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