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 vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
31 pte = pte_offset_kernel(pmd, addr);
33 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
34 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
35 } while (pte++, addr += PAGE_SIZE, addr != end);
38 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
44 pmd = pmd_offset(pud, addr);
46 next = pmd_addr_end(addr, end);
47 if (pmd_none_or_clear_bad(pmd))
49 vunmap_pte_range(pmd, addr, next);
50 } while (pmd++, addr = next, addr != end);
53 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
59 pud = pud_offset(pgd, addr);
61 next = pud_addr_end(addr, end);
62 if (pud_none_or_clear_bad(pud))
64 vunmap_pmd_range(pud, addr, next);
65 } while (pud++, addr = next, addr != end);
68 void unmap_vm_area(struct vm_struct *area)
72 unsigned long addr = (unsigned long) area->addr;
73 unsigned long end = addr + area->size;
76 pgd = pgd_offset_k(addr);
77 flush_cache_vunmap(addr, end);
79 next = pgd_addr_end(addr, end);
80 if (pgd_none_or_clear_bad(pgd))
82 vunmap_pud_range(pgd, addr, next);
83 } while (pgd++, addr = next, addr != end);
84 flush_tlb_kernel_range((unsigned long) area->addr, end);
87 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
88 unsigned long end, pgprot_t prot, struct page ***pages)
92 pte = pte_alloc_kernel(pmd, addr);
96 struct page *page = **pages;
97 WARN_ON(!pte_none(*pte));
100 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
102 } while (pte++, addr += PAGE_SIZE, addr != end);
106 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
107 unsigned long end, pgprot_t prot, struct page ***pages)
112 pmd = pmd_alloc(&init_mm, pud, addr);
116 next = pmd_addr_end(addr, end);
117 if (vmap_pte_range(pmd, addr, next, prot, pages))
119 } while (pmd++, addr = next, addr != end);
123 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
124 unsigned long end, pgprot_t prot, struct page ***pages)
129 pud = pud_alloc(&init_mm, pgd, addr);
133 next = pud_addr_end(addr, end);
134 if (vmap_pmd_range(pud, addr, next, prot, pages))
136 } while (pud++, addr = next, addr != end);
140 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
144 unsigned long addr = (unsigned long) area->addr;
145 unsigned long end = addr + area->size - PAGE_SIZE;
149 pgd = pgd_offset_k(addr);
151 next = pgd_addr_end(addr, end);
152 err = vmap_pud_range(pgd, addr, next, prot, pages);
155 } while (pgd++, addr = next, addr != end);
156 flush_cache_vmap((unsigned long) area->addr, end);
160 struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
161 unsigned long start, unsigned long end, int node)
163 struct vm_struct **p, *tmp, *area;
164 unsigned long align = 1;
167 if (flags & VM_IOREMAP) {
170 if (bit > IOREMAP_MAX_ORDER)
171 bit = IOREMAP_MAX_ORDER;
172 else if (bit < PAGE_SHIFT)
177 addr = ALIGN(start, align);
178 size = PAGE_ALIGN(size);
180 area = kmalloc_node(sizeof(*area), GFP_KERNEL, node);
184 if (unlikely(!size)) {
190 * We always allocate a guard page.
194 write_lock(&vmlist_lock);
195 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
196 if ((unsigned long)tmp->addr < addr) {
197 if((unsigned long)tmp->addr + tmp->size >= addr)
198 addr = ALIGN(tmp->size +
199 (unsigned long)tmp->addr, align);
202 if ((size + addr) < addr)
204 if (size + addr <= (unsigned long)tmp->addr)
206 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
207 if (addr > end - size)
216 area->addr = (void *)addr;
221 write_unlock(&vmlist_lock);
226 write_unlock(&vmlist_lock);
228 if (printk_ratelimit())
229 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
233 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
234 unsigned long start, unsigned long end)
236 return __get_vm_area_node(size, flags, start, end, -1);
240 * get_vm_area - reserve a contingous kernel virtual area
242 * @size: size of the area
243 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
245 * Search an area of @size in the kernel virtual mapping area,
246 * and reserved it for out purposes. Returns the area descriptor
247 * on success or %NULL on failure.
249 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
251 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
254 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
256 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
259 /* Caller must hold vmlist_lock */
260 static struct vm_struct *__find_vm_area(void *addr)
262 struct vm_struct *tmp;
264 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
265 if (tmp->addr == addr)
272 /* Caller must hold vmlist_lock */
273 struct vm_struct *__remove_vm_area(void *addr)
275 struct vm_struct **p, *tmp;
277 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
278 if (tmp->addr == addr)
288 * Remove the guard page.
290 tmp->size -= PAGE_SIZE;
295 * remove_vm_area - find and remove a contingous kernel virtual area
297 * @addr: base address
299 * Search for the kernel VM area starting at @addr, and remove it.
300 * This function returns the found VM area, but using it is NOT safe
301 * on SMP machines, except for its size or flags.
303 struct vm_struct *remove_vm_area(void *addr)
306 write_lock(&vmlist_lock);
307 v = __remove_vm_area(addr);
308 write_unlock(&vmlist_lock);
312 void __vunmap(void *addr, int deallocate_pages)
314 struct vm_struct *area;
319 if ((PAGE_SIZE-1) & (unsigned long)addr) {
320 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
325 area = remove_vm_area(addr);
326 if (unlikely(!area)) {
327 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
333 debug_check_no_locks_freed(addr, area->size);
335 if (deallocate_pages) {
338 for (i = 0; i < area->nr_pages; i++) {
339 BUG_ON(!area->pages[i]);
340 __free_page(area->pages[i]);
343 if (area->flags & VM_VPAGES)
354 * vfree - release memory allocated by vmalloc()
356 * @addr: memory base address
358 * Free the virtually contiguous memory area starting at @addr, as
359 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
360 * NULL, no operation is performed.
362 * Must not be called in interrupt context.
364 void vfree(void *addr)
366 BUG_ON(in_interrupt());
369 EXPORT_SYMBOL(vfree);
372 * vunmap - release virtual mapping obtained by vmap()
374 * @addr: memory base address
376 * Free the virtually contiguous memory area starting at @addr,
377 * which was created from the page array passed to vmap().
379 * Must not be called in interrupt context.
381 void vunmap(void *addr)
383 BUG_ON(in_interrupt());
386 EXPORT_SYMBOL(vunmap);
389 * 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, (gfp_mask & ~__GFP_HIGHMEM), node);
437 remove_vm_area(area->addr);
441 memset(area->pages, 0, array_size);
443 for (i = 0; i < area->nr_pages; i++) {
445 area->pages[i] = alloc_page(gfp_mask);
447 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
448 if (unlikely(!area->pages[i])) {
449 /* Successfully allocated i pages, free them in __vunmap() */
455 if (map_vm_area(area, prot, &pages))
464 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
466 return __vmalloc_area_node(area, gfp_mask, prot, -1);
470 * __vmalloc_node - allocate virtually contiguous memory
472 * @size: allocation size
473 * @gfp_mask: flags for the page level allocator
474 * @prot: protection mask for the allocated pages
475 * @node: node to use for allocation or -1
477 * Allocate enough pages to cover @size from the page level
478 * allocator with @gfp_mask flags. Map them into contiguous
479 * kernel virtual space, using a pagetable protection of @prot.
481 void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
484 struct vm_struct *area;
486 size = PAGE_ALIGN(size);
487 if (!size || (size >> PAGE_SHIFT) > num_physpages)
490 area = get_vm_area_node(size, VM_ALLOC, node);
494 return __vmalloc_area_node(area, gfp_mask, prot, node);
496 EXPORT_SYMBOL(__vmalloc_node);
498 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
500 return __vmalloc_node(size, gfp_mask, prot, -1);
502 EXPORT_SYMBOL(__vmalloc);
505 * vmalloc - allocate virtually contiguous memory
507 * @size: allocation size
509 * Allocate enough pages to cover @size from the page level
510 * allocator and map them into contiguous kernel virtual space.
512 * For tight cotrol over page level allocator and protection flags
513 * use __vmalloc() instead.
515 void *vmalloc(unsigned long size)
517 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
519 EXPORT_SYMBOL(vmalloc);
522 * vmalloc_user - allocate virtually contiguous memory which has
523 * been zeroed so it can be mapped to userspace without
526 * @size: allocation size
528 void *vmalloc_user(unsigned long size)
530 struct vm_struct *area;
533 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
546 * @size: allocation size
549 * Allocate enough pages to cover @size from the page level
550 * allocator and map them into contiguous kernel virtual space.
552 * For tight cotrol over page level allocator and protection flags
553 * use __vmalloc() instead.
555 void *vmalloc_node(unsigned long size, int node)
557 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
559 EXPORT_SYMBOL(vmalloc_node);
561 #ifndef PAGE_KERNEL_EXEC
562 # define PAGE_KERNEL_EXEC PAGE_KERNEL
566 * vmalloc_exec - allocate virtually contiguous, executable memory
568 * @size: allocation size
570 * Kernel-internal function to allocate enough pages to cover @size
571 * the page level allocator and map them into contiguous and
572 * executable kernel virtual space.
574 * For tight cotrol over page level allocator and protection flags
575 * use __vmalloc() instead.
578 void *vmalloc_exec(unsigned long size)
580 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
584 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
586 * @size: allocation size
588 * Allocate enough 32bit PA addressable pages to cover @size from the
589 * page level allocator and map them into contiguous kernel virtual space.
591 void *vmalloc_32(unsigned long size)
593 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
595 EXPORT_SYMBOL(vmalloc_32);
598 * vmalloc_32_user - allocate virtually contiguous memory (32bit
599 * addressable) which is zeroed so it can be
600 * mapped to userspace without leaking data.
602 * @size: allocation size
604 void *vmalloc_32_user(unsigned long size)
606 struct vm_struct *area;
609 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
610 write_lock(&vmlist_lock);
611 area = __find_vm_area(ret);
612 area->flags |= VM_USERMAP;
613 write_unlock(&vmlist_lock);
617 EXPORT_SYMBOL(vmalloc_32_user);
619 long vread(char *buf, char *addr, unsigned long count)
621 struct vm_struct *tmp;
622 char *vaddr, *buf_start = buf;
625 /* Don't allow overflow */
626 if ((unsigned long) addr + count < count)
627 count = -(unsigned long) addr;
629 read_lock(&vmlist_lock);
630 for (tmp = vmlist; tmp; tmp = tmp->next) {
631 vaddr = (char *) tmp->addr;
632 if (addr >= vaddr + tmp->size - PAGE_SIZE)
634 while (addr < vaddr) {
642 n = vaddr + tmp->size - PAGE_SIZE - addr;
653 read_unlock(&vmlist_lock);
654 return buf - buf_start;
657 long vwrite(char *buf, char *addr, unsigned long count)
659 struct vm_struct *tmp;
660 char *vaddr, *buf_start = buf;
663 /* Don't allow overflow */
664 if ((unsigned long) addr + count < count)
665 count = -(unsigned long) addr;
667 read_lock(&vmlist_lock);
668 for (tmp = vmlist; tmp; tmp = tmp->next) {
669 vaddr = (char *) tmp->addr;
670 if (addr >= vaddr + tmp->size - PAGE_SIZE)
672 while (addr < vaddr) {
679 n = vaddr + tmp->size - PAGE_SIZE - addr;
690 read_unlock(&vmlist_lock);
691 return buf - buf_start;
695 * remap_vmalloc_range - map vmalloc pages to userspace
697 * @vma: vma to cover (map full range of vma)
698 * @addr: vmalloc memory
699 * @pgoff: number of pages into addr before first page to map
700 * @returns: 0 for success, -Exxx on failure
702 * This function checks that addr is a valid vmalloc'ed area, and
703 * that it is big enough to cover the vma. Will return failure if
704 * that criteria isn't met.
706 * Similar to remap_pfn_range (see mm/memory.c)
708 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
711 struct vm_struct *area;
712 unsigned long uaddr = vma->vm_start;
713 unsigned long usize = vma->vm_end - vma->vm_start;
716 if ((PAGE_SIZE-1) & (unsigned long)addr)
719 read_lock(&vmlist_lock);
720 area = __find_vm_area(addr);
722 goto out_einval_locked;
724 if (!(area->flags & VM_USERMAP))
725 goto out_einval_locked;
727 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
728 goto out_einval_locked;
729 read_unlock(&vmlist_lock);
731 addr += pgoff << PAGE_SHIFT;
733 struct page *page = vmalloc_to_page(addr);
734 ret = vm_insert_page(vma, uaddr, page);
743 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
744 vma->vm_flags |= VM_RESERVED;
749 read_unlock(&vmlist_lock);
752 EXPORT_SYMBOL(remap_vmalloc_range);