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/compiler.h>
31 #include <linux/llist.h>
32 #include <linux/bitops.h>
34 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
36 #include <asm/shmparam.h>
38 struct vfree_deferred {
39 struct llist_head list;
40 struct work_struct wq;
42 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
44 static void __vunmap(const void *, int);
46 static void free_work(struct work_struct *w)
48 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
49 struct llist_node *llnode = llist_del_all(&p->list);
52 llnode = llist_next(llnode);
57 /*** Page table manipulation functions ***/
59 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
63 pte = pte_offset_kernel(pmd, addr);
65 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
66 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
67 } while (pte++, addr += PAGE_SIZE, addr != end);
70 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
75 pmd = pmd_offset(pud, addr);
77 next = pmd_addr_end(addr, end);
78 if (pmd_clear_huge(pmd))
80 if (pmd_none_or_clear_bad(pmd))
82 vunmap_pte_range(pmd, addr, next);
83 } while (pmd++, addr = next, addr != end);
86 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
91 pud = pud_offset(pgd, addr);
93 next = pud_addr_end(addr, end);
94 if (pud_clear_huge(pud))
96 if (pud_none_or_clear_bad(pud))
98 vunmap_pmd_range(pud, addr, next);
99 } while (pud++, addr = next, addr != end);
102 static void vunmap_page_range(unsigned long addr, unsigned long end)
108 pgd = pgd_offset_k(addr);
110 next = pgd_addr_end(addr, end);
111 if (pgd_none_or_clear_bad(pgd))
113 vunmap_pud_range(pgd, addr, next);
114 } while (pgd++, addr = next, addr != end);
117 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
118 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
123 * nr is a running index into the array which helps higher level
124 * callers keep track of where we're up to.
127 pte = pte_alloc_kernel(pmd, addr);
131 struct page *page = pages[*nr];
133 if (WARN_ON(!pte_none(*pte)))
137 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
139 } while (pte++, addr += PAGE_SIZE, addr != end);
143 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
144 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
149 pmd = pmd_alloc(&init_mm, pud, addr);
153 next = pmd_addr_end(addr, end);
154 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
156 } while (pmd++, addr = next, addr != end);
160 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
161 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
166 pud = pud_alloc(&init_mm, pgd, addr);
170 next = pud_addr_end(addr, end);
171 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
173 } while (pud++, addr = next, addr != end);
178 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
179 * will have pfns corresponding to the "pages" array.
181 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
183 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
184 pgprot_t prot, struct page **pages)
188 unsigned long addr = start;
193 pgd = pgd_offset_k(addr);
195 next = pgd_addr_end(addr, end);
196 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
199 } while (pgd++, addr = next, addr != end);
204 static int vmap_page_range(unsigned long start, unsigned long end,
205 pgprot_t prot, struct page **pages)
209 ret = vmap_page_range_noflush(start, end, prot, pages);
210 flush_cache_vmap(start, end);
214 int is_vmalloc_or_module_addr(const void *x)
217 * ARM, x86-64 and sparc64 put modules in a special place,
218 * and fall back on vmalloc() if that fails. Others
219 * just put it in the vmalloc space.
221 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
222 unsigned long addr = (unsigned long)x;
223 if (addr >= MODULES_VADDR && addr < MODULES_END)
226 return is_vmalloc_addr(x);
230 * Walk a vmap address to the struct page it maps.
232 struct page *vmalloc_to_page(const void *vmalloc_addr)
234 unsigned long addr = (unsigned long) vmalloc_addr;
235 struct page *page = NULL;
236 pgd_t *pgd = pgd_offset_k(addr);
239 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
240 * architectures that do not vmalloc module space
242 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
244 if (!pgd_none(*pgd)) {
245 pud_t *pud = pud_offset(pgd, addr);
246 if (!pud_none(*pud)) {
247 pmd_t *pmd = pmd_offset(pud, addr);
248 if (!pmd_none(*pmd)) {
251 ptep = pte_offset_map(pmd, addr);
253 if (pte_present(pte))
254 page = pte_page(pte);
261 EXPORT_SYMBOL(vmalloc_to_page);
264 * Map a vmalloc()-space virtual address to the physical page frame number.
266 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
268 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
270 EXPORT_SYMBOL(vmalloc_to_pfn);
273 /*** Global kva allocator ***/
275 #define VM_LAZY_FREE 0x01
276 #define VM_LAZY_FREEING 0x02
277 #define VM_VM_AREA 0x04
279 static DEFINE_SPINLOCK(vmap_area_lock);
280 /* Export for kexec only */
281 LIST_HEAD(vmap_area_list);
282 static struct rb_root vmap_area_root = RB_ROOT;
284 /* The vmap cache globals are protected by vmap_area_lock */
285 static struct rb_node *free_vmap_cache;
286 static unsigned long cached_hole_size;
287 static unsigned long cached_vstart;
288 static unsigned long cached_align;
290 static unsigned long vmap_area_pcpu_hole;
292 static struct vmap_area *__find_vmap_area(unsigned long addr)
294 struct rb_node *n = vmap_area_root.rb_node;
297 struct vmap_area *va;
299 va = rb_entry(n, struct vmap_area, rb_node);
300 if (addr < va->va_start)
302 else if (addr >= va->va_end)
311 static void __insert_vmap_area(struct vmap_area *va)
313 struct rb_node **p = &vmap_area_root.rb_node;
314 struct rb_node *parent = NULL;
318 struct vmap_area *tmp_va;
321 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
322 if (va->va_start < tmp_va->va_end)
324 else if (va->va_end > tmp_va->va_start)
330 rb_link_node(&va->rb_node, parent, p);
331 rb_insert_color(&va->rb_node, &vmap_area_root);
333 /* address-sort this list */
334 tmp = rb_prev(&va->rb_node);
336 struct vmap_area *prev;
337 prev = rb_entry(tmp, struct vmap_area, rb_node);
338 list_add_rcu(&va->list, &prev->list);
340 list_add_rcu(&va->list, &vmap_area_list);
343 static void purge_vmap_area_lazy(void);
346 * Allocate a region of KVA of the specified size and alignment, within the
349 static struct vmap_area *alloc_vmap_area(unsigned long size,
351 unsigned long vstart, unsigned long vend,
352 int node, gfp_t gfp_mask)
354 struct vmap_area *va;
358 struct vmap_area *first;
361 BUG_ON(size & ~PAGE_MASK);
362 BUG_ON(!is_power_of_2(align));
364 va = kmalloc_node(sizeof(struct vmap_area),
365 gfp_mask & GFP_RECLAIM_MASK, node);
367 return ERR_PTR(-ENOMEM);
370 * Only scan the relevant parts containing pointers to other objects
371 * to avoid false negatives.
373 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
376 spin_lock(&vmap_area_lock);
378 * Invalidate cache if we have more permissive parameters.
379 * cached_hole_size notes the largest hole noticed _below_
380 * the vmap_area cached in free_vmap_cache: if size fits
381 * into that hole, we want to scan from vstart to reuse
382 * the hole instead of allocating above free_vmap_cache.
383 * Note that __free_vmap_area may update free_vmap_cache
384 * without updating cached_hole_size or cached_align.
386 if (!free_vmap_cache ||
387 size < cached_hole_size ||
388 vstart < cached_vstart ||
389 align < cached_align) {
391 cached_hole_size = 0;
392 free_vmap_cache = NULL;
394 /* record if we encounter less permissive parameters */
395 cached_vstart = vstart;
396 cached_align = align;
398 /* find starting point for our search */
399 if (free_vmap_cache) {
400 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
401 addr = ALIGN(first->va_end, align);
404 if (addr + size < addr)
408 addr = ALIGN(vstart, align);
409 if (addr + size < addr)
412 n = vmap_area_root.rb_node;
416 struct vmap_area *tmp;
417 tmp = rb_entry(n, struct vmap_area, rb_node);
418 if (tmp->va_end >= addr) {
420 if (tmp->va_start <= addr)
431 /* from the starting point, walk areas until a suitable hole is found */
432 while (addr + size > first->va_start && addr + size <= vend) {
433 if (addr + cached_hole_size < first->va_start)
434 cached_hole_size = first->va_start - addr;
435 addr = ALIGN(first->va_end, align);
436 if (addr + size < addr)
439 if (list_is_last(&first->list, &vmap_area_list))
442 first = list_entry(first->list.next,
443 struct vmap_area, list);
447 if (addr + size > vend)
451 va->va_end = addr + size;
453 __insert_vmap_area(va);
454 free_vmap_cache = &va->rb_node;
455 spin_unlock(&vmap_area_lock);
457 BUG_ON(va->va_start & (align-1));
458 BUG_ON(va->va_start < vstart);
459 BUG_ON(va->va_end > vend);
464 spin_unlock(&vmap_area_lock);
466 purge_vmap_area_lazy();
470 if (printk_ratelimit())
471 pr_warn("vmap allocation for size %lu failed: "
472 "use vmalloc=<size> to increase size.\n", size);
474 return ERR_PTR(-EBUSY);
477 static void __free_vmap_area(struct vmap_area *va)
479 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
481 if (free_vmap_cache) {
482 if (va->va_end < cached_vstart) {
483 free_vmap_cache = NULL;
485 struct vmap_area *cache;
486 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
487 if (va->va_start <= cache->va_start) {
488 free_vmap_cache = rb_prev(&va->rb_node);
490 * We don't try to update cached_hole_size or
491 * cached_align, but it won't go very wrong.
496 rb_erase(&va->rb_node, &vmap_area_root);
497 RB_CLEAR_NODE(&va->rb_node);
498 list_del_rcu(&va->list);
501 * Track the highest possible candidate for pcpu area
502 * allocation. Areas outside of vmalloc area can be returned
503 * here too, consider only end addresses which fall inside
504 * vmalloc area proper.
506 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
507 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
509 kfree_rcu(va, rcu_head);
513 * Free a region of KVA allocated by alloc_vmap_area
515 static void free_vmap_area(struct vmap_area *va)
517 spin_lock(&vmap_area_lock);
518 __free_vmap_area(va);
519 spin_unlock(&vmap_area_lock);
523 * Clear the pagetable entries of a given vmap_area
525 static void unmap_vmap_area(struct vmap_area *va)
527 vunmap_page_range(va->va_start, va->va_end);
530 static void vmap_debug_free_range(unsigned long start, unsigned long end)
533 * Unmap page tables and force a TLB flush immediately if
534 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
535 * bugs similarly to those in linear kernel virtual address
536 * space after a page has been freed.
538 * All the lazy freeing logic is still retained, in order to
539 * minimise intrusiveness of this debugging feature.
541 * This is going to be *slow* (linear kernel virtual address
542 * debugging doesn't do a broadcast TLB flush so it is a lot
545 #ifdef CONFIG_DEBUG_PAGEALLOC
546 vunmap_page_range(start, end);
547 flush_tlb_kernel_range(start, end);
552 * lazy_max_pages is the maximum amount of virtual address space we gather up
553 * before attempting to purge with a TLB flush.
555 * There is a tradeoff here: a larger number will cover more kernel page tables
556 * and take slightly longer to purge, but it will linearly reduce the number of
557 * global TLB flushes that must be performed. It would seem natural to scale
558 * this number up linearly with the number of CPUs (because vmapping activity
559 * could also scale linearly with the number of CPUs), however it is likely
560 * that in practice, workloads might be constrained in other ways that mean
561 * vmap activity will not scale linearly with CPUs. Also, I want to be
562 * conservative and not introduce a big latency on huge systems, so go with
563 * a less aggressive log scale. It will still be an improvement over the old
564 * code, and it will be simple to change the scale factor if we find that it
565 * becomes a problem on bigger systems.
567 static unsigned long lazy_max_pages(void)
571 log = fls(num_online_cpus());
573 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
576 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
578 /* for per-CPU blocks */
579 static void purge_fragmented_blocks_allcpus(void);
582 * called before a call to iounmap() if the caller wants vm_area_struct's
585 void set_iounmap_nonlazy(void)
587 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
591 * Purges all lazily-freed vmap areas.
593 * If sync is 0 then don't purge if there is already a purge in progress.
594 * If force_flush is 1, then flush kernel TLBs between *start and *end even
595 * if we found no lazy vmap areas to unmap (callers can use this to optimise
596 * their own TLB flushing).
597 * Returns with *start = min(*start, lowest purged address)
598 * *end = max(*end, highest purged address)
600 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
601 int sync, int force_flush)
603 static DEFINE_SPINLOCK(purge_lock);
605 struct vmap_area *va;
606 struct vmap_area *n_va;
610 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
611 * should not expect such behaviour. This just simplifies locking for
612 * the case that isn't actually used at the moment anyway.
614 if (!sync && !force_flush) {
615 if (!spin_trylock(&purge_lock))
618 spin_lock(&purge_lock);
621 purge_fragmented_blocks_allcpus();
624 list_for_each_entry_rcu(va, &vmap_area_list, list) {
625 if (va->flags & VM_LAZY_FREE) {
626 if (va->va_start < *start)
627 *start = va->va_start;
628 if (va->va_end > *end)
630 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
631 list_add_tail(&va->purge_list, &valist);
632 va->flags |= VM_LAZY_FREEING;
633 va->flags &= ~VM_LAZY_FREE;
639 atomic_sub(nr, &vmap_lazy_nr);
641 if (nr || force_flush)
642 flush_tlb_kernel_range(*start, *end);
645 spin_lock(&vmap_area_lock);
646 list_for_each_entry_safe(va, n_va, &valist, purge_list)
647 __free_vmap_area(va);
648 spin_unlock(&vmap_area_lock);
650 spin_unlock(&purge_lock);
654 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
655 * is already purging.
657 static void try_purge_vmap_area_lazy(void)
659 unsigned long start = ULONG_MAX, end = 0;
661 __purge_vmap_area_lazy(&start, &end, 0, 0);
665 * Kick off a purge of the outstanding lazy areas.
667 static void purge_vmap_area_lazy(void)
669 unsigned long start = ULONG_MAX, end = 0;
671 __purge_vmap_area_lazy(&start, &end, 1, 0);
675 * Free a vmap area, caller ensuring that the area has been unmapped
676 * and flush_cache_vunmap had been called for the correct range
679 static void free_vmap_area_noflush(struct vmap_area *va)
681 va->flags |= VM_LAZY_FREE;
682 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
683 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
684 try_purge_vmap_area_lazy();
688 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
689 * called for the correct range previously.
691 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
694 free_vmap_area_noflush(va);
698 * Free and unmap a vmap area
700 static void free_unmap_vmap_area(struct vmap_area *va)
702 flush_cache_vunmap(va->va_start, va->va_end);
703 free_unmap_vmap_area_noflush(va);
706 static struct vmap_area *find_vmap_area(unsigned long addr)
708 struct vmap_area *va;
710 spin_lock(&vmap_area_lock);
711 va = __find_vmap_area(addr);
712 spin_unlock(&vmap_area_lock);
717 static void free_unmap_vmap_area_addr(unsigned long addr)
719 struct vmap_area *va;
721 va = find_vmap_area(addr);
723 free_unmap_vmap_area(va);
727 /*** Per cpu kva allocator ***/
730 * vmap space is limited especially on 32 bit architectures. Ensure there is
731 * room for at least 16 percpu vmap blocks per CPU.
734 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
735 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
736 * instead (we just need a rough idea)
738 #if BITS_PER_LONG == 32
739 #define VMALLOC_SPACE (128UL*1024*1024)
741 #define VMALLOC_SPACE (128UL*1024*1024*1024)
744 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
745 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
746 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
747 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
748 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
749 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
750 #define VMAP_BBMAP_BITS \
751 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
752 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
753 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
755 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
757 static bool vmap_initialized __read_mostly = false;
759 struct vmap_block_queue {
761 struct list_head free;
766 struct vmap_area *va;
767 unsigned long free, dirty;
768 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
769 struct list_head free_list;
770 struct rcu_head rcu_head;
771 struct list_head purge;
774 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
775 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
778 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
779 * in the free path. Could get rid of this if we change the API to return a
780 * "cookie" from alloc, to be passed to free. But no big deal yet.
782 static DEFINE_SPINLOCK(vmap_block_tree_lock);
783 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
786 * We should probably have a fallback mechanism to allocate virtual memory
787 * out of partially filled vmap blocks. However vmap block sizing should be
788 * fairly reasonable according to the vmalloc size, so it shouldn't be a
792 static unsigned long addr_to_vb_idx(unsigned long addr)
794 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
795 addr /= VMAP_BLOCK_SIZE;
799 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
803 addr = va_start + (pages_off << PAGE_SHIFT);
804 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
809 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
810 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
811 * @order: how many 2^order pages should be occupied in newly allocated block
812 * @gfp_mask: flags for the page level allocator
814 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
816 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
818 struct vmap_block_queue *vbq;
819 struct vmap_block *vb;
820 struct vmap_area *va;
821 unsigned long vb_idx;
825 node = numa_node_id();
827 vb = kmalloc_node(sizeof(struct vmap_block),
828 gfp_mask & GFP_RECLAIM_MASK, node);
830 return ERR_PTR(-ENOMEM);
832 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
833 VMALLOC_START, VMALLOC_END,
840 err = radix_tree_preload(gfp_mask);
847 vaddr = vmap_block_vaddr(va->va_start, 0);
848 spin_lock_init(&vb->lock);
850 /* At least something should be left free */
851 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
852 vb->free = VMAP_BBMAP_BITS - (1UL << order);
854 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
855 INIT_LIST_HEAD(&vb->free_list);
857 vb_idx = addr_to_vb_idx(va->va_start);
858 spin_lock(&vmap_block_tree_lock);
859 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
860 spin_unlock(&vmap_block_tree_lock);
862 radix_tree_preload_end();
864 vbq = &get_cpu_var(vmap_block_queue);
865 spin_lock(&vbq->lock);
866 list_add_tail_rcu(&vb->free_list, &vbq->free);
867 spin_unlock(&vbq->lock);
868 put_cpu_var(vmap_block_queue);
873 static void free_vmap_block(struct vmap_block *vb)
875 struct vmap_block *tmp;
876 unsigned long vb_idx;
878 vb_idx = addr_to_vb_idx(vb->va->va_start);
879 spin_lock(&vmap_block_tree_lock);
880 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
881 spin_unlock(&vmap_block_tree_lock);
884 free_vmap_area_noflush(vb->va);
885 kfree_rcu(vb, rcu_head);
888 static void purge_fragmented_blocks(int cpu)
891 struct vmap_block *vb;
892 struct vmap_block *n_vb;
893 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
896 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
898 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
901 spin_lock(&vb->lock);
902 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
903 vb->free = 0; /* prevent further allocs after releasing lock */
904 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
905 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
906 spin_lock(&vbq->lock);
907 list_del_rcu(&vb->free_list);
908 spin_unlock(&vbq->lock);
909 spin_unlock(&vb->lock);
910 list_add_tail(&vb->purge, &purge);
912 spin_unlock(&vb->lock);
916 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
917 list_del(&vb->purge);
922 static void purge_fragmented_blocks_allcpus(void)
926 for_each_possible_cpu(cpu)
927 purge_fragmented_blocks(cpu);
930 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
932 struct vmap_block_queue *vbq;
933 struct vmap_block *vb;
937 BUG_ON(size & ~PAGE_MASK);
938 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
939 if (WARN_ON(size == 0)) {
941 * Allocating 0 bytes isn't what caller wants since
942 * get_order(0) returns funny result. Just warn and terminate
947 order = get_order(size);
950 vbq = &get_cpu_var(vmap_block_queue);
951 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
952 unsigned long pages_off;
954 spin_lock(&vb->lock);
955 if (vb->free < (1UL << order)) {
956 spin_unlock(&vb->lock);
960 pages_off = VMAP_BBMAP_BITS - vb->free;
961 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
962 vb->free -= 1UL << order;
964 spin_lock(&vbq->lock);
965 list_del_rcu(&vb->free_list);
966 spin_unlock(&vbq->lock);
969 spin_unlock(&vb->lock);
973 put_cpu_var(vmap_block_queue);
976 /* Allocate new block if nothing was found */
978 vaddr = new_vmap_block(order, gfp_mask);
983 static void vb_free(const void *addr, unsigned long size)
985 unsigned long offset;
986 unsigned long vb_idx;
988 struct vmap_block *vb;
990 BUG_ON(size & ~PAGE_MASK);
991 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
993 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
995 order = get_order(size);
997 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
999 vb_idx = addr_to_vb_idx((unsigned long)addr);
1001 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1005 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1007 spin_lock(&vb->lock);
1008 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
1010 vb->dirty += 1UL << order;
1011 if (vb->dirty == VMAP_BBMAP_BITS) {
1013 spin_unlock(&vb->lock);
1014 free_vmap_block(vb);
1016 spin_unlock(&vb->lock);
1020 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1022 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1023 * to amortize TLB flushing overheads. What this means is that any page you
1024 * have now, may, in a former life, have been mapped into kernel virtual
1025 * address by the vmap layer and so there might be some CPUs with TLB entries
1026 * still referencing that page (additional to the regular 1:1 kernel mapping).
1028 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1029 * be sure that none of the pages we have control over will have any aliases
1030 * from the vmap layer.
1032 void vm_unmap_aliases(void)
1034 unsigned long start = ULONG_MAX, end = 0;
1038 if (unlikely(!vmap_initialized))
1041 for_each_possible_cpu(cpu) {
1042 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1043 struct vmap_block *vb;
1046 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1049 spin_lock(&vb->lock);
1050 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1051 if (i < VMAP_BBMAP_BITS) {
1054 j = find_last_bit(vb->dirty_map,
1056 j = j + 1; /* need exclusive index */
1058 s = vb->va->va_start + (i << PAGE_SHIFT);
1059 e = vb->va->va_start + (j << PAGE_SHIFT);
1067 spin_unlock(&vb->lock);
1072 __purge_vmap_area_lazy(&start, &end, 1, flush);
1074 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1077 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1078 * @mem: the pointer returned by vm_map_ram
1079 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1081 void vm_unmap_ram(const void *mem, unsigned int count)
1083 unsigned long size = count << PAGE_SHIFT;
1084 unsigned long addr = (unsigned long)mem;
1087 BUG_ON(addr < VMALLOC_START);
1088 BUG_ON(addr > VMALLOC_END);
1089 BUG_ON(addr & (PAGE_SIZE-1));
1091 debug_check_no_locks_freed(mem, size);
1092 vmap_debug_free_range(addr, addr+size);
1094 if (likely(count <= VMAP_MAX_ALLOC))
1097 free_unmap_vmap_area_addr(addr);
1099 EXPORT_SYMBOL(vm_unmap_ram);
1102 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1103 * @pages: an array of pointers to the pages to be mapped
1104 * @count: number of pages
1105 * @node: prefer to allocate data structures on this node
1106 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1108 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1109 * faster than vmap so it's good. But if you mix long-life and short-life
1110 * objects with vm_map_ram(), it could consume lots of address space through
1111 * fragmentation (especially on a 32bit machine). You could see failures in
1112 * the end. Please use this function for short-lived objects.
1114 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1116 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1118 unsigned long size = count << PAGE_SHIFT;
1122 if (likely(count <= VMAP_MAX_ALLOC)) {
1123 mem = vb_alloc(size, GFP_KERNEL);
1126 addr = (unsigned long)mem;
1128 struct vmap_area *va;
1129 va = alloc_vmap_area(size, PAGE_SIZE,
1130 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1134 addr = va->va_start;
1137 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1138 vm_unmap_ram(mem, count);
1143 EXPORT_SYMBOL(vm_map_ram);
1145 static struct vm_struct *vmlist __initdata;
1147 * vm_area_add_early - add vmap area early during boot
1148 * @vm: vm_struct to add
1150 * This function is used to add fixed kernel vm area to vmlist before
1151 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1152 * should contain proper values and the other fields should be zero.
1154 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1156 void __init vm_area_add_early(struct vm_struct *vm)
1158 struct vm_struct *tmp, **p;
1160 BUG_ON(vmap_initialized);
1161 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1162 if (tmp->addr >= vm->addr) {
1163 BUG_ON(tmp->addr < vm->addr + vm->size);
1166 BUG_ON(tmp->addr + tmp->size > vm->addr);
1173 * vm_area_register_early - register vmap area early during boot
1174 * @vm: vm_struct to register
1175 * @align: requested alignment
1177 * This function is used to register kernel vm area before
1178 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1179 * proper values on entry and other fields should be zero. On return,
1180 * vm->addr contains the allocated address.
1182 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1184 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1186 static size_t vm_init_off __initdata;
1189 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1190 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1192 vm->addr = (void *)addr;
1194 vm_area_add_early(vm);
1197 void __init vmalloc_init(void)
1199 struct vmap_area *va;
1200 struct vm_struct *tmp;
1203 for_each_possible_cpu(i) {
1204 struct vmap_block_queue *vbq;
1205 struct vfree_deferred *p;
1207 vbq = &per_cpu(vmap_block_queue, i);
1208 spin_lock_init(&vbq->lock);
1209 INIT_LIST_HEAD(&vbq->free);
1210 p = &per_cpu(vfree_deferred, i);
1211 init_llist_head(&p->list);
1212 INIT_WORK(&p->wq, free_work);
1215 /* Import existing vmlist entries. */
1216 for (tmp = vmlist; tmp; tmp = tmp->next) {
1217 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1218 va->flags = VM_VM_AREA;
1219 va->va_start = (unsigned long)tmp->addr;
1220 va->va_end = va->va_start + tmp->size;
1222 __insert_vmap_area(va);
1225 vmap_area_pcpu_hole = VMALLOC_END;
1227 vmap_initialized = true;
1231 * map_kernel_range_noflush - map kernel VM area with the specified pages
1232 * @addr: start of the VM area to map
1233 * @size: size of the VM area to map
1234 * @prot: page protection flags to use
1235 * @pages: pages to map
1237 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1238 * specify should have been allocated using get_vm_area() and its
1242 * This function does NOT do any cache flushing. The caller is
1243 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1244 * before calling this function.
1247 * The number of pages mapped on success, -errno on failure.
1249 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1250 pgprot_t prot, struct page **pages)
1252 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1256 * unmap_kernel_range_noflush - unmap kernel VM area
1257 * @addr: start of the VM area to unmap
1258 * @size: size of the VM area to unmap
1260 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1261 * specify should have been allocated using get_vm_area() and its
1265 * This function does NOT do any cache flushing. The caller is
1266 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1267 * before calling this function and flush_tlb_kernel_range() after.
1269 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1271 vunmap_page_range(addr, addr + size);
1273 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1276 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1277 * @addr: start of the VM area to unmap
1278 * @size: size of the VM area to unmap
1280 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1281 * the unmapping and tlb after.
1283 void unmap_kernel_range(unsigned long addr, unsigned long size)
1285 unsigned long end = addr + size;
1287 flush_cache_vunmap(addr, end);
1288 vunmap_page_range(addr, end);
1289 flush_tlb_kernel_range(addr, end);
1291 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1293 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1295 unsigned long addr = (unsigned long)area->addr;
1296 unsigned long end = addr + get_vm_area_size(area);
1299 err = vmap_page_range(addr, end, prot, pages);
1301 return err > 0 ? 0 : err;
1303 EXPORT_SYMBOL_GPL(map_vm_area);
1305 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1306 unsigned long flags, const void *caller)
1308 spin_lock(&vmap_area_lock);
1310 vm->addr = (void *)va->va_start;
1311 vm->size = va->va_end - va->va_start;
1312 vm->caller = caller;
1314 va->flags |= VM_VM_AREA;
1315 spin_unlock(&vmap_area_lock);
1318 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1321 * Before removing VM_UNINITIALIZED,
1322 * we should make sure that vm has proper values.
1323 * Pair with smp_rmb() in show_numa_info().
1326 vm->flags &= ~VM_UNINITIALIZED;
1329 static struct vm_struct *__get_vm_area_node(unsigned long size,
1330 unsigned long align, unsigned long flags, unsigned long start,
1331 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1333 struct vmap_area *va;
1334 struct vm_struct *area;
1336 BUG_ON(in_interrupt());
1337 if (flags & VM_IOREMAP)
1338 align = 1ul << clamp_t(int, fls_long(size),
1339 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1341 size = PAGE_ALIGN(size);
1342 if (unlikely(!size))
1345 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1346 if (unlikely(!area))
1349 if (!(flags & VM_NO_GUARD))
1352 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1358 setup_vmalloc_vm(area, va, flags, caller);
1363 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1364 unsigned long start, unsigned long end)
1366 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1367 GFP_KERNEL, __builtin_return_address(0));
1369 EXPORT_SYMBOL_GPL(__get_vm_area);
1371 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1372 unsigned long start, unsigned long end,
1375 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1376 GFP_KERNEL, caller);
1380 * get_vm_area - reserve a contiguous kernel virtual area
1381 * @size: size of the area
1382 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1384 * Search an area of @size in the kernel virtual mapping area,
1385 * and reserved it for out purposes. Returns the area descriptor
1386 * on success or %NULL on failure.
1388 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1390 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1391 NUMA_NO_NODE, GFP_KERNEL,
1392 __builtin_return_address(0));
1395 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1398 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1399 NUMA_NO_NODE, GFP_KERNEL, caller);
1403 * find_vm_area - find a continuous kernel virtual area
1404 * @addr: base address
1406 * Search for the kernel VM area starting at @addr, and return it.
1407 * It is up to the caller to do all required locking to keep the returned
1410 struct vm_struct *find_vm_area(const void *addr)
1412 struct vmap_area *va;
1414 va = find_vmap_area((unsigned long)addr);
1415 if (va && va->flags & VM_VM_AREA)
1422 * remove_vm_area - find and remove a continuous kernel virtual area
1423 * @addr: base address
1425 * Search for the kernel VM area starting at @addr, and remove it.
1426 * This function returns the found VM area, but using it is NOT safe
1427 * on SMP machines, except for its size or flags.
1429 struct vm_struct *remove_vm_area(const void *addr)
1431 struct vmap_area *va;
1433 va = find_vmap_area((unsigned long)addr);
1434 if (va && va->flags & VM_VM_AREA) {
1435 struct vm_struct *vm = va->vm;
1437 spin_lock(&vmap_area_lock);
1439 va->flags &= ~VM_VM_AREA;
1440 spin_unlock(&vmap_area_lock);
1442 vmap_debug_free_range(va->va_start, va->va_end);
1443 kasan_free_shadow(vm);
1444 free_unmap_vmap_area(va);
1445 vm->size -= PAGE_SIZE;
1452 static void __vunmap(const void *addr, int deallocate_pages)
1454 struct vm_struct *area;
1459 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1463 area = remove_vm_area(addr);
1464 if (unlikely(!area)) {
1465 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1470 debug_check_no_locks_freed(addr, area->size);
1471 debug_check_no_obj_freed(addr, area->size);
1473 if (deallocate_pages) {
1476 for (i = 0; i < area->nr_pages; i++) {
1477 struct page *page = area->pages[i];
1483 if (area->flags & VM_VPAGES)
1494 * vfree - release memory allocated by vmalloc()
1495 * @addr: memory base address
1497 * Free the virtually continuous memory area starting at @addr, as
1498 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1499 * NULL, no operation is performed.
1501 * Must not be called in NMI context (strictly speaking, only if we don't
1502 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1503 * conventions for vfree() arch-depenedent would be a really bad idea)
1505 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1507 void vfree(const void *addr)
1511 kmemleak_free(addr);
1515 if (unlikely(in_interrupt())) {
1516 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1517 if (llist_add((struct llist_node *)addr, &p->list))
1518 schedule_work(&p->wq);
1522 EXPORT_SYMBOL(vfree);
1525 * vunmap - release virtual mapping obtained by vmap()
1526 * @addr: memory base address
1528 * Free the virtually contiguous memory area starting at @addr,
1529 * which was created from the page array passed to vmap().
1531 * Must not be called in interrupt context.
1533 void vunmap(const void *addr)
1535 BUG_ON(in_interrupt());
1540 EXPORT_SYMBOL(vunmap);
1543 * vmap - map an array of pages into virtually contiguous space
1544 * @pages: array of page pointers
1545 * @count: number of pages to map
1546 * @flags: vm_area->flags
1547 * @prot: page protection for the mapping
1549 * Maps @count pages from @pages into contiguous kernel virtual
1552 void *vmap(struct page **pages, unsigned int count,
1553 unsigned long flags, pgprot_t prot)
1555 struct vm_struct *area;
1559 if (count > totalram_pages)
1562 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1563 __builtin_return_address(0));
1567 if (map_vm_area(area, prot, pages)) {
1574 EXPORT_SYMBOL(vmap);
1576 static void *__vmalloc_node(unsigned long size, unsigned long align,
1577 gfp_t gfp_mask, pgprot_t prot,
1578 int node, const void *caller);
1579 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1580 pgprot_t prot, int node)
1582 const int order = 0;
1583 struct page **pages;
1584 unsigned int nr_pages, array_size, i;
1585 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1586 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1588 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1589 array_size = (nr_pages * sizeof(struct page *));
1591 area->nr_pages = nr_pages;
1592 /* Please note that the recursion is strictly bounded. */
1593 if (array_size > PAGE_SIZE) {
1594 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1595 PAGE_KERNEL, node, area->caller);
1596 area->flags |= VM_VPAGES;
1598 pages = kmalloc_node(array_size, nested_gfp, node);
1600 area->pages = pages;
1602 remove_vm_area(area->addr);
1607 for (i = 0; i < area->nr_pages; i++) {
1610 if (node == NUMA_NO_NODE)
1611 page = alloc_page(alloc_mask);
1613 page = alloc_pages_node(node, alloc_mask, order);
1615 if (unlikely(!page)) {
1616 /* Successfully allocated i pages, free them in __vunmap() */
1620 area->pages[i] = page;
1621 if (gfp_mask & __GFP_WAIT)
1625 if (map_vm_area(area, prot, pages))
1630 warn_alloc_failed(gfp_mask, order,
1631 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1632 (area->nr_pages*PAGE_SIZE), area->size);
1638 * __vmalloc_node_range - allocate virtually contiguous memory
1639 * @size: allocation size
1640 * @align: desired alignment
1641 * @start: vm area range start
1642 * @end: vm area range end
1643 * @gfp_mask: flags for the page level allocator
1644 * @prot: protection mask for the allocated pages
1645 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1646 * @node: node to use for allocation or NUMA_NO_NODE
1647 * @caller: caller's return address
1649 * Allocate enough pages to cover @size from the page level
1650 * allocator with @gfp_mask flags. Map them into contiguous
1651 * kernel virtual space, using a pagetable protection of @prot.
1653 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1654 unsigned long start, unsigned long end, gfp_t gfp_mask,
1655 pgprot_t prot, unsigned long vm_flags, int node,
1658 struct vm_struct *area;
1660 unsigned long real_size = size;
1662 size = PAGE_ALIGN(size);
1663 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1666 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1667 vm_flags, start, end, node, gfp_mask, caller);
1671 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1676 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1677 * flag. It means that vm_struct is not fully initialized.
1678 * Now, it is fully initialized, so remove this flag here.
1680 clear_vm_uninitialized_flag(area);
1683 * A ref_count = 2 is needed because vm_struct allocated in
1684 * __get_vm_area_node() contains a reference to the virtual address of
1685 * the vmalloc'ed block.
1687 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1692 warn_alloc_failed(gfp_mask, 0,
1693 "vmalloc: allocation failure: %lu bytes\n",
1699 * __vmalloc_node - allocate virtually contiguous memory
1700 * @size: allocation size
1701 * @align: desired alignment
1702 * @gfp_mask: flags for the page level allocator
1703 * @prot: protection mask for the allocated pages
1704 * @node: node to use for allocation or NUMA_NO_NODE
1705 * @caller: caller's return address
1707 * Allocate enough pages to cover @size from the page level
1708 * allocator with @gfp_mask flags. Map them into contiguous
1709 * kernel virtual space, using a pagetable protection of @prot.
1711 static void *__vmalloc_node(unsigned long size, unsigned long align,
1712 gfp_t gfp_mask, pgprot_t prot,
1713 int node, const void *caller)
1715 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1716 gfp_mask, prot, 0, node, caller);
1719 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1721 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1722 __builtin_return_address(0));
1724 EXPORT_SYMBOL(__vmalloc);
1726 static inline void *__vmalloc_node_flags(unsigned long size,
1727 int node, gfp_t flags)
1729 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1730 node, __builtin_return_address(0));
1734 * vmalloc - allocate virtually contiguous memory
1735 * @size: allocation size
1736 * Allocate enough pages to cover @size from the page level
1737 * allocator and map them into contiguous kernel virtual space.
1739 * For tight control over page level allocator and protection flags
1740 * use __vmalloc() instead.
1742 void *vmalloc(unsigned long size)
1744 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1745 GFP_KERNEL | __GFP_HIGHMEM);
1747 EXPORT_SYMBOL(vmalloc);
1750 * vzalloc - allocate virtually contiguous memory with zero fill
1751 * @size: allocation size
1752 * Allocate enough pages to cover @size from the page level
1753 * allocator and map them into contiguous kernel virtual space.
1754 * The memory allocated is set to zero.
1756 * For tight control over page level allocator and protection flags
1757 * use __vmalloc() instead.
1759 void *vzalloc(unsigned long size)
1761 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1762 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1764 EXPORT_SYMBOL(vzalloc);
1767 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1768 * @size: allocation size
1770 * The resulting memory area is zeroed so it can be mapped to userspace
1771 * without leaking data.
1773 void *vmalloc_user(unsigned long size)
1775 struct vm_struct *area;
1778 ret = __vmalloc_node(size, SHMLBA,
1779 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1780 PAGE_KERNEL, NUMA_NO_NODE,
1781 __builtin_return_address(0));
1783 area = find_vm_area(ret);
1784 area->flags |= VM_USERMAP;
1788 EXPORT_SYMBOL(vmalloc_user);
1791 * vmalloc_node - allocate memory on a specific node
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.
1798 * For tight control over page level allocator and protection flags
1799 * use __vmalloc() instead.
1801 void *vmalloc_node(unsigned long size, int node)
1803 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1804 node, __builtin_return_address(0));
1806 EXPORT_SYMBOL(vmalloc_node);
1809 * vzalloc_node - allocate memory on a specific node with zero fill
1810 * @size: allocation size
1813 * Allocate enough pages to cover @size from the page level
1814 * allocator and map them into contiguous kernel virtual space.
1815 * The memory allocated is set to zero.
1817 * For tight control over page level allocator and protection flags
1818 * use __vmalloc_node() instead.
1820 void *vzalloc_node(unsigned long size, int node)
1822 return __vmalloc_node_flags(size, node,
1823 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1825 EXPORT_SYMBOL(vzalloc_node);
1827 #ifndef PAGE_KERNEL_EXEC
1828 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1832 * vmalloc_exec - allocate virtually contiguous, executable memory
1833 * @size: allocation size
1835 * Kernel-internal function to allocate enough pages to cover @size
1836 * the page level allocator and map them into contiguous and
1837 * executable kernel virtual space.
1839 * For tight control over page level allocator and protection flags
1840 * use __vmalloc() instead.
1843 void *vmalloc_exec(unsigned long size)
1845 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1846 NUMA_NO_NODE, __builtin_return_address(0));
1849 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1850 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1851 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1852 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1854 #define GFP_VMALLOC32 GFP_KERNEL
1858 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1859 * @size: allocation size
1861 * Allocate enough 32bit PA addressable pages to cover @size from the
1862 * page level allocator and map them into contiguous kernel virtual space.
1864 void *vmalloc_32(unsigned long size)
1866 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1867 NUMA_NO_NODE, __builtin_return_address(0));
1869 EXPORT_SYMBOL(vmalloc_32);
1872 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1873 * @size: allocation size
1875 * The resulting memory area is 32bit addressable and zeroed so it can be
1876 * mapped to userspace without leaking data.
1878 void *vmalloc_32_user(unsigned long size)
1880 struct vm_struct *area;
1883 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1884 NUMA_NO_NODE, __builtin_return_address(0));
1886 area = find_vm_area(ret);
1887 area->flags |= VM_USERMAP;
1891 EXPORT_SYMBOL(vmalloc_32_user);
1894 * small helper routine , copy contents to buf from addr.
1895 * If the page is not present, fill zero.
1898 static int aligned_vread(char *buf, char *addr, unsigned long count)
1904 unsigned long offset, length;
1906 offset = (unsigned long)addr & ~PAGE_MASK;
1907 length = PAGE_SIZE - offset;
1910 p = vmalloc_to_page(addr);
1912 * To do safe access to this _mapped_ area, we need
1913 * lock. But adding lock here means that we need to add
1914 * overhead of vmalloc()/vfree() calles for this _debug_
1915 * interface, rarely used. Instead of that, we'll use
1916 * kmap() and get small overhead in this access function.
1920 * we can expect USER0 is not used (see vread/vwrite's
1921 * function description)
1923 void *map = kmap_atomic(p);
1924 memcpy(buf, map + offset, length);
1927 memset(buf, 0, length);
1937 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1943 unsigned long offset, length;
1945 offset = (unsigned long)addr & ~PAGE_MASK;
1946 length = PAGE_SIZE - offset;
1949 p = vmalloc_to_page(addr);
1951 * To do safe access to this _mapped_ area, we need
1952 * lock. But adding lock here means that we need to add
1953 * overhead of vmalloc()/vfree() calles for this _debug_
1954 * interface, rarely used. Instead of that, we'll use
1955 * kmap() and get small overhead in this access function.
1959 * we can expect USER0 is not used (see vread/vwrite's
1960 * function description)
1962 void *map = kmap_atomic(p);
1963 memcpy(map + offset, buf, length);
1975 * vread() - read vmalloc area in a safe way.
1976 * @buf: buffer for reading data
1977 * @addr: vm address.
1978 * @count: number of bytes to be read.
1980 * Returns # of bytes which addr and buf should be increased.
1981 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1982 * includes any intersect with alive vmalloc area.
1984 * This function checks that addr is a valid vmalloc'ed area, and
1985 * copy data from that area to a given buffer. If the given memory range
1986 * of [addr...addr+count) includes some valid address, data is copied to
1987 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1988 * IOREMAP area is treated as memory hole and no copy is done.
1990 * If [addr...addr+count) doesn't includes any intersects with alive
1991 * vm_struct area, returns 0. @buf should be kernel's buffer.
1993 * Note: In usual ops, vread() is never necessary because the caller
1994 * should know vmalloc() area is valid and can use memcpy().
1995 * This is for routines which have to access vmalloc area without
1996 * any informaion, as /dev/kmem.
2000 long vread(char *buf, char *addr, unsigned long count)
2002 struct vmap_area *va;
2003 struct vm_struct *vm;
2004 char *vaddr, *buf_start = buf;
2005 unsigned long buflen = count;
2008 /* Don't allow overflow */
2009 if ((unsigned long) addr + count < count)
2010 count = -(unsigned long) addr;
2012 spin_lock(&vmap_area_lock);
2013 list_for_each_entry(va, &vmap_area_list, list) {
2017 if (!(va->flags & VM_VM_AREA))
2021 vaddr = (char *) vm->addr;
2022 if (addr >= vaddr + get_vm_area_size(vm))
2024 while (addr < vaddr) {
2032 n = vaddr + get_vm_area_size(vm) - addr;
2035 if (!(vm->flags & VM_IOREMAP))
2036 aligned_vread(buf, addr, n);
2037 else /* IOREMAP area is treated as memory hole */
2044 spin_unlock(&vmap_area_lock);
2046 if (buf == buf_start)
2048 /* zero-fill memory holes */
2049 if (buf != buf_start + buflen)
2050 memset(buf, 0, buflen - (buf - buf_start));
2056 * vwrite() - write vmalloc area in a safe way.
2057 * @buf: buffer for source data
2058 * @addr: vm address.
2059 * @count: number of bytes to be read.
2061 * Returns # of bytes which addr and buf should be incresed.
2062 * (same number to @count).
2063 * If [addr...addr+count) doesn't includes any intersect with valid
2064 * vmalloc area, returns 0.
2066 * This function checks that addr is a valid vmalloc'ed area, and
2067 * copy data from a buffer to the given addr. If specified range of
2068 * [addr...addr+count) includes some valid address, data is copied from
2069 * proper area of @buf. If there are memory holes, no copy to hole.
2070 * IOREMAP area is treated as memory hole and no copy is done.
2072 * If [addr...addr+count) doesn't includes any intersects with alive
2073 * vm_struct area, returns 0. @buf should be kernel's buffer.
2075 * Note: In usual ops, vwrite() is never necessary because the caller
2076 * should know vmalloc() area is valid and can use memcpy().
2077 * This is for routines which have to access vmalloc area without
2078 * any informaion, as /dev/kmem.
2081 long vwrite(char *buf, char *addr, unsigned long count)
2083 struct vmap_area *va;
2084 struct vm_struct *vm;
2086 unsigned long n, buflen;
2089 /* Don't allow overflow */
2090 if ((unsigned long) addr + count < count)
2091 count = -(unsigned long) addr;
2094 spin_lock(&vmap_area_lock);
2095 list_for_each_entry(va, &vmap_area_list, list) {
2099 if (!(va->flags & VM_VM_AREA))
2103 vaddr = (char *) vm->addr;
2104 if (addr >= vaddr + get_vm_area_size(vm))
2106 while (addr < vaddr) {
2113 n = vaddr + get_vm_area_size(vm) - addr;
2116 if (!(vm->flags & VM_IOREMAP)) {
2117 aligned_vwrite(buf, addr, n);
2125 spin_unlock(&vmap_area_lock);
2132 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2133 * @vma: vma to cover
2134 * @uaddr: target user address to start at
2135 * @kaddr: virtual address of vmalloc kernel memory
2136 * @size: size of map area
2138 * Returns: 0 for success, -Exxx on failure
2140 * This function checks that @kaddr is a valid vmalloc'ed area,
2141 * and that it is big enough to cover the range starting at
2142 * @uaddr in @vma. Will return failure if that criteria isn't
2145 * Similar to remap_pfn_range() (see mm/memory.c)
2147 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2148 void *kaddr, unsigned long size)
2150 struct vm_struct *area;
2152 size = PAGE_ALIGN(size);
2154 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2157 area = find_vm_area(kaddr);
2161 if (!(area->flags & VM_USERMAP))
2164 if (kaddr + size > area->addr + area->size)
2168 struct page *page = vmalloc_to_page(kaddr);
2171 ret = vm_insert_page(vma, uaddr, page);
2180 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2184 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2187 * remap_vmalloc_range - map vmalloc pages to userspace
2188 * @vma: vma to cover (map full range of vma)
2189 * @addr: vmalloc memory
2190 * @pgoff: number of pages into addr before first page to map
2192 * Returns: 0 for success, -Exxx on failure
2194 * This function checks that addr is a valid vmalloc'ed area, and
2195 * that it is big enough to cover the vma. Will return failure if
2196 * that criteria isn't met.
2198 * Similar to remap_pfn_range() (see mm/memory.c)
2200 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2201 unsigned long pgoff)
2203 return remap_vmalloc_range_partial(vma, vma->vm_start,
2204 addr + (pgoff << PAGE_SHIFT),
2205 vma->vm_end - vma->vm_start);
2207 EXPORT_SYMBOL(remap_vmalloc_range);
2210 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2213 void __weak vmalloc_sync_all(void)
2218 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2230 * alloc_vm_area - allocate a range of kernel address space
2231 * @size: size of the area
2232 * @ptes: returns the PTEs for the address space
2234 * Returns: NULL on failure, vm_struct on success
2236 * This function reserves a range of kernel address space, and
2237 * allocates pagetables to map that range. No actual mappings
2240 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2241 * allocated for the VM area are returned.
2243 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2245 struct vm_struct *area;
2247 area = get_vm_area_caller(size, VM_IOREMAP,
2248 __builtin_return_address(0));
2253 * This ensures that page tables are constructed for this region
2254 * of kernel virtual address space and mapped into init_mm.
2256 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2257 size, f, ptes ? &ptes : NULL)) {
2264 EXPORT_SYMBOL_GPL(alloc_vm_area);
2266 void free_vm_area(struct vm_struct *area)
2268 struct vm_struct *ret;
2269 ret = remove_vm_area(area->addr);
2270 BUG_ON(ret != area);
2273 EXPORT_SYMBOL_GPL(free_vm_area);
2276 static struct vmap_area *node_to_va(struct rb_node *n)
2278 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2282 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2283 * @end: target address
2284 * @pnext: out arg for the next vmap_area
2285 * @pprev: out arg for the previous vmap_area
2287 * Returns: %true if either or both of next and prev are found,
2288 * %false if no vmap_area exists
2290 * Find vmap_areas end addresses of which enclose @end. ie. if not
2291 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2293 static bool pvm_find_next_prev(unsigned long end,
2294 struct vmap_area **pnext,
2295 struct vmap_area **pprev)
2297 struct rb_node *n = vmap_area_root.rb_node;
2298 struct vmap_area *va = NULL;
2301 va = rb_entry(n, struct vmap_area, rb_node);
2302 if (end < va->va_end)
2304 else if (end > va->va_end)
2313 if (va->va_end > end) {
2315 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2318 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2324 * pvm_determine_end - find the highest aligned address between two vmap_areas
2325 * @pnext: in/out arg for the next vmap_area
2326 * @pprev: in/out arg for the previous vmap_area
2329 * Returns: determined end address
2331 * Find the highest aligned address between *@pnext and *@pprev below
2332 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2333 * down address is between the end addresses of the two vmap_areas.
2335 * Please note that the address returned by this function may fall
2336 * inside *@pnext vmap_area. The caller is responsible for checking
2339 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2340 struct vmap_area **pprev,
2341 unsigned long align)
2343 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2347 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2351 while (*pprev && (*pprev)->va_end > addr) {
2353 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2360 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2361 * @offsets: array containing offset of each area
2362 * @sizes: array containing size of each area
2363 * @nr_vms: the number of areas to allocate
2364 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2366 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2367 * vm_structs on success, %NULL on failure
2369 * Percpu allocator wants to use congruent vm areas so that it can
2370 * maintain the offsets among percpu areas. This function allocates
2371 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2372 * be scattered pretty far, distance between two areas easily going up
2373 * to gigabytes. To avoid interacting with regular vmallocs, these
2374 * areas are allocated from top.
2376 * Despite its complicated look, this allocator is rather simple. It
2377 * does everything top-down and scans areas from the end looking for
2378 * matching slot. While scanning, if any of the areas overlaps with
2379 * existing vmap_area, the base address is pulled down to fit the
2380 * area. Scanning is repeated till all the areas fit and then all
2381 * necessary data structres are inserted and the result is returned.
2383 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2384 const size_t *sizes, int nr_vms,
2387 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2388 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2389 struct vmap_area **vas, *prev, *next;
2390 struct vm_struct **vms;
2391 int area, area2, last_area, term_area;
2392 unsigned long base, start, end, last_end;
2393 bool purged = false;
2395 /* verify parameters and allocate data structures */
2396 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2397 for (last_area = 0, area = 0; area < nr_vms; area++) {
2398 start = offsets[area];
2399 end = start + sizes[area];
2401 /* is everything aligned properly? */
2402 BUG_ON(!IS_ALIGNED(offsets[area], align));
2403 BUG_ON(!IS_ALIGNED(sizes[area], align));
2405 /* detect the area with the highest address */
2406 if (start > offsets[last_area])
2409 for (area2 = 0; area2 < nr_vms; area2++) {
2410 unsigned long start2 = offsets[area2];
2411 unsigned long end2 = start2 + sizes[area2];
2416 BUG_ON(start2 >= start && start2 < end);
2417 BUG_ON(end2 <= end && end2 > start);
2420 last_end = offsets[last_area] + sizes[last_area];
2422 if (vmalloc_end - vmalloc_start < last_end) {
2427 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2428 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2432 for (area = 0; area < nr_vms; area++) {
2433 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2434 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2435 if (!vas[area] || !vms[area])
2439 spin_lock(&vmap_area_lock);
2441 /* start scanning - we scan from the top, begin with the last area */
2442 area = term_area = last_area;
2443 start = offsets[area];
2444 end = start + sizes[area];
2446 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2447 base = vmalloc_end - last_end;
2450 base = pvm_determine_end(&next, &prev, align) - end;
2453 BUG_ON(next && next->va_end <= base + end);
2454 BUG_ON(prev && prev->va_end > base + end);
2457 * base might have underflowed, add last_end before
2460 if (base + last_end < vmalloc_start + last_end) {
2461 spin_unlock(&vmap_area_lock);
2463 purge_vmap_area_lazy();
2471 * If next overlaps, move base downwards so that it's
2472 * right below next and then recheck.
2474 if (next && next->va_start < base + end) {
2475 base = pvm_determine_end(&next, &prev, align) - end;
2481 * If prev overlaps, shift down next and prev and move
2482 * base so that it's right below new next and then
2485 if (prev && prev->va_end > base + start) {
2487 prev = node_to_va(rb_prev(&next->rb_node));
2488 base = pvm_determine_end(&next, &prev, align) - end;
2494 * This area fits, move on to the previous one. If
2495 * the previous one is the terminal one, we're done.
2497 area = (area + nr_vms - 1) % nr_vms;
2498 if (area == term_area)
2500 start = offsets[area];
2501 end = start + sizes[area];
2502 pvm_find_next_prev(base + end, &next, &prev);
2505 /* we've found a fitting base, insert all va's */
2506 for (area = 0; area < nr_vms; area++) {
2507 struct vmap_area *va = vas[area];
2509 va->va_start = base + offsets[area];
2510 va->va_end = va->va_start + sizes[area];
2511 __insert_vmap_area(va);
2514 vmap_area_pcpu_hole = base + offsets[last_area];
2516 spin_unlock(&vmap_area_lock);
2518 /* insert all vm's */
2519 for (area = 0; area < nr_vms; area++)
2520 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2527 for (area = 0; area < nr_vms; area++) {
2538 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2539 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2540 * @nr_vms: the number of allocated areas
2542 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2544 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2548 for (i = 0; i < nr_vms; i++)
2549 free_vm_area(vms[i]);
2552 #endif /* CONFIG_SMP */
2554 #ifdef CONFIG_PROC_FS
2555 static void *s_start(struct seq_file *m, loff_t *pos)
2556 __acquires(&vmap_area_lock)
2559 struct vmap_area *va;
2561 spin_lock(&vmap_area_lock);
2562 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2563 while (n > 0 && &va->list != &vmap_area_list) {
2565 va = list_entry(va->list.next, typeof(*va), list);
2567 if (!n && &va->list != &vmap_area_list)
2574 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2576 struct vmap_area *va = p, *next;
2579 next = list_entry(va->list.next, typeof(*va), list);
2580 if (&next->list != &vmap_area_list)
2586 static void s_stop(struct seq_file *m, void *p)
2587 __releases(&vmap_area_lock)
2589 spin_unlock(&vmap_area_lock);
2592 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2594 if (IS_ENABLED(CONFIG_NUMA)) {
2595 unsigned int nr, *counters = m->private;
2600 if (v->flags & VM_UNINITIALIZED)
2602 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2605 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2607 for (nr = 0; nr < v->nr_pages; nr++)
2608 counters[page_to_nid(v->pages[nr])]++;
2610 for_each_node_state(nr, N_HIGH_MEMORY)
2612 seq_printf(m, " N%u=%u", nr, counters[nr]);
2616 static int s_show(struct seq_file *m, void *p)
2618 struct vmap_area *va = p;
2619 struct vm_struct *v;
2622 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2623 * behalf of vmap area is being tear down or vm_map_ram allocation.
2625 if (!(va->flags & VM_VM_AREA))
2630 seq_printf(m, "0x%pK-0x%pK %7ld",
2631 v->addr, v->addr + v->size, v->size);
2634 seq_printf(m, " %pS", v->caller);
2637 seq_printf(m, " pages=%d", v->nr_pages);
2640 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2642 if (v->flags & VM_IOREMAP)
2643 seq_puts(m, " ioremap");
2645 if (v->flags & VM_ALLOC)
2646 seq_puts(m, " vmalloc");
2648 if (v->flags & VM_MAP)
2649 seq_puts(m, " vmap");
2651 if (v->flags & VM_USERMAP)
2652 seq_puts(m, " user");
2654 if (v->flags & VM_VPAGES)
2655 seq_puts(m, " vpages");
2657 show_numa_info(m, v);
2662 static const struct seq_operations vmalloc_op = {
2669 static int vmalloc_open(struct inode *inode, struct file *file)
2671 if (IS_ENABLED(CONFIG_NUMA))
2672 return seq_open_private(file, &vmalloc_op,
2673 nr_node_ids * sizeof(unsigned int));
2675 return seq_open(file, &vmalloc_op);
2678 static const struct file_operations proc_vmalloc_operations = {
2679 .open = vmalloc_open,
2681 .llseek = seq_lseek,
2682 .release = seq_release_private,
2685 static int __init proc_vmalloc_init(void)
2687 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2690 module_init(proc_vmalloc_init);
2692 void get_vmalloc_info(struct vmalloc_info *vmi)
2694 struct vmap_area *va;
2695 unsigned long free_area_size;
2696 unsigned long prev_end;
2699 vmi->largest_chunk = 0;
2701 prev_end = VMALLOC_START;
2705 if (list_empty(&vmap_area_list)) {
2706 vmi->largest_chunk = VMALLOC_TOTAL;
2710 list_for_each_entry_rcu(va, &vmap_area_list, list) {
2711 unsigned long addr = va->va_start;
2714 * Some archs keep another range for modules in vmalloc space
2716 if (addr < VMALLOC_START)
2718 if (addr >= VMALLOC_END)
2721 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2724 vmi->used += (va->va_end - va->va_start);
2726 free_area_size = addr - prev_end;
2727 if (vmi->largest_chunk < free_area_size)
2728 vmi->largest_chunk = free_area_size;
2730 prev_end = va->va_end;
2733 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2734 vmi->largest_chunk = VMALLOC_END - prev_end;