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);
462 if (addr + size > vend)
466 va->va_end = addr + size;
468 __insert_vmap_area(va);
469 free_vmap_cache = &va->rb_node;
470 spin_unlock(&vmap_area_lock);
472 BUG_ON(!IS_ALIGNED(va->va_start, align));
473 BUG_ON(va->va_start < vstart);
474 BUG_ON(va->va_end > vend);
479 spin_unlock(&vmap_area_lock);
481 purge_vmap_area_lazy();
486 if (gfpflags_allow_blocking(gfp_mask)) {
487 unsigned long freed = 0;
488 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
495 if (printk_ratelimit())
496 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
499 return ERR_PTR(-EBUSY);
502 int register_vmap_purge_notifier(struct notifier_block *nb)
504 return blocking_notifier_chain_register(&vmap_notify_list, nb);
506 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
508 int unregister_vmap_purge_notifier(struct notifier_block *nb)
510 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
512 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
514 static void __free_vmap_area(struct vmap_area *va)
516 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
518 if (free_vmap_cache) {
519 if (va->va_end < cached_vstart) {
520 free_vmap_cache = NULL;
522 struct vmap_area *cache;
523 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
524 if (va->va_start <= cache->va_start) {
525 free_vmap_cache = rb_prev(&va->rb_node);
527 * We don't try to update cached_hole_size or
528 * cached_align, but it won't go very wrong.
533 rb_erase(&va->rb_node, &vmap_area_root);
534 RB_CLEAR_NODE(&va->rb_node);
535 list_del_rcu(&va->list);
538 * Track the highest possible candidate for pcpu area
539 * allocation. Areas outside of vmalloc area can be returned
540 * here too, consider only end addresses which fall inside
541 * vmalloc area proper.
543 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
544 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
546 kfree_rcu(va, rcu_head);
550 * Free a region of KVA allocated by alloc_vmap_area
552 static void free_vmap_area(struct vmap_area *va)
554 spin_lock(&vmap_area_lock);
555 __free_vmap_area(va);
556 spin_unlock(&vmap_area_lock);
560 * Clear the pagetable entries of a given vmap_area
562 static void unmap_vmap_area(struct vmap_area *va)
564 vunmap_page_range(va->va_start, va->va_end);
567 static void vmap_debug_free_range(unsigned long start, unsigned long end)
570 * Unmap page tables and force a TLB flush immediately if pagealloc
571 * debugging is enabled. This catches use after free bugs similarly to
572 * those in linear kernel virtual address space after a page has been
575 * All the lazy freeing logic is still retained, in order to minimise
576 * intrusiveness of this debugging feature.
578 * This is going to be *slow* (linear kernel virtual address debugging
579 * doesn't do a broadcast TLB flush so it is a lot faster).
581 if (debug_pagealloc_enabled()) {
582 vunmap_page_range(start, end);
583 flush_tlb_kernel_range(start, end);
588 * lazy_max_pages is the maximum amount of virtual address space we gather up
589 * before attempting to purge with a TLB flush.
591 * There is a tradeoff here: a larger number will cover more kernel page tables
592 * and take slightly longer to purge, but it will linearly reduce the number of
593 * global TLB flushes that must be performed. It would seem natural to scale
594 * this number up linearly with the number of CPUs (because vmapping activity
595 * could also scale linearly with the number of CPUs), however it is likely
596 * that in practice, workloads might be constrained in other ways that mean
597 * vmap activity will not scale linearly with CPUs. Also, I want to be
598 * conservative and not introduce a big latency on huge systems, so go with
599 * a less aggressive log scale. It will still be an improvement over the old
600 * code, and it will be simple to change the scale factor if we find that it
601 * becomes a problem on bigger systems.
603 static unsigned long lazy_max_pages(void)
607 log = fls(num_online_cpus());
609 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
612 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
614 /* for per-CPU blocks */
615 static void purge_fragmented_blocks_allcpus(void);
618 * called before a call to iounmap() if the caller wants vm_area_struct's
621 void set_iounmap_nonlazy(void)
623 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
627 * Purges all lazily-freed vmap areas.
629 * If sync is 0 then don't purge if there is already a purge in progress.
630 * If force_flush is 1, then flush kernel TLBs between *start and *end even
631 * if we found no lazy vmap areas to unmap (callers can use this to optimise
632 * their own TLB flushing).
633 * Returns with *start = min(*start, lowest purged address)
634 * *end = max(*end, highest purged address)
636 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
637 int sync, int force_flush)
639 static DEFINE_SPINLOCK(purge_lock);
640 struct llist_node *valist;
641 struct vmap_area *va;
642 struct vmap_area *n_va;
646 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
647 * should not expect such behaviour. This just simplifies locking for
648 * the case that isn't actually used at the moment anyway.
650 if (!sync && !force_flush) {
651 if (!spin_trylock(&purge_lock))
654 spin_lock(&purge_lock);
657 purge_fragmented_blocks_allcpus();
659 valist = llist_del_all(&vmap_purge_list);
660 llist_for_each_entry(va, valist, purge_list) {
661 if (va->va_start < *start)
662 *start = va->va_start;
663 if (va->va_end > *end)
665 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
669 atomic_sub(nr, &vmap_lazy_nr);
671 if (nr || force_flush)
672 flush_tlb_kernel_range(*start, *end);
675 spin_lock(&vmap_area_lock);
676 llist_for_each_entry_safe(va, n_va, valist, purge_list)
677 __free_vmap_area(va);
678 spin_unlock(&vmap_area_lock);
680 spin_unlock(&purge_lock);
684 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
685 * is already purging.
687 static void try_purge_vmap_area_lazy(void)
689 unsigned long start = ULONG_MAX, end = 0;
691 __purge_vmap_area_lazy(&start, &end, 0, 0);
695 * Kick off a purge of the outstanding lazy areas.
697 static void purge_vmap_area_lazy(void)
699 unsigned long start = ULONG_MAX, end = 0;
701 __purge_vmap_area_lazy(&start, &end, 1, 0);
705 * Free a vmap area, caller ensuring that the area has been unmapped
706 * and flush_cache_vunmap had been called for the correct range
709 static void free_vmap_area_noflush(struct vmap_area *va)
713 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
716 /* After this point, we may free va at any time */
717 llist_add(&va->purge_list, &vmap_purge_list);
719 if (unlikely(nr_lazy > lazy_max_pages()))
720 try_purge_vmap_area_lazy();
724 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
725 * called for the correct range previously.
727 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
730 free_vmap_area_noflush(va);
734 * Free and unmap a vmap area
736 static void free_unmap_vmap_area(struct vmap_area *va)
738 flush_cache_vunmap(va->va_start, va->va_end);
739 free_unmap_vmap_area_noflush(va);
742 static struct vmap_area *find_vmap_area(unsigned long addr)
744 struct vmap_area *va;
746 spin_lock(&vmap_area_lock);
747 va = __find_vmap_area(addr);
748 spin_unlock(&vmap_area_lock);
753 static void free_unmap_vmap_area_addr(unsigned long addr)
755 struct vmap_area *va;
757 va = find_vmap_area(addr);
759 free_unmap_vmap_area(va);
763 /*** Per cpu kva allocator ***/
766 * vmap space is limited especially on 32 bit architectures. Ensure there is
767 * room for at least 16 percpu vmap blocks per CPU.
770 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
771 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
772 * instead (we just need a rough idea)
774 #if BITS_PER_LONG == 32
775 #define VMALLOC_SPACE (128UL*1024*1024)
777 #define VMALLOC_SPACE (128UL*1024*1024*1024)
780 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
781 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
782 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
783 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
784 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
785 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
786 #define VMAP_BBMAP_BITS \
787 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
788 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
789 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
791 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
793 static bool vmap_initialized __read_mostly = false;
795 struct vmap_block_queue {
797 struct list_head free;
802 struct vmap_area *va;
803 unsigned long free, dirty;
804 unsigned long dirty_min, dirty_max; /*< dirty range */
805 struct list_head free_list;
806 struct rcu_head rcu_head;
807 struct list_head purge;
810 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
811 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
814 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
815 * in the free path. Could get rid of this if we change the API to return a
816 * "cookie" from alloc, to be passed to free. But no big deal yet.
818 static DEFINE_SPINLOCK(vmap_block_tree_lock);
819 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
822 * We should probably have a fallback mechanism to allocate virtual memory
823 * out of partially filled vmap blocks. However vmap block sizing should be
824 * fairly reasonable according to the vmalloc size, so it shouldn't be a
828 static unsigned long addr_to_vb_idx(unsigned long addr)
830 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
831 addr /= VMAP_BLOCK_SIZE;
835 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
839 addr = va_start + (pages_off << PAGE_SHIFT);
840 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
845 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
846 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
847 * @order: how many 2^order pages should be occupied in newly allocated block
848 * @gfp_mask: flags for the page level allocator
850 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
852 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
854 struct vmap_block_queue *vbq;
855 struct vmap_block *vb;
856 struct vmap_area *va;
857 unsigned long vb_idx;
861 node = numa_node_id();
863 vb = kmalloc_node(sizeof(struct vmap_block),
864 gfp_mask & GFP_RECLAIM_MASK, node);
866 return ERR_PTR(-ENOMEM);
868 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
869 VMALLOC_START, VMALLOC_END,
876 err = radix_tree_preload(gfp_mask);
883 vaddr = vmap_block_vaddr(va->va_start, 0);
884 spin_lock_init(&vb->lock);
886 /* At least something should be left free */
887 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
888 vb->free = VMAP_BBMAP_BITS - (1UL << order);
890 vb->dirty_min = VMAP_BBMAP_BITS;
892 INIT_LIST_HEAD(&vb->free_list);
894 vb_idx = addr_to_vb_idx(va->va_start);
895 spin_lock(&vmap_block_tree_lock);
896 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
897 spin_unlock(&vmap_block_tree_lock);
899 radix_tree_preload_end();
901 vbq = &get_cpu_var(vmap_block_queue);
902 spin_lock(&vbq->lock);
903 list_add_tail_rcu(&vb->free_list, &vbq->free);
904 spin_unlock(&vbq->lock);
905 put_cpu_var(vmap_block_queue);
910 static void free_vmap_block(struct vmap_block *vb)
912 struct vmap_block *tmp;
913 unsigned long vb_idx;
915 vb_idx = addr_to_vb_idx(vb->va->va_start);
916 spin_lock(&vmap_block_tree_lock);
917 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
918 spin_unlock(&vmap_block_tree_lock);
921 free_vmap_area_noflush(vb->va);
922 kfree_rcu(vb, rcu_head);
925 static void purge_fragmented_blocks(int cpu)
928 struct vmap_block *vb;
929 struct vmap_block *n_vb;
930 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
933 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
935 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
938 spin_lock(&vb->lock);
939 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
940 vb->free = 0; /* prevent further allocs after releasing lock */
941 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
943 vb->dirty_max = VMAP_BBMAP_BITS;
944 spin_lock(&vbq->lock);
945 list_del_rcu(&vb->free_list);
946 spin_unlock(&vbq->lock);
947 spin_unlock(&vb->lock);
948 list_add_tail(&vb->purge, &purge);
950 spin_unlock(&vb->lock);
954 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
955 list_del(&vb->purge);
960 static void purge_fragmented_blocks_allcpus(void)
964 for_each_possible_cpu(cpu)
965 purge_fragmented_blocks(cpu);
968 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
970 struct vmap_block_queue *vbq;
971 struct vmap_block *vb;
975 BUG_ON(offset_in_page(size));
976 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
977 if (WARN_ON(size == 0)) {
979 * Allocating 0 bytes isn't what caller wants since
980 * get_order(0) returns funny result. Just warn and terminate
985 order = get_order(size);
988 vbq = &get_cpu_var(vmap_block_queue);
989 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
990 unsigned long pages_off;
992 spin_lock(&vb->lock);
993 if (vb->free < (1UL << order)) {
994 spin_unlock(&vb->lock);
998 pages_off = VMAP_BBMAP_BITS - vb->free;
999 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
1000 vb->free -= 1UL << order;
1001 if (vb->free == 0) {
1002 spin_lock(&vbq->lock);
1003 list_del_rcu(&vb->free_list);
1004 spin_unlock(&vbq->lock);
1007 spin_unlock(&vb->lock);
1011 put_cpu_var(vmap_block_queue);
1014 /* Allocate new block if nothing was found */
1016 vaddr = new_vmap_block(order, gfp_mask);
1021 static void vb_free(const void *addr, unsigned long size)
1023 unsigned long offset;
1024 unsigned long vb_idx;
1026 struct vmap_block *vb;
1028 BUG_ON(offset_in_page(size));
1029 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1031 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1033 order = get_order(size);
1035 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1036 offset >>= PAGE_SHIFT;
1038 vb_idx = addr_to_vb_idx((unsigned long)addr);
1040 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1044 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1046 spin_lock(&vb->lock);
1048 /* Expand dirty range */
1049 vb->dirty_min = min(vb->dirty_min, offset);
1050 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1052 vb->dirty += 1UL << order;
1053 if (vb->dirty == VMAP_BBMAP_BITS) {
1055 spin_unlock(&vb->lock);
1056 free_vmap_block(vb);
1058 spin_unlock(&vb->lock);
1062 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1064 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1065 * to amortize TLB flushing overheads. What this means is that any page you
1066 * have now, may, in a former life, have been mapped into kernel virtual
1067 * address by the vmap layer and so there might be some CPUs with TLB entries
1068 * still referencing that page (additional to the regular 1:1 kernel mapping).
1070 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1071 * be sure that none of the pages we have control over will have any aliases
1072 * from the vmap layer.
1074 void vm_unmap_aliases(void)
1076 unsigned long start = ULONG_MAX, end = 0;
1080 if (unlikely(!vmap_initialized))
1083 for_each_possible_cpu(cpu) {
1084 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1085 struct vmap_block *vb;
1088 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1089 spin_lock(&vb->lock);
1091 unsigned long va_start = vb->va->va_start;
1094 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1095 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1097 start = min(s, start);
1102 spin_unlock(&vb->lock);
1107 __purge_vmap_area_lazy(&start, &end, 1, flush);
1109 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1112 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1113 * @mem: the pointer returned by vm_map_ram
1114 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1116 void vm_unmap_ram(const void *mem, unsigned int count)
1118 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1119 unsigned long addr = (unsigned long)mem;
1122 BUG_ON(addr < VMALLOC_START);
1123 BUG_ON(addr > VMALLOC_END);
1124 BUG_ON(!PAGE_ALIGNED(addr));
1126 debug_check_no_locks_freed(mem, size);
1127 vmap_debug_free_range(addr, addr+size);
1129 if (likely(count <= VMAP_MAX_ALLOC))
1132 free_unmap_vmap_area_addr(addr);
1134 EXPORT_SYMBOL(vm_unmap_ram);
1137 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1138 * @pages: an array of pointers to the pages to be mapped
1139 * @count: number of pages
1140 * @node: prefer to allocate data structures on this node
1141 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1143 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1144 * faster than vmap so it's good. But if you mix long-life and short-life
1145 * objects with vm_map_ram(), it could consume lots of address space through
1146 * fragmentation (especially on a 32bit machine). You could see failures in
1147 * the end. Please use this function for short-lived objects.
1149 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1151 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1153 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1157 if (likely(count <= VMAP_MAX_ALLOC)) {
1158 mem = vb_alloc(size, GFP_KERNEL);
1161 addr = (unsigned long)mem;
1163 struct vmap_area *va;
1164 va = alloc_vmap_area(size, PAGE_SIZE,
1165 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1169 addr = va->va_start;
1172 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1173 vm_unmap_ram(mem, count);
1178 EXPORT_SYMBOL(vm_map_ram);
1180 static struct vm_struct *vmlist __initdata;
1182 * vm_area_add_early - add vmap area early during boot
1183 * @vm: vm_struct to add
1185 * This function is used to add fixed kernel vm area to vmlist before
1186 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1187 * should contain proper values and the other fields should be zero.
1189 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1191 void __init vm_area_add_early(struct vm_struct *vm)
1193 struct vm_struct *tmp, **p;
1195 BUG_ON(vmap_initialized);
1196 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1197 if (tmp->addr >= vm->addr) {
1198 BUG_ON(tmp->addr < vm->addr + vm->size);
1201 BUG_ON(tmp->addr + tmp->size > vm->addr);
1208 * vm_area_register_early - register vmap area early during boot
1209 * @vm: vm_struct to register
1210 * @align: requested alignment
1212 * This function is used to register kernel vm area before
1213 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1214 * proper values on entry and other fields should be zero. On return,
1215 * vm->addr contains the allocated address.
1217 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1219 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1221 static size_t vm_init_off __initdata;
1224 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1225 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1227 vm->addr = (void *)addr;
1229 vm_area_add_early(vm);
1232 void __init vmalloc_init(void)
1234 struct vmap_area *va;
1235 struct vm_struct *tmp;
1238 for_each_possible_cpu(i) {
1239 struct vmap_block_queue *vbq;
1240 struct vfree_deferred *p;
1242 vbq = &per_cpu(vmap_block_queue, i);
1243 spin_lock_init(&vbq->lock);
1244 INIT_LIST_HEAD(&vbq->free);
1245 p = &per_cpu(vfree_deferred, i);
1246 init_llist_head(&p->list);
1247 INIT_WORK(&p->wq, free_work);
1250 /* Import existing vmlist entries. */
1251 for (tmp = vmlist; tmp; tmp = tmp->next) {
1252 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1253 va->flags = VM_VM_AREA;
1254 va->va_start = (unsigned long)tmp->addr;
1255 va->va_end = va->va_start + tmp->size;
1257 __insert_vmap_area(va);
1260 vmap_area_pcpu_hole = VMALLOC_END;
1262 vmap_initialized = true;
1266 * map_kernel_range_noflush - map kernel VM area with the specified pages
1267 * @addr: start of the VM area to map
1268 * @size: size of the VM area to map
1269 * @prot: page protection flags to use
1270 * @pages: pages to map
1272 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1273 * specify should have been allocated using get_vm_area() and its
1277 * This function does NOT do any cache flushing. The caller is
1278 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1279 * before calling this function.
1282 * The number of pages mapped on success, -errno on failure.
1284 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1285 pgprot_t prot, struct page **pages)
1287 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1291 * unmap_kernel_range_noflush - unmap kernel VM area
1292 * @addr: start of the VM area to unmap
1293 * @size: size of the VM area to unmap
1295 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1296 * specify should have been allocated using get_vm_area() and its
1300 * This function does NOT do any cache flushing. The caller is
1301 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1302 * before calling this function and flush_tlb_kernel_range() after.
1304 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1306 vunmap_page_range(addr, addr + size);
1308 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1311 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1312 * @addr: start of the VM area to unmap
1313 * @size: size of the VM area to unmap
1315 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1316 * the unmapping and tlb after.
1318 void unmap_kernel_range(unsigned long addr, unsigned long size)
1320 unsigned long end = addr + size;
1322 flush_cache_vunmap(addr, end);
1323 vunmap_page_range(addr, end);
1324 flush_tlb_kernel_range(addr, end);
1326 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1328 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1330 unsigned long addr = (unsigned long)area->addr;
1331 unsigned long end = addr + get_vm_area_size(area);
1334 err = vmap_page_range(addr, end, prot, pages);
1336 return err > 0 ? 0 : err;
1338 EXPORT_SYMBOL_GPL(map_vm_area);
1340 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1341 unsigned long flags, const void *caller)
1343 spin_lock(&vmap_area_lock);
1345 vm->addr = (void *)va->va_start;
1346 vm->size = va->va_end - va->va_start;
1347 vm->caller = caller;
1349 va->flags |= VM_VM_AREA;
1350 spin_unlock(&vmap_area_lock);
1353 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1356 * Before removing VM_UNINITIALIZED,
1357 * we should make sure that vm has proper values.
1358 * Pair with smp_rmb() in show_numa_info().
1361 vm->flags &= ~VM_UNINITIALIZED;
1364 static struct vm_struct *__get_vm_area_node(unsigned long size,
1365 unsigned long align, unsigned long flags, unsigned long start,
1366 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1368 struct vmap_area *va;
1369 struct vm_struct *area;
1371 BUG_ON(in_interrupt());
1372 size = PAGE_ALIGN(size);
1373 if (unlikely(!size))
1376 if (flags & VM_IOREMAP)
1377 align = 1ul << clamp_t(int, get_count_order_long(size),
1378 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1380 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1381 if (unlikely(!area))
1384 if (!(flags & VM_NO_GUARD))
1387 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1393 setup_vmalloc_vm(area, va, flags, caller);
1398 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1399 unsigned long start, unsigned long end)
1401 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1402 GFP_KERNEL, __builtin_return_address(0));
1404 EXPORT_SYMBOL_GPL(__get_vm_area);
1406 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1407 unsigned long start, unsigned long end,
1410 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1411 GFP_KERNEL, caller);
1415 * get_vm_area - reserve a contiguous kernel virtual area
1416 * @size: size of the area
1417 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1419 * Search an area of @size in the kernel virtual mapping area,
1420 * and reserved it for out purposes. Returns the area descriptor
1421 * on success or %NULL on failure.
1423 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1425 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1426 NUMA_NO_NODE, GFP_KERNEL,
1427 __builtin_return_address(0));
1430 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1433 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1434 NUMA_NO_NODE, GFP_KERNEL, caller);
1438 * find_vm_area - find a continuous kernel virtual area
1439 * @addr: base address
1441 * Search for the kernel VM area starting at @addr, and return it.
1442 * It is up to the caller to do all required locking to keep the returned
1445 struct vm_struct *find_vm_area(const void *addr)
1447 struct vmap_area *va;
1449 va = find_vmap_area((unsigned long)addr);
1450 if (va && va->flags & VM_VM_AREA)
1457 * remove_vm_area - find and remove a continuous kernel virtual area
1458 * @addr: base address
1460 * Search for the kernel VM area starting at @addr, and remove it.
1461 * This function returns the found VM area, but using it is NOT safe
1462 * on SMP machines, except for its size or flags.
1464 struct vm_struct *remove_vm_area(const void *addr)
1466 struct vmap_area *va;
1468 va = find_vmap_area((unsigned long)addr);
1469 if (va && va->flags & VM_VM_AREA) {
1470 struct vm_struct *vm = va->vm;
1472 spin_lock(&vmap_area_lock);
1474 va->flags &= ~VM_VM_AREA;
1475 spin_unlock(&vmap_area_lock);
1477 vmap_debug_free_range(va->va_start, va->va_end);
1478 kasan_free_shadow(vm);
1479 free_unmap_vmap_area(va);
1486 static void __vunmap(const void *addr, int deallocate_pages)
1488 struct vm_struct *area;
1493 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1497 area = find_vmap_area((unsigned long)addr)->vm;
1498 if (unlikely(!area)) {
1499 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1504 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1505 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1507 remove_vm_area(addr);
1508 if (deallocate_pages) {
1511 for (i = 0; i < area->nr_pages; i++) {
1512 struct page *page = area->pages[i];
1515 __free_pages(page, 0);
1518 kvfree(area->pages);
1526 * vfree - release memory allocated by vmalloc()
1527 * @addr: memory base address
1529 * Free the virtually continuous memory area starting at @addr, as
1530 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1531 * NULL, no operation is performed.
1533 * Must not be called in NMI context (strictly speaking, only if we don't
1534 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1535 * conventions for vfree() arch-depenedent would be a really bad idea)
1537 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1539 void vfree(const void *addr)
1543 kmemleak_free(addr);
1547 if (unlikely(in_interrupt())) {
1548 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1549 if (llist_add((struct llist_node *)addr, &p->list))
1550 schedule_work(&p->wq);
1554 EXPORT_SYMBOL(vfree);
1557 * vunmap - release virtual mapping obtained by vmap()
1558 * @addr: memory base address
1560 * Free the virtually contiguous memory area starting at @addr,
1561 * which was created from the page array passed to vmap().
1563 * Must not be called in interrupt context.
1565 void vunmap(const void *addr)
1567 BUG_ON(in_interrupt());
1572 EXPORT_SYMBOL(vunmap);
1575 * vmap - map an array of pages into virtually contiguous space
1576 * @pages: array of page pointers
1577 * @count: number of pages to map
1578 * @flags: vm_area->flags
1579 * @prot: page protection for the mapping
1581 * Maps @count pages from @pages into contiguous kernel virtual
1584 void *vmap(struct page **pages, unsigned int count,
1585 unsigned long flags, pgprot_t prot)
1587 struct vm_struct *area;
1588 unsigned long size; /* In bytes */
1592 if (count > totalram_pages)
1595 size = (unsigned long)count << PAGE_SHIFT;
1596 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1600 if (map_vm_area(area, prot, pages)) {
1607 EXPORT_SYMBOL(vmap);
1609 static void *__vmalloc_node(unsigned long size, unsigned long align,
1610 gfp_t gfp_mask, pgprot_t prot,
1611 int node, const void *caller);
1612 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1613 pgprot_t prot, int node)
1615 struct page **pages;
1616 unsigned int nr_pages, array_size, i;
1617 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1618 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1620 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1621 array_size = (nr_pages * sizeof(struct page *));
1623 area->nr_pages = nr_pages;
1624 /* Please note that the recursion is strictly bounded. */
1625 if (array_size > PAGE_SIZE) {
1626 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1627 PAGE_KERNEL, node, area->caller);
1629 pages = kmalloc_node(array_size, nested_gfp, node);
1631 area->pages = pages;
1633 remove_vm_area(area->addr);
1638 for (i = 0; i < area->nr_pages; i++) {
1641 if (node == NUMA_NO_NODE)
1642 page = alloc_page(alloc_mask);
1644 page = alloc_pages_node(node, alloc_mask, 0);
1646 if (unlikely(!page)) {
1647 /* Successfully allocated i pages, free them in __vunmap() */
1651 area->pages[i] = page;
1652 if (gfpflags_allow_blocking(gfp_mask))
1656 if (map_vm_area(area, prot, pages))
1661 warn_alloc(gfp_mask,
1662 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1663 (area->nr_pages*PAGE_SIZE), area->size);
1669 * __vmalloc_node_range - allocate virtually contiguous memory
1670 * @size: allocation size
1671 * @align: desired alignment
1672 * @start: vm area range start
1673 * @end: vm area range end
1674 * @gfp_mask: flags for the page level allocator
1675 * @prot: protection mask for the allocated pages
1676 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1677 * @node: node to use for allocation or NUMA_NO_NODE
1678 * @caller: caller's return address
1680 * Allocate enough pages to cover @size from the page level
1681 * allocator with @gfp_mask flags. Map them into contiguous
1682 * kernel virtual space, using a pagetable protection of @prot.
1684 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1685 unsigned long start, unsigned long end, gfp_t gfp_mask,
1686 pgprot_t prot, unsigned long vm_flags, int node,
1689 struct vm_struct *area;
1691 unsigned long real_size = size;
1693 size = PAGE_ALIGN(size);
1694 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1697 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1698 vm_flags, start, end, node, gfp_mask, caller);
1702 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1707 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1708 * flag. It means that vm_struct is not fully initialized.
1709 * Now, it is fully initialized, so remove this flag here.
1711 clear_vm_uninitialized_flag(area);
1714 * A ref_count = 2 is needed because vm_struct allocated in
1715 * __get_vm_area_node() contains a reference to the virtual address of
1716 * the vmalloc'ed block.
1718 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1723 warn_alloc(gfp_mask,
1724 "vmalloc: allocation failure: %lu bytes", real_size);
1729 * __vmalloc_node - allocate virtually contiguous memory
1730 * @size: allocation size
1731 * @align: desired alignment
1732 * @gfp_mask: flags for the page level allocator
1733 * @prot: protection mask for the allocated pages
1734 * @node: node to use for allocation or NUMA_NO_NODE
1735 * @caller: caller's return address
1737 * Allocate enough pages to cover @size from the page level
1738 * allocator with @gfp_mask flags. Map them into contiguous
1739 * kernel virtual space, using a pagetable protection of @prot.
1741 static void *__vmalloc_node(unsigned long size, unsigned long align,
1742 gfp_t gfp_mask, pgprot_t prot,
1743 int node, const void *caller)
1745 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1746 gfp_mask, prot, 0, node, caller);
1749 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1751 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1752 __builtin_return_address(0));
1754 EXPORT_SYMBOL(__vmalloc);
1756 static inline void *__vmalloc_node_flags(unsigned long size,
1757 int node, gfp_t flags)
1759 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1760 node, __builtin_return_address(0));
1764 * vmalloc - allocate virtually contiguous memory
1765 * @size: allocation size
1766 * Allocate enough pages to cover @size from the page level
1767 * allocator and map them into contiguous kernel virtual space.
1769 * For tight control over page level allocator and protection flags
1770 * use __vmalloc() instead.
1772 void *vmalloc(unsigned long size)
1774 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1775 GFP_KERNEL | __GFP_HIGHMEM);
1777 EXPORT_SYMBOL(vmalloc);
1780 * vzalloc - allocate virtually contiguous memory with zero fill
1781 * @size: allocation size
1782 * Allocate enough pages to cover @size from the page level
1783 * allocator and map them into contiguous kernel virtual space.
1784 * The memory allocated is set to zero.
1786 * For tight control over page level allocator and protection flags
1787 * use __vmalloc() instead.
1789 void *vzalloc(unsigned long size)
1791 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1792 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1794 EXPORT_SYMBOL(vzalloc);
1797 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1798 * @size: allocation size
1800 * The resulting memory area is zeroed so it can be mapped to userspace
1801 * without leaking data.
1803 void *vmalloc_user(unsigned long size)
1805 struct vm_struct *area;
1808 ret = __vmalloc_node(size, SHMLBA,
1809 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1810 PAGE_KERNEL, NUMA_NO_NODE,
1811 __builtin_return_address(0));
1813 area = find_vm_area(ret);
1814 area->flags |= VM_USERMAP;
1818 EXPORT_SYMBOL(vmalloc_user);
1821 * vmalloc_node - allocate memory on a specific node
1822 * @size: allocation size
1825 * Allocate enough pages to cover @size from the page level
1826 * allocator and map them into contiguous kernel virtual space.
1828 * For tight control over page level allocator and protection flags
1829 * use __vmalloc() instead.
1831 void *vmalloc_node(unsigned long size, int node)
1833 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1834 node, __builtin_return_address(0));
1836 EXPORT_SYMBOL(vmalloc_node);
1839 * vzalloc_node - allocate memory on a specific node with zero fill
1840 * @size: allocation size
1843 * Allocate enough pages to cover @size from the page level
1844 * allocator and map them into contiguous kernel virtual space.
1845 * The memory allocated is set to zero.
1847 * For tight control over page level allocator and protection flags
1848 * use __vmalloc_node() instead.
1850 void *vzalloc_node(unsigned long size, int node)
1852 return __vmalloc_node_flags(size, node,
1853 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1855 EXPORT_SYMBOL(vzalloc_node);
1857 #ifndef PAGE_KERNEL_EXEC
1858 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1862 * vmalloc_exec - allocate virtually contiguous, executable memory
1863 * @size: allocation size
1865 * Kernel-internal function to allocate enough pages to cover @size
1866 * the page level allocator and map them into contiguous and
1867 * executable kernel virtual space.
1869 * For tight control over page level allocator and protection flags
1870 * use __vmalloc() instead.
1873 void *vmalloc_exec(unsigned long size)
1875 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1876 NUMA_NO_NODE, __builtin_return_address(0));
1879 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1880 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1881 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1882 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1884 #define GFP_VMALLOC32 GFP_KERNEL
1888 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1889 * @size: allocation size
1891 * Allocate enough 32bit PA addressable pages to cover @size from the
1892 * page level allocator and map them into contiguous kernel virtual space.
1894 void *vmalloc_32(unsigned long size)
1896 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1897 NUMA_NO_NODE, __builtin_return_address(0));
1899 EXPORT_SYMBOL(vmalloc_32);
1902 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1903 * @size: allocation size
1905 * The resulting memory area is 32bit addressable and zeroed so it can be
1906 * mapped to userspace without leaking data.
1908 void *vmalloc_32_user(unsigned long size)
1910 struct vm_struct *area;
1913 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1914 NUMA_NO_NODE, __builtin_return_address(0));
1916 area = find_vm_area(ret);
1917 area->flags |= VM_USERMAP;
1921 EXPORT_SYMBOL(vmalloc_32_user);
1924 * small helper routine , copy contents to buf from addr.
1925 * If the page is not present, fill zero.
1928 static int aligned_vread(char *buf, char *addr, unsigned long count)
1934 unsigned long offset, length;
1936 offset = offset_in_page(addr);
1937 length = PAGE_SIZE - offset;
1940 p = vmalloc_to_page(addr);
1942 * To do safe access to this _mapped_ area, we need
1943 * lock. But adding lock here means that we need to add
1944 * overhead of vmalloc()/vfree() calles for this _debug_
1945 * interface, rarely used. Instead of that, we'll use
1946 * kmap() and get small overhead in this access function.
1950 * we can expect USER0 is not used (see vread/vwrite's
1951 * function description)
1953 void *map = kmap_atomic(p);
1954 memcpy(buf, map + offset, length);
1957 memset(buf, 0, length);
1967 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1973 unsigned long offset, length;
1975 offset = offset_in_page(addr);
1976 length = PAGE_SIZE - offset;
1979 p = vmalloc_to_page(addr);
1981 * To do safe access to this _mapped_ area, we need
1982 * lock. But adding lock here means that we need to add
1983 * overhead of vmalloc()/vfree() calles for this _debug_
1984 * interface, rarely used. Instead of that, we'll use
1985 * kmap() and get small overhead in this access function.
1989 * we can expect USER0 is not used (see vread/vwrite's
1990 * function description)
1992 void *map = kmap_atomic(p);
1993 memcpy(map + offset, buf, length);
2005 * vread() - read vmalloc area in a safe way.
2006 * @buf: buffer for reading data
2007 * @addr: vm address.
2008 * @count: number of bytes to be read.
2010 * Returns # of bytes which addr and buf should be increased.
2011 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2012 * includes any intersect with alive vmalloc area.
2014 * This function checks that addr is a valid vmalloc'ed area, and
2015 * copy data from that area to a given buffer. If the given memory range
2016 * of [addr...addr+count) includes some valid address, data is copied to
2017 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2018 * IOREMAP area is treated as memory hole and no copy is done.
2020 * If [addr...addr+count) doesn't includes any intersects with alive
2021 * vm_struct area, returns 0. @buf should be kernel's buffer.
2023 * Note: In usual ops, vread() is never necessary because the caller
2024 * should know vmalloc() area is valid and can use memcpy().
2025 * This is for routines which have to access vmalloc area without
2026 * any informaion, as /dev/kmem.
2030 long vread(char *buf, char *addr, unsigned long count)
2032 struct vmap_area *va;
2033 struct vm_struct *vm;
2034 char *vaddr, *buf_start = buf;
2035 unsigned long buflen = count;
2038 /* Don't allow overflow */
2039 if ((unsigned long) addr + count < count)
2040 count = -(unsigned long) addr;
2042 spin_lock(&vmap_area_lock);
2043 list_for_each_entry(va, &vmap_area_list, list) {
2047 if (!(va->flags & VM_VM_AREA))
2051 vaddr = (char *) vm->addr;
2052 if (addr >= vaddr + get_vm_area_size(vm))
2054 while (addr < vaddr) {
2062 n = vaddr + get_vm_area_size(vm) - addr;
2065 if (!(vm->flags & VM_IOREMAP))
2066 aligned_vread(buf, addr, n);
2067 else /* IOREMAP area is treated as memory hole */
2074 spin_unlock(&vmap_area_lock);
2076 if (buf == buf_start)
2078 /* zero-fill memory holes */
2079 if (buf != buf_start + buflen)
2080 memset(buf, 0, buflen - (buf - buf_start));
2086 * vwrite() - write vmalloc area in a safe way.
2087 * @buf: buffer for source data
2088 * @addr: vm address.
2089 * @count: number of bytes to be read.
2091 * Returns # of bytes which addr and buf should be incresed.
2092 * (same number to @count).
2093 * If [addr...addr+count) doesn't includes any intersect with valid
2094 * vmalloc area, returns 0.
2096 * This function checks that addr is a valid vmalloc'ed area, and
2097 * copy data from a buffer to the given addr. If specified range of
2098 * [addr...addr+count) includes some valid address, data is copied from
2099 * proper area of @buf. If there are memory holes, no copy to hole.
2100 * IOREMAP area is treated as memory hole and no copy is done.
2102 * If [addr...addr+count) doesn't includes any intersects with alive
2103 * vm_struct area, returns 0. @buf should be kernel's buffer.
2105 * Note: In usual ops, vwrite() is never necessary because the caller
2106 * should know vmalloc() area is valid and can use memcpy().
2107 * This is for routines which have to access vmalloc area without
2108 * any informaion, as /dev/kmem.
2111 long vwrite(char *buf, char *addr, unsigned long count)
2113 struct vmap_area *va;
2114 struct vm_struct *vm;
2116 unsigned long n, buflen;
2119 /* Don't allow overflow */
2120 if ((unsigned long) addr + count < count)
2121 count = -(unsigned long) addr;
2124 spin_lock(&vmap_area_lock);
2125 list_for_each_entry(va, &vmap_area_list, list) {
2129 if (!(va->flags & VM_VM_AREA))
2133 vaddr = (char *) vm->addr;
2134 if (addr >= vaddr + get_vm_area_size(vm))
2136 while (addr < vaddr) {
2143 n = vaddr + get_vm_area_size(vm) - addr;
2146 if (!(vm->flags & VM_IOREMAP)) {
2147 aligned_vwrite(buf, addr, n);
2155 spin_unlock(&vmap_area_lock);
2162 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2163 * @vma: vma to cover
2164 * @uaddr: target user address to start at
2165 * @kaddr: virtual address of vmalloc kernel memory
2166 * @size: size of map area
2168 * Returns: 0 for success, -Exxx on failure
2170 * This function checks that @kaddr is a valid vmalloc'ed area,
2171 * and that it is big enough to cover the range starting at
2172 * @uaddr in @vma. Will return failure if that criteria isn't
2175 * Similar to remap_pfn_range() (see mm/memory.c)
2177 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2178 void *kaddr, unsigned long size)
2180 struct vm_struct *area;
2182 size = PAGE_ALIGN(size);
2184 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2187 area = find_vm_area(kaddr);
2191 if (!(area->flags & VM_USERMAP))
2194 if (kaddr + size > area->addr + get_vm_area_size(area))
2198 struct page *page = vmalloc_to_page(kaddr);
2201 ret = vm_insert_page(vma, uaddr, page);
2210 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2214 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2217 * remap_vmalloc_range - map vmalloc pages to userspace
2218 * @vma: vma to cover (map full range of vma)
2219 * @addr: vmalloc memory
2220 * @pgoff: number of pages into addr before first page to map
2222 * Returns: 0 for success, -Exxx on failure
2224 * This function checks that addr is a valid vmalloc'ed area, and
2225 * that it is big enough to cover the vma. Will return failure if
2226 * that criteria isn't met.
2228 * Similar to remap_pfn_range() (see mm/memory.c)
2230 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2231 unsigned long pgoff)
2233 return remap_vmalloc_range_partial(vma, vma->vm_start,
2234 addr + (pgoff << PAGE_SHIFT),
2235 vma->vm_end - vma->vm_start);
2237 EXPORT_SYMBOL(remap_vmalloc_range);
2240 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2243 void __weak vmalloc_sync_all(void)
2248 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2260 * alloc_vm_area - allocate a range of kernel address space
2261 * @size: size of the area
2262 * @ptes: returns the PTEs for the address space
2264 * Returns: NULL on failure, vm_struct on success
2266 * This function reserves a range of kernel address space, and
2267 * allocates pagetables to map that range. No actual mappings
2270 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2271 * allocated for the VM area are returned.
2273 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2275 struct vm_struct *area;
2277 area = get_vm_area_caller(size, VM_IOREMAP,
2278 __builtin_return_address(0));
2283 * This ensures that page tables are constructed for this region
2284 * of kernel virtual address space and mapped into init_mm.
2286 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2287 size, f, ptes ? &ptes : NULL)) {
2294 EXPORT_SYMBOL_GPL(alloc_vm_area);
2296 void free_vm_area(struct vm_struct *area)
2298 struct vm_struct *ret;
2299 ret = remove_vm_area(area->addr);
2300 BUG_ON(ret != area);
2303 EXPORT_SYMBOL_GPL(free_vm_area);
2306 static struct vmap_area *node_to_va(struct rb_node *n)
2308 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2312 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2313 * @end: target address
2314 * @pnext: out arg for the next vmap_area
2315 * @pprev: out arg for the previous vmap_area
2317 * Returns: %true if either or both of next and prev are found,
2318 * %false if no vmap_area exists
2320 * Find vmap_areas end addresses of which enclose @end. ie. if not
2321 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2323 static bool pvm_find_next_prev(unsigned long end,
2324 struct vmap_area **pnext,
2325 struct vmap_area **pprev)
2327 struct rb_node *n = vmap_area_root.rb_node;
2328 struct vmap_area *va = NULL;
2331 va = rb_entry(n, struct vmap_area, rb_node);
2332 if (end < va->va_end)
2334 else if (end > va->va_end)
2343 if (va->va_end > end) {
2345 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2348 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2354 * pvm_determine_end - find the highest aligned address between two vmap_areas
2355 * @pnext: in/out arg for the next vmap_area
2356 * @pprev: in/out arg for the previous vmap_area
2359 * Returns: determined end address
2361 * Find the highest aligned address between *@pnext and *@pprev below
2362 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2363 * down address is between the end addresses of the two vmap_areas.
2365 * Please note that the address returned by this function may fall
2366 * inside *@pnext vmap_area. The caller is responsible for checking
2369 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2370 struct vmap_area **pprev,
2371 unsigned long align)
2373 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2377 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2381 while (*pprev && (*pprev)->va_end > addr) {
2383 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2390 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2391 * @offsets: array containing offset of each area
2392 * @sizes: array containing size of each area
2393 * @nr_vms: the number of areas to allocate
2394 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2396 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2397 * vm_structs on success, %NULL on failure
2399 * Percpu allocator wants to use congruent vm areas so that it can
2400 * maintain the offsets among percpu areas. This function allocates
2401 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2402 * be scattered pretty far, distance between two areas easily going up
2403 * to gigabytes. To avoid interacting with regular vmallocs, these
2404 * areas are allocated from top.
2406 * Despite its complicated look, this allocator is rather simple. It
2407 * does everything top-down and scans areas from the end looking for
2408 * matching slot. While scanning, if any of the areas overlaps with
2409 * existing vmap_area, the base address is pulled down to fit the
2410 * area. Scanning is repeated till all the areas fit and then all
2411 * necessary data structres are inserted and the result is returned.
2413 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2414 const size_t *sizes, int nr_vms,
2417 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2418 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2419 struct vmap_area **vas, *prev, *next;
2420 struct vm_struct **vms;
2421 int area, area2, last_area, term_area;
2422 unsigned long base, start, end, last_end;
2423 bool purged = false;
2425 /* verify parameters and allocate data structures */
2426 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2427 for (last_area = 0, area = 0; area < nr_vms; area++) {
2428 start = offsets[area];
2429 end = start + sizes[area];
2431 /* is everything aligned properly? */
2432 BUG_ON(!IS_ALIGNED(offsets[area], align));
2433 BUG_ON(!IS_ALIGNED(sizes[area], align));
2435 /* detect the area with the highest address */
2436 if (start > offsets[last_area])
2439 for (area2 = 0; area2 < nr_vms; area2++) {
2440 unsigned long start2 = offsets[area2];
2441 unsigned long end2 = start2 + sizes[area2];
2446 BUG_ON(start2 >= start && start2 < end);
2447 BUG_ON(end2 <= end && end2 > start);
2450 last_end = offsets[last_area] + sizes[last_area];
2452 if (vmalloc_end - vmalloc_start < last_end) {
2457 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2458 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2462 for (area = 0; area < nr_vms; area++) {
2463 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2464 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2465 if (!vas[area] || !vms[area])
2469 spin_lock(&vmap_area_lock);
2471 /* start scanning - we scan from the top, begin with the last area */
2472 area = term_area = last_area;
2473 start = offsets[area];
2474 end = start + sizes[area];
2476 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2477 base = vmalloc_end - last_end;
2480 base = pvm_determine_end(&next, &prev, align) - end;
2483 BUG_ON(next && next->va_end <= base + end);
2484 BUG_ON(prev && prev->va_end > base + end);
2487 * base might have underflowed, add last_end before
2490 if (base + last_end < vmalloc_start + last_end) {
2491 spin_unlock(&vmap_area_lock);
2493 purge_vmap_area_lazy();
2501 * If next overlaps, move base downwards so that it's
2502 * right below next and then recheck.
2504 if (next && next->va_start < base + end) {
2505 base = pvm_determine_end(&next, &prev, align) - end;
2511 * If prev overlaps, shift down next and prev and move
2512 * base so that it's right below new next and then
2515 if (prev && prev->va_end > base + start) {
2517 prev = node_to_va(rb_prev(&next->rb_node));
2518 base = pvm_determine_end(&next, &prev, align) - end;
2524 * This area fits, move on to the previous one. If
2525 * the previous one is the terminal one, we're done.
2527 area = (area + nr_vms - 1) % nr_vms;
2528 if (area == term_area)
2530 start = offsets[area];
2531 end = start + sizes[area];
2532 pvm_find_next_prev(base + end, &next, &prev);
2535 /* we've found a fitting base, insert all va's */
2536 for (area = 0; area < nr_vms; area++) {
2537 struct vmap_area *va = vas[area];
2539 va->va_start = base + offsets[area];
2540 va->va_end = va->va_start + sizes[area];
2541 __insert_vmap_area(va);
2544 vmap_area_pcpu_hole = base + offsets[last_area];
2546 spin_unlock(&vmap_area_lock);
2548 /* insert all vm's */
2549 for (area = 0; area < nr_vms; area++)
2550 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2557 for (area = 0; area < nr_vms; area++) {
2568 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2569 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2570 * @nr_vms: the number of allocated areas
2572 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2574 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2578 for (i = 0; i < nr_vms; i++)
2579 free_vm_area(vms[i]);
2582 #endif /* CONFIG_SMP */
2584 #ifdef CONFIG_PROC_FS
2585 static void *s_start(struct seq_file *m, loff_t *pos)
2586 __acquires(&vmap_area_lock)
2589 struct vmap_area *va;
2591 spin_lock(&vmap_area_lock);
2592 va = list_first_entry(&vmap_area_list, typeof(*va), list);
2593 while (n > 0 && &va->list != &vmap_area_list) {
2595 va = list_next_entry(va, list);
2597 if (!n && &va->list != &vmap_area_list)
2604 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2606 struct vmap_area *va = p, *next;
2609 next = list_next_entry(va, list);
2610 if (&next->list != &vmap_area_list)
2616 static void s_stop(struct seq_file *m, void *p)
2617 __releases(&vmap_area_lock)
2619 spin_unlock(&vmap_area_lock);
2622 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2624 if (IS_ENABLED(CONFIG_NUMA)) {
2625 unsigned int nr, *counters = m->private;
2630 if (v->flags & VM_UNINITIALIZED)
2632 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2635 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2637 for (nr = 0; nr < v->nr_pages; nr++)
2638 counters[page_to_nid(v->pages[nr])]++;
2640 for_each_node_state(nr, N_HIGH_MEMORY)
2642 seq_printf(m, " N%u=%u", nr, counters[nr]);
2646 static int s_show(struct seq_file *m, void *p)
2648 struct vmap_area *va = p;
2649 struct vm_struct *v;
2652 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2653 * behalf of vmap area is being tear down or vm_map_ram allocation.
2655 if (!(va->flags & VM_VM_AREA))
2660 seq_printf(m, "0x%pK-0x%pK %7ld",
2661 v->addr, v->addr + v->size, v->size);
2664 seq_printf(m, " %pS", v->caller);
2667 seq_printf(m, " pages=%d", v->nr_pages);
2670 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2672 if (v->flags & VM_IOREMAP)
2673 seq_puts(m, " ioremap");
2675 if (v->flags & VM_ALLOC)
2676 seq_puts(m, " vmalloc");
2678 if (v->flags & VM_MAP)
2679 seq_puts(m, " vmap");
2681 if (v->flags & VM_USERMAP)
2682 seq_puts(m, " user");
2684 if (is_vmalloc_addr(v->pages))
2685 seq_puts(m, " vpages");
2687 show_numa_info(m, v);
2692 static const struct seq_operations vmalloc_op = {
2699 static int vmalloc_open(struct inode *inode, struct file *file)
2701 if (IS_ENABLED(CONFIG_NUMA))
2702 return seq_open_private(file, &vmalloc_op,
2703 nr_node_ids * sizeof(unsigned int));
2705 return seq_open(file, &vmalloc_op);
2708 static const struct file_operations proc_vmalloc_operations = {
2709 .open = vmalloc_open,
2711 .llseek = seq_lseek,
2712 .release = seq_release_private,
2715 static int __init proc_vmalloc_init(void)
2717 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2720 module_init(proc_vmalloc_init);