| /************************************************************************* |
| | COPYRIGHT (c) 2000 BY ABATRON AG |
| |************************************************************************* |
| | |
| | PROJECT NAME: Linux Image to S-record Conversion Utility |
| | FILENAME : img2srec.c |
| | |
| | COMPILER : GCC |
| | |
| | TARGET OS : LINUX / UNIX |
| | TARGET HW : - |
| | |
| | PROGRAMMER : Abatron / RD |
| | CREATION : 07.07.00 |
| | |
| |************************************************************************* |
| | |
| | DESCRIPTION : |
| | |
| | Utility to convert a Linux Boot Image to S-record: |
| | ================================================== |
| | |
| | This command line utility can be used to convert a Linux boot image |
| | (zimage.initrd) to S-Record format used for flash programming. |
| | This conversion takes care of the special sections "IMAGE" and INITRD". |
| | |
| | img2srec [-o offset] image > image.srec |
| | |
| | |
| | Build the utility: |
| | ================== |
| | |
| | To build the utility use GCC as follows: |
| | |
| | gcc img2srec.c -o img2srec |
| | |
| | |
| |************************************************************************* |
| | |
| | |
| | UPDATES : |
| | |
| | DATE NAME CHANGES |
| | ----------------------------------------------------------- |
| | Latest update |
| | |
| | 07.07.00 aba Initial release |
| | |
| |*************************************************************************/ |
| |
| /************************************************************************* |
| | INCLUDES |
| |*************************************************************************/ |
| |
| #include <stddef.h> |
| #include <stdio.h> |
| #include <stdlib.h> |
| #include <ctype.h> |
| #include <string.h> |
| #include <elf.h> |
| #include <unistd.h> |
| #include <errno.h> |
| |
| extern int errno; |
| |
| /************************************************************************* |
| | DEFINES |
| |*************************************************************************/ |
| |
| #define FALSE 0 |
| #define TRUE 1 |
| |
| /************************************************************************* |
| | MACROS |
| |*************************************************************************/ |
| |
| /************************************************************************* |
| | TYPEDEFS |
| |*************************************************************************/ |
| |
| typedef uint8_t CHAR; |
| typedef uint8_t BYTE; |
| typedef uint16_t WORD; |
| typedef uint32_t DWORD; |
| typedef int BOOL; |
| |
| /************************************************************************* |
| | LOCALS |
| |*************************************************************************/ |
| |
| /************************************************************************* |
| | PROTOTYPES |
| |*************************************************************************/ |
| |
| static char *ExtractHex(DWORD *value, char *getPtr); |
| static char *ExtractDecimal(DWORD *value, char *getPtr); |
| static void ExtractNumber(DWORD *value, char *getPtr); |
| static BYTE *ExtractWord(WORD *value, BYTE *buffer); |
| static BYTE *ExtractLong(DWORD *value, BYTE *buffer); |
| static BYTE *ExtractBlock(WORD count, BYTE *data, BYTE *buffer); |
| static char *WriteHex(char *pa, BYTE value, WORD *pCheckSum); |
| static char *BuildSRecord(char *pa, WORD sType, DWORD addr, |
| const BYTE *data, int nCount); |
| static void ConvertELF(char *fileName, DWORD loadOffset); |
| int main(int argc, char *argv[]); |
| |
| /************************************************************************* |
| | FUNCTIONS |
| |*************************************************************************/ |
| |
| static char* ExtractHex (DWORD* value, char* getPtr) |
| { |
| DWORD num; |
| DWORD digit; |
| BYTE c; |
| |
| while (*getPtr == ' ') getPtr++; |
| num = 0; |
| for (;;) { |
| c = *getPtr; |
| if ((c >= '0') && (c <= '9')) digit = (DWORD)(c - '0'); |
| else if ((c >= 'A') && (c <= 'F')) digit = (DWORD)(c - 'A' + 10); |
| else if ((c >= 'a') && (c <= 'f')) digit = (DWORD)(c - 'a' + 10); |
| else break; |
| num <<= 4; |
| num += digit; |
| getPtr++; |
| } /* for */ |
| *value = num; |
| return getPtr; |
| } /* ExtractHex */ |
| |
| static char* ExtractDecimal (DWORD* value, char* getPtr) |
| { |
| DWORD num; |
| DWORD digit; |
| BYTE c; |
| |
| while (*getPtr == ' ') getPtr++; |
| num = 0; |
| for (;;) { |
| c = *getPtr; |
| if ((c >= '0') && (c <= '9')) digit = (DWORD)(c - '0'); |
| else break; |
| num *= 10; |
| num += digit; |
| getPtr++; |
| } /* for */ |
| *value = num; |
| return getPtr; |
| } /* ExtractDecimal */ |
| |
| |
| static void ExtractNumber (DWORD* value, char* getPtr) |
| { |
| BOOL neg = FALSE;; |
| |
| while (*getPtr == ' ') getPtr++; |
| if (*getPtr == '-') { |
| neg = TRUE; |
| getPtr++; |
| } /* if */ |
| if ((*getPtr == '0') && ((*(getPtr+1) == 'x') || (*(getPtr+1) == 'X'))) { |
| getPtr +=2; |
| (void)ExtractHex(value, getPtr); |
| } /* if */ |
| else { |
| (void)ExtractDecimal(value, getPtr); |
| } /* else */ |
| if (neg) *value = -(*value); |
| } /* ExtractNumber */ |
| |
| |
| static BYTE* ExtractWord(WORD* value, BYTE* buffer) |
| { |
| WORD x; |
| x = (WORD)*buffer++; |
| x = (x<<8) + (WORD)*buffer++; |
| *value = x; |
| return buffer; |
| } /* ExtractWord */ |
| |
| |
| static BYTE* ExtractLong(DWORD* value, BYTE* buffer) |
| { |
| DWORD x; |
| x = (DWORD)*buffer++; |
| x = (x<<8) + (DWORD)*buffer++; |
| x = (x<<8) + (DWORD)*buffer++; |
| x = (x<<8) + (DWORD)*buffer++; |
| *value = x; |
| return buffer; |
| } /* ExtractLong */ |
| |
| |
| static BYTE* ExtractBlock(WORD count, BYTE* data, BYTE* buffer) |
| { |
| while (count--) *data++ = *buffer++; |
| return buffer; |
| } /* ExtractBlock */ |
| |
| |
| static char* WriteHex(char* pa, BYTE value, WORD* pCheckSum) |
| { |
| WORD temp; |
| |
| static char ByteToHex[] = "0123456789ABCDEF"; |
| |
| *pCheckSum += value; |
| temp = value / 16; |
| *pa++ = ByteToHex[temp]; |
| temp = value % 16; |
| *pa++ = ByteToHex[temp]; |
| return pa; |
| } |
| |
| |
| static char* BuildSRecord(char* pa, WORD sType, DWORD addr, |
| const BYTE* data, int nCount) |
| { |
| WORD addrLen; |
| WORD sRLen; |
| WORD checkSum; |
| WORD i; |
| |
| switch (sType) { |
| case 0: |
| case 1: |
| case 9: |
| addrLen = 2; |
| break; |
| case 2: |
| case 8: |
| addrLen = 3; |
| break; |
| case 3: |
| case 7: |
| addrLen = 4; |
| break; |
| default: |
| return pa; |
| } /* switch */ |
| |
| *pa++ = 'S'; |
| *pa++ = (char)(sType + '0'); |
| sRLen = addrLen + nCount + 1; |
| checkSum = 0; |
| pa = WriteHex(pa, (BYTE)sRLen, &checkSum); |
| |
| /* Write address field */ |
| for (i = 1; i <= addrLen; i++) { |
| pa = WriteHex(pa, (BYTE)(addr >> (8 * (addrLen - i))), &checkSum); |
| } /* for */ |
| |
| /* Write code/data fields */ |
| for (i = 0; i < nCount; i++) { |
| pa = WriteHex(pa, *data++, &checkSum); |
| } /* for */ |
| |
| /* Write checksum field */ |
| checkSum = ~checkSum; |
| pa = WriteHex(pa, (BYTE)checkSum, &checkSum); |
| *pa++ = '\0'; |
| return pa; |
| } |
| |
| |
| static void ConvertELF(char* fileName, DWORD loadOffset) |
| { |
| FILE* file; |
| int i; |
| int rxCount; |
| BYTE rxBlock[1024]; |
| DWORD loadSize; |
| DWORD firstAddr; |
| DWORD loadAddr; |
| DWORD loadDiff = 0; |
| Elf32_Ehdr elfHeader; |
| Elf32_Shdr sectHeader[32]; |
| BYTE* getPtr; |
| char srecLine[128]; |
| char *hdr_name; |
| |
| |
| /* open file */ |
| if ((file = fopen(fileName,"rb")) == NULL) { |
| fprintf (stderr, "Can't open %s: %s\n", fileName, strerror(errno)); |
| return; |
| } /* if */ |
| |
| /* read ELF header */ |
| rxCount = fread(rxBlock, 1, sizeof elfHeader, file); |
| getPtr = ExtractBlock(sizeof elfHeader.e_ident, elfHeader.e_ident, rxBlock); |
| getPtr = ExtractWord(&elfHeader.e_type, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_machine, getPtr); |
| getPtr = ExtractLong((DWORD *)&elfHeader.e_version, getPtr); |
| getPtr = ExtractLong((DWORD *)&elfHeader.e_entry, getPtr); |
| getPtr = ExtractLong((DWORD *)&elfHeader.e_phoff, getPtr); |
| getPtr = ExtractLong((DWORD *)&elfHeader.e_shoff, getPtr); |
| getPtr = ExtractLong((DWORD *)&elfHeader.e_flags, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_ehsize, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_phentsize, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_phnum, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_shentsize, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_shnum, getPtr); |
| getPtr = ExtractWord(&elfHeader.e_shstrndx, getPtr); |
| if ( (rxCount != sizeof elfHeader) |
| || (elfHeader.e_ident[0] != ELFMAG0) |
| || (elfHeader.e_ident[1] != ELFMAG1) |
| || (elfHeader.e_ident[2] != ELFMAG2) |
| || (elfHeader.e_ident[3] != ELFMAG3) |
| || (elfHeader.e_type != ET_EXEC) |
| ) { |
| fclose(file); |
| fprintf (stderr, "*** illegal file format\n"); |
| return; |
| } /* if */ |
| |
| /* read all section headers */ |
| fseek(file, elfHeader.e_shoff, SEEK_SET); |
| for (i = 0; i < elfHeader.e_shnum; i++) { |
| rxCount = fread(rxBlock, 1, sizeof sectHeader[0], file); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_name, rxBlock); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_type, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_flags, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_addr, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_offset, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_size, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_link, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_info, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_addralign, getPtr); |
| getPtr = ExtractLong((DWORD *)§Header[i].sh_entsize, getPtr); |
| if (rxCount != sizeof sectHeader[0]) { |
| fclose(file); |
| fprintf (stderr, "*** illegal file format\n"); |
| return; |
| } /* if */ |
| } /* for */ |
| |
| if ((hdr_name = strrchr(fileName, '/')) == NULL) { |
| hdr_name = fileName; |
| } else { |
| ++hdr_name; |
| } |
| /* write start record */ |
| (void)BuildSRecord(srecLine, 0, 0, (BYTE *)hdr_name, strlen(hdr_name)); |
| printf("%s\r\n",srecLine); |
| |
| /* write data records */ |
| firstAddr = ~0; |
| loadAddr = 0; |
| for (i = 0; i < elfHeader.e_shnum; i++) { |
| if ( (sectHeader[i].sh_type == SHT_PROGBITS) |
| && (sectHeader[i].sh_size != 0) |
| ) { |
| loadSize = sectHeader[i].sh_size; |
| if (sectHeader[i].sh_flags != 0) { |
| loadAddr = sectHeader[i].sh_addr; |
| loadDiff = loadAddr - sectHeader[i].sh_offset; |
| } /* if */ |
| else { |
| loadAddr = sectHeader[i].sh_offset + loadDiff; |
| } /* else */ |
| |
| if (loadAddr < firstAddr) |
| firstAddr = loadAddr; |
| |
| /* build s-records */ |
| loadSize = sectHeader[i].sh_size; |
| fseek(file, sectHeader[i].sh_offset, SEEK_SET); |
| while (loadSize) { |
| rxCount = fread(rxBlock, 1, (loadSize > 32) ? 32 : loadSize, file); |
| if (rxCount < 0) { |
| fclose(file); |
| fprintf (stderr, "*** illegal file format\n"); |
| return; |
| } /* if */ |
| (void)BuildSRecord(srecLine, 3, loadAddr + loadOffset, rxBlock, rxCount); |
| loadSize -= rxCount; |
| loadAddr += rxCount; |
| printf("%s\r\n",srecLine); |
| } /* while */ |
| } /* if */ |
| } /* for */ |
| |
| /* add end record */ |
| (void)BuildSRecord(srecLine, 7, firstAddr + loadOffset, 0, 0); |
| printf("%s\r\n",srecLine); |
| fclose(file); |
| } /* ConvertELF */ |
| |
| |
| /************************************************************************* |
| | MAIN |
| |*************************************************************************/ |
| |
| int main( int argc, char *argv[ ]) |
| { |
| DWORD offset; |
| |
| if (argc == 2) { |
| ConvertELF(argv[1], 0); |
| } /* if */ |
| else if ((argc == 4) && (strcmp(argv[1], "-o") == 0)) { |
| ExtractNumber(&offset, argv[2]); |
| ConvertELF(argv[3], offset); |
| } /* if */ |
| else { |
| fprintf (stderr, "Usage: img2srec [-o offset] <image>\n"); |
| } /* if */ |
| |
| return 0; |
| } /* main */ |