4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
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>
18 #include <linux/vmalloc.h>
20 #include <asm/uaccess.h>
21 #include <asm/tlbflush.h>
24 DEFINE_RWLOCK(vmlist_lock);
25 struct vm_struct *vmlist;
27 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
30 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
34 pte = pte_offset_kernel(pmd, addr);
36 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
37 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
38 } while (pte++, addr += PAGE_SIZE, addr != end);
41 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
47 pmd = pmd_offset(pud, addr);
49 next = pmd_addr_end(addr, end);
50 if (pmd_none_or_clear_bad(pmd))
52 vunmap_pte_range(pmd, addr, next);
53 } while (pmd++, addr = next, addr != end);
56 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
62 pud = pud_offset(pgd, addr);
64 next = pud_addr_end(addr, end);
65 if (pud_none_or_clear_bad(pud))
67 vunmap_pmd_range(pud, addr, next);
68 } while (pud++, addr = next, addr != end);
71 void unmap_kernel_range(unsigned long addr, unsigned long size)
75 unsigned long start = addr;
76 unsigned long end = addr + size;
79 pgd = pgd_offset_k(addr);
80 flush_cache_vunmap(addr, end);
82 next = pgd_addr_end(addr, end);
83 if (pgd_none_or_clear_bad(pgd))
85 vunmap_pud_range(pgd, addr, next);
86 } while (pgd++, addr = next, addr != end);
87 flush_tlb_kernel_range(start, end);
90 static void unmap_vm_area(struct vm_struct *area)
92 unmap_kernel_range((unsigned long)area->addr, area->size);
95 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
96 unsigned long end, pgprot_t prot, struct page ***pages)
100 pte = pte_alloc_kernel(pmd, addr);
104 struct page *page = **pages;
105 WARN_ON(!pte_none(*pte));
108 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
110 } while (pte++, addr += PAGE_SIZE, addr != end);
114 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
115 unsigned long end, pgprot_t prot, struct page ***pages)
120 pmd = pmd_alloc(&init_mm, pud, addr);
124 next = pmd_addr_end(addr, end);
125 if (vmap_pte_range(pmd, addr, next, prot, pages))
127 } while (pmd++, addr = next, addr != end);
131 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
132 unsigned long end, pgprot_t prot, struct page ***pages)
137 pud = pud_alloc(&init_mm, pgd, addr);
141 next = pud_addr_end(addr, end);
142 if (vmap_pmd_range(pud, addr, next, prot, pages))
144 } while (pud++, addr = next, addr != end);
148 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
152 unsigned long addr = (unsigned long) area->addr;
153 unsigned long end = addr + area->size - PAGE_SIZE;
157 pgd = pgd_offset_k(addr);
159 next = pgd_addr_end(addr, end);
160 err = vmap_pud_range(pgd, addr, next, prot, pages);
163 } while (pgd++, addr = next, addr != end);
164 flush_cache_vmap((unsigned long) area->addr, end);
167 EXPORT_SYMBOL_GPL(map_vm_area);
170 * Map a vmalloc()-space virtual address to the physical page.
172 struct page *vmalloc_to_page(const void *vmalloc_addr)
174 unsigned long addr = (unsigned long) vmalloc_addr;
175 struct page *page = NULL;
176 pgd_t *pgd = pgd_offset_k(addr);
181 if (!pgd_none(*pgd)) {
182 pud = pud_offset(pgd, addr);
183 if (!pud_none(*pud)) {
184 pmd = pmd_offset(pud, addr);
185 if (!pmd_none(*pmd)) {
186 ptep = pte_offset_map(pmd, addr);
188 if (pte_present(pte))
189 page = pte_page(pte);
196 EXPORT_SYMBOL(vmalloc_to_page);
199 * Map a vmalloc()-space virtual address to the physical page frame number.
201 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
203 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
205 EXPORT_SYMBOL(vmalloc_to_pfn);
207 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
208 unsigned long start, unsigned long end,
209 int node, gfp_t gfp_mask)
211 struct vm_struct **p, *tmp, *area;
212 unsigned long align = 1;
215 BUG_ON(in_interrupt());
216 if (flags & VM_IOREMAP) {
219 if (bit > IOREMAP_MAX_ORDER)
220 bit = IOREMAP_MAX_ORDER;
221 else if (bit < PAGE_SHIFT)
226 addr = ALIGN(start, align);
227 size = PAGE_ALIGN(size);
231 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
237 * We always allocate a guard page.
241 write_lock(&vmlist_lock);
242 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
243 if ((unsigned long)tmp->addr < addr) {
244 if((unsigned long)tmp->addr + tmp->size >= addr)
245 addr = ALIGN(tmp->size +
246 (unsigned long)tmp->addr, align);
249 if ((size + addr) < addr)
251 if (size + addr <= (unsigned long)tmp->addr)
253 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
254 if (addr > end - size)
263 area->addr = (void *)addr;
268 write_unlock(&vmlist_lock);
273 write_unlock(&vmlist_lock);
275 if (printk_ratelimit())
276 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
280 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
281 unsigned long start, unsigned long end)
283 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
285 EXPORT_SYMBOL_GPL(__get_vm_area);
288 * get_vm_area - reserve a contiguous kernel virtual area
289 * @size: size of the area
290 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
292 * Search an area of @size in the kernel virtual mapping area,
293 * and reserved it for out purposes. Returns the area descriptor
294 * on success or %NULL on failure.
296 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
298 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
301 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
302 int node, gfp_t gfp_mask)
304 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
308 /* Caller must hold vmlist_lock */
309 static struct vm_struct *__find_vm_area(const void *addr)
311 struct vm_struct *tmp;
313 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
314 if (tmp->addr == addr)
321 /* Caller must hold vmlist_lock */
322 static struct vm_struct *__remove_vm_area(const void *addr)
324 struct vm_struct **p, *tmp;
326 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
327 if (tmp->addr == addr)
337 * Remove the guard page.
339 tmp->size -= PAGE_SIZE;
344 * remove_vm_area - find and remove a continuous kernel virtual area
345 * @addr: base address
347 * Search for the kernel VM area starting at @addr, and remove it.
348 * This function returns the found VM area, but using it is NOT safe
349 * on SMP machines, except for its size or flags.
351 struct vm_struct *remove_vm_area(const void *addr)
354 write_lock(&vmlist_lock);
355 v = __remove_vm_area(addr);
356 write_unlock(&vmlist_lock);
360 static void __vunmap(const void *addr, int deallocate_pages)
362 struct vm_struct *area;
367 if ((PAGE_SIZE-1) & (unsigned long)addr) {
368 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
373 area = remove_vm_area(addr);
374 if (unlikely(!area)) {
375 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
381 debug_check_no_locks_freed(addr, area->size);
383 if (deallocate_pages) {
386 for (i = 0; i < area->nr_pages; i++) {
387 struct page *page = area->pages[i];
393 if (area->flags & VM_VPAGES)
404 * vfree - release memory allocated by vmalloc()
405 * @addr: memory base address
407 * Free the virtually continuous memory area starting at @addr, as
408 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
409 * NULL, no operation is performed.
411 * Must not be called in interrupt context.
413 void vfree(const void *addr)
415 BUG_ON(in_interrupt());
418 EXPORT_SYMBOL(vfree);
421 * vunmap - release virtual mapping obtained by vmap()
422 * @addr: memory base address
424 * Free the virtually contiguous memory area starting at @addr,
425 * which was created from the page array passed to vmap().
427 * Must not be called in interrupt context.
429 void vunmap(const void *addr)
431 BUG_ON(in_interrupt());
434 EXPORT_SYMBOL(vunmap);
437 * vmap - map an array of pages into virtually contiguous space
438 * @pages: array of page pointers
439 * @count: number of pages to map
440 * @flags: vm_area->flags
441 * @prot: page protection for the mapping
443 * Maps @count pages from @pages into contiguous kernel virtual
446 void *vmap(struct page **pages, unsigned int count,
447 unsigned long flags, pgprot_t prot)
449 struct vm_struct *area;
451 if (count > num_physpages)
454 area = get_vm_area((count << PAGE_SHIFT), flags);
457 if (map_vm_area(area, prot, &pages)) {
466 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
467 pgprot_t prot, int node)
470 unsigned int nr_pages, array_size, i;
472 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
473 array_size = (nr_pages * sizeof(struct page *));
475 area->nr_pages = nr_pages;
476 /* Please note that the recursion is strictly bounded. */
477 if (array_size > PAGE_SIZE) {
478 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
480 area->flags |= VM_VPAGES;
482 pages = kmalloc_node(array_size,
483 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
488 remove_vm_area(area->addr);
493 for (i = 0; i < area->nr_pages; i++) {
497 page = alloc_page(gfp_mask);
499 page = alloc_pages_node(node, gfp_mask, 0);
501 if (unlikely(!page)) {
502 /* Successfully allocated i pages, free them in __vunmap() */
506 area->pages[i] = page;
509 if (map_vm_area(area, prot, &pages))
518 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
520 return __vmalloc_area_node(area, gfp_mask, prot, -1);
524 * __vmalloc_node - allocate virtually contiguous memory
525 * @size: allocation size
526 * @gfp_mask: flags for the page level allocator
527 * @prot: protection mask for the allocated pages
528 * @node: node to use for allocation or -1
530 * Allocate enough pages to cover @size from the page level
531 * allocator with @gfp_mask flags. Map them into contiguous
532 * kernel virtual space, using a pagetable protection of @prot.
534 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
537 struct vm_struct *area;
539 size = PAGE_ALIGN(size);
540 if (!size || (size >> PAGE_SHIFT) > num_physpages)
543 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
547 return __vmalloc_area_node(area, gfp_mask, prot, node);
550 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
552 return __vmalloc_node(size, gfp_mask, prot, -1);
554 EXPORT_SYMBOL(__vmalloc);
557 * vmalloc - allocate virtually contiguous memory
558 * @size: allocation size
559 * Allocate enough pages to cover @size from the page level
560 * allocator and map them into contiguous kernel virtual space.
562 * For tight control over page level allocator and protection flags
563 * use __vmalloc() instead.
565 void *vmalloc(unsigned long size)
567 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
569 EXPORT_SYMBOL(vmalloc);
572 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
573 * @size: allocation size
575 * The resulting memory area is zeroed so it can be mapped to userspace
576 * without leaking data.
578 void *vmalloc_user(unsigned long size)
580 struct vm_struct *area;
583 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
585 write_lock(&vmlist_lock);
586 area = __find_vm_area(ret);
587 area->flags |= VM_USERMAP;
588 write_unlock(&vmlist_lock);
592 EXPORT_SYMBOL(vmalloc_user);
595 * vmalloc_node - allocate memory on a specific node
596 * @size: allocation size
599 * Allocate enough pages to cover @size from the page level
600 * allocator and map them into contiguous kernel virtual space.
602 * For tight control over page level allocator and protection flags
603 * use __vmalloc() instead.
605 void *vmalloc_node(unsigned long size, int node)
607 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
609 EXPORT_SYMBOL(vmalloc_node);
611 #ifndef PAGE_KERNEL_EXEC
612 # define PAGE_KERNEL_EXEC PAGE_KERNEL
616 * vmalloc_exec - allocate virtually contiguous, executable memory
617 * @size: allocation size
619 * Kernel-internal function to allocate enough pages to cover @size
620 * the page level allocator and map them into contiguous and
621 * executable kernel virtual space.
623 * For tight control over page level allocator and protection flags
624 * use __vmalloc() instead.
627 void *vmalloc_exec(unsigned long size)
629 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
632 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
633 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
634 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
635 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
637 #define GFP_VMALLOC32 GFP_KERNEL
641 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
642 * @size: allocation size
644 * Allocate enough 32bit PA addressable pages to cover @size from the
645 * page level allocator and map them into contiguous kernel virtual space.
647 void *vmalloc_32(unsigned long size)
649 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
651 EXPORT_SYMBOL(vmalloc_32);
654 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
655 * @size: allocation size
657 * The resulting memory area is 32bit addressable and zeroed so it can be
658 * mapped to userspace without leaking data.
660 void *vmalloc_32_user(unsigned long size)
662 struct vm_struct *area;
665 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
667 write_lock(&vmlist_lock);
668 area = __find_vm_area(ret);
669 area->flags |= VM_USERMAP;
670 write_unlock(&vmlist_lock);
674 EXPORT_SYMBOL(vmalloc_32_user);
676 long vread(char *buf, char *addr, unsigned long count)
678 struct vm_struct *tmp;
679 char *vaddr, *buf_start = buf;
682 /* Don't allow overflow */
683 if ((unsigned long) addr + count < count)
684 count = -(unsigned long) addr;
686 read_lock(&vmlist_lock);
687 for (tmp = vmlist; tmp; tmp = tmp->next) {
688 vaddr = (char *) tmp->addr;
689 if (addr >= vaddr + tmp->size - PAGE_SIZE)
691 while (addr < vaddr) {
699 n = vaddr + tmp->size - PAGE_SIZE - addr;
710 read_unlock(&vmlist_lock);
711 return buf - buf_start;
714 long vwrite(char *buf, char *addr, unsigned long count)
716 struct vm_struct *tmp;
717 char *vaddr, *buf_start = buf;
720 /* Don't allow overflow */
721 if ((unsigned long) addr + count < count)
722 count = -(unsigned long) addr;
724 read_lock(&vmlist_lock);
725 for (tmp = vmlist; tmp; tmp = tmp->next) {
726 vaddr = (char *) tmp->addr;
727 if (addr >= vaddr + tmp->size - PAGE_SIZE)
729 while (addr < vaddr) {
736 n = vaddr + tmp->size - PAGE_SIZE - addr;
747 read_unlock(&vmlist_lock);
748 return buf - buf_start;
752 * remap_vmalloc_range - map vmalloc pages to userspace
753 * @vma: vma to cover (map full range of vma)
754 * @addr: vmalloc memory
755 * @pgoff: number of pages into addr before first page to map
756 * @returns: 0 for success, -Exxx on failure
758 * This function checks that addr is a valid vmalloc'ed area, and
759 * that it is big enough to cover the vma. Will return failure if
760 * that criteria isn't met.
762 * Similar to remap_pfn_range() (see mm/memory.c)
764 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
767 struct vm_struct *area;
768 unsigned long uaddr = vma->vm_start;
769 unsigned long usize = vma->vm_end - vma->vm_start;
772 if ((PAGE_SIZE-1) & (unsigned long)addr)
775 read_lock(&vmlist_lock);
776 area = __find_vm_area(addr);
778 goto out_einval_locked;
780 if (!(area->flags & VM_USERMAP))
781 goto out_einval_locked;
783 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
784 goto out_einval_locked;
785 read_unlock(&vmlist_lock);
787 addr += pgoff << PAGE_SHIFT;
789 struct page *page = vmalloc_to_page(addr);
790 ret = vm_insert_page(vma, uaddr, page);
799 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
800 vma->vm_flags |= VM_RESERVED;
805 read_unlock(&vmlist_lock);
808 EXPORT_SYMBOL(remap_vmalloc_range);
811 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
814 void __attribute__((weak)) vmalloc_sync_all(void)
819 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
821 /* apply_to_page_range() does all the hard work. */
826 * alloc_vm_area - allocate a range of kernel address space
827 * @size: size of the area
828 * @returns: NULL on failure, vm_struct on success
830 * This function reserves a range of kernel address space, and
831 * allocates pagetables to map that range. No actual mappings
832 * are created. If the kernel address space is not shared
833 * between processes, it syncs the pagetable across all
836 struct vm_struct *alloc_vm_area(size_t size)
838 struct vm_struct *area;
840 area = get_vm_area(size, VM_IOREMAP);
845 * This ensures that page tables are constructed for this region
846 * of kernel virtual address space and mapped into init_mm.
848 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
849 area->size, f, NULL)) {
854 /* Make sure the pagetables are constructed in process kernel
860 EXPORT_SYMBOL_GPL(alloc_vm_area);
862 void free_vm_area(struct vm_struct *area)
864 struct vm_struct *ret;
865 ret = remove_vm_area(area->addr);
869 EXPORT_SYMBOL_GPL(free_vm_area);