2261 unsigned short uninitialized_var(mode_flags
);
2263 char *nodelist
= strchr(str
, ':');
2264 char *flags
= strchr(str
, '=');
2268 /* NUL-terminate mode or flags string */
2270 if (nodelist_parse(nodelist
, nodes
))
2272 if (!nodes_subset(nodes
, node_states
[N_HIGH_MEMORY
]))
2278 *flags
++ = '\0'; /* terminate mode string */
2280 for (mode
= 0; mode
<= MPOL_LOCAL
; mode
++) {
2281 if (!strcmp(str
, policy_modes
[mode
])) {
2285 if (mode
> MPOL_LOCAL
)
2289 case MPOL_PREFERRED
:
2291 * Insist on a nodelist of one node only
2294 char *rest
= nodelist
;
2295 while (isdigit(*rest
))
2301 case MPOL_INTERLEAVE
:
2303 * Default to online nodes with memory if no nodelist
2306 nodes
= node_states
[N_HIGH_MEMORY
];
2310 * Don't allow a nodelist; mpol_new() checks flags
2314 mode
= MPOL_PREFERRED
;
2318 * Insist on a empty nodelist
2325 * Insist on a nodelist
2334 * Currently, we only support two mutually exclusive
2337 if (!strcmp(flags
, "static"))
2338 mode_flags
|= MPOL_F_STATIC_NODES
;
2339 else if (!strcmp(flags
, "relative"))
2340 mode_flags
|= MPOL_F_RELATIVE_NODES
;
2345 new = mpol_new(mode
, mode_flags
, &nodes
);
2350 /* save for contextualization */
2351 new->w
.user_nodemask
= nodes
;
2354 NODEMASK_SCRATCH(scratch
);
2357 ret
= mpol_set_nodemask(new, &nodes
, scratch
);
2358 task_unlock(current
);
2361 NODEMASK_SCRATCH_FREE(scratch
);
2370 /* Restore string for error message */
2379 #endif /* CONFIG_TMPFS */
2382 * mpol_to_str - format a mempolicy structure for printing
2383 * @buffer: to contain formatted mempolicy string
2384 * @maxlen: length of @buffer
2385 * @pol: pointer to mempolicy to be formatted
2386 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2388 * Convert a mempolicy into a string.
2389 * Returns the number of characters in buffer (if positive)
2390 * or an error (negative)
2392 int mpol_to_str(char *buffer
, int maxlen
, struct mempolicy
*pol
, int no_context
)
2397 unsigned short mode
;
2398 unsigned short flags
= pol
? pol
->flags
: 0;
2401 * Sanity check: room for longest mode, flag and some nodes
2403 VM_BUG_ON(maxlen
< strlen("interleave") + strlen("relative") + 16);
2405 if (!pol
|| pol
== &default_policy
)
2406 mode
= MPOL_DEFAULT
;
2415 case MPOL_PREFERRED
:
2417 if (flags
& MPOL_F_LOCAL
)
2418 mode
= MPOL_LOCAL
; /* pseudo-policy */
2420 node_set(pol
->v
.preferred_node
, nodes
);
2425 case MPOL_INTERLEAVE
:
2427 nodes
= pol
->w
.user_nodemask
;
2429 nodes
= pol
->v
.nodes
;
2436 l
= strlen(policy_modes
[mode
]);
2437 if (buffer
+ maxlen
< p
+ l
+ 1)
2440 strcpy(p
, policy_modes
[mode
]);
2443 if (flags
& MPOL_MODE_FLAGS
) {
2444 if (buffer
+ maxlen
< p
+ 2)
2449 * Currently, the only defined flags are mutually exclusive
2451 if (flags
& MPOL_F_STATIC_NODES
)
2452 p
+= snprintf(p
, buffer
+ maxlen
- p
, "static");
2453 else if (flags
& MPOL_F_RELATIVE_NODES
)
2454 p
+= snprintf(p
, buffer
+ maxlen
- p
, "relative");
2457 if (!nodes_empty(nodes
)) {
2458 if (buffer
+ maxlen
< p
+ 2)
2461 p
+= nodelist_scnprintf(p
, buffer
+ maxlen
- p
, nodes
);
2467 unsigned long pages
;
2469 unsigned long active
;
2470 unsigned long writeback
;
2471 unsigned long mapcount_max
;
2472 unsigned long dirty
;
2473 unsigned long swapcache
;
2474 unsigned long node
[MAX_NUMNODES
];
2477 static void gather_stats(struct page
*page
, void *private, int pte_dirty
)
2479 struct numa_maps
*md
= private;
2480 int count
= page_mapcount(page
);
2483 if (pte_dirty
|| PageDirty(page
))
2486 if (PageSwapCache(page
))
2489 if (PageActive(page
) || PageUnevictable(page
))
2492 if (PageWriteback(page
))
2498 if (count
> md
->mapcount_max
)
2499 md
->mapcount_max
= count
;
2501 md
->node
[page_to_nid(page
)]++;
2504 #ifdef CONFIG_HUGETLB_PAGE
2505 static void check_huge_range(struct vm_area_struct
*vma
,
2506 unsigned long start
, unsigned long end
,
2507 struct numa_maps
*md
)
2511 struct hstate
*h
= hstate_vma(vma
);
2512 unsigned long sz
= huge_page_size(h
);
2514 for (addr
= start
; addr
< end
; addr
+= sz
) {
2515 pte_t
*ptep
= huge_pte_offset(vma
->vm_mm
,
2516 addr
& huge_page_mask(h
));
2526 page
= pte_page(pte
);
2530 gather_stats(page
, md
, pte_dirty(*ptep
));
2534 static inline void check_huge_range(struct vm_area_struct
*vma
,
2535 unsigned long start
, unsigned long end
,
2536 struct numa_maps
*md
)
2542 * Display pages allocated per node and memory policy via /proc.
2544 int show_numa_map(struct seq_file
*m
, void *v
)
2546 struct proc_maps_private
*priv
= m
->private;
2547 struct vm_area_struct
*vma
= v
;
2548 struct numa_maps
*md
;
2549 struct file
*file
= vma
->vm_file
;
2550 struct mm_struct
*mm
= vma
->vm_mm
;
2551 struct mempolicy
*pol
;
2558 md
= kzalloc(sizeof(struct numa_maps
), GFP_KERNEL
);
2562 pol
= get_vma_policy(priv
->task
, vma
, vma
->vm_start
);
2563 mpol_to_str(buffer
, sizeof(buffer
), pol
, 0);
2566 seq_printf(m
, "%08lx %s", vma
->vm_start
, buffer
);
2569 seq_printf(m
, " file=");
2570 seq_path(m
, &file
->f_path
, "\n\t= ");
2571 } else if (vma
->vm_start
<= mm
->brk
&& vma
->vm_end
>= mm
->start_brk
) {
2572 seq_printf(m
, " heap");
2573 } else if (vma
->vm_start
<= mm
->start_stack
&&
2574 vma
->vm_end
>= mm
->start_stack
) {
2575 seq_printf(m
, " stack");
2578 if (is_vm_hugetlb_page(vma
)) {
2579 check_huge_range(vma
, vma
->vm_start
, vma
->vm_end
, md
);
2580 seq_printf(m
, " huge");
2582 check_pgd_range(vma
, vma
->vm_start
, vma
->vm_end
,
2583 &node_states
[N_HIGH_MEMORY
], MPOL_MF_STATS
, md
);
2590 seq_printf(m
," anon=%lu",md
->anon
);
2593 seq_printf(m
," dirty=%lu",md
->dirty
);
2595 if (md
->pages
!= md
->anon
&& md
->pages
!= md
->dirty
)
2596 seq_printf(m
, " mapped=%lu", md
->pages
);
2598 if (md
->mapcount_max
> 1)
2599 seq_printf(m
, " mapmax=%lu", md
->mapcount_max
);
2602 seq_printf(m
," swapcache=%lu", md
->swapcache
);
2604 if (md
->active
< md
->pages
&& !is_vm_hugetlb_page(vma
))
2605 seq_printf(m
," active=%lu", md
->active
);
2608 seq_printf(m
," writeback=%lu", md
->writeback
);
2610 for_each_node_state(n
, N_HIGH_MEMORY
)
2612 seq_printf(m
, " N%d=%lu", n
, md
->node
[n
]);
2617 if (m
->count
< m
->size
)
2618 m
->version
= (vma
!= priv
->tail_vma
) ? vma
->vm_start
: 0;