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/slab.h>
16 #include <linux/spinlock.h>
17 #include <linux/interrupt.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/debugobjects.h>
21 #include <linux/kallsyms.h>
22 #include <linux/list.h>
23 #include <linux/rbtree.h>
24 #include <linux/radix-tree.h>
25 #include <linux/rcupdate.h>
26 #include <linux/pfn.h>
27 #include <linux/kmemleak.h>
29 #include <asm/atomic.h>
30 #include <asm/uaccess.h>
31 #include <asm/tlbflush.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
);
179 static int vmap_page_range(unsigned long start
, unsigned long end
,
180 pgprot_t prot
, struct page
**pages
)
184 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
185 flush_cache_vmap(start
, end
);
189 static inline int is_vmalloc_or_module_addr(const void *x
)
192 * ARM, x86-64 and sparc64 put modules in a special place,
193 * and fall back on vmalloc() if that fails. Others
194 * just put it in the vmalloc space.
196 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
197 unsigned long addr
= (unsigned long)x
;
198 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
201 return is_vmalloc_addr(x
);
205 * Walk a vmap address to the struct page it maps.
207 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
209 unsigned long addr
= (unsigned long) vmalloc_addr
;
210 struct page
*page
= NULL
;
211 pgd_t
*pgd
= pgd_offset_k(addr
);
214 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
215 * architectures that do not vmalloc module space
217 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
219 if (!pgd_none(*pgd
)) {
220 pud_t
*pud
= pud_offset(pgd
, addr
);
221 if (!pud_none(*pud
)) {
222 pmd_t
*pmd
= pmd_offset(pud
, addr
);
223 if (!pmd_none(*pmd
)) {
226 ptep
= pte_offset_map(pmd
, addr
);
228 if (pte_present(pte
))
229 page
= pte_page(pte
);
236 EXPORT_SYMBOL(vmalloc_to_page
);
239 * Map a vmalloc()-space virtual address to the physical page frame number.
241 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
243 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
245 EXPORT_SYMBOL(vmalloc_to_pfn
);
248 /*** Global kva allocator ***/
250 #define VM_LAZY_FREE 0x01
251 #define VM_LAZY_FREEING 0x02
252 #define VM_VM_AREA 0x04
255 unsigned long va_start
;
256 unsigned long va_end
;
258 struct rb_node rb_node
; /* address sorted rbtree */
259 struct list_head list
; /* address sorted list */
260 struct list_head purge_list
; /* "lazy purge" list */
262 struct rcu_head rcu_head
;
265 static DEFINE_SPINLOCK(vmap_area_lock
);
266 static struct rb_root vmap_area_root
= RB_ROOT
;
267 static LIST_HEAD(vmap_area_list
);
269 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
271 struct rb_node
*n
= vmap_area_root
.rb_node
;
274 struct vmap_area
*va
;
276 va
= rb_entry(n
, struct vmap_area
, rb_node
);
277 if (addr
< va
->va_start
)
279 else if (addr
> va
->va_start
)
288 static void __insert_vmap_area(struct vmap_area
*va
)
290 struct rb_node
**p
= &vmap_area_root
.rb_node
;
291 struct rb_node
*parent
= NULL
;
295 struct vmap_area
*tmp
;
298 tmp
= rb_entry(parent
, struct vmap_area
, rb_node
);
299 if (va
->va_start
< tmp
->va_end
)
301 else if (va
->va_end
> tmp
->va_start
)
307 rb_link_node(&va
->rb_node
, parent
, p
);
308 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
310 /* address-sort this list so it is usable like the vmlist */
311 tmp
= rb_prev(&va
->rb_node
);
313 struct vmap_area
*prev
;
314 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
315 list_add_rcu(&va
->list
, &prev
->list
);
317 list_add_rcu(&va
->list
, &vmap_area_list
);
320 static void purge_vmap_area_lazy(void);
323 * Allocate a region of KVA of the specified size and alignment, within the
326 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
328 unsigned long vstart
, unsigned long vend
,
329 int node
, gfp_t gfp_mask
)
331 struct vmap_area
*va
;
337 BUG_ON(size
& ~PAGE_MASK
);
339 va
= kmalloc_node(sizeof(struct vmap_area
),
340 gfp_mask
& GFP_RECLAIM_MASK
, node
);
342 return ERR_PTR(-ENOMEM
);
345 addr
= ALIGN(vstart
, align
);
347 spin_lock(&vmap_area_lock
);
348 if (addr
+ size
- 1 < addr
)
351 /* XXX: could have a last_hole cache */
352 n
= vmap_area_root
.rb_node
;
354 struct vmap_area
*first
= NULL
;
357 struct vmap_area
*tmp
;
358 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
359 if (tmp
->va_end
>= addr
) {
360 if (!first
&& tmp
->va_start
< addr
+ size
)
372 if (first
->va_end
< addr
) {
373 n
= rb_next(&first
->rb_node
);
375 first
= rb_entry(n
, struct vmap_area
, rb_node
);
380 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
381 addr
= ALIGN(first
->va_end
+ PAGE_SIZE
, align
);
382 if (addr
+ size
- 1 < addr
)
385 n
= rb_next(&first
->rb_node
);
387 first
= rb_entry(n
, struct vmap_area
, rb_node
);
393 if (addr
+ size
> vend
) {
395 spin_unlock(&vmap_area_lock
);
397 purge_vmap_area_lazy();
401 if (printk_ratelimit())
403 "vmap allocation for size %lu failed: "
404 "use vmalloc=<size> to increase size.\n", size
);
406 return ERR_PTR(-EBUSY
);
409 BUG_ON(addr
& (align
-1));
412 va
->va_end
= addr
+ size
;
414 __insert_vmap_area(va
);
415 spin_unlock(&vmap_area_lock
);
420 static void rcu_free_va(struct rcu_head
*head
)
422 struct vmap_area
*va
= container_of(head
, struct vmap_area
, rcu_head
);
427 static void __free_vmap_area(struct vmap_area
*va
)
429 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
430 rb_erase(&va
->rb_node
, &vmap_area_root
);
431 RB_CLEAR_NODE(&va
->rb_node
);
432 list_del_rcu(&va
->list
);
434 call_rcu(&va
->rcu_head
, rcu_free_va
);
438 * Free a region of KVA allocated by alloc_vmap_area
440 static void free_vmap_area(struct vmap_area
*va
)
442 spin_lock(&vmap_area_lock
);
443 __free_vmap_area(va
);
444 spin_unlock(&vmap_area_lock
);
448 * Clear the pagetable entries of a given vmap_area
450 static void unmap_vmap_area(struct vmap_area
*va
)
452 vunmap_page_range(va
->va_start
, va
->va_end
);
455 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
458 * Unmap page tables and force a TLB flush immediately if
459 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
460 * bugs similarly to those in linear kernel virtual address
461 * space after a page has been freed.
463 * All the lazy freeing logic is still retained, in order to
464 * minimise intrusiveness of this debugging feature.
466 * This is going to be *slow* (linear kernel virtual address
467 * debugging doesn't do a broadcast TLB flush so it is a lot
470 #ifdef CONFIG_DEBUG_PAGEALLOC
471 vunmap_page_range(start
, end
);
472 flush_tlb_kernel_range(start
, end
);
477 * lazy_max_pages is the maximum amount of virtual address space we gather up
478 * before attempting to purge with a TLB flush.
480 * There is a tradeoff here: a larger number will cover more kernel page tables
481 * and take slightly longer to purge, but it will linearly reduce the number of
482 * global TLB flushes that must be performed. It would seem natural to scale
483 * this number up linearly with the number of CPUs (because vmapping activity
484 * could also scale linearly with the number of CPUs), however it is likely
485 * that in practice, workloads might be constrained in other ways that mean
486 * vmap activity will not scale linearly with CPUs. Also, I want to be
487 * conservative and not introduce a big latency on huge systems, so go with
488 * a less aggressive log scale. It will still be an improvement over the old
489 * code, and it will be simple to change the scale factor if we find that it
490 * becomes a problem on bigger systems.
492 static unsigned long lazy_max_pages(void)
496 log
= fls(num_online_cpus());
498 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
501 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
504 * Purges all lazily-freed vmap areas.
506 * If sync is 0 then don't purge if there is already a purge in progress.
507 * If force_flush is 1, then flush kernel TLBs between *start and *end even
508 * if we found no lazy vmap areas to unmap (callers can use this to optimise
509 * their own TLB flushing).
510 * Returns with *start = min(*start, lowest purged address)
511 * *end = max(*end, highest purged address)
513 static void __purge_vmap_area_lazy(unsigned long *start
, unsigned long *end
,
514 int sync
, int force_flush
)
516 static DEFINE_SPINLOCK(purge_lock
);
518 struct vmap_area
*va
;
519 struct vmap_area
*n_va
;
523 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
524 * should not expect such behaviour. This just simplifies locking for
525 * the case that isn't actually used at the moment anyway.
527 if (!sync
&& !force_flush
) {
528 if (!spin_trylock(&purge_lock
))
531 spin_lock(&purge_lock
);
534 list_for_each_entry_rcu(va
, &vmap_area_list
, list
) {
535 if (va
->flags
& VM_LAZY_FREE
) {
536 if (va
->va_start
< *start
)
537 *start
= va
->va_start
;
538 if (va
->va_end
> *end
)
540 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
542 list_add_tail(&va
->purge_list
, &valist
);
543 va
->flags
|= VM_LAZY_FREEING
;
544 va
->flags
&= ~VM_LAZY_FREE
;
550 atomic_sub(nr
, &vmap_lazy_nr
);
552 if (nr
|| force_flush
)
553 flush_tlb_kernel_range(*start
, *end
);
556 spin_lock(&vmap_area_lock
);
557 list_for_each_entry_safe(va
, n_va
, &valist
, purge_list
)
558 __free_vmap_area(va
);
559 spin_unlock(&vmap_area_lock
);
561 spin_unlock(&purge_lock
);
565 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
566 * is already purging.
568 static void try_purge_vmap_area_lazy(void)
570 unsigned long start
= ULONG_MAX
, end
= 0;
572 __purge_vmap_area_lazy(&start
, &end
, 0, 0);
576 * Kick off a purge of the outstanding lazy areas.
578 static void purge_vmap_area_lazy(void)
580 unsigned long start
= ULONG_MAX
, end
= 0;
582 __purge_vmap_area_lazy(&start
, &end
, 1, 0);
586 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
587 * called for the correct range previously.
589 static void free_unmap_vmap_area_noflush(struct vmap_area
*va
)
591 va
->flags
|= VM_LAZY_FREE
;
592 atomic_add((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
, &vmap_lazy_nr
);
593 if (unlikely(atomic_read(&vmap_lazy_nr
) > lazy_max_pages()))
594 try_purge_vmap_area_lazy();
598 * Free and unmap a vmap area
600 static void free_unmap_vmap_area(struct vmap_area
*va
)
602 flush_cache_vunmap(va
->va_start
, va
->va_end
);
603 free_unmap_vmap_area_noflush(va
);
606 static struct vmap_area
*find_vmap_area(unsigned long addr
)
608 struct vmap_area
*va
;
610 spin_lock(&vmap_area_lock
);
611 va
= __find_vmap_area(addr
);
612 spin_unlock(&vmap_area_lock
);
617 static void free_unmap_vmap_area_addr(unsigned long addr
)
619 struct vmap_area
*va
;
621 va
= find_vmap_area(addr
);
623 free_unmap_vmap_area(va
);
627 /*** Per cpu kva allocator ***/
630 * vmap space is limited especially on 32 bit architectures. Ensure there is
631 * room for at least 16 percpu vmap blocks per CPU.
634 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
635 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
636 * instead (we just need a rough idea)
638 #if BITS_PER_LONG == 32
639 #define VMALLOC_SPACE (128UL*1024*1024)
641 #define VMALLOC_SPACE (128UL*1024*1024*1024)
644 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
645 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
646 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
647 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
648 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
649 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
650 #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
651 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
652 VMALLOC_PAGES / NR_CPUS / 16))
654 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
656 static bool vmap_initialized __read_mostly
= false;
658 struct vmap_block_queue
{
660 struct list_head free
;
661 struct list_head dirty
;
662 unsigned int nr_dirty
;
667 struct vmap_area
*va
;
668 struct vmap_block_queue
*vbq
;
669 unsigned long free
, dirty
;
670 DECLARE_BITMAP(alloc_map
, VMAP_BBMAP_BITS
);
671 DECLARE_BITMAP(dirty_map
, VMAP_BBMAP_BITS
);
673 struct list_head free_list
;
674 struct rcu_head rcu_head
;
678 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
679 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
682 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
683 * in the free path. Could get rid of this if we change the API to return a
684 * "cookie" from alloc, to be passed to free. But no big deal yet.
686 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
687 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
690 * We should probably have a fallback mechanism to allocate virtual memory
691 * out of partially filled vmap blocks. However vmap block sizing should be
692 * fairly reasonable according to the vmalloc size, so it shouldn't be a
696 static unsigned long addr_to_vb_idx(unsigned long addr
)
698 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
699 addr
/= VMAP_BLOCK_SIZE
;
703 static struct vmap_block
*new_vmap_block(gfp_t gfp_mask
)
705 struct vmap_block_queue
*vbq
;
706 struct vmap_block
*vb
;
707 struct vmap_area
*va
;
708 unsigned long vb_idx
;
711 node
= numa_node_id();
713 vb
= kmalloc_node(sizeof(struct vmap_block
),
714 gfp_mask
& GFP_RECLAIM_MASK
, node
);
716 return ERR_PTR(-ENOMEM
);
718 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
719 VMALLOC_START
, VMALLOC_END
,
721 if (unlikely(IS_ERR(va
))) {
723 return ERR_PTR(PTR_ERR(va
));
726 err
= radix_tree_preload(gfp_mask
);
733 spin_lock_init(&vb
->lock
);
735 vb
->free
= VMAP_BBMAP_BITS
;
737 bitmap_zero(vb
->alloc_map
, VMAP_BBMAP_BITS
);
738 bitmap_zero(vb
->dirty_map
, VMAP_BBMAP_BITS
);
739 INIT_LIST_HEAD(&vb
->free_list
);
741 vb_idx
= addr_to_vb_idx(va
->va_start
);
742 spin_lock(&vmap_block_tree_lock
);
743 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
744 spin_unlock(&vmap_block_tree_lock
);
746 radix_tree_preload_end();
748 vbq
= &get_cpu_var(vmap_block_queue
);
750 spin_lock(&vbq
->lock
);
751 list_add(&vb
->free_list
, &vbq
->free
);
752 spin_unlock(&vbq
->lock
);
753 put_cpu_var(vmap_cpu_blocks
);
758 static void rcu_free_vb(struct rcu_head
*head
)
760 struct vmap_block
*vb
= container_of(head
, struct vmap_block
, rcu_head
);
765 static void free_vmap_block(struct vmap_block
*vb
)
767 struct vmap_block
*tmp
;
768 unsigned long vb_idx
;
770 BUG_ON(!list_empty(&vb
->free_list
));
772 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
773 spin_lock(&vmap_block_tree_lock
);
774 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
775 spin_unlock(&vmap_block_tree_lock
);
778 free_unmap_vmap_area_noflush(vb
->va
);
779 call_rcu(&vb
->rcu_head
, rcu_free_vb
);
782 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
784 struct vmap_block_queue
*vbq
;
785 struct vmap_block
*vb
;
786 unsigned long addr
= 0;
789 BUG_ON(size
& ~PAGE_MASK
);
790 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
791 order
= get_order(size
);
795 vbq
= &get_cpu_var(vmap_block_queue
);
796 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
799 spin_lock(&vb
->lock
);
800 i
= bitmap_find_free_region(vb
->alloc_map
,
801 VMAP_BBMAP_BITS
, order
);
804 addr
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
805 BUG_ON(addr_to_vb_idx(addr
) !=
806 addr_to_vb_idx(vb
->va
->va_start
));
807 vb
->free
-= 1UL << order
;
809 spin_lock(&vbq
->lock
);
810 list_del_init(&vb
->free_list
);
811 spin_unlock(&vbq
->lock
);
813 spin_unlock(&vb
->lock
);
816 spin_unlock(&vb
->lock
);
818 put_cpu_var(vmap_cpu_blocks
);
822 vb
= new_vmap_block(gfp_mask
);
831 static void vb_free(const void *addr
, unsigned long size
)
833 unsigned long offset
;
834 unsigned long vb_idx
;
836 struct vmap_block
*vb
;
838 BUG_ON(size
& ~PAGE_MASK
);
839 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
841 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
843 order
= get_order(size
);
845 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
847 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
849 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
853 spin_lock(&vb
->lock
);
854 bitmap_allocate_region(vb
->dirty_map
, offset
>> PAGE_SHIFT
, order
);
856 vb
->dirty
+= 1UL << order
;
857 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
858 BUG_ON(vb
->free
|| !list_empty(&vb
->free_list
));
859 spin_unlock(&vb
->lock
);
862 spin_unlock(&vb
->lock
);
866 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
868 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
869 * to amortize TLB flushing overheads. What this means is that any page you
870 * have now, may, in a former life, have been mapped into kernel virtual
871 * address by the vmap layer and so there might be some CPUs with TLB entries
872 * still referencing that page (additional to the regular 1:1 kernel mapping).
874 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
875 * be sure that none of the pages we have control over will have any aliases
876 * from the vmap layer.
878 void vm_unmap_aliases(void)
880 unsigned long start
= ULONG_MAX
, end
= 0;
884 if (unlikely(!vmap_initialized
))
887 for_each_possible_cpu(cpu
) {
888 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
889 struct vmap_block
*vb
;
892 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
895 spin_lock(&vb
->lock
);
896 i
= find_first_bit(vb
->dirty_map
, VMAP_BBMAP_BITS
);
897 while (i
< VMAP_BBMAP_BITS
) {
900 j
= find_next_zero_bit(vb
->dirty_map
,
903 s
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
904 e
= vb
->va
->va_start
+ (j
<< PAGE_SHIFT
);
905 vunmap_page_range(s
, e
);
914 i
= find_next_bit(vb
->dirty_map
,
917 spin_unlock(&vb
->lock
);
922 __purge_vmap_area_lazy(&start
, &end
, 1, flush
);
924 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
927 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
928 * @mem: the pointer returned by vm_map_ram
929 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
931 void vm_unmap_ram(const void *mem
, unsigned int count
)
933 unsigned long size
= count
<< PAGE_SHIFT
;
934 unsigned long addr
= (unsigned long)mem
;
937 BUG_ON(addr
< VMALLOC_START
);
938 BUG_ON(addr
> VMALLOC_END
);
939 BUG_ON(addr
& (PAGE_SIZE
-1));
941 debug_check_no_locks_freed(mem
, size
);
942 vmap_debug_free_range(addr
, addr
+size
);
944 if (likely(count
<= VMAP_MAX_ALLOC
))
947 free_unmap_vmap_area_addr(addr
);
949 EXPORT_SYMBOL(vm_unmap_ram
);
952 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
953 * @pages: an array of pointers to the pages to be mapped
954 * @count: number of pages
955 * @node: prefer to allocate data structures on this node
956 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
958 * Returns: a pointer to the address that has been mapped, or %NULL on failure
960 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
962 unsigned long size
= count
<< PAGE_SHIFT
;
966 if (likely(count
<= VMAP_MAX_ALLOC
)) {
967 mem
= vb_alloc(size
, GFP_KERNEL
);
970 addr
= (unsigned long)mem
;
972 struct vmap_area
*va
;
973 va
= alloc_vmap_area(size
, PAGE_SIZE
,
974 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
981 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
982 vm_unmap_ram(mem
, count
);
987 EXPORT_SYMBOL(vm_map_ram
);
990 * vm_area_register_early - register vmap area early during boot
991 * @vm: vm_struct to register
992 * @align: requested alignment
994 * This function is used to register kernel vm area before
995 * vmalloc_init() is called. @vm->size and @vm->flags should contain
996 * proper values on entry and other fields should be zero. On return,
997 * vm->addr contains the allocated address.
999 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1001 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1003 static size_t vm_init_off __initdata
;
1006 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1007 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1009 vm
->addr
= (void *)addr
;
1015 void __init
vmalloc_init(void)
1017 struct vmap_area
*va
;
1018 struct vm_struct
*tmp
;
1021 for_each_possible_cpu(i
) {
1022 struct vmap_block_queue
*vbq
;
1024 vbq
= &per_cpu(vmap_block_queue
, i
);
1025 spin_lock_init(&vbq
->lock
);
1026 INIT_LIST_HEAD(&vbq
->free
);
1027 INIT_LIST_HEAD(&vbq
->dirty
);
1031 /* Import existing vmlist entries. */
1032 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1033 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1034 va
->flags
= tmp
->flags
| VM_VM_AREA
;
1035 va
->va_start
= (unsigned long)tmp
->addr
;
1036 va
->va_end
= va
->va_start
+ tmp
->size
;
1037 __insert_vmap_area(va
);
1039 vmap_initialized
= true;
1043 * map_kernel_range_noflush - map kernel VM area with the specified pages
1044 * @addr: start of the VM area to map
1045 * @size: size of the VM area to map
1046 * @prot: page protection flags to use
1047 * @pages: pages to map
1049 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1050 * specify should have been allocated using get_vm_area() and its
1054 * This function does NOT do any cache flushing. The caller is
1055 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1056 * before calling this function.
1059 * The number of pages mapped on success, -errno on failure.
1061 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1062 pgprot_t prot
, struct page
**pages
)
1064 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1068 * unmap_kernel_range_noflush - unmap kernel VM area
1069 * @addr: start of the VM area to unmap
1070 * @size: size of the VM area to unmap
1072 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1073 * specify should have been allocated using get_vm_area() and its
1077 * This function does NOT do any cache flushing. The caller is
1078 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1079 * before calling this function and flush_tlb_kernel_range() after.
1081 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1083 vunmap_page_range(addr
, addr
+ size
);
1087 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1088 * @addr: start of the VM area to unmap
1089 * @size: size of the VM area to unmap
1091 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1092 * the unmapping and tlb after.
1094 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1096 unsigned long end
= addr
+ size
;
1098 flush_cache_vunmap(addr
, end
);
1099 vunmap_page_range(addr
, end
);
1100 flush_tlb_kernel_range(addr
, end
);
1103 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
***pages
)
1105 unsigned long addr
= (unsigned long)area
->addr
;
1106 unsigned long end
= addr
+ area
->size
- PAGE_SIZE
;
1109 err
= vmap_page_range(addr
, end
, prot
, *pages
);
1117 EXPORT_SYMBOL_GPL(map_vm_area
);
1119 /*** Old vmalloc interfaces ***/
1120 DEFINE_RWLOCK(vmlist_lock
);
1121 struct vm_struct
*vmlist
;
1123 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1124 unsigned long flags
, unsigned long start
, unsigned long end
,
1125 int node
, gfp_t gfp_mask
, void *caller
)
1127 static struct vmap_area
*va
;
1128 struct vm_struct
*area
;
1129 struct vm_struct
*tmp
, **p
;
1130 unsigned long align
= 1;
1132 BUG_ON(in_interrupt());
1133 if (flags
& VM_IOREMAP
) {
1134 int bit
= fls(size
);
1136 if (bit
> IOREMAP_MAX_ORDER
)
1137 bit
= IOREMAP_MAX_ORDER
;
1138 else if (bit
< PAGE_SHIFT
)
1144 size
= PAGE_ALIGN(size
);
1145 if (unlikely(!size
))
1148 area
= kmalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1149 if (unlikely(!area
))
1153 * We always allocate a guard page.
1157 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1163 area
->flags
= flags
;
1164 area
->addr
= (void *)va
->va_start
;
1168 area
->phys_addr
= 0;
1169 area
->caller
= caller
;
1171 va
->flags
|= VM_VM_AREA
;
1173 write_lock(&vmlist_lock
);
1174 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1175 if (tmp
->addr
>= area
->addr
)
1180 write_unlock(&vmlist_lock
);
1185 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1186 unsigned long start
, unsigned long end
)
1188 return __get_vm_area_node(size
, flags
, start
, end
, -1, GFP_KERNEL
,
1189 __builtin_return_address(0));
1191 EXPORT_SYMBOL_GPL(__get_vm_area
);
1193 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1194 unsigned long start
, unsigned long end
,
1197 return __get_vm_area_node(size
, flags
, start
, end
, -1, GFP_KERNEL
,
1202 * get_vm_area - reserve a contiguous kernel virtual area
1203 * @size: size of the area
1204 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1206 * Search an area of @size in the kernel virtual mapping area,
1207 * and reserved it for out purposes. Returns the area descriptor
1208 * on success or %NULL on failure.
1210 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1212 return __get_vm_area_node(size
, flags
, VMALLOC_START
, VMALLOC_END
,
1213 -1, GFP_KERNEL
, __builtin_return_address(0));
1216 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1219 return __get_vm_area_node(size
, flags
, VMALLOC_START
, VMALLOC_END
,
1220 -1, GFP_KERNEL
, caller
);
1223 struct vm_struct
*get_vm_area_node(unsigned long size
, unsigned long flags
,
1224 int node
, gfp_t gfp_mask
)
1226 return __get_vm_area_node(size
, flags
, VMALLOC_START
, VMALLOC_END
, node
,
1227 gfp_mask
, __builtin_return_address(0));
1230 static struct vm_struct
*find_vm_area(const void *addr
)
1232 struct vmap_area
*va
;
1234 va
= find_vmap_area((unsigned long)addr
);
1235 if (va
&& va
->flags
& VM_VM_AREA
)
1242 * remove_vm_area - find and remove a continuous kernel virtual area
1243 * @addr: base address
1245 * Search for the kernel VM area starting at @addr, and remove it.
1246 * This function returns the found VM area, but using it is NOT safe
1247 * on SMP machines, except for its size or flags.
1249 struct vm_struct
*remove_vm_area(const void *addr
)
1251 struct vmap_area
*va
;
1253 va
= find_vmap_area((unsigned long)addr
);
1254 if (va
&& va
->flags
& VM_VM_AREA
) {
1255 struct vm_struct
*vm
= va
->private;
1256 struct vm_struct
*tmp
, **p
;
1258 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1259 free_unmap_vmap_area(va
);
1260 vm
->size
-= PAGE_SIZE
;
1262 write_lock(&vmlist_lock
);
1263 for (p
= &vmlist
; (tmp
= *p
) != vm
; p
= &tmp
->next
)
1266 write_unlock(&vmlist_lock
);
1273 static void __vunmap(const void *addr
, int deallocate_pages
)
1275 struct vm_struct
*area
;
1280 if ((PAGE_SIZE
-1) & (unsigned long)addr
) {
1281 WARN(1, KERN_ERR
"Trying to vfree() bad address (%p)\n", addr
);
1285 area
= remove_vm_area(addr
);
1286 if (unlikely(!area
)) {
1287 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1292 debug_check_no_locks_freed(addr
, area
->size
);
1293 debug_check_no_obj_freed(addr
, area
->size
);
1295 if (deallocate_pages
) {
1298 for (i
= 0; i
< area
->nr_pages
; i
++) {
1299 struct page
*page
= area
->pages
[i
];
1305 if (area
->flags
& VM_VPAGES
)
1316 * vfree - release memory allocated by vmalloc()
1317 * @addr: memory base address
1319 * Free the virtually continuous memory area starting at @addr, as
1320 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1321 * NULL, no operation is performed.
1323 * Must not be called in interrupt context.
1325 void vfree(const void *addr
)
1327 BUG_ON(in_interrupt());
1329 kmemleak_free(addr
);
1333 EXPORT_SYMBOL(vfree
);
1336 * vunmap - release virtual mapping obtained by vmap()
1337 * @addr: memory base address
1339 * Free the virtually contiguous memory area starting at @addr,
1340 * which was created from the page array passed to vmap().
1342 * Must not be called in interrupt context.
1344 void vunmap(const void *addr
)
1346 BUG_ON(in_interrupt());
1350 EXPORT_SYMBOL(vunmap
);
1353 * vmap - map an array of pages into virtually contiguous space
1354 * @pages: array of page pointers
1355 * @count: number of pages to map
1356 * @flags: vm_area->flags
1357 * @prot: page protection for the mapping
1359 * Maps @count pages from @pages into contiguous kernel virtual
1362 void *vmap(struct page
**pages
, unsigned int count
,
1363 unsigned long flags
, pgprot_t prot
)
1365 struct vm_struct
*area
;
1369 if (count
> totalram_pages
)
1372 area
= get_vm_area_caller((count
<< PAGE_SHIFT
), flags
,
1373 __builtin_return_address(0));
1377 if (map_vm_area(area
, prot
, &pages
)) {
1384 EXPORT_SYMBOL(vmap
);
1386 static void *__vmalloc_node(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
,
1387 int node
, void *caller
);
1388 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1389 pgprot_t prot
, int node
, void *caller
)
1391 struct page
**pages
;
1392 unsigned int nr_pages
, array_size
, i
;
1394 nr_pages
= (area
->size
- PAGE_SIZE
) >> PAGE_SHIFT
;
1395 array_size
= (nr_pages
* sizeof(struct page
*));
1397 area
->nr_pages
= nr_pages
;
1398 /* Please note that the recursion is strictly bounded. */
1399 if (array_size
> PAGE_SIZE
) {
1400 pages
= __vmalloc_node(array_size
, gfp_mask
| __GFP_ZERO
,
1401 PAGE_KERNEL
, node
, caller
);
1402 area
->flags
|= VM_VPAGES
;
1404 pages
= kmalloc_node(array_size
,
1405 (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
,
1408 area
->pages
= pages
;
1409 area
->caller
= caller
;
1411 remove_vm_area(area
->addr
);
1416 for (i
= 0; i
< area
->nr_pages
; i
++) {
1420 page
= alloc_page(gfp_mask
);
1422 page
= alloc_pages_node(node
, gfp_mask
, 0);
1424 if (unlikely(!page
)) {
1425 /* Successfully allocated i pages, free them in __vunmap() */
1429 area
->pages
[i
] = page
;
1432 if (map_vm_area(area
, prot
, &pages
))
1441 void *__vmalloc_area(struct vm_struct
*area
, gfp_t gfp_mask
, pgprot_t prot
)
1443 void *addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, -1,
1444 __builtin_return_address(0));
1447 * A ref_count = 3 is needed because the vm_struct and vmap_area
1448 * structures allocated in the __get_vm_area_node() function contain
1449 * references to the virtual address of the vmalloc'ed block.
1451 kmemleak_alloc(addr
, area
->size
- PAGE_SIZE
, 3, gfp_mask
);
1457 * __vmalloc_node - allocate virtually contiguous memory
1458 * @size: allocation size
1459 * @gfp_mask: flags for the page level allocator
1460 * @prot: protection mask for the allocated pages
1461 * @node: node to use for allocation or -1
1462 * @caller: caller's return address
1464 * Allocate enough pages to cover @size from the page level
1465 * allocator with @gfp_mask flags. Map them into contiguous
1466 * kernel virtual space, using a pagetable protection of @prot.
1468 static void *__vmalloc_node(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
,
1469 int node
, void *caller
)
1471 struct vm_struct
*area
;
1473 unsigned long real_size
= size
;
1475 size
= PAGE_ALIGN(size
);
1476 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1479 area
= __get_vm_area_node(size
, VM_ALLOC
, VMALLOC_START
, VMALLOC_END
,
1480 node
, gfp_mask
, caller
);
1485 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
, caller
);
1488 * A ref_count = 3 is needed because the vm_struct and vmap_area
1489 * structures allocated in the __get_vm_area_node() function contain
1490 * references to the virtual address of the vmalloc'ed block.
1492 kmemleak_alloc(addr
, real_size
, 3, gfp_mask
);
1497 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1499 return __vmalloc_node(size
, gfp_mask
, prot
, -1,
1500 __builtin_return_address(0));
1502 EXPORT_SYMBOL(__vmalloc
);
1505 * vmalloc - allocate virtually contiguous memory
1506 * @size: allocation size
1507 * Allocate enough pages to cover @size from the page level
1508 * allocator and map them into contiguous kernel virtual space.
1510 * For tight control over page level allocator and protection flags
1511 * use __vmalloc() instead.
1513 void *vmalloc(unsigned long size
)
1515 return __vmalloc_node(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1516 -1, __builtin_return_address(0));
1518 EXPORT_SYMBOL(vmalloc
);
1521 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1522 * @size: allocation size
1524 * The resulting memory area is zeroed so it can be mapped to userspace
1525 * without leaking data.
1527 void *vmalloc_user(unsigned long size
)
1529 struct vm_struct
*area
;
1532 ret
= __vmalloc_node(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1533 PAGE_KERNEL
, -1, __builtin_return_address(0));
1535 area
= find_vm_area(ret
);
1536 area
->flags
|= VM_USERMAP
;
1540 EXPORT_SYMBOL(vmalloc_user
);
1543 * vmalloc_node - allocate memory on a specific node
1544 * @size: allocation size
1547 * Allocate enough pages to cover @size from the page level
1548 * allocator and map them into contiguous kernel virtual space.
1550 * For tight control over page level allocator and protection flags
1551 * use __vmalloc() instead.
1553 void *vmalloc_node(unsigned long size
, int node
)
1555 return __vmalloc_node(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1556 node
, __builtin_return_address(0));
1558 EXPORT_SYMBOL(vmalloc_node
);
1560 #ifndef PAGE_KERNEL_EXEC
1561 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1565 * vmalloc_exec - allocate virtually contiguous, executable memory
1566 * @size: allocation size
1568 * Kernel-internal function to allocate enough pages to cover @size
1569 * the page level allocator and map them into contiguous and
1570 * executable kernel virtual space.
1572 * For tight control over page level allocator and protection flags
1573 * use __vmalloc() instead.
1576 void *vmalloc_exec(unsigned long size
)
1578 return __vmalloc_node(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1579 -1, __builtin_return_address(0));
1582 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1583 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1584 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1585 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1587 #define GFP_VMALLOC32 GFP_KERNEL
1591 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1592 * @size: allocation size
1594 * Allocate enough 32bit PA addressable pages to cover @size from the
1595 * page level allocator and map them into contiguous kernel virtual space.
1597 void *vmalloc_32(unsigned long size
)
1599 return __vmalloc_node(size
, GFP_VMALLOC32
, PAGE_KERNEL
,
1600 -1, __builtin_return_address(0));
1602 EXPORT_SYMBOL(vmalloc_32
);
1605 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1606 * @size: allocation size
1608 * The resulting memory area is 32bit addressable and zeroed so it can be
1609 * mapped to userspace without leaking data.
1611 void *vmalloc_32_user(unsigned long size
)
1613 struct vm_struct
*area
;
1616 ret
= __vmalloc_node(size
, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1617 -1, __builtin_return_address(0));
1619 area
= find_vm_area(ret
);
1620 area
->flags
|= VM_USERMAP
;
1624 EXPORT_SYMBOL(vmalloc_32_user
);
1626 long vread(char *buf
, char *addr
, unsigned long count
)
1628 struct vm_struct
*tmp
;
1629 char *vaddr
, *buf_start
= buf
;
1632 /* Don't allow overflow */
1633 if ((unsigned long) addr
+ count
< count
)
1634 count
= -(unsigned long) addr
;
1636 read_lock(&vmlist_lock
);
1637 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1638 vaddr
= (char *) tmp
->addr
;
1639 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
1641 while (addr
< vaddr
) {
1649 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
1660 read_unlock(&vmlist_lock
);
1661 return buf
- buf_start
;
1664 long vwrite(char *buf
, char *addr
, unsigned long count
)
1666 struct vm_struct
*tmp
;
1667 char *vaddr
, *buf_start
= buf
;
1670 /* Don't allow overflow */
1671 if ((unsigned long) addr
+ count
< count
)
1672 count
= -(unsigned long) addr
;
1674 read_lock(&vmlist_lock
);
1675 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1676 vaddr
= (char *) tmp
->addr
;
1677 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
1679 while (addr
< vaddr
) {
1686 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
1697 read_unlock(&vmlist_lock
);
1698 return buf
- buf_start
;
1702 * remap_vmalloc_range - map vmalloc pages to userspace
1703 * @vma: vma to cover (map full range of vma)
1704 * @addr: vmalloc memory
1705 * @pgoff: number of pages into addr before first page to map
1707 * Returns: 0 for success, -Exxx on failure
1709 * This function checks that addr is a valid vmalloc'ed area, and
1710 * that it is big enough to cover the vma. Will return failure if
1711 * that criteria isn't met.
1713 * Similar to remap_pfn_range() (see mm/memory.c)
1715 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1716 unsigned long pgoff
)
1718 struct vm_struct
*area
;
1719 unsigned long uaddr
= vma
->vm_start
;
1720 unsigned long usize
= vma
->vm_end
- vma
->vm_start
;
1722 if ((PAGE_SIZE
-1) & (unsigned long)addr
)
1725 area
= find_vm_area(addr
);
1729 if (!(area
->flags
& VM_USERMAP
))
1732 if (usize
+ (pgoff
<< PAGE_SHIFT
) > area
->size
- PAGE_SIZE
)
1735 addr
+= pgoff
<< PAGE_SHIFT
;
1737 struct page
*page
= vmalloc_to_page(addr
);
1740 ret
= vm_insert_page(vma
, uaddr
, page
);
1747 } while (usize
> 0);
1749 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1750 vma
->vm_flags
|= VM_RESERVED
;
1754 EXPORT_SYMBOL(remap_vmalloc_range
);
1757 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1760 void __attribute__((weak
)) vmalloc_sync_all(void)
1765 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
1767 /* apply_to_page_range() does all the hard work. */
1772 * alloc_vm_area - allocate a range of kernel address space
1773 * @size: size of the area
1775 * Returns: NULL on failure, vm_struct on success
1777 * This function reserves a range of kernel address space, and
1778 * allocates pagetables to map that range. No actual mappings
1779 * are created. If the kernel address space is not shared
1780 * between processes, it syncs the pagetable across all
1783 struct vm_struct
*alloc_vm_area(size_t size
)
1785 struct vm_struct
*area
;
1787 area
= get_vm_area_caller(size
, VM_IOREMAP
,
1788 __builtin_return_address(0));
1793 * This ensures that page tables are constructed for this region
1794 * of kernel virtual address space and mapped into init_mm.
1796 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
1797 area
->size
, f
, NULL
)) {
1802 /* Make sure the pagetables are constructed in process kernel
1808 EXPORT_SYMBOL_GPL(alloc_vm_area
);
1810 void free_vm_area(struct vm_struct
*area
)
1812 struct vm_struct
*ret
;
1813 ret
= remove_vm_area(area
->addr
);
1814 BUG_ON(ret
!= area
);
1817 EXPORT_SYMBOL_GPL(free_vm_area
);
1820 #ifdef CONFIG_PROC_FS
1821 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1824 struct vm_struct
*v
;
1826 read_lock(&vmlist_lock
);
1828 while (n
> 0 && v
) {
1839 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1841 struct vm_struct
*v
= p
;
1847 static void s_stop(struct seq_file
*m
, void *p
)
1849 read_unlock(&vmlist_lock
);
1852 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
1855 unsigned int nr
, *counters
= m
->private;
1860 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
1862 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
1863 counters
[page_to_nid(v
->pages
[nr
])]++;
1865 for_each_node_state(nr
, N_HIGH_MEMORY
)
1867 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
1871 static int s_show(struct seq_file
*m
, void *p
)
1873 struct vm_struct
*v
= p
;
1875 seq_printf(m
, "0x%p-0x%p %7ld",
1876 v
->addr
, v
->addr
+ v
->size
, v
->size
);
1879 char buff
[KSYM_SYMBOL_LEN
];
1882 sprint_symbol(buff
, (unsigned long)v
->caller
);
1887 seq_printf(m
, " pages=%d", v
->nr_pages
);
1890 seq_printf(m
, " phys=%lx", v
->phys_addr
);
1892 if (v
->flags
& VM_IOREMAP
)
1893 seq_printf(m
, " ioremap");
1895 if (v
->flags
& VM_ALLOC
)
1896 seq_printf(m
, " vmalloc");
1898 if (v
->flags
& VM_MAP
)
1899 seq_printf(m
, " vmap");
1901 if (v
->flags
& VM_USERMAP
)
1902 seq_printf(m
, " user");
1904 if (v
->flags
& VM_VPAGES
)
1905 seq_printf(m
, " vpages");
1907 show_numa_info(m
, v
);
1912 static const struct seq_operations vmalloc_op
= {
1919 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
1921 unsigned int *ptr
= NULL
;
1925 ptr
= kmalloc(nr_node_ids
* sizeof(unsigned int), GFP_KERNEL
);
1926 ret
= seq_open(file
, &vmalloc_op
);
1928 struct seq_file
*m
= file
->private_data
;
1935 static const struct file_operations proc_vmalloc_operations
= {
1936 .open
= vmalloc_open
,
1938 .llseek
= seq_lseek
,
1939 .release
= seq_release_private
,
1942 static int __init
proc_vmalloc_init(void)
1944 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
1947 module_init(proc_vmalloc_init
);