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 <asm/uaccess.h>
31 #include <asm/tlbflush.h>
32 #include <asm/shmparam.h>
34 /*** Page table manipulation functions ***/
36 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
40 pte
= pte_offset_kernel(pmd
, addr
);
42 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
43 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
44 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
47 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
52 pmd
= pmd_offset(pud
, addr
);
54 next
= pmd_addr_end(addr
, end
);
55 if (pmd_none_or_clear_bad(pmd
))
57 vunmap_pte_range(pmd
, addr
, next
);
58 } while (pmd
++, addr
= next
, addr
!= end
);
61 static void vunmap_pud_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
66 pud
= pud_offset(pgd
, addr
);
68 next
= pud_addr_end(addr
, end
);
69 if (pud_none_or_clear_bad(pud
))
71 vunmap_pmd_range(pud
, addr
, next
);
72 } while (pud
++, addr
= next
, addr
!= end
);
75 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
81 pgd
= pgd_offset_k(addr
);
83 next
= pgd_addr_end(addr
, end
);
84 if (pgd_none_or_clear_bad(pgd
))
86 vunmap_pud_range(pgd
, addr
, next
);
87 } while (pgd
++, addr
= next
, addr
!= end
);
90 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
91 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
100 pte
= pte_alloc_kernel(pmd
, addr
);
104 struct page
*page
= pages
[*nr
];
106 if (WARN_ON(!pte_none(*pte
)))
110 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
112 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
116 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
117 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
122 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
126 next
= pmd_addr_end(addr
, end
);
127 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
129 } while (pmd
++, addr
= next
, addr
!= end
);
133 static int vmap_pud_range(pgd_t
*pgd
, unsigned long addr
,
134 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
139 pud
= pud_alloc(&init_mm
, pgd
, addr
);
143 next
= pud_addr_end(addr
, end
);
144 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
146 } while (pud
++, addr
= next
, addr
!= end
);
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
156 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
157 pgprot_t prot
, struct page
**pages
)
161 unsigned long addr
= start
;
166 pgd
= pgd_offset_k(addr
);
168 next
= pgd_addr_end(addr
, end
);
169 err
= vmap_pud_range(pgd
, addr
, next
, prot
, pages
, &nr
);
172 } while (pgd
++, addr
= next
, addr
!= end
);
177 static int vmap_page_range(unsigned long start
, unsigned long end
,
178 pgprot_t prot
, struct page
**pages
)
182 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
183 flush_cache_vmap(start
, end
);
187 int is_vmalloc_or_module_addr(const void *x
)
190 * ARM, x86-64 and sparc64 put modules in a special place,
191 * and fall back on vmalloc() if that fails. Others
192 * just put it in the vmalloc space.
194 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
195 unsigned long addr
= (unsigned long)x
;
196 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
199 return is_vmalloc_addr(x
);
203 * Walk a vmap address to the struct page it maps.
205 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
207 unsigned long addr
= (unsigned long) vmalloc_addr
;
208 struct page
*page
= NULL
;
209 pgd_t
*pgd
= pgd_offset_k(addr
);
212 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
213 * architectures that do not vmalloc module space
215 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
217 if (!pgd_none(*pgd
)) {
218 pud_t
*pud
= pud_offset(pgd
, addr
);
219 if (!pud_none(*pud
)) {
220 pmd_t
*pmd
= pmd_offset(pud
, addr
);
221 if (!pmd_none(*pmd
)) {
224 ptep
= pte_offset_map(pmd
, addr
);
226 if (pte_present(pte
))
227 page
= pte_page(pte
);
234 EXPORT_SYMBOL(vmalloc_to_page
);
237 * Map a vmalloc()-space virtual address to the physical page frame number.
239 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
241 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
243 EXPORT_SYMBOL(vmalloc_to_pfn
);
246 /*** Global kva allocator ***/
248 #define VM_LAZY_FREE 0x01
249 #define VM_LAZY_FREEING 0x02
250 #define VM_VM_AREA 0x04
253 unsigned long va_start
;
254 unsigned long va_end
;
256 struct rb_node rb_node
; /* address sorted rbtree */
257 struct list_head list
; /* address sorted list */
258 struct list_head purge_list
; /* "lazy purge" list */
259 struct vm_struct
*vm
;
260 struct rcu_head rcu_head
;
263 static DEFINE_SPINLOCK(vmap_area_lock
);
264 static LIST_HEAD(vmap_area_list
);
265 static struct rb_root vmap_area_root
= RB_ROOT
;
267 /* The vmap cache globals are protected by vmap_area_lock */
268 static struct rb_node
*free_vmap_cache
;
269 static unsigned long cached_hole_size
;
270 static unsigned long cached_vstart
;
271 static unsigned long cached_align
;
273 static unsigned long vmap_area_pcpu_hole
;
275 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
277 struct rb_node
*n
= vmap_area_root
.rb_node
;
280 struct vmap_area
*va
;
282 va
= rb_entry(n
, struct vmap_area
, rb_node
);
283 if (addr
< va
->va_start
)
285 else if (addr
> va
->va_start
)
294 static void __insert_vmap_area(struct vmap_area
*va
)
296 struct rb_node
**p
= &vmap_area_root
.rb_node
;
297 struct rb_node
*parent
= NULL
;
301 struct vmap_area
*tmp_va
;
304 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
305 if (va
->va_start
< tmp_va
->va_end
)
307 else if (va
->va_end
> tmp_va
->va_start
)
313 rb_link_node(&va
->rb_node
, parent
, p
);
314 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
316 /* address-sort this list so it is usable like the vmlist */
317 tmp
= rb_prev(&va
->rb_node
);
319 struct vmap_area
*prev
;
320 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
321 list_add_rcu(&va
->list
, &prev
->list
);
323 list_add_rcu(&va
->list
, &vmap_area_list
);
326 static void purge_vmap_area_lazy(void);
329 * Allocate a region of KVA of the specified size and alignment, within the
332 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
334 unsigned long vstart
, unsigned long vend
,
335 int node
, gfp_t gfp_mask
)
337 struct vmap_area
*va
;
341 struct vmap_area
*first
;
344 BUG_ON(size
& ~PAGE_MASK
);
345 BUG_ON(!is_power_of_2(align
));
347 va
= kmalloc_node(sizeof(struct vmap_area
),
348 gfp_mask
& GFP_RECLAIM_MASK
, node
);
350 return ERR_PTR(-ENOMEM
);
353 spin_lock(&vmap_area_lock
);
355 * Invalidate cache if we have more permissive parameters.
356 * cached_hole_size notes the largest hole noticed _below_
357 * the vmap_area cached in free_vmap_cache: if size fits
358 * into that hole, we want to scan from vstart to reuse
359 * the hole instead of allocating above free_vmap_cache.
360 * Note that __free_vmap_area may update free_vmap_cache
361 * without updating cached_hole_size or cached_align.
363 if (!free_vmap_cache
||
364 size
< cached_hole_size
||
365 vstart
< cached_vstart
||
366 align
< cached_align
) {
368 cached_hole_size
= 0;
369 free_vmap_cache
= NULL
;
371 /* record if we encounter less permissive parameters */
372 cached_vstart
= vstart
;
373 cached_align
= align
;
375 /* find starting point for our search */
376 if (free_vmap_cache
) {
377 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
378 addr
= ALIGN(first
->va_end
, align
);
381 if (addr
+ size
- 1 < addr
)
385 addr
= ALIGN(vstart
, align
);
386 if (addr
+ size
- 1 < addr
)
389 n
= vmap_area_root
.rb_node
;
393 struct vmap_area
*tmp
;
394 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
395 if (tmp
->va_end
>= addr
) {
397 if (tmp
->va_start
<= addr
)
408 /* from the starting point, walk areas until a suitable hole is found */
409 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
410 if (addr
+ cached_hole_size
< first
->va_start
)
411 cached_hole_size
= first
->va_start
- addr
;
412 addr
= ALIGN(first
->va_end
, align
);
413 if (addr
+ size
- 1 < addr
)
416 if (list_is_last(&first
->list
, &vmap_area_list
))
419 first
= list_entry(first
->list
.next
,
420 struct vmap_area
, list
);
424 if (addr
+ size
> vend
)
428 va
->va_end
= addr
+ size
;
430 __insert_vmap_area(va
);
431 free_vmap_cache
= &va
->rb_node
;
432 spin_unlock(&vmap_area_lock
);
434 BUG_ON(va
->va_start
& (align
-1));
435 BUG_ON(va
->va_start
< vstart
);
436 BUG_ON(va
->va_end
> vend
);
441 spin_unlock(&vmap_area_lock
);
443 purge_vmap_area_lazy();
447 if (printk_ratelimit())
449 "vmap allocation for size %lu failed: "
450 "use vmalloc=<size> to increase size.\n", size
);
452 return ERR_PTR(-EBUSY
);
455 static void __free_vmap_area(struct vmap_area
*va
)
457 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
459 if (free_vmap_cache
) {
460 if (va
->va_end
< cached_vstart
) {
461 free_vmap_cache
= NULL
;
463 struct vmap_area
*cache
;
464 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
465 if (va
->va_start
<= cache
->va_start
) {
466 free_vmap_cache
= rb_prev(&va
->rb_node
);
468 * We don't try to update cached_hole_size or
469 * cached_align, but it won't go very wrong.
474 rb_erase(&va
->rb_node
, &vmap_area_root
);
475 RB_CLEAR_NODE(&va
->rb_node
);
476 list_del_rcu(&va
->list
);
479 * Track the highest possible candidate for pcpu area
480 * allocation. Areas outside of vmalloc area can be returned
481 * here too, consider only end addresses which fall inside
482 * vmalloc area proper.
484 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
485 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
487 kfree_rcu(va
, rcu_head
);
491 * Free a region of KVA allocated by alloc_vmap_area
493 static void free_vmap_area(struct vmap_area
*va
)
495 spin_lock(&vmap_area_lock
);
496 __free_vmap_area(va
);
497 spin_unlock(&vmap_area_lock
);
501 * Clear the pagetable entries of a given vmap_area
503 static void unmap_vmap_area(struct vmap_area
*va
)
505 vunmap_page_range(va
->va_start
, va
->va_end
);
508 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
511 * Unmap page tables and force a TLB flush immediately if
512 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
513 * bugs similarly to those in linear kernel virtual address
514 * space after a page has been freed.
516 * All the lazy freeing logic is still retained, in order to
517 * minimise intrusiveness of this debugging feature.
519 * This is going to be *slow* (linear kernel virtual address
520 * debugging doesn't do a broadcast TLB flush so it is a lot
523 #ifdef CONFIG_DEBUG_PAGEALLOC
524 vunmap_page_range(start
, end
);
525 flush_tlb_kernel_range(start
, end
);
530 * lazy_max_pages is the maximum amount of virtual address space we gather up
531 * before attempting to purge with a TLB flush.
533 * There is a tradeoff here: a larger number will cover more kernel page tables
534 * and take slightly longer to purge, but it will linearly reduce the number of
535 * global TLB flushes that must be performed. It would seem natural to scale
536 * this number up linearly with the number of CPUs (because vmapping activity
537 * could also scale linearly with the number of CPUs), however it is likely
538 * that in practice, workloads might be constrained in other ways that mean
539 * vmap activity will not scale linearly with CPUs. Also, I want to be
540 * conservative and not introduce a big latency on huge systems, so go with
541 * a less aggressive log scale. It will still be an improvement over the old
542 * code, and it will be simple to change the scale factor if we find that it
543 * becomes a problem on bigger systems.
545 static unsigned long lazy_max_pages(void)
549 log
= fls(num_online_cpus());
551 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
554 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
556 /* for per-CPU blocks */
557 static void purge_fragmented_blocks_allcpus(void);
560 * called before a call to iounmap() if the caller wants vm_area_struct's
563 void set_iounmap_nonlazy(void)
565 atomic_set(&vmap_lazy_nr
, lazy_max_pages()+1);
569 * Purges all lazily-freed vmap areas.
571 * If sync is 0 then don't purge if there is already a purge in progress.
572 * If force_flush is 1, then flush kernel TLBs between *start and *end even
573 * if we found no lazy vmap areas to unmap (callers can use this to optimise
574 * their own TLB flushing).
575 * Returns with *start = min(*start, lowest purged address)
576 * *end = max(*end, highest purged address)
578 static void __purge_vmap_area_lazy(unsigned long *start
, unsigned long *end
,
579 int sync
, int force_flush
)
581 static DEFINE_SPINLOCK(purge_lock
);
583 struct vmap_area
*va
;
584 struct vmap_area
*n_va
;
588 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
589 * should not expect such behaviour. This just simplifies locking for
590 * the case that isn't actually used at the moment anyway.
592 if (!sync
&& !force_flush
) {
593 if (!spin_trylock(&purge_lock
))
596 spin_lock(&purge_lock
);
599 purge_fragmented_blocks_allcpus();
602 list_for_each_entry_rcu(va
, &vmap_area_list
, list
) {
603 if (va
->flags
& VM_LAZY_FREE
) {
604 if (va
->va_start
< *start
)
605 *start
= va
->va_start
;
606 if (va
->va_end
> *end
)
608 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
609 list_add_tail(&va
->purge_list
, &valist
);
610 va
->flags
|= VM_LAZY_FREEING
;
611 va
->flags
&= ~VM_LAZY_FREE
;
617 atomic_sub(nr
, &vmap_lazy_nr
);
619 if (nr
|| force_flush
)
620 flush_tlb_kernel_range(*start
, *end
);
623 spin_lock(&vmap_area_lock
);
624 list_for_each_entry_safe(va
, n_va
, &valist
, purge_list
)
625 __free_vmap_area(va
);
626 spin_unlock(&vmap_area_lock
);
628 spin_unlock(&purge_lock
);
632 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
633 * is already purging.
635 static void try_purge_vmap_area_lazy(void)
637 unsigned long start
= ULONG_MAX
, end
= 0;
639 __purge_vmap_area_lazy(&start
, &end
, 0, 0);
643 * Kick off a purge of the outstanding lazy areas.
645 static void purge_vmap_area_lazy(void)
647 unsigned long start
= ULONG_MAX
, end
= 0;
649 __purge_vmap_area_lazy(&start
, &end
, 1, 0);
653 * Free a vmap area, caller ensuring that the area has been unmapped
654 * and flush_cache_vunmap had been called for the correct range
657 static void free_vmap_area_noflush(struct vmap_area
*va
)
659 va
->flags
|= VM_LAZY_FREE
;
660 atomic_add((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
, &vmap_lazy_nr
);
661 if (unlikely(atomic_read(&vmap_lazy_nr
) > lazy_max_pages()))
662 try_purge_vmap_area_lazy();
666 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
667 * called for the correct range previously.
669 static void free_unmap_vmap_area_noflush(struct vmap_area
*va
)
672 free_vmap_area_noflush(va
);
676 * Free and unmap a vmap area
678 static void free_unmap_vmap_area(struct vmap_area
*va
)
680 flush_cache_vunmap(va
->va_start
, va
->va_end
);
681 free_unmap_vmap_area_noflush(va
);
684 static struct vmap_area
*find_vmap_area(unsigned long addr
)
686 struct vmap_area
*va
;
688 spin_lock(&vmap_area_lock
);
689 va
= __find_vmap_area(addr
);
690 spin_unlock(&vmap_area_lock
);
695 static void free_unmap_vmap_area_addr(unsigned long addr
)
697 struct vmap_area
*va
;
699 va
= find_vmap_area(addr
);
701 free_unmap_vmap_area(va
);
705 /*** Per cpu kva allocator ***/
708 * vmap space is limited especially on 32 bit architectures. Ensure there is
709 * room for at least 16 percpu vmap blocks per CPU.
712 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
713 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
714 * instead (we just need a rough idea)
716 #if BITS_PER_LONG == 32
717 #define VMALLOC_SPACE (128UL*1024*1024)
719 #define VMALLOC_SPACE (128UL*1024*1024*1024)
722 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
723 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
724 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
725 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
726 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
727 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
728 #define VMAP_BBMAP_BITS \
729 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
730 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
731 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
733 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
735 static bool vmap_initialized __read_mostly
= false;
737 struct vmap_block_queue
{
739 struct list_head free
;
744 struct vmap_area
*va
;
745 struct vmap_block_queue
*vbq
;
746 unsigned long free
, dirty
;
747 DECLARE_BITMAP(alloc_map
, VMAP_BBMAP_BITS
);
748 DECLARE_BITMAP(dirty_map
, VMAP_BBMAP_BITS
);
749 struct list_head free_list
;
750 struct rcu_head rcu_head
;
751 struct list_head purge
;
754 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
755 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
758 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
759 * in the free path. Could get rid of this if we change the API to return a
760 * "cookie" from alloc, to be passed to free. But no big deal yet.
762 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
763 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
766 * We should probably have a fallback mechanism to allocate virtual memory
767 * out of partially filled vmap blocks. However vmap block sizing should be
768 * fairly reasonable according to the vmalloc size, so it shouldn't be a
772 static unsigned long addr_to_vb_idx(unsigned long addr
)
774 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
775 addr
/= VMAP_BLOCK_SIZE
;
779 static struct vmap_block
*new_vmap_block(gfp_t gfp_mask
)
781 struct vmap_block_queue
*vbq
;
782 struct vmap_block
*vb
;
783 struct vmap_area
*va
;
784 unsigned long vb_idx
;
787 node
= numa_node_id();
789 vb
= kmalloc_node(sizeof(struct vmap_block
),
790 gfp_mask
& GFP_RECLAIM_MASK
, node
);
792 return ERR_PTR(-ENOMEM
);
794 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
795 VMALLOC_START
, VMALLOC_END
,
802 err
= radix_tree_preload(gfp_mask
);
809 spin_lock_init(&vb
->lock
);
811 vb
->free
= VMAP_BBMAP_BITS
;
813 bitmap_zero(vb
->alloc_map
, VMAP_BBMAP_BITS
);
814 bitmap_zero(vb
->dirty_map
, VMAP_BBMAP_BITS
);
815 INIT_LIST_HEAD(&vb
->free_list
);
817 vb_idx
= addr_to_vb_idx(va
->va_start
);
818 spin_lock(&vmap_block_tree_lock
);
819 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
820 spin_unlock(&vmap_block_tree_lock
);
822 radix_tree_preload_end();
824 vbq
= &get_cpu_var(vmap_block_queue
);
826 spin_lock(&vbq
->lock
);
827 list_add_rcu(&vb
->free_list
, &vbq
->free
);
828 spin_unlock(&vbq
->lock
);
829 put_cpu_var(vmap_block_queue
);
834 static void free_vmap_block(struct vmap_block
*vb
)
836 struct vmap_block
*tmp
;
837 unsigned long vb_idx
;
839 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
840 spin_lock(&vmap_block_tree_lock
);
841 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
842 spin_unlock(&vmap_block_tree_lock
);
845 free_vmap_area_noflush(vb
->va
);
846 kfree_rcu(vb
, rcu_head
);
849 static void purge_fragmented_blocks(int cpu
)
852 struct vmap_block
*vb
;
853 struct vmap_block
*n_vb
;
854 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
857 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
859 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
862 spin_lock(&vb
->lock
);
863 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
864 vb
->free
= 0; /* prevent further allocs after releasing lock */
865 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
866 bitmap_fill(vb
->alloc_map
, VMAP_BBMAP_BITS
);
867 bitmap_fill(vb
->dirty_map
, VMAP_BBMAP_BITS
);
868 spin_lock(&vbq
->lock
);
869 list_del_rcu(&vb
->free_list
);
870 spin_unlock(&vbq
->lock
);
871 spin_unlock(&vb
->lock
);
872 list_add_tail(&vb
->purge
, &purge
);
874 spin_unlock(&vb
->lock
);
878 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
879 list_del(&vb
->purge
);
884 static void purge_fragmented_blocks_thiscpu(void)
886 purge_fragmented_blocks(smp_processor_id());
889 static void purge_fragmented_blocks_allcpus(void)
893 for_each_possible_cpu(cpu
)
894 purge_fragmented_blocks(cpu
);
897 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
899 struct vmap_block_queue
*vbq
;
900 struct vmap_block
*vb
;
901 unsigned long addr
= 0;
905 BUG_ON(size
& ~PAGE_MASK
);
906 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
907 if (WARN_ON(size
== 0)) {
909 * Allocating 0 bytes isn't what caller wants since
910 * get_order(0) returns funny result. Just warn and terminate
915 order
= get_order(size
);
919 vbq
= &get_cpu_var(vmap_block_queue
);
920 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
923 spin_lock(&vb
->lock
);
924 if (vb
->free
< 1UL << order
)
927 i
= bitmap_find_free_region(vb
->alloc_map
,
928 VMAP_BBMAP_BITS
, order
);
931 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
) {
932 /* fragmented and no outstanding allocations */
933 BUG_ON(vb
->dirty
!= VMAP_BBMAP_BITS
);
938 addr
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
939 BUG_ON(addr_to_vb_idx(addr
) !=
940 addr_to_vb_idx(vb
->va
->va_start
));
941 vb
->free
-= 1UL << order
;
943 spin_lock(&vbq
->lock
);
944 list_del_rcu(&vb
->free_list
);
945 spin_unlock(&vbq
->lock
);
947 spin_unlock(&vb
->lock
);
950 spin_unlock(&vb
->lock
);
954 purge_fragmented_blocks_thiscpu();
956 put_cpu_var(vmap_block_queue
);
960 vb
= new_vmap_block(gfp_mask
);
969 static void vb_free(const void *addr
, unsigned long size
)
971 unsigned long offset
;
972 unsigned long vb_idx
;
974 struct vmap_block
*vb
;
976 BUG_ON(size
& ~PAGE_MASK
);
977 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
979 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
981 order
= get_order(size
);
983 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
985 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
987 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
991 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
993 spin_lock(&vb
->lock
);
994 BUG_ON(bitmap_allocate_region(vb
->dirty_map
, offset
>> PAGE_SHIFT
, order
));
996 vb
->dirty
+= 1UL << order
;
997 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
999 spin_unlock(&vb
->lock
);
1000 free_vmap_block(vb
);
1002 spin_unlock(&vb
->lock
);
1006 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1008 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1009 * to amortize TLB flushing overheads. What this means is that any page you
1010 * have now, may, in a former life, have been mapped into kernel virtual
1011 * address by the vmap layer and so there might be some CPUs with TLB entries
1012 * still referencing that page (additional to the regular 1:1 kernel mapping).
1014 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1015 * be sure that none of the pages we have control over will have any aliases
1016 * from the vmap layer.
1018 void vm_unmap_aliases(void)
1020 unsigned long start
= ULONG_MAX
, end
= 0;
1024 if (unlikely(!vmap_initialized
))
1027 for_each_possible_cpu(cpu
) {
1028 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1029 struct vmap_block
*vb
;
1032 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1035 spin_lock(&vb
->lock
);
1036 i
= find_first_bit(vb
->dirty_map
, VMAP_BBMAP_BITS
);
1037 while (i
< VMAP_BBMAP_BITS
) {
1040 j
= find_next_zero_bit(vb
->dirty_map
,
1041 VMAP_BBMAP_BITS
, i
);
1043 s
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
1044 e
= vb
->va
->va_start
+ (j
<< PAGE_SHIFT
);
1053 i
= find_next_bit(vb
->dirty_map
,
1054 VMAP_BBMAP_BITS
, i
);
1056 spin_unlock(&vb
->lock
);
1061 __purge_vmap_area_lazy(&start
, &end
, 1, flush
);
1063 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1066 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1067 * @mem: the pointer returned by vm_map_ram
1068 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1070 void vm_unmap_ram(const void *mem
, unsigned int count
)
1072 unsigned long size
= count
<< PAGE_SHIFT
;
1073 unsigned long addr
= (unsigned long)mem
;
1076 BUG_ON(addr
< VMALLOC_START
);
1077 BUG_ON(addr
> VMALLOC_END
);
1078 BUG_ON(addr
& (PAGE_SIZE
-1));
1080 debug_check_no_locks_freed(mem
, size
);
1081 vmap_debug_free_range(addr
, addr
+size
);
1083 if (likely(count
<= VMAP_MAX_ALLOC
))
1086 free_unmap_vmap_area_addr(addr
);
1088 EXPORT_SYMBOL(vm_unmap_ram
);
1091 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1092 * @pages: an array of pointers to the pages to be mapped
1093 * @count: number of pages
1094 * @node: prefer to allocate data structures on this node
1095 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1097 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1099 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1101 unsigned long size
= count
<< PAGE_SHIFT
;
1105 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1106 mem
= vb_alloc(size
, GFP_KERNEL
);
1109 addr
= (unsigned long)mem
;
1111 struct vmap_area
*va
;
1112 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1113 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1117 addr
= va
->va_start
;
1120 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1121 vm_unmap_ram(mem
, count
);
1126 EXPORT_SYMBOL(vm_map_ram
);
1129 * vm_area_add_early - add vmap area early during boot
1130 * @vm: vm_struct to add
1132 * This function is used to add fixed kernel vm area to vmlist before
1133 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1134 * should contain proper values and the other fields should be zero.
1136 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1138 void __init
vm_area_add_early(struct vm_struct
*vm
)
1140 struct vm_struct
*tmp
, **p
;
1142 BUG_ON(vmap_initialized
);
1143 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1144 if (tmp
->addr
>= vm
->addr
) {
1145 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1148 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1155 * vm_area_register_early - register vmap area early during boot
1156 * @vm: vm_struct to register
1157 * @align: requested alignment
1159 * This function is used to register kernel vm area before
1160 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1161 * proper values on entry and other fields should be zero. On return,
1162 * vm->addr contains the allocated address.
1164 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1166 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1168 static size_t vm_init_off __initdata
;
1171 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1172 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1174 vm
->addr
= (void *)addr
;
1176 vm_area_add_early(vm
);
1179 void __init
vmalloc_init(void)
1181 struct vmap_area
*va
;
1182 struct vm_struct
*tmp
;
1185 for_each_possible_cpu(i
) {
1186 struct vmap_block_queue
*vbq
;
1188 vbq
= &per_cpu(vmap_block_queue
, i
);
1189 spin_lock_init(&vbq
->lock
);
1190 INIT_LIST_HEAD(&vbq
->free
);
1193 /* Import existing vmlist entries. */
1194 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1195 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1196 va
->flags
= VM_VM_AREA
;
1197 va
->va_start
= (unsigned long)tmp
->addr
;
1198 va
->va_end
= va
->va_start
+ tmp
->size
;
1200 __insert_vmap_area(va
);
1203 vmap_area_pcpu_hole
= VMALLOC_END
;
1205 vmap_initialized
= true;
1209 * map_kernel_range_noflush - map kernel VM area with the specified pages
1210 * @addr: start of the VM area to map
1211 * @size: size of the VM area to map
1212 * @prot: page protection flags to use
1213 * @pages: pages to map
1215 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1216 * specify should have been allocated using get_vm_area() and its
1220 * This function does NOT do any cache flushing. The caller is
1221 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1222 * before calling this function.
1225 * The number of pages mapped on success, -errno on failure.
1227 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1228 pgprot_t prot
, struct page
**pages
)
1230 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1234 * unmap_kernel_range_noflush - unmap kernel VM area
1235 * @addr: start of the VM area to unmap
1236 * @size: size of the VM area to unmap
1238 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1239 * specify should have been allocated using get_vm_area() and its
1243 * This function does NOT do any cache flushing. The caller is
1244 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1245 * before calling this function and flush_tlb_kernel_range() after.
1247 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1249 vunmap_page_range(addr
, addr
+ size
);
1251 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1254 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1255 * @addr: start of the VM area to unmap
1256 * @size: size of the VM area to unmap
1258 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1259 * the unmapping and tlb after.
1261 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1263 unsigned long end
= addr
+ size
;
1265 flush_cache_vunmap(addr
, end
);
1266 vunmap_page_range(addr
, end
);
1267 flush_tlb_kernel_range(addr
, end
);
1270 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
***pages
)
1272 unsigned long addr
= (unsigned long)area
->addr
;
1273 unsigned long end
= addr
+ area
->size
- PAGE_SIZE
;
1276 err
= vmap_page_range(addr
, end
, prot
, *pages
);
1284 EXPORT_SYMBOL_GPL(map_vm_area
);
1286 /*** Old vmalloc interfaces ***/
1287 DEFINE_RWLOCK(vmlist_lock
);
1288 struct vm_struct
*vmlist
;
1290 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1291 unsigned long flags
, const void *caller
)
1294 vm
->addr
= (void *)va
->va_start
;
1295 vm
->size
= va
->va_end
- va
->va_start
;
1296 vm
->caller
= caller
;
1298 va
->flags
|= VM_VM_AREA
;
1301 static void insert_vmalloc_vmlist(struct vm_struct
*vm
)
1303 struct vm_struct
*tmp
, **p
;
1305 vm
->flags
&= ~VM_UNLIST
;
1306 write_lock(&vmlist_lock
);
1307 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1308 if (tmp
->addr
>= vm
->addr
)
1313 write_unlock(&vmlist_lock
);
1316 static void insert_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1317 unsigned long flags
, const void *caller
)
1319 setup_vmalloc_vm(vm
, va
, flags
, caller
);
1320 insert_vmalloc_vmlist(vm
);
1323 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1324 unsigned long align
, unsigned long flags
, unsigned long start
,
1325 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1327 struct vmap_area
*va
;
1328 struct vm_struct
*area
;
1330 BUG_ON(in_interrupt());
1331 if (flags
& VM_IOREMAP
) {
1332 int bit
= fls(size
);
1334 if (bit
> IOREMAP_MAX_ORDER
)
1335 bit
= IOREMAP_MAX_ORDER
;
1336 else if (bit
< PAGE_SHIFT
)
1342 size
= PAGE_ALIGN(size
);
1343 if (unlikely(!size
))
1346 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1347 if (unlikely(!area
))
1351 * We always allocate a guard page.
1355 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1362 * When this function is called from __vmalloc_node_range,
1363 * we do not add vm_struct to vmlist here to avoid
1364 * accessing uninitialized members of vm_struct such as
1365 * pages and nr_pages fields. They will be set later.
1366 * To distinguish it from others, we use a VM_UNLIST flag.
1368 if (flags
& VM_UNLIST
)
1369 setup_vmalloc_vm(area
, va
, flags
, caller
);
1371 insert_vmalloc_vm(area
, va
, flags
, caller
);
1376 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1377 unsigned long start
, unsigned long end
)
1379 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1380 GFP_KERNEL
, __builtin_return_address(0));
1382 EXPORT_SYMBOL_GPL(__get_vm_area
);
1384 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1385 unsigned long start
, unsigned long end
,
1388 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1389 GFP_KERNEL
, caller
);
1393 * get_vm_area - reserve a contiguous kernel virtual area
1394 * @size: size of the area
1395 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1397 * Search an area of @size in the kernel virtual mapping area,
1398 * and reserved it for out purposes. Returns the area descriptor
1399 * on success or %NULL on failure.
1401 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1403 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1404 NUMA_NO_NODE
, GFP_KERNEL
,
1405 __builtin_return_address(0));
1408 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1411 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1412 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1416 * find_vm_area - find a continuous kernel virtual area
1417 * @addr: base address
1419 * Search for the kernel VM area starting at @addr, and return it.
1420 * It is up to the caller to do all required locking to keep the returned
1423 struct vm_struct
*find_vm_area(const void *addr
)
1425 struct vmap_area
*va
;
1427 va
= find_vmap_area((unsigned long)addr
);
1428 if (va
&& va
->flags
& VM_VM_AREA
)
1435 * remove_vm_area - find and remove a continuous kernel virtual area
1436 * @addr: base address
1438 * Search for the kernel VM area starting at @addr, and remove it.
1439 * This function returns the found VM area, but using it is NOT safe
1440 * on SMP machines, except for its size or flags.
1442 struct vm_struct
*remove_vm_area(const void *addr
)
1444 struct vmap_area
*va
;
1446 va
= find_vmap_area((unsigned long)addr
);
1447 if (va
&& va
->flags
& VM_VM_AREA
) {
1448 struct vm_struct
*vm
= va
->vm
;
1450 if (!(vm
->flags
& VM_UNLIST
)) {
1451 struct vm_struct
*tmp
, **p
;
1453 * remove from list and disallow access to
1454 * this vm_struct before unmap. (address range
1455 * confliction is maintained by vmap.)
1457 write_lock(&vmlist_lock
);
1458 for (p
= &vmlist
; (tmp
= *p
) != vm
; p
= &tmp
->next
)
1461 write_unlock(&vmlist_lock
);
1464 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1465 free_unmap_vmap_area(va
);
1466 vm
->size
-= PAGE_SIZE
;
1473 static void __vunmap(const void *addr
, int deallocate_pages
)
1475 struct vm_struct
*area
;
1480 if ((PAGE_SIZE
-1) & (unsigned long)addr
) {
1481 WARN(1, KERN_ERR
"Trying to vfree() bad address (%p)\n", addr
);
1485 area
= remove_vm_area(addr
);
1486 if (unlikely(!area
)) {
1487 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1492 debug_check_no_locks_freed(addr
, area
->size
);
1493 debug_check_no_obj_freed(addr
, area
->size
);
1495 if (deallocate_pages
) {
1498 for (i
= 0; i
< area
->nr_pages
; i
++) {
1499 struct page
*page
= area
->pages
[i
];
1505 if (area
->flags
& VM_VPAGES
)
1516 * vfree - release memory allocated by vmalloc()
1517 * @addr: memory base address
1519 * Free the virtually continuous memory area starting at @addr, as
1520 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1521 * NULL, no operation is performed.
1523 * Must not be called in interrupt context.
1525 void vfree(const void *addr
)
1527 BUG_ON(in_interrupt());
1529 kmemleak_free(addr
);
1533 EXPORT_SYMBOL(vfree
);
1536 * vunmap - release virtual mapping obtained by vmap()
1537 * @addr: memory base address
1539 * Free the virtually contiguous memory area starting at @addr,
1540 * which was created from the page array passed to vmap().
1542 * Must not be called in interrupt context.
1544 void vunmap(const void *addr
)
1546 BUG_ON(in_interrupt());
1550 EXPORT_SYMBOL(vunmap
);
1553 * vmap - map an array of pages into virtually contiguous space
1554 * @pages: array of page pointers
1555 * @count: number of pages to map
1556 * @flags: vm_area->flags
1557 * @prot: page protection for the mapping
1559 * Maps @count pages from @pages into contiguous kernel virtual
1562 void *vmap(struct page
**pages
, unsigned int count
,
1563 unsigned long flags
, pgprot_t prot
)
1565 struct vm_struct
*area
;
1569 if (count
> totalram_pages
)
1572 area
= get_vm_area_caller((count
<< PAGE_SHIFT
), flags
,
1573 __builtin_return_address(0));
1577 if (map_vm_area(area
, prot
, &pages
)) {
1584 EXPORT_SYMBOL(vmap
);
1586 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1587 gfp_t gfp_mask
, pgprot_t prot
,
1588 int node
, const void *caller
);
1589 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1590 pgprot_t prot
, int node
, const void *caller
)
1592 const int order
= 0;
1593 struct page
**pages
;
1594 unsigned int nr_pages
, array_size
, i
;
1595 gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1597 nr_pages
= (area
->size
- PAGE_SIZE
) >> PAGE_SHIFT
;
1598 array_size
= (nr_pages
* sizeof(struct page
*));
1600 area
->nr_pages
= nr_pages
;
1601 /* Please note that the recursion is strictly bounded. */
1602 if (array_size
> PAGE_SIZE
) {
1603 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1604 PAGE_KERNEL
, node
, caller
);
1605 area
->flags
|= VM_VPAGES
;
1607 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1609 area
->pages
= pages
;
1610 area
->caller
= caller
;
1612 remove_vm_area(area
->addr
);
1617 for (i
= 0; i
< area
->nr_pages
; i
++) {
1619 gfp_t tmp_mask
= gfp_mask
| __GFP_NOWARN
;
1622 page
= alloc_page(tmp_mask
);
1624 page
= alloc_pages_node(node
, tmp_mask
, order
);
1626 if (unlikely(!page
)) {
1627 /* Successfully allocated i pages, free them in __vunmap() */
1631 area
->pages
[i
] = page
;
1634 if (map_vm_area(area
, prot
, &pages
))
1639 warn_alloc_failed(gfp_mask
, order
,
1640 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1641 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1647 * __vmalloc_node_range - allocate virtually contiguous memory
1648 * @size: allocation size
1649 * @align: desired alignment
1650 * @start: vm area range start
1651 * @end: vm area range end
1652 * @gfp_mask: flags for the page level allocator
1653 * @prot: protection mask for the allocated pages
1654 * @node: node to use for allocation or NUMA_NO_NODE
1655 * @caller: caller's return address
1657 * Allocate enough pages to cover @size from the page level
1658 * allocator with @gfp_mask flags. Map them into contiguous
1659 * kernel virtual space, using a pagetable protection of @prot.
1661 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1662 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1663 pgprot_t prot
, int node
, const void *caller
)
1665 struct vm_struct
*area
;
1667 unsigned long real_size
= size
;
1669 size
= PAGE_ALIGN(size
);
1670 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1673 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNLIST
,
1674 start
, end
, node
, gfp_mask
, caller
);
1678 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
, caller
);
1683 * In this function, newly allocated vm_struct is not added
1684 * to vmlist at __get_vm_area_node(). so, it is added here.
1686 insert_vmalloc_vmlist(area
);
1689 * A ref_count = 3 is needed because the vm_struct and vmap_area
1690 * structures allocated in the __get_vm_area_node() function contain
1691 * references to the virtual address of the vmalloc'ed block.
1693 kmemleak_alloc(addr
, real_size
, 3, gfp_mask
);
1698 warn_alloc_failed(gfp_mask
, 0,
1699 "vmalloc: allocation failure: %lu bytes\n",
1705 * __vmalloc_node - allocate virtually contiguous memory
1706 * @size: allocation size
1707 * @align: desired alignment
1708 * @gfp_mask: flags for the page level allocator
1709 * @prot: protection mask for the allocated pages
1710 * @node: node to use for allocation or NUMA_NO_NODE
1711 * @caller: caller's return address
1713 * Allocate enough pages to cover @size from the page level
1714 * allocator with @gfp_mask flags. Map them into contiguous
1715 * kernel virtual space, using a pagetable protection of @prot.
1717 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1718 gfp_t gfp_mask
, pgprot_t prot
,
1719 int node
, const void *caller
)
1721 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1722 gfp_mask
, prot
, node
, caller
);
1725 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1727 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1728 __builtin_return_address(0));
1730 EXPORT_SYMBOL(__vmalloc
);
1732 static inline void *__vmalloc_node_flags(unsigned long size
,
1733 int node
, gfp_t flags
)
1735 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
1736 node
, __builtin_return_address(0));
1740 * vmalloc - allocate virtually contiguous memory
1741 * @size: allocation size
1742 * Allocate enough pages to cover @size from the page level
1743 * allocator and map them into contiguous kernel virtual space.
1745 * For tight control over page level allocator and protection flags
1746 * use __vmalloc() instead.
1748 void *vmalloc(unsigned long size
)
1750 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1751 GFP_KERNEL
| __GFP_HIGHMEM
);
1753 EXPORT_SYMBOL(vmalloc
);
1756 * vzalloc - allocate virtually contiguous memory with zero fill
1757 * @size: allocation size
1758 * Allocate enough pages to cover @size from the page level
1759 * allocator and map them into contiguous kernel virtual space.
1760 * The memory allocated is set to zero.
1762 * For tight control over page level allocator and protection flags
1763 * use __vmalloc() instead.
1765 void *vzalloc(unsigned long size
)
1767 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1768 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1770 EXPORT_SYMBOL(vzalloc
);
1773 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1774 * @size: allocation size
1776 * The resulting memory area is zeroed so it can be mapped to userspace
1777 * without leaking data.
1779 void *vmalloc_user(unsigned long size
)
1781 struct vm_struct
*area
;
1784 ret
= __vmalloc_node(size
, SHMLBA
,
1785 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1786 PAGE_KERNEL
, NUMA_NO_NODE
,
1787 __builtin_return_address(0));
1789 area
= find_vm_area(ret
);
1790 area
->flags
|= VM_USERMAP
;
1794 EXPORT_SYMBOL(vmalloc_user
);
1797 * vmalloc_node - allocate memory on a specific node
1798 * @size: allocation size
1801 * Allocate enough pages to cover @size from the page level
1802 * allocator and map them into contiguous kernel virtual space.
1804 * For tight control over page level allocator and protection flags
1805 * use __vmalloc() instead.
1807 void *vmalloc_node(unsigned long size
, int node
)
1809 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1810 node
, __builtin_return_address(0));
1812 EXPORT_SYMBOL(vmalloc_node
);
1815 * vzalloc_node - allocate memory on a specific node with zero fill
1816 * @size: allocation size
1819 * Allocate enough pages to cover @size from the page level
1820 * allocator and map them into contiguous kernel virtual space.
1821 * The memory allocated is set to zero.
1823 * For tight control over page level allocator and protection flags
1824 * use __vmalloc_node() instead.
1826 void *vzalloc_node(unsigned long size
, int node
)
1828 return __vmalloc_node_flags(size
, node
,
1829 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
1831 EXPORT_SYMBOL(vzalloc_node
);
1833 #ifndef PAGE_KERNEL_EXEC
1834 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1838 * vmalloc_exec - allocate virtually contiguous, executable memory
1839 * @size: allocation size
1841 * Kernel-internal function to allocate enough pages to cover @size
1842 * the page level allocator and map them into contiguous and
1843 * executable kernel virtual space.
1845 * For tight control over page level allocator and protection flags
1846 * use __vmalloc() instead.
1849 void *vmalloc_exec(unsigned long size
)
1851 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1852 NUMA_NO_NODE
, __builtin_return_address(0));
1855 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1856 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1857 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1858 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1860 #define GFP_VMALLOC32 GFP_KERNEL
1864 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1865 * @size: allocation size
1867 * Allocate enough 32bit PA addressable pages to cover @size from the
1868 * page level allocator and map them into contiguous kernel virtual space.
1870 void *vmalloc_32(unsigned long size
)
1872 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1873 NUMA_NO_NODE
, __builtin_return_address(0));
1875 EXPORT_SYMBOL(vmalloc_32
);
1878 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1879 * @size: allocation size
1881 * The resulting memory area is 32bit addressable and zeroed so it can be
1882 * mapped to userspace without leaking data.
1884 void *vmalloc_32_user(unsigned long size
)
1886 struct vm_struct
*area
;
1889 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1890 NUMA_NO_NODE
, __builtin_return_address(0));
1892 area
= find_vm_area(ret
);
1893 area
->flags
|= VM_USERMAP
;
1897 EXPORT_SYMBOL(vmalloc_32_user
);
1900 * small helper routine , copy contents to buf from addr.
1901 * If the page is not present, fill zero.
1904 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1910 unsigned long offset
, length
;
1912 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1913 length
= PAGE_SIZE
- offset
;
1916 p
= vmalloc_to_page(addr
);
1918 * To do safe access to this _mapped_ area, we need
1919 * lock. But adding lock here means that we need to add
1920 * overhead of vmalloc()/vfree() calles for this _debug_
1921 * interface, rarely used. Instead of that, we'll use
1922 * kmap() and get small overhead in this access function.
1926 * we can expect USER0 is not used (see vread/vwrite's
1927 * function description)
1929 void *map
= kmap_atomic(p
);
1930 memcpy(buf
, map
+ offset
, length
);
1933 memset(buf
, 0, length
);
1943 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
1949 unsigned long offset
, length
;
1951 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1952 length
= PAGE_SIZE
- offset
;
1955 p
= vmalloc_to_page(addr
);
1957 * To do safe access to this _mapped_ area, we need
1958 * lock. But adding lock here means that we need to add
1959 * overhead of vmalloc()/vfree() calles for this _debug_
1960 * interface, rarely used. Instead of that, we'll use
1961 * kmap() and get small overhead in this access function.
1965 * we can expect USER0 is not used (see vread/vwrite's
1966 * function description)
1968 void *map
= kmap_atomic(p
);
1969 memcpy(map
+ offset
, buf
, length
);
1981 * vread() - read vmalloc area in a safe way.
1982 * @buf: buffer for reading data
1983 * @addr: vm address.
1984 * @count: number of bytes to be read.
1986 * Returns # of bytes which addr and buf should be increased.
1987 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1988 * includes any intersect with alive vmalloc area.
1990 * This function checks that addr is a valid vmalloc'ed area, and
1991 * copy data from that area to a given buffer. If the given memory range
1992 * of [addr...addr+count) includes some valid address, data is copied to
1993 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1994 * IOREMAP area is treated as memory hole and no copy is done.
1996 * If [addr...addr+count) doesn't includes any intersects with alive
1997 * vm_struct area, returns 0. @buf should be kernel's buffer.
1999 * Note: In usual ops, vread() is never necessary because the caller
2000 * should know vmalloc() area is valid and can use memcpy().
2001 * This is for routines which have to access vmalloc area without
2002 * any informaion, as /dev/kmem.
2006 long vread(char *buf
, char *addr
, unsigned long count
)
2008 struct vm_struct
*tmp
;
2009 char *vaddr
, *buf_start
= buf
;
2010 unsigned long buflen
= count
;
2013 /* Don't allow overflow */
2014 if ((unsigned long) addr
+ count
< count
)
2015 count
= -(unsigned long) addr
;
2017 read_lock(&vmlist_lock
);
2018 for (tmp
= vmlist
; count
&& tmp
; tmp
= tmp
->next
) {
2019 vaddr
= (char *) tmp
->addr
;
2020 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
2022 while (addr
< vaddr
) {
2030 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
2033 if (!(tmp
->flags
& VM_IOREMAP
))
2034 aligned_vread(buf
, addr
, n
);
2035 else /* IOREMAP area is treated as memory hole */
2042 read_unlock(&vmlist_lock
);
2044 if (buf
== buf_start
)
2046 /* zero-fill memory holes */
2047 if (buf
!= buf_start
+ buflen
)
2048 memset(buf
, 0, buflen
- (buf
- buf_start
));
2054 * vwrite() - write vmalloc area in a safe way.
2055 * @buf: buffer for source data
2056 * @addr: vm address.
2057 * @count: number of bytes to be read.
2059 * Returns # of bytes which addr and buf should be incresed.
2060 * (same number to @count).
2061 * If [addr...addr+count) doesn't includes any intersect with valid
2062 * vmalloc area, returns 0.
2064 * This function checks that addr is a valid vmalloc'ed area, and
2065 * copy data from a buffer to the given addr. If specified range of
2066 * [addr...addr+count) includes some valid address, data is copied from
2067 * proper area of @buf. If there are memory holes, no copy to hole.
2068 * IOREMAP area is treated as memory hole and no copy is done.
2070 * If [addr...addr+count) doesn't includes any intersects with alive
2071 * vm_struct area, returns 0. @buf should be kernel's buffer.
2073 * Note: In usual ops, vwrite() is never necessary because the caller
2074 * should know vmalloc() area is valid and can use memcpy().
2075 * This is for routines which have to access vmalloc area without
2076 * any informaion, as /dev/kmem.
2079 long vwrite(char *buf
, char *addr
, unsigned long count
)
2081 struct vm_struct
*tmp
;
2083 unsigned long n
, buflen
;
2086 /* Don't allow overflow */
2087 if ((unsigned long) addr
+ count
< count
)
2088 count
= -(unsigned long) addr
;
2091 read_lock(&vmlist_lock
);
2092 for (tmp
= vmlist
; count
&& tmp
; tmp
= tmp
->next
) {
2093 vaddr
= (char *) tmp
->addr
;
2094 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
2096 while (addr
< vaddr
) {
2103 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
2106 if (!(tmp
->flags
& VM_IOREMAP
)) {
2107 aligned_vwrite(buf
, addr
, n
);
2115 read_unlock(&vmlist_lock
);
2122 * remap_vmalloc_range - map vmalloc pages to userspace
2123 * @vma: vma to cover (map full range of vma)
2124 * @addr: vmalloc memory
2125 * @pgoff: number of pages into addr before first page to map
2127 * Returns: 0 for success, -Exxx on failure
2129 * This function checks that addr is a valid vmalloc'ed area, and
2130 * that it is big enough to cover the vma. Will return failure if
2131 * that criteria isn't met.
2133 * Similar to remap_pfn_range() (see mm/memory.c)
2135 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2136 unsigned long pgoff
)
2138 struct vm_struct
*area
;
2139 unsigned long uaddr
= vma
->vm_start
;
2140 unsigned long usize
= vma
->vm_end
- vma
->vm_start
;
2142 if ((PAGE_SIZE
-1) & (unsigned long)addr
)
2145 area
= find_vm_area(addr
);
2149 if (!(area
->flags
& VM_USERMAP
))
2152 if (usize
+ (pgoff
<< PAGE_SHIFT
) > area
->size
- PAGE_SIZE
)
2155 addr
+= pgoff
<< PAGE_SHIFT
;
2157 struct page
*page
= vmalloc_to_page(addr
);
2160 ret
= vm_insert_page(vma
, uaddr
, page
);
2167 } while (usize
> 0);
2169 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2173 EXPORT_SYMBOL(remap_vmalloc_range
);
2176 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2179 void __attribute__((weak
)) vmalloc_sync_all(void)
2184 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2196 * alloc_vm_area - allocate a range of kernel address space
2197 * @size: size of the area
2198 * @ptes: returns the PTEs for the address space
2200 * Returns: NULL on failure, vm_struct on success
2202 * This function reserves a range of kernel address space, and
2203 * allocates pagetables to map that range. No actual mappings
2206 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2207 * allocated for the VM area are returned.
2209 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2211 struct vm_struct
*area
;
2213 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2214 __builtin_return_address(0));
2219 * This ensures that page tables are constructed for this region
2220 * of kernel virtual address space and mapped into init_mm.
2222 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2223 size
, f
, ptes
? &ptes
: NULL
)) {
2230 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2232 void free_vm_area(struct vm_struct
*area
)
2234 struct vm_struct
*ret
;
2235 ret
= remove_vm_area(area
->addr
);
2236 BUG_ON(ret
!= area
);
2239 EXPORT_SYMBOL_GPL(free_vm_area
);
2242 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2244 return n
? rb_entry(n
, struct vmap_area
, rb_node
) : NULL
;
2248 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2249 * @end: target address
2250 * @pnext: out arg for the next vmap_area
2251 * @pprev: out arg for the previous vmap_area
2253 * Returns: %true if either or both of next and prev are found,
2254 * %false if no vmap_area exists
2256 * Find vmap_areas end addresses of which enclose @end. ie. if not
2257 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2259 static bool pvm_find_next_prev(unsigned long end
,
2260 struct vmap_area
**pnext
,
2261 struct vmap_area
**pprev
)
2263 struct rb_node
*n
= vmap_area_root
.rb_node
;
2264 struct vmap_area
*va
= NULL
;
2267 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2268 if (end
< va
->va_end
)
2270 else if (end
> va
->va_end
)
2279 if (va
->va_end
> end
) {
2281 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2284 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2290 * pvm_determine_end - find the highest aligned address between two vmap_areas
2291 * @pnext: in/out arg for the next vmap_area
2292 * @pprev: in/out arg for the previous vmap_area
2295 * Returns: determined end address
2297 * Find the highest aligned address between *@pnext and *@pprev below
2298 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2299 * down address is between the end addresses of the two vmap_areas.
2301 * Please note that the address returned by this function may fall
2302 * inside *@pnext vmap_area. The caller is responsible for checking
2305 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2306 struct vmap_area
**pprev
,
2307 unsigned long align
)
2309 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2313 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2317 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2319 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2326 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2327 * @offsets: array containing offset of each area
2328 * @sizes: array containing size of each area
2329 * @nr_vms: the number of areas to allocate
2330 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2332 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2333 * vm_structs on success, %NULL on failure
2335 * Percpu allocator wants to use congruent vm areas so that it can
2336 * maintain the offsets among percpu areas. This function allocates
2337 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2338 * be scattered pretty far, distance between two areas easily going up
2339 * to gigabytes. To avoid interacting with regular vmallocs, these
2340 * areas are allocated from top.
2342 * Despite its complicated look, this allocator is rather simple. It
2343 * does everything top-down and scans areas from the end looking for
2344 * matching slot. While scanning, if any of the areas overlaps with
2345 * existing vmap_area, the base address is pulled down to fit the
2346 * area. Scanning is repeated till all the areas fit and then all
2347 * necessary data structres are inserted and the result is returned.
2349 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2350 const size_t *sizes
, int nr_vms
,
2353 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2354 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2355 struct vmap_area
**vas
, *prev
, *next
;
2356 struct vm_struct
**vms
;
2357 int area
, area2
, last_area
, term_area
;
2358 unsigned long base
, start
, end
, last_end
;
2359 bool purged
= false;
2361 /* verify parameters and allocate data structures */
2362 BUG_ON(align
& ~PAGE_MASK
|| !is_power_of_2(align
));
2363 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2364 start
= offsets
[area
];
2365 end
= start
+ sizes
[area
];
2367 /* is everything aligned properly? */
2368 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2369 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2371 /* detect the area with the highest address */
2372 if (start
> offsets
[last_area
])
2375 for (area2
= 0; area2
< nr_vms
; area2
++) {
2376 unsigned long start2
= offsets
[area2
];
2377 unsigned long end2
= start2
+ sizes
[area2
];
2382 BUG_ON(start2
>= start
&& start2
< end
);
2383 BUG_ON(end2
<= end
&& end2
> start
);
2386 last_end
= offsets
[last_area
] + sizes
[last_area
];
2388 if (vmalloc_end
- vmalloc_start
< last_end
) {
2393 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2394 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2398 for (area
= 0; area
< nr_vms
; area
++) {
2399 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2400 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2401 if (!vas
[area
] || !vms
[area
])
2405 spin_lock(&vmap_area_lock
);
2407 /* start scanning - we scan from the top, begin with the last area */
2408 area
= term_area
= last_area
;
2409 start
= offsets
[area
];
2410 end
= start
+ sizes
[area
];
2412 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2413 base
= vmalloc_end
- last_end
;
2416 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2419 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2420 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2423 * base might have underflowed, add last_end before
2426 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2427 spin_unlock(&vmap_area_lock
);
2429 purge_vmap_area_lazy();
2437 * If next overlaps, move base downwards so that it's
2438 * right below next and then recheck.
2440 if (next
&& next
->va_start
< base
+ end
) {
2441 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2447 * If prev overlaps, shift down next and prev and move
2448 * base so that it's right below new next and then
2451 if (prev
&& prev
->va_end
> base
+ start
) {
2453 prev
= node_to_va(rb_prev(&next
->rb_node
));
2454 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2460 * This area fits, move on to the previous one. If
2461 * the previous one is the terminal one, we're done.
2463 area
= (area
+ nr_vms
- 1) % nr_vms
;
2464 if (area
== term_area
)
2466 start
= offsets
[area
];
2467 end
= start
+ sizes
[area
];
2468 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2471 /* we've found a fitting base, insert all va's */
2472 for (area
= 0; area
< nr_vms
; area
++) {
2473 struct vmap_area
*va
= vas
[area
];
2475 va
->va_start
= base
+ offsets
[area
];
2476 va
->va_end
= va
->va_start
+ sizes
[area
];
2477 __insert_vmap_area(va
);
2480 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2482 spin_unlock(&vmap_area_lock
);
2484 /* insert all vm's */
2485 for (area
= 0; area
< nr_vms
; area
++)
2486 insert_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2493 for (area
= 0; area
< nr_vms
; area
++) {
2504 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2505 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2506 * @nr_vms: the number of allocated areas
2508 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2510 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2514 for (i
= 0; i
< nr_vms
; i
++)
2515 free_vm_area(vms
[i
]);
2518 #endif /* CONFIG_SMP */
2520 #ifdef CONFIG_PROC_FS
2521 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2522 __acquires(&vmlist_lock
)
2525 struct vm_struct
*v
;
2527 read_lock(&vmlist_lock
);
2529 while (n
> 0 && v
) {
2540 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2542 struct vm_struct
*v
= p
;
2548 static void s_stop(struct seq_file
*m
, void *p
)
2549 __releases(&vmlist_lock
)
2551 read_unlock(&vmlist_lock
);
2554 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2556 if (IS_ENABLED(CONFIG_NUMA
)) {
2557 unsigned int nr
, *counters
= m
->private;
2562 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2564 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2565 counters
[page_to_nid(v
->pages
[nr
])]++;
2567 for_each_node_state(nr
, N_HIGH_MEMORY
)
2569 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2573 static int s_show(struct seq_file
*m
, void *p
)
2575 struct vm_struct
*v
= p
;
2577 seq_printf(m
, "0x%pK-0x%pK %7ld",
2578 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2581 seq_printf(m
, " %pS", v
->caller
);
2584 seq_printf(m
, " pages=%d", v
->nr_pages
);
2587 seq_printf(m
, " phys=%llx", (unsigned long long)v
->phys_addr
);
2589 if (v
->flags
& VM_IOREMAP
)
2590 seq_printf(m
, " ioremap");
2592 if (v
->flags
& VM_ALLOC
)
2593 seq_printf(m
, " vmalloc");
2595 if (v
->flags
& VM_MAP
)
2596 seq_printf(m
, " vmap");
2598 if (v
->flags
& VM_USERMAP
)
2599 seq_printf(m
, " user");
2601 if (v
->flags
& VM_VPAGES
)
2602 seq_printf(m
, " vpages");
2604 show_numa_info(m
, v
);
2609 static const struct seq_operations vmalloc_op
= {
2616 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2618 unsigned int *ptr
= NULL
;
2621 if (IS_ENABLED(CONFIG_NUMA
)) {
2622 ptr
= kmalloc(nr_node_ids
* sizeof(unsigned int), GFP_KERNEL
);
2626 ret
= seq_open(file
, &vmalloc_op
);
2628 struct seq_file
*m
= file
->private_data
;
2635 static const struct file_operations proc_vmalloc_operations
= {
2636 .open
= vmalloc_open
,
2638 .llseek
= seq_lseek
,
2639 .release
= seq_release_private
,
2642 static int __init
proc_vmalloc_init(void)
2644 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2647 module_init(proc_vmalloc_init
);