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.
570 int sysctl_lazy_vfree_pages = 32UL * 1024 * 1024 / PAGE_SIZE;
573 * lazy_vfree_tlb_flush_all_threshold is the maximum size of TLB flush by
574 * area. Beyond that the whole TLB will be flushed.
576 int sysctl_lazy_vfree_tlb_flush_all_threshold = SZ_512M;
578 static unsigned long lazy_max_pages(void)
582 log = fls(num_online_cpus());
584 return log * sysctl_lazy_vfree_pages;
587 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
589 /* for per-CPU blocks */
590 static void purge_fragmented_blocks_allcpus(void);
593 * called before a call to iounmap() if the caller wants vm_area_struct's
596 void set_iounmap_nonlazy(void)
598 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
602 * Purges all lazily-freed vmap areas.
604 * If sync is 0 then don't purge if there is already a purge in progress.
605 * If force_flush is 1, then flush kernel TLBs between *start and *end even
606 * if we found no lazy vmap areas to unmap (callers can use this to optimise
607 * their own TLB flushing).
608 * Returns with *start = min(*start, lowest purged address)
609 * *end = max(*end, highest purged address)
611 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
612 int sync, int force_flush)
614 static DEFINE_SPINLOCK(purge_lock);
616 struct vmap_area *va;
617 struct vmap_area *n_va;
621 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
622 * should not expect such behaviour. This just simplifies locking for
623 * the case that isn't actually used at the moment anyway.
625 if (!sync && !force_flush) {
626 if (!spin_trylock(&purge_lock))
629 spin_lock(&purge_lock);
632 purge_fragmented_blocks_allcpus();
635 list_for_each_entry_rcu(va, &vmap_area_list, list) {
636 if (va->flags & VM_LAZY_FREE) {
637 if (va->va_start < *start)
638 *start = va->va_start;
639 if (va->va_end > *end)
641 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
642 list_add_tail(&va->purge_list, &valist);
643 va->flags |= VM_LAZY_FREEING;
644 va->flags &= ~VM_LAZY_FREE;
650 atomic_sub(nr, &vmap_lazy_nr);
652 if (nr || force_flush) {
653 if (nr > (sysctl_lazy_vfree_tlb_flush_all_threshold >> PAGE_SHIFT))
656 list_for_each_entry(va, &valist, purge_list)
657 flush_tlb_kernel_range(va->va_start, va->va_end);
661 spin_lock(&vmap_area_lock);
662 list_for_each_entry_safe(va, n_va, &valist, purge_list)
663 __free_vmap_area(va);
664 spin_unlock(&vmap_area_lock);
666 spin_unlock(&purge_lock);
670 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
671 * is already purging.
673 static void try_purge_vmap_area_lazy(void)
675 unsigned long start = ULONG_MAX, end = 0;
677 __purge_vmap_area_lazy(&start, &end, 0, 0);
681 * Kick off a purge of the outstanding lazy areas.
683 static void purge_vmap_area_lazy(void)
685 unsigned long start = ULONG_MAX, end = 0;
687 __purge_vmap_area_lazy(&start, &end, 1, 0);
691 * Free a vmap area, caller ensuring that the area has been unmapped
692 * and flush_cache_vunmap had been called for the correct range
695 static void free_vmap_area_noflush(struct vmap_area *va)
697 va->flags |= VM_LAZY_FREE;
698 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
699 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
700 try_purge_vmap_area_lazy();
704 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
705 * called for the correct range previously.
707 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
710 free_vmap_area_noflush(va);
714 * Free and unmap a vmap area
716 static void free_unmap_vmap_area(struct vmap_area *va)
718 flush_cache_vunmap(va->va_start, va->va_end);
719 free_unmap_vmap_area_noflush(va);
722 static struct vmap_area *find_vmap_area(unsigned long addr)
724 struct vmap_area *va;
726 spin_lock(&vmap_area_lock);
727 va = __find_vmap_area(addr);
728 spin_unlock(&vmap_area_lock);
733 static void free_unmap_vmap_area_addr(unsigned long addr)
735 struct vmap_area *va;
737 va = find_vmap_area(addr);
739 free_unmap_vmap_area(va);
743 /*** Per cpu kva allocator ***/
746 * vmap space is limited especially on 32 bit architectures. Ensure there is
747 * room for at least 16 percpu vmap blocks per CPU.
750 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
751 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
752 * instead (we just need a rough idea)
754 #if BITS_PER_LONG == 32
755 #define VMALLOC_SPACE (128UL*1024*1024)
757 #define VMALLOC_SPACE (128UL*1024*1024*1024)
760 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
761 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
762 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
763 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
764 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
765 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
766 #define VMAP_BBMAP_BITS \
767 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
768 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
769 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
771 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
773 static bool vmap_initialized __read_mostly = false;
775 struct vmap_block_queue {
777 struct list_head free;
782 struct vmap_area *va;
783 unsigned long free, dirty;
784 unsigned long dirty_min, dirty_max; /*< dirty range */
785 struct list_head free_list;
786 struct rcu_head rcu_head;
787 struct list_head purge;
790 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
791 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
794 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
795 * in the free path. Could get rid of this if we change the API to return a
796 * "cookie" from alloc, to be passed to free. But no big deal yet.
798 static DEFINE_SPINLOCK(vmap_block_tree_lock);
799 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
802 * We should probably have a fallback mechanism to allocate virtual memory
803 * out of partially filled vmap blocks. However vmap block sizing should be
804 * fairly reasonable according to the vmalloc size, so it shouldn't be a
808 static unsigned long addr_to_vb_idx(unsigned long addr)
810 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
811 addr /= VMAP_BLOCK_SIZE;
815 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
819 addr = va_start + (pages_off << PAGE_SHIFT);
820 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
825 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
826 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
827 * @order: how many 2^order pages should be occupied in newly allocated block
828 * @gfp_mask: flags for the page level allocator
830 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
832 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
834 struct vmap_block_queue *vbq;
835 struct vmap_block *vb;
836 struct vmap_area *va;
837 unsigned long vb_idx;
841 node = numa_node_id();
843 vb = kmalloc_node(sizeof(struct vmap_block),
844 gfp_mask & GFP_RECLAIM_MASK, node);
846 return ERR_PTR(-ENOMEM);
848 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
849 VMALLOC_START, VMALLOC_END,
856 err = radix_tree_preload(gfp_mask);
863 vaddr = vmap_block_vaddr(va->va_start, 0);
864 spin_lock_init(&vb->lock);
866 /* At least something should be left free */
867 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
868 vb->free = VMAP_BBMAP_BITS - (1UL << order);
870 vb->dirty_min = VMAP_BBMAP_BITS;
872 INIT_LIST_HEAD(&vb->free_list);
874 vb_idx = addr_to_vb_idx(va->va_start);
875 spin_lock(&vmap_block_tree_lock);
876 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
877 spin_unlock(&vmap_block_tree_lock);
879 radix_tree_preload_end();
881 vbq = &get_cpu_var(vmap_block_queue);
882 spin_lock(&vbq->lock);
883 list_add_tail_rcu(&vb->free_list, &vbq->free);
884 spin_unlock(&vbq->lock);
885 put_cpu_var(vmap_block_queue);
890 static void free_vmap_block(struct vmap_block *vb)
892 struct vmap_block *tmp;
893 unsigned long vb_idx;
895 vb_idx = addr_to_vb_idx(vb->va->va_start);
896 spin_lock(&vmap_block_tree_lock);
897 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
898 spin_unlock(&vmap_block_tree_lock);
901 free_vmap_area_noflush(vb->va);
902 kfree_rcu(vb, rcu_head);
905 static void purge_fragmented_blocks(int cpu)
908 struct vmap_block *vb;
909 struct vmap_block *n_vb;
910 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
913 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
915 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
918 spin_lock(&vb->lock);
919 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
920 vb->free = 0; /* prevent further allocs after releasing lock */
921 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
923 vb->dirty_max = VMAP_BBMAP_BITS;
924 spin_lock(&vbq->lock);
925 list_del_rcu(&vb->free_list);
926 spin_unlock(&vbq->lock);
927 spin_unlock(&vb->lock);
928 list_add_tail(&vb->purge, &purge);
930 spin_unlock(&vb->lock);
934 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
935 list_del(&vb->purge);
940 static void purge_fragmented_blocks_allcpus(void)
944 for_each_possible_cpu(cpu)
945 purge_fragmented_blocks(cpu);
948 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
950 struct vmap_block_queue *vbq;
951 struct vmap_block *vb;
955 BUG_ON(offset_in_page(size));
956 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
957 if (WARN_ON(size == 0)) {
959 * Allocating 0 bytes isn't what caller wants since
960 * get_order(0) returns funny result. Just warn and terminate
965 order = get_order(size);
968 vbq = &get_cpu_var(vmap_block_queue);
969 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
970 unsigned long pages_off;
972 spin_lock(&vb->lock);
973 if (vb->free < (1UL << order)) {
974 spin_unlock(&vb->lock);
978 pages_off = VMAP_BBMAP_BITS - vb->free;
979 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
980 vb->free -= 1UL << order;
982 spin_lock(&vbq->lock);
983 list_del_rcu(&vb->free_list);
984 spin_unlock(&vbq->lock);
987 spin_unlock(&vb->lock);
991 put_cpu_var(vmap_block_queue);
994 /* Allocate new block if nothing was found */
996 vaddr = new_vmap_block(order, gfp_mask);
1001 static void vb_free(const void *addr, unsigned long size)
1003 unsigned long offset;
1004 unsigned long vb_idx;
1006 struct vmap_block *vb;
1008 BUG_ON(offset_in_page(size));
1009 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1011 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1013 order = get_order(size);
1015 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1016 offset >>= PAGE_SHIFT;
1018 vb_idx = addr_to_vb_idx((unsigned long)addr);
1020 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1024 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1026 spin_lock(&vb->lock);
1028 /* Expand dirty range */
1029 vb->dirty_min = min(vb->dirty_min, offset);
1030 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1032 vb->dirty += 1UL << order;
1033 if (vb->dirty == VMAP_BBMAP_BITS) {
1035 spin_unlock(&vb->lock);
1036 free_vmap_block(vb);
1038 spin_unlock(&vb->lock);
1042 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1044 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1045 * to amortize TLB flushing overheads. What this means is that any page you
1046 * have now, may, in a former life, have been mapped into kernel virtual
1047 * address by the vmap layer and so there might be some CPUs with TLB entries
1048 * still referencing that page (additional to the regular 1:1 kernel mapping).
1050 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1051 * be sure that none of the pages we have control over will have any aliases
1052 * from the vmap layer.
1054 void vm_unmap_aliases(void)
1056 unsigned long start = ULONG_MAX, end = 0;
1060 if (unlikely(!vmap_initialized))
1063 for_each_possible_cpu(cpu) {
1064 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1065 struct vmap_block *vb;
1068 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1069 spin_lock(&vb->lock);
1071 unsigned long va_start = vb->va->va_start;
1074 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1075 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1077 start = min(s, start);
1082 spin_unlock(&vb->lock);
1087 __purge_vmap_area_lazy(&start, &end, 1, flush);
1089 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1092 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1093 * @mem: the pointer returned by vm_map_ram
1094 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1096 void vm_unmap_ram(const void *mem, unsigned int count)
1098 unsigned long size = count << PAGE_SHIFT;
1099 unsigned long addr = (unsigned long)mem;
1102 BUG_ON(addr < VMALLOC_START);
1103 BUG_ON(addr > VMALLOC_END);
1104 BUG_ON(addr & (PAGE_SIZE-1));
1106 debug_check_no_locks_freed(mem, size);
1107 vmap_debug_free_range(addr, addr+size);
1109 if (likely(count <= VMAP_MAX_ALLOC))
1112 free_unmap_vmap_area_addr(addr);
1114 EXPORT_SYMBOL(vm_unmap_ram);
1117 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1118 * @pages: an array of pointers to the pages to be mapped
1119 * @count: number of pages
1120 * @node: prefer to allocate data structures on this node
1121 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1123 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1124 * faster than vmap so it's good. But if you mix long-life and short-life
1125 * objects with vm_map_ram(), it could consume lots of address space through
1126 * fragmentation (especially on a 32bit machine). You could see failures in
1127 * the end. Please use this function for short-lived objects.
1129 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1131 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1133 unsigned long size = count << PAGE_SHIFT;
1137 if (likely(count <= VMAP_MAX_ALLOC)) {
1138 mem = vb_alloc(size, GFP_KERNEL);
1141 addr = (unsigned long)mem;
1143 struct vmap_area *va;
1144 va = alloc_vmap_area(size, PAGE_SIZE,
1145 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1149 addr = va->va_start;
1152 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1153 vm_unmap_ram(mem, count);
1158 EXPORT_SYMBOL(vm_map_ram);
1160 static struct vm_struct *vmlist __initdata;
1162 * vm_area_add_early - add vmap area early during boot
1163 * @vm: vm_struct to add
1165 * This function is used to add fixed kernel vm area to vmlist before
1166 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1167 * should contain proper values and the other fields should be zero.
1169 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1171 void __init vm_area_add_early(struct vm_struct *vm)
1173 struct vm_struct *tmp, **p;
1175 BUG_ON(vmap_initialized);
1176 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1177 if (tmp->addr >= vm->addr) {
1178 BUG_ON(tmp->addr < vm->addr + vm->size);
1181 BUG_ON(tmp->addr + tmp->size > vm->addr);
1188 * vm_area_register_early - register vmap area early during boot
1189 * @vm: vm_struct to register
1190 * @align: requested alignment
1192 * This function is used to register kernel vm area before
1193 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1194 * proper values on entry and other fields should be zero. On return,
1195 * vm->addr contains the allocated address.
1197 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1199 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1201 static size_t vm_init_off __initdata;
1204 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1205 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1207 vm->addr = (void *)addr;
1209 vm_area_add_early(vm);
1212 void __init vmalloc_init(void)
1214 struct vmap_area *va;
1215 struct vm_struct *tmp;
1218 for_each_possible_cpu(i) {
1219 struct vmap_block_queue *vbq;
1220 struct vfree_deferred *p;
1222 vbq = &per_cpu(vmap_block_queue, i);
1223 spin_lock_init(&vbq->lock);
1224 INIT_LIST_HEAD(&vbq->free);
1225 p = &per_cpu(vfree_deferred, i);
1226 init_llist_head(&p->list);
1227 INIT_WORK(&p->wq, free_work);
1230 /* Import existing vmlist entries. */
1231 for (tmp = vmlist; tmp; tmp = tmp->next) {
1232 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1233 va->flags = VM_VM_AREA;
1234 va->va_start = (unsigned long)tmp->addr;
1235 va->va_end = va->va_start + tmp->size;
1237 __insert_vmap_area(va);
1240 vmap_area_pcpu_hole = VMALLOC_END;
1242 vmap_initialized = true;
1246 * map_kernel_range_noflush - map kernel VM area with the specified pages
1247 * @addr: start of the VM area to map
1248 * @size: size of the VM area to map
1249 * @prot: page protection flags to use
1250 * @pages: pages to map
1252 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1253 * specify should have been allocated using get_vm_area() and its
1257 * This function does NOT do any cache flushing. The caller is
1258 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1259 * before calling this function.
1262 * The number of pages mapped on success, -errno on failure.
1264 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1265 pgprot_t prot, struct page **pages)
1267 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1271 * unmap_kernel_range_noflush - unmap kernel VM area
1272 * @addr: start of the VM area to unmap
1273 * @size: size of the VM area to unmap
1275 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1276 * specify should have been allocated using get_vm_area() and its
1280 * This function does NOT do any cache flushing. The caller is
1281 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1282 * before calling this function and flush_tlb_kernel_range() after.
1284 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1286 vunmap_page_range(addr, addr + size);
1288 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1291 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1292 * @addr: start of the VM area to unmap
1293 * @size: size of the VM area to unmap
1295 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1296 * the unmapping and tlb after.
1298 void unmap_kernel_range(unsigned long addr, unsigned long size)
1300 unsigned long end = addr + size;
1302 flush_cache_vunmap(addr, end);
1303 vunmap_page_range(addr, end);
1304 flush_tlb_kernel_range(addr, end);
1306 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1308 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1310 unsigned long addr = (unsigned long)area->addr;
1311 unsigned long end = addr + get_vm_area_size(area);
1314 err = vmap_page_range(addr, end, prot, pages);
1316 return err > 0 ? 0 : err;
1318 EXPORT_SYMBOL_GPL(map_vm_area);
1320 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1321 unsigned long flags, const void *caller)
1323 spin_lock(&vmap_area_lock);
1325 vm->addr = (void *)va->va_start;
1326 vm->size = va->va_end - va->va_start;
1327 vm->caller = caller;
1329 va->flags |= VM_VM_AREA;
1330 spin_unlock(&vmap_area_lock);
1333 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1336 * Before removing VM_UNINITIALIZED,
1337 * we should make sure that vm has proper values.
1338 * Pair with smp_rmb() in show_numa_info().
1341 vm->flags &= ~VM_UNINITIALIZED;
1344 static struct vm_struct *__get_vm_area_node(unsigned long size,
1345 unsigned long align, unsigned long flags, unsigned long start,
1346 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1348 struct vmap_area *va;
1349 struct vm_struct *area;
1351 BUG_ON(in_interrupt());
1352 if (flags & VM_IOREMAP)
1353 align = 1ul << clamp_t(int, fls_long(size),
1354 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1356 size = PAGE_ALIGN(size);
1357 if (unlikely(!size))
1360 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1361 if (unlikely(!area))
1364 if (!(flags & VM_NO_GUARD))
1367 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1373 setup_vmalloc_vm(area, va, flags, caller);
1378 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1379 unsigned long start, unsigned long end)
1381 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1382 GFP_KERNEL, __builtin_return_address(0));
1384 EXPORT_SYMBOL_GPL(__get_vm_area);
1386 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1387 unsigned long start, unsigned long end,
1390 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1391 GFP_KERNEL, caller);
1395 * get_vm_area - reserve a contiguous kernel virtual area
1396 * @size: size of the area
1397 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1399 * Search an area of @size in the kernel virtual mapping area,
1400 * and reserved it for out purposes. Returns the area descriptor
1401 * on success or %NULL on failure.
1403 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1405 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1406 NUMA_NO_NODE, GFP_KERNEL,
1407 __builtin_return_address(0));
1409 EXPORT_SYMBOL_GPL(get_vm_area);
1411 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1414 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1415 NUMA_NO_NODE, GFP_KERNEL, caller);
1419 * find_vm_area - find a continuous kernel virtual area
1420 * @addr: base address
1422 * Search for the kernel VM area starting at @addr, and return it.
1423 * It is up to the caller to do all required locking to keep the returned
1426 struct vm_struct *find_vm_area(const void *addr)
1428 struct vmap_area *va;
1430 va = find_vmap_area((unsigned long)addr);
1431 if (va && va->flags & VM_VM_AREA)
1438 * remove_vm_area - find and remove a continuous kernel virtual area
1439 * @addr: base address
1441 * Search for the kernel VM area starting at @addr, and remove it.
1442 * This function returns the found VM area, but using it is NOT safe
1443 * on SMP machines, except for its size or flags.
1445 struct vm_struct *remove_vm_area(const void *addr)
1447 struct vmap_area *va;
1449 va = find_vmap_area((unsigned long)addr);
1450 if (va && va->flags & VM_VM_AREA) {
1451 struct vm_struct *vm = va->vm;
1453 spin_lock(&vmap_area_lock);
1455 va->flags &= ~VM_VM_AREA;
1456 spin_unlock(&vmap_area_lock);
1458 vmap_debug_free_range(va->va_start, va->va_end);
1459 kasan_free_shadow(vm);
1460 free_unmap_vmap_area(va);
1467 static void __vunmap(const void *addr, int deallocate_pages)
1469 struct vm_struct *area;
1474 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1478 area = remove_vm_area(addr);
1479 if (unlikely(!area)) {
1480 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1485 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1486 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1488 if (deallocate_pages) {
1491 for (i = 0; i < area->nr_pages; i++) {
1492 struct page *page = area->pages[i];
1498 if (area->flags & VM_VPAGES)
1509 * vfree - release memory allocated by vmalloc()
1510 * @addr: memory base address
1512 * Free the virtually continuous memory area starting at @addr, as
1513 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1514 * NULL, no operation is performed.
1516 * Must not be called in NMI context (strictly speaking, only if we don't
1517 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1518 * conventions for vfree() arch-depenedent would be a really bad idea)
1520 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1522 void vfree(const void *addr)
1526 kmemleak_free(addr);
1530 if (unlikely(in_interrupt())) {
1531 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1532 if (llist_add((struct llist_node *)addr, &p->list))
1533 schedule_work(&p->wq);
1537 EXPORT_SYMBOL(vfree);
1540 * vunmap - release virtual mapping obtained by vmap()
1541 * @addr: memory base address
1543 * Free the virtually contiguous memory area starting at @addr,
1544 * which was created from the page array passed to vmap().
1546 * Must not be called in interrupt context.
1548 void vunmap(const void *addr)
1550 BUG_ON(in_interrupt());
1555 EXPORT_SYMBOL(vunmap);
1558 * vmap - map an array of pages into virtually contiguous space
1559 * @pages: array of page pointers
1560 * @count: number of pages to map
1561 * @flags: vm_area->flags
1562 * @prot: page protection for the mapping
1564 * Maps @count pages from @pages into contiguous kernel virtual
1567 void *vmap(struct page **pages, unsigned int count,
1568 unsigned long flags, pgprot_t prot)
1570 struct vm_struct *area;
1574 if (count > totalram_pages)
1577 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1578 __builtin_return_address(0));
1582 if (map_vm_area(area, prot, pages)) {
1589 EXPORT_SYMBOL(vmap);
1591 static void *__vmalloc_node(unsigned long size, unsigned long align,
1592 gfp_t gfp_mask, pgprot_t prot,
1593 int node, const void *caller);
1594 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1595 pgprot_t prot, int node)
1597 const int order = 0;
1598 struct page **pages;
1599 unsigned int nr_pages, array_size, i;
1600 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1601 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1603 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1604 array_size = (nr_pages * sizeof(struct page *));
1606 area->nr_pages = nr_pages;
1607 /* Please note that the recursion is strictly bounded. */
1608 if (array_size > PAGE_SIZE) {
1609 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1610 PAGE_KERNEL, node, area->caller);
1611 area->flags |= VM_VPAGES;
1613 pages = kmalloc_node(array_size, nested_gfp, node);
1615 area->pages = pages;
1617 remove_vm_area(area->addr);
1622 for (i = 0; i < area->nr_pages; i++) {
1625 if (node == NUMA_NO_NODE)
1626 page = alloc_page(alloc_mask);
1628 page = alloc_pages_node(node, alloc_mask, order);
1630 if (unlikely(!page)) {
1631 /* Successfully allocated i pages, free them in __vunmap() */
1635 area->pages[i] = page;
1636 if (gfpflags_allow_blocking(gfp_mask))
1640 if (map_vm_area(area, prot, pages))
1645 warn_alloc_failed(gfp_mask, order,
1646 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1647 (area->nr_pages*PAGE_SIZE), area->size);
1653 * __vmalloc_node_range - allocate virtually contiguous memory
1654 * @size: allocation size
1655 * @align: desired alignment
1656 * @start: vm area range start
1657 * @end: vm area range end
1658 * @gfp_mask: flags for the page level allocator
1659 * @prot: protection mask for the allocated pages
1660 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1661 * @node: node to use for allocation or NUMA_NO_NODE
1662 * @caller: caller's return address
1664 * Allocate enough pages to cover @size from the page level
1665 * allocator with @gfp_mask flags. Map them into contiguous
1666 * kernel virtual space, using a pagetable protection of @prot.
1668 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1669 unsigned long start, unsigned long end, gfp_t gfp_mask,
1670 pgprot_t prot, unsigned long vm_flags, int node,
1673 struct vm_struct *area;
1675 unsigned long real_size = size;
1677 size = PAGE_ALIGN(size);
1678 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1681 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1682 vm_flags, start, end, node, gfp_mask, caller);
1686 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1691 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1692 * flag. It means that vm_struct is not fully initialized.
1693 * Now, it is fully initialized, so remove this flag here.
1695 clear_vm_uninitialized_flag(area);
1698 * A ref_count = 2 is needed because vm_struct allocated in
1699 * __get_vm_area_node() contains a reference to the virtual address of
1700 * the vmalloc'ed block.
1702 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1707 warn_alloc_failed(gfp_mask, 0,
1708 "vmalloc: allocation failure: %lu bytes\n",
1714 * __vmalloc_node - allocate virtually contiguous memory
1715 * @size: allocation size
1716 * @align: desired alignment
1717 * @gfp_mask: flags for the page level allocator
1718 * @prot: protection mask for the allocated pages
1719 * @node: node to use for allocation or NUMA_NO_NODE
1720 * @caller: caller's return address
1722 * Allocate enough pages to cover @size from the page level
1723 * allocator with @gfp_mask flags. Map them into contiguous
1724 * kernel virtual space, using a pagetable protection of @prot.
1726 static void *__vmalloc_node(unsigned long size, unsigned long align,
1727 gfp_t gfp_mask, pgprot_t prot,
1728 int node, const void *caller)
1730 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1731 gfp_mask, prot, 0, node, caller);
1734 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1736 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1737 __builtin_return_address(0));
1739 EXPORT_SYMBOL(__vmalloc);
1741 static inline void *__vmalloc_node_flags(unsigned long size,
1742 int node, gfp_t flags)
1744 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1745 node, __builtin_return_address(0));
1749 * vmalloc - allocate virtually contiguous memory
1750 * @size: allocation size
1751 * Allocate enough pages to cover @size from the page level
1752 * allocator and map them into contiguous kernel virtual space.
1754 * For tight control over page level allocator and protection flags
1755 * use __vmalloc() instead.
1757 void *vmalloc(unsigned long size)
1759 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1760 GFP_KERNEL | __GFP_HIGHMEM);
1762 EXPORT_SYMBOL(vmalloc);
1765 * vzalloc - allocate virtually contiguous memory with zero fill
1766 * @size: allocation size
1767 * Allocate enough pages to cover @size from the page level
1768 * allocator and map them into contiguous kernel virtual space.
1769 * The memory allocated is set to zero.
1771 * For tight control over page level allocator and protection flags
1772 * use __vmalloc() instead.
1774 void *vzalloc(unsigned long size)
1776 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1777 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1779 EXPORT_SYMBOL(vzalloc);
1782 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1783 * @size: allocation size
1785 * The resulting memory area is zeroed so it can be mapped to userspace
1786 * without leaking data.
1788 void *vmalloc_user(unsigned long size)
1790 struct vm_struct *area;
1793 ret = __vmalloc_node(size, SHMLBA,
1794 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1795 PAGE_KERNEL, NUMA_NO_NODE,
1796 __builtin_return_address(0));
1798 area = find_vm_area(ret);
1799 area->flags |= VM_USERMAP;
1803 EXPORT_SYMBOL(vmalloc_user);
1806 * vmalloc_node - allocate memory on a specific node
1807 * @size: allocation size
1810 * Allocate enough pages to cover @size from the page level
1811 * allocator and map them into contiguous kernel virtual space.
1813 * For tight control over page level allocator and protection flags
1814 * use __vmalloc() instead.
1816 void *vmalloc_node(unsigned long size, int node)
1818 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1819 node, __builtin_return_address(0));
1821 EXPORT_SYMBOL(vmalloc_node);
1824 * vzalloc_node - allocate memory on a specific node with zero fill
1825 * @size: allocation size
1828 * Allocate enough pages to cover @size from the page level
1829 * allocator and map them into contiguous kernel virtual space.
1830 * The memory allocated is set to zero.
1832 * For tight control over page level allocator and protection flags
1833 * use __vmalloc_node() instead.
1835 void *vzalloc_node(unsigned long size, int node)
1837 return __vmalloc_node_flags(size, node,
1838 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1840 EXPORT_SYMBOL(vzalloc_node);
1842 #ifndef PAGE_KERNEL_EXEC
1843 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1847 * vmalloc_exec - allocate virtually contiguous, executable memory
1848 * @size: allocation size
1850 * Kernel-internal function to allocate enough pages to cover @size
1851 * the page level allocator and map them into contiguous and
1852 * executable kernel virtual space.
1854 * For tight control over page level allocator and protection flags
1855 * use __vmalloc() instead.
1858 void *vmalloc_exec(unsigned long size)
1860 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1861 NUMA_NO_NODE, __builtin_return_address(0));
1864 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1865 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1866 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1867 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1869 #define GFP_VMALLOC32 GFP_KERNEL
1873 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1874 * @size: allocation size
1876 * Allocate enough 32bit PA addressable pages to cover @size from the
1877 * page level allocator and map them into contiguous kernel virtual space.
1879 void *vmalloc_32(unsigned long size)
1881 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1882 NUMA_NO_NODE, __builtin_return_address(0));
1884 EXPORT_SYMBOL(vmalloc_32);
1887 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1888 * @size: allocation size
1890 * The resulting memory area is 32bit addressable and zeroed so it can be
1891 * mapped to userspace without leaking data.
1893 void *vmalloc_32_user(unsigned long size)
1895 struct vm_struct *area;
1898 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1899 NUMA_NO_NODE, __builtin_return_address(0));
1901 area = find_vm_area(ret);
1902 area->flags |= VM_USERMAP;
1906 EXPORT_SYMBOL(vmalloc_32_user);
1909 * small helper routine , copy contents to buf from addr.
1910 * If the page is not present, fill zero.
1913 static int aligned_vread(char *buf, char *addr, unsigned long count)
1919 unsigned long offset, length;
1921 offset = offset_in_page(addr);
1922 length = PAGE_SIZE - offset;
1925 p = vmalloc_to_page(addr);
1927 * To do safe access to this _mapped_ area, we need
1928 * lock. But adding lock here means that we need to add
1929 * overhead of vmalloc()/vfree() calles for this _debug_
1930 * interface, rarely used. Instead of that, we'll use
1931 * kmap() and get small overhead in this access function.
1935 * we can expect USER0 is not used (see vread/vwrite's
1936 * function description)
1938 void *map = kmap_atomic(p);
1939 memcpy(buf, map + offset, length);
1942 memset(buf, 0, length);
1952 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1958 unsigned long offset, length;
1960 offset = offset_in_page(addr);
1961 length = PAGE_SIZE - offset;
1964 p = vmalloc_to_page(addr);
1966 * To do safe access to this _mapped_ area, we need
1967 * lock. But adding lock here means that we need to add
1968 * overhead of vmalloc()/vfree() calles for this _debug_
1969 * interface, rarely used. Instead of that, we'll use
1970 * kmap() and get small overhead in this access function.
1974 * we can expect USER0 is not used (see vread/vwrite's
1975 * function description)
1977 void *map = kmap_atomic(p);
1978 memcpy(map + offset, buf, length);
1990 * vread() - read vmalloc area in a safe way.
1991 * @buf: buffer for reading data
1992 * @addr: vm address.
1993 * @count: number of bytes to be read.
1995 * Returns # of bytes which addr and buf should be increased.
1996 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1997 * includes any intersect with alive vmalloc area.
1999 * This function checks that addr is a valid vmalloc'ed area, and
2000 * copy data from that area to a given buffer. If the given memory range
2001 * of [addr...addr+count) includes some valid address, data is copied to
2002 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2003 * IOREMAP area is treated as memory hole and no copy is done.
2005 * If [addr...addr+count) doesn't includes any intersects with alive
2006 * vm_struct area, returns 0. @buf should be kernel's buffer.
2008 * Note: In usual ops, vread() is never necessary because the caller
2009 * should know vmalloc() area is valid and can use memcpy().
2010 * This is for routines which have to access vmalloc area without
2011 * any informaion, as /dev/kmem.
2015 long vread(char *buf, char *addr, unsigned long count)
2017 struct vmap_area *va;
2018 struct vm_struct *vm;
2019 char *vaddr, *buf_start = buf;
2020 unsigned long buflen = count;
2023 /* Don't allow overflow */
2024 if ((unsigned long) addr + count < count)
2025 count = -(unsigned long) addr;
2027 spin_lock(&vmap_area_lock);
2028 list_for_each_entry(va, &vmap_area_list, list) {
2032 if (!(va->flags & VM_VM_AREA))
2036 vaddr = (char *) vm->addr;
2037 if (addr >= vaddr + get_vm_area_size(vm))
2039 while (addr < vaddr) {
2047 n = vaddr + get_vm_area_size(vm) - addr;
2050 if (!(vm->flags & VM_IOREMAP))
2051 aligned_vread(buf, addr, n);
2052 else /* IOREMAP area is treated as memory hole */
2059 spin_unlock(&vmap_area_lock);
2061 if (buf == buf_start)
2063 /* zero-fill memory holes */
2064 if (buf != buf_start + buflen)
2065 memset(buf, 0, buflen - (buf - buf_start));
2071 * vwrite() - write vmalloc area in a safe way.
2072 * @buf: buffer for source data
2073 * @addr: vm address.
2074 * @count: number of bytes to be read.
2076 * Returns # of bytes which addr and buf should be incresed.
2077 * (same number to @count).
2078 * If [addr...addr+count) doesn't includes any intersect with valid
2079 * vmalloc area, returns 0.
2081 * This function checks that addr is a valid vmalloc'ed area, and
2082 * copy data from a buffer to the given addr. If specified range of
2083 * [addr...addr+count) includes some valid address, data is copied from
2084 * proper area of @buf. If there are memory holes, no copy to hole.
2085 * IOREMAP area is treated as memory hole and no copy is done.
2087 * If [addr...addr+count) doesn't includes any intersects with alive
2088 * vm_struct area, returns 0. @buf should be kernel's buffer.
2090 * Note: In usual ops, vwrite() is never necessary because the caller
2091 * should know vmalloc() area is valid and can use memcpy().
2092 * This is for routines which have to access vmalloc area without
2093 * any informaion, as /dev/kmem.
2096 long vwrite(char *buf, char *addr, unsigned long count)
2098 struct vmap_area *va;
2099 struct vm_struct *vm;
2101 unsigned long n, buflen;
2104 /* Don't allow overflow */
2105 if ((unsigned long) addr + count < count)
2106 count = -(unsigned long) addr;
2109 spin_lock(&vmap_area_lock);
2110 list_for_each_entry(va, &vmap_area_list, list) {
2114 if (!(va->flags & VM_VM_AREA))
2118 vaddr = (char *) vm->addr;
2119 if (addr >= vaddr + get_vm_area_size(vm))
2121 while (addr < vaddr) {
2128 n = vaddr + get_vm_area_size(vm) - addr;
2131 if (!(vm->flags & VM_IOREMAP)) {
2132 aligned_vwrite(buf, addr, n);
2140 spin_unlock(&vmap_area_lock);
2147 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2148 * @vma: vma to cover
2149 * @uaddr: target user address to start at
2150 * @kaddr: virtual address of vmalloc kernel memory
2151 * @size: size of map area
2153 * Returns: 0 for success, -Exxx on failure
2155 * This function checks that @kaddr is a valid vmalloc'ed area,
2156 * and that it is big enough to cover the range starting at
2157 * @uaddr in @vma. Will return failure if that criteria isn't
2160 * Similar to remap_pfn_range() (see mm/memory.c)
2162 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2163 void *kaddr, unsigned long size)
2165 struct vm_struct *area;
2167 size = PAGE_ALIGN(size);
2169 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2172 area = find_vm_area(kaddr);
2176 if (!(area->flags & VM_USERMAP))
2179 if (kaddr + size > area->addr + area->size)
2183 struct page *page = vmalloc_to_page(kaddr);
2186 ret = vm_insert_page(vma, uaddr, page);
2195 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2199 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2202 * remap_vmalloc_range - map vmalloc pages to userspace
2203 * @vma: vma to cover (map full range of vma)
2204 * @addr: vmalloc memory
2205 * @pgoff: number of pages into addr before first page to map
2207 * Returns: 0 for success, -Exxx on failure
2209 * This function checks that addr is a valid vmalloc'ed area, and
2210 * that it is big enough to cover the vma. Will return failure if
2211 * that criteria isn't met.
2213 * Similar to remap_pfn_range() (see mm/memory.c)
2215 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2216 unsigned long pgoff)
2218 return remap_vmalloc_range_partial(vma, vma->vm_start,
2219 addr + (pgoff << PAGE_SHIFT),
2220 vma->vm_end - vma->vm_start);
2222 EXPORT_SYMBOL(remap_vmalloc_range);
2225 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2228 void __weak vmalloc_sync_all(void)
2233 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2245 * alloc_vm_area - allocate a range of kernel address space
2246 * @size: size of the area
2247 * @ptes: returns the PTEs for the address space
2249 * Returns: NULL on failure, vm_struct on success
2251 * This function reserves a range of kernel address space, and
2252 * allocates pagetables to map that range. No actual mappings
2255 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2256 * allocated for the VM area are returned.
2258 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2260 struct vm_struct *area;
2262 area = get_vm_area_caller(size, VM_IOREMAP,
2263 __builtin_return_address(0));
2268 * This ensures that page tables are constructed for this region
2269 * of kernel virtual address space and mapped into init_mm.
2271 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2272 size, f, ptes ? &ptes : NULL)) {
2279 EXPORT_SYMBOL_GPL(alloc_vm_area);
2281 void free_vm_area(struct vm_struct *area)
2283 struct vm_struct *ret;
2284 ret = remove_vm_area(area->addr);
2285 BUG_ON(ret != area);
2288 EXPORT_SYMBOL_GPL(free_vm_area);
2291 static struct vmap_area *node_to_va(struct rb_node *n)
2293 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2297 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2298 * @end: target address
2299 * @pnext: out arg for the next vmap_area
2300 * @pprev: out arg for the previous vmap_area
2302 * Returns: %true if either or both of next and prev are found,
2303 * %false if no vmap_area exists
2305 * Find vmap_areas end addresses of which enclose @end. ie. if not
2306 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2308 static bool pvm_find_next_prev(unsigned long end,
2309 struct vmap_area **pnext,
2310 struct vmap_area **pprev)
2312 struct rb_node *n = vmap_area_root.rb_node;
2313 struct vmap_area *va = NULL;
2316 va = rb_entry(n, struct vmap_area, rb_node);
2317 if (end < va->va_end)
2319 else if (end > va->va_end)
2328 if (va->va_end > end) {
2330 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2333 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2339 * pvm_determine_end - find the highest aligned address between two vmap_areas
2340 * @pnext: in/out arg for the next vmap_area
2341 * @pprev: in/out arg for the previous vmap_area
2344 * Returns: determined end address
2346 * Find the highest aligned address between *@pnext and *@pprev below
2347 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2348 * down address is between the end addresses of the two vmap_areas.
2350 * Please note that the address returned by this function may fall
2351 * inside *@pnext vmap_area. The caller is responsible for checking
2354 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2355 struct vmap_area **pprev,
2356 unsigned long align)
2358 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2362 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2366 while (*pprev && (*pprev)->va_end > addr) {
2368 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2375 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2376 * @offsets: array containing offset of each area
2377 * @sizes: array containing size of each area
2378 * @nr_vms: the number of areas to allocate
2379 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2381 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2382 * vm_structs on success, %NULL on failure
2384 * Percpu allocator wants to use congruent vm areas so that it can
2385 * maintain the offsets among percpu areas. This function allocates
2386 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2387 * be scattered pretty far, distance between two areas easily going up
2388 * to gigabytes. To avoid interacting with regular vmallocs, these
2389 * areas are allocated from top.
2391 * Despite its complicated look, this allocator is rather simple. It
2392 * does everything top-down and scans areas from the end looking for
2393 * matching slot. While scanning, if any of the areas overlaps with
2394 * existing vmap_area, the base address is pulled down to fit the
2395 * area. Scanning is repeated till all the areas fit and then all
2396 * necessary data structres are inserted and the result is returned.
2398 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2399 const size_t *sizes, int nr_vms,
2402 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2403 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2404 struct vmap_area **vas, *prev, *next;
2405 struct vm_struct **vms;
2406 int area, area2, last_area, term_area;
2407 unsigned long base, start, end, last_end;
2408 bool purged = false;
2410 /* verify parameters and allocate data structures */
2411 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2412 for (last_area = 0, area = 0; area < nr_vms; area++) {
2413 start = offsets[area];
2414 end = start + sizes[area];
2416 /* is everything aligned properly? */
2417 BUG_ON(!IS_ALIGNED(offsets[area], align));
2418 BUG_ON(!IS_ALIGNED(sizes[area], align));
2420 /* detect the area with the highest address */
2421 if (start > offsets[last_area])
2424 for (area2 = 0; area2 < nr_vms; area2++) {
2425 unsigned long start2 = offsets[area2];
2426 unsigned long end2 = start2 + sizes[area2];
2431 BUG_ON(start2 >= start && start2 < end);
2432 BUG_ON(end2 <= end && end2 > start);
2435 last_end = offsets[last_area] + sizes[last_area];
2437 if (vmalloc_end - vmalloc_start < last_end) {
2442 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2443 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2447 for (area = 0; area < nr_vms; area++) {
2448 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2449 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2450 if (!vas[area] || !vms[area])
2454 spin_lock(&vmap_area_lock);
2456 /* start scanning - we scan from the top, begin with the last area */
2457 area = term_area = last_area;
2458 start = offsets[area];
2459 end = start + sizes[area];
2461 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2462 base = vmalloc_end - last_end;
2465 base = pvm_determine_end(&next, &prev, align) - end;
2468 BUG_ON(next && next->va_end <= base + end);
2469 BUG_ON(prev && prev->va_end > base + end);
2472 * base might have underflowed, add last_end before
2475 if (base + last_end < vmalloc_start + last_end) {
2476 spin_unlock(&vmap_area_lock);
2478 purge_vmap_area_lazy();
2486 * If next overlaps, move base downwards so that it's
2487 * right below next and then recheck.
2489 if (next && next->va_start < base + end) {
2490 base = pvm_determine_end(&next, &prev, align) - end;
2496 * If prev overlaps, shift down next and prev and move
2497 * base so that it's right below new next and then
2500 if (prev && prev->va_end > base + start) {
2502 prev = node_to_va(rb_prev(&next->rb_node));
2503 base = pvm_determine_end(&next, &prev, align) - end;
2509 * This area fits, move on to the previous one. If
2510 * the previous one is the terminal one, we're done.
2512 area = (area + nr_vms - 1) % nr_vms;
2513 if (area == term_area)
2515 start = offsets[area];
2516 end = start + sizes[area];
2517 pvm_find_next_prev(base + end, &next, &prev);
2520 /* we've found a fitting base, insert all va's */
2521 for (area = 0; area < nr_vms; area++) {
2522 struct vmap_area *va = vas[area];
2524 va->va_start = base + offsets[area];
2525 va->va_end = va->va_start + sizes[area];
2526 __insert_vmap_area(va);
2529 vmap_area_pcpu_hole = base + offsets[last_area];
2531 spin_unlock(&vmap_area_lock);
2533 /* insert all vm's */
2534 for (area = 0; area < nr_vms; area++)
2535 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2542 for (area = 0; area < nr_vms; area++) {
2553 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2554 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2555 * @nr_vms: the number of allocated areas
2557 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2559 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2563 for (i = 0; i < nr_vms; i++)
2564 free_vm_area(vms[i]);
2567 #endif /* CONFIG_SMP */
2569 #ifdef CONFIG_PROC_FS
2570 static void *s_start(struct seq_file *m, loff_t *pos)
2571 __acquires(&vmap_area_lock)
2574 struct vmap_area *va;
2576 spin_lock(&vmap_area_lock);
2577 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2578 while (n > 0 && &va->list != &vmap_area_list) {
2580 va = list_entry(va->list.next, typeof(*va), list);
2582 if (!n && &va->list != &vmap_area_list)
2589 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2591 struct vmap_area *va = p, *next;
2594 next = list_entry(va->list.next, typeof(*va), list);
2595 if (&next->list != &vmap_area_list)
2601 static void s_stop(struct seq_file *m, void *p)
2602 __releases(&vmap_area_lock)
2604 spin_unlock(&vmap_area_lock);
2607 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2609 if (IS_ENABLED(CONFIG_NUMA)) {
2610 unsigned int nr, *counters = m->private;
2615 if (v->flags & VM_UNINITIALIZED)
2617 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2620 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2622 for (nr = 0; nr < v->nr_pages; nr++)
2623 counters[page_to_nid(v->pages[nr])]++;
2625 for_each_node_state(nr, N_HIGH_MEMORY)
2627 seq_printf(m, " N%u=%u", nr, counters[nr]);
2631 static int s_show(struct seq_file *m, void *p)
2633 struct vmap_area *va = p;
2634 struct vm_struct *v;
2637 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2638 * behalf of vmap area is being tear down or vm_map_ram allocation.
2640 if (!(va->flags & VM_VM_AREA))
2645 seq_printf(m, "0x%p-0x%p %7ld",
2646 v->addr, v->addr + v->size, v->size);
2649 seq_printf(m, " %pS", v->caller);
2652 seq_printf(m, " pages=%d", v->nr_pages);
2655 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2657 if (v->flags & VM_IOREMAP)
2658 seq_puts(m, " ioremap");
2660 if (v->flags & VM_ALLOC)
2661 seq_puts(m, " vmalloc");
2663 if (v->flags & VM_MAP)
2664 seq_puts(m, " vmap");
2666 if (v->flags & VM_USERMAP)
2667 seq_puts(m, " user");
2669 if (v->flags & VM_VPAGES)
2670 seq_puts(m, " vpages");
2672 show_numa_info(m, v);
2677 static const struct seq_operations vmalloc_op = {
2684 static int vmalloc_open(struct inode *inode, struct file *file)
2686 if (IS_ENABLED(CONFIG_NUMA))
2687 return seq_open_private(file, &vmalloc_op,
2688 nr_node_ids * sizeof(unsigned int));
2690 return seq_open(file, &vmalloc_op);
2693 static const struct file_operations proc_vmalloc_operations = {
2694 .open = vmalloc_open,
2696 .llseek = seq_lseek,
2697 .release = seq_release_private,
2700 static int __init proc_vmalloc_init(void)
2702 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2705 module_init(proc_vmalloc_init);