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>
29 #include <asm/atomic.h>
30 #include <asm/uaccess.h>
31 #include <asm/tlbflush.h>
34 /*** Page table manipulation functions ***/
36 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
40 pte = pte_offset_kernel(pmd, addr);
42 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
43 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
44 } while (pte++, addr += PAGE_SIZE, addr != end);
47 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
52 pmd = pmd_offset(pud, addr);
54 next = pmd_addr_end(addr, end);
55 if (pmd_none_or_clear_bad(pmd))
57 vunmap_pte_range(pmd, addr, next);
58 } while (pmd++, addr = next, addr != end);
61 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
66 pud = pud_offset(pgd, addr);
68 next = pud_addr_end(addr, end);
69 if (pud_none_or_clear_bad(pud))
71 vunmap_pmd_range(pud, addr, next);
72 } while (pud++, addr = next, addr != end);
75 static void vunmap_page_range(unsigned long addr, unsigned long end)
81 pgd = pgd_offset_k(addr);
83 next = pgd_addr_end(addr, end);
84 if (pgd_none_or_clear_bad(pgd))
86 vunmap_pud_range(pgd, addr, next);
87 } while (pgd++, addr = next, addr != end);
90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
91 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
100 pte = pte_alloc_kernel(pmd, addr);
104 struct page *page = pages[*nr];
106 if (WARN_ON(!pte_none(*pte)))
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
112 } while (pte++, addr += PAGE_SIZE, addr != end);
116 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
122 pmd = pmd_alloc(&init_mm, pud, addr);
126 next = pmd_addr_end(addr, end);
127 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
129 } while (pmd++, addr = next, addr != end);
133 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
139 pud = pud_alloc(&init_mm, pgd, addr);
143 next = pud_addr_end(addr, end);
144 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
146 } while (pud++, addr = next, addr != end);
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
156 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157 pgprot_t prot, struct page **pages)
161 unsigned long addr = start;
166 pgd = pgd_offset_k(addr);
168 next = pgd_addr_end(addr, end);
169 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
172 } while (pgd++, addr = next, addr != end);
179 static int vmap_page_range(unsigned long start, unsigned long end,
180 pgprot_t prot, struct page **pages)
184 ret = vmap_page_range_noflush(start, end, prot, pages);
185 flush_cache_vmap(start, end);
189 static inline int is_vmalloc_or_module_addr(const void *x)
192 * ARM, x86-64 and sparc64 put modules in a special place,
193 * and fall back on vmalloc() if that fails. Others
194 * just put it in the vmalloc space.
196 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
197 unsigned long addr = (unsigned long)x;
198 if (addr >= MODULES_VADDR && addr < MODULES_END)
201 return is_vmalloc_addr(x);
205 * Walk a vmap address to the struct page it maps.
207 struct page *vmalloc_to_page(const void *vmalloc_addr)
209 unsigned long addr = (unsigned long) vmalloc_addr;
210 struct page *page = NULL;
211 pgd_t *pgd = pgd_offset_k(addr);
214 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
215 * architectures that do not vmalloc module space
217 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
219 if (!pgd_none(*pgd)) {
220 pud_t *pud = pud_offset(pgd, addr);
221 if (!pud_none(*pud)) {
222 pmd_t *pmd = pmd_offset(pud, addr);
223 if (!pmd_none(*pmd)) {
226 ptep = pte_offset_map(pmd, addr);
228 if (pte_present(pte))
229 page = pte_page(pte);
236 EXPORT_SYMBOL(vmalloc_to_page);
239 * Map a vmalloc()-space virtual address to the physical page frame number.
241 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
243 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
245 EXPORT_SYMBOL(vmalloc_to_pfn);
248 /*** Global kva allocator ***/
250 #define VM_LAZY_FREE 0x01
251 #define VM_LAZY_FREEING 0x02
252 #define VM_VM_AREA 0x04
255 unsigned long va_start;
256 unsigned long va_end;
258 struct rb_node rb_node; /* address sorted rbtree */
259 struct list_head list; /* address sorted list */
260 struct list_head purge_list; /* "lazy purge" list */
262 struct rcu_head rcu_head;
265 static DEFINE_SPINLOCK(vmap_area_lock);
266 static struct rb_root vmap_area_root = RB_ROOT;
267 static LIST_HEAD(vmap_area_list);
269 static struct vmap_area *__find_vmap_area(unsigned long addr)
271 struct rb_node *n = vmap_area_root.rb_node;
274 struct vmap_area *va;
276 va = rb_entry(n, struct vmap_area, rb_node);
277 if (addr < va->va_start)
279 else if (addr > va->va_start)
288 static void __insert_vmap_area(struct vmap_area *va)
290 struct rb_node **p = &vmap_area_root.rb_node;
291 struct rb_node *parent = NULL;
295 struct vmap_area *tmp;
298 tmp = rb_entry(parent, struct vmap_area, rb_node);
299 if (va->va_start < tmp->va_end)
301 else if (va->va_end > tmp->va_start)
307 rb_link_node(&va->rb_node, parent, p);
308 rb_insert_color(&va->rb_node, &vmap_area_root);
310 /* address-sort this list so it is usable like the vmlist */
311 tmp = rb_prev(&va->rb_node);
313 struct vmap_area *prev;
314 prev = rb_entry(tmp, struct vmap_area, rb_node);
315 list_add_rcu(&va->list, &prev->list);
317 list_add_rcu(&va->list, &vmap_area_list);
320 static void purge_vmap_area_lazy(void);
323 * Allocate a region of KVA of the specified size and alignment, within the
326 static struct vmap_area *alloc_vmap_area(unsigned long size,
328 unsigned long vstart, unsigned long vend,
329 int node, gfp_t gfp_mask)
331 struct vmap_area *va;
336 BUG_ON(size & ~PAGE_MASK);
338 va = kmalloc_node(sizeof(struct vmap_area),
339 gfp_mask & GFP_RECLAIM_MASK, node);
341 return ERR_PTR(-ENOMEM);
344 addr = ALIGN(vstart, align);
346 spin_lock(&vmap_area_lock);
347 /* XXX: could have a last_hole cache */
348 n = vmap_area_root.rb_node;
350 struct vmap_area *first = NULL;
353 struct vmap_area *tmp;
354 tmp = rb_entry(n, struct vmap_area, rb_node);
355 if (tmp->va_end >= addr) {
356 if (!first && tmp->va_start < addr + size)
368 if (first->va_end < addr) {
369 n = rb_next(&first->rb_node);
371 first = rb_entry(n, struct vmap_area, rb_node);
376 while (addr + size > first->va_start && addr + size <= vend) {
377 addr = ALIGN(first->va_end + PAGE_SIZE, align);
379 n = rb_next(&first->rb_node);
381 first = rb_entry(n, struct vmap_area, rb_node);
387 if (addr + size > vend) {
388 spin_unlock(&vmap_area_lock);
390 purge_vmap_area_lazy();
394 if (printk_ratelimit())
396 "vmap allocation for size %lu failed: "
397 "use vmalloc=<size> to increase size.\n", size);
398 return ERR_PTR(-EBUSY);
401 BUG_ON(addr & (align-1));
404 va->va_end = addr + size;
406 __insert_vmap_area(va);
407 spin_unlock(&vmap_area_lock);
412 static void rcu_free_va(struct rcu_head *head)
414 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
419 static void __free_vmap_area(struct vmap_area *va)
421 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
422 rb_erase(&va->rb_node, &vmap_area_root);
423 RB_CLEAR_NODE(&va->rb_node);
424 list_del_rcu(&va->list);
426 call_rcu(&va->rcu_head, rcu_free_va);
430 * Free a region of KVA allocated by alloc_vmap_area
432 static void free_vmap_area(struct vmap_area *va)
434 spin_lock(&vmap_area_lock);
435 __free_vmap_area(va);
436 spin_unlock(&vmap_area_lock);
440 * Clear the pagetable entries of a given vmap_area
442 static void unmap_vmap_area(struct vmap_area *va)
444 vunmap_page_range(va->va_start, va->va_end);
447 static void vmap_debug_free_range(unsigned long start, unsigned long end)
450 * Unmap page tables and force a TLB flush immediately if
451 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
452 * bugs similarly to those in linear kernel virtual address
453 * space after a page has been freed.
455 * All the lazy freeing logic is still retained, in order to
456 * minimise intrusiveness of this debugging feature.
458 * This is going to be *slow* (linear kernel virtual address
459 * debugging doesn't do a broadcast TLB flush so it is a lot
462 #ifdef CONFIG_DEBUG_PAGEALLOC
463 vunmap_page_range(start, end);
464 flush_tlb_kernel_range(start, end);
469 * lazy_max_pages is the maximum amount of virtual address space we gather up
470 * before attempting to purge with a TLB flush.
472 * There is a tradeoff here: a larger number will cover more kernel page tables
473 * and take slightly longer to purge, but it will linearly reduce the number of
474 * global TLB flushes that must be performed. It would seem natural to scale
475 * this number up linearly with the number of CPUs (because vmapping activity
476 * could also scale linearly with the number of CPUs), however it is likely
477 * that in practice, workloads might be constrained in other ways that mean
478 * vmap activity will not scale linearly with CPUs. Also, I want to be
479 * conservative and not introduce a big latency on huge systems, so go with
480 * a less aggressive log scale. It will still be an improvement over the old
481 * code, and it will be simple to change the scale factor if we find that it
482 * becomes a problem on bigger systems.
484 static unsigned long lazy_max_pages(void)
488 log = fls(num_online_cpus());
490 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
493 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
496 * Purges all lazily-freed vmap areas.
498 * If sync is 0 then don't purge if there is already a purge in progress.
499 * If force_flush is 1, then flush kernel TLBs between *start and *end even
500 * if we found no lazy vmap areas to unmap (callers can use this to optimise
501 * their own TLB flushing).
502 * Returns with *start = min(*start, lowest purged address)
503 * *end = max(*end, highest purged address)
505 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
506 int sync, int force_flush)
508 static DEFINE_SPINLOCK(purge_lock);
510 struct vmap_area *va;
514 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
515 * should not expect such behaviour. This just simplifies locking for
516 * the case that isn't actually used at the moment anyway.
518 if (!sync && !force_flush) {
519 if (!spin_trylock(&purge_lock))
522 spin_lock(&purge_lock);
525 list_for_each_entry_rcu(va, &vmap_area_list, list) {
526 if (va->flags & VM_LAZY_FREE) {
527 if (va->va_start < *start)
528 *start = va->va_start;
529 if (va->va_end > *end)
531 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
533 list_add_tail(&va->purge_list, &valist);
534 va->flags |= VM_LAZY_FREEING;
535 va->flags &= ~VM_LAZY_FREE;
541 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
542 atomic_sub(nr, &vmap_lazy_nr);
545 if (nr || force_flush)
546 flush_tlb_kernel_range(*start, *end);
549 spin_lock(&vmap_area_lock);
550 list_for_each_entry(va, &valist, purge_list)
551 __free_vmap_area(va);
552 spin_unlock(&vmap_area_lock);
554 spin_unlock(&purge_lock);
558 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
559 * is already purging.
561 static void try_purge_vmap_area_lazy(void)
563 unsigned long start = ULONG_MAX, end = 0;
565 __purge_vmap_area_lazy(&start, &end, 0, 0);
569 * Kick off a purge of the outstanding lazy areas.
571 static void purge_vmap_area_lazy(void)
573 unsigned long start = ULONG_MAX, end = 0;
575 __purge_vmap_area_lazy(&start, &end, 1, 0);
579 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
580 * called for the correct range previously.
582 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
584 va->flags |= VM_LAZY_FREE;
585 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
586 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
587 try_purge_vmap_area_lazy();
591 * Free and unmap a vmap area
593 static void free_unmap_vmap_area(struct vmap_area *va)
595 flush_cache_vunmap(va->va_start, va->va_end);
596 free_unmap_vmap_area_noflush(va);
599 static struct vmap_area *find_vmap_area(unsigned long addr)
601 struct vmap_area *va;
603 spin_lock(&vmap_area_lock);
604 va = __find_vmap_area(addr);
605 spin_unlock(&vmap_area_lock);
610 static void free_unmap_vmap_area_addr(unsigned long addr)
612 struct vmap_area *va;
614 va = find_vmap_area(addr);
616 free_unmap_vmap_area(va);
620 /*** Per cpu kva allocator ***/
623 * vmap space is limited especially on 32 bit architectures. Ensure there is
624 * room for at least 16 percpu vmap blocks per CPU.
627 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
628 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
629 * instead (we just need a rough idea)
631 #if BITS_PER_LONG == 32
632 #define VMALLOC_SPACE (128UL*1024*1024)
634 #define VMALLOC_SPACE (128UL*1024*1024*1024)
637 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
638 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
639 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
640 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
641 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
642 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
643 #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
644 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
645 VMALLOC_PAGES / NR_CPUS / 16))
647 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
649 static bool vmap_initialized __read_mostly = false;
651 struct vmap_block_queue {
653 struct list_head free;
654 struct list_head dirty;
655 unsigned int nr_dirty;
660 struct vmap_area *va;
661 struct vmap_block_queue *vbq;
662 unsigned long free, dirty;
663 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
664 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
667 struct list_head free_list;
668 struct list_head dirty_list;
670 struct rcu_head rcu_head;
674 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
675 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
678 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
679 * in the free path. Could get rid of this if we change the API to return a
680 * "cookie" from alloc, to be passed to free. But no big deal yet.
682 static DEFINE_SPINLOCK(vmap_block_tree_lock);
683 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
686 * We should probably have a fallback mechanism to allocate virtual memory
687 * out of partially filled vmap blocks. However vmap block sizing should be
688 * fairly reasonable according to the vmalloc size, so it shouldn't be a
692 static unsigned long addr_to_vb_idx(unsigned long addr)
694 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
695 addr /= VMAP_BLOCK_SIZE;
699 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
701 struct vmap_block_queue *vbq;
702 struct vmap_block *vb;
703 struct vmap_area *va;
704 unsigned long vb_idx;
707 node = numa_node_id();
709 vb = kmalloc_node(sizeof(struct vmap_block),
710 gfp_mask & GFP_RECLAIM_MASK, node);
712 return ERR_PTR(-ENOMEM);
714 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
715 VMALLOC_START, VMALLOC_END,
717 if (unlikely(IS_ERR(va))) {
719 return ERR_PTR(PTR_ERR(va));
722 err = radix_tree_preload(gfp_mask);
729 spin_lock_init(&vb->lock);
731 vb->free = VMAP_BBMAP_BITS;
733 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
734 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
735 INIT_LIST_HEAD(&vb->free_list);
736 INIT_LIST_HEAD(&vb->dirty_list);
738 vb_idx = addr_to_vb_idx(va->va_start);
739 spin_lock(&vmap_block_tree_lock);
740 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
741 spin_unlock(&vmap_block_tree_lock);
743 radix_tree_preload_end();
745 vbq = &get_cpu_var(vmap_block_queue);
747 spin_lock(&vbq->lock);
748 list_add(&vb->free_list, &vbq->free);
749 spin_unlock(&vbq->lock);
750 put_cpu_var(vmap_cpu_blocks);
755 static void rcu_free_vb(struct rcu_head *head)
757 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
762 static void free_vmap_block(struct vmap_block *vb)
764 struct vmap_block *tmp;
765 unsigned long vb_idx;
767 spin_lock(&vb->vbq->lock);
768 if (!list_empty(&vb->free_list))
769 list_del(&vb->free_list);
770 if (!list_empty(&vb->dirty_list))
771 list_del(&vb->dirty_list);
772 spin_unlock(&vb->vbq->lock);
774 vb_idx = addr_to_vb_idx(vb->va->va_start);
775 spin_lock(&vmap_block_tree_lock);
776 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
777 spin_unlock(&vmap_block_tree_lock);
780 free_unmap_vmap_area_noflush(vb->va);
781 call_rcu(&vb->rcu_head, rcu_free_vb);
784 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
786 struct vmap_block_queue *vbq;
787 struct vmap_block *vb;
788 unsigned long addr = 0;
791 BUG_ON(size & ~PAGE_MASK);
792 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
793 order = get_order(size);
797 vbq = &get_cpu_var(vmap_block_queue);
798 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
801 spin_lock(&vb->lock);
802 i = bitmap_find_free_region(vb->alloc_map,
803 VMAP_BBMAP_BITS, order);
806 addr = vb->va->va_start + (i << PAGE_SHIFT);
807 BUG_ON(addr_to_vb_idx(addr) !=
808 addr_to_vb_idx(vb->va->va_start));
809 vb->free -= 1UL << order;
811 spin_lock(&vbq->lock);
812 list_del_init(&vb->free_list);
813 spin_unlock(&vbq->lock);
815 spin_unlock(&vb->lock);
818 spin_unlock(&vb->lock);
820 put_cpu_var(vmap_cpu_blocks);
824 vb = new_vmap_block(gfp_mask);
833 static void vb_free(const void *addr, unsigned long size)
835 unsigned long offset;
836 unsigned long vb_idx;
838 struct vmap_block *vb;
840 BUG_ON(size & ~PAGE_MASK);
841 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
843 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
845 order = get_order(size);
847 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
849 vb_idx = addr_to_vb_idx((unsigned long)addr);
851 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
855 spin_lock(&vb->lock);
856 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
858 spin_lock(&vb->vbq->lock);
859 list_add(&vb->dirty_list, &vb->vbq->dirty);
860 spin_unlock(&vb->vbq->lock);
862 vb->dirty += 1UL << order;
863 if (vb->dirty == VMAP_BBMAP_BITS) {
864 BUG_ON(vb->free || !list_empty(&vb->free_list));
865 spin_unlock(&vb->lock);
868 spin_unlock(&vb->lock);
872 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
874 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
875 * to amortize TLB flushing overheads. What this means is that any page you
876 * have now, may, in a former life, have been mapped into kernel virtual
877 * address by the vmap layer and so there might be some CPUs with TLB entries
878 * still referencing that page (additional to the regular 1:1 kernel mapping).
880 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
881 * be sure that none of the pages we have control over will have any aliases
882 * from the vmap layer.
884 void vm_unmap_aliases(void)
886 unsigned long start = ULONG_MAX, end = 0;
890 if (unlikely(!vmap_initialized))
893 for_each_possible_cpu(cpu) {
894 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
895 struct vmap_block *vb;
898 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
901 spin_lock(&vb->lock);
902 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
903 while (i < VMAP_BBMAP_BITS) {
906 j = find_next_zero_bit(vb->dirty_map,
909 s = vb->va->va_start + (i << PAGE_SHIFT);
910 e = vb->va->va_start + (j << PAGE_SHIFT);
911 vunmap_page_range(s, e);
920 i = find_next_bit(vb->dirty_map,
923 spin_unlock(&vb->lock);
928 __purge_vmap_area_lazy(&start, &end, 1, flush);
930 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
933 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
934 * @mem: the pointer returned by vm_map_ram
935 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
937 void vm_unmap_ram(const void *mem, unsigned int count)
939 unsigned long size = count << PAGE_SHIFT;
940 unsigned long addr = (unsigned long)mem;
943 BUG_ON(addr < VMALLOC_START);
944 BUG_ON(addr > VMALLOC_END);
945 BUG_ON(addr & (PAGE_SIZE-1));
947 debug_check_no_locks_freed(mem, size);
948 vmap_debug_free_range(addr, addr+size);
950 if (likely(count <= VMAP_MAX_ALLOC))
953 free_unmap_vmap_area_addr(addr);
955 EXPORT_SYMBOL(vm_unmap_ram);
958 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
959 * @pages: an array of pointers to the pages to be mapped
960 * @count: number of pages
961 * @node: prefer to allocate data structures on this node
962 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
964 * Returns: a pointer to the address that has been mapped, or %NULL on failure
966 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
968 unsigned long size = count << PAGE_SHIFT;
972 if (likely(count <= VMAP_MAX_ALLOC)) {
973 mem = vb_alloc(size, GFP_KERNEL);
976 addr = (unsigned long)mem;
978 struct vmap_area *va;
979 va = alloc_vmap_area(size, PAGE_SIZE,
980 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
987 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
988 vm_unmap_ram(mem, count);
993 EXPORT_SYMBOL(vm_map_ram);
996 * vm_area_register_early - register vmap area early during boot
997 * @vm: vm_struct to register
998 * @align: requested alignment
1000 * This function is used to register kernel vm area before
1001 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1002 * proper values on entry and other fields should be zero. On return,
1003 * vm->addr contains the allocated address.
1005 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1007 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1009 static size_t vm_init_off __initdata;
1012 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1013 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1015 vm->addr = (void *)addr;
1021 void __init vmalloc_init(void)
1023 struct vmap_area *va;
1024 struct vm_struct *tmp;
1027 for_each_possible_cpu(i) {
1028 struct vmap_block_queue *vbq;
1030 vbq = &per_cpu(vmap_block_queue, i);
1031 spin_lock_init(&vbq->lock);
1032 INIT_LIST_HEAD(&vbq->free);
1033 INIT_LIST_HEAD(&vbq->dirty);
1037 /* Import existing vmlist entries. */
1038 for (tmp = vmlist; tmp; tmp = tmp->next) {
1039 va = alloc_bootmem(sizeof(struct vmap_area));
1040 va->flags = tmp->flags | VM_VM_AREA;
1041 va->va_start = (unsigned long)tmp->addr;
1042 va->va_end = va->va_start + tmp->size;
1043 __insert_vmap_area(va);
1045 vmap_initialized = true;
1049 * map_kernel_range_noflush - map kernel VM area with the specified pages
1050 * @addr: start of the VM area to map
1051 * @size: size of the VM area to map
1052 * @prot: page protection flags to use
1053 * @pages: pages to map
1055 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1056 * specify should have been allocated using get_vm_area() and its
1060 * This function does NOT do any cache flushing. The caller is
1061 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1062 * before calling this function.
1065 * The number of pages mapped on success, -errno on failure.
1067 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1068 pgprot_t prot, struct page **pages)
1070 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1074 * unmap_kernel_range_noflush - unmap kernel VM area
1075 * @addr: start of the VM area to unmap
1076 * @size: size of the VM area to unmap
1078 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1079 * specify should have been allocated using get_vm_area() and its
1083 * This function does NOT do any cache flushing. The caller is
1084 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1085 * before calling this function and flush_tlb_kernel_range() after.
1087 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1089 vunmap_page_range(addr, addr + size);
1093 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1094 * @addr: start of the VM area to unmap
1095 * @size: size of the VM area to unmap
1097 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1098 * the unmapping and tlb after.
1100 void unmap_kernel_range(unsigned long addr, unsigned long size)
1102 unsigned long end = addr + size;
1104 flush_cache_vunmap(addr, end);
1105 vunmap_page_range(addr, end);
1106 flush_tlb_kernel_range(addr, end);
1109 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1111 unsigned long addr = (unsigned long)area->addr;
1112 unsigned long end = addr + area->size - PAGE_SIZE;
1115 err = vmap_page_range(addr, end, prot, *pages);
1123 EXPORT_SYMBOL_GPL(map_vm_area);
1125 /*** Old vmalloc interfaces ***/
1126 DEFINE_RWLOCK(vmlist_lock);
1127 struct vm_struct *vmlist;
1129 static struct vm_struct *__get_vm_area_node(unsigned long size,
1130 unsigned long flags, unsigned long start, unsigned long end,
1131 int node, gfp_t gfp_mask, void *caller)
1133 static struct vmap_area *va;
1134 struct vm_struct *area;
1135 struct vm_struct *tmp, **p;
1136 unsigned long align = 1;
1138 BUG_ON(in_interrupt());
1139 if (flags & VM_IOREMAP) {
1140 int bit = fls(size);
1142 if (bit > IOREMAP_MAX_ORDER)
1143 bit = IOREMAP_MAX_ORDER;
1144 else if (bit < PAGE_SHIFT)
1150 size = PAGE_ALIGN(size);
1151 if (unlikely(!size))
1154 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1155 if (unlikely(!area))
1159 * We always allocate a guard page.
1163 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1169 area->flags = flags;
1170 area->addr = (void *)va->va_start;
1174 area->phys_addr = 0;
1175 area->caller = caller;
1177 va->flags |= VM_VM_AREA;
1179 write_lock(&vmlist_lock);
1180 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1181 if (tmp->addr >= area->addr)
1186 write_unlock(&vmlist_lock);
1191 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1192 unsigned long start, unsigned long end)
1194 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1195 __builtin_return_address(0));
1197 EXPORT_SYMBOL_GPL(__get_vm_area);
1199 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1200 unsigned long start, unsigned long end,
1203 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1208 * get_vm_area - reserve a contiguous kernel virtual area
1209 * @size: size of the area
1210 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1212 * Search an area of @size in the kernel virtual mapping area,
1213 * and reserved it for out purposes. Returns the area descriptor
1214 * on success or %NULL on failure.
1216 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1218 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1219 -1, GFP_KERNEL, __builtin_return_address(0));
1222 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1225 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1226 -1, GFP_KERNEL, caller);
1229 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1230 int node, gfp_t gfp_mask)
1232 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
1233 gfp_mask, __builtin_return_address(0));
1236 static struct vm_struct *find_vm_area(const void *addr)
1238 struct vmap_area *va;
1240 va = find_vmap_area((unsigned long)addr);
1241 if (va && va->flags & VM_VM_AREA)
1248 * remove_vm_area - find and remove a continuous kernel virtual area
1249 * @addr: base address
1251 * Search for the kernel VM area starting at @addr, and remove it.
1252 * This function returns the found VM area, but using it is NOT safe
1253 * on SMP machines, except for its size or flags.
1255 struct vm_struct *remove_vm_area(const void *addr)
1257 struct vmap_area *va;
1259 va = find_vmap_area((unsigned long)addr);
1260 if (va && va->flags & VM_VM_AREA) {
1261 struct vm_struct *vm = va->private;
1262 struct vm_struct *tmp, **p;
1264 vmap_debug_free_range(va->va_start, va->va_end);
1265 free_unmap_vmap_area(va);
1266 vm->size -= PAGE_SIZE;
1268 write_lock(&vmlist_lock);
1269 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1272 write_unlock(&vmlist_lock);
1279 static void __vunmap(const void *addr, int deallocate_pages)
1281 struct vm_struct *area;
1286 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1287 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1291 area = remove_vm_area(addr);
1292 if (unlikely(!area)) {
1293 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1298 debug_check_no_locks_freed(addr, area->size);
1299 debug_check_no_obj_freed(addr, area->size);
1301 if (deallocate_pages) {
1304 for (i = 0; i < area->nr_pages; i++) {
1305 struct page *page = area->pages[i];
1311 if (area->flags & VM_VPAGES)
1322 * vfree - release memory allocated by vmalloc()
1323 * @addr: memory base address
1325 * Free the virtually continuous memory area starting at @addr, as
1326 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1327 * NULL, no operation is performed.
1329 * Must not be called in interrupt context.
1331 void vfree(const void *addr)
1333 BUG_ON(in_interrupt());
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());
1352 EXPORT_SYMBOL(vunmap);
1355 * vmap - map an array of pages into virtually contiguous space
1356 * @pages: array of page pointers
1357 * @count: number of pages to map
1358 * @flags: vm_area->flags
1359 * @prot: page protection for the mapping
1361 * Maps @count pages from @pages into contiguous kernel virtual
1364 void *vmap(struct page **pages, unsigned int count,
1365 unsigned long flags, pgprot_t prot)
1367 struct vm_struct *area;
1369 if (count > num_physpages)
1372 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1373 __builtin_return_address(0));
1377 if (map_vm_area(area, prot, &pages)) {
1384 EXPORT_SYMBOL(vmap);
1386 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1387 int node, void *caller);
1388 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1389 pgprot_t prot, int node, void *caller)
1391 struct page **pages;
1392 unsigned int nr_pages, array_size, i;
1394 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1395 array_size = (nr_pages * sizeof(struct page *));
1397 area->nr_pages = nr_pages;
1398 /* Please note that the recursion is strictly bounded. */
1399 if (array_size > PAGE_SIZE) {
1400 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
1401 PAGE_KERNEL, node, caller);
1402 area->flags |= VM_VPAGES;
1404 pages = kmalloc_node(array_size,
1405 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
1408 area->pages = pages;
1409 area->caller = caller;
1411 remove_vm_area(area->addr);
1416 for (i = 0; i < area->nr_pages; i++) {
1420 page = alloc_page(gfp_mask);
1422 page = alloc_pages_node(node, gfp_mask, 0);
1424 if (unlikely(!page)) {
1425 /* Successfully allocated i pages, free them in __vunmap() */
1429 area->pages[i] = page;
1432 if (map_vm_area(area, prot, &pages))
1441 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1443 return __vmalloc_area_node(area, gfp_mask, prot, -1,
1444 __builtin_return_address(0));
1448 * __vmalloc_node - allocate virtually contiguous memory
1449 * @size: allocation size
1450 * @gfp_mask: flags for the page level allocator
1451 * @prot: protection mask for the allocated pages
1452 * @node: node to use for allocation or -1
1453 * @caller: caller's return address
1455 * Allocate enough pages to cover @size from the page level
1456 * allocator with @gfp_mask flags. Map them into contiguous
1457 * kernel virtual space, using a pagetable protection of @prot.
1459 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1460 int node, void *caller)
1462 struct vm_struct *area;
1464 size = PAGE_ALIGN(size);
1465 if (!size || (size >> PAGE_SHIFT) > num_physpages)
1468 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1469 node, gfp_mask, caller);
1474 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1477 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1479 return __vmalloc_node(size, gfp_mask, prot, -1,
1480 __builtin_return_address(0));
1482 EXPORT_SYMBOL(__vmalloc);
1485 * vmalloc - allocate virtually contiguous memory
1486 * @size: allocation size
1487 * Allocate enough pages to cover @size from the page level
1488 * allocator and map them into contiguous kernel virtual space.
1490 * For tight control over page level allocator and protection flags
1491 * use __vmalloc() instead.
1493 void *vmalloc(unsigned long size)
1495 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1496 -1, __builtin_return_address(0));
1498 EXPORT_SYMBOL(vmalloc);
1501 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1502 * @size: allocation size
1504 * The resulting memory area is zeroed so it can be mapped to userspace
1505 * without leaking data.
1507 void *vmalloc_user(unsigned long size)
1509 struct vm_struct *area;
1512 ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1513 PAGE_KERNEL, -1, __builtin_return_address(0));
1515 area = find_vm_area(ret);
1516 area->flags |= VM_USERMAP;
1520 EXPORT_SYMBOL(vmalloc_user);
1523 * vmalloc_node - allocate memory on a specific node
1524 * @size: allocation size
1527 * Allocate enough pages to cover @size from the page level
1528 * allocator and map them into contiguous kernel virtual space.
1530 * For tight control over page level allocator and protection flags
1531 * use __vmalloc() instead.
1533 void *vmalloc_node(unsigned long size, int node)
1535 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1536 node, __builtin_return_address(0));
1538 EXPORT_SYMBOL(vmalloc_node);
1540 #ifndef PAGE_KERNEL_EXEC
1541 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1545 * vmalloc_exec - allocate virtually contiguous, executable memory
1546 * @size: allocation size
1548 * Kernel-internal function to allocate enough pages to cover @size
1549 * the page level allocator and map them into contiguous and
1550 * executable kernel virtual space.
1552 * For tight control over page level allocator and protection flags
1553 * use __vmalloc() instead.
1556 void *vmalloc_exec(unsigned long size)
1558 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1559 -1, __builtin_return_address(0));
1562 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1563 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1564 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1565 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1567 #define GFP_VMALLOC32 GFP_KERNEL
1571 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1572 * @size: allocation size
1574 * Allocate enough 32bit PA addressable pages to cover @size from the
1575 * page level allocator and map them into contiguous kernel virtual space.
1577 void *vmalloc_32(unsigned long size)
1579 return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1580 -1, __builtin_return_address(0));
1582 EXPORT_SYMBOL(vmalloc_32);
1585 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1586 * @size: allocation size
1588 * The resulting memory area is 32bit addressable and zeroed so it can be
1589 * mapped to userspace without leaking data.
1591 void *vmalloc_32_user(unsigned long size)
1593 struct vm_struct *area;
1596 ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1597 -1, __builtin_return_address(0));
1599 area = find_vm_area(ret);
1600 area->flags |= VM_USERMAP;
1604 EXPORT_SYMBOL(vmalloc_32_user);
1606 long vread(char *buf, char *addr, unsigned long count)
1608 struct vm_struct *tmp;
1609 char *vaddr, *buf_start = buf;
1612 /* Don't allow overflow */
1613 if ((unsigned long) addr + count < count)
1614 count = -(unsigned long) addr;
1616 read_lock(&vmlist_lock);
1617 for (tmp = vmlist; tmp; tmp = tmp->next) {
1618 vaddr = (char *) tmp->addr;
1619 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1621 while (addr < vaddr) {
1629 n = vaddr + tmp->size - PAGE_SIZE - addr;
1640 read_unlock(&vmlist_lock);
1641 return buf - buf_start;
1644 long vwrite(char *buf, char *addr, unsigned long count)
1646 struct vm_struct *tmp;
1647 char *vaddr, *buf_start = buf;
1650 /* Don't allow overflow */
1651 if ((unsigned long) addr + count < count)
1652 count = -(unsigned long) addr;
1654 read_lock(&vmlist_lock);
1655 for (tmp = vmlist; tmp; tmp = tmp->next) {
1656 vaddr = (char *) tmp->addr;
1657 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1659 while (addr < vaddr) {
1666 n = vaddr + tmp->size - PAGE_SIZE - addr;
1677 read_unlock(&vmlist_lock);
1678 return buf - buf_start;
1682 * remap_vmalloc_range - map vmalloc pages to userspace
1683 * @vma: vma to cover (map full range of vma)
1684 * @addr: vmalloc memory
1685 * @pgoff: number of pages into addr before first page to map
1687 * Returns: 0 for success, -Exxx on failure
1689 * This function checks that addr is a valid vmalloc'ed area, and
1690 * that it is big enough to cover the vma. Will return failure if
1691 * that criteria isn't met.
1693 * Similar to remap_pfn_range() (see mm/memory.c)
1695 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1696 unsigned long pgoff)
1698 struct vm_struct *area;
1699 unsigned long uaddr = vma->vm_start;
1700 unsigned long usize = vma->vm_end - vma->vm_start;
1702 if ((PAGE_SIZE-1) & (unsigned long)addr)
1705 area = find_vm_area(addr);
1709 if (!(area->flags & VM_USERMAP))
1712 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
1715 addr += pgoff << PAGE_SHIFT;
1717 struct page *page = vmalloc_to_page(addr);
1720 ret = vm_insert_page(vma, uaddr, page);
1727 } while (usize > 0);
1729 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1730 vma->vm_flags |= VM_RESERVED;
1734 EXPORT_SYMBOL(remap_vmalloc_range);
1737 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1740 void __attribute__((weak)) vmalloc_sync_all(void)
1745 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
1747 /* apply_to_page_range() does all the hard work. */
1752 * alloc_vm_area - allocate a range of kernel address space
1753 * @size: size of the area
1755 * Returns: NULL on failure, vm_struct on success
1757 * This function reserves a range of kernel address space, and
1758 * allocates pagetables to map that range. No actual mappings
1759 * are created. If the kernel address space is not shared
1760 * between processes, it syncs the pagetable across all
1763 struct vm_struct *alloc_vm_area(size_t size)
1765 struct vm_struct *area;
1767 area = get_vm_area_caller(size, VM_IOREMAP,
1768 __builtin_return_address(0));
1773 * This ensures that page tables are constructed for this region
1774 * of kernel virtual address space and mapped into init_mm.
1776 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1777 area->size, f, NULL)) {
1782 /* Make sure the pagetables are constructed in process kernel
1788 EXPORT_SYMBOL_GPL(alloc_vm_area);
1790 void free_vm_area(struct vm_struct *area)
1792 struct vm_struct *ret;
1793 ret = remove_vm_area(area->addr);
1794 BUG_ON(ret != area);
1797 EXPORT_SYMBOL_GPL(free_vm_area);
1800 #ifdef CONFIG_PROC_FS
1801 static void *s_start(struct seq_file *m, loff_t *pos)
1804 struct vm_struct *v;
1806 read_lock(&vmlist_lock);
1808 while (n > 0 && v) {
1819 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1821 struct vm_struct *v = p;
1827 static void s_stop(struct seq_file *m, void *p)
1829 read_unlock(&vmlist_lock);
1832 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1835 unsigned int nr, *counters = m->private;
1840 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1842 for (nr = 0; nr < v->nr_pages; nr++)
1843 counters[page_to_nid(v->pages[nr])]++;
1845 for_each_node_state(nr, N_HIGH_MEMORY)
1847 seq_printf(m, " N%u=%u", nr, counters[nr]);
1851 static int s_show(struct seq_file *m, void *p)
1853 struct vm_struct *v = p;
1855 seq_printf(m, "0x%p-0x%p %7ld",
1856 v->addr, v->addr + v->size, v->size);
1859 char buff[KSYM_SYMBOL_LEN];
1862 sprint_symbol(buff, (unsigned long)v->caller);
1867 seq_printf(m, " pages=%d", v->nr_pages);
1870 seq_printf(m, " phys=%lx", v->phys_addr);
1872 if (v->flags & VM_IOREMAP)
1873 seq_printf(m, " ioremap");
1875 if (v->flags & VM_ALLOC)
1876 seq_printf(m, " vmalloc");
1878 if (v->flags & VM_MAP)
1879 seq_printf(m, " vmap");
1881 if (v->flags & VM_USERMAP)
1882 seq_printf(m, " user");
1884 if (v->flags & VM_VPAGES)
1885 seq_printf(m, " vpages");
1887 show_numa_info(m, v);
1892 static const struct seq_operations vmalloc_op = {
1899 static int vmalloc_open(struct inode *inode, struct file *file)
1901 unsigned int *ptr = NULL;
1905 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1906 ret = seq_open(file, &vmalloc_op);
1908 struct seq_file *m = file->private_data;
1915 static const struct file_operations proc_vmalloc_operations = {
1916 .open = vmalloc_open,
1918 .llseek = seq_lseek,
1919 .release = seq_release_private,
1922 static int __init proc_vmalloc_init(void)
1924 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1927 module_init(proc_vmalloc_init);