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 /* Export for kexec only */
265 LIST_HEAD(vmap_area_list);
266 static struct rb_root vmap_area_root = RB_ROOT;
268 /* The vmap cache globals are protected by vmap_area_lock */
269 static struct rb_node *free_vmap_cache;
270 static unsigned long cached_hole_size;
271 static unsigned long cached_vstart;
272 static unsigned long cached_align;
274 static unsigned long vmap_area_pcpu_hole;
276 /*** Old vmalloc interfaces ***/
277 static DEFINE_RWLOCK(vmlist_lock);
278 static struct vm_struct *vmlist;
280 static struct vmap_area *__find_vmap_area(unsigned long addr)
282 struct rb_node *n = vmap_area_root.rb_node;
285 struct vmap_area *va;
287 va = rb_entry(n, struct vmap_area, rb_node);
288 if (addr < va->va_start)
290 else if (addr > va->va_start)
299 static void __insert_vmap_area(struct vmap_area *va)
301 struct rb_node **p = &vmap_area_root.rb_node;
302 struct rb_node *parent = NULL;
306 struct vmap_area *tmp_va;
309 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
310 if (va->va_start < tmp_va->va_end)
312 else if (va->va_end > tmp_va->va_start)
318 rb_link_node(&va->rb_node, parent, p);
319 rb_insert_color(&va->rb_node, &vmap_area_root);
321 /* address-sort this list so it is usable like the vmlist */
322 tmp = rb_prev(&va->rb_node);
324 struct vmap_area *prev;
325 prev = rb_entry(tmp, struct vmap_area, rb_node);
326 list_add_rcu(&va->list, &prev->list);
328 list_add_rcu(&va->list, &vmap_area_list);
331 static void purge_vmap_area_lazy(void);
334 * Allocate a region of KVA of the specified size and alignment, within the
337 static struct vmap_area *alloc_vmap_area(unsigned long size,
339 unsigned long vstart, unsigned long vend,
340 int node, gfp_t gfp_mask)
342 struct vmap_area *va;
346 struct vmap_area *first;
349 BUG_ON(size & ~PAGE_MASK);
350 BUG_ON(!is_power_of_2(align));
352 va = kmalloc_node(sizeof(struct vmap_area),
353 gfp_mask & GFP_RECLAIM_MASK, node);
355 return ERR_PTR(-ENOMEM);
358 spin_lock(&vmap_area_lock);
360 * Invalidate cache if we have more permissive parameters.
361 * cached_hole_size notes the largest hole noticed _below_
362 * the vmap_area cached in free_vmap_cache: if size fits
363 * into that hole, we want to scan from vstart to reuse
364 * the hole instead of allocating above free_vmap_cache.
365 * Note that __free_vmap_area may update free_vmap_cache
366 * without updating cached_hole_size or cached_align.
368 if (!free_vmap_cache ||
369 size < cached_hole_size ||
370 vstart < cached_vstart ||
371 align < cached_align) {
373 cached_hole_size = 0;
374 free_vmap_cache = NULL;
376 /* record if we encounter less permissive parameters */
377 cached_vstart = vstart;
378 cached_align = align;
380 /* find starting point for our search */
381 if (free_vmap_cache) {
382 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
383 addr = ALIGN(first->va_end, align);
386 if (addr + size - 1 < addr)
390 addr = ALIGN(vstart, align);
391 if (addr + size - 1 < addr)
394 n = vmap_area_root.rb_node;
398 struct vmap_area *tmp;
399 tmp = rb_entry(n, struct vmap_area, rb_node);
400 if (tmp->va_end >= addr) {
402 if (tmp->va_start <= addr)
413 /* from the starting point, walk areas until a suitable hole is found */
414 while (addr + size > first->va_start && addr + size <= vend) {
415 if (addr + cached_hole_size < first->va_start)
416 cached_hole_size = first->va_start - addr;
417 addr = ALIGN(first->va_end, align);
418 if (addr + size - 1 < addr)
421 if (list_is_last(&first->list, &vmap_area_list))
424 first = list_entry(first->list.next,
425 struct vmap_area, list);
429 if (addr + size > vend)
433 va->va_end = addr + size;
435 __insert_vmap_area(va);
436 free_vmap_cache = &va->rb_node;
437 spin_unlock(&vmap_area_lock);
439 BUG_ON(va->va_start & (align-1));
440 BUG_ON(va->va_start < vstart);
441 BUG_ON(va->va_end > vend);
446 spin_unlock(&vmap_area_lock);
448 purge_vmap_area_lazy();
452 if (printk_ratelimit())
454 "vmap allocation for size %lu failed: "
455 "use vmalloc=<size> to increase size.\n", size);
457 return ERR_PTR(-EBUSY);
460 static void __free_vmap_area(struct vmap_area *va)
462 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
464 if (free_vmap_cache) {
465 if (va->va_end < cached_vstart) {
466 free_vmap_cache = NULL;
468 struct vmap_area *cache;
469 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
470 if (va->va_start <= cache->va_start) {
471 free_vmap_cache = rb_prev(&va->rb_node);
473 * We don't try to update cached_hole_size or
474 * cached_align, but it won't go very wrong.
479 rb_erase(&va->rb_node, &vmap_area_root);
480 RB_CLEAR_NODE(&va->rb_node);
481 list_del_rcu(&va->list);
484 * Track the highest possible candidate for pcpu area
485 * allocation. Areas outside of vmalloc area can be returned
486 * here too, consider only end addresses which fall inside
487 * vmalloc area proper.
489 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
490 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
492 kfree_rcu(va, rcu_head);
496 * Free a region of KVA allocated by alloc_vmap_area
498 static void free_vmap_area(struct vmap_area *va)
500 spin_lock(&vmap_area_lock);
501 __free_vmap_area(va);
502 spin_unlock(&vmap_area_lock);
506 * Clear the pagetable entries of a given vmap_area
508 static void unmap_vmap_area(struct vmap_area *va)
510 vunmap_page_range(va->va_start, va->va_end);
513 static void vmap_debug_free_range(unsigned long start, unsigned long end)
516 * Unmap page tables and force a TLB flush immediately if
517 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
518 * bugs similarly to those in linear kernel virtual address
519 * space after a page has been freed.
521 * All the lazy freeing logic is still retained, in order to
522 * minimise intrusiveness of this debugging feature.
524 * This is going to be *slow* (linear kernel virtual address
525 * debugging doesn't do a broadcast TLB flush so it is a lot
528 #ifdef CONFIG_DEBUG_PAGEALLOC
529 vunmap_page_range(start, end);
530 flush_tlb_kernel_range(start, end);
535 * lazy_max_pages is the maximum amount of virtual address space we gather up
536 * before attempting to purge with a TLB flush.
538 * There is a tradeoff here: a larger number will cover more kernel page tables
539 * and take slightly longer to purge, but it will linearly reduce the number of
540 * global TLB flushes that must be performed. It would seem natural to scale
541 * this number up linearly with the number of CPUs (because vmapping activity
542 * could also scale linearly with the number of CPUs), however it is likely
543 * that in practice, workloads might be constrained in other ways that mean
544 * vmap activity will not scale linearly with CPUs. Also, I want to be
545 * conservative and not introduce a big latency on huge systems, so go with
546 * a less aggressive log scale. It will still be an improvement over the old
547 * code, and it will be simple to change the scale factor if we find that it
548 * becomes a problem on bigger systems.
550 static unsigned long lazy_max_pages(void)
554 log = fls(num_online_cpus());
556 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
559 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
561 /* for per-CPU blocks */
562 static void purge_fragmented_blocks_allcpus(void);
565 * called before a call to iounmap() if the caller wants vm_area_struct's
568 void set_iounmap_nonlazy(void)
570 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
574 * Purges all lazily-freed vmap areas.
576 * If sync is 0 then don't purge if there is already a purge in progress.
577 * If force_flush is 1, then flush kernel TLBs between *start and *end even
578 * if we found no lazy vmap areas to unmap (callers can use this to optimise
579 * their own TLB flushing).
580 * Returns with *start = min(*start, lowest purged address)
581 * *end = max(*end, highest purged address)
583 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
584 int sync, int force_flush)
586 static DEFINE_SPINLOCK(purge_lock);
588 struct vmap_area *va;
589 struct vmap_area *n_va;
593 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
594 * should not expect such behaviour. This just simplifies locking for
595 * the case that isn't actually used at the moment anyway.
597 if (!sync && !force_flush) {
598 if (!spin_trylock(&purge_lock))
601 spin_lock(&purge_lock);
604 purge_fragmented_blocks_allcpus();
607 list_for_each_entry_rcu(va, &vmap_area_list, list) {
608 if (va->flags & VM_LAZY_FREE) {
609 if (va->va_start < *start)
610 *start = va->va_start;
611 if (va->va_end > *end)
613 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
614 list_add_tail(&va->purge_list, &valist);
615 va->flags |= VM_LAZY_FREEING;
616 va->flags &= ~VM_LAZY_FREE;
622 atomic_sub(nr, &vmap_lazy_nr);
624 if (nr || force_flush)
625 flush_tlb_kernel_range(*start, *end);
628 spin_lock(&vmap_area_lock);
629 list_for_each_entry_safe(va, n_va, &valist, purge_list)
630 __free_vmap_area(va);
631 spin_unlock(&vmap_area_lock);
633 spin_unlock(&purge_lock);
637 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
638 * is already purging.
640 static void try_purge_vmap_area_lazy(void)
642 unsigned long start = ULONG_MAX, end = 0;
644 __purge_vmap_area_lazy(&start, &end, 0, 0);
648 * Kick off a purge of the outstanding lazy areas.
650 static void purge_vmap_area_lazy(void)
652 unsigned long start = ULONG_MAX, end = 0;
654 __purge_vmap_area_lazy(&start, &end, 1, 0);
658 * Free a vmap area, caller ensuring that the area has been unmapped
659 * and flush_cache_vunmap had been called for the correct range
662 static void free_vmap_area_noflush(struct vmap_area *va)
664 va->flags |= VM_LAZY_FREE;
665 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
666 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
667 try_purge_vmap_area_lazy();
671 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
672 * called for the correct range previously.
674 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
677 free_vmap_area_noflush(va);
681 * Free and unmap a vmap area
683 static void free_unmap_vmap_area(struct vmap_area *va)
685 flush_cache_vunmap(va->va_start, va->va_end);
686 free_unmap_vmap_area_noflush(va);
689 static struct vmap_area *find_vmap_area(unsigned long addr)
691 struct vmap_area *va;
693 spin_lock(&vmap_area_lock);
694 va = __find_vmap_area(addr);
695 spin_unlock(&vmap_area_lock);
700 static void free_unmap_vmap_area_addr(unsigned long addr)
702 struct vmap_area *va;
704 va = find_vmap_area(addr);
706 free_unmap_vmap_area(va);
710 /*** Per cpu kva allocator ***/
713 * vmap space is limited especially on 32 bit architectures. Ensure there is
714 * room for at least 16 percpu vmap blocks per CPU.
717 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
718 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
719 * instead (we just need a rough idea)
721 #if BITS_PER_LONG == 32
722 #define VMALLOC_SPACE (128UL*1024*1024)
724 #define VMALLOC_SPACE (128UL*1024*1024*1024)
727 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
728 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
729 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
730 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
731 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
732 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
733 #define VMAP_BBMAP_BITS \
734 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
735 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
736 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
738 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
740 static bool vmap_initialized __read_mostly = false;
742 struct vmap_block_queue {
744 struct list_head free;
749 struct vmap_area *va;
750 struct vmap_block_queue *vbq;
751 unsigned long free, dirty;
752 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
753 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
754 struct list_head free_list;
755 struct rcu_head rcu_head;
756 struct list_head purge;
759 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
760 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
763 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
764 * in the free path. Could get rid of this if we change the API to return a
765 * "cookie" from alloc, to be passed to free. But no big deal yet.
767 static DEFINE_SPINLOCK(vmap_block_tree_lock);
768 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
771 * We should probably have a fallback mechanism to allocate virtual memory
772 * out of partially filled vmap blocks. However vmap block sizing should be
773 * fairly reasonable according to the vmalloc size, so it shouldn't be a
777 static unsigned long addr_to_vb_idx(unsigned long addr)
779 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
780 addr /= VMAP_BLOCK_SIZE;
784 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
786 struct vmap_block_queue *vbq;
787 struct vmap_block *vb;
788 struct vmap_area *va;
789 unsigned long vb_idx;
792 node = numa_node_id();
794 vb = kmalloc_node(sizeof(struct vmap_block),
795 gfp_mask & GFP_RECLAIM_MASK, node);
797 return ERR_PTR(-ENOMEM);
799 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
800 VMALLOC_START, VMALLOC_END,
807 err = radix_tree_preload(gfp_mask);
814 spin_lock_init(&vb->lock);
816 vb->free = VMAP_BBMAP_BITS;
818 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
819 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
820 INIT_LIST_HEAD(&vb->free_list);
822 vb_idx = addr_to_vb_idx(va->va_start);
823 spin_lock(&vmap_block_tree_lock);
824 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
825 spin_unlock(&vmap_block_tree_lock);
827 radix_tree_preload_end();
829 vbq = &get_cpu_var(vmap_block_queue);
831 spin_lock(&vbq->lock);
832 list_add_rcu(&vb->free_list, &vbq->free);
833 spin_unlock(&vbq->lock);
834 put_cpu_var(vmap_block_queue);
839 static void free_vmap_block(struct vmap_block *vb)
841 struct vmap_block *tmp;
842 unsigned long vb_idx;
844 vb_idx = addr_to_vb_idx(vb->va->va_start);
845 spin_lock(&vmap_block_tree_lock);
846 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
847 spin_unlock(&vmap_block_tree_lock);
850 free_vmap_area_noflush(vb->va);
851 kfree_rcu(vb, rcu_head);
854 static void purge_fragmented_blocks(int cpu)
857 struct vmap_block *vb;
858 struct vmap_block *n_vb;
859 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
862 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
864 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
867 spin_lock(&vb->lock);
868 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
869 vb->free = 0; /* prevent further allocs after releasing lock */
870 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
871 bitmap_fill(vb->alloc_map, VMAP_BBMAP_BITS);
872 bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
873 spin_lock(&vbq->lock);
874 list_del_rcu(&vb->free_list);
875 spin_unlock(&vbq->lock);
876 spin_unlock(&vb->lock);
877 list_add_tail(&vb->purge, &purge);
879 spin_unlock(&vb->lock);
883 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
884 list_del(&vb->purge);
889 static void purge_fragmented_blocks_thiscpu(void)
891 purge_fragmented_blocks(smp_processor_id());
894 static void purge_fragmented_blocks_allcpus(void)
898 for_each_possible_cpu(cpu)
899 purge_fragmented_blocks(cpu);
902 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
904 struct vmap_block_queue *vbq;
905 struct vmap_block *vb;
906 unsigned long addr = 0;
910 BUG_ON(size & ~PAGE_MASK);
911 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
912 if (WARN_ON(size == 0)) {
914 * Allocating 0 bytes isn't what caller wants since
915 * get_order(0) returns funny result. Just warn and terminate
920 order = get_order(size);
924 vbq = &get_cpu_var(vmap_block_queue);
925 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
928 spin_lock(&vb->lock);
929 if (vb->free < 1UL << order)
932 i = bitmap_find_free_region(vb->alloc_map,
933 VMAP_BBMAP_BITS, order);
936 if (vb->free + vb->dirty == VMAP_BBMAP_BITS) {
937 /* fragmented and no outstanding allocations */
938 BUG_ON(vb->dirty != VMAP_BBMAP_BITS);
943 addr = vb->va->va_start + (i << PAGE_SHIFT);
944 BUG_ON(addr_to_vb_idx(addr) !=
945 addr_to_vb_idx(vb->va->va_start));
946 vb->free -= 1UL << order;
948 spin_lock(&vbq->lock);
949 list_del_rcu(&vb->free_list);
950 spin_unlock(&vbq->lock);
952 spin_unlock(&vb->lock);
955 spin_unlock(&vb->lock);
959 purge_fragmented_blocks_thiscpu();
961 put_cpu_var(vmap_block_queue);
965 vb = new_vmap_block(gfp_mask);
974 static void vb_free(const void *addr, unsigned long size)
976 unsigned long offset;
977 unsigned long vb_idx;
979 struct vmap_block *vb;
981 BUG_ON(size & ~PAGE_MASK);
982 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
984 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
986 order = get_order(size);
988 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
990 vb_idx = addr_to_vb_idx((unsigned long)addr);
992 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
996 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
998 spin_lock(&vb->lock);
999 BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
1001 vb->dirty += 1UL << order;
1002 if (vb->dirty == VMAP_BBMAP_BITS) {
1004 spin_unlock(&vb->lock);
1005 free_vmap_block(vb);
1007 spin_unlock(&vb->lock);
1011 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1013 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1014 * to amortize TLB flushing overheads. What this means is that any page you
1015 * have now, may, in a former life, have been mapped into kernel virtual
1016 * address by the vmap layer and so there might be some CPUs with TLB entries
1017 * still referencing that page (additional to the regular 1:1 kernel mapping).
1019 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1020 * be sure that none of the pages we have control over will have any aliases
1021 * from the vmap layer.
1023 void vm_unmap_aliases(void)
1025 unsigned long start = ULONG_MAX, end = 0;
1029 if (unlikely(!vmap_initialized))
1032 for_each_possible_cpu(cpu) {
1033 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1034 struct vmap_block *vb;
1037 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1040 spin_lock(&vb->lock);
1041 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1042 while (i < VMAP_BBMAP_BITS) {
1045 j = find_next_zero_bit(vb->dirty_map,
1046 VMAP_BBMAP_BITS, i);
1048 s = vb->va->va_start + (i << PAGE_SHIFT);
1049 e = vb->va->va_start + (j << PAGE_SHIFT);
1058 i = find_next_bit(vb->dirty_map,
1059 VMAP_BBMAP_BITS, i);
1061 spin_unlock(&vb->lock);
1066 __purge_vmap_area_lazy(&start, &end, 1, flush);
1068 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1071 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1072 * @mem: the pointer returned by vm_map_ram
1073 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1075 void vm_unmap_ram(const void *mem, unsigned int count)
1077 unsigned long size = count << PAGE_SHIFT;
1078 unsigned long addr = (unsigned long)mem;
1081 BUG_ON(addr < VMALLOC_START);
1082 BUG_ON(addr > VMALLOC_END);
1083 BUG_ON(addr & (PAGE_SIZE-1));
1085 debug_check_no_locks_freed(mem, size);
1086 vmap_debug_free_range(addr, addr+size);
1088 if (likely(count <= VMAP_MAX_ALLOC))
1091 free_unmap_vmap_area_addr(addr);
1093 EXPORT_SYMBOL(vm_unmap_ram);
1096 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1097 * @pages: an array of pointers to the pages to be mapped
1098 * @count: number of pages
1099 * @node: prefer to allocate data structures on this node
1100 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1102 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1104 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1106 unsigned long size = count << PAGE_SHIFT;
1110 if (likely(count <= VMAP_MAX_ALLOC)) {
1111 mem = vb_alloc(size, GFP_KERNEL);
1114 addr = (unsigned long)mem;
1116 struct vmap_area *va;
1117 va = alloc_vmap_area(size, PAGE_SIZE,
1118 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1122 addr = va->va_start;
1125 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1126 vm_unmap_ram(mem, count);
1131 EXPORT_SYMBOL(vm_map_ram);
1134 * vm_area_add_early - add vmap area early during boot
1135 * @vm: vm_struct to add
1137 * This function is used to add fixed kernel vm area to vmlist before
1138 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1139 * should contain proper values and the other fields should be zero.
1141 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1143 void __init vm_area_add_early(struct vm_struct *vm)
1145 struct vm_struct *tmp, **p;
1147 BUG_ON(vmap_initialized);
1148 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1149 if (tmp->addr >= vm->addr) {
1150 BUG_ON(tmp->addr < vm->addr + vm->size);
1153 BUG_ON(tmp->addr + tmp->size > vm->addr);
1160 * vm_area_register_early - register vmap area early during boot
1161 * @vm: vm_struct to register
1162 * @align: requested alignment
1164 * This function is used to register kernel vm area before
1165 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1166 * proper values on entry and other fields should be zero. On return,
1167 * vm->addr contains the allocated address.
1169 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1171 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1173 static size_t vm_init_off __initdata;
1176 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1177 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1179 vm->addr = (void *)addr;
1181 vm_area_add_early(vm);
1184 void __init vmalloc_init(void)
1186 struct vmap_area *va;
1187 struct vm_struct *tmp;
1190 for_each_possible_cpu(i) {
1191 struct vmap_block_queue *vbq;
1193 vbq = &per_cpu(vmap_block_queue, i);
1194 spin_lock_init(&vbq->lock);
1195 INIT_LIST_HEAD(&vbq->free);
1198 /* Import existing vmlist entries. */
1199 for (tmp = vmlist; tmp; tmp = tmp->next) {
1200 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1201 va->flags = VM_VM_AREA;
1202 va->va_start = (unsigned long)tmp->addr;
1203 va->va_end = va->va_start + tmp->size;
1205 __insert_vmap_area(va);
1208 vmap_area_pcpu_hole = VMALLOC_END;
1210 vmap_initialized = true;
1214 * map_kernel_range_noflush - map kernel VM area with the specified pages
1215 * @addr: start of the VM area to map
1216 * @size: size of the VM area to map
1217 * @prot: page protection flags to use
1218 * @pages: pages to map
1220 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1221 * specify should have been allocated using get_vm_area() and its
1225 * This function does NOT do any cache flushing. The caller is
1226 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1227 * before calling this function.
1230 * The number of pages mapped on success, -errno on failure.
1232 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1233 pgprot_t prot, struct page **pages)
1235 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1239 * unmap_kernel_range_noflush - unmap kernel VM area
1240 * @addr: start of the VM area to unmap
1241 * @size: size of the VM area to unmap
1243 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1244 * specify should have been allocated using get_vm_area() and its
1248 * This function does NOT do any cache flushing. The caller is
1249 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1250 * before calling this function and flush_tlb_kernel_range() after.
1252 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1254 vunmap_page_range(addr, addr + size);
1256 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1259 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1260 * @addr: start of the VM area to unmap
1261 * @size: size of the VM area to unmap
1263 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1264 * the unmapping and tlb after.
1266 void unmap_kernel_range(unsigned long addr, unsigned long size)
1268 unsigned long end = addr + size;
1270 flush_cache_vunmap(addr, end);
1271 vunmap_page_range(addr, end);
1272 flush_tlb_kernel_range(addr, end);
1275 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1277 unsigned long addr = (unsigned long)area->addr;
1278 unsigned long end = addr + area->size - PAGE_SIZE;
1281 err = vmap_page_range(addr, end, prot, *pages);
1289 EXPORT_SYMBOL_GPL(map_vm_area);
1291 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1292 unsigned long flags, const void *caller)
1294 spin_lock(&vmap_area_lock);
1296 vm->addr = (void *)va->va_start;
1297 vm->size = va->va_end - va->va_start;
1298 vm->caller = caller;
1300 va->flags |= VM_VM_AREA;
1301 spin_unlock(&vmap_area_lock);
1304 static void insert_vmalloc_vmlist(struct vm_struct *vm)
1306 struct vm_struct *tmp, **p;
1309 * Before removing VM_UNLIST,
1310 * we should make sure that vm has proper values.
1311 * Pair with smp_rmb() in show_numa_info().
1314 vm->flags &= ~VM_UNLIST;
1316 write_lock(&vmlist_lock);
1317 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1318 if (tmp->addr >= vm->addr)
1323 write_unlock(&vmlist_lock);
1326 static void insert_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1327 unsigned long flags, const void *caller)
1329 setup_vmalloc_vm(vm, va, flags, caller);
1330 insert_vmalloc_vmlist(vm);
1333 static struct vm_struct *__get_vm_area_node(unsigned long size,
1334 unsigned long align, unsigned long flags, unsigned long start,
1335 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1337 struct vmap_area *va;
1338 struct vm_struct *area;
1340 BUG_ON(in_interrupt());
1341 if (flags & VM_IOREMAP) {
1342 int bit = fls(size);
1344 if (bit > IOREMAP_MAX_ORDER)
1345 bit = IOREMAP_MAX_ORDER;
1346 else if (bit < PAGE_SHIFT)
1352 size = PAGE_ALIGN(size);
1353 if (unlikely(!size))
1356 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1357 if (unlikely(!area))
1361 * We always allocate a guard page.
1365 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1372 * When this function is called from __vmalloc_node_range,
1373 * we do not add vm_struct to vmlist here to avoid
1374 * accessing uninitialized members of vm_struct such as
1375 * pages and nr_pages fields. They will be set later.
1376 * To distinguish it from others, we use a VM_UNLIST flag.
1378 if (flags & VM_UNLIST)
1379 setup_vmalloc_vm(area, va, flags, caller);
1381 insert_vmalloc_vm(area, va, flags, caller);
1386 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1387 unsigned long start, unsigned long end)
1389 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1390 GFP_KERNEL, __builtin_return_address(0));
1392 EXPORT_SYMBOL_GPL(__get_vm_area);
1394 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1395 unsigned long start, unsigned long end,
1398 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1399 GFP_KERNEL, caller);
1403 * get_vm_area - reserve a contiguous kernel virtual area
1404 * @size: size of the area
1405 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1407 * Search an area of @size in the kernel virtual mapping area,
1408 * and reserved it for out purposes. Returns the area descriptor
1409 * on success or %NULL on failure.
1411 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1413 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1414 NUMA_NO_NODE, GFP_KERNEL,
1415 __builtin_return_address(0));
1418 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1421 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1422 NUMA_NO_NODE, GFP_KERNEL, caller);
1426 * find_vm_area - find a continuous kernel virtual area
1427 * @addr: base address
1429 * Search for the kernel VM area starting at @addr, and return it.
1430 * It is up to the caller to do all required locking to keep the returned
1433 struct vm_struct *find_vm_area(const void *addr)
1435 struct vmap_area *va;
1437 va = find_vmap_area((unsigned long)addr);
1438 if (va && va->flags & VM_VM_AREA)
1445 * remove_vm_area - find and remove a continuous kernel virtual area
1446 * @addr: base address
1448 * Search for the kernel VM area starting at @addr, and remove it.
1449 * This function returns the found VM area, but using it is NOT safe
1450 * on SMP machines, except for its size or flags.
1452 struct vm_struct *remove_vm_area(const void *addr)
1454 struct vmap_area *va;
1456 va = find_vmap_area((unsigned long)addr);
1457 if (va && va->flags & VM_VM_AREA) {
1458 struct vm_struct *vm = va->vm;
1460 spin_lock(&vmap_area_lock);
1462 va->flags &= ~VM_VM_AREA;
1463 spin_unlock(&vmap_area_lock);
1465 if (!(vm->flags & VM_UNLIST)) {
1466 struct vm_struct *tmp, **p;
1468 * remove from list and disallow access to
1469 * this vm_struct before unmap. (address range
1470 * confliction is maintained by vmap.)
1472 write_lock(&vmlist_lock);
1473 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1476 write_unlock(&vmlist_lock);
1479 vmap_debug_free_range(va->va_start, va->va_end);
1480 free_unmap_vmap_area(va);
1481 vm->size -= PAGE_SIZE;
1488 static void __vunmap(const void *addr, int deallocate_pages)
1490 struct vm_struct *area;
1495 if ((PAGE_SIZE-1) & (unsigned long)addr) {
1496 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1500 area = remove_vm_area(addr);
1501 if (unlikely(!area)) {
1502 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1507 debug_check_no_locks_freed(addr, area->size);
1508 debug_check_no_obj_freed(addr, area->size);
1510 if (deallocate_pages) {
1513 for (i = 0; i < area->nr_pages; i++) {
1514 struct page *page = area->pages[i];
1520 if (area->flags & VM_VPAGES)
1531 * vfree - release memory allocated by vmalloc()
1532 * @addr: memory base address
1534 * Free the virtually continuous memory area starting at @addr, as
1535 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1536 * NULL, no operation is performed.
1538 * Must not be called in interrupt context.
1540 void vfree(const void *addr)
1542 BUG_ON(in_interrupt());
1544 kmemleak_free(addr);
1548 EXPORT_SYMBOL(vfree);
1551 * vunmap - release virtual mapping obtained by vmap()
1552 * @addr: memory base address
1554 * Free the virtually contiguous memory area starting at @addr,
1555 * which was created from the page array passed to vmap().
1557 * Must not be called in interrupt context.
1559 void vunmap(const void *addr)
1561 BUG_ON(in_interrupt());
1565 EXPORT_SYMBOL(vunmap);
1568 * vmap - map an array of pages into virtually contiguous space
1569 * @pages: array of page pointers
1570 * @count: number of pages to map
1571 * @flags: vm_area->flags
1572 * @prot: page protection for the mapping
1574 * Maps @count pages from @pages into contiguous kernel virtual
1577 void *vmap(struct page **pages, unsigned int count,
1578 unsigned long flags, pgprot_t prot)
1580 struct vm_struct *area;
1584 if (count > totalram_pages)
1587 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1588 __builtin_return_address(0));
1592 if (map_vm_area(area, prot, &pages)) {
1599 EXPORT_SYMBOL(vmap);
1601 static void *__vmalloc_node(unsigned long size, unsigned long align,
1602 gfp_t gfp_mask, pgprot_t prot,
1603 int node, const void *caller);
1604 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1605 pgprot_t prot, int node, const void *caller)
1607 const int order = 0;
1608 struct page **pages;
1609 unsigned int nr_pages, array_size, i;
1610 gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1612 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1613 array_size = (nr_pages * sizeof(struct page *));
1615 area->nr_pages = nr_pages;
1616 /* Please note that the recursion is strictly bounded. */
1617 if (array_size > PAGE_SIZE) {
1618 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1619 PAGE_KERNEL, node, caller);
1620 area->flags |= VM_VPAGES;
1622 pages = kmalloc_node(array_size, nested_gfp, node);
1624 area->pages = pages;
1625 area->caller = caller;
1627 remove_vm_area(area->addr);
1632 for (i = 0; i < area->nr_pages; i++) {
1634 gfp_t tmp_mask = gfp_mask | __GFP_NOWARN;
1637 page = alloc_page(tmp_mask);
1639 page = alloc_pages_node(node, tmp_mask, order);
1641 if (unlikely(!page)) {
1642 /* Successfully allocated i pages, free them in __vunmap() */
1646 area->pages[i] = page;
1649 if (map_vm_area(area, prot, &pages))
1654 warn_alloc_failed(gfp_mask, order,
1655 "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1656 (area->nr_pages*PAGE_SIZE), area->size);
1662 * __vmalloc_node_range - allocate virtually contiguous memory
1663 * @size: allocation size
1664 * @align: desired alignment
1665 * @start: vm area range start
1666 * @end: vm area range end
1667 * @gfp_mask: flags for the page level allocator
1668 * @prot: protection mask for the allocated pages
1669 * @node: node to use for allocation or NUMA_NO_NODE
1670 * @caller: caller's return address
1672 * Allocate enough pages to cover @size from the page level
1673 * allocator with @gfp_mask flags. Map them into contiguous
1674 * kernel virtual space, using a pagetable protection of @prot.
1676 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1677 unsigned long start, unsigned long end, gfp_t gfp_mask,
1678 pgprot_t prot, int node, const void *caller)
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_UNLIST,
1689 start, end, node, gfp_mask, caller);
1693 addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1698 * In this function, newly allocated vm_struct is not added
1699 * to vmlist at __get_vm_area_node(). so, it is added here.
1701 insert_vmalloc_vmlist(area);
1704 * A ref_count = 3 is needed because the vm_struct and vmap_area
1705 * structures allocated in the __get_vm_area_node() function contain
1706 * references to the virtual address of the vmalloc'ed block.
1708 kmemleak_alloc(addr, real_size, 3, gfp_mask);
1713 warn_alloc_failed(gfp_mask, 0,
1714 "vmalloc: allocation failure: %lu bytes\n",
1720 * __vmalloc_node - allocate virtually contiguous memory
1721 * @size: allocation size
1722 * @align: desired alignment
1723 * @gfp_mask: flags for the page level allocator
1724 * @prot: protection mask for the allocated pages
1725 * @node: node to use for allocation or NUMA_NO_NODE
1726 * @caller: caller's return address
1728 * Allocate enough pages to cover @size from the page level
1729 * allocator with @gfp_mask flags. Map them into contiguous
1730 * kernel virtual space, using a pagetable protection of @prot.
1732 static void *__vmalloc_node(unsigned long size, unsigned long align,
1733 gfp_t gfp_mask, pgprot_t prot,
1734 int node, const void *caller)
1736 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1737 gfp_mask, prot, node, caller);
1740 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1742 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1743 __builtin_return_address(0));
1745 EXPORT_SYMBOL(__vmalloc);
1747 static inline void *__vmalloc_node_flags(unsigned long size,
1748 int node, gfp_t flags)
1750 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1751 node, __builtin_return_address(0));
1755 * vmalloc - allocate virtually contiguous memory
1756 * @size: allocation size
1757 * Allocate enough pages to cover @size from the page level
1758 * allocator and map them into contiguous kernel virtual space.
1760 * For tight control over page level allocator and protection flags
1761 * use __vmalloc() instead.
1763 void *vmalloc(unsigned long size)
1765 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1766 GFP_KERNEL | __GFP_HIGHMEM);
1768 EXPORT_SYMBOL(vmalloc);
1771 * vzalloc - allocate virtually contiguous memory with zero fill
1772 * @size: allocation size
1773 * Allocate enough pages to cover @size from the page level
1774 * allocator and map them into contiguous kernel virtual space.
1775 * The memory allocated is set to zero.
1777 * For tight control over page level allocator and protection flags
1778 * use __vmalloc() instead.
1780 void *vzalloc(unsigned long size)
1782 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1783 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1785 EXPORT_SYMBOL(vzalloc);
1788 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1789 * @size: allocation size
1791 * The resulting memory area is zeroed so it can be mapped to userspace
1792 * without leaking data.
1794 void *vmalloc_user(unsigned long size)
1796 struct vm_struct *area;
1799 ret = __vmalloc_node(size, SHMLBA,
1800 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1801 PAGE_KERNEL, NUMA_NO_NODE,
1802 __builtin_return_address(0));
1804 area = find_vm_area(ret);
1805 area->flags |= VM_USERMAP;
1809 EXPORT_SYMBOL(vmalloc_user);
1812 * vmalloc_node - allocate memory on a specific node
1813 * @size: allocation size
1816 * Allocate enough pages to cover @size from the page level
1817 * allocator and map them into contiguous kernel virtual space.
1819 * For tight control over page level allocator and protection flags
1820 * use __vmalloc() instead.
1822 void *vmalloc_node(unsigned long size, int node)
1824 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1825 node, __builtin_return_address(0));
1827 EXPORT_SYMBOL(vmalloc_node);
1830 * vzalloc_node - allocate memory on a specific node with zero fill
1831 * @size: allocation size
1834 * Allocate enough pages to cover @size from the page level
1835 * allocator and map them into contiguous kernel virtual space.
1836 * The memory allocated is set to zero.
1838 * For tight control over page level allocator and protection flags
1839 * use __vmalloc_node() instead.
1841 void *vzalloc_node(unsigned long size, int node)
1843 return __vmalloc_node_flags(size, node,
1844 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1846 EXPORT_SYMBOL(vzalloc_node);
1848 #ifndef PAGE_KERNEL_EXEC
1849 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1853 * vmalloc_exec - allocate virtually contiguous, executable memory
1854 * @size: allocation size
1856 * Kernel-internal function to allocate enough pages to cover @size
1857 * the page level allocator and map them into contiguous and
1858 * executable kernel virtual space.
1860 * For tight control over page level allocator and protection flags
1861 * use __vmalloc() instead.
1864 void *vmalloc_exec(unsigned long size)
1866 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1867 NUMA_NO_NODE, __builtin_return_address(0));
1870 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1871 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1872 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1873 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1875 #define GFP_VMALLOC32 GFP_KERNEL
1879 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1880 * @size: allocation size
1882 * Allocate enough 32bit PA addressable pages to cover @size from the
1883 * page level allocator and map them into contiguous kernel virtual space.
1885 void *vmalloc_32(unsigned long size)
1887 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1888 NUMA_NO_NODE, __builtin_return_address(0));
1890 EXPORT_SYMBOL(vmalloc_32);
1893 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1894 * @size: allocation size
1896 * The resulting memory area is 32bit addressable and zeroed so it can be
1897 * mapped to userspace without leaking data.
1899 void *vmalloc_32_user(unsigned long size)
1901 struct vm_struct *area;
1904 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1905 NUMA_NO_NODE, __builtin_return_address(0));
1907 area = find_vm_area(ret);
1908 area->flags |= VM_USERMAP;
1912 EXPORT_SYMBOL(vmalloc_32_user);
1915 * small helper routine , copy contents to buf from addr.
1916 * If the page is not present, fill zero.
1919 static int aligned_vread(char *buf, char *addr, unsigned long count)
1925 unsigned long offset, length;
1927 offset = (unsigned long)addr & ~PAGE_MASK;
1928 length = PAGE_SIZE - offset;
1931 p = vmalloc_to_page(addr);
1933 * To do safe access to this _mapped_ area, we need
1934 * lock. But adding lock here means that we need to add
1935 * overhead of vmalloc()/vfree() calles for this _debug_
1936 * interface, rarely used. Instead of that, we'll use
1937 * kmap() and get small overhead in this access function.
1941 * we can expect USER0 is not used (see vread/vwrite's
1942 * function description)
1944 void *map = kmap_atomic(p);
1945 memcpy(buf, map + offset, length);
1948 memset(buf, 0, length);
1958 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1964 unsigned long offset, length;
1966 offset = (unsigned long)addr & ~PAGE_MASK;
1967 length = PAGE_SIZE - offset;
1970 p = vmalloc_to_page(addr);
1972 * To do safe access to this _mapped_ area, we need
1973 * lock. But adding lock here means that we need to add
1974 * overhead of vmalloc()/vfree() calles for this _debug_
1975 * interface, rarely used. Instead of that, we'll use
1976 * kmap() and get small overhead in this access function.
1980 * we can expect USER0 is not used (see vread/vwrite's
1981 * function description)
1983 void *map = kmap_atomic(p);
1984 memcpy(map + offset, buf, length);
1996 * vread() - read vmalloc area in a safe way.
1997 * @buf: buffer for reading data
1998 * @addr: vm address.
1999 * @count: number of bytes to be read.
2001 * Returns # of bytes which addr and buf should be increased.
2002 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2003 * includes any intersect with alive vmalloc area.
2005 * This function checks that addr is a valid vmalloc'ed area, and
2006 * copy data from that area to a given buffer. If the given memory range
2007 * of [addr...addr+count) includes some valid address, data is copied to
2008 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2009 * IOREMAP area is treated as memory hole and no copy is done.
2011 * If [addr...addr+count) doesn't includes any intersects with alive
2012 * vm_struct area, returns 0. @buf should be kernel's buffer.
2014 * Note: In usual ops, vread() is never necessary because the caller
2015 * should know vmalloc() area is valid and can use memcpy().
2016 * This is for routines which have to access vmalloc area without
2017 * any informaion, as /dev/kmem.
2021 long vread(char *buf, char *addr, unsigned long count)
2023 struct vmap_area *va;
2024 struct vm_struct *vm;
2025 char *vaddr, *buf_start = buf;
2026 unsigned long buflen = count;
2029 /* Don't allow overflow */
2030 if ((unsigned long) addr + count < count)
2031 count = -(unsigned long) addr;
2033 spin_lock(&vmap_area_lock);
2034 list_for_each_entry(va, &vmap_area_list, list) {
2038 if (!(va->flags & VM_VM_AREA))
2042 vaddr = (char *) vm->addr;
2043 if (addr >= vaddr + vm->size - PAGE_SIZE)
2045 while (addr < vaddr) {
2053 n = vaddr + vm->size - PAGE_SIZE - addr;
2056 if (!(vm->flags & VM_IOREMAP))
2057 aligned_vread(buf, addr, n);
2058 else /* IOREMAP area is treated as memory hole */
2065 spin_unlock(&vmap_area_lock);
2067 if (buf == buf_start)
2069 /* zero-fill memory holes */
2070 if (buf != buf_start + buflen)
2071 memset(buf, 0, buflen - (buf - buf_start));
2077 * vwrite() - write vmalloc area in a safe way.
2078 * @buf: buffer for source data
2079 * @addr: vm address.
2080 * @count: number of bytes to be read.
2082 * Returns # of bytes which addr and buf should be incresed.
2083 * (same number to @count).
2084 * If [addr...addr+count) doesn't includes any intersect with valid
2085 * vmalloc area, returns 0.
2087 * This function checks that addr is a valid vmalloc'ed area, and
2088 * copy data from a buffer to the given addr. If specified range of
2089 * [addr...addr+count) includes some valid address, data is copied from
2090 * proper area of @buf. If there are memory holes, no copy to hole.
2091 * IOREMAP area is treated as memory hole and no copy is done.
2093 * If [addr...addr+count) doesn't includes any intersects with alive
2094 * vm_struct area, returns 0. @buf should be kernel's buffer.
2096 * Note: In usual ops, vwrite() is never necessary because the caller
2097 * should know vmalloc() area is valid and can use memcpy().
2098 * This is for routines which have to access vmalloc area without
2099 * any informaion, as /dev/kmem.
2102 long vwrite(char *buf, char *addr, unsigned long count)
2104 struct vmap_area *va;
2105 struct vm_struct *vm;
2107 unsigned long n, buflen;
2110 /* Don't allow overflow */
2111 if ((unsigned long) addr + count < count)
2112 count = -(unsigned long) addr;
2115 spin_lock(&vmap_area_lock);
2116 list_for_each_entry(va, &vmap_area_list, list) {
2120 if (!(va->flags & VM_VM_AREA))
2124 vaddr = (char *) vm->addr;
2125 if (addr >= vaddr + vm->size - PAGE_SIZE)
2127 while (addr < vaddr) {
2134 n = vaddr + vm->size - PAGE_SIZE - addr;
2137 if (!(vm->flags & VM_IOREMAP)) {
2138 aligned_vwrite(buf, addr, n);
2146 spin_unlock(&vmap_area_lock);
2153 * remap_vmalloc_range - map vmalloc pages to userspace
2154 * @vma: vma to cover (map full range of vma)
2155 * @addr: vmalloc memory
2156 * @pgoff: number of pages into addr before first page to map
2158 * Returns: 0 for success, -Exxx on failure
2160 * This function checks that addr is a valid vmalloc'ed area, and
2161 * that it is big enough to cover the vma. Will return failure if
2162 * that criteria isn't met.
2164 * Similar to remap_pfn_range() (see mm/memory.c)
2166 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2167 unsigned long pgoff)
2169 struct vm_struct *area;
2170 unsigned long uaddr = vma->vm_start;
2171 unsigned long usize = vma->vm_end - vma->vm_start;
2173 if ((PAGE_SIZE-1) & (unsigned long)addr)
2176 area = find_vm_area(addr);
2180 if (!(area->flags & VM_USERMAP))
2183 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
2186 addr += pgoff << PAGE_SHIFT;
2188 struct page *page = vmalloc_to_page(addr);
2191 ret = vm_insert_page(vma, uaddr, page);
2198 } while (usize > 0);
2200 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2204 EXPORT_SYMBOL(remap_vmalloc_range);
2207 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2210 void __attribute__((weak)) vmalloc_sync_all(void)
2215 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2227 * alloc_vm_area - allocate a range of kernel address space
2228 * @size: size of the area
2229 * @ptes: returns the PTEs for the address space
2231 * Returns: NULL on failure, vm_struct on success
2233 * This function reserves a range of kernel address space, and
2234 * allocates pagetables to map that range. No actual mappings
2237 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2238 * allocated for the VM area are returned.
2240 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2242 struct vm_struct *area;
2244 area = get_vm_area_caller(size, VM_IOREMAP,
2245 __builtin_return_address(0));
2250 * This ensures that page tables are constructed for this region
2251 * of kernel virtual address space and mapped into init_mm.
2253 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2254 size, f, ptes ? &ptes : NULL)) {
2261 EXPORT_SYMBOL_GPL(alloc_vm_area);
2263 void free_vm_area(struct vm_struct *area)
2265 struct vm_struct *ret;
2266 ret = remove_vm_area(area->addr);
2267 BUG_ON(ret != area);
2270 EXPORT_SYMBOL_GPL(free_vm_area);
2273 static struct vmap_area *node_to_va(struct rb_node *n)
2275 return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2279 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2280 * @end: target address
2281 * @pnext: out arg for the next vmap_area
2282 * @pprev: out arg for the previous vmap_area
2284 * Returns: %true if either or both of next and prev are found,
2285 * %false if no vmap_area exists
2287 * Find vmap_areas end addresses of which enclose @end. ie. if not
2288 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2290 static bool pvm_find_next_prev(unsigned long end,
2291 struct vmap_area **pnext,
2292 struct vmap_area **pprev)
2294 struct rb_node *n = vmap_area_root.rb_node;
2295 struct vmap_area *va = NULL;
2298 va = rb_entry(n, struct vmap_area, rb_node);
2299 if (end < va->va_end)
2301 else if (end > va->va_end)
2310 if (va->va_end > end) {
2312 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2315 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2321 * pvm_determine_end - find the highest aligned address between two vmap_areas
2322 * @pnext: in/out arg for the next vmap_area
2323 * @pprev: in/out arg for the previous vmap_area
2326 * Returns: determined end address
2328 * Find the highest aligned address between *@pnext and *@pprev below
2329 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2330 * down address is between the end addresses of the two vmap_areas.
2332 * Please note that the address returned by this function may fall
2333 * inside *@pnext vmap_area. The caller is responsible for checking
2336 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2337 struct vmap_area **pprev,
2338 unsigned long align)
2340 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2344 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2348 while (*pprev && (*pprev)->va_end > addr) {
2350 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2357 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2358 * @offsets: array containing offset of each area
2359 * @sizes: array containing size of each area
2360 * @nr_vms: the number of areas to allocate
2361 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2363 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2364 * vm_structs on success, %NULL on failure
2366 * Percpu allocator wants to use congruent vm areas so that it can
2367 * maintain the offsets among percpu areas. This function allocates
2368 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2369 * be scattered pretty far, distance between two areas easily going up
2370 * to gigabytes. To avoid interacting with regular vmallocs, these
2371 * areas are allocated from top.
2373 * Despite its complicated look, this allocator is rather simple. It
2374 * does everything top-down and scans areas from the end looking for
2375 * matching slot. While scanning, if any of the areas overlaps with
2376 * existing vmap_area, the base address is pulled down to fit the
2377 * area. Scanning is repeated till all the areas fit and then all
2378 * necessary data structres are inserted and the result is returned.
2380 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2381 const size_t *sizes, int nr_vms,
2384 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2385 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2386 struct vmap_area **vas, *prev, *next;
2387 struct vm_struct **vms;
2388 int area, area2, last_area, term_area;
2389 unsigned long base, start, end, last_end;
2390 bool purged = false;
2392 /* verify parameters and allocate data structures */
2393 BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2394 for (last_area = 0, area = 0; area < nr_vms; area++) {
2395 start = offsets[area];
2396 end = start + sizes[area];
2398 /* is everything aligned properly? */
2399 BUG_ON(!IS_ALIGNED(offsets[area], align));
2400 BUG_ON(!IS_ALIGNED(sizes[area], align));
2402 /* detect the area with the highest address */
2403 if (start > offsets[last_area])
2406 for (area2 = 0; area2 < nr_vms; area2++) {
2407 unsigned long start2 = offsets[area2];
2408 unsigned long end2 = start2 + sizes[area2];
2413 BUG_ON(start2 >= start && start2 < end);
2414 BUG_ON(end2 <= end && end2 > start);
2417 last_end = offsets[last_area] + sizes[last_area];
2419 if (vmalloc_end - vmalloc_start < last_end) {
2424 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2425 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2429 for (area = 0; area < nr_vms; area++) {
2430 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2431 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2432 if (!vas[area] || !vms[area])
2436 spin_lock(&vmap_area_lock);
2438 /* start scanning - we scan from the top, begin with the last area */
2439 area = term_area = last_area;
2440 start = offsets[area];
2441 end = start + sizes[area];
2443 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2444 base = vmalloc_end - last_end;
2447 base = pvm_determine_end(&next, &prev, align) - end;
2450 BUG_ON(next && next->va_end <= base + end);
2451 BUG_ON(prev && prev->va_end > base + end);
2454 * base might have underflowed, add last_end before
2457 if (base + last_end < vmalloc_start + last_end) {
2458 spin_unlock(&vmap_area_lock);
2460 purge_vmap_area_lazy();
2468 * If next overlaps, move base downwards so that it's
2469 * right below next and then recheck.
2471 if (next && next->va_start < base + end) {
2472 base = pvm_determine_end(&next, &prev, align) - end;
2478 * If prev overlaps, shift down next and prev and move
2479 * base so that it's right below new next and then
2482 if (prev && prev->va_end > base + start) {
2484 prev = node_to_va(rb_prev(&next->rb_node));
2485 base = pvm_determine_end(&next, &prev, align) - end;
2491 * This area fits, move on to the previous one. If
2492 * the previous one is the terminal one, we're done.
2494 area = (area + nr_vms - 1) % nr_vms;
2495 if (area == term_area)
2497 start = offsets[area];
2498 end = start + sizes[area];
2499 pvm_find_next_prev(base + end, &next, &prev);
2502 /* we've found a fitting base, insert all va's */
2503 for (area = 0; area < nr_vms; area++) {
2504 struct vmap_area *va = vas[area];
2506 va->va_start = base + offsets[area];
2507 va->va_end = va->va_start + sizes[area];
2508 __insert_vmap_area(va);
2511 vmap_area_pcpu_hole = base + offsets[last_area];
2513 spin_unlock(&vmap_area_lock);
2515 /* insert all vm's */
2516 for (area = 0; area < nr_vms; area++)
2517 insert_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2524 for (area = 0; area < nr_vms; area++) {
2535 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2536 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2537 * @nr_vms: the number of allocated areas
2539 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2541 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2545 for (i = 0; i < nr_vms; i++)
2546 free_vm_area(vms[i]);
2549 #endif /* CONFIG_SMP */
2551 #ifdef CONFIG_PROC_FS
2552 static void *s_start(struct seq_file *m, loff_t *pos)
2553 __acquires(&vmap_area_lock)
2556 struct vmap_area *va;
2558 spin_lock(&vmap_area_lock);
2559 va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2560 while (n > 0 && &va->list != &vmap_area_list) {
2562 va = list_entry(va->list.next, typeof(*va), list);
2564 if (!n && &va->list != &vmap_area_list)
2571 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2573 struct vmap_area *va = p, *next;
2576 next = list_entry(va->list.next, typeof(*va), list);
2577 if (&next->list != &vmap_area_list)
2583 static void s_stop(struct seq_file *m, void *p)
2584 __releases(&vmap_area_lock)
2586 spin_unlock(&vmap_area_lock);
2589 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2591 if (IS_ENABLED(CONFIG_NUMA)) {
2592 unsigned int nr, *counters = m->private;
2597 /* Pair with smp_wmb() in insert_vmalloc_vmlist() */
2599 if (v->flags & VM_UNLIST)
2602 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2604 for (nr = 0; nr < v->nr_pages; nr++)
2605 counters[page_to_nid(v->pages[nr])]++;
2607 for_each_node_state(nr, N_HIGH_MEMORY)
2609 seq_printf(m, " N%u=%u", nr, counters[nr]);
2613 static int s_show(struct seq_file *m, void *p)
2615 struct vmap_area *va = p;
2616 struct vm_struct *v;
2618 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2621 if (!(va->flags & VM_VM_AREA)) {
2622 seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
2623 (void *)va->va_start, (void *)va->va_end,
2624 va->va_end - va->va_start);
2630 seq_printf(m, "0x%pK-0x%pK %7ld",
2631 v->addr, v->addr + v->size, v->size);
2634 seq_printf(m, " %pS", v->caller);
2637 seq_printf(m, " pages=%d", v->nr_pages);
2640 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2642 if (v->flags & VM_IOREMAP)
2643 seq_printf(m, " ioremap");
2645 if (v->flags & VM_ALLOC)
2646 seq_printf(m, " vmalloc");
2648 if (v->flags & VM_MAP)
2649 seq_printf(m, " vmap");
2651 if (v->flags & VM_USERMAP)
2652 seq_printf(m, " user");
2654 if (v->flags & VM_VPAGES)
2655 seq_printf(m, " vpages");
2657 show_numa_info(m, v);
2662 static const struct seq_operations vmalloc_op = {
2669 static int vmalloc_open(struct inode *inode, struct file *file)
2671 unsigned int *ptr = NULL;
2674 if (IS_ENABLED(CONFIG_NUMA)) {
2675 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
2679 ret = seq_open(file, &vmalloc_op);
2681 struct seq_file *m = file->private_data;
2688 static const struct file_operations proc_vmalloc_operations = {
2689 .open = vmalloc_open,
2691 .llseek = seq_lseek,
2692 .release = seq_release_private,
2695 static int __init proc_vmalloc_init(void)
2697 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2700 module_init(proc_vmalloc_init);
2702 void get_vmalloc_info(struct vmalloc_info *vmi)
2704 struct vmap_area *va;
2705 unsigned long free_area_size;
2706 unsigned long prev_end;
2709 vmi->largest_chunk = 0;
2711 prev_end = VMALLOC_START;
2713 spin_lock(&vmap_area_lock);
2715 if (list_empty(&vmap_area_list)) {
2716 vmi->largest_chunk = VMALLOC_TOTAL;
2720 list_for_each_entry(va, &vmap_area_list, list) {
2721 unsigned long addr = va->va_start;
2724 * Some archs keep another range for modules in vmalloc space
2726 if (addr < VMALLOC_START)
2728 if (addr >= VMALLOC_END)
2731 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2734 vmi->used += (va->va_end - va->va_start);
2736 free_area_size = addr - prev_end;
2737 if (vmi->largest_chunk < free_area_size)
2738 vmi->largest_chunk = free_area_size;
2740 prev_end = va->va_end;
2743 if (VMALLOC_END - prev_end > vmi->largest_chunk)
2744 vmi->largest_chunk = VMALLOC_END - prev_end;
2747 spin_unlock(&vmap_area_lock);