4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8 * Numa awareness, Christoph Lameter, SGI, June 2005
11 #include <linux/vmalloc.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/rbtree.h>
25 #include <linux/radix-tree.h>
26 #include <linux/rcupdate.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
29 #include <linux/atomic.h>
30 #include <asm/uaccess.h>
31 #include <asm/tlbflush.h>
32 #include <asm/shmparam.h>
34 /*** Page table manipulation functions ***/
36 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
40 pte = pte_offset_kernel(pmd, addr);
42 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
43 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
44 } while (pte++, addr += PAGE_SIZE, addr != end);
47 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
52 pmd = pmd_offset(pud, addr);
54 next = pmd_addr_end(addr, end);
55 if (pmd_none_or_clear_bad(pmd))
57 vunmap_pte_range(pmd, addr, next);
58 } while (pmd++, addr = next, addr != end);
61 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
66 pud = pud_offset(pgd, addr);
68 next = pud_addr_end(addr, end);
69 if (pud_none_or_clear_bad(pud))
71 vunmap_pmd_range(pud, addr, next);
72 } while (pud++, addr = next, addr != end);
75 static void vunmap_page_range(unsigned long addr, unsigned long end)
81 pgd = pgd_offset_k(addr);
83 next = pgd_addr_end(addr, end);
84 if (pgd_none_or_clear_bad(pgd))
86 vunmap_pud_range(pgd, addr, next);
87 } while (pgd++, addr = next, addr != end);
90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
91 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
100 pte = pte_alloc_kernel(pmd, addr);
104 struct page *page = pages[*nr];
106 if (WARN_ON(!pte_none(*pte)))
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
112 } while (pte++, addr += PAGE_SIZE, addr != end);
116 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
122 pmd = pmd_alloc(&init_mm, pud, addr);
126 next = pmd_addr_end(addr, end);
127 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
129 } while (pmd++, addr = next, addr != end);
133 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
139 pud = pud_alloc(&init_mm, pgd, addr);
143 next = pud_addr_end(addr, end);
144 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
146 } while (pud++, addr = next, addr != end);
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
156 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157 pgprot_t prot, struct page **pages)
161 unsigned long addr = start;
166 pgd = pgd_offset_k(addr);
168 next = pgd_addr_end(addr, end);
169 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
172 } while (pgd++, addr = next, addr != end);
177 static int vmap_page_range(unsigned long start, unsigned long end,
178 pgprot_t prot, struct page **pages)
182 ret = vmap_page_range_noflush(start, end, prot, pages);
183 flush_cache_vmap(start, end);
187 int is_vmalloc_or_module_addr(const void *x)
190 * ARM, x86-64 and sparc64 put modules in a special place,
191 * and fall back on vmalloc() if that fails. Others
192 * just put it in the vmalloc space.
194 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
195 unsigned long addr = (unsigned long)x;
196 if (addr >= MODULES_VADDR && addr < MODULES_END)
199 return is_vmalloc_addr(x);
203 * Walk a vmap address to the struct page it maps.
205 struct page *vmalloc_to_page(const void *vmalloc_addr)
207 unsigned long addr = (unsigned long) vmalloc_addr;
208 struct page *page = NULL;
209 pgd_t *pgd = pgd_offset_k(addr);
212 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
213 * architectures that do not vmalloc module space
215 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
217 if (!pgd_none(*pgd)) {
218 pud_t *pud = pud_offset(pgd, addr);
219 if (!pud_none(*pud)) {
220 pmd_t *pmd = pmd_offset(pud, addr);
221 if (!pmd_none(*pmd)) {
224 ptep = pte_offset_map(pmd, addr);
226 if (pte_present(pte))
227 page = pte_page(pte);
234 EXPORT_SYMBOL(vmalloc_to_page);
237 * Map a vmalloc()-space virtual address to the physical page frame number.
239 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
241 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
243 EXPORT_SYMBOL(vmalloc_to_pfn);
246 /*** Global kva allocator ***/
248 #define VM_LAZY_FREE 0x01
249 #define VM_LAZY_FREEING 0x02
250 #define VM_VM_AREA 0x04
253 unsigned long va_start;
254 unsigned long va_end;
256 struct rb_node rb_node; /* address sorted rbtree */
257 struct list_head list; /* address sorted list */
258 struct list_head purge_list; /* "lazy purge" list */
259 struct vm_struct *vm;
260 struct rcu_head rcu_head;
263 static DEFINE_SPINLOCK(vmap_area_lock);
264 /* Export for kexec only */
265 LIST_HEAD(vmap_area_list);
266 static struct rb_root vmap_area_root = RB_ROOT;
268 /* The vmap cache globals are protected by vmap_area_lock */
269 static struct rb_node *free_vmap_cache;
270 static unsigned long cached_hole_size;
271 static unsigned long cached_vstart;
272 static unsigned long cached_align;
274 static unsigned long vmap_area_pcpu_hole;
276 static struct vmap_area *__find_vmap_area(unsigned long addr)
278 struct rb_node *n = vmap_area_root.rb_node;
281 struct vmap_area *va;
283 va = rb_entry(n, struct vmap_area, rb_node);
284 if (addr < va->va_start)
286 else if (addr > va->va_start)
295 static void __insert_vmap_area(struct vmap_area *va)
297 struct rb_node **p = &vmap_area_root.rb_node;
298 struct rb_node *parent = NULL;
302 struct vmap_area *tmp_va;
305 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
306 if (va->va_start < tmp_va->va_end)
308 else if (va->va_end > tmp_va->va_start)
314 rb_link_node(&va->rb_node, parent, p);
315 rb_insert_color(&va->rb_node, &vmap_area_root);
317 /* address-sort this list */
318 tmp = rb_prev(&va->rb_node);
320 struct vmap_area *prev;
321 prev = rb_entry(tmp, struct vmap_area, rb_node);
322 list_add_rcu(&va->list, &prev->list);
324 list_add_rcu(&va->list, &vmap_area_list);
327 static void purge_vmap_area_lazy(void);
330 * Allocate a region of KVA of the specified size and alignment, within the
333 static struct vmap_area *alloc_vmap_area(unsigned long size,
335 unsigned long vstart, unsigned long vend,
336 int node, gfp_t gfp_mask)
338 struct vmap_area *va;
342 struct vmap_area *first;
345 BUG_ON(size & ~PAGE_MASK);
346 BUG_ON(!is_power_of_2(align));
348 va = kmalloc_node(sizeof(struct vmap_area),
349 gfp_mask & GFP_RECLAIM_MASK, node);
351 return ERR_PTR(-ENOMEM);
354 spin_lock(&vmap_area_lock);
356 * Invalidate cache if we have more permissive parameters.
357 * cached_hole_size notes the largest hole noticed _below_
358 * the vmap_area cached in free_vmap_cache: if size fits
359 * into that hole, we want to scan from vstart to reuse
360 * the hole instead of allocating above free_vmap_cache.
361 * Note that __free_vmap_area may update free_vmap_cache
362 * without updating cached_hole_size or cached_align.
364 if (!free_vmap_cache ||
365 size < cached_hole_size ||
366 vstart < cached_vstart ||
367 align < cached_align) {
369 cached_hole_size = 0;
370 free_vmap_cache = NULL;
372 /* record if we encounter less permissive parameters */
373 cached_vstart = vstart;
374 cached_align = align;
376 /* find starting point for our search */
377 if (free_vmap_cache) {
378 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
379 addr = ALIGN(first->va_end, align);
382 if (addr + size - 1 < addr)
386 addr = ALIGN(vstart, align);
387 if (addr + size - 1 < addr)
390 n = vmap_area_root.rb_node;
394 struct vmap_area *tmp;
395 tmp = rb_entry(n, struct vmap_area, rb_node);
396 if (tmp->va_end >= addr) {
398 if (tmp->va_start <= addr)
409 /* from the starting point, walk areas until a suitable hole is found */
410 while (addr + size > first->va_start && addr + size <= vend) {
411 if (addr + cached_hole_size < first->va_start)
412 cached_hole_size = first->va_start - addr;
413 addr = ALIGN(first->va_end, align);
414 if (addr + size - 1 < addr)
417 if (list_is_last(&first->list, &vmap_area_list))
420 first = list_entry(first->list.next,
421 struct vmap_area, list);
425 if (addr + size > vend)
429 va->va_end = addr + size;
431 __insert_vmap_area(va);
432 free_vmap_cache = &va->rb_node;
433 spin_unlock(&vmap_area_lock);
435 BUG_ON(va->va_start & (align-1));
436 BUG_ON(va->va_start < vstart);
437 BUG_ON(va->va_end > vend);
442 spin_unlock(&vmap_area_lock);
444 purge_vmap_area_lazy();
448 if (printk_ratelimit())
450 "vmap allocation for size %lu failed: "
451 "use vmalloc=<size> to increase size.\n", size);
453 return ERR_PTR(-EBUSY);
456 static void __free_vmap_area(struct vmap_area *va)
458 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
460 if (free_vmap_cache) {
461 if (va->va_end < cached_vstart) {
462 free_vmap_cache = NULL;
464 struct vmap_area *cache;
465 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
466 if (va->va_start <= cache->va_start) {
467 free_vmap_cache = rb_prev(&va->rb_node);
469 * We don't try to update cached_hole_size or
470 * cached_align, but it won't go very wrong.
475 rb_erase(&va->rb_node, &vmap_area_root);
476 RB_CLEAR_NODE(&va->rb_node);
477 list_del_rcu(&va->list);
480 * Track the highest possible candidate for pcpu area
481 * allocation. Areas outside of vmalloc area can be returned
482 * here too, consider only end addresses which fall inside
483 * vmalloc area proper.
485 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
486 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
488 kfree_rcu(va, rcu_head);
492 * Free a region of KVA allocated by alloc_vmap_area
494 static void free_vmap_area(struct vmap_area *va)
496 spin_lock(&vmap_area_lock);
497 __free_vmap_area(va);
498 spin_unlock(&vmap_area_lock);
502 * Clear the pagetable entries of a given vmap_area
504 static void unmap_vmap_area(struct vmap_area *va)
506 vunmap_page_range(va->va_start, va->va_end);
509 static void vmap_debug_free_range(unsigned long start, unsigned long end)
512 * Unmap page tables and force a TLB flush immediately if
513 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
514 * bugs similarly to those in linear kernel virtual address
515 * space after a page has been freed.
517 * All the lazy freeing logic is still retained, in order to
518 * minimise intrusiveness of this debugging feature.
520 * This is going to be *slow* (linear kernel virtual address
521 * debugging doesn't do a broadcast TLB flush so it is a lot
524 #ifdef CONFIG_DEBUG_PAGEALLOC
525 vunmap_page_range(start, end);
526 flush_tlb_kernel_range(start, end);
531 * lazy_max_pages is the maximum amount of virtual address space we gather up
532 * before attempting to purge with a TLB flush.
534 * There is a tradeoff here: a larger number will cover more kernel page tables
535 * and take slightly longer to purge, but it will linearly reduce the number of
536 * global TLB flushes that must be performed. It would seem natural to scale
537 * this number up linearly with the number of CPUs (because vmapping activity
538 * could also scale linearly with the number of CPUs), however it is likely
539 * that in practice, workloads might be constrained in other ways that mean
540 * vmap activity will not scale linearly with CPUs. Also, I want to be
541 * conservative and not introduce a big latency on huge systems, so go with
542 * a less aggressive log scale. It will still be an improvement over the old
543 * code, and it will be simple to change the scale factor if we find that it
544 * becomes a problem on bigger systems.
546 static unsigned long lazy_max_pages(void)
550 log = fls(num_online_cpus());
552 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
555 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
557 /* for per-CPU blocks */
558 static void purge_fragmented_blocks_allcpus(void);
561 * called before a call to iounmap() if the caller wants vm_area_struct's
564 void set_iounmap_nonlazy(void)
566 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
570 * Purges all lazily-freed vmap areas.
572 * If sync is 0 then don't purge if there is already a purge in progress.
573 * If force_flush is 1, then flush kernel TLBs between *start and *end even
574 * if we found no lazy vmap areas to unmap (callers can use this to optimise
575 * their own TLB flushing).
576 * Returns with *start = min(*start, lowest purged address)
577 * *end = max(*end, highest purged address)
579 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
580 int sync, int force_flush)
582 static DEFINE_SPINLOCK(purge_lock);
584 struct vmap_area *va;
585 struct vmap_area *n_va;
589 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
590 * should not expect such behaviour. This just simplifies locking for
591 * the case that isn't actually used at the moment anyway.
593 if (!sync && !force_flush) {
594 if (!spin_trylock(&purge_lock))
597 spin_lock(&purge_lock);
600 purge_fragmented_blocks_allcpus();
603 list_for_each_entry_rcu(va, &vmap_area_list, list) {
604 if (va->flags & VM_LAZY_FREE) {
605 if (va->va_start < *start)
606 *start = va->va_start;
607 if (va->va_end > *end)
609 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
610 list_add_tail(&va->purge_list, &valist);
611 va->flags |= VM_LAZY_FREEING;
612 va->flags &= ~VM_LAZY_FREE;
618 atomic_sub(nr, &vmap_lazy_nr);
620 if (nr || force_flush)
621 flush_tlb_kernel_range(*start, *end);
624 spin_lock(&vmap_area_lock);
625 list_for_each_entry_safe(va, n_va, &valist, purge_list)
626 __free_vmap_area(va);
627 spin_unlock(&vmap_area_lock);
629 spin_unlock(&purge_lock);
633 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
634 * is already purging.
636 static void try_purge_vmap_area_lazy(void)
638 unsigned long start = ULONG_MAX, end = 0;
640 __purge_vmap_area_lazy(&start, &end, 0, 0);
644 * Kick off a purge of the outstanding lazy areas.
646 static void purge_vmap_area_lazy(void)
648 unsigned long start = ULONG_MAX, end = 0;
650 __purge_vmap_area_lazy(&start, &end, 1, 0);
654 * Free a vmap area, caller ensuring that the area has been unmapped
655 * and flush_cache_vunmap had been called for the correct range
658 static void free_vmap_area_noflush(struct vmap_area *va)
660 va->flags |= VM_LAZY_FREE;
661 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
662 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
663 try_purge_vmap_area_lazy();
667 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
668 * called for the correct range previously.
670 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
673 free_vmap_area_noflush(va);
677 * Free and unmap a vmap area
679 static void free_unmap_vmap_area(struct vmap_area *va)
681 flush_cache_vunmap(va->va_start, va->va_end);
682 free_unmap_vmap_area_noflush(va);
685 static struct vmap_area *find_vmap_area(unsigned long addr)
687 struct vmap_area *va;
689 spin_lock(&vmap_area_lock);
690 va = __find_vmap_area(addr);
691 spin_unlock(&vmap_area_lock);
696 static void free_unmap_vmap_area_addr(unsigned long addr)
698 struct vmap_area *va;
700 va = find_vmap_area(addr);
702 free_unmap_vmap_area(va);
706 /*** Per cpu kva allocator ***/
709 * vmap space is limited especially on 32 bit architectures. Ensure there is
710 * room for at least 16 percpu vmap blocks per CPU.
713 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
714 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
715 * instead (we just need a rough idea)
717 #if BITS_PER_LONG == 32
718 #define VMALLOC_SPACE (128UL*1024*1024)
720 #define VMALLOC_SPACE (128UL*1024*1024*1024)
723 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
724 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
725 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
726 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
727 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
728 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
729 #define VMAP_BBMAP_BITS \
730 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
731 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
732 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
734 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
736 static bool vmap_initialized __read_mostly = false;
738 struct vmap_block_queue {
740 struct list_head free;
745 struct vmap_area *va;
746 struct vmap_block_queue *vbq;
747 unsigned long free, dirty;
748 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
749 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
750 struct list_head free_list;
751 struct rcu_head rcu_head;
752 struct list_head purge;
755 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
756 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
759 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
760 * in the free path. Could get rid of this if we change the API to return a
761 * "cookie" from alloc, to be passed to free. But no big deal yet.
763 static DEFINE_SPINLOCK(vmap_block_tree_lock);
764 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
767 * We should probably have a fallback mechanism to allocate virtual memory
768 * out of partially filled vmap blocks. However vmap block sizing should be
769 * fairly reasonable according to the vmalloc size, so it shouldn't be a
773 static unsigned long addr_to_vb_idx(unsigned long addr)
775 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
776 addr /= VMAP_BLOCK_SIZE;
780 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
782 struct vmap_block_queue *vbq;
783 struct vmap_block *vb;
784 struct vmap_area *va;
785 unsigned long vb_idx;
788 node = numa_node_id();
790 vb = kmalloc_node(sizeof(struct vmap_block),
791 gfp_mask & GFP_RECLAIM_MASK, node);
793 return ERR_PTR(-ENOMEM);
795 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
796 VMALLOC_START, VMALLOC_END,
803 err = radix_tree_preload(gfp_mask);
810 spin_lock_init(&vb->lock);
812 vb->free = VMAP_BBMAP_BITS;
814 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
815 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
816 INIT_LIST_HEAD(&vb->free_list);
818 vb_idx = addr_to_vb_idx(va->va_start);
819 spin_lock(&vmap_block_tree_lock);
820 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
821 spin_unlock(&vmap_block_tree_lock);
823 radix_tree_preload_end();
825 vbq = &get_cpu_var(vmap_block_queue);
827 spin_lock(&vbq->lock);
828 list_add_rcu(&vb->free_list, &vbq->free);
829 spin_unlock(&vbq->lock);
830 put_cpu_var(vmap_block_queue);
835 static void free_vmap_block(struct vmap_block *vb)
837 struct vmap_block *tmp;
838 unsigned long vb_idx;
840 vb_idx = addr_to_vb_idx(vb->va->va_start);
841 spin_lock(&vmap_block_tree_lock);
842 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
843 spin_unlock(&vmap_block_tree_lock);
846 free_vmap_area_noflush(vb->va);
847 kfree_rcu(vb, rcu_head);
850 static void purge_fragmented_blocks(int cpu)
853 struct vmap_block *vb;
854 struct vmap_block *n_vb;
855 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
858 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
860 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
863 spin_lock(&vb->lock);
864 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
865 vb->free = 0; /* prevent further allocs after releasing lock */
866 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
867 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
868 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
869 spin_lock(&vbq->lock);
870 list_del_rcu(&vb->free_list);
871 spin_unlock(&vbq->lock);
872 spin_unlock(&vb->lock);
873 list_add_tail(&vb->purge, &purge);
875 spin_unlock(&vb->lock);
879 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
880 list_del(&vb->purge);
885 static void purge_fragmented_blocks_thiscpu(void)
887 purge_fragmented_blocks(smp_processor_id());
890 static void purge_fragmented_blocks_allcpus(void)
894 for_each_possible_cpu(cpu)
895 purge_fragmented_blocks(cpu);
898 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
900 struct vmap_block_queue *vbq;
901 struct vmap_block *vb;
902 unsigned long addr = 0;
906 BUG_ON(size & ~PAGE_MASK);
907 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
908 if (WARN_ON(size == 0)) {
910 * Allocating 0 bytes isn't what caller wants since
911 * get_order(0) returns funny result. Just warn and terminate
916 order = get_order(size);
920 vbq = &get_cpu_var(vmap_block_queue);
921 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
924 spin_lock(&vb->lock);
925 if (vb->free < 1UL << order)
928 i = bitmap_find_free_region(vb->alloc_map,
929 VMAP_BBMAP_BITS, order);
932 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
933 /* fragmented and no outstanding allocations */
934 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
939 addr = vb->va->va_start + (i << PAGE_SHIFT);
940 BUG_ON(addr_to_vb_idx(addr) !=
941 addr_to_vb_idx(vb->va->va_start));
942 vb->free -= 1UL << order;
944 spin_lock(&vbq->lock);
945 list_del_rcu(&vb->free_list);
946 spin_unlock(&vbq->lock);
948 spin_unlock(&vb->lock);
951 spin_unlock(&vb->lock);
955 purge_fragmented_blocks_thiscpu();
957 put_cpu_var(vmap_block_queue);
961 vb = new_vmap_block(gfp_mask);
970 static void vb_free(const void *addr, unsigned long size)
972 unsigned long offset;
973 unsigned long vb_idx;
975 struct vmap_block *vb;
977 BUG_ON(size & ~PAGE_MASK);
978 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
980 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
982 order = get_order(size);
984 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
986 vb_idx = addr_to_vb_idx((unsigned long)addr);
988 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
992 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
994 spin_lock(&vb->lock);
995 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
997 vb->dirty += 1UL << order;
998 if (vb->dirty == VMAP_BBMAP_BITS) {
1000 spin_unlock(&vb->lock);
1001 free_vmap_block(vb);
1003 spin_unlock(&vb->lock);
1007 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1009 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1010 * to amortize TLB flushing overheads. What this means is that any page you
1011 * have now, may, in a former life, have been mapped into kernel virtual
1012 * address by the vmap layer and so there might be some CPUs with TLB entries
1013 * still referencing that page (additional to the regular 1:1 kernel mapping).
1015 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1016 * be sure that none of the pages we have control over will have any aliases
1017 * from the vmap layer.
1019 void vm_unmap_aliases(void)
1021 unsigned long start = ULONG_MAX, end = 0;
1025 if (unlikely(!vmap_initialized))
1028 for_each_possible_cpu(cpu) {
1029 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1030 struct vmap_block *vb;
1033 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1036 spin_lock(&vb->lock);
1037 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1038 while (i < VMAP_BBMAP_BITS) {
1041 j = find_next_zero_bit(vb->dirty_map,
1042 VMAP_BBMAP_BITS, i);
1044 s = vb->va->va_start + (i << PAGE_SHIFT);
1045 e = vb->va->va_start + (j << PAGE_SHIFT);
1054 i = find_next_bit(vb->dirty_map,
1055 VMAP_BBMAP_BITS, i);
1057 spin_unlock(&vb->lock);
1062 __purge_vmap_area_lazy(&start, &end, 1, flush);
1064 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1067 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1068 * @mem: the pointer returned by vm_map_ram
1069 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1071 void vm_unmap_ram(const void *mem, unsigned int count)
1073 unsigned long size = count << PAGE_SHIFT;
1074 unsigned long addr = (unsigned long)mem;
1077 BUG_ON(addr < VMALLOC_START);
1078 BUG_ON(addr > VMALLOC_END);
1079 BUG_ON(addr & (PAGE_SIZE-1));
1081 debug_check_no_locks_freed(mem, size);
1082 vmap_debug_free_range(addr, addr+size);
1084 if (likely(count <= VMAP_MAX_ALLOC))
1087 free_unmap_vmap_area_addr(addr);
1089 EXPORT_SYMBOL(vm_unmap_ram);
1092 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1093 * @pages: an array of pointers to the pages to be mapped
1094 * @count: number of pages
1095 * @node: prefer to allocate data structures on this node
1096 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1098 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1100 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1102 unsigned long size = count << PAGE_SHIFT;
1106 if (likely(count <= VMAP_MAX_ALLOC)) {
1107 mem = vb_alloc(size, GFP_KERNEL);
1110 addr = (unsigned long)mem;
1112 struct vmap_area *va;
1113 va = alloc_vmap_area(size, PAGE_SIZE,
1114 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1118 addr = va->va_start;
1121 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1122 vm_unmap_ram(mem, count);
1127 EXPORT_SYMBOL(vm_map_ram);
1129 static struct vm_struct *vmlist __initdata;
1131 * vm_area_add_early - add vmap area early during boot
1132 * @vm: vm_struct to add
1134 * This function is used to add fixed kernel vm area to vmlist before
1135 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1136 * should contain proper values and the other fields should be zero.
1138 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1140 void __init vm_area_add_early(struct vm_struct *vm)
1142 struct vm_struct *tmp, **p;
1144 BUG_ON(vmap_initialized);
1145 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1146 if (tmp->addr >= vm->addr) {
1147 BUG_ON(tmp->addr < vm->addr + vm->size);
1150 BUG_ON(tmp->addr + tmp->size > vm->addr);
1157 * vm_area_register_early - register vmap area early during boot
1158 * @vm: vm_struct to register
1159 * @align: requested alignment
1161 * This function is used to register kernel vm area before
1162 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1163 * proper values on entry and other fields should be zero. On return,
1164 * vm->addr contains the allocated address.
1166 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1168 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1170 static size_t vm_init_off __initdata;
1173 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1174 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1176 vm->addr = (void *)addr;
1178 vm_area_add_early(vm);
1181 void __init vmalloc_init(void)
1183 struct vmap_area *va;
1184 struct vm_struct *tmp;
1187 for_each_possible_cpu(i) {
1188 struct vmap_block_queue *vbq;
1190 vbq = &per_cpu(vmap_block_queue, i);
1191 spin_lock_init(&vbq->lock);
1192 INIT_LIST_HEAD(&vbq->free);
1195 /* Import existing vmlist entries. */
1196 for (tmp = vmlist; tmp; tmp = tmp->next) {
1197 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1198 va->flags = VM_VM_AREA;
1199 va->va_start = (unsigned long)tmp->addr;
1200 va->va_end = va->va_start + tmp->size;
1202 __insert_vmap_area(va);
1205 vmap_area_pcpu_hole = VMALLOC_END;
1207 vmap_initialized = true;
1211 * map_kernel_range_noflush - map kernel VM area with the specified pages
1212 * @addr: start of the VM area to map
1213 * @size: size of the VM area to map
1214 * @prot: page protection flags to use
1215 * @pages: pages to map
1217 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1218 * specify should have been allocated using get_vm_area() and its
1222 * This function does NOT do any cache flushing. The caller is
1223 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1224 * before calling this function.
1227 * The number of pages mapped on success, -errno on failure.
1229 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1230 pgprot_t prot, struct page **pages)
1232 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1236 * unmap_kernel_range_noflush - unmap kernel VM area
1237 * @addr: start of the VM area to unmap
1238 * @size: size of the VM area to unmap
1240 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1241 * specify should have been allocated using get_vm_area() and its
1245 * This function does NOT do any cache flushing. The caller is
1246 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1247 * before calling this function and flush_tlb_kernel_range() after.
1249 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1251 vunmap_page_range(addr, addr + size);
1253 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1256 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1257 * @addr: start of the VM area to unmap
1258 * @size: size of the VM area to unmap
1260 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1261 * the unmapping and tlb after.
1263 void unmap_kernel_range(unsigned long addr, unsigned long size)
1265 unsigned long end = addr + size;
1267 flush_cache_vunmap(addr, end);
1268 vunmap_page_range(addr, end);
1269 flush_tlb_kernel_range(addr, end);
1272 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1274 unsigned long addr = (unsigned long)area->addr;
1275 unsigned long end = addr + area->size - PAGE_SIZE;
1278 err = vmap_page_range(addr, end, prot, *pages);
1286 EXPORT_SYMBOL_GPL(map_vm_area);
1288 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1289 unsigned long flags, const void *caller)
1291 spin_lock(&vmap_area_lock);
1293 vm->addr = (void *)va->va_start;
1294 vm->size = va->va_end - va->va_start;
1295 vm->caller = caller;
1297 va->flags |= VM_VM_AREA;
1298 spin_unlock(&vmap_area_lock);
1301 static void clear_vm_unlist(struct vm_struct *vm)
1304 * Before removing VM_UNLIST,
1305 * we should make sure that vm has proper values.
1306 * Pair with smp_rmb() in show_numa_info().
1309 vm->flags &= ~VM_UNLIST;
1312 static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1313 unsigned long flags, const void *caller)
1315 setup_vmalloc_vm(vm, va, flags, caller);
1316 clear_vm_unlist(vm);
1319 static struct vm_struct *__get_vm_area_node(unsigned long size,
1320 unsigned long align, unsigned long flags, unsigned long start,
1321 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1323 struct vmap_area *va;
1324 struct vm_struct *area;
1326 BUG_ON(in_interrupt());
1327 if (flags & VM_IOREMAP) {
1328 int bit = fls(size);
1330 if (bit > IOREMAP_MAX_ORDER)
1331 bit = IOREMAP_MAX_ORDER;
1332 else if (bit < PAGE_SHIFT)
1338 size = PAGE_ALIGN(size);
1339 if (unlikely(!size))
1342 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1343 if (unlikely(!area))
1347 * We always allocate a guard page.
1351 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1358 * When this function is called from __vmalloc_node_range,
1359 * we add VM_UNLIST flag to avoid accessing uninitialized
1360 * members of vm_struct such as pages and nr_pages fields.
1361 * They will be set later.
1363 if (flags & VM_UNLIST)
1364 setup_vmalloc_vm(area, va, flags, caller);
1366 insert_vmalloc_vm(area, va, flags, caller);
1371 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1372 unsigned long start, unsigned long end)
1374 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1375 GFP_KERNEL, __builtin_return_address(0));
1377 EXPORT_SYMBOL_GPL(__get_vm_area);
1379 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1380 unsigned long start, unsigned long end,
1383 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1384 GFP_KERNEL, caller);
1388 * get_vm_area - reserve a contiguous kernel virtual area
1389 * @size: size of the area
1390 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1392 * Search an area of @size in the kernel virtual mapping area,
1393 * and reserved it for out purposes. Returns the area descriptor
1394 * on success or %NULL on failure.
1396 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1398 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1399 NUMA_NO_NODE, GFP_KERNEL,
1400 __builtin_return_address(0));
1403 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1406 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1407 NUMA_NO_NODE, GFP_KERNEL, caller);
1411 * find_vm_area - find a continuous kernel virtual area
1412 * @addr: base address
1414 * Search for the kernel VM area starting at @addr, and return it.
1415 * It is up to the caller to do all required locking to keep the returned
1418 struct vm_struct *find_vm_area(const void *addr)
1420 struct vmap_area *va;
1422 va = find_vmap_area((unsigned long)addr);
1423 if (va && va->flags & VM_VM_AREA)
1430 * remove_vm_area - find and remove a continuous kernel virtual area
1431 * @addr: base address
1433 * Search for the kernel VM area starting at @addr, and remove it.
1434 * This function returns the found VM area, but using it is NOT safe
1435 * on SMP machines, except for its size or flags.
1437 struct vm_struct *remove_vm_area(const void *addr)
1439 struct vmap_area *va;
1441 va = find_vmap_area((unsigned long)addr);
1442 if (va && va->flags & VM_VM_AREA) {
1443 struct vm_struct *vm = va->vm;
1445 spin_lock(&vmap_area_lock);
1447 va->flags &= ~VM_VM_AREA;
1448 spin_unlock(&vmap_area_lock);
1450 vmap_debug_free_range(va->va_start, va->va_end);
1451 free_unmap_vmap_area(va);
1452 vm->size -= PAGE_SIZE;
1459 static void __vunmap(const void *addr, int deallocate_pages)
1461 struct vm_struct *area;
1466 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1467 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1471 area = remove_vm_area(addr);
1472 if (unlikely(!area)) {
1473 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1478 debug_check_no_locks_freed(addr, area->size);
1479 debug_check_no_obj_freed(addr, area->size);
1481 if (deallocate_pages) {
1484 for (i = 0; i < area->nr_pages; i++) {
1485 struct page *page = area->pages[i];
1491 if (area->flags & VM_VPAGES)
1502 * vfree - release memory allocated by vmalloc()
1503 * @addr: memory base address
1505 * Free the virtually continuous memory area starting at @addr, as
1506 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1507 * NULL, no operation is performed.
1509 * Must not be called in interrupt context.
1511 void vfree(const void *addr)
1513 BUG_ON(in_interrupt());
1515 kmemleak_free(addr);
1519 EXPORT_SYMBOL(vfree);
1522 * vunmap - release virtual mapping obtained by vmap()
1523 * @addr: memory base address
1525 * Free the virtually contiguous memory area starting at @addr,
1526 * which was created from the page array passed to vmap().
1528 * Must not be called in interrupt context.
1530 void vunmap(const void *addr)
1532 BUG_ON(in_interrupt());
1536 EXPORT_SYMBOL(vunmap);
1539 * vmap - map an array of pages into virtually contiguous space
1540 * @pages: array of page pointers
1541 * @count: number of pages to map
1542 * @flags: vm_area->flags
1543 * @prot: page protection for the mapping
1545 * Maps @count pages from @pages into contiguous kernel virtual
1548 void *vmap(struct page **pages, unsigned int count,
1549 unsigned long flags, pgprot_t prot)
1551 struct vm_struct *area;
1555 if (count > totalram_pages)
1558 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1559 __builtin_return_address(0));
1563 if (map_vm_area(area, prot, &pages)) {
1570 EXPORT_SYMBOL(vmap);
1572 static void *__vmalloc_node(unsigned long size, unsigned long align,
1573 gfp_t gfp_mask, pgprot_t prot,
1574 int node, const void *caller);
1575 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1576 pgprot_t prot, int node, const void *caller)
1578 const int order = 0;
1579 struct page **pages;
1580 unsigned int nr_pages, array_size, i;
1581 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1583 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1584 array_size = (nr_pages * sizeof(struct page *));
1586 area->nr_pages = nr_pages;
1587 /* Please note that the recursion is strictly bounded. */
1588 if (array_size > PAGE_SIZE) {
1589 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1590 PAGE_KERNEL, node, caller);
1591 area->flags |= VM_VPAGES;
1593 pages = kmalloc_node(array_size, nested_gfp, node);
1595 area->pages = pages;
1596 area->caller = caller;
1598 remove_vm_area(area->addr);
1603 for (i = 0; i < area->nr_pages; i++) {
1605 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
1608 page = alloc_page(tmp_mask);
1610 page = alloc_pages_node(node, tmp_mask, order);
1612 if (unlikely(!page)) {
1613 /* Successfully allocated i pages, free them in __vunmap() */
1617 area->pages[i] = page;
1620 if (map_vm_area(area, prot, &pages))
1625 warn_alloc_failed(gfp_mask, order,
1626 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1627 (area->nr_pages*PAGE_SIZE), area->size);
1633 * __vmalloc_node_range - allocate virtually contiguous memory
1634 * @size: allocation size
1635 * @align: desired alignment
1636 * @start: vm area range start
1637 * @end: vm area range end
1638 * @gfp_mask: flags for the page level allocator
1639 * @prot: protection mask for the allocated pages
1640 * @node: node to use for allocation or NUMA_NO_NODE
1641 * @caller: caller's return address
1643 * Allocate enough pages to cover @size from the page level
1644 * allocator with @gfp_mask flags. Map them into contiguous
1645 * kernel virtual space, using a pagetable protection of @prot.
1647 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1648 unsigned long start, unsigned long end, gfp_t gfp_mask,
1649 pgprot_t prot, int node, const void *caller)
1651 struct vm_struct *area;
1653 unsigned long real_size = size;
1655 size = PAGE_ALIGN(size);
1656 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1659 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1660 start, end, node, gfp_mask, caller);
1664 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1669 * In this function, newly allocated vm_struct has VM_UNLIST flag.
1670 * It means that vm_struct is not fully initialized.
1671 * Now, it is fully initialized, so remove this flag here.
1673 clear_vm_unlist(area);
1676 * A ref_count = 3 is needed because the vm_struct and vmap_area
1677 * structures allocated in the __get_vm_area_node() function contain
1678 * references to the virtual address of the vmalloc'ed block.
1680 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1685 warn_alloc_failed(gfp_mask, 0,
1686 "vmalloc: allocation failure: %lu bytes\n",
1692 * __vmalloc_node - allocate virtually contiguous memory
1693 * @size: allocation size
1694 * @align: desired alignment
1695 * @gfp_mask: flags for the page level allocator
1696 * @prot: protection mask for the allocated pages
1697 * @node: node to use for allocation or NUMA_NO_NODE
1698 * @caller: caller's return address
1700 * Allocate enough pages to cover @size from the page level
1701 * allocator with @gfp_mask flags. Map them into contiguous
1702 * kernel virtual space, using a pagetable protection of @prot.
1704 static void *__vmalloc_node(unsigned long size, unsigned long align,
1705 gfp_t gfp_mask, pgprot_t prot,
1706 int node, const void *caller)
1708 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1709 gfp_mask, prot, node, caller);
1712 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1714 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1715 __builtin_return_address(0));
1717 EXPORT_SYMBOL(__vmalloc);
1719 static inline void *__vmalloc_node_flags(unsigned long size,
1720 int node, gfp_t flags)
1722 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1723 node, __builtin_return_address(0));
1727 * vmalloc - allocate virtually contiguous memory
1728 * @size: allocation size
1729 * Allocate enough pages to cover @size from the page level
1730 * allocator and map them into contiguous kernel virtual space.
1732 * For tight control over page level allocator and protection flags
1733 * use __vmalloc() instead.
1735 void *vmalloc(unsigned long size)
1737 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1738 GFP_KERNEL | __GFP_HIGHMEM);
1740 EXPORT_SYMBOL(vmalloc);
1743 * vzalloc - allocate virtually contiguous memory with zero fill
1744 * @size: allocation size
1745 * Allocate enough pages to cover @size from the page level
1746 * allocator and map them into contiguous kernel virtual space.
1747 * The memory allocated is set to zero.
1749 * For tight control over page level allocator and protection flags
1750 * use __vmalloc() instead.
1752 void *vzalloc(unsigned long size)
1754 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1755 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1757 EXPORT_SYMBOL(vzalloc);
1760 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1761 * @size: allocation size
1763 * The resulting memory area is zeroed so it can be mapped to userspace
1764 * without leaking data.
1766 void *vmalloc_user(unsigned long size)
1768 struct vm_struct *area;
1771 ret = __vmalloc_node(size, SHMLBA,
1772 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1773 PAGE_KERNEL, NUMA_NO_NODE,
1774 __builtin_return_address(0));
1776 area = find_vm_area(ret);
1777 area->flags |= VM_USERMAP;
1781 EXPORT_SYMBOL(vmalloc_user);
1784 * vmalloc_node - allocate memory on a specific node
1785 * @size: allocation size
1788 * Allocate enough pages to cover @size from the page level
1789 * allocator and map them into contiguous kernel virtual space.
1791 * For tight control over page level allocator and protection flags
1792 * use __vmalloc() instead.
1794 void *vmalloc_node(unsigned long size, int node)
1796 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1797 node, __builtin_return_address(0));
1799 EXPORT_SYMBOL(vmalloc_node);
1802 * vzalloc_node - allocate memory on a specific node with zero fill
1803 * @size: allocation size
1806 * Allocate enough pages to cover @size from the page level
1807 * allocator and map them into contiguous kernel virtual space.
1808 * The memory allocated is set to zero.
1810 * For tight control over page level allocator and protection flags
1811 * use __vmalloc_node() instead.
1813 void *vzalloc_node(unsigned long size, int node)
1815 return __vmalloc_node_flags(size, node,
1816 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1818 EXPORT_SYMBOL(vzalloc_node);
1820 #ifndef PAGE_KERNEL_EXEC
1821 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1825 * vmalloc_exec - allocate virtually contiguous, executable memory
1826 * @size: allocation size
1828 * Kernel-internal function to allocate enough pages to cover @size
1829 * the page level allocator and map them into contiguous and
1830 * executable kernel virtual space.
1832 * For tight control over page level allocator and protection flags
1833 * use __vmalloc() instead.
1836 void *vmalloc_exec(unsigned long size)
1838 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1839 NUMA_NO_NODE, __builtin_return_address(0));
1842 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1843 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1844 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1845 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1847 #define GFP_VMALLOC32 GFP_KERNEL
1851 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1852 * @size: allocation size
1854 * Allocate enough 32bit PA addressable pages to cover @size from the
1855 * page level allocator and map them into contiguous kernel virtual space.
1857 void *vmalloc_32(unsigned long size)
1859 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1860 NUMA_NO_NODE, __builtin_return_address(0));
1862 EXPORT_SYMBOL(vmalloc_32);
1865 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1866 * @size: allocation size
1868 * The resulting memory area is 32bit addressable and zeroed so it can be
1869 * mapped to userspace without leaking data.
1871 void *vmalloc_32_user(unsigned long size)
1873 struct vm_struct *area;
1876 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1877 NUMA_NO_NODE, __builtin_return_address(0));
1879 area = find_vm_area(ret);
1880 area->flags |= VM_USERMAP;
1884 EXPORT_SYMBOL(vmalloc_32_user);
1887 * small helper routine , copy contents to buf from addr.
1888 * If the page is not present, fill zero.
1891 static int aligned_vread(char *buf, char *addr, unsigned long count)
1897 unsigned long offset, length;
1899 offset = (unsigned long)addr & ~PAGE_MASK;
1900 length = PAGE_SIZE - offset;
1903 p = vmalloc_to_page(addr);
1905 * To do safe access to this _mapped_ area, we need
1906 * lock. But adding lock here means that we need to add
1907 * overhead of vmalloc()/vfree() calles for this _debug_
1908 * interface, rarely used. Instead of that, we'll use
1909 * kmap() and get small overhead in this access function.
1913 * we can expect USER0 is not used (see vread/vwrite's
1914 * function description)
1916 void *map = kmap_atomic(p);
1917 memcpy(buf, map + offset, length);
1920 memset(buf, 0, length);
1930 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1936 unsigned long offset, length;
1938 offset = (unsigned long)addr & ~PAGE_MASK;
1939 length = PAGE_SIZE - offset;
1942 p = vmalloc_to_page(addr);
1944 * To do safe access to this _mapped_ area, we need
1945 * lock. But adding lock here means that we need to add
1946 * overhead of vmalloc()/vfree() calles for this _debug_
1947 * interface, rarely used. Instead of that, we'll use
1948 * kmap() and get small overhead in this access function.
1952 * we can expect USER0 is not used (see vread/vwrite's
1953 * function description)
1955 void *map = kmap_atomic(p);
1956 memcpy(map + offset, buf, length);
1968 * vread() - read vmalloc area in a safe way.
1969 * @buf: buffer for reading data
1970 * @addr: vm address.
1971 * @count: number of bytes to be read.
1973 * Returns # of bytes which addr and buf should be increased.
1974 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1975 * includes any intersect with alive vmalloc area.
1977 * This function checks that addr is a valid vmalloc'ed area, and
1978 * copy data from that area to a given buffer. If the given memory range
1979 * of [addr...addr+count) includes some valid address, data is copied to
1980 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1981 * IOREMAP area is treated as memory hole and no copy is done.
1983 * If [addr...addr+count) doesn't includes any intersects with alive
1984 * vm_struct area, returns 0. @buf should be kernel's buffer.
1986 * Note: In usual ops, vread() is never necessary because the caller
1987 * should know vmalloc() area is valid and can use memcpy().
1988 * This is for routines which have to access vmalloc area without
1989 * any informaion, as /dev/kmem.
1993 long vread(char *buf, char *addr, unsigned long count)
1995 struct vmap_area *va;
1996 struct vm_struct *vm;
1997 char *vaddr, *buf_start = buf;
1998 unsigned long buflen = count;
2001 /* Don't allow overflow */
2002 if ((unsigned long) addr + count < count)
2003 count = -(unsigned long) addr;
2005 spin_lock(&vmap_area_lock);
2006 list_for_each_entry(va, &vmap_area_list, list) {
2010 if (!(va->flags & VM_VM_AREA))
2014 vaddr = (char *) vm->addr;
2015 if (addr >= vaddr + vm->size - PAGE_SIZE)
2017 while (addr < vaddr) {
2025 n = vaddr + vm->size - PAGE_SIZE - addr;
2028 if (!(vm->flags & VM_IOREMAP))
2029 aligned_vread(buf, addr, n);
2030 else /* IOREMAP area is treated as memory hole */
2037 spin_unlock(&vmap_area_lock);
2039 if (buf == buf_start)
2041 /* zero-fill memory holes */
2042 if (buf != buf_start + buflen)
2043 memset(buf, 0, buflen - (buf - buf_start));
2049 * vwrite() - write vmalloc area in a safe way.
2050 * @buf: buffer for source data
2051 * @addr: vm address.
2052 * @count: number of bytes to be read.
2054 * Returns # of bytes which addr and buf should be incresed.
2055 * (same number to @count).
2056 * If [addr...addr+count) doesn't includes any intersect with valid
2057 * vmalloc area, returns 0.
2059 * This function checks that addr is a valid vmalloc'ed area, and
2060 * copy data from a buffer to the given addr. If specified range of
2061 * [addr...addr+count) includes some valid address, data is copied from
2062 * proper area of @buf. If there are memory holes, no copy to hole.
2063 * IOREMAP area is treated as memory hole and no copy is done.
2065 * If [addr...addr+count) doesn't includes any intersects with alive
2066 * vm_struct area, returns 0. @buf should be kernel's buffer.
2068 * Note: In usual ops, vwrite() is never necessary because the caller
2069 * should know vmalloc() area is valid and can use memcpy().
2070 * This is for routines which have to access vmalloc area without
2071 * any informaion, as /dev/kmem.
2074 long vwrite(char *buf, char *addr, unsigned long count)
2076 struct vmap_area *va;
2077 struct vm_struct *vm;
2079 unsigned long n, buflen;
2082 /* Don't allow overflow */
2083 if ((unsigned long) addr + count < count)
2084 count = -(unsigned long) addr;
2087 spin_lock(&vmap_area_lock);
2088 list_for_each_entry(va, &vmap_area_list, list) {
2092 if (!(va->flags & VM_VM_AREA))
2096 vaddr = (char *) vm->addr;
2097 if (addr >= vaddr + vm->size - PAGE_SIZE)
2099 while (addr < vaddr) {
2106 n = vaddr + vm->size - PAGE_SIZE - addr;
2109 if (!(vm->flags & VM_IOREMAP)) {
2110 aligned_vwrite(buf, addr, n);
2118 spin_unlock(&vmap_area_lock);
2125 * remap_vmalloc_range - map vmalloc pages to userspace
2126 * @vma: vma to cover (map full range of vma)
2127 * @addr: vmalloc memory
2128 * @pgoff: number of pages into addr before first page to map
2130 * Returns: 0 for success, -Exxx on failure
2132 * This function checks that addr is a valid vmalloc'ed area, and
2133 * that it is big enough to cover the vma. Will return failure if
2134 * that criteria isn't met.
2136 * Similar to remap_pfn_range() (see mm/memory.c)
2138 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2139 unsigned long pgoff)
2141 struct vm_struct *area;
2142 unsigned long uaddr = vma->vm_start;
2143 unsigned long usize = vma->vm_end - vma->vm_start;
2145 if ((PAGE_SIZE-1) & (unsigned long)addr)
2148 area = find_vm_area(addr);
2152 if (!(area->flags & VM_USERMAP))
2155 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
2158 addr += pgoff << PAGE_SHIFT;
2160 struct page *page = vmalloc_to_page(addr);
2163 ret = vm_insert_page(vma, uaddr, page);
2170 } while (usize > 0);
2172 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2176 EXPORT_SYMBOL(remap_vmalloc_range);
2179 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2182 void __attribute__((weak)) vmalloc_sync_all(void)
2187 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2199 * alloc_vm_area - allocate a range of kernel address space
2200 * @size: size of the area
2201 * @ptes: returns the PTEs for the address space
2203 * Returns: NULL on failure, vm_struct on success
2205 * This function reserves a range of kernel address space, and
2206 * allocates pagetables to map that range. No actual mappings
2209 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2210 * allocated for the VM area are returned.
2212 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2214 struct vm_struct *area;
2216 area = get_vm_area_caller(size, VM_IOREMAP,
2217 __builtin_return_address(0));
2222 * This ensures that page tables are constructed for this region
2223 * of kernel virtual address space and mapped into init_mm.
2225 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2226 size, f, ptes ? &ptes : NULL)) {
2233 EXPORT_SYMBOL_GPL(alloc_vm_area);
2235 void free_vm_area(struct vm_struct *area)
2237 struct vm_struct *ret;
2238 ret = remove_vm_area(area->addr);
2239 BUG_ON(ret != area);
2242 EXPORT_SYMBOL_GPL(free_vm_area);
2245 static struct vmap_area *node_to_va(struct rb_node *n)
2247 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2251 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2252 * @end: target address
2253 * @pnext: out arg for the next vmap_area
2254 * @pprev: out arg for the previous vmap_area
2256 * Returns: %true if either or both of next and prev are found,
2257 * %false if no vmap_area exists
2259 * Find vmap_areas end addresses of which enclose @end. ie. if not
2260 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2262 static bool pvm_find_next_prev(unsigned long end,
2263 struct vmap_area **pnext,
2264 struct vmap_area **pprev)
2266 struct rb_node *n = vmap_area_root.rb_node;
2267 struct vmap_area *va = NULL;
2270 va = rb_entry(n, struct vmap_area, rb_node);
2271 if (end < va->va_end)
2273 else if (end > va->va_end)
2282 if (va->va_end > end) {
2284 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2287 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2293 * pvm_determine_end - find the highest aligned address between two vmap_areas
2294 * @pnext: in/out arg for the next vmap_area
2295 * @pprev: in/out arg for the previous vmap_area
2298 * Returns: determined end address
2300 * Find the highest aligned address between *@pnext and *@pprev below
2301 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2302 * down address is between the end addresses of the two vmap_areas.
2304 * Please note that the address returned by this function may fall
2305 * inside *@pnext vmap_area. The caller is responsible for checking
2308 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2309 struct vmap_area **pprev,
2310 unsigned long align)
2312 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2316 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2320 while (*pprev && (*pprev)->va_end > addr) {
2322 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2329 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2330 * @offsets: array containing offset of each area
2331 * @sizes: array containing size of each area
2332 * @nr_vms: the number of areas to allocate
2333 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2335 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2336 * vm_structs on success, %NULL on failure
2338 * Percpu allocator wants to use congruent vm areas so that it can
2339 * maintain the offsets among percpu areas. This function allocates
2340 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2341 * be scattered pretty far, distance between two areas easily going up
2342 * to gigabytes. To avoid interacting with regular vmallocs, these
2343 * areas are allocated from top.
2345 * Despite its complicated look, this allocator is rather simple. It
2346 * does everything top-down and scans areas from the end looking for
2347 * matching slot. While scanning, if any of the areas overlaps with
2348 * existing vmap_area, the base address is pulled down to fit the
2349 * area. Scanning is repeated till all the areas fit and then all
2350 * necessary data structres are inserted and the result is returned.
2352 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2353 const size_t *sizes, int nr_vms,
2356 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2357 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2358 struct vmap_area **vas, *prev, *next;
2359 struct vm_struct **vms;
2360 int area, area2, last_area, term_area;
2361 unsigned long base, start, end, last_end;
2362 bool purged = false;
2364 /* verify parameters and allocate data structures */
2365 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2366 for (last_area = 0, area = 0; area < nr_vms; area++) {
2367 start = offsets[area];
2368 end = start + sizes[area];
2370 /* is everything aligned properly? */
2371 BUG_ON(!IS_ALIGNED(offsets[area], align));
2372 BUG_ON(!IS_ALIGNED(sizes[area], align));
2374 /* detect the area with the highest address */
2375 if (start > offsets[last_area])
2378 for (area2 = 0; area2 < nr_vms; area2++) {
2379 unsigned long start2 = offsets[area2];
2380 unsigned long end2 = start2 + sizes[area2];
2385 BUG_ON(start2 >= start && start2 < end);
2386 BUG_ON(end2 <= end && end2 > start);
2389 last_end = offsets[last_area] + sizes[last_area];
2391 if (vmalloc_end - vmalloc_start < last_end) {
2396 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2397 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2401 for (area = 0; area < nr_vms; area++) {
2402 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2403 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2404 if (!vas[area] || !vms[area])
2408 spin_lock(&vmap_area_lock);
2410 /* start scanning - we scan from the top, begin with the last area */
2411 area = term_area = last_area;
2412 start = offsets[area];
2413 end = start + sizes[area];
2415 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2416 base = vmalloc_end - last_end;
2419 base = pvm_determine_end(&next, &prev, align) - end;
2422 BUG_ON(next && next->va_end <= base + end);
2423 BUG_ON(prev && prev->va_end > base + end);
2426 * base might have underflowed, add last_end before
2429 if (base + last_end < vmalloc_start + last_end) {
2430 spin_unlock(&vmap_area_lock);
2432 purge_vmap_area_lazy();
2440 * If next overlaps, move base downwards so that it's
2441 * right below next and then recheck.
2443 if (next && next->va_start < base + end) {
2444 base = pvm_determine_end(&next, &prev, align) - end;
2450 * If prev overlaps, shift down next and prev and move
2451 * base so that it's right below new next and then
2454 if (prev && prev->va_end > base + start) {
2456 prev = node_to_va(rb_prev(&next->rb_node));
2457 base = pvm_determine_end(&next, &prev, align) - end;
2463 * This area fits, move on to the previous one. If
2464 * the previous one is the terminal one, we're done.
2466 area = (area + nr_vms - 1) % nr_vms;
2467 if (area == term_area)
2469 start = offsets[area];
2470 end = start + sizes[area];
2471 pvm_find_next_prev(base + end, &next, &prev);
2474 /* we've found a fitting base, insert all va's */
2475 for (area = 0; area < nr_vms; area++) {
2476 struct vmap_area *va = vas[area];
2478 va->va_start = base + offsets[area];
2479 va->va_end = va->va_start + sizes[area];
2480 __insert_vmap_area(va);
2483 vmap_area_pcpu_hole = base + offsets[last_area];
2485 spin_unlock(&vmap_area_lock);
2487 /* insert all vm's */
2488 for (area = 0; area < nr_vms; area++)
2489 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2496 for (area = 0; area < nr_vms; area++) {
2507 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2508 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2509 * @nr_vms: the number of allocated areas
2511 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2513 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2517 for (i = 0; i < nr_vms; i++)
2518 free_vm_area(vms[i]);
2521 #endif /* CONFIG_SMP */
2523 #ifdef CONFIG_PROC_FS
2524 static void *s_start(struct seq_file *m, loff_t *pos)
2525 __acquires(&vmap_area_lock)
2528 struct vmap_area *va;
2530 spin_lock(&vmap_area_lock);
2531 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2532 while (n > 0 && &va->list != &vmap_area_list) {
2534 va = list_entry(va->list.next, typeof(*va), list);
2536 if (!n && &va->list != &vmap_area_list)
2543 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2545 struct vmap_area *va = p, *next;
2548 next = list_entry(va->list.next, typeof(*va), list);
2549 if (&next->list != &vmap_area_list)
2555 static void s_stop(struct seq_file *m, void *p)
2556 __releases(&vmap_area_lock)
2558 spin_unlock(&vmap_area_lock);
2561 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2563 if (IS_ENABLED(CONFIG_NUMA)) {
2564 unsigned int nr, *counters = m->private;
2569 /* Pair with smp_wmb() in clear_vm_unlist() */
2571 if (v->flags & VM_UNLIST)
2574 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2576 for (nr = 0; nr < v->nr_pages; nr++)
2577 counters[page_to_nid(v->pages[nr])]++;
2579 for_each_node_state(nr, N_HIGH_MEMORY)
2581 seq_printf(m, " N%u=%u", nr, counters[nr]);
2585 static int s_show(struct seq_file *m, void *p)
2587 struct vmap_area *va = p;
2588 struct vm_struct *v;
2590 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2593 if (!(va->flags & VM_VM_AREA)) {
2594 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2595 (void *)va->va_start, (void *)va->va_end,
2596 va->va_end - va->va_start);
2602 seq_printf(m, "0x%pK-0x%pK %7ld",
2603 v->addr, v->addr + v->size, v->size);
2606 seq_printf(m, " %pS", v->caller);
2609 seq_printf(m, " pages=%d", v->nr_pages);
2612 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2614 if (v->flags & VM_IOREMAP)
2615 seq_printf(m, " ioremap");
2617 if (v->flags & VM_ALLOC)
2618 seq_printf(m, " vmalloc");
2620 if (v->flags & VM_MAP)
2621 seq_printf(m, " vmap");
2623 if (v->flags & VM_USERMAP)
2624 seq_printf(m, " user");
2626 if (v->flags & VM_VPAGES)
2627 seq_printf(m, " vpages");
2629 show_numa_info(m, v);
2634 static const struct seq_operations vmalloc_op = {
2641 static int vmalloc_open(struct inode *inode, struct file *file)
2643 unsigned int *ptr = NULL;
2646 if (IS_ENABLED(CONFIG_NUMA)) {
2647 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2651 ret = seq_open(file, &vmalloc_op);
2653 struct seq_file *m = file->private_data;
2660 static const struct file_operations proc_vmalloc_operations = {
2661 .open = vmalloc_open,
2663 .llseek = seq_lseek,
2664 .release = seq_release_private,
2667 static int __init proc_vmalloc_init(void)
2669 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2672 module_init(proc_vmalloc_init);
2674 void get_vmalloc_info(struct vmalloc_info *vmi)
2676 struct vmap_area *va;
2677 unsigned long free_area_size;
2678 unsigned long prev_end;
2681 vmi->largest_chunk = 0;
2683 prev_end = VMALLOC_START;
2685 spin_lock(&vmap_area_lock);
2687 if (list_empty(&vmap_area_list)) {
2688 vmi->largest_chunk = VMALLOC_TOTAL;
2692 list_for_each_entry(va, &vmap_area_list, list) {
2693 unsigned long addr = va->va_start;
2696 * Some archs keep another range for modules in vmalloc space
2698 if (addr < VMALLOC_START)
2700 if (addr >= VMALLOC_END)
2703 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2706 vmi->used += (va->va_end - va->va_start);
2708 free_area_size = addr - prev_end;
2709 if (vmi->largest_chunk < free_area_size)
2710 vmi->largest_chunk = free_area_size;
2712 prev_end = va->va_end;
2715 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2716 vmi->largest_chunk = VMALLOC_END - prev_end;
2719 spin_unlock(&vmap_area_lock);