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/signal.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/notifier.h>
25 #include <linux/rbtree.h>
26 #include <linux/radix-tree.h>
27 #include <linux/rcupdate.h>
28 #include <linux/pfn.h>
29 #include <linux/kmemleak.h>
30 #include <linux/atomic.h>
31 #include <linux/compiler.h>
32 #include <linux/llist.h>
33 #include <linux/bitops.h>
35 #include <linux/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred
{
42 struct llist_head list
;
43 struct work_struct wq
;
45 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct
*w
)
51 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
52 struct llist_node
*llnode
= llist_del_all(&p
->list
);
55 llnode
= llist_next(llnode
);
60 /*** Page table manipulation functions ***/
62 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
66 pte
= pte_offset_kernel(pmd
, addr
);
68 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
69 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
70 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
73 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
78 pmd
= pmd_offset(pud
, addr
);
80 next
= pmd_addr_end(addr
, end
);
81 if (pmd_clear_huge(pmd
))
83 if (pmd_none_or_clear_bad(pmd
))
85 vunmap_pte_range(pmd
, addr
, next
);
86 } while (pmd
++, addr
= next
, addr
!= end
);
89 static void vunmap_pud_range(p4d_t
*p4d
, unsigned long addr
, unsigned long end
)
94 pud
= pud_offset(p4d
, addr
);
96 next
= pud_addr_end(addr
, end
);
97 if (pud_clear_huge(pud
))
99 if (pud_none_or_clear_bad(pud
))
101 vunmap_pmd_range(pud
, addr
, next
);
102 } while (pud
++, addr
= next
, addr
!= end
);
105 static void vunmap_p4d_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
110 p4d
= p4d_offset(pgd
, addr
);
112 next
= p4d_addr_end(addr
, end
);
113 if (p4d_clear_huge(p4d
))
115 if (p4d_none_or_clear_bad(p4d
))
117 vunmap_pud_range(p4d
, addr
, next
);
118 } while (p4d
++, addr
= next
, addr
!= end
);
121 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
127 pgd
= pgd_offset_k(addr
);
129 next
= pgd_addr_end(addr
, end
);
130 if (pgd_none_or_clear_bad(pgd
))
132 vunmap_p4d_range(pgd
, addr
, next
);
133 } while (pgd
++, addr
= next
, addr
!= end
);
136 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
137 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
142 * nr is a running index into the array which helps higher level
143 * callers keep track of where we're up to.
146 pte
= pte_alloc_kernel(pmd
, addr
);
150 struct page
*page
= pages
[*nr
];
152 if (WARN_ON(!pte_none(*pte
)))
156 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
158 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
162 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
163 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
168 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
172 next
= pmd_addr_end(addr
, end
);
173 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
175 } while (pmd
++, addr
= next
, addr
!= end
);
179 static int vmap_pud_range(p4d_t
*p4d
, unsigned long addr
,
180 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
185 pud
= pud_alloc(&init_mm
, p4d
, addr
);
189 next
= pud_addr_end(addr
, end
);
190 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
192 } while (pud
++, addr
= next
, addr
!= end
);
196 static int vmap_p4d_range(pgd_t
*pgd
, unsigned long addr
,
197 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
202 p4d
= p4d_alloc(&init_mm
, pgd
, addr
);
206 next
= p4d_addr_end(addr
, end
);
207 if (vmap_pud_range(p4d
, addr
, next
, prot
, pages
, nr
))
209 } while (p4d
++, addr
= next
, addr
!= end
);
214 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
215 * will have pfns corresponding to the "pages" array.
217 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
219 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
220 pgprot_t prot
, struct page
**pages
)
224 unsigned long addr
= start
;
229 pgd
= pgd_offset_k(addr
);
231 next
= pgd_addr_end(addr
, end
);
232 err
= vmap_p4d_range(pgd
, addr
, next
, prot
, pages
, &nr
);
235 } while (pgd
++, addr
= next
, addr
!= end
);
240 static int vmap_page_range(unsigned long start
, unsigned long end
,
241 pgprot_t prot
, struct page
**pages
)
245 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
246 flush_cache_vmap(start
, end
);
250 int is_vmalloc_or_module_addr(const void *x
)
253 * ARM, x86-64 and sparc64 put modules in a special place,
254 * and fall back on vmalloc() if that fails. Others
255 * just put it in the vmalloc space.
257 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
258 unsigned long addr
= (unsigned long)x
;
259 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
262 return is_vmalloc_addr(x
);
266 * Walk a vmap address to the struct page it maps.
268 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
270 unsigned long addr
= (unsigned long) vmalloc_addr
;
271 struct page
*page
= NULL
;
272 pgd_t
*pgd
= pgd_offset_k(addr
);
279 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
280 * architectures that do not vmalloc module space
282 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
286 p4d
= p4d_offset(pgd
, addr
);
289 pud
= pud_offset(p4d
, addr
);
292 pmd
= pmd_offset(pud
, addr
);
296 ptep
= pte_offset_map(pmd
, addr
);
298 if (pte_present(pte
))
299 page
= pte_page(pte
);
303 EXPORT_SYMBOL(vmalloc_to_page
);
306 * Map a vmalloc()-space virtual address to the physical page frame number.
308 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
310 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
312 EXPORT_SYMBOL(vmalloc_to_pfn
);
315 /*** Global kva allocator ***/
317 #define VM_VM_AREA 0x04
319 static DEFINE_SPINLOCK(vmap_area_lock
);
320 /* Export for kexec only */
321 LIST_HEAD(vmap_area_list
);
322 static LLIST_HEAD(vmap_purge_list
);
323 static struct rb_root vmap_area_root
= RB_ROOT
;
325 /* The vmap cache globals are protected by vmap_area_lock */
326 static struct rb_node
*free_vmap_cache
;
327 static unsigned long cached_hole_size
;
328 static unsigned long cached_vstart
;
329 static unsigned long cached_align
;
331 static unsigned long vmap_area_pcpu_hole
;
333 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
335 struct rb_node
*n
= vmap_area_root
.rb_node
;
338 struct vmap_area
*va
;
340 va
= rb_entry(n
, struct vmap_area
, rb_node
);
341 if (addr
< va
->va_start
)
343 else if (addr
>= va
->va_end
)
352 static void __insert_vmap_area(struct vmap_area
*va
)
354 struct rb_node
**p
= &vmap_area_root
.rb_node
;
355 struct rb_node
*parent
= NULL
;
359 struct vmap_area
*tmp_va
;
362 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
363 if (va
->va_start
< tmp_va
->va_end
)
365 else if (va
->va_end
> tmp_va
->va_start
)
371 rb_link_node(&va
->rb_node
, parent
, p
);
372 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
374 /* address-sort this list */
375 tmp
= rb_prev(&va
->rb_node
);
377 struct vmap_area
*prev
;
378 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
379 list_add_rcu(&va
->list
, &prev
->list
);
381 list_add_rcu(&va
->list
, &vmap_area_list
);
384 static void purge_vmap_area_lazy(void);
386 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list
);
389 * Allocate a region of KVA of the specified size and alignment, within the
392 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
394 unsigned long vstart
, unsigned long vend
,
395 int node
, gfp_t gfp_mask
)
397 struct vmap_area
*va
;
401 struct vmap_area
*first
;
404 BUG_ON(offset_in_page(size
));
405 BUG_ON(!is_power_of_2(align
));
409 va
= kmalloc_node(sizeof(struct vmap_area
),
410 gfp_mask
& GFP_RECLAIM_MASK
, node
);
412 return ERR_PTR(-ENOMEM
);
415 * Only scan the relevant parts containing pointers to other objects
416 * to avoid false negatives.
418 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
421 spin_lock(&vmap_area_lock
);
423 * Invalidate cache if we have more permissive parameters.
424 * cached_hole_size notes the largest hole noticed _below_
425 * the vmap_area cached in free_vmap_cache: if size fits
426 * into that hole, we want to scan from vstart to reuse
427 * the hole instead of allocating above free_vmap_cache.
428 * Note that __free_vmap_area may update free_vmap_cache
429 * without updating cached_hole_size or cached_align.
431 if (!free_vmap_cache
||
432 size
< cached_hole_size
||
433 vstart
< cached_vstart
||
434 align
< cached_align
) {
436 cached_hole_size
= 0;
437 free_vmap_cache
= NULL
;
439 /* record if we encounter less permissive parameters */
440 cached_vstart
= vstart
;
441 cached_align
= align
;
443 /* find starting point for our search */
444 if (free_vmap_cache
) {
445 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
446 addr
= ALIGN(first
->va_end
, align
);
449 if (addr
+ size
< addr
)
453 addr
= ALIGN(vstart
, align
);
454 if (addr
+ size
< addr
)
457 n
= vmap_area_root
.rb_node
;
461 struct vmap_area
*tmp
;
462 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
463 if (tmp
->va_end
>= addr
) {
465 if (tmp
->va_start
<= addr
)
476 /* from the starting point, walk areas until a suitable hole is found */
477 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
478 if (addr
+ cached_hole_size
< first
->va_start
)
479 cached_hole_size
= first
->va_start
- addr
;
480 addr
= ALIGN(first
->va_end
, align
);
481 if (addr
+ size
< addr
)
484 if (list_is_last(&first
->list
, &vmap_area_list
))
487 first
= list_next_entry(first
, list
);
491 if (addr
+ size
> vend
)
495 va
->va_end
= addr
+ size
;
497 __insert_vmap_area(va
);
498 free_vmap_cache
= &va
->rb_node
;
499 spin_unlock(&vmap_area_lock
);
501 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
502 BUG_ON(va
->va_start
< vstart
);
503 BUG_ON(va
->va_end
> vend
);
508 spin_unlock(&vmap_area_lock
);
510 purge_vmap_area_lazy();
515 if (gfpflags_allow_blocking(gfp_mask
)) {
516 unsigned long freed
= 0;
517 blocking_notifier_call_chain(&vmap_notify_list
, 0, &freed
);
524 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit())
525 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
528 return ERR_PTR(-EBUSY
);
531 int register_vmap_purge_notifier(struct notifier_block
*nb
)
533 return blocking_notifier_chain_register(&vmap_notify_list
, nb
);
535 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier
);
537 int unregister_vmap_purge_notifier(struct notifier_block
*nb
)
539 return blocking_notifier_chain_unregister(&vmap_notify_list
, nb
);
541 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier
);
543 static void __free_vmap_area(struct vmap_area
*va
)
545 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
547 if (free_vmap_cache
) {
548 if (va
->va_end
< cached_vstart
) {
549 free_vmap_cache
= NULL
;
551 struct vmap_area
*cache
;
552 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
553 if (va
->va_start
<= cache
->va_start
) {
554 free_vmap_cache
= rb_prev(&va
->rb_node
);
556 * We don't try to update cached_hole_size or
557 * cached_align, but it won't go very wrong.
562 rb_erase(&va
->rb_node
, &vmap_area_root
);
563 RB_CLEAR_NODE(&va
->rb_node
);
564 list_del_rcu(&va
->list
);
567 * Track the highest possible candidate for pcpu area
568 * allocation. Areas outside of vmalloc area can be returned
569 * here too, consider only end addresses which fall inside
570 * vmalloc area proper.
572 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
573 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
575 kfree_rcu(va
, rcu_head
);
579 * Free a region of KVA allocated by alloc_vmap_area
581 static void free_vmap_area(struct vmap_area
*va
)
583 spin_lock(&vmap_area_lock
);
584 __free_vmap_area(va
);
585 spin_unlock(&vmap_area_lock
);
589 * Clear the pagetable entries of a given vmap_area
591 static void unmap_vmap_area(struct vmap_area
*va
)
593 vunmap_page_range(va
->va_start
, va
->va_end
);
596 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
599 * Unmap page tables and force a TLB flush immediately if pagealloc
600 * debugging is enabled. This catches use after free bugs similarly to
601 * those in linear kernel virtual address space after a page has been
604 * All the lazy freeing logic is still retained, in order to minimise
605 * intrusiveness of this debugging feature.
607 * This is going to be *slow* (linear kernel virtual address debugging
608 * doesn't do a broadcast TLB flush so it is a lot faster).
610 if (debug_pagealloc_enabled()) {
611 vunmap_page_range(start
, end
);
612 flush_tlb_kernel_range(start
, end
);
617 * lazy_max_pages is the maximum amount of virtual address space we gather up
618 * before attempting to purge with a TLB flush.
620 * There is a tradeoff here: a larger number will cover more kernel page tables
621 * and take slightly longer to purge, but it will linearly reduce the number of
622 * global TLB flushes that must be performed. It would seem natural to scale
623 * this number up linearly with the number of CPUs (because vmapping activity
624 * could also scale linearly with the number of CPUs), however it is likely
625 * that in practice, workloads might be constrained in other ways that mean
626 * vmap activity will not scale linearly with CPUs. Also, I want to be
627 * conservative and not introduce a big latency on huge systems, so go with
628 * a less aggressive log scale. It will still be an improvement over the old
629 * code, and it will be simple to change the scale factor if we find that it
630 * becomes a problem on bigger systems.
632 static unsigned long lazy_max_pages(void)
636 log
= fls(num_online_cpus());
638 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
641 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
644 * Serialize vmap purging. There is no actual criticial section protected
645 * by this look, but we want to avoid concurrent calls for performance
646 * reasons and to make the pcpu_get_vm_areas more deterministic.
648 static DEFINE_MUTEX(vmap_purge_lock
);
650 /* for per-CPU blocks */
651 static void purge_fragmented_blocks_allcpus(void);
654 * called before a call to iounmap() if the caller wants vm_area_struct's
657 void set_iounmap_nonlazy(void)
659 atomic_set(&vmap_lazy_nr
, lazy_max_pages()+1);
663 * Purges all lazily-freed vmap areas.
665 static bool __purge_vmap_area_lazy(unsigned long start
, unsigned long end
)
667 struct llist_node
*valist
;
668 struct vmap_area
*va
;
669 struct vmap_area
*n_va
;
670 bool do_free
= false;
672 lockdep_assert_held(&vmap_purge_lock
);
674 valist
= llist_del_all(&vmap_purge_list
);
675 llist_for_each_entry(va
, valist
, purge_list
) {
676 if (va
->va_start
< start
)
677 start
= va
->va_start
;
678 if (va
->va_end
> end
)
686 flush_tlb_kernel_range(start
, end
);
688 spin_lock(&vmap_area_lock
);
689 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
) {
690 int nr
= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
692 __free_vmap_area(va
);
693 atomic_sub(nr
, &vmap_lazy_nr
);
694 cond_resched_lock(&vmap_area_lock
);
696 spin_unlock(&vmap_area_lock
);
701 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
702 * is already purging.
704 static void try_purge_vmap_area_lazy(void)
706 if (mutex_trylock(&vmap_purge_lock
)) {
707 __purge_vmap_area_lazy(ULONG_MAX
, 0);
708 mutex_unlock(&vmap_purge_lock
);
713 * Kick off a purge of the outstanding lazy areas.
715 static void purge_vmap_area_lazy(void)
717 mutex_lock(&vmap_purge_lock
);
718 purge_fragmented_blocks_allcpus();
719 __purge_vmap_area_lazy(ULONG_MAX
, 0);
720 mutex_unlock(&vmap_purge_lock
);
724 * Free a vmap area, caller ensuring that the area has been unmapped
725 * and flush_cache_vunmap had been called for the correct range
728 static void free_vmap_area_noflush(struct vmap_area
*va
)
732 nr_lazy
= atomic_add_return((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
,
735 /* After this point, we may free va at any time */
736 llist_add(&va
->purge_list
, &vmap_purge_list
);
738 if (unlikely(nr_lazy
> lazy_max_pages()))
739 try_purge_vmap_area_lazy();
743 * Free and unmap a vmap area
745 static void free_unmap_vmap_area(struct vmap_area
*va
)
747 flush_cache_vunmap(va
->va_start
, va
->va_end
);
749 free_vmap_area_noflush(va
);
752 static struct vmap_area
*find_vmap_area(unsigned long addr
)
754 struct vmap_area
*va
;
756 spin_lock(&vmap_area_lock
);
757 va
= __find_vmap_area(addr
);
758 spin_unlock(&vmap_area_lock
);
763 /*** Per cpu kva allocator ***/
766 * vmap space is limited especially on 32 bit architectures. Ensure there is
767 * room for at least 16 percpu vmap blocks per CPU.
770 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
771 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
772 * instead (we just need a rough idea)
774 #if BITS_PER_LONG == 32
775 #define VMALLOC_SPACE (128UL*1024*1024)
777 #define VMALLOC_SPACE (128UL*1024*1024*1024)
780 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
781 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
782 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
783 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
784 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
785 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
786 #define VMAP_BBMAP_BITS \
787 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
788 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
789 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
791 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
793 static bool vmap_initialized __read_mostly
= false;
795 struct vmap_block_queue
{
797 struct list_head free
;
802 struct vmap_area
*va
;
803 unsigned long free
, dirty
;
804 unsigned long dirty_min
, dirty_max
; /*< dirty range */
805 struct list_head free_list
;
806 struct rcu_head rcu_head
;
807 struct list_head purge
;
810 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
811 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
814 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
815 * in the free path. Could get rid of this if we change the API to return a
816 * "cookie" from alloc, to be passed to free. But no big deal yet.
818 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
819 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
822 * We should probably have a fallback mechanism to allocate virtual memory
823 * out of partially filled vmap blocks. However vmap block sizing should be
824 * fairly reasonable according to the vmalloc size, so it shouldn't be a
828 static unsigned long addr_to_vb_idx(unsigned long addr
)
830 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
831 addr
/= VMAP_BLOCK_SIZE
;
835 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
839 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
840 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
845 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
846 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
847 * @order: how many 2^order pages should be occupied in newly allocated block
848 * @gfp_mask: flags for the page level allocator
850 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
852 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
854 struct vmap_block_queue
*vbq
;
855 struct vmap_block
*vb
;
856 struct vmap_area
*va
;
857 unsigned long vb_idx
;
861 node
= numa_node_id();
863 vb
= kmalloc_node(sizeof(struct vmap_block
),
864 gfp_mask
& GFP_RECLAIM_MASK
, node
);
866 return ERR_PTR(-ENOMEM
);
868 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
869 VMALLOC_START
, VMALLOC_END
,
876 err
= radix_tree_preload(gfp_mask
);
883 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
884 spin_lock_init(&vb
->lock
);
886 /* At least something should be left free */
887 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
888 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
890 vb
->dirty_min
= VMAP_BBMAP_BITS
;
892 INIT_LIST_HEAD(&vb
->free_list
);
894 vb_idx
= addr_to_vb_idx(va
->va_start
);
895 spin_lock(&vmap_block_tree_lock
);
896 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
897 spin_unlock(&vmap_block_tree_lock
);
899 radix_tree_preload_end();
901 vbq
= &get_cpu_var(vmap_block_queue
);
902 spin_lock(&vbq
->lock
);
903 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
904 spin_unlock(&vbq
->lock
);
905 put_cpu_var(vmap_block_queue
);
910 static void free_vmap_block(struct vmap_block
*vb
)
912 struct vmap_block
*tmp
;
913 unsigned long vb_idx
;
915 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
916 spin_lock(&vmap_block_tree_lock
);
917 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
918 spin_unlock(&vmap_block_tree_lock
);
921 free_vmap_area_noflush(vb
->va
);
922 kfree_rcu(vb
, rcu_head
);
925 static void purge_fragmented_blocks(int cpu
)
928 struct vmap_block
*vb
;
929 struct vmap_block
*n_vb
;
930 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
933 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
935 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
938 spin_lock(&vb
->lock
);
939 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
940 vb
->free
= 0; /* prevent further allocs after releasing lock */
941 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
943 vb
->dirty_max
= VMAP_BBMAP_BITS
;
944 spin_lock(&vbq
->lock
);
945 list_del_rcu(&vb
->free_list
);
946 spin_unlock(&vbq
->lock
);
947 spin_unlock(&vb
->lock
);
948 list_add_tail(&vb
->purge
, &purge
);
950 spin_unlock(&vb
->lock
);
954 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
955 list_del(&vb
->purge
);
960 static void purge_fragmented_blocks_allcpus(void)
964 for_each_possible_cpu(cpu
)
965 purge_fragmented_blocks(cpu
);
968 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
970 struct vmap_block_queue
*vbq
;
971 struct vmap_block
*vb
;
975 BUG_ON(offset_in_page(size
));
976 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
977 if (WARN_ON(size
== 0)) {
979 * Allocating 0 bytes isn't what caller wants since
980 * get_order(0) returns funny result. Just warn and terminate
985 order
= get_order(size
);
988 vbq
= &get_cpu_var(vmap_block_queue
);
989 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
990 unsigned long pages_off
;
992 spin_lock(&vb
->lock
);
993 if (vb
->free
< (1UL << order
)) {
994 spin_unlock(&vb
->lock
);
998 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
999 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
1000 vb
->free
-= 1UL << order
;
1001 if (vb
->free
== 0) {
1002 spin_lock(&vbq
->lock
);
1003 list_del_rcu(&vb
->free_list
);
1004 spin_unlock(&vbq
->lock
);
1007 spin_unlock(&vb
->lock
);
1011 put_cpu_var(vmap_block_queue
);
1014 /* Allocate new block if nothing was found */
1016 vaddr
= new_vmap_block(order
, gfp_mask
);
1021 static void vb_free(const void *addr
, unsigned long size
)
1023 unsigned long offset
;
1024 unsigned long vb_idx
;
1026 struct vmap_block
*vb
;
1028 BUG_ON(offset_in_page(size
));
1029 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1031 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
1033 order
= get_order(size
);
1035 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
1036 offset
>>= PAGE_SHIFT
;
1038 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
1040 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1044 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1046 spin_lock(&vb
->lock
);
1048 /* Expand dirty range */
1049 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1050 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1052 vb
->dirty
+= 1UL << order
;
1053 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1055 spin_unlock(&vb
->lock
);
1056 free_vmap_block(vb
);
1058 spin_unlock(&vb
->lock
);
1062 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1064 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1065 * to amortize TLB flushing overheads. What this means is that any page you
1066 * have now, may, in a former life, have been mapped into kernel virtual
1067 * address by the vmap layer and so there might be some CPUs with TLB entries
1068 * still referencing that page (additional to the regular 1:1 kernel mapping).
1070 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1071 * be sure that none of the pages we have control over will have any aliases
1072 * from the vmap layer.
1074 void vm_unmap_aliases(void)
1076 unsigned long start
= ULONG_MAX
, end
= 0;
1080 if (unlikely(!vmap_initialized
))
1085 for_each_possible_cpu(cpu
) {
1086 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1087 struct vmap_block
*vb
;
1090 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1091 spin_lock(&vb
->lock
);
1093 unsigned long va_start
= vb
->va
->va_start
;
1096 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1097 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1099 start
= min(s
, start
);
1104 spin_unlock(&vb
->lock
);
1109 mutex_lock(&vmap_purge_lock
);
1110 purge_fragmented_blocks_allcpus();
1111 if (!__purge_vmap_area_lazy(start
, end
) && flush
)
1112 flush_tlb_kernel_range(start
, end
);
1113 mutex_unlock(&vmap_purge_lock
);
1115 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1118 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1119 * @mem: the pointer returned by vm_map_ram
1120 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1122 void vm_unmap_ram(const void *mem
, unsigned int count
)
1124 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1125 unsigned long addr
= (unsigned long)mem
;
1126 struct vmap_area
*va
;
1130 BUG_ON(addr
< VMALLOC_START
);
1131 BUG_ON(addr
> VMALLOC_END
);
1132 BUG_ON(!PAGE_ALIGNED(addr
));
1134 debug_check_no_locks_freed(mem
, size
);
1135 vmap_debug_free_range(addr
, addr
+size
);
1137 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1142 va
= find_vmap_area(addr
);
1144 free_unmap_vmap_area(va
);
1146 EXPORT_SYMBOL(vm_unmap_ram
);
1149 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1150 * @pages: an array of pointers to the pages to be mapped
1151 * @count: number of pages
1152 * @node: prefer to allocate data structures on this node
1153 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1155 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1156 * faster than vmap so it's good. But if you mix long-life and short-life
1157 * objects with vm_map_ram(), it could consume lots of address space through
1158 * fragmentation (especially on a 32bit machine). You could see failures in
1159 * the end. Please use this function for short-lived objects.
1161 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1163 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1165 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1169 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1170 mem
= vb_alloc(size
, GFP_KERNEL
);
1173 addr
= (unsigned long)mem
;
1175 struct vmap_area
*va
;
1176 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1177 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1181 addr
= va
->va_start
;
1184 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1185 vm_unmap_ram(mem
, count
);
1190 EXPORT_SYMBOL(vm_map_ram
);
1192 static struct vm_struct
*vmlist __initdata
;
1194 * vm_area_add_early - add vmap area early during boot
1195 * @vm: vm_struct to add
1197 * This function is used to add fixed kernel vm area to vmlist before
1198 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1199 * should contain proper values and the other fields should be zero.
1201 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1203 void __init
vm_area_add_early(struct vm_struct
*vm
)
1205 struct vm_struct
*tmp
, **p
;
1207 BUG_ON(vmap_initialized
);
1208 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1209 if (tmp
->addr
>= vm
->addr
) {
1210 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1213 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1220 * vm_area_register_early - register vmap area early during boot
1221 * @vm: vm_struct to register
1222 * @align: requested alignment
1224 * This function is used to register kernel vm area before
1225 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1226 * proper values on entry and other fields should be zero. On return,
1227 * vm->addr contains the allocated address.
1229 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1231 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1233 static size_t vm_init_off __initdata
;
1236 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1237 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1239 vm
->addr
= (void *)addr
;
1241 vm_area_add_early(vm
);
1244 void __init
vmalloc_init(void)
1246 struct vmap_area
*va
;
1247 struct vm_struct
*tmp
;
1250 for_each_possible_cpu(i
) {
1251 struct vmap_block_queue
*vbq
;
1252 struct vfree_deferred
*p
;
1254 vbq
= &per_cpu(vmap_block_queue
, i
);
1255 spin_lock_init(&vbq
->lock
);
1256 INIT_LIST_HEAD(&vbq
->free
);
1257 p
= &per_cpu(vfree_deferred
, i
);
1258 init_llist_head(&p
->list
);
1259 INIT_WORK(&p
->wq
, free_work
);
1262 /* Import existing vmlist entries. */
1263 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1264 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1265 va
->flags
= VM_VM_AREA
;
1266 va
->va_start
= (unsigned long)tmp
->addr
;
1267 va
->va_end
= va
->va_start
+ tmp
->size
;
1269 __insert_vmap_area(va
);
1272 vmap_area_pcpu_hole
= VMALLOC_END
;
1274 vmap_initialized
= true;
1278 * map_kernel_range_noflush - map kernel VM area with the specified pages
1279 * @addr: start of the VM area to map
1280 * @size: size of the VM area to map
1281 * @prot: page protection flags to use
1282 * @pages: pages to map
1284 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1285 * specify should have been allocated using get_vm_area() and its
1289 * This function does NOT do any cache flushing. The caller is
1290 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1291 * before calling this function.
1294 * The number of pages mapped on success, -errno on failure.
1296 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1297 pgprot_t prot
, struct page
**pages
)
1299 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1303 * unmap_kernel_range_noflush - unmap kernel VM area
1304 * @addr: start of the VM area to unmap
1305 * @size: size of the VM area to unmap
1307 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1308 * specify should have been allocated using get_vm_area() and its
1312 * This function does NOT do any cache flushing. The caller is
1313 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1314 * before calling this function and flush_tlb_kernel_range() after.
1316 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1318 vunmap_page_range(addr
, addr
+ size
);
1320 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1323 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1324 * @addr: start of the VM area to unmap
1325 * @size: size of the VM area to unmap
1327 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1328 * the unmapping and tlb after.
1330 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1332 unsigned long end
= addr
+ size
;
1334 flush_cache_vunmap(addr
, end
);
1335 vunmap_page_range(addr
, end
);
1336 flush_tlb_kernel_range(addr
, end
);
1338 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
1340 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
1342 unsigned long addr
= (unsigned long)area
->addr
;
1343 unsigned long end
= addr
+ get_vm_area_size(area
);
1346 err
= vmap_page_range(addr
, end
, prot
, pages
);
1348 return err
> 0 ? 0 : err
;
1350 EXPORT_SYMBOL_GPL(map_vm_area
);
1352 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1353 unsigned long flags
, const void *caller
)
1355 spin_lock(&vmap_area_lock
);
1357 vm
->addr
= (void *)va
->va_start
;
1358 vm
->size
= va
->va_end
- va
->va_start
;
1359 vm
->caller
= caller
;
1361 va
->flags
|= VM_VM_AREA
;
1362 spin_unlock(&vmap_area_lock
);
1365 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1368 * Before removing VM_UNINITIALIZED,
1369 * we should make sure that vm has proper values.
1370 * Pair with smp_rmb() in show_numa_info().
1373 vm
->flags
&= ~VM_UNINITIALIZED
;
1376 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1377 unsigned long align
, unsigned long flags
, unsigned long start
,
1378 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1380 struct vmap_area
*va
;
1381 struct vm_struct
*area
;
1383 BUG_ON(in_interrupt());
1384 size
= PAGE_ALIGN(size
);
1385 if (unlikely(!size
))
1388 if (flags
& VM_IOREMAP
)
1389 align
= 1ul << clamp_t(int, get_count_order_long(size
),
1390 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1392 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1393 if (unlikely(!area
))
1396 if (!(flags
& VM_NO_GUARD
))
1399 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1405 setup_vmalloc_vm(area
, va
, flags
, caller
);
1410 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1411 unsigned long start
, unsigned long end
)
1413 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1414 GFP_KERNEL
, __builtin_return_address(0));
1416 EXPORT_SYMBOL_GPL(__get_vm_area
);
1418 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1419 unsigned long start
, unsigned long end
,
1422 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1423 GFP_KERNEL
, caller
);
1427 * get_vm_area - reserve a contiguous kernel virtual area
1428 * @size: size of the area
1429 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1431 * Search an area of @size in the kernel virtual mapping area,
1432 * and reserved it for out purposes. Returns the area descriptor
1433 * on success or %NULL on failure.
1435 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1437 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1438 NUMA_NO_NODE
, GFP_KERNEL
,
1439 __builtin_return_address(0));
1442 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1445 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1446 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1450 * find_vm_area - find a continuous kernel virtual area
1451 * @addr: base address
1453 * Search for the kernel VM area starting at @addr, and return it.
1454 * It is up to the caller to do all required locking to keep the returned
1457 struct vm_struct
*find_vm_area(const void *addr
)
1459 struct vmap_area
*va
;
1461 va
= find_vmap_area((unsigned long)addr
);
1462 if (va
&& va
->flags
& VM_VM_AREA
)
1469 * remove_vm_area - find and remove a continuous kernel virtual area
1470 * @addr: base address
1472 * Search for the kernel VM area starting at @addr, and remove it.
1473 * This function returns the found VM area, but using it is NOT safe
1474 * on SMP machines, except for its size or flags.
1476 struct vm_struct
*remove_vm_area(const void *addr
)
1478 struct vmap_area
*va
;
1482 va
= find_vmap_area((unsigned long)addr
);
1483 if (va
&& va
->flags
& VM_VM_AREA
) {
1484 struct vm_struct
*vm
= va
->vm
;
1486 spin_lock(&vmap_area_lock
);
1488 va
->flags
&= ~VM_VM_AREA
;
1489 spin_unlock(&vmap_area_lock
);
1491 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1492 kasan_free_shadow(vm
);
1493 free_unmap_vmap_area(va
);
1500 static void __vunmap(const void *addr
, int deallocate_pages
)
1502 struct vm_struct
*area
;
1507 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
1511 area
= remove_vm_area(addr
);
1512 if (unlikely(!area
)) {
1513 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1518 debug_check_no_locks_freed(addr
, get_vm_area_size(area
));
1519 debug_check_no_obj_freed(addr
, get_vm_area_size(area
));
1521 if (deallocate_pages
) {
1524 for (i
= 0; i
< area
->nr_pages
; i
++) {
1525 struct page
*page
= area
->pages
[i
];
1528 __free_pages(page
, 0);
1531 kvfree(area
->pages
);
1538 static inline void __vfree_deferred(const void *addr
)
1541 * Use raw_cpu_ptr() because this can be called from preemptible
1542 * context. Preemption is absolutely fine here, because the llist_add()
1543 * implementation is lockless, so it works even if we are adding to
1544 * nother cpu's list. schedule_work() should be fine with this too.
1546 struct vfree_deferred
*p
= raw_cpu_ptr(&vfree_deferred
);
1548 if (llist_add((struct llist_node
*)addr
, &p
->list
))
1549 schedule_work(&p
->wq
);
1553 * vfree_atomic - release memory allocated by vmalloc()
1554 * @addr: memory base address
1556 * This one is just like vfree() but can be called in any atomic context
1559 void vfree_atomic(const void *addr
)
1563 kmemleak_free(addr
);
1567 __vfree_deferred(addr
);
1571 * vfree - release memory allocated by vmalloc()
1572 * @addr: memory base address
1574 * Free the virtually continuous memory area starting at @addr, as
1575 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1576 * NULL, no operation is performed.
1578 * Must not be called in NMI context (strictly speaking, only if we don't
1579 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1580 * conventions for vfree() arch-depenedent would be a really bad idea)
1582 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1584 void vfree(const void *addr
)
1588 kmemleak_free(addr
);
1592 if (unlikely(in_interrupt()))
1593 __vfree_deferred(addr
);
1597 EXPORT_SYMBOL(vfree
);
1600 * vunmap - release virtual mapping obtained by vmap()
1601 * @addr: memory base address
1603 * Free the virtually contiguous memory area starting at @addr,
1604 * which was created from the page array passed to vmap().
1606 * Must not be called in interrupt context.
1608 void vunmap(const void *addr
)
1610 BUG_ON(in_interrupt());
1615 EXPORT_SYMBOL(vunmap
);
1618 * vmap - map an array of pages into virtually contiguous space
1619 * @pages: array of page pointers
1620 * @count: number of pages to map
1621 * @flags: vm_area->flags
1622 * @prot: page protection for the mapping
1624 * Maps @count pages from @pages into contiguous kernel virtual
1627 void *vmap(struct page
**pages
, unsigned int count
,
1628 unsigned long flags
, pgprot_t prot
)
1630 struct vm_struct
*area
;
1631 unsigned long size
; /* In bytes */
1635 if (count
> totalram_pages
)
1638 size
= (unsigned long)count
<< PAGE_SHIFT
;
1639 area
= get_vm_area_caller(size
, flags
, __builtin_return_address(0));
1643 if (map_vm_area(area
, prot
, pages
)) {
1650 EXPORT_SYMBOL(vmap
);
1652 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1653 pgprot_t prot
, int node
)
1655 struct page
**pages
;
1656 unsigned int nr_pages
, array_size
, i
;
1657 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1658 const gfp_t alloc_mask
= gfp_mask
| __GFP_HIGHMEM
| __GFP_NOWARN
;
1660 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
1661 array_size
= (nr_pages
* sizeof(struct page
*));
1663 area
->nr_pages
= nr_pages
;
1664 /* Please note that the recursion is strictly bounded. */
1665 if (array_size
> PAGE_SIZE
) {
1666 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1667 PAGE_KERNEL
, node
, area
->caller
);
1669 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1671 area
->pages
= pages
;
1673 remove_vm_area(area
->addr
);
1678 for (i
= 0; i
< area
->nr_pages
; i
++) {
1681 if (fatal_signal_pending(current
)) {
1686 if (node
== NUMA_NO_NODE
)
1687 page
= alloc_page(alloc_mask
);
1689 page
= alloc_pages_node(node
, alloc_mask
, 0);
1691 if (unlikely(!page
)) {
1692 /* Successfully allocated i pages, free them in __vunmap() */
1696 area
->pages
[i
] = page
;
1697 if (gfpflags_allow_blocking(gfp_mask
))
1701 if (map_vm_area(area
, prot
, pages
))
1706 warn_alloc(gfp_mask
, NULL
,
1707 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1708 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1715 * __vmalloc_node_range - allocate virtually contiguous memory
1716 * @size: allocation size
1717 * @align: desired alignment
1718 * @start: vm area range start
1719 * @end: vm area range end
1720 * @gfp_mask: flags for the page level allocator
1721 * @prot: protection mask for the allocated pages
1722 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1723 * @node: node to use for allocation or NUMA_NO_NODE
1724 * @caller: caller's return address
1726 * Allocate enough pages to cover @size from the page level
1727 * allocator with @gfp_mask flags. Map them into contiguous
1728 * kernel virtual space, using a pagetable protection of @prot.
1730 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1731 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1732 pgprot_t prot
, unsigned long vm_flags
, int node
,
1735 struct vm_struct
*area
;
1737 unsigned long real_size
= size
;
1739 size
= PAGE_ALIGN(size
);
1740 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1743 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
1744 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
1748 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
1753 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1754 * flag. It means that vm_struct is not fully initialized.
1755 * Now, it is fully initialized, so remove this flag here.
1757 clear_vm_uninitialized_flag(area
);
1760 * A ref_count = 2 is needed because vm_struct allocated in
1761 * __get_vm_area_node() contains a reference to the virtual address of
1762 * the vmalloc'ed block.
1764 kmemleak_alloc(addr
, real_size
, 2, gfp_mask
);
1769 warn_alloc(gfp_mask
, NULL
,
1770 "vmalloc: allocation failure: %lu bytes", real_size
);
1775 * __vmalloc_node - allocate virtually contiguous memory
1776 * @size: allocation size
1777 * @align: desired alignment
1778 * @gfp_mask: flags for the page level allocator
1779 * @prot: protection mask for the allocated pages
1780 * @node: node to use for allocation or NUMA_NO_NODE
1781 * @caller: caller's return address
1783 * Allocate enough pages to cover @size from the page level
1784 * allocator with @gfp_mask flags. Map them into contiguous
1785 * kernel virtual space, using a pagetable protection of @prot.
1787 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_REPEAT
1788 * and __GFP_NOFAIL are not supported
1790 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1794 void *__vmalloc_node(unsigned long size
, unsigned long align
,
1795 gfp_t gfp_mask
, pgprot_t prot
,
1796 int node
, const void *caller
)
1798 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1799 gfp_mask
, prot
, 0, node
, caller
);
1802 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1804 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1805 __builtin_return_address(0));
1807 EXPORT_SYMBOL(__vmalloc
);
1810 * vmalloc - allocate virtually contiguous memory
1811 * @size: allocation size
1812 * Allocate enough pages to cover @size from the page level
1813 * allocator and map them into contiguous kernel virtual space.
1815 * For tight control over page level allocator and protection flags
1816 * use __vmalloc() instead.
1818 void *vmalloc(unsigned long size
)
1820 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1823 EXPORT_SYMBOL(vmalloc
);
1826 * vzalloc - allocate virtually contiguous memory with zero fill
1827 * @size: allocation size
1828 * Allocate enough pages to cover @size from the page level
1829 * allocator and map them into contiguous kernel virtual space.
1830 * The memory allocated is set to zero.
1832 * For tight control over page level allocator and protection flags
1833 * use __vmalloc() instead.
1835 void *vzalloc(unsigned long size
)
1837 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1838 GFP_KERNEL
| __GFP_ZERO
);
1840 EXPORT_SYMBOL(vzalloc
);
1843 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1844 * @size: allocation size
1846 * The resulting memory area is zeroed so it can be mapped to userspace
1847 * without leaking data.
1849 void *vmalloc_user(unsigned long size
)
1851 struct vm_struct
*area
;
1854 ret
= __vmalloc_node(size
, SHMLBA
,
1855 GFP_KERNEL
| __GFP_ZERO
,
1856 PAGE_KERNEL
, NUMA_NO_NODE
,
1857 __builtin_return_address(0));
1859 area
= find_vm_area(ret
);
1860 area
->flags
|= VM_USERMAP
;
1864 EXPORT_SYMBOL(vmalloc_user
);
1867 * vmalloc_node - allocate memory on a specific node
1868 * @size: allocation size
1871 * Allocate enough pages to cover @size from the page level
1872 * allocator and map them into contiguous kernel virtual space.
1874 * For tight control over page level allocator and protection flags
1875 * use __vmalloc() instead.
1877 void *vmalloc_node(unsigned long size
, int node
)
1879 return __vmalloc_node(size
, 1, GFP_KERNEL
, PAGE_KERNEL
,
1880 node
, __builtin_return_address(0));
1882 EXPORT_SYMBOL(vmalloc_node
);
1885 * vzalloc_node - allocate memory on a specific node with zero fill
1886 * @size: allocation size
1889 * Allocate enough pages to cover @size from the page level
1890 * allocator and map them into contiguous kernel virtual space.
1891 * The memory allocated is set to zero.
1893 * For tight control over page level allocator and protection flags
1894 * use __vmalloc_node() instead.
1896 void *vzalloc_node(unsigned long size
, int node
)
1898 return __vmalloc_node_flags(size
, node
,
1899 GFP_KERNEL
| __GFP_ZERO
);
1901 EXPORT_SYMBOL(vzalloc_node
);
1903 #ifndef PAGE_KERNEL_EXEC
1904 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1908 * vmalloc_exec - allocate virtually contiguous, executable memory
1909 * @size: allocation size
1911 * Kernel-internal function to allocate enough pages to cover @size
1912 * the page level allocator and map them into contiguous and
1913 * executable kernel virtual space.
1915 * For tight control over page level allocator and protection flags
1916 * use __vmalloc() instead.
1919 void *vmalloc_exec(unsigned long size
)
1921 return __vmalloc_node(size
, 1, GFP_KERNEL
, PAGE_KERNEL_EXEC
,
1922 NUMA_NO_NODE
, __builtin_return_address(0));
1925 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1926 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1927 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1928 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1930 #define GFP_VMALLOC32 GFP_KERNEL
1934 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1935 * @size: allocation size
1937 * Allocate enough 32bit PA addressable pages to cover @size from the
1938 * page level allocator and map them into contiguous kernel virtual space.
1940 void *vmalloc_32(unsigned long size
)
1942 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1943 NUMA_NO_NODE
, __builtin_return_address(0));
1945 EXPORT_SYMBOL(vmalloc_32
);
1948 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1949 * @size: allocation size
1951 * The resulting memory area is 32bit addressable and zeroed so it can be
1952 * mapped to userspace without leaking data.
1954 void *vmalloc_32_user(unsigned long size
)
1956 struct vm_struct
*area
;
1959 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1960 NUMA_NO_NODE
, __builtin_return_address(0));
1962 area
= find_vm_area(ret
);
1963 area
->flags
|= VM_USERMAP
;
1967 EXPORT_SYMBOL(vmalloc_32_user
);
1970 * small helper routine , copy contents to buf from addr.
1971 * If the page is not present, fill zero.
1974 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1980 unsigned long offset
, length
;
1982 offset
= offset_in_page(addr
);
1983 length
= PAGE_SIZE
- offset
;
1986 p
= vmalloc_to_page(addr
);
1988 * To do safe access to this _mapped_ area, we need
1989 * lock. But adding lock here means that we need to add
1990 * overhead of vmalloc()/vfree() calles for this _debug_
1991 * interface, rarely used. Instead of that, we'll use
1992 * kmap() and get small overhead in this access function.
1996 * we can expect USER0 is not used (see vread/vwrite's
1997 * function description)
1999 void *map
= kmap_atomic(p
);
2000 memcpy(buf
, map
+ offset
, length
);
2003 memset(buf
, 0, length
);
2013 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
2019 unsigned long offset
, length
;
2021 offset
= offset_in_page(addr
);
2022 length
= PAGE_SIZE
- offset
;
2025 p
= vmalloc_to_page(addr
);
2027 * To do safe access to this _mapped_ area, we need
2028 * lock. But adding lock here means that we need to add
2029 * overhead of vmalloc()/vfree() calles for this _debug_
2030 * interface, rarely used. Instead of that, we'll use
2031 * kmap() and get small overhead in this access function.
2035 * we can expect USER0 is not used (see vread/vwrite's
2036 * function description)
2038 void *map
= kmap_atomic(p
);
2039 memcpy(map
+ offset
, buf
, length
);
2051 * vread() - read vmalloc area in a safe way.
2052 * @buf: buffer for reading data
2053 * @addr: vm address.
2054 * @count: number of bytes to be read.
2056 * Returns # of bytes which addr and buf should be increased.
2057 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2058 * includes any intersect with alive vmalloc area.
2060 * This function checks that addr is a valid vmalloc'ed area, and
2061 * copy data from that area to a given buffer. If the given memory range
2062 * of [addr...addr+count) includes some valid address, data is copied to
2063 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2064 * IOREMAP area is treated as memory hole and no copy is done.
2066 * If [addr...addr+count) doesn't includes any intersects with alive
2067 * vm_struct area, returns 0. @buf should be kernel's buffer.
2069 * Note: In usual ops, vread() is never necessary because the caller
2070 * should know vmalloc() area is valid and can use memcpy().
2071 * This is for routines which have to access vmalloc area without
2072 * any informaion, as /dev/kmem.
2076 long vread(char *buf
, char *addr
, unsigned long count
)
2078 struct vmap_area
*va
;
2079 struct vm_struct
*vm
;
2080 char *vaddr
, *buf_start
= buf
;
2081 unsigned long buflen
= count
;
2084 /* Don't allow overflow */
2085 if ((unsigned long) addr
+ count
< count
)
2086 count
= -(unsigned long) addr
;
2088 spin_lock(&vmap_area_lock
);
2089 list_for_each_entry(va
, &vmap_area_list
, list
) {
2093 if (!(va
->flags
& VM_VM_AREA
))
2097 vaddr
= (char *) vm
->addr
;
2098 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2100 while (addr
< vaddr
) {
2108 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2111 if (!(vm
->flags
& VM_IOREMAP
))
2112 aligned_vread(buf
, addr
, n
);
2113 else /* IOREMAP area is treated as memory hole */
2120 spin_unlock(&vmap_area_lock
);
2122 if (buf
== buf_start
)
2124 /* zero-fill memory holes */
2125 if (buf
!= buf_start
+ buflen
)
2126 memset(buf
, 0, buflen
- (buf
- buf_start
));
2132 * vwrite() - write vmalloc area in a safe way.
2133 * @buf: buffer for source data
2134 * @addr: vm address.
2135 * @count: number of bytes to be read.
2137 * Returns # of bytes which addr and buf should be incresed.
2138 * (same number to @count).
2139 * If [addr...addr+count) doesn't includes any intersect with valid
2140 * vmalloc area, returns 0.
2142 * This function checks that addr is a valid vmalloc'ed area, and
2143 * copy data from a buffer to the given addr. If specified range of
2144 * [addr...addr+count) includes some valid address, data is copied from
2145 * proper area of @buf. If there are memory holes, no copy to hole.
2146 * IOREMAP area is treated as memory hole and no copy is done.
2148 * If [addr...addr+count) doesn't includes any intersects with alive
2149 * vm_struct area, returns 0. @buf should be kernel's buffer.
2151 * Note: In usual ops, vwrite() is never necessary because the caller
2152 * should know vmalloc() area is valid and can use memcpy().
2153 * This is for routines which have to access vmalloc area without
2154 * any informaion, as /dev/kmem.
2157 long vwrite(char *buf
, char *addr
, unsigned long count
)
2159 struct vmap_area
*va
;
2160 struct vm_struct
*vm
;
2162 unsigned long n
, buflen
;
2165 /* Don't allow overflow */
2166 if ((unsigned long) addr
+ count
< count
)
2167 count
= -(unsigned long) addr
;
2170 spin_lock(&vmap_area_lock
);
2171 list_for_each_entry(va
, &vmap_area_list
, list
) {
2175 if (!(va
->flags
& VM_VM_AREA
))
2179 vaddr
= (char *) vm
->addr
;
2180 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2182 while (addr
< vaddr
) {
2189 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2192 if (!(vm
->flags
& VM_IOREMAP
)) {
2193 aligned_vwrite(buf
, addr
, n
);
2201 spin_unlock(&vmap_area_lock
);
2208 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2209 * @vma: vma to cover
2210 * @uaddr: target user address to start at
2211 * @kaddr: virtual address of vmalloc kernel memory
2212 * @size: size of map area
2214 * Returns: 0 for success, -Exxx on failure
2216 * This function checks that @kaddr is a valid vmalloc'ed area,
2217 * and that it is big enough to cover the range starting at
2218 * @uaddr in @vma. Will return failure if that criteria isn't
2221 * Similar to remap_pfn_range() (see mm/memory.c)
2223 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2224 void *kaddr
, unsigned long size
)
2226 struct vm_struct
*area
;
2228 size
= PAGE_ALIGN(size
);
2230 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2233 area
= find_vm_area(kaddr
);
2237 if (!(area
->flags
& VM_USERMAP
))
2240 if (kaddr
+ size
> area
->addr
+ area
->size
)
2244 struct page
*page
= vmalloc_to_page(kaddr
);
2247 ret
= vm_insert_page(vma
, uaddr
, page
);
2256 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2260 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2263 * remap_vmalloc_range - map vmalloc pages to userspace
2264 * @vma: vma to cover (map full range of vma)
2265 * @addr: vmalloc memory
2266 * @pgoff: number of pages into addr before first page to map
2268 * Returns: 0 for success, -Exxx on failure
2270 * This function checks that addr is a valid vmalloc'ed area, and
2271 * that it is big enough to cover the vma. Will return failure if
2272 * that criteria isn't met.
2274 * Similar to remap_pfn_range() (see mm/memory.c)
2276 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2277 unsigned long pgoff
)
2279 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2280 addr
+ (pgoff
<< PAGE_SHIFT
),
2281 vma
->vm_end
- vma
->vm_start
);
2283 EXPORT_SYMBOL(remap_vmalloc_range
);
2286 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2289 void __weak
vmalloc_sync_all(void)
2294 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2306 * alloc_vm_area - allocate a range of kernel address space
2307 * @size: size of the area
2308 * @ptes: returns the PTEs for the address space
2310 * Returns: NULL on failure, vm_struct on success
2312 * This function reserves a range of kernel address space, and
2313 * allocates pagetables to map that range. No actual mappings
2316 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2317 * allocated for the VM area are returned.
2319 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2321 struct vm_struct
*area
;
2323 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2324 __builtin_return_address(0));
2329 * This ensures that page tables are constructed for this region
2330 * of kernel virtual address space and mapped into init_mm.
2332 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2333 size
, f
, ptes
? &ptes
: NULL
)) {
2340 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2342 void free_vm_area(struct vm_struct
*area
)
2344 struct vm_struct
*ret
;
2345 ret
= remove_vm_area(area
->addr
);
2346 BUG_ON(ret
!= area
);
2349 EXPORT_SYMBOL_GPL(free_vm_area
);
2352 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2354 return rb_entry_safe(n
, struct vmap_area
, rb_node
);
2358 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2359 * @end: target address
2360 * @pnext: out arg for the next vmap_area
2361 * @pprev: out arg for the previous vmap_area
2363 * Returns: %true if either or both of next and prev are found,
2364 * %false if no vmap_area exists
2366 * Find vmap_areas end addresses of which enclose @end. ie. if not
2367 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2369 static bool pvm_find_next_prev(unsigned long end
,
2370 struct vmap_area
**pnext
,
2371 struct vmap_area
**pprev
)
2373 struct rb_node
*n
= vmap_area_root
.rb_node
;
2374 struct vmap_area
*va
= NULL
;
2377 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2378 if (end
< va
->va_end
)
2380 else if (end
> va
->va_end
)
2389 if (va
->va_end
> end
) {
2391 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2394 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2400 * pvm_determine_end - find the highest aligned address between two vmap_areas
2401 * @pnext: in/out arg for the next vmap_area
2402 * @pprev: in/out arg for the previous vmap_area
2405 * Returns: determined end address
2407 * Find the highest aligned address between *@pnext and *@pprev below
2408 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2409 * down address is between the end addresses of the two vmap_areas.
2411 * Please note that the address returned by this function may fall
2412 * inside *@pnext vmap_area. The caller is responsible for checking
2415 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2416 struct vmap_area
**pprev
,
2417 unsigned long align
)
2419 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2423 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2427 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2429 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2436 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2437 * @offsets: array containing offset of each area
2438 * @sizes: array containing size of each area
2439 * @nr_vms: the number of areas to allocate
2440 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2442 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2443 * vm_structs on success, %NULL on failure
2445 * Percpu allocator wants to use congruent vm areas so that it can
2446 * maintain the offsets among percpu areas. This function allocates
2447 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2448 * be scattered pretty far, distance between two areas easily going up
2449 * to gigabytes. To avoid interacting with regular vmallocs, these
2450 * areas are allocated from top.
2452 * Despite its complicated look, this allocator is rather simple. It
2453 * does everything top-down and scans areas from the end looking for
2454 * matching slot. While scanning, if any of the areas overlaps with
2455 * existing vmap_area, the base address is pulled down to fit the
2456 * area. Scanning is repeated till all the areas fit and then all
2457 * necessary data structres are inserted and the result is returned.
2459 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2460 const size_t *sizes
, int nr_vms
,
2463 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2464 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2465 struct vmap_area
**vas
, *prev
, *next
;
2466 struct vm_struct
**vms
;
2467 int area
, area2
, last_area
, term_area
;
2468 unsigned long base
, start
, end
, last_end
;
2469 bool purged
= false;
2471 /* verify parameters and allocate data structures */
2472 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
2473 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2474 start
= offsets
[area
];
2475 end
= start
+ sizes
[area
];
2477 /* is everything aligned properly? */
2478 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2479 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2481 /* detect the area with the highest address */
2482 if (start
> offsets
[last_area
])
2485 for (area2
= 0; area2
< nr_vms
; area2
++) {
2486 unsigned long start2
= offsets
[area2
];
2487 unsigned long end2
= start2
+ sizes
[area2
];
2492 BUG_ON(start2
>= start
&& start2
< end
);
2493 BUG_ON(end2
<= end
&& end2
> start
);
2496 last_end
= offsets
[last_area
] + sizes
[last_area
];
2498 if (vmalloc_end
- vmalloc_start
< last_end
) {
2503 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2504 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2508 for (area
= 0; area
< nr_vms
; area
++) {
2509 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2510 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2511 if (!vas
[area
] || !vms
[area
])
2515 spin_lock(&vmap_area_lock
);
2517 /* start scanning - we scan from the top, begin with the last area */
2518 area
= term_area
= last_area
;
2519 start
= offsets
[area
];
2520 end
= start
+ sizes
[area
];
2522 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2523 base
= vmalloc_end
- last_end
;
2526 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2529 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2530 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2533 * base might have underflowed, add last_end before
2536 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2537 spin_unlock(&vmap_area_lock
);
2539 purge_vmap_area_lazy();
2547 * If next overlaps, move base downwards so that it's
2548 * right below next and then recheck.
2550 if (next
&& next
->va_start
< base
+ end
) {
2551 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2557 * If prev overlaps, shift down next and prev and move
2558 * base so that it's right below new next and then
2561 if (prev
&& prev
->va_end
> base
+ start
) {
2563 prev
= node_to_va(rb_prev(&next
->rb_node
));
2564 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2570 * This area fits, move on to the previous one. If
2571 * the previous one is the terminal one, we're done.
2573 area
= (area
+ nr_vms
- 1) % nr_vms
;
2574 if (area
== term_area
)
2576 start
= offsets
[area
];
2577 end
= start
+ sizes
[area
];
2578 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2581 /* we've found a fitting base, insert all va's */
2582 for (area
= 0; area
< nr_vms
; area
++) {
2583 struct vmap_area
*va
= vas
[area
];
2585 va
->va_start
= base
+ offsets
[area
];
2586 va
->va_end
= va
->va_start
+ sizes
[area
];
2587 __insert_vmap_area(va
);
2590 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2592 spin_unlock(&vmap_area_lock
);
2594 /* insert all vm's */
2595 for (area
= 0; area
< nr_vms
; area
++)
2596 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2603 for (area
= 0; area
< nr_vms
; area
++) {
2614 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2615 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2616 * @nr_vms: the number of allocated areas
2618 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2620 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2624 for (i
= 0; i
< nr_vms
; i
++)
2625 free_vm_area(vms
[i
]);
2628 #endif /* CONFIG_SMP */
2630 #ifdef CONFIG_PROC_FS
2631 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2632 __acquires(&vmap_area_lock
)
2634 spin_lock(&vmap_area_lock
);
2635 return seq_list_start(&vmap_area_list
, *pos
);
2638 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2640 return seq_list_next(p
, &vmap_area_list
, pos
);
2643 static void s_stop(struct seq_file
*m
, void *p
)
2644 __releases(&vmap_area_lock
)
2646 spin_unlock(&vmap_area_lock
);
2649 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2651 if (IS_ENABLED(CONFIG_NUMA
)) {
2652 unsigned int nr
, *counters
= m
->private;
2657 if (v
->flags
& VM_UNINITIALIZED
)
2659 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2662 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2664 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2665 counters
[page_to_nid(v
->pages
[nr
])]++;
2667 for_each_node_state(nr
, N_HIGH_MEMORY
)
2669 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2673 static int s_show(struct seq_file
*m
, void *p
)
2675 struct vmap_area
*va
;
2676 struct vm_struct
*v
;
2678 va
= list_entry(p
, struct vmap_area
, list
);
2681 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2682 * behalf of vmap area is being tear down or vm_map_ram allocation.
2684 if (!(va
->flags
& VM_VM_AREA
))
2689 seq_printf(m
, "0x%pK-0x%pK %7ld",
2690 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2693 seq_printf(m
, " %pS", v
->caller
);
2696 seq_printf(m
, " pages=%d", v
->nr_pages
);
2699 seq_printf(m
, " phys=%pa", &v
->phys_addr
);
2701 if (v
->flags
& VM_IOREMAP
)
2702 seq_puts(m
, " ioremap");
2704 if (v
->flags
& VM_ALLOC
)
2705 seq_puts(m
, " vmalloc");
2707 if (v
->flags
& VM_MAP
)
2708 seq_puts(m
, " vmap");
2710 if (v
->flags
& VM_USERMAP
)
2711 seq_puts(m
, " user");
2713 if (is_vmalloc_addr(v
->pages
))
2714 seq_puts(m
, " vpages");
2716 show_numa_info(m
, v
);
2721 static const struct seq_operations vmalloc_op
= {
2728 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2730 if (IS_ENABLED(CONFIG_NUMA
))
2731 return seq_open_private(file
, &vmalloc_op
,
2732 nr_node_ids
* sizeof(unsigned int));
2734 return seq_open(file
, &vmalloc_op
);
2737 static const struct file_operations proc_vmalloc_operations
= {
2738 .open
= vmalloc_open
,
2740 .llseek
= seq_lseek
,
2741 .release
= seq_release_private
,
2744 static int __init
proc_vmalloc_init(void)
2746 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2749 module_init(proc_vmalloc_init
);