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/set_memory.h>
22 #include <linux/debugobjects.h>
23 #include <linux/kallsyms.h>
24 #include <linux/list.h>
25 #include <linux/notifier.h>
26 #include <linux/rbtree.h>
27 #include <linux/radix-tree.h>
28 #include <linux/rcupdate.h>
29 #include <linux/pfn.h>
30 #include <linux/kmemleak.h>
31 #include <linux/atomic.h>
32 #include <linux/compiler.h>
33 #include <linux/llist.h>
34 #include <linux/bitops.h>
36 #include <linux/uaccess.h>
37 #include <asm/tlbflush.h>
38 #include <asm/shmparam.h>
42 struct vfree_deferred
{
43 struct llist_head list
;
44 struct work_struct wq
;
46 static DEFINE_PER_CPU(struct vfree_deferred
, vfree_deferred
);
48 static void __vunmap(const void *, int);
50 static void free_work(struct work_struct
*w
)
52 struct vfree_deferred
*p
= container_of(w
, struct vfree_deferred
, wq
);
53 struct llist_node
*t
, *llnode
;
55 llist_for_each_safe(llnode
, t
, llist_del_all(&p
->list
))
56 __vunmap((void *)llnode
, 1);
59 /*** Page table manipulation functions ***/
61 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
65 pte
= pte_offset_kernel(pmd
, addr
);
67 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
68 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
69 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
72 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
77 pmd
= pmd_offset(pud
, addr
);
79 next
= pmd_addr_end(addr
, end
);
80 if (pmd_clear_huge(pmd
))
82 if (pmd_none_or_clear_bad(pmd
))
84 vunmap_pte_range(pmd
, addr
, next
);
85 } while (pmd
++, addr
= next
, addr
!= end
);
88 static void vunmap_pud_range(p4d_t
*p4d
, unsigned long addr
, unsigned long end
)
93 pud
= pud_offset(p4d
, addr
);
95 next
= pud_addr_end(addr
, end
);
96 if (pud_clear_huge(pud
))
98 if (pud_none_or_clear_bad(pud
))
100 vunmap_pmd_range(pud
, addr
, next
);
101 } while (pud
++, addr
= next
, addr
!= end
);
104 static void vunmap_p4d_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
109 p4d
= p4d_offset(pgd
, addr
);
111 next
= p4d_addr_end(addr
, end
);
112 if (p4d_clear_huge(p4d
))
114 if (p4d_none_or_clear_bad(p4d
))
116 vunmap_pud_range(p4d
, addr
, next
);
117 } while (p4d
++, addr
= next
, addr
!= end
);
120 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
126 pgd
= pgd_offset_k(addr
);
128 next
= pgd_addr_end(addr
, end
);
129 if (pgd_none_or_clear_bad(pgd
))
131 vunmap_p4d_range(pgd
, addr
, next
);
132 } while (pgd
++, addr
= next
, addr
!= end
);
135 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
136 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
141 * nr is a running index into the array which helps higher level
142 * callers keep track of where we're up to.
145 pte
= pte_alloc_kernel(pmd
, addr
);
149 struct page
*page
= pages
[*nr
];
151 if (WARN_ON(!pte_none(*pte
)))
155 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
157 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
161 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
162 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
167 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
171 next
= pmd_addr_end(addr
, end
);
172 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
174 } while (pmd
++, addr
= next
, addr
!= end
);
178 static int vmap_pud_range(p4d_t
*p4d
, unsigned long addr
,
179 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
184 pud
= pud_alloc(&init_mm
, p4d
, addr
);
188 next
= pud_addr_end(addr
, end
);
189 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
191 } while (pud
++, addr
= next
, addr
!= end
);
195 static int vmap_p4d_range(pgd_t
*pgd
, unsigned long addr
,
196 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
201 p4d
= p4d_alloc(&init_mm
, pgd
, addr
);
205 next
= p4d_addr_end(addr
, end
);
206 if (vmap_pud_range(p4d
, addr
, next
, prot
, pages
, nr
))
208 } while (p4d
++, addr
= next
, addr
!= end
);
213 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
214 * will have pfns corresponding to the "pages" array.
216 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
218 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
219 pgprot_t prot
, struct page
**pages
)
223 unsigned long addr
= start
;
228 pgd
= pgd_offset_k(addr
);
230 next
= pgd_addr_end(addr
, end
);
231 err
= vmap_p4d_range(pgd
, addr
, next
, prot
, pages
, &nr
);
234 } while (pgd
++, addr
= next
, addr
!= end
);
239 static int vmap_page_range(unsigned long start
, unsigned long end
,
240 pgprot_t prot
, struct page
**pages
)
244 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
245 flush_cache_vmap(start
, end
);
249 int is_vmalloc_or_module_addr(const void *x
)
252 * ARM, x86-64 and sparc64 put modules in a special place,
253 * and fall back on vmalloc() if that fails. Others
254 * just put it in the vmalloc space.
256 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
257 unsigned long addr
= (unsigned long)x
;
258 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
261 return is_vmalloc_addr(x
);
265 * Walk a vmap address to the struct page it maps.
267 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
269 unsigned long addr
= (unsigned long) vmalloc_addr
;
270 struct page
*page
= NULL
;
271 pgd_t
*pgd
= pgd_offset_k(addr
);
278 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
279 * architectures that do not vmalloc module space
281 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
285 p4d
= p4d_offset(pgd
, addr
);
288 pud
= pud_offset(p4d
, addr
);
291 * Don't dereference bad PUD or PMD (below) entries. This will also
292 * identify huge mappings, which we may encounter on architectures
293 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
294 * identified as vmalloc addresses by is_vmalloc_addr(), but are
295 * not [unambiguously] associated with a struct page, so there is
296 * no correct value to return for them.
298 WARN_ON_ONCE(pud_bad(*pud
));
299 if (pud_none(*pud
) || pud_bad(*pud
))
301 pmd
= pmd_offset(pud
, addr
);
302 WARN_ON_ONCE(pmd_bad(*pmd
));
303 if (pmd_none(*pmd
) || pmd_bad(*pmd
))
306 ptep
= pte_offset_map(pmd
, addr
);
308 if (pte_present(pte
))
309 page
= pte_page(pte
);
313 EXPORT_SYMBOL(vmalloc_to_page
);
316 * Map a vmalloc()-space virtual address to the physical page frame number.
318 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
320 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
322 EXPORT_SYMBOL(vmalloc_to_pfn
);
325 /*** Global kva allocator ***/
327 #define VM_LAZY_FREE 0x02
328 #define VM_VM_AREA 0x04
330 static DEFINE_SPINLOCK(vmap_area_lock
);
331 /* Export for kexec only */
332 LIST_HEAD(vmap_area_list
);
333 static LLIST_HEAD(vmap_purge_list
);
334 static struct rb_root vmap_area_root
= RB_ROOT
;
336 /* The vmap cache globals are protected by vmap_area_lock */
337 static struct rb_node
*free_vmap_cache
;
338 static unsigned long cached_hole_size
;
339 static unsigned long cached_vstart
;
340 static unsigned long cached_align
;
342 static unsigned long vmap_area_pcpu_hole
;
344 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
346 struct rb_node
*n
= vmap_area_root
.rb_node
;
349 struct vmap_area
*va
;
351 va
= rb_entry(n
, struct vmap_area
, rb_node
);
352 if (addr
< va
->va_start
)
354 else if (addr
>= va
->va_end
)
363 static void __insert_vmap_area(struct vmap_area
*va
)
365 struct rb_node
**p
= &vmap_area_root
.rb_node
;
366 struct rb_node
*parent
= NULL
;
370 struct vmap_area
*tmp_va
;
373 tmp_va
= rb_entry(parent
, struct vmap_area
, rb_node
);
374 if (va
->va_start
< tmp_va
->va_end
)
376 else if (va
->va_end
> tmp_va
->va_start
)
382 rb_link_node(&va
->rb_node
, parent
, p
);
383 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
385 /* address-sort this list */
386 tmp
= rb_prev(&va
->rb_node
);
388 struct vmap_area
*prev
;
389 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
390 list_add_rcu(&va
->list
, &prev
->list
);
392 list_add_rcu(&va
->list
, &vmap_area_list
);
395 static void purge_vmap_area_lazy(void);
397 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list
);
400 * Allocate a region of KVA of the specified size and alignment, within the
403 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
405 unsigned long vstart
, unsigned long vend
,
406 int node
, gfp_t gfp_mask
)
408 struct vmap_area
*va
;
412 struct vmap_area
*first
;
415 BUG_ON(offset_in_page(size
));
416 BUG_ON(!is_power_of_2(align
));
420 va
= kmalloc_node(sizeof(struct vmap_area
),
421 gfp_mask
& GFP_RECLAIM_MASK
, node
);
423 return ERR_PTR(-ENOMEM
);
426 * Only scan the relevant parts containing pointers to other objects
427 * to avoid false negatives.
429 kmemleak_scan_area(&va
->rb_node
, SIZE_MAX
, gfp_mask
& GFP_RECLAIM_MASK
);
432 spin_lock(&vmap_area_lock
);
434 * Invalidate cache if we have more permissive parameters.
435 * cached_hole_size notes the largest hole noticed _below_
436 * the vmap_area cached in free_vmap_cache: if size fits
437 * into that hole, we want to scan from vstart to reuse
438 * the hole instead of allocating above free_vmap_cache.
439 * Note that __free_vmap_area may update free_vmap_cache
440 * without updating cached_hole_size or cached_align.
442 if (!free_vmap_cache
||
443 size
< cached_hole_size
||
444 vstart
< cached_vstart
||
445 align
< cached_align
) {
447 cached_hole_size
= 0;
448 free_vmap_cache
= NULL
;
450 /* record if we encounter less permissive parameters */
451 cached_vstart
= vstart
;
452 cached_align
= align
;
454 /* find starting point for our search */
455 if (free_vmap_cache
) {
456 first
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
457 addr
= ALIGN(first
->va_end
, align
);
460 if (addr
+ size
< addr
)
464 addr
= ALIGN(vstart
, align
);
465 if (addr
+ size
< addr
)
468 n
= vmap_area_root
.rb_node
;
472 struct vmap_area
*tmp
;
473 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
474 if (tmp
->va_end
>= addr
) {
476 if (tmp
->va_start
<= addr
)
487 /* from the starting point, walk areas until a suitable hole is found */
488 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
489 if (addr
+ cached_hole_size
< first
->va_start
)
490 cached_hole_size
= first
->va_start
- addr
;
491 addr
= ALIGN(first
->va_end
, align
);
492 if (addr
+ size
< addr
)
495 if (list_is_last(&first
->list
, &vmap_area_list
))
498 first
= list_next_entry(first
, list
);
503 * Check also calculated address against the vstart,
504 * because it can be 0 because of big align request.
506 if (addr
+ size
> vend
|| addr
< vstart
)
510 va
->va_end
= addr
+ size
;
512 __insert_vmap_area(va
);
513 free_vmap_cache
= &va
->rb_node
;
514 spin_unlock(&vmap_area_lock
);
516 BUG_ON(!IS_ALIGNED(va
->va_start
, align
));
517 BUG_ON(va
->va_start
< vstart
);
518 BUG_ON(va
->va_end
> vend
);
523 spin_unlock(&vmap_area_lock
);
525 purge_vmap_area_lazy();
530 if (gfpflags_allow_blocking(gfp_mask
)) {
531 unsigned long freed
= 0;
532 blocking_notifier_call_chain(&vmap_notify_list
, 0, &freed
);
539 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit())
540 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
543 return ERR_PTR(-EBUSY
);
546 int register_vmap_purge_notifier(struct notifier_block
*nb
)
548 return blocking_notifier_chain_register(&vmap_notify_list
, nb
);
550 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier
);
552 int unregister_vmap_purge_notifier(struct notifier_block
*nb
)
554 return blocking_notifier_chain_unregister(&vmap_notify_list
, nb
);
556 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier
);
558 static void __free_vmap_area(struct vmap_area
*va
)
560 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
562 if (free_vmap_cache
) {
563 if (va
->va_end
< cached_vstart
) {
564 free_vmap_cache
= NULL
;
566 struct vmap_area
*cache
;
567 cache
= rb_entry(free_vmap_cache
, struct vmap_area
, rb_node
);
568 if (va
->va_start
<= cache
->va_start
) {
569 free_vmap_cache
= rb_prev(&va
->rb_node
);
571 * We don't try to update cached_hole_size or
572 * cached_align, but it won't go very wrong.
577 rb_erase(&va
->rb_node
, &vmap_area_root
);
578 RB_CLEAR_NODE(&va
->rb_node
);
579 list_del_rcu(&va
->list
);
582 * Track the highest possible candidate for pcpu area
583 * allocation. Areas outside of vmalloc area can be returned
584 * here too, consider only end addresses which fall inside
585 * vmalloc area proper.
587 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
588 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
590 kfree_rcu(va
, rcu_head
);
594 * Free a region of KVA allocated by alloc_vmap_area
596 static void free_vmap_area(struct vmap_area
*va
)
598 spin_lock(&vmap_area_lock
);
599 __free_vmap_area(va
);
600 spin_unlock(&vmap_area_lock
);
604 * Clear the pagetable entries of a given vmap_area
606 static void unmap_vmap_area(struct vmap_area
*va
)
608 vunmap_page_range(va
->va_start
, va
->va_end
);
612 * lazy_max_pages is the maximum amount of virtual address space we gather up
613 * before attempting to purge with a TLB flush.
615 * There is a tradeoff here: a larger number will cover more kernel page tables
616 * and take slightly longer to purge, but it will linearly reduce the number of
617 * global TLB flushes that must be performed. It would seem natural to scale
618 * this number up linearly with the number of CPUs (because vmapping activity
619 * could also scale linearly with the number of CPUs), however it is likely
620 * that in practice, workloads might be constrained in other ways that mean
621 * vmap activity will not scale linearly with CPUs. Also, I want to be
622 * conservative and not introduce a big latency on huge systems, so go with
623 * a less aggressive log scale. It will still be an improvement over the old
624 * code, and it will be simple to change the scale factor if we find that it
625 * becomes a problem on bigger systems.
627 static unsigned long lazy_max_pages(void)
631 log
= fls(num_online_cpus());
633 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
636 static atomic_long_t vmap_lazy_nr
= ATOMIC_LONG_INIT(0);
639 * Serialize vmap purging. There is no actual criticial section protected
640 * by this look, but we want to avoid concurrent calls for performance
641 * reasons and to make the pcpu_get_vm_areas more deterministic.
643 static DEFINE_MUTEX(vmap_purge_lock
);
645 /* for per-CPU blocks */
646 static void purge_fragmented_blocks_allcpus(void);
649 * called before a call to iounmap() if the caller wants vm_area_struct's
652 void set_iounmap_nonlazy(void)
654 atomic_long_set(&vmap_lazy_nr
, lazy_max_pages()+1);
658 * Purges all lazily-freed vmap areas.
660 static bool __purge_vmap_area_lazy(unsigned long start
, unsigned long end
)
662 unsigned long resched_threshold
;
663 struct llist_node
*valist
;
664 struct vmap_area
*va
;
665 struct vmap_area
*n_va
;
667 lockdep_assert_held(&vmap_purge_lock
);
669 valist
= llist_del_all(&vmap_purge_list
);
670 if (unlikely(valist
== NULL
))
674 * TODO: to calculate a flush range without looping.
675 * The list can be up to lazy_max_pages() elements.
677 llist_for_each_entry(va
, valist
, purge_list
) {
678 if (va
->va_start
< start
)
679 start
= va
->va_start
;
680 if (va
->va_end
> end
)
684 flush_tlb_kernel_range(start
, end
);
685 resched_threshold
= lazy_max_pages() << 1;
687 spin_lock(&vmap_area_lock
);
688 llist_for_each_entry_safe(va
, n_va
, valist
, purge_list
) {
689 unsigned long nr
= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
691 __free_vmap_area(va
);
692 atomic_long_sub(nr
, &vmap_lazy_nr
);
694 if (atomic_long_read(&vmap_lazy_nr
) < resched_threshold
)
695 cond_resched_lock(&vmap_area_lock
);
697 spin_unlock(&vmap_area_lock
);
702 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
703 * is already purging.
705 static void try_purge_vmap_area_lazy(void)
707 if (mutex_trylock(&vmap_purge_lock
)) {
708 __purge_vmap_area_lazy(ULONG_MAX
, 0);
709 mutex_unlock(&vmap_purge_lock
);
714 * Kick off a purge of the outstanding lazy areas.
716 static void purge_vmap_area_lazy(void)
718 mutex_lock(&vmap_purge_lock
);
719 purge_fragmented_blocks_allcpus();
720 __purge_vmap_area_lazy(ULONG_MAX
, 0);
721 mutex_unlock(&vmap_purge_lock
);
725 * Free a vmap area, caller ensuring that the area has been unmapped
726 * and flush_cache_vunmap had been called for the correct range
729 static void free_vmap_area_noflush(struct vmap_area
*va
)
731 unsigned long nr_lazy
;
733 nr_lazy
= atomic_long_add_return((va
->va_end
- va
->va_start
) >>
734 PAGE_SHIFT
, &vmap_lazy_nr
);
736 /* After this point, we may free va at any time */
737 llist_add(&va
->purge_list
, &vmap_purge_list
);
739 if (unlikely(nr_lazy
> lazy_max_pages()))
740 try_purge_vmap_area_lazy();
744 * Free and unmap a vmap area
746 static void free_unmap_vmap_area(struct vmap_area
*va
)
748 flush_cache_vunmap(va
->va_start
, va
->va_end
);
750 if (debug_pagealloc_enabled())
751 flush_tlb_kernel_range(va
->va_start
, va
->va_end
);
753 free_vmap_area_noflush(va
);
756 static struct vmap_area
*find_vmap_area(unsigned long addr
)
758 struct vmap_area
*va
;
760 spin_lock(&vmap_area_lock
);
761 va
= __find_vmap_area(addr
);
762 spin_unlock(&vmap_area_lock
);
767 /*** Per cpu kva allocator ***/
770 * vmap space is limited especially on 32 bit architectures. Ensure there is
771 * room for at least 16 percpu vmap blocks per CPU.
774 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
775 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
776 * instead (we just need a rough idea)
778 #if BITS_PER_LONG == 32
779 #define VMALLOC_SPACE (128UL*1024*1024)
781 #define VMALLOC_SPACE (128UL*1024*1024*1024)
784 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
785 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
786 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
787 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
788 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
789 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
790 #define VMAP_BBMAP_BITS \
791 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
792 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
793 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
795 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
797 static bool vmap_initialized __read_mostly
= false;
799 struct vmap_block_queue
{
801 struct list_head free
;
806 struct vmap_area
*va
;
807 unsigned long free
, dirty
;
808 unsigned long dirty_min
, dirty_max
; /*< dirty range */
809 struct list_head free_list
;
810 struct rcu_head rcu_head
;
811 struct list_head purge
;
814 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
815 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
818 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
819 * in the free path. Could get rid of this if we change the API to return a
820 * "cookie" from alloc, to be passed to free. But no big deal yet.
822 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
823 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
826 * We should probably have a fallback mechanism to allocate virtual memory
827 * out of partially filled vmap blocks. However vmap block sizing should be
828 * fairly reasonable according to the vmalloc size, so it shouldn't be a
832 static unsigned long addr_to_vb_idx(unsigned long addr
)
834 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
835 addr
/= VMAP_BLOCK_SIZE
;
839 static void *vmap_block_vaddr(unsigned long va_start
, unsigned long pages_off
)
843 addr
= va_start
+ (pages_off
<< PAGE_SHIFT
);
844 BUG_ON(addr_to_vb_idx(addr
) != addr_to_vb_idx(va_start
));
849 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
850 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
851 * @order: how many 2^order pages should be occupied in newly allocated block
852 * @gfp_mask: flags for the page level allocator
854 * Return: virtual address in a newly allocated block or ERR_PTR(-errno)
856 static void *new_vmap_block(unsigned int order
, gfp_t gfp_mask
)
858 struct vmap_block_queue
*vbq
;
859 struct vmap_block
*vb
;
860 struct vmap_area
*va
;
861 unsigned long vb_idx
;
865 node
= numa_node_id();
867 vb
= kmalloc_node(sizeof(struct vmap_block
),
868 gfp_mask
& GFP_RECLAIM_MASK
, node
);
870 return ERR_PTR(-ENOMEM
);
872 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
873 VMALLOC_START
, VMALLOC_END
,
880 err
= radix_tree_preload(gfp_mask
);
887 vaddr
= vmap_block_vaddr(va
->va_start
, 0);
888 spin_lock_init(&vb
->lock
);
890 /* At least something should be left free */
891 BUG_ON(VMAP_BBMAP_BITS
<= (1UL << order
));
892 vb
->free
= VMAP_BBMAP_BITS
- (1UL << order
);
894 vb
->dirty_min
= VMAP_BBMAP_BITS
;
896 INIT_LIST_HEAD(&vb
->free_list
);
898 vb_idx
= addr_to_vb_idx(va
->va_start
);
899 spin_lock(&vmap_block_tree_lock
);
900 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
901 spin_unlock(&vmap_block_tree_lock
);
903 radix_tree_preload_end();
905 vbq
= &get_cpu_var(vmap_block_queue
);
906 spin_lock(&vbq
->lock
);
907 list_add_tail_rcu(&vb
->free_list
, &vbq
->free
);
908 spin_unlock(&vbq
->lock
);
909 put_cpu_var(vmap_block_queue
);
914 static void free_vmap_block(struct vmap_block
*vb
)
916 struct vmap_block
*tmp
;
917 unsigned long vb_idx
;
919 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
920 spin_lock(&vmap_block_tree_lock
);
921 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
922 spin_unlock(&vmap_block_tree_lock
);
925 free_vmap_area_noflush(vb
->va
);
926 kfree_rcu(vb
, rcu_head
);
929 static void purge_fragmented_blocks(int cpu
)
932 struct vmap_block
*vb
;
933 struct vmap_block
*n_vb
;
934 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
937 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
939 if (!(vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
))
942 spin_lock(&vb
->lock
);
943 if (vb
->free
+ vb
->dirty
== VMAP_BBMAP_BITS
&& vb
->dirty
!= VMAP_BBMAP_BITS
) {
944 vb
->free
= 0; /* prevent further allocs after releasing lock */
945 vb
->dirty
= VMAP_BBMAP_BITS
; /* prevent purging it again */
947 vb
->dirty_max
= VMAP_BBMAP_BITS
;
948 spin_lock(&vbq
->lock
);
949 list_del_rcu(&vb
->free_list
);
950 spin_unlock(&vbq
->lock
);
951 spin_unlock(&vb
->lock
);
952 list_add_tail(&vb
->purge
, &purge
);
954 spin_unlock(&vb
->lock
);
958 list_for_each_entry_safe(vb
, n_vb
, &purge
, purge
) {
959 list_del(&vb
->purge
);
964 static void purge_fragmented_blocks_allcpus(void)
968 for_each_possible_cpu(cpu
)
969 purge_fragmented_blocks(cpu
);
972 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
974 struct vmap_block_queue
*vbq
;
975 struct vmap_block
*vb
;
979 BUG_ON(offset_in_page(size
));
980 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
981 if (WARN_ON(size
== 0)) {
983 * Allocating 0 bytes isn't what caller wants since
984 * get_order(0) returns funny result. Just warn and terminate
989 order
= get_order(size
);
992 vbq
= &get_cpu_var(vmap_block_queue
);
993 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
994 unsigned long pages_off
;
996 spin_lock(&vb
->lock
);
997 if (vb
->free
< (1UL << order
)) {
998 spin_unlock(&vb
->lock
);
1002 pages_off
= VMAP_BBMAP_BITS
- vb
->free
;
1003 vaddr
= vmap_block_vaddr(vb
->va
->va_start
, pages_off
);
1004 vb
->free
-= 1UL << order
;
1005 if (vb
->free
== 0) {
1006 spin_lock(&vbq
->lock
);
1007 list_del_rcu(&vb
->free_list
);
1008 spin_unlock(&vbq
->lock
);
1011 spin_unlock(&vb
->lock
);
1015 put_cpu_var(vmap_block_queue
);
1018 /* Allocate new block if nothing was found */
1020 vaddr
= new_vmap_block(order
, gfp_mask
);
1025 static void vb_free(const void *addr
, unsigned long size
)
1027 unsigned long offset
;
1028 unsigned long vb_idx
;
1030 struct vmap_block
*vb
;
1032 BUG_ON(offset_in_page(size
));
1033 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
1035 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
1037 order
= get_order(size
);
1039 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
1040 offset
>>= PAGE_SHIFT
;
1042 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
1044 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
1048 vunmap_page_range((unsigned long)addr
, (unsigned long)addr
+ size
);
1050 if (debug_pagealloc_enabled())
1051 flush_tlb_kernel_range((unsigned long)addr
,
1052 (unsigned long)addr
+ size
);
1054 spin_lock(&vb
->lock
);
1056 /* Expand dirty range */
1057 vb
->dirty_min
= min(vb
->dirty_min
, offset
);
1058 vb
->dirty_max
= max(vb
->dirty_max
, offset
+ (1UL << order
));
1060 vb
->dirty
+= 1UL << order
;
1061 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
1063 spin_unlock(&vb
->lock
);
1064 free_vmap_block(vb
);
1066 spin_unlock(&vb
->lock
);
1069 static void _vm_unmap_aliases(unsigned long start
, unsigned long end
, int flush
)
1073 if (unlikely(!vmap_initialized
))
1078 for_each_possible_cpu(cpu
) {
1079 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
1080 struct vmap_block
*vb
;
1083 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
1084 spin_lock(&vb
->lock
);
1086 unsigned long va_start
= vb
->va
->va_start
;
1089 s
= va_start
+ (vb
->dirty_min
<< PAGE_SHIFT
);
1090 e
= va_start
+ (vb
->dirty_max
<< PAGE_SHIFT
);
1092 start
= min(s
, start
);
1097 spin_unlock(&vb
->lock
);
1102 mutex_lock(&vmap_purge_lock
);
1103 purge_fragmented_blocks_allcpus();
1104 if (!__purge_vmap_area_lazy(start
, end
) && flush
)
1105 flush_tlb_kernel_range(start
, end
);
1106 mutex_unlock(&vmap_purge_lock
);
1110 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1112 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1113 * to amortize TLB flushing overheads. What this means is that any page you
1114 * have now, may, in a former life, have been mapped into kernel virtual
1115 * address by the vmap layer and so there might be some CPUs with TLB entries
1116 * still referencing that page (additional to the regular 1:1 kernel mapping).
1118 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1119 * be sure that none of the pages we have control over will have any aliases
1120 * from the vmap layer.
1122 void vm_unmap_aliases(void)
1124 unsigned long start
= ULONG_MAX
, end
= 0;
1127 _vm_unmap_aliases(start
, end
, flush
);
1129 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
1132 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1133 * @mem: the pointer returned by vm_map_ram
1134 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1136 void vm_unmap_ram(const void *mem
, unsigned int count
)
1138 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1139 unsigned long addr
= (unsigned long)mem
;
1140 struct vmap_area
*va
;
1144 BUG_ON(addr
< VMALLOC_START
);
1145 BUG_ON(addr
> VMALLOC_END
);
1146 BUG_ON(!PAGE_ALIGNED(addr
));
1148 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1149 debug_check_no_locks_freed(mem
, size
);
1154 va
= find_vmap_area(addr
);
1156 debug_check_no_locks_freed((void *)va
->va_start
,
1157 (va
->va_end
- va
->va_start
));
1158 free_unmap_vmap_area(va
);
1160 EXPORT_SYMBOL(vm_unmap_ram
);
1163 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1164 * @pages: an array of pointers to the pages to be mapped
1165 * @count: number of pages
1166 * @node: prefer to allocate data structures on this node
1167 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1169 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1170 * faster than vmap so it's good. But if you mix long-life and short-life
1171 * objects with vm_map_ram(), it could consume lots of address space through
1172 * fragmentation (especially on a 32bit machine). You could see failures in
1173 * the end. Please use this function for short-lived objects.
1175 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1177 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
1179 unsigned long size
= (unsigned long)count
<< PAGE_SHIFT
;
1183 if (likely(count
<= VMAP_MAX_ALLOC
)) {
1184 mem
= vb_alloc(size
, GFP_KERNEL
);
1187 addr
= (unsigned long)mem
;
1189 struct vmap_area
*va
;
1190 va
= alloc_vmap_area(size
, PAGE_SIZE
,
1191 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
1195 addr
= va
->va_start
;
1198 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
1199 vm_unmap_ram(mem
, count
);
1204 EXPORT_SYMBOL(vm_map_ram
);
1206 static struct vm_struct
*vmlist __initdata
;
1209 * vm_area_add_early - add vmap area early during boot
1210 * @vm: vm_struct to add
1212 * This function is used to add fixed kernel vm area to vmlist before
1213 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1214 * should contain proper values and the other fields should be zero.
1216 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1218 void __init
vm_area_add_early(struct vm_struct
*vm
)
1220 struct vm_struct
*tmp
, **p
;
1222 BUG_ON(vmap_initialized
);
1223 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1224 if (tmp
->addr
>= vm
->addr
) {
1225 BUG_ON(tmp
->addr
< vm
->addr
+ vm
->size
);
1228 BUG_ON(tmp
->addr
+ tmp
->size
> vm
->addr
);
1235 * vm_area_register_early - register vmap area early during boot
1236 * @vm: vm_struct to register
1237 * @align: requested alignment
1239 * This function is used to register kernel vm area before
1240 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1241 * proper values on entry and other fields should be zero. On return,
1242 * vm->addr contains the allocated address.
1244 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1246 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1248 static size_t vm_init_off __initdata
;
1251 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1252 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1254 vm
->addr
= (void *)addr
;
1256 vm_area_add_early(vm
);
1259 void __init
vmalloc_init(void)
1261 struct vmap_area
*va
;
1262 struct vm_struct
*tmp
;
1265 for_each_possible_cpu(i
) {
1266 struct vmap_block_queue
*vbq
;
1267 struct vfree_deferred
*p
;
1269 vbq
= &per_cpu(vmap_block_queue
, i
);
1270 spin_lock_init(&vbq
->lock
);
1271 INIT_LIST_HEAD(&vbq
->free
);
1272 p
= &per_cpu(vfree_deferred
, i
);
1273 init_llist_head(&p
->list
);
1274 INIT_WORK(&p
->wq
, free_work
);
1277 /* Import existing vmlist entries. */
1278 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1279 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1280 va
->flags
= VM_VM_AREA
;
1281 va
->va_start
= (unsigned long)tmp
->addr
;
1282 va
->va_end
= va
->va_start
+ tmp
->size
;
1284 __insert_vmap_area(va
);
1287 vmap_area_pcpu_hole
= VMALLOC_END
;
1289 vmap_initialized
= true;
1293 * map_kernel_range_noflush - map kernel VM area with the specified pages
1294 * @addr: start of the VM area to map
1295 * @size: size of the VM area to map
1296 * @prot: page protection flags to use
1297 * @pages: pages to map
1299 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1300 * specify should have been allocated using get_vm_area() and its
1304 * This function does NOT do any cache flushing. The caller is
1305 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1306 * before calling this function.
1309 * The number of pages mapped on success, -errno on failure.
1311 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1312 pgprot_t prot
, struct page
**pages
)
1314 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1318 * unmap_kernel_range_noflush - unmap kernel VM area
1319 * @addr: start of the VM area to unmap
1320 * @size: size of the VM area to unmap
1322 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1323 * specify should have been allocated using get_vm_area() and its
1327 * This function does NOT do any cache flushing. The caller is
1328 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1329 * before calling this function and flush_tlb_kernel_range() after.
1331 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1333 vunmap_page_range(addr
, addr
+ size
);
1335 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush
);
1338 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1339 * @addr: start of the VM area to unmap
1340 * @size: size of the VM area to unmap
1342 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1343 * the unmapping and tlb after.
1345 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1347 unsigned long end
= addr
+ size
;
1349 flush_cache_vunmap(addr
, end
);
1350 vunmap_page_range(addr
, end
);
1351 flush_tlb_kernel_range(addr
, end
);
1353 EXPORT_SYMBOL_GPL(unmap_kernel_range
);
1355 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
**pages
)
1357 unsigned long addr
= (unsigned long)area
->addr
;
1358 unsigned long end
= addr
+ get_vm_area_size(area
);
1361 err
= vmap_page_range(addr
, end
, prot
, pages
);
1363 return err
> 0 ? 0 : err
;
1365 EXPORT_SYMBOL_GPL(map_vm_area
);
1367 static void setup_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1368 unsigned long flags
, const void *caller
)
1370 spin_lock(&vmap_area_lock
);
1372 vm
->addr
= (void *)va
->va_start
;
1373 vm
->size
= va
->va_end
- va
->va_start
;
1374 vm
->caller
= caller
;
1376 va
->flags
|= VM_VM_AREA
;
1377 spin_unlock(&vmap_area_lock
);
1380 static void clear_vm_uninitialized_flag(struct vm_struct
*vm
)
1383 * Before removing VM_UNINITIALIZED,
1384 * we should make sure that vm has proper values.
1385 * Pair with smp_rmb() in show_numa_info().
1388 vm
->flags
&= ~VM_UNINITIALIZED
;
1391 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1392 unsigned long align
, unsigned long flags
, unsigned long start
,
1393 unsigned long end
, int node
, gfp_t gfp_mask
, const void *caller
)
1395 struct vmap_area
*va
;
1396 struct vm_struct
*area
;
1398 BUG_ON(in_interrupt());
1399 size
= PAGE_ALIGN(size
);
1400 if (unlikely(!size
))
1403 if (flags
& VM_IOREMAP
)
1404 align
= 1ul << clamp_t(int, get_count_order_long(size
),
1405 PAGE_SHIFT
, IOREMAP_MAX_ORDER
);
1407 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1408 if (unlikely(!area
))
1411 if (!(flags
& VM_NO_GUARD
))
1414 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1420 setup_vmalloc_vm(area
, va
, flags
, caller
);
1425 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1426 unsigned long start
, unsigned long end
)
1428 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1429 GFP_KERNEL
, __builtin_return_address(0));
1431 EXPORT_SYMBOL_GPL(__get_vm_area
);
1433 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1434 unsigned long start
, unsigned long end
,
1437 return __get_vm_area_node(size
, 1, flags
, start
, end
, NUMA_NO_NODE
,
1438 GFP_KERNEL
, caller
);
1442 * get_vm_area - reserve a contiguous kernel virtual area
1443 * @size: size of the area
1444 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1446 * Search an area of @size in the kernel virtual mapping area,
1447 * and reserved it for out purposes. Returns the area descriptor
1448 * on success or %NULL on failure.
1450 * Return: the area descriptor on success or %NULL on failure.
1452 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1454 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1455 NUMA_NO_NODE
, GFP_KERNEL
,
1456 __builtin_return_address(0));
1459 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1462 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1463 NUMA_NO_NODE
, GFP_KERNEL
, caller
);
1467 * find_vm_area - find a continuous kernel virtual area
1468 * @addr: base address
1470 * Search for the kernel VM area starting at @addr, and return it.
1471 * It is up to the caller to do all required locking to keep the returned
1474 * Return: pointer to the found area or %NULL on faulure
1476 struct vm_struct
*find_vm_area(const void *addr
)
1478 struct vmap_area
*va
;
1480 va
= find_vmap_area((unsigned long)addr
);
1481 if (va
&& va
->flags
& VM_VM_AREA
)
1488 * remove_vm_area - find and remove a continuous kernel virtual area
1489 * @addr: base address
1491 * Search for the kernel VM area starting at @addr, and remove it.
1492 * This function returns the found VM area, but using it is NOT safe
1493 * on SMP machines, except for its size or flags.
1495 * Return: pointer to the found area or %NULL on faulure
1497 struct vm_struct
*remove_vm_area(const void *addr
)
1499 struct vmap_area
*va
;
1503 va
= find_vmap_area((unsigned long)addr
);
1504 if (va
&& va
->flags
& VM_VM_AREA
) {
1505 struct vm_struct
*vm
= va
->vm
;
1507 spin_lock(&vmap_area_lock
);
1509 va
->flags
&= ~VM_VM_AREA
;
1510 va
->flags
|= VM_LAZY_FREE
;
1511 spin_unlock(&vmap_area_lock
);
1513 kasan_free_shadow(vm
);
1514 free_unmap_vmap_area(va
);
1521 static inline void set_area_direct_map(const struct vm_struct
*area
,
1522 int (*set_direct_map
)(struct page
*page
))
1526 for (i
= 0; i
< area
->nr_pages
; i
++)
1527 if (page_address(area
->pages
[i
]))
1528 set_direct_map(area
->pages
[i
]);
1531 /* Handle removing and resetting vm mappings related to the vm_struct. */
1532 static void vm_remove_mappings(struct vm_struct
*area
, int deallocate_pages
)
1534 unsigned long addr
= (unsigned long)area
->addr
;
1535 unsigned long start
= ULONG_MAX
, end
= 0;
1536 int flush_reset
= area
->flags
& VM_FLUSH_RESET_PERMS
;
1540 * The below block can be removed when all architectures that have
1541 * direct map permissions also have set_direct_map_() implementations.
1542 * This is concerned with resetting the direct map any an vm alias with
1543 * execute permissions, without leaving a RW+X window.
1545 if (flush_reset
&& !IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP
)) {
1546 set_memory_nx(addr
, area
->nr_pages
);
1547 set_memory_rw(addr
, area
->nr_pages
);
1550 remove_vm_area(area
->addr
);
1552 /* If this is not VM_FLUSH_RESET_PERMS memory, no need for the below. */
1557 * If not deallocating pages, just do the flush of the VM area and
1560 if (!deallocate_pages
) {
1566 * If execution gets here, flush the vm mapping and reset the direct
1567 * map. Find the start and end range of the direct mappings to make sure
1568 * the vm_unmap_aliases() flush includes the direct map.
1570 for (i
= 0; i
< area
->nr_pages
; i
++) {
1571 if (page_address(area
->pages
[i
])) {
1572 start
= min(addr
, start
);
1573 end
= max(addr
, end
);
1578 * Set direct map to something invalid so that it won't be cached if
1579 * there are any accesses after the TLB flush, then flush the TLB and
1580 * reset the direct map permissions to the default.
1582 set_area_direct_map(area
, set_direct_map_invalid_noflush
);
1583 _vm_unmap_aliases(start
, end
, 1);
1584 set_area_direct_map(area
, set_direct_map_default_noflush
);
1587 static void __vunmap(const void *addr
, int deallocate_pages
)
1589 struct vm_struct
*area
;
1594 if (WARN(!PAGE_ALIGNED(addr
), "Trying to vfree() bad address (%p)\n",
1598 area
= find_vm_area(addr
);
1599 if (unlikely(!area
)) {
1600 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1605 debug_check_no_locks_freed(area
->addr
, get_vm_area_size(area
));
1606 debug_check_no_obj_freed(area
->addr
, get_vm_area_size(area
));
1608 vm_remove_mappings(area
, deallocate_pages
);
1610 if (deallocate_pages
) {
1613 for (i
= 0; i
< area
->nr_pages
; i
++) {
1614 struct page
*page
= area
->pages
[i
];
1617 __free_pages(page
, 0);
1620 kvfree(area
->pages
);
1627 static inline void __vfree_deferred(const void *addr
)
1630 * Use raw_cpu_ptr() because this can be called from preemptible
1631 * context. Preemption is absolutely fine here, because the llist_add()
1632 * implementation is lockless, so it works even if we are adding to
1633 * nother cpu's list. schedule_work() should be fine with this too.
1635 struct vfree_deferred
*p
= raw_cpu_ptr(&vfree_deferred
);
1637 if (llist_add((struct llist_node
*)addr
, &p
->list
))
1638 schedule_work(&p
->wq
);
1642 * vfree_atomic - release memory allocated by vmalloc()
1643 * @addr: memory base address
1645 * This one is just like vfree() but can be called in any atomic context
1648 void vfree_atomic(const void *addr
)
1652 kmemleak_free(addr
);
1656 __vfree_deferred(addr
);
1659 static void __vfree(const void *addr
)
1661 if (unlikely(in_interrupt()))
1662 __vfree_deferred(addr
);
1668 * vfree - release memory allocated by vmalloc()
1669 * @addr: memory base address
1671 * Free the virtually continuous memory area starting at @addr, as
1672 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1673 * NULL, no operation is performed.
1675 * Must not be called in NMI context (strictly speaking, only if we don't
1676 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1677 * conventions for vfree() arch-depenedent would be a really bad idea)
1679 * May sleep if called *not* from interrupt context.
1681 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1683 void vfree(const void *addr
)
1687 kmemleak_free(addr
);
1689 might_sleep_if(!in_interrupt());
1696 EXPORT_SYMBOL(vfree
);
1699 * vunmap - release virtual mapping obtained by vmap()
1700 * @addr: memory base address
1702 * Free the virtually contiguous memory area starting at @addr,
1703 * which was created from the page array passed to vmap().
1705 * Must not be called in interrupt context.
1707 void vunmap(const void *addr
)
1709 BUG_ON(in_interrupt());
1714 EXPORT_SYMBOL(vunmap
);
1717 * vmap - map an array of pages into virtually contiguous space
1718 * @pages: array of page pointers
1719 * @count: number of pages to map
1720 * @flags: vm_area->flags
1721 * @prot: page protection for the mapping
1723 * Maps @count pages from @pages into contiguous kernel virtual
1726 * Return: the address of the area or %NULL on failure
1728 void *vmap(struct page
**pages
, unsigned int count
,
1729 unsigned long flags
, pgprot_t prot
)
1731 struct vm_struct
*area
;
1732 unsigned long size
; /* In bytes */
1736 if (count
> totalram_pages())
1739 size
= (unsigned long)count
<< PAGE_SHIFT
;
1740 area
= get_vm_area_caller(size
, flags
, __builtin_return_address(0));
1744 if (map_vm_area(area
, prot
, pages
)) {
1751 EXPORT_SYMBOL(vmap
);
1753 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1754 gfp_t gfp_mask
, pgprot_t prot
,
1755 int node
, const void *caller
);
1756 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1757 pgprot_t prot
, int node
)
1759 struct page
**pages
;
1760 unsigned int nr_pages
, array_size
, i
;
1761 const gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1762 const gfp_t alloc_mask
= gfp_mask
| __GFP_NOWARN
;
1763 const gfp_t highmem_mask
= (gfp_mask
& (GFP_DMA
| GFP_DMA32
)) ?
1767 nr_pages
= get_vm_area_size(area
) >> PAGE_SHIFT
;
1768 array_size
= (nr_pages
* sizeof(struct page
*));
1770 area
->nr_pages
= nr_pages
;
1771 /* Please note that the recursion is strictly bounded. */
1772 if (array_size
> PAGE_SIZE
) {
1773 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|highmem_mask
,
1774 PAGE_KERNEL
, node
, area
->caller
);
1776 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1778 area
->pages
= pages
;
1780 remove_vm_area(area
->addr
);
1785 for (i
= 0; i
< area
->nr_pages
; i
++) {
1788 if (node
== NUMA_NO_NODE
)
1789 page
= alloc_page(alloc_mask
|highmem_mask
);
1791 page
= alloc_pages_node(node
, alloc_mask
|highmem_mask
, 0);
1793 if (unlikely(!page
)) {
1794 /* Successfully allocated i pages, free them in __vunmap() */
1798 area
->pages
[i
] = page
;
1799 if (gfpflags_allow_blocking(gfp_mask
|highmem_mask
))
1803 if (map_vm_area(area
, prot
, pages
))
1808 warn_alloc(gfp_mask
, NULL
,
1809 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1810 (area
->nr_pages
*PAGE_SIZE
), area
->size
);
1811 __vfree(area
->addr
);
1816 * __vmalloc_node_range - allocate virtually contiguous memory
1817 * @size: allocation size
1818 * @align: desired alignment
1819 * @start: vm area range start
1820 * @end: vm area range end
1821 * @gfp_mask: flags for the page level allocator
1822 * @prot: protection mask for the allocated pages
1823 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1824 * @node: node to use for allocation or NUMA_NO_NODE
1825 * @caller: caller's return address
1827 * Allocate enough pages to cover @size from the page level
1828 * allocator with @gfp_mask flags. Map them into contiguous
1829 * kernel virtual space, using a pagetable protection of @prot.
1831 * Return: the address of the area or %NULL on failure
1833 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
1834 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
1835 pgprot_t prot
, unsigned long vm_flags
, int node
,
1838 struct vm_struct
*area
;
1840 unsigned long real_size
= size
;
1842 size
= PAGE_ALIGN(size
);
1843 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages())
1846 area
= __get_vm_area_node(size
, align
, VM_ALLOC
| VM_UNINITIALIZED
|
1847 vm_flags
, start
, end
, node
, gfp_mask
, caller
);
1851 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
);
1856 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1857 * flag. It means that vm_struct is not fully initialized.
1858 * Now, it is fully initialized, so remove this flag here.
1860 clear_vm_uninitialized_flag(area
);
1862 kmemleak_vmalloc(area
, size
, gfp_mask
);
1867 warn_alloc(gfp_mask
, NULL
,
1868 "vmalloc: allocation failure: %lu bytes", real_size
);
1873 * This is only for performance analysis of vmalloc and stress purpose.
1874 * It is required by vmalloc test module, therefore do not use it other
1877 #ifdef CONFIG_TEST_VMALLOC_MODULE
1878 EXPORT_SYMBOL_GPL(__vmalloc_node_range
);
1882 * __vmalloc_node - allocate virtually contiguous memory
1883 * @size: allocation size
1884 * @align: desired alignment
1885 * @gfp_mask: flags for the page level allocator
1886 * @prot: protection mask for the allocated pages
1887 * @node: node to use for allocation or NUMA_NO_NODE
1888 * @caller: caller's return address
1890 * Allocate enough pages to cover @size from the page level
1891 * allocator with @gfp_mask flags. Map them into contiguous
1892 * kernel virtual space, using a pagetable protection of @prot.
1894 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
1895 * and __GFP_NOFAIL are not supported
1897 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1900 * Return: pointer to the allocated memory or %NULL on error
1902 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1903 gfp_t gfp_mask
, pgprot_t prot
,
1904 int node
, const void *caller
)
1906 return __vmalloc_node_range(size
, align
, VMALLOC_START
, VMALLOC_END
,
1907 gfp_mask
, prot
, 0, node
, caller
);
1910 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1912 return __vmalloc_node(size
, 1, gfp_mask
, prot
, NUMA_NO_NODE
,
1913 __builtin_return_address(0));
1915 EXPORT_SYMBOL(__vmalloc
);
1917 static inline void *__vmalloc_node_flags(unsigned long size
,
1918 int node
, gfp_t flags
)
1920 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
,
1921 node
, __builtin_return_address(0));
1925 void *__vmalloc_node_flags_caller(unsigned long size
, int node
, gfp_t flags
,
1928 return __vmalloc_node(size
, 1, flags
, PAGE_KERNEL
, node
, caller
);
1932 * vmalloc - allocate virtually contiguous memory
1933 * @size: allocation size
1935 * Allocate enough pages to cover @size from the page level
1936 * allocator and map them into contiguous kernel virtual space.
1938 * For tight control over page level allocator and protection flags
1939 * use __vmalloc() instead.
1941 * Return: pointer to the allocated memory or %NULL on error
1943 void *vmalloc(unsigned long size
)
1945 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1948 EXPORT_SYMBOL(vmalloc
);
1951 * vzalloc - allocate virtually contiguous memory with zero fill
1952 * @size: allocation size
1954 * Allocate enough pages to cover @size from the page level
1955 * allocator and map them into contiguous kernel virtual space.
1956 * The memory allocated is set to zero.
1958 * For tight control over page level allocator and protection flags
1959 * use __vmalloc() instead.
1961 * Return: pointer to the allocated memory or %NULL on error
1963 void *vzalloc(unsigned long size
)
1965 return __vmalloc_node_flags(size
, NUMA_NO_NODE
,
1966 GFP_KERNEL
| __GFP_ZERO
);
1968 EXPORT_SYMBOL(vzalloc
);
1971 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1972 * @size: allocation size
1974 * The resulting memory area is zeroed so it can be mapped to userspace
1975 * without leaking data.
1977 * Return: pointer to the allocated memory or %NULL on error
1979 void *vmalloc_user(unsigned long size
)
1981 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
1982 GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
,
1983 VM_USERMAP
, NUMA_NO_NODE
,
1984 __builtin_return_address(0));
1986 EXPORT_SYMBOL(vmalloc_user
);
1989 * vmalloc_node - allocate memory on a specific node
1990 * @size: allocation size
1993 * Allocate enough pages to cover @size from the page level
1994 * allocator and map them into contiguous kernel virtual space.
1996 * For tight control over page level allocator and protection flags
1997 * use __vmalloc() instead.
1999 * Return: pointer to the allocated memory or %NULL on error
2001 void *vmalloc_node(unsigned long size
, int node
)
2003 return __vmalloc_node(size
, 1, GFP_KERNEL
, PAGE_KERNEL
,
2004 node
, __builtin_return_address(0));
2006 EXPORT_SYMBOL(vmalloc_node
);
2009 * vzalloc_node - allocate memory on a specific node with zero fill
2010 * @size: allocation size
2013 * Allocate enough pages to cover @size from the page level
2014 * allocator and map them into contiguous kernel virtual space.
2015 * The memory allocated is set to zero.
2017 * For tight control over page level allocator and protection flags
2018 * use __vmalloc_node() instead.
2020 * Return: pointer to the allocated memory or %NULL on error
2022 void *vzalloc_node(unsigned long size
, int node
)
2024 return __vmalloc_node_flags(size
, node
,
2025 GFP_KERNEL
| __GFP_ZERO
);
2027 EXPORT_SYMBOL(vzalloc_node
);
2030 * vmalloc_exec - allocate virtually contiguous, executable memory
2031 * @size: allocation size
2033 * Kernel-internal function to allocate enough pages to cover @size
2034 * the page level allocator and map them into contiguous and
2035 * executable kernel virtual space.
2037 * For tight control over page level allocator and protection flags
2038 * use __vmalloc() instead.
2040 * Return: pointer to the allocated memory or %NULL on error
2042 void *vmalloc_exec(unsigned long size
)
2044 return __vmalloc_node_range(size
, 1, VMALLOC_START
, VMALLOC_END
,
2045 GFP_KERNEL
, PAGE_KERNEL_EXEC
, VM_FLUSH_RESET_PERMS
,
2046 NUMA_NO_NODE
, __builtin_return_address(0));
2049 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
2050 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
2051 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
2052 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
2055 * 64b systems should always have either DMA or DMA32 zones. For others
2056 * GFP_DMA32 should do the right thing and use the normal zone.
2058 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
2062 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
2063 * @size: allocation size
2065 * Allocate enough 32bit PA addressable pages to cover @size from the
2066 * page level allocator and map them into contiguous kernel virtual space.
2068 * Return: pointer to the allocated memory or %NULL on error
2070 void *vmalloc_32(unsigned long size
)
2072 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
2073 NUMA_NO_NODE
, __builtin_return_address(0));
2075 EXPORT_SYMBOL(vmalloc_32
);
2078 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
2079 * @size: allocation size
2081 * The resulting memory area is 32bit addressable and zeroed so it can be
2082 * mapped to userspace without leaking data.
2084 * Return: pointer to the allocated memory or %NULL on error
2086 void *vmalloc_32_user(unsigned long size
)
2088 return __vmalloc_node_range(size
, SHMLBA
, VMALLOC_START
, VMALLOC_END
,
2089 GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
2090 VM_USERMAP
, NUMA_NO_NODE
,
2091 __builtin_return_address(0));
2093 EXPORT_SYMBOL(vmalloc_32_user
);
2096 * small helper routine , copy contents to buf from addr.
2097 * If the page is not present, fill zero.
2100 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
2106 unsigned long offset
, length
;
2108 offset
= offset_in_page(addr
);
2109 length
= PAGE_SIZE
- offset
;
2112 p
= vmalloc_to_page(addr
);
2114 * To do safe access to this _mapped_ area, we need
2115 * lock. But adding lock here means that we need to add
2116 * overhead of vmalloc()/vfree() calles for this _debug_
2117 * interface, rarely used. Instead of that, we'll use
2118 * kmap() and get small overhead in this access function.
2122 * we can expect USER0 is not used (see vread/vwrite's
2123 * function description)
2125 void *map
= kmap_atomic(p
);
2126 memcpy(buf
, map
+ offset
, length
);
2129 memset(buf
, 0, length
);
2139 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
2145 unsigned long offset
, length
;
2147 offset
= offset_in_page(addr
);
2148 length
= PAGE_SIZE
- offset
;
2151 p
= vmalloc_to_page(addr
);
2153 * To do safe access to this _mapped_ area, we need
2154 * lock. But adding lock here means that we need to add
2155 * overhead of vmalloc()/vfree() calles for this _debug_
2156 * interface, rarely used. Instead of that, we'll use
2157 * kmap() and get small overhead in this access function.
2161 * we can expect USER0 is not used (see vread/vwrite's
2162 * function description)
2164 void *map
= kmap_atomic(p
);
2165 memcpy(map
+ offset
, buf
, length
);
2177 * vread() - read vmalloc area in a safe way.
2178 * @buf: buffer for reading data
2179 * @addr: vm address.
2180 * @count: number of bytes to be read.
2182 * This function checks that addr is a valid vmalloc'ed area, and
2183 * copy data from that area to a given buffer. If the given memory range
2184 * of [addr...addr+count) includes some valid address, data is copied to
2185 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2186 * IOREMAP area is treated as memory hole and no copy is done.
2188 * If [addr...addr+count) doesn't includes any intersects with alive
2189 * vm_struct area, returns 0. @buf should be kernel's buffer.
2191 * Note: In usual ops, vread() is never necessary because the caller
2192 * should know vmalloc() area is valid and can use memcpy().
2193 * This is for routines which have to access vmalloc area without
2194 * any informaion, as /dev/kmem.
2196 * Return: number of bytes for which addr and buf should be increased
2197 * (same number as @count) or %0 if [addr...addr+count) doesn't
2198 * include any intersection with valid vmalloc area
2200 long vread(char *buf
, char *addr
, unsigned long count
)
2202 struct vmap_area
*va
;
2203 struct vm_struct
*vm
;
2204 char *vaddr
, *buf_start
= buf
;
2205 unsigned long buflen
= count
;
2208 /* Don't allow overflow */
2209 if ((unsigned long) addr
+ count
< count
)
2210 count
= -(unsigned long) addr
;
2212 spin_lock(&vmap_area_lock
);
2213 list_for_each_entry(va
, &vmap_area_list
, list
) {
2217 if (!(va
->flags
& VM_VM_AREA
))
2221 vaddr
= (char *) vm
->addr
;
2222 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2224 while (addr
< vaddr
) {
2232 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2235 if (!(vm
->flags
& VM_IOREMAP
))
2236 aligned_vread(buf
, addr
, n
);
2237 else /* IOREMAP area is treated as memory hole */
2244 spin_unlock(&vmap_area_lock
);
2246 if (buf
== buf_start
)
2248 /* zero-fill memory holes */
2249 if (buf
!= buf_start
+ buflen
)
2250 memset(buf
, 0, buflen
- (buf
- buf_start
));
2256 * vwrite() - write vmalloc area in a safe way.
2257 * @buf: buffer for source data
2258 * @addr: vm address.
2259 * @count: number of bytes to be read.
2261 * This function checks that addr is a valid vmalloc'ed area, and
2262 * copy data from a buffer to the given addr. If specified range of
2263 * [addr...addr+count) includes some valid address, data is copied from
2264 * proper area of @buf. If there are memory holes, no copy to hole.
2265 * IOREMAP area is treated as memory hole and no copy is done.
2267 * If [addr...addr+count) doesn't includes any intersects with alive
2268 * vm_struct area, returns 0. @buf should be kernel's buffer.
2270 * Note: In usual ops, vwrite() is never necessary because the caller
2271 * should know vmalloc() area is valid and can use memcpy().
2272 * This is for routines which have to access vmalloc area without
2273 * any informaion, as /dev/kmem.
2275 * Return: number of bytes for which addr and buf should be
2276 * increased (same number as @count) or %0 if [addr...addr+count)
2277 * doesn't include any intersection with valid vmalloc area
2279 long vwrite(char *buf
, char *addr
, unsigned long count
)
2281 struct vmap_area
*va
;
2282 struct vm_struct
*vm
;
2284 unsigned long n
, buflen
;
2287 /* Don't allow overflow */
2288 if ((unsigned long) addr
+ count
< count
)
2289 count
= -(unsigned long) addr
;
2292 spin_lock(&vmap_area_lock
);
2293 list_for_each_entry(va
, &vmap_area_list
, list
) {
2297 if (!(va
->flags
& VM_VM_AREA
))
2301 vaddr
= (char *) vm
->addr
;
2302 if (addr
>= vaddr
+ get_vm_area_size(vm
))
2304 while (addr
< vaddr
) {
2311 n
= vaddr
+ get_vm_area_size(vm
) - addr
;
2314 if (!(vm
->flags
& VM_IOREMAP
)) {
2315 aligned_vwrite(buf
, addr
, n
);
2323 spin_unlock(&vmap_area_lock
);
2330 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2331 * @vma: vma to cover
2332 * @uaddr: target user address to start at
2333 * @kaddr: virtual address of vmalloc kernel memory
2334 * @size: size of map area
2336 * Returns: 0 for success, -Exxx on failure
2338 * This function checks that @kaddr is a valid vmalloc'ed area,
2339 * and that it is big enough to cover the range starting at
2340 * @uaddr in @vma. Will return failure if that criteria isn't
2343 * Similar to remap_pfn_range() (see mm/memory.c)
2345 int remap_vmalloc_range_partial(struct vm_area_struct
*vma
, unsigned long uaddr
,
2346 void *kaddr
, unsigned long size
)
2348 struct vm_struct
*area
;
2350 size
= PAGE_ALIGN(size
);
2352 if (!PAGE_ALIGNED(uaddr
) || !PAGE_ALIGNED(kaddr
))
2355 area
= find_vm_area(kaddr
);
2359 if (!(area
->flags
& VM_USERMAP
))
2362 if (kaddr
+ size
> area
->addr
+ get_vm_area_size(area
))
2366 struct page
*page
= vmalloc_to_page(kaddr
);
2369 ret
= vm_insert_page(vma
, uaddr
, page
);
2378 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
2382 EXPORT_SYMBOL(remap_vmalloc_range_partial
);
2385 * remap_vmalloc_range - map vmalloc pages to userspace
2386 * @vma: vma to cover (map full range of vma)
2387 * @addr: vmalloc memory
2388 * @pgoff: number of pages into addr before first page to map
2390 * Returns: 0 for success, -Exxx on failure
2392 * This function checks that addr is a valid vmalloc'ed area, and
2393 * that it is big enough to cover the vma. Will return failure if
2394 * that criteria isn't met.
2396 * Similar to remap_pfn_range() (see mm/memory.c)
2398 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
2399 unsigned long pgoff
)
2401 return remap_vmalloc_range_partial(vma
, vma
->vm_start
,
2402 addr
+ (pgoff
<< PAGE_SHIFT
),
2403 vma
->vm_end
- vma
->vm_start
);
2405 EXPORT_SYMBOL(remap_vmalloc_range
);
2408 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2411 void __weak
vmalloc_sync_all(void)
2416 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
2428 * alloc_vm_area - allocate a range of kernel address space
2429 * @size: size of the area
2430 * @ptes: returns the PTEs for the address space
2432 * Returns: NULL on failure, vm_struct on success
2434 * This function reserves a range of kernel address space, and
2435 * allocates pagetables to map that range. No actual mappings
2438 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2439 * allocated for the VM area are returned.
2441 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
2443 struct vm_struct
*area
;
2445 area
= get_vm_area_caller(size
, VM_IOREMAP
,
2446 __builtin_return_address(0));
2451 * This ensures that page tables are constructed for this region
2452 * of kernel virtual address space and mapped into init_mm.
2454 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
2455 size
, f
, ptes
? &ptes
: NULL
)) {
2462 EXPORT_SYMBOL_GPL(alloc_vm_area
);
2464 void free_vm_area(struct vm_struct
*area
)
2466 struct vm_struct
*ret
;
2467 ret
= remove_vm_area(area
->addr
);
2468 BUG_ON(ret
!= area
);
2471 EXPORT_SYMBOL_GPL(free_vm_area
);
2474 static struct vmap_area
*node_to_va(struct rb_node
*n
)
2476 return rb_entry_safe(n
, struct vmap_area
, rb_node
);
2480 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2481 * @end: target address
2482 * @pnext: out arg for the next vmap_area
2483 * @pprev: out arg for the previous vmap_area
2485 * Returns: %true if either or both of next and prev are found,
2486 * %false if no vmap_area exists
2488 * Find vmap_areas end addresses of which enclose @end. ie. if not
2489 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2491 static bool pvm_find_next_prev(unsigned long end
,
2492 struct vmap_area
**pnext
,
2493 struct vmap_area
**pprev
)
2495 struct rb_node
*n
= vmap_area_root
.rb_node
;
2496 struct vmap_area
*va
= NULL
;
2499 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2500 if (end
< va
->va_end
)
2502 else if (end
> va
->va_end
)
2511 if (va
->va_end
> end
) {
2513 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2516 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2522 * pvm_determine_end - find the highest aligned address between two vmap_areas
2523 * @pnext: in/out arg for the next vmap_area
2524 * @pprev: in/out arg for the previous vmap_area
2527 * Returns: determined end address
2529 * Find the highest aligned address between *@pnext and *@pprev below
2530 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2531 * down address is between the end addresses of the two vmap_areas.
2533 * Please note that the address returned by this function may fall
2534 * inside *@pnext vmap_area. The caller is responsible for checking
2537 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2538 struct vmap_area
**pprev
,
2539 unsigned long align
)
2541 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2545 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2549 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2551 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2558 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2559 * @offsets: array containing offset of each area
2560 * @sizes: array containing size of each area
2561 * @nr_vms: the number of areas to allocate
2562 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2564 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2565 * vm_structs on success, %NULL on failure
2567 * Percpu allocator wants to use congruent vm areas so that it can
2568 * maintain the offsets among percpu areas. This function allocates
2569 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2570 * be scattered pretty far, distance between two areas easily going up
2571 * to gigabytes. To avoid interacting with regular vmallocs, these
2572 * areas are allocated from top.
2574 * Despite its complicated look, this allocator is rather simple. It
2575 * does everything top-down and scans areas from the end looking for
2576 * matching slot. While scanning, if any of the areas overlaps with
2577 * existing vmap_area, the base address is pulled down to fit the
2578 * area. Scanning is repeated till all the areas fit and then all
2579 * necessary data structures are inserted and the result is returned.
2581 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2582 const size_t *sizes
, int nr_vms
,
2585 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2586 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2587 struct vmap_area
**vas
, *prev
, *next
;
2588 struct vm_struct
**vms
;
2589 int area
, area2
, last_area
, term_area
;
2590 unsigned long base
, start
, end
, last_end
;
2591 bool purged
= false;
2593 /* verify parameters and allocate data structures */
2594 BUG_ON(offset_in_page(align
) || !is_power_of_2(align
));
2595 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2596 start
= offsets
[area
];
2597 end
= start
+ sizes
[area
];
2599 /* is everything aligned properly? */
2600 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2601 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2603 /* detect the area with the highest address */
2604 if (start
> offsets
[last_area
])
2607 for (area2
= area
+ 1; area2
< nr_vms
; area2
++) {
2608 unsigned long start2
= offsets
[area2
];
2609 unsigned long end2
= start2
+ sizes
[area2
];
2611 BUG_ON(start2
< end
&& start
< end2
);
2614 last_end
= offsets
[last_area
] + sizes
[last_area
];
2616 if (vmalloc_end
- vmalloc_start
< last_end
) {
2621 vms
= kcalloc(nr_vms
, sizeof(vms
[0]), GFP_KERNEL
);
2622 vas
= kcalloc(nr_vms
, sizeof(vas
[0]), GFP_KERNEL
);
2626 for (area
= 0; area
< nr_vms
; area
++) {
2627 vas
[area
] = kzalloc(sizeof(struct vmap_area
), GFP_KERNEL
);
2628 vms
[area
] = kzalloc(sizeof(struct vm_struct
), GFP_KERNEL
);
2629 if (!vas
[area
] || !vms
[area
])
2633 spin_lock(&vmap_area_lock
);
2635 /* start scanning - we scan from the top, begin with the last area */
2636 area
= term_area
= last_area
;
2637 start
= offsets
[area
];
2638 end
= start
+ sizes
[area
];
2640 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2641 base
= vmalloc_end
- last_end
;
2644 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2647 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2648 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2651 * base might have underflowed, add last_end before
2654 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2655 spin_unlock(&vmap_area_lock
);
2657 purge_vmap_area_lazy();
2665 * If next overlaps, move base downwards so that it's
2666 * right below next and then recheck.
2668 if (next
&& next
->va_start
< base
+ end
) {
2669 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2675 * If prev overlaps, shift down next and prev and move
2676 * base so that it's right below new next and then
2679 if (prev
&& prev
->va_end
> base
+ start
) {
2681 prev
= node_to_va(rb_prev(&next
->rb_node
));
2682 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2688 * This area fits, move on to the previous one. If
2689 * the previous one is the terminal one, we're done.
2691 area
= (area
+ nr_vms
- 1) % nr_vms
;
2692 if (area
== term_area
)
2694 start
= offsets
[area
];
2695 end
= start
+ sizes
[area
];
2696 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2699 /* we've found a fitting base, insert all va's */
2700 for (area
= 0; area
< nr_vms
; area
++) {
2701 struct vmap_area
*va
= vas
[area
];
2703 va
->va_start
= base
+ offsets
[area
];
2704 va
->va_end
= va
->va_start
+ sizes
[area
];
2705 __insert_vmap_area(va
);
2708 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2710 spin_unlock(&vmap_area_lock
);
2712 /* insert all vm's */
2713 for (area
= 0; area
< nr_vms
; area
++)
2714 setup_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2721 for (area
= 0; area
< nr_vms
; area
++) {
2732 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2733 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2734 * @nr_vms: the number of allocated areas
2736 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2738 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2742 for (i
= 0; i
< nr_vms
; i
++)
2743 free_vm_area(vms
[i
]);
2746 #endif /* CONFIG_SMP */
2748 #ifdef CONFIG_PROC_FS
2749 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2750 __acquires(&vmap_area_lock
)
2752 spin_lock(&vmap_area_lock
);
2753 return seq_list_start(&vmap_area_list
, *pos
);
2756 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2758 return seq_list_next(p
, &vmap_area_list
, pos
);
2761 static void s_stop(struct seq_file
*m
, void *p
)
2762 __releases(&vmap_area_lock
)
2764 spin_unlock(&vmap_area_lock
);
2767 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2769 if (IS_ENABLED(CONFIG_NUMA
)) {
2770 unsigned int nr
, *counters
= m
->private;
2775 if (v
->flags
& VM_UNINITIALIZED
)
2777 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2780 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2782 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2783 counters
[page_to_nid(v
->pages
[nr
])]++;
2785 for_each_node_state(nr
, N_HIGH_MEMORY
)
2787 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2791 static int s_show(struct seq_file
*m
, void *p
)
2793 struct vmap_area
*va
;
2794 struct vm_struct
*v
;
2796 va
= list_entry(p
, struct vmap_area
, list
);
2799 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2800 * behalf of vmap area is being tear down or vm_map_ram allocation.
2802 if (!(va
->flags
& VM_VM_AREA
)) {
2803 seq_printf(m
, "0x%pK-0x%pK %7ld %s\n",
2804 (void *)va
->va_start
, (void *)va
->va_end
,
2805 va
->va_end
- va
->va_start
,
2806 va
->flags
& VM_LAZY_FREE
? "unpurged vm_area" : "vm_map_ram");
2813 seq_printf(m
, "0x%pK-0x%pK %7ld",
2814 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2817 seq_printf(m
, " %pS", v
->caller
);
2820 seq_printf(m
, " pages=%d", v
->nr_pages
);
2823 seq_printf(m
, " phys=%pa", &v
->phys_addr
);
2825 if (v
->flags
& VM_IOREMAP
)
2826 seq_puts(m
, " ioremap");
2828 if (v
->flags
& VM_ALLOC
)
2829 seq_puts(m
, " vmalloc");
2831 if (v
->flags
& VM_MAP
)
2832 seq_puts(m
, " vmap");
2834 if (v
->flags
& VM_USERMAP
)
2835 seq_puts(m
, " user");
2837 if (is_vmalloc_addr(v
->pages
))
2838 seq_puts(m
, " vpages");
2840 show_numa_info(m
, v
);
2845 static const struct seq_operations vmalloc_op
= {
2852 static int __init
proc_vmalloc_init(void)
2854 if (IS_ENABLED(CONFIG_NUMA
))
2855 proc_create_seq_private("vmallocinfo", 0400, NULL
,
2857 nr_node_ids
* sizeof(unsigned int), NULL
);
2859 proc_create_seq("vmallocinfo", 0400, NULL
, &vmalloc_op
);
2862 module_init(proc_vmalloc_init
);