4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8 * Numa awareness, Christoph Lameter, SGI, June 2005
11 #include <linux/vmalloc.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/notifier.h>
25 #include <linux/rbtree.h>
26 #include <linux/radix-tree.h>
27 #include <linux/rcupdate.h>
28 #include <linux/pfn.h>
29 #include <linux/kmemleak.h>
30 #include <linux/atomic.h>
31 #include <linux/compiler.h>
32 #include <linux/llist.h>
33 #include <linux/bitops.h>
35 #include <asm/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
45 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct *w)
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
52 struct llist_node *llnode = llist_del_all(&p->list);
55 llnode = llist_next(llnode);
60 /*** Page table manipulation functions ***/
62 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
66 pte = pte_offset_kernel(pmd, addr);
68 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
69 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
70 } while (pte++, addr += PAGE_SIZE, addr != end);
73 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
78 pmd = pmd_offset(pud, addr);
80 next = pmd_addr_end(addr, end);
81 if (pmd_clear_huge(pmd))
83 if (pmd_none_or_clear_bad(pmd))
85 vunmap_pte_range(pmd, addr, next);
86 } while (pmd++, addr = next, addr != end);
89 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
94 pud = pud_offset(pgd, addr);
96 next = pud_addr_end(addr, end);
97 if (pud_clear_huge(pud))
99 if (pud_none_or_clear_bad(pud))
101 vunmap_pmd_range(pud, addr, next);
102 } while (pud++, addr = next, addr != end);
105 static void vunmap_page_range(unsigned long addr, unsigned long end)
111 pgd = pgd_offset_k(addr);
113 next = pgd_addr_end(addr, end);
114 if (pgd_none_or_clear_bad(pgd))
116 vunmap_pud_range(pgd, addr, next);
117 } while (pgd++, addr = next, addr != end);
120 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
121 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
126 * nr is a running index into the array which helps higher level
127 * callers keep track of where we're up to.
130 pte = pte_alloc_kernel(pmd, addr);
134 struct page *page = pages[*nr];
136 if (WARN_ON(!pte_none(*pte)))
140 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
142 } while (pte++, addr += PAGE_SIZE, addr != end);
146 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
147 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
152 pmd = pmd_alloc(&init_mm, pud, addr);
156 next = pmd_addr_end(addr, end);
157 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
159 } while (pmd++, addr = next, addr != end);
163 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
164 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
169 pud = pud_alloc(&init_mm, pgd, addr);
173 next = pud_addr_end(addr, end);
174 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
176 } while (pud++, addr = next, addr != end);
181 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
182 * will have pfns corresponding to the "pages" array.
184 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
186 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
187 pgprot_t prot, struct page **pages)
191 unsigned long addr = start;
196 pgd = pgd_offset_k(addr);
198 next = pgd_addr_end(addr, end);
199 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
202 } while (pgd++, addr = next, addr != end);
207 static int vmap_page_range(unsigned long start, unsigned long end,
208 pgprot_t prot, struct page **pages)
212 ret = vmap_page_range_noflush(start, end, prot, pages);
213 flush_cache_vmap(start, end);
217 int is_vmalloc_or_module_addr(const void *x)
220 * ARM, x86-64 and sparc64 put modules in a special place,
221 * and fall back on vmalloc() if that fails. Others
222 * just put it in the vmalloc space.
224 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
225 unsigned long addr = (unsigned long)x;
226 if (addr >= MODULES_VADDR && addr < MODULES_END)
229 return is_vmalloc_addr(x);
233 * Walk a vmap address to the struct page it maps.
235 struct page *vmalloc_to_page(const void *vmalloc_addr)
237 unsigned long addr = (unsigned long) vmalloc_addr;
238 struct page *page = NULL;
239 pgd_t *pgd = pgd_offset_k(addr);
242 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
243 * architectures that do not vmalloc module space
245 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
247 if (!pgd_none(*pgd)) {
248 pud_t *pud = pud_offset(pgd, addr);
249 if (!pud_none(*pud)) {
250 pmd_t *pmd = pmd_offset(pud, addr);
251 if (!pmd_none(*pmd)) {
254 ptep = pte_offset_map(pmd, addr);
256 if (pte_present(pte))
257 page = pte_page(pte);
264 EXPORT_SYMBOL(vmalloc_to_page);
267 * Map a vmalloc()-space virtual address to the physical page frame number.
269 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
271 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
273 EXPORT_SYMBOL(vmalloc_to_pfn);
276 /*** Global kva allocator ***/
278 #define VM_LAZY_FREE 0x01
279 #define VM_LAZY_FREEING 0x02
280 #define VM_VM_AREA 0x04
282 static DEFINE_SPINLOCK(vmap_area_lock);
283 /* Export for kexec only */
284 LIST_HEAD(vmap_area_list);
285 static struct rb_root vmap_area_root = RB_ROOT;
287 /* The vmap cache globals are protected by vmap_area_lock */
288 static struct rb_node *free_vmap_cache;
289 static unsigned long cached_hole_size;
290 static unsigned long cached_vstart;
291 static unsigned long cached_align;
293 static unsigned long vmap_area_pcpu_hole;
295 static struct vmap_area *__find_vmap_area(unsigned long addr)
297 struct rb_node *n = vmap_area_root.rb_node;
300 struct vmap_area *va;
302 va = rb_entry(n, struct vmap_area, rb_node);
303 if (addr < va->va_start)
305 else if (addr >= va->va_end)
314 static void __insert_vmap_area(struct vmap_area *va)
316 struct rb_node **p = &vmap_area_root.rb_node;
317 struct rb_node *parent = NULL;
321 struct vmap_area *tmp_va;
324 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
325 if (va->va_start < tmp_va->va_end)
327 else if (va->va_end > tmp_va->va_start)
333 rb_link_node(&va->rb_node, parent, p);
334 rb_insert_color(&va->rb_node, &vmap_area_root);
336 /* address-sort this list */
337 tmp = rb_prev(&va->rb_node);
339 struct vmap_area *prev;
340 prev = rb_entry(tmp, struct vmap_area, rb_node);
341 list_add_rcu(&va->list, &prev->list);
343 list_add_rcu(&va->list, &vmap_area_list);
346 static void purge_vmap_area_lazy(void);
348 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
351 * Allocate a region of KVA of the specified size and alignment, within the
354 static struct vmap_area *alloc_vmap_area(unsigned long size,
356 unsigned long vstart, unsigned long vend,
357 int node, gfp_t gfp_mask)
359 struct vmap_area *va;
363 struct vmap_area *first;
366 BUG_ON(offset_in_page(size));
367 BUG_ON(!is_power_of_2(align));
369 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
371 va = kmalloc_node(sizeof(struct vmap_area),
372 gfp_mask & GFP_RECLAIM_MASK, node);
374 return ERR_PTR(-ENOMEM);
377 * Only scan the relevant parts containing pointers to other objects
378 * to avoid false negatives.
380 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
383 spin_lock(&vmap_area_lock);
385 * Invalidate cache if we have more permissive parameters.
386 * cached_hole_size notes the largest hole noticed _below_
387 * the vmap_area cached in free_vmap_cache: if size fits
388 * into that hole, we want to scan from vstart to reuse
389 * the hole instead of allocating above free_vmap_cache.
390 * Note that __free_vmap_area may update free_vmap_cache
391 * without updating cached_hole_size or cached_align.
393 if (!free_vmap_cache ||
394 size < cached_hole_size ||
395 vstart < cached_vstart ||
396 align < cached_align) {
398 cached_hole_size = 0;
399 free_vmap_cache = NULL;
401 /* record if we encounter less permissive parameters */
402 cached_vstart = vstart;
403 cached_align = align;
405 /* find starting point for our search */
406 if (free_vmap_cache) {
407 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
408 addr = ALIGN(first->va_end, align);
411 if (addr + size < addr)
415 addr = ALIGN(vstart, align);
416 if (addr + size < addr)
419 n = vmap_area_root.rb_node;
423 struct vmap_area *tmp;
424 tmp = rb_entry(n, struct vmap_area, rb_node);
425 if (tmp->va_end >= addr) {
427 if (tmp->va_start <= addr)
438 /* from the starting point, walk areas until a suitable hole is found */
439 while (addr + size > first->va_start && addr + size <= vend) {
440 if (addr + cached_hole_size < first->va_start)
441 cached_hole_size = first->va_start - addr;
442 addr = ALIGN(first->va_end, align);
443 if (addr + size < addr)
446 if (list_is_last(&first->list, &vmap_area_list))
449 first = list_next_entry(first, list);
453 if (addr + size > vend)
457 va->va_end = addr + size;
459 __insert_vmap_area(va);
460 free_vmap_cache = &va->rb_node;
461 spin_unlock(&vmap_area_lock);
463 BUG_ON(!IS_ALIGNED(va->va_start, align));
464 BUG_ON(va->va_start < vstart);
465 BUG_ON(va->va_end > vend);
470 spin_unlock(&vmap_area_lock);
472 purge_vmap_area_lazy();
477 if (gfpflags_allow_blocking(gfp_mask)) {
478 unsigned long freed = 0;
479 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
486 if (printk_ratelimit())
487 pr_warn("vmap allocation for size %lu failed: "
488 "use vmalloc=<size> to increase size.\n", size);
490 return ERR_PTR(-EBUSY);
493 int register_vmap_purge_notifier(struct notifier_block *nb)
495 return blocking_notifier_chain_register(&vmap_notify_list, nb);
497 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
499 int unregister_vmap_purge_notifier(struct notifier_block *nb)
501 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
503 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
505 static void __free_vmap_area(struct vmap_area *va)
507 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
509 if (free_vmap_cache) {
510 if (va->va_end < cached_vstart) {
511 free_vmap_cache = NULL;
513 struct vmap_area *cache;
514 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
515 if (va->va_start <= cache->va_start) {
516 free_vmap_cache = rb_prev(&va->rb_node);
518 * We don't try to update cached_hole_size or
519 * cached_align, but it won't go very wrong.
524 rb_erase(&va->rb_node, &vmap_area_root);
525 RB_CLEAR_NODE(&va->rb_node);
526 list_del_rcu(&va->list);
529 * Track the highest possible candidate for pcpu area
530 * allocation. Areas outside of vmalloc area can be returned
531 * here too, consider only end addresses which fall inside
532 * vmalloc area proper.
534 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
535 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
537 kfree_rcu(va, rcu_head);
541 * Free a region of KVA allocated by alloc_vmap_area
543 static void free_vmap_area(struct vmap_area *va)
545 spin_lock(&vmap_area_lock);
546 __free_vmap_area(va);
547 spin_unlock(&vmap_area_lock);
551 * Clear the pagetable entries of a given vmap_area
553 static void unmap_vmap_area(struct vmap_area *va)
555 vunmap_page_range(va->va_start, va->va_end);
558 static void vmap_debug_free_range(unsigned long start, unsigned long end)
561 * Unmap page tables and force a TLB flush immediately if
562 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
563 * bugs similarly to those in linear kernel virtual address
564 * space after a page has been freed.
566 * All the lazy freeing logic is still retained, in order to
567 * minimise intrusiveness of this debugging feature.
569 * This is going to be *slow* (linear kernel virtual address
570 * debugging doesn't do a broadcast TLB flush so it is a lot
573 #ifdef CONFIG_DEBUG_PAGEALLOC
574 vunmap_page_range(start, end);
575 flush_tlb_kernel_range(start, end);
580 * lazy_max_pages is the maximum amount of virtual address space we gather up
581 * before attempting to purge with a TLB flush.
583 * There is a tradeoff here: a larger number will cover more kernel page tables
584 * and take slightly longer to purge, but it will linearly reduce the number of
585 * global TLB flushes that must be performed. It would seem natural to scale
586 * this number up linearly with the number of CPUs (because vmapping activity
587 * could also scale linearly with the number of CPUs), however it is likely
588 * that in practice, workloads might be constrained in other ways that mean
589 * vmap activity will not scale linearly with CPUs. Also, I want to be
590 * conservative and not introduce a big latency on huge systems, so go with
591 * a less aggressive log scale. It will still be an improvement over the old
592 * code, and it will be simple to change the scale factor if we find that it
593 * becomes a problem on bigger systems.
595 static unsigned long lazy_max_pages(void)
599 log = fls(num_online_cpus());
601 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
604 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
606 /* for per-CPU blocks */
607 static void purge_fragmented_blocks_allcpus(void);
610 * called before a call to iounmap() if the caller wants vm_area_struct's
613 void set_iounmap_nonlazy(void)
615 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
619 * Purges all lazily-freed vmap areas.
621 * If sync is 0 then don't purge if there is already a purge in progress.
622 * If force_flush is 1, then flush kernel TLBs between *start and *end even
623 * if we found no lazy vmap areas to unmap (callers can use this to optimise
624 * their own TLB flushing).
625 * Returns with *start = min(*start, lowest purged address)
626 * *end = max(*end, highest purged address)
628 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
629 int sync, int force_flush)
631 static DEFINE_SPINLOCK(purge_lock);
633 struct vmap_area *va;
634 struct vmap_area *n_va;
638 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
639 * should not expect such behaviour. This just simplifies locking for
640 * the case that isn't actually used at the moment anyway.
642 if (!sync && !force_flush) {
643 if (!spin_trylock(&purge_lock))
646 spin_lock(&purge_lock);
649 purge_fragmented_blocks_allcpus();
652 list_for_each_entry_rcu(va, &vmap_area_list, list) {
653 if (va->flags & VM_LAZY_FREE) {
654 if (va->va_start < *start)
655 *start = va->va_start;
656 if (va->va_end > *end)
658 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
659 list_add_tail(&va->purge_list, &valist);
660 va->flags |= VM_LAZY_FREEING;
661 va->flags &= ~VM_LAZY_FREE;
667 atomic_sub(nr, &vmap_lazy_nr);
669 if (nr || force_flush)
670 flush_tlb_kernel_range(*start, *end);
673 spin_lock(&vmap_area_lock);
674 list_for_each_entry_safe(va, n_va, &valist, purge_list)
675 __free_vmap_area(va);
676 spin_unlock(&vmap_area_lock);
678 spin_unlock(&purge_lock);
682 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
683 * is already purging.
685 static void try_purge_vmap_area_lazy(void)
687 unsigned long start = ULONG_MAX, end = 0;
689 __purge_vmap_area_lazy(&start, &end, 0, 0);
693 * Kick off a purge of the outstanding lazy areas.
695 static void purge_vmap_area_lazy(void)
697 unsigned long start = ULONG_MAX, end = 0;
699 __purge_vmap_area_lazy(&start, &end, 1, 0);
703 * Free a vmap area, caller ensuring that the area has been unmapped
704 * and flush_cache_vunmap had been called for the correct range
707 static void free_vmap_area_noflush(struct vmap_area *va)
709 va->flags |= VM_LAZY_FREE;
710 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
711 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
712 try_purge_vmap_area_lazy();
716 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
717 * called for the correct range previously.
719 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
722 free_vmap_area_noflush(va);
726 * Free and unmap a vmap area
728 static void free_unmap_vmap_area(struct vmap_area *va)
730 flush_cache_vunmap(va->va_start, va->va_end);
731 free_unmap_vmap_area_noflush(va);
734 static struct vmap_area *find_vmap_area(unsigned long addr)
736 struct vmap_area *va;
738 spin_lock(&vmap_area_lock);
739 va = __find_vmap_area(addr);
740 spin_unlock(&vmap_area_lock);
745 static void free_unmap_vmap_area_addr(unsigned long addr)
747 struct vmap_area *va;
749 va = find_vmap_area(addr);
751 free_unmap_vmap_area(va);
755 /*** Per cpu kva allocator ***/
758 * vmap space is limited especially on 32 bit architectures. Ensure there is
759 * room for at least 16 percpu vmap blocks per CPU.
762 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
763 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
764 * instead (we just need a rough idea)
766 #if BITS_PER_LONG == 32
767 #define VMALLOC_SPACE (128UL*1024*1024)
769 #define VMALLOC_SPACE (128UL*1024*1024*1024)
772 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
773 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
774 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
775 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
776 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
777 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
778 #define VMAP_BBMAP_BITS \
779 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
780 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
781 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
783 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
785 static bool vmap_initialized __read_mostly = false;
787 struct vmap_block_queue {
789 struct list_head free;
794 struct vmap_area *va;
795 unsigned long free, dirty;
796 unsigned long dirty_min, dirty_max; /*< dirty range */
797 struct list_head free_list;
798 struct rcu_head rcu_head;
799 struct list_head purge;
802 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
803 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
806 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
807 * in the free path. Could get rid of this if we change the API to return a
808 * "cookie" from alloc, to be passed to free. But no big deal yet.
810 static DEFINE_SPINLOCK(vmap_block_tree_lock);
811 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
814 * We should probably have a fallback mechanism to allocate virtual memory
815 * out of partially filled vmap blocks. However vmap block sizing should be
816 * fairly reasonable according to the vmalloc size, so it shouldn't be a
820 static unsigned long addr_to_vb_idx(unsigned long addr)
822 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
823 addr /= VMAP_BLOCK_SIZE;
827 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
831 addr = va_start + (pages_off << PAGE_SHIFT);
832 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
837 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
838 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
839 * @order: how many 2^order pages should be occupied in newly allocated block
840 * @gfp_mask: flags for the page level allocator
842 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
844 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
846 struct vmap_block_queue *vbq;
847 struct vmap_block *vb;
848 struct vmap_area *va;
849 unsigned long vb_idx;
853 node = numa_node_id();
855 vb = kmalloc_node(sizeof(struct vmap_block),
856 gfp_mask & GFP_RECLAIM_MASK, node);
858 return ERR_PTR(-ENOMEM);
860 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
861 VMALLOC_START, VMALLOC_END,
868 err = radix_tree_preload(gfp_mask);
875 vaddr = vmap_block_vaddr(va->va_start, 0);
876 spin_lock_init(&vb->lock);
878 /* At least something should be left free */
879 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
880 vb->free = VMAP_BBMAP_BITS - (1UL << order);
882 vb->dirty_min = VMAP_BBMAP_BITS;
884 INIT_LIST_HEAD(&vb->free_list);
886 vb_idx = addr_to_vb_idx(va->va_start);
887 spin_lock(&vmap_block_tree_lock);
888 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
889 spin_unlock(&vmap_block_tree_lock);
891 radix_tree_preload_end();
893 vbq = &get_cpu_var(vmap_block_queue);
894 spin_lock(&vbq->lock);
895 list_add_tail_rcu(&vb->free_list, &vbq->free);
896 spin_unlock(&vbq->lock);
897 put_cpu_var(vmap_block_queue);
902 static void free_vmap_block(struct vmap_block *vb)
904 struct vmap_block *tmp;
905 unsigned long vb_idx;
907 vb_idx = addr_to_vb_idx(vb->va->va_start);
908 spin_lock(&vmap_block_tree_lock);
909 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
910 spin_unlock(&vmap_block_tree_lock);
913 free_vmap_area_noflush(vb->va);
914 kfree_rcu(vb, rcu_head);
917 static void purge_fragmented_blocks(int cpu)
920 struct vmap_block *vb;
921 struct vmap_block *n_vb;
922 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
925 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
927 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
930 spin_lock(&vb->lock);
931 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
932 vb->free = 0; /* prevent further allocs after releasing lock */
933 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
935 vb->dirty_max = VMAP_BBMAP_BITS;
936 spin_lock(&vbq->lock);
937 list_del_rcu(&vb->free_list);
938 spin_unlock(&vbq->lock);
939 spin_unlock(&vb->lock);
940 list_add_tail(&vb->purge, &purge);
942 spin_unlock(&vb->lock);
946 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
947 list_del(&vb->purge);
952 static void purge_fragmented_blocks_allcpus(void)
956 for_each_possible_cpu(cpu)
957 purge_fragmented_blocks(cpu);
960 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
962 struct vmap_block_queue *vbq;
963 struct vmap_block *vb;
967 BUG_ON(offset_in_page(size));
968 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
969 if (WARN_ON(size == 0)) {
971 * Allocating 0 bytes isn't what caller wants since
972 * get_order(0) returns funny result. Just warn and terminate
977 order = get_order(size);
980 vbq = &get_cpu_var(vmap_block_queue);
981 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
982 unsigned long pages_off;
984 spin_lock(&vb->lock);
985 if (vb->free < (1UL << order)) {
986 spin_unlock(&vb->lock);
990 pages_off = VMAP_BBMAP_BITS - vb->free;
991 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
992 vb->free -= 1UL << order;
994 spin_lock(&vbq->lock);
995 list_del_rcu(&vb->free_list);
996 spin_unlock(&vbq->lock);
999 spin_unlock(&vb->lock);
1003 put_cpu_var(vmap_block_queue);
1006 /* Allocate new block if nothing was found */
1008 vaddr = new_vmap_block(order, gfp_mask);
1013 static void vb_free(const void *addr, unsigned long size)
1015 unsigned long offset;
1016 unsigned long vb_idx;
1018 struct vmap_block *vb;
1020 BUG_ON(offset_in_page(size));
1021 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1023 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1025 order = get_order(size);
1027 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1028 offset >>= PAGE_SHIFT;
1030 vb_idx = addr_to_vb_idx((unsigned long)addr);
1032 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1036 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1038 spin_lock(&vb->lock);
1040 /* Expand dirty range */
1041 vb->dirty_min = min(vb->dirty_min, offset);
1042 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1044 vb->dirty += 1UL << order;
1045 if (vb->dirty == VMAP_BBMAP_BITS) {
1047 spin_unlock(&vb->lock);
1048 free_vmap_block(vb);
1050 spin_unlock(&vb->lock);
1054 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1056 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1057 * to amortize TLB flushing overheads. What this means is that any page you
1058 * have now, may, in a former life, have been mapped into kernel virtual
1059 * address by the vmap layer and so there might be some CPUs with TLB entries
1060 * still referencing that page (additional to the regular 1:1 kernel mapping).
1062 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1063 * be sure that none of the pages we have control over will have any aliases
1064 * from the vmap layer.
1066 void vm_unmap_aliases(void)
1068 unsigned long start = ULONG_MAX, end = 0;
1072 if (unlikely(!vmap_initialized))
1075 for_each_possible_cpu(cpu) {
1076 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1077 struct vmap_block *vb;
1080 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1081 spin_lock(&vb->lock);
1083 unsigned long va_start = vb->va->va_start;
1086 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1087 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1089 start = min(s, start);
1094 spin_unlock(&vb->lock);
1099 __purge_vmap_area_lazy(&start, &end, 1, flush);
1101 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1104 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1105 * @mem: the pointer returned by vm_map_ram
1106 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1108 void vm_unmap_ram(const void *mem, unsigned int count)
1110 unsigned long size = count << PAGE_SHIFT;
1111 unsigned long addr = (unsigned long)mem;
1114 BUG_ON(addr < VMALLOC_START);
1115 BUG_ON(addr > VMALLOC_END);
1116 BUG_ON(!IS_ALIGNED(addr, PAGE_SIZE));
1118 debug_check_no_locks_freed(mem, size);
1119 vmap_debug_free_range(addr, addr+size);
1121 if (likely(count <= VMAP_MAX_ALLOC))
1124 free_unmap_vmap_area_addr(addr);
1126 EXPORT_SYMBOL(vm_unmap_ram);
1129 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1130 * @pages: an array of pointers to the pages to be mapped
1131 * @count: number of pages
1132 * @node: prefer to allocate data structures on this node
1133 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1135 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1136 * faster than vmap so it's good. But if you mix long-life and short-life
1137 * objects with vm_map_ram(), it could consume lots of address space through
1138 * fragmentation (especially on a 32bit machine). You could see failures in
1139 * the end. Please use this function for short-lived objects.
1141 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1143 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1145 unsigned long size = count << PAGE_SHIFT;
1149 if (likely(count <= VMAP_MAX_ALLOC)) {
1150 mem = vb_alloc(size, GFP_KERNEL);
1153 addr = (unsigned long)mem;
1155 struct vmap_area *va;
1156 va = alloc_vmap_area(size, PAGE_SIZE,
1157 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1161 addr = va->va_start;
1164 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1165 vm_unmap_ram(mem, count);
1170 EXPORT_SYMBOL(vm_map_ram);
1172 static struct vm_struct *vmlist __initdata;
1174 * vm_area_add_early - add vmap area early during boot
1175 * @vm: vm_struct to add
1177 * This function is used to add fixed kernel vm area to vmlist before
1178 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1179 * should contain proper values and the other fields should be zero.
1181 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1183 void __init vm_area_add_early(struct vm_struct *vm)
1185 struct vm_struct *tmp, **p;
1187 BUG_ON(vmap_initialized);
1188 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1189 if (tmp->addr >= vm->addr) {
1190 BUG_ON(tmp->addr < vm->addr + vm->size);
1193 BUG_ON(tmp->addr + tmp->size > vm->addr);
1200 * vm_area_register_early - register vmap area early during boot
1201 * @vm: vm_struct to register
1202 * @align: requested alignment
1204 * This function is used to register kernel vm area before
1205 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1206 * proper values on entry and other fields should be zero. On return,
1207 * vm->addr contains the allocated address.
1209 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1211 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1213 static size_t vm_init_off __initdata;
1216 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1217 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1219 vm->addr = (void *)addr;
1221 vm_area_add_early(vm);
1224 void __init vmalloc_init(void)
1226 struct vmap_area *va;
1227 struct vm_struct *tmp;
1230 for_each_possible_cpu(i) {
1231 struct vmap_block_queue *vbq;
1232 struct vfree_deferred *p;
1234 vbq = &per_cpu(vmap_block_queue, i);
1235 spin_lock_init(&vbq->lock);
1236 INIT_LIST_HEAD(&vbq->free);
1237 p = &per_cpu(vfree_deferred, i);
1238 init_llist_head(&p->list);
1239 INIT_WORK(&p->wq, free_work);
1242 /* Import existing vmlist entries. */
1243 for (tmp = vmlist; tmp; tmp = tmp->next) {
1244 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1245 va->flags = VM_VM_AREA;
1246 va->va_start = (unsigned long)tmp->addr;
1247 va->va_end = va->va_start + tmp->size;
1249 __insert_vmap_area(va);
1252 vmap_area_pcpu_hole = VMALLOC_END;
1254 vmap_initialized = true;
1258 * map_kernel_range_noflush - map kernel VM area with the specified pages
1259 * @addr: start of the VM area to map
1260 * @size: size of the VM area to map
1261 * @prot: page protection flags to use
1262 * @pages: pages to map
1264 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1265 * specify should have been allocated using get_vm_area() and its
1269 * This function does NOT do any cache flushing. The caller is
1270 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1271 * before calling this function.
1274 * The number of pages mapped on success, -errno on failure.
1276 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1277 pgprot_t prot, struct page **pages)
1279 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1283 * unmap_kernel_range_noflush - unmap kernel VM area
1284 * @addr: start of the VM area to unmap
1285 * @size: size of the VM area to unmap
1287 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1288 * specify should have been allocated using get_vm_area() and its
1292 * This function does NOT do any cache flushing. The caller is
1293 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1294 * before calling this function and flush_tlb_kernel_range() after.
1296 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1298 vunmap_page_range(addr, addr + size);
1300 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1303 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1304 * @addr: start of the VM area to unmap
1305 * @size: size of the VM area to unmap
1307 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1308 * the unmapping and tlb after.
1310 void unmap_kernel_range(unsigned long addr, unsigned long size)
1312 unsigned long end = addr + size;
1314 flush_cache_vunmap(addr, end);
1315 vunmap_page_range(addr, end);
1316 flush_tlb_kernel_range(addr, end);
1318 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1320 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1322 unsigned long addr = (unsigned long)area->addr;
1323 unsigned long end = addr + get_vm_area_size(area);
1326 err = vmap_page_range(addr, end, prot, pages);
1328 return err > 0 ? 0 : err;
1330 EXPORT_SYMBOL_GPL(map_vm_area);
1332 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1333 unsigned long flags, const void *caller)
1335 spin_lock(&vmap_area_lock);
1337 vm->addr = (void *)va->va_start;
1338 vm->size = va->va_end - va->va_start;
1339 vm->caller = caller;
1341 va->flags |= VM_VM_AREA;
1342 spin_unlock(&vmap_area_lock);
1345 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1348 * Before removing VM_UNINITIALIZED,
1349 * we should make sure that vm has proper values.
1350 * Pair with smp_rmb() in show_numa_info().
1353 vm->flags &= ~VM_UNINITIALIZED;
1356 static struct vm_struct *__get_vm_area_node(unsigned long size,
1357 unsigned long align, unsigned long flags, unsigned long start,
1358 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1360 struct vmap_area *va;
1361 struct vm_struct *area;
1363 BUG_ON(in_interrupt());
1364 if (flags & VM_IOREMAP)
1365 align = 1ul << clamp_t(int, fls_long(size),
1366 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1368 size = PAGE_ALIGN(size);
1369 if (unlikely(!size))
1372 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1373 if (unlikely(!area))
1376 if (!(flags & VM_NO_GUARD))
1379 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1385 setup_vmalloc_vm(area, va, flags, caller);
1390 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1391 unsigned long start, unsigned long end)
1393 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1394 GFP_KERNEL, __builtin_return_address(0));
1396 EXPORT_SYMBOL_GPL(__get_vm_area);
1398 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1399 unsigned long start, unsigned long end,
1402 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1403 GFP_KERNEL, caller);
1407 * get_vm_area - reserve a contiguous kernel virtual area
1408 * @size: size of the area
1409 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1411 * Search an area of @size in the kernel virtual mapping area,
1412 * and reserved it for out purposes. Returns the area descriptor
1413 * on success or %NULL on failure.
1415 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1417 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1418 NUMA_NO_NODE, GFP_KERNEL,
1419 __builtin_return_address(0));
1422 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1425 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1426 NUMA_NO_NODE, GFP_KERNEL, caller);
1430 * find_vm_area - find a continuous kernel virtual area
1431 * @addr: base address
1433 * Search for the kernel VM area starting at @addr, and return it.
1434 * It is up to the caller to do all required locking to keep the returned
1437 struct vm_struct *find_vm_area(const void *addr)
1439 struct vmap_area *va;
1441 va = find_vmap_area((unsigned long)addr);
1442 if (va && va->flags & VM_VM_AREA)
1449 * remove_vm_area - find and remove a continuous kernel virtual area
1450 * @addr: base address
1452 * Search for the kernel VM area starting at @addr, and remove it.
1453 * This function returns the found VM area, but using it is NOT safe
1454 * on SMP machines, except for its size or flags.
1456 struct vm_struct *remove_vm_area(const void *addr)
1458 struct vmap_area *va;
1460 va = find_vmap_area((unsigned long)addr);
1461 if (va && va->flags & VM_VM_AREA) {
1462 struct vm_struct *vm = va->vm;
1464 spin_lock(&vmap_area_lock);
1466 va->flags &= ~VM_VM_AREA;
1467 spin_unlock(&vmap_area_lock);
1469 vmap_debug_free_range(va->va_start, va->va_end);
1470 kasan_free_shadow(vm);
1471 free_unmap_vmap_area(va);
1478 static void __vunmap(const void *addr, int deallocate_pages)
1480 struct vm_struct *area;
1485 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1489 area = remove_vm_area(addr);
1490 if (unlikely(!area)) {
1491 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1496 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1497 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1499 if (deallocate_pages) {
1502 for (i = 0; i < area->nr_pages; i++) {
1503 struct page *page = area->pages[i];
1506 __free_kmem_pages(page, 0);
1509 kvfree(area->pages);
1517 * vfree - release memory allocated by vmalloc()
1518 * @addr: memory base address
1520 * Free the virtually continuous memory area starting at @addr, as
1521 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1522 * NULL, no operation is performed.
1524 * Must not be called in NMI context (strictly speaking, only if we don't
1525 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1526 * conventions for vfree() arch-depenedent would be a really bad idea)
1528 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1530 void vfree(const void *addr)
1534 kmemleak_free(addr);
1538 if (unlikely(in_interrupt())) {
1539 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1540 if (llist_add((struct llist_node *)addr, &p->list))
1541 schedule_work(&p->wq);
1545 EXPORT_SYMBOL(vfree);
1548 * vunmap - release virtual mapping obtained by vmap()
1549 * @addr: memory base address
1551 * Free the virtually contiguous memory area starting at @addr,
1552 * which was created from the page array passed to vmap().
1554 * Must not be called in interrupt context.
1556 void vunmap(const void *addr)
1558 BUG_ON(in_interrupt());
1563 EXPORT_SYMBOL(vunmap);
1566 * vmap - map an array of pages into virtually contiguous space
1567 * @pages: array of page pointers
1568 * @count: number of pages to map
1569 * @flags: vm_area->flags
1570 * @prot: page protection for the mapping
1572 * Maps @count pages from @pages into contiguous kernel virtual
1575 void *vmap(struct page **pages, unsigned int count,
1576 unsigned long flags, pgprot_t prot)
1578 struct vm_struct *area;
1582 if (count > totalram_pages)
1585 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1586 __builtin_return_address(0));
1590 if (map_vm_area(area, prot, pages)) {
1597 EXPORT_SYMBOL(vmap);
1599 static void *__vmalloc_node(unsigned long size, unsigned long align,
1600 gfp_t gfp_mask, pgprot_t prot,
1601 int node, const void *caller);
1602 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1603 pgprot_t prot, int node)
1605 const int order = 0;
1606 struct page **pages;
1607 unsigned int nr_pages, array_size, i;
1608 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1609 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1611 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1612 array_size = (nr_pages * sizeof(struct page *));
1614 area->nr_pages = nr_pages;
1615 /* Please note that the recursion is strictly bounded. */
1616 if (array_size > PAGE_SIZE) {
1617 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1618 PAGE_KERNEL, node, area->caller);
1620 pages = kmalloc_node(array_size, nested_gfp, node);
1622 area->pages = pages;
1624 remove_vm_area(area->addr);
1629 for (i = 0; i < area->nr_pages; i++) {
1632 if (node == NUMA_NO_NODE)
1633 page = alloc_kmem_pages(alloc_mask, order);
1635 page = alloc_kmem_pages_node(node, alloc_mask, order);
1637 if (unlikely(!page)) {
1638 /* Successfully allocated i pages, free them in __vunmap() */
1642 area->pages[i] = page;
1643 if (gfpflags_allow_blocking(gfp_mask))
1647 if (map_vm_area(area, prot, pages))
1652 warn_alloc_failed(gfp_mask, order,
1653 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1654 (area->nr_pages*PAGE_SIZE), area->size);
1660 * __vmalloc_node_range - allocate virtually contiguous memory
1661 * @size: allocation size
1662 * @align: desired alignment
1663 * @start: vm area range start
1664 * @end: vm area range end
1665 * @gfp_mask: flags for the page level allocator
1666 * @prot: protection mask for the allocated pages
1667 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1668 * @node: node to use for allocation or NUMA_NO_NODE
1669 * @caller: caller's return address
1671 * Allocate enough pages to cover @size from the page level
1672 * allocator with @gfp_mask flags. Map them into contiguous
1673 * kernel virtual space, using a pagetable protection of @prot.
1675 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1676 unsigned long start, unsigned long end, gfp_t gfp_mask,
1677 pgprot_t prot, unsigned long vm_flags, int node,
1680 struct vm_struct *area;
1682 unsigned long real_size = size;
1684 size = PAGE_ALIGN(size);
1685 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1688 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1689 vm_flags, start, end, node, gfp_mask, caller);
1693 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1698 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1699 * flag. It means that vm_struct is not fully initialized.
1700 * Now, it is fully initialized, so remove this flag here.
1702 clear_vm_uninitialized_flag(area);
1705 * A ref_count = 2 is needed because vm_struct allocated in
1706 * __get_vm_area_node() contains a reference to the virtual address of
1707 * the vmalloc'ed block.
1709 kmemleak_alloc(addr, real_size, 2, gfp_mask);
1714 warn_alloc_failed(gfp_mask, 0,
1715 "vmalloc: allocation failure: %lu bytes\n",
1721 * __vmalloc_node - allocate virtually contiguous memory
1722 * @size: allocation size
1723 * @align: desired alignment
1724 * @gfp_mask: flags for the page level allocator
1725 * @prot: protection mask for the allocated pages
1726 * @node: node to use for allocation or NUMA_NO_NODE
1727 * @caller: caller's return address
1729 * Allocate enough pages to cover @size from the page level
1730 * allocator with @gfp_mask flags. Map them into contiguous
1731 * kernel virtual space, using a pagetable protection of @prot.
1733 static void *__vmalloc_node(unsigned long size, unsigned long align,
1734 gfp_t gfp_mask, pgprot_t prot,
1735 int node, const void *caller)
1737 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1738 gfp_mask, prot, 0, node, caller);
1741 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1743 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1744 __builtin_return_address(0));
1746 EXPORT_SYMBOL(__vmalloc);
1748 static inline void *__vmalloc_node_flags(unsigned long size,
1749 int node, gfp_t flags)
1751 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1752 node, __builtin_return_address(0));
1756 * vmalloc - allocate virtually contiguous memory
1757 * @size: allocation size
1758 * Allocate enough pages to cover @size from the page level
1759 * allocator and map them into contiguous kernel virtual space.
1761 * For tight control over page level allocator and protection flags
1762 * use __vmalloc() instead.
1764 void *vmalloc(unsigned long size)
1766 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1767 GFP_KERNEL | __GFP_HIGHMEM);
1769 EXPORT_SYMBOL(vmalloc);
1772 * vzalloc - allocate virtually contiguous memory with zero fill
1773 * @size: allocation size
1774 * Allocate enough pages to cover @size from the page level
1775 * allocator and map them into contiguous kernel virtual space.
1776 * The memory allocated is set to zero.
1778 * For tight control over page level allocator and protection flags
1779 * use __vmalloc() instead.
1781 void *vzalloc(unsigned long size)
1783 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1784 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1786 EXPORT_SYMBOL(vzalloc);
1789 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1790 * @size: allocation size
1792 * The resulting memory area is zeroed so it can be mapped to userspace
1793 * without leaking data.
1795 void *vmalloc_user(unsigned long size)
1797 struct vm_struct *area;
1800 ret = __vmalloc_node(size, SHMLBA,
1801 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1802 PAGE_KERNEL, NUMA_NO_NODE,
1803 __builtin_return_address(0));
1805 area = find_vm_area(ret);
1806 area->flags |= VM_USERMAP;
1810 EXPORT_SYMBOL(vmalloc_user);
1813 * vmalloc_node - allocate memory on a specific node
1814 * @size: allocation size
1817 * Allocate enough pages to cover @size from the page level
1818 * allocator and map them into contiguous kernel virtual space.
1820 * For tight control over page level allocator and protection flags
1821 * use __vmalloc() instead.
1823 void *vmalloc_node(unsigned long size, int node)
1825 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1826 node, __builtin_return_address(0));
1828 EXPORT_SYMBOL(vmalloc_node);
1831 * vzalloc_node - allocate memory on a specific node with zero fill
1832 * @size: allocation size
1835 * Allocate enough pages to cover @size from the page level
1836 * allocator and map them into contiguous kernel virtual space.
1837 * The memory allocated is set to zero.
1839 * For tight control over page level allocator and protection flags
1840 * use __vmalloc_node() instead.
1842 void *vzalloc_node(unsigned long size, int node)
1844 return __vmalloc_node_flags(size, node,
1845 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1847 EXPORT_SYMBOL(vzalloc_node);
1849 #ifndef PAGE_KERNEL_EXEC
1850 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1854 * vmalloc_exec - allocate virtually contiguous, executable memory
1855 * @size: allocation size
1857 * Kernel-internal function to allocate enough pages to cover @size
1858 * the page level allocator and map them into contiguous and
1859 * executable kernel virtual space.
1861 * For tight control over page level allocator and protection flags
1862 * use __vmalloc() instead.
1865 void *vmalloc_exec(unsigned long size)
1867 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1868 NUMA_NO_NODE, __builtin_return_address(0));
1871 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1872 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1873 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1874 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1876 #define GFP_VMALLOC32 GFP_KERNEL
1880 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1881 * @size: allocation size
1883 * Allocate enough 32bit PA addressable pages to cover @size from the
1884 * page level allocator and map them into contiguous kernel virtual space.
1886 void *vmalloc_32(unsigned long size)
1888 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1889 NUMA_NO_NODE, __builtin_return_address(0));
1891 EXPORT_SYMBOL(vmalloc_32);
1894 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1895 * @size: allocation size
1897 * The resulting memory area is 32bit addressable and zeroed so it can be
1898 * mapped to userspace without leaking data.
1900 void *vmalloc_32_user(unsigned long size)
1902 struct vm_struct *area;
1905 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1906 NUMA_NO_NODE, __builtin_return_address(0));
1908 area = find_vm_area(ret);
1909 area->flags |= VM_USERMAP;
1913 EXPORT_SYMBOL(vmalloc_32_user);
1916 * small helper routine , copy contents to buf from addr.
1917 * If the page is not present, fill zero.
1920 static int aligned_vread(char *buf, char *addr, unsigned long count)
1926 unsigned long offset, length;
1928 offset = offset_in_page(addr);
1929 length = PAGE_SIZE - offset;
1932 p = vmalloc_to_page(addr);
1934 * To do safe access to this _mapped_ area, we need
1935 * lock. But adding lock here means that we need to add
1936 * overhead of vmalloc()/vfree() calles for this _debug_
1937 * interface, rarely used. Instead of that, we'll use
1938 * kmap() and get small overhead in this access function.
1942 * we can expect USER0 is not used (see vread/vwrite's
1943 * function description)
1945 void *map = kmap_atomic(p);
1946 memcpy(buf, map + offset, length);
1949 memset(buf, 0, length);
1959 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1965 unsigned long offset, length;
1967 offset = offset_in_page(addr);
1968 length = PAGE_SIZE - offset;
1971 p = vmalloc_to_page(addr);
1973 * To do safe access to this _mapped_ area, we need
1974 * lock. But adding lock here means that we need to add
1975 * overhead of vmalloc()/vfree() calles for this _debug_
1976 * interface, rarely used. Instead of that, we'll use
1977 * kmap() and get small overhead in this access function.
1981 * we can expect USER0 is not used (see vread/vwrite's
1982 * function description)
1984 void *map = kmap_atomic(p);
1985 memcpy(map + offset, buf, length);
1997 * vread() - read vmalloc area in a safe way.
1998 * @buf: buffer for reading data
1999 * @addr: vm address.
2000 * @count: number of bytes to be read.
2002 * Returns # of bytes which addr and buf should be increased.
2003 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2004 * includes any intersect with alive vmalloc area.
2006 * This function checks that addr is a valid vmalloc'ed area, and
2007 * copy data from that area to a given buffer. If the given memory range
2008 * of [addr...addr+count) includes some valid address, data is copied to
2009 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2010 * IOREMAP area is treated as memory hole and no copy is done.
2012 * If [addr...addr+count) doesn't includes any intersects with alive
2013 * vm_struct area, returns 0. @buf should be kernel's buffer.
2015 * Note: In usual ops, vread() is never necessary because the caller
2016 * should know vmalloc() area is valid and can use memcpy().
2017 * This is for routines which have to access vmalloc area without
2018 * any informaion, as /dev/kmem.
2022 long vread(char *buf, char *addr, unsigned long count)
2024 struct vmap_area *va;
2025 struct vm_struct *vm;
2026 char *vaddr, *buf_start = buf;
2027 unsigned long buflen = count;
2030 /* Don't allow overflow */
2031 if ((unsigned long) addr + count < count)
2032 count = -(unsigned long) addr;
2034 spin_lock(&vmap_area_lock);
2035 list_for_each_entry(va, &vmap_area_list, list) {
2039 if (!(va->flags & VM_VM_AREA))
2043 vaddr = (char *) vm->addr;
2044 if (addr >= vaddr + get_vm_area_size(vm))
2046 while (addr < vaddr) {
2054 n = vaddr + get_vm_area_size(vm) - addr;
2057 if (!(vm->flags & VM_IOREMAP))
2058 aligned_vread(buf, addr, n);
2059 else /* IOREMAP area is treated as memory hole */
2066 spin_unlock(&vmap_area_lock);
2068 if (buf == buf_start)
2070 /* zero-fill memory holes */
2071 if (buf != buf_start + buflen)
2072 memset(buf, 0, buflen - (buf - buf_start));
2078 * vwrite() - write vmalloc area in a safe way.
2079 * @buf: buffer for source data
2080 * @addr: vm address.
2081 * @count: number of bytes to be read.
2083 * Returns # of bytes which addr and buf should be incresed.
2084 * (same number to @count).
2085 * If [addr...addr+count) doesn't includes any intersect with valid
2086 * vmalloc area, returns 0.
2088 * This function checks that addr is a valid vmalloc'ed area, and
2089 * copy data from a buffer to the given addr. If specified range of
2090 * [addr...addr+count) includes some valid address, data is copied from
2091 * proper area of @buf. If there are memory holes, no copy to hole.
2092 * IOREMAP area is treated as memory hole and no copy is done.
2094 * If [addr...addr+count) doesn't includes any intersects with alive
2095 * vm_struct area, returns 0. @buf should be kernel's buffer.
2097 * Note: In usual ops, vwrite() is never necessary because the caller
2098 * should know vmalloc() area is valid and can use memcpy().
2099 * This is for routines which have to access vmalloc area without
2100 * any informaion, as /dev/kmem.
2103 long vwrite(char *buf, char *addr, unsigned long count)
2105 struct vmap_area *va;
2106 struct vm_struct *vm;
2108 unsigned long n, buflen;
2111 /* Don't allow overflow */
2112 if ((unsigned long) addr + count < count)
2113 count = -(unsigned long) addr;
2116 spin_lock(&vmap_area_lock);
2117 list_for_each_entry(va, &vmap_area_list, list) {
2121 if (!(va->flags & VM_VM_AREA))
2125 vaddr = (char *) vm->addr;
2126 if (addr >= vaddr + get_vm_area_size(vm))
2128 while (addr < vaddr) {
2135 n = vaddr + get_vm_area_size(vm) - addr;
2138 if (!(vm->flags & VM_IOREMAP)) {
2139 aligned_vwrite(buf, addr, n);
2147 spin_unlock(&vmap_area_lock);
2154 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2155 * @vma: vma to cover
2156 * @uaddr: target user address to start at
2157 * @kaddr: virtual address of vmalloc kernel memory
2158 * @size: size of map area
2160 * Returns: 0 for success, -Exxx on failure
2162 * This function checks that @kaddr is a valid vmalloc'ed area,
2163 * and that it is big enough to cover the range starting at
2164 * @uaddr in @vma. Will return failure if that criteria isn't
2167 * Similar to remap_pfn_range() (see mm/memory.c)
2169 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2170 void *kaddr, unsigned long size)
2172 struct vm_struct *area;
2174 size = PAGE_ALIGN(size);
2176 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2179 area = find_vm_area(kaddr);
2183 if (!(area->flags & VM_USERMAP))
2186 if (kaddr + size > area->addr + area->size)
2190 struct page *page = vmalloc_to_page(kaddr);
2193 ret = vm_insert_page(vma, uaddr, page);
2202 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2206 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2209 * remap_vmalloc_range - map vmalloc pages to userspace
2210 * @vma: vma to cover (map full range of vma)
2211 * @addr: vmalloc memory
2212 * @pgoff: number of pages into addr before first page to map
2214 * Returns: 0 for success, -Exxx on failure
2216 * This function checks that addr is a valid vmalloc'ed area, and
2217 * that it is big enough to cover the vma. Will return failure if
2218 * that criteria isn't met.
2220 * Similar to remap_pfn_range() (see mm/memory.c)
2222 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2223 unsigned long pgoff)
2225 return remap_vmalloc_range_partial(vma, vma->vm_start,
2226 addr + (pgoff << PAGE_SHIFT),
2227 vma->vm_end - vma->vm_start);
2229 EXPORT_SYMBOL(remap_vmalloc_range);
2232 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2235 void __weak vmalloc_sync_all(void)
2240 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2252 * alloc_vm_area - allocate a range of kernel address space
2253 * @size: size of the area
2254 * @ptes: returns the PTEs for the address space
2256 * Returns: NULL on failure, vm_struct on success
2258 * This function reserves a range of kernel address space, and
2259 * allocates pagetables to map that range. No actual mappings
2262 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2263 * allocated for the VM area are returned.
2265 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2267 struct vm_struct *area;
2269 area = get_vm_area_caller(size, VM_IOREMAP,
2270 __builtin_return_address(0));
2275 * This ensures that page tables are constructed for this region
2276 * of kernel virtual address space and mapped into init_mm.
2278 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2279 size, f, ptes ? &ptes : NULL)) {
2286 EXPORT_SYMBOL_GPL(alloc_vm_area);
2288 void free_vm_area(struct vm_struct *area)
2290 struct vm_struct *ret;
2291 ret = remove_vm_area(area->addr);
2292 BUG_ON(ret != area);
2295 EXPORT_SYMBOL_GPL(free_vm_area);
2298 static struct vmap_area *node_to_va(struct rb_node *n)
2300 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2304 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2305 * @end: target address
2306 * @pnext: out arg for the next vmap_area
2307 * @pprev: out arg for the previous vmap_area
2309 * Returns: %true if either or both of next and prev are found,
2310 * %false if no vmap_area exists
2312 * Find vmap_areas end addresses of which enclose @end. ie. if not
2313 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2315 static bool pvm_find_next_prev(unsigned long end,
2316 struct vmap_area **pnext,
2317 struct vmap_area **pprev)
2319 struct rb_node *n = vmap_area_root.rb_node;
2320 struct vmap_area *va = NULL;
2323 va = rb_entry(n, struct vmap_area, rb_node);
2324 if (end < va->va_end)
2326 else if (end > va->va_end)
2335 if (va->va_end > end) {
2337 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2340 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2346 * pvm_determine_end - find the highest aligned address between two vmap_areas
2347 * @pnext: in/out arg for the next vmap_area
2348 * @pprev: in/out arg for the previous vmap_area
2351 * Returns: determined end address
2353 * Find the highest aligned address between *@pnext and *@pprev below
2354 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2355 * down address is between the end addresses of the two vmap_areas.
2357 * Please note that the address returned by this function may fall
2358 * inside *@pnext vmap_area. The caller is responsible for checking
2361 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2362 struct vmap_area **pprev,
2363 unsigned long align)
2365 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2369 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2373 while (*pprev && (*pprev)->va_end > addr) {
2375 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2382 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2383 * @offsets: array containing offset of each area
2384 * @sizes: array containing size of each area
2385 * @nr_vms: the number of areas to allocate
2386 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2388 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2389 * vm_structs on success, %NULL on failure
2391 * Percpu allocator wants to use congruent vm areas so that it can
2392 * maintain the offsets among percpu areas. This function allocates
2393 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2394 * be scattered pretty far, distance between two areas easily going up
2395 * to gigabytes. To avoid interacting with regular vmallocs, these
2396 * areas are allocated from top.
2398 * Despite its complicated look, this allocator is rather simple. It
2399 * does everything top-down and scans areas from the end looking for
2400 * matching slot. While scanning, if any of the areas overlaps with
2401 * existing vmap_area, the base address is pulled down to fit the
2402 * area. Scanning is repeated till all the areas fit and then all
2403 * necessary data structres are inserted and the result is returned.
2405 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2406 const size_t *sizes, int nr_vms,
2409 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2410 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2411 struct vmap_area **vas, *prev, *next;
2412 struct vm_struct **vms;
2413 int area, area2, last_area, term_area;
2414 unsigned long base, start, end, last_end;
2415 bool purged = false;
2417 /* verify parameters and allocate data structures */
2418 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2419 for (last_area = 0, area = 0; area < nr_vms; area++) {
2420 start = offsets[area];
2421 end = start + sizes[area];
2423 /* is everything aligned properly? */
2424 BUG_ON(!IS_ALIGNED(offsets[area], align));
2425 BUG_ON(!IS_ALIGNED(sizes[area], align));
2427 /* detect the area with the highest address */
2428 if (start > offsets[last_area])
2431 for (area2 = 0; area2 < nr_vms; area2++) {
2432 unsigned long start2 = offsets[area2];
2433 unsigned long end2 = start2 + sizes[area2];
2438 BUG_ON(start2 >= start && start2 < end);
2439 BUG_ON(end2 <= end && end2 > start);
2442 last_end = offsets[last_area] + sizes[last_area];
2444 if (vmalloc_end - vmalloc_start < last_end) {
2449 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2450 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2454 for (area = 0; area < nr_vms; area++) {
2455 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2456 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2457 if (!vas[area] || !vms[area])
2461 spin_lock(&vmap_area_lock);
2463 /* start scanning - we scan from the top, begin with the last area */
2464 area = term_area = last_area;
2465 start = offsets[area];
2466 end = start + sizes[area];
2468 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2469 base = vmalloc_end - last_end;
2472 base = pvm_determine_end(&next, &prev, align) - end;
2475 BUG_ON(next && next->va_end <= base + end);
2476 BUG_ON(prev && prev->va_end > base + end);
2479 * base might have underflowed, add last_end before
2482 if (base + last_end < vmalloc_start + last_end) {
2483 spin_unlock(&vmap_area_lock);
2485 purge_vmap_area_lazy();
2493 * If next overlaps, move base downwards so that it's
2494 * right below next and then recheck.
2496 if (next && next->va_start < base + end) {
2497 base = pvm_determine_end(&next, &prev, align) - end;
2503 * If prev overlaps, shift down next and prev and move
2504 * base so that it's right below new next and then
2507 if (prev && prev->va_end > base + start) {
2509 prev = node_to_va(rb_prev(&next->rb_node));
2510 base = pvm_determine_end(&next, &prev, align) - end;
2516 * This area fits, move on to the previous one. If
2517 * the previous one is the terminal one, we're done.
2519 area = (area + nr_vms - 1) % nr_vms;
2520 if (area == term_area)
2522 start = offsets[area];
2523 end = start + sizes[area];
2524 pvm_find_next_prev(base + end, &next, &prev);
2527 /* we've found a fitting base, insert all va's */
2528 for (area = 0; area < nr_vms; area++) {
2529 struct vmap_area *va = vas[area];
2531 va->va_start = base + offsets[area];
2532 va->va_end = va->va_start + sizes[area];
2533 __insert_vmap_area(va);
2536 vmap_area_pcpu_hole = base + offsets[last_area];
2538 spin_unlock(&vmap_area_lock);
2540 /* insert all vm's */
2541 for (area = 0; area < nr_vms; area++)
2542 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2549 for (area = 0; area < nr_vms; area++) {
2560 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2561 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2562 * @nr_vms: the number of allocated areas
2564 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2566 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2570 for (i = 0; i < nr_vms; i++)
2571 free_vm_area(vms[i]);
2574 #endif /* CONFIG_SMP */
2576 #ifdef CONFIG_PROC_FS
2577 static void *s_start(struct seq_file *m, loff_t *pos)
2578 __acquires(&vmap_area_lock)
2581 struct vmap_area *va;
2583 spin_lock(&vmap_area_lock);
2584 va = list_first_entry(&vmap_area_list, typeof(*va), list);
2585 while (n > 0 && &va->list != &vmap_area_list) {
2587 va = list_next_entry(va, list);
2589 if (!n && &va->list != &vmap_area_list)
2596 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2598 struct vmap_area *va = p, *next;
2601 next = list_next_entry(va, list);
2602 if (&next->list != &vmap_area_list)
2608 static void s_stop(struct seq_file *m, void *p)
2609 __releases(&vmap_area_lock)
2611 spin_unlock(&vmap_area_lock);
2614 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2616 if (IS_ENABLED(CONFIG_NUMA)) {
2617 unsigned int nr, *counters = m->private;
2622 if (v->flags & VM_UNINITIALIZED)
2624 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2627 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2629 for (nr = 0; nr < v->nr_pages; nr++)
2630 counters[page_to_nid(v->pages[nr])]++;
2632 for_each_node_state(nr, N_HIGH_MEMORY)
2634 seq_printf(m, " N%u=%u", nr, counters[nr]);
2638 static int s_show(struct seq_file *m, void *p)
2640 struct vmap_area *va = p;
2641 struct vm_struct *v;
2644 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2645 * behalf of vmap area is being tear down or vm_map_ram allocation.
2647 if (!(va->flags & VM_VM_AREA))
2652 seq_printf(m, "0x%pK-0x%pK %7ld",
2653 v->addr, v->addr + v->size, v->size);
2656 seq_printf(m, " %pS", v->caller);
2659 seq_printf(m, " pages=%d", v->nr_pages);
2662 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2664 if (v->flags & VM_IOREMAP)
2665 seq_puts(m, " ioremap");
2667 if (v->flags & VM_ALLOC)
2668 seq_puts(m, " vmalloc");
2670 if (v->flags & VM_MAP)
2671 seq_puts(m, " vmap");
2673 if (v->flags & VM_USERMAP)
2674 seq_puts(m, " user");
2676 if (is_vmalloc_addr(v->pages))
2677 seq_puts(m, " vpages");
2679 show_numa_info(m, v);
2684 static const struct seq_operations vmalloc_op = {
2691 static int vmalloc_open(struct inode *inode, struct file *file)
2693 if (IS_ENABLED(CONFIG_NUMA))
2694 return seq_open_private(file, &vmalloc_op,
2695 nr_node_ids * sizeof(unsigned int));
2697 return seq_open(file, &vmalloc_op);
2700 static const struct file_operations proc_vmalloc_operations = {
2701 .open = vmalloc_open,
2703 .llseek = seq_lseek,
2704 .release = seq_release_private,
2707 static int __init proc_vmalloc_init(void)
2709 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2712 module_init(proc_vmalloc_init);