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>
40 struct vfree_deferred {
41 struct llist_head list;
42 struct work_struct wq;
44 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
46 static void __vunmap(const void *, int);
48 static void free_work(struct work_struct *w)
50 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
51 struct llist_node *llnode = llist_del_all(&p->list);
54 llnode = llist_next(llnode);
59 /*** Page table manipulation functions ***/
61 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
65 pte = pte_offset_kernel(pmd, addr);
67 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
68 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
69 } while (pte++, addr += PAGE_SIZE, addr != end);
72 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
77 pmd = pmd_offset(pud, addr);
79 next = pmd_addr_end(addr, end);
80 if (pmd_clear_huge(pmd))
82 if (pmd_none_or_clear_bad(pmd))
84 vunmap_pte_range(pmd, addr, next);
85 } while (pmd++, addr = next, addr != end);
88 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
93 pud = pud_offset(pgd, addr);
95 next = pud_addr_end(addr, end);
96 if (pud_clear_huge(pud))
98 if (pud_none_or_clear_bad(pud))
100 vunmap_pmd_range(pud, addr, next);
101 } while (pud++, addr = next, addr != end);
104 static void vunmap_page_range(unsigned long addr, unsigned long end)
110 pgd = pgd_offset_k(addr);
112 next = pgd_addr_end(addr, end);
113 if (pgd_none_or_clear_bad(pgd))
115 vunmap_pud_range(pgd, addr, next);
116 } while (pgd++, addr = next, addr != end);
119 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
120 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
125 * nr is a running index into the array which helps higher level
126 * callers keep track of where we're up to.
129 pte = pte_alloc_kernel(pmd, addr);
133 struct page *page = pages[*nr];
135 if (WARN_ON(!pte_none(*pte)))
139 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
141 } while (pte++, addr += PAGE_SIZE, addr != end);
145 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
146 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
151 pmd = pmd_alloc(&init_mm, pud, addr);
155 next = pmd_addr_end(addr, end);
156 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
158 } while (pmd++, addr = next, addr != end);
162 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
163 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
168 pud = pud_alloc(&init_mm, pgd, addr);
172 next = pud_addr_end(addr, end);
173 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
175 } while (pud++, addr = next, addr != end);
180 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
181 * will have pfns corresponding to the "pages" array.
183 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
185 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
186 pgprot_t prot, struct page **pages)
190 unsigned long addr = start;
195 pgd = pgd_offset_k(addr);
197 next = pgd_addr_end(addr, end);
198 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
201 } while (pgd++, addr = next, addr != end);
206 static int vmap_page_range(unsigned long start, unsigned long end,
207 pgprot_t prot, struct page **pages)
211 ret = vmap_page_range_noflush(start, end, prot, pages);
212 flush_cache_vmap(start, end);
216 int is_vmalloc_or_module_addr(const void *x)
219 * ARM, x86-64 and sparc64 put modules in a special place,
220 * and fall back on vmalloc() if that fails. Others
221 * just put it in the vmalloc space.
223 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
224 unsigned long addr = (unsigned long)x;
225 if (addr >= MODULES_VADDR && addr < MODULES_END)
228 return is_vmalloc_addr(x);
232 * Walk a vmap address to the struct page it maps.
234 struct page *vmalloc_to_page(const void *vmalloc_addr)
236 unsigned long addr = (unsigned long) vmalloc_addr;
237 struct page *page = NULL;
238 pgd_t *pgd = pgd_offset_k(addr);
241 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
242 * architectures that do not vmalloc module space
244 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
246 if (!pgd_none(*pgd)) {
247 pud_t *pud = pud_offset(pgd, addr);
248 if (!pud_none(*pud)) {
249 pmd_t *pmd = pmd_offset(pud, addr);
250 if (!pmd_none(*pmd)) {
253 ptep = pte_offset_map(pmd, addr);
255 if (pte_present(pte))
256 page = pte_page(pte);
263 EXPORT_SYMBOL(vmalloc_to_page);
266 * Map a vmalloc()-space virtual address to the physical page frame number.
268 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
270 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
272 EXPORT_SYMBOL(vmalloc_to_pfn);
275 /*** Global kva allocator ***/
277 #define VM_LAZY_FREE 0x01
278 #define VM_LAZY_FREEING 0x02
279 #define VM_VM_AREA 0x04
281 static DEFINE_SPINLOCK(vmap_area_lock);
282 /* Export for kexec only */
283 LIST_HEAD(vmap_area_list);
284 static struct rb_root vmap_area_root = RB_ROOT;
286 /* The vmap cache globals are protected by vmap_area_lock */
287 static struct rb_node *free_vmap_cache;
288 static unsigned long cached_hole_size;
289 static unsigned long cached_vstart;
290 static unsigned long cached_align;
292 static unsigned long vmap_area_pcpu_hole;
294 static struct vmap_area *__find_vmap_area(unsigned long addr)
296 struct rb_node *n = vmap_area_root.rb_node;
299 struct vmap_area *va;
301 va = rb_entry(n, struct vmap_area, rb_node);
302 if (addr < va->va_start)
304 else if (addr >= va->va_end)
313 static void __insert_vmap_area(struct vmap_area *va)
315 struct rb_node **p = &vmap_area_root.rb_node;
316 struct rb_node *parent = NULL;
320 struct vmap_area *tmp_va;
323 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
324 if (va->va_start < tmp_va->va_end)
326 else if (va->va_end > tmp_va->va_start)
332 rb_link_node(&va->rb_node, parent, p);
333 rb_insert_color(&va->rb_node, &vmap_area_root);
335 /* address-sort this list */
336 tmp = rb_prev(&va->rb_node);
338 struct vmap_area *prev;
339 prev = rb_entry(tmp, struct vmap_area, rb_node);
340 list_add_rcu(&va->list, &prev->list);
342 list_add_rcu(&va->list, &vmap_area_list);
345 static void purge_vmap_area_lazy(void);
348 * Allocate a region of KVA of the specified size and alignment, within the
351 static struct vmap_area *alloc_vmap_area(unsigned long size,
353 unsigned long vstart, unsigned long vend,
354 int node, gfp_t gfp_mask)
356 struct vmap_area *va;
360 struct vmap_area *first;
363 BUG_ON(offset_in_page(size));
364 BUG_ON(!is_power_of_2(align));
366 va = kmalloc_node(sizeof(struct vmap_area),
367 gfp_mask & GFP_RECLAIM_MASK, node);
369 return ERR_PTR(-ENOMEM);
372 * Only scan the relevant parts containing pointers to other objects
373 * to avoid false negatives.
375 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
378 spin_lock(&vmap_area_lock);
380 * Invalidate cache if we have more permissive parameters.
381 * cached_hole_size notes the largest hole noticed _below_
382 * the vmap_area cached in free_vmap_cache: if size fits
383 * into that hole, we want to scan from vstart to reuse
384 * the hole instead of allocating above free_vmap_cache.
385 * Note that __free_vmap_area may update free_vmap_cache
386 * without updating cached_hole_size or cached_align.
388 if (!free_vmap_cache ||
389 size < cached_hole_size ||
390 vstart < cached_vstart ||
391 align < cached_align) {
393 cached_hole_size = 0;
394 free_vmap_cache = NULL;
396 /* record if we encounter less permissive parameters */
397 cached_vstart = vstart;
398 cached_align = align;
400 /* find starting point for our search */
401 if (free_vmap_cache) {
402 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
403 addr = ALIGN(first->va_end, align);
406 if (addr + size < addr)
410 addr = ALIGN(vstart, align);
411 if (addr + size < addr)
414 n = vmap_area_root.rb_node;
418 struct vmap_area *tmp;
419 tmp = rb_entry(n, struct vmap_area, rb_node);
420 if (tmp->va_end >= addr) {
422 if (tmp->va_start <= addr)
433 /* from the starting point, walk areas until a suitable hole is found */
434 while (addr + size > first->va_start && addr + size <= vend) {
435 if (addr + cached_hole_size < first->va_start)
436 cached_hole_size = first->va_start - addr;
437 addr = ALIGN(first->va_end, align);
438 if (addr + size < addr)
441 if (list_is_last(&first->list, &vmap_area_list))
444 first = list_entry(first->list.next,
445 struct vmap_area, list);
450 * Check also calculated address against the vstart,
451 * because it can be 0 because of big align request.
453 if (addr + size > vend || addr < vstart)
457 va->va_end = addr + size;
459 __insert_vmap_area(va);
460 free_vmap_cache = &va->rb_node;
461 spin_unlock(&vmap_area_lock);
463 BUG_ON(va->va_start & (align-1));
464 BUG_ON(va->va_start < vstart);
465 BUG_ON(va->va_end > vend);
470 spin_unlock(&vmap_area_lock);
472 purge_vmap_area_lazy();
476 if (printk_ratelimit())
477 pr_warn("vmap allocation for size %lu failed: "
478 "use vmalloc=<size> to increase size.\n", size);
480 return ERR_PTR(-EBUSY);
483 static void __free_vmap_area(struct vmap_area *va)
485 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
487 if (free_vmap_cache) {
488 if (va->va_end < cached_vstart) {
489 free_vmap_cache = NULL;
491 struct vmap_area *cache;
492 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
493 if (va->va_start <= cache->va_start) {
494 free_vmap_cache = rb_prev(&va->rb_node);
496 * We don't try to update cached_hole_size or
497 * cached_align, but it won't go very wrong.
502 rb_erase(&va->rb_node, &vmap_area_root);
503 RB_CLEAR_NODE(&va->rb_node);
504 list_del_rcu(&va->list);
507 * Track the highest possible candidate for pcpu area
508 * allocation. Areas outside of vmalloc area can be returned
509 * here too, consider only end addresses which fall inside
510 * vmalloc area proper.
512 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
513 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
515 kfree_rcu(va, rcu_head);
519 * Free a region of KVA allocated by alloc_vmap_area
521 static void free_vmap_area(struct vmap_area *va)
523 spin_lock(&vmap_area_lock);
524 __free_vmap_area(va);
525 spin_unlock(&vmap_area_lock);
529 * Clear the pagetable entries of a given vmap_area
531 static void unmap_vmap_area(struct vmap_area *va)
533 vunmap_page_range(va->va_start, va->va_end);
536 static void vmap_debug_free_range(unsigned long start, unsigned long end)
539 * Unmap page tables and force a TLB flush immediately if
540 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
541 * bugs similarly to those in linear kernel virtual address
542 * space after a page has been freed.
544 * All the lazy freeing logic is still retained, in order to
545 * minimise intrusiveness of this debugging feature.
547 * This is going to be *slow* (linear kernel virtual address
548 * debugging doesn't do a broadcast TLB flush so it is a lot
551 #ifdef CONFIG_DEBUG_PAGEALLOC
552 vunmap_page_range(start, end);
553 flush_tlb_kernel_range(start, end);
558 * lazy_max_pages is the maximum amount of virtual address space we gather up
559 * before attempting to purge with a TLB flush.
561 * There is a tradeoff here: a larger number will cover more kernel page tables
562 * and take slightly longer to purge, but it will linearly reduce the number of
563 * global TLB flushes that must be performed. It would seem natural to scale
564 * this number up linearly with the number of CPUs (because vmapping activity
565 * could also scale linearly with the number of CPUs), however it is likely
566 * that in practice, workloads might be constrained in other ways that mean
567 * vmap activity will not scale linearly with CPUs. Also, I want to be
568 * conservative and not introduce a big latency on huge systems, so go with
569 * a less aggressive log scale. It will still be an improvement over the old
570 * code, and it will be simple to change the scale factor if we find that it
571 * becomes a problem on bigger systems.
573 static unsigned long lazy_max_pages(void)
577 log = fls(num_online_cpus());
579 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
582 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
584 /* for per-CPU blocks */
585 static void purge_fragmented_blocks_allcpus(void);
588 * called before a call to iounmap() if the caller wants vm_area_struct's
591 void set_iounmap_nonlazy(void)
593 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
597 * Purges all lazily-freed vmap areas.
599 * If sync is 0 then don't purge if there is already a purge in progress.
600 * If force_flush is 1, then flush kernel TLBs between *start and *end even
601 * if we found no lazy vmap areas to unmap (callers can use this to optimise
602 * their own TLB flushing).
603 * Returns with *start = min(*start, lowest purged address)
604 * *end = max(*end, highest purged address)
606 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
607 int sync, int force_flush)
609 static DEFINE_SPINLOCK(purge_lock);
611 struct vmap_area *va;
612 struct vmap_area *n_va;
616 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
617 * should not expect such behaviour. This just simplifies locking for
618 * the case that isn't actually used at the moment anyway.
620 if (!sync && !force_flush) {
621 if (!spin_trylock(&purge_lock))
624 spin_lock(&purge_lock);
627 purge_fragmented_blocks_allcpus();
630 list_for_each_entry_rcu(va, &vmap_area_list, list) {
631 if (va->flags & VM_LAZY_FREE) {
632 if (va->va_start < *start)
633 *start = va->va_start;
634 if (va->va_end > *end)
636 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
637 list_add_tail(&va->purge_list, &valist);
638 va->flags |= VM_LAZY_FREEING;
639 va->flags &= ~VM_LAZY_FREE;
645 atomic_sub(nr, &vmap_lazy_nr);
647 if (nr || force_flush)
648 flush_tlb_kernel_range(*start, *end);
651 spin_lock(&vmap_area_lock);
652 list_for_each_entry_safe(va, n_va, &valist, purge_list)
653 __free_vmap_area(va);
654 spin_unlock(&vmap_area_lock);
656 spin_unlock(&purge_lock);
660 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
661 * is already purging.
663 static void try_purge_vmap_area_lazy(void)
665 unsigned long start = ULONG_MAX, end = 0;
667 __purge_vmap_area_lazy(&start, &end, 0, 0);
671 * Kick off a purge of the outstanding lazy areas.
673 static void purge_vmap_area_lazy(void)
675 unsigned long start = ULONG_MAX, end = 0;
677 __purge_vmap_area_lazy(&start, &end, 1, 0);
681 * Free a vmap area, caller ensuring that the area has been unmapped
682 * and flush_cache_vunmap had been called for the correct range
685 static void free_vmap_area_noflush(struct vmap_area *va)
687 va->flags |= VM_LAZY_FREE;
688 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
689 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
690 try_purge_vmap_area_lazy();
694 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
695 * called for the correct range previously.
697 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
700 free_vmap_area_noflush(va);
704 * Free and unmap a vmap area
706 static void free_unmap_vmap_area(struct vmap_area *va)
708 flush_cache_vunmap(va->va_start, va->va_end);
709 free_unmap_vmap_area_noflush(va);
712 static struct vmap_area *find_vmap_area(unsigned long addr)
714 struct vmap_area *va;
716 spin_lock(&vmap_area_lock);
717 va = __find_vmap_area(addr);
718 spin_unlock(&vmap_area_lock);
723 static void free_unmap_vmap_area_addr(unsigned long addr)
725 struct vmap_area *va;
727 va = find_vmap_area(addr);
729 free_unmap_vmap_area(va);
733 /*** Per cpu kva allocator ***/
736 * vmap space is limited especially on 32 bit architectures. Ensure there is
737 * room for at least 16 percpu vmap blocks per CPU.
740 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
741 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
742 * instead (we just need a rough idea)
744 #if BITS_PER_LONG == 32
745 #define VMALLOC_SPACE (128UL*1024*1024)
747 #define VMALLOC_SPACE (128UL*1024*1024*1024)
750 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
751 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
752 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
753 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
754 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
755 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
756 #define VMAP_BBMAP_BITS \
757 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
758 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
759 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
761 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
763 static bool vmap_initialized __read_mostly = false;
765 struct vmap_block_queue {
767 struct list_head free;
772 struct vmap_area *va;
773 unsigned long free, dirty;
774 unsigned long dirty_min, dirty_max; /*< dirty range */
775 struct list_head free_list;
776 struct rcu_head rcu_head;
777 struct list_head purge;
780 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
781 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
784 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
785 * in the free path. Could get rid of this if we change the API to return a
786 * "cookie" from alloc, to be passed to free. But no big deal yet.
788 static DEFINE_SPINLOCK(vmap_block_tree_lock);
789 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
792 * We should probably have a fallback mechanism to allocate virtual memory
793 * out of partially filled vmap blocks. However vmap block sizing should be
794 * fairly reasonable according to the vmalloc size, so it shouldn't be a
798 static unsigned long addr_to_vb_idx(unsigned long addr)
800 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
801 addr /= VMAP_BLOCK_SIZE;
805 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
809 addr = va_start + (pages_off << PAGE_SHIFT);
810 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
815 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
816 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
817 * @order: how many 2^order pages should be occupied in newly allocated block
818 * @gfp_mask: flags for the page level allocator
820 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
822 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
824 struct vmap_block_queue *vbq;
825 struct vmap_block *vb;
826 struct vmap_area *va;
827 unsigned long vb_idx;
831 node = numa_node_id();
833 vb = kmalloc_node(sizeof(struct vmap_block),
834 gfp_mask & GFP_RECLAIM_MASK, node);
836 return ERR_PTR(-ENOMEM);
838 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
839 VMALLOC_START, VMALLOC_END,
846 err = radix_tree_preload(gfp_mask);
853 vaddr = vmap_block_vaddr(va->va_start, 0);
854 spin_lock_init(&vb->lock);
856 /* At least something should be left free */
857 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
858 vb->free = VMAP_BBMAP_BITS - (1UL << order);
860 vb->dirty_min = VMAP_BBMAP_BITS;
862 INIT_LIST_HEAD(&vb->free_list);
864 vb_idx = addr_to_vb_idx(va->va_start);
865 spin_lock(&vmap_block_tree_lock);
866 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
867 spin_unlock(&vmap_block_tree_lock);
869 radix_tree_preload_end();
871 vbq = &get_cpu_var(vmap_block_queue);
872 spin_lock(&vbq->lock);
873 list_add_tail_rcu(&vb->free_list, &vbq->free);
874 spin_unlock(&vbq->lock);
875 put_cpu_var(vmap_block_queue);
880 static void free_vmap_block(struct vmap_block *vb)
882 struct vmap_block *tmp;
883 unsigned long vb_idx;
885 vb_idx = addr_to_vb_idx(vb->va->va_start);
886 spin_lock(&vmap_block_tree_lock);
887 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
888 spin_unlock(&vmap_block_tree_lock);
891 free_vmap_area_noflush(vb->va);
892 kfree_rcu(vb, rcu_head);
895 static void purge_fragmented_blocks(int cpu)
898 struct vmap_block *vb;
899 struct vmap_block *n_vb;
900 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
903 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
905 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
908 spin_lock(&vb->lock);
909 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
910 vb->free = 0; /* prevent further allocs after releasing lock */
911 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
913 vb->dirty_max = VMAP_BBMAP_BITS;
914 spin_lock(&vbq->lock);
915 list_del_rcu(&vb->free_list);
916 spin_unlock(&vbq->lock);
917 spin_unlock(&vb->lock);
918 list_add_tail(&vb->purge, &purge);
920 spin_unlock(&vb->lock);
924 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
925 list_del(&vb->purge);
930 static void purge_fragmented_blocks_allcpus(void)
934 for_each_possible_cpu(cpu)
935 purge_fragmented_blocks(cpu);
938 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
940 struct vmap_block_queue *vbq;
941 struct vmap_block *vb;
945 BUG_ON(offset_in_page(size));
946 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
947 if (WARN_ON(size == 0)) {
949 * Allocating 0 bytes isn't what caller wants since
950 * get_order(0) returns funny result. Just warn and terminate
955 order = get_order(size);
958 vbq = &get_cpu_var(vmap_block_queue);
959 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
960 unsigned long pages_off;
962 spin_lock(&vb->lock);
963 if (vb->free < (1UL << order)) {
964 spin_unlock(&vb->lock);
968 pages_off = VMAP_BBMAP_BITS - vb->free;
969 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
970 vb->free -= 1UL << order;
972 spin_lock(&vbq->lock);
973 list_del_rcu(&vb->free_list);
974 spin_unlock(&vbq->lock);
977 spin_unlock(&vb->lock);
981 put_cpu_var(vmap_block_queue);
984 /* Allocate new block if nothing was found */
986 vaddr = new_vmap_block(order, gfp_mask);
991 static void vb_free(const void *addr, unsigned long size)
993 unsigned long offset;
994 unsigned long vb_idx;
996 struct vmap_block *vb;
998 BUG_ON(offset_in_page(size));
999 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1001 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1003 order = get_order(size);
1005 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1006 offset >>= PAGE_SHIFT;
1008 vb_idx = addr_to_vb_idx((unsigned long)addr);
1010 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1014 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1016 spin_lock(&vb->lock);
1018 /* Expand dirty range */
1019 vb->dirty_min = min(vb->dirty_min, offset);
1020 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1022 vb->dirty += 1UL << order;
1023 if (vb->dirty == VMAP_BBMAP_BITS) {
1025 spin_unlock(&vb->lock);
1026 free_vmap_block(vb);
1028 spin_unlock(&vb->lock);
1032 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1034 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1035 * to amortize TLB flushing overheads. What this means is that any page you
1036 * have now, may, in a former life, have been mapped into kernel virtual
1037 * address by the vmap layer and so there might be some CPUs with TLB entries
1038 * still referencing that page (additional to the regular 1:1 kernel mapping).
1040 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1041 * be sure that none of the pages we have control over will have any aliases
1042 * from the vmap layer.
1044 void vm_unmap_aliases(void)
1046 unsigned long start = ULONG_MAX, end = 0;
1050 if (unlikely(!vmap_initialized))
1053 for_each_possible_cpu(cpu) {
1054 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1055 struct vmap_block *vb;
1058 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1059 spin_lock(&vb->lock);
1061 unsigned long va_start = vb->va->va_start;
1064 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1065 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1067 start = min(s, start);
1072 spin_unlock(&vb->lock);
1077 __purge_vmap_area_lazy(&start, &end, 1, flush);
1079 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1082 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1083 * @mem: the pointer returned by vm_map_ram
1084 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1086 void vm_unmap_ram(const void *mem, unsigned int count)
1088 unsigned long size = count << PAGE_SHIFT;
1089 unsigned long addr = (unsigned long)mem;
1092 BUG_ON(addr < VMALLOC_START);
1093 BUG_ON(addr > VMALLOC_END);
1094 BUG_ON(addr & (PAGE_SIZE-1));
1096 debug_check_no_locks_freed(mem, size);
1097 vmap_debug_free_range(addr, addr+size);
1099 if (likely(count <= VMAP_MAX_ALLOC))
1102 free_unmap_vmap_area_addr(addr);
1104 EXPORT_SYMBOL(vm_unmap_ram);
1107 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1108 * @pages: an array of pointers to the pages to be mapped
1109 * @count: number of pages
1110 * @node: prefer to allocate data structures on this node
1111 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1113 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1114 * faster than vmap so it's good. But if you mix long-life and short-life
1115 * objects with vm_map_ram(), it could consume lots of address space through
1116 * fragmentation (especially on a 32bit machine). You could see failures in
1117 * the end. Please use this function for short-lived objects.
1119 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1121 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1123 unsigned long size = count << PAGE_SHIFT;
1127 if (likely(count <= VMAP_MAX_ALLOC)) {
1128 mem = vb_alloc(size, GFP_KERNEL);
1131 addr = (unsigned long)mem;
1133 struct vmap_area *va;
1134 va = alloc_vmap_area(size, PAGE_SIZE,
1135 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1139 addr = va->va_start;
1142 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1143 vm_unmap_ram(mem, count);
1148 EXPORT_SYMBOL(vm_map_ram);
1150 static struct vm_struct *vmlist __initdata;
1152 * vm_area_add_early - add vmap area early during boot
1153 * @vm: vm_struct to add
1155 * This function is used to add fixed kernel vm area to vmlist before
1156 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1157 * should contain proper values and the other fields should be zero.
1159 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1161 void __init vm_area_add_early(struct vm_struct *vm)
1163 struct vm_struct *tmp, **p;
1165 BUG_ON(vmap_initialized);
1166 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1167 if (tmp->addr >= vm->addr) {
1168 BUG_ON(tmp->addr < vm->addr + vm->size);
1171 BUG_ON(tmp->addr + tmp->size > vm->addr);
1178 * vm_area_register_early - register vmap area early during boot
1179 * @vm: vm_struct to register
1180 * @align: requested alignment
1182 * This function is used to register kernel vm area before
1183 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1184 * proper values on entry and other fields should be zero. On return,
1185 * vm->addr contains the allocated address.
1187 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1189 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1191 static size_t vm_init_off __initdata;
1194 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1195 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1197 vm->addr = (void *)addr;
1199 vm_area_add_early(vm);
1202 void __init vmalloc_init(void)
1204 struct vmap_area *va;
1205 struct vm_struct *tmp;
1208 for_each_possible_cpu(i) {
1209 struct vmap_block_queue *vbq;
1210 struct vfree_deferred *p;
1212 vbq = &per_cpu(vmap_block_queue, i);
1213 spin_lock_init(&vbq->lock);
1214 INIT_LIST_HEAD(&vbq->free);
1215 p = &per_cpu(vfree_deferred, i);
1216 init_llist_head(&p->list);
1217 INIT_WORK(&p->wq, free_work);
1220 /* Import existing vmlist entries. */
1221 for (tmp = vmlist; tmp; tmp = tmp->next) {
1222 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1223 va->flags = VM_VM_AREA;
1224 va->va_start = (unsigned long)tmp->addr;
1225 va->va_end = va->va_start + tmp->size;
1227 __insert_vmap_area(va);
1230 vmap_area_pcpu_hole = VMALLOC_END;
1232 vmap_initialized = true;
1236 * map_kernel_range_noflush - map kernel VM area with the specified pages
1237 * @addr: start of the VM area to map
1238 * @size: size of the VM area to map
1239 * @prot: page protection flags to use
1240 * @pages: pages to map
1242 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1243 * specify should have been allocated using get_vm_area() and its
1247 * This function does NOT do any cache flushing. The caller is
1248 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1249 * before calling this function.
1252 * The number of pages mapped on success, -errno on failure.
1254 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1255 pgprot_t prot, struct page **pages)
1257 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1261 * unmap_kernel_range_noflush - unmap kernel VM area
1262 * @addr: start of the VM area to unmap
1263 * @size: size of the VM area to unmap
1265 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1266 * specify should have been allocated using get_vm_area() and its
1270 * This function does NOT do any cache flushing. The caller is
1271 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1272 * before calling this function and flush_tlb_kernel_range() after.
1274 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1276 vunmap_page_range(addr, addr + size);
1278 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1281 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1282 * @addr: start of the VM area to unmap
1283 * @size: size of the VM area to unmap
1285 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1286 * the unmapping and tlb after.
1288 void unmap_kernel_range(unsigned long addr, unsigned long size)
1290 unsigned long end = addr + size;
1292 flush_cache_vunmap(addr, end);
1293 vunmap_page_range(addr, end);
1294 flush_tlb_kernel_range(addr, end);
1296 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1298 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1300 unsigned long addr = (unsigned long)area->addr;
1301 unsigned long end = addr + get_vm_area_size(area);
1304 err = vmap_page_range(addr, end, prot, pages);
1306 return err > 0 ? 0 : err;
1308 EXPORT_SYMBOL_GPL(map_vm_area);
1310 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1311 unsigned long flags, const void *caller)
1313 spin_lock(&vmap_area_lock);
1315 vm->addr = (void *)va->va_start;
1316 vm->size = va->va_end - va->va_start;
1317 vm->caller = caller;
1319 va->flags |= VM_VM_AREA;
1320 spin_unlock(&vmap_area_lock);
1323 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1326 * Before removing VM_UNINITIALIZED,
1327 * we should make sure that vm has proper values.
1328 * Pair with smp_rmb() in show_numa_info().
1331 vm->flags &= ~VM_UNINITIALIZED;
1334 static struct vm_struct *__get_vm_area_node(unsigned long size,
1335 unsigned long align, unsigned long flags, unsigned long start,
1336 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1338 struct vmap_area *va;
1339 struct vm_struct *area;
1341 BUG_ON(in_interrupt());
1342 if (flags & VM_IOREMAP)
1343 align = 1ul << clamp_t(int, fls_long(size),
1344 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1346 size = PAGE_ALIGN(size);
1347 if (unlikely(!size))
1350 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1351 if (unlikely(!area))
1354 if (!(flags & VM_NO_GUARD))
1357 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1363 setup_vmalloc_vm(area, va, flags, caller);
1368 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1369 unsigned long start, unsigned long end)
1371 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1372 GFP_KERNEL, __builtin_return_address(0));
1374 EXPORT_SYMBOL_GPL(__get_vm_area);
1376 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1377 unsigned long start, unsigned long end,
1380 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1381 GFP_KERNEL, caller);
1385 * get_vm_area - reserve a contiguous kernel virtual area
1386 * @size: size of the area
1387 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1389 * Search an area of @size in the kernel virtual mapping area,
1390 * and reserved it for out purposes. Returns the area descriptor
1391 * on success or %NULL on failure.
1393 struct vm_struct *get_vm_area(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,
1397 __builtin_return_address(0));
1400 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1403 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1404 NUMA_NO_NODE, GFP_KERNEL, caller);
1408 * find_vm_area - find a continuous kernel virtual area
1409 * @addr: base address
1411 * Search for the kernel VM area starting at @addr, and return it.
1412 * It is up to the caller to do all required locking to keep the returned
1415 struct vm_struct *find_vm_area(const void *addr)
1417 struct vmap_area *va;
1419 va = find_vmap_area((unsigned long)addr);
1420 if (va && va->flags & VM_VM_AREA)
1427 * remove_vm_area - find and remove a continuous kernel virtual area
1428 * @addr: base address
1430 * Search for the kernel VM area starting at @addr, and remove it.
1431 * This function returns the found VM area, but using it is NOT safe
1432 * on SMP machines, except for its size or flags.
1434 struct vm_struct *remove_vm_area(const void *addr)
1436 struct vmap_area *va;
1438 va = find_vmap_area((unsigned long)addr);
1439 if (va && va->flags & VM_VM_AREA) {
1440 struct vm_struct *vm = va->vm;
1442 spin_lock(&vmap_area_lock);
1444 va->flags &= ~VM_VM_AREA;
1445 spin_unlock(&vmap_area_lock);
1447 vmap_debug_free_range(va->va_start, va->va_end);
1448 kasan_free_shadow(vm);
1449 free_unmap_vmap_area(va);
1456 static void __vunmap(const void *addr, int deallocate_pages)
1458 struct vm_struct *area;
1463 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1467 area = find_vmap_area((unsigned long)addr)->vm;
1468 if (unlikely(!area)) {
1469 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1474 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1475 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1477 remove_vm_area(addr);
1478 if (deallocate_pages) {
1481 for (i = 0; i < area->nr_pages; i++) {
1482 struct page *page = area->pages[i];
1488 if (area->flags & VM_VPAGES)
1499 * vfree - release memory allocated by vmalloc()
1500 * @addr: memory base address
1502 * Free the virtually continuous memory area starting at @addr, as
1503 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1504 * NULL, no operation is performed.
1506 * Must not be called in NMI context (strictly speaking, only if we don't
1507 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1508 * conventions for vfree() arch-depenedent would be a really bad idea)
1510 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1512 void vfree(const void *addr)
1516 kmemleak_free(addr);
1520 if (unlikely(in_interrupt())) {
1521 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1522 if (llist_add((struct llist_node *)addr, &p->list))
1523 schedule_work(&p->wq);
1527 EXPORT_SYMBOL(vfree);
1530 * vunmap - release virtual mapping obtained by vmap()
1531 * @addr: memory base address
1533 * Free the virtually contiguous memory area starting at @addr,
1534 * which was created from the page array passed to vmap().
1536 * Must not be called in interrupt context.
1538 void vunmap(const void *addr)
1540 BUG_ON(in_interrupt());
1545 EXPORT_SYMBOL(vunmap);
1548 * vmap - map an array of pages into virtually contiguous space
1549 * @pages: array of page pointers
1550 * @count: number of pages to map
1551 * @flags: vm_area->flags
1552 * @prot: page protection for the mapping
1554 * Maps @count pages from @pages into contiguous kernel virtual
1557 void *vmap(struct page **pages, unsigned int count,
1558 unsigned long flags, pgprot_t prot)
1560 struct vm_struct *area;
1564 if (count > totalram_pages)
1567 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1568 __builtin_return_address(0));
1572 if (map_vm_area(area, prot, pages)) {
1579 EXPORT_SYMBOL(vmap);
1581 static void *__vmalloc_node(unsigned long size, unsigned long align,
1582 gfp_t gfp_mask, pgprot_t prot,
1583 int node, const void *caller);
1584 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1585 pgprot_t prot, int node)
1587 const int order = 0;
1588 struct page **pages;
1589 unsigned int nr_pages, array_size, i;
1590 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1591 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1593 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1594 array_size = (nr_pages * sizeof(struct page *));
1596 area->nr_pages = nr_pages;
1597 /* Please note that the recursion is strictly bounded. */
1598 if (array_size > PAGE_SIZE) {
1599 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1600 PAGE_KERNEL, node, area->caller);
1601 area->flags |= VM_VPAGES;
1603 pages = kmalloc_node(array_size, nested_gfp, node);
1605 area->pages = pages;
1607 remove_vm_area(area->addr);
1612 for (i = 0; i < area->nr_pages; i++) {
1615 if (node == NUMA_NO_NODE)
1616 page = alloc_page(alloc_mask);
1618 page = alloc_pages_node(node, alloc_mask, order);
1620 if (unlikely(!page)) {
1621 /* Successfully allocated i pages, free them in __vunmap() */
1625 area->pages[i] = page;
1626 if (gfpflags_allow_blocking(gfp_mask))
1630 if (map_vm_area(area, prot, pages))
1635 warn_alloc_failed(gfp_mask, order,
1636 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1637 (area->nr_pages*PAGE_SIZE), area->size);
1643 * __vmalloc_node_range - allocate virtually contiguous memory
1644 * @size: allocation size
1645 * @align: desired alignment
1646 * @start: vm area range start
1647 * @end: vm area range end
1648 * @gfp_mask: flags for the page level allocator
1649 * @prot: protection mask for the allocated pages
1650 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1651 * @node: node to use for allocation or NUMA_NO_NODE
1652 * @caller: caller's return address
1654 * Allocate enough pages to cover @size from the page level
1655 * allocator with @gfp_mask flags. Map them into contiguous
1656 * kernel virtual space, using a pagetable protection of @prot.
1658 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1659 unsigned long start, unsigned long end, gfp_t gfp_mask,
1660 pgprot_t prot, unsigned long vm_flags, int node,
1663 struct vm_struct *area;
1665 unsigned long real_size = size;
1667 size = PAGE_ALIGN(size);
1668 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1671 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1672 vm_flags, start, end, node, gfp_mask, caller);
1676 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1681 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1682 * flag. It means that vm_struct is not fully initialized.
1683 * Now, it is fully initialized, so remove this flag here.
1685 clear_vm_uninitialized_flag(area);
1688 * A ref_count = 2 is needed because vm_struct allocated in
1689 * __get_vm_area_node() contains a reference to the virtual address of
1690 * the vmalloc'ed block.
1692 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1697 warn_alloc_failed(gfp_mask, 0,
1698 "vmalloc: allocation failure: %lu bytes\n",
1704 * __vmalloc_node - allocate virtually contiguous memory
1705 * @size: allocation size
1706 * @align: desired alignment
1707 * @gfp_mask: flags for the page level allocator
1708 * @prot: protection mask for the allocated pages
1709 * @node: node to use for allocation or NUMA_NO_NODE
1710 * @caller: caller's return address
1712 * Allocate enough pages to cover @size from the page level
1713 * allocator with @gfp_mask flags. Map them into contiguous
1714 * kernel virtual space, using a pagetable protection of @prot.
1716 static void *__vmalloc_node(unsigned long size, unsigned long align,
1717 gfp_t gfp_mask, pgprot_t prot,
1718 int node, const void *caller)
1720 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1721 gfp_mask, prot, 0, node, caller);
1724 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1726 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1727 __builtin_return_address(0));
1729 EXPORT_SYMBOL(__vmalloc);
1731 static inline void *__vmalloc_node_flags(unsigned long size,
1732 int node, gfp_t flags)
1734 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1735 node, __builtin_return_address(0));
1739 * vmalloc - allocate virtually contiguous memory
1740 * @size: allocation size
1741 * Allocate enough pages to cover @size from the page level
1742 * allocator and map them into contiguous kernel virtual space.
1744 * For tight control over page level allocator and protection flags
1745 * use __vmalloc() instead.
1747 void *vmalloc(unsigned long size)
1749 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1750 GFP_KERNEL | __GFP_HIGHMEM);
1752 EXPORT_SYMBOL(vmalloc);
1755 * vzalloc - allocate virtually contiguous memory with zero fill
1756 * @size: allocation size
1757 * Allocate enough pages to cover @size from the page level
1758 * allocator and map them into contiguous kernel virtual space.
1759 * The memory allocated is set to zero.
1761 * For tight control over page level allocator and protection flags
1762 * use __vmalloc() instead.
1764 void *vzalloc(unsigned long size)
1766 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1767 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1769 EXPORT_SYMBOL(vzalloc);
1772 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1773 * @size: allocation size
1775 * The resulting memory area is zeroed so it can be mapped to userspace
1776 * without leaking data.
1778 void *vmalloc_user(unsigned long size)
1780 struct vm_struct *area;
1783 ret = __vmalloc_node(size, SHMLBA,
1784 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1785 PAGE_KERNEL, NUMA_NO_NODE,
1786 __builtin_return_address(0));
1788 area = find_vm_area(ret);
1789 area->flags |= VM_USERMAP;
1793 EXPORT_SYMBOL(vmalloc_user);
1796 * vmalloc_node - allocate memory on a specific node
1797 * @size: allocation size
1800 * Allocate enough pages to cover @size from the page level
1801 * allocator and map them into contiguous kernel virtual space.
1803 * For tight control over page level allocator and protection flags
1804 * use __vmalloc() instead.
1806 void *vmalloc_node(unsigned long size, int node)
1808 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1809 node, __builtin_return_address(0));
1811 EXPORT_SYMBOL(vmalloc_node);
1814 * vzalloc_node - allocate memory on a specific node with zero fill
1815 * @size: allocation size
1818 * Allocate enough pages to cover @size from the page level
1819 * allocator and map them into contiguous kernel virtual space.
1820 * The memory allocated is set to zero.
1822 * For tight control over page level allocator and protection flags
1823 * use __vmalloc_node() instead.
1825 void *vzalloc_node(unsigned long size, int node)
1827 return __vmalloc_node_flags(size, node,
1828 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1830 EXPORT_SYMBOL(vzalloc_node);
1832 #ifndef PAGE_KERNEL_EXEC
1833 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1837 * vmalloc_exec - allocate virtually contiguous, executable memory
1838 * @size: allocation size
1840 * Kernel-internal function to allocate enough pages to cover @size
1841 * the page level allocator and map them into contiguous and
1842 * executable kernel virtual space.
1844 * For tight control over page level allocator and protection flags
1845 * use __vmalloc() instead.
1848 void *vmalloc_exec(unsigned long size)
1850 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1851 NUMA_NO_NODE, __builtin_return_address(0));
1854 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1855 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1856 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1857 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1859 #define GFP_VMALLOC32 GFP_KERNEL
1863 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1864 * @size: allocation size
1866 * Allocate enough 32bit PA addressable pages to cover @size from the
1867 * page level allocator and map them into contiguous kernel virtual space.
1869 void *vmalloc_32(unsigned long size)
1871 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1872 NUMA_NO_NODE, __builtin_return_address(0));
1874 EXPORT_SYMBOL(vmalloc_32);
1877 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1878 * @size: allocation size
1880 * The resulting memory area is 32bit addressable and zeroed so it can be
1881 * mapped to userspace without leaking data.
1883 void *vmalloc_32_user(unsigned long size)
1885 struct vm_struct *area;
1888 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1889 NUMA_NO_NODE, __builtin_return_address(0));
1891 area = find_vm_area(ret);
1892 area->flags |= VM_USERMAP;
1896 EXPORT_SYMBOL(vmalloc_32_user);
1899 * small helper routine , copy contents to buf from addr.
1900 * If the page is not present, fill zero.
1903 static int aligned_vread(char *buf, char *addr, unsigned long count)
1909 unsigned long offset, length;
1911 offset = offset_in_page(addr);
1912 length = PAGE_SIZE - offset;
1915 p = vmalloc_to_page(addr);
1917 * To do safe access to this _mapped_ area, we need
1918 * lock. But adding lock here means that we need to add
1919 * overhead of vmalloc()/vfree() calles for this _debug_
1920 * interface, rarely used. Instead of that, we'll use
1921 * kmap() and get small overhead in this access function.
1925 * we can expect USER0 is not used (see vread/vwrite's
1926 * function description)
1928 void *map = kmap_atomic(p);
1929 memcpy(buf, map + offset, length);
1932 memset(buf, 0, length);
1942 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1948 unsigned long offset, length;
1950 offset = offset_in_page(addr);
1951 length = PAGE_SIZE - offset;
1954 p = vmalloc_to_page(addr);
1956 * To do safe access to this _mapped_ area, we need
1957 * lock. But adding lock here means that we need to add
1958 * overhead of vmalloc()/vfree() calles for this _debug_
1959 * interface, rarely used. Instead of that, we'll use
1960 * kmap() and get small overhead in this access function.
1964 * we can expect USER0 is not used (see vread/vwrite's
1965 * function description)
1967 void *map = kmap_atomic(p);
1968 memcpy(map + offset, buf, length);
1980 * vread() - read vmalloc area in a safe way.
1981 * @buf: buffer for reading data
1982 * @addr: vm address.
1983 * @count: number of bytes to be read.
1985 * Returns # of bytes which addr and buf should be increased.
1986 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1987 * includes any intersect with alive vmalloc area.
1989 * This function checks that addr is a valid vmalloc'ed area, and
1990 * copy data from that area to a given buffer. If the given memory range
1991 * of [addr...addr+count) includes some valid address, data is copied to
1992 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1993 * IOREMAP area is treated as memory hole and no copy is done.
1995 * If [addr...addr+count) doesn't includes any intersects with alive
1996 * vm_struct area, returns 0. @buf should be kernel's buffer.
1998 * Note: In usual ops, vread() is never necessary because the caller
1999 * should know vmalloc() area is valid and can use memcpy().
2000 * This is for routines which have to access vmalloc area without
2001 * any informaion, as /dev/kmem.
2005 long vread(char *buf, char *addr, unsigned long count)
2007 struct vmap_area *va;
2008 struct vm_struct *vm;
2009 char *vaddr, *buf_start = buf;
2010 unsigned long buflen = count;
2013 /* Don't allow overflow */
2014 if ((unsigned long) addr + count < count)
2015 count = -(unsigned long) addr;
2017 spin_lock(&vmap_area_lock);
2018 list_for_each_entry(va, &vmap_area_list, list) {
2022 if (!(va->flags & VM_VM_AREA))
2026 vaddr = (char *) vm->addr;
2027 if (addr >= vaddr + get_vm_area_size(vm))
2029 while (addr < vaddr) {
2037 n = vaddr + get_vm_area_size(vm) - addr;
2040 if (!(vm->flags & VM_IOREMAP))
2041 aligned_vread(buf, addr, n);
2042 else /* IOREMAP area is treated as memory hole */
2049 spin_unlock(&vmap_area_lock);
2051 if (buf == buf_start)
2053 /* zero-fill memory holes */
2054 if (buf != buf_start + buflen)
2055 memset(buf, 0, buflen - (buf - buf_start));
2061 * vwrite() - write vmalloc area in a safe way.
2062 * @buf: buffer for source data
2063 * @addr: vm address.
2064 * @count: number of bytes to be read.
2066 * Returns # of bytes which addr and buf should be incresed.
2067 * (same number to @count).
2068 * If [addr...addr+count) doesn't includes any intersect with valid
2069 * vmalloc area, returns 0.
2071 * This function checks that addr is a valid vmalloc'ed area, and
2072 * copy data from a buffer to the given addr. If specified range of
2073 * [addr...addr+count) includes some valid address, data is copied from
2074 * proper area of @buf. If there are memory holes, no copy to hole.
2075 * IOREMAP area is treated as memory hole and no copy is done.
2077 * If [addr...addr+count) doesn't includes any intersects with alive
2078 * vm_struct area, returns 0. @buf should be kernel's buffer.
2080 * Note: In usual ops, vwrite() is never necessary because the caller
2081 * should know vmalloc() area is valid and can use memcpy().
2082 * This is for routines which have to access vmalloc area without
2083 * any informaion, as /dev/kmem.
2086 long vwrite(char *buf, char *addr, unsigned long count)
2088 struct vmap_area *va;
2089 struct vm_struct *vm;
2091 unsigned long n, buflen;
2094 /* Don't allow overflow */
2095 if ((unsigned long) addr + count < count)
2096 count = -(unsigned long) addr;
2099 spin_lock(&vmap_area_lock);
2100 list_for_each_entry(va, &vmap_area_list, list) {
2104 if (!(va->flags & VM_VM_AREA))
2108 vaddr = (char *) vm->addr;
2109 if (addr >= vaddr + get_vm_area_size(vm))
2111 while (addr < vaddr) {
2118 n = vaddr + get_vm_area_size(vm) - addr;
2121 if (!(vm->flags & VM_IOREMAP)) {
2122 aligned_vwrite(buf, addr, n);
2130 spin_unlock(&vmap_area_lock);
2137 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2138 * @vma: vma to cover
2139 * @uaddr: target user address to start at
2140 * @kaddr: virtual address of vmalloc kernel memory
2141 * @size: size of map area
2143 * Returns: 0 for success, -Exxx on failure
2145 * This function checks that @kaddr is a valid vmalloc'ed area,
2146 * and that it is big enough to cover the range starting at
2147 * @uaddr in @vma. Will return failure if that criteria isn't
2150 * Similar to remap_pfn_range() (see mm/memory.c)
2152 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2153 void *kaddr, unsigned long size)
2155 struct vm_struct *area;
2157 size = PAGE_ALIGN(size);
2159 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2162 area = find_vm_area(kaddr);
2166 if (!(area->flags & VM_USERMAP))
2169 if (kaddr + size > area->addr + get_vm_area_size(area))
2173 struct page *page = vmalloc_to_page(kaddr);
2176 ret = vm_insert_page(vma, uaddr, page);
2185 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2189 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2192 * remap_vmalloc_range - map vmalloc pages to userspace
2193 * @vma: vma to cover (map full range of vma)
2194 * @addr: vmalloc memory
2195 * @pgoff: number of pages into addr before first page to map
2197 * Returns: 0 for success, -Exxx on failure
2199 * This function checks that addr is a valid vmalloc'ed area, and
2200 * that it is big enough to cover the vma. Will return failure if
2201 * that criteria isn't met.
2203 * Similar to remap_pfn_range() (see mm/memory.c)
2205 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2206 unsigned long pgoff)
2208 return remap_vmalloc_range_partial(vma, vma->vm_start,
2209 addr + (pgoff << PAGE_SHIFT),
2210 vma->vm_end - vma->vm_start);
2212 EXPORT_SYMBOL(remap_vmalloc_range);
2215 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2218 void __weak vmalloc_sync_all(void)
2223 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2235 * alloc_vm_area - allocate a range of kernel address space
2236 * @size: size of the area
2237 * @ptes: returns the PTEs for the address space
2239 * Returns: NULL on failure, vm_struct on success
2241 * This function reserves a range of kernel address space, and
2242 * allocates pagetables to map that range. No actual mappings
2245 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2246 * allocated for the VM area are returned.
2248 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2250 struct vm_struct *area;
2252 area = get_vm_area_caller(size, VM_IOREMAP,
2253 __builtin_return_address(0));
2258 * This ensures that page tables are constructed for this region
2259 * of kernel virtual address space and mapped into init_mm.
2261 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2262 size, f, ptes ? &ptes : NULL)) {
2269 EXPORT_SYMBOL_GPL(alloc_vm_area);
2271 void free_vm_area(struct vm_struct *area)
2273 struct vm_struct *ret;
2274 ret = remove_vm_area(area->addr);
2275 BUG_ON(ret != area);
2278 EXPORT_SYMBOL_GPL(free_vm_area);
2281 static struct vmap_area *node_to_va(struct rb_node *n)
2283 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2287 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2288 * @end: target address
2289 * @pnext: out arg for the next vmap_area
2290 * @pprev: out arg for the previous vmap_area
2292 * Returns: %true if either or both of next and prev are found,
2293 * %false if no vmap_area exists
2295 * Find vmap_areas end addresses of which enclose @end. ie. if not
2296 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2298 static bool pvm_find_next_prev(unsigned long end,
2299 struct vmap_area **pnext,
2300 struct vmap_area **pprev)
2302 struct rb_node *n = vmap_area_root.rb_node;
2303 struct vmap_area *va = NULL;
2306 va = rb_entry(n, struct vmap_area, rb_node);
2307 if (end < va->va_end)
2309 else if (end > va->va_end)
2318 if (va->va_end > end) {
2320 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2323 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2329 * pvm_determine_end - find the highest aligned address between two vmap_areas
2330 * @pnext: in/out arg for the next vmap_area
2331 * @pprev: in/out arg for the previous vmap_area
2334 * Returns: determined end address
2336 * Find the highest aligned address between *@pnext and *@pprev below
2337 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2338 * down address is between the end addresses of the two vmap_areas.
2340 * Please note that the address returned by this function may fall
2341 * inside *@pnext vmap_area. The caller is responsible for checking
2344 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2345 struct vmap_area **pprev,
2346 unsigned long align)
2348 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2352 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2356 while (*pprev && (*pprev)->va_end > addr) {
2358 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2365 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2366 * @offsets: array containing offset of each area
2367 * @sizes: array containing size of each area
2368 * @nr_vms: the number of areas to allocate
2369 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2371 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2372 * vm_structs on success, %NULL on failure
2374 * Percpu allocator wants to use congruent vm areas so that it can
2375 * maintain the offsets among percpu areas. This function allocates
2376 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2377 * be scattered pretty far, distance between two areas easily going up
2378 * to gigabytes. To avoid interacting with regular vmallocs, these
2379 * areas are allocated from top.
2381 * Despite its complicated look, this allocator is rather simple. It
2382 * does everything top-down and scans areas from the end looking for
2383 * matching slot. While scanning, if any of the areas overlaps with
2384 * existing vmap_area, the base address is pulled down to fit the
2385 * area. Scanning is repeated till all the areas fit and then all
2386 * necessary data structres are inserted and the result is returned.
2388 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2389 const size_t *sizes, int nr_vms,
2392 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2393 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2394 struct vmap_area **vas, *prev, *next;
2395 struct vm_struct **vms;
2396 int area, area2, last_area, term_area;
2397 unsigned long base, start, end, last_end;
2398 bool purged = false;
2400 /* verify parameters and allocate data structures */
2401 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2402 for (last_area = 0, area = 0; area < nr_vms; area++) {
2403 start = offsets[area];
2404 end = start + sizes[area];
2406 /* is everything aligned properly? */
2407 BUG_ON(!IS_ALIGNED(offsets[area], align));
2408 BUG_ON(!IS_ALIGNED(sizes[area], align));
2410 /* detect the area with the highest address */
2411 if (start > offsets[last_area])
2414 for (area2 = 0; area2 < nr_vms; area2++) {
2415 unsigned long start2 = offsets[area2];
2416 unsigned long end2 = start2 + sizes[area2];
2421 BUG_ON(start2 >= start && start2 < end);
2422 BUG_ON(end2 <= end && end2 > start);
2425 last_end = offsets[last_area] + sizes[last_area];
2427 if (vmalloc_end - vmalloc_start < last_end) {
2432 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2433 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2437 for (area = 0; area < nr_vms; area++) {
2438 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2439 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2440 if (!vas[area] || !vms[area])
2444 spin_lock(&vmap_area_lock);
2446 /* start scanning - we scan from the top, begin with the last area */
2447 area = term_area = last_area;
2448 start = offsets[area];
2449 end = start + sizes[area];
2451 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2452 base = vmalloc_end - last_end;
2455 base = pvm_determine_end(&next, &prev, align) - end;
2458 BUG_ON(next && next->va_end <= base + end);
2459 BUG_ON(prev && prev->va_end > base + end);
2462 * base might have underflowed, add last_end before
2465 if (base + last_end < vmalloc_start + last_end) {
2466 spin_unlock(&vmap_area_lock);
2468 purge_vmap_area_lazy();
2476 * If next overlaps, move base downwards so that it's
2477 * right below next and then recheck.
2479 if (next && next->va_start < base + end) {
2480 base = pvm_determine_end(&next, &prev, align) - end;
2486 * If prev overlaps, shift down next and prev and move
2487 * base so that it's right below new next and then
2490 if (prev && prev->va_end > base + start) {
2492 prev = node_to_va(rb_prev(&next->rb_node));
2493 base = pvm_determine_end(&next, &prev, align) - end;
2499 * This area fits, move on to the previous one. If
2500 * the previous one is the terminal one, we're done.
2502 area = (area + nr_vms - 1) % nr_vms;
2503 if (area == term_area)
2505 start = offsets[area];
2506 end = start + sizes[area];
2507 pvm_find_next_prev(base + end, &next, &prev);
2510 /* we've found a fitting base, insert all va's */
2511 for (area = 0; area < nr_vms; area++) {
2512 struct vmap_area *va = vas[area];
2514 va->va_start = base + offsets[area];
2515 va->va_end = va->va_start + sizes[area];
2516 __insert_vmap_area(va);
2519 vmap_area_pcpu_hole = base + offsets[last_area];
2521 spin_unlock(&vmap_area_lock);
2523 /* insert all vm's */
2524 for (area = 0; area < nr_vms; area++)
2525 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2532 for (area = 0; area < nr_vms; area++) {
2543 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2544 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2545 * @nr_vms: the number of allocated areas
2547 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2549 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2553 for (i = 0; i < nr_vms; i++)
2554 free_vm_area(vms[i]);
2557 #endif /* CONFIG_SMP */
2559 #ifdef CONFIG_PROC_FS
2560 static void *s_start(struct seq_file *m, loff_t *pos)
2561 __acquires(&vmap_area_lock)
2564 struct vmap_area *va;
2566 spin_lock(&vmap_area_lock);
2567 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2568 while (n > 0 && &va->list != &vmap_area_list) {
2570 va = list_entry(va->list.next, typeof(*va), list);
2572 if (!n && &va->list != &vmap_area_list)
2579 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2581 struct vmap_area *va = p, *next;
2584 next = list_entry(va->list.next, typeof(*va), list);
2585 if (&next->list != &vmap_area_list)
2591 static void s_stop(struct seq_file *m, void *p)
2592 __releases(&vmap_area_lock)
2594 spin_unlock(&vmap_area_lock);
2597 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2599 if (IS_ENABLED(CONFIG_NUMA)) {
2600 unsigned int nr, *counters = m->private;
2605 if (v->flags & VM_UNINITIALIZED)
2607 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2610 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2612 for (nr = 0; nr < v->nr_pages; nr++)
2613 counters[page_to_nid(v->pages[nr])]++;
2615 for_each_node_state(nr, N_HIGH_MEMORY)
2617 seq_printf(m, " N%u=%u", nr, counters[nr]);
2621 static int s_show(struct seq_file *m, void *p)
2623 struct vmap_area *va = p;
2624 struct vm_struct *v;
2627 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2628 * behalf of vmap area is being tear down or vm_map_ram allocation.
2630 if (!(va->flags & VM_VM_AREA))
2635 seq_printf(m, "0x%pK-0x%pK %7ld",
2636 v->addr, v->addr + v->size, v->size);
2639 seq_printf(m, " %pS", v->caller);
2642 seq_printf(m, " pages=%d", v->nr_pages);
2645 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2647 if (v->flags & VM_IOREMAP)
2648 seq_puts(m, " ioremap");
2650 if (v->flags & VM_ALLOC)
2651 seq_puts(m, " vmalloc");
2653 if (v->flags & VM_MAP)
2654 seq_puts(m, " vmap");
2656 if (v->flags & VM_USERMAP)
2657 seq_puts(m, " user");
2659 if (v->flags & VM_VPAGES)
2660 seq_puts(m, " vpages");
2662 show_numa_info(m, v);
2667 static const struct seq_operations vmalloc_op = {
2674 static int vmalloc_open(struct inode *inode, struct file *file)
2676 if (IS_ENABLED(CONFIG_NUMA))
2677 return seq_open_private(file, &vmalloc_op,
2678 nr_node_ids * sizeof(unsigned int));
2680 return seq_open(file, &vmalloc_op);
2683 static const struct file_operations proc_vmalloc_operations = {
2684 .open = vmalloc_open,
2686 .llseek = seq_lseek,
2687 .release = seq_release_private,
2690 static int __init proc_vmalloc_init(void)
2692 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2695 module_init(proc_vmalloc_init);