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 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)
 172 {
 173         struct vm_struct **p, *tmp, *area;
 174         unsigned long align = 1;
 175         unsigned long addr;
 176
 177         BUG_ON(in_interrupt());
 178         if (flags & VM_IOREMAP) {
 179                 int bit = fls(size);
 180
 181                 if (bit > IOREMAP_MAX_ORDER)
 182                         bit = IOREMAP_MAX_ORDER;
 183                 else if (bit < PAGE_SHIFT)
 184                         bit = PAGE_SHIFT;
 185
 186                 align = 1ul << bit;
 187         }
 188         addr = ALIGN(start, align);
 189         size = PAGE_ALIGN(size);
 190         if (unlikely(!size))
 191                 return NULL;
 192
 193         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 194
 195         if (unlikely(!area))
 196                 return NULL;
 197
 198         /*
 199          * We always allocate a guard page.
 200          */
 201         size += PAGE_SIZE;
 202
 203         write_lock(&vmlist_lock);
 204         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 205                 if ((unsigned long)tmp->addr < addr) {
 206                         if((unsigned long)tmp->addr + tmp->size >= addr)
 207                                 addr = ALIGN(tmp->size +
 208                                              (unsigned long)tmp->addr, align);
 209                         continue;
 210                 }
 211                 if ((size + addr) < addr)
 212                         goto out;
 213                 if (size + addr <= (unsigned long)tmp->addr)
 214                         goto found;
 215                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 216                 if (addr > end - size)
 217                         goto out;
 218         }
 219
 220 found:
 221         area->next = *p;
 222         *p = area;
 223
 224         area->flags = flags;
 225         area->addr = (void *)addr;
 226         area->size = size;
 227         area->pages = NULL;
 228         area->nr_pages = 0;
 229         area->phys_addr = 0;
 230         write_unlock(&vmlist_lock);
 231
 232         return area;
 233
 234 out:
 235         write_unlock(&vmlist_lock);
 236         kfree(area);
 237         if (printk_ratelimit())
 238                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 239         return NULL;
 240 }
 241
 242 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 243                                 unsigned long start, unsigned long end)
 244 {
 245         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
 246 }
 247 EXPORT_SYMBOL_GPL(__get_vm_area);
 248
 249 /**
 250  *      get_vm_area  -  reserve a contiguous kernel virtual area
 251  *      @size:          size of the area
 252  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 253  *
 254  *      Search an area of @size in the kernel virtual mapping area,
 255  *      and reserved it for out purposes.  Returns the area descriptor
 256  *      on success or %NULL on failure.
 257  */
 258 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 259 {
 260         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 261 }
 262
 263 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 264                                    int node, gfp_t gfp_mask)
 265 {
 266         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 267                                   gfp_mask);
 268 }
 269
 270 /* Caller must hold vmlist_lock */
 271 static struct vm_struct *__find_vm_area(void *addr)
 272 {
 273         struct vm_struct *tmp;
 274
 275         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 276                  if (tmp->addr == addr)
 277                         break;
 278         }
 279
 280         return tmp;
 281 }
 282
 283 /* Caller must hold vmlist_lock */
 284 static struct vm_struct *__remove_vm_area(void *addr)
 285 {
 286         struct vm_struct **p, *tmp;
 287
 288         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 289                  if (tmp->addr == addr)
 290                          goto found;
 291         }
 292         return NULL;
 293
 294 found:
 295         unmap_vm_area(tmp);
 296         *p = tmp->next;
 297
 298         /*
 299          * Remove the guard page.
 300          */
 301         tmp->size -= PAGE_SIZE;
 302         return tmp;
 303 }
 304
 305 /**
 306  *      remove_vm_area  -  find and remove a continuous kernel virtual area
 307  *      @addr:          base address
 308  *
 309  *      Search for the kernel VM area starting at @addr, and remove it.
 310  *      This function returns the found VM area, but using it is NOT safe
 311  *      on SMP machines, except for its size or flags.
 312  */
 313 struct vm_struct *remove_vm_area(void *addr)
 314 {
 315         struct vm_struct *v;
 316         write_lock(&vmlist_lock);
 317         v = __remove_vm_area(addr);
 318         write_unlock(&vmlist_lock);
 319         return v;
 320 }
 321
 322 static void __vunmap(void *addr, int deallocate_pages)
 323 {
 324         struct vm_struct *area;
 325
 326         if (!addr)
 327                 return;
 328
 329         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 330                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 331                 WARN_ON(1);
 332                 return;
 333         }
 334
 335         area = remove_vm_area(addr);
 336         if (unlikely(!area)) {
 337                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 338                                 addr);
 339                 WARN_ON(1);
 340                 return;
 341         }
 342
 343         debug_check_no_locks_freed(addr, area->size);
 344
 345         if (deallocate_pages) {
 346                 int i;
 347
 348                 for (i = 0; i < area->nr_pages; i++) {
 349                         BUG_ON(!area->pages[i]);
 350                         __free_page(area->pages[i]);
 351                 }
 352
 353                 if (area->flags & VM_VPAGES)
 354                         vfree(area->pages);
 355                 else
 356                         kfree(area->pages);
 357         }
 358
 359         kfree(area);
 360         return;
 361 }
 362
 363 /**
 364  *      vfree  -  release memory allocated by vmalloc()
 365  *      @addr:          memory base address
 366  *
 367  *      Free the virtually continuous memory area starting at @addr, as
 368  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 369  *      NULL, no operation is performed.
 370  *
 371  *      Must not be called in interrupt context.
 372  */
 373 void vfree(void *addr)
 374 {
 375         BUG_ON(in_interrupt());
 376         __vunmap(addr, 1);
 377 }
 378 EXPORT_SYMBOL(vfree);
 379
 380 /**
 381  *      vunmap  -  release virtual mapping obtained by vmap()
 382  *      @addr:          memory base address
 383  *
 384  *      Free the virtually contiguous memory area starting at @addr,
 385  *      which was created from the page array passed to vmap().
 386  *
 387  *      Must not be called in interrupt context.
 388  */
 389 void vunmap(void *addr)
 390 {
 391         BUG_ON(in_interrupt());
 392         __vunmap(addr, 0);
 393 }
 394 EXPORT_SYMBOL(vunmap);
 395
 396 /**
 397  *      vmap  -  map an array of pages into virtually contiguous space
 398  *      @pages:         array of page pointers
 399  *      @count:         number of pages to map
 400  *      @flags:         vm_area->flags
 401  *      @prot:          page protection for the mapping
 402  *
 403  *      Maps @count pages from @pages into contiguous kernel virtual
 404  *      space.
 405  */
 406 void *vmap(struct page **pages, unsigned int count,
 407                 unsigned long flags, pgprot_t prot)
 408 {
 409         struct vm_struct *area;
 410
 411         if (count > num_physpages)
 412                 return NULL;
 413
 414         area = get_vm_area((count << PAGE_SHIFT), flags);
 415         if (!area)
 416                 return NULL;
 417         if (map_vm_area(area, prot, &pages)) {
 418                 vunmap(area->addr);
 419                 return NULL;
 420         }
 421
 422         return area->addr;
 423 }
 424 EXPORT_SYMBOL(vmap);
 425
 426 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 427                                 pgprot_t prot, int node)
 428 {
 429         struct page **pages;
 430         unsigned int nr_pages, array_size, i;
 431
 432         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 433         array_size = (nr_pages * sizeof(struct page *));
 434
 435         area->nr_pages = nr_pages;
 436         /* Please note that the recursion is strictly bounded. */
 437         if (array_size > PAGE_SIZE) {
 438                 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
 439                                         PAGE_KERNEL, node);
 440                 area->flags |= VM_VPAGES;
 441         } else {
 442                 pages = kmalloc_node(array_size,
 443                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
 444                                 node);
 445         }
 446         area->pages = pages;
 447         if (!area->pages) {
 448                 remove_vm_area(area->addr);
 449                 kfree(area);
 450                 return NULL;
 451         }
 452
 453         for (i = 0; i < area->nr_pages; i++) {
 454                 if (node < 0)
 455                         area->pages[i] = alloc_page(gfp_mask);
 456                 else
 457                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 458                 if (unlikely(!area->pages[i])) {
 459                         /* Successfully allocated i pages, free them in __vunmap() */
 460                         area->nr_pages = i;
 461                         goto fail;
 462                 }
 463         }
 464
 465         if (map_vm_area(area, prot, &pages))
 466                 goto fail;
 467         return area->addr;
 468
 469 fail:
 470         vfree(area->addr);
 471         return NULL;
 472 }
 473
 474 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 475 {
 476         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 477 }
 478
 479 /**
 480  *      __vmalloc_node  -  allocate virtually contiguous memory
 481  *      @size:          allocation size
 482  *      @gfp_mask:      flags for the page level allocator
 483  *      @prot:          protection mask for the allocated pages
 484  *      @node:          node to use for allocation or -1
 485  *
 486  *      Allocate enough pages to cover @size from the page level
 487  *      allocator with @gfp_mask flags.  Map them into contiguous
 488  *      kernel virtual space, using a pagetable protection of @prot.
 489  */
 490 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 491                             int node)
 492 {
 493         struct vm_struct *area;
 494
 495         size = PAGE_ALIGN(size);
 496         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 497                 return NULL;
 498
 499         area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
 500         if (!area)
 501                 return NULL;
 502
 503         return __vmalloc_area_node(area, gfp_mask, prot, node);
 504 }
 505
 506 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 507 {
 508         return __vmalloc_node(size, gfp_mask, prot, -1);
 509 }
 510 EXPORT_SYMBOL(__vmalloc);
 511
 512 /**
 513  *      vmalloc  -  allocate virtually contiguous memory
 514  *      @size:          allocation size
 515  *      Allocate enough pages to cover @size from the page level
 516  *      allocator and map them into contiguous kernel virtual space.
 517  *
 518  *      For tight control over page level allocator and protection flags
 519  *      use __vmalloc() instead.
 520  */
 521 void *vmalloc(unsigned long size)
 522 {
 523         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 524 }
 525 EXPORT_SYMBOL(vmalloc);
 526
 527 /**
 528  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 529  * @size: allocation size
 530  *
 531  * The resulting memory area is zeroed so it can be mapped to userspace
 532  * without leaking data.
 533  */
 534 void *vmalloc_user(unsigned long size)
 535 {
 536         struct vm_struct *area;
 537         void *ret;
 538
 539         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 540         if (ret) {
 541                 write_lock(&vmlist_lock);
 542                 area = __find_vm_area(ret);
 543                 area->flags |= VM_USERMAP;
 544                 write_unlock(&vmlist_lock);
 545         }
 546         return ret;
 547 }
 548 EXPORT_SYMBOL(vmalloc_user);
 549
 550 /**
 551  *      vmalloc_node  -  allocate memory on a specific node
 552  *      @size:          allocation size
 553  *      @node:          numa node
 554  *
 555  *      Allocate enough pages to cover @size from the page level
 556  *      allocator and map them into contiguous kernel virtual space.
 557  *
 558  *      For tight control over page level allocator and protection flags
 559  *      use __vmalloc() instead.
 560  */
 561 void *vmalloc_node(unsigned long size, int node)
 562 {
 563         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 564 }
 565 EXPORT_SYMBOL(vmalloc_node);
 566
 567 #ifndef PAGE_KERNEL_EXEC
 568 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 569 #endif
 570
 571 /**
 572  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 573  *      @size:          allocation size
 574  *
 575  *      Kernel-internal function to allocate enough pages to cover @size
 576  *      the page level allocator and map them into contiguous and
 577  *      executable kernel virtual space.
 578  *
 579  *      For tight control over page level allocator and protection flags
 580  *      use __vmalloc() instead.
 581  */
 582
 583 void *vmalloc_exec(unsigned long size)
 584 {
 585         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 586 }
 587
 588 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
 589 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
 590 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
 591 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
 592 #else
 593 #define GFP_VMALLOC32 GFP_KERNEL
 594 #endif
 595
 596 /**
 597  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 598  *      @size:          allocation size
 599  *
 600  *      Allocate enough 32bit PA addressable pages to cover @size from the
 601  *      page level allocator and map them into contiguous kernel virtual space.
 602  */
 603 void *vmalloc_32(unsigned long size)
 604 {
 605         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 606 }
 607 EXPORT_SYMBOL(vmalloc_32);
 608
 609 /**
 610  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 611  *      @size:          allocation size
 612  *
 613  * The resulting memory area is 32bit addressable and zeroed so it can be
 614  * mapped to userspace without leaking data.
 615  */
 616 void *vmalloc_32_user(unsigned long size)
 617 {
 618         struct vm_struct *area;
 619         void *ret;
 620
 621         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 622         if (ret) {
 623                 write_lock(&vmlist_lock);
 624                 area = __find_vm_area(ret);
 625                 area->flags |= VM_USERMAP;
 626                 write_unlock(&vmlist_lock);
 627         }
 628         return ret;
 629 }
 630 EXPORT_SYMBOL(vmalloc_32_user);
 631
 632 long vread(char *buf, char *addr, unsigned long count)
 633 {
 634         struct vm_struct *tmp;
 635         char *vaddr, *buf_start = buf;
 636         unsigned long n;
 637
 638         /* Don't allow overflow */
 639         if ((unsigned long) addr + count < count)
 640                 count = -(unsigned long) addr;
 641
 642         read_lock(&vmlist_lock);
 643         for (tmp = vmlist; tmp; tmp = tmp->next) {
 644                 vaddr = (char *) tmp->addr;
 645                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 646                         continue;
 647                 while (addr < vaddr) {
 648                         if (count == 0)
 649                                 goto finished;
 650                         *buf = '\0';
 651                         buf++;
 652                         addr++;
 653                         count--;
 654                 }
 655                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 656                 do {
 657                         if (count == 0)
 658                                 goto finished;
 659                         *buf = *addr;
 660                         buf++;
 661                         addr++;
 662                         count--;
 663                 } while (--n > 0);
 664         }
 665 finished:
 666         read_unlock(&vmlist_lock);
 667         return buf - buf_start;
 668 }
 669
 670 long vwrite(char *buf, char *addr, unsigned long count)
 671 {
 672         struct vm_struct *tmp;
 673         char *vaddr, *buf_start = buf;
 674         unsigned long n;
 675
 676         /* Don't allow overflow */
 677         if ((unsigned long) addr + count < count)
 678                 count = -(unsigned long) addr;
 679
 680         read_lock(&vmlist_lock);
 681         for (tmp = vmlist; tmp; tmp = tmp->next) {
 682                 vaddr = (char *) tmp->addr;
 683                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 684                         continue;
 685                 while (addr < vaddr) {
 686                         if (count == 0)
 687                                 goto finished;
 688                         buf++;
 689                         addr++;
 690                         count--;
 691                 }
 692                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 693                 do {
 694                         if (count == 0)
 695                                 goto finished;
 696                         *addr = *buf;
 697                         buf++;
 698                         addr++;
 699                         count--;
 700                 } while (--n > 0);
 701         }
 702 finished:
 703         read_unlock(&vmlist_lock);
 704         return buf - buf_start;
 705 }
 706
 707 /**
 708  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 709  *      @vma:           vma to cover (map full range of vma)
 710  *      @addr:          vmalloc memory
 711  *      @pgoff:         number of pages into addr before first page to map
 712  *      @returns:       0 for success, -Exxx on failure
 713  *
 714  *      This function checks that addr is a valid vmalloc'ed area, and
 715  *      that it is big enough to cover the vma. Will return failure if
 716  *      that criteria isn't met.
 717  *
 718  *      Similar to remap_pfn_range() (see mm/memory.c)
 719  */
 720 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 721                                                 unsigned long pgoff)
 722 {
 723         struct vm_struct *area;
 724         unsigned long uaddr = vma->vm_start;
 725         unsigned long usize = vma->vm_end - vma->vm_start;
 726         int ret;
 727
 728         if ((PAGE_SIZE-1) & (unsigned long)addr)
 729                 return -EINVAL;
 730
 731         read_lock(&vmlist_lock);
 732         area = __find_vm_area(addr);
 733         if (!area)
 734                 goto out_einval_locked;
 735
 736         if (!(area->flags & VM_USERMAP))
 737                 goto out_einval_locked;
 738
 739         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 740                 goto out_einval_locked;
 741         read_unlock(&vmlist_lock);
 742
 743         addr += pgoff << PAGE_SHIFT;
 744         do {
 745                 struct page *page = vmalloc_to_page(addr);
 746                 ret = vm_insert_page(vma, uaddr, page);
 747                 if (ret)
 748                         return ret;
 749
 750                 uaddr += PAGE_SIZE;
 751                 addr += PAGE_SIZE;
 752                 usize -= PAGE_SIZE;
 753         } while (usize > 0);
 754
 755         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 756         vma->vm_flags |= VM_RESERVED;
 757
 758         return ret;
 759
 760 out_einval_locked:
 761         read_unlock(&vmlist_lock);
 762         return -EINVAL;
 763 }
 764 EXPORT_SYMBOL(remap_vmalloc_range);
 765
 766 /*
 767  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 768  * have one.
 769  */
 770 void  __attribute__((weak)) vmalloc_sync_all(void)
 771 {
 772 }
 773
 774
 775 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
 776 {
 777         /* apply_to_page_range() does all the hard work. */
 778         return 0;
 779 }
 780
 781 /**
 782  *      alloc_vm_area - allocate a range of kernel address space
 783  *      @size:          size of the area
 784  *      @returns:       NULL on failure, vm_struct on success
 785  *
 786  *      This function reserves a range of kernel address space, and
 787  *      allocates pagetables to map that range.  No actual mappings
 788  *      are created.  If the kernel address space is not shared
 789  *      between processes, it syncs the pagetable across all
 790  *      processes.
 791  */
 792 struct vm_struct *alloc_vm_area(size_t size)
 793 {
 794         struct vm_struct *area;
 795
 796         area = get_vm_area(size, VM_IOREMAP);
 797         if (area == NULL)
 798                 return NULL;
 799
 800         /*
 801          * This ensures that page tables are constructed for this region
 802          * of kernel virtual address space and mapped into init_mm.
 803          */
 804         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 805                                 area->size, f, NULL)) {
 806                 free_vm_area(area);
 807                 return NULL;
 808         }
 809
 810         /* Make sure the pagetables are constructed in process kernel
 811            mappings */
 812         vmalloc_sync_all();
 813
 814         return area;
 815 }
 816 EXPORT_SYMBOL_GPL(alloc_vm_area);
 817
 818 void free_vm_area(struct vm_struct *area)
 819 {
 820         struct vm_struct *ret;
 821         ret = remove_vm_area(area->addr);
 822         BUG_ON(ret != area);
 823         kfree(area);
 824 }
 825 EXPORT_SYMBOL_GPL(free_vm_area);