I have posed this query to the Hi-tech C 'Other Compilers' forum; unfortunately it has had no traffic for more than one year.
I would appreciate a description of the object file format and the output of 'dump.exe' for any Hi-Tech C compiler package (for any arch) dated around 1990 - 1991 or so.
With a little work is is possible to deduce some basic fields in the object file by inspecting hex dumps and correlating these with some of the fields in the 'dump.exe' output but there are details which would take much work to discern.
I know from time to time users of Hi-Tech C read this NG and some even have the old toolchains; your help would be much appreciated.
I stumbled upon the description in an archive related to symbol files; I post it here for future reference as Usenet archives are more permanent than most other Web resources:
HI-TECH Software Object Code Format
Introduction This describes the format of the object code recognized by the linker "link". Note that later versions of the object code have identifying version numbers in IDENT records. Early version of the object code did not have these version numbers. The presence of the version number can be determined from the length of the ident record. Certain features will be identified as being present in versions greater than some value by the notation (>=m.n) where m is the major ver-sion number and n is the minor version number, e.g. (>=2.1).
Object Code Structure The object code (OCODE) is essentially binary, comprising a sequence of records, each with an outer envelope thus:
| Length (16 bits) | Record type (8 bits) | Data (Length*8 bits) |
i.e. a 16 bit length, followed by an 8 bit type, then a sequence of 8 bit bytes. The length is the number of bytes, exluding the length and type. In practice, the length of the data portion of any record will not exceed 512 bytes. This places an upper limit on the size of the buffer required to hold any record.
2.1. TEXT Record The major record type is TEXT (== 1). Its format (i.e. the data part of the basic record envelope) is thus:
| Offset (32 bits) | Psect name (null term.) | data bytes |
The Offset is the offset (in bytes) within the named psect (i.e. program section) at which the data bytes are to be placed. The name, like all names in this OCODE, is null terminated. The number of data bytes is found by subtracting the length of the name + 4 from the record length in the outer envelope.
2.2. PSECT Record The PSECT (== 2) record is used to define program sections. Each PSECT record describes one psect.
Note that as a consistency check, the null byte terminating the name must be the last byte in the record. The flag bits are: GLOBAL 020 Psect is global PURE 040 Psect is to be read-only OVRLD 0100 Each modules data for this psect starts at relative location 0. The default is concatenation. ABS 0200 This psect is to be loaded at absolute 0. BIGSEG 0400 Means the relocatibility of this psect is multiplied by 65535 BPAGE 01000 This is a base page psect - ignore overflows in relocations
2.3. RELOC Record The RELOC (== 3) type record contains relocation information relating to the last TEXT record. It consists of a sequence of offset/descriptor pairs, the offset being a 16 bit offset within the last TEXT record, and the descriptor providing the relocation information. The type of the relocation may be simple relocation within a psect, relocation by the value of an external name, or whatever. Complex relocations may also appear, which are a sequence of operators and operands, to be evaluated by a stack machine. Each simple relocation record will appear as:
| Offset (16 bits) | Reloc. type (8 bits) | Psect or external name |
The name is, as usual, null terminated. The actual RELOC record consists of an arbitrary number of these entries. The relocation type byte is interpreted as follows:
bits |7 4|3 0| | type | size |
The size is the size in bytes of the quantity to be relocated. The type is one of: RABS 0 Absolute - no relocation (i.e. no change) RPSECT 1 Relocation within named psect RNAME 2 Relocation by value of name RRPSECT 5 Relocation within psect, less the current pc. This is required for machines using relative addressing. val = val + psect.base - code.location RRNAME 6 As for RRPSECT, but relative to a symbol rather than a psect. RSPSECT 9 Relocation by the segment value for the psect RSNAME 10 Relocation by the segment value for the symbol RCPLX 48 Complex relocation follows A complex relocation entry has, in place of the name, a sequence of operator bytes and operand bytes with further information. Where binary operations are performed, the top stack entry is the left hand operand, and the next to top stack value is the right hand operand. Both values are removed from the stack and the result pushed back on. Unary operators act on the top stack value. Operands like RC_VAL and RPSECT push their value onto the stack. The possible type bytes are as follows: RC_END 0 End of complex relocations RC_LOW 1 And with 0xFF RC_HI 2 Right shift 8 bits, and with 0xFF RC_VAL 3 32 bit constant follows RC_ADD 4 Addition RC_SUB 5 Subtraction RC_MUL 6 Multiplication RC_DIV 7 Division RC_SHL 8 Shift left RC_SHR 9 Shift right RC_AND 10 Bitwise AND RC_OR 11 Bitwise OR RC_XOR 12 Bitwise XOR RC_CPL 13 Bitwise complement RC_NEG 14 Negate RC_BITF 15 Bitfield extract - operands are value, bit offset, and length RPSECT 16 Value of psect - null-terminated name follows RC_MOD 17 Modulus RNAME 32 Value of symbol - null-terminated name follows RSPSECT 144 Segment selector of psect RSNAME 160 Segment selector of symbol
2.4. SYM Record The SYM (== 4) record defines external and internal names. It consists of a sequence of name definitions. Note that it is not essential to mention an external name here if it is referenced in a RELOC record. Each entry is as follows:
| Value (32 bits) | flags (16 bits) | psect name | symbol name |
Note that the psect name will be null if the symbol is not being defined. That is, the psect name will be a single null byte. The flag word contains the same flags as defined above for PSECT records. In the low order 4 bits is one of the following values: NULL 0 The normal case when defining local or global symbols STACK 1 Symbol is a stack (auto variable) symbol COMM 2 Symbol is a common symbol - if it is defined elsewhere then this is the same as EXTERN, otherwise the linker will allocate space in the psect named, of size equal to the maximum value of any instance of this symbol encountered. REGNAM 3 Symbol refers to a register LINENO 4 This is a line number in the source code FILNAM 5 This is the name of a source file EXTERN 6 This is a reference to an externally defined symbol The psect name must be null. STACK, REGNAM, FILNAM and LINENO symbols are purely for use by a debugger. These symbols are not handled by the linker at all (other than to correctly relocate their values) but will be passed through to a symbol table unless suppressed with a -x option.
2.5. START Record The START (== 5) record defines a start address. Only one START record may appear in a module.
| offset (32 bits) | psect name (null term) |
This defines the start address to be at the specified offset in the named psect.
2.6. END Record The END (== 6) record is as follows:
| flags (16 bits) |
The flag value is currently unused.
2.7. IDENT Record The IDENT (= 7) record identifies the target machine and specifies the ordering of bytes in 32 and 16 bit quantities. This affects both offset and symbol values as well as relocatable values in TEXT records. The IDENT record is optional, and the default ordering is strictly low byte first. If the IDENT record is present, it must be the first record in the file. The length field in every record's outer envelope is always stored low byte first, irrespective of the content of any IDENT record.
| byte order (32 bits) | byte order (16 bits) | machine name | | version number (16 bits) |
The byte order fields list the relative position of successively more significant bytes within fields of 32 and 16 bits respectively. For example, the ident record for a PDP11 would look like this:
| 02 03 00 01 | 00 01 | PDP11 |
The machine name is null terminated. The ident record need not be present, if more than one ident record is encountered, all records must have identical byte ordering. The version number (>=2.0) occupies 2 bytes after the machine name. This is present only if the record length is long enough to contain these extra 2 bytes. If absent, the version number may be assumed to be 1.0. If present, the first byte is the major version number, and the second the minor version number.
2.8. XPSECT Record The XPSECT (==8) record contains further information about a psect. Some object files may not contain any XPSECT records. The layout is as follows:
| max size (32 bits) | relocatability (16 bits) | selector (16 bits) | | reserved (24 bits) | type (8 bits) | psect name |
The max size field is a long integer specifying the maximum size of this psect. The relocatability field, if nonzero, specifies a boundary on which the psect must start, e.g. for the 8086 this field is usually 16, since 8086 segment must start on a 16 byte boundary. The reserved field should be all zeros. The psect name is the null terminated name of the psect. The selector (>=2.0) is non-zero if there is a segment selector to be associated with this psect. This feature is normally only used with segmented architecture processors like the 8086. In object code versions prior to 2.0 this field was always zero. The type field (>=2.1) allows the linker to verify that all module's contributions to a given psect are of the same type. This field must match in all modules for this psect. This is typically used by the 8086 assembler to encode the state of the USE32 flag for a psect.
2.9. SEGMENT Record The SEGMENT (== 9) record (>= 2.0) is present in fully linked object files only, i.e. those produced by the linker. This defines a segment, which is a contiguous set of psects. The record structure is as follows:
The size field is the size of the segment in bytes. The base address is the physical base address of the segment. This is not necessarily the same as the origin, which is the address used when references to this segment were relocated. These correspond to the load and link addresses respectively of the psects comprising the segment. The selector is the segment selector value used in referring to this segment. This is of relevance mainly for 8086 family processors. The segment name is the name of the first psect comprising this segment.
2.10. XSYM Record The XSYM (== 10) record (>= 2.1) is similar to the SYM record but also defines a selector value for the symbol. The record layout is as follows:
| value (32 bits) | flags (16 bits) | selector (16 bits) | name |
The selector value is the selector of the segment within which the symbol is located.
2.11. SIGNAT Record The SIGNAT (== 11) record (>=2.1) defines a signature for a symbol. At link time all signatures for a symbol are compared and must be identical. If a signature is not the same, an error message is issued. The record layout is as follows:
| signature (16 bits) | symbol name |
2.12. FNINFO Record The FNINFO (== 12) record (>= 2.3) provides information necessary for call-graphing and local data allocation for non-reentrant functions. Within a record are concatenated one or more instances of the following layouts. Each instance is variable length, depending on name lengths.
| FNCALL (1) | caller name | callee name | | FNARG (2) | caller name | callee name | | FNINDIR (3) | caller name | callee signature (16 bits) | | FNADDR (4) | function name | signature (16 bits) | | FNSIZE (5) | func name | local size (32 bits) | arg size (32 bits) | | FNROOT (6) | func name |
The FNCALL instance specifies that one function directly calls another. The FNARG instance specifies that one function will have arguments built while another is called. The FNINDIR instance specifies that a function calls indirectly another (unknown) function, whose signature is as specified. The FNADDR instance specifies that the function has its address taken (and is thus a candidate for being called indirectly) and its signature. The FNSIZE instance specifies the local variable block size and the argument block size for a function. The FNROOT instance specifies that this function is the root of a call graph (either the main function or an interrupt function).
2.13. FNCONF Record The FNCONF (== 13) record (>=2.3) provides environmental information for call-graphing and local data allocation purposes. The layout comprises three null-terminated strings as follows:
| psect name | local data prefix | argument prefix |
The psect name is the name of the psect in which local data and argument blocks should be allocated by the linker. The local data and argument prefixes are the strings which should be prepended to a function name to derive the names for the local data and argument blocks for a given function.
NOTES No padding or fill bytes are stored in the object file, i.e. no alignment of the records is done. Do not attempt to read or write object files using structures, i.e. all records must be built up byte-by-byte in a char array.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.