4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8 * Numa awareness, Christoph Lameter, SGI, June 2005
11 #include <linux/vmalloc.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched/signal.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/set_memory.h>
22 #include <linux/debugobjects.h>
23 #include <linux/kallsyms.h>
24 #include <linux/list.h>
25 #include <linux/notifier.h>
26 #include <linux/rbtree.h>
27 #include <linux/radix-tree.h>
28 #include <linux/rcupdate.h>
29 #include <linux/pfn.h>
30 #include <linux/kmemleak.h>
31 #include <linux/atomic.h>
32 #include <linux/compiler.h>
33 #include <linux/llist.h>
34 #include <linux/bitops.h>
35 #include <linux/rbtree_augmented.h>
37 #include <linux/uaccess.h>
38 #include <asm/tlbflush.h>
39 #include <asm/shmparam.h>
43 struct vfree_deferred
{
44 struct llist_head list
;
45 struct work_struct wq
;
47 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
49 static void __vunmap(const void *, int);
51 static void free_work(struct work_struct
*w
)
53 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
54 struct llist_node
*t
, *llnode
;
56 llist_for_each_safe(llnode
, t
, llist_del_all(&p
->list
))
57 __vunmap((void *)llnode
, 1);
60 /*** Page table manipulation functions ***/
62 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
66 pte
= pte_offset_kernel(pmd
, addr
);
68 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
69 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
70 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
73 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
78 pmd
= pmd_offset(pud
, addr
);
80 next
= pmd_addr_end(addr
, end
);
81 if (pmd_clear_huge(pmd
))
83 if (pmd_none_or_clear_bad(pmd
))
85 vunmap_pte_range(pmd
, addr
, next
);
86 } while (pmd
++, addr
= next
, addr
!= end
);
89 static void vunmap_pud_range(p4d_t
*p4d
, unsigned long addr
, unsigned long end
)
94 pud
= pud_offset(p4d
, addr
);
96 next
= pud_addr_end(addr
, end
);
97 if (pud_clear_huge(pud
))
99 if (pud_none_or_clear_bad(pud
))
101 vunmap_pmd_range(pud
, addr
, next
);
102 } while (pud
++, addr
= next
, addr
!= end
);
105 static void vunmap_p4d_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
110 p4d
= p4d_offset(pgd
, addr
);
112 next
= p4d_addr_end(addr
, end
);
113 if (p4d_clear_huge(p4d
))
115 if (p4d_none_or_clear_bad(p4d
))
117 vunmap_pud_range(p4d
, addr
, next
);
118 } while (p4d
++, addr
= next
, addr
!= end
);
121 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
127 pgd
= pgd_offset_k(addr
);
129 next
= pgd_addr_end(addr
, end
);
130 if (pgd_none_or_clear_bad(pgd
))
132 vunmap_p4d_range(pgd
, addr
, next
);
133 } while (pgd
++, addr
= next
, addr
!= end
);
136 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
137 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
142 * nr is a running index into the array which helps higher level
143 * callers keep track of where we're up to.
146 pte
= pte_alloc_kernel(pmd
, addr
);
150 struct page
*page
= pages
[*nr
];
152 if (WARN_ON(!pte_none(*pte
)))
156 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
158 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
162 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
163 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
168 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
172 next
= pmd_addr_end(addr
, end
);
173 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
175 } while (pmd
++, addr
= next
, addr
!= end
);
179 static int vmap_pud_range(p4d_t
*p4d
, unsigned long addr
,
180 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
185 pud
= pud_alloc(&init_mm
, p4d
, addr
);
189 next
= pud_addr_end(addr
, end
);
190 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
192 } while (pud
++, addr
= next
, addr
!= end
);
196 static int vmap_p4d_range(pgd_t
*pgd
, unsigned long addr
,
197 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
202 p4d
= p4d_alloc(&init_mm
, pgd
, addr
);
206 next
= p4d_addr_end(addr
, end
);
207 if (vmap_pud_range(p4d
, addr
, next
, prot
, pages
, nr
))
209 } while (p4d
++, addr
= next
, addr
!= end
);
214 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
215 * will have pfns corresponding to the "pages" array.
217 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
219 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
220 pgprot_t prot
, struct page
**pages
)
224 unsigned long addr
= start
;
229 pgd
= pgd_offset_k(addr
);
231 next
= pgd_addr_end(addr
, end
);
232 err
= vmap_p4d_range(pgd
, addr
, next
, prot
, pages
, &nr
);
235 } while (pgd
++, addr
= next
, addr
!= end
);
240 static int vmap_page_range(unsigned long start
, unsigned long end
,
241 pgprot_t prot
, struct page
**pages
)
245 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
246 flush_cache_vmap(start
, end
);
250 int is_vmalloc_or_module_addr(const void *x
)
253 * ARM, x86-64 and sparc64 put modules in a special place,
254 * and fall back on vmalloc() if that fails. Others
255 * just put it in the vmalloc space.
257 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
258 unsigned long addr
= (unsigned long)x
;
259 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
262 return is_vmalloc_addr(x
);
266 * Walk a vmap address to the struct page it maps.
268 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
270 unsigned long addr
= (unsigned long) vmalloc_addr
;
271 struct page
*page
= NULL
;
272 pgd_t
*pgd
= pgd_offset_k(addr
);
279 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
280 * architectures that do not vmalloc module space
282 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
286 p4d
= p4d_offset(pgd
, addr
);
289 pud
= pud_offset(p4d
, addr
);
292 * Don't dereference bad PUD or PMD (below) entries. This will also
293 * identify huge mappings, which we may encounter on architectures
294 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
295 * identified as vmalloc addresses by is_vmalloc_addr(), but are
296 * not [unambiguously] associated with a struct page, so there is
297 * no correct value to return for them.
299 WARN_ON_ONCE(pud_bad(*pud
));
300 if (pud_none(*pud
) || pud_bad(*pud
))
302 pmd
= pmd_offset(pud
, addr
);
303 WARN_ON_ONCE(pmd_bad(*pmd
));
304 if (pmd_none(*pmd
) || pmd_bad(*pmd
))
307 ptep
= pte_offset_map(pmd
, addr
);
309 if (pte_present(pte
))
310 page
= pte_page(pte
);
314 EXPORT_SYMBOL(vmalloc_to_page
);
317 * Map a vmalloc()-space virtual address to the physical page frame number.
319 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
321 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
323 EXPORT_SYMBOL(vmalloc_to_pfn
);
326 /*** Global kva allocator ***/
328 #define VM_LAZY_FREE 0x02
329 #define VM_VM_AREA 0x04
331 static DEFINE_SPINLOCK(vmap_area_lock
);
332 /* Export for kexec only */
333 LIST_HEAD(vmap_area_list
);
334 static LLIST_HEAD(vmap_purge_list
);
335 static struct rb_root vmap_area_root
= RB_ROOT
;
336 static bool vmap_initialized __read_mostly
;
339 * This kmem_cache is used for vmap_area objects. Instead of
340 * allocating from slab we reuse an object from this cache to
341 * make things faster. Especially in "no edge" splitting of
344 static struct kmem_cache
*vmap_area_cachep
;
347 * This linked list is used in pair with free_vmap_area_root.
348 * It gives O(1) access to prev/next to perform fast coalescing.
350 static LIST_HEAD(free_vmap_area_list
);
353 * This augment red-black tree represents the free vmap space.
354 * All vmap_area objects in this tree are sorted by va->va_start
355 * address. It is used for allocation and merging when a vmap
356 * object is released.
358 * Each vmap_area node contains a maximum available free block
359 * of its sub-tree, right or left. Therefore it is possible to
360 * find a lowest match of free area.
362 static struct rb_root free_vmap_area_root
= RB_ROOT
;
364 static __always_inline
unsigned long
365 va_size(struct vmap_area
*va
)
367 return (va
->va_end
- va
->va_start
);
370 static __always_inline
unsigned long
371 get_subtree_max_size(struct rb_node
*node
)
373 struct vmap_area
*va
;
375 va
= rb_entry_safe(node
, struct vmap_area
, rb_node
);
376 return va
? va
->subtree_max_size
: 0;
380 * Gets called when remove the node and rotate.
382 static __always_inline
unsigned long
383 compute_subtree_max_size(struct vmap_area
*va
)
385 return max3(va_size(va
),
386 get_subtree_max_size(va
->rb_node
.rb_left
),
387 get_subtree_max_size(va
->rb_node
.rb_right
));
390 RB_DECLARE_CALLBACKS(static, free_vmap_area_rb_augment_cb
,
391 struct vmap_area
, rb_node
, unsigned long, subtree_max_size
,
392 compute_subtree_max_size
)
394 static void purge_vmap_area_lazy(void);
395 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list
);
396 static unsigned long lazy_max_pages(void);
398 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
400 struct rb_node
*n
= vmap_area_root
.rb_node
;
403 struct vmap_area
*va
;
405 va
= rb_entry(n
, struct vmap_area
, rb_node
);
406 if (addr
< va
->va_start
)
408 else if (addr
>= va
->va_end
)
418 * This function returns back addresses of parent node
419 * and its left or right link for further processing.
421 static __always_inline
struct rb_node
**
422 find_va_links(struct vmap_area
*va
,
423 struct rb_root
*root
, struct rb_node
*from
,
424 struct rb_node
**parent
)
426 struct vmap_area
*tmp_va
;
427 struct rb_node
**link
;
430 link
= &root
->rb_node
;
431 if (unlikely(!*link
)) {
440 * Go to the bottom of the tree. When we hit the last point
441 * we end up with parent rb_node and correct direction, i name
442 * it link, where the new va->rb_node will be attached to.
445 tmp_va
= rb_entry(*link
, struct vmap_area
, rb_node
);
448 * During the traversal we also do some sanity check.
449 * Trigger the BUG() if there are sides(left/right)
452 if (va
->va_start
< tmp_va
->va_end
&&
453 va
->va_end
<= tmp_va
->va_start
)
454 link
= &(*link
)->rb_left
;
455 else if (va
->va_end
> tmp_va
->va_start
&&
456 va
->va_start
>= tmp_va
->va_end
)
457 link
= &(*link
)->rb_right
;
462 *parent
= &tmp_va
->rb_node
;
466 static __always_inline
struct list_head
*
467 get_va_next_sibling(struct rb_node
*parent
, struct rb_node
**link
)
469 struct list_head
*list
;
471 if (unlikely(!parent
))
473 * The red-black tree where we try to find VA neighbors
474 * before merging or inserting is empty, i.e. it means
475 * there is no free vmap space. Normally it does not
476 * happen but we handle this case anyway.
480 list
= &rb_entry(parent
, struct vmap_area
, rb_node
)->list
;
481 return (&parent
->rb_right
== link
? list
->next
: list
);
484 static __always_inline
void
485 link_va(struct vmap_area
*va
, struct rb_root
*root
,
486 struct rb_node
*parent
, struct rb_node
**link
, struct list_head
*head
)
489 * VA is still not in the list, but we can
490 * identify its future previous list_head node.
492 if (likely(parent
)) {
493 head
= &rb_entry(parent
, struct vmap_area
, rb_node
)->list
;
494 if (&parent
->rb_right
!= link
)
498 /* Insert to the rb-tree */
499 rb_link_node(&va
->rb_node
, parent
, link
);
500 if (root
== &free_vmap_area_root
) {
502 * Some explanation here. Just perform simple insertion
503 * to the tree. We do not set va->subtree_max_size to
504 * its current size before calling rb_insert_augmented().
505 * It is because of we populate the tree from the bottom
506 * to parent levels when the node _is_ in the tree.
508 * Therefore we set subtree_max_size to zero after insertion,
509 * to let __augment_tree_propagate_from() puts everything to
510 * the correct order later on.
512 rb_insert_augmented(&va
->rb_node
,
513 root
, &free_vmap_area_rb_augment_cb
);
514 va
->subtree_max_size
= 0;
516 rb_insert_color(&va
->rb_node
, root
);
519 /* Address-sort this list */
520 list_add(&va
->list
, head
);
523 static __always_inline
void
524 unlink_va(struct vmap_area
*va
, struct rb_root
*root
)
527 * During merging a VA node can be empty, therefore
528 * not linked with the tree nor list. Just check it.
530 if (!RB_EMPTY_NODE(&va
->rb_node
)) {
531 if (root
== &free_vmap_area_root
)
532 rb_erase_augmented(&va
->rb_node
,
533 root
, &free_vmap_area_rb_augment_cb
);
535 rb_erase(&va
->rb_node
, root
);
538 RB_CLEAR_NODE(&va
->rb_node
);
543 * This function populates subtree_max_size from bottom to upper
544 * levels starting from VA point. The propagation must be done
545 * when VA size is modified by changing its va_start/va_end. Or
546 * in case of newly inserting of VA to the tree.
548 * It means that __augment_tree_propagate_from() must be called:
549 * - After VA has been inserted to the tree(free path);
550 * - After VA has been shrunk(allocation path);
551 * - After VA has been increased(merging path).
553 * Please note that, it does not mean that upper parent nodes
554 * and their subtree_max_size are recalculated all the time up
563 * For example if we modify the node 4, shrinking it to 2, then
564 * no any modification is required. If we shrink the node 2 to 1
565 * its subtree_max_size is updated only, and set to 1. If we shrink
566 * the node 8 to 6, then its subtree_max_size is set to 6 and parent
569 static __always_inline
void
570 augment_tree_propagate_from(struct vmap_area
*va
)
572 struct rb_node
*node
= &va
->rb_node
;
573 unsigned long new_va_sub_max_size
;
576 va
= rb_entry(node
, struct vmap_area
, rb_node
);
577 new_va_sub_max_size
= compute_subtree_max_size(va
);
580 * If the newly calculated maximum available size of the
581 * subtree is equal to the current one, then it means that
582 * the tree is propagated correctly. So we have to stop at
583 * this point to save cycles.
585 if (va
->subtree_max_size
== new_va_sub_max_size
)
588 va
->subtree_max_size
= new_va_sub_max_size
;
589 node
= rb_parent(&va
->rb_node
);
594 insert_vmap_area(struct vmap_area
*va
,
595 struct rb_root
*root
, struct list_head
*head
)
597 struct rb_node
**link
;
598 struct rb_node
*parent
;
600 link
= find_va_links(va
, root
, NULL
, &parent
);
601 link_va(va
, root
, parent
, link
, head
);
605 insert_vmap_area_augment(struct vmap_area
*va
,
606 struct rb_node
*from
, struct rb_root
*root
,
607 struct list_head
*head
)
609 struct rb_node
**link
;
610 struct rb_node
*parent
;
613 link
= find_va_links(va
, NULL
, from
, &parent
);
615 link
= find_va_links(va
, root
, NULL
, &parent
);
617 link_va(va
, root
, parent
, link
, head
);
618 augment_tree_propagate_from(va
);
622 * Merge de-allocated chunk of VA memory with previous
623 * and next free blocks. If coalesce is not done a new
624 * free area is inserted. If VA has been merged, it is
627 static __always_inline
void
628 merge_or_add_vmap_area(struct vmap_area
*va
,
629 struct rb_root
*root
, struct list_head
*head
)
631 struct vmap_area
*sibling
;
632 struct list_head
*next
;
633 struct rb_node
**link
;
634 struct rb_node
*parent
;
638 * Find a place in the tree where VA potentially will be
639 * inserted, unless it is merged with its sibling/siblings.
641 link
= find_va_links(va
, root
, NULL
, &parent
);
644 * Get next node of VA to check if merging can be done.
646 next
= get_va_next_sibling(parent
, link
);
647 if (unlikely(next
== NULL
))
653 * |<------VA------>|<-----Next----->|
658 sibling
= list_entry(next
, struct vmap_area
, list
);
659 if (sibling
->va_start
== va
->va_end
) {
660 sibling
->va_start
= va
->va_start
;
662 /* Check and update the tree if needed. */
663 augment_tree_propagate_from(sibling
);
665 /* Remove this VA, it has been merged. */
668 /* Free vmap_area object. */
669 kmem_cache_free(vmap_area_cachep
, va
);
671 /* Point to the new merged area. */
680 * |<-----Prev----->|<------VA------>|
684 if (next
->prev
!= head
) {
685 sibling
= list_entry(next
->prev
, struct vmap_area
, list
);
686 if (sibling
->va_end
== va
->va_start
) {
687 sibling
->va_end
= va
->va_end
;
689 /* Check and update the tree if needed. */
690 augment_tree_propagate_from(sibling
);
692 /* Remove this VA, it has been merged. */
695 /* Free vmap_area object. */
696 kmem_cache_free(vmap_area_cachep
, va
);
704 link_va(va
, root
, parent
, link
, head
);
705 augment_tree_propagate_from(va
);
709 static __always_inline
bool
710 is_within_this_va(struct vmap_area
*va
, unsigned long size
,
711 unsigned long align
, unsigned long vstart
)
713 unsigned long nva_start_addr
;
715 if (va
->va_start
> vstart
)
716 nva_start_addr
= ALIGN(va
->va_start
, align
);
718 nva_start_addr
= ALIGN(vstart
, align
);
720 /* Can be overflowed due to big size or alignment. */
721 if (nva_start_addr
+ size
< nva_start_addr
||
722 nva_start_addr
< vstart
)
725 return (nva_start_addr
+ size
<= va
->va_end
);
729 * Find the first free block(lowest start address) in the tree,
730 * that will accomplish the request corresponding to passing
733 static __always_inline
struct vmap_area
*
734 find_vmap_lowest_match(unsigned long size
,
735 unsigned long align
, unsigned long vstart
)
737 struct vmap_area
*va
;
738 struct rb_node
*node
;
739 unsigned long length
;
741 /* Start from the root. */
742 node
= free_vmap_area_root
.rb_node
;
744 /* Adjust the search size for alignment overhead. */
745 length
= size
+ align
- 1;
748 va
= rb_entry(node
, struct vmap_area
, rb_node
);
750 if (get_subtree_max_size(node
->rb_left
) >= length
&&
751 vstart
< va
->va_start
) {
752 node
= node
->rb_left
;
754 if (is_within_this_va(va
, size
, align
, vstart
))
758 * Does not make sense to go deeper towards the right
759 * sub-tree if it does not have a free block that is
760 * equal or bigger to the requested search length.
762 if (get_subtree_max_size(node
->rb_right
) >= length
) {
763 node
= node
->rb_right
;
768 * OK. We roll back and find the fist right sub-tree,
769 * that will satisfy the search criteria. It can happen
770 * only once due to "vstart" restriction.
772 while ((node
= rb_parent(node
))) {
773 va
= rb_entry(node
, struct vmap_area
, rb_node
);
774 if (is_within_this_va(va
, size
, align
, vstart
))
777 if (get_subtree_max_size(node
->rb_right
) >= length
&&
778 vstart
<= va
->va_start
) {
779 node
= node
->rb_right
;
791 FL_FIT_TYPE
= 1, /* full fit */
792 LE_FIT_TYPE
= 2, /* left edge fit */
793 RE_FIT_TYPE
= 3, /* right edge fit */
794 NE_FIT_TYPE
= 4 /* no edge fit */
797 static __always_inline
enum fit_type
798 classify_va_fit_type(struct vmap_area
*va
,
799 unsigned long nva_start_addr
, unsigned long size
)
803 /* Check if it is within VA. */
804 if (nva_start_addr
< va
->va_start
||
805 nva_start_addr
+ size
> va
->va_end
)
809 if (va
->va_start
== nva_start_addr
) {
810 if (va
->va_end
== nva_start_addr
+ size
)
814 } else if (va
->va_end
== nva_start_addr
+ size
) {
823 static __always_inline
int
824 adjust_va_to_fit_type(struct vmap_area
*va
,
825 unsigned long nva_start_addr
, unsigned long size
,
828 struct vmap_area
*lva
;
830 if (type
== FL_FIT_TYPE
) {
832 * No need to split VA, it fully fits.
838 unlink_va(va
, &free_vmap_area_root
);
839 kmem_cache_free(vmap_area_cachep
, va
);
840 } else if (type
== LE_FIT_TYPE
) {
842 * Split left edge of fit VA.
848 va
->va_start
+= size
;
849 } else if (type
== RE_FIT_TYPE
) {
851 * Split right edge of fit VA.
857 va
->va_end
= nva_start_addr
;
858 } else if (type
== NE_FIT_TYPE
) {
860 * Split no edge of fit VA.
866 lva
= kmem_cache_alloc(vmap_area_cachep
, GFP_NOWAIT
);
871 * Build the remainder.
873 lva
->va_start
= va
->va_start
;
874 lva
->va_end
= nva_start_addr
;
877 * Shrink this VA to remaining size.
879 va
->va_start
= nva_start_addr
+ size
;
884 if (type
!= FL_FIT_TYPE
) {
885 augment_tree_propagate_from(va
);
887 if (type
== NE_FIT_TYPE
)
888 insert_vmap_area_augment(lva
, &va
->rb_node
,
889 &free_vmap_area_root
, &free_vmap_area_list
);
896 * Returns a start address of the newly allocated area, if success.
897 * Otherwise a vend is returned that indicates failure.
899 static __always_inline
unsigned long
900 __alloc_vmap_area(unsigned long size
, unsigned long align
,
901 unsigned long vstart
, unsigned long vend
, int node
)
903 unsigned long nva_start_addr
;
904 struct vmap_area
*va
;
908 va
= find_vmap_lowest_match(size
, align
, vstart
);
912 if (va
->va_start
> vstart
)
913 nva_start_addr
= ALIGN(va
->va_start
, align
);
915 nva_start_addr
= ALIGN(vstart
, align
);
917 /* Check the "vend" restriction. */
918 if (nva_start_addr
+ size
> vend
)
921 /* Classify what we have found. */
922 type
= classify_va_fit_type(va
, nva_start_addr
, size
);
923 if (WARN_ON_ONCE(type
== NOTHING_FIT
))
926 /* Update the free vmap_area. */
927 ret
= adjust_va_to_fit_type(va
, nva_start_addr
, size
, type
);
931 return nva_start_addr
;
935 * Allocate a region of KVA of the specified size and alignment, within the
938 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
940 unsigned long vstart
, unsigned long vend
,
941 int node
, gfp_t gfp_mask
)
943 struct vmap_area
*va
;
948 BUG_ON(offset_in_page(size
));
949 BUG_ON(!is_power_of_2(align
));
951 if (unlikely(!vmap_initialized
))
952 return ERR_PTR(-EBUSY
);
956 va
= kmem_cache_alloc_node(vmap_area_cachep
,
957 gfp_mask
& GFP_RECLAIM_MASK
, node
);
959 return ERR_PTR(-ENOMEM
);
962 * Only scan the relevant parts containing pointers to other objects
963 * to avoid false negatives.
965 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
968 spin_lock(&vmap_area_lock
);
971 * If an allocation fails, the "vend" address is
972 * returned. Therefore trigger the overflow path.
974 addr
= __alloc_vmap_area(size
, align
, vstart
, vend
, node
);
975 if (unlikely(addr
== vend
))
979 va
->va_end
= addr
+ size
;
981 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
983 spin_unlock(&vmap_area_lock
);
985 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
986 BUG_ON(va
->va_start
< vstart
);
987 BUG_ON(va
->va_end
> vend
);
992 spin_unlock(&vmap_area_lock
);
994 purge_vmap_area_lazy();
999 if (gfpflags_allow_blocking(gfp_mask
)) {
1000 unsigned long freed
= 0;
1001 blocking_notifier_call_chain(&vmap_notify_list
, 0, &freed
);
1008 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit())
1009 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
1012 kmem_cache_free(vmap_area_cachep
, va
);
1013 return ERR_PTR(-EBUSY
);
1016 int register_vmap_purge_notifier(struct notifier_block
*nb
)
1018 return blocking_notifier_chain_register(&vmap_notify_list
, nb
);
1020 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier
);
1022 int unregister_vmap_purge_notifier(struct notifier_block
*nb
)
1024 return blocking_notifier_chain_unregister(&vmap_notify_list
, nb
);
1026 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier
);
1028 static void __free_vmap_area(struct vmap_area
*va
)
1030 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
1033 * Remove from the busy tree/list.
1035 unlink_va(va
, &vmap_area_root
);
1038 * Merge VA with its neighbors, otherwise just add it.
1040 merge_or_add_vmap_area(va
,
1041 &free_vmap_area_root
, &free_vmap_area_list
);
1045 * Free a region of KVA allocated by alloc_vmap_area
1047 static void free_vmap_area(struct vmap_area
*va
)
1049 spin_lock(&vmap_area_lock
);
1050 __free_vmap_area(va
);
1051 spin_unlock(&vmap_area_lock
);
1055 * Clear the pagetable entries of a given vmap_area
1057 static void unmap_vmap_area(struct vmap_area
*va
)
1059 vunmap_page_range(va
->va_start
, va
->va_end
);
1063 * lazy_max_pages is the maximum amount of virtual address space we gather up
1064 * before attempting to purge with a TLB flush.
1066 * There is a tradeoff here: a larger number will cover more kernel page tables
1067 * and take slightly longer to purge, but it will linearly reduce the number of
1068 * global TLB flushes that must be performed. It would seem natural to scale
1069 * this number up linearly with the number of CPUs (because vmapping activity
1070 * could also scale linearly with the number of CPUs), however it is likely
1071 * that in practice, workloads might be constrained in other ways that mean
1072 * vmap activity will not scale linearly with CPUs. Also, I want to be
1073 * conservative and not introduce a big latency on huge systems, so go with
1074 * a less aggressive log scale. It will still be an improvement over the old
1075 * code, and it will be simple to change the scale factor if we find that it
1076 * becomes a problem on bigger systems.
1078 static unsigned long lazy_max_pages(void)
1082 log
= fls(num_online_cpus());
1084 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
1087 static atomic_long_t vmap_lazy_nr
= ATOMIC_LONG_INIT(0);
1090 * Serialize vmap purging. There is no actual criticial section protected
1091 * by this look, but we want to avoid concurrent calls for performance
1092 * reasons and to make the pcpu_get_vm_areas more deterministic.
1094 static DEFINE_MUTEX(vmap_purge_lock
);
1096 /* for per-CPU blocks */
1097 static void purge_fragmented_blocks_allcpus(void);
1100 * called before a call to iounmap() if the caller wants vm_area_struct's
1101 * immediately freed.
1103 void set_iounmap_nonlazy(void)
1105 atomic_long_set(&vmap_lazy_nr
, lazy_max_pages()+1);
1109 * Purges all lazily-freed vmap areas.
1111 static bool __purge_vmap_area_lazy(unsigned long start
, unsigned long end
)
1113 unsigned long resched_threshold
;
1114 struct llist_node
*valist
;
1115 struct vmap_area
*va
;
1116 struct vmap_area
*n_va
;
1118 lockdep_assert_held(&vmap_purge_lock
);
1120 valist
= llist_del_all(&vmap_purge_list
);
1121 if (unlikely(valist
== NULL
))
1125 * TODO: to calculate a flush range without looping.
1126 * The list can be up to lazy_max_pages() elements.
1128 llist_for_each_entry(va
, valist
, purge_list
) {
1129 if (va
->va_start
< start
)
1130 start
= va
->va_start
;
1131 if (va
->va_end
> end
)
1135 flush_tlb_kernel_range(start
, end
);
1136 resched_threshold
= lazy_max_pages() << 1;
1138 spin_lock(&vmap_area_lock
);
1139 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
) {
1140 unsigned long nr
= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
1142 __free_vmap_area(va
);
1143 atomic_long_sub(nr
, &vmap_lazy_nr
);
1145 if (atomic_long_read(&vmap_lazy_nr
) < resched_threshold
)
1146 cond_resched_lock(&vmap_area_lock
);
1148 spin_unlock(&vmap_area_lock
);
1153 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
1154 * is already purging.
1156 static void try_purge_vmap_area_lazy(void)
1158 if (mutex_trylock(&vmap_purge_lock
)) {
1159 __purge_vmap_area_lazy(ULONG_MAX
, 0);
1160 mutex_unlock(&vmap_purge_lock
);
1165 * Kick off a purge of the outstanding lazy areas.
1167 static void purge_vmap_area_lazy(void)
1169 mutex_lock(&vmap_purge_lock
);
1170 purge_fragmented_blocks_allcpus();
1171 __purge_vmap_area_lazy(ULONG_MAX
, 0);
1172 mutex_unlock(&vmap_purge_lock
);
1176 * Free a vmap area, caller ensuring that the area has been unmapped
1177 * and flush_cache_vunmap had been called for the correct range
1180 static void free_vmap_area_noflush(struct vmap_area
*va
)
1182 unsigned long nr_lazy
;
1184 nr_lazy
= atomic_long_add_return((va
->va_end
- va
->va_start
) >>
1185 PAGE_SHIFT
, &vmap_lazy_nr
);
1187 /* After this point, we may free va at any time */
1188 llist_add(&va
->purge_list
, &vmap_purge_list
);
1190 if (unlikely(nr_lazy
> lazy_max_pages()))
1191 try_purge_vmap_area_lazy();
1195 * Free and unmap a vmap area
1197 static void free_unmap_vmap_area(struct vmap_area
*va
)
1199 flush_cache_vunmap(va
->va_start
, va
->va_end
);
1200 unmap_vmap_area(va
);
1201 if (debug_pagealloc_enabled())
1202 flush_tlb_kernel_range(va
->va_start
, va
->va_end
);
1204 free_vmap_area_noflush(va
);
1207 static struct vmap_area
*find_vmap_area(unsigned long addr
)
1209 struct vmap_area
*va
;
1211 spin_lock(&vmap_area_lock
);
1212 va
= __find_vmap_area(addr
);
1213 spin_unlock(&vmap_area_lock
);
1218 /*** Per cpu kva allocator ***/
1221 * vmap space is limited especially on 32 bit architectures. Ensure there is
1222 * room for at least 16 percpu vmap blocks per CPU.
1225 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
1226 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
1227 * instead (we just need a rough idea)
1229 #if BITS_PER_LONG == 32
1230 #define VMALLOC_SPACE (128UL*1024*1024)
1232 #define VMALLOC_SPACE (128UL*1024*1024*1024)
1235 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
1236 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
1237 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
1238 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
1239 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
1240 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
1241 #define VMAP_BBMAP_BITS \
1242 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
1243 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
1244 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
1246 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
1248 struct vmap_block_queue
{
1250 struct list_head free
;
1255 struct vmap_area
*va
;
1256 unsigned long free
, dirty
;
1257 unsigned long dirty_min
, dirty_max
; /*< dirty range */
1258 struct list_head free_list
;
1259 struct rcu_head rcu_head
;
1260 struct list_head purge
;
1263 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
1264 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
1267 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
1268 * in the free path. Could get rid of this if we change the API to return a
1269 * "cookie" from alloc, to be passed to free. But no big deal yet.
1271 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
1272 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
1275 * We should probably have a fallback mechanism to allocate virtual memory
1276 * out of partially filled vmap blocks. However vmap block sizing should be
1277 * fairly reasonable according to the vmalloc size, so it shouldn't be a
1281 static unsigned long addr_to_vb_idx(unsigned long addr
)
1283 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
1284 addr
/= VMAP_BLOCK_SIZE
;
1288 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
1292 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
1293 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
1294 return (void *)addr
;
1298 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
1299 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
1300 * @order: how many 2^order pages should be occupied in newly allocated block
1301 * @gfp_mask: flags for the page level allocator
1303 * Return: virtual address in a newly allocated block or ERR_PTR(-errno)
1305 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
1307 struct vmap_block_queue
*vbq
;
1308 struct vmap_block
*vb
;
1309 struct vmap_area
*va
;
1310 unsigned long vb_idx
;
1314 node
= numa_node_id();
1316 vb
= kmalloc_node(sizeof(struct vmap_block
),
1317 gfp_mask
& GFP_RECLAIM_MASK
, node
);
1319 return ERR_PTR(-ENOMEM
);
1321 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
1322 VMALLOC_START
, VMALLOC_END
,
1326 return ERR_CAST(va
);
1329 err
= radix_tree_preload(gfp_mask
);
1330 if (unlikely(err
)) {
1333 return ERR_PTR(err
);
1336 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
1337 spin_lock_init(&vb
->lock
);
1339 /* At least something should be left free */
1340 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
1341 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
1343 vb
->dirty_min
= VMAP_BBMAP_BITS
;
1345 INIT_LIST_HEAD(&vb
->free_list
);
1347 vb_idx
= addr_to_vb_idx(va
->va_start
);
1348 spin_lock(&vmap_block_tree_lock
);
1349 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
1350 spin_unlock(&vmap_block_tree_lock
);
1352 radix_tree_preload_end();
1354 vbq
= &get_cpu_var(vmap_block_queue
);
1355 spin_lock(&vbq
->lock
);
1356 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
1357 spin_unlock(&vbq
->lock
);
1358 put_cpu_var(vmap_block_queue
);
1363 static void free_vmap_block(struct vmap_block
*vb
)
1365 struct vmap_block
*tmp
;
1366 unsigned long vb_idx
;
1368 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
1369 spin_lock(&vmap_block_tree_lock
);
1370 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
1371 spin_unlock(&vmap_block_tree_lock
);
1374 free_vmap_area_noflush(vb
->va
);
1375 kfree_rcu(vb
, rcu_head
);
1378 static void purge_fragmented_blocks(int cpu
)
1381 struct vmap_block
*vb
;
1382 struct vmap_block
*n_vb
;
1383 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1386 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1388 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
1391 spin_lock(&vb
->lock
);
1392 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
1393 vb
->free
= 0; /* prevent further allocs after releasing lock */
1394 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
1396 vb
->dirty_max
= VMAP_BBMAP_BITS
;
1397 spin_lock(&vbq
->lock
);
1398 list_del_rcu(&vb
->free_list
);
1399 spin_unlock(&vbq
->lock
);
1400 spin_unlock(&vb
->lock
);
1401 list_add_tail(&vb
->purge
, &purge
);
1403 spin_unlock(&vb
->lock
);
1407 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
1408 list_del(&vb
->purge
);
1409 free_vmap_block(vb
);
1413 static void purge_fragmented_blocks_allcpus(void)
1417 for_each_possible_cpu(cpu
)
1418 purge_fragmented_blocks(cpu
);
1421 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
1423 struct vmap_block_queue
*vbq
;
1424 struct vmap_block
*vb
;
1428 BUG_ON(offset_in_page(size
));
1429 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1430 if (WARN_ON(size
== 0)) {
1432 * Allocating 0 bytes isn't what caller wants since
1433 * get_order(0) returns funny result. Just warn and terminate
1438 order
= get_order(size
);
1441 vbq
= &get_cpu_var(vmap_block_queue
);
1442 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1443 unsigned long pages_off
;
1445 spin_lock(&vb
->lock
);
1446 if (vb
->free
< (1UL << order
)) {
1447 spin_unlock(&vb
->lock
);
1451 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
1452 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
1453 vb
->free
-= 1UL << order
;
1454 if (vb
->free
== 0) {
1455 spin_lock(&vbq
->lock
);
1456 list_del_rcu(&vb
->free_list
);
1457 spin_unlock(&vbq
->lock
);
1460 spin_unlock(&vb
->lock
);
1464 put_cpu_var(vmap_block_queue
);
1467 /* Allocate new block if nothing was found */
1469 vaddr
= new_vmap_block(order
, gfp_mask
);
1474 static void vb_free(const void *addr
, unsigned long size
)
1476 unsigned long offset
;
1477 unsigned long vb_idx
;
1479 struct vmap_block
*vb
;
1481 BUG_ON(offset_in_page(size
));
1482 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1484 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
1486 order
= get_order(size
);
1488 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
1489 offset
>>= PAGE_SHIFT
;
1491 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
1493 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1497 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1499 if (debug_pagealloc_enabled())
1500 flush_tlb_kernel_range((unsigned long)addr
,
1501 (unsigned long)addr
+ size
);
1503 spin_lock(&vb
->lock
);
1505 /* Expand dirty range */
1506 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1507 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1509 vb
->dirty
+= 1UL << order
;
1510 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1512 spin_unlock(&vb
->lock
);
1513 free_vmap_block(vb
);
1515 spin_unlock(&vb
->lock
);
1518 static void _vm_unmap_aliases(unsigned long start
, unsigned long end
, int flush
)
1522 if (unlikely(!vmap_initialized
))
1527 for_each_possible_cpu(cpu
) {
1528 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1529 struct vmap_block
*vb
;
1532 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1533 spin_lock(&vb
->lock
);
1535 unsigned long va_start
= vb
->va
->va_start
;
1538 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1539 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1541 start
= min(s
, start
);
1546 spin_unlock(&vb
->lock
);
1551 mutex_lock(&vmap_purge_lock
);
1552 purge_fragmented_blocks_allcpus();
1553 if (!__purge_vmap_area_lazy(start
, end
) && flush
)
1554 flush_tlb_kernel_range(start
, end
);
1555 mutex_unlock(&vmap_purge_lock
);
1559 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1561 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1562 * to amortize TLB flushing overheads. What this means is that any page you
1563 * have now, may, in a former life, have been mapped into kernel virtual
1564 * address by the vmap layer and so there might be some CPUs with TLB entries
1565 * still referencing that page (additional to the regular 1:1 kernel mapping).
1567 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1568 * be sure that none of the pages we have control over will have any aliases
1569 * from the vmap layer.
1571 void vm_unmap_aliases(void)
1573 unsigned long start
= ULONG_MAX
, end
= 0;
1576 _vm_unmap_aliases(start
, end
, flush
);
1578 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1581 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1582 * @mem: the pointer returned by vm_map_ram
1583 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1585 void vm_unmap_ram(const void *mem
, unsigned int count
)
1587 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1588 unsigned long addr
= (unsigned long)mem
;
1589 struct vmap_area
*va
;
1593 BUG_ON(addr
< VMALLOC_START
);
1594 BUG_ON(addr
> VMALLOC_END
);
1595 BUG_ON(!PAGE_ALIGNED(addr
));
1597 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1598 debug_check_no_locks_freed(mem
, size
);
1603 va
= find_vmap_area(addr
);
1605 debug_check_no_locks_freed((void *)va
->va_start
,
1606 (va
->va_end
- va
->va_start
));
1607 free_unmap_vmap_area(va
);
1609 EXPORT_SYMBOL(vm_unmap_ram
);
1612 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1613 * @pages: an array of pointers to the pages to be mapped
1614 * @count: number of pages
1615 * @node: prefer to allocate data structures on this node
1616 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1618 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1619 * faster than vmap so it's good. But if you mix long-life and short-life
1620 * objects with vm_map_ram(), it could consume lots of address space through
1621 * fragmentation (especially on a 32bit machine). You could see failures in
1622 * the end. Please use this function for short-lived objects.
1624 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1626 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1628 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1632 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1633 mem
= vb_alloc(size
, GFP_KERNEL
);
1636 addr
= (unsigned long)mem
;
1638 struct vmap_area
*va
;
1639 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1640 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1644 addr
= va
->va_start
;
1647 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1648 vm_unmap_ram(mem
, count
);
1653 EXPORT_SYMBOL(vm_map_ram
);
1655 static struct vm_struct
*vmlist __initdata
;
1658 * vm_area_add_early - add vmap area early during boot
1659 * @vm: vm_struct to add
1661 * This function is used to add fixed kernel vm area to vmlist before
1662 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1663 * should contain proper values and the other fields should be zero.
1665 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1667 void __init
vm_area_add_early(struct vm_struct
*vm
)
1669 struct vm_struct
*tmp
, **p
;
1671 BUG_ON(vmap_initialized
);
1672 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1673 if (tmp
->addr
>= vm
->addr
) {
1674 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1677 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1684 * vm_area_register_early - register vmap area early during boot
1685 * @vm: vm_struct to register
1686 * @align: requested alignment
1688 * This function is used to register kernel vm area before
1689 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1690 * proper values on entry and other fields should be zero. On return,
1691 * vm->addr contains the allocated address.
1693 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1695 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1697 static size_t vm_init_off __initdata
;
1700 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1701 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1703 vm
->addr
= (void *)addr
;
1705 vm_area_add_early(vm
);
1708 static void vmap_init_free_space(void)
1710 unsigned long vmap_start
= 1;
1711 const unsigned long vmap_end
= ULONG_MAX
;
1712 struct vmap_area
*busy
, *free
;
1716 * -|-----|.....|-----|-----|-----|.....|-
1718 * |<--------------------------------->|
1720 list_for_each_entry(busy
, &vmap_area_list
, list
) {
1721 if (busy
->va_start
- vmap_start
> 0) {
1722 free
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1723 if (!WARN_ON_ONCE(!free
)) {
1724 free
->va_start
= vmap_start
;
1725 free
->va_end
= busy
->va_start
;
1727 insert_vmap_area_augment(free
, NULL
,
1728 &free_vmap_area_root
,
1729 &free_vmap_area_list
);
1733 vmap_start
= busy
->va_end
;
1736 if (vmap_end
- vmap_start
> 0) {
1737 free
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1738 if (!WARN_ON_ONCE(!free
)) {
1739 free
->va_start
= vmap_start
;
1740 free
->va_end
= vmap_end
;
1742 insert_vmap_area_augment(free
, NULL
,
1743 &free_vmap_area_root
,
1744 &free_vmap_area_list
);
1749 void __init
vmalloc_init(void)
1751 struct vmap_area
*va
;
1752 struct vm_struct
*tmp
;
1756 * Create the cache for vmap_area objects.
1758 vmap_area_cachep
= KMEM_CACHE(vmap_area
, SLAB_PANIC
);
1760 for_each_possible_cpu(i
) {
1761 struct vmap_block_queue
*vbq
;
1762 struct vfree_deferred
*p
;
1764 vbq
= &per_cpu(vmap_block_queue
, i
);
1765 spin_lock_init(&vbq
->lock
);
1766 INIT_LIST_HEAD(&vbq
->free
);
1767 p
= &per_cpu(vfree_deferred
, i
);
1768 init_llist_head(&p
->list
);
1769 INIT_WORK(&p
->wq
, free_work
);
1772 /* Import existing vmlist entries. */
1773 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1774 va
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1775 if (WARN_ON_ONCE(!va
))
1778 va
->flags
= VM_VM_AREA
;
1779 va
->va_start
= (unsigned long)tmp
->addr
;
1780 va
->va_end
= va
->va_start
+ tmp
->size
;
1782 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
1786 * Now we can initialize a free vmap space.
1788 vmap_init_free_space();
1789 vmap_initialized
= true;
1793 * map_kernel_range_noflush - map kernel VM area with the specified pages
1794 * @addr: start of the VM area to map
1795 * @size: size of the VM area to map
1796 * @prot: page protection flags to use
1797 * @pages: pages to map
1799 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1800 * specify should have been allocated using get_vm_area() and its
1804 * This function does NOT do any cache flushing. The caller is
1805 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1806 * before calling this function.
1809 * The number of pages mapped on success, -errno on failure.
1811 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1812 pgprot_t prot
, struct page
**pages
)
1814 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1818 * unmap_kernel_range_noflush - unmap kernel VM area
1819 * @addr: start of the VM area to unmap
1820 * @size: size of the VM area to unmap
1822 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1823 * specify should have been allocated using get_vm_area() and its
1827 * This function does NOT do any cache flushing. The caller is
1828 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1829 * before calling this function and flush_tlb_kernel_range() after.
1831 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1833 vunmap_page_range(addr
, addr
+ size
);
1835 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1838 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1839 * @addr: start of the VM area to unmap
1840 * @size: size of the VM area to unmap
1842 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1843 * the unmapping and tlb after.
1845 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1847 unsigned long end
= addr
+ size
;
1849 flush_cache_vunmap(addr
, end
);
1850 vunmap_page_range(addr
, end
);
1851 flush_tlb_kernel_range(addr
, end
);
1853 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
1855 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
1857 unsigned long addr
= (unsigned long)area
->addr
;
1858 unsigned long end
= addr
+ get_vm_area_size(area
);
1861 err
= vmap_page_range(addr
, end
, prot
, pages
);
1863 return err
> 0 ? 0 : err
;
1865 EXPORT_SYMBOL_GPL(map_vm_area
);
1867 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1868 unsigned long flags
, const void *caller
)
1870 spin_lock(&vmap_area_lock
);
1872 vm
->addr
= (void *)va
->va_start
;
1873 vm
->size
= va
->va_end
- va
->va_start
;
1874 vm
->caller
= caller
;
1876 va
->flags
|= VM_VM_AREA
;
1877 spin_unlock(&vmap_area_lock
);
1880 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1883 * Before removing VM_UNINITIALIZED,
1884 * we should make sure that vm has proper values.
1885 * Pair with smp_rmb() in show_numa_info().
1888 vm
->flags
&= ~VM_UNINITIALIZED
;
1891 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1892 unsigned long align
, unsigned long flags
, unsigned long start
,
1893 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1895 struct vmap_area
*va
;
1896 struct vm_struct
*area
;
1898 BUG_ON(in_interrupt());
1899 size
= PAGE_ALIGN(size
);
1900 if (unlikely(!size
))
1903 if (flags
& VM_IOREMAP
)
1904 align
= 1ul << clamp_t(int, get_count_order_long(size
),
1905 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1907 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1908 if (unlikely(!area
))
1911 if (!(flags
& VM_NO_GUARD
))
1914 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1920 setup_vmalloc_vm(area
, va
, flags
, caller
);
1925 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1926 unsigned long start
, unsigned long end
)
1928 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1929 GFP_KERNEL
, __builtin_return_address(0));
1931 EXPORT_SYMBOL_GPL(__get_vm_area
);
1933 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1934 unsigned long start
, unsigned long end
,
1937 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1938 GFP_KERNEL
, caller
);
1942 * get_vm_area - reserve a contiguous kernel virtual area
1943 * @size: size of the area
1944 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1946 * Search an area of @size in the kernel virtual mapping area,
1947 * and reserved it for out purposes. Returns the area descriptor
1948 * on success or %NULL on failure.
1950 * Return: the area descriptor on success or %NULL on failure.
1952 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1954 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1955 NUMA_NO_NODE
, GFP_KERNEL
,
1956 __builtin_return_address(0));
1959 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1962 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1963 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1967 * find_vm_area - find a continuous kernel virtual area
1968 * @addr: base address
1970 * Search for the kernel VM area starting at @addr, and return it.
1971 * It is up to the caller to do all required locking to keep the returned
1974 * Return: pointer to the found area or %NULL on faulure
1976 struct vm_struct
*find_vm_area(const void *addr
)
1978 struct vmap_area
*va
;
1980 va
= find_vmap_area((unsigned long)addr
);
1981 if (va
&& va
->flags
& VM_VM_AREA
)
1988 * remove_vm_area - find and remove a continuous kernel virtual area
1989 * @addr: base address
1991 * Search for the kernel VM area starting at @addr, and remove it.
1992 * This function returns the found VM area, but using it is NOT safe
1993 * on SMP machines, except for its size or flags.
1995 * Return: pointer to the found area or %NULL on faulure
1997 struct vm_struct
*remove_vm_area(const void *addr
)
1999 struct vmap_area
*va
;
2003 va
= find_vmap_area((unsigned long)addr
);
2004 if (va
&& va
->flags
& VM_VM_AREA
) {
2005 struct vm_struct
*vm
= va
->vm
;
2007 spin_lock(&vmap_area_lock
);
2009 va
->flags
&= ~VM_VM_AREA
;
2010 va
->flags
|= VM_LAZY_FREE
;
2011 spin_unlock(&vmap_area_lock
);
2013 kasan_free_shadow(vm
);
2014 free_unmap_vmap_area(va
);
2021 static inline void set_area_direct_map(const struct vm_struct
*area
,
2022 int (*set_direct_map
)(struct page
*page
))
2026 for (i
= 0; i
< area
->nr_pages
; i
++)
2027 if (page_address(area
->pages
[i
]))
2028 set_direct_map(area
->pages
[i
]);
2031 /* Handle removing and resetting vm mappings related to the vm_struct. */
2032 static void vm_remove_mappings(struct vm_struct
*area
, int deallocate_pages
)
2034 unsigned long addr
= (unsigned long)area
->addr
;
2035 unsigned long start
= ULONG_MAX
, end
= 0;
2036 int flush_reset
= area
->flags
& VM_FLUSH_RESET_PERMS
;
2040 * The below block can be removed when all architectures that have
2041 * direct map permissions also have set_direct_map_() implementations.
2042 * This is concerned with resetting the direct map any an vm alias with
2043 * execute permissions, without leaving a RW+X window.
2045 if (flush_reset
&& !IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP
)) {
2046 set_memory_nx(addr
, area
->nr_pages
);
2047 set_memory_rw(addr
, area
->nr_pages
);
2050 remove_vm_area(area
->addr
);
2052 /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */
2057 * If not deallocating pages, just do the flush of the VM area and
2060 if (!deallocate_pages
) {
2066 * If execution gets here, flush the vm mapping and reset the direct
2067 * map. Find the start and end range of the direct mappings to make sure
2068 * the vm_unmap_aliases() flush includes the direct map.
2070 for (i
= 0; i
< area
->nr_pages
; i
++) {
2071 if (page_address(area
->pages
[i
])) {
2072 start
= min(addr
, start
);
2073 end
= max(addr
, end
);
2078 * Set direct map to something invalid so that it won't be cached if
2079 * there are any accesses after the TLB flush, then flush the TLB and
2080 * reset the direct map permissions to the default.
2082 set_area_direct_map(area
, set_direct_map_invalid_noflush
);
2083 _vm_unmap_aliases(start
, end
, 1);
2084 set_area_direct_map(area
, set_direct_map_default_noflush
);
2087 static void __vunmap(const void *addr
, int deallocate_pages
)
2089 struct vm_struct
*area
;
2094 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
2098 area
= find_vm_area(addr
);
2099 if (unlikely(!area
)) {
2100 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
2105 debug_check_no_locks_freed(area
->addr
, get_vm_area_size(area
));
2106 debug_check_no_obj_freed(area
->addr
, get_vm_area_size(area
));
2108 vm_remove_mappings(area
, deallocate_pages
);
2110 if (deallocate_pages
) {
2113 for (i
= 0; i
< area
->nr_pages
; i
++) {
2114 struct page
*page
= area
->pages
[i
];
2117 __free_pages(page
, 0);
2120 kvfree(area
->pages
);
2127 static inline void __vfree_deferred(const void *addr
)
2130 * Use raw_cpu_ptr() because this can be called from preemptible
2131 * context. Preemption is absolutely fine here, because the llist_add()
2132 * implementation is lockless, so it works even if we are adding to
2133 * nother cpu's list. schedule_work() should be fine with this too.
2135 struct vfree_deferred
*p
= raw_cpu_ptr(&vfree_deferred
);
2137 if (llist_add((struct llist_node
*)addr
, &p
->list
))
2138 schedule_work(&p
->wq
);
2142 * vfree_atomic - release memory allocated by vmalloc()
2143 * @addr: memory base address
2145 * This one is just like vfree() but can be called in any atomic context
2148 void vfree_atomic(const void *addr
)
2152 kmemleak_free(addr
);
2156 __vfree_deferred(addr
);
2159 static void __vfree(const void *addr
)
2161 if (unlikely(in_interrupt()))
2162 __vfree_deferred(addr
);
2168 * vfree - release memory allocated by vmalloc()
2169 * @addr: memory base address
2171 * Free the virtually continuous memory area starting at @addr, as
2172 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
2173 * NULL, no operation is performed.
2175 * Must not be called in NMI context (strictly speaking, only if we don't
2176 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
2177 * conventions for vfree() arch-depenedent would be a really bad idea)
2179 * May sleep if called *not* from interrupt context.
2181 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
2183 void vfree(const void *addr
)
2187 kmemleak_free(addr
);
2189 might_sleep_if(!in_interrupt());
2196 EXPORT_SYMBOL(vfree
);
2199 * vunmap - release virtual mapping obtained by vmap()
2200 * @addr: memory base address
2202 * Free the virtually contiguous memory area starting at @addr,
2203 * which was created from the page array passed to vmap().
2205 * Must not be called in interrupt context.
2207 void vunmap(const void *addr
)
2209 BUG_ON(in_interrupt());
2214 EXPORT_SYMBOL(vunmap
);
2217 * vmap - map an array of pages into virtually contiguous space
2218 * @pages: array of page pointers
2219 * @count: number of pages to map
2220 * @flags: vm_area->flags
2221 * @prot: page protection for the mapping
2223 * Maps @count pages from @pages into contiguous kernel virtual
2226 * Return: the address of the area or %NULL on failure
2228 void *vmap(struct page
**pages
, unsigned int count
,
2229 unsigned long flags
, pgprot_t prot
)
2231 struct vm_struct
*area
;
2232 unsigned long size
; /* In bytes */
2236 if (count
> totalram_pages())
2239 size
= (unsigned long)count
<< PAGE_SHIFT
;
2240 area
= get_vm_area_caller(size
, flags
, __builtin_return_address(0));
2244 if (map_vm_area(area
, prot
, pages
)) {
2251 EXPORT_SYMBOL(vmap
);
2253 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
2254 gfp_t gfp_mask
, pgprot_t prot
,
2255 int node
, const void *caller
);
2256 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
2257 pgprot_t prot
, int node
)
2259 struct page
**pages
;
2260 unsigned int nr_pages
, array_size
, i
;
2261 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
2262 const gfp_t alloc_mask
= gfp_mask
| __GFP_NOWARN
;
2263 const gfp_t highmem_mask
= (gfp_mask
& (GFP_DMA
| GFP_DMA32
)) ?
2267 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
2268 array_size
= (nr_pages
* sizeof(struct page
*));
2270 area
->nr_pages
= nr_pages
;
2271 /* Please note that the recursion is strictly bounded. */
2272 if (array_size
> PAGE_SIZE
) {
2273 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|highmem_mask
,
2274 PAGE_KERNEL
, node
, area
->caller
);
2276 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
2278 area
->pages
= pages
;
2280 remove_vm_area(area
->addr
);
2285 for (i
= 0; i
< area
->nr_pages
; i
++) {
2288 if (node
== NUMA_NO_NODE
)
2289 page
= alloc_page(alloc_mask
|highmem_mask
);
2291 page
= alloc_pages_node(node
, alloc_mask
|highmem_mask
, 0);
2293 if (unlikely(!page
)) {
2294 /* Successfully allocated i pages, free them in __vunmap() */
2298 area
->pages
[i
] = page
;
2299 if (gfpflags_allow_blocking(gfp_mask
|highmem_mask
))
2303 if (map_vm_area(area
, prot
, pages
))
2308 warn_alloc(gfp_mask
, NULL
,
2309 "vmalloc: allocation failure, allocated %ld of %ld bytes",
2310 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
2311 __vfree(area
->addr
);
2316 * __vmalloc_node_range - allocate virtually contiguous memory
2317 * @size: allocation size
2318 * @align: desired alignment
2319 * @start: vm area range start
2320 * @end: vm area range end
2321 * @gfp_mask: flags for the page level allocator
2322 * @prot: protection mask for the allocated pages
2323 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
2324 * @node: node to use for allocation or NUMA_NO_NODE
2325 * @caller: caller's return address
2327 * Allocate enough pages to cover @size from the page level
2328 * allocator with @gfp_mask flags. Map them into contiguous
2329 * kernel virtual space, using a pagetable protection of @prot.
2331 * Return: the address of the area or %NULL on failure
2333 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
2334 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
2335 pgprot_t prot
, unsigned long vm_flags
, int node
,
2338 struct vm_struct
*area
;
2340 unsigned long real_size
= size
;
2342 size
= PAGE_ALIGN(size
);
2343 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages())
2346 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
2347 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
2351 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
2356 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
2357 * flag. It means that vm_struct is not fully initialized.
2358 * Now, it is fully initialized, so remove this flag here.
2360 clear_vm_uninitialized_flag(area
);
2362 kmemleak_vmalloc(area
, size
, gfp_mask
);
2367 warn_alloc(gfp_mask
, NULL
,
2368 "vmalloc: allocation failure: %lu bytes", real_size
);
2373 * This is only for performance analysis of vmalloc and stress purpose.
2374 * It is required by vmalloc test module, therefore do not use it other
2377 #ifdef CONFIG_TEST_VMALLOC_MODULE
2378 EXPORT_SYMBOL_GPL(__vmalloc_node_range
);
2382 * __vmalloc_node - allocate virtually contiguous memory
2383 * @size: allocation size
2384 * @align: desired alignment
2385 * @gfp_mask: flags for the page level allocator
2386 * @prot: protection mask for the allocated pages
2387 * @node: node to use for allocation or NUMA_NO_NODE
2388 * @caller: caller's return address
2390 * Allocate enough pages to cover @size from the page level
2391 * allocator with @gfp_mask flags. Map them into contiguous
2392 * kernel virtual space, using a pagetable protection of @prot.
2394 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
2395 * and __GFP_NOFAIL are not supported
2397 * Any use of gfp flags outside of GFP_KERNEL should be consulted
2400 * Return: pointer to the allocated memory or %NULL on error
2402 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
2403 gfp_t gfp_mask
, pgprot_t prot
,
2404 int node
, const void *caller
)
2406 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
2407 gfp_mask
, prot
, 0, node
, caller
);
2410 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
2412 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
2413 __builtin_return_address(0));
2415 EXPORT_SYMBOL(__vmalloc
);
2417 static inline void *__vmalloc_node_flags(unsigned long size
,
2418 int node
, gfp_t flags
)
2420 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
2421 node
, __builtin_return_address(0));
2425 void *__vmalloc_node_flags_caller(unsigned long size
, int node
, gfp_t flags
,
2428 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
, node
, caller
);
2432 * vmalloc - allocate virtually contiguous memory
2433 * @size: allocation size
2435 * Allocate enough pages to cover @size from the page level
2436 * allocator and map them into contiguous kernel virtual space.
2438 * For tight control over page level allocator and protection flags
2439 * use __vmalloc() instead.
2441 * Return: pointer to the allocated memory or %NULL on error
2443 void *vmalloc(unsigned long size
)
2445 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
2448 EXPORT_SYMBOL(vmalloc
);
2451 * vzalloc - allocate virtually contiguous memory with zero fill
2452 * @size: allocation size
2454 * Allocate enough pages to cover @size from the page level
2455 * allocator and map them into contiguous kernel virtual space.
2456 * The memory allocated is set to zero.
2458 * For tight control over page level allocator and protection flags
2459 * use __vmalloc() instead.
2461 * Return: pointer to the allocated memory or %NULL on error
2463 void *vzalloc(unsigned long size
)
2465 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
2466 GFP_KERNEL
| __GFP_ZERO
);
2468 EXPORT_SYMBOL(vzalloc
);
2471 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
2472 * @size: allocation size
2474 * The resulting memory area is zeroed so it can be mapped to userspace
2475 * without leaking data.
2477 * Return: pointer to the allocated memory or %NULL on error
2479 void *vmalloc_user(unsigned long size
)
2481 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
2482 GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
,
2483 VM_USERMAP
, NUMA_NO_NODE
,
2484 __builtin_return_address(0));
2486 EXPORT_SYMBOL(vmalloc_user
);
2489 * vmalloc_node - allocate memory on a specific node
2490 * @size: allocation size
2493 * Allocate enough pages to cover @size from the page level
2494 * allocator and map them into contiguous kernel virtual space.
2496 * For tight control over page level allocator and protection flags
2497 * use __vmalloc() instead.
2499 * Return: pointer to the allocated memory or %NULL on error
2501 void *vmalloc_node(unsigned long size
, int node
)
2503 return __vmalloc_node(size
, 1, GFP_KERNEL
, PAGE_KERNEL
,
2504 node
, __builtin_return_address(0));
2506 EXPORT_SYMBOL(vmalloc_node
);
2509 * vzalloc_node - allocate memory on a specific node with zero fill
2510 * @size: allocation size
2513 * Allocate enough pages to cover @size from the page level
2514 * allocator and map them into contiguous kernel virtual space.
2515 * The memory allocated is set to zero.
2517 * For tight control over page level allocator and protection flags
2518 * use __vmalloc_node() instead.
2520 * Return: pointer to the allocated memory or %NULL on error
2522 void *vzalloc_node(unsigned long size
, int node
)
2524 return __vmalloc_node_flags(size
, node
,
2525 GFP_KERNEL
| __GFP_ZERO
);
2527 EXPORT_SYMBOL(vzalloc_node
);
2530 * vmalloc_exec - allocate virtually contiguous, executable memory
2531 * @size: allocation size
2533 * Kernel-internal function to allocate enough pages to cover @size
2534 * the page level allocator and map them into contiguous and
2535 * executable kernel virtual space.
2537 * For tight control over page level allocator and protection flags
2538 * use __vmalloc() instead.
2540 * Return: pointer to the allocated memory or %NULL on error
2542 void *vmalloc_exec(unsigned long size
)
2544 return __vmalloc_node_range(size
, 1, VMALLOC_START
, VMALLOC_END
,
2545 GFP_KERNEL
, PAGE_KERNEL_EXEC
, VM_FLUSH_RESET_PERMS
,
2546 NUMA_NO_NODE
, __builtin_return_address(0));
2549 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
2550 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
2551 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
2552 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
2555 * 64b systems should always have either DMA or DMA32 zones. For others
2556 * GFP_DMA32 should do the right thing and use the normal zone.
2558 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
2562 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
2563 * @size: allocation size
2565 * Allocate enough 32bit PA addressable pages to cover @size from the
2566 * page level allocator and map them into contiguous kernel virtual space.
2568 * Return: pointer to the allocated memory or %NULL on error
2570 void *vmalloc_32(unsigned long size
)
2572 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
2573 NUMA_NO_NODE
, __builtin_return_address(0));
2575 EXPORT_SYMBOL(vmalloc_32
);
2578 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
2579 * @size: allocation size
2581 * The resulting memory area is 32bit addressable and zeroed so it can be
2582 * mapped to userspace without leaking data.
2584 * Return: pointer to the allocated memory or %NULL on error
2586 void *vmalloc_32_user(unsigned long size
)
2588 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
2589 GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
2590 VM_USERMAP
, NUMA_NO_NODE
,
2591 __builtin_return_address(0));
2593 EXPORT_SYMBOL(vmalloc_32_user
);
2596 * small helper routine , copy contents to buf from addr.
2597 * If the page is not present, fill zero.
2600 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
2606 unsigned long offset
, length
;
2608 offset
= offset_in_page(addr
);
2609 length
= PAGE_SIZE
- offset
;
2612 p
= vmalloc_to_page(addr
);
2614 * To do safe access to this _mapped_ area, we need
2615 * lock. But adding lock here means that we need to add
2616 * overhead of vmalloc()/vfree() calles for this _debug_
2617 * interface, rarely used. Instead of that, we'll use
2618 * kmap() and get small overhead in this access function.
2622 * we can expect USER0 is not used (see vread/vwrite's
2623 * function description)
2625 void *map
= kmap_atomic(p
);
2626 memcpy(buf
, map
+ offset
, length
);
2629 memset(buf
, 0, length
);
2639 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
2645 unsigned long offset
, length
;
2647 offset
= offset_in_page(addr
);
2648 length
= PAGE_SIZE
- offset
;
2651 p
= vmalloc_to_page(addr
);
2653 * To do safe access to this _mapped_ area, we need
2654 * lock. But adding lock here means that we need to add
2655 * overhead of vmalloc()/vfree() calles for this _debug_
2656 * interface, rarely used. Instead of that, we'll use
2657 * kmap() and get small overhead in this access function.
2661 * we can expect USER0 is not used (see vread/vwrite's
2662 * function description)
2664 void *map
= kmap_atomic(p
);
2665 memcpy(map
+ offset
, buf
, length
);
2677 * vread() - read vmalloc area in a safe way.
2678 * @buf: buffer for reading data
2679 * @addr: vm address.
2680 * @count: number of bytes to be read.
2682 * This function checks that addr is a valid vmalloc'ed area, and
2683 * copy data from that area to a given buffer. If the given memory range
2684 * of [addr...addr+count) includes some valid address, data is copied to
2685 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2686 * IOREMAP area is treated as memory hole and no copy is done.
2688 * If [addr...addr+count) doesn't includes any intersects with alive
2689 * vm_struct area, returns 0. @buf should be kernel's buffer.
2691 * Note: In usual ops, vread() is never necessary because the caller
2692 * should know vmalloc() area is valid and can use memcpy().
2693 * This is for routines which have to access vmalloc area without
2694 * any informaion, as /dev/kmem.
2696 * Return: number of bytes for which addr and buf should be increased
2697 * (same number as @count) or %0 if [addr...addr+count) doesn't
2698 * include any intersection with valid vmalloc area
2700 long vread(char *buf
, char *addr
, unsigned long count
)
2702 struct vmap_area
*va
;
2703 struct vm_struct
*vm
;
2704 char *vaddr
, *buf_start
= buf
;
2705 unsigned long buflen
= count
;
2708 /* Don't allow overflow */
2709 if ((unsigned long) addr
+ count
< count
)
2710 count
= -(unsigned long) addr
;
2712 spin_lock(&vmap_area_lock
);
2713 list_for_each_entry(va
, &vmap_area_list
, list
) {
2717 if (!(va
->flags
& VM_VM_AREA
))
2721 vaddr
= (char *) vm
->addr
;
2722 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2724 while (addr
< vaddr
) {
2732 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2735 if (!(vm
->flags
& VM_IOREMAP
))
2736 aligned_vread(buf
, addr
, n
);
2737 else /* IOREMAP area is treated as memory hole */
2744 spin_unlock(&vmap_area_lock
);
2746 if (buf
== buf_start
)
2748 /* zero-fill memory holes */
2749 if (buf
!= buf_start
+ buflen
)
2750 memset(buf
, 0, buflen
- (buf
- buf_start
));
2756 * vwrite() - write vmalloc area in a safe way.
2757 * @buf: buffer for source data
2758 * @addr: vm address.
2759 * @count: number of bytes to be read.
2761 * This function checks that addr is a valid vmalloc'ed area, and
2762 * copy data from a buffer to the given addr. If specified range of
2763 * [addr...addr+count) includes some valid address, data is copied from
2764 * proper area of @buf. If there are memory holes, no copy to hole.
2765 * IOREMAP area is treated as memory hole and no copy is done.
2767 * If [addr...addr+count) doesn't includes any intersects with alive
2768 * vm_struct area, returns 0. @buf should be kernel's buffer.
2770 * Note: In usual ops, vwrite() is never necessary because the caller
2771 * should know vmalloc() area is valid and can use memcpy().
2772 * This is for routines which have to access vmalloc area without
2773 * any informaion, as /dev/kmem.
2775 * Return: number of bytes for which addr and buf should be
2776 * increased (same number as @count) or %0 if [addr...addr+count)
2777 * doesn't include any intersection with valid vmalloc area
2779 long vwrite(char *buf
, char *addr
, unsigned long count
)
2781 struct vmap_area
*va
;
2782 struct vm_struct
*vm
;
2784 unsigned long n
, buflen
;
2787 /* Don't allow overflow */
2788 if ((unsigned long) addr
+ count
< count
)
2789 count
= -(unsigned long) addr
;
2792 spin_lock(&vmap_area_lock
);
2793 list_for_each_entry(va
, &vmap_area_list
, list
) {
2797 if (!(va
->flags
& VM_VM_AREA
))
2801 vaddr
= (char *) vm
->addr
;
2802 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2804 while (addr
< vaddr
) {
2811 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2814 if (!(vm
->flags
& VM_IOREMAP
)) {
2815 aligned_vwrite(buf
, addr
, n
);
2823 spin_unlock(&vmap_area_lock
);
2830 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2831 * @vma: vma to cover
2832 * @uaddr: target user address to start at
2833 * @kaddr: virtual address of vmalloc kernel memory
2834 * @size: size of map area
2836 * Returns: 0 for success, -Exxx on failure
2838 * This function checks that @kaddr is a valid vmalloc'ed area,
2839 * and that it is big enough to cover the range starting at
2840 * @uaddr in @vma. Will return failure if that criteria isn't
2843 * Similar to remap_pfn_range() (see mm/memory.c)
2845 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2846 void *kaddr
, unsigned long size
)
2848 struct vm_struct
*area
;
2850 size
= PAGE_ALIGN(size
);
2852 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2855 area
= find_vm_area(kaddr
);
2859 if (!(area
->flags
& VM_USERMAP
))
2862 if (kaddr
+ size
> area
->addr
+ get_vm_area_size(area
))
2866 struct page
*page
= vmalloc_to_page(kaddr
);
2869 ret
= vm_insert_page(vma
, uaddr
, page
);
2878 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2882 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2885 * remap_vmalloc_range - map vmalloc pages to userspace
2886 * @vma: vma to cover (map full range of vma)
2887 * @addr: vmalloc memory
2888 * @pgoff: number of pages into addr before first page to map
2890 * Returns: 0 for success, -Exxx on failure
2892 * This function checks that addr is a valid vmalloc'ed area, and
2893 * that it is big enough to cover the vma. Will return failure if
2894 * that criteria isn't met.
2896 * Similar to remap_pfn_range() (see mm/memory.c)
2898 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2899 unsigned long pgoff
)
2901 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2902 addr
+ (pgoff
<< PAGE_SHIFT
),
2903 vma
->vm_end
- vma
->vm_start
);
2905 EXPORT_SYMBOL(remap_vmalloc_range
);
2908 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2911 void __weak
vmalloc_sync_all(void)
2916 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2928 * alloc_vm_area - allocate a range of kernel address space
2929 * @size: size of the area
2930 * @ptes: returns the PTEs for the address space
2932 * Returns: NULL on failure, vm_struct on success
2934 * This function reserves a range of kernel address space, and
2935 * allocates pagetables to map that range. No actual mappings
2938 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2939 * allocated for the VM area are returned.
2941 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2943 struct vm_struct
*area
;
2945 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2946 __builtin_return_address(0));
2951 * This ensures that page tables are constructed for this region
2952 * of kernel virtual address space and mapped into init_mm.
2954 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2955 size
, f
, ptes
? &ptes
: NULL
)) {
2962 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2964 void free_vm_area(struct vm_struct
*area
)
2966 struct vm_struct
*ret
;
2967 ret
= remove_vm_area(area
->addr
);
2968 BUG_ON(ret
!= area
);
2971 EXPORT_SYMBOL_GPL(free_vm_area
);
2974 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2976 return rb_entry_safe(n
, struct vmap_area
, rb_node
);
2980 * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to
2981 * @addr: target address
2983 * Returns: vmap_area if it is found. If there is no such area
2984 * the first highest(reverse order) vmap_area is returned
2985 * i.e. va->va_start < addr && va->va_end < addr or NULL
2986 * if there are no any areas before @addr.
2988 static struct vmap_area
*
2989 pvm_find_va_enclose_addr(unsigned long addr
)
2991 struct vmap_area
*va
, *tmp
;
2994 n
= free_vmap_area_root
.rb_node
;
2998 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
2999 if (tmp
->va_start
<= addr
) {
3001 if (tmp
->va_end
>= addr
)
3014 * pvm_determine_end_from_reverse - find the highest aligned address
3015 * of free block below VMALLOC_END
3017 * in - the VA we start the search(reverse order);
3018 * out - the VA with the highest aligned end address.
3020 * Returns: determined end address within vmap_area
3022 static unsigned long
3023 pvm_determine_end_from_reverse(struct vmap_area
**va
, unsigned long align
)
3025 unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
3029 list_for_each_entry_from_reverse((*va
),
3030 &free_vmap_area_list
, list
) {
3031 addr
= min((*va
)->va_end
& ~(align
- 1), vmalloc_end
);
3032 if ((*va
)->va_start
< addr
)
3041 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
3042 * @offsets: array containing offset of each area
3043 * @sizes: array containing size of each area
3044 * @nr_vms: the number of areas to allocate
3045 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
3047 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
3048 * vm_structs on success, %NULL on failure
3050 * Percpu allocator wants to use congruent vm areas so that it can
3051 * maintain the offsets among percpu areas. This function allocates
3052 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
3053 * be scattered pretty far, distance between two areas easily going up
3054 * to gigabytes. To avoid interacting with regular vmallocs, these
3055 * areas are allocated from top.
3057 * Despite its complicated look, this allocator is rather simple. It
3058 * does everything top-down and scans free blocks from the end looking
3059 * for matching base. While scanning, if any of the areas do not fit the
3060 * base address is pulled down to fit the area. Scanning is repeated till
3061 * all the areas fit and then all necessary data structures are inserted
3062 * and the result is returned.
3064 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
3065 const size_t *sizes
, int nr_vms
,
3068 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
3069 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
3070 struct vmap_area
**vas
, *va
;
3071 struct vm_struct
**vms
;
3072 int area
, area2
, last_area
, term_area
;
3073 unsigned long base
, start
, size
, end
, last_end
;
3074 bool purged
= false;
3077 /* verify parameters and allocate data structures */
3078 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
3079 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
3080 start
= offsets
[area
];
3081 end
= start
+ sizes
[area
];
3083 /* is everything aligned properly? */
3084 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
3085 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
3087 /* detect the area with the highest address */
3088 if (start
> offsets
[last_area
])
3091 for (area2
= area
+ 1; area2
< nr_vms
; area2
++) {
3092 unsigned long start2
= offsets
[area2
];
3093 unsigned long end2
= start2
+ sizes
[area2
];
3095 BUG_ON(start2
< end
&& start
< end2
);
3098 last_end
= offsets
[last_area
] + sizes
[last_area
];
3100 if (vmalloc_end
- vmalloc_start
< last_end
) {
3105 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
3106 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
3110 for (area
= 0; area
< nr_vms
; area
++) {
3111 vas
[area
] = kmem_cache_zalloc(vmap_area_cachep
, GFP_KERNEL
);
3112 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
3113 if (!vas
[area
] || !vms
[area
])
3117 spin_lock(&vmap_area_lock
);
3119 /* start scanning - we scan from the top, begin with the last area */
3120 area
= term_area
= last_area
;
3121 start
= offsets
[area
];
3122 end
= start
+ sizes
[area
];
3124 va
= pvm_find_va_enclose_addr(vmalloc_end
);
3125 base
= pvm_determine_end_from_reverse(&va
, align
) - end
;
3129 * base might have underflowed, add last_end before
3132 if (base
+ last_end
< vmalloc_start
+ last_end
)
3136 * Fitting base has not been found.
3142 * If this VA does not fit, move base downwards and recheck.
3144 if (base
+ start
< va
->va_start
|| base
+ end
> va
->va_end
) {
3145 va
= node_to_va(rb_prev(&va
->rb_node
));
3146 base
= pvm_determine_end_from_reverse(&va
, align
) - end
;
3152 * This area fits, move on to the previous one. If
3153 * the previous one is the terminal one, we're done.
3155 area
= (area
+ nr_vms
- 1) % nr_vms
;
3156 if (area
== term_area
)
3159 start
= offsets
[area
];
3160 end
= start
+ sizes
[area
];
3161 va
= pvm_find_va_enclose_addr(base
+ end
);
3164 /* we've found a fitting base, insert all va's */
3165 for (area
= 0; area
< nr_vms
; area
++) {
3168 start
= base
+ offsets
[area
];
3171 va
= pvm_find_va_enclose_addr(start
);
3172 if (WARN_ON_ONCE(va
== NULL
))
3173 /* It is a BUG(), but trigger recovery instead. */
3176 type
= classify_va_fit_type(va
, start
, size
);
3177 if (WARN_ON_ONCE(type
== NOTHING_FIT
))
3178 /* It is a BUG(), but trigger recovery instead. */
3181 ret
= adjust_va_to_fit_type(va
, start
, size
, type
);
3185 /* Allocated area. */
3187 va
->va_start
= start
;
3188 va
->va_end
= start
+ size
;
3190 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
3193 spin_unlock(&vmap_area_lock
);
3195 /* insert all vm's */
3196 for (area
= 0; area
< nr_vms
; area
++)
3197 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
3204 /* Remove previously inserted areas. */
3206 __free_vmap_area(vas
[area
]);
3211 spin_unlock(&vmap_area_lock
);
3213 purge_vmap_area_lazy();
3216 /* Before "retry", check if we recover. */
3217 for (area
= 0; area
< nr_vms
; area
++) {
3221 vas
[area
] = kmem_cache_zalloc(
3222 vmap_area_cachep
, GFP_KERNEL
);
3231 for (area
= 0; area
< nr_vms
; area
++) {
3233 kmem_cache_free(vmap_area_cachep
, vas
[area
]);
3244 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
3245 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
3246 * @nr_vms: the number of allocated areas
3248 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
3250 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
3254 for (i
= 0; i
< nr_vms
; i
++)
3255 free_vm_area(vms
[i
]);
3258 #endif /* CONFIG_SMP */
3260 #ifdef CONFIG_PROC_FS
3261 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
3262 __acquires(&vmap_area_lock
)
3264 spin_lock(&vmap_area_lock
);
3265 return seq_list_start(&vmap_area_list
, *pos
);
3268 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
3270 return seq_list_next(p
, &vmap_area_list
, pos
);
3273 static void s_stop(struct seq_file
*m
, void *p
)
3274 __releases(&vmap_area_lock
)
3276 spin_unlock(&vmap_area_lock
);
3279 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
3281 if (IS_ENABLED(CONFIG_NUMA
)) {
3282 unsigned int nr
, *counters
= m
->private;
3287 if (v
->flags
& VM_UNINITIALIZED
)
3289 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
3292 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
3294 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
3295 counters
[page_to_nid(v
->pages
[nr
])]++;
3297 for_each_node_state(nr
, N_HIGH_MEMORY
)
3299 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
3303 static int s_show(struct seq_file
*m
, void *p
)
3305 struct vmap_area
*va
;
3306 struct vm_struct
*v
;
3308 va
= list_entry(p
, struct vmap_area
, list
);
3311 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
3312 * behalf of vmap area is being tear down or vm_map_ram allocation.
3314 if (!(va
->flags
& VM_VM_AREA
)) {
3315 seq_printf(m
, "0x%pK-0x%pK %7ld %s\n",
3316 (void *)va
->va_start
, (void *)va
->va_end
,
3317 va
->va_end
- va
->va_start
,
3318 va
->flags
& VM_LAZY_FREE
? "unpurged vm_area" : "vm_map_ram");
3325 seq_printf(m
, "0x%pK-0x%pK %7ld",
3326 v
->addr
, v
->addr
+ v
->size
, v
->size
);
3329 seq_printf(m
, " %pS", v
->caller
);
3332 seq_printf(m
, " pages=%d", v
->nr_pages
);
3335 seq_printf(m
, " phys=%pa", &v
->phys_addr
);
3337 if (v
->flags
& VM_IOREMAP
)
3338 seq_puts(m
, " ioremap");
3340 if (v
->flags
& VM_ALLOC
)
3341 seq_puts(m
, " vmalloc");
3343 if (v
->flags
& VM_MAP
)
3344 seq_puts(m
, " vmap");
3346 if (v
->flags
& VM_USERMAP
)
3347 seq_puts(m
, " user");
3349 if (is_vmalloc_addr(v
->pages
))
3350 seq_puts(m
, " vpages");
3352 show_numa_info(m
, v
);
3357 static const struct seq_operations vmalloc_op
= {
3364 static int __init
proc_vmalloc_init(void)
3366 if (IS_ENABLED(CONFIG_NUMA
))
3367 proc_create_seq_private("vmallocinfo", 0400, NULL
,
3369 nr_node_ids
* sizeof(unsigned int), NULL
);
3371 proc_create_seq("vmallocinfo", 0400, NULL
, &vmalloc_op
);
3374 module_init(proc_vmalloc_init
);