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 <linux/llist.h>
31 #include <asm/uaccess.h>
32 #include <asm/tlbflush.h>
33 #include <asm/shmparam.h>
35 struct vfree_deferred {
36 struct llist_head list;
37 struct work_struct wq;
39 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
41 static void __vunmap(const void *, int);
43 static void free_work(struct work_struct *w)
45 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
46 struct llist_node *llnode = llist_del_all(&p->list);
49 llnode = llist_next(llnode);
54 /*** Page table manipulation functions ***/
56 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
60 pte = pte_offset_kernel(pmd, addr);
62 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
63 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
64 } while (pte++, addr += PAGE_SIZE, addr != end);
67 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
72 pmd = pmd_offset(pud, addr);
74 next = pmd_addr_end(addr, end);
75 if (pmd_none_or_clear_bad(pmd))
77 vunmap_pte_range(pmd, addr, next);
78 } while (pmd++, addr = next, addr != end);
81 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
86 pud = pud_offset(pgd, addr);
88 next = pud_addr_end(addr, end);
89 if (pud_none_or_clear_bad(pud))
91 vunmap_pmd_range(pud, addr, next);
92 } while (pud++, addr = next, addr != end);
95 static void vunmap_page_range(unsigned long addr, unsigned long end)
101 pgd = pgd_offset_k(addr);
103 next = pgd_addr_end(addr, end);
104 if (pgd_none_or_clear_bad(pgd))
106 vunmap_pud_range(pgd, addr, next);
107 } while (pgd++, addr = next, addr != end);
110 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
111 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
116 * nr is a running index into the array which helps higher level
117 * callers keep track of where we're up to.
120 pte = pte_alloc_kernel(pmd, addr);
124 struct page *page = pages[*nr];
126 if (WARN_ON(!pte_none(*pte)))
130 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
132 } while (pte++, addr += PAGE_SIZE, addr != end);
136 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
137 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
142 pmd = pmd_alloc(&init_mm, pud, addr);
146 next = pmd_addr_end(addr, end);
147 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
149 } while (pmd++, addr = next, addr != end);
153 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
154 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
159 pud = pud_alloc(&init_mm, pgd, addr);
163 next = pud_addr_end(addr, end);
164 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
166 } while (pud++, addr = next, addr != end);
171 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
172 * will have pfns corresponding to the "pages" array.
174 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
176 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
177 pgprot_t prot, struct page **pages)
181 unsigned long addr = start;
186 pgd = pgd_offset_k(addr);
188 next = pgd_addr_end(addr, end);
189 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
192 } while (pgd++, addr = next, addr != end);
197 static int vmap_page_range(unsigned long start, unsigned long end,
198 pgprot_t prot, struct page **pages)
202 ret = vmap_page_range_noflush(start, end, prot, pages);
203 flush_cache_vmap(start, end);
207 int is_vmalloc_or_module_addr(const void *x)
210 * ARM, x86-64 and sparc64 put modules in a special place,
211 * and fall back on vmalloc() if that fails. Others
212 * just put it in the vmalloc space.
214 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
215 unsigned long addr = (unsigned long)x;
216 if (addr >= MODULES_VADDR && addr < MODULES_END)
219 return is_vmalloc_addr(x);
223 * Walk a vmap address to the struct page it maps.
225 struct page *vmalloc_to_page(const void *vmalloc_addr)
227 unsigned long addr = (unsigned long) vmalloc_addr;
228 struct page *page = NULL;
229 pgd_t *pgd = pgd_offset_k(addr);
232 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
233 * architectures that do not vmalloc module space
235 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
237 if (!pgd_none(*pgd)) {
238 pud_t *pud = pud_offset(pgd, addr);
239 if (!pud_none(*pud)) {
240 pmd_t *pmd = pmd_offset(pud, addr);
241 if (!pmd_none(*pmd)) {
244 ptep = pte_offset_map(pmd, addr);
246 if (pte_present(pte))
247 page = pte_page(pte);
254 EXPORT_SYMBOL(vmalloc_to_page);
257 * Map a vmalloc()-space virtual address to the physical page frame number.
259 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
261 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
263 EXPORT_SYMBOL(vmalloc_to_pfn);
266 /*** Global kva allocator ***/
268 #define VM_LAZY_FREE 0x01
269 #define VM_LAZY_FREEING 0x02
270 #define VM_VM_AREA 0x04
272 static DEFINE_SPINLOCK(vmap_area_lock);
273 /* Export for kexec only */
274 LIST_HEAD(vmap_area_list);
275 static struct rb_root vmap_area_root = RB_ROOT;
277 /* The vmap cache globals are protected by vmap_area_lock */
278 static struct rb_node *free_vmap_cache;
279 static unsigned long cached_hole_size;
280 static unsigned long cached_vstart;
281 static unsigned long cached_align;
283 static unsigned long vmap_area_pcpu_hole;
285 static struct vmap_area *__find_vmap_area(unsigned long addr)
287 struct rb_node *n = vmap_area_root.rb_node;
290 struct vmap_area *va;
292 va = rb_entry(n, struct vmap_area, rb_node);
293 if (addr < va->va_start)
295 else if (addr >= va->va_end)
304 static void __insert_vmap_area(struct vmap_area *va)
306 struct rb_node **p = &vmap_area_root.rb_node;
307 struct rb_node *parent = NULL;
311 struct vmap_area *tmp_va;
314 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
315 if (va->va_start < tmp_va->va_end)
317 else if (va->va_end > tmp_va->va_start)
323 rb_link_node(&va->rb_node, parent, p);
324 rb_insert_color(&va->rb_node, &vmap_area_root);
326 /* address-sort this list */
327 tmp = rb_prev(&va->rb_node);
329 struct vmap_area *prev;
330 prev = rb_entry(tmp, struct vmap_area, rb_node);
331 list_add_rcu(&va->list, &prev->list);
333 list_add_rcu(&va->list, &vmap_area_list);
336 static void purge_vmap_area_lazy(void);
339 * Allocate a region of KVA of the specified size and alignment, within the
342 static struct vmap_area *alloc_vmap_area(unsigned long size,
344 unsigned long vstart, unsigned long vend,
345 int node, gfp_t gfp_mask)
347 struct vmap_area *va;
351 struct vmap_area *first;
354 BUG_ON(size & ~PAGE_MASK);
355 BUG_ON(!is_power_of_2(align));
357 va = kmalloc_node(sizeof(struct vmap_area),
358 gfp_mask & GFP_RECLAIM_MASK, node);
360 return ERR_PTR(-ENOMEM);
363 spin_lock(&vmap_area_lock);
365 * Invalidate cache if we have more permissive parameters.
366 * cached_hole_size notes the largest hole noticed _below_
367 * the vmap_area cached in free_vmap_cache: if size fits
368 * into that hole, we want to scan from vstart to reuse
369 * the hole instead of allocating above free_vmap_cache.
370 * Note that __free_vmap_area may update free_vmap_cache
371 * without updating cached_hole_size or cached_align.
373 if (!free_vmap_cache ||
374 size < cached_hole_size ||
375 vstart < cached_vstart ||
376 align < cached_align) {
378 cached_hole_size = 0;
379 free_vmap_cache = NULL;
381 /* record if we encounter less permissive parameters */
382 cached_vstart = vstart;
383 cached_align = align;
385 /* find starting point for our search */
386 if (free_vmap_cache) {
387 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
388 addr = ALIGN(first->va_end, align);
391 if (addr + size - 1 < addr)
395 addr = ALIGN(vstart, align);
396 if (addr + size - 1 < addr)
399 n = vmap_area_root.rb_node;
403 struct vmap_area *tmp;
404 tmp = rb_entry(n, struct vmap_area, rb_node);
405 if (tmp->va_end >= addr) {
407 if (tmp->va_start <= addr)
418 /* from the starting point, walk areas until a suitable hole is found */
419 while (addr + size > first->va_start && addr + size <= vend) {
420 if (addr + cached_hole_size < first->va_start)
421 cached_hole_size = first->va_start - addr;
422 addr = ALIGN(first->va_end, align);
423 if (addr + size - 1 < addr)
426 if (list_is_last(&first->list, &vmap_area_list))
429 first = list_entry(first->list.next,
430 struct vmap_area, list);
434 if (addr + size > vend)
438 va->va_end = addr + size;
440 __insert_vmap_area(va);
441 free_vmap_cache = &va->rb_node;
442 spin_unlock(&vmap_area_lock);
444 BUG_ON(va->va_start & (align-1));
445 BUG_ON(va->va_start < vstart);
446 BUG_ON(va->va_end > vend);
451 spin_unlock(&vmap_area_lock);
453 purge_vmap_area_lazy();
457 if (printk_ratelimit())
459 "vmap allocation for size %lu failed: "
460 "use vmalloc=<size> to increase size.\n", size);
462 return ERR_PTR(-EBUSY);
465 static void __free_vmap_area(struct vmap_area *va)
467 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
469 if (free_vmap_cache) {
470 if (va->va_end < cached_vstart) {
471 free_vmap_cache = NULL;
473 struct vmap_area *cache;
474 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
475 if (va->va_start <= cache->va_start) {
476 free_vmap_cache = rb_prev(&va->rb_node);
478 * We don't try to update cached_hole_size or
479 * cached_align, but it won't go very wrong.
484 rb_erase(&va->rb_node, &vmap_area_root);
485 RB_CLEAR_NODE(&va->rb_node);
486 list_del_rcu(&va->list);
489 * Track the highest possible candidate for pcpu area
490 * allocation. Areas outside of vmalloc area can be returned
491 * here too, consider only end addresses which fall inside
492 * vmalloc area proper.
494 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
495 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
497 kfree_rcu(va, rcu_head);
501 * Free a region of KVA allocated by alloc_vmap_area
503 static void free_vmap_area(struct vmap_area *va)
505 spin_lock(&vmap_area_lock);
506 __free_vmap_area(va);
507 spin_unlock(&vmap_area_lock);
511 * Clear the pagetable entries of a given vmap_area
513 static void unmap_vmap_area(struct vmap_area *va)
515 vunmap_page_range(va->va_start, va->va_end);
518 static void vmap_debug_free_range(unsigned long start, unsigned long end)
521 * Unmap page tables and force a TLB flush immediately if
522 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
523 * bugs similarly to those in linear kernel virtual address
524 * space after a page has been freed.
526 * All the lazy freeing logic is still retained, in order to
527 * minimise intrusiveness of this debugging feature.
529 * This is going to be *slow* (linear kernel virtual address
530 * debugging doesn't do a broadcast TLB flush so it is a lot
533 #ifdef CONFIG_DEBUG_PAGEALLOC
534 vunmap_page_range(start, end);
535 flush_tlb_kernel_range(start, end);
540 * lazy_max_pages is the maximum amount of virtual address space we gather up
541 * before attempting to purge with a TLB flush.
543 * There is a tradeoff here: a larger number will cover more kernel page tables
544 * and take slightly longer to purge, but it will linearly reduce the number of
545 * global TLB flushes that must be performed. It would seem natural to scale
546 * this number up linearly with the number of CPUs (because vmapping activity
547 * could also scale linearly with the number of CPUs), however it is likely
548 * that in practice, workloads might be constrained in other ways that mean
549 * vmap activity will not scale linearly with CPUs. Also, I want to be
550 * conservative and not introduce a big latency on huge systems, so go with
551 * a less aggressive log scale. It will still be an improvement over the old
552 * code, and it will be simple to change the scale factor if we find that it
553 * becomes a problem on bigger systems.
555 static unsigned long lazy_max_pages(void)
559 log = fls(num_online_cpus());
561 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
564 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
566 /* for per-CPU blocks */
567 static void purge_fragmented_blocks_allcpus(void);
570 * called before a call to iounmap() if the caller wants vm_area_struct's
573 void set_iounmap_nonlazy(void)
575 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
579 * Purges all lazily-freed vmap areas.
581 * If sync is 0 then don't purge if there is already a purge in progress.
582 * If force_flush is 1, then flush kernel TLBs between *start and *end even
583 * if we found no lazy vmap areas to unmap (callers can use this to optimise
584 * their own TLB flushing).
585 * Returns with *start = min(*start, lowest purged address)
586 * *end = max(*end, highest purged address)
588 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
589 int sync, int force_flush)
591 static DEFINE_SPINLOCK(purge_lock);
593 struct vmap_area *va;
594 struct vmap_area *n_va;
598 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
599 * should not expect such behaviour. This just simplifies locking for
600 * the case that isn't actually used at the moment anyway.
602 if (!sync && !force_flush) {
603 if (!spin_trylock(&purge_lock))
606 spin_lock(&purge_lock);
609 purge_fragmented_blocks_allcpus();
612 list_for_each_entry_rcu(va, &vmap_area_list, list) {
613 if (va->flags & VM_LAZY_FREE) {
614 if (va->va_start < *start)
615 *start = va->va_start;
616 if (va->va_end > *end)
618 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
619 list_add_tail(&va->purge_list, &valist);
620 va->flags |= VM_LAZY_FREEING;
621 va->flags &= ~VM_LAZY_FREE;
627 atomic_sub(nr, &vmap_lazy_nr);
629 if (nr || force_flush)
630 flush_tlb_kernel_range(*start, *end);
633 spin_lock(&vmap_area_lock);
634 list_for_each_entry_safe(va, n_va, &valist, purge_list)
635 __free_vmap_area(va);
636 spin_unlock(&vmap_area_lock);
638 spin_unlock(&purge_lock);
642 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
643 * is already purging.
645 static void try_purge_vmap_area_lazy(void)
647 unsigned long start = ULONG_MAX, end = 0;
649 __purge_vmap_area_lazy(&start, &end, 0, 0);
653 * Kick off a purge of the outstanding lazy areas.
655 static void purge_vmap_area_lazy(void)
657 unsigned long start = ULONG_MAX, end = 0;
659 __purge_vmap_area_lazy(&start, &end, 1, 0);
663 * Free a vmap area, caller ensuring that the area has been unmapped
664 * and flush_cache_vunmap had been called for the correct range
667 static void free_vmap_area_noflush(struct vmap_area *va)
669 va->flags |= VM_LAZY_FREE;
670 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
671 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
672 try_purge_vmap_area_lazy();
676 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
677 * called for the correct range previously.
679 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
682 free_vmap_area_noflush(va);
686 * Free and unmap a vmap area
688 static void free_unmap_vmap_area(struct vmap_area *va)
690 flush_cache_vunmap(va->va_start, va->va_end);
691 free_unmap_vmap_area_noflush(va);
694 static struct vmap_area *find_vmap_area(unsigned long addr)
696 struct vmap_area *va;
698 spin_lock(&vmap_area_lock);
699 va = __find_vmap_area(addr);
700 spin_unlock(&vmap_area_lock);
705 static void free_unmap_vmap_area_addr(unsigned long addr)
707 struct vmap_area *va;
709 va = find_vmap_area(addr);
711 free_unmap_vmap_area(va);
715 /*** Per cpu kva allocator ***/
718 * vmap space is limited especially on 32 bit architectures. Ensure there is
719 * room for at least 16 percpu vmap blocks per CPU.
722 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
723 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
724 * instead (we just need a rough idea)
726 #if BITS_PER_LONG == 32
727 #define VMALLOC_SPACE (128UL*1024*1024)
729 #define VMALLOC_SPACE (128UL*1024*1024*1024)
732 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
733 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
734 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
735 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
736 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
737 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
738 #define VMAP_BBMAP_BITS \
739 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
740 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
741 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
743 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
745 static bool vmap_initialized __read_mostly = false;
747 struct vmap_block_queue {
749 struct list_head free;
754 struct vmap_area *va;
755 struct vmap_block_queue *vbq;
756 unsigned long free, dirty;
757 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
758 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
759 struct list_head free_list;
760 struct rcu_head rcu_head;
761 struct list_head purge;
764 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
765 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
768 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
769 * in the free path. Could get rid of this if we change the API to return a
770 * "cookie" from alloc, to be passed to free. But no big deal yet.
772 static DEFINE_SPINLOCK(vmap_block_tree_lock);
773 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
776 * We should probably have a fallback mechanism to allocate virtual memory
777 * out of partially filled vmap blocks. However vmap block sizing should be
778 * fairly reasonable according to the vmalloc size, so it shouldn't be a
782 static unsigned long addr_to_vb_idx(unsigned long addr)
784 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
785 addr /= VMAP_BLOCK_SIZE;
789 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
791 struct vmap_block_queue *vbq;
792 struct vmap_block *vb;
793 struct vmap_area *va;
794 unsigned long vb_idx;
797 node = numa_node_id();
799 vb = kmalloc_node(sizeof(struct vmap_block),
800 gfp_mask & GFP_RECLAIM_MASK, node);
802 return ERR_PTR(-ENOMEM);
804 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
805 VMALLOC_START, VMALLOC_END,
812 err = radix_tree_preload(gfp_mask);
819 spin_lock_init(&vb->lock);
821 vb->free = VMAP_BBMAP_BITS;
823 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
824 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
825 INIT_LIST_HEAD(&vb->free_list);
827 vb_idx = addr_to_vb_idx(va->va_start);
828 spin_lock(&vmap_block_tree_lock);
829 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
830 spin_unlock(&vmap_block_tree_lock);
832 radix_tree_preload_end();
834 vbq = &get_cpu_var(vmap_block_queue);
836 spin_lock(&vbq->lock);
837 list_add_rcu(&vb->free_list, &vbq->free);
838 spin_unlock(&vbq->lock);
839 put_cpu_var(vmap_block_queue);
844 static void free_vmap_block(struct vmap_block *vb)
846 struct vmap_block *tmp;
847 unsigned long vb_idx;
849 vb_idx = addr_to_vb_idx(vb->va->va_start);
850 spin_lock(&vmap_block_tree_lock);
851 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
852 spin_unlock(&vmap_block_tree_lock);
855 free_vmap_area_noflush(vb->va);
856 kfree_rcu(vb, rcu_head);
859 static void purge_fragmented_blocks(int cpu)
862 struct vmap_block *vb;
863 struct vmap_block *n_vb;
864 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
867 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
869 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
872 spin_lock(&vb->lock);
873 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
874 vb->free = 0; /* prevent further allocs after releasing lock */
875 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
876 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
877 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
878 spin_lock(&vbq->lock);
879 list_del_rcu(&vb->free_list);
880 spin_unlock(&vbq->lock);
881 spin_unlock(&vb->lock);
882 list_add_tail(&vb->purge, &purge);
884 spin_unlock(&vb->lock);
888 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
889 list_del(&vb->purge);
894 static void purge_fragmented_blocks_thiscpu(void)
896 purge_fragmented_blocks(smp_processor_id());
899 static void purge_fragmented_blocks_allcpus(void)
903 for_each_possible_cpu(cpu)
904 purge_fragmented_blocks(cpu);
907 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
909 struct vmap_block_queue *vbq;
910 struct vmap_block *vb;
911 unsigned long addr = 0;
914 BUG_ON(size & ~PAGE_MASK);
915 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
916 if (WARN_ON(size == 0)) {
918 * Allocating 0 bytes isn't what caller wants since
919 * get_order(0) returns funny result. Just warn and terminate
924 order = get_order(size);
928 vbq = &get_cpu_var(vmap_block_queue);
929 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
932 spin_lock(&vb->lock);
933 if (vb->free < 1UL << order)
936 i = VMAP_BBMAP_BITS - vb->free;
937 addr = vb->va->va_start + (i << PAGE_SHIFT);
938 BUG_ON(addr_to_vb_idx(addr) !=
939 addr_to_vb_idx(vb->va->va_start));
940 vb->free -= 1UL << order;
942 spin_lock(&vbq->lock);
943 list_del_rcu(&vb->free_list);
944 spin_unlock(&vbq->lock);
946 spin_unlock(&vb->lock);
949 spin_unlock(&vb->lock);
952 put_cpu_var(vmap_block_queue);
956 vb = new_vmap_block(gfp_mask);
965 static void vb_free(const void *addr, unsigned long size)
967 unsigned long offset;
968 unsigned long vb_idx;
970 struct vmap_block *vb;
972 BUG_ON(size & ~PAGE_MASK);
973 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
975 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
977 order = get_order(size);
979 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
981 vb_idx = addr_to_vb_idx((unsigned long)addr);
983 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
987 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
989 spin_lock(&vb->lock);
990 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
992 vb->dirty += 1UL << order;
993 if (vb->dirty == VMAP_BBMAP_BITS) {
995 spin_unlock(&vb->lock);
998 spin_unlock(&vb->lock);
1002 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1004 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1005 * to amortize TLB flushing overheads. What this means is that any page you
1006 * have now, may, in a former life, have been mapped into kernel virtual
1007 * address by the vmap layer and so there might be some CPUs with TLB entries
1008 * still referencing that page (additional to the regular 1:1 kernel mapping).
1010 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1011 * be sure that none of the pages we have control over will have any aliases
1012 * from the vmap layer.
1014 void vm_unmap_aliases(void)
1016 unsigned long start = ULONG_MAX, end = 0;
1020 if (unlikely(!vmap_initialized))
1023 for_each_possible_cpu(cpu) {
1024 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1025 struct vmap_block *vb;
1028 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1031 spin_lock(&vb->lock);
1032 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1033 while (i < VMAP_BBMAP_BITS) {
1036 j = find_next_zero_bit(vb->dirty_map,
1037 VMAP_BBMAP_BITS, i);
1039 s = vb->va->va_start + (i << PAGE_SHIFT);
1040 e = vb->va->va_start + (j << PAGE_SHIFT);
1049 i = find_next_bit(vb->dirty_map,
1050 VMAP_BBMAP_BITS, i);
1052 spin_unlock(&vb->lock);
1057 __purge_vmap_area_lazy(&start, &end, 1, flush);
1059 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1062 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1063 * @mem: the pointer returned by vm_map_ram
1064 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1066 void vm_unmap_ram(const void *mem, unsigned int count)
1068 unsigned long size = count << PAGE_SHIFT;
1069 unsigned long addr = (unsigned long)mem;
1072 BUG_ON(addr < VMALLOC_START);
1073 BUG_ON(addr > VMALLOC_END);
1074 BUG_ON(addr & (PAGE_SIZE-1));
1076 debug_check_no_locks_freed(mem, size);
1077 vmap_debug_free_range(addr, addr+size);
1079 if (likely(count <= VMAP_MAX_ALLOC))
1082 free_unmap_vmap_area_addr(addr);
1084 EXPORT_SYMBOL(vm_unmap_ram);
1087 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1088 * @pages: an array of pointers to the pages to be mapped
1089 * @count: number of pages
1090 * @node: prefer to allocate data structures on this node
1091 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1093 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1095 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1097 unsigned long size = count << PAGE_SHIFT;
1101 if (likely(count <= VMAP_MAX_ALLOC)) {
1102 mem = vb_alloc(size, GFP_KERNEL);
1105 addr = (unsigned long)mem;
1107 struct vmap_area *va;
1108 va = alloc_vmap_area(size, PAGE_SIZE,
1109 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1113 addr = va->va_start;
1116 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1117 vm_unmap_ram(mem, count);
1122 EXPORT_SYMBOL(vm_map_ram);
1124 static struct vm_struct *vmlist __initdata;
1126 * vm_area_add_early - add vmap area early during boot
1127 * @vm: vm_struct to add
1129 * This function is used to add fixed kernel vm area to vmlist before
1130 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1131 * should contain proper values and the other fields should be zero.
1133 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1135 void __init vm_area_add_early(struct vm_struct *vm)
1137 struct vm_struct *tmp, **p;
1139 BUG_ON(vmap_initialized);
1140 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1141 if (tmp->addr >= vm->addr) {
1142 BUG_ON(tmp->addr < vm->addr + vm->size);
1145 BUG_ON(tmp->addr + tmp->size > vm->addr);
1152 * vm_area_register_early - register vmap area early during boot
1153 * @vm: vm_struct to register
1154 * @align: requested alignment
1156 * This function is used to register kernel vm area before
1157 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1158 * proper values on entry and other fields should be zero. On return,
1159 * vm->addr contains the allocated address.
1161 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1163 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1165 static size_t vm_init_off __initdata;
1168 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1169 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1171 vm->addr = (void *)addr;
1173 vm_area_add_early(vm);
1176 void __init vmalloc_init(void)
1178 struct vmap_area *va;
1179 struct vm_struct *tmp;
1182 for_each_possible_cpu(i) {
1183 struct vmap_block_queue *vbq;
1184 struct vfree_deferred *p;
1186 vbq = &per_cpu(vmap_block_queue, i);
1187 spin_lock_init(&vbq->lock);
1188 INIT_LIST_HEAD(&vbq->free);
1189 p = &per_cpu(vfree_deferred, i);
1190 init_llist_head(&p->list);
1191 INIT_WORK(&p->wq, free_work);
1194 /* Import existing vmlist entries. */
1195 for (tmp = vmlist; tmp; tmp = tmp->next) {
1196 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1197 va->flags = VM_VM_AREA;
1198 va->va_start = (unsigned long)tmp->addr;
1199 va->va_end = va->va_start + tmp->size;
1201 __insert_vmap_area(va);
1204 vmap_area_pcpu_hole = VMALLOC_END;
1206 vmap_initialized = true;
1210 * map_kernel_range_noflush - map kernel VM area with the specified pages
1211 * @addr: start of the VM area to map
1212 * @size: size of the VM area to map
1213 * @prot: page protection flags to use
1214 * @pages: pages to map
1216 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1217 * specify should have been allocated using get_vm_area() and its
1221 * This function does NOT do any cache flushing. The caller is
1222 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1223 * before calling this function.
1226 * The number of pages mapped on success, -errno on failure.
1228 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1229 pgprot_t prot, struct page **pages)
1231 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1235 * unmap_kernel_range_noflush - unmap kernel VM area
1236 * @addr: start of the VM area to unmap
1237 * @size: size of the VM area to unmap
1239 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1240 * specify should have been allocated using get_vm_area() and its
1244 * This function does NOT do any cache flushing. The caller is
1245 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1246 * before calling this function and flush_tlb_kernel_range() after.
1248 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1250 vunmap_page_range(addr, addr + size);
1252 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1255 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1256 * @addr: start of the VM area to unmap
1257 * @size: size of the VM area to unmap
1259 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1260 * the unmapping and tlb after.
1262 void unmap_kernel_range(unsigned long addr, unsigned long size)
1264 unsigned long end = addr + size;
1266 flush_cache_vunmap(addr, end);
1267 vunmap_page_range(addr, end);
1268 flush_tlb_kernel_range(addr, end);
1271 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1273 unsigned long addr = (unsigned long)area->addr;
1274 unsigned long end = addr + area->size - PAGE_SIZE;
1277 err = vmap_page_range(addr, end, prot, *pages);
1285 EXPORT_SYMBOL_GPL(map_vm_area);
1287 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1288 unsigned long flags, const void *caller)
1290 spin_lock(&vmap_area_lock);
1292 vm->addr = (void *)va->va_start;
1293 vm->size = va->va_end - va->va_start;
1294 vm->caller = caller;
1296 va->flags |= VM_VM_AREA;
1297 spin_unlock(&vmap_area_lock);
1300 static void clear_vm_unlist(struct vm_struct *vm)
1303 * Before removing VM_UNLIST,
1304 * we should make sure that vm has proper values.
1305 * Pair with smp_rmb() in show_numa_info().
1308 vm->flags &= ~VM_UNLIST;
1311 static struct vm_struct *__get_vm_area_node(unsigned long size,
1312 unsigned long align, unsigned long flags, unsigned long start,
1313 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1315 struct vmap_area *va;
1316 struct vm_struct *area;
1318 BUG_ON(in_interrupt());
1319 if (flags & VM_IOREMAP)
1320 align = 1ul << clamp(fls(size), PAGE_SHIFT, IOREMAP_MAX_ORDER);
1322 size = PAGE_ALIGN(size);
1323 if (unlikely(!size))
1326 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1327 if (unlikely(!area))
1331 * We always allocate a guard page.
1335 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1341 setup_vmalloc_vm(area, va, flags, caller);
1346 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1347 unsigned long start, unsigned long end)
1349 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1350 GFP_KERNEL, __builtin_return_address(0));
1352 EXPORT_SYMBOL_GPL(__get_vm_area);
1354 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1355 unsigned long start, unsigned long end,
1358 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1359 GFP_KERNEL, caller);
1363 * get_vm_area - reserve a contiguous kernel virtual area
1364 * @size: size of the area
1365 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1367 * Search an area of @size in the kernel virtual mapping area,
1368 * and reserved it for out purposes. Returns the area descriptor
1369 * on success or %NULL on failure.
1371 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1373 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1374 NUMA_NO_NODE, GFP_KERNEL,
1375 __builtin_return_address(0));
1378 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1381 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1382 NUMA_NO_NODE, GFP_KERNEL, caller);
1386 * find_vm_area - find a continuous kernel virtual area
1387 * @addr: base address
1389 * Search for the kernel VM area starting at @addr, and return it.
1390 * It is up to the caller to do all required locking to keep the returned
1393 struct vm_struct *find_vm_area(const void *addr)
1395 struct vmap_area *va;
1397 va = find_vmap_area((unsigned long)addr);
1398 if (va && va->flags & VM_VM_AREA)
1405 * remove_vm_area - find and remove a continuous kernel virtual area
1406 * @addr: base address
1408 * Search for the kernel VM area starting at @addr, and remove it.
1409 * This function returns the found VM area, but using it is NOT safe
1410 * on SMP machines, except for its size or flags.
1412 struct vm_struct *remove_vm_area(const void *addr)
1414 struct vmap_area *va;
1416 va = find_vmap_area((unsigned long)addr);
1417 if (va && va->flags & VM_VM_AREA) {
1418 struct vm_struct *vm = va->vm;
1420 spin_lock(&vmap_area_lock);
1422 va->flags &= ~VM_VM_AREA;
1423 spin_unlock(&vmap_area_lock);
1425 vmap_debug_free_range(va->va_start, va->va_end);
1426 free_unmap_vmap_area(va);
1427 vm->size -= PAGE_SIZE;
1434 static void __vunmap(const void *addr, int deallocate_pages)
1436 struct vm_struct *area;
1441 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1445 area = remove_vm_area(addr);
1446 if (unlikely(!area)) {
1447 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1452 debug_check_no_locks_freed(addr, area->size);
1453 debug_check_no_obj_freed(addr, area->size);
1455 if (deallocate_pages) {
1458 for (i = 0; i < area->nr_pages; i++) {
1459 struct page *page = area->pages[i];
1465 if (area->flags & VM_VPAGES)
1476 * vfree - release memory allocated by vmalloc()
1477 * @addr: memory base address
1479 * Free the virtually continuous memory area starting at @addr, as
1480 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1481 * NULL, no operation is performed.
1483 * Must not be called in NMI context (strictly speaking, only if we don't
1484 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1485 * conventions for vfree() arch-depenedent would be a really bad idea)
1487 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1490 void vfree(const void *addr)
1494 kmemleak_free(addr);
1498 if (unlikely(in_interrupt())) {
1499 struct vfree_deferred *p = &__get_cpu_var(vfree_deferred);
1500 llist_add((struct llist_node *)addr, &p->list);
1501 schedule_work(&p->wq);
1505 EXPORT_SYMBOL(vfree);
1508 * vunmap - release virtual mapping obtained by vmap()
1509 * @addr: memory base address
1511 * Free the virtually contiguous memory area starting at @addr,
1512 * which was created from the page array passed to vmap().
1514 * Must not be called in interrupt context.
1516 void vunmap(const void *addr)
1518 BUG_ON(in_interrupt());
1523 EXPORT_SYMBOL(vunmap);
1526 * vmap - map an array of pages into virtually contiguous space
1527 * @pages: array of page pointers
1528 * @count: number of pages to map
1529 * @flags: vm_area->flags
1530 * @prot: page protection for the mapping
1532 * Maps @count pages from @pages into contiguous kernel virtual
1535 void *vmap(struct page **pages, unsigned int count,
1536 unsigned long flags, pgprot_t prot)
1538 struct vm_struct *area;
1542 if (count > totalram_pages)
1545 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1546 __builtin_return_address(0));
1550 if (map_vm_area(area, prot, &pages)) {
1557 EXPORT_SYMBOL(vmap);
1559 static void *__vmalloc_node(unsigned long size, unsigned long align,
1560 gfp_t gfp_mask, pgprot_t prot,
1561 int node, const void *caller);
1562 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1563 pgprot_t prot, int node, const void *caller)
1565 const int order = 0;
1566 struct page **pages;
1567 unsigned int nr_pages, array_size, i;
1568 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1570 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1571 array_size = (nr_pages * sizeof(struct page *));
1573 area->nr_pages = nr_pages;
1574 /* Please note that the recursion is strictly bounded. */
1575 if (array_size > PAGE_SIZE) {
1576 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1577 PAGE_KERNEL, node, caller);
1578 area->flags |= VM_VPAGES;
1580 pages = kmalloc_node(array_size, nested_gfp, node);
1582 area->pages = pages;
1583 area->caller = caller;
1585 remove_vm_area(area->addr);
1590 for (i = 0; i < area->nr_pages; i++) {
1592 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
1595 page = alloc_page(tmp_mask);
1597 page = alloc_pages_node(node, tmp_mask, order);
1599 if (unlikely(!page)) {
1600 /* Successfully allocated i pages, free them in __vunmap() */
1604 area->pages[i] = page;
1607 if (map_vm_area(area, prot, &pages))
1612 warn_alloc_failed(gfp_mask, order,
1613 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1614 (area->nr_pages*PAGE_SIZE), area->size);
1620 * __vmalloc_node_range - allocate virtually contiguous memory
1621 * @size: allocation size
1622 * @align: desired alignment
1623 * @start: vm area range start
1624 * @end: vm area range end
1625 * @gfp_mask: flags for the page level allocator
1626 * @prot: protection mask for the allocated pages
1627 * @node: node to use for allocation or NUMA_NO_NODE
1628 * @caller: caller's return address
1630 * Allocate enough pages to cover @size from the page level
1631 * allocator with @gfp_mask flags. Map them into contiguous
1632 * kernel virtual space, using a pagetable protection of @prot.
1634 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1635 unsigned long start, unsigned long end, gfp_t gfp_mask,
1636 pgprot_t prot, int node, const void *caller)
1638 struct vm_struct *area;
1640 unsigned long real_size = size;
1642 size = PAGE_ALIGN(size);
1643 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1646 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1647 start, end, node, gfp_mask, caller);
1651 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1656 * In this function, newly allocated vm_struct has VM_UNLIST flag.
1657 * It means that vm_struct is not fully initialized.
1658 * Now, it is fully initialized, so remove this flag here.
1660 clear_vm_unlist(area);
1663 * A ref_count = 3 is needed because the vm_struct and vmap_area
1664 * structures allocated in the __get_vm_area_node() function contain
1665 * references to the virtual address of the vmalloc'ed block.
1667 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1672 warn_alloc_failed(gfp_mask, 0,
1673 "vmalloc: allocation failure: %lu bytes\n",
1679 * __vmalloc_node - allocate virtually contiguous memory
1680 * @size: allocation size
1681 * @align: desired alignment
1682 * @gfp_mask: flags for the page level allocator
1683 * @prot: protection mask for the allocated pages
1684 * @node: node to use for allocation or NUMA_NO_NODE
1685 * @caller: caller's return address
1687 * Allocate enough pages to cover @size from the page level
1688 * allocator with @gfp_mask flags. Map them into contiguous
1689 * kernel virtual space, using a pagetable protection of @prot.
1691 static void *__vmalloc_node(unsigned long size, unsigned long align,
1692 gfp_t gfp_mask, pgprot_t prot,
1693 int node, const void *caller)
1695 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1696 gfp_mask, prot, node, caller);
1699 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1701 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1702 __builtin_return_address(0));
1704 EXPORT_SYMBOL(__vmalloc);
1706 static inline void *__vmalloc_node_flags(unsigned long size,
1707 int node, gfp_t flags)
1709 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1710 node, __builtin_return_address(0));
1714 * vmalloc - allocate virtually contiguous memory
1715 * @size: allocation size
1716 * Allocate enough pages to cover @size from the page level
1717 * allocator and map them into contiguous kernel virtual space.
1719 * For tight control over page level allocator and protection flags
1720 * use __vmalloc() instead.
1722 void *vmalloc(unsigned long size)
1724 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1725 GFP_KERNEL | __GFP_HIGHMEM);
1727 EXPORT_SYMBOL(vmalloc);
1730 * vzalloc - allocate virtually contiguous memory with zero fill
1731 * @size: allocation size
1732 * Allocate enough pages to cover @size from the page level
1733 * allocator and map them into contiguous kernel virtual space.
1734 * The memory allocated is set to zero.
1736 * For tight control over page level allocator and protection flags
1737 * use __vmalloc() instead.
1739 void *vzalloc(unsigned long size)
1741 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1742 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1744 EXPORT_SYMBOL(vzalloc);
1747 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1748 * @size: allocation size
1750 * The resulting memory area is zeroed so it can be mapped to userspace
1751 * without leaking data.
1753 void *vmalloc_user(unsigned long size)
1755 struct vm_struct *area;
1758 ret = __vmalloc_node(size, SHMLBA,
1759 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1760 PAGE_KERNEL, NUMA_NO_NODE,
1761 __builtin_return_address(0));
1763 area = find_vm_area(ret);
1764 area->flags |= VM_USERMAP;
1768 EXPORT_SYMBOL(vmalloc_user);
1771 * vmalloc_node - allocate memory on a specific node
1772 * @size: allocation size
1775 * Allocate enough pages to cover @size from the page level
1776 * allocator and map them into contiguous kernel virtual space.
1778 * For tight control over page level allocator and protection flags
1779 * use __vmalloc() instead.
1781 void *vmalloc_node(unsigned long size, int node)
1783 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1784 node, __builtin_return_address(0));
1786 EXPORT_SYMBOL(vmalloc_node);
1789 * vzalloc_node - allocate memory on a specific node with zero fill
1790 * @size: allocation size
1793 * Allocate enough pages to cover @size from the page level
1794 * allocator and map them into contiguous kernel virtual space.
1795 * The memory allocated is set to zero.
1797 * For tight control over page level allocator and protection flags
1798 * use __vmalloc_node() instead.
1800 void *vzalloc_node(unsigned long size, int node)
1802 return __vmalloc_node_flags(size, node,
1803 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1805 EXPORT_SYMBOL(vzalloc_node);
1807 #ifndef PAGE_KERNEL_EXEC
1808 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1812 * vmalloc_exec - allocate virtually contiguous, executable memory
1813 * @size: allocation size
1815 * Kernel-internal function to allocate enough pages to cover @size
1816 * the page level allocator and map them into contiguous and
1817 * executable kernel virtual space.
1819 * For tight control over page level allocator and protection flags
1820 * use __vmalloc() instead.
1823 void *vmalloc_exec(unsigned long size)
1825 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1826 NUMA_NO_NODE, __builtin_return_address(0));
1829 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1830 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1831 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1832 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1834 #define GFP_VMALLOC32 GFP_KERNEL
1838 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1839 * @size: allocation size
1841 * Allocate enough 32bit PA addressable pages to cover @size from the
1842 * page level allocator and map them into contiguous kernel virtual space.
1844 void *vmalloc_32(unsigned long size)
1846 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1847 NUMA_NO_NODE, __builtin_return_address(0));
1849 EXPORT_SYMBOL(vmalloc_32);
1852 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1853 * @size: allocation size
1855 * The resulting memory area is 32bit addressable and zeroed so it can be
1856 * mapped to userspace without leaking data.
1858 void *vmalloc_32_user(unsigned long size)
1860 struct vm_struct *area;
1863 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1864 NUMA_NO_NODE, __builtin_return_address(0));
1866 area = find_vm_area(ret);
1867 area->flags |= VM_USERMAP;
1871 EXPORT_SYMBOL(vmalloc_32_user);
1874 * small helper routine , copy contents to buf from addr.
1875 * If the page is not present, fill zero.
1878 static int aligned_vread(char *buf, char *addr, unsigned long count)
1884 unsigned long offset, length;
1886 offset = (unsigned long)addr & ~PAGE_MASK;
1887 length = PAGE_SIZE - offset;
1890 p = vmalloc_to_page(addr);
1892 * To do safe access to this _mapped_ area, we need
1893 * lock. But adding lock here means that we need to add
1894 * overhead of vmalloc()/vfree() calles for this _debug_
1895 * interface, rarely used. Instead of that, we'll use
1896 * kmap() and get small overhead in this access function.
1900 * we can expect USER0 is not used (see vread/vwrite's
1901 * function description)
1903 void *map = kmap_atomic(p);
1904 memcpy(buf, map + offset, length);
1907 memset(buf, 0, length);
1917 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1923 unsigned long offset, length;
1925 offset = (unsigned long)addr & ~PAGE_MASK;
1926 length = PAGE_SIZE - offset;
1929 p = vmalloc_to_page(addr);
1931 * To do safe access to this _mapped_ area, we need
1932 * lock. But adding lock here means that we need to add
1933 * overhead of vmalloc()/vfree() calles for this _debug_
1934 * interface, rarely used. Instead of that, we'll use
1935 * kmap() and get small overhead in this access function.
1939 * we can expect USER0 is not used (see vread/vwrite's
1940 * function description)
1942 void *map = kmap_atomic(p);
1943 memcpy(map + offset, buf, length);
1955 * vread() - read vmalloc area in a safe way.
1956 * @buf: buffer for reading data
1957 * @addr: vm address.
1958 * @count: number of bytes to be read.
1960 * Returns # of bytes which addr and buf should be increased.
1961 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1962 * includes any intersect with alive vmalloc area.
1964 * This function checks that addr is a valid vmalloc'ed area, and
1965 * copy data from that area to a given buffer. If the given memory range
1966 * of [addr...addr+count) includes some valid address, data is copied to
1967 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1968 * IOREMAP area is treated as memory hole and no copy is done.
1970 * If [addr...addr+count) doesn't includes any intersects with alive
1971 * vm_struct area, returns 0. @buf should be kernel's buffer.
1973 * Note: In usual ops, vread() is never necessary because the caller
1974 * should know vmalloc() area is valid and can use memcpy().
1975 * This is for routines which have to access vmalloc area without
1976 * any informaion, as /dev/kmem.
1980 long vread(char *buf, char *addr, unsigned long count)
1982 struct vmap_area *va;
1983 struct vm_struct *vm;
1984 char *vaddr, *buf_start = buf;
1985 unsigned long buflen = count;
1988 /* Don't allow overflow */
1989 if ((unsigned long) addr + count < count)
1990 count = -(unsigned long) addr;
1992 spin_lock(&vmap_area_lock);
1993 list_for_each_entry(va, &vmap_area_list, list) {
1997 if (!(va->flags & VM_VM_AREA))
2001 vaddr = (char *) vm->addr;
2002 if (addr >= vaddr + vm->size - PAGE_SIZE)
2004 while (addr < vaddr) {
2012 n = vaddr + vm->size - PAGE_SIZE - addr;
2015 if (!(vm->flags & VM_IOREMAP))
2016 aligned_vread(buf, addr, n);
2017 else /* IOREMAP area is treated as memory hole */
2024 spin_unlock(&vmap_area_lock);
2026 if (buf == buf_start)
2028 /* zero-fill memory holes */
2029 if (buf != buf_start + buflen)
2030 memset(buf, 0, buflen - (buf - buf_start));
2036 * vwrite() - write vmalloc area in a safe way.
2037 * @buf: buffer for source data
2038 * @addr: vm address.
2039 * @count: number of bytes to be read.
2041 * Returns # of bytes which addr and buf should be incresed.
2042 * (same number to @count).
2043 * If [addr...addr+count) doesn't includes any intersect with valid
2044 * vmalloc area, returns 0.
2046 * This function checks that addr is a valid vmalloc'ed area, and
2047 * copy data from a buffer to the given addr. If specified range of
2048 * [addr...addr+count) includes some valid address, data is copied from
2049 * proper area of @buf. If there are memory holes, no copy to hole.
2050 * IOREMAP area is treated as memory hole and no copy is done.
2052 * If [addr...addr+count) doesn't includes any intersects with alive
2053 * vm_struct area, returns 0. @buf should be kernel's buffer.
2055 * Note: In usual ops, vwrite() is never necessary because the caller
2056 * should know vmalloc() area is valid and can use memcpy().
2057 * This is for routines which have to access vmalloc area without
2058 * any informaion, as /dev/kmem.
2061 long vwrite(char *buf, char *addr, unsigned long count)
2063 struct vmap_area *va;
2064 struct vm_struct *vm;
2066 unsigned long n, buflen;
2069 /* Don't allow overflow */
2070 if ((unsigned long) addr + count < count)
2071 count = -(unsigned long) addr;
2074 spin_lock(&vmap_area_lock);
2075 list_for_each_entry(va, &vmap_area_list, list) {
2079 if (!(va->flags & VM_VM_AREA))
2083 vaddr = (char *) vm->addr;
2084 if (addr >= vaddr + vm->size - PAGE_SIZE)
2086 while (addr < vaddr) {
2093 n = vaddr + vm->size - PAGE_SIZE - addr;
2096 if (!(vm->flags & VM_IOREMAP)) {
2097 aligned_vwrite(buf, addr, n);
2105 spin_unlock(&vmap_area_lock);
2112 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2113 * @vma: vma to cover
2114 * @uaddr: target user address to start at
2115 * @kaddr: virtual address of vmalloc kernel memory
2116 * @size: size of map area
2118 * Returns: 0 for success, -Exxx on failure
2120 * This function checks that @kaddr is a valid vmalloc'ed area,
2121 * and that it is big enough to cover the range starting at
2122 * @uaddr in @vma. Will return failure if that criteria isn't
2125 * Similar to remap_pfn_range() (see mm/memory.c)
2127 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2128 void *kaddr, unsigned long size)
2130 struct vm_struct *area;
2132 size = PAGE_ALIGN(size);
2134 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2137 area = find_vm_area(kaddr);
2141 if (!(area->flags & VM_USERMAP))
2144 if (kaddr + size > area->addr + area->size)
2148 struct page *page = vmalloc_to_page(kaddr);
2151 ret = vm_insert_page(vma, uaddr, page);
2160 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2164 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2167 * remap_vmalloc_range - map vmalloc pages to userspace
2168 * @vma: vma to cover (map full range of vma)
2169 * @addr: vmalloc memory
2170 * @pgoff: number of pages into addr before first page to map
2172 * Returns: 0 for success, -Exxx on failure
2174 * This function checks that addr is a valid vmalloc'ed area, and
2175 * that it is big enough to cover the vma. Will return failure if
2176 * that criteria isn't met.
2178 * Similar to remap_pfn_range() (see mm/memory.c)
2180 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2181 unsigned long pgoff)
2183 return remap_vmalloc_range_partial(vma, vma->vm_start,
2184 addr + (pgoff << PAGE_SHIFT),
2185 vma->vm_end - vma->vm_start);
2187 EXPORT_SYMBOL(remap_vmalloc_range);
2190 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2193 void __attribute__((weak)) vmalloc_sync_all(void)
2198 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2210 * alloc_vm_area - allocate a range of kernel address space
2211 * @size: size of the area
2212 * @ptes: returns the PTEs for the address space
2214 * Returns: NULL on failure, vm_struct on success
2216 * This function reserves a range of kernel address space, and
2217 * allocates pagetables to map that range. No actual mappings
2220 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2221 * allocated for the VM area are returned.
2223 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2225 struct vm_struct *area;
2227 area = get_vm_area_caller(size, VM_IOREMAP,
2228 __builtin_return_address(0));
2233 * This ensures that page tables are constructed for this region
2234 * of kernel virtual address space and mapped into init_mm.
2236 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2237 size, f, ptes ? &ptes : NULL)) {
2244 EXPORT_SYMBOL_GPL(alloc_vm_area);
2246 void free_vm_area(struct vm_struct *area)
2248 struct vm_struct *ret;
2249 ret = remove_vm_area(area->addr);
2250 BUG_ON(ret != area);
2253 EXPORT_SYMBOL_GPL(free_vm_area);
2256 static struct vmap_area *node_to_va(struct rb_node *n)
2258 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2262 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2263 * @end: target address
2264 * @pnext: out arg for the next vmap_area
2265 * @pprev: out arg for the previous vmap_area
2267 * Returns: %true if either or both of next and prev are found,
2268 * %false if no vmap_area exists
2270 * Find vmap_areas end addresses of which enclose @end. ie. if not
2271 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2273 static bool pvm_find_next_prev(unsigned long end,
2274 struct vmap_area **pnext,
2275 struct vmap_area **pprev)
2277 struct rb_node *n = vmap_area_root.rb_node;
2278 struct vmap_area *va = NULL;
2281 va = rb_entry(n, struct vmap_area, rb_node);
2282 if (end < va->va_end)
2284 else if (end > va->va_end)
2293 if (va->va_end > end) {
2295 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2298 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2304 * pvm_determine_end - find the highest aligned address between two vmap_areas
2305 * @pnext: in/out arg for the next vmap_area
2306 * @pprev: in/out arg for the previous vmap_area
2309 * Returns: determined end address
2311 * Find the highest aligned address between *@pnext and *@pprev below
2312 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2313 * down address is between the end addresses of the two vmap_areas.
2315 * Please note that the address returned by this function may fall
2316 * inside *@pnext vmap_area. The caller is responsible for checking
2319 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2320 struct vmap_area **pprev,
2321 unsigned long align)
2323 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2327 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2331 while (*pprev && (*pprev)->va_end > addr) {
2333 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2340 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2341 * @offsets: array containing offset of each area
2342 * @sizes: array containing size of each area
2343 * @nr_vms: the number of areas to allocate
2344 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2346 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2347 * vm_structs on success, %NULL on failure
2349 * Percpu allocator wants to use congruent vm areas so that it can
2350 * maintain the offsets among percpu areas. This function allocates
2351 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2352 * be scattered pretty far, distance between two areas easily going up
2353 * to gigabytes. To avoid interacting with regular vmallocs, these
2354 * areas are allocated from top.
2356 * Despite its complicated look, this allocator is rather simple. It
2357 * does everything top-down and scans areas from the end looking for
2358 * matching slot. While scanning, if any of the areas overlaps with
2359 * existing vmap_area, the base address is pulled down to fit the
2360 * area. Scanning is repeated till all the areas fit and then all
2361 * necessary data structres are inserted and the result is returned.
2363 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2364 const size_t *sizes, int nr_vms,
2367 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2368 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2369 struct vmap_area **vas, *prev, *next;
2370 struct vm_struct **vms;
2371 int area, area2, last_area, term_area;
2372 unsigned long base, start, end, last_end;
2373 bool purged = false;
2375 /* verify parameters and allocate data structures */
2376 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2377 for (last_area = 0, area = 0; area < nr_vms; area++) {
2378 start = offsets[area];
2379 end = start + sizes[area];
2381 /* is everything aligned properly? */
2382 BUG_ON(!IS_ALIGNED(offsets[area], align));
2383 BUG_ON(!IS_ALIGNED(sizes[area], align));
2385 /* detect the area with the highest address */
2386 if (start > offsets[last_area])
2389 for (area2 = 0; area2 < nr_vms; area2++) {
2390 unsigned long start2 = offsets[area2];
2391 unsigned long end2 = start2 + sizes[area2];
2396 BUG_ON(start2 >= start && start2 < end);
2397 BUG_ON(end2 <= end && end2 > start);
2400 last_end = offsets[last_area] + sizes[last_area];
2402 if (vmalloc_end - vmalloc_start < last_end) {
2407 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2408 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2412 for (area = 0; area < nr_vms; area++) {
2413 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2414 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2415 if (!vas[area] || !vms[area])
2419 spin_lock(&vmap_area_lock);
2421 /* start scanning - we scan from the top, begin with the last area */
2422 area = term_area = last_area;
2423 start = offsets[area];
2424 end = start + sizes[area];
2426 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2427 base = vmalloc_end - last_end;
2430 base = pvm_determine_end(&next, &prev, align) - end;
2433 BUG_ON(next && next->va_end <= base + end);
2434 BUG_ON(prev && prev->va_end > base + end);
2437 * base might have underflowed, add last_end before
2440 if (base + last_end < vmalloc_start + last_end) {
2441 spin_unlock(&vmap_area_lock);
2443 purge_vmap_area_lazy();
2451 * If next overlaps, move base downwards so that it's
2452 * right below next and then recheck.
2454 if (next && next->va_start < base + end) {
2455 base = pvm_determine_end(&next, &prev, align) - end;
2461 * If prev overlaps, shift down next and prev and move
2462 * base so that it's right below new next and then
2465 if (prev && prev->va_end > base + start) {
2467 prev = node_to_va(rb_prev(&next->rb_node));
2468 base = pvm_determine_end(&next, &prev, align) - end;
2474 * This area fits, move on to the previous one. If
2475 * the previous one is the terminal one, we're done.
2477 area = (area + nr_vms - 1) % nr_vms;
2478 if (area == term_area)
2480 start = offsets[area];
2481 end = start + sizes[area];
2482 pvm_find_next_prev(base + end, &next, &prev);
2485 /* we've found a fitting base, insert all va's */
2486 for (area = 0; area < nr_vms; area++) {
2487 struct vmap_area *va = vas[area];
2489 va->va_start = base + offsets[area];
2490 va->va_end = va->va_start + sizes[area];
2491 __insert_vmap_area(va);
2494 vmap_area_pcpu_hole = base + offsets[last_area];
2496 spin_unlock(&vmap_area_lock);
2498 /* insert all vm's */
2499 for (area = 0; area < nr_vms; area++)
2500 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2507 for (area = 0; area < nr_vms; area++) {
2518 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2519 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2520 * @nr_vms: the number of allocated areas
2522 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2524 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2528 for (i = 0; i < nr_vms; i++)
2529 free_vm_area(vms[i]);
2532 #endif /* CONFIG_SMP */
2534 #ifdef CONFIG_PROC_FS
2535 static void *s_start(struct seq_file *m, loff_t *pos)
2536 __acquires(&vmap_area_lock)
2539 struct vmap_area *va;
2541 spin_lock(&vmap_area_lock);
2542 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2543 while (n > 0 && &va->list != &vmap_area_list) {
2545 va = list_entry(va->list.next, typeof(*va), list);
2547 if (!n && &va->list != &vmap_area_list)
2554 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2556 struct vmap_area *va = p, *next;
2559 next = list_entry(va->list.next, typeof(*va), list);
2560 if (&next->list != &vmap_area_list)
2566 static void s_stop(struct seq_file *m, void *p)
2567 __releases(&vmap_area_lock)
2569 spin_unlock(&vmap_area_lock);
2572 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2574 if (IS_ENABLED(CONFIG_NUMA)) {
2575 unsigned int nr, *counters = m->private;
2580 /* Pair with smp_wmb() in clear_vm_unlist() */
2582 if (v->flags & VM_UNLIST)
2585 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2587 for (nr = 0; nr < v->nr_pages; nr++)
2588 counters[page_to_nid(v->pages[nr])]++;
2590 for_each_node_state(nr, N_HIGH_MEMORY)
2592 seq_printf(m, " N%u=%u", nr, counters[nr]);
2596 static int s_show(struct seq_file *m, void *p)
2598 struct vmap_area *va = p;
2599 struct vm_struct *v;
2601 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2604 if (!(va->flags & VM_VM_AREA)) {
2605 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2606 (void *)va->va_start, (void *)va->va_end,
2607 va->va_end - va->va_start);
2613 seq_printf(m, "0x%pK-0x%pK %7ld",
2614 v->addr, v->addr + v->size, v->size);
2617 seq_printf(m, " %pS", v->caller);
2620 seq_printf(m, " pages=%d", v->nr_pages);
2623 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2625 if (v->flags & VM_IOREMAP)
2626 seq_printf(m, " ioremap");
2628 if (v->flags & VM_ALLOC)
2629 seq_printf(m, " vmalloc");
2631 if (v->flags & VM_MAP)
2632 seq_printf(m, " vmap");
2634 if (v->flags & VM_USERMAP)
2635 seq_printf(m, " user");
2637 if (v->flags & VM_VPAGES)
2638 seq_printf(m, " vpages");
2640 show_numa_info(m, v);
2645 static const struct seq_operations vmalloc_op = {
2652 static int vmalloc_open(struct inode *inode, struct file *file)
2654 unsigned int *ptr = NULL;
2657 if (IS_ENABLED(CONFIG_NUMA)) {
2658 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2662 ret = seq_open(file, &vmalloc_op);
2664 struct seq_file *m = file->private_data;
2671 static const struct file_operations proc_vmalloc_operations = {
2672 .open = vmalloc_open,
2674 .llseek = seq_lseek,
2675 .release = seq_release_private,
2678 static int __init proc_vmalloc_init(void)
2680 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2683 module_init(proc_vmalloc_init);
2685 void get_vmalloc_info(struct vmalloc_info *vmi)
2687 struct vmap_area *va;
2688 unsigned long free_area_size;
2689 unsigned long prev_end;
2692 vmi->largest_chunk = 0;
2694 prev_end = VMALLOC_START;
2696 spin_lock(&vmap_area_lock);
2698 if (list_empty(&vmap_area_list)) {
2699 vmi->largest_chunk = VMALLOC_TOTAL;
2703 list_for_each_entry(va, &vmap_area_list, list) {
2704 unsigned long addr = va->va_start;
2707 * Some archs keep another range for modules in vmalloc space
2709 if (addr < VMALLOC_START)
2711 if (addr >= VMALLOC_END)
2714 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2717 vmi->used += (va->va_end - va->va_start);
2719 free_area_size = addr - prev_end;
2720 if (vmi->largest_chunk < free_area_size)
2721 vmi->largest_chunk = free_area_size;
2723 prev_end = va->va_end;
2726 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2727 vmi->largest_chunk = VMALLOC_END - prev_end;
2730 spin_unlock(&vmap_area_lock);