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.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/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/rbtree.h>
25 #include <linux/radix-tree.h>
26 #include <linux/rcupdate.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
29 #include <linux/atomic.h>
30 #include <linux/compiler.h>
31 #include <linux/llist.h>
32 #include <linux/bitops.h>
34 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
36 #include <asm/shmparam.h>
40 struct vfree_deferred
{
41 struct llist_head list
;
42 struct work_struct wq
;
44 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
46 static void __vunmap(const void *, int);
48 static void free_work(struct work_struct
*w
)
50 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
51 struct llist_node
*llnode
= llist_del_all(&p
->list
);
54 llnode
= llist_next(llnode
);
59 /*** Page table manipulation functions ***/
61 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
65 pte
= pte_offset_kernel(pmd
, addr
);
67 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
68 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
69 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
72 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
77 pmd
= pmd_offset(pud
, addr
);
79 next
= pmd_addr_end(addr
, end
);
80 if (pmd_clear_huge(pmd
))
82 if (pmd_none_or_clear_bad(pmd
))
84 vunmap_pte_range(pmd
, addr
, next
);
85 } while (pmd
++, addr
= next
, addr
!= end
);
88 static void vunmap_pud_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
93 pud
= pud_offset(pgd
, addr
);
95 next
= pud_addr_end(addr
, end
);
96 if (pud_clear_huge(pud
))
98 if (pud_none_or_clear_bad(pud
))
100 vunmap_pmd_range(pud
, addr
, next
);
101 } while (pud
++, addr
= next
, addr
!= end
);
104 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
110 pgd
= pgd_offset_k(addr
);
112 next
= pgd_addr_end(addr
, end
);
113 if (pgd_none_or_clear_bad(pgd
))
115 vunmap_pud_range(pgd
, addr
, next
);
116 } while (pgd
++, addr
= next
, addr
!= end
);
119 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
120 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
125 * nr is a running index into the array which helps higher level
126 * callers keep track of where we're up to.
129 pte
= pte_alloc_kernel(pmd
, addr
);
133 struct page
*page
= pages
[*nr
];
135 if (WARN_ON(!pte_none(*pte
)))
139 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
141 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
145 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
146 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
151 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
155 next
= pmd_addr_end(addr
, end
);
156 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
158 } while (pmd
++, addr
= next
, addr
!= end
);
162 static int vmap_pud_range(pgd_t
*pgd
, unsigned long addr
,
163 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
168 pud
= pud_alloc(&init_mm
, pgd
, addr
);
172 next
= pud_addr_end(addr
, end
);
173 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
175 } while (pud
++, addr
= next
, addr
!= end
);
180 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
181 * will have pfns corresponding to the "pages" array.
183 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
185 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
186 pgprot_t prot
, struct page
**pages
)
190 unsigned long addr
= start
;
195 pgd
= pgd_offset_k(addr
);
197 next
= pgd_addr_end(addr
, end
);
198 err
= vmap_pud_range(pgd
, addr
, next
, prot
, pages
, &nr
);
201 } while (pgd
++, addr
= next
, addr
!= end
);
206 static int vmap_page_range(unsigned long start
, unsigned long end
,
207 pgprot_t prot
, struct page
**pages
)
211 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
212 flush_cache_vmap(start
, end
);
216 int is_vmalloc_or_module_addr(const void *x
)
219 * ARM, x86-64 and sparc64 put modules in a special place,
220 * and fall back on vmalloc() if that fails. Others
221 * just put it in the vmalloc space.
223 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
224 unsigned long addr
= (unsigned long)x
;
225 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
228 return is_vmalloc_addr(x
);
232 * Walk a vmap address to the struct page it maps.
234 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
236 unsigned long addr
= (unsigned long) vmalloc_addr
;
237 struct page
*page
= NULL
;
238 pgd_t
*pgd
= pgd_offset_k(addr
);
241 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
242 * architectures that do not vmalloc module space
244 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
246 if (!pgd_none(*pgd
)) {
247 pud_t
*pud
= pud_offset(pgd
, addr
);
248 if (!pud_none(*pud
)) {
249 pmd_t
*pmd
= pmd_offset(pud
, addr
);
250 if (!pmd_none(*pmd
)) {
253 ptep
= pte_offset_map(pmd
, addr
);
255 if (pte_present(pte
))
256 page
= pte_page(pte
);
263 EXPORT_SYMBOL(vmalloc_to_page
);
266 * Map a vmalloc()-space virtual address to the physical page frame number.
268 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
270 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
272 EXPORT_SYMBOL(vmalloc_to_pfn
);
275 /*** Global kva allocator ***/
277 #define VM_VM_AREA 0x04
279 static DEFINE_SPINLOCK(vmap_area_lock
);
280 /* Export for kexec only */
281 LIST_HEAD(vmap_area_list
);
282 static LLIST_HEAD(vmap_purge_list
);
283 static struct rb_root vmap_area_root
= RB_ROOT
;
285 /* The vmap cache globals are protected by vmap_area_lock */
286 static struct rb_node
*free_vmap_cache
;
287 static unsigned long cached_hole_size
;
288 static unsigned long cached_vstart
;
289 static unsigned long cached_align
;
291 static unsigned long vmap_area_pcpu_hole
;
293 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
295 struct rb_node
*n
= vmap_area_root
.rb_node
;
298 struct vmap_area
*va
;
300 va
= rb_entry(n
, struct vmap_area
, rb_node
);
301 if (addr
< va
->va_start
)
303 else if (addr
>= va
->va_end
)
312 static void __insert_vmap_area(struct vmap_area
*va
)
314 struct rb_node
**p
= &vmap_area_root
.rb_node
;
315 struct rb_node
*parent
= NULL
;
319 struct vmap_area
*tmp_va
;
322 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
323 if (va
->va_start
< tmp_va
->va_end
)
325 else if (va
->va_end
> tmp_va
->va_start
)
331 rb_link_node(&va
->rb_node
, parent
, p
);
332 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
334 /* address-sort this list */
335 tmp
= rb_prev(&va
->rb_node
);
337 struct vmap_area
*prev
;
338 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
339 list_add_rcu(&va
->list
, &prev
->list
);
341 list_add_rcu(&va
->list
, &vmap_area_list
);
344 static void purge_vmap_area_lazy(void);
347 * Allocate a region of KVA of the specified size and alignment, within the
350 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
352 unsigned long vstart
, unsigned long vend
,
353 int node
, gfp_t gfp_mask
)
355 struct vmap_area
*va
;
359 struct vmap_area
*first
;
362 BUG_ON(offset_in_page(size
));
363 BUG_ON(!is_power_of_2(align
));
365 va
= kmalloc_node(sizeof(struct vmap_area
),
366 gfp_mask
& GFP_RECLAIM_MASK
, node
);
368 return ERR_PTR(-ENOMEM
);
371 * Only scan the relevant parts containing pointers to other objects
372 * to avoid false negatives.
374 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
377 spin_lock(&vmap_area_lock
);
379 * Invalidate cache if we have more permissive parameters.
380 * cached_hole_size notes the largest hole noticed _below_
381 * the vmap_area cached in free_vmap_cache: if size fits
382 * into that hole, we want to scan from vstart to reuse
383 * the hole instead of allocating above free_vmap_cache.
384 * Note that __free_vmap_area may update free_vmap_cache
385 * without updating cached_hole_size or cached_align.
387 if (!free_vmap_cache
||
388 size
< cached_hole_size
||
389 vstart
< cached_vstart
||
390 align
< cached_align
) {
392 cached_hole_size
= 0;
393 free_vmap_cache
= NULL
;
395 /* record if we encounter less permissive parameters */
396 cached_vstart
= vstart
;
397 cached_align
= align
;
399 /* find starting point for our search */
400 if (free_vmap_cache
) {
401 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
402 addr
= ALIGN(first
->va_end
, align
);
405 if (addr
+ size
< addr
)
409 addr
= ALIGN(vstart
, align
);
410 if (addr
+ size
< addr
)
413 n
= vmap_area_root
.rb_node
;
417 struct vmap_area
*tmp
;
418 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
419 if (tmp
->va_end
>= addr
) {
421 if (tmp
->va_start
<= addr
)
432 /* from the starting point, walk areas until a suitable hole is found */
433 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
434 if (addr
+ cached_hole_size
< first
->va_start
)
435 cached_hole_size
= first
->va_start
- addr
;
436 addr
= ALIGN(first
->va_end
, align
);
437 if (addr
+ size
< addr
)
440 if (list_is_last(&first
->list
, &vmap_area_list
))
443 first
= list_next_entry(first
, list
);
447 if (addr
+ size
> vend
)
451 va
->va_end
= addr
+ size
;
453 __insert_vmap_area(va
);
454 free_vmap_cache
= &va
->rb_node
;
455 spin_unlock(&vmap_area_lock
);
457 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
458 BUG_ON(va
->va_start
< vstart
);
459 BUG_ON(va
->va_end
> vend
);
464 spin_unlock(&vmap_area_lock
);
466 purge_vmap_area_lazy();
470 if (printk_ratelimit())
471 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
474 return ERR_PTR(-EBUSY
);
477 static void __free_vmap_area(struct vmap_area
*va
)
479 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
481 if (free_vmap_cache
) {
482 if (va
->va_end
< cached_vstart
) {
483 free_vmap_cache
= NULL
;
485 struct vmap_area
*cache
;
486 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
487 if (va
->va_start
<= cache
->va_start
) {
488 free_vmap_cache
= rb_prev(&va
->rb_node
);
490 * We don't try to update cached_hole_size or
491 * cached_align, but it won't go very wrong.
496 rb_erase(&va
->rb_node
, &vmap_area_root
);
497 RB_CLEAR_NODE(&va
->rb_node
);
498 list_del_rcu(&va
->list
);
501 * Track the highest possible candidate for pcpu area
502 * allocation. Areas outside of vmalloc area can be returned
503 * here too, consider only end addresses which fall inside
504 * vmalloc area proper.
506 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
507 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
509 kfree_rcu(va
, rcu_head
);
513 * Free a region of KVA allocated by alloc_vmap_area
515 static void free_vmap_area(struct vmap_area
*va
)
517 spin_lock(&vmap_area_lock
);
518 __free_vmap_area(va
);
519 spin_unlock(&vmap_area_lock
);
523 * Clear the pagetable entries of a given vmap_area
525 static void unmap_vmap_area(struct vmap_area
*va
)
527 vunmap_page_range(va
->va_start
, va
->va_end
);
530 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
533 * Unmap page tables and force a TLB flush immediately if pagealloc
534 * debugging is enabled. This catches use after free bugs similarly to
535 * those in linear kernel virtual address space after a page has been
538 * All the lazy freeing logic is still retained, in order to minimise
539 * intrusiveness of this debugging feature.
541 * This is going to be *slow* (linear kernel virtual address debugging
542 * doesn't do a broadcast TLB flush so it is a lot faster).
544 if (debug_pagealloc_enabled()) {
545 vunmap_page_range(start
, end
);
546 flush_tlb_kernel_range(start
, end
);
551 * lazy_max_pages is the maximum amount of virtual address space we gather up
552 * before attempting to purge with a TLB flush.
554 * There is a tradeoff here: a larger number will cover more kernel page tables
555 * and take slightly longer to purge, but it will linearly reduce the number of
556 * global TLB flushes that must be performed. It would seem natural to scale
557 * this number up linearly with the number of CPUs (because vmapping activity
558 * could also scale linearly with the number of CPUs), however it is likely
559 * that in practice, workloads might be constrained in other ways that mean
560 * vmap activity will not scale linearly with CPUs. Also, I want to be
561 * conservative and not introduce a big latency on huge systems, so go with
562 * a less aggressive log scale. It will still be an improvement over the old
563 * code, and it will be simple to change the scale factor if we find that it
564 * becomes a problem on bigger systems.
566 static unsigned long lazy_max_pages(void)
570 log
= fls(num_online_cpus());
572 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
575 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
577 /* for per-CPU blocks */
578 static void purge_fragmented_blocks_allcpus(void);
581 * called before a call to iounmap() if the caller wants vm_area_struct's
584 void set_iounmap_nonlazy(void)
586 atomic_set(&vmap_lazy_nr
, lazy_max_pages()+1);
590 * Purges all lazily-freed vmap areas.
592 * If sync is 0 then don't purge if there is already a purge in progress.
593 * If force_flush is 1, then flush kernel TLBs between *start and *end even
594 * if we found no lazy vmap areas to unmap (callers can use this to optimise
595 * their own TLB flushing).
596 * Returns with *start = min(*start, lowest purged address)
597 * *end = max(*end, highest purged address)
599 static void __purge_vmap_area_lazy(unsigned long *start
, unsigned long *end
,
600 int sync
, int force_flush
)
602 static DEFINE_SPINLOCK(purge_lock
);
603 struct llist_node
*valist
;
604 struct vmap_area
*va
;
605 struct vmap_area
*n_va
;
609 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
610 * should not expect such behaviour. This just simplifies locking for
611 * the case that isn't actually used at the moment anyway.
613 if (!sync
&& !force_flush
) {
614 if (!spin_trylock(&purge_lock
))
617 spin_lock(&purge_lock
);
620 purge_fragmented_blocks_allcpus();
622 valist
= llist_del_all(&vmap_purge_list
);
623 llist_for_each_entry(va
, valist
, purge_list
) {
624 if (va
->va_start
< *start
)
625 *start
= va
->va_start
;
626 if (va
->va_end
> *end
)
628 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
632 atomic_sub(nr
, &vmap_lazy_nr
);
634 if (nr
|| force_flush
)
635 flush_tlb_kernel_range(*start
, *end
);
638 spin_lock(&vmap_area_lock
);
639 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
)
640 __free_vmap_area(va
);
641 spin_unlock(&vmap_area_lock
);
643 spin_unlock(&purge_lock
);
647 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
648 * is already purging.
650 static void try_purge_vmap_area_lazy(void)
652 unsigned long start
= ULONG_MAX
, end
= 0;
654 __purge_vmap_area_lazy(&start
, &end
, 0, 0);
658 * Kick off a purge of the outstanding lazy areas.
660 static void purge_vmap_area_lazy(void)
662 unsigned long start
= ULONG_MAX
, end
= 0;
664 __purge_vmap_area_lazy(&start
, &end
, 1, 0);
668 * Free a vmap area, caller ensuring that the area has been unmapped
669 * and flush_cache_vunmap had been called for the correct range
672 static void free_vmap_area_noflush(struct vmap_area
*va
)
676 nr_lazy
= atomic_add_return((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
,
679 /* After this point, we may free va at any time */
680 llist_add(&va
->purge_list
, &vmap_purge_list
);
682 if (unlikely(nr_lazy
> lazy_max_pages()))
683 try_purge_vmap_area_lazy();
687 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
688 * called for the correct range previously.
690 static void free_unmap_vmap_area_noflush(struct vmap_area
*va
)
693 free_vmap_area_noflush(va
);
697 * Free and unmap a vmap area
699 static void free_unmap_vmap_area(struct vmap_area
*va
)
701 flush_cache_vunmap(va
->va_start
, va
->va_end
);
702 free_unmap_vmap_area_noflush(va
);
705 static struct vmap_area
*find_vmap_area(unsigned long addr
)
707 struct vmap_area
*va
;
709 spin_lock(&vmap_area_lock
);
710 va
= __find_vmap_area(addr
);
711 spin_unlock(&vmap_area_lock
);
716 static void free_unmap_vmap_area_addr(unsigned long addr
)
718 struct vmap_area
*va
;
720 va
= find_vmap_area(addr
);
722 free_unmap_vmap_area(va
);
726 /*** Per cpu kva allocator ***/
729 * vmap space is limited especially on 32 bit architectures. Ensure there is
730 * room for at least 16 percpu vmap blocks per CPU.
733 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
734 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
735 * instead (we just need a rough idea)
737 #if BITS_PER_LONG == 32
738 #define VMALLOC_SPACE (128UL*1024*1024)
740 #define VMALLOC_SPACE (128UL*1024*1024*1024)
743 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
744 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
745 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
746 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
747 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
748 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
749 #define VMAP_BBMAP_BITS \
750 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
751 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
752 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
754 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
756 static bool vmap_initialized __read_mostly
= false;
758 struct vmap_block_queue
{
760 struct list_head free
;
765 struct vmap_area
*va
;
766 unsigned long free
, dirty
;
767 unsigned long dirty_min
, dirty_max
; /*< dirty range */
768 struct list_head free_list
;
769 struct rcu_head rcu_head
;
770 struct list_head purge
;
773 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
774 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
777 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
778 * in the free path. Could get rid of this if we change the API to return a
779 * "cookie" from alloc, to be passed to free. But no big deal yet.
781 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
782 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
785 * We should probably have a fallback mechanism to allocate virtual memory
786 * out of partially filled vmap blocks. However vmap block sizing should be
787 * fairly reasonable according to the vmalloc size, so it shouldn't be a
791 static unsigned long addr_to_vb_idx(unsigned long addr
)
793 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
794 addr
/= VMAP_BLOCK_SIZE
;
798 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
802 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
803 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
808 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
809 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
810 * @order: how many 2^order pages should be occupied in newly allocated block
811 * @gfp_mask: flags for the page level allocator
813 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
815 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
817 struct vmap_block_queue
*vbq
;
818 struct vmap_block
*vb
;
819 struct vmap_area
*va
;
820 unsigned long vb_idx
;
824 node
= numa_node_id();
826 vb
= kmalloc_node(sizeof(struct vmap_block
),
827 gfp_mask
& GFP_RECLAIM_MASK
, node
);
829 return ERR_PTR(-ENOMEM
);
831 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
832 VMALLOC_START
, VMALLOC_END
,
839 err
= radix_tree_preload(gfp_mask
);
846 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
847 spin_lock_init(&vb
->lock
);
849 /* At least something should be left free */
850 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
851 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
853 vb
->dirty_min
= VMAP_BBMAP_BITS
;
855 INIT_LIST_HEAD(&vb
->free_list
);
857 vb_idx
= addr_to_vb_idx(va
->va_start
);
858 spin_lock(&vmap_block_tree_lock
);
859 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
860 spin_unlock(&vmap_block_tree_lock
);
862 radix_tree_preload_end();
864 vbq
= &get_cpu_var(vmap_block_queue
);
865 spin_lock(&vbq
->lock
);
866 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
867 spin_unlock(&vbq
->lock
);
868 put_cpu_var(vmap_block_queue
);
873 static void free_vmap_block(struct vmap_block
*vb
)
875 struct vmap_block
*tmp
;
876 unsigned long vb_idx
;
878 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
879 spin_lock(&vmap_block_tree_lock
);
880 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
881 spin_unlock(&vmap_block_tree_lock
);
884 free_vmap_area_noflush(vb
->va
);
885 kfree_rcu(vb
, rcu_head
);
888 static void purge_fragmented_blocks(int cpu
)
891 struct vmap_block
*vb
;
892 struct vmap_block
*n_vb
;
893 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
896 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
898 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
901 spin_lock(&vb
->lock
);
902 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
903 vb
->free
= 0; /* prevent further allocs after releasing lock */
904 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
906 vb
->dirty_max
= VMAP_BBMAP_BITS
;
907 spin_lock(&vbq
->lock
);
908 list_del_rcu(&vb
->free_list
);
909 spin_unlock(&vbq
->lock
);
910 spin_unlock(&vb
->lock
);
911 list_add_tail(&vb
->purge
, &purge
);
913 spin_unlock(&vb
->lock
);
917 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
918 list_del(&vb
->purge
);
923 static void purge_fragmented_blocks_allcpus(void)
927 for_each_possible_cpu(cpu
)
928 purge_fragmented_blocks(cpu
);
931 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
933 struct vmap_block_queue
*vbq
;
934 struct vmap_block
*vb
;
938 BUG_ON(offset_in_page(size
));
939 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
940 if (WARN_ON(size
== 0)) {
942 * Allocating 0 bytes isn't what caller wants since
943 * get_order(0) returns funny result. Just warn and terminate
948 order
= get_order(size
);
951 vbq
= &get_cpu_var(vmap_block_queue
);
952 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
953 unsigned long pages_off
;
955 spin_lock(&vb
->lock
);
956 if (vb
->free
< (1UL << order
)) {
957 spin_unlock(&vb
->lock
);
961 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
962 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
963 vb
->free
-= 1UL << order
;
965 spin_lock(&vbq
->lock
);
966 list_del_rcu(&vb
->free_list
);
967 spin_unlock(&vbq
->lock
);
970 spin_unlock(&vb
->lock
);
974 put_cpu_var(vmap_block_queue
);
977 /* Allocate new block if nothing was found */
979 vaddr
= new_vmap_block(order
, gfp_mask
);
984 static void vb_free(const void *addr
, unsigned long size
)
986 unsigned long offset
;
987 unsigned long vb_idx
;
989 struct vmap_block
*vb
;
991 BUG_ON(offset_in_page(size
));
992 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
994 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
996 order
= get_order(size
);
998 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
999 offset
>>= PAGE_SHIFT
;
1001 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
1003 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1007 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1009 spin_lock(&vb
->lock
);
1011 /* Expand dirty range */
1012 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1013 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1015 vb
->dirty
+= 1UL << order
;
1016 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1018 spin_unlock(&vb
->lock
);
1019 free_vmap_block(vb
);
1021 spin_unlock(&vb
->lock
);
1025 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1027 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1028 * to amortize TLB flushing overheads. What this means is that any page you
1029 * have now, may, in a former life, have been mapped into kernel virtual
1030 * address by the vmap layer and so there might be some CPUs with TLB entries
1031 * still referencing that page (additional to the regular 1:1 kernel mapping).
1033 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1034 * be sure that none of the pages we have control over will have any aliases
1035 * from the vmap layer.
1037 void vm_unmap_aliases(void)
1039 unsigned long start
= ULONG_MAX
, end
= 0;
1043 if (unlikely(!vmap_initialized
))
1046 for_each_possible_cpu(cpu
) {
1047 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1048 struct vmap_block
*vb
;
1051 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1052 spin_lock(&vb
->lock
);
1054 unsigned long va_start
= vb
->va
->va_start
;
1057 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1058 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1060 start
= min(s
, start
);
1065 spin_unlock(&vb
->lock
);
1070 __purge_vmap_area_lazy(&start
, &end
, 1, flush
);
1072 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1075 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1076 * @mem: the pointer returned by vm_map_ram
1077 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1079 void vm_unmap_ram(const void *mem
, unsigned int count
)
1081 unsigned long size
= count
<< PAGE_SHIFT
;
1082 unsigned long addr
= (unsigned long)mem
;
1085 BUG_ON(addr
< VMALLOC_START
);
1086 BUG_ON(addr
> VMALLOC_END
);
1087 BUG_ON(!PAGE_ALIGNED(addr
));
1089 debug_check_no_locks_freed(mem
, size
);
1090 vmap_debug_free_range(addr
, addr
+size
);
1092 if (likely(count
<= VMAP_MAX_ALLOC
))
1095 free_unmap_vmap_area_addr(addr
);
1097 EXPORT_SYMBOL(vm_unmap_ram
);
1100 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1101 * @pages: an array of pointers to the pages to be mapped
1102 * @count: number of pages
1103 * @node: prefer to allocate data structures on this node
1104 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1106 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1107 * faster than vmap so it's good. But if you mix long-life and short-life
1108 * objects with vm_map_ram(), it could consume lots of address space through
1109 * fragmentation (especially on a 32bit machine). You could see failures in
1110 * the end. Please use this function for short-lived objects.
1112 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1114 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1116 unsigned long size
= count
<< PAGE_SHIFT
;
1120 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1121 mem
= vb_alloc(size
, GFP_KERNEL
);
1124 addr
= (unsigned long)mem
;
1126 struct vmap_area
*va
;
1127 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1128 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1132 addr
= va
->va_start
;
1135 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1136 vm_unmap_ram(mem
, count
);
1141 EXPORT_SYMBOL(vm_map_ram
);
1143 static struct vm_struct
*vmlist __initdata
;
1145 * vm_area_add_early - add vmap area early during boot
1146 * @vm: vm_struct to add
1148 * This function is used to add fixed kernel vm area to vmlist before
1149 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1150 * should contain proper values and the other fields should be zero.
1152 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1154 void __init
vm_area_add_early(struct vm_struct
*vm
)
1156 struct vm_struct
*tmp
, **p
;
1158 BUG_ON(vmap_initialized
);
1159 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1160 if (tmp
->addr
>= vm
->addr
) {
1161 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1164 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1171 * vm_area_register_early - register vmap area early during boot
1172 * @vm: vm_struct to register
1173 * @align: requested alignment
1175 * This function is used to register kernel vm area before
1176 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1177 * proper values on entry and other fields should be zero. On return,
1178 * vm->addr contains the allocated address.
1180 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1182 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1184 static size_t vm_init_off __initdata
;
1187 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1188 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1190 vm
->addr
= (void *)addr
;
1192 vm_area_add_early(vm
);
1195 void __init
vmalloc_init(void)
1197 struct vmap_area
*va
;
1198 struct vm_struct
*tmp
;
1201 for_each_possible_cpu(i
) {
1202 struct vmap_block_queue
*vbq
;
1203 struct vfree_deferred
*p
;
1205 vbq
= &per_cpu(vmap_block_queue
, i
);
1206 spin_lock_init(&vbq
->lock
);
1207 INIT_LIST_HEAD(&vbq
->free
);
1208 p
= &per_cpu(vfree_deferred
, i
);
1209 init_llist_head(&p
->list
);
1210 INIT_WORK(&p
->wq
, free_work
);
1213 /* Import existing vmlist entries. */
1214 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1215 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1216 va
->flags
= VM_VM_AREA
;
1217 va
->va_start
= (unsigned long)tmp
->addr
;
1218 va
->va_end
= va
->va_start
+ tmp
->size
;
1220 __insert_vmap_area(va
);
1223 vmap_area_pcpu_hole
= VMALLOC_END
;
1225 vmap_initialized
= true;
1229 * map_kernel_range_noflush - map kernel VM area with the specified pages
1230 * @addr: start of the VM area to map
1231 * @size: size of the VM area to map
1232 * @prot: page protection flags to use
1233 * @pages: pages to map
1235 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1236 * specify should have been allocated using get_vm_area() and its
1240 * This function does NOT do any cache flushing. The caller is
1241 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1242 * before calling this function.
1245 * The number of pages mapped on success, -errno on failure.
1247 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1248 pgprot_t prot
, struct page
**pages
)
1250 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1254 * unmap_kernel_range_noflush - unmap kernel VM area
1255 * @addr: start of the VM area to unmap
1256 * @size: size of the VM area to unmap
1258 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1259 * specify should have been allocated using get_vm_area() and its
1263 * This function does NOT do any cache flushing. The caller is
1264 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1265 * before calling this function and flush_tlb_kernel_range() after.
1267 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1269 vunmap_page_range(addr
, addr
+ size
);
1271 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1274 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1275 * @addr: start of the VM area to unmap
1276 * @size: size of the VM area to unmap
1278 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1279 * the unmapping and tlb after.
1281 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1283 unsigned long end
= addr
+ size
;
1285 flush_cache_vunmap(addr
, end
);
1286 vunmap_page_range(addr
, end
);
1287 flush_tlb_kernel_range(addr
, end
);
1289 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
1291 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
1293 unsigned long addr
= (unsigned long)area
->addr
;
1294 unsigned long end
= addr
+ get_vm_area_size(area
);
1297 err
= vmap_page_range(addr
, end
, prot
, pages
);
1299 return err
> 0 ? 0 : err
;
1301 EXPORT_SYMBOL_GPL(map_vm_area
);
1303 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1304 unsigned long flags
, const void *caller
)
1306 spin_lock(&vmap_area_lock
);
1308 vm
->addr
= (void *)va
->va_start
;
1309 vm
->size
= va
->va_end
- va
->va_start
;
1310 vm
->caller
= caller
;
1312 va
->flags
|= VM_VM_AREA
;
1313 spin_unlock(&vmap_area_lock
);
1316 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1319 * Before removing VM_UNINITIALIZED,
1320 * we should make sure that vm has proper values.
1321 * Pair with smp_rmb() in show_numa_info().
1324 vm
->flags
&= ~VM_UNINITIALIZED
;
1327 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1328 unsigned long align
, unsigned long flags
, unsigned long start
,
1329 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1331 struct vmap_area
*va
;
1332 struct vm_struct
*area
;
1334 BUG_ON(in_interrupt());
1335 if (flags
& VM_IOREMAP
)
1336 align
= 1ul << clamp_t(int, fls_long(size
),
1337 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1339 size
= PAGE_ALIGN(size
);
1340 if (unlikely(!size
))
1343 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1344 if (unlikely(!area
))
1347 if (!(flags
& VM_NO_GUARD
))
1350 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1356 setup_vmalloc_vm(area
, va
, flags
, caller
);
1361 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1362 unsigned long start
, unsigned long end
)
1364 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1365 GFP_KERNEL
, __builtin_return_address(0));
1367 EXPORT_SYMBOL_GPL(__get_vm_area
);
1369 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1370 unsigned long start
, unsigned long end
,
1373 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1374 GFP_KERNEL
, caller
);
1378 * get_vm_area - reserve a contiguous kernel virtual area
1379 * @size: size of the area
1380 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1382 * Search an area of @size in the kernel virtual mapping area,
1383 * and reserved it for out purposes. Returns the area descriptor
1384 * on success or %NULL on failure.
1386 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1388 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1389 NUMA_NO_NODE
, GFP_KERNEL
,
1390 __builtin_return_address(0));
1393 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1396 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1397 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1401 * find_vm_area - find a continuous kernel virtual area
1402 * @addr: base address
1404 * Search for the kernel VM area starting at @addr, and return it.
1405 * It is up to the caller to do all required locking to keep the returned
1408 struct vm_struct
*find_vm_area(const void *addr
)
1410 struct vmap_area
*va
;
1412 va
= find_vmap_area((unsigned long)addr
);
1413 if (va
&& va
->flags
& VM_VM_AREA
)
1420 * remove_vm_area - find and remove a continuous kernel virtual area
1421 * @addr: base address
1423 * Search for the kernel VM area starting at @addr, and remove it.
1424 * This function returns the found VM area, but using it is NOT safe
1425 * on SMP machines, except for its size or flags.
1427 struct vm_struct
*remove_vm_area(const void *addr
)
1429 struct vmap_area
*va
;
1431 va
= find_vmap_area((unsigned long)addr
);
1432 if (va
&& va
->flags
& VM_VM_AREA
) {
1433 struct vm_struct
*vm
= va
->vm
;
1435 spin_lock(&vmap_area_lock
);
1437 va
->flags
&= ~VM_VM_AREA
;
1438 spin_unlock(&vmap_area_lock
);
1440 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1441 kasan_free_shadow(vm
);
1442 free_unmap_vmap_area(va
);
1449 static void __vunmap(const void *addr
, int deallocate_pages
)
1451 struct vm_struct
*area
;
1456 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
1460 area
= remove_vm_area(addr
);
1461 if (unlikely(!area
)) {
1462 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1467 debug_check_no_locks_freed(addr
, get_vm_area_size(area
));
1468 debug_check_no_obj_freed(addr
, get_vm_area_size(area
));
1470 if (deallocate_pages
) {
1473 for (i
= 0; i
< area
->nr_pages
; i
++) {
1474 struct page
*page
= area
->pages
[i
];
1477 __free_kmem_pages(page
, 0);
1480 kvfree(area
->pages
);
1488 * vfree - release memory allocated by vmalloc()
1489 * @addr: memory base address
1491 * Free the virtually continuous memory area starting at @addr, as
1492 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1493 * NULL, no operation is performed.
1495 * Must not be called in NMI context (strictly speaking, only if we don't
1496 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1497 * conventions for vfree() arch-depenedent would be a really bad idea)
1499 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1501 void vfree(const void *addr
)
1505 kmemleak_free(addr
);
1509 if (unlikely(in_interrupt())) {
1510 struct vfree_deferred
*p
= this_cpu_ptr(&vfree_deferred
);
1511 if (llist_add((struct llist_node
*)addr
, &p
->list
))
1512 schedule_work(&p
->wq
);
1516 EXPORT_SYMBOL(vfree
);
1519 * vunmap - release virtual mapping obtained by vmap()
1520 * @addr: memory base address
1522 * Free the virtually contiguous memory area starting at @addr,
1523 * which was created from the page array passed to vmap().
1525 * Must not be called in interrupt context.
1527 void vunmap(const void *addr
)
1529 BUG_ON(in_interrupt());
1534 EXPORT_SYMBOL(vunmap
);
1537 * vmap - map an array of pages into virtually contiguous space
1538 * @pages: array of page pointers
1539 * @count: number of pages to map
1540 * @flags: vm_area->flags
1541 * @prot: page protection for the mapping
1543 * Maps @count pages from @pages into contiguous kernel virtual
1546 void *vmap(struct page
**pages
, unsigned int count
,
1547 unsigned long flags
, pgprot_t prot
)
1549 struct vm_struct
*area
;
1553 if (count
> totalram_pages
)
1556 area
= get_vm_area_caller((count
<< PAGE_SHIFT
), flags
,
1557 __builtin_return_address(0));
1561 if (map_vm_area(area
, prot
, pages
)) {
1568 EXPORT_SYMBOL(vmap
);
1570 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1571 gfp_t gfp_mask
, pgprot_t prot
,
1572 int node
, const void *caller
);
1573 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1574 pgprot_t prot
, int node
)
1576 const int order
= 0;
1577 struct page
**pages
;
1578 unsigned int nr_pages
, array_size
, i
;
1579 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1580 const gfp_t alloc_mask
= gfp_mask
| __GFP_NOWARN
;
1582 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
1583 array_size
= (nr_pages
* sizeof(struct page
*));
1585 area
->nr_pages
= nr_pages
;
1586 /* Please note that the recursion is strictly bounded. */
1587 if (array_size
> PAGE_SIZE
) {
1588 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1589 PAGE_KERNEL
, node
, area
->caller
);
1591 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1593 area
->pages
= pages
;
1595 remove_vm_area(area
->addr
);
1600 for (i
= 0; i
< area
->nr_pages
; i
++) {
1603 if (node
== NUMA_NO_NODE
)
1604 page
= alloc_kmem_pages(alloc_mask
, order
);
1606 page
= alloc_kmem_pages_node(node
, alloc_mask
, order
);
1608 if (unlikely(!page
)) {
1609 /* Successfully allocated i pages, free them in __vunmap() */
1613 area
->pages
[i
] = page
;
1614 if (gfpflags_allow_blocking(gfp_mask
))
1618 if (map_vm_area(area
, prot
, pages
))
1623 warn_alloc_failed(gfp_mask
, order
,
1624 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1625 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1631 * __vmalloc_node_range - allocate virtually contiguous memory
1632 * @size: allocation size
1633 * @align: desired alignment
1634 * @start: vm area range start
1635 * @end: vm area range end
1636 * @gfp_mask: flags for the page level allocator
1637 * @prot: protection mask for the allocated pages
1638 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1639 * @node: node to use for allocation or NUMA_NO_NODE
1640 * @caller: caller's return address
1642 * Allocate enough pages to cover @size from the page level
1643 * allocator with @gfp_mask flags. Map them into contiguous
1644 * kernel virtual space, using a pagetable protection of @prot.
1646 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1647 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1648 pgprot_t prot
, unsigned long vm_flags
, int node
,
1651 struct vm_struct
*area
;
1653 unsigned long real_size
= size
;
1655 size
= PAGE_ALIGN(size
);
1656 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1659 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
1660 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
1664 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
1669 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1670 * flag. It means that vm_struct is not fully initialized.
1671 * Now, it is fully initialized, so remove this flag here.
1673 clear_vm_uninitialized_flag(area
);
1676 * A ref_count = 2 is needed because vm_struct allocated in
1677 * __get_vm_area_node() contains a reference to the virtual address of
1678 * the vmalloc'ed block.
1680 kmemleak_alloc(addr
, real_size
, 2, gfp_mask
);
1685 warn_alloc_failed(gfp_mask
, 0,
1686 "vmalloc: allocation failure: %lu bytes\n",
1692 * __vmalloc_node - allocate virtually contiguous memory
1693 * @size: allocation size
1694 * @align: desired alignment
1695 * @gfp_mask: flags for the page level allocator
1696 * @prot: protection mask for the allocated pages
1697 * @node: node to use for allocation or NUMA_NO_NODE
1698 * @caller: caller's return address
1700 * Allocate enough pages to cover @size from the page level
1701 * allocator with @gfp_mask flags. Map them into contiguous
1702 * kernel virtual space, using a pagetable protection of @prot.
1704 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1705 gfp_t gfp_mask
, pgprot_t prot
,
1706 int node
, const void *caller
)
1708 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1709 gfp_mask
, prot
, 0, node
, caller
);
1712 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1714 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1715 __builtin_return_address(0));
1717 EXPORT_SYMBOL(__vmalloc
);
1719 static inline void *__vmalloc_node_flags(unsigned long size
,
1720 int node
, gfp_t flags
)
1722 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
1723 node
, __builtin_return_address(0));
1727 * vmalloc - allocate virtually contiguous memory
1728 * @size: allocation size
1729 * Allocate enough pages to cover @size from the page level
1730 * allocator and map them into contiguous kernel virtual space.
1732 * For tight control over page level allocator and protection flags
1733 * use __vmalloc() instead.
1735 void *vmalloc(unsigned long size
)
1737 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1738 GFP_KERNEL
| __GFP_HIGHMEM
);
1740 EXPORT_SYMBOL(vmalloc
);
1743 * vzalloc - allocate virtually contiguous memory with zero fill
1744 * @size: allocation size
1745 * Allocate enough pages to cover @size from the page level
1746 * allocator and map them into contiguous kernel virtual space.
1747 * The memory allocated is set to zero.
1749 * For tight control over page level allocator and protection flags
1750 * use __vmalloc() instead.
1752 void *vzalloc(unsigned long size
)
1754 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1755 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1757 EXPORT_SYMBOL(vzalloc
);
1760 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1761 * @size: allocation size
1763 * The resulting memory area is zeroed so it can be mapped to userspace
1764 * without leaking data.
1766 void *vmalloc_user(unsigned long size
)
1768 struct vm_struct
*area
;
1771 ret
= __vmalloc_node(size
, SHMLBA
,
1772 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1773 PAGE_KERNEL
, NUMA_NO_NODE
,
1774 __builtin_return_address(0));
1776 area
= find_vm_area(ret
);
1777 area
->flags
|= VM_USERMAP
;
1781 EXPORT_SYMBOL(vmalloc_user
);
1784 * vmalloc_node - allocate memory on a specific node
1785 * @size: allocation size
1788 * Allocate enough pages to cover @size from the page level
1789 * allocator and map them into contiguous kernel virtual space.
1791 * For tight control over page level allocator and protection flags
1792 * use __vmalloc() instead.
1794 void *vmalloc_node(unsigned long size
, int node
)
1796 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1797 node
, __builtin_return_address(0));
1799 EXPORT_SYMBOL(vmalloc_node
);
1802 * vzalloc_node - allocate memory on a specific node with zero fill
1803 * @size: allocation size
1806 * Allocate enough pages to cover @size from the page level
1807 * allocator and map them into contiguous kernel virtual space.
1808 * The memory allocated is set to zero.
1810 * For tight control over page level allocator and protection flags
1811 * use __vmalloc_node() instead.
1813 void *vzalloc_node(unsigned long size
, int node
)
1815 return __vmalloc_node_flags(size
, node
,
1816 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1818 EXPORT_SYMBOL(vzalloc_node
);
1820 #ifndef PAGE_KERNEL_EXEC
1821 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1825 * vmalloc_exec - allocate virtually contiguous, executable memory
1826 * @size: allocation size
1828 * Kernel-internal function to allocate enough pages to cover @size
1829 * the page level allocator and map them into contiguous and
1830 * executable kernel virtual space.
1832 * For tight control over page level allocator and protection flags
1833 * use __vmalloc() instead.
1836 void *vmalloc_exec(unsigned long size
)
1838 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1839 NUMA_NO_NODE
, __builtin_return_address(0));
1842 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1843 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1844 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1845 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1847 #define GFP_VMALLOC32 GFP_KERNEL
1851 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1852 * @size: allocation size
1854 * Allocate enough 32bit PA addressable pages to cover @size from the
1855 * page level allocator and map them into contiguous kernel virtual space.
1857 void *vmalloc_32(unsigned long size
)
1859 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1860 NUMA_NO_NODE
, __builtin_return_address(0));
1862 EXPORT_SYMBOL(vmalloc_32
);
1865 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1866 * @size: allocation size
1868 * The resulting memory area is 32bit addressable and zeroed so it can be
1869 * mapped to userspace without leaking data.
1871 void *vmalloc_32_user(unsigned long size
)
1873 struct vm_struct
*area
;
1876 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1877 NUMA_NO_NODE
, __builtin_return_address(0));
1879 area
= find_vm_area(ret
);
1880 area
->flags
|= VM_USERMAP
;
1884 EXPORT_SYMBOL(vmalloc_32_user
);
1887 * small helper routine , copy contents to buf from addr.
1888 * If the page is not present, fill zero.
1891 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1897 unsigned long offset
, length
;
1899 offset
= offset_in_page(addr
);
1900 length
= PAGE_SIZE
- offset
;
1903 p
= vmalloc_to_page(addr
);
1905 * To do safe access to this _mapped_ area, we need
1906 * lock. But adding lock here means that we need to add
1907 * overhead of vmalloc()/vfree() calles for this _debug_
1908 * interface, rarely used. Instead of that, we'll use
1909 * kmap() and get small overhead in this access function.
1913 * we can expect USER0 is not used (see vread/vwrite's
1914 * function description)
1916 void *map
= kmap_atomic(p
);
1917 memcpy(buf
, map
+ offset
, length
);
1920 memset(buf
, 0, length
);
1930 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
1936 unsigned long offset
, length
;
1938 offset
= offset_in_page(addr
);
1939 length
= PAGE_SIZE
- offset
;
1942 p
= vmalloc_to_page(addr
);
1944 * To do safe access to this _mapped_ area, we need
1945 * lock. But adding lock here means that we need to add
1946 * overhead of vmalloc()/vfree() calles for this _debug_
1947 * interface, rarely used. Instead of that, we'll use
1948 * kmap() and get small overhead in this access function.
1952 * we can expect USER0 is not used (see vread/vwrite's
1953 * function description)
1955 void *map
= kmap_atomic(p
);
1956 memcpy(map
+ offset
, buf
, length
);
1968 * vread() - read vmalloc area in a safe way.
1969 * @buf: buffer for reading data
1970 * @addr: vm address.
1971 * @count: number of bytes to be read.
1973 * Returns # of bytes which addr and buf should be increased.
1974 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1975 * includes any intersect with alive vmalloc area.
1977 * This function checks that addr is a valid vmalloc'ed area, and
1978 * copy data from that area to a given buffer. If the given memory range
1979 * of [addr...addr+count) includes some valid address, data is copied to
1980 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1981 * IOREMAP area is treated as memory hole and no copy is done.
1983 * If [addr...addr+count) doesn't includes any intersects with alive
1984 * vm_struct area, returns 0. @buf should be kernel's buffer.
1986 * Note: In usual ops, vread() is never necessary because the caller
1987 * should know vmalloc() area is valid and can use memcpy().
1988 * This is for routines which have to access vmalloc area without
1989 * any informaion, as /dev/kmem.
1993 long vread(char *buf
, char *addr
, unsigned long count
)
1995 struct vmap_area
*va
;
1996 struct vm_struct
*vm
;
1997 char *vaddr
, *buf_start
= buf
;
1998 unsigned long buflen
= count
;
2001 /* Don't allow overflow */
2002 if ((unsigned long) addr
+ count
< count
)
2003 count
= -(unsigned long) addr
;
2005 spin_lock(&vmap_area_lock
);
2006 list_for_each_entry(va
, &vmap_area_list
, list
) {
2010 if (!(va
->flags
& VM_VM_AREA
))
2014 vaddr
= (char *) vm
->addr
;
2015 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2017 while (addr
< vaddr
) {
2025 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2028 if (!(vm
->flags
& VM_IOREMAP
))
2029 aligned_vread(buf
, addr
, n
);
2030 else /* IOREMAP area is treated as memory hole */
2037 spin_unlock(&vmap_area_lock
);
2039 if (buf
== buf_start
)
2041 /* zero-fill memory holes */
2042 if (buf
!= buf_start
+ buflen
)
2043 memset(buf
, 0, buflen
- (buf
- buf_start
));
2049 * vwrite() - write vmalloc area in a safe way.
2050 * @buf: buffer for source data
2051 * @addr: vm address.
2052 * @count: number of bytes to be read.
2054 * Returns # of bytes which addr and buf should be incresed.
2055 * (same number to @count).
2056 * If [addr...addr+count) doesn't includes any intersect with valid
2057 * vmalloc area, returns 0.
2059 * This function checks that addr is a valid vmalloc'ed area, and
2060 * copy data from a buffer to the given addr. If specified range of
2061 * [addr...addr+count) includes some valid address, data is copied from
2062 * proper area of @buf. If there are memory holes, no copy to hole.
2063 * IOREMAP area is treated as memory hole and no copy is done.
2065 * If [addr...addr+count) doesn't includes any intersects with alive
2066 * vm_struct area, returns 0. @buf should be kernel's buffer.
2068 * Note: In usual ops, vwrite() is never necessary because the caller
2069 * should know vmalloc() area is valid and can use memcpy().
2070 * This is for routines which have to access vmalloc area without
2071 * any informaion, as /dev/kmem.
2074 long vwrite(char *buf
, char *addr
, unsigned long count
)
2076 struct vmap_area
*va
;
2077 struct vm_struct
*vm
;
2079 unsigned long n
, buflen
;
2082 /* Don't allow overflow */
2083 if ((unsigned long) addr
+ count
< count
)
2084 count
= -(unsigned long) addr
;
2087 spin_lock(&vmap_area_lock
);
2088 list_for_each_entry(va
, &vmap_area_list
, list
) {
2092 if (!(va
->flags
& VM_VM_AREA
))
2096 vaddr
= (char *) vm
->addr
;
2097 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2099 while (addr
< vaddr
) {
2106 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2109 if (!(vm
->flags
& VM_IOREMAP
)) {
2110 aligned_vwrite(buf
, addr
, n
);
2118 spin_unlock(&vmap_area_lock
);
2125 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2126 * @vma: vma to cover
2127 * @uaddr: target user address to start at
2128 * @kaddr: virtual address of vmalloc kernel memory
2129 * @size: size of map area
2131 * Returns: 0 for success, -Exxx on failure
2133 * This function checks that @kaddr is a valid vmalloc'ed area,
2134 * and that it is big enough to cover the range starting at
2135 * @uaddr in @vma. Will return failure if that criteria isn't
2138 * Similar to remap_pfn_range() (see mm/memory.c)
2140 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2141 void *kaddr
, unsigned long size
)
2143 struct vm_struct
*area
;
2145 size
= PAGE_ALIGN(size
);
2147 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2150 area
= find_vm_area(kaddr
);
2154 if (!(area
->flags
& VM_USERMAP
))
2157 if (kaddr
+ size
> area
->addr
+ area
->size
)
2161 struct page
*page
= vmalloc_to_page(kaddr
);
2164 ret
= vm_insert_page(vma
, uaddr
, page
);
2173 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2177 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2180 * remap_vmalloc_range - map vmalloc pages to userspace
2181 * @vma: vma to cover (map full range of vma)
2182 * @addr: vmalloc memory
2183 * @pgoff: number of pages into addr before first page to map
2185 * Returns: 0 for success, -Exxx on failure
2187 * This function checks that addr is a valid vmalloc'ed area, and
2188 * that it is big enough to cover the vma. Will return failure if
2189 * that criteria isn't met.
2191 * Similar to remap_pfn_range() (see mm/memory.c)
2193 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2194 unsigned long pgoff
)
2196 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2197 addr
+ (pgoff
<< PAGE_SHIFT
),
2198 vma
->vm_end
- vma
->vm_start
);
2200 EXPORT_SYMBOL(remap_vmalloc_range
);
2203 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2206 void __weak
vmalloc_sync_all(void)
2211 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2223 * alloc_vm_area - allocate a range of kernel address space
2224 * @size: size of the area
2225 * @ptes: returns the PTEs for the address space
2227 * Returns: NULL on failure, vm_struct on success
2229 * This function reserves a range of kernel address space, and
2230 * allocates pagetables to map that range. No actual mappings
2233 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2234 * allocated for the VM area are returned.
2236 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2238 struct vm_struct
*area
;
2240 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2241 __builtin_return_address(0));
2246 * This ensures that page tables are constructed for this region
2247 * of kernel virtual address space and mapped into init_mm.
2249 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2250 size
, f
, ptes
? &ptes
: NULL
)) {
2257 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2259 void free_vm_area(struct vm_struct
*area
)
2261 struct vm_struct
*ret
;
2262 ret
= remove_vm_area(area
->addr
);
2263 BUG_ON(ret
!= area
);
2266 EXPORT_SYMBOL_GPL(free_vm_area
);
2269 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2271 return n
? rb_entry(n
, struct vmap_area
, rb_node
) : NULL
;
2275 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2276 * @end: target address
2277 * @pnext: out arg for the next vmap_area
2278 * @pprev: out arg for the previous vmap_area
2280 * Returns: %true if either or both of next and prev are found,
2281 * %false if no vmap_area exists
2283 * Find vmap_areas end addresses of which enclose @end. ie. if not
2284 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2286 static bool pvm_find_next_prev(unsigned long end
,
2287 struct vmap_area
**pnext
,
2288 struct vmap_area
**pprev
)
2290 struct rb_node
*n
= vmap_area_root
.rb_node
;
2291 struct vmap_area
*va
= NULL
;
2294 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2295 if (end
< va
->va_end
)
2297 else if (end
> va
->va_end
)
2306 if (va
->va_end
> end
) {
2308 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2311 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2317 * pvm_determine_end - find the highest aligned address between two vmap_areas
2318 * @pnext: in/out arg for the next vmap_area
2319 * @pprev: in/out arg for the previous vmap_area
2322 * Returns: determined end address
2324 * Find the highest aligned address between *@pnext and *@pprev below
2325 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2326 * down address is between the end addresses of the two vmap_areas.
2328 * Please note that the address returned by this function may fall
2329 * inside *@pnext vmap_area. The caller is responsible for checking
2332 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2333 struct vmap_area
**pprev
,
2334 unsigned long align
)
2336 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2340 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2344 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2346 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2353 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2354 * @offsets: array containing offset of each area
2355 * @sizes: array containing size of each area
2356 * @nr_vms: the number of areas to allocate
2357 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2359 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2360 * vm_structs on success, %NULL on failure
2362 * Percpu allocator wants to use congruent vm areas so that it can
2363 * maintain the offsets among percpu areas. This function allocates
2364 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2365 * be scattered pretty far, distance between two areas easily going up
2366 * to gigabytes. To avoid interacting with regular vmallocs, these
2367 * areas are allocated from top.
2369 * Despite its complicated look, this allocator is rather simple. It
2370 * does everything top-down and scans areas from the end looking for
2371 * matching slot. While scanning, if any of the areas overlaps with
2372 * existing vmap_area, the base address is pulled down to fit the
2373 * area. Scanning is repeated till all the areas fit and then all
2374 * necessary data structres are inserted and the result is returned.
2376 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2377 const size_t *sizes
, int nr_vms
,
2380 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2381 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2382 struct vmap_area
**vas
, *prev
, *next
;
2383 struct vm_struct
**vms
;
2384 int area
, area2
, last_area
, term_area
;
2385 unsigned long base
, start
, end
, last_end
;
2386 bool purged
= false;
2388 /* verify parameters and allocate data structures */
2389 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
2390 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2391 start
= offsets
[area
];
2392 end
= start
+ sizes
[area
];
2394 /* is everything aligned properly? */
2395 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2396 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2398 /* detect the area with the highest address */
2399 if (start
> offsets
[last_area
])
2402 for (area2
= 0; area2
< nr_vms
; area2
++) {
2403 unsigned long start2
= offsets
[area2
];
2404 unsigned long end2
= start2
+ sizes
[area2
];
2409 BUG_ON(start2
>= start
&& start2
< end
);
2410 BUG_ON(end2
<= end
&& end2
> start
);
2413 last_end
= offsets
[last_area
] + sizes
[last_area
];
2415 if (vmalloc_end
- vmalloc_start
< last_end
) {
2420 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2421 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2425 for (area
= 0; area
< nr_vms
; area
++) {
2426 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2427 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2428 if (!vas
[area
] || !vms
[area
])
2432 spin_lock(&vmap_area_lock
);
2434 /* start scanning - we scan from the top, begin with the last area */
2435 area
= term_area
= last_area
;
2436 start
= offsets
[area
];
2437 end
= start
+ sizes
[area
];
2439 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2440 base
= vmalloc_end
- last_end
;
2443 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2446 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2447 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2450 * base might have underflowed, add last_end before
2453 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2454 spin_unlock(&vmap_area_lock
);
2456 purge_vmap_area_lazy();
2464 * If next overlaps, move base downwards so that it's
2465 * right below next and then recheck.
2467 if (next
&& next
->va_start
< base
+ end
) {
2468 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2474 * If prev overlaps, shift down next and prev and move
2475 * base so that it's right below new next and then
2478 if (prev
&& prev
->va_end
> base
+ start
) {
2480 prev
= node_to_va(rb_prev(&next
->rb_node
));
2481 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2487 * This area fits, move on to the previous one. If
2488 * the previous one is the terminal one, we're done.
2490 area
= (area
+ nr_vms
- 1) % nr_vms
;
2491 if (area
== term_area
)
2493 start
= offsets
[area
];
2494 end
= start
+ sizes
[area
];
2495 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2498 /* we've found a fitting base, insert all va's */
2499 for (area
= 0; area
< nr_vms
; area
++) {
2500 struct vmap_area
*va
= vas
[area
];
2502 va
->va_start
= base
+ offsets
[area
];
2503 va
->va_end
= va
->va_start
+ sizes
[area
];
2504 __insert_vmap_area(va
);
2507 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2509 spin_unlock(&vmap_area_lock
);
2511 /* insert all vm's */
2512 for (area
= 0; area
< nr_vms
; area
++)
2513 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2520 for (area
= 0; area
< nr_vms
; area
++) {
2531 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2532 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2533 * @nr_vms: the number of allocated areas
2535 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2537 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2541 for (i
= 0; i
< nr_vms
; i
++)
2542 free_vm_area(vms
[i
]);
2545 #endif /* CONFIG_SMP */
2547 #ifdef CONFIG_PROC_FS
2548 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2549 __acquires(&vmap_area_lock
)
2552 struct vmap_area
*va
;
2554 spin_lock(&vmap_area_lock
);
2555 va
= list_first_entry(&vmap_area_list
, typeof(*va
), list
);
2556 while (n
> 0 && &va
->list
!= &vmap_area_list
) {
2558 va
= list_next_entry(va
, list
);
2560 if (!n
&& &va
->list
!= &vmap_area_list
)
2567 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2569 struct vmap_area
*va
= p
, *next
;
2572 next
= list_next_entry(va
, list
);
2573 if (&next
->list
!= &vmap_area_list
)
2579 static void s_stop(struct seq_file
*m
, void *p
)
2580 __releases(&vmap_area_lock
)
2582 spin_unlock(&vmap_area_lock
);
2585 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2587 if (IS_ENABLED(CONFIG_NUMA
)) {
2588 unsigned int nr
, *counters
= m
->private;
2593 if (v
->flags
& VM_UNINITIALIZED
)
2595 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2598 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2600 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2601 counters
[page_to_nid(v
->pages
[nr
])]++;
2603 for_each_node_state(nr
, N_HIGH_MEMORY
)
2605 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2609 static int s_show(struct seq_file
*m
, void *p
)
2611 struct vmap_area
*va
= p
;
2612 struct vm_struct
*v
;
2615 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2616 * behalf of vmap area is being tear down or vm_map_ram allocation.
2618 if (!(va
->flags
& VM_VM_AREA
))
2623 seq_printf(m
, "0x%pK-0x%pK %7ld",
2624 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2627 seq_printf(m
, " %pS", v
->caller
);
2630 seq_printf(m
, " pages=%d", v
->nr_pages
);
2633 seq_printf(m
, " phys=%llx", (unsigned long long)v
->phys_addr
);
2635 if (v
->flags
& VM_IOREMAP
)
2636 seq_puts(m
, " ioremap");
2638 if (v
->flags
& VM_ALLOC
)
2639 seq_puts(m
, " vmalloc");
2641 if (v
->flags
& VM_MAP
)
2642 seq_puts(m
, " vmap");
2644 if (v
->flags
& VM_USERMAP
)
2645 seq_puts(m
, " user");
2647 if (is_vmalloc_addr(v
->pages
))
2648 seq_puts(m
, " vpages");
2650 show_numa_info(m
, v
);
2655 static const struct seq_operations vmalloc_op
= {
2662 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2664 if (IS_ENABLED(CONFIG_NUMA
))
2665 return seq_open_private(file
, &vmalloc_op
,
2666 nr_node_ids
* sizeof(unsigned int));
2668 return seq_open(file
, &vmalloc_op
);
2671 static const struct file_operations proc_vmalloc_operations
= {
2672 .open
= vmalloc_open
,
2674 .llseek
= seq_lseek
,
2675 .release
= seq_release_private
,
2678 static int __init
proc_vmalloc_init(void)
2680 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2683 module_init(proc_vmalloc_init
);