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/notifier.h>
25 #include <linux/rbtree.h>
26 #include <linux/radix-tree.h>
27 #include <linux/rcupdate.h>
28 #include <linux/pfn.h>
29 #include <linux/kmemleak.h>
30 #include <linux/atomic.h>
31 #include <linux/compiler.h>
32 #include <linux/llist.h>
33 #include <linux/bitops.h>
35 #include <asm/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred
{
42 struct llist_head list
;
43 struct work_struct wq
;
45 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct
*w
)
51 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
52 struct llist_node
*llnode
= llist_del_all(&p
->list
);
55 llnode
= llist_next(llnode
);
60 /*** Page table manipulation functions ***/
62 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
66 pte
= pte_offset_kernel(pmd
, addr
);
68 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
69 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
70 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
73 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
78 pmd
= pmd_offset(pud
, addr
);
80 next
= pmd_addr_end(addr
, end
);
81 if (pmd_clear_huge(pmd
))
83 if (pmd_none_or_clear_bad(pmd
))
85 vunmap_pte_range(pmd
, addr
, next
);
86 } while (pmd
++, addr
= next
, addr
!= end
);
89 static void vunmap_pud_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
94 pud
= pud_offset(pgd
, addr
);
96 next
= pud_addr_end(addr
, end
);
97 if (pud_clear_huge(pud
))
99 if (pud_none_or_clear_bad(pud
))
101 vunmap_pmd_range(pud
, addr
, next
);
102 } while (pud
++, addr
= next
, addr
!= end
);
105 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
111 pgd
= pgd_offset_k(addr
);
113 next
= pgd_addr_end(addr
, end
);
114 if (pgd_none_or_clear_bad(pgd
))
116 vunmap_pud_range(pgd
, addr
, next
);
117 } while (pgd
++, addr
= next
, addr
!= end
);
120 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
121 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
126 * nr is a running index into the array which helps higher level
127 * callers keep track of where we're up to.
130 pte
= pte_alloc_kernel(pmd
, addr
);
134 struct page
*page
= pages
[*nr
];
136 if (WARN_ON(!pte_none(*pte
)))
140 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
142 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
146 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
147 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
152 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
156 next
= pmd_addr_end(addr
, end
);
157 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
159 } while (pmd
++, addr
= next
, addr
!= end
);
163 static int vmap_pud_range(pgd_t
*pgd
, unsigned long addr
,
164 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
169 pud
= pud_alloc(&init_mm
, pgd
, addr
);
173 next
= pud_addr_end(addr
, end
);
174 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
176 } while (pud
++, addr
= next
, addr
!= end
);
181 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
182 * will have pfns corresponding to the "pages" array.
184 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
186 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
187 pgprot_t prot
, struct page
**pages
)
191 unsigned long addr
= start
;
196 pgd
= pgd_offset_k(addr
);
198 next
= pgd_addr_end(addr
, end
);
199 err
= vmap_pud_range(pgd
, addr
, next
, prot
, pages
, &nr
);
202 } while (pgd
++, addr
= next
, addr
!= end
);
207 static int vmap_page_range(unsigned long start
, unsigned long end
,
208 pgprot_t prot
, struct page
**pages
)
212 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
213 flush_cache_vmap(start
, end
);
217 int is_vmalloc_or_module_addr(const void *x
)
220 * ARM, x86-64 and sparc64 put modules in a special place,
221 * and fall back on vmalloc() if that fails. Others
222 * just put it in the vmalloc space.
224 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
225 unsigned long addr
= (unsigned long)x
;
226 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
229 return is_vmalloc_addr(x
);
233 * Walk a vmap address to the struct page it maps.
235 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
237 unsigned long addr
= (unsigned long) vmalloc_addr
;
238 struct page
*page
= NULL
;
239 pgd_t
*pgd
= pgd_offset_k(addr
);
242 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
243 * architectures that do not vmalloc module space
245 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
247 if (!pgd_none(*pgd
)) {
248 pud_t
*pud
= pud_offset(pgd
, addr
);
249 if (!pud_none(*pud
)) {
250 pmd_t
*pmd
= pmd_offset(pud
, addr
);
251 if (!pmd_none(*pmd
)) {
254 ptep
= pte_offset_map(pmd
, addr
);
256 if (pte_present(pte
))
257 page
= pte_page(pte
);
264 EXPORT_SYMBOL(vmalloc_to_page
);
267 * Map a vmalloc()-space virtual address to the physical page frame number.
269 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
271 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
273 EXPORT_SYMBOL(vmalloc_to_pfn
);
276 /*** Global kva allocator ***/
278 #define VM_VM_AREA 0x04
280 static DEFINE_SPINLOCK(vmap_area_lock
);
281 /* Export for kexec only */
282 LIST_HEAD(vmap_area_list
);
283 static LLIST_HEAD(vmap_purge_list
);
284 static struct rb_root vmap_area_root
= RB_ROOT
;
286 /* The vmap cache globals are protected by vmap_area_lock */
287 static struct rb_node
*free_vmap_cache
;
288 static unsigned long cached_hole_size
;
289 static unsigned long cached_vstart
;
290 static unsigned long cached_align
;
292 static unsigned long vmap_area_pcpu_hole
;
294 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
296 struct rb_node
*n
= vmap_area_root
.rb_node
;
299 struct vmap_area
*va
;
301 va
= rb_entry(n
, struct vmap_area
, rb_node
);
302 if (addr
< va
->va_start
)
304 else if (addr
>= va
->va_end
)
313 static void __insert_vmap_area(struct vmap_area
*va
)
315 struct rb_node
**p
= &vmap_area_root
.rb_node
;
316 struct rb_node
*parent
= NULL
;
320 struct vmap_area
*tmp_va
;
323 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
324 if (va
->va_start
< tmp_va
->va_end
)
326 else if (va
->va_end
> tmp_va
->va_start
)
332 rb_link_node(&va
->rb_node
, parent
, p
);
333 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
335 /* address-sort this list */
336 tmp
= rb_prev(&va
->rb_node
);
338 struct vmap_area
*prev
;
339 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
340 list_add_rcu(&va
->list
, &prev
->list
);
342 list_add_rcu(&va
->list
, &vmap_area_list
);
345 static void purge_vmap_area_lazy(void);
347 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list
);
350 * Allocate a region of KVA of the specified size and alignment, within the
353 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
355 unsigned long vstart
, unsigned long vend
,
356 int node
, gfp_t gfp_mask
)
358 struct vmap_area
*va
;
362 struct vmap_area
*first
;
365 BUG_ON(offset_in_page(size
));
366 BUG_ON(!is_power_of_2(align
));
370 va
= kmalloc_node(sizeof(struct vmap_area
),
371 gfp_mask
& GFP_RECLAIM_MASK
, node
);
373 return ERR_PTR(-ENOMEM
);
376 * Only scan the relevant parts containing pointers to other objects
377 * to avoid false negatives.
379 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
382 spin_lock(&vmap_area_lock
);
384 * Invalidate cache if we have more permissive parameters.
385 * cached_hole_size notes the largest hole noticed _below_
386 * the vmap_area cached in free_vmap_cache: if size fits
387 * into that hole, we want to scan from vstart to reuse
388 * the hole instead of allocating above free_vmap_cache.
389 * Note that __free_vmap_area may update free_vmap_cache
390 * without updating cached_hole_size or cached_align.
392 if (!free_vmap_cache
||
393 size
< cached_hole_size
||
394 vstart
< cached_vstart
||
395 align
< cached_align
) {
397 cached_hole_size
= 0;
398 free_vmap_cache
= NULL
;
400 /* record if we encounter less permissive parameters */
401 cached_vstart
= vstart
;
402 cached_align
= align
;
404 /* find starting point for our search */
405 if (free_vmap_cache
) {
406 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
407 addr
= ALIGN(first
->va_end
, align
);
410 if (addr
+ size
< addr
)
414 addr
= ALIGN(vstart
, align
);
415 if (addr
+ size
< addr
)
418 n
= vmap_area_root
.rb_node
;
422 struct vmap_area
*tmp
;
423 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
424 if (tmp
->va_end
>= addr
) {
426 if (tmp
->va_start
<= addr
)
437 /* from the starting point, walk areas until a suitable hole is found */
438 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
439 if (addr
+ cached_hole_size
< first
->va_start
)
440 cached_hole_size
= first
->va_start
- addr
;
441 addr
= ALIGN(first
->va_end
, align
);
442 if (addr
+ size
< addr
)
445 if (list_is_last(&first
->list
, &vmap_area_list
))
448 first
= list_next_entry(first
, list
);
452 if (addr
+ size
> vend
)
456 va
->va_end
= addr
+ size
;
458 __insert_vmap_area(va
);
459 free_vmap_cache
= &va
->rb_node
;
460 spin_unlock(&vmap_area_lock
);
462 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
463 BUG_ON(va
->va_start
< vstart
);
464 BUG_ON(va
->va_end
> vend
);
469 spin_unlock(&vmap_area_lock
);
471 purge_vmap_area_lazy();
476 if (gfpflags_allow_blocking(gfp_mask
)) {
477 unsigned long freed
= 0;
478 blocking_notifier_call_chain(&vmap_notify_list
, 0, &freed
);
485 if (printk_ratelimit())
486 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
489 return ERR_PTR(-EBUSY
);
492 int register_vmap_purge_notifier(struct notifier_block
*nb
)
494 return blocking_notifier_chain_register(&vmap_notify_list
, nb
);
496 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier
);
498 int unregister_vmap_purge_notifier(struct notifier_block
*nb
)
500 return blocking_notifier_chain_unregister(&vmap_notify_list
, nb
);
502 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier
);
504 static void __free_vmap_area(struct vmap_area
*va
)
506 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
508 if (free_vmap_cache
) {
509 if (va
->va_end
< cached_vstart
) {
510 free_vmap_cache
= NULL
;
512 struct vmap_area
*cache
;
513 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
514 if (va
->va_start
<= cache
->va_start
) {
515 free_vmap_cache
= rb_prev(&va
->rb_node
);
517 * We don't try to update cached_hole_size or
518 * cached_align, but it won't go very wrong.
523 rb_erase(&va
->rb_node
, &vmap_area_root
);
524 RB_CLEAR_NODE(&va
->rb_node
);
525 list_del_rcu(&va
->list
);
528 * Track the highest possible candidate for pcpu area
529 * allocation. Areas outside of vmalloc area can be returned
530 * here too, consider only end addresses which fall inside
531 * vmalloc area proper.
533 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
534 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
536 kfree_rcu(va
, rcu_head
);
540 * Free a region of KVA allocated by alloc_vmap_area
542 static void free_vmap_area(struct vmap_area
*va
)
544 spin_lock(&vmap_area_lock
);
545 __free_vmap_area(va
);
546 spin_unlock(&vmap_area_lock
);
550 * Clear the pagetable entries of a given vmap_area
552 static void unmap_vmap_area(struct vmap_area
*va
)
554 vunmap_page_range(va
->va_start
, va
->va_end
);
557 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
560 * Unmap page tables and force a TLB flush immediately if pagealloc
561 * debugging is enabled. This catches use after free bugs similarly to
562 * those in linear kernel virtual address space after a page has been
565 * All the lazy freeing logic is still retained, in order to minimise
566 * intrusiveness of this debugging feature.
568 * This is going to be *slow* (linear kernel virtual address debugging
569 * doesn't do a broadcast TLB flush so it is a lot faster).
571 if (debug_pagealloc_enabled()) {
572 vunmap_page_range(start
, end
);
573 flush_tlb_kernel_range(start
, end
);
578 * lazy_max_pages is the maximum amount of virtual address space we gather up
579 * before attempting to purge with a TLB flush.
581 * There is a tradeoff here: a larger number will cover more kernel page tables
582 * and take slightly longer to purge, but it will linearly reduce the number of
583 * global TLB flushes that must be performed. It would seem natural to scale
584 * this number up linearly with the number of CPUs (because vmapping activity
585 * could also scale linearly with the number of CPUs), however it is likely
586 * that in practice, workloads might be constrained in other ways that mean
587 * vmap activity will not scale linearly with CPUs. Also, I want to be
588 * conservative and not introduce a big latency on huge systems, so go with
589 * a less aggressive log scale. It will still be an improvement over the old
590 * code, and it will be simple to change the scale factor if we find that it
591 * becomes a problem on bigger systems.
593 static unsigned long lazy_max_pages(void)
597 log
= fls(num_online_cpus());
599 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
602 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
605 * Serialize vmap purging. There is no actual criticial section protected
606 * by this look, but we want to avoid concurrent calls for performance
607 * reasons and to make the pcpu_get_vm_areas more deterministic.
609 static DEFINE_SPINLOCK(vmap_purge_lock
);
611 /* for per-CPU blocks */
612 static void purge_fragmented_blocks_allcpus(void);
615 * called before a call to iounmap() if the caller wants vm_area_struct's
618 void set_iounmap_nonlazy(void)
620 atomic_set(&vmap_lazy_nr
, lazy_max_pages()+1);
624 * Purges all lazily-freed vmap areas.
626 static bool __purge_vmap_area_lazy(unsigned long start
, unsigned long end
)
628 struct llist_node
*valist
;
629 struct vmap_area
*va
;
630 struct vmap_area
*n_va
;
633 lockdep_assert_held(&vmap_purge_lock
);
635 valist
= llist_del_all(&vmap_purge_list
);
636 llist_for_each_entry(va
, valist
, purge_list
) {
637 if (va
->va_start
< start
)
638 start
= va
->va_start
;
639 if (va
->va_end
> end
)
641 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
647 atomic_sub(nr
, &vmap_lazy_nr
);
648 flush_tlb_kernel_range(start
, end
);
650 spin_lock(&vmap_area_lock
);
651 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
)
652 __free_vmap_area(va
);
653 spin_unlock(&vmap_area_lock
);
658 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
659 * is already purging.
661 static void try_purge_vmap_area_lazy(void)
663 if (spin_trylock(&vmap_purge_lock
)) {
664 __purge_vmap_area_lazy(ULONG_MAX
, 0);
665 spin_unlock(&vmap_purge_lock
);
670 * Kick off a purge of the outstanding lazy areas.
672 static void purge_vmap_area_lazy(void)
674 spin_lock(&vmap_purge_lock
);
675 purge_fragmented_blocks_allcpus();
676 __purge_vmap_area_lazy(ULONG_MAX
, 0);
677 spin_unlock(&vmap_purge_lock
);
681 * Free a vmap area, caller ensuring that the area has been unmapped
682 * and flush_cache_vunmap had been called for the correct range
685 static void free_vmap_area_noflush(struct vmap_area
*va
)
689 nr_lazy
= atomic_add_return((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
,
692 /* After this point, we may free va at any time */
693 llist_add(&va
->purge_list
, &vmap_purge_list
);
695 if (unlikely(nr_lazy
> lazy_max_pages()))
696 try_purge_vmap_area_lazy();
700 * Free and unmap a vmap area
702 static void free_unmap_vmap_area(struct vmap_area
*va
)
704 flush_cache_vunmap(va
->va_start
, va
->va_end
);
706 free_vmap_area_noflush(va
);
709 static struct vmap_area
*find_vmap_area(unsigned long addr
)
711 struct vmap_area
*va
;
713 spin_lock(&vmap_area_lock
);
714 va
= __find_vmap_area(addr
);
715 spin_unlock(&vmap_area_lock
);
720 /*** Per cpu kva allocator ***/
723 * vmap space is limited especially on 32 bit architectures. Ensure there is
724 * room for at least 16 percpu vmap blocks per CPU.
727 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
728 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
729 * instead (we just need a rough idea)
731 #if BITS_PER_LONG == 32
732 #define VMALLOC_SPACE (128UL*1024*1024)
734 #define VMALLOC_SPACE (128UL*1024*1024*1024)
737 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
738 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
739 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
740 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
741 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
742 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
743 #define VMAP_BBMAP_BITS \
744 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
745 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
746 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
748 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
750 static bool vmap_initialized __read_mostly
= false;
752 struct vmap_block_queue
{
754 struct list_head free
;
759 struct vmap_area
*va
;
760 unsigned long free
, dirty
;
761 unsigned long dirty_min
, dirty_max
; /*< dirty range */
762 struct list_head free_list
;
763 struct rcu_head rcu_head
;
764 struct list_head purge
;
767 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
768 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
771 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
772 * in the free path. Could get rid of this if we change the API to return a
773 * "cookie" from alloc, to be passed to free. But no big deal yet.
775 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
776 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
779 * We should probably have a fallback mechanism to allocate virtual memory
780 * out of partially filled vmap blocks. However vmap block sizing should be
781 * fairly reasonable according to the vmalloc size, so it shouldn't be a
785 static unsigned long addr_to_vb_idx(unsigned long addr
)
787 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
788 addr
/= VMAP_BLOCK_SIZE
;
792 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
796 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
797 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
802 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
803 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
804 * @order: how many 2^order pages should be occupied in newly allocated block
805 * @gfp_mask: flags for the page level allocator
807 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
809 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
811 struct vmap_block_queue
*vbq
;
812 struct vmap_block
*vb
;
813 struct vmap_area
*va
;
814 unsigned long vb_idx
;
818 node
= numa_node_id();
820 vb
= kmalloc_node(sizeof(struct vmap_block
),
821 gfp_mask
& GFP_RECLAIM_MASK
, node
);
823 return ERR_PTR(-ENOMEM
);
825 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
826 VMALLOC_START
, VMALLOC_END
,
833 err
= radix_tree_preload(gfp_mask
);
840 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
841 spin_lock_init(&vb
->lock
);
843 /* At least something should be left free */
844 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
845 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
847 vb
->dirty_min
= VMAP_BBMAP_BITS
;
849 INIT_LIST_HEAD(&vb
->free_list
);
851 vb_idx
= addr_to_vb_idx(va
->va_start
);
852 spin_lock(&vmap_block_tree_lock
);
853 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
854 spin_unlock(&vmap_block_tree_lock
);
856 radix_tree_preload_end();
858 vbq
= &get_cpu_var(vmap_block_queue
);
859 spin_lock(&vbq
->lock
);
860 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
861 spin_unlock(&vbq
->lock
);
862 put_cpu_var(vmap_block_queue
);
867 static void free_vmap_block(struct vmap_block
*vb
)
869 struct vmap_block
*tmp
;
870 unsigned long vb_idx
;
872 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
873 spin_lock(&vmap_block_tree_lock
);
874 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
875 spin_unlock(&vmap_block_tree_lock
);
878 free_vmap_area_noflush(vb
->va
);
879 kfree_rcu(vb
, rcu_head
);
882 static void purge_fragmented_blocks(int cpu
)
885 struct vmap_block
*vb
;
886 struct vmap_block
*n_vb
;
887 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
890 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
892 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
895 spin_lock(&vb
->lock
);
896 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
897 vb
->free
= 0; /* prevent further allocs after releasing lock */
898 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
900 vb
->dirty_max
= VMAP_BBMAP_BITS
;
901 spin_lock(&vbq
->lock
);
902 list_del_rcu(&vb
->free_list
);
903 spin_unlock(&vbq
->lock
);
904 spin_unlock(&vb
->lock
);
905 list_add_tail(&vb
->purge
, &purge
);
907 spin_unlock(&vb
->lock
);
911 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
912 list_del(&vb
->purge
);
917 static void purge_fragmented_blocks_allcpus(void)
921 for_each_possible_cpu(cpu
)
922 purge_fragmented_blocks(cpu
);
925 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
927 struct vmap_block_queue
*vbq
;
928 struct vmap_block
*vb
;
932 BUG_ON(offset_in_page(size
));
933 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
934 if (WARN_ON(size
== 0)) {
936 * Allocating 0 bytes isn't what caller wants since
937 * get_order(0) returns funny result. Just warn and terminate
942 order
= get_order(size
);
945 vbq
= &get_cpu_var(vmap_block_queue
);
946 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
947 unsigned long pages_off
;
949 spin_lock(&vb
->lock
);
950 if (vb
->free
< (1UL << order
)) {
951 spin_unlock(&vb
->lock
);
955 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
956 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
957 vb
->free
-= 1UL << order
;
959 spin_lock(&vbq
->lock
);
960 list_del_rcu(&vb
->free_list
);
961 spin_unlock(&vbq
->lock
);
964 spin_unlock(&vb
->lock
);
968 put_cpu_var(vmap_block_queue
);
971 /* Allocate new block if nothing was found */
973 vaddr
= new_vmap_block(order
, gfp_mask
);
978 static void vb_free(const void *addr
, unsigned long size
)
980 unsigned long offset
;
981 unsigned long vb_idx
;
983 struct vmap_block
*vb
;
985 BUG_ON(offset_in_page(size
));
986 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
988 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
990 order
= get_order(size
);
992 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
993 offset
>>= PAGE_SHIFT
;
995 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
997 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1001 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1003 spin_lock(&vb
->lock
);
1005 /* Expand dirty range */
1006 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1007 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1009 vb
->dirty
+= 1UL << order
;
1010 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1012 spin_unlock(&vb
->lock
);
1013 free_vmap_block(vb
);
1015 spin_unlock(&vb
->lock
);
1019 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1021 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1022 * to amortize TLB flushing overheads. What this means is that any page you
1023 * have now, may, in a former life, have been mapped into kernel virtual
1024 * address by the vmap layer and so there might be some CPUs with TLB entries
1025 * still referencing that page (additional to the regular 1:1 kernel mapping).
1027 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1028 * be sure that none of the pages we have control over will have any aliases
1029 * from the vmap layer.
1031 void vm_unmap_aliases(void)
1033 unsigned long start
= ULONG_MAX
, end
= 0;
1037 if (unlikely(!vmap_initialized
))
1042 for_each_possible_cpu(cpu
) {
1043 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1044 struct vmap_block
*vb
;
1047 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1048 spin_lock(&vb
->lock
);
1050 unsigned long va_start
= vb
->va
->va_start
;
1053 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1054 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1056 start
= min(s
, start
);
1061 spin_unlock(&vb
->lock
);
1066 spin_lock(&vmap_purge_lock
);
1067 purge_fragmented_blocks_allcpus();
1068 if (!__purge_vmap_area_lazy(start
, end
) && flush
)
1069 flush_tlb_kernel_range(start
, end
);
1070 spin_unlock(&vmap_purge_lock
);
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
= (unsigned long)count
<< PAGE_SHIFT
;
1082 unsigned long addr
= (unsigned long)mem
;
1083 struct vmap_area
*va
;
1087 BUG_ON(addr
< VMALLOC_START
);
1088 BUG_ON(addr
> VMALLOC_END
);
1089 BUG_ON(!PAGE_ALIGNED(addr
));
1091 debug_check_no_locks_freed(mem
, size
);
1092 vmap_debug_free_range(addr
, addr
+size
);
1094 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1099 va
= find_vmap_area(addr
);
1101 free_unmap_vmap_area(va
);
1103 EXPORT_SYMBOL(vm_unmap_ram
);
1106 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1107 * @pages: an array of pointers to the pages to be mapped
1108 * @count: number of pages
1109 * @node: prefer to allocate data structures on this node
1110 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1112 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1113 * faster than vmap so it's good. But if you mix long-life and short-life
1114 * objects with vm_map_ram(), it could consume lots of address space through
1115 * fragmentation (especially on a 32bit machine). You could see failures in
1116 * the end. Please use this function for short-lived objects.
1118 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1120 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1122 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1126 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1127 mem
= vb_alloc(size
, GFP_KERNEL
);
1130 addr
= (unsigned long)mem
;
1132 struct vmap_area
*va
;
1133 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1134 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1138 addr
= va
->va_start
;
1141 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1142 vm_unmap_ram(mem
, count
);
1147 EXPORT_SYMBOL(vm_map_ram
);
1149 static struct vm_struct
*vmlist __initdata
;
1151 * vm_area_add_early - add vmap area early during boot
1152 * @vm: vm_struct to add
1154 * This function is used to add fixed kernel vm area to vmlist before
1155 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1156 * should contain proper values and the other fields should be zero.
1158 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1160 void __init
vm_area_add_early(struct vm_struct
*vm
)
1162 struct vm_struct
*tmp
, **p
;
1164 BUG_ON(vmap_initialized
);
1165 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1166 if (tmp
->addr
>= vm
->addr
) {
1167 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1170 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1177 * vm_area_register_early - register vmap area early during boot
1178 * @vm: vm_struct to register
1179 * @align: requested alignment
1181 * This function is used to register kernel vm area before
1182 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1183 * proper values on entry and other fields should be zero. On return,
1184 * vm->addr contains the allocated address.
1186 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1188 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1190 static size_t vm_init_off __initdata
;
1193 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1194 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1196 vm
->addr
= (void *)addr
;
1198 vm_area_add_early(vm
);
1201 void __init
vmalloc_init(void)
1203 struct vmap_area
*va
;
1204 struct vm_struct
*tmp
;
1207 for_each_possible_cpu(i
) {
1208 struct vmap_block_queue
*vbq
;
1209 struct vfree_deferred
*p
;
1211 vbq
= &per_cpu(vmap_block_queue
, i
);
1212 spin_lock_init(&vbq
->lock
);
1213 INIT_LIST_HEAD(&vbq
->free
);
1214 p
= &per_cpu(vfree_deferred
, i
);
1215 init_llist_head(&p
->list
);
1216 INIT_WORK(&p
->wq
, free_work
);
1219 /* Import existing vmlist entries. */
1220 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1221 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1222 va
->flags
= VM_VM_AREA
;
1223 va
->va_start
= (unsigned long)tmp
->addr
;
1224 va
->va_end
= va
->va_start
+ tmp
->size
;
1226 __insert_vmap_area(va
);
1229 vmap_area_pcpu_hole
= VMALLOC_END
;
1231 vmap_initialized
= true;
1235 * map_kernel_range_noflush - map kernel VM area with the specified pages
1236 * @addr: start of the VM area to map
1237 * @size: size of the VM area to map
1238 * @prot: page protection flags to use
1239 * @pages: pages to map
1241 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1242 * specify should have been allocated using get_vm_area() and its
1246 * This function does NOT do any cache flushing. The caller is
1247 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1248 * before calling this function.
1251 * The number of pages mapped on success, -errno on failure.
1253 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1254 pgprot_t prot
, struct page
**pages
)
1256 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1260 * unmap_kernel_range_noflush - unmap kernel VM area
1261 * @addr: start of the VM area to unmap
1262 * @size: size of the VM area to unmap
1264 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1265 * specify should have been allocated using get_vm_area() and its
1269 * This function does NOT do any cache flushing. The caller is
1270 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1271 * before calling this function and flush_tlb_kernel_range() after.
1273 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1275 vunmap_page_range(addr
, addr
+ size
);
1277 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1280 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1281 * @addr: start of the VM area to unmap
1282 * @size: size of the VM area to unmap
1284 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1285 * the unmapping and tlb after.
1287 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1289 unsigned long end
= addr
+ size
;
1291 flush_cache_vunmap(addr
, end
);
1292 vunmap_page_range(addr
, end
);
1293 flush_tlb_kernel_range(addr
, end
);
1295 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
1297 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
1299 unsigned long addr
= (unsigned long)area
->addr
;
1300 unsigned long end
= addr
+ get_vm_area_size(area
);
1303 err
= vmap_page_range(addr
, end
, prot
, pages
);
1305 return err
> 0 ? 0 : err
;
1307 EXPORT_SYMBOL_GPL(map_vm_area
);
1309 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1310 unsigned long flags
, const void *caller
)
1312 spin_lock(&vmap_area_lock
);
1314 vm
->addr
= (void *)va
->va_start
;
1315 vm
->size
= va
->va_end
- va
->va_start
;
1316 vm
->caller
= caller
;
1318 va
->flags
|= VM_VM_AREA
;
1319 spin_unlock(&vmap_area_lock
);
1322 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1325 * Before removing VM_UNINITIALIZED,
1326 * we should make sure that vm has proper values.
1327 * Pair with smp_rmb() in show_numa_info().
1330 vm
->flags
&= ~VM_UNINITIALIZED
;
1333 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1334 unsigned long align
, unsigned long flags
, unsigned long start
,
1335 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1337 struct vmap_area
*va
;
1338 struct vm_struct
*area
;
1340 BUG_ON(in_interrupt());
1341 size
= PAGE_ALIGN(size
);
1342 if (unlikely(!size
))
1345 if (flags
& VM_IOREMAP
)
1346 align
= 1ul << clamp_t(int, get_count_order_long(size
),
1347 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1349 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1350 if (unlikely(!area
))
1353 if (!(flags
& VM_NO_GUARD
))
1356 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1362 setup_vmalloc_vm(area
, va
, flags
, caller
);
1367 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1368 unsigned long start
, unsigned long end
)
1370 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1371 GFP_KERNEL
, __builtin_return_address(0));
1373 EXPORT_SYMBOL_GPL(__get_vm_area
);
1375 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1376 unsigned long start
, unsigned long end
,
1379 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1380 GFP_KERNEL
, caller
);
1384 * get_vm_area - reserve a contiguous kernel virtual area
1385 * @size: size of the area
1386 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1388 * Search an area of @size in the kernel virtual mapping area,
1389 * and reserved it for out purposes. Returns the area descriptor
1390 * on success or %NULL on failure.
1392 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1394 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1395 NUMA_NO_NODE
, GFP_KERNEL
,
1396 __builtin_return_address(0));
1399 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1402 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1403 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1407 * find_vm_area - find a continuous kernel virtual area
1408 * @addr: base address
1410 * Search for the kernel VM area starting at @addr, and return it.
1411 * It is up to the caller to do all required locking to keep the returned
1414 struct vm_struct
*find_vm_area(const void *addr
)
1416 struct vmap_area
*va
;
1418 va
= find_vmap_area((unsigned long)addr
);
1419 if (va
&& va
->flags
& VM_VM_AREA
)
1426 * remove_vm_area - find and remove a continuous kernel virtual area
1427 * @addr: base address
1429 * Search for the kernel VM area starting at @addr, and remove it.
1430 * This function returns the found VM area, but using it is NOT safe
1431 * on SMP machines, except for its size or flags.
1433 struct vm_struct
*remove_vm_area(const void *addr
)
1435 struct vmap_area
*va
;
1439 va
= find_vmap_area((unsigned long)addr
);
1440 if (va
&& va
->flags
& VM_VM_AREA
) {
1441 struct vm_struct
*vm
= va
->vm
;
1443 spin_lock(&vmap_area_lock
);
1445 va
->flags
&= ~VM_VM_AREA
;
1446 spin_unlock(&vmap_area_lock
);
1448 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1449 kasan_free_shadow(vm
);
1450 free_unmap_vmap_area(va
);
1457 static void __vunmap(const void *addr
, int deallocate_pages
)
1459 struct vm_struct
*area
;
1464 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
1468 area
= remove_vm_area(addr
);
1469 if (unlikely(!area
)) {
1470 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1475 debug_check_no_locks_freed(addr
, get_vm_area_size(area
));
1476 debug_check_no_obj_freed(addr
, get_vm_area_size(area
));
1478 if (deallocate_pages
) {
1481 for (i
= 0; i
< area
->nr_pages
; i
++) {
1482 struct page
*page
= area
->pages
[i
];
1485 __free_pages(page
, 0);
1488 kvfree(area
->pages
);
1495 static inline void __vfree_deferred(const void *addr
)
1498 * Use raw_cpu_ptr() because this can be called from preemptible
1499 * context. Preemption is absolutely fine here, because the llist_add()
1500 * implementation is lockless, so it works even if we are adding to
1501 * nother cpu's list. schedule_work() should be fine with this too.
1503 struct vfree_deferred
*p
= raw_cpu_ptr(&vfree_deferred
);
1505 if (llist_add((struct llist_node
*)addr
, &p
->list
))
1506 schedule_work(&p
->wq
);
1510 * vfree_atomic - release memory allocated by vmalloc()
1511 * @addr: memory base address
1513 * This one is just like vfree() but can be called in any atomic context
1516 void vfree_atomic(const void *addr
)
1520 kmemleak_free(addr
);
1524 __vfree_deferred(addr
);
1528 * vfree - release memory allocated by vmalloc()
1529 * @addr: memory base address
1531 * Free the virtually continuous memory area starting at @addr, as
1532 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1533 * NULL, no operation is performed.
1535 * Must not be called in NMI context (strictly speaking, only if we don't
1536 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1537 * conventions for vfree() arch-depenedent would be a really bad idea)
1539 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1541 void vfree(const void *addr
)
1545 kmemleak_free(addr
);
1549 if (unlikely(in_interrupt()))
1550 __vfree_deferred(addr
);
1554 EXPORT_SYMBOL(vfree
);
1557 * vunmap - release virtual mapping obtained by vmap()
1558 * @addr: memory base address
1560 * Free the virtually contiguous memory area starting at @addr,
1561 * which was created from the page array passed to vmap().
1563 * Must not be called in interrupt context.
1565 void vunmap(const void *addr
)
1567 BUG_ON(in_interrupt());
1572 EXPORT_SYMBOL(vunmap
);
1575 * vmap - map an array of pages into virtually contiguous space
1576 * @pages: array of page pointers
1577 * @count: number of pages to map
1578 * @flags: vm_area->flags
1579 * @prot: page protection for the mapping
1581 * Maps @count pages from @pages into contiguous kernel virtual
1584 void *vmap(struct page
**pages
, unsigned int count
,
1585 unsigned long flags
, pgprot_t prot
)
1587 struct vm_struct
*area
;
1588 unsigned long size
; /* In bytes */
1592 if (count
> totalram_pages
)
1595 size
= (unsigned long)count
<< PAGE_SHIFT
;
1596 area
= get_vm_area_caller(size
, flags
, __builtin_return_address(0));
1600 if (map_vm_area(area
, prot
, pages
)) {
1607 EXPORT_SYMBOL(vmap
);
1609 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1610 gfp_t gfp_mask
, pgprot_t prot
,
1611 int node
, const void *caller
);
1612 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1613 pgprot_t prot
, int node
)
1615 struct page
**pages
;
1616 unsigned int nr_pages
, array_size
, i
;
1617 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1618 const gfp_t alloc_mask
= gfp_mask
| __GFP_NOWARN
;
1620 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
1621 array_size
= (nr_pages
* sizeof(struct page
*));
1623 area
->nr_pages
= nr_pages
;
1624 /* Please note that the recursion is strictly bounded. */
1625 if (array_size
> PAGE_SIZE
) {
1626 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1627 PAGE_KERNEL
, node
, area
->caller
);
1629 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1631 area
->pages
= pages
;
1633 remove_vm_area(area
->addr
);
1638 for (i
= 0; i
< area
->nr_pages
; i
++) {
1641 if (node
== NUMA_NO_NODE
)
1642 page
= alloc_page(alloc_mask
);
1644 page
= alloc_pages_node(node
, alloc_mask
, 0);
1646 if (unlikely(!page
)) {
1647 /* Successfully allocated i pages, free them in __vunmap() */
1651 area
->pages
[i
] = page
;
1652 if (gfpflags_allow_blocking(gfp_mask
))
1656 if (map_vm_area(area
, prot
, pages
))
1661 warn_alloc(gfp_mask
,
1662 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1663 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1669 * __vmalloc_node_range - allocate virtually contiguous memory
1670 * @size: allocation size
1671 * @align: desired alignment
1672 * @start: vm area range start
1673 * @end: vm area range end
1674 * @gfp_mask: flags for the page level allocator
1675 * @prot: protection mask for the allocated pages
1676 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
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 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1685 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1686 pgprot_t prot
, unsigned long vm_flags
, int node
,
1689 struct vm_struct
*area
;
1691 unsigned long real_size
= size
;
1693 size
= PAGE_ALIGN(size
);
1694 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1697 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
1698 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
1702 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
1707 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1708 * flag. It means that vm_struct is not fully initialized.
1709 * Now, it is fully initialized, so remove this flag here.
1711 clear_vm_uninitialized_flag(area
);
1714 * A ref_count = 2 is needed because vm_struct allocated in
1715 * __get_vm_area_node() contains a reference to the virtual address of
1716 * the vmalloc'ed block.
1718 kmemleak_alloc(addr
, real_size
, 2, gfp_mask
);
1723 warn_alloc(gfp_mask
,
1724 "vmalloc: allocation failure: %lu bytes", real_size
);
1729 * __vmalloc_node - allocate virtually contiguous memory
1730 * @size: allocation size
1731 * @align: desired alignment
1732 * @gfp_mask: flags for the page level allocator
1733 * @prot: protection mask for the allocated pages
1734 * @node: node to use for allocation or NUMA_NO_NODE
1735 * @caller: caller's return address
1737 * Allocate enough pages to cover @size from the page level
1738 * allocator with @gfp_mask flags. Map them into contiguous
1739 * kernel virtual space, using a pagetable protection of @prot.
1741 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1742 gfp_t gfp_mask
, pgprot_t prot
,
1743 int node
, const void *caller
)
1745 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1746 gfp_mask
, prot
, 0, node
, caller
);
1749 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1751 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1752 __builtin_return_address(0));
1754 EXPORT_SYMBOL(__vmalloc
);
1756 static inline void *__vmalloc_node_flags(unsigned long size
,
1757 int node
, gfp_t flags
)
1759 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
1760 node
, __builtin_return_address(0));
1764 * vmalloc - allocate virtually contiguous memory
1765 * @size: allocation size
1766 * Allocate enough pages to cover @size from the page level
1767 * allocator and map them into contiguous kernel virtual space.
1769 * For tight control over page level allocator and protection flags
1770 * use __vmalloc() instead.
1772 void *vmalloc(unsigned long size
)
1774 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1775 GFP_KERNEL
| __GFP_HIGHMEM
);
1777 EXPORT_SYMBOL(vmalloc
);
1780 * vzalloc - allocate virtually contiguous memory with zero fill
1781 * @size: allocation size
1782 * Allocate enough pages to cover @size from the page level
1783 * allocator and map them into contiguous kernel virtual space.
1784 * The memory allocated is set to zero.
1786 * For tight control over page level allocator and protection flags
1787 * use __vmalloc() instead.
1789 void *vzalloc(unsigned long size
)
1791 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1792 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1794 EXPORT_SYMBOL(vzalloc
);
1797 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1798 * @size: allocation size
1800 * The resulting memory area is zeroed so it can be mapped to userspace
1801 * without leaking data.
1803 void *vmalloc_user(unsigned long size
)
1805 struct vm_struct
*area
;
1808 ret
= __vmalloc_node(size
, SHMLBA
,
1809 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1810 PAGE_KERNEL
, NUMA_NO_NODE
,
1811 __builtin_return_address(0));
1813 area
= find_vm_area(ret
);
1814 area
->flags
|= VM_USERMAP
;
1818 EXPORT_SYMBOL(vmalloc_user
);
1821 * vmalloc_node - allocate memory on a specific node
1822 * @size: allocation size
1825 * Allocate enough pages to cover @size from the page level
1826 * allocator and map them into contiguous kernel virtual space.
1828 * For tight control over page level allocator and protection flags
1829 * use __vmalloc() instead.
1831 void *vmalloc_node(unsigned long size
, int node
)
1833 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1834 node
, __builtin_return_address(0));
1836 EXPORT_SYMBOL(vmalloc_node
);
1839 * vzalloc_node - allocate memory on a specific node with zero fill
1840 * @size: allocation size
1843 * Allocate enough pages to cover @size from the page level
1844 * allocator and map them into contiguous kernel virtual space.
1845 * The memory allocated is set to zero.
1847 * For tight control over page level allocator and protection flags
1848 * use __vmalloc_node() instead.
1850 void *vzalloc_node(unsigned long size
, int node
)
1852 return __vmalloc_node_flags(size
, node
,
1853 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1855 EXPORT_SYMBOL(vzalloc_node
);
1857 #ifndef PAGE_KERNEL_EXEC
1858 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1862 * vmalloc_exec - allocate virtually contiguous, executable memory
1863 * @size: allocation size
1865 * Kernel-internal function to allocate enough pages to cover @size
1866 * the page level allocator and map them into contiguous and
1867 * executable kernel virtual space.
1869 * For tight control over page level allocator and protection flags
1870 * use __vmalloc() instead.
1873 void *vmalloc_exec(unsigned long size
)
1875 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1876 NUMA_NO_NODE
, __builtin_return_address(0));
1879 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1880 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1881 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1882 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1884 #define GFP_VMALLOC32 GFP_KERNEL
1888 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1889 * @size: allocation size
1891 * Allocate enough 32bit PA addressable pages to cover @size from the
1892 * page level allocator and map them into contiguous kernel virtual space.
1894 void *vmalloc_32(unsigned long size
)
1896 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1897 NUMA_NO_NODE
, __builtin_return_address(0));
1899 EXPORT_SYMBOL(vmalloc_32
);
1902 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1903 * @size: allocation size
1905 * The resulting memory area is 32bit addressable and zeroed so it can be
1906 * mapped to userspace without leaking data.
1908 void *vmalloc_32_user(unsigned long size
)
1910 struct vm_struct
*area
;
1913 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1914 NUMA_NO_NODE
, __builtin_return_address(0));
1916 area
= find_vm_area(ret
);
1917 area
->flags
|= VM_USERMAP
;
1921 EXPORT_SYMBOL(vmalloc_32_user
);
1924 * small helper routine , copy contents to buf from addr.
1925 * If the page is not present, fill zero.
1928 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1934 unsigned long offset
, length
;
1936 offset
= offset_in_page(addr
);
1937 length
= PAGE_SIZE
- offset
;
1940 p
= vmalloc_to_page(addr
);
1942 * To do safe access to this _mapped_ area, we need
1943 * lock. But adding lock here means that we need to add
1944 * overhead of vmalloc()/vfree() calles for this _debug_
1945 * interface, rarely used. Instead of that, we'll use
1946 * kmap() and get small overhead in this access function.
1950 * we can expect USER0 is not used (see vread/vwrite's
1951 * function description)
1953 void *map
= kmap_atomic(p
);
1954 memcpy(buf
, map
+ offset
, length
);
1957 memset(buf
, 0, length
);
1967 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
1973 unsigned long offset
, length
;
1975 offset
= offset_in_page(addr
);
1976 length
= PAGE_SIZE
- offset
;
1979 p
= vmalloc_to_page(addr
);
1981 * To do safe access to this _mapped_ area, we need
1982 * lock. But adding lock here means that we need to add
1983 * overhead of vmalloc()/vfree() calles for this _debug_
1984 * interface, rarely used. Instead of that, we'll use
1985 * kmap() and get small overhead in this access function.
1989 * we can expect USER0 is not used (see vread/vwrite's
1990 * function description)
1992 void *map
= kmap_atomic(p
);
1993 memcpy(map
+ offset
, buf
, length
);
2005 * vread() - read vmalloc area in a safe way.
2006 * @buf: buffer for reading data
2007 * @addr: vm address.
2008 * @count: number of bytes to be read.
2010 * Returns # of bytes which addr and buf should be increased.
2011 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2012 * includes any intersect with alive vmalloc area.
2014 * This function checks that addr is a valid vmalloc'ed area, and
2015 * copy data from that area to a given buffer. If the given memory range
2016 * of [addr...addr+count) includes some valid address, data is copied to
2017 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2018 * IOREMAP area is treated as memory hole and no copy is done.
2020 * If [addr...addr+count) doesn't includes any intersects with alive
2021 * vm_struct area, returns 0. @buf should be kernel's buffer.
2023 * Note: In usual ops, vread() is never necessary because the caller
2024 * should know vmalloc() area is valid and can use memcpy().
2025 * This is for routines which have to access vmalloc area without
2026 * any informaion, as /dev/kmem.
2030 long vread(char *buf
, char *addr
, unsigned long count
)
2032 struct vmap_area
*va
;
2033 struct vm_struct
*vm
;
2034 char *vaddr
, *buf_start
= buf
;
2035 unsigned long buflen
= count
;
2038 /* Don't allow overflow */
2039 if ((unsigned long) addr
+ count
< count
)
2040 count
= -(unsigned long) addr
;
2042 spin_lock(&vmap_area_lock
);
2043 list_for_each_entry(va
, &vmap_area_list
, list
) {
2047 if (!(va
->flags
& VM_VM_AREA
))
2051 vaddr
= (char *) vm
->addr
;
2052 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2054 while (addr
< vaddr
) {
2062 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2065 if (!(vm
->flags
& VM_IOREMAP
))
2066 aligned_vread(buf
, addr
, n
);
2067 else /* IOREMAP area is treated as memory hole */
2074 spin_unlock(&vmap_area_lock
);
2076 if (buf
== buf_start
)
2078 /* zero-fill memory holes */
2079 if (buf
!= buf_start
+ buflen
)
2080 memset(buf
, 0, buflen
- (buf
- buf_start
));
2086 * vwrite() - write vmalloc area in a safe way.
2087 * @buf: buffer for source data
2088 * @addr: vm address.
2089 * @count: number of bytes to be read.
2091 * Returns # of bytes which addr and buf should be incresed.
2092 * (same number to @count).
2093 * If [addr...addr+count) doesn't includes any intersect with valid
2094 * vmalloc area, returns 0.
2096 * This function checks that addr is a valid vmalloc'ed area, and
2097 * copy data from a buffer to the given addr. If specified range of
2098 * [addr...addr+count) includes some valid address, data is copied from
2099 * proper area of @buf. If there are memory holes, no copy to hole.
2100 * IOREMAP area is treated as memory hole and no copy is done.
2102 * If [addr...addr+count) doesn't includes any intersects with alive
2103 * vm_struct area, returns 0. @buf should be kernel's buffer.
2105 * Note: In usual ops, vwrite() is never necessary because the caller
2106 * should know vmalloc() area is valid and can use memcpy().
2107 * This is for routines which have to access vmalloc area without
2108 * any informaion, as /dev/kmem.
2111 long vwrite(char *buf
, char *addr
, unsigned long count
)
2113 struct vmap_area
*va
;
2114 struct vm_struct
*vm
;
2116 unsigned long n
, buflen
;
2119 /* Don't allow overflow */
2120 if ((unsigned long) addr
+ count
< count
)
2121 count
= -(unsigned long) addr
;
2124 spin_lock(&vmap_area_lock
);
2125 list_for_each_entry(va
, &vmap_area_list
, list
) {
2129 if (!(va
->flags
& VM_VM_AREA
))
2133 vaddr
= (char *) vm
->addr
;
2134 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2136 while (addr
< vaddr
) {
2143 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2146 if (!(vm
->flags
& VM_IOREMAP
)) {
2147 aligned_vwrite(buf
, addr
, n
);
2155 spin_unlock(&vmap_area_lock
);
2162 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2163 * @vma: vma to cover
2164 * @uaddr: target user address to start at
2165 * @kaddr: virtual address of vmalloc kernel memory
2166 * @size: size of map area
2168 * Returns: 0 for success, -Exxx on failure
2170 * This function checks that @kaddr is a valid vmalloc'ed area,
2171 * and that it is big enough to cover the range starting at
2172 * @uaddr in @vma. Will return failure if that criteria isn't
2175 * Similar to remap_pfn_range() (see mm/memory.c)
2177 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2178 void *kaddr
, unsigned long size
)
2180 struct vm_struct
*area
;
2182 size
= PAGE_ALIGN(size
);
2184 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2187 area
= find_vm_area(kaddr
);
2191 if (!(area
->flags
& VM_USERMAP
))
2194 if (kaddr
+ size
> area
->addr
+ area
->size
)
2198 struct page
*page
= vmalloc_to_page(kaddr
);
2201 ret
= vm_insert_page(vma
, uaddr
, page
);
2210 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2214 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2217 * remap_vmalloc_range - map vmalloc pages to userspace
2218 * @vma: vma to cover (map full range of vma)
2219 * @addr: vmalloc memory
2220 * @pgoff: number of pages into addr before first page to map
2222 * Returns: 0 for success, -Exxx on failure
2224 * This function checks that addr is a valid vmalloc'ed area, and
2225 * that it is big enough to cover the vma. Will return failure if
2226 * that criteria isn't met.
2228 * Similar to remap_pfn_range() (see mm/memory.c)
2230 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2231 unsigned long pgoff
)
2233 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2234 addr
+ (pgoff
<< PAGE_SHIFT
),
2235 vma
->vm_end
- vma
->vm_start
);
2237 EXPORT_SYMBOL(remap_vmalloc_range
);
2240 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2243 void __weak
vmalloc_sync_all(void)
2248 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2260 * alloc_vm_area - allocate a range of kernel address space
2261 * @size: size of the area
2262 * @ptes: returns the PTEs for the address space
2264 * Returns: NULL on failure, vm_struct on success
2266 * This function reserves a range of kernel address space, and
2267 * allocates pagetables to map that range. No actual mappings
2270 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2271 * allocated for the VM area are returned.
2273 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2275 struct vm_struct
*area
;
2277 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2278 __builtin_return_address(0));
2283 * This ensures that page tables are constructed for this region
2284 * of kernel virtual address space and mapped into init_mm.
2286 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2287 size
, f
, ptes
? &ptes
: NULL
)) {
2294 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2296 void free_vm_area(struct vm_struct
*area
)
2298 struct vm_struct
*ret
;
2299 ret
= remove_vm_area(area
->addr
);
2300 BUG_ON(ret
!= area
);
2303 EXPORT_SYMBOL_GPL(free_vm_area
);
2306 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2308 return n
? rb_entry(n
, struct vmap_area
, rb_node
) : NULL
;
2312 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2313 * @end: target address
2314 * @pnext: out arg for the next vmap_area
2315 * @pprev: out arg for the previous vmap_area
2317 * Returns: %true if either or both of next and prev are found,
2318 * %false if no vmap_area exists
2320 * Find vmap_areas end addresses of which enclose @end. ie. if not
2321 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2323 static bool pvm_find_next_prev(unsigned long end
,
2324 struct vmap_area
**pnext
,
2325 struct vmap_area
**pprev
)
2327 struct rb_node
*n
= vmap_area_root
.rb_node
;
2328 struct vmap_area
*va
= NULL
;
2331 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2332 if (end
< va
->va_end
)
2334 else if (end
> va
->va_end
)
2343 if (va
->va_end
> end
) {
2345 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2348 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2354 * pvm_determine_end - find the highest aligned address between two vmap_areas
2355 * @pnext: in/out arg for the next vmap_area
2356 * @pprev: in/out arg for the previous vmap_area
2359 * Returns: determined end address
2361 * Find the highest aligned address between *@pnext and *@pprev below
2362 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2363 * down address is between the end addresses of the two vmap_areas.
2365 * Please note that the address returned by this function may fall
2366 * inside *@pnext vmap_area. The caller is responsible for checking
2369 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2370 struct vmap_area
**pprev
,
2371 unsigned long align
)
2373 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2377 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2381 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2383 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2390 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2391 * @offsets: array containing offset of each area
2392 * @sizes: array containing size of each area
2393 * @nr_vms: the number of areas to allocate
2394 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2396 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2397 * vm_structs on success, %NULL on failure
2399 * Percpu allocator wants to use congruent vm areas so that it can
2400 * maintain the offsets among percpu areas. This function allocates
2401 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2402 * be scattered pretty far, distance between two areas easily going up
2403 * to gigabytes. To avoid interacting with regular vmallocs, these
2404 * areas are allocated from top.
2406 * Despite its complicated look, this allocator is rather simple. It
2407 * does everything top-down and scans areas from the end looking for
2408 * matching slot. While scanning, if any of the areas overlaps with
2409 * existing vmap_area, the base address is pulled down to fit the
2410 * area. Scanning is repeated till all the areas fit and then all
2411 * necessary data structres are inserted and the result is returned.
2413 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2414 const size_t *sizes
, int nr_vms
,
2417 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2418 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2419 struct vmap_area
**vas
, *prev
, *next
;
2420 struct vm_struct
**vms
;
2421 int area
, area2
, last_area
, term_area
;
2422 unsigned long base
, start
, end
, last_end
;
2423 bool purged
= false;
2425 /* verify parameters and allocate data structures */
2426 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
2427 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2428 start
= offsets
[area
];
2429 end
= start
+ sizes
[area
];
2431 /* is everything aligned properly? */
2432 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2433 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2435 /* detect the area with the highest address */
2436 if (start
> offsets
[last_area
])
2439 for (area2
= 0; area2
< nr_vms
; area2
++) {
2440 unsigned long start2
= offsets
[area2
];
2441 unsigned long end2
= start2
+ sizes
[area2
];
2446 BUG_ON(start2
>= start
&& start2
< end
);
2447 BUG_ON(end2
<= end
&& end2
> start
);
2450 last_end
= offsets
[last_area
] + sizes
[last_area
];
2452 if (vmalloc_end
- vmalloc_start
< last_end
) {
2457 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2458 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2462 for (area
= 0; area
< nr_vms
; area
++) {
2463 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2464 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2465 if (!vas
[area
] || !vms
[area
])
2469 spin_lock(&vmap_area_lock
);
2471 /* start scanning - we scan from the top, begin with the last area */
2472 area
= term_area
= last_area
;
2473 start
= offsets
[area
];
2474 end
= start
+ sizes
[area
];
2476 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2477 base
= vmalloc_end
- last_end
;
2480 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2483 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2484 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2487 * base might have underflowed, add last_end before
2490 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2491 spin_unlock(&vmap_area_lock
);
2493 purge_vmap_area_lazy();
2501 * If next overlaps, move base downwards so that it's
2502 * right below next and then recheck.
2504 if (next
&& next
->va_start
< base
+ end
) {
2505 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2511 * If prev overlaps, shift down next and prev and move
2512 * base so that it's right below new next and then
2515 if (prev
&& prev
->va_end
> base
+ start
) {
2517 prev
= node_to_va(rb_prev(&next
->rb_node
));
2518 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2524 * This area fits, move on to the previous one. If
2525 * the previous one is the terminal one, we're done.
2527 area
= (area
+ nr_vms
- 1) % nr_vms
;
2528 if (area
== term_area
)
2530 start
= offsets
[area
];
2531 end
= start
+ sizes
[area
];
2532 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2535 /* we've found a fitting base, insert all va's */
2536 for (area
= 0; area
< nr_vms
; area
++) {
2537 struct vmap_area
*va
= vas
[area
];
2539 va
->va_start
= base
+ offsets
[area
];
2540 va
->va_end
= va
->va_start
+ sizes
[area
];
2541 __insert_vmap_area(va
);
2544 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2546 spin_unlock(&vmap_area_lock
);
2548 /* insert all vm's */
2549 for (area
= 0; area
< nr_vms
; area
++)
2550 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2557 for (area
= 0; area
< nr_vms
; area
++) {
2568 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2569 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2570 * @nr_vms: the number of allocated areas
2572 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2574 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2578 for (i
= 0; i
< nr_vms
; i
++)
2579 free_vm_area(vms
[i
]);
2582 #endif /* CONFIG_SMP */
2584 #ifdef CONFIG_PROC_FS
2585 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2586 __acquires(&vmap_area_lock
)
2588 spin_lock(&vmap_area_lock
);
2589 return seq_list_start(&vmap_area_list
, *pos
);
2592 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2594 return seq_list_next(p
, &vmap_area_list
, pos
);
2597 static void s_stop(struct seq_file
*m
, void *p
)
2598 __releases(&vmap_area_lock
)
2600 spin_unlock(&vmap_area_lock
);
2603 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2605 if (IS_ENABLED(CONFIG_NUMA
)) {
2606 unsigned int nr
, *counters
= m
->private;
2611 if (v
->flags
& VM_UNINITIALIZED
)
2613 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2616 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2618 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2619 counters
[page_to_nid(v
->pages
[nr
])]++;
2621 for_each_node_state(nr
, N_HIGH_MEMORY
)
2623 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2627 static int s_show(struct seq_file
*m
, void *p
)
2629 struct vmap_area
*va
;
2630 struct vm_struct
*v
;
2632 va
= list_entry(p
, struct vmap_area
, list
);
2635 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2636 * behalf of vmap area is being tear down or vm_map_ram allocation.
2638 if (!(va
->flags
& VM_VM_AREA
))
2643 seq_printf(m
, "0x%pK-0x%pK %7ld",
2644 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2647 seq_printf(m
, " %pS", v
->caller
);
2650 seq_printf(m
, " pages=%d", v
->nr_pages
);
2653 seq_printf(m
, " phys=%llx", (unsigned long long)v
->phys_addr
);
2655 if (v
->flags
& VM_IOREMAP
)
2656 seq_puts(m
, " ioremap");
2658 if (v
->flags
& VM_ALLOC
)
2659 seq_puts(m
, " vmalloc");
2661 if (v
->flags
& VM_MAP
)
2662 seq_puts(m
, " vmap");
2664 if (v
->flags
& VM_USERMAP
)
2665 seq_puts(m
, " user");
2667 if (is_vmalloc_addr(v
->pages
))
2668 seq_puts(m
, " vpages");
2670 show_numa_info(m
, v
);
2675 static const struct seq_operations vmalloc_op
= {
2682 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2684 if (IS_ENABLED(CONFIG_NUMA
))
2685 return seq_open_private(file
, &vmalloc_op
,
2686 nr_node_ids
* sizeof(unsigned int));
2688 return seq_open(file
, &vmalloc_op
);
2691 static const struct file_operations proc_vmalloc_operations
= {
2692 .open
= vmalloc_open
,
2694 .llseek
= seq_lseek
,
2695 .release
= seq_release_private
,
2698 static int __init
proc_vmalloc_init(void)
2700 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2703 module_init(proc_vmalloc_init
);