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
12 #include <linux/module.h>
13 #include <linux/highmem.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/seq_file.h>
18 #include <linux/debugobjects.h>
19 #include <linux/vmalloc.h>
20 #include <linux/kallsyms.h>
22 #include <asm/uaccess.h>
23 #include <asm/tlbflush.h>
26 DEFINE_RWLOCK(vmlist_lock);
27 struct vm_struct *vmlist;
29 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
30 int node, void *caller);
32 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
36 pte = pte_offset_kernel(pmd, addr);
38 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
39 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
40 } while (pte++, addr += PAGE_SIZE, addr != end);
43 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
49 pmd = pmd_offset(pud, addr);
51 next = pmd_addr_end(addr, end);
52 if (pmd_none_or_clear_bad(pmd))
54 vunmap_pte_range(pmd, addr, next);
55 } while (pmd++, addr = next, addr != end);
58 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
64 pud = pud_offset(pgd, addr);
66 next = pud_addr_end(addr, end);
67 if (pud_none_or_clear_bad(pud))
69 vunmap_pmd_range(pud, addr, next);
70 } while (pud++, addr = next, addr != end);
73 void unmap_kernel_range(unsigned long addr, unsigned long size)
77 unsigned long start = addr;
78 unsigned long end = addr + size;
81 pgd = pgd_offset_k(addr);
82 flush_cache_vunmap(addr, end);
84 next = pgd_addr_end(addr, end);
85 if (pgd_none_or_clear_bad(pgd))
87 vunmap_pud_range(pgd, addr, next);
88 } while (pgd++, addr = next, addr != end);
89 flush_tlb_kernel_range(start, end);
92 static void unmap_vm_area(struct vm_struct *area)
94 unmap_kernel_range((unsigned long)area->addr, area->size);
97 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
98 unsigned long end, pgprot_t prot, struct page ***pages)
102 pte = pte_alloc_kernel(pmd, addr);
106 struct page *page = **pages;
107 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 inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page ***pages)
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))
129 } while (pmd++, addr = next, addr != end);
133 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page ***pages)
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))
146 } while (pud++, addr = next, addr != end);
150 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
154 unsigned long addr = (unsigned long) area->addr;
155 unsigned long end = addr + area->size - PAGE_SIZE;
159 pgd = pgd_offset_k(addr);
161 next = pgd_addr_end(addr, end);
162 err = vmap_pud_range(pgd, addr, next, prot, pages);
165 } while (pgd++, addr = next, addr != end);
166 flush_cache_vmap((unsigned long) area->addr, end);
169 EXPORT_SYMBOL_GPL(map_vm_area);
172 * Map a vmalloc()-space virtual address to the physical page.
174 struct page *vmalloc_to_page(const void *vmalloc_addr)
176 unsigned long addr = (unsigned long) vmalloc_addr;
177 struct page *page = NULL;
178 pgd_t *pgd = pgd_offset_k(addr);
183 if (!pgd_none(*pgd)) {
184 pud = pud_offset(pgd, addr);
185 if (!pud_none(*pud)) {
186 pmd = pmd_offset(pud, addr);
187 if (!pmd_none(*pmd)) {
188 ptep = pte_offset_map(pmd, addr);
190 if (pte_present(pte))
191 page = pte_page(pte);
198 EXPORT_SYMBOL(vmalloc_to_page);
201 * Map a vmalloc()-space virtual address to the physical page frame number.
203 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
205 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
207 EXPORT_SYMBOL(vmalloc_to_pfn);
209 static struct vm_struct *
210 __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
211 unsigned long end, int node, gfp_t gfp_mask, void *caller)
213 struct vm_struct **p, *tmp, *area;
214 unsigned long align = 1;
217 BUG_ON(in_interrupt());
218 if (flags & VM_IOREMAP) {
221 if (bit > IOREMAP_MAX_ORDER)
222 bit = IOREMAP_MAX_ORDER;
223 else if (bit < PAGE_SHIFT)
228 addr = ALIGN(start, align);
229 size = PAGE_ALIGN(size);
233 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
239 * We always allocate a guard page.
243 write_lock(&vmlist_lock);
244 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
245 if ((unsigned long)tmp->addr < addr) {
246 if((unsigned long)tmp->addr + tmp->size >= addr)
247 addr = ALIGN(tmp->size +
248 (unsigned long)tmp->addr, align);
251 if ((size + addr) < addr)
253 if (size + addr <= (unsigned long)tmp->addr)
255 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
256 if (addr > end - size)
259 if ((size + addr) < addr)
261 if (addr > end - size)
269 area->addr = (void *)addr;
274 area->caller = caller;
275 write_unlock(&vmlist_lock);
280 write_unlock(&vmlist_lock);
282 if (printk_ratelimit())
283 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
287 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
288 unsigned long start, unsigned long end)
290 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
291 __builtin_return_address(0));
293 EXPORT_SYMBOL_GPL(__get_vm_area);
296 * get_vm_area - reserve a contiguous kernel virtual area
297 * @size: size of the area
298 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
300 * Search an area of @size in the kernel virtual mapping area,
301 * and reserved it for out purposes. Returns the area descriptor
302 * on success or %NULL on failure.
304 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
306 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
307 -1, GFP_KERNEL, __builtin_return_address(0));
310 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
313 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
314 -1, GFP_KERNEL, caller);
317 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
318 int node, gfp_t gfp_mask)
320 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
321 gfp_mask, __builtin_return_address(0));
324 /* Caller must hold vmlist_lock */
325 static struct vm_struct *__find_vm_area(const void *addr)
327 struct vm_struct *tmp;
329 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
330 if (tmp->addr == addr)
337 /* Caller must hold vmlist_lock */
338 static struct vm_struct *__remove_vm_area(const void *addr)
340 struct vm_struct **p, *tmp;
342 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
343 if (tmp->addr == addr)
353 * Remove the guard page.
355 tmp->size -= PAGE_SIZE;
360 * remove_vm_area - find and remove a continuous kernel virtual area
361 * @addr: base address
363 * Search for the kernel VM area starting at @addr, and remove it.
364 * This function returns the found VM area, but using it is NOT safe
365 * on SMP machines, except for its size or flags.
367 struct vm_struct *remove_vm_area(const void *addr)
370 write_lock(&vmlist_lock);
371 v = __remove_vm_area(addr);
372 write_unlock(&vmlist_lock);
376 static void __vunmap(const void *addr, int deallocate_pages)
378 struct vm_struct *area;
383 if ((PAGE_SIZE-1) & (unsigned long)addr) {
384 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
389 area = remove_vm_area(addr);
390 if (unlikely(!area)) {
391 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
397 debug_check_no_locks_freed(addr, area->size);
398 debug_check_no_obj_freed(addr, area->size);
400 if (deallocate_pages) {
403 for (i = 0; i < area->nr_pages; i++) {
404 struct page *page = area->pages[i];
410 if (area->flags & VM_VPAGES)
421 * vfree - release memory allocated by vmalloc()
422 * @addr: memory base address
424 * Free the virtually continuous memory area starting at @addr, as
425 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
426 * NULL, no operation is performed.
428 * Must not be called in interrupt context.
430 void vfree(const void *addr)
432 BUG_ON(in_interrupt());
435 EXPORT_SYMBOL(vfree);
438 * vunmap - release virtual mapping obtained by vmap()
439 * @addr: memory base address
441 * Free the virtually contiguous memory area starting at @addr,
442 * which was created from the page array passed to vmap().
444 * Must not be called in interrupt context.
446 void vunmap(const void *addr)
448 BUG_ON(in_interrupt());
451 EXPORT_SYMBOL(vunmap);
454 * vmap - map an array of pages into virtually contiguous space
455 * @pages: array of page pointers
456 * @count: number of pages to map
457 * @flags: vm_area->flags
458 * @prot: page protection for the mapping
460 * Maps @count pages from @pages into contiguous kernel virtual
463 void *vmap(struct page **pages, unsigned int count,
464 unsigned long flags, pgprot_t prot)
466 struct vm_struct *area;
468 if (count > num_physpages)
471 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
472 __builtin_return_address(0));
476 if (map_vm_area(area, prot, &pages)) {
485 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
486 pgprot_t prot, int node, void *caller)
489 unsigned int nr_pages, array_size, i;
491 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
492 array_size = (nr_pages * sizeof(struct page *));
494 area->nr_pages = nr_pages;
495 /* Please note that the recursion is strictly bounded. */
496 if (array_size > PAGE_SIZE) {
497 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
498 PAGE_KERNEL, node, caller);
499 area->flags |= VM_VPAGES;
501 pages = kmalloc_node(array_size,
502 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
506 area->caller = caller;
508 remove_vm_area(area->addr);
513 for (i = 0; i < area->nr_pages; i++) {
517 page = alloc_page(gfp_mask);
519 page = alloc_pages_node(node, gfp_mask, 0);
521 if (unlikely(!page)) {
522 /* Successfully allocated i pages, free them in __vunmap() */
526 area->pages[i] = page;
529 if (map_vm_area(area, prot, &pages))
538 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
540 return __vmalloc_area_node(area, gfp_mask, prot, -1,
541 __builtin_return_address(0));
545 * __vmalloc_node - allocate virtually contiguous memory
546 * @size: allocation size
547 * @gfp_mask: flags for the page level allocator
548 * @prot: protection mask for the allocated pages
549 * @node: node to use for allocation or -1
550 * @caller: caller's return address
552 * Allocate enough pages to cover @size from the page level
553 * allocator with @gfp_mask flags. Map them into contiguous
554 * kernel virtual space, using a pagetable protection of @prot.
556 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
557 int node, void *caller)
559 struct vm_struct *area;
561 size = PAGE_ALIGN(size);
562 if (!size || (size >> PAGE_SHIFT) > num_physpages)
565 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
566 node, gfp_mask, caller);
571 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
574 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
576 return __vmalloc_node(size, gfp_mask, prot, -1,
577 __builtin_return_address(0));
579 EXPORT_SYMBOL(__vmalloc);
582 * vmalloc - allocate virtually contiguous memory
583 * @size: allocation size
584 * Allocate enough pages to cover @size from the page level
585 * allocator and map them into contiguous kernel virtual space.
587 * For tight control over page level allocator and protection flags
588 * use __vmalloc() instead.
590 void *vmalloc(unsigned long size)
592 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
593 -1, __builtin_return_address(0));
595 EXPORT_SYMBOL(vmalloc);
598 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
599 * @size: allocation size
601 * The resulting memory area is zeroed so it can be mapped to userspace
602 * without leaking data.
604 void *vmalloc_user(unsigned long size)
606 struct vm_struct *area;
609 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
611 write_lock(&vmlist_lock);
612 area = __find_vm_area(ret);
613 area->flags |= VM_USERMAP;
614 write_unlock(&vmlist_lock);
618 EXPORT_SYMBOL(vmalloc_user);
621 * vmalloc_node - allocate memory on a specific node
622 * @size: allocation size
625 * Allocate enough pages to cover @size from the page level
626 * allocator and map them into contiguous kernel virtual space.
628 * For tight control over page level allocator and protection flags
629 * use __vmalloc() instead.
631 void *vmalloc_node(unsigned long size, int node)
633 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
634 node, __builtin_return_address(0));
636 EXPORT_SYMBOL(vmalloc_node);
638 #ifndef PAGE_KERNEL_EXEC
639 # define PAGE_KERNEL_EXEC PAGE_KERNEL
643 * vmalloc_exec - allocate virtually contiguous, executable memory
644 * @size: allocation size
646 * Kernel-internal function to allocate enough pages to cover @size
647 * the page level allocator and map them into contiguous and
648 * executable kernel virtual space.
650 * For tight control over page level allocator and protection flags
651 * use __vmalloc() instead.
654 void *vmalloc_exec(unsigned long size)
656 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
659 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
660 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
661 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
662 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
664 #define GFP_VMALLOC32 GFP_KERNEL
668 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
669 * @size: allocation size
671 * Allocate enough 32bit PA addressable pages to cover @size from the
672 * page level allocator and map them into contiguous kernel virtual space.
674 void *vmalloc_32(unsigned long size)
676 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
678 EXPORT_SYMBOL(vmalloc_32);
681 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
682 * @size: allocation size
684 * The resulting memory area is 32bit addressable and zeroed so it can be
685 * mapped to userspace without leaking data.
687 void *vmalloc_32_user(unsigned long size)
689 struct vm_struct *area;
692 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
694 write_lock(&vmlist_lock);
695 area = __find_vm_area(ret);
696 area->flags |= VM_USERMAP;
697 write_unlock(&vmlist_lock);
701 EXPORT_SYMBOL(vmalloc_32_user);
703 long vread(char *buf, char *addr, unsigned long count)
705 struct vm_struct *tmp;
706 char *vaddr, *buf_start = buf;
709 /* Don't allow overflow */
710 if ((unsigned long) addr + count < count)
711 count = -(unsigned long) addr;
713 read_lock(&vmlist_lock);
714 for (tmp = vmlist; tmp; tmp = tmp->next) {
715 vaddr = (char *) tmp->addr;
716 if (addr >= vaddr + tmp->size - PAGE_SIZE)
718 while (addr < vaddr) {
726 n = vaddr + tmp->size - PAGE_SIZE - addr;
737 read_unlock(&vmlist_lock);
738 return buf - buf_start;
741 long vwrite(char *buf, char *addr, unsigned long count)
743 struct vm_struct *tmp;
744 char *vaddr, *buf_start = buf;
747 /* Don't allow overflow */
748 if ((unsigned long) addr + count < count)
749 count = -(unsigned long) addr;
751 read_lock(&vmlist_lock);
752 for (tmp = vmlist; tmp; tmp = tmp->next) {
753 vaddr = (char *) tmp->addr;
754 if (addr >= vaddr + tmp->size - PAGE_SIZE)
756 while (addr < vaddr) {
763 n = vaddr + tmp->size - PAGE_SIZE - addr;
774 read_unlock(&vmlist_lock);
775 return buf - buf_start;
779 * remap_vmalloc_range - map vmalloc pages to userspace
780 * @vma: vma to cover (map full range of vma)
781 * @addr: vmalloc memory
782 * @pgoff: number of pages into addr before first page to map
784 * Returns: 0 for success, -Exxx on failure
786 * This function checks that addr is a valid vmalloc'ed area, and
787 * that it is big enough to cover the vma. Will return failure if
788 * that criteria isn't met.
790 * Similar to remap_pfn_range() (see mm/memory.c)
792 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
795 struct vm_struct *area;
796 unsigned long uaddr = vma->vm_start;
797 unsigned long usize = vma->vm_end - vma->vm_start;
800 if ((PAGE_SIZE-1) & (unsigned long)addr)
803 read_lock(&vmlist_lock);
804 area = __find_vm_area(addr);
806 goto out_einval_locked;
808 if (!(area->flags & VM_USERMAP))
809 goto out_einval_locked;
811 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
812 goto out_einval_locked;
813 read_unlock(&vmlist_lock);
815 addr += pgoff << PAGE_SHIFT;
817 struct page *page = vmalloc_to_page(addr);
818 ret = vm_insert_page(vma, uaddr, page);
827 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
828 vma->vm_flags |= VM_RESERVED;
833 read_unlock(&vmlist_lock);
836 EXPORT_SYMBOL(remap_vmalloc_range);
839 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
842 void __attribute__((weak)) vmalloc_sync_all(void)
847 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
849 /* apply_to_page_range() does all the hard work. */
854 * alloc_vm_area - allocate a range of kernel address space
855 * @size: size of the area
857 * Returns: NULL on failure, vm_struct on success
859 * This function reserves a range of kernel address space, and
860 * allocates pagetables to map that range. No actual mappings
861 * are created. If the kernel address space is not shared
862 * between processes, it syncs the pagetable across all
865 struct vm_struct *alloc_vm_area(size_t size)
867 struct vm_struct *area;
869 area = get_vm_area_caller(size, VM_IOREMAP,
870 __builtin_return_address(0));
875 * This ensures that page tables are constructed for this region
876 * of kernel virtual address space and mapped into init_mm.
878 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
879 area->size, f, NULL)) {
884 /* Make sure the pagetables are constructed in process kernel
890 EXPORT_SYMBOL_GPL(alloc_vm_area);
892 void free_vm_area(struct vm_struct *area)
894 struct vm_struct *ret;
895 ret = remove_vm_area(area->addr);
899 EXPORT_SYMBOL_GPL(free_vm_area);
902 #ifdef CONFIG_PROC_FS
903 static void *s_start(struct seq_file *m, loff_t *pos)
908 read_lock(&vmlist_lock);
921 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
923 struct vm_struct *v = p;
929 static void s_stop(struct seq_file *m, void *p)
931 read_unlock(&vmlist_lock);
934 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
937 unsigned int nr, *counters = m->private;
942 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
944 for (nr = 0; nr < v->nr_pages; nr++)
945 counters[page_to_nid(v->pages[nr])]++;
947 for_each_node_state(nr, N_HIGH_MEMORY)
949 seq_printf(m, " N%u=%u", nr, counters[nr]);
953 static int s_show(struct seq_file *m, void *p)
955 struct vm_struct *v = p;
957 seq_printf(m, "0x%p-0x%p %7ld",
958 v->addr, v->addr + v->size, v->size);
961 char buff[2 * KSYM_NAME_LEN];
964 sprint_symbol(buff, (unsigned long)v->caller);
969 seq_printf(m, " pages=%d", v->nr_pages);
972 seq_printf(m, " phys=%lx", v->phys_addr);
974 if (v->flags & VM_IOREMAP)
975 seq_printf(m, " ioremap");
977 if (v->flags & VM_ALLOC)
978 seq_printf(m, " vmalloc");
980 if (v->flags & VM_MAP)
981 seq_printf(m, " vmap");
983 if (v->flags & VM_USERMAP)
984 seq_printf(m, " user");
986 if (v->flags & VM_VPAGES)
987 seq_printf(m, " vpages");
989 show_numa_info(m, v);
994 const struct seq_operations vmalloc_op = {