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>
33 #include <asm/uaccess.h>
34 #include <asm/tlbflush.h>
35 #include <asm/shmparam.h>
37 struct vfree_deferred
{
38 struct llist_head list
;
39 struct work_struct wq
;
41 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
43 static void __vunmap(const void *, int);
45 static void free_work(struct work_struct
*w
)
47 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
48 struct llist_node
*llnode
= llist_del_all(&p
->list
);
51 llnode
= llist_next(llnode
);
56 /*** Page table manipulation functions ***/
58 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
62 pte
= pte_offset_kernel(pmd
, addr
);
64 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
65 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
66 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
69 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
74 pmd
= pmd_offset(pud
, addr
);
76 next
= pmd_addr_end(addr
, end
);
77 if (pmd_none_or_clear_bad(pmd
))
79 vunmap_pte_range(pmd
, addr
, next
);
80 } while (pmd
++, addr
= next
, addr
!= end
);
83 static void vunmap_pud_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
88 pud
= pud_offset(pgd
, addr
);
90 next
= pud_addr_end(addr
, end
);
91 if (pud_none_or_clear_bad(pud
))
93 vunmap_pmd_range(pud
, addr
, next
);
94 } while (pud
++, addr
= next
, addr
!= end
);
97 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
103 pgd
= pgd_offset_k(addr
);
105 next
= pgd_addr_end(addr
, end
);
106 if (pgd_none_or_clear_bad(pgd
))
108 vunmap_pud_range(pgd
, addr
, next
);
109 } while (pgd
++, addr
= next
, addr
!= end
);
112 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
113 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
118 * nr is a running index into the array which helps higher level
119 * callers keep track of where we're up to.
122 pte
= pte_alloc_kernel(pmd
, addr
);
126 struct page
*page
= pages
[*nr
];
128 if (WARN_ON(!pte_none(*pte
)))
132 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
134 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
138 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
139 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
144 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
148 next
= pmd_addr_end(addr
, end
);
149 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
151 } while (pmd
++, addr
= next
, addr
!= end
);
155 static int vmap_pud_range(pgd_t
*pgd
, unsigned long addr
,
156 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
161 pud
= pud_alloc(&init_mm
, pgd
, addr
);
165 next
= pud_addr_end(addr
, end
);
166 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
168 } while (pud
++, addr
= next
, addr
!= end
);
173 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
174 * will have pfns corresponding to the "pages" array.
176 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
178 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
179 pgprot_t prot
, struct page
**pages
)
183 unsigned long addr
= start
;
188 pgd
= pgd_offset_k(addr
);
190 next
= pgd_addr_end(addr
, end
);
191 err
= vmap_pud_range(pgd
, addr
, next
, prot
, pages
, &nr
);
194 } while (pgd
++, addr
= next
, addr
!= end
);
199 static int vmap_page_range(unsigned long start
, unsigned long end
,
200 pgprot_t prot
, struct page
**pages
)
204 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
205 flush_cache_vmap(start
, end
);
209 int is_vmalloc_or_module_addr(const void *x
)
212 * ARM, x86-64 and sparc64 put modules in a special place,
213 * and fall back on vmalloc() if that fails. Others
214 * just put it in the vmalloc space.
216 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
217 unsigned long addr
= (unsigned long)x
;
218 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
221 return is_vmalloc_addr(x
);
225 * Walk a vmap address to the struct page it maps.
227 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
229 unsigned long addr
= (unsigned long) vmalloc_addr
;
230 struct page
*page
= NULL
;
231 pgd_t
*pgd
= pgd_offset_k(addr
);
234 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
235 * architectures that do not vmalloc module space
237 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
239 if (!pgd_none(*pgd
)) {
240 pud_t
*pud
= pud_offset(pgd
, addr
);
241 if (!pud_none(*pud
)) {
242 pmd_t
*pmd
= pmd_offset(pud
, addr
);
243 if (!pmd_none(*pmd
)) {
246 ptep
= pte_offset_map(pmd
, addr
);
248 if (pte_present(pte
))
249 page
= pte_page(pte
);
256 EXPORT_SYMBOL(vmalloc_to_page
);
259 * Map a vmalloc()-space virtual address to the physical page frame number.
261 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
263 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
265 EXPORT_SYMBOL(vmalloc_to_pfn
);
268 /*** Global kva allocator ***/
270 #define VM_LAZY_FREE 0x01
271 #define VM_LAZY_FREEING 0x02
272 #define VM_VM_AREA 0x04
274 static DEFINE_SPINLOCK(vmap_area_lock
);
275 /* Export for kexec only */
276 LIST_HEAD(vmap_area_list
);
277 static struct rb_root vmap_area_root
= RB_ROOT
;
279 /* The vmap cache globals are protected by vmap_area_lock */
280 static struct rb_node
*free_vmap_cache
;
281 static unsigned long cached_hole_size
;
282 static unsigned long cached_vstart
;
283 static unsigned long cached_align
;
285 static unsigned long vmap_area_pcpu_hole
;
287 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
289 struct rb_node
*n
= vmap_area_root
.rb_node
;
292 struct vmap_area
*va
;
294 va
= rb_entry(n
, struct vmap_area
, rb_node
);
295 if (addr
< va
->va_start
)
297 else if (addr
>= va
->va_end
)
306 static void __insert_vmap_area(struct vmap_area
*va
)
308 struct rb_node
**p
= &vmap_area_root
.rb_node
;
309 struct rb_node
*parent
= NULL
;
313 struct vmap_area
*tmp_va
;
316 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
317 if (va
->va_start
< tmp_va
->va_end
)
319 else if (va
->va_end
> tmp_va
->va_start
)
325 rb_link_node(&va
->rb_node
, parent
, p
);
326 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
328 /* address-sort this list */
329 tmp
= rb_prev(&va
->rb_node
);
331 struct vmap_area
*prev
;
332 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
333 list_add_rcu(&va
->list
, &prev
->list
);
335 list_add_rcu(&va
->list
, &vmap_area_list
);
338 static void purge_vmap_area_lazy(void);
341 * Allocate a region of KVA of the specified size and alignment, within the
344 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
346 unsigned long vstart
, unsigned long vend
,
347 int node
, gfp_t gfp_mask
)
349 struct vmap_area
*va
;
353 struct vmap_area
*first
;
356 BUG_ON(size
& ~PAGE_MASK
);
357 BUG_ON(!is_power_of_2(align
));
359 va
= kmalloc_node(sizeof(struct vmap_area
),
360 gfp_mask
& GFP_RECLAIM_MASK
, node
);
362 return ERR_PTR(-ENOMEM
);
365 * Only scan the relevant parts containing pointers to other objects
366 * to avoid false negatives.
368 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
371 spin_lock(&vmap_area_lock
);
373 * Invalidate cache if we have more permissive parameters.
374 * cached_hole_size notes the largest hole noticed _below_
375 * the vmap_area cached in free_vmap_cache: if size fits
376 * into that hole, we want to scan from vstart to reuse
377 * the hole instead of allocating above free_vmap_cache.
378 * Note that __free_vmap_area may update free_vmap_cache
379 * without updating cached_hole_size or cached_align.
381 if (!free_vmap_cache
||
382 size
< cached_hole_size
||
383 vstart
< cached_vstart
||
384 align
< cached_align
) {
386 cached_hole_size
= 0;
387 free_vmap_cache
= NULL
;
389 /* record if we encounter less permissive parameters */
390 cached_vstart
= vstart
;
391 cached_align
= align
;
393 /* find starting point for our search */
394 if (free_vmap_cache
) {
395 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
396 addr
= ALIGN(first
->va_end
, align
);
399 if (addr
+ size
< addr
)
403 addr
= ALIGN(vstart
, align
);
404 if (addr
+ size
< addr
)
407 n
= vmap_area_root
.rb_node
;
411 struct vmap_area
*tmp
;
412 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
413 if (tmp
->va_end
>= addr
) {
415 if (tmp
->va_start
<= addr
)
426 /* from the starting point, walk areas until a suitable hole is found */
427 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
428 if (addr
+ cached_hole_size
< first
->va_start
)
429 cached_hole_size
= first
->va_start
- addr
;
430 addr
= ALIGN(first
->va_end
, align
);
431 if (addr
+ size
< addr
)
434 if (list_is_last(&first
->list
, &vmap_area_list
))
437 first
= list_entry(first
->list
.next
,
438 struct vmap_area
, list
);
442 if (addr
+ size
> vend
)
446 va
->va_end
= addr
+ size
;
448 __insert_vmap_area(va
);
449 free_vmap_cache
= &va
->rb_node
;
450 spin_unlock(&vmap_area_lock
);
452 BUG_ON(va
->va_start
& (align
-1));
453 BUG_ON(va
->va_start
< vstart
);
454 BUG_ON(va
->va_end
> vend
);
459 spin_unlock(&vmap_area_lock
);
461 purge_vmap_area_lazy();
465 if (printk_ratelimit())
467 "vmap allocation for size %lu failed: "
468 "use vmalloc=<size> to increase size.\n", size
);
470 return ERR_PTR(-EBUSY
);
473 static void __free_vmap_area(struct vmap_area
*va
)
475 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
477 if (free_vmap_cache
) {
478 if (va
->va_end
< cached_vstart
) {
479 free_vmap_cache
= NULL
;
481 struct vmap_area
*cache
;
482 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
483 if (va
->va_start
<= cache
->va_start
) {
484 free_vmap_cache
= rb_prev(&va
->rb_node
);
486 * We don't try to update cached_hole_size or
487 * cached_align, but it won't go very wrong.
492 rb_erase(&va
->rb_node
, &vmap_area_root
);
493 RB_CLEAR_NODE(&va
->rb_node
);
494 list_del_rcu(&va
->list
);
497 * Track the highest possible candidate for pcpu area
498 * allocation. Areas outside of vmalloc area can be returned
499 * here too, consider only end addresses which fall inside
500 * vmalloc area proper.
502 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
503 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
505 kfree_rcu(va
, rcu_head
);
509 * Free a region of KVA allocated by alloc_vmap_area
511 static void free_vmap_area(struct vmap_area
*va
)
513 spin_lock(&vmap_area_lock
);
514 __free_vmap_area(va
);
515 spin_unlock(&vmap_area_lock
);
519 * Clear the pagetable entries of a given vmap_area
521 static void unmap_vmap_area(struct vmap_area
*va
)
523 vunmap_page_range(va
->va_start
, va
->va_end
);
526 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
529 * Unmap page tables and force a TLB flush immediately if
530 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
531 * bugs similarly to those in linear kernel virtual address
532 * space after a page has been freed.
534 * All the lazy freeing logic is still retained, in order to
535 * minimise intrusiveness of this debugging feature.
537 * This is going to be *slow* (linear kernel virtual address
538 * debugging doesn't do a broadcast TLB flush so it is a lot
541 #ifdef CONFIG_DEBUG_PAGEALLOC
542 vunmap_page_range(start
, end
);
543 flush_tlb_kernel_range(start
, end
);
548 * lazy_max_pages is the maximum amount of virtual address space we gather up
549 * before attempting to purge with a TLB flush.
551 * There is a tradeoff here: a larger number will cover more kernel page tables
552 * and take slightly longer to purge, but it will linearly reduce the number of
553 * global TLB flushes that must be performed. It would seem natural to scale
554 * this number up linearly with the number of CPUs (because vmapping activity
555 * could also scale linearly with the number of CPUs), however it is likely
556 * that in practice, workloads might be constrained in other ways that mean
557 * vmap activity will not scale linearly with CPUs. Also, I want to be
558 * conservative and not introduce a big latency on huge systems, so go with
559 * a less aggressive log scale. It will still be an improvement over the old
560 * code, and it will be simple to change the scale factor if we find that it
561 * becomes a problem on bigger systems.
563 static unsigned long lazy_max_pages(void)
567 log
= fls(num_online_cpus());
569 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
572 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
574 /* for per-CPU blocks */
575 static void purge_fragmented_blocks_allcpus(void);
578 * called before a call to iounmap() if the caller wants vm_area_struct's
581 void set_iounmap_nonlazy(void)
583 atomic_set(&vmap_lazy_nr
, lazy_max_pages()+1);
587 * Purges all lazily-freed vmap areas.
589 * If sync is 0 then don't purge if there is already a purge in progress.
590 * If force_flush is 1, then flush kernel TLBs between *start and *end even
591 * if we found no lazy vmap areas to unmap (callers can use this to optimise
592 * their own TLB flushing).
593 * Returns with *start = min(*start, lowest purged address)
594 * *end = max(*end, highest purged address)
596 static void __purge_vmap_area_lazy(unsigned long *start
, unsigned long *end
,
597 int sync
, int force_flush
)
599 static DEFINE_SPINLOCK(purge_lock
);
601 struct vmap_area
*va
;
602 struct vmap_area
*n_va
;
606 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
607 * should not expect such behaviour. This just simplifies locking for
608 * the case that isn't actually used at the moment anyway.
610 if (!sync
&& !force_flush
) {
611 if (!spin_trylock(&purge_lock
))
614 spin_lock(&purge_lock
);
617 purge_fragmented_blocks_allcpus();
620 list_for_each_entry_rcu(va
, &vmap_area_list
, list
) {
621 if (va
->flags
& VM_LAZY_FREE
) {
622 if (va
->va_start
< *start
)
623 *start
= va
->va_start
;
624 if (va
->va_end
> *end
)
626 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
627 list_add_tail(&va
->purge_list
, &valist
);
628 va
->flags
|= VM_LAZY_FREEING
;
629 va
->flags
&= ~VM_LAZY_FREE
;
635 atomic_sub(nr
, &vmap_lazy_nr
);
637 if (nr
|| force_flush
)
638 flush_tlb_kernel_range(*start
, *end
);
641 spin_lock(&vmap_area_lock
);
642 list_for_each_entry_safe(va
, n_va
, &valist
, purge_list
)
643 __free_vmap_area(va
);
644 spin_unlock(&vmap_area_lock
);
646 spin_unlock(&purge_lock
);
650 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
651 * is already purging.
653 static void try_purge_vmap_area_lazy(void)
655 unsigned long start
= ULONG_MAX
, end
= 0;
657 __purge_vmap_area_lazy(&start
, &end
, 0, 0);
661 * Kick off a purge of the outstanding lazy areas.
663 static void purge_vmap_area_lazy(void)
665 unsigned long start
= ULONG_MAX
, end
= 0;
667 __purge_vmap_area_lazy(&start
, &end
, 1, 0);
671 * Free a vmap area, caller ensuring that the area has been unmapped
672 * and flush_cache_vunmap had been called for the correct range
675 static void free_vmap_area_noflush(struct vmap_area
*va
)
677 va
->flags
|= VM_LAZY_FREE
;
678 atomic_add((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
, &vmap_lazy_nr
);
679 if (unlikely(atomic_read(&vmap_lazy_nr
) > lazy_max_pages()))
680 try_purge_vmap_area_lazy();
684 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
685 * called for the correct range previously.
687 static void free_unmap_vmap_area_noflush(struct vmap_area
*va
)
690 free_vmap_area_noflush(va
);
694 * Free and unmap a vmap area
696 static void free_unmap_vmap_area(struct vmap_area
*va
)
698 flush_cache_vunmap(va
->va_start
, va
->va_end
);
699 free_unmap_vmap_area_noflush(va
);
702 static struct vmap_area
*find_vmap_area(unsigned long addr
)
704 struct vmap_area
*va
;
706 spin_lock(&vmap_area_lock
);
707 va
= __find_vmap_area(addr
);
708 spin_unlock(&vmap_area_lock
);
713 static void free_unmap_vmap_area_addr(unsigned long addr
)
715 struct vmap_area
*va
;
717 va
= find_vmap_area(addr
);
719 free_unmap_vmap_area(va
);
723 /*** Per cpu kva allocator ***/
726 * vmap space is limited especially on 32 bit architectures. Ensure there is
727 * room for at least 16 percpu vmap blocks per CPU.
730 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
731 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
732 * instead (we just need a rough idea)
734 #if BITS_PER_LONG == 32
735 #define VMALLOC_SPACE (128UL*1024*1024)
737 #define VMALLOC_SPACE (128UL*1024*1024*1024)
740 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
741 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
742 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
743 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
744 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
745 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
746 #define VMAP_BBMAP_BITS \
747 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
748 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
749 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
751 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
753 static bool vmap_initialized __read_mostly
= false;
755 struct vmap_block_queue
{
757 struct list_head free
;
762 struct vmap_area
*va
;
763 unsigned long free
, dirty
;
764 DECLARE_BITMAP(dirty_map
, VMAP_BBMAP_BITS
);
765 struct list_head free_list
;
766 struct rcu_head rcu_head
;
767 struct list_head purge
;
770 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
771 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
774 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
775 * in the free path. Could get rid of this if we change the API to return a
776 * "cookie" from alloc, to be passed to free. But no big deal yet.
778 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
779 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
782 * We should probably have a fallback mechanism to allocate virtual memory
783 * out of partially filled vmap blocks. However vmap block sizing should be
784 * fairly reasonable according to the vmalloc size, so it shouldn't be a
788 static unsigned long addr_to_vb_idx(unsigned long addr
)
790 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
791 addr
/= VMAP_BLOCK_SIZE
;
795 static struct vmap_block
*new_vmap_block(gfp_t gfp_mask
)
797 struct vmap_block_queue
*vbq
;
798 struct vmap_block
*vb
;
799 struct vmap_area
*va
;
800 unsigned long vb_idx
;
803 node
= numa_node_id();
805 vb
= kmalloc_node(sizeof(struct vmap_block
),
806 gfp_mask
& GFP_RECLAIM_MASK
, node
);
808 return ERR_PTR(-ENOMEM
);
810 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
811 VMALLOC_START
, VMALLOC_END
,
818 err
= radix_tree_preload(gfp_mask
);
825 spin_lock_init(&vb
->lock
);
827 vb
->free
= VMAP_BBMAP_BITS
;
829 bitmap_zero(vb
->dirty_map
, VMAP_BBMAP_BITS
);
830 INIT_LIST_HEAD(&vb
->free_list
);
832 vb_idx
= addr_to_vb_idx(va
->va_start
);
833 spin_lock(&vmap_block_tree_lock
);
834 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
835 spin_unlock(&vmap_block_tree_lock
);
837 radix_tree_preload_end();
839 vbq
= &get_cpu_var(vmap_block_queue
);
840 spin_lock(&vbq
->lock
);
841 list_add_rcu(&vb
->free_list
, &vbq
->free
);
842 spin_unlock(&vbq
->lock
);
843 put_cpu_var(vmap_block_queue
);
848 static void free_vmap_block(struct vmap_block
*vb
)
850 struct vmap_block
*tmp
;
851 unsigned long vb_idx
;
853 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
854 spin_lock(&vmap_block_tree_lock
);
855 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
856 spin_unlock(&vmap_block_tree_lock
);
859 free_vmap_area_noflush(vb
->va
);
860 kfree_rcu(vb
, rcu_head
);
863 static void purge_fragmented_blocks(int cpu
)
866 struct vmap_block
*vb
;
867 struct vmap_block
*n_vb
;
868 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
871 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
873 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
876 spin_lock(&vb
->lock
);
877 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
878 vb
->free
= 0; /* prevent further allocs after releasing lock */
879 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
880 bitmap_fill(vb
->dirty_map
, VMAP_BBMAP_BITS
);
881 spin_lock(&vbq
->lock
);
882 list_del_rcu(&vb
->free_list
);
883 spin_unlock(&vbq
->lock
);
884 spin_unlock(&vb
->lock
);
885 list_add_tail(&vb
->purge
, &purge
);
887 spin_unlock(&vb
->lock
);
891 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
892 list_del(&vb
->purge
);
897 static void purge_fragmented_blocks_allcpus(void)
901 for_each_possible_cpu(cpu
)
902 purge_fragmented_blocks(cpu
);
905 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
907 struct vmap_block_queue
*vbq
;
908 struct vmap_block
*vb
;
909 unsigned long addr
= 0;
912 BUG_ON(size
& ~PAGE_MASK
);
913 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
914 if (WARN_ON(size
== 0)) {
916 * Allocating 0 bytes isn't what caller wants since
917 * get_order(0) returns funny result. Just warn and terminate
922 order
= get_order(size
);
926 vbq
= &get_cpu_var(vmap_block_queue
);
927 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
930 spin_lock(&vb
->lock
);
931 if (vb
->free
< 1UL << order
)
934 i
= VMAP_BBMAP_BITS
- vb
->free
;
935 addr
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
936 BUG_ON(addr_to_vb_idx(addr
) !=
937 addr_to_vb_idx(vb
->va
->va_start
));
938 vb
->free
-= 1UL << order
;
940 spin_lock(&vbq
->lock
);
941 list_del_rcu(&vb
->free_list
);
942 spin_unlock(&vbq
->lock
);
944 spin_unlock(&vb
->lock
);
947 spin_unlock(&vb
->lock
);
950 put_cpu_var(vmap_block_queue
);
954 vb
= new_vmap_block(gfp_mask
);
963 static void vb_free(const void *addr
, unsigned long size
)
965 unsigned long offset
;
966 unsigned long vb_idx
;
968 struct vmap_block
*vb
;
970 BUG_ON(size
& ~PAGE_MASK
);
971 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
973 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
975 order
= get_order(size
);
977 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
979 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
981 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
985 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
987 spin_lock(&vb
->lock
);
988 BUG_ON(bitmap_allocate_region(vb
->dirty_map
, offset
>> PAGE_SHIFT
, order
));
990 vb
->dirty
+= 1UL << order
;
991 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
993 spin_unlock(&vb
->lock
);
996 spin_unlock(&vb
->lock
);
1000 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1002 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1003 * to amortize TLB flushing overheads. What this means is that any page you
1004 * have now, may, in a former life, have been mapped into kernel virtual
1005 * address by the vmap layer and so there might be some CPUs with TLB entries
1006 * still referencing that page (additional to the regular 1:1 kernel mapping).
1008 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1009 * be sure that none of the pages we have control over will have any aliases
1010 * from the vmap layer.
1012 void vm_unmap_aliases(void)
1014 unsigned long start
= ULONG_MAX
, end
= 0;
1018 if (unlikely(!vmap_initialized
))
1021 for_each_possible_cpu(cpu
) {
1022 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1023 struct vmap_block
*vb
;
1026 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1029 spin_lock(&vb
->lock
);
1030 i
= find_first_bit(vb
->dirty_map
, VMAP_BBMAP_BITS
);
1031 if (i
< VMAP_BBMAP_BITS
) {
1034 j
= find_last_bit(vb
->dirty_map
,
1036 j
= j
+ 1; /* need exclusive index */
1038 s
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
1039 e
= vb
->va
->va_start
+ (j
<< PAGE_SHIFT
);
1047 spin_unlock(&vb
->lock
);
1052 __purge_vmap_area_lazy(&start
, &end
, 1, flush
);
1054 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1057 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1058 * @mem: the pointer returned by vm_map_ram
1059 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1061 void vm_unmap_ram(const void *mem
, unsigned int count
)
1063 unsigned long size
= count
<< PAGE_SHIFT
;
1064 unsigned long addr
= (unsigned long)mem
;
1067 BUG_ON(addr
< VMALLOC_START
);
1068 BUG_ON(addr
> VMALLOC_END
);
1069 BUG_ON(addr
& (PAGE_SIZE
-1));
1071 debug_check_no_locks_freed(mem
, size
);
1072 vmap_debug_free_range(addr
, addr
+size
);
1074 if (likely(count
<= VMAP_MAX_ALLOC
))
1077 free_unmap_vmap_area_addr(addr
);
1079 EXPORT_SYMBOL(vm_unmap_ram
);
1082 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1083 * @pages: an array of pointers to the pages to be mapped
1084 * @count: number of pages
1085 * @node: prefer to allocate data structures on this node
1086 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1088 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1090 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1092 unsigned long size
= count
<< PAGE_SHIFT
;
1096 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1097 mem
= vb_alloc(size
, GFP_KERNEL
);
1100 addr
= (unsigned long)mem
;
1102 struct vmap_area
*va
;
1103 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1104 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1108 addr
= va
->va_start
;
1111 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1112 vm_unmap_ram(mem
, count
);
1117 EXPORT_SYMBOL(vm_map_ram
);
1119 static struct vm_struct
*vmlist __initdata
;
1121 * vm_area_add_early - add vmap area early during boot
1122 * @vm: vm_struct to add
1124 * This function is used to add fixed kernel vm area to vmlist before
1125 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1126 * should contain proper values and the other fields should be zero.
1128 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1130 void __init
vm_area_add_early(struct vm_struct
*vm
)
1132 struct vm_struct
*tmp
, **p
;
1134 BUG_ON(vmap_initialized
);
1135 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1136 if (tmp
->addr
>= vm
->addr
) {
1137 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1140 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1147 * vm_area_register_early - register vmap area early during boot
1148 * @vm: vm_struct to register
1149 * @align: requested alignment
1151 * This function is used to register kernel vm area before
1152 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1153 * proper values on entry and other fields should be zero. On return,
1154 * vm->addr contains the allocated address.
1156 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1158 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1160 static size_t vm_init_off __initdata
;
1163 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1164 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1166 vm
->addr
= (void *)addr
;
1168 vm_area_add_early(vm
);
1171 void __init
vmalloc_init(void)
1173 struct vmap_area
*va
;
1174 struct vm_struct
*tmp
;
1177 for_each_possible_cpu(i
) {
1178 struct vmap_block_queue
*vbq
;
1179 struct vfree_deferred
*p
;
1181 vbq
= &per_cpu(vmap_block_queue
, i
);
1182 spin_lock_init(&vbq
->lock
);
1183 INIT_LIST_HEAD(&vbq
->free
);
1184 p
= &per_cpu(vfree_deferred
, i
);
1185 init_llist_head(&p
->list
);
1186 INIT_WORK(&p
->wq
, free_work
);
1189 /* Import existing vmlist entries. */
1190 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1191 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1192 va
->flags
= VM_VM_AREA
;
1193 va
->va_start
= (unsigned long)tmp
->addr
;
1194 va
->va_end
= va
->va_start
+ tmp
->size
;
1196 __insert_vmap_area(va
);
1199 vmap_area_pcpu_hole
= VMALLOC_END
;
1201 vmap_initialized
= true;
1205 * map_kernel_range_noflush - map kernel VM area with the specified pages
1206 * @addr: start of the VM area to map
1207 * @size: size of the VM area to map
1208 * @prot: page protection flags to use
1209 * @pages: pages to map
1211 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1212 * specify should have been allocated using get_vm_area() and its
1216 * This function does NOT do any cache flushing. The caller is
1217 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1218 * before calling this function.
1221 * The number of pages mapped on success, -errno on failure.
1223 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1224 pgprot_t prot
, struct page
**pages
)
1226 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1230 * unmap_kernel_range_noflush - unmap kernel VM area
1231 * @addr: start of the VM area to unmap
1232 * @size: size of the VM area to unmap
1234 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1235 * specify should have been allocated using get_vm_area() and its
1239 * This function does NOT do any cache flushing. The caller is
1240 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1241 * before calling this function and flush_tlb_kernel_range() after.
1243 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1245 vunmap_page_range(addr
, addr
+ size
);
1247 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1250 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1251 * @addr: start of the VM area to unmap
1252 * @size: size of the VM area to unmap
1254 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1255 * the unmapping and tlb after.
1257 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1259 unsigned long end
= addr
+ size
;
1261 flush_cache_vunmap(addr
, end
);
1262 vunmap_page_range(addr
, end
);
1263 flush_tlb_kernel_range(addr
, end
);
1266 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
***pages
)
1268 unsigned long addr
= (unsigned long)area
->addr
;
1269 unsigned long end
= addr
+ get_vm_area_size(area
);
1272 err
= vmap_page_range(addr
, end
, prot
, *pages
);
1280 EXPORT_SYMBOL_GPL(map_vm_area
);
1282 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1283 unsigned long flags
, const void *caller
)
1285 spin_lock(&vmap_area_lock
);
1287 vm
->addr
= (void *)va
->va_start
;
1288 vm
->size
= va
->va_end
- va
->va_start
;
1289 vm
->caller
= caller
;
1291 va
->flags
|= VM_VM_AREA
;
1292 spin_unlock(&vmap_area_lock
);
1295 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1298 * Before removing VM_UNINITIALIZED,
1299 * we should make sure that vm has proper values.
1300 * Pair with smp_rmb() in show_numa_info().
1303 vm
->flags
&= ~VM_UNINITIALIZED
;
1306 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1307 unsigned long align
, unsigned long flags
, unsigned long start
,
1308 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1310 struct vmap_area
*va
;
1311 struct vm_struct
*area
;
1313 BUG_ON(in_interrupt());
1314 if (flags
& VM_IOREMAP
)
1315 align
= 1ul << clamp(fls(size
), PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1317 size
= PAGE_ALIGN(size
);
1318 if (unlikely(!size
))
1321 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1322 if (unlikely(!area
))
1326 * We always allocate a guard page.
1330 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1336 setup_vmalloc_vm(area
, va
, flags
, caller
);
1341 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1342 unsigned long start
, unsigned long end
)
1344 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1345 GFP_KERNEL
, __builtin_return_address(0));
1347 EXPORT_SYMBOL_GPL(__get_vm_area
);
1349 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1350 unsigned long start
, unsigned long end
,
1353 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1354 GFP_KERNEL
, caller
);
1358 * get_vm_area - reserve a contiguous kernel virtual area
1359 * @size: size of the area
1360 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1362 * Search an area of @size in the kernel virtual mapping area,
1363 * and reserved it for out purposes. Returns the area descriptor
1364 * on success or %NULL on failure.
1366 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1368 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1369 NUMA_NO_NODE
, GFP_KERNEL
,
1370 __builtin_return_address(0));
1373 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1376 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1377 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1381 * find_vm_area - find a continuous kernel virtual area
1382 * @addr: base address
1384 * Search for the kernel VM area starting at @addr, and return it.
1385 * It is up to the caller to do all required locking to keep the returned
1388 struct vm_struct
*find_vm_area(const void *addr
)
1390 struct vmap_area
*va
;
1392 va
= find_vmap_area((unsigned long)addr
);
1393 if (va
&& va
->flags
& VM_VM_AREA
)
1400 * remove_vm_area - find and remove a continuous kernel virtual area
1401 * @addr: base address
1403 * Search for the kernel VM area starting at @addr, and remove it.
1404 * This function returns the found VM area, but using it is NOT safe
1405 * on SMP machines, except for its size or flags.
1407 struct vm_struct
*remove_vm_area(const void *addr
)
1409 struct vmap_area
*va
;
1411 va
= find_vmap_area((unsigned long)addr
);
1412 if (va
&& va
->flags
& VM_VM_AREA
) {
1413 struct vm_struct
*vm
= va
->vm
;
1415 spin_lock(&vmap_area_lock
);
1417 va
->flags
&= ~VM_VM_AREA
;
1418 spin_unlock(&vmap_area_lock
);
1420 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1421 free_unmap_vmap_area(va
);
1422 vm
->size
-= PAGE_SIZE
;
1429 static void __vunmap(const void *addr
, int deallocate_pages
)
1431 struct vm_struct
*area
;
1436 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
1440 area
= remove_vm_area(addr
);
1441 if (unlikely(!area
)) {
1442 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1447 debug_check_no_locks_freed(addr
, area
->size
);
1448 debug_check_no_obj_freed(addr
, area
->size
);
1450 if (deallocate_pages
) {
1453 for (i
= 0; i
< area
->nr_pages
; i
++) {
1454 struct page
*page
= area
->pages
[i
];
1460 if (area
->flags
& VM_VPAGES
)
1471 * vfree - release memory allocated by vmalloc()
1472 * @addr: memory base address
1474 * Free the virtually continuous memory area starting at @addr, as
1475 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1476 * NULL, no operation is performed.
1478 * Must not be called in NMI context (strictly speaking, only if we don't
1479 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1480 * conventions for vfree() arch-depenedent would be a really bad idea)
1482 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1484 void vfree(const void *addr
)
1488 kmemleak_free(addr
);
1492 if (unlikely(in_interrupt())) {
1493 struct vfree_deferred
*p
= &__get_cpu_var(vfree_deferred
);
1494 if (llist_add((struct llist_node
*)addr
, &p
->list
))
1495 schedule_work(&p
->wq
);
1499 EXPORT_SYMBOL(vfree
);
1502 * vunmap - release virtual mapping obtained by vmap()
1503 * @addr: memory base address
1505 * Free the virtually contiguous memory area starting at @addr,
1506 * which was created from the page array passed to vmap().
1508 * Must not be called in interrupt context.
1510 void vunmap(const void *addr
)
1512 BUG_ON(in_interrupt());
1517 EXPORT_SYMBOL(vunmap
);
1520 * vmap - map an array of pages into virtually contiguous space
1521 * @pages: array of page pointers
1522 * @count: number of pages to map
1523 * @flags: vm_area->flags
1524 * @prot: page protection for the mapping
1526 * Maps @count pages from @pages into contiguous kernel virtual
1529 void *vmap(struct page
**pages
, unsigned int count
,
1530 unsigned long flags
, pgprot_t prot
)
1532 struct vm_struct
*area
;
1536 if (count
> totalram_pages
)
1539 area
= get_vm_area_caller((count
<< PAGE_SHIFT
), flags
,
1540 __builtin_return_address(0));
1544 if (map_vm_area(area
, prot
, &pages
)) {
1551 EXPORT_SYMBOL(vmap
);
1553 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1554 gfp_t gfp_mask
, pgprot_t prot
,
1555 int node
, const void *caller
);
1556 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1557 pgprot_t prot
, int node
)
1559 const int order
= 0;
1560 struct page
**pages
;
1561 unsigned int nr_pages
, array_size
, i
;
1562 gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1564 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
1565 array_size
= (nr_pages
* sizeof(struct page
*));
1567 area
->nr_pages
= nr_pages
;
1568 /* Please note that the recursion is strictly bounded. */
1569 if (array_size
> PAGE_SIZE
) {
1570 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1571 PAGE_KERNEL
, node
, area
->caller
);
1572 area
->flags
|= VM_VPAGES
;
1574 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1576 area
->pages
= pages
;
1578 remove_vm_area(area
->addr
);
1583 for (i
= 0; i
< area
->nr_pages
; i
++) {
1585 gfp_t tmp_mask
= gfp_mask
| __GFP_NOWARN
;
1587 if (node
== NUMA_NO_NODE
)
1588 page
= alloc_page(tmp_mask
);
1590 page
= alloc_pages_node(node
, tmp_mask
, order
);
1592 if (unlikely(!page
)) {
1593 /* Successfully allocated i pages, free them in __vunmap() */
1597 area
->pages
[i
] = page
;
1600 if (map_vm_area(area
, prot
, &pages
))
1605 warn_alloc_failed(gfp_mask
, order
,
1606 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1607 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1613 * __vmalloc_node_range - allocate virtually contiguous memory
1614 * @size: allocation size
1615 * @align: desired alignment
1616 * @start: vm area range start
1617 * @end: vm area range end
1618 * @gfp_mask: flags for the page level allocator
1619 * @prot: protection mask for the allocated pages
1620 * @node: node to use for allocation or NUMA_NO_NODE
1621 * @caller: caller's return address
1623 * Allocate enough pages to cover @size from the page level
1624 * allocator with @gfp_mask flags. Map them into contiguous
1625 * kernel virtual space, using a pagetable protection of @prot.
1627 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1628 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1629 pgprot_t prot
, int node
, const void *caller
)
1631 struct vm_struct
*area
;
1633 unsigned long real_size
= size
;
1635 size
= PAGE_ALIGN(size
);
1636 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1639 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
,
1640 start
, end
, node
, gfp_mask
, caller
);
1644 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
1649 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1650 * flag. It means that vm_struct is not fully initialized.
1651 * Now, it is fully initialized, so remove this flag here.
1653 clear_vm_uninitialized_flag(area
);
1656 * A ref_count = 2 is needed because vm_struct allocated in
1657 * __get_vm_area_node() contains a reference to the virtual address of
1658 * the vmalloc'ed block.
1660 kmemleak_alloc(addr
, real_size
, 2, gfp_mask
);
1665 warn_alloc_failed(gfp_mask
, 0,
1666 "vmalloc: allocation failure: %lu bytes\n",
1672 * __vmalloc_node - allocate virtually contiguous memory
1673 * @size: allocation size
1674 * @align: desired alignment
1675 * @gfp_mask: flags for the page level allocator
1676 * @prot: protection mask for the allocated pages
1677 * @node: node to use for allocation or NUMA_NO_NODE
1678 * @caller: caller's return address
1680 * Allocate enough pages to cover @size from the page level
1681 * allocator with @gfp_mask flags. Map them into contiguous
1682 * kernel virtual space, using a pagetable protection of @prot.
1684 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1685 gfp_t gfp_mask
, pgprot_t prot
,
1686 int node
, const void *caller
)
1688 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1689 gfp_mask
, prot
, node
, caller
);
1692 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1694 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1695 __builtin_return_address(0));
1697 EXPORT_SYMBOL(__vmalloc
);
1699 static inline void *__vmalloc_node_flags(unsigned long size
,
1700 int node
, gfp_t flags
)
1702 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
1703 node
, __builtin_return_address(0));
1707 * vmalloc - allocate virtually contiguous memory
1708 * @size: allocation size
1709 * Allocate enough pages to cover @size from the page level
1710 * allocator and map them into contiguous kernel virtual space.
1712 * For tight control over page level allocator and protection flags
1713 * use __vmalloc() instead.
1715 void *vmalloc(unsigned long size
)
1717 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1718 GFP_KERNEL
| __GFP_HIGHMEM
);
1720 EXPORT_SYMBOL(vmalloc
);
1723 * vzalloc - allocate virtually contiguous memory with zero fill
1724 * @size: allocation size
1725 * Allocate enough pages to cover @size from the page level
1726 * allocator and map them into contiguous kernel virtual space.
1727 * The memory allocated is set to zero.
1729 * For tight control over page level allocator and protection flags
1730 * use __vmalloc() instead.
1732 void *vzalloc(unsigned long size
)
1734 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1735 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1737 EXPORT_SYMBOL(vzalloc
);
1740 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1741 * @size: allocation size
1743 * The resulting memory area is zeroed so it can be mapped to userspace
1744 * without leaking data.
1746 void *vmalloc_user(unsigned long size
)
1748 struct vm_struct
*area
;
1751 ret
= __vmalloc_node(size
, SHMLBA
,
1752 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1753 PAGE_KERNEL
, NUMA_NO_NODE
,
1754 __builtin_return_address(0));
1756 area
= find_vm_area(ret
);
1757 area
->flags
|= VM_USERMAP
;
1761 EXPORT_SYMBOL(vmalloc_user
);
1764 * vmalloc_node - allocate memory on a specific node
1765 * @size: allocation size
1768 * Allocate enough pages to cover @size from the page level
1769 * allocator and map them into contiguous kernel virtual space.
1771 * For tight control over page level allocator and protection flags
1772 * use __vmalloc() instead.
1774 void *vmalloc_node(unsigned long size
, int node
)
1776 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1777 node
, __builtin_return_address(0));
1779 EXPORT_SYMBOL(vmalloc_node
);
1782 * vzalloc_node - allocate memory on a specific node with zero fill
1783 * @size: allocation size
1786 * Allocate enough pages to cover @size from the page level
1787 * allocator and map them into contiguous kernel virtual space.
1788 * The memory allocated is set to zero.
1790 * For tight control over page level allocator and protection flags
1791 * use __vmalloc_node() instead.
1793 void *vzalloc_node(unsigned long size
, int node
)
1795 return __vmalloc_node_flags(size
, node
,
1796 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1798 EXPORT_SYMBOL(vzalloc_node
);
1800 #ifndef PAGE_KERNEL_EXEC
1801 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1805 * vmalloc_exec - allocate virtually contiguous, executable memory
1806 * @size: allocation size
1808 * Kernel-internal function to allocate enough pages to cover @size
1809 * the page level allocator and map them into contiguous and
1810 * executable kernel virtual space.
1812 * For tight control over page level allocator and protection flags
1813 * use __vmalloc() instead.
1816 void *vmalloc_exec(unsigned long size
)
1818 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1819 NUMA_NO_NODE
, __builtin_return_address(0));
1822 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1823 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1824 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1825 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1827 #define GFP_VMALLOC32 GFP_KERNEL
1831 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1832 * @size: allocation size
1834 * Allocate enough 32bit PA addressable pages to cover @size from the
1835 * page level allocator and map them into contiguous kernel virtual space.
1837 void *vmalloc_32(unsigned long size
)
1839 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1840 NUMA_NO_NODE
, __builtin_return_address(0));
1842 EXPORT_SYMBOL(vmalloc_32
);
1845 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1846 * @size: allocation size
1848 * The resulting memory area is 32bit addressable and zeroed so it can be
1849 * mapped to userspace without leaking data.
1851 void *vmalloc_32_user(unsigned long size
)
1853 struct vm_struct
*area
;
1856 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1857 NUMA_NO_NODE
, __builtin_return_address(0));
1859 area
= find_vm_area(ret
);
1860 area
->flags
|= VM_USERMAP
;
1864 EXPORT_SYMBOL(vmalloc_32_user
);
1867 * small helper routine , copy contents to buf from addr.
1868 * If the page is not present, fill zero.
1871 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1877 unsigned long offset
, length
;
1879 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1880 length
= PAGE_SIZE
- offset
;
1883 p
= vmalloc_to_page(addr
);
1885 * To do safe access to this _mapped_ area, we need
1886 * lock. But adding lock here means that we need to add
1887 * overhead of vmalloc()/vfree() calles for this _debug_
1888 * interface, rarely used. Instead of that, we'll use
1889 * kmap() and get small overhead in this access function.
1893 * we can expect USER0 is not used (see vread/vwrite's
1894 * function description)
1896 void *map
= kmap_atomic(p
);
1897 memcpy(buf
, map
+ offset
, length
);
1900 memset(buf
, 0, length
);
1910 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
1916 unsigned long offset
, length
;
1918 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1919 length
= PAGE_SIZE
- offset
;
1922 p
= vmalloc_to_page(addr
);
1924 * To do safe access to this _mapped_ area, we need
1925 * lock. But adding lock here means that we need to add
1926 * overhead of vmalloc()/vfree() calles for this _debug_
1927 * interface, rarely used. Instead of that, we'll use
1928 * kmap() and get small overhead in this access function.
1932 * we can expect USER0 is not used (see vread/vwrite's
1933 * function description)
1935 void *map
= kmap_atomic(p
);
1936 memcpy(map
+ offset
, buf
, length
);
1948 * vread() - read vmalloc area in a safe way.
1949 * @buf: buffer for reading data
1950 * @addr: vm address.
1951 * @count: number of bytes to be read.
1953 * Returns # of bytes which addr and buf should be increased.
1954 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1955 * includes any intersect with alive vmalloc area.
1957 * This function checks that addr is a valid vmalloc'ed area, and
1958 * copy data from that area to a given buffer. If the given memory range
1959 * of [addr...addr+count) includes some valid address, data is copied to
1960 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1961 * IOREMAP area is treated as memory hole and no copy is done.
1963 * If [addr...addr+count) doesn't includes any intersects with alive
1964 * vm_struct area, returns 0. @buf should be kernel's buffer.
1966 * Note: In usual ops, vread() is never necessary because the caller
1967 * should know vmalloc() area is valid and can use memcpy().
1968 * This is for routines which have to access vmalloc area without
1969 * any informaion, as /dev/kmem.
1973 long vread(char *buf
, char *addr
, unsigned long count
)
1975 struct vmap_area
*va
;
1976 struct vm_struct
*vm
;
1977 char *vaddr
, *buf_start
= buf
;
1978 unsigned long buflen
= count
;
1981 /* Don't allow overflow */
1982 if ((unsigned long) addr
+ count
< count
)
1983 count
= -(unsigned long) addr
;
1985 spin_lock(&vmap_area_lock
);
1986 list_for_each_entry(va
, &vmap_area_list
, list
) {
1990 if (!(va
->flags
& VM_VM_AREA
))
1994 vaddr
= (char *) vm
->addr
;
1995 if (addr
>= vaddr
+ get_vm_area_size(vm
))
1997 while (addr
< vaddr
) {
2005 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2008 if (!(vm
->flags
& VM_IOREMAP
))
2009 aligned_vread(buf
, addr
, n
);
2010 else /* IOREMAP area is treated as memory hole */
2017 spin_unlock(&vmap_area_lock
);
2019 if (buf
== buf_start
)
2021 /* zero-fill memory holes */
2022 if (buf
!= buf_start
+ buflen
)
2023 memset(buf
, 0, buflen
- (buf
- buf_start
));
2029 * vwrite() - write vmalloc area in a safe way.
2030 * @buf: buffer for source data
2031 * @addr: vm address.
2032 * @count: number of bytes to be read.
2034 * Returns # of bytes which addr and buf should be incresed.
2035 * (same number to @count).
2036 * If [addr...addr+count) doesn't includes any intersect with valid
2037 * vmalloc area, returns 0.
2039 * This function checks that addr is a valid vmalloc'ed area, and
2040 * copy data from a buffer to the given addr. If specified range of
2041 * [addr...addr+count) includes some valid address, data is copied from
2042 * proper area of @buf. If there are memory holes, no copy to hole.
2043 * IOREMAP area is treated as memory hole and no copy is done.
2045 * If [addr...addr+count) doesn't includes any intersects with alive
2046 * vm_struct area, returns 0. @buf should be kernel's buffer.
2048 * Note: In usual ops, vwrite() is never necessary because the caller
2049 * should know vmalloc() area is valid and can use memcpy().
2050 * This is for routines which have to access vmalloc area without
2051 * any informaion, as /dev/kmem.
2054 long vwrite(char *buf
, char *addr
, unsigned long count
)
2056 struct vmap_area
*va
;
2057 struct vm_struct
*vm
;
2059 unsigned long n
, buflen
;
2062 /* Don't allow overflow */
2063 if ((unsigned long) addr
+ count
< count
)
2064 count
= -(unsigned long) addr
;
2067 spin_lock(&vmap_area_lock
);
2068 list_for_each_entry(va
, &vmap_area_list
, list
) {
2072 if (!(va
->flags
& VM_VM_AREA
))
2076 vaddr
= (char *) vm
->addr
;
2077 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2079 while (addr
< vaddr
) {
2086 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2089 if (!(vm
->flags
& VM_IOREMAP
)) {
2090 aligned_vwrite(buf
, addr
, n
);
2098 spin_unlock(&vmap_area_lock
);
2105 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2106 * @vma: vma to cover
2107 * @uaddr: target user address to start at
2108 * @kaddr: virtual address of vmalloc kernel memory
2109 * @size: size of map area
2111 * Returns: 0 for success, -Exxx on failure
2113 * This function checks that @kaddr is a valid vmalloc'ed area,
2114 * and that it is big enough to cover the range starting at
2115 * @uaddr in @vma. Will return failure if that criteria isn't
2118 * Similar to remap_pfn_range() (see mm/memory.c)
2120 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2121 void *kaddr
, unsigned long size
)
2123 struct vm_struct
*area
;
2125 size
= PAGE_ALIGN(size
);
2127 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2130 area
= find_vm_area(kaddr
);
2134 if (!(area
->flags
& VM_USERMAP
))
2137 if (kaddr
+ size
> area
->addr
+ area
->size
)
2141 struct page
*page
= vmalloc_to_page(kaddr
);
2144 ret
= vm_insert_page(vma
, uaddr
, page
);
2153 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2157 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2160 * remap_vmalloc_range - map vmalloc pages to userspace
2161 * @vma: vma to cover (map full range of vma)
2162 * @addr: vmalloc memory
2163 * @pgoff: number of pages into addr before first page to map
2165 * Returns: 0 for success, -Exxx on failure
2167 * This function checks that addr is a valid vmalloc'ed area, and
2168 * that it is big enough to cover the vma. Will return failure if
2169 * that criteria isn't met.
2171 * Similar to remap_pfn_range() (see mm/memory.c)
2173 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2174 unsigned long pgoff
)
2176 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2177 addr
+ (pgoff
<< PAGE_SHIFT
),
2178 vma
->vm_end
- vma
->vm_start
);
2180 EXPORT_SYMBOL(remap_vmalloc_range
);
2183 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2186 void __weak
vmalloc_sync_all(void)
2191 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2203 * alloc_vm_area - allocate a range of kernel address space
2204 * @size: size of the area
2205 * @ptes: returns the PTEs for the address space
2207 * Returns: NULL on failure, vm_struct on success
2209 * This function reserves a range of kernel address space, and
2210 * allocates pagetables to map that range. No actual mappings
2213 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2214 * allocated for the VM area are returned.
2216 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2218 struct vm_struct
*area
;
2220 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2221 __builtin_return_address(0));
2226 * This ensures that page tables are constructed for this region
2227 * of kernel virtual address space and mapped into init_mm.
2229 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2230 size
, f
, ptes
? &ptes
: NULL
)) {
2237 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2239 void free_vm_area(struct vm_struct
*area
)
2241 struct vm_struct
*ret
;
2242 ret
= remove_vm_area(area
->addr
);
2243 BUG_ON(ret
!= area
);
2246 EXPORT_SYMBOL_GPL(free_vm_area
);
2249 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2251 return n
? rb_entry(n
, struct vmap_area
, rb_node
) : NULL
;
2255 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2256 * @end: target address
2257 * @pnext: out arg for the next vmap_area
2258 * @pprev: out arg for the previous vmap_area
2260 * Returns: %true if either or both of next and prev are found,
2261 * %false if no vmap_area exists
2263 * Find vmap_areas end addresses of which enclose @end. ie. if not
2264 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2266 static bool pvm_find_next_prev(unsigned long end
,
2267 struct vmap_area
**pnext
,
2268 struct vmap_area
**pprev
)
2270 struct rb_node
*n
= vmap_area_root
.rb_node
;
2271 struct vmap_area
*va
= NULL
;
2274 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2275 if (end
< va
->va_end
)
2277 else if (end
> va
->va_end
)
2286 if (va
->va_end
> end
) {
2288 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2291 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2297 * pvm_determine_end - find the highest aligned address between two vmap_areas
2298 * @pnext: in/out arg for the next vmap_area
2299 * @pprev: in/out arg for the previous vmap_area
2302 * Returns: determined end address
2304 * Find the highest aligned address between *@pnext and *@pprev below
2305 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2306 * down address is between the end addresses of the two vmap_areas.
2308 * Please note that the address returned by this function may fall
2309 * inside *@pnext vmap_area. The caller is responsible for checking
2312 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2313 struct vmap_area
**pprev
,
2314 unsigned long align
)
2316 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2320 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2324 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2326 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2333 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2334 * @offsets: array containing offset of each area
2335 * @sizes: array containing size of each area
2336 * @nr_vms: the number of areas to allocate
2337 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2339 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2340 * vm_structs on success, %NULL on failure
2342 * Percpu allocator wants to use congruent vm areas so that it can
2343 * maintain the offsets among percpu areas. This function allocates
2344 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2345 * be scattered pretty far, distance between two areas easily going up
2346 * to gigabytes. To avoid interacting with regular vmallocs, these
2347 * areas are allocated from top.
2349 * Despite its complicated look, this allocator is rather simple. It
2350 * does everything top-down and scans areas from the end looking for
2351 * matching slot. While scanning, if any of the areas overlaps with
2352 * existing vmap_area, the base address is pulled down to fit the
2353 * area. Scanning is repeated till all the areas fit and then all
2354 * necessary data structres are inserted and the result is returned.
2356 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2357 const size_t *sizes
, int nr_vms
,
2360 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2361 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2362 struct vmap_area
**vas
, *prev
, *next
;
2363 struct vm_struct
**vms
;
2364 int area
, area2
, last_area
, term_area
;
2365 unsigned long base
, start
, end
, last_end
;
2366 bool purged
= false;
2368 /* verify parameters and allocate data structures */
2369 BUG_ON(align
& ~PAGE_MASK
|| !is_power_of_2(align
));
2370 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2371 start
= offsets
[area
];
2372 end
= start
+ sizes
[area
];
2374 /* is everything aligned properly? */
2375 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2376 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2378 /* detect the area with the highest address */
2379 if (start
> offsets
[last_area
])
2382 for (area2
= 0; area2
< nr_vms
; area2
++) {
2383 unsigned long start2
= offsets
[area2
];
2384 unsigned long end2
= start2
+ sizes
[area2
];
2389 BUG_ON(start2
>= start
&& start2
< end
);
2390 BUG_ON(end2
<= end
&& end2
> start
);
2393 last_end
= offsets
[last_area
] + sizes
[last_area
];
2395 if (vmalloc_end
- vmalloc_start
< last_end
) {
2400 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2401 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2405 for (area
= 0; area
< nr_vms
; area
++) {
2406 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2407 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2408 if (!vas
[area
] || !vms
[area
])
2412 spin_lock(&vmap_area_lock
);
2414 /* start scanning - we scan from the top, begin with the last area */
2415 area
= term_area
= last_area
;
2416 start
= offsets
[area
];
2417 end
= start
+ sizes
[area
];
2419 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2420 base
= vmalloc_end
- last_end
;
2423 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2426 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2427 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2430 * base might have underflowed, add last_end before
2433 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2434 spin_unlock(&vmap_area_lock
);
2436 purge_vmap_area_lazy();
2444 * If next overlaps, move base downwards so that it's
2445 * right below next and then recheck.
2447 if (next
&& next
->va_start
< base
+ end
) {
2448 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2454 * If prev overlaps, shift down next and prev and move
2455 * base so that it's right below new next and then
2458 if (prev
&& prev
->va_end
> base
+ start
) {
2460 prev
= node_to_va(rb_prev(&next
->rb_node
));
2461 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2467 * This area fits, move on to the previous one. If
2468 * the previous one is the terminal one, we're done.
2470 area
= (area
+ nr_vms
- 1) % nr_vms
;
2471 if (area
== term_area
)
2473 start
= offsets
[area
];
2474 end
= start
+ sizes
[area
];
2475 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2478 /* we've found a fitting base, insert all va's */
2479 for (area
= 0; area
< nr_vms
; area
++) {
2480 struct vmap_area
*va
= vas
[area
];
2482 va
->va_start
= base
+ offsets
[area
];
2483 va
->va_end
= va
->va_start
+ sizes
[area
];
2484 __insert_vmap_area(va
);
2487 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2489 spin_unlock(&vmap_area_lock
);
2491 /* insert all vm's */
2492 for (area
= 0; area
< nr_vms
; area
++)
2493 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2500 for (area
= 0; area
< nr_vms
; area
++) {
2511 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2512 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2513 * @nr_vms: the number of allocated areas
2515 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2517 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2521 for (i
= 0; i
< nr_vms
; i
++)
2522 free_vm_area(vms
[i
]);
2525 #endif /* CONFIG_SMP */
2527 #ifdef CONFIG_PROC_FS
2528 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2529 __acquires(&vmap_area_lock
)
2532 struct vmap_area
*va
;
2534 spin_lock(&vmap_area_lock
);
2535 va
= list_entry((&vmap_area_list
)->next
, typeof(*va
), list
);
2536 while (n
> 0 && &va
->list
!= &vmap_area_list
) {
2538 va
= list_entry(va
->list
.next
, typeof(*va
), list
);
2540 if (!n
&& &va
->list
!= &vmap_area_list
)
2547 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2549 struct vmap_area
*va
= p
, *next
;
2552 next
= list_entry(va
->list
.next
, typeof(*va
), list
);
2553 if (&next
->list
!= &vmap_area_list
)
2559 static void s_stop(struct seq_file
*m
, void *p
)
2560 __releases(&vmap_area_lock
)
2562 spin_unlock(&vmap_area_lock
);
2565 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2567 if (IS_ENABLED(CONFIG_NUMA
)) {
2568 unsigned int nr
, *counters
= m
->private;
2573 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2575 if (v
->flags
& VM_UNINITIALIZED
)
2578 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2580 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2581 counters
[page_to_nid(v
->pages
[nr
])]++;
2583 for_each_node_state(nr
, N_HIGH_MEMORY
)
2585 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2589 static int s_show(struct seq_file
*m
, void *p
)
2591 struct vmap_area
*va
= p
;
2592 struct vm_struct
*v
;
2595 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2596 * behalf of vmap area is being tear down or vm_map_ram allocation.
2598 if (!(va
->flags
& VM_VM_AREA
))
2603 seq_printf(m
, "0x%pK-0x%pK %7ld",
2604 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2607 seq_printf(m
, " %pS", v
->caller
);
2610 seq_printf(m
, " pages=%d", v
->nr_pages
);
2613 seq_printf(m
, " phys=%llx", (unsigned long long)v
->phys_addr
);
2615 if (v
->flags
& VM_IOREMAP
)
2616 seq_printf(m
, " ioremap");
2618 if (v
->flags
& VM_ALLOC
)
2619 seq_printf(m
, " vmalloc");
2621 if (v
->flags
& VM_MAP
)
2622 seq_printf(m
, " vmap");
2624 if (v
->flags
& VM_USERMAP
)
2625 seq_printf(m
, " user");
2627 if (v
->flags
& VM_VPAGES
)
2628 seq_printf(m
, " vpages");
2630 show_numa_info(m
, v
);
2635 static const struct seq_operations vmalloc_op
= {
2642 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2644 unsigned int *ptr
= NULL
;
2647 if (IS_ENABLED(CONFIG_NUMA
)) {
2648 ptr
= kmalloc(nr_node_ids
* sizeof(unsigned int), GFP_KERNEL
);
2652 ret
= seq_open(file
, &vmalloc_op
);
2654 struct seq_file
*m
= file
->private_data
;
2661 static const struct file_operations proc_vmalloc_operations
= {
2662 .open
= vmalloc_open
,
2664 .llseek
= seq_lseek
,
2665 .release
= seq_release_private
,
2668 static int __init
proc_vmalloc_init(void)
2670 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2673 module_init(proc_vmalloc_init
);
2675 void get_vmalloc_info(struct vmalloc_info
*vmi
)
2677 struct vmap_area
*va
;
2678 unsigned long free_area_size
;
2679 unsigned long prev_end
;
2682 vmi
->largest_chunk
= 0;
2684 prev_end
= VMALLOC_START
;
2686 spin_lock(&vmap_area_lock
);
2688 if (list_empty(&vmap_area_list
)) {
2689 vmi
->largest_chunk
= VMALLOC_TOTAL
;
2693 list_for_each_entry(va
, &vmap_area_list
, list
) {
2694 unsigned long addr
= va
->va_start
;
2697 * Some archs keep another range for modules in vmalloc space
2699 if (addr
< VMALLOC_START
)
2701 if (addr
>= VMALLOC_END
)
2704 if (va
->flags
& (VM_LAZY_FREE
| VM_LAZY_FREEING
))
2707 vmi
->used
+= (va
->va_end
- va
->va_start
);
2709 free_area_size
= addr
- prev_end
;
2710 if (vmi
->largest_chunk
< free_area_size
)
2711 vmi
->largest_chunk
= free_area_size
;
2713 prev_end
= va
->va_end
;
2716 if (VMALLOC_END
- prev_end
> vmi
->largest_chunk
)
2717 vmi
->largest_chunk
= VMALLOC_END
- prev_end
;
2720 spin_unlock(&vmap_area_lock
);