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/notifier.h>
25 #include <linux/rbtree.h>
26 #include <linux/radix-tree.h>
27 #include <linux/rcupdate.h>
28 #include <linux/pfn.h>
29 #include <linux/kmemleak.h>
30 #include <linux/atomic.h>
31 #include <linux/compiler.h>
32 #include <linux/llist.h>
33 #include <linux/bitops.h>
35 #include <asm/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
45 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct *w)
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
52 struct llist_node *llnode = llist_del_all(&p->list);
55 llnode = llist_next(llnode);
60 /*** Page table manipulation functions ***/
62 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
66 pte = pte_offset_kernel(pmd, addr);
68 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
69 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
70 } while (pte++, addr += PAGE_SIZE, addr != end);
73 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
78 pmd = pmd_offset(pud, addr);
80 next = pmd_addr_end(addr, end);
81 if (pmd_clear_huge(pmd))
83 if (pmd_none_or_clear_bad(pmd))
85 vunmap_pte_range(pmd, addr, next);
86 } while (pmd++, addr = next, addr != end);
89 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
94 pud = pud_offset(pgd, addr);
96 next = pud_addr_end(addr, end);
97 if (pud_clear_huge(pud))
99 if (pud_none_or_clear_bad(pud))
101 vunmap_pmd_range(pud, addr, next);
102 } while (pud++, addr = next, addr != end);
105 static void vunmap_page_range(unsigned long addr, unsigned long end)
111 pgd = pgd_offset_k(addr);
113 next = pgd_addr_end(addr, end);
114 if (pgd_none_or_clear_bad(pgd))
116 vunmap_pud_range(pgd, addr, next);
117 } while (pgd++, addr = next, addr != end);
120 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
121 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
126 * nr is a running index into the array which helps higher level
127 * callers keep track of where we're up to.
130 pte = pte_alloc_kernel(pmd, addr);
134 struct page *page = pages[*nr];
136 if (WARN_ON(!pte_none(*pte)))
140 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
142 } while (pte++, addr += PAGE_SIZE, addr != end);
146 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
147 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
152 pmd = pmd_alloc(&init_mm, pud, addr);
156 next = pmd_addr_end(addr, end);
157 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
159 } while (pmd++, addr = next, addr != end);
163 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
164 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
169 pud = pud_alloc(&init_mm, pgd, addr);
173 next = pud_addr_end(addr, end);
174 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
176 } while (pud++, addr = next, addr != end);
181 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
182 * will have pfns corresponding to the "pages" array.
184 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
186 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
187 pgprot_t prot, struct page **pages)
191 unsigned long addr = start;
196 pgd = pgd_offset_k(addr);
198 next = pgd_addr_end(addr, end);
199 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
202 } while (pgd++, addr = next, addr != end);
207 static int vmap_page_range(unsigned long start, unsigned long end,
208 pgprot_t prot, struct page **pages)
212 ret = vmap_page_range_noflush(start, end, prot, pages);
213 flush_cache_vmap(start, end);
217 int is_vmalloc_or_module_addr(const void *x)
220 * ARM, x86-64 and sparc64 put modules in a special place,
221 * and fall back on vmalloc() if that fails. Others
222 * just put it in the vmalloc space.
224 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
225 unsigned long addr = (unsigned long)x;
226 if (addr >= MODULES_VADDR && addr < MODULES_END)
229 return is_vmalloc_addr(x);
233 * Walk a vmap address to the struct page it maps.
235 struct page *vmalloc_to_page(const void *vmalloc_addr)
237 unsigned long addr = (unsigned long) vmalloc_addr;
238 struct page *page = NULL;
239 pgd_t *pgd = pgd_offset_k(addr);
242 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
243 * architectures that do not vmalloc module space
245 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
248 * Don't dereference bad PUD or PMD (below) entries. This will also
249 * identify huge mappings, which we may encounter on architectures
250 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
251 * identified as vmalloc addresses by is_vmalloc_addr(), but are
252 * not [unambiguously] associated with a struct page, so there is
253 * no correct value to return for them.
255 if (!pgd_none(*pgd)) {
256 pud_t *pud = pud_offset(pgd, addr);
257 WARN_ON_ONCE(pud_bad(*pud));
258 if (!pud_none(*pud) && !pud_bad(*pud)) {
259 pmd_t *pmd = pmd_offset(pud, addr);
260 WARN_ON_ONCE(pmd_bad(*pmd));
261 if (!pmd_none(*pmd) && !pmd_bad(*pmd)) {
264 ptep = pte_offset_map(pmd, addr);
266 if (pte_present(pte))
267 page = pte_page(pte);
274 EXPORT_SYMBOL(vmalloc_to_page);
277 * Map a vmalloc()-space virtual address to the physical page frame number.
279 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
281 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
283 EXPORT_SYMBOL(vmalloc_to_pfn);
286 /*** Global kva allocator ***/
288 #define VM_VM_AREA 0x04
290 static DEFINE_SPINLOCK(vmap_area_lock);
291 /* Export for kexec only */
292 LIST_HEAD(vmap_area_list);
293 static LLIST_HEAD(vmap_purge_list);
294 static struct rb_root vmap_area_root = RB_ROOT;
296 /* The vmap cache globals are protected by vmap_area_lock */
297 static struct rb_node *free_vmap_cache;
298 static unsigned long cached_hole_size;
299 static unsigned long cached_vstart;
300 static unsigned long cached_align;
302 static unsigned long vmap_area_pcpu_hole;
304 static struct vmap_area *__find_vmap_area(unsigned long addr)
306 struct rb_node *n = vmap_area_root.rb_node;
309 struct vmap_area *va;
311 va = rb_entry(n, struct vmap_area, rb_node);
312 if (addr < va->va_start)
314 else if (addr >= va->va_end)
323 static void __insert_vmap_area(struct vmap_area *va)
325 struct rb_node **p = &vmap_area_root.rb_node;
326 struct rb_node *parent = NULL;
330 struct vmap_area *tmp_va;
333 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
334 if (va->va_start < tmp_va->va_end)
336 else if (va->va_end > tmp_va->va_start)
342 rb_link_node(&va->rb_node, parent, p);
343 rb_insert_color(&va->rb_node, &vmap_area_root);
345 /* address-sort this list */
346 tmp = rb_prev(&va->rb_node);
348 struct vmap_area *prev;
349 prev = rb_entry(tmp, struct vmap_area, rb_node);
350 list_add_rcu(&va->list, &prev->list);
352 list_add_rcu(&va->list, &vmap_area_list);
355 static void purge_vmap_area_lazy(void);
357 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
360 * Allocate a region of KVA of the specified size and alignment, within the
363 static struct vmap_area *alloc_vmap_area(unsigned long size,
365 unsigned long vstart, unsigned long vend,
366 int node, gfp_t gfp_mask)
368 struct vmap_area *va;
372 struct vmap_area *first;
375 BUG_ON(offset_in_page(size));
376 BUG_ON(!is_power_of_2(align));
378 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
380 va = kmalloc_node(sizeof(struct vmap_area),
381 gfp_mask & GFP_RECLAIM_MASK, node);
383 return ERR_PTR(-ENOMEM);
386 * Only scan the relevant parts containing pointers to other objects
387 * to avoid false negatives.
389 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
392 spin_lock(&vmap_area_lock);
394 * Invalidate cache if we have more permissive parameters.
395 * cached_hole_size notes the largest hole noticed _below_
396 * the vmap_area cached in free_vmap_cache: if size fits
397 * into that hole, we want to scan from vstart to reuse
398 * the hole instead of allocating above free_vmap_cache.
399 * Note that __free_vmap_area may update free_vmap_cache
400 * without updating cached_hole_size or cached_align.
402 if (!free_vmap_cache ||
403 size < cached_hole_size ||
404 vstart < cached_vstart ||
405 align < cached_align) {
407 cached_hole_size = 0;
408 free_vmap_cache = NULL;
410 /* record if we encounter less permissive parameters */
411 cached_vstart = vstart;
412 cached_align = align;
414 /* find starting point for our search */
415 if (free_vmap_cache) {
416 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
417 addr = ALIGN(first->va_end, align);
420 if (addr + size < addr)
424 addr = ALIGN(vstart, align);
425 if (addr + size < addr)
428 n = vmap_area_root.rb_node;
432 struct vmap_area *tmp;
433 tmp = rb_entry(n, struct vmap_area, rb_node);
434 if (tmp->va_end >= addr) {
436 if (tmp->va_start <= addr)
447 /* from the starting point, walk areas until a suitable hole is found */
448 while (addr + size > first->va_start && addr + size <= vend) {
449 if (addr + cached_hole_size < first->va_start)
450 cached_hole_size = first->va_start - addr;
451 addr = ALIGN(first->va_end, align);
452 if (addr + size < addr)
455 if (list_is_last(&first->list, &vmap_area_list))
458 first = list_next_entry(first, list);
463 * Check also calculated address against the vstart,
464 * because it can be 0 because of big align request.
466 if (addr + size > vend || addr < vstart)
470 va->va_end = addr + size;
472 __insert_vmap_area(va);
473 free_vmap_cache = &va->rb_node;
474 spin_unlock(&vmap_area_lock);
476 BUG_ON(!IS_ALIGNED(va->va_start, align));
477 BUG_ON(va->va_start < vstart);
478 BUG_ON(va->va_end > vend);
483 spin_unlock(&vmap_area_lock);
485 purge_vmap_area_lazy();
490 if (gfpflags_allow_blocking(gfp_mask)) {
491 unsigned long freed = 0;
492 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
499 if (printk_ratelimit())
500 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
503 return ERR_PTR(-EBUSY);
506 int register_vmap_purge_notifier(struct notifier_block *nb)
508 return blocking_notifier_chain_register(&vmap_notify_list, nb);
510 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
512 int unregister_vmap_purge_notifier(struct notifier_block *nb)
514 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
516 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
518 static void __free_vmap_area(struct vmap_area *va)
520 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
522 if (free_vmap_cache) {
523 if (va->va_end < cached_vstart) {
524 free_vmap_cache = NULL;
526 struct vmap_area *cache;
527 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
528 if (va->va_start <= cache->va_start) {
529 free_vmap_cache = rb_prev(&va->rb_node);
531 * We don't try to update cached_hole_size or
532 * cached_align, but it won't go very wrong.
537 rb_erase(&va->rb_node, &vmap_area_root);
538 RB_CLEAR_NODE(&va->rb_node);
539 list_del_rcu(&va->list);
542 * Track the highest possible candidate for pcpu area
543 * allocation. Areas outside of vmalloc area can be returned
544 * here too, consider only end addresses which fall inside
545 * vmalloc area proper.
547 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
548 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
550 kfree_rcu(va, rcu_head);
554 * Free a region of KVA allocated by alloc_vmap_area
556 static void free_vmap_area(struct vmap_area *va)
558 spin_lock(&vmap_area_lock);
559 __free_vmap_area(va);
560 spin_unlock(&vmap_area_lock);
564 * Clear the pagetable entries of a given vmap_area
566 static void unmap_vmap_area(struct vmap_area *va)
568 vunmap_page_range(va->va_start, va->va_end);
571 static void vmap_debug_free_range(unsigned long start, unsigned long end)
574 * Unmap page tables and force a TLB flush immediately if pagealloc
575 * debugging is enabled. This catches use after free bugs similarly to
576 * those in linear kernel virtual address space after a page has been
579 * All the lazy freeing logic is still retained, in order to minimise
580 * intrusiveness of this debugging feature.
582 * This is going to be *slow* (linear kernel virtual address debugging
583 * doesn't do a broadcast TLB flush so it is a lot faster).
585 if (debug_pagealloc_enabled()) {
586 vunmap_page_range(start, end);
587 flush_tlb_kernel_range(start, end);
592 * lazy_max_pages is the maximum amount of virtual address space we gather up
593 * before attempting to purge with a TLB flush.
595 * There is a tradeoff here: a larger number will cover more kernel page tables
596 * and take slightly longer to purge, but it will linearly reduce the number of
597 * global TLB flushes that must be performed. It would seem natural to scale
598 * this number up linearly with the number of CPUs (because vmapping activity
599 * could also scale linearly with the number of CPUs), however it is likely
600 * that in practice, workloads might be constrained in other ways that mean
601 * vmap activity will not scale linearly with CPUs. Also, I want to be
602 * conservative and not introduce a big latency on huge systems, so go with
603 * a less aggressive log scale. It will still be an improvement over the old
604 * code, and it will be simple to change the scale factor if we find that it
605 * becomes a problem on bigger systems.
607 static unsigned long lazy_max_pages(void)
611 log = fls(num_online_cpus());
613 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
616 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
618 /* for per-CPU blocks */
619 static void purge_fragmented_blocks_allcpus(void);
622 * called before a call to iounmap() if the caller wants vm_area_struct's
625 void set_iounmap_nonlazy(void)
627 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
631 * Purges all lazily-freed vmap areas.
633 * If sync is 0 then don't purge if there is already a purge in progress.
634 * If force_flush is 1, then flush kernel TLBs between *start and *end even
635 * if we found no lazy vmap areas to unmap (callers can use this to optimise
636 * their own TLB flushing).
637 * Returns with *start = min(*start, lowest purged address)
638 * *end = max(*end, highest purged address)
640 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
641 int sync, int force_flush)
643 static DEFINE_SPINLOCK(purge_lock);
644 struct llist_node *valist;
645 struct vmap_area *va;
646 struct vmap_area *n_va;
650 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
651 * should not expect such behaviour. This just simplifies locking for
652 * the case that isn't actually used at the moment anyway.
654 if (!sync && !force_flush) {
655 if (!spin_trylock(&purge_lock))
658 spin_lock(&purge_lock);
661 purge_fragmented_blocks_allcpus();
663 valist = llist_del_all(&vmap_purge_list);
664 llist_for_each_entry(va, valist, purge_list) {
665 if (va->va_start < *start)
666 *start = va->va_start;
667 if (va->va_end > *end)
669 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
673 atomic_sub(nr, &vmap_lazy_nr);
675 if (nr || force_flush)
676 flush_tlb_kernel_range(*start, *end);
679 spin_lock(&vmap_area_lock);
680 llist_for_each_entry_safe(va, n_va, valist, purge_list)
681 __free_vmap_area(va);
682 spin_unlock(&vmap_area_lock);
684 spin_unlock(&purge_lock);
688 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
689 * is already purging.
691 static void try_purge_vmap_area_lazy(void)
693 unsigned long start = ULONG_MAX, end = 0;
695 __purge_vmap_area_lazy(&start, &end, 0, 0);
699 * Kick off a purge of the outstanding lazy areas.
701 static void purge_vmap_area_lazy(void)
703 unsigned long start = ULONG_MAX, end = 0;
705 __purge_vmap_area_lazy(&start, &end, 1, 0);
709 * Free a vmap area, caller ensuring that the area has been unmapped
710 * and flush_cache_vunmap had been called for the correct range
713 static void free_vmap_area_noflush(struct vmap_area *va)
717 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
720 /* After this point, we may free va at any time */
721 llist_add(&va->purge_list, &vmap_purge_list);
723 if (unlikely(nr_lazy > lazy_max_pages()))
724 try_purge_vmap_area_lazy();
728 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
729 * called for the correct range previously.
731 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
734 free_vmap_area_noflush(va);
738 * Free and unmap a vmap area
740 static void free_unmap_vmap_area(struct vmap_area *va)
742 flush_cache_vunmap(va->va_start, va->va_end);
743 free_unmap_vmap_area_noflush(va);
746 static struct vmap_area *find_vmap_area(unsigned long addr)
748 struct vmap_area *va;
750 spin_lock(&vmap_area_lock);
751 va = __find_vmap_area(addr);
752 spin_unlock(&vmap_area_lock);
757 static void free_unmap_vmap_area_addr(unsigned long addr)
759 struct vmap_area *va;
761 va = find_vmap_area(addr);
763 free_unmap_vmap_area(va);
767 /*** Per cpu kva allocator ***/
770 * vmap space is limited especially on 32 bit architectures. Ensure there is
771 * room for at least 16 percpu vmap blocks per CPU.
774 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
775 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
776 * instead (we just need a rough idea)
778 #if BITS_PER_LONG == 32
779 #define VMALLOC_SPACE (128UL*1024*1024)
781 #define VMALLOC_SPACE (128UL*1024*1024*1024)
784 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
785 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
786 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
787 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
788 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
789 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
790 #define VMAP_BBMAP_BITS \
791 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
792 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
793 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
795 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
797 static bool vmap_initialized __read_mostly = false;
799 struct vmap_block_queue {
801 struct list_head free;
806 struct vmap_area *va;
807 unsigned long free, dirty;
808 unsigned long dirty_min, dirty_max; /*< dirty range */
809 struct list_head free_list;
810 struct rcu_head rcu_head;
811 struct list_head purge;
814 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
815 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
818 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
819 * in the free path. Could get rid of this if we change the API to return a
820 * "cookie" from alloc, to be passed to free. But no big deal yet.
822 static DEFINE_SPINLOCK(vmap_block_tree_lock);
823 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
826 * We should probably have a fallback mechanism to allocate virtual memory
827 * out of partially filled vmap blocks. However vmap block sizing should be
828 * fairly reasonable according to the vmalloc size, so it shouldn't be a
832 static unsigned long addr_to_vb_idx(unsigned long addr)
834 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
835 addr /= VMAP_BLOCK_SIZE;
839 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
843 addr = va_start + (pages_off << PAGE_SHIFT);
844 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
849 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
850 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
851 * @order: how many 2^order pages should be occupied in newly allocated block
852 * @gfp_mask: flags for the page level allocator
854 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
856 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
858 struct vmap_block_queue *vbq;
859 struct vmap_block *vb;
860 struct vmap_area *va;
861 unsigned long vb_idx;
865 node = numa_node_id();
867 vb = kmalloc_node(sizeof(struct vmap_block),
868 gfp_mask & GFP_RECLAIM_MASK, node);
870 return ERR_PTR(-ENOMEM);
872 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
873 VMALLOC_START, VMALLOC_END,
880 err = radix_tree_preload(gfp_mask);
887 vaddr = vmap_block_vaddr(va->va_start, 0);
888 spin_lock_init(&vb->lock);
890 /* At least something should be left free */
891 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
892 vb->free = VMAP_BBMAP_BITS - (1UL << order);
894 vb->dirty_min = VMAP_BBMAP_BITS;
896 INIT_LIST_HEAD(&vb->free_list);
898 vb_idx = addr_to_vb_idx(va->va_start);
899 spin_lock(&vmap_block_tree_lock);
900 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
901 spin_unlock(&vmap_block_tree_lock);
903 radix_tree_preload_end();
905 vbq = &get_cpu_var(vmap_block_queue);
906 spin_lock(&vbq->lock);
907 list_add_tail_rcu(&vb->free_list, &vbq->free);
908 spin_unlock(&vbq->lock);
909 put_cpu_var(vmap_block_queue);
914 static void free_vmap_block(struct vmap_block *vb)
916 struct vmap_block *tmp;
917 unsigned long vb_idx;
919 vb_idx = addr_to_vb_idx(vb->va->va_start);
920 spin_lock(&vmap_block_tree_lock);
921 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
922 spin_unlock(&vmap_block_tree_lock);
925 free_vmap_area_noflush(vb->va);
926 kfree_rcu(vb, rcu_head);
929 static void purge_fragmented_blocks(int cpu)
932 struct vmap_block *vb;
933 struct vmap_block *n_vb;
934 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
937 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
939 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
942 spin_lock(&vb->lock);
943 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
944 vb->free = 0; /* prevent further allocs after releasing lock */
945 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
947 vb->dirty_max = VMAP_BBMAP_BITS;
948 spin_lock(&vbq->lock);
949 list_del_rcu(&vb->free_list);
950 spin_unlock(&vbq->lock);
951 spin_unlock(&vb->lock);
952 list_add_tail(&vb->purge, &purge);
954 spin_unlock(&vb->lock);
958 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
959 list_del(&vb->purge);
964 static void purge_fragmented_blocks_allcpus(void)
968 for_each_possible_cpu(cpu)
969 purge_fragmented_blocks(cpu);
972 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
974 struct vmap_block_queue *vbq;
975 struct vmap_block *vb;
979 BUG_ON(offset_in_page(size));
980 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
981 if (WARN_ON(size == 0)) {
983 * Allocating 0 bytes isn't what caller wants since
984 * get_order(0) returns funny result. Just warn and terminate
989 order = get_order(size);
992 vbq = &get_cpu_var(vmap_block_queue);
993 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
994 unsigned long pages_off;
996 spin_lock(&vb->lock);
997 if (vb->free < (1UL << order)) {
998 spin_unlock(&vb->lock);
1002 pages_off = VMAP_BBMAP_BITS - vb->free;
1003 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
1004 vb->free -= 1UL << order;
1005 if (vb->free == 0) {
1006 spin_lock(&vbq->lock);
1007 list_del_rcu(&vb->free_list);
1008 spin_unlock(&vbq->lock);
1011 spin_unlock(&vb->lock);
1015 put_cpu_var(vmap_block_queue);
1018 /* Allocate new block if nothing was found */
1020 vaddr = new_vmap_block(order, gfp_mask);
1025 static void vb_free(const void *addr, unsigned long size)
1027 unsigned long offset;
1028 unsigned long vb_idx;
1030 struct vmap_block *vb;
1032 BUG_ON(offset_in_page(size));
1033 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1035 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1037 order = get_order(size);
1039 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1040 offset >>= PAGE_SHIFT;
1042 vb_idx = addr_to_vb_idx((unsigned long)addr);
1044 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1048 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1050 spin_lock(&vb->lock);
1052 /* Expand dirty range */
1053 vb->dirty_min = min(vb->dirty_min, offset);
1054 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1056 vb->dirty += 1UL << order;
1057 if (vb->dirty == VMAP_BBMAP_BITS) {
1059 spin_unlock(&vb->lock);
1060 free_vmap_block(vb);
1062 spin_unlock(&vb->lock);
1066 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1068 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1069 * to amortize TLB flushing overheads. What this means is that any page you
1070 * have now, may, in a former life, have been mapped into kernel virtual
1071 * address by the vmap layer and so there might be some CPUs with TLB entries
1072 * still referencing that page (additional to the regular 1:1 kernel mapping).
1074 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1075 * be sure that none of the pages we have control over will have any aliases
1076 * from the vmap layer.
1078 void vm_unmap_aliases(void)
1080 unsigned long start = ULONG_MAX, end = 0;
1084 if (unlikely(!vmap_initialized))
1087 for_each_possible_cpu(cpu) {
1088 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1089 struct vmap_block *vb;
1092 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1093 spin_lock(&vb->lock);
1095 unsigned long va_start = vb->va->va_start;
1098 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1099 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1101 start = min(s, start);
1106 spin_unlock(&vb->lock);
1111 __purge_vmap_area_lazy(&start, &end, 1, flush);
1113 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1116 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1117 * @mem: the pointer returned by vm_map_ram
1118 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1120 void vm_unmap_ram(const void *mem, unsigned int count)
1122 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1123 unsigned long addr = (unsigned long)mem;
1126 BUG_ON(addr < VMALLOC_START);
1127 BUG_ON(addr > VMALLOC_END);
1128 BUG_ON(!PAGE_ALIGNED(addr));
1130 debug_check_no_locks_freed(mem, size);
1131 vmap_debug_free_range(addr, addr+size);
1133 if (likely(count <= VMAP_MAX_ALLOC))
1136 free_unmap_vmap_area_addr(addr);
1138 EXPORT_SYMBOL(vm_unmap_ram);
1141 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1142 * @pages: an array of pointers to the pages to be mapped
1143 * @count: number of pages
1144 * @node: prefer to allocate data structures on this node
1145 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1147 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1148 * faster than vmap so it's good. But if you mix long-life and short-life
1149 * objects with vm_map_ram(), it could consume lots of address space through
1150 * fragmentation (especially on a 32bit machine). You could see failures in
1151 * the end. Please use this function for short-lived objects.
1153 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1155 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1157 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1161 if (likely(count <= VMAP_MAX_ALLOC)) {
1162 mem = vb_alloc(size, GFP_KERNEL);
1165 addr = (unsigned long)mem;
1167 struct vmap_area *va;
1168 va = alloc_vmap_area(size, PAGE_SIZE,
1169 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1173 addr = va->va_start;
1176 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1177 vm_unmap_ram(mem, count);
1182 EXPORT_SYMBOL(vm_map_ram);
1184 static struct vm_struct *vmlist __initdata;
1186 * vm_area_add_early - add vmap area early during boot
1187 * @vm: vm_struct to add
1189 * This function is used to add fixed kernel vm area to vmlist before
1190 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1191 * should contain proper values and the other fields should be zero.
1193 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1195 void __init vm_area_add_early(struct vm_struct *vm)
1197 struct vm_struct *tmp, **p;
1199 BUG_ON(vmap_initialized);
1200 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1201 if (tmp->addr >= vm->addr) {
1202 BUG_ON(tmp->addr < vm->addr + vm->size);
1205 BUG_ON(tmp->addr + tmp->size > vm->addr);
1212 * vm_area_register_early - register vmap area early during boot
1213 * @vm: vm_struct to register
1214 * @align: requested alignment
1216 * This function is used to register kernel vm area before
1217 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1218 * proper values on entry and other fields should be zero. On return,
1219 * vm->addr contains the allocated address.
1221 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1223 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1225 static size_t vm_init_off __initdata;
1228 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1229 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1231 vm->addr = (void *)addr;
1233 vm_area_add_early(vm);
1236 void __init vmalloc_init(void)
1238 struct vmap_area *va;
1239 struct vm_struct *tmp;
1242 for_each_possible_cpu(i) {
1243 struct vmap_block_queue *vbq;
1244 struct vfree_deferred *p;
1246 vbq = &per_cpu(vmap_block_queue, i);
1247 spin_lock_init(&vbq->lock);
1248 INIT_LIST_HEAD(&vbq->free);
1249 p = &per_cpu(vfree_deferred, i);
1250 init_llist_head(&p->list);
1251 INIT_WORK(&p->wq, free_work);
1254 /* Import existing vmlist entries. */
1255 for (tmp = vmlist; tmp; tmp = tmp->next) {
1256 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1257 va->flags = VM_VM_AREA;
1258 va->va_start = (unsigned long)tmp->addr;
1259 va->va_end = va->va_start + tmp->size;
1261 __insert_vmap_area(va);
1264 vmap_area_pcpu_hole = VMALLOC_END;
1266 vmap_initialized = true;
1270 * map_kernel_range_noflush - map kernel VM area with the specified pages
1271 * @addr: start of the VM area to map
1272 * @size: size of the VM area to map
1273 * @prot: page protection flags to use
1274 * @pages: pages to map
1276 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1277 * specify should have been allocated using get_vm_area() and its
1281 * This function does NOT do any cache flushing. The caller is
1282 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1283 * before calling this function.
1286 * The number of pages mapped on success, -errno on failure.
1288 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1289 pgprot_t prot, struct page **pages)
1291 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1295 * unmap_kernel_range_noflush - unmap kernel VM area
1296 * @addr: start of the VM area to unmap
1297 * @size: size of the VM area to unmap
1299 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1300 * specify should have been allocated using get_vm_area() and its
1304 * This function does NOT do any cache flushing. The caller is
1305 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1306 * before calling this function and flush_tlb_kernel_range() after.
1308 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1310 vunmap_page_range(addr, addr + size);
1312 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1315 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1316 * @addr: start of the VM area to unmap
1317 * @size: size of the VM area to unmap
1319 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1320 * the unmapping and tlb after.
1322 void unmap_kernel_range(unsigned long addr, unsigned long size)
1324 unsigned long end = addr + size;
1326 flush_cache_vunmap(addr, end);
1327 vunmap_page_range(addr, end);
1328 flush_tlb_kernel_range(addr, end);
1330 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1332 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1334 unsigned long addr = (unsigned long)area->addr;
1335 unsigned long end = addr + get_vm_area_size(area);
1338 err = vmap_page_range(addr, end, prot, pages);
1340 return err > 0 ? 0 : err;
1342 EXPORT_SYMBOL_GPL(map_vm_area);
1344 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1345 unsigned long flags, const void *caller)
1347 spin_lock(&vmap_area_lock);
1349 vm->addr = (void *)va->va_start;
1350 vm->size = va->va_end - va->va_start;
1351 vm->caller = caller;
1353 va->flags |= VM_VM_AREA;
1354 spin_unlock(&vmap_area_lock);
1357 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1360 * Before removing VM_UNINITIALIZED,
1361 * we should make sure that vm has proper values.
1362 * Pair with smp_rmb() in show_numa_info().
1365 vm->flags &= ~VM_UNINITIALIZED;
1368 static struct vm_struct *__get_vm_area_node(unsigned long size,
1369 unsigned long align, unsigned long flags, unsigned long start,
1370 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1372 struct vmap_area *va;
1373 struct vm_struct *area;
1375 BUG_ON(in_interrupt());
1376 size = PAGE_ALIGN(size);
1377 if (unlikely(!size))
1380 if (flags & VM_IOREMAP)
1381 align = 1ul << clamp_t(int, get_count_order_long(size),
1382 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1384 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1385 if (unlikely(!area))
1388 if (!(flags & VM_NO_GUARD))
1391 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1397 setup_vmalloc_vm(area, va, flags, caller);
1402 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1403 unsigned long start, unsigned long end)
1405 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1406 GFP_KERNEL, __builtin_return_address(0));
1408 EXPORT_SYMBOL_GPL(__get_vm_area);
1410 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1411 unsigned long start, unsigned long end,
1414 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1415 GFP_KERNEL, caller);
1419 * get_vm_area - reserve a contiguous kernel virtual area
1420 * @size: size of the area
1421 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1423 * Search an area of @size in the kernel virtual mapping area,
1424 * and reserved it for out purposes. Returns the area descriptor
1425 * on success or %NULL on failure.
1427 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1429 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1430 NUMA_NO_NODE, GFP_KERNEL,
1431 __builtin_return_address(0));
1434 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1437 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1438 NUMA_NO_NODE, GFP_KERNEL, caller);
1442 * find_vm_area - find a continuous kernel virtual area
1443 * @addr: base address
1445 * Search for the kernel VM area starting at @addr, and return it.
1446 * It is up to the caller to do all required locking to keep the returned
1449 struct vm_struct *find_vm_area(const void *addr)
1451 struct vmap_area *va;
1453 va = find_vmap_area((unsigned long)addr);
1454 if (va && va->flags & VM_VM_AREA)
1461 * remove_vm_area - find and remove a continuous kernel virtual area
1462 * @addr: base address
1464 * Search for the kernel VM area starting at @addr, and remove it.
1465 * This function returns the found VM area, but using it is NOT safe
1466 * on SMP machines, except for its size or flags.
1468 struct vm_struct *remove_vm_area(const void *addr)
1470 struct vmap_area *va;
1472 va = find_vmap_area((unsigned long)addr);
1473 if (va && va->flags & VM_VM_AREA) {
1474 struct vm_struct *vm = va->vm;
1476 spin_lock(&vmap_area_lock);
1478 va->flags &= ~VM_VM_AREA;
1479 spin_unlock(&vmap_area_lock);
1481 vmap_debug_free_range(va->va_start, va->va_end);
1482 kasan_free_shadow(vm);
1483 free_unmap_vmap_area(va);
1490 static void __vunmap(const void *addr, int deallocate_pages)
1492 struct vm_struct *area;
1497 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1501 area = find_vmap_area((unsigned long)addr)->vm;
1502 if (unlikely(!area)) {
1503 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1508 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1509 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1511 remove_vm_area(addr);
1512 if (deallocate_pages) {
1515 for (i = 0; i < area->nr_pages; i++) {
1516 struct page *page = area->pages[i];
1519 __free_pages(page, 0);
1522 kvfree(area->pages);
1530 * vfree - release memory allocated by vmalloc()
1531 * @addr: memory base address
1533 * Free the virtually continuous memory area starting at @addr, as
1534 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1535 * NULL, no operation is performed.
1537 * Must not be called in NMI context (strictly speaking, only if we don't
1538 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1539 * conventions for vfree() arch-depenedent would be a really bad idea)
1541 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1543 void vfree(const void *addr)
1547 kmemleak_free(addr);
1551 if (unlikely(in_interrupt())) {
1552 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1553 if (llist_add((struct llist_node *)addr, &p->list))
1554 schedule_work(&p->wq);
1558 EXPORT_SYMBOL(vfree);
1561 * vunmap - release virtual mapping obtained by vmap()
1562 * @addr: memory base address
1564 * Free the virtually contiguous memory area starting at @addr,
1565 * which was created from the page array passed to vmap().
1567 * Must not be called in interrupt context.
1569 void vunmap(const void *addr)
1571 BUG_ON(in_interrupt());
1576 EXPORT_SYMBOL(vunmap);
1579 * vmap - map an array of pages into virtually contiguous space
1580 * @pages: array of page pointers
1581 * @count: number of pages to map
1582 * @flags: vm_area->flags
1583 * @prot: page protection for the mapping
1585 * Maps @count pages from @pages into contiguous kernel virtual
1588 void *vmap(struct page **pages, unsigned int count,
1589 unsigned long flags, pgprot_t prot)
1591 struct vm_struct *area;
1592 unsigned long size; /* In bytes */
1596 if (count > totalram_pages)
1599 size = (unsigned long)count << PAGE_SHIFT;
1600 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1604 if (map_vm_area(area, prot, pages)) {
1611 EXPORT_SYMBOL(vmap);
1613 static void *__vmalloc_node(unsigned long size, unsigned long align,
1614 gfp_t gfp_mask, pgprot_t prot,
1615 int node, const void *caller);
1616 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1617 pgprot_t prot, int node)
1619 struct page **pages;
1620 unsigned int nr_pages, array_size, i;
1621 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1622 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1624 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1625 array_size = (nr_pages * sizeof(struct page *));
1627 area->nr_pages = nr_pages;
1628 /* Please note that the recursion is strictly bounded. */
1629 if (array_size > PAGE_SIZE) {
1630 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1631 PAGE_KERNEL, node, area->caller);
1633 pages = kmalloc_node(array_size, nested_gfp, node);
1635 area->pages = pages;
1637 remove_vm_area(area->addr);
1642 for (i = 0; i < area->nr_pages; i++) {
1645 if (node == NUMA_NO_NODE)
1646 page = alloc_page(alloc_mask);
1648 page = alloc_pages_node(node, alloc_mask, 0);
1650 if (unlikely(!page)) {
1651 /* Successfully allocated i pages, free them in __vunmap() */
1655 area->pages[i] = page;
1656 if (gfpflags_allow_blocking(gfp_mask))
1660 if (map_vm_area(area, prot, pages))
1665 warn_alloc(gfp_mask,
1666 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1667 (area->nr_pages*PAGE_SIZE), area->size);
1673 * __vmalloc_node_range - allocate virtually contiguous memory
1674 * @size: allocation size
1675 * @align: desired alignment
1676 * @start: vm area range start
1677 * @end: vm area range end
1678 * @gfp_mask: flags for the page level allocator
1679 * @prot: protection mask for the allocated pages
1680 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1681 * @node: node to use for allocation or NUMA_NO_NODE
1682 * @caller: caller's return address
1684 * Allocate enough pages to cover @size from the page level
1685 * allocator with @gfp_mask flags. Map them into contiguous
1686 * kernel virtual space, using a pagetable protection of @prot.
1688 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1689 unsigned long start, unsigned long end, gfp_t gfp_mask,
1690 pgprot_t prot, unsigned long vm_flags, int node,
1693 struct vm_struct *area;
1695 unsigned long real_size = size;
1697 size = PAGE_ALIGN(size);
1698 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1701 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1702 vm_flags, start, end, node, gfp_mask, caller);
1706 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1711 * First make sure the mappings are removed from all page-tables
1712 * before they are freed.
1717 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1718 * flag. It means that vm_struct is not fully initialized.
1719 * Now, it is fully initialized, so remove this flag here.
1721 clear_vm_uninitialized_flag(area);
1724 * A ref_count = 2 is needed because vm_struct allocated in
1725 * __get_vm_area_node() contains a reference to the virtual address of
1726 * the vmalloc'ed block.
1728 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1733 warn_alloc(gfp_mask,
1734 "vmalloc: allocation failure: %lu bytes", real_size);
1739 * __vmalloc_node - allocate virtually contiguous memory
1740 * @size: allocation size
1741 * @align: desired alignment
1742 * @gfp_mask: flags for the page level allocator
1743 * @prot: protection mask for the allocated pages
1744 * @node: node to use for allocation or NUMA_NO_NODE
1745 * @caller: caller's return address
1747 * Allocate enough pages to cover @size from the page level
1748 * allocator with @gfp_mask flags. Map them into contiguous
1749 * kernel virtual space, using a pagetable protection of @prot.
1751 static void *__vmalloc_node(unsigned long size, unsigned long align,
1752 gfp_t gfp_mask, pgprot_t prot,
1753 int node, const void *caller)
1755 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1756 gfp_mask, prot, 0, node, caller);
1759 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1761 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1762 __builtin_return_address(0));
1764 EXPORT_SYMBOL(__vmalloc);
1766 static inline void *__vmalloc_node_flags(unsigned long size,
1767 int node, gfp_t flags)
1769 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1770 node, __builtin_return_address(0));
1774 * vmalloc - allocate virtually contiguous memory
1775 * @size: allocation size
1776 * Allocate enough pages to cover @size from the page level
1777 * allocator and map them into contiguous kernel virtual space.
1779 * For tight control over page level allocator and protection flags
1780 * use __vmalloc() instead.
1782 void *vmalloc(unsigned long size)
1784 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1785 GFP_KERNEL | __GFP_HIGHMEM);
1787 EXPORT_SYMBOL(vmalloc);
1790 * vzalloc - allocate virtually contiguous memory with zero fill
1791 * @size: allocation size
1792 * Allocate enough pages to cover @size from the page level
1793 * allocator and map them into contiguous kernel virtual space.
1794 * The memory allocated is set to zero.
1796 * For tight control over page level allocator and protection flags
1797 * use __vmalloc() instead.
1799 void *vzalloc(unsigned long size)
1801 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1802 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1804 EXPORT_SYMBOL(vzalloc);
1807 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1808 * @size: allocation size
1810 * The resulting memory area is zeroed so it can be mapped to userspace
1811 * without leaking data.
1813 void *vmalloc_user(unsigned long size)
1815 struct vm_struct *area;
1818 ret = __vmalloc_node(size, SHMLBA,
1819 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1820 PAGE_KERNEL, NUMA_NO_NODE,
1821 __builtin_return_address(0));
1823 area = find_vm_area(ret);
1824 area->flags |= VM_USERMAP;
1828 EXPORT_SYMBOL(vmalloc_user);
1831 * vmalloc_node - allocate memory on a specific node
1832 * @size: allocation size
1835 * Allocate enough pages to cover @size from the page level
1836 * allocator and map them into contiguous kernel virtual space.
1838 * For tight control over page level allocator and protection flags
1839 * use __vmalloc() instead.
1841 void *vmalloc_node(unsigned long size, int node)
1843 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1844 node, __builtin_return_address(0));
1846 EXPORT_SYMBOL(vmalloc_node);
1849 * vzalloc_node - allocate memory on a specific node with zero fill
1850 * @size: allocation size
1853 * Allocate enough pages to cover @size from the page level
1854 * allocator and map them into contiguous kernel virtual space.
1855 * The memory allocated is set to zero.
1857 * For tight control over page level allocator and protection flags
1858 * use __vmalloc_node() instead.
1860 void *vzalloc_node(unsigned long size, int node)
1862 return __vmalloc_node_flags(size, node,
1863 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1865 EXPORT_SYMBOL(vzalloc_node);
1867 #ifndef PAGE_KERNEL_EXEC
1868 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1872 * vmalloc_exec - allocate virtually contiguous, executable memory
1873 * @size: allocation size
1875 * Kernel-internal function to allocate enough pages to cover @size
1876 * the page level allocator and map them into contiguous and
1877 * executable kernel virtual space.
1879 * For tight control over page level allocator and protection flags
1880 * use __vmalloc() instead.
1883 void *vmalloc_exec(unsigned long size)
1885 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1886 NUMA_NO_NODE, __builtin_return_address(0));
1889 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1890 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1891 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1892 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1894 #define GFP_VMALLOC32 GFP_KERNEL
1898 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1899 * @size: allocation size
1901 * Allocate enough 32bit PA addressable pages to cover @size from the
1902 * page level allocator and map them into contiguous kernel virtual space.
1904 void *vmalloc_32(unsigned long size)
1906 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1907 NUMA_NO_NODE, __builtin_return_address(0));
1909 EXPORT_SYMBOL(vmalloc_32);
1912 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1913 * @size: allocation size
1915 * The resulting memory area is 32bit addressable and zeroed so it can be
1916 * mapped to userspace without leaking data.
1918 void *vmalloc_32_user(unsigned long size)
1920 struct vm_struct *area;
1923 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1924 NUMA_NO_NODE, __builtin_return_address(0));
1926 area = find_vm_area(ret);
1927 area->flags |= VM_USERMAP;
1931 EXPORT_SYMBOL(vmalloc_32_user);
1934 * small helper routine , copy contents to buf from addr.
1935 * If the page is not present, fill zero.
1938 static int aligned_vread(char *buf, char *addr, unsigned long count)
1944 unsigned long offset, length;
1946 offset = offset_in_page(addr);
1947 length = PAGE_SIZE - offset;
1950 p = vmalloc_to_page(addr);
1952 * To do safe access to this _mapped_ area, we need
1953 * lock. But adding lock here means that we need to add
1954 * overhead of vmalloc()/vfree() calles for this _debug_
1955 * interface, rarely used. Instead of that, we'll use
1956 * kmap() and get small overhead in this access function.
1960 * we can expect USER0 is not used (see vread/vwrite's
1961 * function description)
1963 void *map = kmap_atomic(p);
1964 memcpy(buf, map + offset, length);
1967 memset(buf, 0, length);
1977 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1983 unsigned long offset, length;
1985 offset = offset_in_page(addr);
1986 length = PAGE_SIZE - offset;
1989 p = vmalloc_to_page(addr);
1991 * To do safe access to this _mapped_ area, we need
1992 * lock. But adding lock here means that we need to add
1993 * overhead of vmalloc()/vfree() calles for this _debug_
1994 * interface, rarely used. Instead of that, we'll use
1995 * kmap() and get small overhead in this access function.
1999 * we can expect USER0 is not used (see vread/vwrite's
2000 * function description)
2002 void *map = kmap_atomic(p);
2003 memcpy(map + offset, buf, length);
2015 * vread() - read vmalloc area in a safe way.
2016 * @buf: buffer for reading data
2017 * @addr: vm address.
2018 * @count: number of bytes to be read.
2020 * Returns # of bytes which addr and buf should be increased.
2021 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2022 * includes any intersect with alive vmalloc area.
2024 * This function checks that addr is a valid vmalloc'ed area, and
2025 * copy data from that area to a given buffer. If the given memory range
2026 * of [addr...addr+count) includes some valid address, data is copied to
2027 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2028 * IOREMAP area is treated as memory hole and no copy is done.
2030 * If [addr...addr+count) doesn't includes any intersects with alive
2031 * vm_struct area, returns 0. @buf should be kernel's buffer.
2033 * Note: In usual ops, vread() is never necessary because the caller
2034 * should know vmalloc() area is valid and can use memcpy().
2035 * This is for routines which have to access vmalloc area without
2036 * any informaion, as /dev/kmem.
2040 long vread(char *buf, char *addr, unsigned long count)
2042 struct vmap_area *va;
2043 struct vm_struct *vm;
2044 char *vaddr, *buf_start = buf;
2045 unsigned long buflen = count;
2048 /* Don't allow overflow */
2049 if ((unsigned long) addr + count < count)
2050 count = -(unsigned long) addr;
2052 spin_lock(&vmap_area_lock);
2053 list_for_each_entry(va, &vmap_area_list, list) {
2057 if (!(va->flags & VM_VM_AREA))
2061 vaddr = (char *) vm->addr;
2062 if (addr >= vaddr + get_vm_area_size(vm))
2064 while (addr < vaddr) {
2072 n = vaddr + get_vm_area_size(vm) - addr;
2075 if (!(vm->flags & VM_IOREMAP))
2076 aligned_vread(buf, addr, n);
2077 else /* IOREMAP area is treated as memory hole */
2084 spin_unlock(&vmap_area_lock);
2086 if (buf == buf_start)
2088 /* zero-fill memory holes */
2089 if (buf != buf_start + buflen)
2090 memset(buf, 0, buflen - (buf - buf_start));
2096 * vwrite() - write vmalloc area in a safe way.
2097 * @buf: buffer for source data
2098 * @addr: vm address.
2099 * @count: number of bytes to be read.
2101 * Returns # of bytes which addr and buf should be incresed.
2102 * (same number to @count).
2103 * If [addr...addr+count) doesn't includes any intersect with valid
2104 * vmalloc area, returns 0.
2106 * This function checks that addr is a valid vmalloc'ed area, and
2107 * copy data from a buffer to the given addr. If specified range of
2108 * [addr...addr+count) includes some valid address, data is copied from
2109 * proper area of @buf. If there are memory holes, no copy to hole.
2110 * IOREMAP area is treated as memory hole and no copy is done.
2112 * If [addr...addr+count) doesn't includes any intersects with alive
2113 * vm_struct area, returns 0. @buf should be kernel's buffer.
2115 * Note: In usual ops, vwrite() is never necessary because the caller
2116 * should know vmalloc() area is valid and can use memcpy().
2117 * This is for routines which have to access vmalloc area without
2118 * any informaion, as /dev/kmem.
2121 long vwrite(char *buf, char *addr, unsigned long count)
2123 struct vmap_area *va;
2124 struct vm_struct *vm;
2126 unsigned long n, buflen;
2129 /* Don't allow overflow */
2130 if ((unsigned long) addr + count < count)
2131 count = -(unsigned long) addr;
2134 spin_lock(&vmap_area_lock);
2135 list_for_each_entry(va, &vmap_area_list, list) {
2139 if (!(va->flags & VM_VM_AREA))
2143 vaddr = (char *) vm->addr;
2144 if (addr >= vaddr + get_vm_area_size(vm))
2146 while (addr < vaddr) {
2153 n = vaddr + get_vm_area_size(vm) - addr;
2156 if (!(vm->flags & VM_IOREMAP)) {
2157 aligned_vwrite(buf, addr, n);
2165 spin_unlock(&vmap_area_lock);
2172 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2173 * @vma: vma to cover
2174 * @uaddr: target user address to start at
2175 * @kaddr: virtual address of vmalloc kernel memory
2176 * @size: size of map area
2178 * Returns: 0 for success, -Exxx on failure
2180 * This function checks that @kaddr is a valid vmalloc'ed area,
2181 * and that it is big enough to cover the range starting at
2182 * @uaddr in @vma. Will return failure if that criteria isn't
2185 * Similar to remap_pfn_range() (see mm/memory.c)
2187 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2188 void *kaddr, unsigned long size)
2190 struct vm_struct *area;
2192 size = PAGE_ALIGN(size);
2194 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2197 area = find_vm_area(kaddr);
2201 if (!(area->flags & VM_USERMAP))
2204 if (kaddr + size > area->addr + get_vm_area_size(area))
2208 struct page *page = vmalloc_to_page(kaddr);
2211 ret = vm_insert_page(vma, uaddr, page);
2220 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2224 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2227 * remap_vmalloc_range - map vmalloc pages to userspace
2228 * @vma: vma to cover (map full range of vma)
2229 * @addr: vmalloc memory
2230 * @pgoff: number of pages into addr before first page to map
2232 * Returns: 0 for success, -Exxx on failure
2234 * This function checks that addr is a valid vmalloc'ed area, and
2235 * that it is big enough to cover the vma. Will return failure if
2236 * that criteria isn't met.
2238 * Similar to remap_pfn_range() (see mm/memory.c)
2240 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2241 unsigned long pgoff)
2243 return remap_vmalloc_range_partial(vma, vma->vm_start,
2244 addr + (pgoff << PAGE_SHIFT),
2245 vma->vm_end - vma->vm_start);
2247 EXPORT_SYMBOL(remap_vmalloc_range);
2250 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2253 * The purpose of this function is to make sure the vmalloc area
2254 * mappings are identical in all page-tables in the system.
2256 void __weak vmalloc_sync_all(void)
2261 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2273 * alloc_vm_area - allocate a range of kernel address space
2274 * @size: size of the area
2275 * @ptes: returns the PTEs for the address space
2277 * Returns: NULL on failure, vm_struct on success
2279 * This function reserves a range of kernel address space, and
2280 * allocates pagetables to map that range. No actual mappings
2283 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2284 * allocated for the VM area are returned.
2286 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2288 struct vm_struct *area;
2290 area = get_vm_area_caller(size, VM_IOREMAP,
2291 __builtin_return_address(0));
2296 * This ensures that page tables are constructed for this region
2297 * of kernel virtual address space and mapped into init_mm.
2299 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2300 size, f, ptes ? &ptes : NULL)) {
2307 EXPORT_SYMBOL_GPL(alloc_vm_area);
2309 void free_vm_area(struct vm_struct *area)
2311 struct vm_struct *ret;
2312 ret = remove_vm_area(area->addr);
2313 BUG_ON(ret != area);
2316 EXPORT_SYMBOL_GPL(free_vm_area);
2319 static struct vmap_area *node_to_va(struct rb_node *n)
2321 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2325 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2326 * @end: target address
2327 * @pnext: out arg for the next vmap_area
2328 * @pprev: out arg for the previous vmap_area
2330 * Returns: %true if either or both of next and prev are found,
2331 * %false if no vmap_area exists
2333 * Find vmap_areas end addresses of which enclose @end. ie. if not
2334 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2336 static bool pvm_find_next_prev(unsigned long end,
2337 struct vmap_area **pnext,
2338 struct vmap_area **pprev)
2340 struct rb_node *n = vmap_area_root.rb_node;
2341 struct vmap_area *va = NULL;
2344 va = rb_entry(n, struct vmap_area, rb_node);
2345 if (end < va->va_end)
2347 else if (end > va->va_end)
2356 if (va->va_end > end) {
2358 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2361 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2367 * pvm_determine_end - find the highest aligned address between two vmap_areas
2368 * @pnext: in/out arg for the next vmap_area
2369 * @pprev: in/out arg for the previous vmap_area
2372 * Returns: determined end address
2374 * Find the highest aligned address between *@pnext and *@pprev below
2375 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2376 * down address is between the end addresses of the two vmap_areas.
2378 * Please note that the address returned by this function may fall
2379 * inside *@pnext vmap_area. The caller is responsible for checking
2382 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2383 struct vmap_area **pprev,
2384 unsigned long align)
2386 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2390 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2394 while (*pprev && (*pprev)->va_end > addr) {
2396 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2403 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2404 * @offsets: array containing offset of each area
2405 * @sizes: array containing size of each area
2406 * @nr_vms: the number of areas to allocate
2407 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2409 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2410 * vm_structs on success, %NULL on failure
2412 * Percpu allocator wants to use congruent vm areas so that it can
2413 * maintain the offsets among percpu areas. This function allocates
2414 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2415 * be scattered pretty far, distance between two areas easily going up
2416 * to gigabytes. To avoid interacting with regular vmallocs, these
2417 * areas are allocated from top.
2419 * Despite its complicated look, this allocator is rather simple. It
2420 * does everything top-down and scans areas from the end looking for
2421 * matching slot. While scanning, if any of the areas overlaps with
2422 * existing vmap_area, the base address is pulled down to fit the
2423 * area. Scanning is repeated till all the areas fit and then all
2424 * necessary data structres are inserted and the result is returned.
2426 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2427 const size_t *sizes, int nr_vms,
2430 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2431 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2432 struct vmap_area **vas, *prev, *next;
2433 struct vm_struct **vms;
2434 int area, area2, last_area, term_area;
2435 unsigned long base, start, end, last_end;
2436 bool purged = false;
2438 /* verify parameters and allocate data structures */
2439 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2440 for (last_area = 0, area = 0; area < nr_vms; area++) {
2441 start = offsets[area];
2442 end = start + sizes[area];
2444 /* is everything aligned properly? */
2445 BUG_ON(!IS_ALIGNED(offsets[area], align));
2446 BUG_ON(!IS_ALIGNED(sizes[area], align));
2448 /* detect the area with the highest address */
2449 if (start > offsets[last_area])
2452 for (area2 = 0; area2 < nr_vms; area2++) {
2453 unsigned long start2 = offsets[area2];
2454 unsigned long end2 = start2 + sizes[area2];
2459 BUG_ON(start2 >= start && start2 < end);
2460 BUG_ON(end2 <= end && end2 > start);
2463 last_end = offsets[last_area] + sizes[last_area];
2465 if (vmalloc_end - vmalloc_start < last_end) {
2470 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2471 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2475 for (area = 0; area < nr_vms; area++) {
2476 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2477 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2478 if (!vas[area] || !vms[area])
2482 spin_lock(&vmap_area_lock);
2484 /* start scanning - we scan from the top, begin with the last area */
2485 area = term_area = last_area;
2486 start = offsets[area];
2487 end = start + sizes[area];
2489 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2490 base = vmalloc_end - last_end;
2493 base = pvm_determine_end(&next, &prev, align) - end;
2496 BUG_ON(next && next->va_end <= base + end);
2497 BUG_ON(prev && prev->va_end > base + end);
2500 * base might have underflowed, add last_end before
2503 if (base + last_end < vmalloc_start + last_end) {
2504 spin_unlock(&vmap_area_lock);
2506 purge_vmap_area_lazy();
2514 * If next overlaps, move base downwards so that it's
2515 * right below next and then recheck.
2517 if (next && next->va_start < base + end) {
2518 base = pvm_determine_end(&next, &prev, align) - end;
2524 * If prev overlaps, shift down next and prev and move
2525 * base so that it's right below new next and then
2528 if (prev && prev->va_end > base + start) {
2530 prev = node_to_va(rb_prev(&next->rb_node));
2531 base = pvm_determine_end(&next, &prev, align) - end;
2537 * This area fits, move on to the previous one. If
2538 * the previous one is the terminal one, we're done.
2540 area = (area + nr_vms - 1) % nr_vms;
2541 if (area == term_area)
2543 start = offsets[area];
2544 end = start + sizes[area];
2545 pvm_find_next_prev(base + end, &next, &prev);
2548 /* we've found a fitting base, insert all va's */
2549 for (area = 0; area < nr_vms; area++) {
2550 struct vmap_area *va = vas[area];
2552 va->va_start = base + offsets[area];
2553 va->va_end = va->va_start + sizes[area];
2554 __insert_vmap_area(va);
2557 vmap_area_pcpu_hole = base + offsets[last_area];
2559 spin_unlock(&vmap_area_lock);
2561 /* insert all vm's */
2562 for (area = 0; area < nr_vms; area++)
2563 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2570 for (area = 0; area < nr_vms; area++) {
2581 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2582 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2583 * @nr_vms: the number of allocated areas
2585 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2587 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2591 for (i = 0; i < nr_vms; i++)
2592 free_vm_area(vms[i]);
2595 #endif /* CONFIG_SMP */
2597 #ifdef CONFIG_PROC_FS
2598 static void *s_start(struct seq_file *m, loff_t *pos)
2599 __acquires(&vmap_area_lock)
2602 struct vmap_area *va;
2604 spin_lock(&vmap_area_lock);
2605 va = list_first_entry(&vmap_area_list, typeof(*va), list);
2606 while (n > 0 && &va->list != &vmap_area_list) {
2608 va = list_next_entry(va, list);
2610 if (!n && &va->list != &vmap_area_list)
2617 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2619 struct vmap_area *va = p, *next;
2622 next = list_next_entry(va, list);
2623 if (&next->list != &vmap_area_list)
2629 static void s_stop(struct seq_file *m, void *p)
2630 __releases(&vmap_area_lock)
2632 spin_unlock(&vmap_area_lock);
2635 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2637 if (IS_ENABLED(CONFIG_NUMA)) {
2638 unsigned int nr, *counters = m->private;
2643 if (v->flags & VM_UNINITIALIZED)
2645 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2648 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2650 for (nr = 0; nr < v->nr_pages; nr++)
2651 counters[page_to_nid(v->pages[nr])]++;
2653 for_each_node_state(nr, N_HIGH_MEMORY)
2655 seq_printf(m, " N%u=%u", nr, counters[nr]);
2659 static int s_show(struct seq_file *m, void *p)
2661 struct vmap_area *va = p;
2662 struct vm_struct *v;
2665 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2666 * behalf of vmap area is being tear down or vm_map_ram allocation.
2668 if (!(va->flags & VM_VM_AREA))
2673 seq_printf(m, "0x%pK-0x%pK %7ld",
2674 v->addr, v->addr + v->size, v->size);
2677 seq_printf(m, " %pS", v->caller);
2680 seq_printf(m, " pages=%d", v->nr_pages);
2683 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2685 if (v->flags & VM_IOREMAP)
2686 seq_puts(m, " ioremap");
2688 if (v->flags & VM_ALLOC)
2689 seq_puts(m, " vmalloc");
2691 if (v->flags & VM_MAP)
2692 seq_puts(m, " vmap");
2694 if (v->flags & VM_USERMAP)
2695 seq_puts(m, " user");
2697 if (is_vmalloc_addr(v->pages))
2698 seq_puts(m, " vpages");
2700 show_numa_info(m, v);
2705 static const struct seq_operations vmalloc_op = {
2712 static int vmalloc_open(struct inode *inode, struct file *file)
2714 if (IS_ENABLED(CONFIG_NUMA))
2715 return seq_open_private(file, &vmalloc_op,
2716 nr_node_ids * sizeof(unsigned int));
2718 return seq_open(file, &vmalloc_op);
2721 static const struct file_operations proc_vmalloc_operations = {
2722 .open = vmalloc_open,
2724 .llseek = seq_lseek,
2725 .release = seq_release_private,
2728 static int __init proc_vmalloc_init(void)
2730 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2733 module_init(proc_vmalloc_init);