mm/nommu.c

   1 /*
   2  *  linux/mm/nommu.c
   3  *
   4  *  Replacement code for mm functions to support CPU's that don't
   5  *  have any form of memory management unit (thus no virtual memory).
   6  *
   7  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
   8  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
   9  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/mman.h>
  14 #include <linux/swap.h>
  15 #include <linux/smp_lock.h>
  16 #include <linux/highmem.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/slab.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/backing-dev.h>
  22
  23 #include <asm/uaccess.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26
  27 void *high_memory;
  28 struct page *mem_map;
  29 unsigned long max_mapnr;
  30 unsigned long num_physpages;
  31 unsigned long askedalloc, realalloc;
  32 atomic_t vm_committed_space = ATOMIC_INIT(0);
  33 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  34 int sysctl_overcommit_ratio = 50; /* default is 50% */
  35 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  36
  37 EXPORT_SYMBOL(sysctl_max_map_count);
  38 EXPORT_SYMBOL(mem_map);
  39
  40 EXPORT_SYMBOL(vmalloc);
  41 EXPORT_SYMBOL(vfree);
  42 EXPORT_SYMBOL(vmalloc_to_page);
  43 EXPORT_SYMBOL(vmalloc_32);
  44
  45 /*
  46  * Handle all mappings that got truncated by a "truncate()"
  47  * system call.
  48  *
  49  * NOTE! We have to be ready to update the memory sharing
  50  * between the file and the memory map for a potential last
  51  * incomplete page.  Ugly, but necessary.
  52  */
  53 int vmtruncate(struct inode *inode, loff_t offset)
  54 {
  55         struct address_space *mapping = inode->i_mapping;
  56         unsigned long limit;
  57
  58         if (inode->i_size < offset)
  59                 goto do_expand;
  60         i_size_write(inode, offset);
  61
  62         truncate_inode_pages(mapping, offset);
  63         goto out_truncate;
  64
  65 do_expand:
  66         limit = current->rlim[RLIMIT_FSIZE].rlim_cur;
  67         if (limit != RLIM_INFINITY && offset > limit)
  68                 goto out_sig;
  69         if (offset > inode->i_sb->s_maxbytes)
  70                 goto out;
  71         i_size_write(inode, offset);
  72
  73 out_truncate:
  74         if (inode->i_op && inode->i_op->truncate)
  75                 inode->i_op->truncate(inode);
  76         return 0;
  77 out_sig:
  78         send_sig(SIGXFSZ, current, 0);
  79 out:
  80         return -EFBIG;
  81 }
  82
  83 EXPORT_SYMBOL(vmtruncate);
  84
  85 /*
  86  * Return the total memory allocated for this pointer, not
  87  * just what the caller asked for.
  88  *
  89  * Doesn't have to be accurate, i.e. may have races.
  90  */
  91 unsigned int kobjsize(const void *objp)
  92 {
  93         struct page *page;
  94
  95         if (!objp || !((page = virt_to_page(objp))))
  96                 return 0;
  97
  98         if (PageSlab(page))
  99                 return ksize(objp);
 100
 101         BUG_ON(page->index < 0);
 102         BUG_ON(page->index >= MAX_ORDER);
 103
 104         return (PAGE_SIZE << page->index);
 105 }
 106
 107 /*
 108  * The nommu dodgy version :-)
 109  */
 110 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 111         unsigned long start, int len, int write, int force,
 112         struct page **pages, struct vm_area_struct **vmas)
 113 {
 114         int i;
 115         static struct vm_area_struct dummy_vma;
 116
 117         for (i = 0; i < len; i++) {
 118                 if (pages) {
 119                         pages[i] = virt_to_page(start);
 120                         if (pages[i])
 121                                 page_cache_get(pages[i]);
 122                 }
 123                 if (vmas)
 124                         vmas[i] = &dummy_vma;
 125                 start += PAGE_SIZE;
 126         }
 127         return(i);
 128 }
 129
 130 EXPORT_SYMBOL(get_user_pages);
 131
 132 rwlock_t vmlist_lock = RW_LOCK_UNLOCKED;
 133 struct vm_struct *vmlist;
 134
 135 void vfree(void *addr)
 136 {
 137         kfree(addr);
 138 }
 139
 140 void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
 141 {
 142         /*
 143          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 144          */
 145         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 146 }
 147
 148 struct page * vmalloc_to_page(void *addr)
 149 {
 150         return virt_to_page(addr);
 151 }
 152
 153 long vread(char *buf, char *addr, unsigned long count)
 154 {
 155         memcpy(buf, addr, count);
 156         return count;
 157 }
 158
 159 long vwrite(char *buf, char *addr, unsigned long count)
 160 {
 161         /* Don't allow overflow */
 162         if ((unsigned long) addr + count < count)
 163                 count = -(unsigned long) addr;
 164
 165         memcpy(addr, buf, count);
 166         return(count);
 167 }
 168
 169 /*
 170  *      vmalloc  -  allocate virtually continguos memory
 171  *
 172  *      @size:          allocation size
 173  *
 174  *      Allocate enough pages to cover @size from the page level
 175  *      allocator and map them into continguos kernel virtual space.
 176  *
 177  *      For tight cotrol over page level allocator and protection flags
 178  *      use __vmalloc() instead.
 179  */
 180 void *vmalloc(unsigned long size)
 181 {
 182        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 183 }
 184
 185 /*
 186  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 187  *
 188  *      @size:          allocation size
 189  *
 190  *      Allocate enough 32bit PA addressable pages to cover @size from the
 191  *      page level allocator and map them into continguos kernel virtual space.
 192  */
 193 void *vmalloc_32(unsigned long size)
 194 {
 195         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 196 }
 197
 198 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 199 {
 200         BUG();
 201         return NULL;
 202 }
 203
 204 void vunmap(void *addr)
 205 {
 206         BUG();
 207 }
 208
 209 /*
 210  *  sys_brk() for the most part doesn't need the global kernel
 211  *  lock, except when an application is doing something nasty
 212  *  like trying to un-brk an area that has already been mapped
 213  *  to a regular file.  in this case, the unmapping will need
 214  *  to invoke file system routines that need the global lock.
 215  */
 216 asmlinkage unsigned long sys_brk(unsigned long brk)
 217 {
 218         struct mm_struct *mm = current->mm;
 219
 220         if (brk < mm->end_code || brk < mm->start_brk || brk > mm->context.end_brk)
 221                 return mm->brk;
 222
 223         if (mm->brk == brk)
 224                 return mm->brk;
 225
 226         /*
 227          * Always allow shrinking brk
 228          */
 229         if (brk <= mm->brk) {
 230                 mm->brk = brk;
 231                 return brk;
 232         }
 233
 234         /*
 235          * Ok, looks good - let it rip.
 236          */
 237         return mm->brk = brk;
 238 }
 239
 240 /*
 241  * Combine the mmap "prot" and "flags" argument into one "vm_flags" used
 242  * internally. Essentially, translate the "PROT_xxx" and "MAP_xxx" bits
 243  * into "VM_xxx".
 244  */
 245 static inline unsigned long calc_vm_flags(unsigned long prot, unsigned long flags)
 246 {
 247 #define _trans(x,bit1,bit2) \
 248 ((bit1==bit2)?(x&bit1):(x&bit1)?bit2:0)
 249
 250         unsigned long prot_bits, flag_bits;
 251         prot_bits =
 252                 _trans(prot, PROT_READ, VM_READ) |
 253                 _trans(prot, PROT_WRITE, VM_WRITE) |
 254                 _trans(prot, PROT_EXEC, VM_EXEC);
 255         flag_bits =
 256                 _trans(flags, MAP_GROWSDOWN, VM_GROWSDOWN) |
 257                 _trans(flags, MAP_DENYWRITE, VM_DENYWRITE) |
 258                 _trans(flags, MAP_EXECUTABLE, VM_EXECUTABLE);
 259         return prot_bits | flag_bits;
 260 #undef _trans
 261 }
 262
 263 #ifdef DEBUG
 264 static void show_process_blocks(struct mm_struct * mm)
 265 {
 266         struct mm_tblock_struct *tblock;
 267
 268         printk("Process blocks %d:", current->pid);
 269
 270         for (tblock = &mm->context.tblock; tblock; tblock = tblock->next) {
 271                 printk(" %p: %p", tblock, tblock->rblock);
 272                 if (tblock->rblock)
 273                         printk(" (%d @%p #%d)", kobjsize(tblock->rblock->kblock), tblock->rblock->kblock, tblock->rblock->refcount);
 274                 printk(tblock->next ? " ->" : ".\n");
 275         }
 276 }
 277 #endif /* DEBUG */
 278
 279 unsigned long do_mmap_pgoff(
 280         struct file * file,
 281         unsigned long addr,
 282         unsigned long len,
 283         unsigned long prot,
 284         unsigned long flags,
 285         unsigned long pgoff)
 286 {
 287         void * result;
 288         struct mm_tblock_struct * tblock;
 289         unsigned int vm_flags;
 290
 291         /*
 292          * Get the !CONFIG_MMU specific checks done first
 293          */
 294         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && (file)) {
 295                 printk("MAP_SHARED not supported (cannot write mappings to disk)\n");
 296                 return -EINVAL;
 297         }
 298
 299         if ((prot & PROT_WRITE) && (flags & MAP_PRIVATE)) {
 300                 printk("Private writable mappings not supported\n");
 301                 return -EINVAL;
 302         }
 303
 304         /*
 305          *      now all the standard checks
 306          */
 307         if (file && (!file->f_op || !file->f_op->mmap))
 308                 return -ENODEV;
 309
 310         if (PAGE_ALIGN(len) == 0)
 311                 return addr;
 312
 313         if (len > TASK_SIZE)
 314                 return -EINVAL;
 315
 316         /* offset overflow? */
 317         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 318                 return -EINVAL;
 319
 320         /* Do simple checking here so the lower-level routines won't have
 321          * to. we assume access permissions have been handled by the open
 322          * of the memory object, so we don't do any here.
 323          */
 324         vm_flags = calc_vm_flags(prot,flags) /* | mm->def_flags */ | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 325
 326         /*
 327          * determine the object being mapped and call the appropriate
 328          * specific mapper.
 329          */
 330         if (file) {
 331                 struct vm_area_struct vma;
 332                 int error;
 333
 334                 if (!file->f_op)
 335                         return -ENODEV;
 336
 337                 vma.vm_start = addr;
 338                 vma.vm_end = addr + len;
 339                 vma.vm_flags = vm_flags;
 340                 vma.vm_pgoff = pgoff;
 341
 342 #ifdef CONFIG_MAGIC_ROM_PTR
 343                 /* First, try simpler routine designed to give us a ROM pointer. */
 344
 345                 if (file->f_op->romptr && !(prot & PROT_WRITE)) {
 346                         error = file->f_op->romptr(file, &vma);
 347 #ifdef DEBUG
 348                         printk("romptr mmap returned %d, start 0x%.8x\n", error,
 349                                         vma.vm_start);
 350 #endif
 351                         if (!error)
 352                                 return vma.vm_start;
 353                         else if (error != -ENOSYS)
 354                                 return error;
 355                 } else
 356 #endif /* CONFIG_MAGIC_ROM_PTR */
 357                 /* Then try full mmap routine, which might return a RAM pointer,
 358                    or do something truly complicated. */
 359
 360                 if (file->f_op->mmap) {
 361                         error = file->f_op->mmap(file, &vma);
 362
 363 #ifdef DEBUG
 364                         printk("f_op->mmap() returned %d/%lx\n", error, vma.vm_start);
 365 #endif
 366                         if (!error)
 367                                 return vma.vm_start;
 368                         else if (error != -ENOSYS)
 369                                 return error;
 370                 } else
 371                         return -ENODEV; /* No mapping operations defined */
 372
 373                 /* An ENOSYS error indicates that mmap isn't possible (as opposed to
 374                    tried but failed) so we'll fall through to the copy. */
 375         }
 376
 377         tblock = (struct mm_tblock_struct *)
 378                         kmalloc(sizeof(struct mm_tblock_struct), GFP_KERNEL);
 379         if (!tblock) {
 380                 printk("Allocation of tblock for %lu byte allocation from process %d failed\n", len, current->pid);
 381                 show_free_areas();
 382                 return -ENOMEM;
 383         }
 384
 385         tblock->rblock = (struct mm_rblock_struct *)
 386                         kmalloc(sizeof(struct mm_rblock_struct), GFP_KERNEL);
 387
 388         if (!tblock->rblock) {
 389                 printk("Allocation of rblock for %lu byte allocation from process %d failed\n", len, current->pid);
 390                 show_free_areas();
 391                 kfree(tblock);
 392                 return -ENOMEM;
 393         }
 394
 395         result = kmalloc(len, GFP_KERNEL);
 396         if (!result) {
 397                 printk("Allocation of length %lu from process %d failed\n", len,
 398                                 current->pid);
 399                 show_free_areas();
 400                 kfree(tblock->rblock);
 401                 kfree(tblock);
 402                 return -ENOMEM;
 403         }
 404
 405         tblock->rblock->refcount = 1;
 406         tblock->rblock->kblock = result;
 407         tblock->rblock->size = len;
 408
 409         realalloc += kobjsize(result);
 410         askedalloc += len;
 411
 412 #ifdef WARN_ON_SLACK
 413         if ((len+WARN_ON_SLACK) <= kobjsize(result))
 414                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", len, current->pid, kobjsize(result)-len);
 415 #endif
 416
 417         if (file) {
 418                 int error;
 419                 mm_segment_t old_fs = get_fs();
 420                 set_fs(KERNEL_DS);
 421                 error = file->f_op->read(file, (char *) result, len, &file->f_pos);
 422                 set_fs(old_fs);
 423                 if (error < 0) {
 424                         kfree(result);
 425                         kfree(tblock->rblock);
 426                         kfree(tblock);
 427                         return error;
 428                 }
 429                 if (error < len)
 430                         memset(result+error, '\0', len-error);
 431         } else {
 432                 memset(result, '\0', len);
 433         }
 434
 435         realalloc += kobjsize(tblock);
 436         askedalloc += sizeof(struct mm_tblock_struct);
 437
 438         realalloc += kobjsize(tblock->rblock);
 439         askedalloc += sizeof(struct mm_rblock_struct);
 440
 441         tblock->next = current->mm->context.tblock.next;
 442         current->mm->context.tblock.next = tblock;
 443         current->mm->total_vm += len >> PAGE_SHIFT;
 444
 445 #ifdef DEBUG
 446         printk("do_mmap:\n");
 447         show_process_blocks(current->mm);
 448 #endif
 449
 450         return (unsigned long)result;
 451 }
 452
 453 int do_munmap(struct mm_struct * mm, unsigned long addr, size_t len)
 454 {
 455         struct mm_tblock_struct * tblock, *tmp;
 456
 457 #ifdef CONFIG_MAGIC_ROM_PTR
 458         /*
 459          * For efficiency's sake, if the pointer is obviously in ROM,
 460          * don't bother walking the lists to free it.
 461          */
 462         if (is_in_rom(addr))
 463                 return 0;
 464 #endif /* CONFIG_MAGIC_ROM_PTR */
 465
 466 #ifdef DEBUG
 467         printk("do_munmap:\n");
 468 #endif
 469
 470         tmp = &mm->context.tblock; /* dummy head */
 471         while ((tblock=tmp->next) && tblock->rblock &&
 472                         tblock->rblock->kblock != (void*)addr)
 473                 tmp = tblock;
 474
 475         if (!tblock) {
 476                 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 477                                 current->pid, current->comm, (void*)addr);
 478                 return -EINVAL;
 479         }
 480         if (tblock->rblock) {
 481                 if (!--tblock->rblock->refcount) {
 482                         if (tblock->rblock->kblock) {
 483                                 realalloc -= kobjsize(tblock->rblock->kblock);
 484                                 askedalloc -= tblock->rblock->size;
 485                                 kfree(tblock->rblock->kblock);
 486                         }
 487
 488                         realalloc -= kobjsize(tblock->rblock);
 489                         askedalloc -= sizeof(struct mm_rblock_struct);
 490                         kfree(tblock->rblock);
 491                 }
 492         }
 493         tmp->next = tblock->next;
 494         realalloc -= kobjsize(tblock);
 495         askedalloc -= sizeof(struct mm_tblock_struct);
 496         kfree(tblock);
 497         mm->total_vm -= len >> PAGE_SHIFT;
 498
 499 #ifdef DEBUG
 500         show_process_blocks(mm);
 501 #endif
 502
 503         return 0;
 504 }
 505
 506 /* Release all mmaps. */
 507 void exit_mmap(struct mm_struct * mm)
 508 {
 509         struct mm_tblock_struct *tmp;
 510         mm->total_vm = 0;
 511
 512         if (!mm)
 513                 return;
 514
 515 #ifdef DEBUG
 516         printk("Exit_mmap:\n");
 517 #endif
 518
 519         while((tmp = mm->context.tblock.next)) {
 520                 if (tmp->rblock) {
 521                         if (!--tmp->rblock->refcount) {
 522                                 if (tmp->rblock->kblock) {
 523                                         realalloc -= kobjsize(tmp->rblock->kblock);
 524                                         askedalloc -= tmp->rblock->size;
 525                                         kfree(tmp->rblock->kblock);
 526                                 }
 527                                 realalloc -= kobjsize(tmp->rblock);
 528                                 askedalloc -= sizeof(struct mm_rblock_struct);
 529                                 kfree(tmp->rblock);
 530                         }
 531                         tmp->rblock = 0;
 532                 }
 533                 mm->context.tblock.next = tmp->next;
 534                 realalloc -= kobjsize(tmp);
 535                 askedalloc -= sizeof(struct mm_tblock_struct);
 536                 kfree(tmp);
 537         }
 538
 539 #ifdef DEBUG
 540         show_process_blocks(mm);
 541 #endif
 542 }
 543
 544 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 545 {
 546         int ret;
 547         struct mm_struct *mm = current->mm;
 548
 549         down_write(&mm->mmap_sem);
 550         ret = do_munmap(mm, addr, len);
 551         up_write(&mm->mmap_sem);
 552         return ret;
 553 }
 554
 555 unsigned long do_brk(unsigned long addr, unsigned long len)
 556 {
 557         return -ENOMEM;
 558 }
 559
 560 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
 561 {
 562         return NULL;
 563 }
 564
 565 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
 566 {
 567         return NULL;
 568 }
 569
 570 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 571 {
 572         return NULL;
 573 }
 574
 575 int remap_page_range(struct vm_area_struct *vma, unsigned long from,
 576                 unsigned long to, unsigned long size, pgprot_t prot)
 577 {
 578         return -EPERM;
 579 }
 580
 581 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
 582         unsigned long len, unsigned long pgoff, unsigned long flags)
 583 {
 584         return -ENOMEM;
 585 }
 586
 587 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 588 {
 589 }
 590
 591 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
 592         unsigned long len, unsigned long pgoff, unsigned long flags)
 593 {
 594         return -ENOMEM;
 595 }
 596
 597 void arch_unmap_area(struct vm_area_struct *area)
 598 {
 599 }
 600