4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8 * Numa awareness, Christoph Lameter, SGI, June 2005
11 #include <linux/vmalloc.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/rbtree.h>
25 #include <linux/radix-tree.h>
26 #include <linux/rcupdate.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
29 #include <linux/atomic.h>
30 #include <linux/compiler.h>
31 #include <linux/llist.h>
32 #include <linux/bitops.h>
34 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
36 #include <asm/shmparam.h>
40 struct vfree_deferred {
41 struct llist_head list;
42 struct work_struct wq;
44 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
46 static void __vunmap(const void *, int);
48 static void free_work(struct work_struct *w)
50 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
51 struct llist_node *llnode = llist_del_all(&p->list);
54 llnode = llist_next(llnode);
59 /*** Page table manipulation functions ***/
61 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
65 pte = pte_offset_kernel(pmd, addr);
67 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
68 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
69 } while (pte++, addr += PAGE_SIZE, addr != end);
72 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
77 pmd = pmd_offset(pud, addr);
79 next = pmd_addr_end(addr, end);
80 if (pmd_clear_huge(pmd))
82 if (pmd_none_or_clear_bad(pmd))
84 vunmap_pte_range(pmd, addr, next);
85 } while (pmd++, addr = next, addr != end);
88 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
93 pud = pud_offset(pgd, addr);
95 next = pud_addr_end(addr, end);
96 if (pud_clear_huge(pud))
98 if (pud_none_or_clear_bad(pud))
100 vunmap_pmd_range(pud, addr, next);
101 } while (pud++, addr = next, addr != end);
104 static void vunmap_page_range(unsigned long addr, unsigned long end)
110 pgd = pgd_offset_k(addr);
112 next = pgd_addr_end(addr, end);
113 if (pgd_none_or_clear_bad(pgd))
115 vunmap_pud_range(pgd, addr, next);
116 } while (pgd++, addr = next, addr != end);
119 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
120 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
125 * nr is a running index into the array which helps higher level
126 * callers keep track of where we're up to.
129 pte = pte_alloc_kernel(pmd, addr);
133 struct page *page = pages[*nr];
135 if (WARN_ON(!pte_none(*pte)))
139 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
141 } while (pte++, addr += PAGE_SIZE, addr != end);
145 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
146 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
151 pmd = pmd_alloc(&init_mm, pud, addr);
155 next = pmd_addr_end(addr, end);
156 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
158 } while (pmd++, addr = next, addr != end);
162 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
163 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
168 pud = pud_alloc(&init_mm, pgd, addr);
172 next = pud_addr_end(addr, end);
173 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
175 } while (pud++, addr = next, addr != end);
180 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
181 * will have pfns corresponding to the "pages" array.
183 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
185 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
186 pgprot_t prot, struct page **pages)
190 unsigned long addr = start;
195 pgd = pgd_offset_k(addr);
197 next = pgd_addr_end(addr, end);
198 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
201 } while (pgd++, addr = next, addr != end);
206 static int vmap_page_range(unsigned long start, unsigned long end,
207 pgprot_t prot, struct page **pages)
211 ret = vmap_page_range_noflush(start, end, prot, pages);
212 flush_cache_vmap(start, end);
216 int is_vmalloc_or_module_addr(const void *x)
219 * ARM, x86-64 and sparc64 put modules in a special place,
220 * and fall back on vmalloc() if that fails. Others
221 * just put it in the vmalloc space.
223 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
224 unsigned long addr = (unsigned long)x;
225 if (addr >= MODULES_VADDR && addr < MODULES_END)
228 return is_vmalloc_addr(x);
232 * Walk a vmap address to the struct page it maps.
234 struct page *vmalloc_to_page(const void *vmalloc_addr)
236 unsigned long addr = (unsigned long) vmalloc_addr;
237 struct page *page = NULL;
238 pgd_t *pgd = pgd_offset_k(addr);
241 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
242 * architectures that do not vmalloc module space
244 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
246 if (!pgd_none(*pgd)) {
247 pud_t *pud = pud_offset(pgd, addr);
248 if (!pud_none(*pud)) {
249 pmd_t *pmd = pmd_offset(pud, addr);
250 if (!pmd_none(*pmd)) {
253 ptep = pte_offset_map(pmd, addr);
255 if (pte_present(pte))
256 page = pte_page(pte);
263 EXPORT_SYMBOL(vmalloc_to_page);
266 * Map a vmalloc()-space virtual address to the physical page frame number.
268 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
270 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
272 EXPORT_SYMBOL(vmalloc_to_pfn);
275 /*** Global kva allocator ***/
277 #define VM_LAZY_FREE 0x01
278 #define VM_LAZY_FREEING 0x02
279 #define VM_VM_AREA 0x04
281 static DEFINE_SPINLOCK(vmap_area_lock);
282 /* Export for kexec only */
283 LIST_HEAD(vmap_area_list);
284 static struct rb_root vmap_area_root = RB_ROOT;
286 /* The vmap cache globals are protected by vmap_area_lock */
287 static struct rb_node *free_vmap_cache;
288 static unsigned long cached_hole_size;
289 static unsigned long cached_vstart;
290 static unsigned long cached_align;
292 static unsigned long vmap_area_pcpu_hole;
294 static struct vmap_area *__find_vmap_area(unsigned long addr)
296 struct rb_node *n = vmap_area_root.rb_node;
299 struct vmap_area *va;
301 va = rb_entry(n, struct vmap_area, rb_node);
302 if (addr < va->va_start)
304 else if (addr >= va->va_end)
313 static void __insert_vmap_area(struct vmap_area *va)
315 struct rb_node **p = &vmap_area_root.rb_node;
316 struct rb_node *parent = NULL;
320 struct vmap_area *tmp_va;
323 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
324 if (va->va_start < tmp_va->va_end)
326 else if (va->va_end > tmp_va->va_start)
332 rb_link_node(&va->rb_node, parent, p);
333 rb_insert_color(&va->rb_node, &vmap_area_root);
335 /* address-sort this list */
336 tmp = rb_prev(&va->rb_node);
338 struct vmap_area *prev;
339 prev = rb_entry(tmp, struct vmap_area, rb_node);
340 list_add_rcu(&va->list, &prev->list);
342 list_add_rcu(&va->list, &vmap_area_list);
345 static void purge_vmap_area_lazy(void);
348 * Allocate a region of KVA of the specified size and alignment, within the
351 static struct vmap_area *alloc_vmap_area(unsigned long size,
353 unsigned long vstart, unsigned long vend,
354 int node, gfp_t gfp_mask)
356 struct vmap_area *va;
360 struct vmap_area *first;
363 BUG_ON(offset_in_page(size));
364 BUG_ON(!is_power_of_2(align));
366 va = kmalloc_node(sizeof(struct vmap_area),
367 gfp_mask & GFP_RECLAIM_MASK, node);
369 return ERR_PTR(-ENOMEM);
372 * Only scan the relevant parts containing pointers to other objects
373 * to avoid false negatives.
375 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
378 spin_lock(&vmap_area_lock);
380 * Invalidate cache if we have more permissive parameters.
381 * cached_hole_size notes the largest hole noticed _below_
382 * the vmap_area cached in free_vmap_cache: if size fits
383 * into that hole, we want to scan from vstart to reuse
384 * the hole instead of allocating above free_vmap_cache.
385 * Note that __free_vmap_area may update free_vmap_cache
386 * without updating cached_hole_size or cached_align.
388 if (!free_vmap_cache ||
389 size < cached_hole_size ||
390 vstart < cached_vstart ||
391 align < cached_align) {
393 cached_hole_size = 0;
394 free_vmap_cache = NULL;
396 /* record if we encounter less permissive parameters */
397 cached_vstart = vstart;
398 cached_align = align;
400 /* find starting point for our search */
401 if (free_vmap_cache) {
402 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
403 addr = ALIGN(first->va_end, align);
406 if (addr + size < addr)
410 addr = ALIGN(vstart, align);
411 if (addr + size < addr)
414 n = vmap_area_root.rb_node;
418 struct vmap_area *tmp;
419 tmp = rb_entry(n, struct vmap_area, rb_node);
420 if (tmp->va_end >= addr) {
422 if (tmp->va_start <= addr)
433 /* from the starting point, walk areas until a suitable hole is found */
434 while (addr + size > first->va_start && addr + size <= vend) {
435 if (addr + cached_hole_size < first->va_start)
436 cached_hole_size = first->va_start - addr;
437 addr = ALIGN(first->va_end, align);
438 if (addr + size < addr)
441 if (list_is_last(&first->list, &vmap_area_list))
444 first = list_entry(first->list.next,
445 struct vmap_area, list);
449 if (addr + size > vend)
453 va->va_end = addr + size;
455 __insert_vmap_area(va);
456 free_vmap_cache = &va->rb_node;
457 spin_unlock(&vmap_area_lock);
459 BUG_ON(va->va_start & (align-1));
460 BUG_ON(va->va_start < vstart);
461 BUG_ON(va->va_end > vend);
466 spin_unlock(&vmap_area_lock);
468 purge_vmap_area_lazy();
472 if (printk_ratelimit())
473 pr_warn("vmap allocation for size %lu failed: "
474 "use vmalloc=<size> to increase size.\n", size);
476 return ERR_PTR(-EBUSY);
479 static void __free_vmap_area(struct vmap_area *va)
481 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
483 if (free_vmap_cache) {
484 if (va->va_end < cached_vstart) {
485 free_vmap_cache = NULL;
487 struct vmap_area *cache;
488 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
489 if (va->va_start <= cache->va_start) {
490 free_vmap_cache = rb_prev(&va->rb_node);
492 * We don't try to update cached_hole_size or
493 * cached_align, but it won't go very wrong.
498 rb_erase(&va->rb_node, &vmap_area_root);
499 RB_CLEAR_NODE(&va->rb_node);
500 list_del_rcu(&va->list);
503 * Track the highest possible candidate for pcpu area
504 * allocation. Areas outside of vmalloc area can be returned
505 * here too, consider only end addresses which fall inside
506 * vmalloc area proper.
508 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
509 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
511 kfree_rcu(va, rcu_head);
515 * Free a region of KVA allocated by alloc_vmap_area
517 static void free_vmap_area(struct vmap_area *va)
519 spin_lock(&vmap_area_lock);
520 __free_vmap_area(va);
521 spin_unlock(&vmap_area_lock);
525 * Clear the pagetable entries of a given vmap_area
527 static void unmap_vmap_area(struct vmap_area *va)
529 vunmap_page_range(va->va_start, va->va_end);
532 static void vmap_debug_free_range(unsigned long start, unsigned long end)
535 * Unmap page tables and force a TLB flush immediately if
536 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
537 * bugs similarly to those in linear kernel virtual address
538 * space after a page has been freed.
540 * All the lazy freeing logic is still retained, in order to
541 * minimise intrusiveness of this debugging feature.
543 * This is going to be *slow* (linear kernel virtual address
544 * debugging doesn't do a broadcast TLB flush so it is a lot
547 #ifdef CONFIG_DEBUG_PAGEALLOC
548 vunmap_page_range(start, end);
549 flush_tlb_kernel_range(start, end);
554 * lazy_max_pages is the maximum amount of virtual address space we gather up
555 * before attempting to purge with a TLB flush.
557 * There is a tradeoff here: a larger number will cover more kernel page tables
558 * and take slightly longer to purge, but it will linearly reduce the number of
559 * global TLB flushes that must be performed. It would seem natural to scale
560 * this number up linearly with the number of CPUs (because vmapping activity
561 * could also scale linearly with the number of CPUs), however it is likely
562 * that in practice, workloads might be constrained in other ways that mean
563 * vmap activity will not scale linearly with CPUs. Also, I want to be
564 * conservative and not introduce a big latency on huge systems, so go with
565 * a less aggressive log scale. It will still be an improvement over the old
566 * code, and it will be simple to change the scale factor if we find that it
567 * becomes a problem on bigger systems.
569 static unsigned long lazy_max_pages(void)
573 log = fls(num_online_cpus());
575 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
578 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
580 /* for per-CPU blocks */
581 static void purge_fragmented_blocks_allcpus(void);
584 * called before a call to iounmap() if the caller wants vm_area_struct's
587 void set_iounmap_nonlazy(void)
589 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
593 * Purges all lazily-freed vmap areas.
595 * If sync is 0 then don't purge if there is already a purge in progress.
596 * If force_flush is 1, then flush kernel TLBs between *start and *end even
597 * if we found no lazy vmap areas to unmap (callers can use this to optimise
598 * their own TLB flushing).
599 * Returns with *start = min(*start, lowest purged address)
600 * *end = max(*end, highest purged address)
602 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
603 int sync, int force_flush)
605 static DEFINE_SPINLOCK(purge_lock);
607 struct vmap_area *va;
608 struct vmap_area *n_va;
612 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
613 * should not expect such behaviour. This just simplifies locking for
614 * the case that isn't actually used at the moment anyway.
616 if (!sync && !force_flush) {
617 if (!spin_trylock(&purge_lock))
620 spin_lock(&purge_lock);
623 purge_fragmented_blocks_allcpus();
626 list_for_each_entry_rcu(va, &vmap_area_list, list) {
627 if (va->flags & VM_LAZY_FREE) {
628 if (va->va_start < *start)
629 *start = va->va_start;
630 if (va->va_end > *end)
632 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
633 list_add_tail(&va->purge_list, &valist);
634 va->flags |= VM_LAZY_FREEING;
635 va->flags &= ~VM_LAZY_FREE;
641 atomic_sub(nr, &vmap_lazy_nr);
643 if (nr || force_flush)
644 flush_tlb_kernel_range(*start, *end);
647 spin_lock(&vmap_area_lock);
648 list_for_each_entry_safe(va, n_va, &valist, purge_list)
649 __free_vmap_area(va);
650 spin_unlock(&vmap_area_lock);
652 spin_unlock(&purge_lock);
656 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
657 * is already purging.
659 static void try_purge_vmap_area_lazy(void)
661 unsigned long start = ULONG_MAX, end = 0;
663 __purge_vmap_area_lazy(&start, &end, 0, 0);
667 * Kick off a purge of the outstanding lazy areas.
669 static void purge_vmap_area_lazy(void)
671 unsigned long start = ULONG_MAX, end = 0;
673 __purge_vmap_area_lazy(&start, &end, 1, 0);
677 * Free a vmap area, caller ensuring that the area has been unmapped
678 * and flush_cache_vunmap had been called for the correct range
681 static void free_vmap_area_noflush(struct vmap_area *va)
683 va->flags |= VM_LAZY_FREE;
684 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
685 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
686 try_purge_vmap_area_lazy();
690 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
691 * called for the correct range previously.
693 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
696 free_vmap_area_noflush(va);
700 * Free and unmap a vmap area
702 static void free_unmap_vmap_area(struct vmap_area *va)
704 flush_cache_vunmap(va->va_start, va->va_end);
705 free_unmap_vmap_area_noflush(va);
708 static struct vmap_area *find_vmap_area(unsigned long addr)
710 struct vmap_area *va;
712 spin_lock(&vmap_area_lock);
713 va = __find_vmap_area(addr);
714 spin_unlock(&vmap_area_lock);
719 static void free_unmap_vmap_area_addr(unsigned long addr)
721 struct vmap_area *va;
723 va = find_vmap_area(addr);
725 free_unmap_vmap_area(va);
729 /*** Per cpu kva allocator ***/
732 * vmap space is limited especially on 32 bit architectures. Ensure there is
733 * room for at least 16 percpu vmap blocks per CPU.
736 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
737 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
738 * instead (we just need a rough idea)
740 #if BITS_PER_LONG == 32
741 #define VMALLOC_SPACE (128UL*1024*1024)
743 #define VMALLOC_SPACE (128UL*1024*1024*1024)
746 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
747 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
748 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
749 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
750 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
751 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
752 #define VMAP_BBMAP_BITS \
753 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
754 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
755 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
757 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
759 static bool vmap_initialized __read_mostly = false;
761 struct vmap_block_queue {
763 struct list_head free;
768 struct vmap_area *va;
769 unsigned long free, dirty;
770 unsigned long dirty_min, dirty_max; /*< dirty range */
771 struct list_head free_list;
772 struct rcu_head rcu_head;
773 struct list_head purge;
776 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
777 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
780 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
781 * in the free path. Could get rid of this if we change the API to return a
782 * "cookie" from alloc, to be passed to free. But no big deal yet.
784 static DEFINE_SPINLOCK(vmap_block_tree_lock);
785 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
788 * We should probably have a fallback mechanism to allocate virtual memory
789 * out of partially filled vmap blocks. However vmap block sizing should be
790 * fairly reasonable according to the vmalloc size, so it shouldn't be a
794 static unsigned long addr_to_vb_idx(unsigned long addr)
796 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
797 addr /= VMAP_BLOCK_SIZE;
801 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
805 addr = va_start + (pages_off << PAGE_SHIFT);
806 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
811 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
812 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
813 * @order: how many 2^order pages should be occupied in newly allocated block
814 * @gfp_mask: flags for the page level allocator
816 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
818 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
820 struct vmap_block_queue *vbq;
821 struct vmap_block *vb;
822 struct vmap_area *va;
823 unsigned long vb_idx;
827 node = numa_node_id();
829 vb = kmalloc_node(sizeof(struct vmap_block),
830 gfp_mask & GFP_RECLAIM_MASK, node);
832 return ERR_PTR(-ENOMEM);
834 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
835 VMALLOC_START, VMALLOC_END,
842 err = radix_tree_preload(gfp_mask);
849 vaddr = vmap_block_vaddr(va->va_start, 0);
850 spin_lock_init(&vb->lock);
852 /* At least something should be left free */
853 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
854 vb->free = VMAP_BBMAP_BITS - (1UL << order);
856 vb->dirty_min = VMAP_BBMAP_BITS;
858 INIT_LIST_HEAD(&vb->free_list);
860 vb_idx = addr_to_vb_idx(va->va_start);
861 spin_lock(&vmap_block_tree_lock);
862 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
863 spin_unlock(&vmap_block_tree_lock);
865 radix_tree_preload_end();
867 cpu = get_cpu_light();
868 vbq = this_cpu_ptr(&vmap_block_queue);
869 spin_lock(&vbq->lock);
870 list_add_tail_rcu(&vb->free_list, &vbq->free);
871 spin_unlock(&vbq->lock);
877 static void free_vmap_block(struct vmap_block *vb)
879 struct vmap_block *tmp;
880 unsigned long vb_idx;
882 vb_idx = addr_to_vb_idx(vb->va->va_start);
883 spin_lock(&vmap_block_tree_lock);
884 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
885 spin_unlock(&vmap_block_tree_lock);
888 free_vmap_area_noflush(vb->va);
889 kfree_rcu(vb, rcu_head);
892 static void purge_fragmented_blocks(int cpu)
895 struct vmap_block *vb;
896 struct vmap_block *n_vb;
897 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
900 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
902 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
905 spin_lock(&vb->lock);
906 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
907 vb->free = 0; /* prevent further allocs after releasing lock */
908 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
910 vb->dirty_max = VMAP_BBMAP_BITS;
911 spin_lock(&vbq->lock);
912 list_del_rcu(&vb->free_list);
913 spin_unlock(&vbq->lock);
914 spin_unlock(&vb->lock);
915 list_add_tail(&vb->purge, &purge);
917 spin_unlock(&vb->lock);
921 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
922 list_del(&vb->purge);
927 static void purge_fragmented_blocks_allcpus(void)
931 for_each_possible_cpu(cpu)
932 purge_fragmented_blocks(cpu);
935 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
937 struct vmap_block_queue *vbq;
938 struct vmap_block *vb;
943 BUG_ON(offset_in_page(size));
944 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
945 if (WARN_ON(size == 0)) {
947 * Allocating 0 bytes isn't what caller wants since
948 * get_order(0) returns funny result. Just warn and terminate
953 order = get_order(size);
956 cpu = get_cpu_light();
957 vbq = this_cpu_ptr(&vmap_block_queue);
958 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
959 unsigned long pages_off;
961 spin_lock(&vb->lock);
962 if (vb->free < (1UL << order)) {
963 spin_unlock(&vb->lock);
967 pages_off = VMAP_BBMAP_BITS - vb->free;
968 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
969 vb->free -= 1UL << order;
971 spin_lock(&vbq->lock);
972 list_del_rcu(&vb->free_list);
973 spin_unlock(&vbq->lock);
976 spin_unlock(&vb->lock);
983 /* Allocate new block if nothing was found */
985 vaddr = new_vmap_block(order, gfp_mask);
990 static void vb_free(const void *addr, unsigned long size)
992 unsigned long offset;
993 unsigned long vb_idx;
995 struct vmap_block *vb;
997 BUG_ON(offset_in_page(size));
998 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1000 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1002 order = get_order(size);
1004 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1005 offset >>= PAGE_SHIFT;
1007 vb_idx = addr_to_vb_idx((unsigned long)addr);
1009 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1013 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1015 spin_lock(&vb->lock);
1017 /* Expand dirty range */
1018 vb->dirty_min = min(vb->dirty_min, offset);
1019 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1021 vb->dirty += 1UL << order;
1022 if (vb->dirty == VMAP_BBMAP_BITS) {
1024 spin_unlock(&vb->lock);
1025 free_vmap_block(vb);
1027 spin_unlock(&vb->lock);
1031 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1033 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1034 * to amortize TLB flushing overheads. What this means is that any page you
1035 * have now, may, in a former life, have been mapped into kernel virtual
1036 * address by the vmap layer and so there might be some CPUs with TLB entries
1037 * still referencing that page (additional to the regular 1:1 kernel mapping).
1039 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1040 * be sure that none of the pages we have control over will have any aliases
1041 * from the vmap layer.
1043 void vm_unmap_aliases(void)
1045 unsigned long start = ULONG_MAX, end = 0;
1049 if (unlikely(!vmap_initialized))
1052 for_each_possible_cpu(cpu) {
1053 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1054 struct vmap_block *vb;
1057 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1058 spin_lock(&vb->lock);
1060 unsigned long va_start = vb->va->va_start;
1063 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1064 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1066 start = min(s, start);
1071 spin_unlock(&vb->lock);
1076 __purge_vmap_area_lazy(&start, &end, 1, flush);
1078 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1081 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1082 * @mem: the pointer returned by vm_map_ram
1083 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1085 void vm_unmap_ram(const void *mem, unsigned int count)
1087 unsigned long size = count << PAGE_SHIFT;
1088 unsigned long addr = (unsigned long)mem;
1091 BUG_ON(addr < VMALLOC_START);
1092 BUG_ON(addr > VMALLOC_END);
1093 BUG_ON(addr & (PAGE_SIZE-1));
1095 debug_check_no_locks_freed(mem, size);
1096 vmap_debug_free_range(addr, addr+size);
1098 if (likely(count <= VMAP_MAX_ALLOC))
1101 free_unmap_vmap_area_addr(addr);
1103 EXPORT_SYMBOL(vm_unmap_ram);
1106 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1107 * @pages: an array of pointers to the pages to be mapped
1108 * @count: number of pages
1109 * @node: prefer to allocate data structures on this node
1110 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1112 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1113 * faster than vmap so it's good. But if you mix long-life and short-life
1114 * objects with vm_map_ram(), it could consume lots of address space through
1115 * fragmentation (especially on a 32bit machine). You could see failures in
1116 * the end. Please use this function for short-lived objects.
1118 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1120 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1122 unsigned long size = count << PAGE_SHIFT;
1126 if (likely(count <= VMAP_MAX_ALLOC)) {
1127 mem = vb_alloc(size, GFP_KERNEL);
1130 addr = (unsigned long)mem;
1132 struct vmap_area *va;
1133 va = alloc_vmap_area(size, PAGE_SIZE,
1134 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1138 addr = va->va_start;
1141 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1142 vm_unmap_ram(mem, count);
1147 EXPORT_SYMBOL(vm_map_ram);
1149 static struct vm_struct *vmlist __initdata;
1151 * vm_area_add_early - add vmap area early during boot
1152 * @vm: vm_struct to add
1154 * This function is used to add fixed kernel vm area to vmlist before
1155 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1156 * should contain proper values and the other fields should be zero.
1158 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1160 void __init vm_area_add_early(struct vm_struct *vm)
1162 struct vm_struct *tmp, **p;
1164 BUG_ON(vmap_initialized);
1165 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1166 if (tmp->addr >= vm->addr) {
1167 BUG_ON(tmp->addr < vm->addr + vm->size);
1170 BUG_ON(tmp->addr + tmp->size > vm->addr);
1177 * vm_area_register_early - register vmap area early during boot
1178 * @vm: vm_struct to register
1179 * @align: requested alignment
1181 * This function is used to register kernel vm area before
1182 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1183 * proper values on entry and other fields should be zero. On return,
1184 * vm->addr contains the allocated address.
1186 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1188 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1190 static size_t vm_init_off __initdata;
1193 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1194 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1196 vm->addr = (void *)addr;
1198 vm_area_add_early(vm);
1201 void __init vmalloc_init(void)
1203 struct vmap_area *va;
1204 struct vm_struct *tmp;
1207 for_each_possible_cpu(i) {
1208 struct vmap_block_queue *vbq;
1209 struct vfree_deferred *p;
1211 vbq = &per_cpu(vmap_block_queue, i);
1212 spin_lock_init(&vbq->lock);
1213 INIT_LIST_HEAD(&vbq->free);
1214 p = &per_cpu(vfree_deferred, i);
1215 init_llist_head(&p->list);
1216 INIT_WORK(&p->wq, free_work);
1219 /* Import existing vmlist entries. */
1220 for (tmp = vmlist; tmp; tmp = tmp->next) {
1221 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1222 va->flags = VM_VM_AREA;
1223 va->va_start = (unsigned long)tmp->addr;
1224 va->va_end = va->va_start + tmp->size;
1226 __insert_vmap_area(va);
1229 vmap_area_pcpu_hole = VMALLOC_END;
1231 vmap_initialized = true;
1235 * map_kernel_range_noflush - map kernel VM area with the specified pages
1236 * @addr: start of the VM area to map
1237 * @size: size of the VM area to map
1238 * @prot: page protection flags to use
1239 * @pages: pages to map
1241 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1242 * specify should have been allocated using get_vm_area() and its
1246 * This function does NOT do any cache flushing. The caller is
1247 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1248 * before calling this function.
1251 * The number of pages mapped on success, -errno on failure.
1253 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1254 pgprot_t prot, struct page **pages)
1256 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1260 * unmap_kernel_range_noflush - unmap kernel VM area
1261 * @addr: start of the VM area to unmap
1262 * @size: size of the VM area to unmap
1264 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1265 * specify should have been allocated using get_vm_area() and its
1269 * This function does NOT do any cache flushing. The caller is
1270 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1271 * before calling this function and flush_tlb_kernel_range() after.
1273 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1275 vunmap_page_range(addr, addr + size);
1277 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1280 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1281 * @addr: start of the VM area to unmap
1282 * @size: size of the VM area to unmap
1284 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1285 * the unmapping and tlb after.
1287 void unmap_kernel_range(unsigned long addr, unsigned long size)
1289 unsigned long end = addr + size;
1291 flush_cache_vunmap(addr, end);
1292 vunmap_page_range(addr, end);
1293 flush_tlb_kernel_range(addr, end);
1295 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1297 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1299 unsigned long addr = (unsigned long)area->addr;
1300 unsigned long end = addr + get_vm_area_size(area);
1303 err = vmap_page_range(addr, end, prot, pages);
1305 return err > 0 ? 0 : err;
1307 EXPORT_SYMBOL_GPL(map_vm_area);
1309 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1310 unsigned long flags, const void *caller)
1312 spin_lock(&vmap_area_lock);
1314 vm->addr = (void *)va->va_start;
1315 vm->size = va->va_end - va->va_start;
1316 vm->caller = caller;
1318 va->flags |= VM_VM_AREA;
1319 spin_unlock(&vmap_area_lock);
1322 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1325 * Before removing VM_UNINITIALIZED,
1326 * we should make sure that vm has proper values.
1327 * Pair with smp_rmb() in show_numa_info().
1330 vm->flags &= ~VM_UNINITIALIZED;
1333 static struct vm_struct *__get_vm_area_node(unsigned long size,
1334 unsigned long align, unsigned long flags, unsigned long start,
1335 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1337 struct vmap_area *va;
1338 struct vm_struct *area;
1340 BUG_ON(in_interrupt());
1341 if (flags & VM_IOREMAP)
1342 align = 1ul << clamp_t(int, fls_long(size),
1343 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1345 size = PAGE_ALIGN(size);
1346 if (unlikely(!size))
1349 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1350 if (unlikely(!area))
1353 if (!(flags & VM_NO_GUARD))
1356 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1362 setup_vmalloc_vm(area, va, flags, caller);
1367 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1368 unsigned long start, unsigned long end)
1370 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1371 GFP_KERNEL, __builtin_return_address(0));
1373 EXPORT_SYMBOL_GPL(__get_vm_area);
1375 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1376 unsigned long start, unsigned long end,
1379 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1380 GFP_KERNEL, caller);
1384 * get_vm_area - reserve a contiguous kernel virtual area
1385 * @size: size of the area
1386 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1388 * Search an area of @size in the kernel virtual mapping area,
1389 * and reserved it for out purposes. Returns the area descriptor
1390 * on success or %NULL on failure.
1392 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1394 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1395 NUMA_NO_NODE, GFP_KERNEL,
1396 __builtin_return_address(0));
1399 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1402 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1403 NUMA_NO_NODE, GFP_KERNEL, caller);
1407 * find_vm_area - find a continuous kernel virtual area
1408 * @addr: base address
1410 * Search for the kernel VM area starting at @addr, and return it.
1411 * It is up to the caller to do all required locking to keep the returned
1414 struct vm_struct *find_vm_area(const void *addr)
1416 struct vmap_area *va;
1418 va = find_vmap_area((unsigned long)addr);
1419 if (va && va->flags & VM_VM_AREA)
1426 * remove_vm_area - find and remove a continuous kernel virtual area
1427 * @addr: base address
1429 * Search for the kernel VM area starting at @addr, and remove it.
1430 * This function returns the found VM area, but using it is NOT safe
1431 * on SMP machines, except for its size or flags.
1433 struct vm_struct *remove_vm_area(const void *addr)
1435 struct vmap_area *va;
1437 va = find_vmap_area((unsigned long)addr);
1438 if (va && va->flags & VM_VM_AREA) {
1439 struct vm_struct *vm = va->vm;
1441 spin_lock(&vmap_area_lock);
1443 va->flags &= ~VM_VM_AREA;
1444 spin_unlock(&vmap_area_lock);
1446 vmap_debug_free_range(va->va_start, va->va_end);
1447 kasan_free_shadow(vm);
1448 free_unmap_vmap_area(va);
1455 static void __vunmap(const void *addr, int deallocate_pages)
1457 struct vm_struct *area;
1462 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1466 area = remove_vm_area(addr);
1467 if (unlikely(!area)) {
1468 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1473 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1474 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1476 if (deallocate_pages) {
1479 for (i = 0; i < area->nr_pages; i++) {
1480 struct page *page = area->pages[i];
1486 if (area->flags & VM_VPAGES)
1497 * vfree - release memory allocated by vmalloc()
1498 * @addr: memory base address
1500 * Free the virtually continuous memory area starting at @addr, as
1501 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1502 * NULL, no operation is performed.
1504 * Must not be called in NMI context (strictly speaking, only if we don't
1505 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1506 * conventions for vfree() arch-depenedent would be a really bad idea)
1508 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1510 void vfree(const void *addr)
1514 kmemleak_free(addr);
1518 if (unlikely(in_interrupt())) {
1519 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1520 if (llist_add((struct llist_node *)addr, &p->list))
1521 schedule_work(&p->wq);
1525 EXPORT_SYMBOL(vfree);
1528 * vunmap - release virtual mapping obtained by vmap()
1529 * @addr: memory base address
1531 * Free the virtually contiguous memory area starting at @addr,
1532 * which was created from the page array passed to vmap().
1534 * Must not be called in interrupt context.
1536 void vunmap(const void *addr)
1538 BUG_ON(in_interrupt());
1543 EXPORT_SYMBOL(vunmap);
1546 * vmap - map an array of pages into virtually contiguous space
1547 * @pages: array of page pointers
1548 * @count: number of pages to map
1549 * @flags: vm_area->flags
1550 * @prot: page protection for the mapping
1552 * Maps @count pages from @pages into contiguous kernel virtual
1555 void *vmap(struct page **pages, unsigned int count,
1556 unsigned long flags, pgprot_t prot)
1558 struct vm_struct *area;
1562 if (count > totalram_pages)
1565 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1566 __builtin_return_address(0));
1570 if (map_vm_area(area, prot, pages)) {
1577 EXPORT_SYMBOL(vmap);
1579 static void *__vmalloc_node(unsigned long size, unsigned long align,
1580 gfp_t gfp_mask, pgprot_t prot,
1581 int node, const void *caller);
1582 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1583 pgprot_t prot, int node)
1585 const int order = 0;
1586 struct page **pages;
1587 unsigned int nr_pages, array_size, i;
1588 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1589 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1591 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1592 array_size = (nr_pages * sizeof(struct page *));
1594 area->nr_pages = nr_pages;
1595 /* Please note that the recursion is strictly bounded. */
1596 if (array_size > PAGE_SIZE) {
1597 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1598 PAGE_KERNEL, node, area->caller);
1599 area->flags |= VM_VPAGES;
1601 pages = kmalloc_node(array_size, nested_gfp, node);
1603 area->pages = pages;
1605 remove_vm_area(area->addr);
1610 for (i = 0; i < area->nr_pages; i++) {
1613 if (node == NUMA_NO_NODE)
1614 page = alloc_page(alloc_mask);
1616 page = alloc_pages_node(node, alloc_mask, order);
1618 if (unlikely(!page)) {
1619 /* Successfully allocated i pages, free them in __vunmap() */
1623 area->pages[i] = page;
1624 if (gfpflags_allow_blocking(gfp_mask))
1628 if (map_vm_area(area, prot, pages))
1633 warn_alloc_failed(gfp_mask, order,
1634 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1635 (area->nr_pages*PAGE_SIZE), area->size);
1641 * __vmalloc_node_range - allocate virtually contiguous memory
1642 * @size: allocation size
1643 * @align: desired alignment
1644 * @start: vm area range start
1645 * @end: vm area range end
1646 * @gfp_mask: flags for the page level allocator
1647 * @prot: protection mask for the allocated pages
1648 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1649 * @node: node to use for allocation or NUMA_NO_NODE
1650 * @caller: caller's return address
1652 * Allocate enough pages to cover @size from the page level
1653 * allocator with @gfp_mask flags. Map them into contiguous
1654 * kernel virtual space, using a pagetable protection of @prot.
1656 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1657 unsigned long start, unsigned long end, gfp_t gfp_mask,
1658 pgprot_t prot, unsigned long vm_flags, int node,
1661 struct vm_struct *area;
1663 unsigned long real_size = size;
1665 size = PAGE_ALIGN(size);
1666 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1669 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1670 vm_flags, start, end, node, gfp_mask, caller);
1674 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1679 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1680 * flag. It means that vm_struct is not fully initialized.
1681 * Now, it is fully initialized, so remove this flag here.
1683 clear_vm_uninitialized_flag(area);
1686 * A ref_count = 2 is needed because vm_struct allocated in
1687 * __get_vm_area_node() contains a reference to the virtual address of
1688 * the vmalloc'ed block.
1690 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1695 warn_alloc_failed(gfp_mask, 0,
1696 "vmalloc: allocation failure: %lu bytes\n",
1702 * __vmalloc_node - allocate virtually contiguous memory
1703 * @size: allocation size
1704 * @align: desired alignment
1705 * @gfp_mask: flags for the page level allocator
1706 * @prot: protection mask for the allocated pages
1707 * @node: node to use for allocation or NUMA_NO_NODE
1708 * @caller: caller's return address
1710 * Allocate enough pages to cover @size from the page level
1711 * allocator with @gfp_mask flags. Map them into contiguous
1712 * kernel virtual space, using a pagetable protection of @prot.
1714 static void *__vmalloc_node(unsigned long size, unsigned long align,
1715 gfp_t gfp_mask, pgprot_t prot,
1716 int node, const void *caller)
1718 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1719 gfp_mask, prot, 0, node, caller);
1722 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1724 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1725 __builtin_return_address(0));
1727 EXPORT_SYMBOL(__vmalloc);
1729 static inline void *__vmalloc_node_flags(unsigned long size,
1730 int node, gfp_t flags)
1732 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1733 node, __builtin_return_address(0));
1737 * vmalloc - allocate virtually contiguous memory
1738 * @size: allocation size
1739 * Allocate enough pages to cover @size from the page level
1740 * allocator and map them into contiguous kernel virtual space.
1742 * For tight control over page level allocator and protection flags
1743 * use __vmalloc() instead.
1745 void *vmalloc(unsigned long size)
1747 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1748 GFP_KERNEL | __GFP_HIGHMEM);
1750 EXPORT_SYMBOL(vmalloc);
1753 * vzalloc - allocate virtually contiguous memory with zero fill
1754 * @size: allocation size
1755 * Allocate enough pages to cover @size from the page level
1756 * allocator and map them into contiguous kernel virtual space.
1757 * The memory allocated is set to zero.
1759 * For tight control over page level allocator and protection flags
1760 * use __vmalloc() instead.
1762 void *vzalloc(unsigned long size)
1764 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1765 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1767 EXPORT_SYMBOL(vzalloc);
1770 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1771 * @size: allocation size
1773 * The resulting memory area is zeroed so it can be mapped to userspace
1774 * without leaking data.
1776 void *vmalloc_user(unsigned long size)
1778 struct vm_struct *area;
1781 ret = __vmalloc_node(size, SHMLBA,
1782 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1783 PAGE_KERNEL, NUMA_NO_NODE,
1784 __builtin_return_address(0));
1786 area = find_vm_area(ret);
1787 area->flags |= VM_USERMAP;
1791 EXPORT_SYMBOL(vmalloc_user);
1794 * vmalloc_node - allocate memory on a specific node
1795 * @size: allocation size
1798 * Allocate enough pages to cover @size from the page level
1799 * allocator and map them into contiguous kernel virtual space.
1801 * For tight control over page level allocator and protection flags
1802 * use __vmalloc() instead.
1804 void *vmalloc_node(unsigned long size, int node)
1806 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1807 node, __builtin_return_address(0));
1809 EXPORT_SYMBOL(vmalloc_node);
1812 * vzalloc_node - allocate memory on a specific node with zero fill
1813 * @size: allocation size
1816 * Allocate enough pages to cover @size from the page level
1817 * allocator and map them into contiguous kernel virtual space.
1818 * The memory allocated is set to zero.
1820 * For tight control over page level allocator and protection flags
1821 * use __vmalloc_node() instead.
1823 void *vzalloc_node(unsigned long size, int node)
1825 return __vmalloc_node_flags(size, node,
1826 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1828 EXPORT_SYMBOL(vzalloc_node);
1830 #ifndef PAGE_KERNEL_EXEC
1831 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1835 * vmalloc_exec - allocate virtually contiguous, executable memory
1836 * @size: allocation size
1838 * Kernel-internal function to allocate enough pages to cover @size
1839 * the page level allocator and map them into contiguous and
1840 * executable kernel virtual space.
1842 * For tight control over page level allocator and protection flags
1843 * use __vmalloc() instead.
1846 void *vmalloc_exec(unsigned long size)
1848 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1849 NUMA_NO_NODE, __builtin_return_address(0));
1852 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1853 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1854 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1855 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1857 #define GFP_VMALLOC32 GFP_KERNEL
1861 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1862 * @size: allocation size
1864 * Allocate enough 32bit PA addressable pages to cover @size from the
1865 * page level allocator and map them into contiguous kernel virtual space.
1867 void *vmalloc_32(unsigned long size)
1869 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1870 NUMA_NO_NODE, __builtin_return_address(0));
1872 EXPORT_SYMBOL(vmalloc_32);
1875 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1876 * @size: allocation size
1878 * The resulting memory area is 32bit addressable and zeroed so it can be
1879 * mapped to userspace without leaking data.
1881 void *vmalloc_32_user(unsigned long size)
1883 struct vm_struct *area;
1886 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1887 NUMA_NO_NODE, __builtin_return_address(0));
1889 area = find_vm_area(ret);
1890 area->flags |= VM_USERMAP;
1894 EXPORT_SYMBOL(vmalloc_32_user);
1897 * small helper routine , copy contents to buf from addr.
1898 * If the page is not present, fill zero.
1901 static int aligned_vread(char *buf, char *addr, unsigned long count)
1907 unsigned long offset, length;
1909 offset = offset_in_page(addr);
1910 length = PAGE_SIZE - offset;
1913 p = vmalloc_to_page(addr);
1915 * To do safe access to this _mapped_ area, we need
1916 * lock. But adding lock here means that we need to add
1917 * overhead of vmalloc()/vfree() calles for this _debug_
1918 * interface, rarely used. Instead of that, we'll use
1919 * kmap() and get small overhead in this access function.
1923 * we can expect USER0 is not used (see vread/vwrite's
1924 * function description)
1926 void *map = kmap_atomic(p);
1927 memcpy(buf, map + offset, length);
1930 memset(buf, 0, length);
1940 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1946 unsigned long offset, length;
1948 offset = offset_in_page(addr);
1949 length = PAGE_SIZE - offset;
1952 p = vmalloc_to_page(addr);
1954 * To do safe access to this _mapped_ area, we need
1955 * lock. But adding lock here means that we need to add
1956 * overhead of vmalloc()/vfree() calles for this _debug_
1957 * interface, rarely used. Instead of that, we'll use
1958 * kmap() and get small overhead in this access function.
1962 * we can expect USER0 is not used (see vread/vwrite's
1963 * function description)
1965 void *map = kmap_atomic(p);
1966 memcpy(map + offset, buf, length);
1978 * vread() - read vmalloc area in a safe way.
1979 * @buf: buffer for reading data
1980 * @addr: vm address.
1981 * @count: number of bytes to be read.
1983 * Returns # of bytes which addr and buf should be increased.
1984 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1985 * includes any intersect with alive vmalloc area.
1987 * This function checks that addr is a valid vmalloc'ed area, and
1988 * copy data from that area to a given buffer. If the given memory range
1989 * of [addr...addr+count) includes some valid address, data is copied to
1990 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1991 * IOREMAP area is treated as memory hole and no copy is done.
1993 * If [addr...addr+count) doesn't includes any intersects with alive
1994 * vm_struct area, returns 0. @buf should be kernel's buffer.
1996 * Note: In usual ops, vread() is never necessary because the caller
1997 * should know vmalloc() area is valid and can use memcpy().
1998 * This is for routines which have to access vmalloc area without
1999 * any informaion, as /dev/kmem.
2003 long vread(char *buf, char *addr, unsigned long count)
2005 struct vmap_area *va;
2006 struct vm_struct *vm;
2007 char *vaddr, *buf_start = buf;
2008 unsigned long buflen = count;
2011 /* Don't allow overflow */
2012 if ((unsigned long) addr + count < count)
2013 count = -(unsigned long) addr;
2015 spin_lock(&vmap_area_lock);
2016 list_for_each_entry(va, &vmap_area_list, list) {
2020 if (!(va->flags & VM_VM_AREA))
2024 vaddr = (char *) vm->addr;
2025 if (addr >= vaddr + get_vm_area_size(vm))
2027 while (addr < vaddr) {
2035 n = vaddr + get_vm_area_size(vm) - addr;
2038 if (!(vm->flags & VM_IOREMAP))
2039 aligned_vread(buf, addr, n);
2040 else /* IOREMAP area is treated as memory hole */
2047 spin_unlock(&vmap_area_lock);
2049 if (buf == buf_start)
2051 /* zero-fill memory holes */
2052 if (buf != buf_start + buflen)
2053 memset(buf, 0, buflen - (buf - buf_start));
2059 * vwrite() - write vmalloc area in a safe way.
2060 * @buf: buffer for source data
2061 * @addr: vm address.
2062 * @count: number of bytes to be read.
2064 * Returns # of bytes which addr and buf should be incresed.
2065 * (same number to @count).
2066 * If [addr...addr+count) doesn't includes any intersect with valid
2067 * vmalloc area, returns 0.
2069 * This function checks that addr is a valid vmalloc'ed area, and
2070 * copy data from a buffer to the given addr. If specified range of
2071 * [addr...addr+count) includes some valid address, data is copied from
2072 * proper area of @buf. If there are memory holes, no copy to hole.
2073 * IOREMAP area is treated as memory hole and no copy is done.
2075 * If [addr...addr+count) doesn't includes any intersects with alive
2076 * vm_struct area, returns 0. @buf should be kernel's buffer.
2078 * Note: In usual ops, vwrite() is never necessary because the caller
2079 * should know vmalloc() area is valid and can use memcpy().
2080 * This is for routines which have to access vmalloc area without
2081 * any informaion, as /dev/kmem.
2084 long vwrite(char *buf, char *addr, unsigned long count)
2086 struct vmap_area *va;
2087 struct vm_struct *vm;
2089 unsigned long n, buflen;
2092 /* Don't allow overflow */
2093 if ((unsigned long) addr + count < count)
2094 count = -(unsigned long) addr;
2097 spin_lock(&vmap_area_lock);
2098 list_for_each_entry(va, &vmap_area_list, list) {
2102 if (!(va->flags & VM_VM_AREA))
2106 vaddr = (char *) vm->addr;
2107 if (addr >= vaddr + get_vm_area_size(vm))
2109 while (addr < vaddr) {
2116 n = vaddr + get_vm_area_size(vm) - addr;
2119 if (!(vm->flags & VM_IOREMAP)) {
2120 aligned_vwrite(buf, addr, n);
2128 spin_unlock(&vmap_area_lock);
2135 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2136 * @vma: vma to cover
2137 * @uaddr: target user address to start at
2138 * @kaddr: virtual address of vmalloc kernel memory
2139 * @size: size of map area
2141 * Returns: 0 for success, -Exxx on failure
2143 * This function checks that @kaddr is a valid vmalloc'ed area,
2144 * and that it is big enough to cover the range starting at
2145 * @uaddr in @vma. Will return failure if that criteria isn't
2148 * Similar to remap_pfn_range() (see mm/memory.c)
2150 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2151 void *kaddr, unsigned long size)
2153 struct vm_struct *area;
2155 size = PAGE_ALIGN(size);
2157 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2160 area = find_vm_area(kaddr);
2164 if (!(area->flags & VM_USERMAP))
2167 if (kaddr + size > area->addr + area->size)
2171 struct page *page = vmalloc_to_page(kaddr);
2174 ret = vm_insert_page(vma, uaddr, page);
2183 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2187 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2190 * remap_vmalloc_range - map vmalloc pages to userspace
2191 * @vma: vma to cover (map full range of vma)
2192 * @addr: vmalloc memory
2193 * @pgoff: number of pages into addr before first page to map
2195 * Returns: 0 for success, -Exxx on failure
2197 * This function checks that addr is a valid vmalloc'ed area, and
2198 * that it is big enough to cover the vma. Will return failure if
2199 * that criteria isn't met.
2201 * Similar to remap_pfn_range() (see mm/memory.c)
2203 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2204 unsigned long pgoff)
2206 return remap_vmalloc_range_partial(vma, vma->vm_start,
2207 addr + (pgoff << PAGE_SHIFT),
2208 vma->vm_end - vma->vm_start);
2210 EXPORT_SYMBOL(remap_vmalloc_range);
2213 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2216 void __weak vmalloc_sync_all(void)
2221 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2233 * alloc_vm_area - allocate a range of kernel address space
2234 * @size: size of the area
2235 * @ptes: returns the PTEs for the address space
2237 * Returns: NULL on failure, vm_struct on success
2239 * This function reserves a range of kernel address space, and
2240 * allocates pagetables to map that range. No actual mappings
2243 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2244 * allocated for the VM area are returned.
2246 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2248 struct vm_struct *area;
2250 area = get_vm_area_caller(size, VM_IOREMAP,
2251 __builtin_return_address(0));
2256 * This ensures that page tables are constructed for this region
2257 * of kernel virtual address space and mapped into init_mm.
2259 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2260 size, f, ptes ? &ptes : NULL)) {
2267 EXPORT_SYMBOL_GPL(alloc_vm_area);
2269 void free_vm_area(struct vm_struct *area)
2271 struct vm_struct *ret;
2272 ret = remove_vm_area(area->addr);
2273 BUG_ON(ret != area);
2276 EXPORT_SYMBOL_GPL(free_vm_area);
2279 static struct vmap_area *node_to_va(struct rb_node *n)
2281 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2285 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2286 * @end: target address
2287 * @pnext: out arg for the next vmap_area
2288 * @pprev: out arg for the previous vmap_area
2290 * Returns: %true if either or both of next and prev are found,
2291 * %false if no vmap_area exists
2293 * Find vmap_areas end addresses of which enclose @end. ie. if not
2294 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2296 static bool pvm_find_next_prev(unsigned long end,
2297 struct vmap_area **pnext,
2298 struct vmap_area **pprev)
2300 struct rb_node *n = vmap_area_root.rb_node;
2301 struct vmap_area *va = NULL;
2304 va = rb_entry(n, struct vmap_area, rb_node);
2305 if (end < va->va_end)
2307 else if (end > va->va_end)
2316 if (va->va_end > end) {
2318 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2321 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2327 * pvm_determine_end - find the highest aligned address between two vmap_areas
2328 * @pnext: in/out arg for the next vmap_area
2329 * @pprev: in/out arg for the previous vmap_area
2332 * Returns: determined end address
2334 * Find the highest aligned address between *@pnext and *@pprev below
2335 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2336 * down address is between the end addresses of the two vmap_areas.
2338 * Please note that the address returned by this function may fall
2339 * inside *@pnext vmap_area. The caller is responsible for checking
2342 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2343 struct vmap_area **pprev,
2344 unsigned long align)
2346 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2350 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2354 while (*pprev && (*pprev)->va_end > addr) {
2356 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2363 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2364 * @offsets: array containing offset of each area
2365 * @sizes: array containing size of each area
2366 * @nr_vms: the number of areas to allocate
2367 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2369 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2370 * vm_structs on success, %NULL on failure
2372 * Percpu allocator wants to use congruent vm areas so that it can
2373 * maintain the offsets among percpu areas. This function allocates
2374 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2375 * be scattered pretty far, distance between two areas easily going up
2376 * to gigabytes. To avoid interacting with regular vmallocs, these
2377 * areas are allocated from top.
2379 * Despite its complicated look, this allocator is rather simple. It
2380 * does everything top-down and scans areas from the end looking for
2381 * matching slot. While scanning, if any of the areas overlaps with
2382 * existing vmap_area, the base address is pulled down to fit the
2383 * area. Scanning is repeated till all the areas fit and then all
2384 * necessary data structres are inserted and the result is returned.
2386 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2387 const size_t *sizes, int nr_vms,
2390 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2391 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2392 struct vmap_area **vas, *prev, *next;
2393 struct vm_struct **vms;
2394 int area, area2, last_area, term_area;
2395 unsigned long base, start, end, last_end;
2396 bool purged = false;
2398 /* verify parameters and allocate data structures */
2399 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2400 for (last_area = 0, area = 0; area < nr_vms; area++) {
2401 start = offsets[area];
2402 end = start + sizes[area];
2404 /* is everything aligned properly? */
2405 BUG_ON(!IS_ALIGNED(offsets[area], align));
2406 BUG_ON(!IS_ALIGNED(sizes[area], align));
2408 /* detect the area with the highest address */
2409 if (start > offsets[last_area])
2412 for (area2 = 0; area2 < nr_vms; area2++) {
2413 unsigned long start2 = offsets[area2];
2414 unsigned long end2 = start2 + sizes[area2];
2419 BUG_ON(start2 >= start && start2 < end);
2420 BUG_ON(end2 <= end && end2 > start);
2423 last_end = offsets[last_area] + sizes[last_area];
2425 if (vmalloc_end - vmalloc_start < last_end) {
2430 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2431 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2435 for (area = 0; area < nr_vms; area++) {
2436 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2437 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2438 if (!vas[area] || !vms[area])
2442 spin_lock(&vmap_area_lock);
2444 /* start scanning - we scan from the top, begin with the last area */
2445 area = term_area = last_area;
2446 start = offsets[area];
2447 end = start + sizes[area];
2449 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2450 base = vmalloc_end - last_end;
2453 base = pvm_determine_end(&next, &prev, align) - end;
2456 BUG_ON(next && next->va_end <= base + end);
2457 BUG_ON(prev && prev->va_end > base + end);
2460 * base might have underflowed, add last_end before
2463 if (base + last_end < vmalloc_start + last_end) {
2464 spin_unlock(&vmap_area_lock);
2466 purge_vmap_area_lazy();
2474 * If next overlaps, move base downwards so that it's
2475 * right below next and then recheck.
2477 if (next && next->va_start < base + end) {
2478 base = pvm_determine_end(&next, &prev, align) - end;
2484 * If prev overlaps, shift down next and prev and move
2485 * base so that it's right below new next and then
2488 if (prev && prev->va_end > base + start) {
2490 prev = node_to_va(rb_prev(&next->rb_node));
2491 base = pvm_determine_end(&next, &prev, align) - end;
2497 * This area fits, move on to the previous one. If
2498 * the previous one is the terminal one, we're done.
2500 area = (area + nr_vms - 1) % nr_vms;
2501 if (area == term_area)
2503 start = offsets[area];
2504 end = start + sizes[area];
2505 pvm_find_next_prev(base + end, &next, &prev);
2508 /* we've found a fitting base, insert all va's */
2509 for (area = 0; area < nr_vms; area++) {
2510 struct vmap_area *va = vas[area];
2512 va->va_start = base + offsets[area];
2513 va->va_end = va->va_start + sizes[area];
2514 __insert_vmap_area(va);
2517 vmap_area_pcpu_hole = base + offsets[last_area];
2519 spin_unlock(&vmap_area_lock);
2521 /* insert all vm's */
2522 for (area = 0; area < nr_vms; area++)
2523 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2530 for (area = 0; area < nr_vms; area++) {
2541 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2542 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2543 * @nr_vms: the number of allocated areas
2545 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2547 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2551 for (i = 0; i < nr_vms; i++)
2552 free_vm_area(vms[i]);
2555 #endif /* CONFIG_SMP */
2557 #ifdef CONFIG_PROC_FS
2558 static void *s_start(struct seq_file *m, loff_t *pos)
2559 __acquires(&vmap_area_lock)
2562 struct vmap_area *va;
2564 spin_lock(&vmap_area_lock);
2565 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2566 while (n > 0 && &va->list != &vmap_area_list) {
2568 va = list_entry(va->list.next, typeof(*va), list);
2570 if (!n && &va->list != &vmap_area_list)
2577 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2579 struct vmap_area *va = p, *next;
2582 next = list_entry(va->list.next, typeof(*va), list);
2583 if (&next->list != &vmap_area_list)
2589 static void s_stop(struct seq_file *m, void *p)
2590 __releases(&vmap_area_lock)
2592 spin_unlock(&vmap_area_lock);
2595 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2597 if (IS_ENABLED(CONFIG_NUMA)) {
2598 unsigned int nr, *counters = m->private;
2603 if (v->flags & VM_UNINITIALIZED)
2605 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2608 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2610 for (nr = 0; nr < v->nr_pages; nr++)
2611 counters[page_to_nid(v->pages[nr])]++;
2613 for_each_node_state(nr, N_HIGH_MEMORY)
2615 seq_printf(m, " N%u=%u", nr, counters[nr]);
2619 static int s_show(struct seq_file *m, void *p)
2621 struct vmap_area *va = p;
2622 struct vm_struct *v;
2625 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2626 * behalf of vmap area is being tear down or vm_map_ram allocation.
2628 if (!(va->flags & VM_VM_AREA))
2633 seq_printf(m, "0x%pK-0x%pK %7ld",
2634 v->addr, v->addr + v->size, v->size);
2637 seq_printf(m, " %pS", v->caller);
2640 seq_printf(m, " pages=%d", v->nr_pages);
2643 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2645 if (v->flags & VM_IOREMAP)
2646 seq_puts(m, " ioremap");
2648 if (v->flags & VM_ALLOC)
2649 seq_puts(m, " vmalloc");
2651 if (v->flags & VM_MAP)
2652 seq_puts(m, " vmap");
2654 if (v->flags & VM_USERMAP)
2655 seq_puts(m, " user");
2657 if (v->flags & VM_VPAGES)
2658 seq_puts(m, " vpages");
2660 show_numa_info(m, v);
2665 static const struct seq_operations vmalloc_op = {
2672 static int vmalloc_open(struct inode *inode, struct file *file)
2674 if (IS_ENABLED(CONFIG_NUMA))
2675 return seq_open_private(file, &vmalloc_op,
2676 nr_node_ids * sizeof(unsigned int));
2678 return seq_open(file, &vmalloc_op);
2681 static const struct file_operations proc_vmalloc_operations = {
2682 .open = vmalloc_open,
2684 .llseek = seq_lseek,
2685 .release = seq_release_private,
2688 static int __init proc_vmalloc_init(void)
2690 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2693 module_init(proc_vmalloc_init);