package/linux/kernel-source/arch/mips/brcm-boards/bcm947xx/compressed/decompress_bunzip2.c

   1 /* vi: set sw=4 ts=4: */
   2 /*      Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
   3
   4         Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
   5         which also acknowledges contributions by Mike Burrows, David Wheeler,
   6         Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
   7         Robert Sedgewick, and Jon L. Bentley.
   8
   9         This code is licensed under the LGPLv2:
  10                 LGPL (http://www.gnu.org/copyleft/lgpl.html
  11 */
  12
  13 /*
  14         Size and speed optimizations by Manuel Novoa III  (mjn3@codepoet.org).
  15
  16         More efficient reading of huffman codes, a streamlined read_bunzip()
  17         function, and various other tweaks.  In (limited) tests, approximately
  18         20% faster than bzcat on x86 and about 10% faster on arm.
  19
  20         Note that about 2/3 of the time is spent in read_unzip() reversing
  21         the Burrows-Wheeler transformation.  Much of that time is delay
  22         resulting from cache misses.
  23
  24         I would ask that anyone benefiting from this work, especially those
  25         using it in commercial products, consider making a donation to my local
  26         non-profit hospice organization in the name of the woman I loved, who
  27         passed away Feb. 12, 2003.
  28
  29                 In memory of Toni W. Hagan
  30
  31                 Hospice of Acadiana, Inc.
  32                 2600 Johnston St., Suite 200
  33                 Lafayette, LA 70503-3240
  34
  35                 Phone (337) 232-1234 or 1-800-738-2226
  36                 Fax   (337) 232-1297
  37
  38                 http://www.hospiceacadiana.com/
  39
  40         Manuel
  41  */
  42
  43 /* May 21, 2004        Manuel Novoa III
  44  * Modified to load a bzip'd kernel on the linksys wrt54g.
  45  *
  46  * May 30, 2004
  47  * Further size reduction via inlining and disabling len check code.
  48  */
  49
  50 /**********************************************************************/
  51
  52 /* Note... the LED code is specific to the v2.0 (and GS?) unit. */
  53 #undef  ENABLE_LEDS
  54 /* #define ENABLE_LEDS  1 */
  55
  56 /* Do we want to bother with checking the bzip'd data for errors? */
  57 #undef  ENABLE_BUNZIP_CHECKING
  58 /* #define ENABLE_BUNZIP_CHECKING  1 */
  59
  60 /**********************************************************************/
  61 /* #include <bcm4710.h> */
  62 #define BCM4710_FLASH           0x1fc00000      /* Flash */
  63
  64 #define KSEG0                   0x80000000
  65 #define KSEG1                   0xa0000000
  66
  67 #define KSEG1ADDR(a)            ((((unsigned)(a)) & 0x1fffffffU) | KSEG1)
  68
  69 /* The following cache code was taken from the file bcm4710_cache.h
  70  * which was necessarily GPL as it was used to build the linksys
  71  * kernel for the wrt54g. */
  72
  73 #warning icache [l]size hardcoded
  74
  75 #define icache_size                             8192
  76 #define ic_lsize                                  16
  77
  78 #define Index_Invalidate_I              0x00
  79
  80 #define cache_unroll(base,op)   \
  81         __asm__ __volatile__(           \
  82                 ".set noreorder;\n"             \
  83                 ".set mips3;\n"                 \
  84                 "cache %1, (%0);\n"             \
  85                 ".set mips0;\n"                 \
  86                 ".set reorder\n"                \
  87                 :                                               \
  88                 : "r" (base),                   \
  89                   "i" (op));
  90
  91 static __inline__ void blast_icache(void)
  92 {
  93         unsigned long start = KSEG0;
  94         unsigned long end = (start + icache_size);
  95
  96         while(start < end) {
  97                 cache_unroll(start,Index_Invalidate_I);
  98                 start += ic_lsize;
  99         }
 100 }
 101
 102 /**********************************************************************/
 103 #ifndef INT_MAX
 104 #define INT_MAX (((1 << 30)-1)*2 + 1)
 105 #endif
 106 /**********************************************************************/
 107 #ifdef ENABLE_BUNZIP_CHECKING
 108
 109 #define REBOOT          do {} while (1)
 110
 111 #else
 112
 113 #define REBOOT          ((void) 0)
 114
 115 #endif
 116 /**********************************************************************/
 117 #ifdef ENABLE_LEDS
 118
 119 #define LED_POWER_ON            0x02 /* OFF == flashing */
 120 #define LED_DMZ_OFF                     0x80
 121 #define LED_WLAN_OFF            0x01
 122
 123 #define LED_CODE_0                      (LED_POWER_ON | LED_DMZ_OFF | LED_WLAN_OFF)
 124 #define LED_CODE_1                      (LED_POWER_ON | LED_DMZ_OFF)
 125 #define LED_CODE_2                      (LED_POWER_ON               | LED_WLAN_OFF)
 126 #define LED_CODE_3                      (LED_POWER_ON)
 127
 128 #define SET_LED_ERROR(X) \
 129         do { \
 130         *(volatile u8*)(KSEG1ADDR(0x18000064))=(X & ~LED_POWER_ON); \
 131         *(volatile u8*)(KSEG1ADDR(0x18000068))=0; /* Disable changes */ \
 132         REBOOT; \
 133         } while (0)
 134
 135 #define SET_LED(X)      *(volatile u8*)(KSEG1ADDR(0x18000064))=X;
 136
 137
 138 typedef unsigned char u8;
 139
 140 #else
 141
 142 #define SET_LED_ERROR(X)                REBOOT
 143 #define SET_LED(X)                              ((void)0)
 144
 145 #endif
 146
 147 /**********************************************************************/
 148
 149 /* Constants for huffman coding */
 150 #define MAX_GROUPS                      6
 151 #define GROUP_SIZE              50              /* 64 would have been more efficient */
 152 #define MAX_HUFCODE_BITS        20              /* Longest huffman code allowed */
 153 #define MAX_SYMBOLS             258             /* 256 literals + RUNA + RUNB */
 154 #define SYMBOL_RUNA                     0
 155 #define SYMBOL_RUNB                     1
 156
 157 /* Status return values */
 158 #define RETVAL_OK                                               0
 159 #define RETVAL_LAST_BLOCK                               (-1)
 160 #define RETVAL_NOT_BZIP_DATA                    (-2)
 161 #define RETVAL_UNEXPECTED_INPUT_EOF             (-3)
 162 #define RETVAL_UNEXPECTED_OUTPUT_EOF    (-4)
 163 #define RETVAL_DATA_ERROR                               (-5)
 164 #define RETVAL_OUT_OF_MEMORY                    (-6)
 165 #define RETVAL_OBSOLETE_INPUT                   (-7)
 166
 167 /* Other housekeeping constants */
 168 #define IOBUF_SIZE                      4096
 169
 170 /* This is what we know about each huffman coding group */
 171 struct group_data {
 172         /* We have an extra slot at the end of limit[] for a sentinal value. */
 173         int limit[MAX_HUFCODE_BITS+1],base[MAX_HUFCODE_BITS],permute[MAX_SYMBOLS];
 174         int minLen, maxLen;
 175 };
 176
 177 /* Structure holding all the housekeeping data, including IO buffers and
 178    memory that persists between calls to bunzip */
 179 typedef struct {
 180         /* State for interrupting output loop */
 181         int writeCopies,writePos,writeRunCountdown,writeCount,writeCurrent;
 182         /* I/O tracking data (file handles, buffers, positions, etc.) */
 183 #if defined(ENABLE_BUNZIP_CHECKING)
 184         int /*in_fd,out_fd,*/ inbufCount,inbufPos /*,outbufPos*/;
 185 #else
 186         int /*in_fd,out_fd,inbufCount,*/ inbufPos /*,outbufPos*/;
 187 #endif
 188         unsigned char *inbuf /*,*outbuf*/;
 189         unsigned int inbufBitCount, inbufBits;
 190         /* The CRC values stored in the block header and calculated from the data */
 191 #ifdef ENABLE_BUNZIP_CHECKING
 192         unsigned int crc32Table[256],headerCRC, totalCRC, writeCRC;
 193         /* Intermediate buffer and its size (in bytes) */
 194         unsigned int *dbuf, dbufSize;
 195 #else
 196         unsigned int *dbuf;
 197 #endif
 198         /* These things are a bit too big to go on the stack */
 199         unsigned char selectors[32768];                 /* nSelectors=15 bits */
 200         struct group_data groups[MAX_GROUPS];   /* huffman coding tables */
 201 } bunzip_data;
 202
 203 static int get_next_block(bunzip_data *bd);
 204
 205 /**********************************************************************/
 206 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
 207    If start_bunzip was initialized with out_fd=-1, then up to len bytes of
 208    data are written to outbuf.  Return value is number of bytes written or
 209    error (all errors are negative numbers).  If out_fd!=-1, outbuf and len
 210    are ignored, data is written to out_fd and return is RETVAL_OK or error.
 211 */
 212
 213 static __inline__ int read_bunzip(bunzip_data *bd, char *outbuf, int len)
 214 {
 215         const unsigned int *dbuf;
 216         int pos,current,previous,gotcount;
 217 #ifdef ENABLE_LEDS
 218         int led_state = LED_CODE_2;
 219 #endif
 220
 221         /* If last read was short due to end of file, return last block now */
 222         if(bd->writeCount<0) return bd->writeCount;
 223
 224         gotcount = 0;
 225         dbuf=bd->dbuf;
 226         pos=bd->writePos;
 227         current=bd->writeCurrent;
 228
 229         /* We will always have pending decoded data to write into the output
 230            buffer unless this is the very first call (in which case we haven't
 231            huffman-decoded a block into the intermediate buffer yet). */
 232
 233         if (bd->writeCopies) {
 234                 /* Inside the loop, writeCopies means extra copies (beyond 1) */
 235                 --bd->writeCopies;
 236                 /* Loop outputting bytes */
 237                 for(;;) {
 238 #if 0           /* Might want to enable this if passing a limiting size. */
 239 /* #ifdef ENABLE_BUNZIP_CHECKING */
 240                         /* If the output buffer is full, snapshot state and return */
 241                         if(gotcount >= len) {
 242                                 bd->writePos=pos;
 243                                 bd->writeCurrent=current;
 244                                 bd->writeCopies++;
 245                                 return len;
 246                         }
 247 #endif
 248                         /* Write next byte into output buffer, updating CRC */
 249                         outbuf[gotcount++] = current;
 250 #ifdef ENABLE_BUNZIP_CHECKING
 251                         bd->writeCRC=(((bd->writeCRC)<<8)
 252                                                   ^bd->crc32Table[((bd->writeCRC)>>24)^current]);
 253 #endif
 254                         /* Loop now if we're outputting multiple copies of this byte */
 255                         if (bd->writeCopies) {
 256                                 --bd->writeCopies;
 257                                 continue;
 258                         }
 259 decode_next_byte:
 260                         if (!bd->writeCount--) break;
 261                         /* Follow sequence vector to undo Burrows-Wheeler transform */
 262                         previous=current;
 263                         pos=dbuf[pos];
 264                         current=pos&0xff;
 265                         pos>>=8;
 266                         /* After 3 consecutive copies of the same byte, the 4th is a repeat
 267                            count.  We count down from 4 instead
 268                          * of counting up because testing for non-zero is faster */
 269                         if(--bd->writeRunCountdown) {
 270                                 if(current!=previous) bd->writeRunCountdown=4;
 271                         } else {
 272                                 /* We have a repeated run, this byte indicates the count */
 273                                 bd->writeCopies=current;
 274                                 current=previous;
 275                                 bd->writeRunCountdown=5;
 276                                 /* Sometimes there are just 3 bytes (run length 0) */
 277                                 if(!bd->writeCopies) goto decode_next_byte;
 278                                 /* Subtract the 1 copy we'd output anyway to get extras */
 279                                 --bd->writeCopies;
 280                         }
 281                 }
 282 #ifdef ENABLE_BUNZIP_CHECKING
 283                 /* Decompression of this block completed successfully */
 284                 bd->writeCRC=~bd->writeCRC;
 285                 bd->totalCRC=((bd->totalCRC<<1) | (bd->totalCRC>>31)) ^ bd->writeCRC;
 286                 /* If this block had a CRC error, force file level CRC error. */
 287                 if(bd->writeCRC!=bd->headerCRC) {
 288                         bd->totalCRC=bd->headerCRC+1;
 289                         return RETVAL_LAST_BLOCK;
 290                 }
 291 #endif
 292         }
 293
 294 #ifdef ENABLE_LEDS
 295         if (led_state == LED_CODE_2) {
 296                 led_state = LED_CODE_1;
 297         } else {
 298                 led_state = LED_CODE_2;
 299         }
 300         SET_LED(led_state);
 301 #endif
 302
 303         /* Refill the intermediate buffer by huffman-decoding next block of input */
 304         /* (previous is just a convenient unused temp variable here) */
 305         previous=get_next_block(bd);
 306 #ifdef ENABLE_BUNZIP_CHECKING
 307         if(previous) {
 308                 bd->writeCount=previous;
 309                 return (previous!=RETVAL_LAST_BLOCK) ? previous : gotcount;
 310         }
 311         bd->writeCRC=0xffffffffUL;
 312 #else
 313         if (previous) return gotcount;
 314 #endif
 315         pos=bd->writePos;
 316         current=bd->writeCurrent;
 317         goto decode_next_byte;
 318 }
 319
 320 /**********************************************************************/
 321 /* WARNING!!! Must be the first function!!! */
 322
 323 void load_and_run(unsigned long ra)
 324 {
 325         int dbuf[900000];       /* Maximum requred */
 326         bunzip_data bd;
 327
 328         unsigned int i;
 329 #ifdef ENABLE_BUNZIP_CHECKING
 330         unsigned int j, c;
 331         int r;
 332 #endif
 333         char *p;
 334
 335 #ifdef ENABLE_LEDS
 336         *(volatile u8*)(KSEG1ADDR(0x18000068))=0x83; /* Allow all bits to change */
 337         SET_LED(LED_CODE_0);
 338 #endif
 339
 340 /*      memset(&bd,0,sizeof(bunzip_data)); */
 341         p = (char *) &bd;
 342         for (i = 0 ; i < sizeof(bunzip_data) ; i++) {
 343                 p[i] = 0;
 344         }
 345
 346         /* Find start of flash and adjust for pmon partition. */
 347         p = ((char *) KSEG1ADDR(BCM4710_FLASH)) + 0x10000;
 348
 349         SET_LED(LED_CODE_1);
 350         /* Find the start of the bzip'd data. */
 351         while ((p[0]!='B') || (p[1]!='Z') || (p[2]!='h') /*|| (p[3]!='9')*/) ++p;
 352         SET_LED(LED_CODE_2);
 353
 354         /* Setup input buffer */
 355         bd.inbuf=p+4;           /* Skip the "BZh9" header. */
 356 #ifdef ENABLE_BUNZIP_CHECKING
 357         bd.inbufCount=INT_MAX;
 358         /* Init the CRC32 table (big endian) */
 359         for(i=0;i<256;i++) {
 360                 c=i<<24;
 361                 for(j=8;j;j--)
 362                         c=c&0x80000000 ? (c<<1)^0x04c11db7 : (c<<1);
 363                 bd.crc32Table[i]=c;
 364         }
 365
 366         bd.dbufSize=900000;
 367 #endif
 368         bd.dbuf=dbuf;
 369
 370         /* Actually do the bunzip */
 371 #ifdef ENABLE_BUNZIP_CHECKING
 372         r = read_bunzip(&bd, ((char *) LOADADDR), INT_MAX);
 373         if (r > 0) {
 374                 if (bd.headerCRC==bd.totalCRC) {
 375                         SET_LED(LED_CODE_3);
 376                         {
 377                                 int code = LED_WLAN_OFF;
 378                                 int i, j;
 379                                 for (j=0 ; j < 4 ; j++) {
 380                                         for (i=0; i<(1<<27) ; i++) {}
 381                                         SET_LED(code);
 382                                         code ^= LED_WLAN_OFF;
 383                                 }
 384                         }
 385                         blast_icache();
 386                         /* Jump to load address */
 387                         ((void (*)(void)) LOADADDR)();
 388                 } else {
 389                         SET_LED_ERROR(LED_CODE_3);
 390                 }
 391         } else {
 392                 SET_LED_ERROR(LED_CODE_2);
 393         }
 394 #else
 395         read_bunzip(&bd, ((char *) LOADADDR), INT_MAX);
 396         blast_icache();
 397         /* Jump to load address */
 398         ((void (*)(void)) LOADADDR)();
 399 #endif
 400 }
 401
 402 /**********************************************************************/
 403 /* Return the next nnn bits of input.  All reads from the compressed input
 404    are done through this function.  All reads are big endian */
 405 static unsigned int get_bits(bunzip_data *bd, char bits_wanted)
 406 {
 407         unsigned int bits=0;
 408
 409         /* If we need to get more data from the byte buffer, do so.  (Loop getting
 410            one byte at a time to enforce endianness and avoid unaligned access.) */
 411         while (bd->inbufBitCount<bits_wanted) {
 412                 /* If we need to read more data from file into byte buffer, do so */
 413 #ifdef ENABLE_BUNZIP_CHECKING
 414                 if(bd->inbufPos==bd->inbufCount) {
 415                         SET_LED_ERROR(LED_CODE_0);
 416                 }
 417 #endif
 418                 /* Avoid 32-bit overflow (dump bit buffer to top of output) */
 419                 if(bd->inbufBitCount>=24) {
 420                         bits=bd->inbufBits&((1<<bd->inbufBitCount)-1);
 421                         bits_wanted-=bd->inbufBitCount;
 422                         bits<<=bits_wanted;
 423                         bd->inbufBitCount=0;
 424                 }
 425                 /* Grab next 8 bits of input from buffer. */
 426                 bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
 427                 bd->inbufBitCount+=8;
 428         }
 429         /* Calculate result */
 430         bd->inbufBitCount-=bits_wanted;
 431         bits|=(bd->inbufBits>>bd->inbufBitCount)&((1<<bits_wanted)-1);
 432
 433         return bits;
 434 }
 435
 436 /* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
 437
 438 static int get_next_block(bunzip_data *bd)
 439 {
 440         struct group_data *hufGroup;
 441 #ifdef ENABLE_BUNZIP_CHECKING
 442         int dbufCount,nextSym,dbufSize,groupCount,*base,*limit,selector,
 443                 i,j,k,t,runPos,symCount,symTotal,nSelectors,byteCount[256];
 444 #else
 445         int dbufCount,nextSym,/*dbufSize,*/groupCount,*base,*limit,selector,
 446                 i,j,k,t,runPos,symCount,symTotal,nSelectors,byteCount[256];
 447 #endif
 448         unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
 449         unsigned int *dbuf,origPtr;
 450
 451         dbuf=bd->dbuf;
 452 #ifdef ENABLE_BUNZIP_CHECKING
 453         dbufSize=bd->dbufSize;
 454 #endif
 455         selectors=bd->selectors;
 456         /* Read in header signature and CRC, then validate signature.
 457            (last block signature means CRC is for whole file, return now) */
 458         i = get_bits(bd,24);
 459         j = get_bits(bd,24);
 460 #ifdef ENABLE_BUNZIP_CHECKING
 461         bd->headerCRC=get_bits(bd,32);
 462         if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK;
 463         if ((i != 0x314159) || (j != 0x265359)) return RETVAL_NOT_BZIP_DATA;
 464         /* We can add support for blockRandomised if anybody complains.  There was
 465            some code for this in busybox 1.0.0-pre3, but nobody ever noticed that
 466            it didn't actually work. */
 467         if(get_bits(bd,1)) return RETVAL_OBSOLETE_INPUT;
 468         if((origPtr=get_bits(bd,24)) > dbufSize) return RETVAL_DATA_ERROR;
 469 #else
 470         get_bits(bd,32);
 471         if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK;
 472         get_bits(bd,1);
 473         origPtr=get_bits(bd,24);
 474 #endif
 475         /* mapping table: if some byte values are never used (encoding things
 476            like ascii text), the compression code removes the gaps to have fewer
 477            symbols to deal with, and writes a sparse bitfield indicating which
 478            values were present.  We make a translation table to convert the symbols
 479            back to the corresponding bytes. */
 480         t=get_bits(bd, 16);
 481         symTotal=0;
 482         for (i=0;i<16;i++) {
 483                 if(t&(1<<(15-i))) {
 484                         k=get_bits(bd,16);
 485                         for(j=0;j<16;j++)
 486                                 if(k&(1<<(15-j))) symToByte[symTotal++]=(16*i)+j;
 487                 }
 488         }
 489         /* How many different huffman coding groups does this block use? */
 490         groupCount=get_bits(bd,3);
 491 #ifdef ENABLE_BUNZIP_CHECKING
 492         if (groupCount<2 || groupCount>MAX_GROUPS) return RETVAL_DATA_ERROR;
 493 #endif
 494         /* nSelectors: Every GROUP_SIZE many symbols we select a new huffman coding
 495            group.  Read in the group selector list, which is stored as MTF encoded
 496            bit runs.  (MTF=Move To Front, as each value is used it's moved to the
 497            start of the list.) */
 498 #ifdef ENABLE_BUNZIP_CHECKING
 499         if(!(nSelectors=get_bits(bd, 15))) return RETVAL_DATA_ERROR;
 500 #else
 501         nSelectors=get_bits(bd, 15);
 502 #endif
 503         for(i=0; i<groupCount; i++) mtfSymbol[i] = i;
 504         for(i=0; i<nSelectors; i++) {
 505                 /* Get next value */
 506 #ifdef ENABLE_BUNZIP_CHECKING
 507                 for(j=0;get_bits(bd,1);j++) if (j>=groupCount) return RETVAL_DATA_ERROR;
 508 #else
 509                 for(j=0;get_bits(bd,1);j++) ;
 510 #endif
 511                 /* Decode MTF to get the next selector */
 512                 uc = mtfSymbol[j];
 513                 for(;j;j--) mtfSymbol[j] = mtfSymbol[j-1];
 514                 mtfSymbol[0]=selectors[i]=uc;
 515         }
 516         /* Read the huffman coding tables for each group, which code for symTotal
 517            literal symbols, plus two run symbols (RUNA, RUNB) */
 518         symCount=symTotal+2;
 519         for (j=0; j<groupCount; j++) {
 520                 unsigned char length[MAX_SYMBOLS],temp[MAX_HUFCODE_BITS+1];
 521                 int     minLen, maxLen, pp;
 522                 /* Read huffman code lengths for each symbol.  They're stored in
 523                    a way similar to mtf; record a starting value for the first symbol,
 524                    and an offset from the previous value for everys symbol after that.
 525                    (Subtracting 1 before the loop and then adding it back at the end is
 526                    an optimization that makes the test inside the loop simpler: symbol
 527                    length 0 becomes negative, so an unsigned inequality catches it.) */
 528                 t=get_bits(bd, 5)-1;
 529                 for (i = 0; i < symCount; i++) {
 530                         for(;;) {
 531 #ifdef ENABLE_BUNZIP_CHECKING
 532                                 if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
 533                                         return RETVAL_DATA_ERROR;
 534 #endif
 535                                 /* If first bit is 0, stop.  Else second bit indicates whether
 536                                    to increment or decrement the value.  Optimization: grab 2
 537                                    bits and unget the second if the first was 0. */
 538                                 k = get_bits(bd,2);
 539                                 if (k < 2) {
 540                                         bd->inbufBitCount++;
 541                                         break;
 542                                 }
 543                                 /* Add one if second bit 1, else subtract 1.  Avoids if/else */
 544                                 t+=(((k+1)&2)-1);
 545                         }
 546                         /* Correct for the initial -1, to get the final symbol length */
 547                         length[i]=t+1;
 548                 }
 549                 /* Find largest and smallest lengths in this group */
 550                 minLen=maxLen=length[0];
 551                 for(i = 1; i < symCount; i++) {
 552                         if(length[i] > maxLen) maxLen = length[i];
 553                         else if(length[i] < minLen) minLen = length[i];
 554                 }
 555                 /* Calculate permute[], base[], and limit[] tables from length[].
 556                  *
 557                  * permute[] is the lookup table for converting huffman coded symbols
 558                  * into decoded symbols.  base[] is the amount to subtract from the
 559                  * value of a huffman symbol of a given length when using permute[].
 560                  *
 561                  * limit[] indicates the largest numerical value a symbol with a given
 562                  * number of bits can have.  This is how the huffman codes can vary in
 563                  * length: each code with a value>limit[length] needs another bit.
 564                  */
 565                 hufGroup=bd->groups+j;
 566                 hufGroup->minLen = minLen;
 567                 hufGroup->maxLen = maxLen;
 568                 /* Note that minLen can't be smaller than 1, so we adjust the base
 569                    and limit array pointers so we're not always wasting the first
 570                    entry.  We do this again when using them (during symbol decoding).*/
 571                 base=hufGroup->base-1;
 572                 limit=hufGroup->limit-1;
 573                 /* Calculate permute[].  Concurently, initialize temp[] and limit[]. */
 574                 pp=0;
 575                 for(i=minLen;i<=maxLen;i++) {
 576                         temp[i]=limit[i]=0;
 577                         for(t=0;t<symCount;t++)
 578                                 if(length[t]==i) hufGroup->permute[pp++] = t;
 579                 }
 580                 /* Count symbols coded for at each bit length */
 581                 for (i=0;i<symCount;i++) temp[length[i]]++;
 582                 /* Calculate limit[] (the largest symbol-coding value at each bit
 583                  * length, which is (previous limit<<1)+symbols at this level), and
 584                  * base[] (number of symbols to ignore at each bit length, which is
 585                  * limit minus the cumulative count of symbols coded for already). */
 586                 pp=t=0;
 587                 for (i=minLen; i<maxLen; i++) {
 588                         pp+=temp[i];
 589                         /* We read the largest possible symbol size and then unget bits
 590                            after determining how many we need, and those extra bits could
 591                            be set to anything.  (They're noise from future symbols.)  At
 592                            each level we're really only interested in the first few bits,
 593                            so here we set all the trailing to-be-ignored bits to 1 so they
 594                            don't affect the value>limit[length] comparison. */
 595                         limit[i]= (pp << (maxLen - i)) - 1;
 596                         pp<<=1;
 597                         base[i+1]=pp-(t+=temp[i]);
 598                 }
 599                 limit[maxLen+1] = INT_MAX; /* Sentinal value for reading next sym. */
 600                 limit[maxLen]=pp+temp[maxLen]-1;
 601                 base[minLen]=0;
 602         }
 603         /* We've finished reading and digesting the block header.  Now read this
 604            block's huffman coded symbols from the file and undo the huffman coding
 605            and run length encoding, saving the result into dbuf[dbufCount++]=uc */
 606
 607         /* Initialize symbol occurrence counters and symbol Move To Front table */
 608         for(i=0;i<256;i++) {
 609                 byteCount[i] = 0;
 610                 mtfSymbol[i]=(unsigned char)i;
 611         }
 612         /* Loop through compressed symbols. */
 613         runPos=dbufCount=symCount=selector=0;
 614         for(;;) {
 615                 /* Determine which huffman coding group to use. */
 616                 if(!(symCount--)) {
 617                         symCount=GROUP_SIZE-1;
 618 #ifdef ENABLE_BUNZIP_CHECKING
 619                         if(selector>=nSelectors) return RETVAL_DATA_ERROR;
 620 #endif
 621                         hufGroup=bd->groups+selectors[selector++];
 622                         base=hufGroup->base-1;
 623                         limit=hufGroup->limit-1;
 624                 }
 625                 /* Read next huffman-coded symbol. */
 626                 /* Note: It is far cheaper to read maxLen bits and back up than it is
 627                    to read minLen bits and then an additional bit at a time, testing
 628                    as we go.  Because there is a trailing last block (with file CRC),
 629                    there is no danger of the overread causing an unexpected EOF for a
 630                    valid compressed file.  As a further optimization, we do the read
 631                    inline (falling back to a call to get_bits if the buffer runs
 632                    dry).  The following (up to got_huff_bits:) is equivalent to
 633                    j=get_bits(bd,hufGroup->maxLen);
 634                  */
 635                 while (bd->inbufBitCount<hufGroup->maxLen) {
 636 #ifdef ENABLE_BUNZIP_CHECKING
 637                         if(bd->inbufPos==bd->inbufCount) {
 638                                 j = get_bits(bd,hufGroup->maxLen);
 639                                 goto got_huff_bits;
 640                         }
 641 #endif
 642                         bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
 643                         bd->inbufBitCount+=8;
 644                 };
 645                 bd->inbufBitCount-=hufGroup->maxLen;
 646                 j = (bd->inbufBits>>bd->inbufBitCount)&((1<<hufGroup->maxLen)-1);
 647 got_huff_bits:
 648                 /* Figure how how many bits are in next symbol and unget extras */
 649                 i=hufGroup->minLen;
 650                 while(j>limit[i]) ++i;
 651                 bd->inbufBitCount += (hufGroup->maxLen - i);
 652                 /* Huffman decode value to get nextSym (with bounds checking) */
 653 #ifdef ENABLE_BUNZIP_CHECKING
 654                 if ((i > hufGroup->maxLen)
 655                         || (((unsigned)(j=(j>>(hufGroup->maxLen-i))-base[i]))
 656                                 >= MAX_SYMBOLS))
 657                         return RETVAL_DATA_ERROR;
 658 #else
 659                 j=(j>>(hufGroup->maxLen-i))-base[i];
 660 #endif
 661                 nextSym = hufGroup->permute[j];
 662                 /* We have now decoded the symbol, which indicates either a new literal
 663                    byte, or a repeated run of the most recent literal byte.  First,
 664                    check if nextSym indicates a repeated run, and if so loop collecting
 665                    how many times to repeat the last literal. */
 666                 if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
 667                         /* If this is the start of a new run, zero out counter */
 668                         if(!runPos) {
 669                                 runPos = 1;
 670                                 t = 0;
 671                         }
 672                         /* Neat trick that saves 1 symbol: instead of or-ing 0 or 1 at
 673                            each bit position, add 1 or 2 instead.  For example,
 674                            1011 is 1<<0 + 1<<1 + 2<<2.  1010 is 2<<0 + 2<<1 + 1<<2.
 675                            You can make any bit pattern that way using 1 less symbol than
 676                            the basic or 0/1 method (except all bits 0, which would use no
 677                            symbols, but a run of length 0 doesn't mean anything in this
 678                            context).  Thus space is saved. */
 679                         t += (runPos << nextSym); /* +runPos if RUNA; +2*runPos if RUNB */
 680                         runPos <<= 1;
 681                         continue;
 682                 }
 683                 /* When we hit the first non-run symbol after a run, we now know
 684                    how many times to repeat the last literal, so append that many
 685                    copies to our buffer of decoded symbols (dbuf) now.  (The last
 686                    literal used is the one at the head of the mtfSymbol array.) */
 687                 if(runPos) {
 688                         runPos=0;
 689 #ifdef ENABLE_BUNZIP_CHECKING
 690                         if(dbufCount+t>=dbufSize) return RETVAL_DATA_ERROR;
 691 #endif
 692
 693                         uc = symToByte[mtfSymbol[0]];
 694                         byteCount[uc] += t;
 695                         while(t--) dbuf[dbufCount++]=uc;
 696                 }
 697                 /* Is this the terminating symbol? */
 698                 if(nextSym>symTotal) break;
 699                 /* At this point, nextSym indicates a new literal character.  Subtract
 700                    one to get the position in the MTF array at which this literal is
 701                    currently to be found.  (Note that the result can't be -1 or 0,
 702                    because 0 and 1 are RUNA and RUNB.  But another instance of the
 703                    first symbol in the mtf array, position 0, would have been handled
 704                    as part of a run above.  Therefore 1 unused mtf position minus
 705                    2 non-literal nextSym values equals -1.) */
 706 #ifdef ENABLE_BUNZIP_CHECKING
 707                 if(dbufCount>=dbufSize) return RETVAL_DATA_ERROR;
 708 #endif
 709                 i = nextSym - 1;
 710                 uc = mtfSymbol[i];
 711                 /* Adjust the MTF array.  Since we typically expect to move only a
 712                  * small number of symbols, and are bound by 256 in any case, using
 713                  * memmove here would typically be bigger and slower due to function
 714                  * call overhead and other assorted setup costs. */
 715                 do {
 716                         mtfSymbol[i] = mtfSymbol[i-1];
 717                 } while (--i);
 718                 mtfSymbol[0] = uc;
 719                 uc=symToByte[uc];
 720                 /* We have our literal byte.  Save it into dbuf. */
 721                 byteCount[uc]++;
 722                 dbuf[dbufCount++] = (unsigned int)uc;
 723         }
 724         /* At this point, we've read all the huffman-coded symbols (and repeated
 725        runs) for this block from the input stream, and decoded them into the
 726            intermediate buffer.  There are dbufCount many decoded bytes in dbuf[].
 727            Now undo the Burrows-Wheeler transform on dbuf.
 728            See http://dogma.net/markn/articles/bwt/bwt.htm
 729          */
 730         /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
 731         j=0;
 732         for(i=0;i<256;i++) {
 733                 k=j+byteCount[i];
 734                 byteCount[i] = j;
 735                 j=k;
 736         }
 737         /* Figure out what order dbuf would be in if we sorted it. */
 738         for (i=0;i<dbufCount;i++) {
 739                 uc=(unsigned char)(dbuf[i] & 0xff);
 740                 dbuf[byteCount[uc]] |= (i << 8);
 741                 byteCount[uc]++;
 742         }
 743         /* Decode first byte by hand to initialize "previous" byte.  Note that it
 744            doesn't get output, and if the first three characters are identical
 745            it doesn't qualify as a run (hence writeRunCountdown=5). */
 746         if(dbufCount) {
 747 #ifdef ENABLE_BUNZIP_CHECKING
 748                 if(origPtr>=dbufCount) return RETVAL_DATA_ERROR;
 749 #endif
 750                 bd->writePos=dbuf[origPtr];
 751             bd->writeCurrent=(unsigned char)(bd->writePos&0xff);
 752                 bd->writePos>>=8;
 753                 bd->writeRunCountdown=5;
 754         }
 755         bd->writeCount=dbufCount;
 756
 757         return RETVAL_OK;
 758 }