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/slab.h>
16 #include <linux/spinlock.h>
17 #include <linux/interrupt.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/debugobjects.h>
21 #include <linux/kallsyms.h>
22 #include <linux/list.h>
23 #include <linux/rbtree.h>
24 #include <linux/radix-tree.h>
25 #include <linux/rcupdate.h>
26 #include <linux/bootmem.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
30 #include <asm/atomic.h>
31 #include <asm/uaccess.h>
32 #include <asm/tlbflush.h>
35 /*** Page table manipulation functions ***/
37 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
41 pte = pte_offset_kernel(pmd, addr);
43 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
44 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
45 } while (pte++, addr += PAGE_SIZE, addr != end);
48 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
53 pmd = pmd_offset(pud, addr);
55 next = pmd_addr_end(addr, end);
56 if (pmd_none_or_clear_bad(pmd))
58 vunmap_pte_range(pmd, addr, next);
59 } while (pmd++, addr = next, addr != end);
62 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
67 pud = pud_offset(pgd, addr);
69 next = pud_addr_end(addr, end);
70 if (pud_none_or_clear_bad(pud))
72 vunmap_pmd_range(pud, addr, next);
73 } while (pud++, addr = next, addr != end);
76 static void vunmap_page_range(unsigned long addr, unsigned long end)
82 pgd = pgd_offset_k(addr);
84 next = pgd_addr_end(addr, end);
85 if (pgd_none_or_clear_bad(pgd))
87 vunmap_pud_range(pgd, addr, next);
88 } while (pgd++, addr = next, addr != end);
91 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
92 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
97 * nr is a running index into the array which helps higher level
98 * callers keep track of where we're up to.
101 pte = pte_alloc_kernel(pmd, addr);
105 struct page *page = pages[*nr];
107 if (WARN_ON(!pte_none(*pte)))
111 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
113 } while (pte++, addr += PAGE_SIZE, addr != end);
117 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
118 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
123 pmd = pmd_alloc(&init_mm, pud, addr);
127 next = pmd_addr_end(addr, end);
128 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
130 } while (pmd++, addr = next, addr != end);
134 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
135 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
140 pud = pud_alloc(&init_mm, pgd, addr);
144 next = pud_addr_end(addr, end);
145 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
147 } while (pud++, addr = next, addr != end);
152 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
153 * will have pfns corresponding to the "pages" array.
155 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
157 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
158 pgprot_t prot, struct page **pages)
162 unsigned long addr = start;
167 pgd = pgd_offset_k(addr);
169 next = pgd_addr_end(addr, end);
170 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
173 } while (pgd++, addr = next, addr != end);
180 static int vmap_page_range(unsigned long start, unsigned long end,
181 pgprot_t prot, struct page **pages)
185 ret = vmap_page_range_noflush(start, end, prot, pages);
186 flush_cache_vmap(start, end);
190 static inline int is_vmalloc_or_module_addr(const void *x)
193 * ARM, x86-64 and sparc64 put modules in a special place,
194 * and fall back on vmalloc() if that fails. Others
195 * just put it in the vmalloc space.
197 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
198 unsigned long addr = (unsigned long)x;
199 if (addr >= MODULES_VADDR && addr < MODULES_END)
202 return is_vmalloc_addr(x);
206 * Walk a vmap address to the struct page it maps.
208 struct page *vmalloc_to_page(const void *vmalloc_addr)
210 unsigned long addr = (unsigned long) vmalloc_addr;
211 struct page *page = NULL;
212 pgd_t *pgd = pgd_offset_k(addr);
215 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
216 * architectures that do not vmalloc module space
218 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
220 if (!pgd_none(*pgd)) {
221 pud_t *pud = pud_offset(pgd, addr);
222 if (!pud_none(*pud)) {
223 pmd_t *pmd = pmd_offset(pud, addr);
224 if (!pmd_none(*pmd)) {
227 ptep = pte_offset_map(pmd, addr);
229 if (pte_present(pte))
230 page = pte_page(pte);
237 EXPORT_SYMBOL(vmalloc_to_page);
240 * Map a vmalloc()-space virtual address to the physical page frame number.
242 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
244 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
246 EXPORT_SYMBOL(vmalloc_to_pfn);
249 /*** Global kva allocator ***/
251 #define VM_LAZY_FREE 0x01
252 #define VM_LAZY_FREEING 0x02
253 #define VM_VM_AREA 0x04
256 unsigned long va_start;
257 unsigned long va_end;
259 struct rb_node rb_node; /* address sorted rbtree */
260 struct list_head list; /* address sorted list */
261 struct list_head purge_list; /* "lazy purge" list */
263 struct rcu_head rcu_head;
266 static DEFINE_SPINLOCK(vmap_area_lock);
267 static struct rb_root vmap_area_root = RB_ROOT;
268 static LIST_HEAD(vmap_area_list);
270 static struct vmap_area *__find_vmap_area(unsigned long addr)
272 struct rb_node *n = vmap_area_root.rb_node;
275 struct vmap_area *va;
277 va = rb_entry(n, struct vmap_area, rb_node);
278 if (addr < va->va_start)
280 else if (addr > va->va_start)
289 static void __insert_vmap_area(struct vmap_area *va)
291 struct rb_node **p = &vmap_area_root.rb_node;
292 struct rb_node *parent = NULL;
296 struct vmap_area *tmp;
299 tmp = rb_entry(parent, struct vmap_area, rb_node);
300 if (va->va_start < tmp->va_end)
302 else if (va->va_end > tmp->va_start)
308 rb_link_node(&va->rb_node, parent, p);
309 rb_insert_color(&va->rb_node, &vmap_area_root);
311 /* address-sort this list so it is usable like the vmlist */
312 tmp = rb_prev(&va->rb_node);
314 struct vmap_area *prev;
315 prev = rb_entry(tmp, struct vmap_area, rb_node);
316 list_add_rcu(&va->list, &prev->list);
318 list_add_rcu(&va->list, &vmap_area_list);
321 static void purge_vmap_area_lazy(void);
324 * Allocate a region of KVA of the specified size and alignment, within the
327 static struct vmap_area *alloc_vmap_area(unsigned long size,
329 unsigned long vstart, unsigned long vend,
330 int node, gfp_t gfp_mask)
332 struct vmap_area *va;
338 BUG_ON(size & ~PAGE_MASK);
340 va = kmalloc_node(sizeof(struct vmap_area),
341 gfp_mask & GFP_RECLAIM_MASK, node);
343 return ERR_PTR(-ENOMEM);
346 addr = ALIGN(vstart, align);
348 spin_lock(&vmap_area_lock);
349 if (addr + size - 1 < addr)
352 /* XXX: could have a last_hole cache */
353 n = vmap_area_root.rb_node;
355 struct vmap_area *first = NULL;
358 struct vmap_area *tmp;
359 tmp = rb_entry(n, struct vmap_area, rb_node);
360 if (tmp->va_end >= addr) {
361 if (!first && tmp->va_start < addr + size)
373 if (first->va_end < addr) {
374 n = rb_next(&first->rb_node);
376 first = rb_entry(n, struct vmap_area, rb_node);
381 while (addr + size > first->va_start && addr + size <= vend) {
382 addr = ALIGN(first->va_end + PAGE_SIZE, align);
383 if (addr + size - 1 < addr)
386 n = rb_next(&first->rb_node);
388 first = rb_entry(n, struct vmap_area, rb_node);
394 if (addr + size > vend) {
396 spin_unlock(&vmap_area_lock);
398 purge_vmap_area_lazy();
402 if (printk_ratelimit())
404 "vmap allocation for size %lu failed: "
405 "use vmalloc=<size> to increase size.\n", size);
407 return ERR_PTR(-EBUSY);
410 BUG_ON(addr & (align-1));
413 va->va_end = addr + size;
415 __insert_vmap_area(va);
416 spin_unlock(&vmap_area_lock);
421 static void rcu_free_va(struct rcu_head *head)
423 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
428 static void __free_vmap_area(struct vmap_area *va)
430 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
431 rb_erase(&va->rb_node, &vmap_area_root);
432 RB_CLEAR_NODE(&va->rb_node);
433 list_del_rcu(&va->list);
435 call_rcu(&va->rcu_head, rcu_free_va);
439 * Free a region of KVA allocated by alloc_vmap_area
441 static void free_vmap_area(struct vmap_area *va)
443 spin_lock(&vmap_area_lock);
444 __free_vmap_area(va);
445 spin_unlock(&vmap_area_lock);
449 * Clear the pagetable entries of a given vmap_area
451 static void unmap_vmap_area(struct vmap_area *va)
453 vunmap_page_range(va->va_start, va->va_end);
456 static void vmap_debug_free_range(unsigned long start, unsigned long end)
459 * Unmap page tables and force a TLB flush immediately if
460 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
461 * bugs similarly to those in linear kernel virtual address
462 * space after a page has been freed.
464 * All the lazy freeing logic is still retained, in order to
465 * minimise intrusiveness of this debugging feature.
467 * This is going to be *slow* (linear kernel virtual address
468 * debugging doesn't do a broadcast TLB flush so it is a lot
471 #ifdef CONFIG_DEBUG_PAGEALLOC
472 vunmap_page_range(start, end);
473 flush_tlb_kernel_range(start, end);
478 * lazy_max_pages is the maximum amount of virtual address space we gather up
479 * before attempting to purge with a TLB flush.
481 * There is a tradeoff here: a larger number will cover more kernel page tables
482 * and take slightly longer to purge, but it will linearly reduce the number of
483 * global TLB flushes that must be performed. It would seem natural to scale
484 * this number up linearly with the number of CPUs (because vmapping activity
485 * could also scale linearly with the number of CPUs), however it is likely
486 * that in practice, workloads might be constrained in other ways that mean
487 * vmap activity will not scale linearly with CPUs. Also, I want to be
488 * conservative and not introduce a big latency on huge systems, so go with
489 * a less aggressive log scale. It will still be an improvement over the old
490 * code, and it will be simple to change the scale factor if we find that it
491 * becomes a problem on bigger systems.
493 static unsigned long lazy_max_pages(void)
497 log = fls(num_online_cpus());
499 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
502 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
505 * Purges all lazily-freed vmap areas.
507 * If sync is 0 then don't purge if there is already a purge in progress.
508 * If force_flush is 1, then flush kernel TLBs between *start and *end even
509 * if we found no lazy vmap areas to unmap (callers can use this to optimise
510 * their own TLB flushing).
511 * Returns with *start = min(*start, lowest purged address)
512 * *end = max(*end, highest purged address)
514 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
515 int sync, int force_flush)
517 static DEFINE_SPINLOCK(purge_lock);
519 struct vmap_area *va;
520 struct vmap_area *n_va;
524 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
525 * should not expect such behaviour. This just simplifies locking for
526 * the case that isn't actually used at the moment anyway.
528 if (!sync && !force_flush) {
529 if (!spin_trylock(&purge_lock))
532 spin_lock(&purge_lock);
535 list_for_each_entry_rcu(va, &vmap_area_list, list) {
536 if (va->flags & VM_LAZY_FREE) {
537 if (va->va_start < *start)
538 *start = va->va_start;
539 if (va->va_end > *end)
541 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
543 list_add_tail(&va->purge_list, &valist);
544 va->flags |= VM_LAZY_FREEING;
545 va->flags &= ~VM_LAZY_FREE;
551 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
552 atomic_sub(nr, &vmap_lazy_nr);
555 if (nr || force_flush)
556 flush_tlb_kernel_range(*start, *end);
559 spin_lock(&vmap_area_lock);
560 list_for_each_entry_safe(va, n_va, &valist, purge_list)
561 __free_vmap_area(va);
562 spin_unlock(&vmap_area_lock);
564 spin_unlock(&purge_lock);
568 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
569 * is already purging.
571 static void try_purge_vmap_area_lazy(void)
573 unsigned long start = ULONG_MAX, end = 0;
575 __purge_vmap_area_lazy(&start, &end, 0, 0);
579 * Kick off a purge of the outstanding lazy areas.
581 static void purge_vmap_area_lazy(void)
583 unsigned long start = ULONG_MAX, end = 0;
585 __purge_vmap_area_lazy(&start, &end, 1, 0);
589 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
590 * called for the correct range previously.
592 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
594 va->flags |= VM_LAZY_FREE;
595 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
596 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
597 try_purge_vmap_area_lazy();
601 * Free and unmap a vmap area
603 static void free_unmap_vmap_area(struct vmap_area *va)
605 flush_cache_vunmap(va->va_start, va->va_end);
606 free_unmap_vmap_area_noflush(va);
609 static struct vmap_area *find_vmap_area(unsigned long addr)
611 struct vmap_area *va;
613 spin_lock(&vmap_area_lock);
614 va = __find_vmap_area(addr);
615 spin_unlock(&vmap_area_lock);
620 static void free_unmap_vmap_area_addr(unsigned long addr)
622 struct vmap_area *va;
624 va = find_vmap_area(addr);
626 free_unmap_vmap_area(va);
630 /*** Per cpu kva allocator ***/
633 * vmap space is limited especially on 32 bit architectures. Ensure there is
634 * room for at least 16 percpu vmap blocks per CPU.
637 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
638 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
639 * instead (we just need a rough idea)
641 #if BITS_PER_LONG == 32
642 #define VMALLOC_SPACE (128UL*1024*1024)
644 #define VMALLOC_SPACE (128UL*1024*1024*1024)
647 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
648 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
649 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
650 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
651 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
652 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
653 #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
654 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
655 VMALLOC_PAGES / NR_CPUS / 16))
657 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
659 static bool vmap_initialized __read_mostly = false;
661 struct vmap_block_queue {
663 struct list_head free;
664 struct list_head dirty;
665 unsigned int nr_dirty;
670 struct vmap_area *va;
671 struct vmap_block_queue *vbq;
672 unsigned long free, dirty;
673 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
674 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
676 struct list_head free_list;
677 struct rcu_head rcu_head;
681 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
682 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
685 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
686 * in the free path. Could get rid of this if we change the API to return a
687 * "cookie" from alloc, to be passed to free. But no big deal yet.
689 static DEFINE_SPINLOCK(vmap_block_tree_lock);
690 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
693 * We should probably have a fallback mechanism to allocate virtual memory
694 * out of partially filled vmap blocks. However vmap block sizing should be
695 * fairly reasonable according to the vmalloc size, so it shouldn't be a
699 static unsigned long addr_to_vb_idx(unsigned long addr)
701 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
702 addr /= VMAP_BLOCK_SIZE;
706 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
708 struct vmap_block_queue *vbq;
709 struct vmap_block *vb;
710 struct vmap_area *va;
711 unsigned long vb_idx;
714 node = numa_node_id();
716 vb = kmalloc_node(sizeof(struct vmap_block),
717 gfp_mask & GFP_RECLAIM_MASK, node);
719 return ERR_PTR(-ENOMEM);
721 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
722 VMALLOC_START, VMALLOC_END,
724 if (unlikely(IS_ERR(va))) {
726 return ERR_PTR(PTR_ERR(va));
729 err = radix_tree_preload(gfp_mask);
736 spin_lock_init(&vb->lock);
738 vb->free = VMAP_BBMAP_BITS;
740 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
741 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
742 INIT_LIST_HEAD(&vb->free_list);
744 vb_idx = addr_to_vb_idx(va->va_start);
745 spin_lock(&vmap_block_tree_lock);
746 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
747 spin_unlock(&vmap_block_tree_lock);
749 radix_tree_preload_end();
751 vbq = &get_cpu_var(vmap_block_queue);
753 spin_lock(&vbq->lock);
754 list_add(&vb->free_list, &vbq->free);
755 spin_unlock(&vbq->lock);
756 put_cpu_var(vmap_cpu_blocks);
761 static void rcu_free_vb(struct rcu_head *head)
763 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
768 static void free_vmap_block(struct vmap_block *vb)
770 struct vmap_block *tmp;
771 unsigned long vb_idx;
773 BUG_ON(!list_empty(&vb->free_list));
775 vb_idx = addr_to_vb_idx(vb->va->va_start);
776 spin_lock(&vmap_block_tree_lock);
777 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
778 spin_unlock(&vmap_block_tree_lock);
781 free_unmap_vmap_area_noflush(vb->va);
782 call_rcu(&vb->rcu_head, rcu_free_vb);
785 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
787 struct vmap_block_queue *vbq;
788 struct vmap_block *vb;
789 unsigned long addr = 0;
792 BUG_ON(size & ~PAGE_MASK);
793 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
794 order = get_order(size);
798 vbq = &get_cpu_var(vmap_block_queue);
799 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
802 spin_lock(&vb->lock);
803 i = bitmap_find_free_region(vb->alloc_map,
804 VMAP_BBMAP_BITS, order);
807 addr = vb->va->va_start + (i << PAGE_SHIFT);
808 BUG_ON(addr_to_vb_idx(addr) !=
809 addr_to_vb_idx(vb->va->va_start));
810 vb->free -= 1UL << order;
812 spin_lock(&vbq->lock);
813 list_del_init(&vb->free_list);
814 spin_unlock(&vbq->lock);
816 spin_unlock(&vb->lock);
819 spin_unlock(&vb->lock);
821 put_cpu_var(vmap_cpu_blocks);
825 vb = new_vmap_block(gfp_mask);
834 static void vb_free(const void *addr, unsigned long size)
836 unsigned long offset;
837 unsigned long vb_idx;
839 struct vmap_block *vb;
841 BUG_ON(size & ~PAGE_MASK);
842 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
844 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
846 order = get_order(size);
848 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
850 vb_idx = addr_to_vb_idx((unsigned long)addr);
852 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
856 spin_lock(&vb->lock);
857 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
859 vb->dirty += 1UL << order;
860 if (vb->dirty == VMAP_BBMAP_BITS) {
861 BUG_ON(vb->free || !list_empty(&vb->free_list));
862 spin_unlock(&vb->lock);
865 spin_unlock(&vb->lock);
869 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
871 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
872 * to amortize TLB flushing overheads. What this means is that any page you
873 * have now, may, in a former life, have been mapped into kernel virtual
874 * address by the vmap layer and so there might be some CPUs with TLB entries
875 * still referencing that page (additional to the regular 1:1 kernel mapping).
877 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
878 * be sure that none of the pages we have control over will have any aliases
879 * from the vmap layer.
881 void vm_unmap_aliases(void)
883 unsigned long start = ULONG_MAX, end = 0;
887 if (unlikely(!vmap_initialized))
890 for_each_possible_cpu(cpu) {
891 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
892 struct vmap_block *vb;
895 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
898 spin_lock(&vb->lock);
899 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
900 while (i < VMAP_BBMAP_BITS) {
903 j = find_next_zero_bit(vb->dirty_map,
906 s = vb->va->va_start + (i << PAGE_SHIFT);
907 e = vb->va->va_start + (j << PAGE_SHIFT);
908 vunmap_page_range(s, e);
917 i = find_next_bit(vb->dirty_map,
920 spin_unlock(&vb->lock);
925 __purge_vmap_area_lazy(&start, &end, 1, flush);
927 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
930 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
931 * @mem: the pointer returned by vm_map_ram
932 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
934 void vm_unmap_ram(const void *mem, unsigned int count)
936 unsigned long size = count << PAGE_SHIFT;
937 unsigned long addr = (unsigned long)mem;
940 BUG_ON(addr < VMALLOC_START);
941 BUG_ON(addr > VMALLOC_END);
942 BUG_ON(addr & (PAGE_SIZE-1));
944 debug_check_no_locks_freed(mem, size);
945 vmap_debug_free_range(addr, addr+size);
947 if (likely(count <= VMAP_MAX_ALLOC))
950 free_unmap_vmap_area_addr(addr);
952 EXPORT_SYMBOL(vm_unmap_ram);
955 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
956 * @pages: an array of pointers to the pages to be mapped
957 * @count: number of pages
958 * @node: prefer to allocate data structures on this node
959 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
961 * Returns: a pointer to the address that has been mapped, or %NULL on failure
963 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
965 unsigned long size = count << PAGE_SHIFT;
969 if (likely(count <= VMAP_MAX_ALLOC)) {
970 mem = vb_alloc(size, GFP_KERNEL);
973 addr = (unsigned long)mem;
975 struct vmap_area *va;
976 va = alloc_vmap_area(size, PAGE_SIZE,
977 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
984 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
985 vm_unmap_ram(mem, count);
990 EXPORT_SYMBOL(vm_map_ram);
993 * vm_area_register_early - register vmap area early during boot
994 * @vm: vm_struct to register
995 * @align: requested alignment
997 * This function is used to register kernel vm area before
998 * vmalloc_init() is called. @vm->size and @vm->flags should contain
999 * proper values on entry and other fields should be zero. On return,
1000 * vm->addr contains the allocated address.
1002 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1004 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1006 static size_t vm_init_off __initdata;
1009 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1010 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1012 vm->addr = (void *)addr;
1018 void __init vmalloc_init(void)
1020 struct vmap_area *va;
1021 struct vm_struct *tmp;
1024 for_each_possible_cpu(i) {
1025 struct vmap_block_queue *vbq;
1027 vbq = &per_cpu(vmap_block_queue, i);
1028 spin_lock_init(&vbq->lock);
1029 INIT_LIST_HEAD(&vbq->free);
1030 INIT_LIST_HEAD(&vbq->dirty);
1034 /* Import existing vmlist entries. */
1035 for (tmp = vmlist; tmp; tmp = tmp->next) {
1036 va = alloc_bootmem(sizeof(struct vmap_area));
1037 va->flags = tmp->flags | VM_VM_AREA;
1038 va->va_start = (unsigned long)tmp->addr;
1039 va->va_end = va->va_start + tmp->size;
1040 __insert_vmap_area(va);
1042 vmap_initialized = true;
1046 * map_kernel_range_noflush - map kernel VM area with the specified pages
1047 * @addr: start of the VM area to map
1048 * @size: size of the VM area to map
1049 * @prot: page protection flags to use
1050 * @pages: pages to map
1052 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1053 * specify should have been allocated using get_vm_area() and its
1057 * This function does NOT do any cache flushing. The caller is
1058 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1059 * before calling this function.
1062 * The number of pages mapped on success, -errno on failure.
1064 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1065 pgprot_t prot, struct page **pages)
1067 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1071 * unmap_kernel_range_noflush - unmap kernel VM area
1072 * @addr: start of the VM area to unmap
1073 * @size: size of the VM area to unmap
1075 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1076 * specify should have been allocated using get_vm_area() and its
1080 * This function does NOT do any cache flushing. The caller is
1081 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1082 * before calling this function and flush_tlb_kernel_range() after.
1084 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1086 vunmap_page_range(addr, addr + size);
1090 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1091 * @addr: start of the VM area to unmap
1092 * @size: size of the VM area to unmap
1094 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1095 * the unmapping and tlb after.
1097 void unmap_kernel_range(unsigned long addr, unsigned long size)
1099 unsigned long end = addr + size;
1101 flush_cache_vunmap(addr, end);
1102 vunmap_page_range(addr, end);
1103 flush_tlb_kernel_range(addr, end);
1106 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1108 unsigned long addr = (unsigned long)area->addr;
1109 unsigned long end = addr + area->size - PAGE_SIZE;
1112 err = vmap_page_range(addr, end, prot, *pages);
1120 EXPORT_SYMBOL_GPL(map_vm_area);
1122 /*** Old vmalloc interfaces ***/
1123 DEFINE_RWLOCK(vmlist_lock);
1124 struct vm_struct *vmlist;
1126 static struct vm_struct *__get_vm_area_node(unsigned long size,
1127 unsigned long flags, unsigned long start, unsigned long end,
1128 int node, gfp_t gfp_mask, void *caller)
1130 static struct vmap_area *va;
1131 struct vm_struct *area;
1132 struct vm_struct *tmp, **p;
1133 unsigned long align = 1;
1135 BUG_ON(in_interrupt());
1136 if (flags & VM_IOREMAP) {
1137 int bit = fls(size);
1139 if (bit > IOREMAP_MAX_ORDER)
1140 bit = IOREMAP_MAX_ORDER;
1141 else if (bit < PAGE_SHIFT)
1147 size = PAGE_ALIGN(size);
1148 if (unlikely(!size))
1151 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1152 if (unlikely(!area))
1156 * We always allocate a guard page.
1160 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1166 area->flags = flags;
1167 area->addr = (void *)va->va_start;
1171 area->phys_addr = 0;
1172 area->caller = caller;
1174 va->flags |= VM_VM_AREA;
1176 write_lock(&vmlist_lock);
1177 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1178 if (tmp->addr >= area->addr)
1183 write_unlock(&vmlist_lock);
1188 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1189 unsigned long start, unsigned long end)
1191 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1192 __builtin_return_address(0));
1194 EXPORT_SYMBOL_GPL(__get_vm_area);
1196 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1197 unsigned long start, unsigned long end,
1200 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1205 * get_vm_area - reserve a contiguous kernel virtual area
1206 * @size: size of the area
1207 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1209 * Search an area of @size in the kernel virtual mapping area,
1210 * and reserved it for out purposes. Returns the area descriptor
1211 * on success or %NULL on failure.
1213 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1215 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1216 -1, GFP_KERNEL, __builtin_return_address(0));
1219 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1222 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1223 -1, GFP_KERNEL, caller);
1226 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1227 int node, gfp_t gfp_mask)
1229 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
1230 gfp_mask, __builtin_return_address(0));
1233 static struct vm_struct *find_vm_area(const void *addr)
1235 struct vmap_area *va;
1237 va = find_vmap_area((unsigned long)addr);
1238 if (va && va->flags & VM_VM_AREA)
1245 * remove_vm_area - find and remove a continuous kernel virtual area
1246 * @addr: base address
1248 * Search for the kernel VM area starting at @addr, and remove it.
1249 * This function returns the found VM area, but using it is NOT safe
1250 * on SMP machines, except for its size or flags.
1252 struct vm_struct *remove_vm_area(const void *addr)
1254 struct vmap_area *va;
1256 va = find_vmap_area((unsigned long)addr);
1257 if (va && va->flags & VM_VM_AREA) {
1258 struct vm_struct *vm = va->private;
1259 struct vm_struct *tmp, **p;
1261 vmap_debug_free_range(va->va_start, va->va_end);
1262 free_unmap_vmap_area(va);
1263 vm->size -= PAGE_SIZE;
1265 write_lock(&vmlist_lock);
1266 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1269 write_unlock(&vmlist_lock);
1276 static void __vunmap(const void *addr, int deallocate_pages)
1278 struct vm_struct *area;
1283 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1284 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1288 area = remove_vm_area(addr);
1289 if (unlikely(!area)) {
1290 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1295 debug_check_no_locks_freed(addr, area->size);
1296 debug_check_no_obj_freed(addr, area->size);
1298 if (deallocate_pages) {
1301 for (i = 0; i < area->nr_pages; i++) {
1302 struct page *page = area->pages[i];
1308 if (area->flags & VM_VPAGES)
1319 * vfree - release memory allocated by vmalloc()
1320 * @addr: memory base address
1322 * Free the virtually continuous memory area starting at @addr, as
1323 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1324 * NULL, no operation is performed.
1326 * Must not be called in interrupt context.
1328 void vfree(const void *addr)
1330 BUG_ON(in_interrupt());
1332 kmemleak_free(addr);
1336 EXPORT_SYMBOL(vfree);
1339 * vunmap - release virtual mapping obtained by vmap()
1340 * @addr: memory base address
1342 * Free the virtually contiguous memory area starting at @addr,
1343 * which was created from the page array passed to vmap().
1345 * Must not be called in interrupt context.
1347 void vunmap(const void *addr)
1349 BUG_ON(in_interrupt());
1353 EXPORT_SYMBOL(vunmap);
1356 * vmap - map an array of pages into virtually contiguous space
1357 * @pages: array of page pointers
1358 * @count: number of pages to map
1359 * @flags: vm_area->flags
1360 * @prot: page protection for the mapping
1362 * Maps @count pages from @pages into contiguous kernel virtual
1365 void *vmap(struct page **pages, unsigned int count,
1366 unsigned long flags, pgprot_t prot)
1368 struct vm_struct *area;
1372 if (count > num_physpages)
1375 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1376 __builtin_return_address(0));
1380 if (map_vm_area(area, prot, &pages)) {
1387 EXPORT_SYMBOL(vmap);
1389 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1390 int node, void *caller);
1391 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1392 pgprot_t prot, int node, void *caller)
1394 struct page **pages;
1395 unsigned int nr_pages, array_size, i;
1397 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1398 array_size = (nr_pages * sizeof(struct page *));
1400 area->nr_pages = nr_pages;
1401 /* Please note that the recursion is strictly bounded. */
1402 if (array_size > PAGE_SIZE) {
1403 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
1404 PAGE_KERNEL, node, caller);
1405 area->flags |= VM_VPAGES;
1407 pages = kmalloc_node(array_size,
1408 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
1411 area->pages = pages;
1412 area->caller = caller;
1414 remove_vm_area(area->addr);
1419 for (i = 0; i < area->nr_pages; i++) {
1423 page = alloc_page(gfp_mask);
1425 page = alloc_pages_node(node, gfp_mask, 0);
1427 if (unlikely(!page)) {
1428 /* Successfully allocated i pages, free them in __vunmap() */
1432 area->pages[i] = page;
1435 if (map_vm_area(area, prot, &pages))
1444 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1446 void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1,
1447 __builtin_return_address(0));
1450 * A ref_count = 3 is needed because the vm_struct and vmap_area
1451 * structures allocated in the __get_vm_area_node() function contain
1452 * references to the virtual address of the vmalloc'ed block.
1454 kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask);
1460 * __vmalloc_node - allocate virtually contiguous memory
1461 * @size: allocation size
1462 * @gfp_mask: flags for the page level allocator
1463 * @prot: protection mask for the allocated pages
1464 * @node: node to use for allocation or -1
1465 * @caller: caller's return address
1467 * Allocate enough pages to cover @size from the page level
1468 * allocator with @gfp_mask flags. Map them into contiguous
1469 * kernel virtual space, using a pagetable protection of @prot.
1471 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1472 int node, void *caller)
1474 struct vm_struct *area;
1476 unsigned long real_size = size;
1478 size = PAGE_ALIGN(size);
1479 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1482 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1483 node, gfp_mask, caller);
1488 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1491 * A ref_count = 3 is needed because the vm_struct and vmap_area
1492 * structures allocated in the __get_vm_area_node() function contain
1493 * references to the virtual address of the vmalloc'ed block.
1495 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1500 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1502 return __vmalloc_node(size, gfp_mask, prot, -1,
1503 __builtin_return_address(0));
1505 EXPORT_SYMBOL(__vmalloc);
1508 * vmalloc - allocate virtually contiguous memory
1509 * @size: allocation size
1510 * Allocate enough pages to cover @size from the page level
1511 * allocator and map them into contiguous kernel virtual space.
1513 * For tight control over page level allocator and protection flags
1514 * use __vmalloc() instead.
1516 void *vmalloc(unsigned long size)
1518 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1519 -1, __builtin_return_address(0));
1521 EXPORT_SYMBOL(vmalloc);
1524 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1525 * @size: allocation size
1527 * The resulting memory area is zeroed so it can be mapped to userspace
1528 * without leaking data.
1530 void *vmalloc_user(unsigned long size)
1532 struct vm_struct *area;
1535 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1536 PAGE_KERNEL, -1, __builtin_return_address(0));
1538 area = find_vm_area(ret);
1539 area->flags |= VM_USERMAP;
1543 EXPORT_SYMBOL(vmalloc_user);
1546 * vmalloc_node - allocate memory on a specific node
1547 * @size: allocation size
1550 * Allocate enough pages to cover @size from the page level
1551 * allocator and map them into contiguous kernel virtual space.
1553 * For tight control over page level allocator and protection flags
1554 * use __vmalloc() instead.
1556 void *vmalloc_node(unsigned long size, int node)
1558 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1559 node, __builtin_return_address(0));
1561 EXPORT_SYMBOL(vmalloc_node);
1563 #ifndef PAGE_KERNEL_EXEC
1564 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1568 * vmalloc_exec - allocate virtually contiguous, executable memory
1569 * @size: allocation size
1571 * Kernel-internal function to allocate enough pages to cover @size
1572 * the page level allocator and map them into contiguous and
1573 * executable kernel virtual space.
1575 * For tight control over page level allocator and protection flags
1576 * use __vmalloc() instead.
1579 void *vmalloc_exec(unsigned long size)
1581 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1582 -1, __builtin_return_address(0));
1585 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1586 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1587 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1588 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1590 #define GFP_VMALLOC32 GFP_KERNEL
1594 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1595 * @size: allocation size
1597 * Allocate enough 32bit PA addressable pages to cover @size from the
1598 * page level allocator and map them into contiguous kernel virtual space.
1600 void *vmalloc_32(unsigned long size)
1602 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1603 -1, __builtin_return_address(0));
1605 EXPORT_SYMBOL(vmalloc_32);
1608 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1609 * @size: allocation size
1611 * The resulting memory area is 32bit addressable and zeroed so it can be
1612 * mapped to userspace without leaking data.
1614 void *vmalloc_32_user(unsigned long size)
1616 struct vm_struct *area;
1619 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1620 -1, __builtin_return_address(0));
1622 area = find_vm_area(ret);
1623 area->flags |= VM_USERMAP;
1627 EXPORT_SYMBOL(vmalloc_32_user);
1629 long vread(char *buf, char *addr, unsigned long count)
1631 struct vm_struct *tmp;
1632 char *vaddr, *buf_start = buf;
1635 /* Don't allow overflow */
1636 if ((unsigned long) addr + count < count)
1637 count = -(unsigned long) addr;
1639 read_lock(&vmlist_lock);
1640 for (tmp = vmlist; tmp; tmp = tmp->next) {
1641 vaddr = (char *) tmp->addr;
1642 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1644 while (addr < vaddr) {
1652 n = vaddr + tmp->size - PAGE_SIZE - addr;
1663 read_unlock(&vmlist_lock);
1664 return buf - buf_start;
1667 long vwrite(char *buf, char *addr, unsigned long count)
1669 struct vm_struct *tmp;
1670 char *vaddr, *buf_start = buf;
1673 /* Don't allow overflow */
1674 if ((unsigned long) addr + count < count)
1675 count = -(unsigned long) addr;
1677 read_lock(&vmlist_lock);
1678 for (tmp = vmlist; tmp; tmp = tmp->next) {
1679 vaddr = (char *) tmp->addr;
1680 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1682 while (addr < vaddr) {
1689 n = vaddr + tmp->size - PAGE_SIZE - addr;
1700 read_unlock(&vmlist_lock);
1701 return buf - buf_start;
1705 * remap_vmalloc_range - map vmalloc pages to userspace
1706 * @vma: vma to cover (map full range of vma)
1707 * @addr: vmalloc memory
1708 * @pgoff: number of pages into addr before first page to map
1710 * Returns: 0 for success, -Exxx on failure
1712 * This function checks that addr is a valid vmalloc'ed area, and
1713 * that it is big enough to cover the vma. Will return failure if
1714 * that criteria isn't met.
1716 * Similar to remap_pfn_range() (see mm/memory.c)
1718 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1719 unsigned long pgoff)
1721 struct vm_struct *area;
1722 unsigned long uaddr = vma->vm_start;
1723 unsigned long usize = vma->vm_end - vma->vm_start;
1725 if ((PAGE_SIZE-1) & (unsigned long)addr)
1728 area = find_vm_area(addr);
1732 if (!(area->flags & VM_USERMAP))
1735 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
1738 addr += pgoff << PAGE_SHIFT;
1740 struct page *page = vmalloc_to_page(addr);
1743 ret = vm_insert_page(vma, uaddr, page);
1750 } while (usize > 0);
1752 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1753 vma->vm_flags |= VM_RESERVED;
1757 EXPORT_SYMBOL(remap_vmalloc_range);
1760 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1763 void __attribute__((weak)) vmalloc_sync_all(void)
1768 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
1770 /* apply_to_page_range() does all the hard work. */
1775 * alloc_vm_area - allocate a range of kernel address space
1776 * @size: size of the area
1778 * Returns: NULL on failure, vm_struct on success
1780 * This function reserves a range of kernel address space, and
1781 * allocates pagetables to map that range. No actual mappings
1782 * are created. If the kernel address space is not shared
1783 * between processes, it syncs the pagetable across all
1786 struct vm_struct *alloc_vm_area(size_t size)
1788 struct vm_struct *area;
1790 area = get_vm_area_caller(size, VM_IOREMAP,
1791 __builtin_return_address(0));
1796 * This ensures that page tables are constructed for this region
1797 * of kernel virtual address space and mapped into init_mm.
1799 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1800 area->size, f, NULL)) {
1805 /* Make sure the pagetables are constructed in process kernel
1811 EXPORT_SYMBOL_GPL(alloc_vm_area);
1813 void free_vm_area(struct vm_struct *area)
1815 struct vm_struct *ret;
1816 ret = remove_vm_area(area->addr);
1817 BUG_ON(ret != area);
1820 EXPORT_SYMBOL_GPL(free_vm_area);
1823 #ifdef CONFIG_PROC_FS
1824 static void *s_start(struct seq_file *m, loff_t *pos)
1827 struct vm_struct *v;
1829 read_lock(&vmlist_lock);
1831 while (n > 0 && v) {
1842 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1844 struct vm_struct *v = p;
1850 static void s_stop(struct seq_file *m, void *p)
1852 read_unlock(&vmlist_lock);
1855 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1858 unsigned int nr, *counters = m->private;
1863 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1865 for (nr = 0; nr < v->nr_pages; nr++)
1866 counters[page_to_nid(v->pages[nr])]++;
1868 for_each_node_state(nr, N_HIGH_MEMORY)
1870 seq_printf(m, " N%u=%u", nr, counters[nr]);
1874 static int s_show(struct seq_file *m, void *p)
1876 struct vm_struct *v = p;
1878 seq_printf(m, "0x%p-0x%p %7ld",
1879 v->addr, v->addr + v->size, v->size);
1882 char buff[KSYM_SYMBOL_LEN];
1885 sprint_symbol(buff, (unsigned long)v->caller);
1890 seq_printf(m, " pages=%d", v->nr_pages);
1893 seq_printf(m, " phys=%lx", v->phys_addr);
1895 if (v->flags & VM_IOREMAP)
1896 seq_printf(m, " ioremap");
1898 if (v->flags & VM_ALLOC)
1899 seq_printf(m, " vmalloc");
1901 if (v->flags & VM_MAP)
1902 seq_printf(m, " vmap");
1904 if (v->flags & VM_USERMAP)
1905 seq_printf(m, " user");
1907 if (v->flags & VM_VPAGES)
1908 seq_printf(m, " vpages");
1910 show_numa_info(m, v);
1915 static const struct seq_operations vmalloc_op = {
1922 static int vmalloc_open(struct inode *inode, struct file *file)
1924 unsigned int *ptr = NULL;
1928 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1929 ret = seq_open(file, &vmalloc_op);
1931 struct seq_file *m = file->private_data;
1938 static const struct file_operations proc_vmalloc_operations = {
1939 .open = vmalloc_open,
1941 .llseek = seq_lseek,
1942 .release = seq_release_private,
1945 static int __init proc_vmalloc_init(void)
1947 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1950 module_init(proc_vmalloc_init);