1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1993 Linus Torvalds
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
7 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
8 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
9 * Numa awareness, Christoph Lameter, SGI, June 2005
12 #include <linux/vmalloc.h>
14 #include <linux/module.h>
15 #include <linux/highmem.h>
16 #include <linux/sched/signal.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <linux/interrupt.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/set_memory.h>
23 #include <linux/debugobjects.h>
24 #include <linux/kallsyms.h>
25 #include <linux/list.h>
26 #include <linux/notifier.h>
27 #include <linux/rbtree.h>
28 #include <linux/radix-tree.h>
29 #include <linux/rcupdate.h>
30 #include <linux/pfn.h>
31 #include <linux/kmemleak.h>
32 #include <linux/atomic.h>
33 #include <linux/compiler.h>
34 #include <linux/llist.h>
35 #include <linux/bitops.h>
36 #include <linux/rbtree_augmented.h>
38 #include <linux/uaccess.h>
39 #include <asm/tlbflush.h>
40 #include <asm/shmparam.h>
44 struct vfree_deferred
{
45 struct llist_head list
;
46 struct work_struct wq
;
48 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
50 static void __vunmap(const void *, int);
52 static void free_work(struct work_struct
*w
)
54 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
55 struct llist_node
*t
, *llnode
;
57 llist_for_each_safe(llnode
, t
, llist_del_all(&p
->list
))
58 __vunmap((void *)llnode
, 1);
61 /*** Page table manipulation functions ***/
63 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
67 pte
= pte_offset_kernel(pmd
, addr
);
69 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
70 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
71 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
74 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
79 pmd
= pmd_offset(pud
, addr
);
81 next
= pmd_addr_end(addr
, end
);
82 if (pmd_clear_huge(pmd
))
84 if (pmd_none_or_clear_bad(pmd
))
86 vunmap_pte_range(pmd
, addr
, next
);
87 } while (pmd
++, addr
= next
, addr
!= end
);
90 static void vunmap_pud_range(p4d_t
*p4d
, unsigned long addr
, unsigned long end
)
95 pud
= pud_offset(p4d
, addr
);
97 next
= pud_addr_end(addr
, end
);
98 if (pud_clear_huge(pud
))
100 if (pud_none_or_clear_bad(pud
))
102 vunmap_pmd_range(pud
, addr
, next
);
103 } while (pud
++, addr
= next
, addr
!= end
);
106 static void vunmap_p4d_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
111 p4d
= p4d_offset(pgd
, addr
);
113 next
= p4d_addr_end(addr
, end
);
114 if (p4d_clear_huge(p4d
))
116 if (p4d_none_or_clear_bad(p4d
))
118 vunmap_pud_range(p4d
, addr
, next
);
119 } while (p4d
++, addr
= next
, addr
!= end
);
122 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
128 pgd
= pgd_offset_k(addr
);
130 next
= pgd_addr_end(addr
, end
);
131 if (pgd_none_or_clear_bad(pgd
))
133 vunmap_p4d_range(pgd
, addr
, next
);
134 } while (pgd
++, addr
= next
, addr
!= end
);
137 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
138 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
143 * nr is a running index into the array which helps higher level
144 * callers keep track of where we're up to.
147 pte
= pte_alloc_kernel(pmd
, addr
);
151 struct page
*page
= pages
[*nr
];
153 if (WARN_ON(!pte_none(*pte
)))
157 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
159 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
163 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
164 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
169 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
173 next
= pmd_addr_end(addr
, end
);
174 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
176 } while (pmd
++, addr
= next
, addr
!= end
);
180 static int vmap_pud_range(p4d_t
*p4d
, unsigned long addr
,
181 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
186 pud
= pud_alloc(&init_mm
, p4d
, addr
);
190 next
= pud_addr_end(addr
, end
);
191 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
193 } while (pud
++, addr
= next
, addr
!= end
);
197 static int vmap_p4d_range(pgd_t
*pgd
, unsigned long addr
,
198 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
203 p4d
= p4d_alloc(&init_mm
, pgd
, addr
);
207 next
= p4d_addr_end(addr
, end
);
208 if (vmap_pud_range(p4d
, addr
, next
, prot
, pages
, nr
))
210 } while (p4d
++, addr
= next
, addr
!= end
);
215 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
216 * will have pfns corresponding to the "pages" array.
218 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
220 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
221 pgprot_t prot
, struct page
**pages
)
225 unsigned long addr
= start
;
230 pgd
= pgd_offset_k(addr
);
232 next
= pgd_addr_end(addr
, end
);
233 err
= vmap_p4d_range(pgd
, addr
, next
, prot
, pages
, &nr
);
236 } while (pgd
++, addr
= next
, addr
!= end
);
241 static int vmap_page_range(unsigned long start
, unsigned long end
,
242 pgprot_t prot
, struct page
**pages
)
246 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
247 flush_cache_vmap(start
, end
);
251 int is_vmalloc_or_module_addr(const void *x
)
254 * ARM, x86-64 and sparc64 put modules in a special place,
255 * and fall back on vmalloc() if that fails. Others
256 * just put it in the vmalloc space.
258 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
259 unsigned long addr
= (unsigned long)x
;
260 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
263 return is_vmalloc_addr(x
);
267 * Walk a vmap address to the struct page it maps.
269 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
271 unsigned long addr
= (unsigned long) vmalloc_addr
;
272 struct page
*page
= NULL
;
273 pgd_t
*pgd
= pgd_offset_k(addr
);
280 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
281 * architectures that do not vmalloc module space
283 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
287 p4d
= p4d_offset(pgd
, addr
);
290 pud
= pud_offset(p4d
, addr
);
293 * Don't dereference bad PUD or PMD (below) entries. This will also
294 * identify huge mappings, which we may encounter on architectures
295 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
296 * identified as vmalloc addresses by is_vmalloc_addr(), but are
297 * not [unambiguously] associated with a struct page, so there is
298 * no correct value to return for them.
300 WARN_ON_ONCE(pud_bad(*pud
));
301 if (pud_none(*pud
) || pud_bad(*pud
))
303 pmd
= pmd_offset(pud
, addr
);
304 WARN_ON_ONCE(pmd_bad(*pmd
));
305 if (pmd_none(*pmd
) || pmd_bad(*pmd
))
308 ptep
= pte_offset_map(pmd
, addr
);
310 if (pte_present(pte
))
311 page
= pte_page(pte
);
315 EXPORT_SYMBOL(vmalloc_to_page
);
318 * Map a vmalloc()-space virtual address to the physical page frame number.
320 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
322 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
324 EXPORT_SYMBOL(vmalloc_to_pfn
);
327 /*** Global kva allocator ***/
329 #define DEBUG_AUGMENT_PROPAGATE_CHECK 0
330 #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0
333 static DEFINE_SPINLOCK(vmap_area_lock
);
334 /* Export for kexec only */
335 LIST_HEAD(vmap_area_list
);
336 static LLIST_HEAD(vmap_purge_list
);
337 static struct rb_root vmap_area_root
= RB_ROOT
;
338 static bool vmap_initialized __read_mostly
;
341 * This kmem_cache is used for vmap_area objects. Instead of
342 * allocating from slab we reuse an object from this cache to
343 * make things faster. Especially in "no edge" splitting of
346 static struct kmem_cache
*vmap_area_cachep
;
349 * This linked list is used in pair with free_vmap_area_root.
350 * It gives O(1) access to prev/next to perform fast coalescing.
352 static LIST_HEAD(free_vmap_area_list
);
355 * This augment red-black tree represents the free vmap space.
356 * All vmap_area objects in this tree are sorted by va->va_start
357 * address. It is used for allocation and merging when a vmap
358 * object is released.
360 * Each vmap_area node contains a maximum available free block
361 * of its sub-tree, right or left. Therefore it is possible to
362 * find a lowest match of free area.
364 static struct rb_root free_vmap_area_root
= RB_ROOT
;
367 * Preload a CPU with one object for "no edge" split case. The
368 * aim is to get rid of allocations from the atomic context, thus
369 * to use more permissive allocation masks.
371 static DEFINE_PER_CPU(struct vmap_area
*, ne_fit_preload_node
);
373 static __always_inline
unsigned long
374 va_size(struct vmap_area
*va
)
376 return (va
->va_end
- va
->va_start
);
379 static __always_inline
unsigned long
380 get_subtree_max_size(struct rb_node
*node
)
382 struct vmap_area
*va
;
384 va
= rb_entry_safe(node
, struct vmap_area
, rb_node
);
385 return va
? va
->subtree_max_size
: 0;
389 * Gets called when remove the node and rotate.
391 static __always_inline
unsigned long
392 compute_subtree_max_size(struct vmap_area
*va
)
394 return max3(va_size(va
),
395 get_subtree_max_size(va
->rb_node
.rb_left
),
396 get_subtree_max_size(va
->rb_node
.rb_right
));
399 RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb
,
400 struct vmap_area
, rb_node
, unsigned long, subtree_max_size
, va_size
)
402 static void purge_vmap_area_lazy(void);
403 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list
);
404 static unsigned long lazy_max_pages(void);
406 static atomic_long_t nr_vmalloc_pages
;
408 unsigned long vmalloc_nr_pages(void)
410 return atomic_long_read(&nr_vmalloc_pages
);
413 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
415 struct rb_node
*n
= vmap_area_root
.rb_node
;
418 struct vmap_area
*va
;
420 va
= rb_entry(n
, struct vmap_area
, rb_node
);
421 if (addr
< va
->va_start
)
423 else if (addr
>= va
->va_end
)
433 * This function returns back addresses of parent node
434 * and its left or right link for further processing.
436 static __always_inline
struct rb_node
**
437 find_va_links(struct vmap_area
*va
,
438 struct rb_root
*root
, struct rb_node
*from
,
439 struct rb_node
**parent
)
441 struct vmap_area
*tmp_va
;
442 struct rb_node
**link
;
445 link
= &root
->rb_node
;
446 if (unlikely(!*link
)) {
455 * Go to the bottom of the tree. When we hit the last point
456 * we end up with parent rb_node and correct direction, i name
457 * it link, where the new va->rb_node will be attached to.
460 tmp_va
= rb_entry(*link
, struct vmap_area
, rb_node
);
463 * During the traversal we also do some sanity check.
464 * Trigger the BUG() if there are sides(left/right)
467 if (va
->va_start
< tmp_va
->va_end
&&
468 va
->va_end
<= tmp_va
->va_start
)
469 link
= &(*link
)->rb_left
;
470 else if (va
->va_end
> tmp_va
->va_start
&&
471 va
->va_start
>= tmp_va
->va_end
)
472 link
= &(*link
)->rb_right
;
477 *parent
= &tmp_va
->rb_node
;
481 static __always_inline
struct list_head
*
482 get_va_next_sibling(struct rb_node
*parent
, struct rb_node
**link
)
484 struct list_head
*list
;
486 if (unlikely(!parent
))
488 * The red-black tree where we try to find VA neighbors
489 * before merging or inserting is empty, i.e. it means
490 * there is no free vmap space. Normally it does not
491 * happen but we handle this case anyway.
495 list
= &rb_entry(parent
, struct vmap_area
, rb_node
)->list
;
496 return (&parent
->rb_right
== link
? list
->next
: list
);
499 static __always_inline
void
500 link_va(struct vmap_area
*va
, struct rb_root
*root
,
501 struct rb_node
*parent
, struct rb_node
**link
, struct list_head
*head
)
504 * VA is still not in the list, but we can
505 * identify its future previous list_head node.
507 if (likely(parent
)) {
508 head
= &rb_entry(parent
, struct vmap_area
, rb_node
)->list
;
509 if (&parent
->rb_right
!= link
)
513 /* Insert to the rb-tree */
514 rb_link_node(&va
->rb_node
, parent
, link
);
515 if (root
== &free_vmap_area_root
) {
517 * Some explanation here. Just perform simple insertion
518 * to the tree. We do not set va->subtree_max_size to
519 * its current size before calling rb_insert_augmented().
520 * It is because of we populate the tree from the bottom
521 * to parent levels when the node _is_ in the tree.
523 * Therefore we set subtree_max_size to zero after insertion,
524 * to let __augment_tree_propagate_from() puts everything to
525 * the correct order later on.
527 rb_insert_augmented(&va
->rb_node
,
528 root
, &free_vmap_area_rb_augment_cb
);
529 va
->subtree_max_size
= 0;
531 rb_insert_color(&va
->rb_node
, root
);
534 /* Address-sort this list */
535 list_add(&va
->list
, head
);
538 static __always_inline
void
539 unlink_va(struct vmap_area
*va
, struct rb_root
*root
)
541 if (WARN_ON(RB_EMPTY_NODE(&va
->rb_node
)))
544 if (root
== &free_vmap_area_root
)
545 rb_erase_augmented(&va
->rb_node
,
546 root
, &free_vmap_area_rb_augment_cb
);
548 rb_erase(&va
->rb_node
, root
);
551 RB_CLEAR_NODE(&va
->rb_node
);
554 #if DEBUG_AUGMENT_PROPAGATE_CHECK
556 augment_tree_propagate_check(struct rb_node
*n
)
558 struct vmap_area
*va
;
559 struct rb_node
*node
;
566 va
= rb_entry(n
, struct vmap_area
, rb_node
);
567 size
= va
->subtree_max_size
;
571 va
= rb_entry(node
, struct vmap_area
, rb_node
);
573 if (get_subtree_max_size(node
->rb_left
) == size
) {
574 node
= node
->rb_left
;
576 if (va_size(va
) == size
) {
581 node
= node
->rb_right
;
586 va
= rb_entry(n
, struct vmap_area
, rb_node
);
587 pr_emerg("tree is corrupted: %lu, %lu\n",
588 va_size(va
), va
->subtree_max_size
);
591 augment_tree_propagate_check(n
->rb_left
);
592 augment_tree_propagate_check(n
->rb_right
);
597 * This function populates subtree_max_size from bottom to upper
598 * levels starting from VA point. The propagation must be done
599 * when VA size is modified by changing its va_start/va_end. Or
600 * in case of newly inserting of VA to the tree.
602 * It means that __augment_tree_propagate_from() must be called:
603 * - After VA has been inserted to the tree(free path);
604 * - After VA has been shrunk(allocation path);
605 * - After VA has been increased(merging path).
607 * Please note that, it does not mean that upper parent nodes
608 * and their subtree_max_size are recalculated all the time up
617 * For example if we modify the node 4, shrinking it to 2, then
618 * no any modification is required. If we shrink the node 2 to 1
619 * its subtree_max_size is updated only, and set to 1. If we shrink
620 * the node 8 to 6, then its subtree_max_size is set to 6 and parent
623 static __always_inline
void
624 augment_tree_propagate_from(struct vmap_area
*va
)
626 struct rb_node
*node
= &va
->rb_node
;
627 unsigned long new_va_sub_max_size
;
630 va
= rb_entry(node
, struct vmap_area
, rb_node
);
631 new_va_sub_max_size
= compute_subtree_max_size(va
);
634 * If the newly calculated maximum available size of the
635 * subtree is equal to the current one, then it means that
636 * the tree is propagated correctly. So we have to stop at
637 * this point to save cycles.
639 if (va
->subtree_max_size
== new_va_sub_max_size
)
642 va
->subtree_max_size
= new_va_sub_max_size
;
643 node
= rb_parent(&va
->rb_node
);
646 #if DEBUG_AUGMENT_PROPAGATE_CHECK
647 augment_tree_propagate_check(free_vmap_area_root
.rb_node
);
652 insert_vmap_area(struct vmap_area
*va
,
653 struct rb_root
*root
, struct list_head
*head
)
655 struct rb_node
**link
;
656 struct rb_node
*parent
;
658 link
= find_va_links(va
, root
, NULL
, &parent
);
659 link_va(va
, root
, parent
, link
, head
);
663 insert_vmap_area_augment(struct vmap_area
*va
,
664 struct rb_node
*from
, struct rb_root
*root
,
665 struct list_head
*head
)
667 struct rb_node
**link
;
668 struct rb_node
*parent
;
671 link
= find_va_links(va
, NULL
, from
, &parent
);
673 link
= find_va_links(va
, root
, NULL
, &parent
);
675 link_va(va
, root
, parent
, link
, head
);
676 augment_tree_propagate_from(va
);
680 * Merge de-allocated chunk of VA memory with previous
681 * and next free blocks. If coalesce is not done a new
682 * free area is inserted. If VA has been merged, it is
685 static __always_inline
void
686 merge_or_add_vmap_area(struct vmap_area
*va
,
687 struct rb_root
*root
, struct list_head
*head
)
689 struct vmap_area
*sibling
;
690 struct list_head
*next
;
691 struct rb_node
**link
;
692 struct rb_node
*parent
;
696 * Find a place in the tree where VA potentially will be
697 * inserted, unless it is merged with its sibling/siblings.
699 link
= find_va_links(va
, root
, NULL
, &parent
);
702 * Get next node of VA to check if merging can be done.
704 next
= get_va_next_sibling(parent
, link
);
705 if (unlikely(next
== NULL
))
711 * |<------VA------>|<-----Next----->|
716 sibling
= list_entry(next
, struct vmap_area
, list
);
717 if (sibling
->va_start
== va
->va_end
) {
718 sibling
->va_start
= va
->va_start
;
720 /* Check and update the tree if needed. */
721 augment_tree_propagate_from(sibling
);
723 /* Free vmap_area object. */
724 kmem_cache_free(vmap_area_cachep
, va
);
726 /* Point to the new merged area. */
735 * |<-----Prev----->|<------VA------>|
739 if (next
->prev
!= head
) {
740 sibling
= list_entry(next
->prev
, struct vmap_area
, list
);
741 if (sibling
->va_end
== va
->va_start
) {
742 sibling
->va_end
= va
->va_end
;
744 /* Check and update the tree if needed. */
745 augment_tree_propagate_from(sibling
);
750 /* Free vmap_area object. */
751 kmem_cache_free(vmap_area_cachep
, va
);
758 link_va(va
, root
, parent
, link
, head
);
759 augment_tree_propagate_from(va
);
763 static __always_inline
bool
764 is_within_this_va(struct vmap_area
*va
, unsigned long size
,
765 unsigned long align
, unsigned long vstart
)
767 unsigned long nva_start_addr
;
769 if (va
->va_start
> vstart
)
770 nva_start_addr
= ALIGN(va
->va_start
, align
);
772 nva_start_addr
= ALIGN(vstart
, align
);
774 /* Can be overflowed due to big size or alignment. */
775 if (nva_start_addr
+ size
< nva_start_addr
||
776 nva_start_addr
< vstart
)
779 return (nva_start_addr
+ size
<= va
->va_end
);
783 * Find the first free block(lowest start address) in the tree,
784 * that will accomplish the request corresponding to passing
787 static __always_inline
struct vmap_area
*
788 find_vmap_lowest_match(unsigned long size
,
789 unsigned long align
, unsigned long vstart
)
791 struct vmap_area
*va
;
792 struct rb_node
*node
;
793 unsigned long length
;
795 /* Start from the root. */
796 node
= free_vmap_area_root
.rb_node
;
798 /* Adjust the search size for alignment overhead. */
799 length
= size
+ align
- 1;
802 va
= rb_entry(node
, struct vmap_area
, rb_node
);
804 if (get_subtree_max_size(node
->rb_left
) >= length
&&
805 vstart
< va
->va_start
) {
806 node
= node
->rb_left
;
808 if (is_within_this_va(va
, size
, align
, vstart
))
812 * Does not make sense to go deeper towards the right
813 * sub-tree if it does not have a free block that is
814 * equal or bigger to the requested search length.
816 if (get_subtree_max_size(node
->rb_right
) >= length
) {
817 node
= node
->rb_right
;
822 * OK. We roll back and find the first right sub-tree,
823 * that will satisfy the search criteria. It can happen
824 * only once due to "vstart" restriction.
826 while ((node
= rb_parent(node
))) {
827 va
= rb_entry(node
, struct vmap_area
, rb_node
);
828 if (is_within_this_va(va
, size
, align
, vstart
))
831 if (get_subtree_max_size(node
->rb_right
) >= length
&&
832 vstart
<= va
->va_start
) {
833 node
= node
->rb_right
;
843 #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
844 #include <linux/random.h>
846 static struct vmap_area
*
847 find_vmap_lowest_linear_match(unsigned long size
,
848 unsigned long align
, unsigned long vstart
)
850 struct vmap_area
*va
;
852 list_for_each_entry(va
, &free_vmap_area_list
, list
) {
853 if (!is_within_this_va(va
, size
, align
, vstart
))
863 find_vmap_lowest_match_check(unsigned long size
)
865 struct vmap_area
*va_1
, *va_2
;
866 unsigned long vstart
;
869 get_random_bytes(&rnd
, sizeof(rnd
));
870 vstart
= VMALLOC_START
+ rnd
;
872 va_1
= find_vmap_lowest_match(size
, 1, vstart
);
873 va_2
= find_vmap_lowest_linear_match(size
, 1, vstart
);
876 pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n",
883 FL_FIT_TYPE
= 1, /* full fit */
884 LE_FIT_TYPE
= 2, /* left edge fit */
885 RE_FIT_TYPE
= 3, /* right edge fit */
886 NE_FIT_TYPE
= 4 /* no edge fit */
889 static __always_inline
enum fit_type
890 classify_va_fit_type(struct vmap_area
*va
,
891 unsigned long nva_start_addr
, unsigned long size
)
895 /* Check if it is within VA. */
896 if (nva_start_addr
< va
->va_start
||
897 nva_start_addr
+ size
> va
->va_end
)
901 if (va
->va_start
== nva_start_addr
) {
902 if (va
->va_end
== nva_start_addr
+ size
)
906 } else if (va
->va_end
== nva_start_addr
+ size
) {
915 static __always_inline
int
916 adjust_va_to_fit_type(struct vmap_area
*va
,
917 unsigned long nva_start_addr
, unsigned long size
,
920 struct vmap_area
*lva
= NULL
;
922 if (type
== FL_FIT_TYPE
) {
924 * No need to split VA, it fully fits.
930 unlink_va(va
, &free_vmap_area_root
);
931 kmem_cache_free(vmap_area_cachep
, va
);
932 } else if (type
== LE_FIT_TYPE
) {
934 * Split left edge of fit VA.
940 va
->va_start
+= size
;
941 } else if (type
== RE_FIT_TYPE
) {
943 * Split right edge of fit VA.
949 va
->va_end
= nva_start_addr
;
950 } else if (type
== NE_FIT_TYPE
) {
952 * Split no edge of fit VA.
958 lva
= __this_cpu_xchg(ne_fit_preload_node
, NULL
);
959 if (unlikely(!lva
)) {
961 * For percpu allocator we do not do any pre-allocation
962 * and leave it as it is. The reason is it most likely
963 * never ends up with NE_FIT_TYPE splitting. In case of
964 * percpu allocations offsets and sizes are aligned to
965 * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE
966 * are its main fitting cases.
968 * There are a few exceptions though, as an example it is
969 * a first allocation (early boot up) when we have "one"
970 * big free space that has to be split.
972 lva
= kmem_cache_alloc(vmap_area_cachep
, GFP_NOWAIT
);
978 * Build the remainder.
980 lva
->va_start
= va
->va_start
;
981 lva
->va_end
= nva_start_addr
;
984 * Shrink this VA to remaining size.
986 va
->va_start
= nva_start_addr
+ size
;
991 if (type
!= FL_FIT_TYPE
) {
992 augment_tree_propagate_from(va
);
994 if (lva
) /* type == NE_FIT_TYPE */
995 insert_vmap_area_augment(lva
, &va
->rb_node
,
996 &free_vmap_area_root
, &free_vmap_area_list
);
1003 * Returns a start address of the newly allocated area, if success.
1004 * Otherwise a vend is returned that indicates failure.
1006 static __always_inline
unsigned long
1007 __alloc_vmap_area(unsigned long size
, unsigned long align
,
1008 unsigned long vstart
, unsigned long vend
)
1010 unsigned long nva_start_addr
;
1011 struct vmap_area
*va
;
1015 va
= find_vmap_lowest_match(size
, align
, vstart
);
1019 if (va
->va_start
> vstart
)
1020 nva_start_addr
= ALIGN(va
->va_start
, align
);
1022 nva_start_addr
= ALIGN(vstart
, align
);
1024 /* Check the "vend" restriction. */
1025 if (nva_start_addr
+ size
> vend
)
1028 /* Classify what we have found. */
1029 type
= classify_va_fit_type(va
, nva_start_addr
, size
);
1030 if (WARN_ON_ONCE(type
== NOTHING_FIT
))
1033 /* Update the free vmap_area. */
1034 ret
= adjust_va_to_fit_type(va
, nva_start_addr
, size
, type
);
1038 #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
1039 find_vmap_lowest_match_check(size
);
1042 return nva_start_addr
;
1046 * Allocate a region of KVA of the specified size and alignment, within the
1049 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
1050 unsigned long align
,
1051 unsigned long vstart
, unsigned long vend
,
1052 int node
, gfp_t gfp_mask
)
1054 struct vmap_area
*va
, *pva
;
1059 BUG_ON(offset_in_page(size
));
1060 BUG_ON(!is_power_of_2(align
));
1062 if (unlikely(!vmap_initialized
))
1063 return ERR_PTR(-EBUSY
);
1067 va
= kmem_cache_alloc_node(vmap_area_cachep
,
1068 gfp_mask
& GFP_RECLAIM_MASK
, node
);
1070 return ERR_PTR(-ENOMEM
);
1073 * Only scan the relevant parts containing pointers to other objects
1074 * to avoid false negatives.
1076 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
1080 * Preload this CPU with one extra vmap_area object to ensure
1081 * that we have it available when fit type of free area is
1084 * The preload is done in non-atomic context, thus it allows us
1085 * to use more permissive allocation masks to be more stable under
1086 * low memory condition and high memory pressure.
1088 * Even if it fails we do not really care about that. Just proceed
1089 * as it is. "overflow" path will refill the cache we allocate from.
1092 if (!__this_cpu_read(ne_fit_preload_node
)) {
1094 pva
= kmem_cache_alloc_node(vmap_area_cachep
, GFP_KERNEL
, node
);
1097 if (__this_cpu_cmpxchg(ne_fit_preload_node
, NULL
, pva
)) {
1099 kmem_cache_free(vmap_area_cachep
, pva
);
1103 spin_lock(&vmap_area_lock
);
1107 * If an allocation fails, the "vend" address is
1108 * returned. Therefore trigger the overflow path.
1110 addr
= __alloc_vmap_area(size
, align
, vstart
, vend
);
1111 if (unlikely(addr
== vend
))
1114 va
->va_start
= addr
;
1115 va
->va_end
= addr
+ size
;
1117 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
1119 spin_unlock(&vmap_area_lock
);
1121 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
1122 BUG_ON(va
->va_start
< vstart
);
1123 BUG_ON(va
->va_end
> vend
);
1128 spin_unlock(&vmap_area_lock
);
1130 purge_vmap_area_lazy();
1135 if (gfpflags_allow_blocking(gfp_mask
)) {
1136 unsigned long freed
= 0;
1137 blocking_notifier_call_chain(&vmap_notify_list
, 0, &freed
);
1144 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit())
1145 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
1148 kmem_cache_free(vmap_area_cachep
, va
);
1149 return ERR_PTR(-EBUSY
);
1152 int register_vmap_purge_notifier(struct notifier_block
*nb
)
1154 return blocking_notifier_chain_register(&vmap_notify_list
, nb
);
1156 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier
);
1158 int unregister_vmap_purge_notifier(struct notifier_block
*nb
)
1160 return blocking_notifier_chain_unregister(&vmap_notify_list
, nb
);
1162 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier
);
1164 static void __free_vmap_area(struct vmap_area
*va
)
1167 * Remove from the busy tree/list.
1169 unlink_va(va
, &vmap_area_root
);
1172 * Merge VA with its neighbors, otherwise just add it.
1174 merge_or_add_vmap_area(va
,
1175 &free_vmap_area_root
, &free_vmap_area_list
);
1179 * Free a region of KVA allocated by alloc_vmap_area
1181 static void free_vmap_area(struct vmap_area
*va
)
1183 spin_lock(&vmap_area_lock
);
1184 __free_vmap_area(va
);
1185 spin_unlock(&vmap_area_lock
);
1189 * Clear the pagetable entries of a given vmap_area
1191 static void unmap_vmap_area(struct vmap_area
*va
)
1193 vunmap_page_range(va
->va_start
, va
->va_end
);
1197 * lazy_max_pages is the maximum amount of virtual address space we gather up
1198 * before attempting to purge with a TLB flush.
1200 * There is a tradeoff here: a larger number will cover more kernel page tables
1201 * and take slightly longer to purge, but it will linearly reduce the number of
1202 * global TLB flushes that must be performed. It would seem natural to scale
1203 * this number up linearly with the number of CPUs (because vmapping activity
1204 * could also scale linearly with the number of CPUs), however it is likely
1205 * that in practice, workloads might be constrained in other ways that mean
1206 * vmap activity will not scale linearly with CPUs. Also, I want to be
1207 * conservative and not introduce a big latency on huge systems, so go with
1208 * a less aggressive log scale. It will still be an improvement over the old
1209 * code, and it will be simple to change the scale factor if we find that it
1210 * becomes a problem on bigger systems.
1212 static unsigned long lazy_max_pages(void)
1216 log
= fls(num_online_cpus());
1218 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
1221 static atomic_long_t vmap_lazy_nr
= ATOMIC_LONG_INIT(0);
1224 * Serialize vmap purging. There is no actual criticial section protected
1225 * by this look, but we want to avoid concurrent calls for performance
1226 * reasons and to make the pcpu_get_vm_areas more deterministic.
1228 static DEFINE_MUTEX(vmap_purge_lock
);
1230 /* for per-CPU blocks */
1231 static void purge_fragmented_blocks_allcpus(void);
1234 * called before a call to iounmap() if the caller wants vm_area_struct's
1235 * immediately freed.
1237 void set_iounmap_nonlazy(void)
1239 atomic_long_set(&vmap_lazy_nr
, lazy_max_pages()+1);
1243 * Purges all lazily-freed vmap areas.
1245 static bool __purge_vmap_area_lazy(unsigned long start
, unsigned long end
)
1247 unsigned long resched_threshold
;
1248 struct llist_node
*valist
;
1249 struct vmap_area
*va
;
1250 struct vmap_area
*n_va
;
1252 lockdep_assert_held(&vmap_purge_lock
);
1254 valist
= llist_del_all(&vmap_purge_list
);
1255 if (unlikely(valist
== NULL
))
1259 * First make sure the mappings are removed from all page-tables
1260 * before they are freed.
1265 * TODO: to calculate a flush range without looping.
1266 * The list can be up to lazy_max_pages() elements.
1268 llist_for_each_entry(va
, valist
, purge_list
) {
1269 if (va
->va_start
< start
)
1270 start
= va
->va_start
;
1271 if (va
->va_end
> end
)
1275 flush_tlb_kernel_range(start
, end
);
1276 resched_threshold
= lazy_max_pages() << 1;
1278 spin_lock(&vmap_area_lock
);
1279 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
) {
1280 unsigned long nr
= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
1283 * Finally insert or merge lazily-freed area. It is
1284 * detached and there is no need to "unlink" it from
1287 merge_or_add_vmap_area(va
,
1288 &free_vmap_area_root
, &free_vmap_area_list
);
1290 atomic_long_sub(nr
, &vmap_lazy_nr
);
1292 if (atomic_long_read(&vmap_lazy_nr
) < resched_threshold
)
1293 cond_resched_lock(&vmap_area_lock
);
1295 spin_unlock(&vmap_area_lock
);
1300 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
1301 * is already purging.
1303 static void try_purge_vmap_area_lazy(void)
1305 if (mutex_trylock(&vmap_purge_lock
)) {
1306 __purge_vmap_area_lazy(ULONG_MAX
, 0);
1307 mutex_unlock(&vmap_purge_lock
);
1312 * Kick off a purge of the outstanding lazy areas.
1314 static void purge_vmap_area_lazy(void)
1316 mutex_lock(&vmap_purge_lock
);
1317 purge_fragmented_blocks_allcpus();
1318 __purge_vmap_area_lazy(ULONG_MAX
, 0);
1319 mutex_unlock(&vmap_purge_lock
);
1323 * Free a vmap area, caller ensuring that the area has been unmapped
1324 * and flush_cache_vunmap had been called for the correct range
1327 static void free_vmap_area_noflush(struct vmap_area
*va
)
1329 unsigned long nr_lazy
;
1331 spin_lock(&vmap_area_lock
);
1332 unlink_va(va
, &vmap_area_root
);
1333 spin_unlock(&vmap_area_lock
);
1335 nr_lazy
= atomic_long_add_return((va
->va_end
- va
->va_start
) >>
1336 PAGE_SHIFT
, &vmap_lazy_nr
);
1338 /* After this point, we may free va at any time */
1339 llist_add(&va
->purge_list
, &vmap_purge_list
);
1341 if (unlikely(nr_lazy
> lazy_max_pages()))
1342 try_purge_vmap_area_lazy();
1346 * Free and unmap a vmap area
1348 static void free_unmap_vmap_area(struct vmap_area
*va
)
1350 flush_cache_vunmap(va
->va_start
, va
->va_end
);
1351 unmap_vmap_area(va
);
1352 if (debug_pagealloc_enabled())
1353 flush_tlb_kernel_range(va
->va_start
, va
->va_end
);
1355 free_vmap_area_noflush(va
);
1358 static struct vmap_area
*find_vmap_area(unsigned long addr
)
1360 struct vmap_area
*va
;
1362 spin_lock(&vmap_area_lock
);
1363 va
= __find_vmap_area(addr
);
1364 spin_unlock(&vmap_area_lock
);
1369 /*** Per cpu kva allocator ***/
1372 * vmap space is limited especially on 32 bit architectures. Ensure there is
1373 * room for at least 16 percpu vmap blocks per CPU.
1376 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
1377 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
1378 * instead (we just need a rough idea)
1380 #if BITS_PER_LONG == 32
1381 #define VMALLOC_SPACE (128UL*1024*1024)
1383 #define VMALLOC_SPACE (128UL*1024*1024*1024)
1386 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
1387 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
1388 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
1389 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
1390 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
1391 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
1392 #define VMAP_BBMAP_BITS \
1393 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
1394 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
1395 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
1397 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
1399 struct vmap_block_queue
{
1401 struct list_head free
;
1406 struct vmap_area
*va
;
1407 unsigned long free
, dirty
;
1408 unsigned long dirty_min
, dirty_max
; /*< dirty range */
1409 struct list_head free_list
;
1410 struct rcu_head rcu_head
;
1411 struct list_head purge
;
1414 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
1415 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
1418 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
1419 * in the free path. Could get rid of this if we change the API to return a
1420 * "cookie" from alloc, to be passed to free. But no big deal yet.
1422 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
1423 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
1426 * We should probably have a fallback mechanism to allocate virtual memory
1427 * out of partially filled vmap blocks. However vmap block sizing should be
1428 * fairly reasonable according to the vmalloc size, so it shouldn't be a
1432 static unsigned long addr_to_vb_idx(unsigned long addr
)
1434 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
1435 addr
/= VMAP_BLOCK_SIZE
;
1439 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
1443 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
1444 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
1445 return (void *)addr
;
1449 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
1450 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
1451 * @order: how many 2^order pages should be occupied in newly allocated block
1452 * @gfp_mask: flags for the page level allocator
1454 * Return: virtual address in a newly allocated block or ERR_PTR(-errno)
1456 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
1458 struct vmap_block_queue
*vbq
;
1459 struct vmap_block
*vb
;
1460 struct vmap_area
*va
;
1461 unsigned long vb_idx
;
1465 node
= numa_node_id();
1467 vb
= kmalloc_node(sizeof(struct vmap_block
),
1468 gfp_mask
& GFP_RECLAIM_MASK
, node
);
1470 return ERR_PTR(-ENOMEM
);
1472 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
1473 VMALLOC_START
, VMALLOC_END
,
1477 return ERR_CAST(va
);
1480 err
= radix_tree_preload(gfp_mask
);
1481 if (unlikely(err
)) {
1484 return ERR_PTR(err
);
1487 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
1488 spin_lock_init(&vb
->lock
);
1490 /* At least something should be left free */
1491 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
1492 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
1494 vb
->dirty_min
= VMAP_BBMAP_BITS
;
1496 INIT_LIST_HEAD(&vb
->free_list
);
1498 vb_idx
= addr_to_vb_idx(va
->va_start
);
1499 spin_lock(&vmap_block_tree_lock
);
1500 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
1501 spin_unlock(&vmap_block_tree_lock
);
1503 radix_tree_preload_end();
1505 vbq
= &get_cpu_var(vmap_block_queue
);
1506 spin_lock(&vbq
->lock
);
1507 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
1508 spin_unlock(&vbq
->lock
);
1509 put_cpu_var(vmap_block_queue
);
1514 static void free_vmap_block(struct vmap_block
*vb
)
1516 struct vmap_block
*tmp
;
1517 unsigned long vb_idx
;
1519 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
1520 spin_lock(&vmap_block_tree_lock
);
1521 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
1522 spin_unlock(&vmap_block_tree_lock
);
1525 free_vmap_area_noflush(vb
->va
);
1526 kfree_rcu(vb
, rcu_head
);
1529 static void purge_fragmented_blocks(int cpu
)
1532 struct vmap_block
*vb
;
1533 struct vmap_block
*n_vb
;
1534 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1537 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1539 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
1542 spin_lock(&vb
->lock
);
1543 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
1544 vb
->free
= 0; /* prevent further allocs after releasing lock */
1545 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
1547 vb
->dirty_max
= VMAP_BBMAP_BITS
;
1548 spin_lock(&vbq
->lock
);
1549 list_del_rcu(&vb
->free_list
);
1550 spin_unlock(&vbq
->lock
);
1551 spin_unlock(&vb
->lock
);
1552 list_add_tail(&vb
->purge
, &purge
);
1554 spin_unlock(&vb
->lock
);
1558 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
1559 list_del(&vb
->purge
);
1560 free_vmap_block(vb
);
1564 static void purge_fragmented_blocks_allcpus(void)
1568 for_each_possible_cpu(cpu
)
1569 purge_fragmented_blocks(cpu
);
1572 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
1574 struct vmap_block_queue
*vbq
;
1575 struct vmap_block
*vb
;
1579 BUG_ON(offset_in_page(size
));
1580 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1581 if (WARN_ON(size
== 0)) {
1583 * Allocating 0 bytes isn't what caller wants since
1584 * get_order(0) returns funny result. Just warn and terminate
1589 order
= get_order(size
);
1592 vbq
= &get_cpu_var(vmap_block_queue
);
1593 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1594 unsigned long pages_off
;
1596 spin_lock(&vb
->lock
);
1597 if (vb
->free
< (1UL << order
)) {
1598 spin_unlock(&vb
->lock
);
1602 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
1603 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
1604 vb
->free
-= 1UL << order
;
1605 if (vb
->free
== 0) {
1606 spin_lock(&vbq
->lock
);
1607 list_del_rcu(&vb
->free_list
);
1608 spin_unlock(&vbq
->lock
);
1611 spin_unlock(&vb
->lock
);
1615 put_cpu_var(vmap_block_queue
);
1618 /* Allocate new block if nothing was found */
1620 vaddr
= new_vmap_block(order
, gfp_mask
);
1625 static void vb_free(const void *addr
, unsigned long size
)
1627 unsigned long offset
;
1628 unsigned long vb_idx
;
1630 struct vmap_block
*vb
;
1632 BUG_ON(offset_in_page(size
));
1633 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1635 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
1637 order
= get_order(size
);
1639 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
1640 offset
>>= PAGE_SHIFT
;
1642 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
1644 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1648 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1650 if (debug_pagealloc_enabled())
1651 flush_tlb_kernel_range((unsigned long)addr
,
1652 (unsigned long)addr
+ size
);
1654 spin_lock(&vb
->lock
);
1656 /* Expand dirty range */
1657 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1658 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1660 vb
->dirty
+= 1UL << order
;
1661 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1663 spin_unlock(&vb
->lock
);
1664 free_vmap_block(vb
);
1666 spin_unlock(&vb
->lock
);
1669 static void _vm_unmap_aliases(unsigned long start
, unsigned long end
, int flush
)
1673 if (unlikely(!vmap_initialized
))
1678 for_each_possible_cpu(cpu
) {
1679 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1680 struct vmap_block
*vb
;
1683 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1684 spin_lock(&vb
->lock
);
1686 unsigned long va_start
= vb
->va
->va_start
;
1689 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1690 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1692 start
= min(s
, start
);
1697 spin_unlock(&vb
->lock
);
1702 mutex_lock(&vmap_purge_lock
);
1703 purge_fragmented_blocks_allcpus();
1704 if (!__purge_vmap_area_lazy(start
, end
) && flush
)
1705 flush_tlb_kernel_range(start
, end
);
1706 mutex_unlock(&vmap_purge_lock
);
1710 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1712 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1713 * to amortize TLB flushing overheads. What this means is that any page you
1714 * have now, may, in a former life, have been mapped into kernel virtual
1715 * address by the vmap layer and so there might be some CPUs with TLB entries
1716 * still referencing that page (additional to the regular 1:1 kernel mapping).
1718 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1719 * be sure that none of the pages we have control over will have any aliases
1720 * from the vmap layer.
1722 void vm_unmap_aliases(void)
1724 unsigned long start
= ULONG_MAX
, end
= 0;
1727 _vm_unmap_aliases(start
, end
, flush
);
1729 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1732 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1733 * @mem: the pointer returned by vm_map_ram
1734 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1736 void vm_unmap_ram(const void *mem
, unsigned int count
)
1738 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1739 unsigned long addr
= (unsigned long)mem
;
1740 struct vmap_area
*va
;
1744 BUG_ON(addr
< VMALLOC_START
);
1745 BUG_ON(addr
> VMALLOC_END
);
1746 BUG_ON(!PAGE_ALIGNED(addr
));
1748 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1749 debug_check_no_locks_freed(mem
, size
);
1754 va
= find_vmap_area(addr
);
1756 debug_check_no_locks_freed((void *)va
->va_start
,
1757 (va
->va_end
- va
->va_start
));
1758 free_unmap_vmap_area(va
);
1760 EXPORT_SYMBOL(vm_unmap_ram
);
1763 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1764 * @pages: an array of pointers to the pages to be mapped
1765 * @count: number of pages
1766 * @node: prefer to allocate data structures on this node
1767 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1769 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1770 * faster than vmap so it's good. But if you mix long-life and short-life
1771 * objects with vm_map_ram(), it could consume lots of address space through
1772 * fragmentation (especially on a 32bit machine). You could see failures in
1773 * the end. Please use this function for short-lived objects.
1775 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1777 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1779 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1783 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1784 mem
= vb_alloc(size
, GFP_KERNEL
);
1787 addr
= (unsigned long)mem
;
1789 struct vmap_area
*va
;
1790 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1791 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1795 addr
= va
->va_start
;
1798 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1799 vm_unmap_ram(mem
, count
);
1804 EXPORT_SYMBOL(vm_map_ram
);
1806 static struct vm_struct
*vmlist __initdata
;
1809 * vm_area_add_early - add vmap area early during boot
1810 * @vm: vm_struct to add
1812 * This function is used to add fixed kernel vm area to vmlist before
1813 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1814 * should contain proper values and the other fields should be zero.
1816 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1818 void __init
vm_area_add_early(struct vm_struct
*vm
)
1820 struct vm_struct
*tmp
, **p
;
1822 BUG_ON(vmap_initialized
);
1823 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1824 if (tmp
->addr
>= vm
->addr
) {
1825 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1828 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1835 * vm_area_register_early - register vmap area early during boot
1836 * @vm: vm_struct to register
1837 * @align: requested alignment
1839 * This function is used to register kernel vm area before
1840 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1841 * proper values on entry and other fields should be zero. On return,
1842 * vm->addr contains the allocated address.
1844 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1846 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1848 static size_t vm_init_off __initdata
;
1851 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1852 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1854 vm
->addr
= (void *)addr
;
1856 vm_area_add_early(vm
);
1859 static void vmap_init_free_space(void)
1861 unsigned long vmap_start
= 1;
1862 const unsigned long vmap_end
= ULONG_MAX
;
1863 struct vmap_area
*busy
, *free
;
1867 * -|-----|.....|-----|-----|-----|.....|-
1869 * |<--------------------------------->|
1871 list_for_each_entry(busy
, &vmap_area_list
, list
) {
1872 if (busy
->va_start
- vmap_start
> 0) {
1873 free
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1874 if (!WARN_ON_ONCE(!free
)) {
1875 free
->va_start
= vmap_start
;
1876 free
->va_end
= busy
->va_start
;
1878 insert_vmap_area_augment(free
, NULL
,
1879 &free_vmap_area_root
,
1880 &free_vmap_area_list
);
1884 vmap_start
= busy
->va_end
;
1887 if (vmap_end
- vmap_start
> 0) {
1888 free
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1889 if (!WARN_ON_ONCE(!free
)) {
1890 free
->va_start
= vmap_start
;
1891 free
->va_end
= vmap_end
;
1893 insert_vmap_area_augment(free
, NULL
,
1894 &free_vmap_area_root
,
1895 &free_vmap_area_list
);
1900 void __init
vmalloc_init(void)
1902 struct vmap_area
*va
;
1903 struct vm_struct
*tmp
;
1907 * Create the cache for vmap_area objects.
1909 vmap_area_cachep
= KMEM_CACHE(vmap_area
, SLAB_PANIC
);
1911 for_each_possible_cpu(i
) {
1912 struct vmap_block_queue
*vbq
;
1913 struct vfree_deferred
*p
;
1915 vbq
= &per_cpu(vmap_block_queue
, i
);
1916 spin_lock_init(&vbq
->lock
);
1917 INIT_LIST_HEAD(&vbq
->free
);
1918 p
= &per_cpu(vfree_deferred
, i
);
1919 init_llist_head(&p
->list
);
1920 INIT_WORK(&p
->wq
, free_work
);
1923 /* Import existing vmlist entries. */
1924 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1925 va
= kmem_cache_zalloc(vmap_area_cachep
, GFP_NOWAIT
);
1926 if (WARN_ON_ONCE(!va
))
1929 va
->va_start
= (unsigned long)tmp
->addr
;
1930 va
->va_end
= va
->va_start
+ tmp
->size
;
1932 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
1936 * Now we can initialize a free vmap space.
1938 vmap_init_free_space();
1939 vmap_initialized
= true;
1943 * map_kernel_range_noflush - map kernel VM area with the specified pages
1944 * @addr: start of the VM area to map
1945 * @size: size of the VM area to map
1946 * @prot: page protection flags to use
1947 * @pages: pages to map
1949 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1950 * specify should have been allocated using get_vm_area() and its
1954 * This function does NOT do any cache flushing. The caller is
1955 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1956 * before calling this function.
1959 * The number of pages mapped on success, -errno on failure.
1961 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1962 pgprot_t prot
, struct page
**pages
)
1964 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1968 * unmap_kernel_range_noflush - unmap kernel VM area
1969 * @addr: start of the VM area to unmap
1970 * @size: size of the VM area to unmap
1972 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1973 * specify should have been allocated using get_vm_area() and its
1977 * This function does NOT do any cache flushing. The caller is
1978 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1979 * before calling this function and flush_tlb_kernel_range() after.
1981 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1983 vunmap_page_range(addr
, addr
+ size
);
1985 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1988 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1989 * @addr: start of the VM area to unmap
1990 * @size: size of the VM area to unmap
1992 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1993 * the unmapping and tlb after.
1995 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1997 unsigned long end
= addr
+ size
;
1999 flush_cache_vunmap(addr
, end
);
2000 vunmap_page_range(addr
, end
);
2001 flush_tlb_kernel_range(addr
, end
);
2003 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
2005 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
2007 unsigned long addr
= (unsigned long)area
->addr
;
2008 unsigned long end
= addr
+ get_vm_area_size(area
);
2011 err
= vmap_page_range(addr
, end
, prot
, pages
);
2013 return err
> 0 ? 0 : err
;
2015 EXPORT_SYMBOL_GPL(map_vm_area
);
2017 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
2018 unsigned long flags
, const void *caller
)
2020 spin_lock(&vmap_area_lock
);
2022 vm
->addr
= (void *)va
->va_start
;
2023 vm
->size
= va
->va_end
- va
->va_start
;
2024 vm
->caller
= caller
;
2026 spin_unlock(&vmap_area_lock
);
2029 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
2032 * Before removing VM_UNINITIALIZED,
2033 * we should make sure that vm has proper values.
2034 * Pair with smp_rmb() in show_numa_info().
2037 vm
->flags
&= ~VM_UNINITIALIZED
;
2040 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
2041 unsigned long align
, unsigned long flags
, unsigned long start
,
2042 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
2044 struct vmap_area
*va
;
2045 struct vm_struct
*area
;
2047 BUG_ON(in_interrupt());
2048 size
= PAGE_ALIGN(size
);
2049 if (unlikely(!size
))
2052 if (flags
& VM_IOREMAP
)
2053 align
= 1ul << clamp_t(int, get_count_order_long(size
),
2054 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
2056 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
2057 if (unlikely(!area
))
2060 if (!(flags
& VM_NO_GUARD
))
2063 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
2069 setup_vmalloc_vm(area
, va
, flags
, caller
);
2074 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
2075 unsigned long start
, unsigned long end
)
2077 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
2078 GFP_KERNEL
, __builtin_return_address(0));
2080 EXPORT_SYMBOL_GPL(__get_vm_area
);
2082 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
2083 unsigned long start
, unsigned long end
,
2086 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
2087 GFP_KERNEL
, caller
);
2091 * get_vm_area - reserve a contiguous kernel virtual area
2092 * @size: size of the area
2093 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
2095 * Search an area of @size in the kernel virtual mapping area,
2096 * and reserved it for out purposes. Returns the area descriptor
2097 * on success or %NULL on failure.
2099 * Return: the area descriptor on success or %NULL on failure.
2101 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
2103 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
2104 NUMA_NO_NODE
, GFP_KERNEL
,
2105 __builtin_return_address(0));
2108 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
2111 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
2112 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
2116 * find_vm_area - find a continuous kernel virtual area
2117 * @addr: base address
2119 * Search for the kernel VM area starting at @addr, and return it.
2120 * It is up to the caller to do all required locking to keep the returned
2123 * Return: pointer to the found area or %NULL on faulure
2125 struct vm_struct
*find_vm_area(const void *addr
)
2127 struct vmap_area
*va
;
2129 va
= find_vmap_area((unsigned long)addr
);
2137 * remove_vm_area - find and remove a continuous kernel virtual area
2138 * @addr: base address
2140 * Search for the kernel VM area starting at @addr, and remove it.
2141 * This function returns the found VM area, but using it is NOT safe
2142 * on SMP machines, except for its size or flags.
2144 * Return: pointer to the found area or %NULL on faulure
2146 struct vm_struct
*remove_vm_area(const void *addr
)
2148 struct vmap_area
*va
;
2152 spin_lock(&vmap_area_lock
);
2153 va
= __find_vmap_area((unsigned long)addr
);
2155 struct vm_struct
*vm
= va
->vm
;
2158 spin_unlock(&vmap_area_lock
);
2160 kasan_free_shadow(vm
);
2161 free_unmap_vmap_area(va
);
2166 spin_unlock(&vmap_area_lock
);
2170 static inline void set_area_direct_map(const struct vm_struct
*area
,
2171 int (*set_direct_map
)(struct page
*page
))
2175 for (i
= 0; i
< area
->nr_pages
; i
++)
2176 if (page_address(area
->pages
[i
]))
2177 set_direct_map(area
->pages
[i
]);
2180 /* Handle removing and resetting vm mappings related to the vm_struct. */
2181 static void vm_remove_mappings(struct vm_struct
*area
, int deallocate_pages
)
2183 unsigned long start
= ULONG_MAX
, end
= 0;
2184 int flush_reset
= area
->flags
& VM_FLUSH_RESET_PERMS
;
2188 remove_vm_area(area
->addr
);
2190 /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */
2195 * If not deallocating pages, just do the flush of the VM area and
2198 if (!deallocate_pages
) {
2204 * If execution gets here, flush the vm mapping and reset the direct
2205 * map. Find the start and end range of the direct mappings to make sure
2206 * the vm_unmap_aliases() flush includes the direct map.
2208 for (i
= 0; i
< area
->nr_pages
; i
++) {
2209 unsigned long addr
= (unsigned long)page_address(area
->pages
[i
]);
2211 start
= min(addr
, start
);
2212 end
= max(addr
+ PAGE_SIZE
, end
);
2218 * Set direct map to something invalid so that it won't be cached if
2219 * there are any accesses after the TLB flush, then flush the TLB and
2220 * reset the direct map permissions to the default.
2222 set_area_direct_map(area
, set_direct_map_invalid_noflush
);
2223 _vm_unmap_aliases(start
, end
, flush_dmap
);
2224 set_area_direct_map(area
, set_direct_map_default_noflush
);
2227 static void __vunmap(const void *addr
, int deallocate_pages
)
2229 struct vm_struct
*area
;
2234 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
2238 area
= find_vm_area(addr
);
2239 if (unlikely(!area
)) {
2240 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
2245 debug_check_no_locks_freed(area
->addr
, get_vm_area_size(area
));
2246 debug_check_no_obj_freed(area
->addr
, get_vm_area_size(area
));
2248 vm_remove_mappings(area
, deallocate_pages
);
2250 if (deallocate_pages
) {
2253 for (i
= 0; i
< area
->nr_pages
; i
++) {
2254 struct page
*page
= area
->pages
[i
];
2257 __free_pages(page
, 0);
2259 atomic_long_sub(area
->nr_pages
, &nr_vmalloc_pages
);
2261 kvfree(area
->pages
);
2268 static inline void __vfree_deferred(const void *addr
)
2271 * Use raw_cpu_ptr() because this can be called from preemptible
2272 * context. Preemption is absolutely fine here, because the llist_add()
2273 * implementation is lockless, so it works even if we are adding to
2274 * nother cpu's list. schedule_work() should be fine with this too.
2276 struct vfree_deferred
*p
= raw_cpu_ptr(&vfree_deferred
);
2278 if (llist_add((struct llist_node
*)addr
, &p
->list
))
2279 schedule_work(&p
->wq
);
2283 * vfree_atomic - release memory allocated by vmalloc()
2284 * @addr: memory base address
2286 * This one is just like vfree() but can be called in any atomic context
2289 void vfree_atomic(const void *addr
)
2293 kmemleak_free(addr
);
2297 __vfree_deferred(addr
);
2300 static void __vfree(const void *addr
)
2302 if (unlikely(in_interrupt()))
2303 __vfree_deferred(addr
);
2309 * vfree - release memory allocated by vmalloc()
2310 * @addr: memory base address
2312 * Free the virtually continuous memory area starting at @addr, as
2313 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
2314 * NULL, no operation is performed.
2316 * Must not be called in NMI context (strictly speaking, only if we don't
2317 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
2318 * conventions for vfree() arch-depenedent would be a really bad idea)
2320 * May sleep if called *not* from interrupt context.
2322 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
2324 void vfree(const void *addr
)
2328 kmemleak_free(addr
);
2330 might_sleep_if(!in_interrupt());
2337 EXPORT_SYMBOL(vfree
);
2340 * vunmap - release virtual mapping obtained by vmap()
2341 * @addr: memory base address
2343 * Free the virtually contiguous memory area starting at @addr,
2344 * which was created from the page array passed to vmap().
2346 * Must not be called in interrupt context.
2348 void vunmap(const void *addr
)
2350 BUG_ON(in_interrupt());
2355 EXPORT_SYMBOL(vunmap
);
2358 * vmap - map an array of pages into virtually contiguous space
2359 * @pages: array of page pointers
2360 * @count: number of pages to map
2361 * @flags: vm_area->flags
2362 * @prot: page protection for the mapping
2364 * Maps @count pages from @pages into contiguous kernel virtual
2367 * Return: the address of the area or %NULL on failure
2369 void *vmap(struct page
**pages
, unsigned int count
,
2370 unsigned long flags
, pgprot_t prot
)
2372 struct vm_struct
*area
;
2373 unsigned long size
; /* In bytes */
2377 if (count
> totalram_pages())
2380 size
= (unsigned long)count
<< PAGE_SHIFT
;
2381 area
= get_vm_area_caller(size
, flags
, __builtin_return_address(0));
2385 if (map_vm_area(area
, prot
, pages
)) {
2392 EXPORT_SYMBOL(vmap
);
2394 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
2395 gfp_t gfp_mask
, pgprot_t prot
,
2396 int node
, const void *caller
);
2397 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
2398 pgprot_t prot
, int node
)
2400 struct page
**pages
;
2401 unsigned int nr_pages
, array_size
, i
;
2402 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
2403 const gfp_t alloc_mask
= gfp_mask
| __GFP_NOWARN
;
2404 const gfp_t highmem_mask
= (gfp_mask
& (GFP_DMA
| GFP_DMA32
)) ?
2408 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
2409 array_size
= (nr_pages
* sizeof(struct page
*));
2411 /* Please note that the recursion is strictly bounded. */
2412 if (array_size
> PAGE_SIZE
) {
2413 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|highmem_mask
,
2414 PAGE_KERNEL
, node
, area
->caller
);
2416 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
2420 remove_vm_area(area
->addr
);
2425 area
->pages
= pages
;
2426 area
->nr_pages
= nr_pages
;
2428 for (i
= 0; i
< area
->nr_pages
; i
++) {
2431 if (node
== NUMA_NO_NODE
)
2432 page
= alloc_page(alloc_mask
|highmem_mask
);
2434 page
= alloc_pages_node(node
, alloc_mask
|highmem_mask
, 0);
2436 if (unlikely(!page
)) {
2437 /* Successfully allocated i pages, free them in __vunmap() */
2439 atomic_long_add(area
->nr_pages
, &nr_vmalloc_pages
);
2442 area
->pages
[i
] = page
;
2443 if (gfpflags_allow_blocking(gfp_mask
|highmem_mask
))
2446 atomic_long_add(area
->nr_pages
, &nr_vmalloc_pages
);
2448 if (map_vm_area(area
, prot
, pages
))
2453 warn_alloc(gfp_mask
, NULL
,
2454 "vmalloc: allocation failure, allocated %ld of %ld bytes",
2455 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
2456 __vfree(area
->addr
);
2461 * __vmalloc_node_range - allocate virtually contiguous memory
2462 * @size: allocation size
2463 * @align: desired alignment
2464 * @start: vm area range start
2465 * @end: vm area range end
2466 * @gfp_mask: flags for the page level allocator
2467 * @prot: protection mask for the allocated pages
2468 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
2469 * @node: node to use for allocation or NUMA_NO_NODE
2470 * @caller: caller's return address
2472 * Allocate enough pages to cover @size from the page level
2473 * allocator with @gfp_mask flags. Map them into contiguous
2474 * kernel virtual space, using a pagetable protection of @prot.
2476 * Return: the address of the area or %NULL on failure
2478 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
2479 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
2480 pgprot_t prot
, unsigned long vm_flags
, int node
,
2483 struct vm_struct
*area
;
2485 unsigned long real_size
= size
;
2487 size
= PAGE_ALIGN(size
);
2488 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages())
2491 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
2492 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
2496 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
2501 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
2502 * flag. It means that vm_struct is not fully initialized.
2503 * Now, it is fully initialized, so remove this flag here.
2505 clear_vm_uninitialized_flag(area
);
2507 kmemleak_vmalloc(area
, size
, gfp_mask
);
2512 warn_alloc(gfp_mask
, NULL
,
2513 "vmalloc: allocation failure: %lu bytes", real_size
);
2518 * This is only for performance analysis of vmalloc and stress purpose.
2519 * It is required by vmalloc test module, therefore do not use it other
2522 #ifdef CONFIG_TEST_VMALLOC_MODULE
2523 EXPORT_SYMBOL_GPL(__vmalloc_node_range
);
2527 * __vmalloc_node - allocate virtually contiguous memory
2528 * @size: allocation size
2529 * @align: desired alignment
2530 * @gfp_mask: flags for the page level allocator
2531 * @prot: protection mask for the allocated pages
2532 * @node: node to use for allocation or NUMA_NO_NODE
2533 * @caller: caller's return address
2535 * Allocate enough pages to cover @size from the page level
2536 * allocator with @gfp_mask flags. Map them into contiguous
2537 * kernel virtual space, using a pagetable protection of @prot.
2539 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
2540 * and __GFP_NOFAIL are not supported
2542 * Any use of gfp flags outside of GFP_KERNEL should be consulted
2545 * Return: pointer to the allocated memory or %NULL on error
2547 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
2548 gfp_t gfp_mask
, pgprot_t prot
,
2549 int node
, const void *caller
)
2551 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
2552 gfp_mask
, prot
, 0, node
, caller
);
2555 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
2557 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
2558 __builtin_return_address(0));
2560 EXPORT_SYMBOL(__vmalloc
);
2562 static inline void *__vmalloc_node_flags(unsigned long size
,
2563 int node
, gfp_t flags
)
2565 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
2566 node
, __builtin_return_address(0));
2570 void *__vmalloc_node_flags_caller(unsigned long size
, int node
, gfp_t flags
,
2573 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
, node
, caller
);
2577 * vmalloc - allocate virtually contiguous memory
2578 * @size: allocation size
2580 * Allocate enough pages to cover @size from the page level
2581 * allocator and map them into contiguous kernel virtual space.
2583 * For tight control over page level allocator and protection flags
2584 * use __vmalloc() instead.
2586 * Return: pointer to the allocated memory or %NULL on error
2588 void *vmalloc(unsigned long size
)
2590 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
2593 EXPORT_SYMBOL(vmalloc
);
2596 * vzalloc - allocate virtually contiguous memory with zero fill
2597 * @size: allocation size
2599 * Allocate enough pages to cover @size from the page level
2600 * allocator and map them into contiguous kernel virtual space.
2601 * The memory allocated is set to zero.
2603 * For tight control over page level allocator and protection flags
2604 * use __vmalloc() instead.
2606 * Return: pointer to the allocated memory or %NULL on error
2608 void *vzalloc(unsigned long size
)
2610 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
2611 GFP_KERNEL
| __GFP_ZERO
);
2613 EXPORT_SYMBOL(vzalloc
);
2616 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
2617 * @size: allocation size
2619 * The resulting memory area is zeroed so it can be mapped to userspace
2620 * without leaking data.
2622 * Return: pointer to the allocated memory or %NULL on error
2624 void *vmalloc_user(unsigned long size
)
2626 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
2627 GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
,
2628 VM_USERMAP
, NUMA_NO_NODE
,
2629 __builtin_return_address(0));
2631 EXPORT_SYMBOL(vmalloc_user
);
2634 * vmalloc_node - allocate memory on a specific node
2635 * @size: allocation size
2638 * Allocate enough pages to cover @size from the page level
2639 * allocator and map them into contiguous kernel virtual space.
2641 * For tight control over page level allocator and protection flags
2642 * use __vmalloc() instead.
2644 * Return: pointer to the allocated memory or %NULL on error
2646 void *vmalloc_node(unsigned long size
, int node
)
2648 return __vmalloc_node(size
, 1, GFP_KERNEL
, PAGE_KERNEL
,
2649 node
, __builtin_return_address(0));
2651 EXPORT_SYMBOL(vmalloc_node
);
2654 * vzalloc_node - allocate memory on a specific node with zero fill
2655 * @size: allocation size
2658 * Allocate enough pages to cover @size from the page level
2659 * allocator and map them into contiguous kernel virtual space.
2660 * The memory allocated is set to zero.
2662 * For tight control over page level allocator and protection flags
2663 * use __vmalloc_node() instead.
2665 * Return: pointer to the allocated memory or %NULL on error
2667 void *vzalloc_node(unsigned long size
, int node
)
2669 return __vmalloc_node_flags(size
, node
,
2670 GFP_KERNEL
| __GFP_ZERO
);
2672 EXPORT_SYMBOL(vzalloc_node
);
2675 * vmalloc_exec - allocate virtually contiguous, executable memory
2676 * @size: allocation size
2678 * Kernel-internal function to allocate enough pages to cover @size
2679 * the page level allocator and map them into contiguous and
2680 * executable kernel virtual space.
2682 * For tight control over page level allocator and protection flags
2683 * use __vmalloc() instead.
2685 * Return: pointer to the allocated memory or %NULL on error
2687 void *vmalloc_exec(unsigned long size
)
2689 return __vmalloc_node_range(size
, 1, VMALLOC_START
, VMALLOC_END
,
2690 GFP_KERNEL
, PAGE_KERNEL_EXEC
, VM_FLUSH_RESET_PERMS
,
2691 NUMA_NO_NODE
, __builtin_return_address(0));
2694 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
2695 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
2696 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
2697 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
2700 * 64b systems should always have either DMA or DMA32 zones. For others
2701 * GFP_DMA32 should do the right thing and use the normal zone.
2703 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
2707 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
2708 * @size: allocation size
2710 * Allocate enough 32bit PA addressable pages to cover @size from the
2711 * page level allocator and map them into contiguous kernel virtual space.
2713 * Return: pointer to the allocated memory or %NULL on error
2715 void *vmalloc_32(unsigned long size
)
2717 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
2718 NUMA_NO_NODE
, __builtin_return_address(0));
2720 EXPORT_SYMBOL(vmalloc_32
);
2723 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
2724 * @size: allocation size
2726 * The resulting memory area is 32bit addressable and zeroed so it can be
2727 * mapped to userspace without leaking data.
2729 * Return: pointer to the allocated memory or %NULL on error
2731 void *vmalloc_32_user(unsigned long size
)
2733 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
2734 GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
2735 VM_USERMAP
, NUMA_NO_NODE
,
2736 __builtin_return_address(0));
2738 EXPORT_SYMBOL(vmalloc_32_user
);
2741 * small helper routine , copy contents to buf from addr.
2742 * If the page is not present, fill zero.
2745 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
2751 unsigned long offset
, length
;
2753 offset
= offset_in_page(addr
);
2754 length
= PAGE_SIZE
- offset
;
2757 p
= vmalloc_to_page(addr
);
2759 * To do safe access to this _mapped_ area, we need
2760 * lock. But adding lock here means that we need to add
2761 * overhead of vmalloc()/vfree() calles for this _debug_
2762 * interface, rarely used. Instead of that, we'll use
2763 * kmap() and get small overhead in this access function.
2767 * we can expect USER0 is not used (see vread/vwrite's
2768 * function description)
2770 void *map
= kmap_atomic(p
);
2771 memcpy(buf
, map
+ offset
, length
);
2774 memset(buf
, 0, length
);
2784 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
2790 unsigned long offset
, length
;
2792 offset
= offset_in_page(addr
);
2793 length
= PAGE_SIZE
- offset
;
2796 p
= vmalloc_to_page(addr
);
2798 * To do safe access to this _mapped_ area, we need
2799 * lock. But adding lock here means that we need to add
2800 * overhead of vmalloc()/vfree() calles for this _debug_
2801 * interface, rarely used. Instead of that, we'll use
2802 * kmap() and get small overhead in this access function.
2806 * we can expect USER0 is not used (see vread/vwrite's
2807 * function description)
2809 void *map
= kmap_atomic(p
);
2810 memcpy(map
+ offset
, buf
, length
);
2822 * vread() - read vmalloc area in a safe way.
2823 * @buf: buffer for reading data
2824 * @addr: vm address.
2825 * @count: number of bytes to be read.
2827 * This function checks that addr is a valid vmalloc'ed area, and
2828 * copy data from that area to a given buffer. If the given memory range
2829 * of [addr...addr+count) includes some valid address, data is copied to
2830 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2831 * IOREMAP area is treated as memory hole and no copy is done.
2833 * If [addr...addr+count) doesn't includes any intersects with alive
2834 * vm_struct area, returns 0. @buf should be kernel's buffer.
2836 * Note: In usual ops, vread() is never necessary because the caller
2837 * should know vmalloc() area is valid and can use memcpy().
2838 * This is for routines which have to access vmalloc area without
2839 * any information, as /dev/kmem.
2841 * Return: number of bytes for which addr and buf should be increased
2842 * (same number as @count) or %0 if [addr...addr+count) doesn't
2843 * include any intersection with valid vmalloc area
2845 long vread(char *buf
, char *addr
, unsigned long count
)
2847 struct vmap_area
*va
;
2848 struct vm_struct
*vm
;
2849 char *vaddr
, *buf_start
= buf
;
2850 unsigned long buflen
= count
;
2853 /* Don't allow overflow */
2854 if ((unsigned long) addr
+ count
< count
)
2855 count
= -(unsigned long) addr
;
2857 spin_lock(&vmap_area_lock
);
2858 list_for_each_entry(va
, &vmap_area_list
, list
) {
2866 vaddr
= (char *) vm
->addr
;
2867 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2869 while (addr
< vaddr
) {
2877 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2880 if (!(vm
->flags
& VM_IOREMAP
))
2881 aligned_vread(buf
, addr
, n
);
2882 else /* IOREMAP area is treated as memory hole */
2889 spin_unlock(&vmap_area_lock
);
2891 if (buf
== buf_start
)
2893 /* zero-fill memory holes */
2894 if (buf
!= buf_start
+ buflen
)
2895 memset(buf
, 0, buflen
- (buf
- buf_start
));
2901 * vwrite() - write vmalloc area in a safe way.
2902 * @buf: buffer for source data
2903 * @addr: vm address.
2904 * @count: number of bytes to be read.
2906 * This function checks that addr is a valid vmalloc'ed area, and
2907 * copy data from a buffer to the given addr. If specified range of
2908 * [addr...addr+count) includes some valid address, data is copied from
2909 * proper area of @buf. If there are memory holes, no copy to hole.
2910 * IOREMAP area is treated as memory hole and no copy is done.
2912 * If [addr...addr+count) doesn't includes any intersects with alive
2913 * vm_struct area, returns 0. @buf should be kernel's buffer.
2915 * Note: In usual ops, vwrite() is never necessary because the caller
2916 * should know vmalloc() area is valid and can use memcpy().
2917 * This is for routines which have to access vmalloc area without
2918 * any information, as /dev/kmem.
2920 * Return: number of bytes for which addr and buf should be
2921 * increased (same number as @count) or %0 if [addr...addr+count)
2922 * doesn't include any intersection with valid vmalloc area
2924 long vwrite(char *buf
, char *addr
, unsigned long count
)
2926 struct vmap_area
*va
;
2927 struct vm_struct
*vm
;
2929 unsigned long n
, buflen
;
2932 /* Don't allow overflow */
2933 if ((unsigned long) addr
+ count
< count
)
2934 count
= -(unsigned long) addr
;
2937 spin_lock(&vmap_area_lock
);
2938 list_for_each_entry(va
, &vmap_area_list
, list
) {
2946 vaddr
= (char *) vm
->addr
;
2947 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2949 while (addr
< vaddr
) {
2956 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2959 if (!(vm
->flags
& VM_IOREMAP
)) {
2960 aligned_vwrite(buf
, addr
, n
);
2968 spin_unlock(&vmap_area_lock
);
2975 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2976 * @vma: vma to cover
2977 * @uaddr: target user address to start at
2978 * @kaddr: virtual address of vmalloc kernel memory
2979 * @size: size of map area
2981 * Returns: 0 for success, -Exxx on failure
2983 * This function checks that @kaddr is a valid vmalloc'ed area,
2984 * and that it is big enough to cover the range starting at
2985 * @uaddr in @vma. Will return failure if that criteria isn't
2988 * Similar to remap_pfn_range() (see mm/memory.c)
2990 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2991 void *kaddr
, unsigned long size
)
2993 struct vm_struct
*area
;
2995 size
= PAGE_ALIGN(size
);
2997 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
3000 area
= find_vm_area(kaddr
);
3004 if (!(area
->flags
& (VM_USERMAP
| VM_DMA_COHERENT
)))
3007 if (kaddr
+ size
> area
->addr
+ get_vm_area_size(area
))
3011 struct page
*page
= vmalloc_to_page(kaddr
);
3014 ret
= vm_insert_page(vma
, uaddr
, page
);
3023 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
3027 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
3030 * remap_vmalloc_range - map vmalloc pages to userspace
3031 * @vma: vma to cover (map full range of vma)
3032 * @addr: vmalloc memory
3033 * @pgoff: number of pages into addr before first page to map
3035 * Returns: 0 for success, -Exxx on failure
3037 * This function checks that addr is a valid vmalloc'ed area, and
3038 * that it is big enough to cover the vma. Will return failure if
3039 * that criteria isn't met.
3041 * Similar to remap_pfn_range() (see mm/memory.c)
3043 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
3044 unsigned long pgoff
)
3046 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
3047 addr
+ (pgoff
<< PAGE_SHIFT
),
3048 vma
->vm_end
- vma
->vm_start
);
3050 EXPORT_SYMBOL(remap_vmalloc_range
);
3053 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
3056 * The purpose of this function is to make sure the vmalloc area
3057 * mappings are identical in all page-tables in the system.
3059 void __weak
vmalloc_sync_all(void)
3064 static int f(pte_t
*pte
, unsigned long addr
, void *data
)
3076 * alloc_vm_area - allocate a range of kernel address space
3077 * @size: size of the area
3078 * @ptes: returns the PTEs for the address space
3080 * Returns: NULL on failure, vm_struct on success
3082 * This function reserves a range of kernel address space, and
3083 * allocates pagetables to map that range. No actual mappings
3086 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
3087 * allocated for the VM area are returned.
3089 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
3091 struct vm_struct
*area
;
3093 area
= get_vm_area_caller(size
, VM_IOREMAP
,
3094 __builtin_return_address(0));
3099 * This ensures that page tables are constructed for this region
3100 * of kernel virtual address space and mapped into init_mm.
3102 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
3103 size
, f
, ptes
? &ptes
: NULL
)) {
3110 EXPORT_SYMBOL_GPL(alloc_vm_area
);
3112 void free_vm_area(struct vm_struct
*area
)
3114 struct vm_struct
*ret
;
3115 ret
= remove_vm_area(area
->addr
);
3116 BUG_ON(ret
!= area
);
3119 EXPORT_SYMBOL_GPL(free_vm_area
);
3122 static struct vmap_area
*node_to_va(struct rb_node
*n
)
3124 return rb_entry_safe(n
, struct vmap_area
, rb_node
);
3128 * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to
3129 * @addr: target address
3131 * Returns: vmap_area if it is found. If there is no such area
3132 * the first highest(reverse order) vmap_area is returned
3133 * i.e. va->va_start < addr && va->va_end < addr or NULL
3134 * if there are no any areas before @addr.
3136 static struct vmap_area
*
3137 pvm_find_va_enclose_addr(unsigned long addr
)
3139 struct vmap_area
*va
, *tmp
;
3142 n
= free_vmap_area_root
.rb_node
;
3146 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
3147 if (tmp
->va_start
<= addr
) {
3149 if (tmp
->va_end
>= addr
)
3162 * pvm_determine_end_from_reverse - find the highest aligned address
3163 * of free block below VMALLOC_END
3165 * in - the VA we start the search(reverse order);
3166 * out - the VA with the highest aligned end address.
3168 * Returns: determined end address within vmap_area
3170 static unsigned long
3171 pvm_determine_end_from_reverse(struct vmap_area
**va
, unsigned long align
)
3173 unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
3177 list_for_each_entry_from_reverse((*va
),
3178 &free_vmap_area_list
, list
) {
3179 addr
= min((*va
)->va_end
& ~(align
- 1), vmalloc_end
);
3180 if ((*va
)->va_start
< addr
)
3189 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
3190 * @offsets: array containing offset of each area
3191 * @sizes: array containing size of each area
3192 * @nr_vms: the number of areas to allocate
3193 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
3195 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
3196 * vm_structs on success, %NULL on failure
3198 * Percpu allocator wants to use congruent vm areas so that it can
3199 * maintain the offsets among percpu areas. This function allocates
3200 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
3201 * be scattered pretty far, distance between two areas easily going up
3202 * to gigabytes. To avoid interacting with regular vmallocs, these
3203 * areas are allocated from top.
3205 * Despite its complicated look, this allocator is rather simple. It
3206 * does everything top-down and scans free blocks from the end looking
3207 * for matching base. While scanning, if any of the areas do not fit the
3208 * base address is pulled down to fit the area. Scanning is repeated till
3209 * all the areas fit and then all necessary data structures are inserted
3210 * and the result is returned.
3212 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
3213 const size_t *sizes
, int nr_vms
,
3216 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
3217 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
3218 struct vmap_area
**vas
, *va
;
3219 struct vm_struct
**vms
;
3220 int area
, area2
, last_area
, term_area
;
3221 unsigned long base
, start
, size
, end
, last_end
;
3222 bool purged
= false;
3225 /* verify parameters and allocate data structures */
3226 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
3227 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
3228 start
= offsets
[area
];
3229 end
= start
+ sizes
[area
];
3231 /* is everything aligned properly? */
3232 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
3233 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
3235 /* detect the area with the highest address */
3236 if (start
> offsets
[last_area
])
3239 for (area2
= area
+ 1; area2
< nr_vms
; area2
++) {
3240 unsigned long start2
= offsets
[area2
];
3241 unsigned long end2
= start2
+ sizes
[area2
];
3243 BUG_ON(start2
< end
&& start
< end2
);
3246 last_end
= offsets
[last_area
] + sizes
[last_area
];
3248 if (vmalloc_end
- vmalloc_start
< last_end
) {
3253 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
3254 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
3258 for (area
= 0; area
< nr_vms
; area
++) {
3259 vas
[area
] = kmem_cache_zalloc(vmap_area_cachep
, GFP_KERNEL
);
3260 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
3261 if (!vas
[area
] || !vms
[area
])
3265 spin_lock(&vmap_area_lock
);
3267 /* start scanning - we scan from the top, begin with the last area */
3268 area
= term_area
= last_area
;
3269 start
= offsets
[area
];
3270 end
= start
+ sizes
[area
];
3272 va
= pvm_find_va_enclose_addr(vmalloc_end
);
3273 base
= pvm_determine_end_from_reverse(&va
, align
) - end
;
3277 * base might have underflowed, add last_end before
3280 if (base
+ last_end
< vmalloc_start
+ last_end
)
3284 * Fitting base has not been found.
3290 * If required width exeeds current VA block, move
3291 * base downwards and then recheck.
3293 if (base
+ end
> va
->va_end
) {
3294 base
= pvm_determine_end_from_reverse(&va
, align
) - end
;
3300 * If this VA does not fit, move base downwards and recheck.
3302 if (base
+ start
< va
->va_start
) {
3303 va
= node_to_va(rb_prev(&va
->rb_node
));
3304 base
= pvm_determine_end_from_reverse(&va
, align
) - end
;
3310 * This area fits, move on to the previous one. If
3311 * the previous one is the terminal one, we're done.
3313 area
= (area
+ nr_vms
- 1) % nr_vms
;
3314 if (area
== term_area
)
3317 start
= offsets
[area
];
3318 end
= start
+ sizes
[area
];
3319 va
= pvm_find_va_enclose_addr(base
+ end
);
3322 /* we've found a fitting base, insert all va's */
3323 for (area
= 0; area
< nr_vms
; area
++) {
3326 start
= base
+ offsets
[area
];
3329 va
= pvm_find_va_enclose_addr(start
);
3330 if (WARN_ON_ONCE(va
== NULL
))
3331 /* It is a BUG(), but trigger recovery instead. */
3334 type
= classify_va_fit_type(va
, start
, size
);
3335 if (WARN_ON_ONCE(type
== NOTHING_FIT
))
3336 /* It is a BUG(), but trigger recovery instead. */
3339 ret
= adjust_va_to_fit_type(va
, start
, size
, type
);
3343 /* Allocated area. */
3345 va
->va_start
= start
;
3346 va
->va_end
= start
+ size
;
3348 insert_vmap_area(va
, &vmap_area_root
, &vmap_area_list
);
3351 spin_unlock(&vmap_area_lock
);
3353 /* insert all vm's */
3354 for (area
= 0; area
< nr_vms
; area
++)
3355 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
3362 /* Remove previously inserted areas. */
3364 __free_vmap_area(vas
[area
]);
3369 spin_unlock(&vmap_area_lock
);
3371 purge_vmap_area_lazy();
3374 /* Before "retry", check if we recover. */
3375 for (area
= 0; area
< nr_vms
; area
++) {
3379 vas
[area
] = kmem_cache_zalloc(
3380 vmap_area_cachep
, GFP_KERNEL
);
3389 for (area
= 0; area
< nr_vms
; area
++) {
3391 kmem_cache_free(vmap_area_cachep
, vas
[area
]);
3402 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
3403 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
3404 * @nr_vms: the number of allocated areas
3406 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
3408 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
3412 for (i
= 0; i
< nr_vms
; i
++)
3413 free_vm_area(vms
[i
]);
3416 #endif /* CONFIG_SMP */
3418 #ifdef CONFIG_PROC_FS
3419 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
3420 __acquires(&vmap_area_lock
)
3422 spin_lock(&vmap_area_lock
);
3423 return seq_list_start(&vmap_area_list
, *pos
);
3426 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
3428 return seq_list_next(p
, &vmap_area_list
, pos
);
3431 static void s_stop(struct seq_file
*m
, void *p
)
3432 __releases(&vmap_area_lock
)
3434 spin_unlock(&vmap_area_lock
);
3437 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
3439 if (IS_ENABLED(CONFIG_NUMA
)) {
3440 unsigned int nr
, *counters
= m
->private;
3445 if (v
->flags
& VM_UNINITIALIZED
)
3447 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
3450 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
3452 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
3453 counters
[page_to_nid(v
->pages
[nr
])]++;
3455 for_each_node_state(nr
, N_HIGH_MEMORY
)
3457 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
3461 static void show_purge_info(struct seq_file
*m
)
3463 struct llist_node
*head
;
3464 struct vmap_area
*va
;
3466 head
= READ_ONCE(vmap_purge_list
.first
);
3470 llist_for_each_entry(va
, head
, purge_list
) {
3471 seq_printf(m
, "0x%pK-0x%pK %7ld unpurged vm_area\n",
3472 (void *)va
->va_start
, (void *)va
->va_end
,
3473 va
->va_end
- va
->va_start
);
3477 static int s_show(struct seq_file
*m
, void *p
)
3479 struct vmap_area
*va
;
3480 struct vm_struct
*v
;
3482 va
= list_entry(p
, struct vmap_area
, list
);
3485 * s_show can encounter race with remove_vm_area, !vm on behalf
3486 * of vmap area is being tear down or vm_map_ram allocation.
3489 seq_printf(m
, "0x%pK-0x%pK %7ld vm_map_ram\n",
3490 (void *)va
->va_start
, (void *)va
->va_end
,
3491 va
->va_end
- va
->va_start
);
3498 seq_printf(m
, "0x%pK-0x%pK %7ld",
3499 v
->addr
, v
->addr
+ v
->size
, v
->size
);
3502 seq_printf(m
, " %pS", v
->caller
);
3505 seq_printf(m
, " pages=%d", v
->nr_pages
);
3508 seq_printf(m
, " phys=%pa", &v
->phys_addr
);
3510 if (v
->flags
& VM_IOREMAP
)
3511 seq_puts(m
, " ioremap");
3513 if (v
->flags
& VM_ALLOC
)
3514 seq_puts(m
, " vmalloc");
3516 if (v
->flags
& VM_MAP
)
3517 seq_puts(m
, " vmap");
3519 if (v
->flags
& VM_USERMAP
)
3520 seq_puts(m
, " user");
3522 if (v
->flags
& VM_DMA_COHERENT
)
3523 seq_puts(m
, " dma-coherent");
3525 if (is_vmalloc_addr(v
->pages
))
3526 seq_puts(m
, " vpages");
3528 show_numa_info(m
, v
);
3532 * As a final step, dump "unpurged" areas. Note,
3533 * that entire "/proc/vmallocinfo" output will not
3534 * be address sorted, because the purge list is not
3537 if (list_is_last(&va
->list
, &vmap_area_list
))
3543 static const struct seq_operations vmalloc_op
= {
3550 static int __init
proc_vmalloc_init(void)
3552 if (IS_ENABLED(CONFIG_NUMA
))
3553 proc_create_seq_private("vmallocinfo", 0400, NULL
,
3555 nr_node_ids
* sizeof(unsigned int), NULL
);
3557 proc_create_seq("vmallocinfo", 0400, NULL
, &vmalloc_op
);
3560 module_init(proc_vmalloc_init
);