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 #include <linux/seq_file.h>
  18 #include <linux/debugobjects.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/kallsyms.h>
  21
  22 #include <asm/uaccess.h>
  23 #include <asm/tlbflush.h>
  24
  25
  26 DEFINE_RWLOCK(vmlist_lock);
  27 struct vm_struct *vmlist;
  28
  29 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  30                             int node, void *caller);
  31
  32 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  33 {
  34         pte_t *pte;
  35
  36         pte = pte_offset_kernel(pmd, addr);
  37         do {
  38                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  39                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  40         } while (pte++, addr += PAGE_SIZE, addr != end);
  41 }
  42
  43 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  44                                                 unsigned long end)
  45 {
  46         pmd_t *pmd;
  47         unsigned long next;
  48
  49         pmd = pmd_offset(pud, addr);
  50         do {
  51                 next = pmd_addr_end(addr, end);
  52                 if (pmd_none_or_clear_bad(pmd))
  53                         continue;
  54                 vunmap_pte_range(pmd, addr, next);
  55         } while (pmd++, addr = next, addr != end);
  56 }
  57
  58 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  59                                                 unsigned long end)
  60 {
  61         pud_t *pud;
  62         unsigned long next;
  63
  64         pud = pud_offset(pgd, addr);
  65         do {
  66                 next = pud_addr_end(addr, end);
  67                 if (pud_none_or_clear_bad(pud))
  68                         continue;
  69                 vunmap_pmd_range(pud, addr, next);
  70         } while (pud++, addr = next, addr != end);
  71 }
  72
  73 void unmap_kernel_range(unsigned long addr, unsigned long size)
  74 {
  75         pgd_t *pgd;
  76         unsigned long next;
  77         unsigned long start = addr;
  78         unsigned long end = addr + size;
  79
  80         BUG_ON(addr >= end);
  81         pgd = pgd_offset_k(addr);
  82         flush_cache_vunmap(addr, end);
  83         do {
  84                 next = pgd_addr_end(addr, end);
  85                 if (pgd_none_or_clear_bad(pgd))
  86                         continue;
  87                 vunmap_pud_range(pgd, addr, next);
  88         } while (pgd++, addr = next, addr != end);
  89         flush_tlb_kernel_range(start, end);
  90 }
  91
  92 static void unmap_vm_area(struct vm_struct *area)
  93 {
  94         unmap_kernel_range((unsigned long)area->addr, area->size);
  95 }
  96
  97 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  98                         unsigned long end, pgprot_t prot, struct page ***pages)
  99 {
 100         pte_t *pte;
 101
 102         pte = pte_alloc_kernel(pmd, addr);
 103         if (!pte)
 104                 return -ENOMEM;
 105         do {
 106                 struct page *page = **pages;
 107                 WARN_ON(!pte_none(*pte));
 108                 if (!page)
 109                         return -ENOMEM;
 110                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 111                 (*pages)++;
 112         } while (pte++, addr += PAGE_SIZE, addr != end);
 113         return 0;
 114 }
 115
 116 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 117                         unsigned long end, pgprot_t prot, struct page ***pages)
 118 {
 119         pmd_t *pmd;
 120         unsigned long next;
 121
 122         pmd = pmd_alloc(&init_mm, pud, addr);
 123         if (!pmd)
 124                 return -ENOMEM;
 125         do {
 126                 next = pmd_addr_end(addr, end);
 127                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 128                         return -ENOMEM;
 129         } while (pmd++, addr = next, addr != end);
 130         return 0;
 131 }
 132
 133 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 134                         unsigned long end, pgprot_t prot, struct page ***pages)
 135 {
 136         pud_t *pud;
 137         unsigned long next;
 138
 139         pud = pud_alloc(&init_mm, pgd, addr);
 140         if (!pud)
 141                 return -ENOMEM;
 142         do {
 143                 next = pud_addr_end(addr, end);
 144                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 145                         return -ENOMEM;
 146         } while (pud++, addr = next, addr != end);
 147         return 0;
 148 }
 149
 150 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 151 {
 152         pgd_t *pgd;
 153         unsigned long next;
 154         unsigned long addr = (unsigned long) area->addr;
 155         unsigned long end = addr + area->size - PAGE_SIZE;
 156         int err;
 157
 158         BUG_ON(addr >= end);
 159         pgd = pgd_offset_k(addr);
 160         do {
 161                 next = pgd_addr_end(addr, end);
 162                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 163                 if (err)
 164                         break;
 165         } while (pgd++, addr = next, addr != end);
 166         flush_cache_vmap((unsigned long) area->addr, end);
 167         return err;
 168 }
 169 EXPORT_SYMBOL_GPL(map_vm_area);
 170
 171 /*
 172  * Map a vmalloc()-space virtual address to the physical page.
 173  */
 174 struct page *vmalloc_to_page(const void *vmalloc_addr)
 175 {
 176         unsigned long addr = (unsigned long) vmalloc_addr;
 177         struct page *page = NULL;
 178         pgd_t *pgd = pgd_offset_k(addr);
 179         pud_t *pud;
 180         pmd_t *pmd;
 181         pte_t *ptep, pte;
 182
 183         /*
 184          * XXX we might need to change this if we add VIRTUAL_BUG_ON for
 185          * architectures that do not vmalloc module space
 186          */
 187         VIRTUAL_BUG_ON(!is_vmalloc_addr(vmalloc_addr) &&
 188                         !is_module_address(addr));
 189
 190         if (!pgd_none(*pgd)) {
 191                 pud = pud_offset(pgd, addr);
 192                 if (!pud_none(*pud)) {
 193                         pmd = pmd_offset(pud, addr);
 194                         if (!pmd_none(*pmd)) {
 195                                 ptep = pte_offset_map(pmd, addr);
 196                                 pte = *ptep;
 197                                 if (pte_present(pte))
 198                                         page = pte_page(pte);
 199                                 pte_unmap(ptep);
 200                         }
 201                 }
 202         }
 203         return page;
 204 }
 205 EXPORT_SYMBOL(vmalloc_to_page);
 206
 207 /*
 208  * Map a vmalloc()-space virtual address to the physical page frame number.
 209  */
 210 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
 211 {
 212         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
 213 }
 214 EXPORT_SYMBOL(vmalloc_to_pfn);
 215
 216 static struct vm_struct *
 217 __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
 218                 unsigned long end, int node, gfp_t gfp_mask, void *caller)
 219 {
 220         struct vm_struct **p, *tmp, *area;
 221         unsigned long align = 1;
 222         unsigned long addr;
 223
 224         BUG_ON(in_interrupt());
 225         if (flags & VM_IOREMAP) {
 226                 int bit = fls(size);
 227
 228                 if (bit > IOREMAP_MAX_ORDER)
 229                         bit = IOREMAP_MAX_ORDER;
 230                 else if (bit < PAGE_SHIFT)
 231                         bit = PAGE_SHIFT;
 232
 233                 align = 1ul << bit;
 234         }
 235         addr = ALIGN(start, align);
 236         size = PAGE_ALIGN(size);
 237         if (unlikely(!size))
 238                 return NULL;
 239
 240         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 241
 242         if (unlikely(!area))
 243                 return NULL;
 244
 245         /*
 246          * We always allocate a guard page.
 247          */
 248         size += PAGE_SIZE;
 249
 250         write_lock(&vmlist_lock);
 251         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 252                 if ((unsigned long)tmp->addr < addr) {
 253                         if((unsigned long)tmp->addr + tmp->size >= addr)
 254                                 addr = ALIGN(tmp->size +
 255                                              (unsigned long)tmp->addr, align);
 256                         continue;
 257                 }
 258                 if ((size + addr) < addr)
 259                         goto out;
 260                 if (size + addr <= (unsigned long)tmp->addr)
 261                         goto found;
 262                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 263                 if (addr > end - size)
 264                         goto out;
 265         }
 266         if ((size + addr) < addr)
 267                 goto out;
 268         if (addr > end - size)
 269                 goto out;
 270
 271 found:
 272         area->next = *p;
 273         *p = area;
 274
 275         area->flags = flags;
 276         area->addr = (void *)addr;
 277         area->size = size;
 278         area->pages = NULL;
 279         area->nr_pages = 0;
 280         area->phys_addr = 0;
 281         area->caller = caller;
 282         write_unlock(&vmlist_lock);
 283
 284         return area;
 285
 286 out:
 287         write_unlock(&vmlist_lock);
 288         kfree(area);
 289         if (printk_ratelimit())
 290                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 291         return NULL;
 292 }
 293
 294 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 295                                 unsigned long start, unsigned long end)
 296 {
 297         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
 298                                                 __builtin_return_address(0));
 299 }
 300 EXPORT_SYMBOL_GPL(__get_vm_area);
 301
 302 /**
 303  *      get_vm_area  -  reserve a contiguous kernel virtual area
 304  *      @size:          size of the area
 305  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 306  *
 307  *      Search an area of @size in the kernel virtual mapping area,
 308  *      and reserved it for out purposes.  Returns the area descriptor
 309  *      on success or %NULL on failure.
 310  */
 311 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 312 {
 313         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 314                                 -1, GFP_KERNEL, __builtin_return_address(0));
 315 }
 316
 317 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
 318                                 void *caller)
 319 {
 320         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 321                                                 -1, GFP_KERNEL, caller);
 322 }
 323
 324 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 325                                    int node, gfp_t gfp_mask)
 326 {
 327         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 328                                   gfp_mask, __builtin_return_address(0));
 329 }
 330
 331 /* Caller must hold vmlist_lock */
 332 static struct vm_struct *__find_vm_area(const void *addr)
 333 {
 334         struct vm_struct *tmp;
 335
 336         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 337                  if (tmp->addr == addr)
 338                         break;
 339         }
 340
 341         return tmp;
 342 }
 343
 344 /* Caller must hold vmlist_lock */
 345 static struct vm_struct *__remove_vm_area(const void *addr)
 346 {
 347         struct vm_struct **p, *tmp;
 348
 349         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 350                  if (tmp->addr == addr)
 351                          goto found;
 352         }
 353         return NULL;
 354
 355 found:
 356         unmap_vm_area(tmp);
 357         *p = tmp->next;
 358
 359         /*
 360          * Remove the guard page.
 361          */
 362         tmp->size -= PAGE_SIZE;
 363         return tmp;
 364 }
 365
 366 /**
 367  *      remove_vm_area  -  find and remove a continuous kernel virtual area
 368  *      @addr:          base address
 369  *
 370  *      Search for the kernel VM area starting at @addr, and remove it.
 371  *      This function returns the found VM area, but using it is NOT safe
 372  *      on SMP machines, except for its size or flags.
 373  */
 374 struct vm_struct *remove_vm_area(const void *addr)
 375 {
 376         struct vm_struct *v;
 377         write_lock(&vmlist_lock);
 378         v = __remove_vm_area(addr);
 379         write_unlock(&vmlist_lock);
 380         return v;
 381 }
 382
 383 static void __vunmap(const void *addr, int deallocate_pages)
 384 {
 385         struct vm_struct *area;
 386
 387         if (!addr)
 388                 return;
 389
 390         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 391                 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 392                 return;
 393         }
 394
 395         area = remove_vm_area(addr);
 396         if (unlikely(!area)) {
 397                 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 398                                 addr);
 399                 return;
 400         }
 401
 402         debug_check_no_locks_freed(addr, area->size);
 403         debug_check_no_obj_freed(addr, area->size);
 404
 405         if (deallocate_pages) {
 406                 int i;
 407
 408                 for (i = 0; i < area->nr_pages; i++) {
 409                         struct page *page = area->pages[i];
 410
 411                         BUG_ON(!page);
 412                         __free_page(page);
 413                 }
 414
 415                 if (area->flags & VM_VPAGES)
 416                         vfree(area->pages);
 417                 else
 418                         kfree(area->pages);
 419         }
 420
 421         kfree(area);
 422         return;
 423 }
 424
 425 /**
 426  *      vfree  -  release memory allocated by vmalloc()
 427  *      @addr:          memory base address
 428  *
 429  *      Free the virtually continuous memory area starting at @addr, as
 430  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 431  *      NULL, no operation is performed.
 432  *
 433  *      Must not be called in interrupt context.
 434  */
 435 void vfree(const void *addr)
 436 {
 437         BUG_ON(in_interrupt());
 438         __vunmap(addr, 1);
 439 }
 440 EXPORT_SYMBOL(vfree);
 441
 442 /**
 443  *      vunmap  -  release virtual mapping obtained by vmap()
 444  *      @addr:          memory base address
 445  *
 446  *      Free the virtually contiguous memory area starting at @addr,
 447  *      which was created from the page array passed to vmap().
 448  *
 449  *      Must not be called in interrupt context.
 450  */
 451 void vunmap(const void *addr)
 452 {
 453         BUG_ON(in_interrupt());
 454         __vunmap(addr, 0);
 455 }
 456 EXPORT_SYMBOL(vunmap);
 457
 458 /**
 459  *      vmap  -  map an array of pages into virtually contiguous space
 460  *      @pages:         array of page pointers
 461  *      @count:         number of pages to map
 462  *      @flags:         vm_area->flags
 463  *      @prot:          page protection for the mapping
 464  *
 465  *      Maps @count pages from @pages into contiguous kernel virtual
 466  *      space.
 467  */
 468 void *vmap(struct page **pages, unsigned int count,
 469                 unsigned long flags, pgprot_t prot)
 470 {
 471         struct vm_struct *area;
 472
 473         if (count > num_physpages)
 474                 return NULL;
 475
 476         area = get_vm_area_caller((count << PAGE_SHIFT), flags,
 477                                         __builtin_return_address(0));
 478         if (!area)
 479                 return NULL;
 480
 481         if (map_vm_area(area, prot, &pages)) {
 482                 vunmap(area->addr);
 483                 return NULL;
 484         }
 485
 486         return area->addr;
 487 }
 488 EXPORT_SYMBOL(vmap);
 489
 490 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 491                                  pgprot_t prot, int node, void *caller)
 492 {
 493         struct page **pages;
 494         unsigned int nr_pages, array_size, i;
 495
 496         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 497         array_size = (nr_pages * sizeof(struct page *));
 498
 499         area->nr_pages = nr_pages;
 500         /* Please note that the recursion is strictly bounded. */
 501         if (array_size > PAGE_SIZE) {
 502                 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
 503                                 PAGE_KERNEL, node, caller);
 504                 area->flags |= VM_VPAGES;
 505         } else {
 506                 pages = kmalloc_node(array_size,
 507                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
 508                                 node);
 509         }
 510         area->pages = pages;
 511         area->caller = caller;
 512         if (!area->pages) {
 513                 remove_vm_area(area->addr);
 514                 kfree(area);
 515                 return NULL;
 516         }
 517
 518         for (i = 0; i < area->nr_pages; i++) {
 519                 struct page *page;
 520
 521                 if (node < 0)
 522                         page = alloc_page(gfp_mask);
 523                 else
 524                         page = alloc_pages_node(node, gfp_mask, 0);
 525
 526                 if (unlikely(!page)) {
 527                         /* Successfully allocated i pages, free them in __vunmap() */
 528                         area->nr_pages = i;
 529                         goto fail;
 530                 }
 531                 area->pages[i] = page;
 532         }
 533
 534         if (map_vm_area(area, prot, &pages))
 535                 goto fail;
 536         return area->addr;
 537
 538 fail:
 539         vfree(area->addr);
 540         return NULL;
 541 }
 542
 543 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 544 {
 545         return __vmalloc_area_node(area, gfp_mask, prot, -1,
 546                                         __builtin_return_address(0));
 547 }
 548
 549 /**
 550  *      __vmalloc_node  -  allocate virtually contiguous memory
 551  *      @size:          allocation size
 552  *      @gfp_mask:      flags for the page level allocator
 553  *      @prot:          protection mask for the allocated pages
 554  *      @node:          node to use for allocation or -1
 555  *      @caller:        caller's return address
 556  *
 557  *      Allocate enough pages to cover @size from the page level
 558  *      allocator with @gfp_mask flags.  Map them into contiguous
 559  *      kernel virtual space, using a pagetable protection of @prot.
 560  */
 561 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 562                                                 int node, void *caller)
 563 {
 564         struct vm_struct *area;
 565
 566         size = PAGE_ALIGN(size);
 567         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 568                 return NULL;
 569
 570         area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
 571                                                 node, gfp_mask, caller);
 572
 573         if (!area)
 574                 return NULL;
 575
 576         return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
 577 }
 578
 579 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 580 {
 581         return __vmalloc_node(size, gfp_mask, prot, -1,
 582                                 __builtin_return_address(0));
 583 }
 584 EXPORT_SYMBOL(__vmalloc);
 585
 586 /**
 587  *      vmalloc  -  allocate virtually contiguous memory
 588  *      @size:          allocation size
 589  *      Allocate enough pages to cover @size from the page level
 590  *      allocator and map them into contiguous kernel virtual space.
 591  *
 592  *      For tight control over page level allocator and protection flags
 593  *      use __vmalloc() instead.
 594  */
 595 void *vmalloc(unsigned long size)
 596 {
 597         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 598                                         -1, __builtin_return_address(0));
 599 }
 600 EXPORT_SYMBOL(vmalloc);
 601
 602 /**
 603  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 604  * @size: allocation size
 605  *
 606  * The resulting memory area is zeroed so it can be mapped to userspace
 607  * without leaking data.
 608  */
 609 void *vmalloc_user(unsigned long size)
 610 {
 611         struct vm_struct *area;
 612         void *ret;
 613
 614         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 615         if (ret) {
 616                 write_lock(&vmlist_lock);
 617                 area = __find_vm_area(ret);
 618                 area->flags |= VM_USERMAP;
 619                 write_unlock(&vmlist_lock);
 620         }
 621         return ret;
 622 }
 623 EXPORT_SYMBOL(vmalloc_user);
 624
 625 /**
 626  *      vmalloc_node  -  allocate memory on a specific node
 627  *      @size:          allocation size
 628  *      @node:          numa node
 629  *
 630  *      Allocate enough pages to cover @size from the page level
 631  *      allocator and map them into contiguous kernel virtual space.
 632  *
 633  *      For tight control over page level allocator and protection flags
 634  *      use __vmalloc() instead.
 635  */
 636 void *vmalloc_node(unsigned long size, int node)
 637 {
 638         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 639                                         node, __builtin_return_address(0));
 640 }
 641 EXPORT_SYMBOL(vmalloc_node);
 642
 643 #ifndef PAGE_KERNEL_EXEC
 644 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 645 #endif
 646
 647 /**
 648  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 649  *      @size:          allocation size
 650  *
 651  *      Kernel-internal function to allocate enough pages to cover @size
 652  *      the page level allocator and map them into contiguous and
 653  *      executable kernel virtual space.
 654  *
 655  *      For tight control over page level allocator and protection flags
 656  *      use __vmalloc() instead.
 657  */
 658
 659 void *vmalloc_exec(unsigned long size)
 660 {
 661         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 662 }
 663
 664 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
 665 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
 666 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
 667 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
 668 #else
 669 #define GFP_VMALLOC32 GFP_KERNEL
 670 #endif
 671
 672 /**
 673  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 674  *      @size:          allocation size
 675  *
 676  *      Allocate enough 32bit PA addressable pages to cover @size from the
 677  *      page level allocator and map them into contiguous kernel virtual space.
 678  */
 679 void *vmalloc_32(unsigned long size)
 680 {
 681         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 682 }
 683 EXPORT_SYMBOL(vmalloc_32);
 684
 685 /**
 686  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 687  *      @size:          allocation size
 688  *
 689  * The resulting memory area is 32bit addressable and zeroed so it can be
 690  * mapped to userspace without leaking data.
 691  */
 692 void *vmalloc_32_user(unsigned long size)
 693 {
 694         struct vm_struct *area;
 695         void *ret;
 696
 697         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 698         if (ret) {
 699                 write_lock(&vmlist_lock);
 700                 area = __find_vm_area(ret);
 701                 area->flags |= VM_USERMAP;
 702                 write_unlock(&vmlist_lock);
 703         }
 704         return ret;
 705 }
 706 EXPORT_SYMBOL(vmalloc_32_user);
 707
 708 long vread(char *buf, char *addr, unsigned long count)
 709 {
 710         struct vm_struct *tmp;
 711         char *vaddr, *buf_start = buf;
 712         unsigned long n;
 713
 714         /* Don't allow overflow */
 715         if ((unsigned long) addr + count < count)
 716                 count = -(unsigned long) addr;
 717
 718         read_lock(&vmlist_lock);
 719         for (tmp = vmlist; tmp; tmp = tmp->next) {
 720                 vaddr = (char *) tmp->addr;
 721                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 722                         continue;
 723                 while (addr < vaddr) {
 724                         if (count == 0)
 725                                 goto finished;
 726                         *buf = '\0';
 727                         buf++;
 728                         addr++;
 729                         count--;
 730                 }
 731                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 732                 do {
 733                         if (count == 0)
 734                                 goto finished;
 735                         *buf = *addr;
 736                         buf++;
 737                         addr++;
 738                         count--;
 739                 } while (--n > 0);
 740         }
 741 finished:
 742         read_unlock(&vmlist_lock);
 743         return buf - buf_start;
 744 }
 745
 746 long vwrite(char *buf, char *addr, unsigned long count)
 747 {
 748         struct vm_struct *tmp;
 749         char *vaddr, *buf_start = buf;
 750         unsigned long n;
 751
 752         /* Don't allow overflow */
 753         if ((unsigned long) addr + count < count)
 754                 count = -(unsigned long) addr;
 755
 756         read_lock(&vmlist_lock);
 757         for (tmp = vmlist; tmp; tmp = tmp->next) {
 758                 vaddr = (char *) tmp->addr;
 759                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 760                         continue;
 761                 while (addr < vaddr) {
 762                         if (count == 0)
 763                                 goto finished;
 764                         buf++;
 765                         addr++;
 766                         count--;
 767                 }
 768                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 769                 do {
 770                         if (count == 0)
 771                                 goto finished;
 772                         *addr = *buf;
 773                         buf++;
 774                         addr++;
 775                         count--;
 776                 } while (--n > 0);
 777         }
 778 finished:
 779         read_unlock(&vmlist_lock);
 780         return buf - buf_start;
 781 }
 782
 783 /**
 784  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 785  *      @vma:           vma to cover (map full range of vma)
 786  *      @addr:          vmalloc memory
 787  *      @pgoff:         number of pages into addr before first page to map
 788  *
 789  *      Returns:        0 for success, -Exxx on failure
 790  *
 791  *      This function checks that addr is a valid vmalloc'ed area, and
 792  *      that it is big enough to cover the vma. Will return failure if
 793  *      that criteria isn't met.
 794  *
 795  *      Similar to remap_pfn_range() (see mm/memory.c)
 796  */
 797 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 798                                                 unsigned long pgoff)
 799 {
 800         struct vm_struct *area;
 801         unsigned long uaddr = vma->vm_start;
 802         unsigned long usize = vma->vm_end - vma->vm_start;
 803         int ret;
 804
 805         if ((PAGE_SIZE-1) & (unsigned long)addr)
 806                 return -EINVAL;
 807
 808         read_lock(&vmlist_lock);
 809         area = __find_vm_area(addr);
 810         if (!area)
 811                 goto out_einval_locked;
 812
 813         if (!(area->flags & VM_USERMAP))
 814                 goto out_einval_locked;
 815
 816         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 817                 goto out_einval_locked;
 818         read_unlock(&vmlist_lock);
 819
 820         addr += pgoff << PAGE_SHIFT;
 821         do {
 822                 struct page *page = vmalloc_to_page(addr);
 823                 ret = vm_insert_page(vma, uaddr, page);
 824                 if (ret)
 825                         return ret;
 826
 827                 uaddr += PAGE_SIZE;
 828                 addr += PAGE_SIZE;
 829                 usize -= PAGE_SIZE;
 830         } while (usize > 0);
 831
 832         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 833         vma->vm_flags |= VM_RESERVED;
 834
 835         return ret;
 836
 837 out_einval_locked:
 838         read_unlock(&vmlist_lock);
 839         return -EINVAL;
 840 }
 841 EXPORT_SYMBOL(remap_vmalloc_range);
 842
 843 /*
 844  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 845  * have one.
 846  */
 847 void  __attribute__((weak)) vmalloc_sync_all(void)
 848 {
 849 }
 850
 851
 852 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
 853 {
 854         /* apply_to_page_range() does all the hard work. */
 855         return 0;
 856 }
 857
 858 /**
 859  *      alloc_vm_area - allocate a range of kernel address space
 860  *      @size:          size of the area
 861  *
 862  *      Returns:        NULL on failure, vm_struct on success
 863  *
 864  *      This function reserves a range of kernel address space, and
 865  *      allocates pagetables to map that range.  No actual mappings
 866  *      are created.  If the kernel address space is not shared
 867  *      between processes, it syncs the pagetable across all
 868  *      processes.
 869  */
 870 struct vm_struct *alloc_vm_area(size_t size)
 871 {
 872         struct vm_struct *area;
 873
 874         area = get_vm_area_caller(size, VM_IOREMAP,
 875                                 __builtin_return_address(0));
 876         if (area == NULL)
 877                 return NULL;
 878
 879         /*
 880          * This ensures that page tables are constructed for this region
 881          * of kernel virtual address space and mapped into init_mm.
 882          */
 883         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 884                                 area->size, f, NULL)) {
 885                 free_vm_area(area);
 886                 return NULL;
 887         }
 888
 889         /* Make sure the pagetables are constructed in process kernel
 890            mappings */
 891         vmalloc_sync_all();
 892
 893         return area;
 894 }
 895 EXPORT_SYMBOL_GPL(alloc_vm_area);
 896
 897 void free_vm_area(struct vm_struct *area)
 898 {
 899         struct vm_struct *ret;
 900         ret = remove_vm_area(area->addr);
 901         BUG_ON(ret != area);
 902         kfree(area);
 903 }
 904 EXPORT_SYMBOL_GPL(free_vm_area);
 905
 906
 907 #ifdef CONFIG_PROC_FS
 908 static void *s_start(struct seq_file *m, loff_t *pos)
 909 {
 910         loff_t n = *pos;
 911         struct vm_struct *v;
 912
 913         read_lock(&vmlist_lock);
 914         v = vmlist;
 915         while (n > 0 && v) {
 916                 n--;
 917                 v = v->next;
 918         }
 919         if (!n)
 920                 return v;
 921
 922         return NULL;
 923
 924 }
 925
 926 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
 927 {
 928         struct vm_struct *v = p;
 929
 930         ++*pos;
 931         return v->next;
 932 }
 933
 934 static void s_stop(struct seq_file *m, void *p)
 935 {
 936         read_unlock(&vmlist_lock);
 937 }
 938
 939 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
 940 {
 941         if (NUMA_BUILD) {
 942                 unsigned int nr, *counters = m->private;
 943
 944                 if (!counters)
 945                         return;
 946
 947                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
 948
 949                 for (nr = 0; nr < v->nr_pages; nr++)
 950                         counters[page_to_nid(v->pages[nr])]++;
 951
 952                 for_each_node_state(nr, N_HIGH_MEMORY)
 953                         if (counters[nr])
 954                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
 955         }
 956 }
 957
 958 static int s_show(struct seq_file *m, void *p)
 959 {
 960         struct vm_struct *v = p;
 961
 962         seq_printf(m, "0x%p-0x%p %7ld",
 963                 v->addr, v->addr + v->size, v->size);
 964
 965         if (v->caller) {
 966                 char buff[2 * KSYM_NAME_LEN];
 967
 968                 seq_putc(m, ' ');
 969                 sprint_symbol(buff, (unsigned long)v->caller);
 970                 seq_puts(m, buff);
 971         }
 972
 973         if (v->nr_pages)
 974                 seq_printf(m, " pages=%d", v->nr_pages);
 975
 976         if (v->phys_addr)
 977                 seq_printf(m, " phys=%lx", v->phys_addr);
 978
 979         if (v->flags & VM_IOREMAP)
 980                 seq_printf(m, " ioremap");
 981
 982         if (v->flags & VM_ALLOC)
 983                 seq_printf(m, " vmalloc");
 984
 985         if (v->flags & VM_MAP)
 986                 seq_printf(m, " vmap");
 987
 988         if (v->flags & VM_USERMAP)
 989                 seq_printf(m, " user");
 990
 991         if (v->flags & VM_VPAGES)
 992                 seq_printf(m, " vpages");
 993
 994         show_numa_info(m, v);
 995         seq_putc(m, '\n');
 996         return 0;
 997 }
 998
 999 const struct seq_operations vmalloc_op = {
1000         .start = s_start,
1001         .next = s_next,
1002         .stop = s_stop,
1003         .show = s_show,
1004 };
1005 #endif
1006