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
252 static DEFINE_SPINLOCK(vmap_area_lock);
253 /* Export for kexec only */
254 LIST_HEAD(vmap_area_list);
255 static struct rb_root vmap_area_root = RB_ROOT;
257 /* The vmap cache globals are protected by vmap_area_lock */
258 static struct rb_node *free_vmap_cache;
259 static unsigned long cached_hole_size;
260 static unsigned long cached_vstart;
261 static unsigned long cached_align;
263 static unsigned long vmap_area_pcpu_hole;
265 static struct vmap_area *__find_vmap_area(unsigned long addr)
267 struct rb_node *n = vmap_area_root.rb_node;
270 struct vmap_area *va;
272 va = rb_entry(n, struct vmap_area, rb_node);
273 if (addr < va->va_start)
275 else if (addr > va->va_start)
284 static void __insert_vmap_area(struct vmap_area *va)
286 struct rb_node **p = &vmap_area_root.rb_node;
287 struct rb_node *parent = NULL;
291 struct vmap_area *tmp_va;
294 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
295 if (va->va_start < tmp_va->va_end)
297 else if (va->va_end > tmp_va->va_start)
303 rb_link_node(&va->rb_node, parent, p);
304 rb_insert_color(&va->rb_node, &vmap_area_root);
306 /* address-sort this list */
307 tmp = rb_prev(&va->rb_node);
309 struct vmap_area *prev;
310 prev = rb_entry(tmp, struct vmap_area, rb_node);
311 list_add_rcu(&va->list, &prev->list);
313 list_add_rcu(&va->list, &vmap_area_list);
316 static void purge_vmap_area_lazy(void);
319 * Allocate a region of KVA of the specified size and alignment, within the
322 static struct vmap_area *alloc_vmap_area(unsigned long size,
324 unsigned long vstart, unsigned long vend,
325 int node, gfp_t gfp_mask)
327 struct vmap_area *va;
331 struct vmap_area *first;
334 BUG_ON(size & ~PAGE_MASK);
335 BUG_ON(!is_power_of_2(align));
337 va = kmalloc_node(sizeof(struct vmap_area),
338 gfp_mask & GFP_RECLAIM_MASK, node);
340 return ERR_PTR(-ENOMEM);
343 spin_lock(&vmap_area_lock);
345 * Invalidate cache if we have more permissive parameters.
346 * cached_hole_size notes the largest hole noticed _below_
347 * the vmap_area cached in free_vmap_cache: if size fits
348 * into that hole, we want to scan from vstart to reuse
349 * the hole instead of allocating above free_vmap_cache.
350 * Note that __free_vmap_area may update free_vmap_cache
351 * without updating cached_hole_size or cached_align.
353 if (!free_vmap_cache ||
354 size < cached_hole_size ||
355 vstart < cached_vstart ||
356 align < cached_align) {
358 cached_hole_size = 0;
359 free_vmap_cache = NULL;
361 /* record if we encounter less permissive parameters */
362 cached_vstart = vstart;
363 cached_align = align;
365 /* find starting point for our search */
366 if (free_vmap_cache) {
367 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
368 addr = ALIGN(first->va_end, align);
371 if (addr + size - 1 < addr)
375 addr = ALIGN(vstart, align);
376 if (addr + size - 1 < addr)
379 n = vmap_area_root.rb_node;
383 struct vmap_area *tmp;
384 tmp = rb_entry(n, struct vmap_area, rb_node);
385 if (tmp->va_end >= addr) {
387 if (tmp->va_start <= addr)
398 /* from the starting point, walk areas until a suitable hole is found */
399 while (addr + size > first->va_start && addr + size <= vend) {
400 if (addr + cached_hole_size < first->va_start)
401 cached_hole_size = first->va_start - addr;
402 addr = ALIGN(first->va_end, align);
403 if (addr + size - 1 < addr)
406 if (list_is_last(&first->list, &vmap_area_list))
409 first = list_entry(first->list.next,
410 struct vmap_area, list);
414 if (addr + size > vend)
418 va->va_end = addr + size;
420 __insert_vmap_area(va);
421 free_vmap_cache = &va->rb_node;
422 spin_unlock(&vmap_area_lock);
424 BUG_ON(va->va_start & (align-1));
425 BUG_ON(va->va_start < vstart);
426 BUG_ON(va->va_end > vend);
431 spin_unlock(&vmap_area_lock);
433 purge_vmap_area_lazy();
437 if (printk_ratelimit())
439 "vmap allocation for size %lu failed: "
440 "use vmalloc=<size> to increase size.\n", size);
442 return ERR_PTR(-EBUSY);
445 static void __free_vmap_area(struct vmap_area *va)
447 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
449 if (free_vmap_cache) {
450 if (va->va_end < cached_vstart) {
451 free_vmap_cache = NULL;
453 struct vmap_area *cache;
454 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
455 if (va->va_start <= cache->va_start) {
456 free_vmap_cache = rb_prev(&va->rb_node);
458 * We don't try to update cached_hole_size or
459 * cached_align, but it won't go very wrong.
464 rb_erase(&va->rb_node, &vmap_area_root);
465 RB_CLEAR_NODE(&va->rb_node);
466 list_del_rcu(&va->list);
469 * Track the highest possible candidate for pcpu area
470 * allocation. Areas outside of vmalloc area can be returned
471 * here too, consider only end addresses which fall inside
472 * vmalloc area proper.
474 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
475 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
477 kfree_rcu(va, rcu_head);
481 * Free a region of KVA allocated by alloc_vmap_area
483 static void free_vmap_area(struct vmap_area *va)
485 spin_lock(&vmap_area_lock);
486 __free_vmap_area(va);
487 spin_unlock(&vmap_area_lock);
491 * Clear the pagetable entries of a given vmap_area
493 static void unmap_vmap_area(struct vmap_area *va)
495 vunmap_page_range(va->va_start, va->va_end);
498 static void vmap_debug_free_range(unsigned long start, unsigned long end)
501 * Unmap page tables and force a TLB flush immediately if
502 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
503 * bugs similarly to those in linear kernel virtual address
504 * space after a page has been freed.
506 * All the lazy freeing logic is still retained, in order to
507 * minimise intrusiveness of this debugging feature.
509 * This is going to be *slow* (linear kernel virtual address
510 * debugging doesn't do a broadcast TLB flush so it is a lot
513 #ifdef CONFIG_DEBUG_PAGEALLOC
514 vunmap_page_range(start, end);
515 flush_tlb_kernel_range(start, end);
520 * lazy_max_pages is the maximum amount of virtual address space we gather up
521 * before attempting to purge with a TLB flush.
523 * There is a tradeoff here: a larger number will cover more kernel page tables
524 * and take slightly longer to purge, but it will linearly reduce the number of
525 * global TLB flushes that must be performed. It would seem natural to scale
526 * this number up linearly with the number of CPUs (because vmapping activity
527 * could also scale linearly with the number of CPUs), however it is likely
528 * that in practice, workloads might be constrained in other ways that mean
529 * vmap activity will not scale linearly with CPUs. Also, I want to be
530 * conservative and not introduce a big latency on huge systems, so go with
531 * a less aggressive log scale. It will still be an improvement over the old
532 * code, and it will be simple to change the scale factor if we find that it
533 * becomes a problem on bigger systems.
535 static unsigned long lazy_max_pages(void)
539 log = fls(num_online_cpus());
541 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
544 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
546 /* for per-CPU blocks */
547 static void purge_fragmented_blocks_allcpus(void);
550 * called before a call to iounmap() if the caller wants vm_area_struct's
553 void set_iounmap_nonlazy(void)
555 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
559 * Purges all lazily-freed vmap areas.
561 * If sync is 0 then don't purge if there is already a purge in progress.
562 * If force_flush is 1, then flush kernel TLBs between *start and *end even
563 * if we found no lazy vmap areas to unmap (callers can use this to optimise
564 * their own TLB flushing).
565 * Returns with *start = min(*start, lowest purged address)
566 * *end = max(*end, highest purged address)
568 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
569 int sync, int force_flush)
571 static DEFINE_SPINLOCK(purge_lock);
573 struct vmap_area *va;
574 struct vmap_area *n_va;
578 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
579 * should not expect such behaviour. This just simplifies locking for
580 * the case that isn't actually used at the moment anyway.
582 if (!sync && !force_flush) {
583 if (!spin_trylock(&purge_lock))
586 spin_lock(&purge_lock);
589 purge_fragmented_blocks_allcpus();
592 list_for_each_entry_rcu(va, &vmap_area_list, list) {
593 if (va->flags & VM_LAZY_FREE) {
594 if (va->va_start < *start)
595 *start = va->va_start;
596 if (va->va_end > *end)
598 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
599 list_add_tail(&va->purge_list, &valist);
600 va->flags |= VM_LAZY_FREEING;
601 va->flags &= ~VM_LAZY_FREE;
607 atomic_sub(nr, &vmap_lazy_nr);
609 if (nr || force_flush)
610 flush_tlb_kernel_range(*start, *end);
613 spin_lock(&vmap_area_lock);
614 list_for_each_entry_safe(va, n_va, &valist, purge_list)
615 __free_vmap_area(va);
616 spin_unlock(&vmap_area_lock);
618 spin_unlock(&purge_lock);
622 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
623 * is already purging.
625 static void try_purge_vmap_area_lazy(void)
627 unsigned long start = ULONG_MAX, end = 0;
629 __purge_vmap_area_lazy(&start, &end, 0, 0);
633 * Kick off a purge of the outstanding lazy areas.
635 static void purge_vmap_area_lazy(void)
637 unsigned long start = ULONG_MAX, end = 0;
639 __purge_vmap_area_lazy(&start, &end, 1, 0);
643 * Free a vmap area, caller ensuring that the area has been unmapped
644 * and flush_cache_vunmap had been called for the correct range
647 static void free_vmap_area_noflush(struct vmap_area *va)
649 va->flags |= VM_LAZY_FREE;
650 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
651 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
652 try_purge_vmap_area_lazy();
656 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
657 * called for the correct range previously.
659 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
662 free_vmap_area_noflush(va);
666 * Free and unmap a vmap area
668 static void free_unmap_vmap_area(struct vmap_area *va)
670 flush_cache_vunmap(va->va_start, va->va_end);
671 free_unmap_vmap_area_noflush(va);
674 static struct vmap_area *find_vmap_area(unsigned long addr)
676 struct vmap_area *va;
678 spin_lock(&vmap_area_lock);
679 va = __find_vmap_area(addr);
680 spin_unlock(&vmap_area_lock);
685 static void free_unmap_vmap_area_addr(unsigned long addr)
687 struct vmap_area *va;
689 va = find_vmap_area(addr);
691 free_unmap_vmap_area(va);
695 /*** Per cpu kva allocator ***/
698 * vmap space is limited especially on 32 bit architectures. Ensure there is
699 * room for at least 16 percpu vmap blocks per CPU.
702 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
703 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
704 * instead (we just need a rough idea)
706 #if BITS_PER_LONG == 32
707 #define VMALLOC_SPACE (128UL*1024*1024)
709 #define VMALLOC_SPACE (128UL*1024*1024*1024)
712 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
713 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
714 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
715 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
716 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
717 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
718 #define VMAP_BBMAP_BITS \
719 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
720 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
721 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
723 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
725 static bool vmap_initialized __read_mostly = false;
727 struct vmap_block_queue {
729 struct list_head free;
734 struct vmap_area *va;
735 struct vmap_block_queue *vbq;
736 unsigned long free, dirty;
737 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
738 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
739 struct list_head free_list;
740 struct rcu_head rcu_head;
741 struct list_head purge;
744 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
745 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
748 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
749 * in the free path. Could get rid of this if we change the API to return a
750 * "cookie" from alloc, to be passed to free. But no big deal yet.
752 static DEFINE_SPINLOCK(vmap_block_tree_lock);
753 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
756 * We should probably have a fallback mechanism to allocate virtual memory
757 * out of partially filled vmap blocks. However vmap block sizing should be
758 * fairly reasonable according to the vmalloc size, so it shouldn't be a
762 static unsigned long addr_to_vb_idx(unsigned long addr)
764 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
765 addr /= VMAP_BLOCK_SIZE;
769 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
771 struct vmap_block_queue *vbq;
772 struct vmap_block *vb;
773 struct vmap_area *va;
774 unsigned long vb_idx;
777 node = numa_node_id();
779 vb = kmalloc_node(sizeof(struct vmap_block),
780 gfp_mask & GFP_RECLAIM_MASK, node);
782 return ERR_PTR(-ENOMEM);
784 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
785 VMALLOC_START, VMALLOC_END,
792 err = radix_tree_preload(gfp_mask);
799 spin_lock_init(&vb->lock);
801 vb->free = VMAP_BBMAP_BITS;
803 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
804 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
805 INIT_LIST_HEAD(&vb->free_list);
807 vb_idx = addr_to_vb_idx(va->va_start);
808 spin_lock(&vmap_block_tree_lock);
809 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
810 spin_unlock(&vmap_block_tree_lock);
812 radix_tree_preload_end();
814 vbq = &get_cpu_var(vmap_block_queue);
816 spin_lock(&vbq->lock);
817 list_add_rcu(&vb->free_list, &vbq->free);
818 spin_unlock(&vbq->lock);
819 put_cpu_var(vmap_block_queue);
824 static void free_vmap_block(struct vmap_block *vb)
826 struct vmap_block *tmp;
827 unsigned long vb_idx;
829 vb_idx = addr_to_vb_idx(vb->va->va_start);
830 spin_lock(&vmap_block_tree_lock);
831 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
832 spin_unlock(&vmap_block_tree_lock);
835 free_vmap_area_noflush(vb->va);
836 kfree_rcu(vb, rcu_head);
839 static void purge_fragmented_blocks(int cpu)
842 struct vmap_block *vb;
843 struct vmap_block *n_vb;
844 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
847 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
849 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
852 spin_lock(&vb->lock);
853 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
854 vb->free = 0; /* prevent further allocs after releasing lock */
855 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
856 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
857 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
858 spin_lock(&vbq->lock);
859 list_del_rcu(&vb->free_list);
860 spin_unlock(&vbq->lock);
861 spin_unlock(&vb->lock);
862 list_add_tail(&vb->purge, &purge);
864 spin_unlock(&vb->lock);
868 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
869 list_del(&vb->purge);
874 static void purge_fragmented_blocks_thiscpu(void)
876 purge_fragmented_blocks(smp_processor_id());
879 static void purge_fragmented_blocks_allcpus(void)
883 for_each_possible_cpu(cpu)
884 purge_fragmented_blocks(cpu);
887 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
889 struct vmap_block_queue *vbq;
890 struct vmap_block *vb;
891 unsigned long addr = 0;
895 BUG_ON(size & ~PAGE_MASK);
896 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
897 if (WARN_ON(size == 0)) {
899 * Allocating 0 bytes isn't what caller wants since
900 * get_order(0) returns funny result. Just warn and terminate
905 order = get_order(size);
909 vbq = &get_cpu_var(vmap_block_queue);
910 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
913 spin_lock(&vb->lock);
914 if (vb->free < 1UL << order)
917 i = bitmap_find_free_region(vb->alloc_map,
918 VMAP_BBMAP_BITS, order);
921 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
922 /* fragmented and no outstanding allocations */
923 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
928 addr = vb->va->va_start + (i << PAGE_SHIFT);
929 BUG_ON(addr_to_vb_idx(addr) !=
930 addr_to_vb_idx(vb->va->va_start));
931 vb->free -= 1UL << order;
933 spin_lock(&vbq->lock);
934 list_del_rcu(&vb->free_list);
935 spin_unlock(&vbq->lock);
937 spin_unlock(&vb->lock);
940 spin_unlock(&vb->lock);
944 purge_fragmented_blocks_thiscpu();
946 put_cpu_var(vmap_block_queue);
950 vb = new_vmap_block(gfp_mask);
959 static void vb_free(const void *addr, unsigned long size)
961 unsigned long offset;
962 unsigned long vb_idx;
964 struct vmap_block *vb;
966 BUG_ON(size & ~PAGE_MASK);
967 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
969 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
971 order = get_order(size);
973 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
975 vb_idx = addr_to_vb_idx((unsigned long)addr);
977 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
981 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
983 spin_lock(&vb->lock);
984 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
986 vb->dirty += 1UL << order;
987 if (vb->dirty == VMAP_BBMAP_BITS) {
989 spin_unlock(&vb->lock);
992 spin_unlock(&vb->lock);
996 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
998 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
999 * to amortize TLB flushing overheads. What this means is that any page you
1000 * have now, may, in a former life, have been mapped into kernel virtual
1001 * address by the vmap layer and so there might be some CPUs with TLB entries
1002 * still referencing that page (additional to the regular 1:1 kernel mapping).
1004 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1005 * be sure that none of the pages we have control over will have any aliases
1006 * from the vmap layer.
1008 void vm_unmap_aliases(void)
1010 unsigned long start = ULONG_MAX, end = 0;
1014 if (unlikely(!vmap_initialized))
1017 for_each_possible_cpu(cpu) {
1018 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1019 struct vmap_block *vb;
1022 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1025 spin_lock(&vb->lock);
1026 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1027 while (i < VMAP_BBMAP_BITS) {
1030 j = find_next_zero_bit(vb->dirty_map,
1031 VMAP_BBMAP_BITS, i);
1033 s = vb->va->va_start + (i << PAGE_SHIFT);
1034 e = vb->va->va_start + (j << PAGE_SHIFT);
1043 i = find_next_bit(vb->dirty_map,
1044 VMAP_BBMAP_BITS, i);
1046 spin_unlock(&vb->lock);
1051 __purge_vmap_area_lazy(&start, &end, 1, flush);
1053 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1056 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1057 * @mem: the pointer returned by vm_map_ram
1058 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1060 void vm_unmap_ram(const void *mem, unsigned int count)
1062 unsigned long size = count << PAGE_SHIFT;
1063 unsigned long addr = (unsigned long)mem;
1066 BUG_ON(addr < VMALLOC_START);
1067 BUG_ON(addr > VMALLOC_END);
1068 BUG_ON(addr & (PAGE_SIZE-1));
1070 debug_check_no_locks_freed(mem, size);
1071 vmap_debug_free_range(addr, addr+size);
1073 if (likely(count <= VMAP_MAX_ALLOC))
1076 free_unmap_vmap_area_addr(addr);
1078 EXPORT_SYMBOL(vm_unmap_ram);
1081 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1082 * @pages: an array of pointers to the pages to be mapped
1083 * @count: number of pages
1084 * @node: prefer to allocate data structures on this node
1085 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1087 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1089 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1091 unsigned long size = count << PAGE_SHIFT;
1095 if (likely(count <= VMAP_MAX_ALLOC)) {
1096 mem = vb_alloc(size, GFP_KERNEL);
1099 addr = (unsigned long)mem;
1101 struct vmap_area *va;
1102 va = alloc_vmap_area(size, PAGE_SIZE,
1103 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1107 addr = va->va_start;
1110 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1111 vm_unmap_ram(mem, count);
1116 EXPORT_SYMBOL(vm_map_ram);
1118 static struct vm_struct *vmlist __initdata;
1120 * vm_area_add_early - add vmap area early during boot
1121 * @vm: vm_struct to add
1123 * This function is used to add fixed kernel vm area to vmlist before
1124 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1125 * should contain proper values and the other fields should be zero.
1127 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1129 void __init vm_area_add_early(struct vm_struct *vm)
1131 struct vm_struct *tmp, **p;
1133 BUG_ON(vmap_initialized);
1134 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1135 if (tmp->addr >= vm->addr) {
1136 BUG_ON(tmp->addr < vm->addr + vm->size);
1139 BUG_ON(tmp->addr + tmp->size > vm->addr);
1146 * vm_area_register_early - register vmap area early during boot
1147 * @vm: vm_struct to register
1148 * @align: requested alignment
1150 * This function is used to register kernel vm area before
1151 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1152 * proper values on entry and other fields should be zero. On return,
1153 * vm->addr contains the allocated address.
1155 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1157 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1159 static size_t vm_init_off __initdata;
1162 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1163 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1165 vm->addr = (void *)addr;
1167 vm_area_add_early(vm);
1170 void __init vmalloc_init(void)
1172 struct vmap_area *va;
1173 struct vm_struct *tmp;
1176 for_each_possible_cpu(i) {
1177 struct vmap_block_queue *vbq;
1179 vbq = &per_cpu(vmap_block_queue, i);
1180 spin_lock_init(&vbq->lock);
1181 INIT_LIST_HEAD(&vbq->free);
1184 /* Import existing vmlist entries. */
1185 for (tmp = vmlist; tmp; tmp = tmp->next) {
1186 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1187 va->flags = VM_VM_AREA;
1188 va->va_start = (unsigned long)tmp->addr;
1189 va->va_end = va->va_start + tmp->size;
1191 __insert_vmap_area(va);
1194 vmap_area_pcpu_hole = VMALLOC_END;
1196 vmap_initialized = true;
1200 * map_kernel_range_noflush - map kernel VM area with the specified pages
1201 * @addr: start of the VM area to map
1202 * @size: size of the VM area to map
1203 * @prot: page protection flags to use
1204 * @pages: pages to map
1206 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1207 * specify should have been allocated using get_vm_area() and its
1211 * This function does NOT do any cache flushing. The caller is
1212 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1213 * before calling this function.
1216 * The number of pages mapped on success, -errno on failure.
1218 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1219 pgprot_t prot, struct page **pages)
1221 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1225 * unmap_kernel_range_noflush - unmap kernel VM area
1226 * @addr: start of the VM area to unmap
1227 * @size: size of the VM area to unmap
1229 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1230 * specify should have been allocated using get_vm_area() and its
1234 * This function does NOT do any cache flushing. The caller is
1235 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1236 * before calling this function and flush_tlb_kernel_range() after.
1238 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1240 vunmap_page_range(addr, addr + size);
1242 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1245 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1246 * @addr: start of the VM area to unmap
1247 * @size: size of the VM area to unmap
1249 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1250 * the unmapping and tlb after.
1252 void unmap_kernel_range(unsigned long addr, unsigned long size)
1254 unsigned long end = addr + size;
1256 flush_cache_vunmap(addr, end);
1257 vunmap_page_range(addr, end);
1258 flush_tlb_kernel_range(addr, end);
1261 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1263 unsigned long addr = (unsigned long)area->addr;
1264 unsigned long end = addr + area->size - PAGE_SIZE;
1267 err = vmap_page_range(addr, end, prot, *pages);
1275 EXPORT_SYMBOL_GPL(map_vm_area);
1277 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1278 unsigned long flags, const void *caller)
1280 spin_lock(&vmap_area_lock);
1282 vm->addr = (void *)va->va_start;
1283 vm->size = va->va_end - va->va_start;
1284 vm->caller = caller;
1286 va->flags |= VM_VM_AREA;
1287 spin_unlock(&vmap_area_lock);
1290 static void clear_vm_unlist(struct vm_struct *vm)
1293 * Before removing VM_UNLIST,
1294 * we should make sure that vm has proper values.
1295 * Pair with smp_rmb() in show_numa_info().
1298 vm->flags &= ~VM_UNLIST;
1301 static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1302 unsigned long flags, const void *caller)
1304 setup_vmalloc_vm(vm, va, flags, caller);
1305 clear_vm_unlist(vm);
1308 static struct vm_struct *__get_vm_area_node(unsigned long size,
1309 unsigned long align, unsigned long flags, unsigned long start,
1310 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1312 struct vmap_area *va;
1313 struct vm_struct *area;
1315 BUG_ON(in_interrupt());
1316 if (flags & VM_IOREMAP) {
1317 int bit = fls(size);
1319 if (bit > IOREMAP_MAX_ORDER)
1320 bit = IOREMAP_MAX_ORDER;
1321 else if (bit < PAGE_SHIFT)
1327 size = PAGE_ALIGN(size);
1328 if (unlikely(!size))
1331 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1332 if (unlikely(!area))
1336 * We always allocate a guard page.
1340 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1347 * When this function is called from __vmalloc_node_range,
1348 * we add VM_UNLIST flag to avoid accessing uninitialized
1349 * members of vm_struct such as pages and nr_pages fields.
1350 * They will be set later.
1352 if (flags & VM_UNLIST)
1353 setup_vmalloc_vm(area, va, flags, caller);
1355 insert_vmalloc_vm(area, va, flags, caller);
1360 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1361 unsigned long start, unsigned long end)
1363 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1364 GFP_KERNEL, __builtin_return_address(0));
1366 EXPORT_SYMBOL_GPL(__get_vm_area);
1368 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1369 unsigned long start, unsigned long end,
1372 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1373 GFP_KERNEL, caller);
1377 * get_vm_area - reserve a contiguous kernel virtual area
1378 * @size: size of the area
1379 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1381 * Search an area of @size in the kernel virtual mapping area,
1382 * and reserved it for out purposes. Returns the area descriptor
1383 * on success or %NULL on failure.
1385 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1387 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1388 NUMA_NO_NODE, GFP_KERNEL,
1389 __builtin_return_address(0));
1392 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1395 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1396 NUMA_NO_NODE, GFP_KERNEL, caller);
1400 * find_vm_area - find a continuous kernel virtual area
1401 * @addr: base address
1403 * Search for the kernel VM area starting at @addr, and return it.
1404 * It is up to the caller to do all required locking to keep the returned
1407 struct vm_struct *find_vm_area(const void *addr)
1409 struct vmap_area *va;
1411 va = find_vmap_area((unsigned long)addr);
1412 if (va && va->flags & VM_VM_AREA)
1419 * remove_vm_area - find and remove a continuous kernel virtual area
1420 * @addr: base address
1422 * Search for the kernel VM area starting at @addr, and remove it.
1423 * This function returns the found VM area, but using it is NOT safe
1424 * on SMP machines, except for its size or flags.
1426 struct vm_struct *remove_vm_area(const void *addr)
1428 struct vmap_area *va;
1430 va = find_vmap_area((unsigned long)addr);
1431 if (va && va->flags & VM_VM_AREA) {
1432 struct vm_struct *vm = va->vm;
1434 spin_lock(&vmap_area_lock);
1436 va->flags &= ~VM_VM_AREA;
1437 spin_unlock(&vmap_area_lock);
1439 vmap_debug_free_range(va->va_start, va->va_end);
1440 free_unmap_vmap_area(va);
1441 vm->size -= PAGE_SIZE;
1448 static void __vunmap(const void *addr, int deallocate_pages)
1450 struct vm_struct *area;
1455 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1456 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1460 area = remove_vm_area(addr);
1461 if (unlikely(!area)) {
1462 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1467 debug_check_no_locks_freed(addr, area->size);
1468 debug_check_no_obj_freed(addr, area->size);
1470 if (deallocate_pages) {
1473 for (i = 0; i < area->nr_pages; i++) {
1474 struct page *page = area->pages[i];
1480 if (area->flags & VM_VPAGES)
1491 * vfree - release memory allocated by vmalloc()
1492 * @addr: memory base address
1494 * Free the virtually continuous memory area starting at @addr, as
1495 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1496 * NULL, no operation is performed.
1498 * Must not be called in interrupt context.
1500 void vfree(const void *addr)
1502 BUG_ON(in_interrupt());
1504 kmemleak_free(addr);
1508 EXPORT_SYMBOL(vfree);
1511 * vunmap - release virtual mapping obtained by vmap()
1512 * @addr: memory base address
1514 * Free the virtually contiguous memory area starting at @addr,
1515 * which was created from the page array passed to vmap().
1517 * Must not be called in interrupt context.
1519 void vunmap(const void *addr)
1521 BUG_ON(in_interrupt());
1525 EXPORT_SYMBOL(vunmap);
1528 * vmap - map an array of pages into virtually contiguous space
1529 * @pages: array of page pointers
1530 * @count: number of pages to map
1531 * @flags: vm_area->flags
1532 * @prot: page protection for the mapping
1534 * Maps @count pages from @pages into contiguous kernel virtual
1537 void *vmap(struct page **pages, unsigned int count,
1538 unsigned long flags, pgprot_t prot)
1540 struct vm_struct *area;
1544 if (count > totalram_pages)
1547 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1548 __builtin_return_address(0));
1552 if (map_vm_area(area, prot, &pages)) {
1559 EXPORT_SYMBOL(vmap);
1561 static void *__vmalloc_node(unsigned long size, unsigned long align,
1562 gfp_t gfp_mask, pgprot_t prot,
1563 int node, const void *caller);
1564 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1565 pgprot_t prot, int node, const void *caller)
1567 const int order = 0;
1568 struct page **pages;
1569 unsigned int nr_pages, array_size, i;
1570 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1572 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1573 array_size = (nr_pages * sizeof(struct page *));
1575 area->nr_pages = nr_pages;
1576 /* Please note that the recursion is strictly bounded. */
1577 if (array_size > PAGE_SIZE) {
1578 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1579 PAGE_KERNEL, node, caller);
1580 area->flags |= VM_VPAGES;
1582 pages = kmalloc_node(array_size, nested_gfp, node);
1584 area->pages = pages;
1585 area->caller = caller;
1587 remove_vm_area(area->addr);
1592 for (i = 0; i < area->nr_pages; i++) {
1594 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
1597 page = alloc_page(tmp_mask);
1599 page = alloc_pages_node(node, tmp_mask, order);
1601 if (unlikely(!page)) {
1602 /* Successfully allocated i pages, free them in __vunmap() */
1606 area->pages[i] = page;
1609 if (map_vm_area(area, prot, &pages))
1614 warn_alloc_failed(gfp_mask, order,
1615 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1616 (area->nr_pages*PAGE_SIZE), area->size);
1622 * __vmalloc_node_range - allocate virtually contiguous memory
1623 * @size: allocation size
1624 * @align: desired alignment
1625 * @start: vm area range start
1626 * @end: vm area range end
1627 * @gfp_mask: flags for the page level allocator
1628 * @prot: protection mask for the allocated pages
1629 * @node: node to use for allocation or NUMA_NO_NODE
1630 * @caller: caller's return address
1632 * Allocate enough pages to cover @size from the page level
1633 * allocator with @gfp_mask flags. Map them into contiguous
1634 * kernel virtual space, using a pagetable protection of @prot.
1636 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1637 unsigned long start, unsigned long end, gfp_t gfp_mask,
1638 pgprot_t prot, int node, const void *caller)
1640 struct vm_struct *area;
1642 unsigned long real_size = size;
1644 size = PAGE_ALIGN(size);
1645 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1648 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1649 start, end, node, gfp_mask, caller);
1653 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1658 * In this function, newly allocated vm_struct has VM_UNLIST flag.
1659 * It means that vm_struct is not fully initialized.
1660 * Now, it is fully initialized, so remove this flag here.
1662 clear_vm_unlist(area);
1665 * A ref_count = 3 is needed because the vm_struct and vmap_area
1666 * structures allocated in the __get_vm_area_node() function contain
1667 * references to the virtual address of the vmalloc'ed block.
1669 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1674 warn_alloc_failed(gfp_mask, 0,
1675 "vmalloc: allocation failure: %lu bytes\n",
1681 * __vmalloc_node - allocate virtually contiguous memory
1682 * @size: allocation size
1683 * @align: desired alignment
1684 * @gfp_mask: flags for the page level allocator
1685 * @prot: protection mask for the allocated pages
1686 * @node: node to use for allocation or NUMA_NO_NODE
1687 * @caller: caller's return address
1689 * Allocate enough pages to cover @size from the page level
1690 * allocator with @gfp_mask flags. Map them into contiguous
1691 * kernel virtual space, using a pagetable protection of @prot.
1693 static void *__vmalloc_node(unsigned long size, unsigned long align,
1694 gfp_t gfp_mask, pgprot_t prot,
1695 int node, const void *caller)
1697 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1698 gfp_mask, prot, node, caller);
1701 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1703 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1704 __builtin_return_address(0));
1706 EXPORT_SYMBOL(__vmalloc);
1708 static inline void *__vmalloc_node_flags(unsigned long size,
1709 int node, gfp_t flags)
1711 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1712 node, __builtin_return_address(0));
1716 * vmalloc - allocate virtually contiguous memory
1717 * @size: allocation size
1718 * Allocate enough pages to cover @size from the page level
1719 * allocator and map them into contiguous kernel virtual space.
1721 * For tight control over page level allocator and protection flags
1722 * use __vmalloc() instead.
1724 void *vmalloc(unsigned long size)
1726 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1727 GFP_KERNEL | __GFP_HIGHMEM);
1729 EXPORT_SYMBOL(vmalloc);
1732 * vzalloc - allocate virtually contiguous memory with zero fill
1733 * @size: allocation size
1734 * Allocate enough pages to cover @size from the page level
1735 * allocator and map them into contiguous kernel virtual space.
1736 * The memory allocated is set to zero.
1738 * For tight control over page level allocator and protection flags
1739 * use __vmalloc() instead.
1741 void *vzalloc(unsigned long size)
1743 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1744 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1746 EXPORT_SYMBOL(vzalloc);
1749 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1750 * @size: allocation size
1752 * The resulting memory area is zeroed so it can be mapped to userspace
1753 * without leaking data.
1755 void *vmalloc_user(unsigned long size)
1757 struct vm_struct *area;
1760 ret = __vmalloc_node(size, SHMLBA,
1761 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1762 PAGE_KERNEL, NUMA_NO_NODE,
1763 __builtin_return_address(0));
1765 area = find_vm_area(ret);
1766 area->flags |= VM_USERMAP;
1770 EXPORT_SYMBOL(vmalloc_user);
1773 * vmalloc_node - allocate memory on a specific node
1774 * @size: allocation size
1777 * Allocate enough pages to cover @size from the page level
1778 * allocator and map them into contiguous kernel virtual space.
1780 * For tight control over page level allocator and protection flags
1781 * use __vmalloc() instead.
1783 void *vmalloc_node(unsigned long size, int node)
1785 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1786 node, __builtin_return_address(0));
1788 EXPORT_SYMBOL(vmalloc_node);
1791 * vzalloc_node - allocate memory on a specific node with zero fill
1792 * @size: allocation size
1795 * Allocate enough pages to cover @size from the page level
1796 * allocator and map them into contiguous kernel virtual space.
1797 * The memory allocated is set to zero.
1799 * For tight control over page level allocator and protection flags
1800 * use __vmalloc_node() instead.
1802 void *vzalloc_node(unsigned long size, int node)
1804 return __vmalloc_node_flags(size, node,
1805 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1807 EXPORT_SYMBOL(vzalloc_node);
1809 #ifndef PAGE_KERNEL_EXEC
1810 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1814 * vmalloc_exec - allocate virtually contiguous, executable memory
1815 * @size: allocation size
1817 * Kernel-internal function to allocate enough pages to cover @size
1818 * the page level allocator and map them into contiguous and
1819 * executable kernel virtual space.
1821 * For tight control over page level allocator and protection flags
1822 * use __vmalloc() instead.
1825 void *vmalloc_exec(unsigned long size)
1827 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1828 NUMA_NO_NODE, __builtin_return_address(0));
1831 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1832 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1833 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1834 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1836 #define GFP_VMALLOC32 GFP_KERNEL
1840 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1841 * @size: allocation size
1843 * Allocate enough 32bit PA addressable pages to cover @size from the
1844 * page level allocator and map them into contiguous kernel virtual space.
1846 void *vmalloc_32(unsigned long size)
1848 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1849 NUMA_NO_NODE, __builtin_return_address(0));
1851 EXPORT_SYMBOL(vmalloc_32);
1854 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1855 * @size: allocation size
1857 * The resulting memory area is 32bit addressable and zeroed so it can be
1858 * mapped to userspace without leaking data.
1860 void *vmalloc_32_user(unsigned long size)
1862 struct vm_struct *area;
1865 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1866 NUMA_NO_NODE, __builtin_return_address(0));
1868 area = find_vm_area(ret);
1869 area->flags |= VM_USERMAP;
1873 EXPORT_SYMBOL(vmalloc_32_user);
1876 * small helper routine , copy contents to buf from addr.
1877 * If the page is not present, fill zero.
1880 static int aligned_vread(char *buf, char *addr, unsigned long count)
1886 unsigned long offset, length;
1888 offset = (unsigned long)addr & ~PAGE_MASK;
1889 length = PAGE_SIZE - offset;
1892 p = vmalloc_to_page(addr);
1894 * To do safe access to this _mapped_ area, we need
1895 * lock. But adding lock here means that we need to add
1896 * overhead of vmalloc()/vfree() calles for this _debug_
1897 * interface, rarely used. Instead of that, we'll use
1898 * kmap() and get small overhead in this access function.
1902 * we can expect USER0 is not used (see vread/vwrite's
1903 * function description)
1905 void *map = kmap_atomic(p);
1906 memcpy(buf, map + offset, length);
1909 memset(buf, 0, length);
1919 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1925 unsigned long offset, length;
1927 offset = (unsigned long)addr & ~PAGE_MASK;
1928 length = PAGE_SIZE - offset;
1931 p = vmalloc_to_page(addr);
1933 * To do safe access to this _mapped_ area, we need
1934 * lock. But adding lock here means that we need to add
1935 * overhead of vmalloc()/vfree() calles for this _debug_
1936 * interface, rarely used. Instead of that, we'll use
1937 * kmap() and get small overhead in this access function.
1941 * we can expect USER0 is not used (see vread/vwrite's
1942 * function description)
1944 void *map = kmap_atomic(p);
1945 memcpy(map + offset, buf, length);
1957 * vread() - read vmalloc area in a safe way.
1958 * @buf: buffer for reading data
1959 * @addr: vm address.
1960 * @count: number of bytes to be read.
1962 * Returns # of bytes which addr and buf should be increased.
1963 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1964 * includes any intersect with alive vmalloc area.
1966 * This function checks that addr is a valid vmalloc'ed area, and
1967 * copy data from that area to a given buffer. If the given memory range
1968 * of [addr...addr+count) includes some valid address, data is copied to
1969 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1970 * IOREMAP area is treated as memory hole and no copy is done.
1972 * If [addr...addr+count) doesn't includes any intersects with alive
1973 * vm_struct area, returns 0. @buf should be kernel's buffer.
1975 * Note: In usual ops, vread() is never necessary because the caller
1976 * should know vmalloc() area is valid and can use memcpy().
1977 * This is for routines which have to access vmalloc area without
1978 * any informaion, as /dev/kmem.
1982 long vread(char *buf, char *addr, unsigned long count)
1984 struct vmap_area *va;
1985 struct vm_struct *vm;
1986 char *vaddr, *buf_start = buf;
1987 unsigned long buflen = count;
1990 /* Don't allow overflow */
1991 if ((unsigned long) addr + count < count)
1992 count = -(unsigned long) addr;
1994 spin_lock(&vmap_area_lock);
1995 list_for_each_entry(va, &vmap_area_list, list) {
1999 if (!(va->flags & VM_VM_AREA))
2003 vaddr = (char *) vm->addr;
2004 if (addr >= vaddr + vm->size - PAGE_SIZE)
2006 while (addr < vaddr) {
2014 n = vaddr + vm->size - PAGE_SIZE - addr;
2017 if (!(vm->flags & VM_IOREMAP))
2018 aligned_vread(buf, addr, n);
2019 else /* IOREMAP area is treated as memory hole */
2026 spin_unlock(&vmap_area_lock);
2028 if (buf == buf_start)
2030 /* zero-fill memory holes */
2031 if (buf != buf_start + buflen)
2032 memset(buf, 0, buflen - (buf - buf_start));
2038 * vwrite() - write vmalloc area in a safe way.
2039 * @buf: buffer for source data
2040 * @addr: vm address.
2041 * @count: number of bytes to be read.
2043 * Returns # of bytes which addr and buf should be incresed.
2044 * (same number to @count).
2045 * If [addr...addr+count) doesn't includes any intersect with valid
2046 * vmalloc area, returns 0.
2048 * This function checks that addr is a valid vmalloc'ed area, and
2049 * copy data from a buffer to the given addr. If specified range of
2050 * [addr...addr+count) includes some valid address, data is copied from
2051 * proper area of @buf. If there are memory holes, no copy to hole.
2052 * IOREMAP area is treated as memory hole and no copy is done.
2054 * If [addr...addr+count) doesn't includes any intersects with alive
2055 * vm_struct area, returns 0. @buf should be kernel's buffer.
2057 * Note: In usual ops, vwrite() is never necessary because the caller
2058 * should know vmalloc() area is valid and can use memcpy().
2059 * This is for routines which have to access vmalloc area without
2060 * any informaion, as /dev/kmem.
2063 long vwrite(char *buf, char *addr, unsigned long count)
2065 struct vmap_area *va;
2066 struct vm_struct *vm;
2068 unsigned long n, buflen;
2071 /* Don't allow overflow */
2072 if ((unsigned long) addr + count < count)
2073 count = -(unsigned long) addr;
2076 spin_lock(&vmap_area_lock);
2077 list_for_each_entry(va, &vmap_area_list, list) {
2081 if (!(va->flags & VM_VM_AREA))
2085 vaddr = (char *) vm->addr;
2086 if (addr >= vaddr + vm->size - PAGE_SIZE)
2088 while (addr < vaddr) {
2095 n = vaddr + vm->size - PAGE_SIZE - addr;
2098 if (!(vm->flags & VM_IOREMAP)) {
2099 aligned_vwrite(buf, addr, n);
2107 spin_unlock(&vmap_area_lock);
2114 * remap_vmalloc_range - map vmalloc pages to userspace
2115 * @vma: vma to cover (map full range of vma)
2116 * @addr: vmalloc memory
2117 * @pgoff: number of pages into addr before first page to map
2119 * Returns: 0 for success, -Exxx on failure
2121 * This function checks that addr is a valid vmalloc'ed area, and
2122 * that it is big enough to cover the vma. Will return failure if
2123 * that criteria isn't met.
2125 * Similar to remap_pfn_range() (see mm/memory.c)
2127 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2128 unsigned long pgoff)
2130 struct vm_struct *area;
2131 unsigned long uaddr = vma->vm_start;
2132 unsigned long usize = vma->vm_end - vma->vm_start;
2134 if ((PAGE_SIZE-1) & (unsigned long)addr)
2137 area = find_vm_area(addr);
2141 if (!(area->flags & VM_USERMAP))
2144 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
2147 addr += pgoff << PAGE_SHIFT;
2149 struct page *page = vmalloc_to_page(addr);
2152 ret = vm_insert_page(vma, uaddr, page);
2159 } while (usize > 0);
2161 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2165 EXPORT_SYMBOL(remap_vmalloc_range);
2168 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2171 void __attribute__((weak)) vmalloc_sync_all(void)
2176 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2188 * alloc_vm_area - allocate a range of kernel address space
2189 * @size: size of the area
2190 * @ptes: returns the PTEs for the address space
2192 * Returns: NULL on failure, vm_struct on success
2194 * This function reserves a range of kernel address space, and
2195 * allocates pagetables to map that range. No actual mappings
2198 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2199 * allocated for the VM area are returned.
2201 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2203 struct vm_struct *area;
2205 area = get_vm_area_caller(size, VM_IOREMAP,
2206 __builtin_return_address(0));
2211 * This ensures that page tables are constructed for this region
2212 * of kernel virtual address space and mapped into init_mm.
2214 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2215 size, f, ptes ? &ptes : NULL)) {
2222 EXPORT_SYMBOL_GPL(alloc_vm_area);
2224 void free_vm_area(struct vm_struct *area)
2226 struct vm_struct *ret;
2227 ret = remove_vm_area(area->addr);
2228 BUG_ON(ret != area);
2231 EXPORT_SYMBOL_GPL(free_vm_area);
2234 static struct vmap_area *node_to_va(struct rb_node *n)
2236 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2240 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2241 * @end: target address
2242 * @pnext: out arg for the next vmap_area
2243 * @pprev: out arg for the previous vmap_area
2245 * Returns: %true if either or both of next and prev are found,
2246 * %false if no vmap_area exists
2248 * Find vmap_areas end addresses of which enclose @end. ie. if not
2249 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2251 static bool pvm_find_next_prev(unsigned long end,
2252 struct vmap_area **pnext,
2253 struct vmap_area **pprev)
2255 struct rb_node *n = vmap_area_root.rb_node;
2256 struct vmap_area *va = NULL;
2259 va = rb_entry(n, struct vmap_area, rb_node);
2260 if (end < va->va_end)
2262 else if (end > va->va_end)
2271 if (va->va_end > end) {
2273 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2276 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2282 * pvm_determine_end - find the highest aligned address between two vmap_areas
2283 * @pnext: in/out arg for the next vmap_area
2284 * @pprev: in/out arg for the previous vmap_area
2287 * Returns: determined end address
2289 * Find the highest aligned address between *@pnext and *@pprev below
2290 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2291 * down address is between the end addresses of the two vmap_areas.
2293 * Please note that the address returned by this function may fall
2294 * inside *@pnext vmap_area. The caller is responsible for checking
2297 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2298 struct vmap_area **pprev,
2299 unsigned long align)
2301 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2305 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2309 while (*pprev && (*pprev)->va_end > addr) {
2311 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2318 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2319 * @offsets: array containing offset of each area
2320 * @sizes: array containing size of each area
2321 * @nr_vms: the number of areas to allocate
2322 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2324 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2325 * vm_structs on success, %NULL on failure
2327 * Percpu allocator wants to use congruent vm areas so that it can
2328 * maintain the offsets among percpu areas. This function allocates
2329 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2330 * be scattered pretty far, distance between two areas easily going up
2331 * to gigabytes. To avoid interacting with regular vmallocs, these
2332 * areas are allocated from top.
2334 * Despite its complicated look, this allocator is rather simple. It
2335 * does everything top-down and scans areas from the end looking for
2336 * matching slot. While scanning, if any of the areas overlaps with
2337 * existing vmap_area, the base address is pulled down to fit the
2338 * area. Scanning is repeated till all the areas fit and then all
2339 * necessary data structres are inserted and the result is returned.
2341 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2342 const size_t *sizes, int nr_vms,
2345 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2346 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2347 struct vmap_area **vas, *prev, *next;
2348 struct vm_struct **vms;
2349 int area, area2, last_area, term_area;
2350 unsigned long base, start, end, last_end;
2351 bool purged = false;
2353 /* verify parameters and allocate data structures */
2354 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2355 for (last_area = 0, area = 0; area < nr_vms; area++) {
2356 start = offsets[area];
2357 end = start + sizes[area];
2359 /* is everything aligned properly? */
2360 BUG_ON(!IS_ALIGNED(offsets[area], align));
2361 BUG_ON(!IS_ALIGNED(sizes[area], align));
2363 /* detect the area with the highest address */
2364 if (start > offsets[last_area])
2367 for (area2 = 0; area2 < nr_vms; area2++) {
2368 unsigned long start2 = offsets[area2];
2369 unsigned long end2 = start2 + sizes[area2];
2374 BUG_ON(start2 >= start && start2 < end);
2375 BUG_ON(end2 <= end && end2 > start);
2378 last_end = offsets[last_area] + sizes[last_area];
2380 if (vmalloc_end - vmalloc_start < last_end) {
2385 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2386 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2390 for (area = 0; area < nr_vms; area++) {
2391 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2392 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2393 if (!vas[area] || !vms[area])
2397 spin_lock(&vmap_area_lock);
2399 /* start scanning - we scan from the top, begin with the last area */
2400 area = term_area = last_area;
2401 start = offsets[area];
2402 end = start + sizes[area];
2404 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2405 base = vmalloc_end - last_end;
2408 base = pvm_determine_end(&next, &prev, align) - end;
2411 BUG_ON(next && next->va_end <= base + end);
2412 BUG_ON(prev && prev->va_end > base + end);
2415 * base might have underflowed, add last_end before
2418 if (base + last_end < vmalloc_start + last_end) {
2419 spin_unlock(&vmap_area_lock);
2421 purge_vmap_area_lazy();
2429 * If next overlaps, move base downwards so that it's
2430 * right below next and then recheck.
2432 if (next && next->va_start < base + end) {
2433 base = pvm_determine_end(&next, &prev, align) - end;
2439 * If prev overlaps, shift down next and prev and move
2440 * base so that it's right below new next and then
2443 if (prev && prev->va_end > base + start) {
2445 prev = node_to_va(rb_prev(&next->rb_node));
2446 base = pvm_determine_end(&next, &prev, align) - end;
2452 * This area fits, move on to the previous one. If
2453 * the previous one is the terminal one, we're done.
2455 area = (area + nr_vms - 1) % nr_vms;
2456 if (area == term_area)
2458 start = offsets[area];
2459 end = start + sizes[area];
2460 pvm_find_next_prev(base + end, &next, &prev);
2463 /* we've found a fitting base, insert all va's */
2464 for (area = 0; area < nr_vms; area++) {
2465 struct vmap_area *va = vas[area];
2467 va->va_start = base + offsets[area];
2468 va->va_end = va->va_start + sizes[area];
2469 __insert_vmap_area(va);
2472 vmap_area_pcpu_hole = base + offsets[last_area];
2474 spin_unlock(&vmap_area_lock);
2476 /* insert all vm's */
2477 for (area = 0; area < nr_vms; area++)
2478 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2485 for (area = 0; area < nr_vms; area++) {
2496 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2497 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2498 * @nr_vms: the number of allocated areas
2500 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2502 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2506 for (i = 0; i < nr_vms; i++)
2507 free_vm_area(vms[i]);
2510 #endif /* CONFIG_SMP */
2512 #ifdef CONFIG_PROC_FS
2513 static void *s_start(struct seq_file *m, loff_t *pos)
2514 __acquires(&vmap_area_lock)
2517 struct vmap_area *va;
2519 spin_lock(&vmap_area_lock);
2520 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2521 while (n > 0 && &va->list != &vmap_area_list) {
2523 va = list_entry(va->list.next, typeof(*va), list);
2525 if (!n && &va->list != &vmap_area_list)
2532 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2534 struct vmap_area *va = p, *next;
2537 next = list_entry(va->list.next, typeof(*va), list);
2538 if (&next->list != &vmap_area_list)
2544 static void s_stop(struct seq_file *m, void *p)
2545 __releases(&vmap_area_lock)
2547 spin_unlock(&vmap_area_lock);
2550 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2552 if (IS_ENABLED(CONFIG_NUMA)) {
2553 unsigned int nr, *counters = m->private;
2558 /* Pair with smp_wmb() in clear_vm_unlist() */
2560 if (v->flags & VM_UNLIST)
2563 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2565 for (nr = 0; nr < v->nr_pages; nr++)
2566 counters[page_to_nid(v->pages[nr])]++;
2568 for_each_node_state(nr, N_HIGH_MEMORY)
2570 seq_printf(m, " N%u=%u", nr, counters[nr]);
2574 static int s_show(struct seq_file *m, void *p)
2576 struct vmap_area *va = p;
2577 struct vm_struct *v;
2579 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2582 if (!(va->flags & VM_VM_AREA)) {
2583 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2584 (void *)va->va_start, (void *)va->va_end,
2585 va->va_end - va->va_start);
2591 seq_printf(m, "0x%pK-0x%pK %7ld",
2592 v->addr, v->addr + v->size, v->size);
2595 seq_printf(m, " %pS", v->caller);
2598 seq_printf(m, " pages=%d", v->nr_pages);
2601 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2603 if (v->flags & VM_IOREMAP)
2604 seq_printf(m, " ioremap");
2606 if (v->flags & VM_ALLOC)
2607 seq_printf(m, " vmalloc");
2609 if (v->flags & VM_MAP)
2610 seq_printf(m, " vmap");
2612 if (v->flags & VM_USERMAP)
2613 seq_printf(m, " user");
2615 if (v->flags & VM_VPAGES)
2616 seq_printf(m, " vpages");
2618 show_numa_info(m, v);
2623 static const struct seq_operations vmalloc_op = {
2630 static int vmalloc_open(struct inode *inode, struct file *file)
2632 unsigned int *ptr = NULL;
2635 if (IS_ENABLED(CONFIG_NUMA)) {
2636 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2640 ret = seq_open(file, &vmalloc_op);
2642 struct seq_file *m = file->private_data;
2649 static const struct file_operations proc_vmalloc_operations = {
2650 .open = vmalloc_open,
2652 .llseek = seq_lseek,
2653 .release = seq_release_private,
2656 static int __init proc_vmalloc_init(void)
2658 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2661 module_init(proc_vmalloc_init);
2663 void get_vmalloc_info(struct vmalloc_info *vmi)
2665 struct vmap_area *va;
2666 unsigned long free_area_size;
2667 unsigned long prev_end;
2670 vmi->largest_chunk = 0;
2672 prev_end = VMALLOC_START;
2674 spin_lock(&vmap_area_lock);
2676 if (list_empty(&vmap_area_list)) {
2677 vmi->largest_chunk = VMALLOC_TOTAL;
2681 list_for_each_entry(va, &vmap_area_list, list) {
2682 unsigned long addr = va->va_start;
2685 * Some archs keep another range for modules in vmalloc space
2687 if (addr < VMALLOC_START)
2689 if (addr >= VMALLOC_END)
2692 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2695 vmi->used += (va->va_end - va->va_start);
2697 free_area_size = addr - prev_end;
2698 if (vmi->largest_chunk < free_area_size)
2699 vmi->largest_chunk = free_area_size;
2701 prev_end = va->va_end;
2704 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2705 vmi->largest_chunk = VMALLOC_END - prev_end;
2708 spin_unlock(&vmap_area_lock);