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/signal.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/notifier.h>
25 #include <linux/rbtree.h>
26 #include <linux/radix-tree.h>
27 #include <linux/rcupdate.h>
28 #include <linux/pfn.h>
29 #include <linux/kmemleak.h>
30 #include <linux/atomic.h>
31 #include <linux/compiler.h>
32 #include <linux/llist.h>
33 #include <linux/bitops.h>
35 #include <linux/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
45 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct *w)
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
52 struct llist_node *t, *llnode;
54 llist_for_each_safe(llnode, t, llist_del_all(&p->list))
55 __vunmap((void *)llnode, 1);
58 /*** Page table manipulation functions ***/
60 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
64 pte = pte_offset_kernel(pmd, addr);
66 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
67 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
68 } while (pte++, addr += PAGE_SIZE, addr != end);
71 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
76 pmd = pmd_offset(pud, addr);
78 next = pmd_addr_end(addr, end);
79 if (pmd_clear_huge(pmd))
81 if (pmd_none_or_clear_bad(pmd))
83 vunmap_pte_range(pmd, addr, next);
84 } while (pmd++, addr = next, addr != end);
87 static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
92 pud = pud_offset(p4d, addr);
94 next = pud_addr_end(addr, end);
95 if (pud_clear_huge(pud))
97 if (pud_none_or_clear_bad(pud))
99 vunmap_pmd_range(pud, addr, next);
100 } while (pud++, addr = next, addr != end);
103 static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
108 p4d = p4d_offset(pgd, addr);
110 next = p4d_addr_end(addr, end);
111 if (p4d_clear_huge(p4d))
113 if (p4d_none_or_clear_bad(p4d))
115 vunmap_pud_range(p4d, addr, next);
116 } while (p4d++, addr = next, addr != end);
119 static void vunmap_page_range(unsigned long addr, unsigned long end)
125 pgd = pgd_offset_k(addr);
127 next = pgd_addr_end(addr, end);
128 if (pgd_none_or_clear_bad(pgd))
130 vunmap_p4d_range(pgd, addr, next);
131 } while (pgd++, addr = next, addr != end);
134 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
135 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
140 * nr is a running index into the array which helps higher level
141 * callers keep track of where we're up to.
144 pte = pte_alloc_kernel(pmd, addr);
148 struct page *page = pages[*nr];
150 if (WARN_ON(!pte_none(*pte)))
154 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
156 } while (pte++, addr += PAGE_SIZE, addr != end);
160 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
161 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
166 pmd = pmd_alloc(&init_mm, pud, addr);
170 next = pmd_addr_end(addr, end);
171 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
173 } while (pmd++, addr = next, addr != end);
177 static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
178 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
183 pud = pud_alloc(&init_mm, p4d, addr);
187 next = pud_addr_end(addr, end);
188 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
190 } while (pud++, addr = next, addr != end);
194 static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
195 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
200 p4d = p4d_alloc(&init_mm, pgd, addr);
204 next = p4d_addr_end(addr, end);
205 if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
207 } while (p4d++, addr = next, addr != end);
212 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
213 * will have pfns corresponding to the "pages" array.
215 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
217 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
218 pgprot_t prot, struct page **pages)
222 unsigned long addr = start;
227 pgd = pgd_offset_k(addr);
229 next = pgd_addr_end(addr, end);
230 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
233 } while (pgd++, addr = next, addr != end);
238 static int vmap_page_range(unsigned long start, unsigned long end,
239 pgprot_t prot, struct page **pages)
243 ret = vmap_page_range_noflush(start, end, prot, pages);
244 flush_cache_vmap(start, end);
248 int is_vmalloc_or_module_addr(const void *x)
251 * ARM, x86-64 and sparc64 put modules in a special place,
252 * and fall back on vmalloc() if that fails. Others
253 * just put it in the vmalloc space.
255 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
256 unsigned long addr = (unsigned long)x;
257 if (addr >= MODULES_VADDR && addr < MODULES_END)
260 return is_vmalloc_addr(x);
264 * Walk a vmap address to the struct page it maps.
266 struct page *vmalloc_to_page(const void *vmalloc_addr)
268 unsigned long addr = (unsigned long) vmalloc_addr;
269 struct page *page = NULL;
270 pgd_t *pgd = pgd_offset_k(addr);
277 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
278 * architectures that do not vmalloc module space
280 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
284 p4d = p4d_offset(pgd, addr);
287 pud = pud_offset(p4d, addr);
290 * Don't dereference bad PUD or PMD (below) entries. This will also
291 * identify huge mappings, which we may encounter on architectures
292 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
293 * identified as vmalloc addresses by is_vmalloc_addr(), but are
294 * not [unambiguously] associated with a struct page, so there is
295 * no correct value to return for them.
297 WARN_ON_ONCE(pud_bad(*pud));
298 if (pud_none(*pud) || pud_bad(*pud))
300 pmd = pmd_offset(pud, addr);
301 WARN_ON_ONCE(pmd_bad(*pmd));
302 if (pmd_none(*pmd) || pmd_bad(*pmd))
305 ptep = pte_offset_map(pmd, addr);
307 if (pte_present(pte))
308 page = pte_page(pte);
312 EXPORT_SYMBOL(vmalloc_to_page);
315 * Map a vmalloc()-space virtual address to the physical page frame number.
317 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
319 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
321 EXPORT_SYMBOL(vmalloc_to_pfn);
324 /*** Global kva allocator ***/
326 #define VM_LAZY_FREE 0x02
327 #define VM_VM_AREA 0x04
329 static DEFINE_SPINLOCK(vmap_area_lock);
330 /* Export for kexec only */
331 LIST_HEAD(vmap_area_list);
332 static LLIST_HEAD(vmap_purge_list);
333 static struct rb_root vmap_area_root = RB_ROOT;
335 /* The vmap cache globals are protected by vmap_area_lock */
336 static struct rb_node *free_vmap_cache;
337 static unsigned long cached_hole_size;
338 static unsigned long cached_vstart;
339 static unsigned long cached_align;
341 static unsigned long vmap_area_pcpu_hole;
343 static struct vmap_area *__find_vmap_area(unsigned long addr)
345 struct rb_node *n = vmap_area_root.rb_node;
348 struct vmap_area *va;
350 va = rb_entry(n, struct vmap_area, rb_node);
351 if (addr < va->va_start)
353 else if (addr >= va->va_end)
362 static void __insert_vmap_area(struct vmap_area *va)
364 struct rb_node **p = &vmap_area_root.rb_node;
365 struct rb_node *parent = NULL;
369 struct vmap_area *tmp_va;
372 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
373 if (va->va_start < tmp_va->va_end)
375 else if (va->va_end > tmp_va->va_start)
381 rb_link_node(&va->rb_node, parent, p);
382 rb_insert_color(&va->rb_node, &vmap_area_root);
384 /* address-sort this list */
385 tmp = rb_prev(&va->rb_node);
387 struct vmap_area *prev;
388 prev = rb_entry(tmp, struct vmap_area, rb_node);
389 list_add_rcu(&va->list, &prev->list);
391 list_add_rcu(&va->list, &vmap_area_list);
394 static void purge_vmap_area_lazy(void);
396 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
399 * Allocate a region of KVA of the specified size and alignment, within the
402 static struct vmap_area *alloc_vmap_area(unsigned long size,
404 unsigned long vstart, unsigned long vend,
405 int node, gfp_t gfp_mask)
407 struct vmap_area *va;
411 struct vmap_area *first;
414 BUG_ON(offset_in_page(size));
415 BUG_ON(!is_power_of_2(align));
419 va = kmalloc_node(sizeof(struct vmap_area),
420 gfp_mask & GFP_RECLAIM_MASK, node);
422 return ERR_PTR(-ENOMEM);
425 * Only scan the relevant parts containing pointers to other objects
426 * to avoid false negatives.
428 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
431 spin_lock(&vmap_area_lock);
433 * Invalidate cache if we have more permissive parameters.
434 * cached_hole_size notes the largest hole noticed _below_
435 * the vmap_area cached in free_vmap_cache: if size fits
436 * into that hole, we want to scan from vstart to reuse
437 * the hole instead of allocating above free_vmap_cache.
438 * Note that __free_vmap_area may update free_vmap_cache
439 * without updating cached_hole_size or cached_align.
441 if (!free_vmap_cache ||
442 size < cached_hole_size ||
443 vstart < cached_vstart ||
444 align < cached_align) {
446 cached_hole_size = 0;
447 free_vmap_cache = NULL;
449 /* record if we encounter less permissive parameters */
450 cached_vstart = vstart;
451 cached_align = align;
453 /* find starting point for our search */
454 if (free_vmap_cache) {
455 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
456 addr = ALIGN(first->va_end, align);
459 if (addr + size < addr)
463 addr = ALIGN(vstart, align);
464 if (addr + size < addr)
467 n = vmap_area_root.rb_node;
471 struct vmap_area *tmp;
472 tmp = rb_entry(n, struct vmap_area, rb_node);
473 if (tmp->va_end >= addr) {
475 if (tmp->va_start <= addr)
486 /* from the starting point, walk areas until a suitable hole is found */
487 while (addr + size > first->va_start && addr + size <= vend) {
488 if (addr + cached_hole_size < first->va_start)
489 cached_hole_size = first->va_start - addr;
490 addr = ALIGN(first->va_end, align);
491 if (addr + size < addr)
494 if (list_is_last(&first->list, &vmap_area_list))
497 first = list_next_entry(first, list);
502 * Check also calculated address against the vstart,
503 * because it can be 0 because of big align request.
505 if (addr + size > vend || addr < vstart)
509 va->va_end = addr + size;
511 __insert_vmap_area(va);
512 free_vmap_cache = &va->rb_node;
513 spin_unlock(&vmap_area_lock);
515 BUG_ON(!IS_ALIGNED(va->va_start, align));
516 BUG_ON(va->va_start < vstart);
517 BUG_ON(va->va_end > vend);
522 spin_unlock(&vmap_area_lock);
524 purge_vmap_area_lazy();
529 if (gfpflags_allow_blocking(gfp_mask)) {
530 unsigned long freed = 0;
531 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
538 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
539 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
542 return ERR_PTR(-EBUSY);
545 int register_vmap_purge_notifier(struct notifier_block *nb)
547 return blocking_notifier_chain_register(&vmap_notify_list, nb);
549 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
551 int unregister_vmap_purge_notifier(struct notifier_block *nb)
553 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
555 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
557 static void __free_vmap_area(struct vmap_area *va)
559 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
561 if (free_vmap_cache) {
562 if (va->va_end < cached_vstart) {
563 free_vmap_cache = NULL;
565 struct vmap_area *cache;
566 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
567 if (va->va_start <= cache->va_start) {
568 free_vmap_cache = rb_prev(&va->rb_node);
570 * We don't try to update cached_hole_size or
571 * cached_align, but it won't go very wrong.
576 rb_erase(&va->rb_node, &vmap_area_root);
577 RB_CLEAR_NODE(&va->rb_node);
578 list_del_rcu(&va->list);
581 * Track the highest possible candidate for pcpu area
582 * allocation. Areas outside of vmalloc area can be returned
583 * here too, consider only end addresses which fall inside
584 * vmalloc area proper.
586 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
587 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
589 kfree_rcu(va, rcu_head);
593 * Free a region of KVA allocated by alloc_vmap_area
595 static void free_vmap_area(struct vmap_area *va)
597 spin_lock(&vmap_area_lock);
598 __free_vmap_area(va);
599 spin_unlock(&vmap_area_lock);
603 * Clear the pagetable entries of a given vmap_area
605 static void unmap_vmap_area(struct vmap_area *va)
607 vunmap_page_range(va->va_start, va->va_end);
611 * lazy_max_pages is the maximum amount of virtual address space we gather up
612 * before attempting to purge with a TLB flush.
614 * There is a tradeoff here: a larger number will cover more kernel page tables
615 * and take slightly longer to purge, but it will linearly reduce the number of
616 * global TLB flushes that must be performed. It would seem natural to scale
617 * this number up linearly with the number of CPUs (because vmapping activity
618 * could also scale linearly with the number of CPUs), however it is likely
619 * that in practice, workloads might be constrained in other ways that mean
620 * vmap activity will not scale linearly with CPUs. Also, I want to be
621 * conservative and not introduce a big latency on huge systems, so go with
622 * a less aggressive log scale. It will still be an improvement over the old
623 * code, and it will be simple to change the scale factor if we find that it
624 * becomes a problem on bigger systems.
626 static unsigned long lazy_max_pages(void)
630 log = fls(num_online_cpus());
632 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
635 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
638 * Serialize vmap purging. There is no actual criticial section protected
639 * by this look, but we want to avoid concurrent calls for performance
640 * reasons and to make the pcpu_get_vm_areas more deterministic.
642 static DEFINE_MUTEX(vmap_purge_lock);
644 /* for per-CPU blocks */
645 static void purge_fragmented_blocks_allcpus(void);
648 * called before a call to iounmap() if the caller wants vm_area_struct's
651 void set_iounmap_nonlazy(void)
653 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
657 * Purges all lazily-freed vmap areas.
659 static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
661 struct llist_node *valist;
662 struct vmap_area *va;
663 struct vmap_area *n_va;
664 bool do_free = false;
666 lockdep_assert_held(&vmap_purge_lock);
668 valist = llist_del_all(&vmap_purge_list);
669 llist_for_each_entry(va, valist, purge_list) {
670 if (va->va_start < start)
671 start = va->va_start;
672 if (va->va_end > end)
680 flush_tlb_kernel_range(start, end);
682 spin_lock(&vmap_area_lock);
683 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
684 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
686 __free_vmap_area(va);
687 atomic_sub(nr, &vmap_lazy_nr);
688 cond_resched_lock(&vmap_area_lock);
690 spin_unlock(&vmap_area_lock);
695 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
696 * is already purging.
698 static void try_purge_vmap_area_lazy(void)
700 if (mutex_trylock(&vmap_purge_lock)) {
701 __purge_vmap_area_lazy(ULONG_MAX, 0);
702 mutex_unlock(&vmap_purge_lock);
707 * Kick off a purge of the outstanding lazy areas.
709 static void purge_vmap_area_lazy(void)
711 mutex_lock(&vmap_purge_lock);
712 purge_fragmented_blocks_allcpus();
713 __purge_vmap_area_lazy(ULONG_MAX, 0);
714 mutex_unlock(&vmap_purge_lock);
718 * Free a vmap area, caller ensuring that the area has been unmapped
719 * and flush_cache_vunmap had been called for the correct range
722 static void free_vmap_area_noflush(struct vmap_area *va)
726 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
729 /* After this point, we may free va at any time */
730 llist_add(&va->purge_list, &vmap_purge_list);
732 if (unlikely(nr_lazy > lazy_max_pages()))
733 try_purge_vmap_area_lazy();
737 * Free and unmap a vmap area
739 static void free_unmap_vmap_area(struct vmap_area *va)
741 flush_cache_vunmap(va->va_start, va->va_end);
743 if (debug_pagealloc_enabled())
744 flush_tlb_kernel_range(va->va_start, va->va_end);
746 free_vmap_area_noflush(va);
749 static struct vmap_area *find_vmap_area(unsigned long addr)
751 struct vmap_area *va;
753 spin_lock(&vmap_area_lock);
754 va = __find_vmap_area(addr);
755 spin_unlock(&vmap_area_lock);
760 /*** Per cpu kva allocator ***/
763 * vmap space is limited especially on 32 bit architectures. Ensure there is
764 * room for at least 16 percpu vmap blocks per CPU.
767 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
768 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
769 * instead (we just need a rough idea)
771 #if BITS_PER_LONG == 32
772 #define VMALLOC_SPACE (128UL*1024*1024)
774 #define VMALLOC_SPACE (128UL*1024*1024*1024)
777 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
778 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
779 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
780 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
781 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
782 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
783 #define VMAP_BBMAP_BITS \
784 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
785 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
786 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
788 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
790 static bool vmap_initialized __read_mostly = false;
792 struct vmap_block_queue {
794 struct list_head free;
799 struct vmap_area *va;
800 unsigned long free, dirty;
801 unsigned long dirty_min, dirty_max; /*< dirty range */
802 struct list_head free_list;
803 struct rcu_head rcu_head;
804 struct list_head purge;
807 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
808 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
811 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
812 * in the free path. Could get rid of this if we change the API to return a
813 * "cookie" from alloc, to be passed to free. But no big deal yet.
815 static DEFINE_SPINLOCK(vmap_block_tree_lock);
816 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
819 * We should probably have a fallback mechanism to allocate virtual memory
820 * out of partially filled vmap blocks. However vmap block sizing should be
821 * fairly reasonable according to the vmalloc size, so it shouldn't be a
825 static unsigned long addr_to_vb_idx(unsigned long addr)
827 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
828 addr /= VMAP_BLOCK_SIZE;
832 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
836 addr = va_start + (pages_off << PAGE_SHIFT);
837 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
842 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
843 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
844 * @order: how many 2^order pages should be occupied in newly allocated block
845 * @gfp_mask: flags for the page level allocator
847 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
849 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
851 struct vmap_block_queue *vbq;
852 struct vmap_block *vb;
853 struct vmap_area *va;
854 unsigned long vb_idx;
858 node = numa_node_id();
860 vb = kmalloc_node(sizeof(struct vmap_block),
861 gfp_mask & GFP_RECLAIM_MASK, node);
863 return ERR_PTR(-ENOMEM);
865 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
866 VMALLOC_START, VMALLOC_END,
873 err = radix_tree_preload(gfp_mask);
880 vaddr = vmap_block_vaddr(va->va_start, 0);
881 spin_lock_init(&vb->lock);
883 /* At least something should be left free */
884 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
885 vb->free = VMAP_BBMAP_BITS - (1UL << order);
887 vb->dirty_min = VMAP_BBMAP_BITS;
889 INIT_LIST_HEAD(&vb->free_list);
891 vb_idx = addr_to_vb_idx(va->va_start);
892 spin_lock(&vmap_block_tree_lock);
893 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
894 spin_unlock(&vmap_block_tree_lock);
896 radix_tree_preload_end();
898 vbq = &get_cpu_var(vmap_block_queue);
899 spin_lock(&vbq->lock);
900 list_add_tail_rcu(&vb->free_list, &vbq->free);
901 spin_unlock(&vbq->lock);
902 put_cpu_var(vmap_block_queue);
907 static void free_vmap_block(struct vmap_block *vb)
909 struct vmap_block *tmp;
910 unsigned long vb_idx;
912 vb_idx = addr_to_vb_idx(vb->va->va_start);
913 spin_lock(&vmap_block_tree_lock);
914 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
915 spin_unlock(&vmap_block_tree_lock);
918 free_vmap_area_noflush(vb->va);
919 kfree_rcu(vb, rcu_head);
922 static void purge_fragmented_blocks(int cpu)
925 struct vmap_block *vb;
926 struct vmap_block *n_vb;
927 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
930 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
932 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
935 spin_lock(&vb->lock);
936 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
937 vb->free = 0; /* prevent further allocs after releasing lock */
938 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
940 vb->dirty_max = VMAP_BBMAP_BITS;
941 spin_lock(&vbq->lock);
942 list_del_rcu(&vb->free_list);
943 spin_unlock(&vbq->lock);
944 spin_unlock(&vb->lock);
945 list_add_tail(&vb->purge, &purge);
947 spin_unlock(&vb->lock);
951 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
952 list_del(&vb->purge);
957 static void purge_fragmented_blocks_allcpus(void)
961 for_each_possible_cpu(cpu)
962 purge_fragmented_blocks(cpu);
965 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
967 struct vmap_block_queue *vbq;
968 struct vmap_block *vb;
972 BUG_ON(offset_in_page(size));
973 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
974 if (WARN_ON(size == 0)) {
976 * Allocating 0 bytes isn't what caller wants since
977 * get_order(0) returns funny result. Just warn and terminate
982 order = get_order(size);
985 vbq = &get_cpu_var(vmap_block_queue);
986 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
987 unsigned long pages_off;
989 spin_lock(&vb->lock);
990 if (vb->free < (1UL << order)) {
991 spin_unlock(&vb->lock);
995 pages_off = VMAP_BBMAP_BITS - vb->free;
996 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
997 vb->free -= 1UL << order;
999 spin_lock(&vbq->lock);
1000 list_del_rcu(&vb->free_list);
1001 spin_unlock(&vbq->lock);
1004 spin_unlock(&vb->lock);
1008 put_cpu_var(vmap_block_queue);
1011 /* Allocate new block if nothing was found */
1013 vaddr = new_vmap_block(order, gfp_mask);
1018 static void vb_free(const void *addr, unsigned long size)
1020 unsigned long offset;
1021 unsigned long vb_idx;
1023 struct vmap_block *vb;
1025 BUG_ON(offset_in_page(size));
1026 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1028 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1030 order = get_order(size);
1032 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1033 offset >>= PAGE_SHIFT;
1035 vb_idx = addr_to_vb_idx((unsigned long)addr);
1037 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1041 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1043 if (debug_pagealloc_enabled())
1044 flush_tlb_kernel_range((unsigned long)addr,
1045 (unsigned long)addr + size);
1047 spin_lock(&vb->lock);
1049 /* Expand dirty range */
1050 vb->dirty_min = min(vb->dirty_min, offset);
1051 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1053 vb->dirty += 1UL << order;
1054 if (vb->dirty == VMAP_BBMAP_BITS) {
1056 spin_unlock(&vb->lock);
1057 free_vmap_block(vb);
1059 spin_unlock(&vb->lock);
1063 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1065 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1066 * to amortize TLB flushing overheads. What this means is that any page you
1067 * have now, may, in a former life, have been mapped into kernel virtual
1068 * address by the vmap layer and so there might be some CPUs with TLB entries
1069 * still referencing that page (additional to the regular 1:1 kernel mapping).
1071 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1072 * be sure that none of the pages we have control over will have any aliases
1073 * from the vmap layer.
1075 void vm_unmap_aliases(void)
1077 unsigned long start = ULONG_MAX, end = 0;
1081 if (unlikely(!vmap_initialized))
1086 for_each_possible_cpu(cpu) {
1087 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1088 struct vmap_block *vb;
1091 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1092 spin_lock(&vb->lock);
1094 unsigned long va_start = vb->va->va_start;
1097 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1098 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1100 start = min(s, start);
1105 spin_unlock(&vb->lock);
1110 mutex_lock(&vmap_purge_lock);
1111 purge_fragmented_blocks_allcpus();
1112 if (!__purge_vmap_area_lazy(start, end) && flush)
1113 flush_tlb_kernel_range(start, end);
1114 mutex_unlock(&vmap_purge_lock);
1116 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1119 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1120 * @mem: the pointer returned by vm_map_ram
1121 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1123 void vm_unmap_ram(const void *mem, unsigned int count)
1125 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1126 unsigned long addr = (unsigned long)mem;
1127 struct vmap_area *va;
1131 BUG_ON(addr < VMALLOC_START);
1132 BUG_ON(addr > VMALLOC_END);
1133 BUG_ON(!PAGE_ALIGNED(addr));
1135 if (likely(count <= VMAP_MAX_ALLOC)) {
1136 debug_check_no_locks_freed(mem, size);
1141 va = find_vmap_area(addr);
1143 debug_check_no_locks_freed((void *)va->va_start,
1144 (va->va_end - va->va_start));
1145 free_unmap_vmap_area(va);
1147 EXPORT_SYMBOL(vm_unmap_ram);
1150 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1151 * @pages: an array of pointers to the pages to be mapped
1152 * @count: number of pages
1153 * @node: prefer to allocate data structures on this node
1154 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1156 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1157 * faster than vmap so it's good. But if you mix long-life and short-life
1158 * objects with vm_map_ram(), it could consume lots of address space through
1159 * fragmentation (especially on a 32bit machine). You could see failures in
1160 * the end. Please use this function for short-lived objects.
1162 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1164 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1166 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1170 if (likely(count <= VMAP_MAX_ALLOC)) {
1171 mem = vb_alloc(size, GFP_KERNEL);
1174 addr = (unsigned long)mem;
1176 struct vmap_area *va;
1177 va = alloc_vmap_area(size, PAGE_SIZE,
1178 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1182 addr = va->va_start;
1185 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1186 vm_unmap_ram(mem, count);
1191 EXPORT_SYMBOL(vm_map_ram);
1193 static struct vm_struct *vmlist __initdata;
1195 * vm_area_add_early - add vmap area early during boot
1196 * @vm: vm_struct to add
1198 * This function is used to add fixed kernel vm area to vmlist before
1199 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1200 * should contain proper values and the other fields should be zero.
1202 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1204 void __init vm_area_add_early(struct vm_struct *vm)
1206 struct vm_struct *tmp, **p;
1208 BUG_ON(vmap_initialized);
1209 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1210 if (tmp->addr >= vm->addr) {
1211 BUG_ON(tmp->addr < vm->addr + vm->size);
1214 BUG_ON(tmp->addr + tmp->size > vm->addr);
1221 * vm_area_register_early - register vmap area early during boot
1222 * @vm: vm_struct to register
1223 * @align: requested alignment
1225 * This function is used to register kernel vm area before
1226 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1227 * proper values on entry and other fields should be zero. On return,
1228 * vm->addr contains the allocated address.
1230 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1232 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1234 static size_t vm_init_off __initdata;
1237 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1238 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1240 vm->addr = (void *)addr;
1242 vm_area_add_early(vm);
1245 void __init vmalloc_init(void)
1247 struct vmap_area *va;
1248 struct vm_struct *tmp;
1251 for_each_possible_cpu(i) {
1252 struct vmap_block_queue *vbq;
1253 struct vfree_deferred *p;
1255 vbq = &per_cpu(vmap_block_queue, i);
1256 spin_lock_init(&vbq->lock);
1257 INIT_LIST_HEAD(&vbq->free);
1258 p = &per_cpu(vfree_deferred, i);
1259 init_llist_head(&p->list);
1260 INIT_WORK(&p->wq, free_work);
1263 /* Import existing vmlist entries. */
1264 for (tmp = vmlist; tmp; tmp = tmp->next) {
1265 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1266 va->flags = VM_VM_AREA;
1267 va->va_start = (unsigned long)tmp->addr;
1268 va->va_end = va->va_start + tmp->size;
1270 __insert_vmap_area(va);
1273 vmap_area_pcpu_hole = VMALLOC_END;
1275 vmap_initialized = true;
1279 * map_kernel_range_noflush - map kernel VM area with the specified pages
1280 * @addr: start of the VM area to map
1281 * @size: size of the VM area to map
1282 * @prot: page protection flags to use
1283 * @pages: pages to map
1285 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1286 * specify should have been allocated using get_vm_area() and its
1290 * This function does NOT do any cache flushing. The caller is
1291 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1292 * before calling this function.
1295 * The number of pages mapped on success, -errno on failure.
1297 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1298 pgprot_t prot, struct page **pages)
1300 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1304 * unmap_kernel_range_noflush - unmap kernel VM area
1305 * @addr: start of the VM area to unmap
1306 * @size: size of the VM area to unmap
1308 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1309 * specify should have been allocated using get_vm_area() and its
1313 * This function does NOT do any cache flushing. The caller is
1314 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1315 * before calling this function and flush_tlb_kernel_range() after.
1317 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1319 vunmap_page_range(addr, addr + size);
1321 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1324 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1325 * @addr: start of the VM area to unmap
1326 * @size: size of the VM area to unmap
1328 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1329 * the unmapping and tlb after.
1331 void unmap_kernel_range(unsigned long addr, unsigned long size)
1333 unsigned long end = addr + size;
1335 flush_cache_vunmap(addr, end);
1336 vunmap_page_range(addr, end);
1337 flush_tlb_kernel_range(addr, end);
1339 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1341 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1343 unsigned long addr = (unsigned long)area->addr;
1344 unsigned long end = addr + get_vm_area_size(area);
1347 err = vmap_page_range(addr, end, prot, pages);
1349 return err > 0 ? 0 : err;
1351 EXPORT_SYMBOL_GPL(map_vm_area);
1353 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1354 unsigned long flags, const void *caller)
1356 spin_lock(&vmap_area_lock);
1358 vm->addr = (void *)va->va_start;
1359 vm->size = va->va_end - va->va_start;
1360 vm->caller = caller;
1362 va->flags |= VM_VM_AREA;
1363 spin_unlock(&vmap_area_lock);
1366 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1369 * Before removing VM_UNINITIALIZED,
1370 * we should make sure that vm has proper values.
1371 * Pair with smp_rmb() in show_numa_info().
1374 vm->flags &= ~VM_UNINITIALIZED;
1377 static struct vm_struct *__get_vm_area_node(unsigned long size,
1378 unsigned long align, unsigned long flags, unsigned long start,
1379 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1381 struct vmap_area *va;
1382 struct vm_struct *area;
1384 BUG_ON(in_interrupt());
1385 size = PAGE_ALIGN(size);
1386 if (unlikely(!size))
1389 if (flags & VM_IOREMAP)
1390 align = 1ul << clamp_t(int, get_count_order_long(size),
1391 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1393 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1394 if (unlikely(!area))
1397 if (!(flags & VM_NO_GUARD))
1400 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1406 setup_vmalloc_vm(area, va, flags, caller);
1411 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1412 unsigned long start, unsigned long end)
1414 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1415 GFP_KERNEL, __builtin_return_address(0));
1417 EXPORT_SYMBOL_GPL(__get_vm_area);
1419 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1420 unsigned long start, unsigned long end,
1423 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1424 GFP_KERNEL, caller);
1428 * get_vm_area - reserve a contiguous kernel virtual area
1429 * @size: size of the area
1430 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1432 * Search an area of @size in the kernel virtual mapping area,
1433 * and reserved it for out purposes. Returns the area descriptor
1434 * on success or %NULL on failure.
1436 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1438 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1439 NUMA_NO_NODE, GFP_KERNEL,
1440 __builtin_return_address(0));
1443 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1446 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1447 NUMA_NO_NODE, GFP_KERNEL, caller);
1451 * find_vm_area - find a continuous kernel virtual area
1452 * @addr: base address
1454 * Search for the kernel VM area starting at @addr, and return it.
1455 * It is up to the caller to do all required locking to keep the returned
1458 struct vm_struct *find_vm_area(const void *addr)
1460 struct vmap_area *va;
1462 va = find_vmap_area((unsigned long)addr);
1463 if (va && va->flags & VM_VM_AREA)
1470 * remove_vm_area - find and remove a continuous kernel virtual area
1471 * @addr: base address
1473 * Search for the kernel VM area starting at @addr, and remove it.
1474 * This function returns the found VM area, but using it is NOT safe
1475 * on SMP machines, except for its size or flags.
1477 struct vm_struct *remove_vm_area(const void *addr)
1479 struct vmap_area *va;
1483 va = find_vmap_area((unsigned long)addr);
1484 if (va && va->flags & VM_VM_AREA) {
1485 struct vm_struct *vm = va->vm;
1487 spin_lock(&vmap_area_lock);
1489 va->flags &= ~VM_VM_AREA;
1490 va->flags |= VM_LAZY_FREE;
1491 spin_unlock(&vmap_area_lock);
1493 kasan_free_shadow(vm);
1494 free_unmap_vmap_area(va);
1501 static void __vunmap(const void *addr, int deallocate_pages)
1503 struct vm_struct *area;
1508 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1512 area = find_vm_area(addr);
1513 if (unlikely(!area)) {
1514 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1519 debug_check_no_locks_freed(area->addr, get_vm_area_size(area));
1520 debug_check_no_obj_freed(area->addr, get_vm_area_size(area));
1522 remove_vm_area(addr);
1523 if (deallocate_pages) {
1526 for (i = 0; i < area->nr_pages; i++) {
1527 struct page *page = area->pages[i];
1530 __free_pages(page, 0);
1533 kvfree(area->pages);
1540 static inline void __vfree_deferred(const void *addr)
1543 * Use raw_cpu_ptr() because this can be called from preemptible
1544 * context. Preemption is absolutely fine here, because the llist_add()
1545 * implementation is lockless, so it works even if we are adding to
1546 * nother cpu's list. schedule_work() should be fine with this too.
1548 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1550 if (llist_add((struct llist_node *)addr, &p->list))
1551 schedule_work(&p->wq);
1555 * vfree_atomic - release memory allocated by vmalloc()
1556 * @addr: memory base address
1558 * This one is just like vfree() but can be called in any atomic context
1561 void vfree_atomic(const void *addr)
1565 kmemleak_free(addr);
1569 __vfree_deferred(addr);
1572 static void __vfree(const void *addr)
1574 if (unlikely(in_interrupt()))
1575 __vfree_deferred(addr);
1581 * vfree - release memory allocated by vmalloc()
1582 * @addr: memory base address
1584 * Free the virtually continuous memory area starting at @addr, as
1585 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1586 * NULL, no operation is performed.
1588 * Must not be called in NMI context (strictly speaking, only if we don't
1589 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1590 * conventions for vfree() arch-depenedent would be a really bad idea)
1592 * May sleep if called *not* from interrupt context.
1594 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1596 void vfree(const void *addr)
1600 kmemleak_free(addr);
1602 might_sleep_if(!in_interrupt());
1609 EXPORT_SYMBOL(vfree);
1612 * vunmap - release virtual mapping obtained by vmap()
1613 * @addr: memory base address
1615 * Free the virtually contiguous memory area starting at @addr,
1616 * which was created from the page array passed to vmap().
1618 * Must not be called in interrupt context.
1620 void vunmap(const void *addr)
1622 BUG_ON(in_interrupt());
1627 EXPORT_SYMBOL(vunmap);
1630 * vmap - map an array of pages into virtually contiguous space
1631 * @pages: array of page pointers
1632 * @count: number of pages to map
1633 * @flags: vm_area->flags
1634 * @prot: page protection for the mapping
1636 * Maps @count pages from @pages into contiguous kernel virtual
1639 void *vmap(struct page **pages, unsigned int count,
1640 unsigned long flags, pgprot_t prot)
1642 struct vm_struct *area;
1643 unsigned long size; /* In bytes */
1647 if (count > totalram_pages())
1650 size = (unsigned long)count << PAGE_SHIFT;
1651 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1655 if (map_vm_area(area, prot, pages)) {
1662 EXPORT_SYMBOL(vmap);
1664 static void *__vmalloc_node(unsigned long size, unsigned long align,
1665 gfp_t gfp_mask, pgprot_t prot,
1666 int node, const void *caller);
1667 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1668 pgprot_t prot, int node)
1670 struct page **pages;
1671 unsigned int nr_pages, array_size, i;
1672 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1673 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1674 const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1678 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1679 array_size = (nr_pages * sizeof(struct page *));
1681 area->nr_pages = nr_pages;
1682 /* Please note that the recursion is strictly bounded. */
1683 if (array_size > PAGE_SIZE) {
1684 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
1685 PAGE_KERNEL, node, area->caller);
1687 pages = kmalloc_node(array_size, nested_gfp, node);
1689 area->pages = pages;
1691 remove_vm_area(area->addr);
1696 for (i = 0; i < area->nr_pages; i++) {
1699 if (node == NUMA_NO_NODE)
1700 page = alloc_page(alloc_mask|highmem_mask);
1702 page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
1704 if (unlikely(!page)) {
1705 /* Successfully allocated i pages, free them in __vunmap() */
1709 area->pages[i] = page;
1710 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
1714 if (map_vm_area(area, prot, pages))
1719 warn_alloc(gfp_mask, NULL,
1720 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1721 (area->nr_pages*PAGE_SIZE), area->size);
1722 __vfree(area->addr);
1727 * __vmalloc_node_range - allocate virtually contiguous memory
1728 * @size: allocation size
1729 * @align: desired alignment
1730 * @start: vm area range start
1731 * @end: vm area range end
1732 * @gfp_mask: flags for the page level allocator
1733 * @prot: protection mask for the allocated pages
1734 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1735 * @node: node to use for allocation or NUMA_NO_NODE
1736 * @caller: caller's return address
1738 * Allocate enough pages to cover @size from the page level
1739 * allocator with @gfp_mask flags. Map them into contiguous
1740 * kernel virtual space, using a pagetable protection of @prot.
1742 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1743 unsigned long start, unsigned long end, gfp_t gfp_mask,
1744 pgprot_t prot, unsigned long vm_flags, int node,
1747 struct vm_struct *area;
1749 unsigned long real_size = size;
1751 size = PAGE_ALIGN(size);
1752 if (!size || (size >> PAGE_SHIFT) > totalram_pages())
1755 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1756 vm_flags, start, end, node, gfp_mask, caller);
1760 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1765 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1766 * flag. It means that vm_struct is not fully initialized.
1767 * Now, it is fully initialized, so remove this flag here.
1769 clear_vm_uninitialized_flag(area);
1771 kmemleak_vmalloc(area, size, gfp_mask);
1776 warn_alloc(gfp_mask, NULL,
1777 "vmalloc: allocation failure: %lu bytes", real_size);
1782 * This is only for performance analysis of vmalloc and stress purpose.
1783 * It is required by vmalloc test module, therefore do not use it other
1786 #ifdef CONFIG_TEST_VMALLOC_MODULE
1787 EXPORT_SYMBOL_GPL(__vmalloc_node_range);
1791 * __vmalloc_node - allocate virtually contiguous memory
1792 * @size: allocation size
1793 * @align: desired alignment
1794 * @gfp_mask: flags for the page level allocator
1795 * @prot: protection mask for the allocated pages
1796 * @node: node to use for allocation or NUMA_NO_NODE
1797 * @caller: caller's return address
1799 * Allocate enough pages to cover @size from the page level
1800 * allocator with @gfp_mask flags. Map them into contiguous
1801 * kernel virtual space, using a pagetable protection of @prot.
1803 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
1804 * and __GFP_NOFAIL are not supported
1806 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1810 static void *__vmalloc_node(unsigned long size, unsigned long align,
1811 gfp_t gfp_mask, pgprot_t prot,
1812 int node, const void *caller)
1814 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1815 gfp_mask, prot, 0, node, caller);
1818 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1820 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1821 __builtin_return_address(0));
1823 EXPORT_SYMBOL(__vmalloc);
1825 static inline void *__vmalloc_node_flags(unsigned long size,
1826 int node, gfp_t flags)
1828 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1829 node, __builtin_return_address(0));
1833 void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1836 return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1840 * vmalloc - allocate virtually contiguous memory
1841 * @size: allocation size
1842 * Allocate enough pages to cover @size from the page level
1843 * allocator and map them into contiguous kernel virtual space.
1845 * For tight control over page level allocator and protection flags
1846 * use __vmalloc() instead.
1848 void *vmalloc(unsigned long size)
1850 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1853 EXPORT_SYMBOL(vmalloc);
1856 * vzalloc - allocate virtually contiguous memory with zero fill
1857 * @size: allocation size
1858 * Allocate enough pages to cover @size from the page level
1859 * allocator and map them into contiguous kernel virtual space.
1860 * The memory allocated is set to zero.
1862 * For tight control over page level allocator and protection flags
1863 * use __vmalloc() instead.
1865 void *vzalloc(unsigned long size)
1867 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1868 GFP_KERNEL | __GFP_ZERO);
1870 EXPORT_SYMBOL(vzalloc);
1873 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1874 * @size: allocation size
1876 * The resulting memory area is zeroed so it can be mapped to userspace
1877 * without leaking data.
1879 void *vmalloc_user(unsigned long size)
1881 return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
1882 GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL,
1883 VM_USERMAP, NUMA_NO_NODE,
1884 __builtin_return_address(0));
1886 EXPORT_SYMBOL(vmalloc_user);
1889 * vmalloc_node - allocate memory on a specific node
1890 * @size: allocation size
1893 * Allocate enough pages to cover @size from the page level
1894 * allocator and map them into contiguous kernel virtual space.
1896 * For tight control over page level allocator and protection flags
1897 * use __vmalloc() instead.
1899 void *vmalloc_node(unsigned long size, int node)
1901 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
1902 node, __builtin_return_address(0));
1904 EXPORT_SYMBOL(vmalloc_node);
1907 * vzalloc_node - allocate memory on a specific node with zero fill
1908 * @size: allocation size
1911 * Allocate enough pages to cover @size from the page level
1912 * allocator and map them into contiguous kernel virtual space.
1913 * The memory allocated is set to zero.
1915 * For tight control over page level allocator and protection flags
1916 * use __vmalloc_node() instead.
1918 void *vzalloc_node(unsigned long size, int node)
1920 return __vmalloc_node_flags(size, node,
1921 GFP_KERNEL | __GFP_ZERO);
1923 EXPORT_SYMBOL(vzalloc_node);
1926 * vmalloc_exec - allocate virtually contiguous, executable memory
1927 * @size: allocation size
1929 * Kernel-internal function to allocate enough pages to cover @size
1930 * the page level allocator and map them into contiguous and
1931 * executable kernel virtual space.
1933 * For tight control over page level allocator and protection flags
1934 * use __vmalloc() instead.
1937 void *vmalloc_exec(unsigned long size)
1939 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1940 NUMA_NO_NODE, __builtin_return_address(0));
1943 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1944 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
1945 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1946 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
1949 * 64b systems should always have either DMA or DMA32 zones. For others
1950 * GFP_DMA32 should do the right thing and use the normal zone.
1952 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1956 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1957 * @size: allocation size
1959 * Allocate enough 32bit PA addressable pages to cover @size from the
1960 * page level allocator and map them into contiguous kernel virtual space.
1962 void *vmalloc_32(unsigned long size)
1964 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1965 NUMA_NO_NODE, __builtin_return_address(0));
1967 EXPORT_SYMBOL(vmalloc_32);
1970 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1971 * @size: allocation size
1973 * The resulting memory area is 32bit addressable and zeroed so it can be
1974 * mapped to userspace without leaking data.
1976 void *vmalloc_32_user(unsigned long size)
1978 return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
1979 GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1980 VM_USERMAP, NUMA_NO_NODE,
1981 __builtin_return_address(0));
1983 EXPORT_SYMBOL(vmalloc_32_user);
1986 * small helper routine , copy contents to buf from addr.
1987 * If the page is not present, fill zero.
1990 static int aligned_vread(char *buf, char *addr, unsigned long count)
1996 unsigned long offset, length;
1998 offset = offset_in_page(addr);
1999 length = PAGE_SIZE - offset;
2002 p = vmalloc_to_page(addr);
2004 * To do safe access to this _mapped_ area, we need
2005 * lock. But adding lock here means that we need to add
2006 * overhead of vmalloc()/vfree() calles for this _debug_
2007 * interface, rarely used. Instead of that, we'll use
2008 * kmap() and get small overhead in this access function.
2012 * we can expect USER0 is not used (see vread/vwrite's
2013 * function description)
2015 void *map = kmap_atomic(p);
2016 memcpy(buf, map + offset, length);
2019 memset(buf, 0, length);
2029 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2035 unsigned long offset, length;
2037 offset = offset_in_page(addr);
2038 length = PAGE_SIZE - offset;
2041 p = vmalloc_to_page(addr);
2043 * To do safe access to this _mapped_ area, we need
2044 * lock. But adding lock here means that we need to add
2045 * overhead of vmalloc()/vfree() calles for this _debug_
2046 * interface, rarely used. Instead of that, we'll use
2047 * kmap() and get small overhead in this access function.
2051 * we can expect USER0 is not used (see vread/vwrite's
2052 * function description)
2054 void *map = kmap_atomic(p);
2055 memcpy(map + offset, buf, length);
2067 * vread() - read vmalloc area in a safe way.
2068 * @buf: buffer for reading data
2069 * @addr: vm address.
2070 * @count: number of bytes to be read.
2072 * Returns # of bytes which addr and buf should be increased.
2073 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2074 * includes any intersect with alive vmalloc area.
2076 * This function checks that addr is a valid vmalloc'ed area, and
2077 * copy data from that area to a given buffer. If the given memory range
2078 * of [addr...addr+count) includes some valid address, data is copied to
2079 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2080 * IOREMAP area is treated as memory hole and no copy is done.
2082 * If [addr...addr+count) doesn't includes any intersects with alive
2083 * vm_struct area, returns 0. @buf should be kernel's buffer.
2085 * Note: In usual ops, vread() is never necessary because the caller
2086 * should know vmalloc() area is valid and can use memcpy().
2087 * This is for routines which have to access vmalloc area without
2088 * any informaion, as /dev/kmem.
2092 long vread(char *buf, char *addr, unsigned long count)
2094 struct vmap_area *va;
2095 struct vm_struct *vm;
2096 char *vaddr, *buf_start = buf;
2097 unsigned long buflen = count;
2100 /* Don't allow overflow */
2101 if ((unsigned long) addr + count < count)
2102 count = -(unsigned long) addr;
2104 spin_lock(&vmap_area_lock);
2105 list_for_each_entry(va, &vmap_area_list, list) {
2109 if (!(va->flags & VM_VM_AREA))
2113 vaddr = (char *) vm->addr;
2114 if (addr >= vaddr + get_vm_area_size(vm))
2116 while (addr < vaddr) {
2124 n = vaddr + get_vm_area_size(vm) - addr;
2127 if (!(vm->flags & VM_IOREMAP))
2128 aligned_vread(buf, addr, n);
2129 else /* IOREMAP area is treated as memory hole */
2136 spin_unlock(&vmap_area_lock);
2138 if (buf == buf_start)
2140 /* zero-fill memory holes */
2141 if (buf != buf_start + buflen)
2142 memset(buf, 0, buflen - (buf - buf_start));
2148 * vwrite() - write vmalloc area in a safe way.
2149 * @buf: buffer for source data
2150 * @addr: vm address.
2151 * @count: number of bytes to be read.
2153 * Returns # of bytes which addr and buf should be incresed.
2154 * (same number to @count).
2155 * If [addr...addr+count) doesn't includes any intersect with valid
2156 * vmalloc area, returns 0.
2158 * This function checks that addr is a valid vmalloc'ed area, and
2159 * copy data from a buffer to the given addr. If specified range of
2160 * [addr...addr+count) includes some valid address, data is copied from
2161 * proper area of @buf. If there are memory holes, no copy to hole.
2162 * IOREMAP area is treated as memory hole and no copy is done.
2164 * If [addr...addr+count) doesn't includes any intersects with alive
2165 * vm_struct area, returns 0. @buf should be kernel's buffer.
2167 * Note: In usual ops, vwrite() is never necessary because the caller
2168 * should know vmalloc() area is valid and can use memcpy().
2169 * This is for routines which have to access vmalloc area without
2170 * any informaion, as /dev/kmem.
2173 long vwrite(char *buf, char *addr, unsigned long count)
2175 struct vmap_area *va;
2176 struct vm_struct *vm;
2178 unsigned long n, buflen;
2181 /* Don't allow overflow */
2182 if ((unsigned long) addr + count < count)
2183 count = -(unsigned long) addr;
2186 spin_lock(&vmap_area_lock);
2187 list_for_each_entry(va, &vmap_area_list, list) {
2191 if (!(va->flags & VM_VM_AREA))
2195 vaddr = (char *) vm->addr;
2196 if (addr >= vaddr + get_vm_area_size(vm))
2198 while (addr < vaddr) {
2205 n = vaddr + get_vm_area_size(vm) - addr;
2208 if (!(vm->flags & VM_IOREMAP)) {
2209 aligned_vwrite(buf, addr, n);
2217 spin_unlock(&vmap_area_lock);
2224 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2225 * @vma: vma to cover
2226 * @uaddr: target user address to start at
2227 * @kaddr: virtual address of vmalloc kernel memory
2228 * @size: size of map area
2230 * Returns: 0 for success, -Exxx on failure
2232 * This function checks that @kaddr is a valid vmalloc'ed area,
2233 * and that it is big enough to cover the range starting at
2234 * @uaddr in @vma. Will return failure if that criteria isn't
2237 * Similar to remap_pfn_range() (see mm/memory.c)
2239 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2240 void *kaddr, unsigned long size)
2242 struct vm_struct *area;
2244 size = PAGE_ALIGN(size);
2246 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2249 area = find_vm_area(kaddr);
2253 if (!(area->flags & VM_USERMAP))
2256 if (kaddr + size > area->addr + get_vm_area_size(area))
2260 struct page *page = vmalloc_to_page(kaddr);
2263 ret = vm_insert_page(vma, uaddr, page);
2272 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2276 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2279 * remap_vmalloc_range - map vmalloc pages to userspace
2280 * @vma: vma to cover (map full range of vma)
2281 * @addr: vmalloc memory
2282 * @pgoff: number of pages into addr before first page to map
2284 * Returns: 0 for success, -Exxx on failure
2286 * This function checks that addr is a valid vmalloc'ed area, and
2287 * that it is big enough to cover the vma. Will return failure if
2288 * that criteria isn't met.
2290 * Similar to remap_pfn_range() (see mm/memory.c)
2292 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2293 unsigned long pgoff)
2295 return remap_vmalloc_range_partial(vma, vma->vm_start,
2296 addr + (pgoff << PAGE_SHIFT),
2297 vma->vm_end - vma->vm_start);
2299 EXPORT_SYMBOL(remap_vmalloc_range);
2302 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2305 void __weak vmalloc_sync_all(void)
2310 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2322 * alloc_vm_area - allocate a range of kernel address space
2323 * @size: size of the area
2324 * @ptes: returns the PTEs for the address space
2326 * Returns: NULL on failure, vm_struct on success
2328 * This function reserves a range of kernel address space, and
2329 * allocates pagetables to map that range. No actual mappings
2332 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2333 * allocated for the VM area are returned.
2335 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2337 struct vm_struct *area;
2339 area = get_vm_area_caller(size, VM_IOREMAP,
2340 __builtin_return_address(0));
2345 * This ensures that page tables are constructed for this region
2346 * of kernel virtual address space and mapped into init_mm.
2348 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2349 size, f, ptes ? &ptes : NULL)) {
2356 EXPORT_SYMBOL_GPL(alloc_vm_area);
2358 void free_vm_area(struct vm_struct *area)
2360 struct vm_struct *ret;
2361 ret = remove_vm_area(area->addr);
2362 BUG_ON(ret != area);
2365 EXPORT_SYMBOL_GPL(free_vm_area);
2368 static struct vmap_area *node_to_va(struct rb_node *n)
2370 return rb_entry_safe(n, struct vmap_area, rb_node);
2374 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2375 * @end: target address
2376 * @pnext: out arg for the next vmap_area
2377 * @pprev: out arg for the previous vmap_area
2379 * Returns: %true if either or both of next and prev are found,
2380 * %false if no vmap_area exists
2382 * Find vmap_areas end addresses of which enclose @end. ie. if not
2383 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2385 static bool pvm_find_next_prev(unsigned long end,
2386 struct vmap_area **pnext,
2387 struct vmap_area **pprev)
2389 struct rb_node *n = vmap_area_root.rb_node;
2390 struct vmap_area *va = NULL;
2393 va = rb_entry(n, struct vmap_area, rb_node);
2394 if (end < va->va_end)
2396 else if (end > va->va_end)
2405 if (va->va_end > end) {
2407 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2410 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2416 * pvm_determine_end - find the highest aligned address between two vmap_areas
2417 * @pnext: in/out arg for the next vmap_area
2418 * @pprev: in/out arg for the previous vmap_area
2421 * Returns: determined end address
2423 * Find the highest aligned address between *@pnext and *@pprev below
2424 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2425 * down address is between the end addresses of the two vmap_areas.
2427 * Please note that the address returned by this function may fall
2428 * inside *@pnext vmap_area. The caller is responsible for checking
2431 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2432 struct vmap_area **pprev,
2433 unsigned long align)
2435 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2439 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2443 while (*pprev && (*pprev)->va_end > addr) {
2445 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2452 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2453 * @offsets: array containing offset of each area
2454 * @sizes: array containing size of each area
2455 * @nr_vms: the number of areas to allocate
2456 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2458 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2459 * vm_structs on success, %NULL on failure
2461 * Percpu allocator wants to use congruent vm areas so that it can
2462 * maintain the offsets among percpu areas. This function allocates
2463 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2464 * be scattered pretty far, distance between two areas easily going up
2465 * to gigabytes. To avoid interacting with regular vmallocs, these
2466 * areas are allocated from top.
2468 * Despite its complicated look, this allocator is rather simple. It
2469 * does everything top-down and scans areas from the end looking for
2470 * matching slot. While scanning, if any of the areas overlaps with
2471 * existing vmap_area, the base address is pulled down to fit the
2472 * area. Scanning is repeated till all the areas fit and then all
2473 * necessary data structures are inserted and the result is returned.
2475 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2476 const size_t *sizes, int nr_vms,
2479 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2480 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2481 struct vmap_area **vas, *prev, *next;
2482 struct vm_struct **vms;
2483 int area, area2, last_area, term_area;
2484 unsigned long base, start, end, last_end;
2485 bool purged = false;
2487 /* verify parameters and allocate data structures */
2488 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2489 for (last_area = 0, area = 0; area < nr_vms; area++) {
2490 start = offsets[area];
2491 end = start + sizes[area];
2493 /* is everything aligned properly? */
2494 BUG_ON(!IS_ALIGNED(offsets[area], align));
2495 BUG_ON(!IS_ALIGNED(sizes[area], align));
2497 /* detect the area with the highest address */
2498 if (start > offsets[last_area])
2501 for (area2 = area + 1; area2 < nr_vms; area2++) {
2502 unsigned long start2 = offsets[area2];
2503 unsigned long end2 = start2 + sizes[area2];
2505 BUG_ON(start2 < end && start < end2);
2508 last_end = offsets[last_area] + sizes[last_area];
2510 if (vmalloc_end - vmalloc_start < last_end) {
2515 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2516 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2520 for (area = 0; area < nr_vms; area++) {
2521 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2522 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2523 if (!vas[area] || !vms[area])
2527 spin_lock(&vmap_area_lock);
2529 /* start scanning - we scan from the top, begin with the last area */
2530 area = term_area = last_area;
2531 start = offsets[area];
2532 end = start + sizes[area];
2534 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2535 base = vmalloc_end - last_end;
2538 base = pvm_determine_end(&next, &prev, align) - end;
2541 BUG_ON(next && next->va_end <= base + end);
2542 BUG_ON(prev && prev->va_end > base + end);
2545 * base might have underflowed, add last_end before
2548 if (base + last_end < vmalloc_start + last_end) {
2549 spin_unlock(&vmap_area_lock);
2551 purge_vmap_area_lazy();
2559 * If next overlaps, move base downwards so that it's
2560 * right below next and then recheck.
2562 if (next && next->va_start < base + end) {
2563 base = pvm_determine_end(&next, &prev, align) - end;
2569 * If prev overlaps, shift down next and prev and move
2570 * base so that it's right below new next and then
2573 if (prev && prev->va_end > base + start) {
2575 prev = node_to_va(rb_prev(&next->rb_node));
2576 base = pvm_determine_end(&next, &prev, align) - end;
2582 * This area fits, move on to the previous one. If
2583 * the previous one is the terminal one, we're done.
2585 area = (area + nr_vms - 1) % nr_vms;
2586 if (area == term_area)
2588 start = offsets[area];
2589 end = start + sizes[area];
2590 pvm_find_next_prev(base + end, &next, &prev);
2593 /* we've found a fitting base, insert all va's */
2594 for (area = 0; area < nr_vms; area++) {
2595 struct vmap_area *va = vas[area];
2597 va->va_start = base + offsets[area];
2598 va->va_end = va->va_start + sizes[area];
2599 __insert_vmap_area(va);
2602 vmap_area_pcpu_hole = base + offsets[last_area];
2604 spin_unlock(&vmap_area_lock);
2606 /* insert all vm's */
2607 for (area = 0; area < nr_vms; area++)
2608 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2615 for (area = 0; area < nr_vms; area++) {
2626 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2627 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2628 * @nr_vms: the number of allocated areas
2630 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2632 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2636 for (i = 0; i < nr_vms; i++)
2637 free_vm_area(vms[i]);
2640 #endif /* CONFIG_SMP */
2642 #ifdef CONFIG_PROC_FS
2643 static void *s_start(struct seq_file *m, loff_t *pos)
2644 __acquires(&vmap_area_lock)
2646 spin_lock(&vmap_area_lock);
2647 return seq_list_start(&vmap_area_list, *pos);
2650 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2652 return seq_list_next(p, &vmap_area_list, pos);
2655 static void s_stop(struct seq_file *m, void *p)
2656 __releases(&vmap_area_lock)
2658 spin_unlock(&vmap_area_lock);
2661 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2663 if (IS_ENABLED(CONFIG_NUMA)) {
2664 unsigned int nr, *counters = m->private;
2669 if (v->flags & VM_UNINITIALIZED)
2671 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2674 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2676 for (nr = 0; nr < v->nr_pages; nr++)
2677 counters[page_to_nid(v->pages[nr])]++;
2679 for_each_node_state(nr, N_HIGH_MEMORY)
2681 seq_printf(m, " N%u=%u", nr, counters[nr]);
2685 static int s_show(struct seq_file *m, void *p)
2687 struct vmap_area *va;
2688 struct vm_struct *v;
2690 va = list_entry(p, struct vmap_area, list);
2693 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2694 * behalf of vmap area is being tear down or vm_map_ram allocation.
2696 if (!(va->flags & VM_VM_AREA)) {
2697 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2698 (void *)va->va_start, (void *)va->va_end,
2699 va->va_end - va->va_start,
2700 va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2707 seq_printf(m, "0x%pK-0x%pK %7ld",
2708 v->addr, v->addr + v->size, v->size);
2711 seq_printf(m, " %pS", v->caller);
2714 seq_printf(m, " pages=%d", v->nr_pages);
2717 seq_printf(m, " phys=%pa", &v->phys_addr);
2719 if (v->flags & VM_IOREMAP)
2720 seq_puts(m, " ioremap");
2722 if (v->flags & VM_ALLOC)
2723 seq_puts(m, " vmalloc");
2725 if (v->flags & VM_MAP)
2726 seq_puts(m, " vmap");
2728 if (v->flags & VM_USERMAP)
2729 seq_puts(m, " user");
2731 if (is_vmalloc_addr(v->pages))
2732 seq_puts(m, " vpages");
2734 show_numa_info(m, v);
2739 static const struct seq_operations vmalloc_op = {
2746 static int __init proc_vmalloc_init(void)
2748 if (IS_ENABLED(CONFIG_NUMA))
2749 proc_create_seq_private("vmallocinfo", 0400, NULL,
2751 nr_node_ids * sizeof(unsigned int), NULL);
2753 proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op);
2756 module_init(proc_vmalloc_init);