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>
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);
169 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
170 unsigned long start, unsigned long end,
171 int node, gfp_t gfp_mask)
173 struct vm_struct **p, *tmp, *area;
174 unsigned long align = 1;
177 BUG_ON(in_interrupt());
178 if (flags & VM_IOREMAP) {
181 if (bit > IOREMAP_MAX_ORDER)
182 bit = IOREMAP_MAX_ORDER;
183 else if (bit < PAGE_SHIFT)
188 addr = ALIGN(start, align);
189 size = PAGE_ALIGN(size);
193 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_LEVEL_MASK, node);
198 * We always allocate a guard page.
202 write_lock(&vmlist_lock);
203 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
204 if ((unsigned long)tmp->addr < addr) {
205 if((unsigned long)tmp->addr + tmp->size >= addr)
206 addr = ALIGN(tmp->size +
207 (unsigned long)tmp->addr, align);
210 if ((size + addr) < addr)
212 if (size + addr <= (unsigned long)tmp->addr)
214 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
215 if (addr > end - size)
224 area->addr = (void *)addr;
229 write_unlock(&vmlist_lock);
234 write_unlock(&vmlist_lock);
236 if (printk_ratelimit())
237 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
241 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
242 unsigned long start, unsigned long end)
244 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
246 EXPORT_SYMBOL_GPL(__get_vm_area);
249 * get_vm_area - reserve a contingous kernel virtual area
250 * @size: size of the area
251 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
253 * Search an area of @size in the kernel virtual mapping area,
254 * and reserved it for out purposes. Returns the area descriptor
255 * on success or %NULL on failure.
257 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
259 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
262 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
263 int node, gfp_t gfp_mask)
265 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
269 /* Caller must hold vmlist_lock */
270 static struct vm_struct *__find_vm_area(void *addr)
272 struct vm_struct *tmp;
274 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
275 if (tmp->addr == addr)
282 /* Caller must hold vmlist_lock */
283 static struct vm_struct *__remove_vm_area(void *addr)
285 struct vm_struct **p, *tmp;
287 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
288 if (tmp->addr == addr)
298 * Remove the guard page.
300 tmp->size -= PAGE_SIZE;
305 * remove_vm_area - find and remove a contingous kernel virtual area
306 * @addr: base address
308 * Search for the kernel VM area starting at @addr, and remove it.
309 * This function returns the found VM area, but using it is NOT safe
310 * on SMP machines, except for its size or flags.
312 struct vm_struct *remove_vm_area(void *addr)
315 write_lock(&vmlist_lock);
316 v = __remove_vm_area(addr);
317 write_unlock(&vmlist_lock);
321 static void __vunmap(void *addr, int deallocate_pages)
323 struct vm_struct *area;
328 if ((PAGE_SIZE-1) & (unsigned long)addr) {
329 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
334 area = remove_vm_area(addr);
335 if (unlikely(!area)) {
336 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
342 debug_check_no_locks_freed(addr, area->size);
344 if (deallocate_pages) {
347 for (i = 0; i < area->nr_pages; i++) {
348 BUG_ON(!area->pages[i]);
349 __free_page(area->pages[i]);
352 if (area->flags & VM_VPAGES)
363 * vfree - release memory allocated by vmalloc()
364 * @addr: memory base address
366 * Free the virtually contiguous memory area starting at @addr, as
367 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
368 * NULL, no operation is performed.
370 * Must not be called in interrupt context.
372 void vfree(void *addr)
374 BUG_ON(in_interrupt());
377 EXPORT_SYMBOL(vfree);
380 * vunmap - release virtual mapping obtained by vmap()
381 * @addr: memory base address
383 * Free the virtually contiguous memory area starting at @addr,
384 * which was created from the page array passed to vmap().
386 * Must not be called in interrupt context.
388 void vunmap(void *addr)
390 BUG_ON(in_interrupt());
393 EXPORT_SYMBOL(vunmap);
396 * vmap - map an array of pages into virtually contiguous space
397 * @pages: array of page pointers
398 * @count: number of pages to map
399 * @flags: vm_area->flags
400 * @prot: page protection for the mapping
402 * Maps @count pages from @pages into contiguous kernel virtual
405 void *vmap(struct page **pages, unsigned int count,
406 unsigned long flags, pgprot_t prot)
408 struct vm_struct *area;
410 if (count > num_physpages)
413 area = get_vm_area((count << PAGE_SHIFT), flags);
416 if (map_vm_area(area, prot, &pages)) {
425 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
426 pgprot_t prot, int node)
429 unsigned int nr_pages, array_size, i;
431 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
432 array_size = (nr_pages * sizeof(struct page *));
434 area->nr_pages = nr_pages;
435 /* Please note that the recursion is strictly bounded. */
436 if (array_size > PAGE_SIZE) {
437 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
439 area->flags |= VM_VPAGES;
441 pages = kmalloc_node(array_size,
442 (gfp_mask & GFP_LEVEL_MASK) | __GFP_ZERO,
447 remove_vm_area(area->addr);
452 for (i = 0; i < area->nr_pages; i++) {
454 area->pages[i] = alloc_page(gfp_mask);
456 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
457 if (unlikely(!area->pages[i])) {
458 /* Successfully allocated i pages, free them in __vunmap() */
464 if (map_vm_area(area, prot, &pages))
473 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
475 return __vmalloc_area_node(area, gfp_mask, prot, -1);
479 * __vmalloc_node - allocate virtually contiguous memory
480 * @size: allocation size
481 * @gfp_mask: flags for the page level allocator
482 * @prot: protection mask for the allocated pages
483 * @node: node to use for allocation or -1
485 * Allocate enough pages to cover @size from the page level
486 * allocator with @gfp_mask flags. Map them into contiguous
487 * kernel virtual space, using a pagetable protection of @prot.
489 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
492 struct vm_struct *area;
494 size = PAGE_ALIGN(size);
495 if (!size || (size >> PAGE_SHIFT) > num_physpages)
498 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
502 return __vmalloc_area_node(area, gfp_mask, prot, node);
505 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
507 return __vmalloc_node(size, gfp_mask, prot, -1);
509 EXPORT_SYMBOL(__vmalloc);
512 * vmalloc - allocate virtually contiguous memory
513 * @size: allocation size
514 * Allocate enough pages to cover @size from the page level
515 * allocator and map them into contiguous kernel virtual space.
517 * For tight control over page level allocator and protection flags
518 * use __vmalloc() instead.
520 void *vmalloc(unsigned long size)
522 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
524 EXPORT_SYMBOL(vmalloc);
527 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
528 * @size: allocation size
530 * The resulting memory area is zeroed so it can be mapped to userspace
531 * without leaking data.
533 void *vmalloc_user(unsigned long size)
535 struct vm_struct *area;
538 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
540 write_lock(&vmlist_lock);
541 area = __find_vm_area(ret);
542 area->flags |= VM_USERMAP;
543 write_unlock(&vmlist_lock);
547 EXPORT_SYMBOL(vmalloc_user);
550 * vmalloc_node - allocate memory on a specific node
551 * @size: allocation size
554 * Allocate enough pages to cover @size from the page level
555 * allocator and map them into contiguous kernel virtual space.
557 * For tight control over page level allocator and protection flags
558 * use __vmalloc() instead.
560 void *vmalloc_node(unsigned long size, int node)
562 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
564 EXPORT_SYMBOL(vmalloc_node);
566 #ifndef PAGE_KERNEL_EXEC
567 # define PAGE_KERNEL_EXEC PAGE_KERNEL
571 * vmalloc_exec - allocate virtually contiguous, executable memory
572 * @size: allocation size
574 * Kernel-internal function to allocate enough pages to cover @size
575 * the page level allocator and map them into contiguous and
576 * executable kernel virtual space.
578 * For tight control over page level allocator and protection flags
579 * use __vmalloc() instead.
582 void *vmalloc_exec(unsigned long size)
584 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
587 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
588 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
589 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
590 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
592 #define GFP_VMALLOC32 GFP_KERNEL
596 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
597 * @size: allocation size
599 * Allocate enough 32bit PA addressable pages to cover @size from the
600 * page level allocator and map them into contiguous kernel virtual space.
602 void *vmalloc_32(unsigned long size)
604 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
606 EXPORT_SYMBOL(vmalloc_32);
609 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
610 * @size: allocation size
612 * The resulting memory area is 32bit addressable and zeroed so it can be
613 * mapped to userspace without leaking data.
615 void *vmalloc_32_user(unsigned long size)
617 struct vm_struct *area;
620 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
622 write_lock(&vmlist_lock);
623 area = __find_vm_area(ret);
624 area->flags |= VM_USERMAP;
625 write_unlock(&vmlist_lock);
629 EXPORT_SYMBOL(vmalloc_32_user);
631 long vread(char *buf, char *addr, unsigned long count)
633 struct vm_struct *tmp;
634 char *vaddr, *buf_start = buf;
637 /* Don't allow overflow */
638 if ((unsigned long) addr + count < count)
639 count = -(unsigned long) addr;
641 read_lock(&vmlist_lock);
642 for (tmp = vmlist; tmp; tmp = tmp->next) {
643 vaddr = (char *) tmp->addr;
644 if (addr >= vaddr + tmp->size - PAGE_SIZE)
646 while (addr < vaddr) {
654 n = vaddr + tmp->size - PAGE_SIZE - addr;
665 read_unlock(&vmlist_lock);
666 return buf - buf_start;
669 long vwrite(char *buf, char *addr, unsigned long count)
671 struct vm_struct *tmp;
672 char *vaddr, *buf_start = buf;
675 /* Don't allow overflow */
676 if ((unsigned long) addr + count < count)
677 count = -(unsigned long) addr;
679 read_lock(&vmlist_lock);
680 for (tmp = vmlist; tmp; tmp = tmp->next) {
681 vaddr = (char *) tmp->addr;
682 if (addr >= vaddr + tmp->size - PAGE_SIZE)
684 while (addr < vaddr) {
691 n = vaddr + tmp->size - PAGE_SIZE - addr;
702 read_unlock(&vmlist_lock);
703 return buf - buf_start;
707 * remap_vmalloc_range - map vmalloc pages to userspace
708 * @vma: vma to cover (map full range of vma)
709 * @addr: vmalloc memory
710 * @pgoff: number of pages into addr before first page to map
711 * @returns: 0 for success, -Exxx on failure
713 * This function checks that addr is a valid vmalloc'ed area, and
714 * that it is big enough to cover the vma. Will return failure if
715 * that criteria isn't met.
717 * Similar to remap_pfn_range() (see mm/memory.c)
719 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
722 struct vm_struct *area;
723 unsigned long uaddr = vma->vm_start;
724 unsigned long usize = vma->vm_end - vma->vm_start;
727 if ((PAGE_SIZE-1) & (unsigned long)addr)
730 read_lock(&vmlist_lock);
731 area = __find_vm_area(addr);
733 goto out_einval_locked;
735 if (!(area->flags & VM_USERMAP))
736 goto out_einval_locked;
738 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
739 goto out_einval_locked;
740 read_unlock(&vmlist_lock);
742 addr += pgoff << PAGE_SHIFT;
744 struct page *page = vmalloc_to_page(addr);
745 ret = vm_insert_page(vma, uaddr, page);
754 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
755 vma->vm_flags |= VM_RESERVED;
760 read_unlock(&vmlist_lock);
763 EXPORT_SYMBOL(remap_vmalloc_range);
766 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
769 void __attribute__((weak)) vmalloc_sync_all(void)
774 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
776 /* apply_to_page_range() does all the hard work. */
781 * alloc_vm_area - allocate a range of kernel address space
782 * @size: size of the area
783 * @returns: NULL on failure, vm_struct on success
785 * This function reserves a range of kernel address space, and
786 * allocates pagetables to map that range. No actual mappings
787 * are created. If the kernel address space is not shared
788 * between processes, it syncs the pagetable across all
791 struct vm_struct *alloc_vm_area(size_t size)
793 struct vm_struct *area;
795 area = get_vm_area(size, VM_IOREMAP);
800 * This ensures that page tables are constructed for this region
801 * of kernel virtual address space and mapped into init_mm.
803 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
804 area->size, f, NULL)) {
809 /* Make sure the pagetables are constructed in process kernel
815 EXPORT_SYMBOL_GPL(alloc_vm_area);
817 void free_vm_area(struct vm_struct *area)
819 struct vm_struct *ret;
820 ret = remove_vm_area(area->addr);
824 EXPORT_SYMBOL_GPL(free_vm_area);