mm/vmalloc.c

   1 /*
   2  *  linux/mm/vmalloc.c
   3  *
   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
   9  */
  10
  11 #include <linux/mm.h>
  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>
  17
  18 #include <linux/vmalloc.h>
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22
  23
  24 DEFINE_RWLOCK(vmlist_lock);
  25 struct vm_struct *vmlist;
  26
  27 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  28                             int node);
  29
  30 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  31 {
  32         pte_t *pte;
  33
  34         pte = pte_offset_kernel(pmd, addr);
  35         do {
  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);
  39 }
  40
  41 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  42                                                 unsigned long end)
  43 {
  44         pmd_t *pmd;
  45         unsigned long next;
  46
  47         pmd = pmd_offset(pud, addr);
  48         do {
  49                 next = pmd_addr_end(addr, end);
  50                 if (pmd_none_or_clear_bad(pmd))
  51                         continue;
  52                 vunmap_pte_range(pmd, addr, next);
  53         } while (pmd++, addr = next, addr != end);
  54 }
  55
  56 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  57                                                 unsigned long end)
  58 {
  59         pud_t *pud;
  60         unsigned long next;
  61
  62         pud = pud_offset(pgd, addr);
  63         do {
  64                 next = pud_addr_end(addr, end);
  65                 if (pud_none_or_clear_bad(pud))
  66                         continue;
  67                 vunmap_pmd_range(pud, addr, next);
  68         } while (pud++, addr = next, addr != end);
  69 }
  70
  71 void unmap_kernel_range(unsigned long addr, unsigned long size)
  72 {
  73         pgd_t *pgd;
  74         unsigned long next;
  75         unsigned long start = addr;
  76         unsigned long end = addr + size;
  77
  78         BUG_ON(addr >= end);
  79         pgd = pgd_offset_k(addr);
  80         flush_cache_vunmap(addr, end);
  81         do {
  82                 next = pgd_addr_end(addr, end);
  83                 if (pgd_none_or_clear_bad(pgd))
  84                         continue;
  85                 vunmap_pud_range(pgd, addr, next);
  86         } while (pgd++, addr = next, addr != end);
  87         flush_tlb_kernel_range(start, end);
  88 }
  89
  90 static void unmap_vm_area(struct vm_struct *area)
  91 {
  92         unmap_kernel_range((unsigned long)area->addr, area->size);
  93 }
  94
  95 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  96                         unsigned long end, pgprot_t prot, struct page ***pages)
  97 {
  98         pte_t *pte;
  99
 100         pte = pte_alloc_kernel(pmd, addr);
 101         if (!pte)
 102                 return -ENOMEM;
 103         do {
 104                 struct page *page = **pages;
 105                 WARN_ON(!pte_none(*pte));
 106                 if (!page)
 107                         return -ENOMEM;
 108                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 109                 (*pages)++;
 110         } while (pte++, addr += PAGE_SIZE, addr != end);
 111         return 0;
 112 }
 113
 114 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 115                         unsigned long end, pgprot_t prot, struct page ***pages)
 116 {
 117         pmd_t *pmd;
 118         unsigned long next;
 119
 120         pmd = pmd_alloc(&init_mm, pud, addr);
 121         if (!pmd)
 122                 return -ENOMEM;
 123         do {
 124                 next = pmd_addr_end(addr, end);
 125                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 126                         return -ENOMEM;
 127         } while (pmd++, addr = next, addr != end);
 128         return 0;
 129 }
 130
 131 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 132                         unsigned long end, pgprot_t prot, struct page ***pages)
 133 {
 134         pud_t *pud;
 135         unsigned long next;
 136
 137         pud = pud_alloc(&init_mm, pgd, addr);
 138         if (!pud)
 139                 return -ENOMEM;
 140         do {
 141                 next = pud_addr_end(addr, end);
 142                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 143                         return -ENOMEM;
 144         } while (pud++, addr = next, addr != end);
 145         return 0;
 146 }
 147
 148 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 149 {
 150         pgd_t *pgd;
 151         unsigned long next;
 152         unsigned long addr = (unsigned long) area->addr;
 153         unsigned long end = addr + area->size - PAGE_SIZE;
 154         int err;
 155
 156         BUG_ON(addr >= end);
 157         pgd = pgd_offset_k(addr);
 158         do {
 159                 next = pgd_addr_end(addr, end);
 160                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 161                 if (err)
 162                         break;
 163         } while (pgd++, addr = next, addr != end);
 164         flush_cache_vmap((unsigned long) area->addr, end);
 165         return err;
 166 }
 167 EXPORT_SYMBOL_GPL(map_vm_area);
 168
 169 /*
 170  * Map a vmalloc()-space virtual address to the physical page.
 171  */
 172 struct page *vmalloc_to_page(const void *vmalloc_addr)
 173 {
 174         unsigned long addr = (unsigned long) vmalloc_addr;
 175         struct page *page = NULL;
 176         pgd_t *pgd = pgd_offset_k(addr);
 177         pud_t *pud;
 178         pmd_t *pmd;
 179         pte_t *ptep, pte;
 180
 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);
 187                                 pte = *ptep;
 188                                 if (pte_present(pte))
 189                                         page = pte_page(pte);
 190                                 pte_unmap(ptep);
 191                         }
 192                 }
 193         }
 194         return page;
 195 }
 196 EXPORT_SYMBOL(vmalloc_to_page);
 197
 198 /*
 199  * Map a vmalloc()-space virtual address to the physical page frame number.
 200  */
 201 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
 202 {
 203         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
 204 }
 205 EXPORT_SYMBOL(vmalloc_to_pfn);
 206
 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)
 210 {
 211         struct vm_struct **p, *tmp, *area;
 212         unsigned long align = 1;
 213         unsigned long addr;
 214
 215         BUG_ON(in_interrupt());
 216         if (flags & VM_IOREMAP) {
 217                 int bit = fls(size);
 218
 219                 if (bit > IOREMAP_MAX_ORDER)
 220                         bit = IOREMAP_MAX_ORDER;
 221                 else if (bit < PAGE_SHIFT)
 222                         bit = PAGE_SHIFT;
 223
 224                 align = 1ul << bit;
 225         }
 226         addr = ALIGN(start, align);
 227         size = PAGE_ALIGN(size);
 228         if (unlikely(!size))
 229                 return NULL;
 230
 231         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 232
 233         if (unlikely(!area))
 234                 return NULL;
 235
 236         /*
 237          * We always allocate a guard page.
 238          */
 239         size += PAGE_SIZE;
 240
 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);
 247                         continue;
 248                 }
 249                 if ((size + addr) < addr)
 250                         goto out;
 251                 if (size + addr <= (unsigned long)tmp->addr)
 252                         goto found;
 253                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 254                 if (addr > end - size)
 255                         goto out;
 256         }
 257
 258 found:
 259         area->next = *p;
 260         *p = area;
 261
 262         area->flags = flags;
 263         area->addr = (void *)addr;
 264         area->size = size;
 265         area->pages = NULL;
 266         area->nr_pages = 0;
 267         area->phys_addr = 0;
 268         write_unlock(&vmlist_lock);
 269
 270         return area;
 271
 272 out:
 273         write_unlock(&vmlist_lock);
 274         kfree(area);
 275         if (printk_ratelimit())
 276                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 277         return NULL;
 278 }
 279
 280 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 281                                 unsigned long start, unsigned long end)
 282 {
 283         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
 284 }
 285 EXPORT_SYMBOL_GPL(__get_vm_area);
 286
 287 /**
 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
 291  *
 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.
 295  */
 296 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 297 {
 298         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 299 }
 300
 301 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 302                                    int node, gfp_t gfp_mask)
 303 {
 304         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 305                                   gfp_mask);
 306 }
 307
 308 /* Caller must hold vmlist_lock */
 309 static struct vm_struct *__find_vm_area(const void *addr)
 310 {
 311         struct vm_struct *tmp;
 312
 313         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 314                  if (tmp->addr == addr)
 315                         break;
 316         }
 317
 318         return tmp;
 319 }
 320
 321 /* Caller must hold vmlist_lock */
 322 static struct vm_struct *__remove_vm_area(const void *addr)
 323 {
 324         struct vm_struct **p, *tmp;
 325
 326         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 327                  if (tmp->addr == addr)
 328                          goto found;
 329         }
 330         return NULL;
 331
 332 found:
 333         unmap_vm_area(tmp);
 334         *p = tmp->next;
 335
 336         /*
 337          * Remove the guard page.
 338          */
 339         tmp->size -= PAGE_SIZE;
 340         return tmp;
 341 }
 342
 343 /**
 344  *      remove_vm_area  -  find and remove a continuous kernel virtual area
 345  *      @addr:          base address
 346  *
 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.
 350  */
 351 struct vm_struct *remove_vm_area(const void *addr)
 352 {
 353         struct vm_struct *v;
 354         write_lock(&vmlist_lock);
 355         v = __remove_vm_area(addr);
 356         write_unlock(&vmlist_lock);
 357         return v;
 358 }
 359
 360 static void __vunmap(const void *addr, int deallocate_pages)
 361 {
 362         struct vm_struct *area;
 363
 364         if (!addr)
 365                 return;
 366
 367         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 368                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 369                 WARN_ON(1);
 370                 return;
 371         }
 372
 373         area = remove_vm_area(addr);
 374         if (unlikely(!area)) {
 375                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 376                                 addr);
 377                 WARN_ON(1);
 378                 return;
 379         }
 380
 381         debug_check_no_locks_freed(addr, area->size);
 382
 383         if (deallocate_pages) {
 384                 int i;
 385
 386                 for (i = 0; i < area->nr_pages; i++) {
 387                         struct page *page = area->pages[i];
 388
 389                         BUG_ON(!page);
 390                         __free_page(page);
 391                 }
 392
 393                 if (area->flags & VM_VPAGES)
 394                         vfree(area->pages);
 395                 else
 396                         kfree(area->pages);
 397         }
 398
 399         kfree(area);
 400         return;
 401 }
 402
 403 /**
 404  *      vfree  -  release memory allocated by vmalloc()
 405  *      @addr:          memory base address
 406  *
 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.
 410  *
 411  *      Must not be called in interrupt context.
 412  */
 413 void vfree(const void *addr)
 414 {
 415         BUG_ON(in_interrupt());
 416         __vunmap(addr, 1);
 417 }
 418 EXPORT_SYMBOL(vfree);
 419
 420 /**
 421  *      vunmap  -  release virtual mapping obtained by vmap()
 422  *      @addr:          memory base address
 423  *
 424  *      Free the virtually contiguous memory area starting at @addr,
 425  *      which was created from the page array passed to vmap().
 426  *
 427  *      Must not be called in interrupt context.
 428  */
 429 void vunmap(const void *addr)
 430 {
 431         BUG_ON(in_interrupt());
 432         __vunmap(addr, 0);
 433 }
 434 EXPORT_SYMBOL(vunmap);
 435
 436 /**
 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
 442  *
 443  *      Maps @count pages from @pages into contiguous kernel virtual
 444  *      space.
 445  */
 446 void *vmap(struct page **pages, unsigned int count,
 447                 unsigned long flags, pgprot_t prot)
 448 {
 449         struct vm_struct *area;
 450
 451         if (count > num_physpages)
 452                 return NULL;
 453
 454         area = get_vm_area((count << PAGE_SHIFT), flags);
 455         if (!area)
 456                 return NULL;
 457         if (map_vm_area(area, prot, &pages)) {
 458                 vunmap(area->addr);
 459                 return NULL;
 460         }
 461
 462         return area->addr;
 463 }
 464 EXPORT_SYMBOL(vmap);
 465
 466 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 467                                 pgprot_t prot, int node)
 468 {
 469         struct page **pages;
 470         unsigned int nr_pages, array_size, i;
 471
 472         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 473         array_size = (nr_pages * sizeof(struct page *));
 474
 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,
 479                                         PAGE_KERNEL, node);
 480                 area->flags |= VM_VPAGES;
 481         } else {
 482                 pages = kmalloc_node(array_size,
 483                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
 484                                 node);
 485         }
 486         area->pages = pages;
 487         if (!area->pages) {
 488                 remove_vm_area(area->addr);
 489                 kfree(area);
 490                 return NULL;
 491         }
 492
 493         for (i = 0; i < area->nr_pages; i++) {
 494                 struct page *page;
 495
 496                 if (node < 0)
 497                         page = alloc_page(gfp_mask);
 498                 else
 499                         page = alloc_pages_node(node, gfp_mask, 0);
 500
 501                 if (unlikely(!page)) {
 502                         /* Successfully allocated i pages, free them in __vunmap() */
 503                         area->nr_pages = i;
 504                         goto fail;
 505                 }
 506                 area->pages[i] = page;
 507         }
 508
 509         if (map_vm_area(area, prot, &pages))
 510                 goto fail;
 511         return area->addr;
 512
 513 fail:
 514         vfree(area->addr);
 515         return NULL;
 516 }
 517
 518 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 519 {
 520         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 521 }
 522
 523 /**
 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
 529  *
 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.
 533  */
 534 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 535                             int node)
 536 {
 537         struct vm_struct *area;
 538
 539         size = PAGE_ALIGN(size);
 540         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 541                 return NULL;
 542
 543         area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
 544         if (!area)
 545                 return NULL;
 546
 547         return __vmalloc_area_node(area, gfp_mask, prot, node);
 548 }
 549
 550 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 551 {
 552         return __vmalloc_node(size, gfp_mask, prot, -1);
 553 }
 554 EXPORT_SYMBOL(__vmalloc);
 555
 556 /**
 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.
 561  *
 562  *      For tight control over page level allocator and protection flags
 563  *      use __vmalloc() instead.
 564  */
 565 void *vmalloc(unsigned long size)
 566 {
 567         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 568 }
 569 EXPORT_SYMBOL(vmalloc);
 570
 571 /**
 572  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 573  * @size: allocation size
 574  *
 575  * The resulting memory area is zeroed so it can be mapped to userspace
 576  * without leaking data.
 577  */
 578 void *vmalloc_user(unsigned long size)
 579 {
 580         struct vm_struct *area;
 581         void *ret;
 582
 583         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 584         if (ret) {
 585                 write_lock(&vmlist_lock);
 586                 area = __find_vm_area(ret);
 587                 area->flags |= VM_USERMAP;
 588                 write_unlock(&vmlist_lock);
 589         }
 590         return ret;
 591 }
 592 EXPORT_SYMBOL(vmalloc_user);
 593
 594 /**
 595  *      vmalloc_node  -  allocate memory on a specific node
 596  *      @size:          allocation size
 597  *      @node:          numa node
 598  *
 599  *      Allocate enough pages to cover @size from the page level
 600  *      allocator and map them into contiguous kernel virtual space.
 601  *
 602  *      For tight control over page level allocator and protection flags
 603  *      use __vmalloc() instead.
 604  */
 605 void *vmalloc_node(unsigned long size, int node)
 606 {
 607         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 608 }
 609 EXPORT_SYMBOL(vmalloc_node);
 610
 611 #ifndef PAGE_KERNEL_EXEC
 612 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 613 #endif
 614
 615 /**
 616  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 617  *      @size:          allocation size
 618  *
 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.
 622  *
 623  *      For tight control over page level allocator and protection flags
 624  *      use __vmalloc() instead.
 625  */
 626
 627 void *vmalloc_exec(unsigned long size)
 628 {
 629         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 630 }
 631
 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
 636 #else
 637 #define GFP_VMALLOC32 GFP_KERNEL
 638 #endif
 639
 640 /**
 641  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 642  *      @size:          allocation size
 643  *
 644  *      Allocate enough 32bit PA addressable pages to cover @size from the
 645  *      page level allocator and map them into contiguous kernel virtual space.
 646  */
 647 void *vmalloc_32(unsigned long size)
 648 {
 649         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 650 }
 651 EXPORT_SYMBOL(vmalloc_32);
 652
 653 /**
 654  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 655  *      @size:          allocation size
 656  *
 657  * The resulting memory area is 32bit addressable and zeroed so it can be
 658  * mapped to userspace without leaking data.
 659  */
 660 void *vmalloc_32_user(unsigned long size)
 661 {
 662         struct vm_struct *area;
 663         void *ret;
 664
 665         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 666         if (ret) {
 667                 write_lock(&vmlist_lock);
 668                 area = __find_vm_area(ret);
 669                 area->flags |= VM_USERMAP;
 670                 write_unlock(&vmlist_lock);
 671         }
 672         return ret;
 673 }
 674 EXPORT_SYMBOL(vmalloc_32_user);
 675
 676 long vread(char *buf, char *addr, unsigned long count)
 677 {
 678         struct vm_struct *tmp;
 679         char *vaddr, *buf_start = buf;
 680         unsigned long n;
 681
 682         /* Don't allow overflow */
 683         if ((unsigned long) addr + count < count)
 684                 count = -(unsigned long) addr;
 685
 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)
 690                         continue;
 691                 while (addr < vaddr) {
 692                         if (count == 0)
 693                                 goto finished;
 694                         *buf = '\0';
 695                         buf++;
 696                         addr++;
 697                         count--;
 698                 }
 699                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 700                 do {
 701                         if (count == 0)
 702                                 goto finished;
 703                         *buf = *addr;
 704                         buf++;
 705                         addr++;
 706                         count--;
 707                 } while (--n > 0);
 708         }
 709 finished:
 710         read_unlock(&vmlist_lock);
 711         return buf - buf_start;
 712 }
 713
 714 long vwrite(char *buf, char *addr, unsigned long count)
 715 {
 716         struct vm_struct *tmp;
 717         char *vaddr, *buf_start = buf;
 718         unsigned long n;
 719
 720         /* Don't allow overflow */
 721         if ((unsigned long) addr + count < count)
 722                 count = -(unsigned long) addr;
 723
 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)
 728                         continue;
 729                 while (addr < vaddr) {
 730                         if (count == 0)
 731                                 goto finished;
 732                         buf++;
 733                         addr++;
 734                         count--;
 735                 }
 736                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 737                 do {
 738                         if (count == 0)
 739                                 goto finished;
 740                         *addr = *buf;
 741                         buf++;
 742                         addr++;
 743                         count--;
 744                 } while (--n > 0);
 745         }
 746 finished:
 747         read_unlock(&vmlist_lock);
 748         return buf - buf_start;
 749 }
 750
 751 /**
 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
 757  *
 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.
 761  *
 762  *      Similar to remap_pfn_range() (see mm/memory.c)
 763  */
 764 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 765                                                 unsigned long pgoff)
 766 {
 767         struct vm_struct *area;
 768         unsigned long uaddr = vma->vm_start;
 769         unsigned long usize = vma->vm_end - vma->vm_start;
 770         int ret;
 771
 772         if ((PAGE_SIZE-1) & (unsigned long)addr)
 773                 return -EINVAL;
 774
 775         read_lock(&vmlist_lock);
 776         area = __find_vm_area(addr);
 777         if (!area)
 778                 goto out_einval_locked;
 779
 780         if (!(area->flags & VM_USERMAP))
 781                 goto out_einval_locked;
 782
 783         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 784                 goto out_einval_locked;
 785         read_unlock(&vmlist_lock);
 786
 787         addr += pgoff << PAGE_SHIFT;
 788         do {
 789                 struct page *page = vmalloc_to_page(addr);
 790                 ret = vm_insert_page(vma, uaddr, page);
 791                 if (ret)
 792                         return ret;
 793
 794                 uaddr += PAGE_SIZE;
 795                 addr += PAGE_SIZE;
 796                 usize -= PAGE_SIZE;
 797         } while (usize > 0);
 798
 799         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 800         vma->vm_flags |= VM_RESERVED;
 801
 802         return ret;
 803
 804 out_einval_locked:
 805         read_unlock(&vmlist_lock);
 806         return -EINVAL;
 807 }
 808 EXPORT_SYMBOL(remap_vmalloc_range);
 809
 810 /*
 811  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 812  * have one.
 813  */
 814 void  __attribute__((weak)) vmalloc_sync_all(void)
 815 {
 816 }
 817
 818
 819 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
 820 {
 821         /* apply_to_page_range() does all the hard work. */
 822         return 0;
 823 }
 824
 825 /**
 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
 829  *
 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
 834  *      processes.
 835  */
 836 struct vm_struct *alloc_vm_area(size_t size)
 837 {
 838         struct vm_struct *area;
 839
 840         area = get_vm_area(size, VM_IOREMAP);
 841         if (area == NULL)
 842                 return NULL;
 843
 844         /*
 845          * This ensures that page tables are constructed for this region
 846          * of kernel virtual address space and mapped into init_mm.
 847          */
 848         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 849                                 area->size, f, NULL)) {
 850                 free_vm_area(area);
 851                 return NULL;
 852         }
 853
 854         /* Make sure the pagetables are constructed in process kernel
 855            mappings */
 856         vmalloc_sync_all();
 857
 858         return area;
 859 }
 860 EXPORT_SYMBOL_GPL(alloc_vm_area);
 861
 862 void free_vm_area(struct vm_struct *area)
 863 {
 864         struct vm_struct *ret;
 865         ret = remove_vm_area(area->addr);
 866         BUG_ON(ret != area);
 867         kfree(area);
 868 }
 869 EXPORT_SYMBOL_GPL(free_vm_area);