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
11 #include <linux/module.h>
12 #include <linux/highmem.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/interrupt.h>
17 #include <linux/vmalloc.h>
19 #include <asm/uaccess.h>
20 #include <asm/tlbflush.h>
23 DEFINE_RWLOCK(vmlist_lock);
24 struct vm_struct *vmlist;
26 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
30 pte = pte_offset_kernel(pmd, addr);
32 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
33 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
34 } while (pte++, addr += PAGE_SIZE, addr != end);
37 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
43 pmd = pmd_offset(pud, addr);
45 next = pmd_addr_end(addr, end);
46 if (pmd_none_or_clear_bad(pmd))
48 vunmap_pte_range(pmd, addr, next);
49 } while (pmd++, addr = next, addr != end);
52 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
58 pud = pud_offset(pgd, addr);
60 next = pud_addr_end(addr, end);
61 if (pud_none_or_clear_bad(pud))
63 vunmap_pmd_range(pud, addr, next);
64 } while (pud++, addr = next, addr != end);
67 void unmap_vm_area(struct vm_struct *area)
71 unsigned long addr = (unsigned long) area->addr;
72 unsigned long end = addr + area->size;
75 pgd = pgd_offset_k(addr);
76 flush_cache_vunmap(addr, end);
78 next = pgd_addr_end(addr, end);
79 if (pgd_none_or_clear_bad(pgd))
81 vunmap_pud_range(pgd, addr, next);
82 } while (pgd++, addr = next, addr != end);
83 flush_tlb_kernel_range((unsigned long) area->addr, end);
86 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
87 unsigned long end, pgprot_t prot, struct page ***pages)
91 pte = pte_alloc_kernel(&init_mm, pmd, addr);
95 struct page *page = **pages;
96 WARN_ON(!pte_none(*pte));
99 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
101 } while (pte++, addr += PAGE_SIZE, addr != end);
105 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
106 unsigned long end, pgprot_t prot, struct page ***pages)
111 pmd = pmd_alloc(&init_mm, pud, addr);
115 next = pmd_addr_end(addr, end);
116 if (vmap_pte_range(pmd, addr, next, prot, pages))
118 } while (pmd++, addr = next, addr != end);
122 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
123 unsigned long end, pgprot_t prot, struct page ***pages)
128 pud = pud_alloc(&init_mm, pgd, addr);
132 next = pud_addr_end(addr, end);
133 if (vmap_pmd_range(pud, addr, next, prot, pages))
135 } while (pud++, addr = next, addr != end);
139 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
143 unsigned long addr = (unsigned long) area->addr;
144 unsigned long end = addr + area->size - PAGE_SIZE;
148 pgd = pgd_offset_k(addr);
149 spin_lock(&init_mm.page_table_lock);
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 spin_unlock(&init_mm.page_table_lock);
157 flush_cache_vmap((unsigned long) area->addr, end);
161 #define IOREMAP_MAX_ORDER (7 + PAGE_SHIFT) /* 128 pages */
163 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
164 unsigned long start, unsigned long end)
166 struct vm_struct **p, *tmp, *area;
167 unsigned long align = 1;
170 if (flags & VM_IOREMAP) {
173 if (bit > IOREMAP_MAX_ORDER)
174 bit = IOREMAP_MAX_ORDER;
175 else if (bit < PAGE_SHIFT)
180 addr = ALIGN(start, align);
181 size = PAGE_ALIGN(size);
183 area = kmalloc(sizeof(*area), GFP_KERNEL);
187 if (unlikely(!size)) {
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");
237 * get_vm_area - reserve a contingous kernel virtual area
239 * @size: size of the area
240 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
242 * Search an area of @size in the kernel virtual mapping area,
243 * and reserved it for out purposes. Returns the area descriptor
244 * on success or %NULL on failure.
246 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
248 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
252 * remove_vm_area - find and remove a contingous kernel virtual area
254 * @addr: base address
256 * Search for the kernel VM area starting at @addr, and remove it.
257 * This function returns the found VM area, but using it is NOT safe
260 struct vm_struct *remove_vm_area(void *addr)
262 struct vm_struct **p, *tmp;
264 write_lock(&vmlist_lock);
265 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
266 if (tmp->addr == addr)
269 write_unlock(&vmlist_lock);
275 write_unlock(&vmlist_lock);
278 * Remove the guard page.
280 tmp->size -= PAGE_SIZE;
284 void __vunmap(void *addr, int deallocate_pages)
286 struct vm_struct *area;
291 if ((PAGE_SIZE-1) & (unsigned long)addr) {
292 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
297 area = remove_vm_area(addr);
298 if (unlikely(!area)) {
299 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
305 if (deallocate_pages) {
308 for (i = 0; i < area->nr_pages; i++) {
309 if (unlikely(!area->pages[i]))
311 __free_page(area->pages[i]);
314 if (area->nr_pages > PAGE_SIZE/sizeof(struct page *))
325 * vfree - release memory allocated by vmalloc()
327 * @addr: memory base address
329 * Free the virtually contiguous memory area starting at @addr, as
330 * obtained from vmalloc(), vmalloc_32() or __vmalloc().
332 * May not be called in interrupt context.
334 void vfree(void *addr)
336 BUG_ON(in_interrupt());
340 EXPORT_SYMBOL(vfree);
343 * vunmap - release virtual mapping obtained by vmap()
345 * @addr: memory base address
347 * Free the virtually contiguous memory area starting at @addr,
348 * which was created from the page array passed to vmap().
350 * May not be called in interrupt context.
352 void vunmap(void *addr)
354 BUG_ON(in_interrupt());
358 EXPORT_SYMBOL(vunmap);
361 * vmap - map an array of pages into virtually contiguous space
363 * @pages: array of page pointers
364 * @count: number of pages to map
365 * @flags: vm_area->flags
366 * @prot: page protection for the mapping
368 * Maps @count pages from @pages into contiguous kernel virtual
371 void *vmap(struct page **pages, unsigned int count,
372 unsigned long flags, pgprot_t prot)
374 struct vm_struct *area;
376 if (count > num_physpages)
379 area = get_vm_area((count << PAGE_SHIFT), flags);
382 if (map_vm_area(area, prot, &pages)) {
392 void *__vmalloc_area(struct vm_struct *area, unsigned int __nocast gfp_mask, pgprot_t prot)
395 unsigned int nr_pages, array_size, i;
397 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
398 array_size = (nr_pages * sizeof(struct page *));
400 area->nr_pages = nr_pages;
401 /* Please note that the recursion is strictly bounded. */
402 if (array_size > PAGE_SIZE)
403 pages = __vmalloc(array_size, gfp_mask, PAGE_KERNEL);
405 pages = kmalloc(array_size, (gfp_mask & ~__GFP_HIGHMEM));
408 remove_vm_area(area->addr);
412 memset(area->pages, 0, array_size);
414 for (i = 0; i < area->nr_pages; i++) {
415 area->pages[i] = alloc_page(gfp_mask);
416 if (unlikely(!area->pages[i])) {
417 /* Successfully allocated i pages, free them in __vunmap() */
423 if (map_vm_area(area, prot, &pages))
433 * __vmalloc - allocate virtually contiguous memory
435 * @size: allocation size
436 * @gfp_mask: flags for the page level allocator
437 * @prot: protection mask for the allocated pages
439 * Allocate enough pages to cover @size from the page level
440 * allocator with @gfp_mask flags. Map them into contiguous
441 * kernel virtual space, using a pagetable protection of @prot.
443 void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask, pgprot_t prot)
445 struct vm_struct *area;
447 size = PAGE_ALIGN(size);
448 if (!size || (size >> PAGE_SHIFT) > num_physpages)
451 area = get_vm_area(size, VM_ALLOC);
455 return __vmalloc_area(area, gfp_mask, prot);
458 EXPORT_SYMBOL(__vmalloc);
461 * vmalloc - allocate virtually contiguous memory
463 * @size: allocation size
465 * Allocate enough pages to cover @size from the page level
466 * allocator and map them into contiguous kernel virtual space.
468 * For tight cotrol over page level allocator and protection flags
469 * use __vmalloc() instead.
471 void *vmalloc(unsigned long size)
473 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
476 EXPORT_SYMBOL(vmalloc);
479 * vmalloc_exec - allocate virtually contiguous, executable memory
481 * @size: allocation size
483 * Kernel-internal function to allocate enough pages to cover @size
484 * the page level allocator and map them into contiguous and
485 * executable kernel virtual space.
487 * For tight cotrol over page level allocator and protection flags
488 * use __vmalloc() instead.
491 #ifndef PAGE_KERNEL_EXEC
492 # define PAGE_KERNEL_EXEC PAGE_KERNEL
495 void *vmalloc_exec(unsigned long size)
497 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
501 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
503 * @size: allocation size
505 * Allocate enough 32bit PA addressable pages to cover @size from the
506 * page level allocator and map them into contiguous kernel virtual space.
508 void *vmalloc_32(unsigned long size)
510 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
513 EXPORT_SYMBOL(vmalloc_32);
515 long vread(char *buf, char *addr, unsigned long count)
517 struct vm_struct *tmp;
518 char *vaddr, *buf_start = buf;
521 /* Don't allow overflow */
522 if ((unsigned long) addr + count < count)
523 count = -(unsigned long) addr;
525 read_lock(&vmlist_lock);
526 for (tmp = vmlist; tmp; tmp = tmp->next) {
527 vaddr = (char *) tmp->addr;
528 if (addr >= vaddr + tmp->size - PAGE_SIZE)
530 while (addr < vaddr) {
538 n = vaddr + tmp->size - PAGE_SIZE - addr;
549 read_unlock(&vmlist_lock);
550 return buf - buf_start;
553 long vwrite(char *buf, char *addr, unsigned long count)
555 struct vm_struct *tmp;
556 char *vaddr, *buf_start = buf;
559 /* Don't allow overflow */
560 if ((unsigned long) addr + count < count)
561 count = -(unsigned long) addr;
563 read_lock(&vmlist_lock);
564 for (tmp = vmlist; tmp; tmp = tmp->next) {
565 vaddr = (char *) tmp->addr;
566 if (addr >= vaddr + tmp->size - PAGE_SIZE)
568 while (addr < vaddr) {
575 n = vaddr + tmp->size - PAGE_SIZE - addr;
586 read_unlock(&vmlist_lock);
587 return buf - buf_start;