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  */
   9
  10 #include <linux/mm.h>
  11 #include <linux/module.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/interrupt.h>
  16
  17 #include <linux/vmalloc.h>
  18
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21
  22
  23 DEFINE_RWLOCK(vmlist_lock);
  24 struct vm_struct *vmlist;
  25
  26 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  27 {
  28         pte_t *pte;
  29
  30         pte = pte_offset_kernel(pmd, addr);
  31         do {
  32                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  33                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  34         } while (pte++, addr += PAGE_SIZE, addr != end);
  35 }
  36
  37 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  38                                                 unsigned long end)
  39 {
  40         pmd_t *pmd;
  41         unsigned long next;
  42
  43         pmd = pmd_offset(pud, addr);
  44         do {
  45                 next = pmd_addr_end(addr, end);
  46                 if (pmd_none_or_clear_bad(pmd))
  47                         continue;
  48                 vunmap_pte_range(pmd, addr, next);
  49         } while (pmd++, addr = next, addr != end);
  50 }
  51
  52 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  53                                                 unsigned long end)
  54 {
  55         pud_t *pud;
  56         unsigned long next;
  57
  58         pud = pud_offset(pgd, addr);
  59         do {
  60                 next = pud_addr_end(addr, end);
  61                 if (pud_none_or_clear_bad(pud))
  62                         continue;
  63                 vunmap_pmd_range(pud, addr, next);
  64         } while (pud++, addr = next, addr != end);
  65 }
  66
  67 void unmap_vm_area(struct vm_struct *area)
  68 {
  69         pgd_t *pgd;
  70         unsigned long next;
  71         unsigned long addr = (unsigned long) area->addr;
  72         unsigned long end = addr + area->size;
  73
  74         BUG_ON(addr >= end);
  75         pgd = pgd_offset_k(addr);
  76         flush_cache_vunmap(addr, end);
  77         do {
  78                 next = pgd_addr_end(addr, end);
  79                 if (pgd_none_or_clear_bad(pgd))
  80                         continue;
  81                 vunmap_pud_range(pgd, addr, next);
  82         } while (pgd++, addr = next, addr != end);
  83         flush_tlb_kernel_range((unsigned long) area->addr, end);
  84 }
  85
  86 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  87                         unsigned long end, pgprot_t prot, struct page ***pages)
  88 {
  89         pte_t *pte;
  90
  91         pte = pte_alloc_kernel(&init_mm, pmd, addr);
  92         if (!pte)
  93                 return -ENOMEM;
  94         do {
  95                 struct page *page = **pages;
  96                 WARN_ON(!pte_none(*pte));
  97                 if (!page)
  98                         return -ENOMEM;
  99                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 100                 (*pages)++;
 101         } while (pte++, addr += PAGE_SIZE, addr != end);
 102         return 0;
 103 }
 104
 105 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 106                         unsigned long end, pgprot_t prot, struct page ***pages)
 107 {
 108         pmd_t *pmd;
 109         unsigned long next;
 110
 111         pmd = pmd_alloc(&init_mm, pud, addr);
 112         if (!pmd)
 113                 return -ENOMEM;
 114         do {
 115                 next = pmd_addr_end(addr, end);
 116                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 117                         return -ENOMEM;
 118         } while (pmd++, addr = next, addr != end);
 119         return 0;
 120 }
 121
 122 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 123                         unsigned long end, pgprot_t prot, struct page ***pages)
 124 {
 125         pud_t *pud;
 126         unsigned long next;
 127
 128         pud = pud_alloc(&init_mm, pgd, addr);
 129         if (!pud)
 130                 return -ENOMEM;
 131         do {
 132                 next = pud_addr_end(addr, end);
 133                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 134                         return -ENOMEM;
 135         } while (pud++, addr = next, addr != end);
 136         return 0;
 137 }
 138
 139 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 140 {
 141         pgd_t *pgd;
 142         unsigned long next;
 143         unsigned long addr = (unsigned long) area->addr;
 144         unsigned long end = addr + area->size - PAGE_SIZE;
 145         int err;
 146
 147         BUG_ON(addr >= end);
 148         pgd = pgd_offset_k(addr);
 149         spin_lock(&init_mm.page_table_lock);
 150         do {
 151                 next = pgd_addr_end(addr, end);
 152                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 153                 if (err)
 154                         break;
 155         } while (pgd++, addr = next, addr != end);
 156         spin_unlock(&init_mm.page_table_lock);
 157         flush_cache_vmap((unsigned long) area->addr, end);
 158         return err;
 159 }
 160
 161 #define IOREMAP_MAX_ORDER       (7 + PAGE_SHIFT)        /* 128 pages */
 162
 163 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 164                                 unsigned long start, unsigned long end)
 165 {
 166         struct vm_struct **p, *tmp, *area;
 167         unsigned long align = 1;
 168         unsigned long addr;
 169
 170         if (flags & VM_IOREMAP) {
 171                 int bit = fls(size);
 172
 173                 if (bit > IOREMAP_MAX_ORDER)
 174                         bit = IOREMAP_MAX_ORDER;
 175                 else if (bit < PAGE_SHIFT)
 176                         bit = PAGE_SHIFT;
 177
 178                 align = 1ul << bit;
 179         }
 180         addr = ALIGN(start, align);
 181         size = PAGE_ALIGN(size);
 182
 183         area = kmalloc(sizeof(*area), GFP_KERNEL);
 184         if (unlikely(!area))
 185                 return NULL;
 186
 187         if (unlikely(!size)) {
 188                 kfree (area);
 189                 return NULL;
 190         }
 191
 192         /*
 193          * We always allocate a guard page.
 194          */
 195         size += PAGE_SIZE;
 196
 197         write_lock(&vmlist_lock);
 198         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 199                 if ((unsigned long)tmp->addr < addr) {
 200                         if((unsigned long)tmp->addr + tmp->size >= addr)
 201                                 addr = ALIGN(tmp->size +
 202                                              (unsigned long)tmp->addr, align);
 203                         continue;
 204                 }
 205                 if ((size + addr) < addr)
 206                         goto out;
 207                 if (size + addr <= (unsigned long)tmp->addr)
 208                         goto found;
 209                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 210                 if (addr > end - size)
 211                         goto out;
 212         }
 213
 214 found:
 215         area->next = *p;
 216         *p = area;
 217
 218         area->flags = flags;
 219         area->addr = (void *)addr;
 220         area->size = size;
 221         area->pages = NULL;
 222         area->nr_pages = 0;
 223         area->phys_addr = 0;
 224         write_unlock(&vmlist_lock);
 225
 226         return area;
 227
 228 out:
 229         write_unlock(&vmlist_lock);
 230         kfree(area);
 231         if (printk_ratelimit())
 232                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 233         return NULL;
 234 }
 235
 236 /**
 237  *      get_vm_area  -  reserve a contingous kernel virtual area
 238  *
 239  *      @size:          size of the area
 240  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 241  *
 242  *      Search an area of @size in the kernel virtual mapping area,
 243  *      and reserved it for out purposes.  Returns the area descriptor
 244  *      on success or %NULL on failure.
 245  */
 246 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 247 {
 248         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 249 }
 250
 251 /**
 252  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 253  *
 254  *      @addr:          base address
 255  *
 256  *      Search for the kernel VM area starting at @addr, and remove it.
 257  *      This function returns the found VM area, but using it is NOT safe
 258  *      on SMP machines.
 259  */
 260 struct vm_struct *remove_vm_area(void *addr)
 261 {
 262         struct vm_struct **p, *tmp;
 263
 264         write_lock(&vmlist_lock);
 265         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 266                  if (tmp->addr == addr)
 267                          goto found;
 268         }
 269         write_unlock(&vmlist_lock);
 270         return NULL;
 271
 272 found:
 273         unmap_vm_area(tmp);
 274         *p = tmp->next;
 275         write_unlock(&vmlist_lock);
 276
 277         /*
 278          * Remove the guard page.
 279          */
 280         tmp->size -= PAGE_SIZE;
 281         return tmp;
 282 }
 283
 284 void __vunmap(void *addr, int deallocate_pages)
 285 {
 286         struct vm_struct *area;
 287
 288         if (!addr)
 289                 return;
 290
 291         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 292                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 293                 WARN_ON(1);
 294                 return;
 295         }
 296
 297         area = remove_vm_area(addr);
 298         if (unlikely(!area)) {
 299                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 300                                 addr);
 301                 WARN_ON(1);
 302                 return;
 303         }
 304
 305         if (deallocate_pages) {
 306                 int i;
 307
 308                 for (i = 0; i < area->nr_pages; i++) {
 309                         if (unlikely(!area->pages[i]))
 310                                 BUG();
 311                         __free_page(area->pages[i]);
 312                 }
 313
 314                 if (area->nr_pages > PAGE_SIZE/sizeof(struct page *))
 315                         vfree(area->pages);
 316                 else
 317                         kfree(area->pages);
 318         }
 319
 320         kfree(area);
 321         return;
 322 }
 323
 324 /**
 325  *      vfree  -  release memory allocated by vmalloc()
 326  *
 327  *      @addr:          memory base address
 328  *
 329  *      Free the virtually contiguous memory area starting at @addr, as
 330  *      obtained from vmalloc(), vmalloc_32() or __vmalloc().
 331  *
 332  *      May not be called in interrupt context.
 333  */
 334 void vfree(void *addr)
 335 {
 336         BUG_ON(in_interrupt());
 337         __vunmap(addr, 1);
 338 }
 339
 340 EXPORT_SYMBOL(vfree);
 341
 342 /**
 343  *      vunmap  -  release virtual mapping obtained by vmap()
 344  *
 345  *      @addr:          memory base address
 346  *
 347  *      Free the virtually contiguous memory area starting at @addr,
 348  *      which was created from the page array passed to vmap().
 349  *
 350  *      May not be called in interrupt context.
 351  */
 352 void vunmap(void *addr)
 353 {
 354         BUG_ON(in_interrupt());
 355         __vunmap(addr, 0);
 356 }
 357
 358 EXPORT_SYMBOL(vunmap);
 359
 360 /**
 361  *      vmap  -  map an array of pages into virtually contiguous space
 362  *
 363  *      @pages:         array of page pointers
 364  *      @count:         number of pages to map
 365  *      @flags:         vm_area->flags
 366  *      @prot:          page protection for the mapping
 367  *
 368  *      Maps @count pages from @pages into contiguous kernel virtual
 369  *      space.
 370  */
 371 void *vmap(struct page **pages, unsigned int count,
 372                 unsigned long flags, pgprot_t prot)
 373 {
 374         struct vm_struct *area;
 375
 376         if (count > num_physpages)
 377                 return NULL;
 378
 379         area = get_vm_area((count << PAGE_SHIFT), flags);
 380         if (!area)
 381                 return NULL;
 382         if (map_vm_area(area, prot, &pages)) {
 383                 vunmap(area->addr);
 384                 return NULL;
 385         }
 386
 387         return area->addr;
 388 }
 389
 390 EXPORT_SYMBOL(vmap);
 391
 392 void *__vmalloc_area(struct vm_struct *area, unsigned int __nocast gfp_mask, pgprot_t prot)
 393 {
 394         struct page **pages;
 395         unsigned int nr_pages, array_size, i;
 396
 397         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 398         array_size = (nr_pages * sizeof(struct page *));
 399
 400         area->nr_pages = nr_pages;
 401         /* Please note that the recursion is strictly bounded. */
 402         if (array_size > PAGE_SIZE)
 403                 pages = __vmalloc(array_size, gfp_mask, PAGE_KERNEL);
 404         else
 405                 pages = kmalloc(array_size, (gfp_mask & ~__GFP_HIGHMEM));
 406         area->pages = pages;
 407         if (!area->pages) {
 408                 remove_vm_area(area->addr);
 409                 kfree(area);
 410                 return NULL;
 411         }
 412         memset(area->pages, 0, array_size);
 413
 414         for (i = 0; i < area->nr_pages; i++) {
 415                 area->pages[i] = alloc_page(gfp_mask);
 416                 if (unlikely(!area->pages[i])) {
 417                         /* Successfully allocated i pages, free them in __vunmap() */
 418                         area->nr_pages = i;
 419                         goto fail;
 420                 }
 421         }
 422
 423         if (map_vm_area(area, prot, &pages))
 424                 goto fail;
 425         return area->addr;
 426
 427 fail:
 428         vfree(area->addr);
 429         return NULL;
 430 }
 431
 432 /**
 433  *      __vmalloc  -  allocate virtually contiguous memory
 434  *
 435  *      @size:          allocation size
 436  *      @gfp_mask:      flags for the page level allocator
 437  *      @prot:          protection mask for the allocated pages
 438  *
 439  *      Allocate enough pages to cover @size from the page level
 440  *      allocator with @gfp_mask flags.  Map them into contiguous
 441  *      kernel virtual space, using a pagetable protection of @prot.
 442  */
 443 void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask, pgprot_t prot)
 444 {
 445         struct vm_struct *area;
 446
 447         size = PAGE_ALIGN(size);
 448         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 449                 return NULL;
 450
 451         area = get_vm_area(size, VM_ALLOC);
 452         if (!area)
 453                 return NULL;
 454
 455         return __vmalloc_area(area, gfp_mask, prot);
 456 }
 457
 458 EXPORT_SYMBOL(__vmalloc);
 459
 460 /**
 461  *      vmalloc  -  allocate virtually contiguous memory
 462  *
 463  *      @size:          allocation size
 464  *
 465  *      Allocate enough pages to cover @size from the page level
 466  *      allocator and map them into contiguous kernel virtual space.
 467  *
 468  *      For tight cotrol over page level allocator and protection flags
 469  *      use __vmalloc() instead.
 470  */
 471 void *vmalloc(unsigned long size)
 472 {
 473        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 474 }
 475
 476 EXPORT_SYMBOL(vmalloc);
 477
 478 /**
 479  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 480  *
 481  *      @size:          allocation size
 482  *
 483  *      Kernel-internal function to allocate enough pages to cover @size
 484  *      the page level allocator and map them into contiguous and
 485  *      executable kernel virtual space.
 486  *
 487  *      For tight cotrol over page level allocator and protection flags
 488  *      use __vmalloc() instead.
 489  */
 490
 491 #ifndef PAGE_KERNEL_EXEC
 492 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 493 #endif
 494
 495 void *vmalloc_exec(unsigned long size)
 496 {
 497         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 498 }
 499
 500 /**
 501  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 502  *
 503  *      @size:          allocation size
 504  *
 505  *      Allocate enough 32bit PA addressable pages to cover @size from the
 506  *      page level allocator and map them into contiguous kernel virtual space.
 507  */
 508 void *vmalloc_32(unsigned long size)
 509 {
 510         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 511 }
 512
 513 EXPORT_SYMBOL(vmalloc_32);
 514
 515 long vread(char *buf, char *addr, unsigned long count)
 516 {
 517         struct vm_struct *tmp;
 518         char *vaddr, *buf_start = buf;
 519         unsigned long n;
 520
 521         /* Don't allow overflow */
 522         if ((unsigned long) addr + count < count)
 523                 count = -(unsigned long) addr;
 524
 525         read_lock(&vmlist_lock);
 526         for (tmp = vmlist; tmp; tmp = tmp->next) {
 527                 vaddr = (char *) tmp->addr;
 528                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 529                         continue;
 530                 while (addr < vaddr) {
 531                         if (count == 0)
 532                                 goto finished;
 533                         *buf = '\0';
 534                         buf++;
 535                         addr++;
 536                         count--;
 537                 }
 538                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 539                 do {
 540                         if (count == 0)
 541                                 goto finished;
 542                         *buf = *addr;
 543                         buf++;
 544                         addr++;
 545                         count--;
 546                 } while (--n > 0);
 547         }
 548 finished:
 549         read_unlock(&vmlist_lock);
 550         return buf - buf_start;
 551 }
 552
 553 long vwrite(char *buf, char *addr, unsigned long count)
 554 {
 555         struct vm_struct *tmp;
 556         char *vaddr, *buf_start = buf;
 557         unsigned long n;
 558
 559         /* Don't allow overflow */
 560         if ((unsigned long) addr + count < count)
 561                 count = -(unsigned long) addr;
 562
 563         read_lock(&vmlist_lock);
 564         for (tmp = vmlist; tmp; tmp = tmp->next) {
 565                 vaddr = (char *) tmp->addr;
 566                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 567                         continue;
 568                 while (addr < vaddr) {
 569                         if (count == 0)
 570                                 goto finished;
 571                         buf++;
 572                         addr++;
 573                         count--;
 574                 }
 575                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 576                 do {
 577                         if (count == 0)
 578                                 goto finished;
 579                         *addr = *buf;
 580                         buf++;
 581                         addr++;
 582                         count--;
 583                 } while (--n > 0);
 584         }
 585 finished:
 586         read_unlock(&vmlist_lock);
 587         return buf - buf_start;
 588 }