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_vm_area(struct vm_struct *area)
75 unsigned long addr = (unsigned long) area->addr;
76 unsigned long end = addr + area->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((unsigned long) area->addr, end);
90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
91 unsigned long end, pgprot_t prot, struct page ***pages)
95 pte = pte_alloc_kernel(pmd, addr);
99 struct page *page = **pages;
100 WARN_ON(!pte_none(*pte));
103 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
105 } while (pte++, addr += PAGE_SIZE, addr != end);
109 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
110 unsigned long end, pgprot_t prot, struct page ***pages)
115 pmd = pmd_alloc(&init_mm, pud, addr);
119 next = pmd_addr_end(addr, end);
120 if (vmap_pte_range(pmd, addr, next, prot, pages))
122 } while (pmd++, addr = next, addr != end);
126 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
127 unsigned long end, pgprot_t prot, struct page ***pages)
132 pud = pud_alloc(&init_mm, pgd, addr);
136 next = pud_addr_end(addr, end);
137 if (vmap_pmd_range(pud, addr, next, prot, pages))
139 } while (pud++, addr = next, addr != end);
143 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
147 unsigned long addr = (unsigned long) area->addr;
148 unsigned long end = addr + area->size - PAGE_SIZE;
152 pgd = pgd_offset_k(addr);
154 next = pgd_addr_end(addr, end);
155 err = vmap_pud_range(pgd, addr, next, prot, pages);
158 } while (pgd++, addr = next, addr != end);
159 flush_cache_vmap((unsigned long) area->addr, end);
163 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
164 unsigned long start, unsigned long end,
165 int node, gfp_t gfp_mask)
167 struct vm_struct **p, *tmp, *area;
168 unsigned long align = 1;
171 BUG_ON(in_interrupt());
172 if (flags & VM_IOREMAP) {
175 if (bit > IOREMAP_MAX_ORDER)
176 bit = IOREMAP_MAX_ORDER;
177 else if (bit < PAGE_SHIFT)
182 addr = ALIGN(start, align);
183 size = PAGE_ALIGN(size);
185 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_LEVEL_MASK, node);
193 * We always allocate a guard page.
197 write_lock(&vmlist_lock);
198 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
199 if ((unsigned long)tmp->addr < addr) {
200 if((unsigned long)tmp->addr + tmp->size >= addr)
201 addr = ALIGN(tmp->size +
202 (unsigned long)tmp->addr, align);
205 if ((size + addr) < addr)
207 if (size + addr <= (unsigned long)tmp->addr)
209 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
210 if (addr > end - size)
219 area->addr = (void *)addr;
224 write_unlock(&vmlist_lock);
229 write_unlock(&vmlist_lock);
231 if (printk_ratelimit())
232 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
236 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
237 unsigned long start, unsigned long end)
239 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
243 * get_vm_area - reserve a contingous kernel virtual area
244 * @size: size of the area
245 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
247 * Search an area of @size in the kernel virtual mapping area,
248 * and reserved it for out purposes. Returns the area descriptor
249 * on success or %NULL on failure.
251 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
253 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
256 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
257 int node, gfp_t gfp_mask)
259 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
263 /* Caller must hold vmlist_lock */
264 static struct vm_struct *__find_vm_area(void *addr)
266 struct vm_struct *tmp;
268 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
269 if (tmp->addr == addr)
276 /* Caller must hold vmlist_lock */
277 static struct vm_struct *__remove_vm_area(void *addr)
279 struct vm_struct **p, *tmp;
281 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
282 if (tmp->addr == addr)
292 * Remove the guard page.
294 tmp->size -= PAGE_SIZE;
299 * remove_vm_area - find and remove a contingous kernel virtual area
300 * @addr: base address
302 * Search for the kernel VM area starting at @addr, and remove it.
303 * This function returns the found VM area, but using it is NOT safe
304 * on SMP machines, except for its size or flags.
306 struct vm_struct *remove_vm_area(void *addr)
309 write_lock(&vmlist_lock);
310 v = __remove_vm_area(addr);
311 write_unlock(&vmlist_lock);
315 void __vunmap(void *addr, int deallocate_pages)
317 struct vm_struct *area;
322 if ((PAGE_SIZE-1) & (unsigned long)addr) {
323 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
328 area = remove_vm_area(addr);
329 if (unlikely(!area)) {
330 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
336 debug_check_no_locks_freed(addr, area->size);
338 if (deallocate_pages) {
341 for (i = 0; i < area->nr_pages; i++) {
342 BUG_ON(!area->pages[i]);
343 __free_page(area->pages[i]);
346 if (area->flags & VM_VPAGES)
357 * vfree - release memory allocated by vmalloc()
358 * @addr: memory base address
360 * Free the virtually contiguous memory area starting at @addr, as
361 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
362 * NULL, no operation is performed.
364 * Must not be called in interrupt context.
366 void vfree(void *addr)
368 BUG_ON(in_interrupt());
371 EXPORT_SYMBOL(vfree);
374 * vunmap - release virtual mapping obtained by vmap()
375 * @addr: memory base address
377 * Free the virtually contiguous memory area starting at @addr,
378 * which was created from the page array passed to vmap().
380 * Must not be called in interrupt context.
382 void vunmap(void *addr)
384 BUG_ON(in_interrupt());
387 EXPORT_SYMBOL(vunmap);
390 * vmap - map an array of pages into virtually contiguous space
391 * @pages: array of page pointers
392 * @count: number of pages to map
393 * @flags: vm_area->flags
394 * @prot: page protection for the mapping
396 * Maps @count pages from @pages into contiguous kernel virtual
399 void *vmap(struct page **pages, unsigned int count,
400 unsigned long flags, pgprot_t prot)
402 struct vm_struct *area;
404 if (count > num_physpages)
407 area = get_vm_area((count << PAGE_SHIFT), flags);
410 if (map_vm_area(area, prot, &pages)) {
419 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
420 pgprot_t prot, int node)
423 unsigned int nr_pages, array_size, i;
425 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
426 array_size = (nr_pages * sizeof(struct page *));
428 area->nr_pages = nr_pages;
429 /* Please note that the recursion is strictly bounded. */
430 if (array_size > PAGE_SIZE) {
431 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
432 area->flags |= VM_VPAGES;
434 pages = kmalloc_node(array_size,
435 (gfp_mask & ~(__GFP_HIGHMEM | __GFP_ZERO)),
440 remove_vm_area(area->addr);
444 memset(area->pages, 0, array_size);
446 for (i = 0; i < area->nr_pages; i++) {
448 area->pages[i] = alloc_page(gfp_mask);
450 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
451 if (unlikely(!area->pages[i])) {
452 /* Successfully allocated i pages, free them in __vunmap() */
458 if (map_vm_area(area, prot, &pages))
467 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
469 return __vmalloc_area_node(area, gfp_mask, prot, -1);
473 * __vmalloc_node - allocate virtually contiguous memory
474 * @size: allocation size
475 * @gfp_mask: flags for the page level allocator
476 * @prot: protection mask for the allocated pages
477 * @node: node to use for allocation or -1
479 * Allocate enough pages to cover @size from the page level
480 * allocator with @gfp_mask flags. Map them into contiguous
481 * kernel virtual space, using a pagetable protection of @prot.
483 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
486 struct vm_struct *area;
488 size = PAGE_ALIGN(size);
489 if (!size || (size >> PAGE_SHIFT) > num_physpages)
492 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
496 return __vmalloc_area_node(area, gfp_mask, prot, node);
499 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
501 return __vmalloc_node(size, gfp_mask, prot, -1);
503 EXPORT_SYMBOL(__vmalloc);
506 * vmalloc - allocate virtually contiguous memory
507 * @size: allocation size
508 * Allocate enough pages to cover @size from the page level
509 * allocator and map them into contiguous kernel virtual space.
511 * For tight control over page level allocator and protection flags
512 * use __vmalloc() instead.
514 void *vmalloc(unsigned long size)
516 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
518 EXPORT_SYMBOL(vmalloc);
521 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
522 * @size: allocation size
524 * The resulting memory area is zeroed so it can be mapped to userspace
525 * without leaking data.
527 void *vmalloc_user(unsigned long size)
529 struct vm_struct *area;
532 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
534 write_lock(&vmlist_lock);
535 area = __find_vm_area(ret);
536 area->flags |= VM_USERMAP;
537 write_unlock(&vmlist_lock);
541 EXPORT_SYMBOL(vmalloc_user);
544 * vmalloc_node - allocate memory on a specific node
545 * @size: allocation size
548 * Allocate enough pages to cover @size from the page level
549 * allocator and map them into contiguous kernel virtual space.
551 * For tight control over page level allocator and protection flags
552 * use __vmalloc() instead.
554 void *vmalloc_node(unsigned long size, int node)
556 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
558 EXPORT_SYMBOL(vmalloc_node);
560 #ifndef PAGE_KERNEL_EXEC
561 # define PAGE_KERNEL_EXEC PAGE_KERNEL
565 * vmalloc_exec - allocate virtually contiguous, executable memory
566 * @size: allocation size
568 * Kernel-internal function to allocate enough pages to cover @size
569 * the page level allocator and map them into contiguous and
570 * executable kernel virtual space.
572 * For tight control over page level allocator and protection flags
573 * use __vmalloc() instead.
576 void *vmalloc_exec(unsigned long size)
578 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
582 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
583 * @size: allocation size
585 * Allocate enough 32bit PA addressable pages to cover @size from the
586 * page level allocator and map them into contiguous kernel virtual space.
588 void *vmalloc_32(unsigned long size)
590 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
592 EXPORT_SYMBOL(vmalloc_32);
595 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
596 * @size: allocation size
598 * The resulting memory area is 32bit addressable and zeroed so it can be
599 * mapped to userspace without leaking data.
601 void *vmalloc_32_user(unsigned long size)
603 struct vm_struct *area;
606 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
608 write_lock(&vmlist_lock);
609 area = __find_vm_area(ret);
610 area->flags |= VM_USERMAP;
611 write_unlock(&vmlist_lock);
615 EXPORT_SYMBOL(vmalloc_32_user);
617 long vread(char *buf, char *addr, unsigned long count)
619 struct vm_struct *tmp;
620 char *vaddr, *buf_start = buf;
623 /* Don't allow overflow */
624 if ((unsigned long) addr + count < count)
625 count = -(unsigned long) addr;
627 read_lock(&vmlist_lock);
628 for (tmp = vmlist; tmp; tmp = tmp->next) {
629 vaddr = (char *) tmp->addr;
630 if (addr >= vaddr + tmp->size - PAGE_SIZE)
632 while (addr < vaddr) {
640 n = vaddr + tmp->size - PAGE_SIZE - addr;
651 read_unlock(&vmlist_lock);
652 return buf - buf_start;
655 long vwrite(char *buf, char *addr, unsigned long count)
657 struct vm_struct *tmp;
658 char *vaddr, *buf_start = buf;
661 /* Don't allow overflow */
662 if ((unsigned long) addr + count < count)
663 count = -(unsigned long) addr;
665 read_lock(&vmlist_lock);
666 for (tmp = vmlist; tmp; tmp = tmp->next) {
667 vaddr = (char *) tmp->addr;
668 if (addr >= vaddr + tmp->size - PAGE_SIZE)
670 while (addr < vaddr) {
677 n = vaddr + tmp->size - PAGE_SIZE - addr;
688 read_unlock(&vmlist_lock);
689 return buf - buf_start;
693 * remap_vmalloc_range - map vmalloc pages to userspace
694 * @vma: vma to cover (map full range of vma)
695 * @addr: vmalloc memory
696 * @pgoff: number of pages into addr before first page to map
697 * @returns: 0 for success, -Exxx on failure
699 * This function checks that addr is a valid vmalloc'ed area, and
700 * that it is big enough to cover the vma. Will return failure if
701 * that criteria isn't met.
703 * Similar to remap_pfn_range (see mm/memory.c)
705 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
708 struct vm_struct *area;
709 unsigned long uaddr = vma->vm_start;
710 unsigned long usize = vma->vm_end - vma->vm_start;
713 if ((PAGE_SIZE-1) & (unsigned long)addr)
716 read_lock(&vmlist_lock);
717 area = __find_vm_area(addr);
719 goto out_einval_locked;
721 if (!(area->flags & VM_USERMAP))
722 goto out_einval_locked;
724 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
725 goto out_einval_locked;
726 read_unlock(&vmlist_lock);
728 addr += pgoff << PAGE_SHIFT;
730 struct page *page = vmalloc_to_page(addr);
731 ret = vm_insert_page(vma, uaddr, page);
740 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
741 vma->vm_flags |= VM_RESERVED;
746 read_unlock(&vmlist_lock);
749 EXPORT_SYMBOL(remap_vmalloc_range);