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 <asm/uaccess.h>
31 #include <asm/tlbflush.h>
32 #include <asm/shmparam.h>
34 /*** Page table manipulation functions ***/
36 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
40 pte = pte_offset_kernel(pmd, addr);
42 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
43 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
44 } while (pte++, addr += PAGE_SIZE, addr != end);
47 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
52 pmd = pmd_offset(pud, addr);
54 next = pmd_addr_end(addr, end);
55 if (pmd_none_or_clear_bad(pmd))
57 vunmap_pte_range(pmd, addr, next);
58 } while (pmd++, addr = next, addr != end);
61 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
66 pud = pud_offset(pgd, addr);
68 next = pud_addr_end(addr, end);
69 if (pud_none_or_clear_bad(pud))
71 vunmap_pmd_range(pud, addr, next);
72 } while (pud++, addr = next, addr != end);
75 static void vunmap_page_range(unsigned long addr, unsigned long end)
81 pgd = pgd_offset_k(addr);
83 next = pgd_addr_end(addr, end);
84 if (pgd_none_or_clear_bad(pgd))
86 vunmap_pud_range(pgd, addr, next);
87 } while (pgd++, addr = next, addr != end);
90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
91 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
96 * nr is a running index into the array which helps higher level
97 * callers keep track of where we're up to.
100 pte = pte_alloc_kernel(pmd, addr);
104 struct page *page = pages[*nr];
106 if (WARN_ON(!pte_none(*pte)))
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
112 } while (pte++, addr += PAGE_SIZE, addr != end);
116 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
122 pmd = pmd_alloc(&init_mm, pud, addr);
126 next = pmd_addr_end(addr, end);
127 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
129 } while (pmd++, addr = next, addr != end);
133 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
139 pud = pud_alloc(&init_mm, pgd, addr);
143 next = pud_addr_end(addr, end);
144 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
146 } while (pud++, addr = next, addr != end);
151 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152 * will have pfns corresponding to the "pages" array.
154 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
156 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157 pgprot_t prot, struct page **pages)
161 unsigned long addr = start;
166 pgd = pgd_offset_k(addr);
168 next = pgd_addr_end(addr, end);
169 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
172 } while (pgd++, addr = next, addr != end);
177 static int vmap_page_range(unsigned long start, unsigned long end,
178 pgprot_t prot, struct page **pages)
182 ret = vmap_page_range_noflush(start, end, prot, pages);
183 flush_cache_vmap(start, end);
187 int is_vmalloc_or_module_addr(const void *x)
190 * ARM, x86-64 and sparc64 put modules in a special place,
191 * and fall back on vmalloc() if that fails. Others
192 * just put it in the vmalloc space.
194 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
195 unsigned long addr = (unsigned long)x;
196 if (addr >= MODULES_VADDR && addr < MODULES_END)
199 return is_vmalloc_addr(x);
203 * Walk a vmap address to the struct page it maps.
205 struct page *vmalloc_to_page(const void *vmalloc_addr)
207 unsigned long addr = (unsigned long) vmalloc_addr;
208 struct page *page = NULL;
209 pgd_t *pgd = pgd_offset_k(addr);
212 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
213 * architectures that do not vmalloc module space
215 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
217 if (!pgd_none(*pgd)) {
218 pud_t *pud = pud_offset(pgd, addr);
219 if (!pud_none(*pud)) {
220 pmd_t *pmd = pmd_offset(pud, addr);
221 if (!pmd_none(*pmd)) {
224 ptep = pte_offset_map(pmd, addr);
226 if (pte_present(pte))
227 page = pte_page(pte);
234 EXPORT_SYMBOL(vmalloc_to_page);
237 * Map a vmalloc()-space virtual address to the physical page frame number.
239 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
241 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
243 EXPORT_SYMBOL(vmalloc_to_pfn);
246 /*** Global kva allocator ***/
248 #define VM_LAZY_FREE 0x01
249 #define VM_LAZY_FREEING 0x02
250 #define VM_VM_AREA 0x04
253 unsigned long va_start;
254 unsigned long va_end;
256 struct rb_node rb_node; /* address sorted rbtree */
257 struct list_head list; /* address sorted list */
258 struct list_head purge_list; /* "lazy purge" list */
259 struct vm_struct *vm;
260 struct rcu_head rcu_head;
263 static DEFINE_SPINLOCK(vmap_area_lock);
264 static LIST_HEAD(vmap_area_list);
265 static struct rb_root vmap_area_root = RB_ROOT;
267 /* The vmap cache globals are protected by vmap_area_lock */
268 static struct rb_node *free_vmap_cache;
269 static unsigned long cached_hole_size;
270 static unsigned long cached_vstart;
271 static unsigned long cached_align;
273 static unsigned long vmap_area_pcpu_hole;
275 static struct vmap_area *__find_vmap_area(unsigned long addr)
277 struct rb_node *n = vmap_area_root.rb_node;
280 struct vmap_area *va;
282 va = rb_entry(n, struct vmap_area, rb_node);
283 if (addr < va->va_start)
285 else if (addr > va->va_start)
294 static void __insert_vmap_area(struct vmap_area *va)
296 struct rb_node **p = &vmap_area_root.rb_node;
297 struct rb_node *parent = NULL;
301 struct vmap_area *tmp_va;
304 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
305 if (va->va_start < tmp_va->va_end)
307 else if (va->va_end > tmp_va->va_start)
313 rb_link_node(&va->rb_node, parent, p);
314 rb_insert_color(&va->rb_node, &vmap_area_root);
316 /* address-sort this list so it is usable like the vmlist */
317 tmp = rb_prev(&va->rb_node);
319 struct vmap_area *prev;
320 prev = rb_entry(tmp, struct vmap_area, rb_node);
321 list_add_rcu(&va->list, &prev->list);
323 list_add_rcu(&va->list, &vmap_area_list);
326 static void purge_vmap_area_lazy(void);
329 * Allocate a region of KVA of the specified size and alignment, within the
332 static struct vmap_area *alloc_vmap_area(unsigned long size,
334 unsigned long vstart, unsigned long vend,
335 int node, gfp_t gfp_mask)
337 struct vmap_area *va;
341 struct vmap_area *first;
344 BUG_ON(size & ~PAGE_MASK);
345 BUG_ON(!is_power_of_2(align));
347 va = kmalloc_node(sizeof(struct vmap_area),
348 gfp_mask & GFP_RECLAIM_MASK, node);
350 return ERR_PTR(-ENOMEM);
353 spin_lock(&vmap_area_lock);
355 * Invalidate cache if we have more permissive parameters.
356 * cached_hole_size notes the largest hole noticed _below_
357 * the vmap_area cached in free_vmap_cache: if size fits
358 * into that hole, we want to scan from vstart to reuse
359 * the hole instead of allocating above free_vmap_cache.
360 * Note that __free_vmap_area may update free_vmap_cache
361 * without updating cached_hole_size or cached_align.
363 if (!free_vmap_cache ||
364 size < cached_hole_size ||
365 vstart < cached_vstart ||
366 align < cached_align) {
368 cached_hole_size = 0;
369 free_vmap_cache = NULL;
371 /* record if we encounter less permissive parameters */
372 cached_vstart = vstart;
373 cached_align = align;
375 /* find starting point for our search */
376 if (free_vmap_cache) {
377 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
378 addr = ALIGN(first->va_end, align);
381 if (addr + size - 1 < addr)
385 addr = ALIGN(vstart, align);
386 if (addr + size - 1 < addr)
389 n = vmap_area_root.rb_node;
393 struct vmap_area *tmp;
394 tmp = rb_entry(n, struct vmap_area, rb_node);
395 if (tmp->va_end >= addr) {
397 if (tmp->va_start <= addr)
408 /* from the starting point, walk areas until a suitable hole is found */
409 while (addr + size > first->va_start && addr + size <= vend) {
410 if (addr + cached_hole_size < first->va_start)
411 cached_hole_size = first->va_start - addr;
412 addr = ALIGN(first->va_end, align);
413 if (addr + size - 1 < addr)
416 if (list_is_last(&first->list, &vmap_area_list))
419 first = list_entry(first->list.next,
420 struct vmap_area, list);
424 if (addr + size > vend)
428 va->va_end = addr + size;
430 __insert_vmap_area(va);
431 free_vmap_cache = &va->rb_node;
432 spin_unlock(&vmap_area_lock);
434 BUG_ON(va->va_start & (align-1));
435 BUG_ON(va->va_start < vstart);
436 BUG_ON(va->va_end > vend);
441 spin_unlock(&vmap_area_lock);
443 purge_vmap_area_lazy();
447 if (printk_ratelimit())
449 "vmap allocation for size %lu failed: "
450 "use vmalloc=<size> to increase size.\n", size);
452 return ERR_PTR(-EBUSY);
455 static void __free_vmap_area(struct vmap_area *va)
457 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
459 if (free_vmap_cache) {
460 if (va->va_end < cached_vstart) {
461 free_vmap_cache = NULL;
463 struct vmap_area *cache;
464 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
465 if (va->va_start <= cache->va_start) {
466 free_vmap_cache = rb_prev(&va->rb_node);
468 * We don't try to update cached_hole_size or
469 * cached_align, but it won't go very wrong.
474 rb_erase(&va->rb_node, &vmap_area_root);
475 RB_CLEAR_NODE(&va->rb_node);
476 list_del_rcu(&va->list);
479 * Track the highest possible candidate for pcpu area
480 * allocation. Areas outside of vmalloc area can be returned
481 * here too, consider only end addresses which fall inside
482 * vmalloc area proper.
484 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
485 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
487 kfree_rcu(va, rcu_head);
491 * Free a region of KVA allocated by alloc_vmap_area
493 static void free_vmap_area(struct vmap_area *va)
495 spin_lock(&vmap_area_lock);
496 __free_vmap_area(va);
497 spin_unlock(&vmap_area_lock);
501 * Clear the pagetable entries of a given vmap_area
503 static void unmap_vmap_area(struct vmap_area *va)
505 vunmap_page_range(va->va_start, va->va_end);
508 static void vmap_debug_free_range(unsigned long start, unsigned long end)
511 * Unmap page tables and force a TLB flush immediately if
512 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
513 * bugs similarly to those in linear kernel virtual address
514 * space after a page has been freed.
516 * All the lazy freeing logic is still retained, in order to
517 * minimise intrusiveness of this debugging feature.
519 * This is going to be *slow* (linear kernel virtual address
520 * debugging doesn't do a broadcast TLB flush so it is a lot
523 #ifdef CONFIG_DEBUG_PAGEALLOC
524 vunmap_page_range(start, end);
525 flush_tlb_kernel_range(start, end);
530 * lazy_max_pages is the maximum amount of virtual address space we gather up
531 * before attempting to purge with a TLB flush.
533 * There is a tradeoff here: a larger number will cover more kernel page tables
534 * and take slightly longer to purge, but it will linearly reduce the number of
535 * global TLB flushes that must be performed. It would seem natural to scale
536 * this number up linearly with the number of CPUs (because vmapping activity
537 * could also scale linearly with the number of CPUs), however it is likely
538 * that in practice, workloads might be constrained in other ways that mean
539 * vmap activity will not scale linearly with CPUs. Also, I want to be
540 * conservative and not introduce a big latency on huge systems, so go with
541 * a less aggressive log scale. It will still be an improvement over the old
542 * code, and it will be simple to change the scale factor if we find that it
543 * becomes a problem on bigger systems.
545 static unsigned long lazy_max_pages(void)
549 log = fls(num_online_cpus());
551 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
554 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
556 /* for per-CPU blocks */
557 static void purge_fragmented_blocks_allcpus(void);
560 * called before a call to iounmap() if the caller wants vm_area_struct's
563 void set_iounmap_nonlazy(void)
565 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
569 * Purges all lazily-freed vmap areas.
571 * If sync is 0 then don't purge if there is already a purge in progress.
572 * If force_flush is 1, then flush kernel TLBs between *start and *end even
573 * if we found no lazy vmap areas to unmap (callers can use this to optimise
574 * their own TLB flushing).
575 * Returns with *start = min(*start, lowest purged address)
576 * *end = max(*end, highest purged address)
578 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
579 int sync, int force_flush)
581 static DEFINE_SPINLOCK(purge_lock);
583 struct vmap_area *va;
584 struct vmap_area *n_va;
588 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
589 * should not expect such behaviour. This just simplifies locking for
590 * the case that isn't actually used at the moment anyway.
592 if (!sync && !force_flush) {
593 if (!spin_trylock(&purge_lock))
596 spin_lock(&purge_lock);
599 purge_fragmented_blocks_allcpus();
602 list_for_each_entry_rcu(va, &vmap_area_list, list) {
603 if (va->flags & VM_LAZY_FREE) {
604 if (va->va_start < *start)
605 *start = va->va_start;
606 if (va->va_end > *end)
608 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
609 list_add_tail(&va->purge_list, &valist);
610 va->flags |= VM_LAZY_FREEING;
611 va->flags &= ~VM_LAZY_FREE;
617 atomic_sub(nr, &vmap_lazy_nr);
619 if (nr || force_flush)
620 flush_tlb_kernel_range(*start, *end);
623 spin_lock(&vmap_area_lock);
624 list_for_each_entry_safe(va, n_va, &valist, purge_list)
625 __free_vmap_area(va);
626 spin_unlock(&vmap_area_lock);
628 spin_unlock(&purge_lock);
632 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
633 * is already purging.
635 static void try_purge_vmap_area_lazy(void)
637 unsigned long start = ULONG_MAX, end = 0;
639 __purge_vmap_area_lazy(&start, &end, 0, 0);
643 * Kick off a purge of the outstanding lazy areas.
645 static void purge_vmap_area_lazy(void)
647 unsigned long start = ULONG_MAX, end = 0;
649 __purge_vmap_area_lazy(&start, &end, 1, 0);
653 * Free a vmap area, caller ensuring that the area has been unmapped
654 * and flush_cache_vunmap had been called for the correct range
657 static void free_vmap_area_noflush(struct vmap_area *va)
659 va->flags |= VM_LAZY_FREE;
660 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
661 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
662 try_purge_vmap_area_lazy();
666 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
667 * called for the correct range previously.
669 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
672 free_vmap_area_noflush(va);
676 * Free and unmap a vmap area
678 static void free_unmap_vmap_area(struct vmap_area *va)
680 flush_cache_vunmap(va->va_start, va->va_end);
681 free_unmap_vmap_area_noflush(va);
684 static struct vmap_area *find_vmap_area(unsigned long addr)
686 struct vmap_area *va;
688 spin_lock(&vmap_area_lock);
689 va = __find_vmap_area(addr);
690 spin_unlock(&vmap_area_lock);
695 static void free_unmap_vmap_area_addr(unsigned long addr)
697 struct vmap_area *va;
699 va = find_vmap_area(addr);
701 free_unmap_vmap_area(va);
705 /*** Per cpu kva allocator ***/
708 * vmap space is limited especially on 32 bit architectures. Ensure there is
709 * room for at least 16 percpu vmap blocks per CPU.
712 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
713 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
714 * instead (we just need a rough idea)
716 #if BITS_PER_LONG == 32
717 #define VMALLOC_SPACE (128UL*1024*1024)
719 #define VMALLOC_SPACE (128UL*1024*1024*1024)
722 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
723 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
724 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
725 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
726 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
727 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
728 #define VMAP_BBMAP_BITS \
729 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
730 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
731 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
733 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
735 static bool vmap_initialized __read_mostly = false;
737 struct vmap_block_queue {
739 struct list_head free;
744 struct vmap_area *va;
745 struct vmap_block_queue *vbq;
746 unsigned long free, dirty;
747 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
748 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
749 struct list_head free_list;
750 struct rcu_head rcu_head;
751 struct list_head purge;
754 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
755 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
758 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
759 * in the free path. Could get rid of this if we change the API to return a
760 * "cookie" from alloc, to be passed to free. But no big deal yet.
762 static DEFINE_SPINLOCK(vmap_block_tree_lock);
763 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
766 * We should probably have a fallback mechanism to allocate virtual memory
767 * out of partially filled vmap blocks. However vmap block sizing should be
768 * fairly reasonable according to the vmalloc size, so it shouldn't be a
772 static unsigned long addr_to_vb_idx(unsigned long addr)
774 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
775 addr /= VMAP_BLOCK_SIZE;
779 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
781 struct vmap_block_queue *vbq;
782 struct vmap_block *vb;
783 struct vmap_area *va;
784 unsigned long vb_idx;
787 node = numa_node_id();
789 vb = kmalloc_node(sizeof(struct vmap_block),
790 gfp_mask & GFP_RECLAIM_MASK, node);
792 return ERR_PTR(-ENOMEM);
794 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
795 VMALLOC_START, VMALLOC_END,
802 err = radix_tree_preload(gfp_mask);
809 spin_lock_init(&vb->lock);
811 vb->free = VMAP_BBMAP_BITS;
813 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
814 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
815 INIT_LIST_HEAD(&vb->free_list);
817 vb_idx = addr_to_vb_idx(va->va_start);
818 spin_lock(&vmap_block_tree_lock);
819 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
820 spin_unlock(&vmap_block_tree_lock);
822 radix_tree_preload_end();
824 vbq = &get_cpu_var(vmap_block_queue);
826 spin_lock(&vbq->lock);
827 list_add_rcu(&vb->free_list, &vbq->free);
828 spin_unlock(&vbq->lock);
829 put_cpu_var(vmap_block_queue);
834 static void free_vmap_block(struct vmap_block *vb)
836 struct vmap_block *tmp;
837 unsigned long vb_idx;
839 vb_idx = addr_to_vb_idx(vb->va->va_start);
840 spin_lock(&vmap_block_tree_lock);
841 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
842 spin_unlock(&vmap_block_tree_lock);
845 free_vmap_area_noflush(vb->va);
846 kfree_rcu(vb, rcu_head);
849 static void purge_fragmented_blocks(int cpu)
852 struct vmap_block *vb;
853 struct vmap_block *n_vb;
854 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
857 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
859 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
862 spin_lock(&vb->lock);
863 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
864 vb->free = 0; /* prevent further allocs after releasing lock */
865 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
866 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
867 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
868 spin_lock(&vbq->lock);
869 list_del_rcu(&vb->free_list);
870 spin_unlock(&vbq->lock);
871 spin_unlock(&vb->lock);
872 list_add_tail(&vb->purge, &purge);
874 spin_unlock(&vb->lock);
878 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
879 list_del(&vb->purge);
884 static void purge_fragmented_blocks_thiscpu(void)
886 purge_fragmented_blocks(smp_processor_id());
889 static void purge_fragmented_blocks_allcpus(void)
893 for_each_possible_cpu(cpu)
894 purge_fragmented_blocks(cpu);
897 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
899 struct vmap_block_queue *vbq;
900 struct vmap_block *vb;
901 unsigned long addr = 0;
905 BUG_ON(size & ~PAGE_MASK);
906 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
907 if (WARN_ON(size == 0)) {
909 * Allocating 0 bytes isn't what caller wants since
910 * get_order(0) returns funny result. Just warn and terminate
915 order = get_order(size);
919 vbq = &get_cpu_var(vmap_block_queue);
920 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
923 spin_lock(&vb->lock);
924 if (vb->free < 1UL << order)
927 i = bitmap_find_free_region(vb->alloc_map,
928 VMAP_BBMAP_BITS, order);
931 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
932 /* fragmented and no outstanding allocations */
933 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
938 addr = vb->va->va_start + (i << PAGE_SHIFT);
939 BUG_ON(addr_to_vb_idx(addr) !=
940 addr_to_vb_idx(vb->va->va_start));
941 vb->free -= 1UL << order;
943 spin_lock(&vbq->lock);
944 list_del_rcu(&vb->free_list);
945 spin_unlock(&vbq->lock);
947 spin_unlock(&vb->lock);
950 spin_unlock(&vb->lock);
954 purge_fragmented_blocks_thiscpu();
956 put_cpu_var(vmap_block_queue);
960 vb = new_vmap_block(gfp_mask);
969 static void vb_free(const void *addr, unsigned long size)
971 unsigned long offset;
972 unsigned long vb_idx;
974 struct vmap_block *vb;
976 BUG_ON(size & ~PAGE_MASK);
977 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
979 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
981 order = get_order(size);
983 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
985 vb_idx = addr_to_vb_idx((unsigned long)addr);
987 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
991 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
993 spin_lock(&vb->lock);
994 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
996 vb->dirty += 1UL << order;
997 if (vb->dirty == VMAP_BBMAP_BITS) {
999 spin_unlock(&vb->lock);
1000 free_vmap_block(vb);
1002 spin_unlock(&vb->lock);
1006 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1008 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1009 * to amortize TLB flushing overheads. What this means is that any page you
1010 * have now, may, in a former life, have been mapped into kernel virtual
1011 * address by the vmap layer and so there might be some CPUs with TLB entries
1012 * still referencing that page (additional to the regular 1:1 kernel mapping).
1014 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1015 * be sure that none of the pages we have control over will have any aliases
1016 * from the vmap layer.
1018 void vm_unmap_aliases(void)
1020 unsigned long start = ULONG_MAX, end = 0;
1024 if (unlikely(!vmap_initialized))
1027 for_each_possible_cpu(cpu) {
1028 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1029 struct vmap_block *vb;
1032 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1035 spin_lock(&vb->lock);
1036 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1037 while (i < VMAP_BBMAP_BITS) {
1040 j = find_next_zero_bit(vb->dirty_map,
1041 VMAP_BBMAP_BITS, i);
1043 s = vb->va->va_start + (i << PAGE_SHIFT);
1044 e = vb->va->va_start + (j << PAGE_SHIFT);
1053 i = find_next_bit(vb->dirty_map,
1054 VMAP_BBMAP_BITS, i);
1056 spin_unlock(&vb->lock);
1061 __purge_vmap_area_lazy(&start, &end, 1, flush);
1063 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1066 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1067 * @mem: the pointer returned by vm_map_ram
1068 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1070 void vm_unmap_ram(const void *mem, unsigned int count)
1072 unsigned long size = count << PAGE_SHIFT;
1073 unsigned long addr = (unsigned long)mem;
1076 BUG_ON(addr < VMALLOC_START);
1077 BUG_ON(addr > VMALLOC_END);
1078 BUG_ON(addr & (PAGE_SIZE-1));
1080 debug_check_no_locks_freed(mem, size);
1081 vmap_debug_free_range(addr, addr+size);
1083 if (likely(count <= VMAP_MAX_ALLOC))
1086 free_unmap_vmap_area_addr(addr);
1088 EXPORT_SYMBOL(vm_unmap_ram);
1091 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1092 * @pages: an array of pointers to the pages to be mapped
1093 * @count: number of pages
1094 * @node: prefer to allocate data structures on this node
1095 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1097 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1099 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1101 unsigned long size = count << PAGE_SHIFT;
1105 if (likely(count <= VMAP_MAX_ALLOC)) {
1106 mem = vb_alloc(size, GFP_KERNEL);
1109 addr = (unsigned long)mem;
1111 struct vmap_area *va;
1112 va = alloc_vmap_area(size, PAGE_SIZE,
1113 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1117 addr = va->va_start;
1120 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1121 vm_unmap_ram(mem, count);
1126 EXPORT_SYMBOL(vm_map_ram);
1129 * vm_area_add_early - add vmap area early during boot
1130 * @vm: vm_struct to add
1132 * This function is used to add fixed kernel vm area to vmlist before
1133 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1134 * should contain proper values and the other fields should be zero.
1136 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1138 void __init vm_area_add_early(struct vm_struct *vm)
1140 struct vm_struct *tmp, **p;
1142 BUG_ON(vmap_initialized);
1143 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1144 if (tmp->addr >= vm->addr) {
1145 BUG_ON(tmp->addr < vm->addr + vm->size);
1148 BUG_ON(tmp->addr + tmp->size > vm->addr);
1155 * vm_area_register_early - register vmap area early during boot
1156 * @vm: vm_struct to register
1157 * @align: requested alignment
1159 * This function is used to register kernel vm area before
1160 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1161 * proper values on entry and other fields should be zero. On return,
1162 * vm->addr contains the allocated address.
1164 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1166 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1168 static size_t vm_init_off __initdata;
1171 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1172 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1174 vm->addr = (void *)addr;
1176 vm_area_add_early(vm);
1179 void __init vmalloc_init(void)
1181 struct vmap_area *va;
1182 struct vm_struct *tmp;
1185 for_each_possible_cpu(i) {
1186 struct vmap_block_queue *vbq;
1188 vbq = &per_cpu(vmap_block_queue, i);
1189 spin_lock_init(&vbq->lock);
1190 INIT_LIST_HEAD(&vbq->free);
1193 /* Import existing vmlist entries. */
1194 for (tmp = vmlist; tmp; tmp = tmp->next) {
1195 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1196 va->flags = VM_VM_AREA;
1197 va->va_start = (unsigned long)tmp->addr;
1198 va->va_end = va->va_start + tmp->size;
1200 __insert_vmap_area(va);
1203 vmap_area_pcpu_hole = VMALLOC_END;
1205 vmap_initialized = true;
1209 * map_kernel_range_noflush - map kernel VM area with the specified pages
1210 * @addr: start of the VM area to map
1211 * @size: size of the VM area to map
1212 * @prot: page protection flags to use
1213 * @pages: pages to map
1215 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1216 * specify should have been allocated using get_vm_area() and its
1220 * This function does NOT do any cache flushing. The caller is
1221 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1222 * before calling this function.
1225 * The number of pages mapped on success, -errno on failure.
1227 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1228 pgprot_t prot, struct page **pages)
1230 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1234 * unmap_kernel_range_noflush - unmap kernel VM area
1235 * @addr: start of the VM area to unmap
1236 * @size: size of the VM area to unmap
1238 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1239 * specify should have been allocated using get_vm_area() and its
1243 * This function does NOT do any cache flushing. The caller is
1244 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1245 * before calling this function and flush_tlb_kernel_range() after.
1247 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1249 vunmap_page_range(addr, addr + size);
1251 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1254 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1255 * @addr: start of the VM area to unmap
1256 * @size: size of the VM area to unmap
1258 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1259 * the unmapping and tlb after.
1261 void unmap_kernel_range(unsigned long addr, unsigned long size)
1263 unsigned long end = addr + size;
1265 flush_cache_vunmap(addr, end);
1266 vunmap_page_range(addr, end);
1267 flush_tlb_kernel_range(addr, end);
1270 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1272 unsigned long addr = (unsigned long)area->addr;
1273 unsigned long end = addr + area->size - PAGE_SIZE;
1276 err = vmap_page_range(addr, end, prot, *pages);
1284 EXPORT_SYMBOL_GPL(map_vm_area);
1286 /*** Old vmalloc interfaces ***/
1287 DEFINE_RWLOCK(vmlist_lock);
1288 struct vm_struct *vmlist;
1290 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1291 unsigned long flags, const void *caller)
1293 spin_lock(&vmap_area_lock);
1295 vm->addr = (void *)va->va_start;
1296 vm->size = va->va_end - va->va_start;
1297 vm->caller = caller;
1299 va->flags |= VM_VM_AREA;
1300 spin_unlock(&vmap_area_lock);
1303 static void insert_vmalloc_vmlist(struct vm_struct *vm)
1305 struct vm_struct *tmp, **p;
1308 * Before removing VM_UNLIST,
1309 * we should make sure that vm has proper values.
1310 * Pair with smp_rmb() in show_numa_info().
1313 vm->flags &= ~VM_UNLIST;
1315 write_lock(&vmlist_lock);
1316 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1317 if (tmp->addr >= vm->addr)
1322 write_unlock(&vmlist_lock);
1325 static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1326 unsigned long flags, const void *caller)
1328 setup_vmalloc_vm(vm, va, flags, caller);
1329 insert_vmalloc_vmlist(vm);
1332 static struct vm_struct *__get_vm_area_node(unsigned long size,
1333 unsigned long align, unsigned long flags, unsigned long start,
1334 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1336 struct vmap_area *va;
1337 struct vm_struct *area;
1339 BUG_ON(in_interrupt());
1340 if (flags & VM_IOREMAP) {
1341 int bit = fls(size);
1343 if (bit > IOREMAP_MAX_ORDER)
1344 bit = IOREMAP_MAX_ORDER;
1345 else if (bit < PAGE_SHIFT)
1351 size = PAGE_ALIGN(size);
1352 if (unlikely(!size))
1355 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1356 if (unlikely(!area))
1360 * We always allocate a guard page.
1364 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1371 * When this function is called from __vmalloc_node_range,
1372 * we do not add vm_struct to vmlist here to avoid
1373 * accessing uninitialized members of vm_struct such as
1374 * pages and nr_pages fields. They will be set later.
1375 * To distinguish it from others, we use a VM_UNLIST flag.
1377 if (flags & VM_UNLIST)
1378 setup_vmalloc_vm(area, va, flags, caller);
1380 insert_vmalloc_vm(area, va, flags, caller);
1385 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1386 unsigned long start, unsigned long end)
1388 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1389 GFP_KERNEL, __builtin_return_address(0));
1391 EXPORT_SYMBOL_GPL(__get_vm_area);
1393 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1394 unsigned long start, unsigned long end,
1397 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1398 GFP_KERNEL, caller);
1402 * get_vm_area - reserve a contiguous kernel virtual area
1403 * @size: size of the area
1404 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1406 * Search an area of @size in the kernel virtual mapping area,
1407 * and reserved it for out purposes. Returns the area descriptor
1408 * on success or %NULL on failure.
1410 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1412 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1413 NUMA_NO_NODE, GFP_KERNEL,
1414 __builtin_return_address(0));
1417 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1420 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1421 NUMA_NO_NODE, GFP_KERNEL, caller);
1425 * find_vm_area - find a continuous kernel virtual area
1426 * @addr: base address
1428 * Search for the kernel VM area starting at @addr, and return it.
1429 * It is up to the caller to do all required locking to keep the returned
1432 struct vm_struct *find_vm_area(const void *addr)
1434 struct vmap_area *va;
1436 va = find_vmap_area((unsigned long)addr);
1437 if (va && va->flags & VM_VM_AREA)
1444 * remove_vm_area - find and remove a continuous kernel virtual area
1445 * @addr: base address
1447 * Search for the kernel VM area starting at @addr, and remove it.
1448 * This function returns the found VM area, but using it is NOT safe
1449 * on SMP machines, except for its size or flags.
1451 struct vm_struct *remove_vm_area(const void *addr)
1453 struct vmap_area *va;
1455 va = find_vmap_area((unsigned long)addr);
1456 if (va && va->flags & VM_VM_AREA) {
1457 struct vm_struct *vm = va->vm;
1459 spin_lock(&vmap_area_lock);
1461 va->flags &= ~VM_VM_AREA;
1462 spin_unlock(&vmap_area_lock);
1464 if (!(vm->flags & VM_UNLIST)) {
1465 struct vm_struct *tmp, **p;
1467 * remove from list and disallow access to
1468 * this vm_struct before unmap. (address range
1469 * confliction is maintained by vmap.)
1471 write_lock(&vmlist_lock);
1472 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1475 write_unlock(&vmlist_lock);
1478 vmap_debug_free_range(va->va_start, va->va_end);
1479 free_unmap_vmap_area(va);
1480 vm->size -= PAGE_SIZE;
1487 static void __vunmap(const void *addr, int deallocate_pages)
1489 struct vm_struct *area;
1494 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1495 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1499 area = remove_vm_area(addr);
1500 if (unlikely(!area)) {
1501 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1506 debug_check_no_locks_freed(addr, area->size);
1507 debug_check_no_obj_freed(addr, area->size);
1509 if (deallocate_pages) {
1512 for (i = 0; i < area->nr_pages; i++) {
1513 struct page *page = area->pages[i];
1519 if (area->flags & VM_VPAGES)
1530 * vfree - release memory allocated by vmalloc()
1531 * @addr: memory base address
1533 * Free the virtually continuous memory area starting at @addr, as
1534 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1535 * NULL, no operation is performed.
1537 * Must not be called in interrupt context.
1539 void vfree(const void *addr)
1541 BUG_ON(in_interrupt());
1543 kmemleak_free(addr);
1547 EXPORT_SYMBOL(vfree);
1550 * vunmap - release virtual mapping obtained by vmap()
1551 * @addr: memory base address
1553 * Free the virtually contiguous memory area starting at @addr,
1554 * which was created from the page array passed to vmap().
1556 * Must not be called in interrupt context.
1558 void vunmap(const void *addr)
1560 BUG_ON(in_interrupt());
1564 EXPORT_SYMBOL(vunmap);
1567 * vmap - map an array of pages into virtually contiguous space
1568 * @pages: array of page pointers
1569 * @count: number of pages to map
1570 * @flags: vm_area->flags
1571 * @prot: page protection for the mapping
1573 * Maps @count pages from @pages into contiguous kernel virtual
1576 void *vmap(struct page **pages, unsigned int count,
1577 unsigned long flags, pgprot_t prot)
1579 struct vm_struct *area;
1583 if (count > totalram_pages)
1586 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1587 __builtin_return_address(0));
1591 if (map_vm_area(area, prot, &pages)) {
1598 EXPORT_SYMBOL(vmap);
1600 static void *__vmalloc_node(unsigned long size, unsigned long align,
1601 gfp_t gfp_mask, pgprot_t prot,
1602 int node, const void *caller);
1603 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1604 pgprot_t prot, int node, const void *caller)
1606 const int order = 0;
1607 struct page **pages;
1608 unsigned int nr_pages, array_size, i;
1609 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1611 nr_pages = (area->size - PAGE_SIZE) >> 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, caller);
1619 area->flags |= VM_VPAGES;
1621 pages = kmalloc_node(array_size, nested_gfp, node);
1623 area->pages = pages;
1624 area->caller = caller;
1626 remove_vm_area(area->addr);
1631 for (i = 0; i < area->nr_pages; i++) {
1633 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
1636 page = alloc_page(tmp_mask);
1638 page = alloc_pages_node(node, tmp_mask, order);
1640 if (unlikely(!page)) {
1641 /* Successfully allocated i pages, free them in __vunmap() */
1645 area->pages[i] = page;
1648 if (map_vm_area(area, prot, &pages))
1653 warn_alloc_failed(gfp_mask, order,
1654 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1655 (area->nr_pages*PAGE_SIZE), area->size);
1661 * __vmalloc_node_range - allocate virtually contiguous memory
1662 * @size: allocation size
1663 * @align: desired alignment
1664 * @start: vm area range start
1665 * @end: vm area range end
1666 * @gfp_mask: flags for the page level allocator
1667 * @prot: protection mask for the allocated pages
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, int node, const void *caller)
1679 struct vm_struct *area;
1681 unsigned long real_size = size;
1683 size = PAGE_ALIGN(size);
1684 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1687 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNLIST,
1688 start, end, node, gfp_mask, caller);
1692 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1697 * In this function, newly allocated vm_struct is not added
1698 * to vmlist at __get_vm_area_node(). so, it is added here.
1700 insert_vmalloc_vmlist(area);
1703 * A ref_count = 3 is needed because the vm_struct and vmap_area
1704 * structures allocated in the __get_vm_area_node() function contain
1705 * references to the virtual address of the vmalloc'ed block.
1707 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1712 warn_alloc_failed(gfp_mask, 0,
1713 "vmalloc: allocation failure: %lu bytes\n",
1719 * __vmalloc_node - allocate virtually contiguous memory
1720 * @size: allocation size
1721 * @align: desired alignment
1722 * @gfp_mask: flags for the page level allocator
1723 * @prot: protection mask for the allocated pages
1724 * @node: node to use for allocation or NUMA_NO_NODE
1725 * @caller: caller's return address
1727 * Allocate enough pages to cover @size from the page level
1728 * allocator with @gfp_mask flags. Map them into contiguous
1729 * kernel virtual space, using a pagetable protection of @prot.
1731 static void *__vmalloc_node(unsigned long size, unsigned long align,
1732 gfp_t gfp_mask, pgprot_t prot,
1733 int node, const void *caller)
1735 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1736 gfp_mask, prot, node, caller);
1739 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1741 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1742 __builtin_return_address(0));
1744 EXPORT_SYMBOL(__vmalloc);
1746 static inline void *__vmalloc_node_flags(unsigned long size,
1747 int node, gfp_t flags)
1749 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1750 node, __builtin_return_address(0));
1754 * vmalloc - allocate virtually contiguous memory
1755 * @size: allocation size
1756 * Allocate enough pages to cover @size from the page level
1757 * allocator and map them into contiguous kernel virtual space.
1759 * For tight control over page level allocator and protection flags
1760 * use __vmalloc() instead.
1762 void *vmalloc(unsigned long size)
1764 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1765 GFP_KERNEL | __GFP_HIGHMEM);
1767 EXPORT_SYMBOL(vmalloc);
1770 * vzalloc - allocate virtually contiguous memory with zero fill
1771 * @size: allocation size
1772 * Allocate enough pages to cover @size from the page level
1773 * allocator and map them into contiguous kernel virtual space.
1774 * The memory allocated is set to zero.
1776 * For tight control over page level allocator and protection flags
1777 * use __vmalloc() instead.
1779 void *vzalloc(unsigned long size)
1781 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1782 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1784 EXPORT_SYMBOL(vzalloc);
1787 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1788 * @size: allocation size
1790 * The resulting memory area is zeroed so it can be mapped to userspace
1791 * without leaking data.
1793 void *vmalloc_user(unsigned long size)
1795 struct vm_struct *area;
1798 ret = __vmalloc_node(size, SHMLBA,
1799 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1800 PAGE_KERNEL, NUMA_NO_NODE,
1801 __builtin_return_address(0));
1803 area = find_vm_area(ret);
1804 area->flags |= VM_USERMAP;
1808 EXPORT_SYMBOL(vmalloc_user);
1811 * vmalloc_node - allocate memory on a specific node
1812 * @size: allocation size
1815 * Allocate enough pages to cover @size from the page level
1816 * allocator and map them into contiguous kernel virtual space.
1818 * For tight control over page level allocator and protection flags
1819 * use __vmalloc() instead.
1821 void *vmalloc_node(unsigned long size, int node)
1823 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1824 node, __builtin_return_address(0));
1826 EXPORT_SYMBOL(vmalloc_node);
1829 * vzalloc_node - allocate memory on a specific node with zero fill
1830 * @size: allocation size
1833 * Allocate enough pages to cover @size from the page level
1834 * allocator and map them into contiguous kernel virtual space.
1835 * The memory allocated is set to zero.
1837 * For tight control over page level allocator and protection flags
1838 * use __vmalloc_node() instead.
1840 void *vzalloc_node(unsigned long size, int node)
1842 return __vmalloc_node_flags(size, node,
1843 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1845 EXPORT_SYMBOL(vzalloc_node);
1847 #ifndef PAGE_KERNEL_EXEC
1848 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1852 * vmalloc_exec - allocate virtually contiguous, executable memory
1853 * @size: allocation size
1855 * Kernel-internal function to allocate enough pages to cover @size
1856 * the page level allocator and map them into contiguous and
1857 * executable kernel virtual space.
1859 * For tight control over page level allocator and protection flags
1860 * use __vmalloc() instead.
1863 void *vmalloc_exec(unsigned long size)
1865 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1866 NUMA_NO_NODE, __builtin_return_address(0));
1869 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1870 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1871 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1872 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1874 #define GFP_VMALLOC32 GFP_KERNEL
1878 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1879 * @size: allocation size
1881 * Allocate enough 32bit PA addressable pages to cover @size from the
1882 * page level allocator and map them into contiguous kernel virtual space.
1884 void *vmalloc_32(unsigned long size)
1886 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1887 NUMA_NO_NODE, __builtin_return_address(0));
1889 EXPORT_SYMBOL(vmalloc_32);
1892 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1893 * @size: allocation size
1895 * The resulting memory area is 32bit addressable and zeroed so it can be
1896 * mapped to userspace without leaking data.
1898 void *vmalloc_32_user(unsigned long size)
1900 struct vm_struct *area;
1903 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1904 NUMA_NO_NODE, __builtin_return_address(0));
1906 area = find_vm_area(ret);
1907 area->flags |= VM_USERMAP;
1911 EXPORT_SYMBOL(vmalloc_32_user);
1914 * small helper routine , copy contents to buf from addr.
1915 * If the page is not present, fill zero.
1918 static int aligned_vread(char *buf, char *addr, unsigned long count)
1924 unsigned long offset, length;
1926 offset = (unsigned long)addr & ~PAGE_MASK;
1927 length = PAGE_SIZE - offset;
1930 p = vmalloc_to_page(addr);
1932 * To do safe access to this _mapped_ area, we need
1933 * lock. But adding lock here means that we need to add
1934 * overhead of vmalloc()/vfree() calles for this _debug_
1935 * interface, rarely used. Instead of that, we'll use
1936 * kmap() and get small overhead in this access function.
1940 * we can expect USER0 is not used (see vread/vwrite's
1941 * function description)
1943 void *map = kmap_atomic(p);
1944 memcpy(buf, map + offset, length);
1947 memset(buf, 0, length);
1957 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1963 unsigned long offset, length;
1965 offset = (unsigned long)addr & ~PAGE_MASK;
1966 length = PAGE_SIZE - offset;
1969 p = vmalloc_to_page(addr);
1971 * To do safe access to this _mapped_ area, we need
1972 * lock. But adding lock here means that we need to add
1973 * overhead of vmalloc()/vfree() calles for this _debug_
1974 * interface, rarely used. Instead of that, we'll use
1975 * kmap() and get small overhead in this access function.
1979 * we can expect USER0 is not used (see vread/vwrite's
1980 * function description)
1982 void *map = kmap_atomic(p);
1983 memcpy(map + offset, buf, length);
1995 * vread() - read vmalloc area in a safe way.
1996 * @buf: buffer for reading data
1997 * @addr: vm address.
1998 * @count: number of bytes to be read.
2000 * Returns # of bytes which addr and buf should be increased.
2001 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2002 * includes any intersect with alive vmalloc area.
2004 * This function checks that addr is a valid vmalloc'ed area, and
2005 * copy data from that area to a given buffer. If the given memory range
2006 * of [addr...addr+count) includes some valid address, data is copied to
2007 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2008 * IOREMAP area is treated as memory hole and no copy is done.
2010 * If [addr...addr+count) doesn't includes any intersects with alive
2011 * vm_struct area, returns 0. @buf should be kernel's buffer.
2013 * Note: In usual ops, vread() is never necessary because the caller
2014 * should know vmalloc() area is valid and can use memcpy().
2015 * This is for routines which have to access vmalloc area without
2016 * any informaion, as /dev/kmem.
2020 long vread(char *buf, char *addr, unsigned long count)
2022 struct vmap_area *va;
2023 struct vm_struct *vm;
2024 char *vaddr, *buf_start = buf;
2025 unsigned long buflen = count;
2028 /* Don't allow overflow */
2029 if ((unsigned long) addr + count < count)
2030 count = -(unsigned long) addr;
2032 spin_lock(&vmap_area_lock);
2033 list_for_each_entry(va, &vmap_area_list, list) {
2037 if (!(va->flags & VM_VM_AREA))
2041 vaddr = (char *) vm->addr;
2042 if (addr >= vaddr + vm->size - PAGE_SIZE)
2044 while (addr < vaddr) {
2052 n = vaddr + vm->size - PAGE_SIZE - addr;
2055 if (!(vm->flags & VM_IOREMAP))
2056 aligned_vread(buf, addr, n);
2057 else /* IOREMAP area is treated as memory hole */
2064 spin_unlock(&vmap_area_lock);
2066 if (buf == buf_start)
2068 /* zero-fill memory holes */
2069 if (buf != buf_start + buflen)
2070 memset(buf, 0, buflen - (buf - buf_start));
2076 * vwrite() - write vmalloc area in a safe way.
2077 * @buf: buffer for source data
2078 * @addr: vm address.
2079 * @count: number of bytes to be read.
2081 * Returns # of bytes which addr and buf should be incresed.
2082 * (same number to @count).
2083 * If [addr...addr+count) doesn't includes any intersect with valid
2084 * vmalloc area, returns 0.
2086 * This function checks that addr is a valid vmalloc'ed area, and
2087 * copy data from a buffer to the given addr. If specified range of
2088 * [addr...addr+count) includes some valid address, data is copied from
2089 * proper area of @buf. If there are memory holes, no copy to hole.
2090 * IOREMAP area is treated as memory hole and no copy is done.
2092 * If [addr...addr+count) doesn't includes any intersects with alive
2093 * vm_struct area, returns 0. @buf should be kernel's buffer.
2095 * Note: In usual ops, vwrite() is never necessary because the caller
2096 * should know vmalloc() area is valid and can use memcpy().
2097 * This is for routines which have to access vmalloc area without
2098 * any informaion, as /dev/kmem.
2101 long vwrite(char *buf, char *addr, unsigned long count)
2103 struct vmap_area *va;
2104 struct vm_struct *vm;
2106 unsigned long n, buflen;
2109 /* Don't allow overflow */
2110 if ((unsigned long) addr + count < count)
2111 count = -(unsigned long) addr;
2114 spin_lock(&vmap_area_lock);
2115 list_for_each_entry(va, &vmap_area_list, list) {
2119 if (!(va->flags & VM_VM_AREA))
2123 vaddr = (char *) vm->addr;
2124 if (addr >= vaddr + vm->size - PAGE_SIZE)
2126 while (addr < vaddr) {
2133 n = vaddr + vm->size - PAGE_SIZE - addr;
2136 if (!(vm->flags & VM_IOREMAP)) {
2137 aligned_vwrite(buf, addr, n);
2145 spin_unlock(&vmap_area_lock);
2152 * remap_vmalloc_range - map vmalloc pages to userspace
2153 * @vma: vma to cover (map full range of vma)
2154 * @addr: vmalloc memory
2155 * @pgoff: number of pages into addr before first page to map
2157 * Returns: 0 for success, -Exxx on failure
2159 * This function checks that addr is a valid vmalloc'ed area, and
2160 * that it is big enough to cover the vma. Will return failure if
2161 * that criteria isn't met.
2163 * Similar to remap_pfn_range() (see mm/memory.c)
2165 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2166 unsigned long pgoff)
2168 struct vm_struct *area;
2169 unsigned long uaddr = vma->vm_start;
2170 unsigned long usize = vma->vm_end - vma->vm_start;
2172 if ((PAGE_SIZE-1) & (unsigned long)addr)
2175 area = find_vm_area(addr);
2179 if (!(area->flags & VM_USERMAP))
2182 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
2185 addr += pgoff << PAGE_SHIFT;
2187 struct page *page = vmalloc_to_page(addr);
2190 ret = vm_insert_page(vma, uaddr, page);
2197 } while (usize > 0);
2199 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2203 EXPORT_SYMBOL(remap_vmalloc_range);
2206 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2209 void __attribute__((weak)) vmalloc_sync_all(void)
2214 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2226 * alloc_vm_area - allocate a range of kernel address space
2227 * @size: size of the area
2228 * @ptes: returns the PTEs for the address space
2230 * Returns: NULL on failure, vm_struct on success
2232 * This function reserves a range of kernel address space, and
2233 * allocates pagetables to map that range. No actual mappings
2236 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2237 * allocated for the VM area are returned.
2239 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2241 struct vm_struct *area;
2243 area = get_vm_area_caller(size, VM_IOREMAP,
2244 __builtin_return_address(0));
2249 * This ensures that page tables are constructed for this region
2250 * of kernel virtual address space and mapped into init_mm.
2252 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2253 size, f, ptes ? &ptes : NULL)) {
2260 EXPORT_SYMBOL_GPL(alloc_vm_area);
2262 void free_vm_area(struct vm_struct *area)
2264 struct vm_struct *ret;
2265 ret = remove_vm_area(area->addr);
2266 BUG_ON(ret != area);
2269 EXPORT_SYMBOL_GPL(free_vm_area);
2272 static struct vmap_area *node_to_va(struct rb_node *n)
2274 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2278 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2279 * @end: target address
2280 * @pnext: out arg for the next vmap_area
2281 * @pprev: out arg for the previous vmap_area
2283 * Returns: %true if either or both of next and prev are found,
2284 * %false if no vmap_area exists
2286 * Find vmap_areas end addresses of which enclose @end. ie. if not
2287 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2289 static bool pvm_find_next_prev(unsigned long end,
2290 struct vmap_area **pnext,
2291 struct vmap_area **pprev)
2293 struct rb_node *n = vmap_area_root.rb_node;
2294 struct vmap_area *va = NULL;
2297 va = rb_entry(n, struct vmap_area, rb_node);
2298 if (end < va->va_end)
2300 else if (end > va->va_end)
2309 if (va->va_end > end) {
2311 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2314 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2320 * pvm_determine_end - find the highest aligned address between two vmap_areas
2321 * @pnext: in/out arg for the next vmap_area
2322 * @pprev: in/out arg for the previous vmap_area
2325 * Returns: determined end address
2327 * Find the highest aligned address between *@pnext and *@pprev below
2328 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2329 * down address is between the end addresses of the two vmap_areas.
2331 * Please note that the address returned by this function may fall
2332 * inside *@pnext vmap_area. The caller is responsible for checking
2335 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2336 struct vmap_area **pprev,
2337 unsigned long align)
2339 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2343 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2347 while (*pprev && (*pprev)->va_end > addr) {
2349 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2356 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2357 * @offsets: array containing offset of each area
2358 * @sizes: array containing size of each area
2359 * @nr_vms: the number of areas to allocate
2360 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2362 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2363 * vm_structs on success, %NULL on failure
2365 * Percpu allocator wants to use congruent vm areas so that it can
2366 * maintain the offsets among percpu areas. This function allocates
2367 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2368 * be scattered pretty far, distance between two areas easily going up
2369 * to gigabytes. To avoid interacting with regular vmallocs, these
2370 * areas are allocated from top.
2372 * Despite its complicated look, this allocator is rather simple. It
2373 * does everything top-down and scans areas from the end looking for
2374 * matching slot. While scanning, if any of the areas overlaps with
2375 * existing vmap_area, the base address is pulled down to fit the
2376 * area. Scanning is repeated till all the areas fit and then all
2377 * necessary data structres are inserted and the result is returned.
2379 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2380 const size_t *sizes, int nr_vms,
2383 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2384 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2385 struct vmap_area **vas, *prev, *next;
2386 struct vm_struct **vms;
2387 int area, area2, last_area, term_area;
2388 unsigned long base, start, end, last_end;
2389 bool purged = false;
2391 /* verify parameters and allocate data structures */
2392 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2393 for (last_area = 0, area = 0; area < nr_vms; area++) {
2394 start = offsets[area];
2395 end = start + sizes[area];
2397 /* is everything aligned properly? */
2398 BUG_ON(!IS_ALIGNED(offsets[area], align));
2399 BUG_ON(!IS_ALIGNED(sizes[area], align));
2401 /* detect the area with the highest address */
2402 if (start > offsets[last_area])
2405 for (area2 = 0; area2 < nr_vms; area2++) {
2406 unsigned long start2 = offsets[area2];
2407 unsigned long end2 = start2 + sizes[area2];
2412 BUG_ON(start2 >= start && start2 < end);
2413 BUG_ON(end2 <= end && end2 > start);
2416 last_end = offsets[last_area] + sizes[last_area];
2418 if (vmalloc_end - vmalloc_start < last_end) {
2423 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2424 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2428 for (area = 0; area < nr_vms; area++) {
2429 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2430 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2431 if (!vas[area] || !vms[area])
2435 spin_lock(&vmap_area_lock);
2437 /* start scanning - we scan from the top, begin with the last area */
2438 area = term_area = last_area;
2439 start = offsets[area];
2440 end = start + sizes[area];
2442 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2443 base = vmalloc_end - last_end;
2446 base = pvm_determine_end(&next, &prev, align) - end;
2449 BUG_ON(next && next->va_end <= base + end);
2450 BUG_ON(prev && prev->va_end > base + end);
2453 * base might have underflowed, add last_end before
2456 if (base + last_end < vmalloc_start + last_end) {
2457 spin_unlock(&vmap_area_lock);
2459 purge_vmap_area_lazy();
2467 * If next overlaps, move base downwards so that it's
2468 * right below next and then recheck.
2470 if (next && next->va_start < base + end) {
2471 base = pvm_determine_end(&next, &prev, align) - end;
2477 * If prev overlaps, shift down next and prev and move
2478 * base so that it's right below new next and then
2481 if (prev && prev->va_end > base + start) {
2483 prev = node_to_va(rb_prev(&next->rb_node));
2484 base = pvm_determine_end(&next, &prev, align) - end;
2490 * This area fits, move on to the previous one. If
2491 * the previous one is the terminal one, we're done.
2493 area = (area + nr_vms - 1) % nr_vms;
2494 if (area == term_area)
2496 start = offsets[area];
2497 end = start + sizes[area];
2498 pvm_find_next_prev(base + end, &next, &prev);
2501 /* we've found a fitting base, insert all va's */
2502 for (area = 0; area < nr_vms; area++) {
2503 struct vmap_area *va = vas[area];
2505 va->va_start = base + offsets[area];
2506 va->va_end = va->va_start + sizes[area];
2507 __insert_vmap_area(va);
2510 vmap_area_pcpu_hole = base + offsets[last_area];
2512 spin_unlock(&vmap_area_lock);
2514 /* insert all vm's */
2515 for (area = 0; area < nr_vms; area++)
2516 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2523 for (area = 0; area < nr_vms; area++) {
2534 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2535 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2536 * @nr_vms: the number of allocated areas
2538 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2540 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2544 for (i = 0; i < nr_vms; i++)
2545 free_vm_area(vms[i]);
2548 #endif /* CONFIG_SMP */
2550 #ifdef CONFIG_PROC_FS
2551 static void *s_start(struct seq_file *m, loff_t *pos)
2552 __acquires(&vmap_area_lock)
2555 struct vmap_area *va;
2557 spin_lock(&vmap_area_lock);
2558 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2559 while (n > 0 && &va->list != &vmap_area_list) {
2561 va = list_entry(va->list.next, typeof(*va), list);
2563 if (!n && &va->list != &vmap_area_list)
2570 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2572 struct vmap_area *va = p, *next;
2575 next = list_entry(va->list.next, typeof(*va), list);
2576 if (&next->list != &vmap_area_list)
2582 static void s_stop(struct seq_file *m, void *p)
2583 __releases(&vmap_area_lock)
2585 spin_unlock(&vmap_area_lock);
2588 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2590 if (IS_ENABLED(CONFIG_NUMA)) {
2591 unsigned int nr, *counters = m->private;
2596 /* Pair with smp_wmb() in insert_vmalloc_vmlist() */
2598 if (v->flags & VM_UNLIST)
2601 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2603 for (nr = 0; nr < v->nr_pages; nr++)
2604 counters[page_to_nid(v->pages[nr])]++;
2606 for_each_node_state(nr, N_HIGH_MEMORY)
2608 seq_printf(m, " N%u=%u", nr, counters[nr]);
2612 static int s_show(struct seq_file *m, void *p)
2614 struct vmap_area *va = p;
2615 struct vm_struct *v;
2617 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2620 if (!(va->flags & VM_VM_AREA)) {
2621 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2622 (void *)va->va_start, (void *)va->va_end,
2623 va->va_end - va->va_start);
2629 seq_printf(m, "0x%pK-0x%pK %7ld",
2630 v->addr, v->addr + v->size, v->size);
2633 seq_printf(m, " %pS", v->caller);
2636 seq_printf(m, " pages=%d", v->nr_pages);
2639 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2641 if (v->flags & VM_IOREMAP)
2642 seq_printf(m, " ioremap");
2644 if (v->flags & VM_ALLOC)
2645 seq_printf(m, " vmalloc");
2647 if (v->flags & VM_MAP)
2648 seq_printf(m, " vmap");
2650 if (v->flags & VM_USERMAP)
2651 seq_printf(m, " user");
2653 if (v->flags & VM_VPAGES)
2654 seq_printf(m, " vpages");
2656 show_numa_info(m, v);
2661 static const struct seq_operations vmalloc_op = {
2668 static int vmalloc_open(struct inode *inode, struct file *file)
2670 unsigned int *ptr = NULL;
2673 if (IS_ENABLED(CONFIG_NUMA)) {
2674 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2678 ret = seq_open(file, &vmalloc_op);
2680 struct seq_file *m = file->private_data;
2687 static const struct file_operations proc_vmalloc_operations = {
2688 .open = vmalloc_open,
2690 .llseek = seq_lseek,
2691 .release = seq_release_private,
2694 static int __init proc_vmalloc_init(void)
2696 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2699 module_init(proc_vmalloc_init);
2701 void get_vmalloc_info(struct vmalloc_info *vmi)
2703 struct vmap_area *va;
2704 unsigned long free_area_size;
2705 unsigned long prev_end;
2708 vmi->largest_chunk = 0;
2710 prev_end = VMALLOC_START;
2712 spin_lock(&vmap_area_lock);
2714 if (list_empty(&vmap_area_list)) {
2715 vmi->largest_chunk = VMALLOC_TOTAL;
2719 list_for_each_entry(va, &vmap_area_list, list) {
2720 unsigned long addr = va->va_start;
2723 * Some archs keep another range for modules in vmalloc space
2725 if (addr < VMALLOC_START)
2727 if (addr >= VMALLOC_END)
2730 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2733 vmi->used += (va->va_end - va->va_start);
2735 free_area_size = addr - prev_end;
2736 if (vmi->largest_chunk < free_area_size)
2737 vmi->largest_chunk = free_area_size;
2739 prev_end = va->va_end;
2742 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2743 vmi->largest_chunk = VMALLOC_END - prev_end;
2746 spin_unlock(&vmap_area_lock);