17 Bit Software 2: Collection B

home *** CD-ROM | disk | FTP | other *** search

/ 17 Bit Software 2: Collection B / 17Bit_Collection_B.iso / files / 2032.dms / in.adf / dis.docs < prev next >

Wrap

Text File | 1978-01-10 | 12.0 KB | 239 lines

Dis - an AmigaDOS File Disassembler/Dumper General Dis is a program which will symbolically disassemble AmigaDOS object files, whether unlinked or fully linked. It can also operate in a non- disassembly mode and just dump the hunks in the files, or even just list which hunks are found. It uses my 'disassemble.library' as the disassembler, so it has all of the features of that library, including support for the full MC68000 family, label generation, controllable output format, etc. It also has special support for the case/switch constructs generated by several compilers, and hooks to allow better disassembly of AmigaDOS libraries and devices. 'Dis' will also generate appropriate output for data and bss hunks. This simplest way to use Dis is as in: Dis <object-file> Dis itself has been compiled with symbolic information left in, so a good sample of what it does can be seen by having it disassemble itself, as in Dis Dis Some interesting things can be seen from the disassembly. First, it was linked with an old version of BLink, since the first hunk is an extra text hunk added to just branch to the actual starting point. Second, it was not linked with 'smalldata', since there are multiple bss sections. Third, it was linked with 'smallcode', since there is only one text hunk (other than the one BLink added). The first few hunks are bss hunks that correspond to file level variables in the source files for Dis. Symbolic information for them was present, so Dis was able to show the individual variables. (This symbolic information is, unfortunately, only available with my Beta 1.3 version of Draco, and not with any version that has been distributed.) Hunk #1 is the space used by the run-time system to save the contents of registers D0 and A0, so that the command-line parameters can be accessed. Hunks 2, 3, 4, 5, and 6 are file-level storage for the various source files of Dis. Hunks 7 and 8 come from the interface libraries for dos.library and exec.library. Hunk 9 is the main code hunk. It will take a while to disassemble. The lines reading "Loop 1: 594 new labels.", etc. need some explanation. They are to show that Dis is actually working and hasn't gone away somehow. In its default mode of operation, it takes multiple passes through each text hunk, initially assuming that only the first part is code. It runs through all parts known to be code, identifying all branches, jumps, calls, etc. The branch and jump targets are identified, and the data there is taken to be code, and, if the branch is unconditional or is a jump, the code immediately after it is not assumed to be code unless some other instruction branches to it. This process is repeated for the entire hunk, until no more targets are defined. A final pass is then used to produce the actual output. See below for methods of altering this behavior. Similar special cases are made for subroutine returns of various forms. The goal behind this special handling is to accurately distinguish between actual instructions in a text hunk and non-instructions in the hunk. Other things that can be in a text hunk are constants of various types, tables of labels used for switch/case statements, etc. As you can see, symbolic information is used to provide labels, which here are procedure names, and symbolic addresses of variables. This type of information can be produced by various compilers - see the instructions for the compiler for how to do it. Often, however, fully linked object files have little or no symbolic information left in them. In these cases Dis will not be able to show symbolic labels and variable names. It will still be able to generate labels for referenced locations, however. Variable or data references (or references to code in another hunk) will be shown as a hunk-number/offset pair. Since unlinked object files contain references, by name, to symbols defined elsewhere, Dis is able to produce symbolic information in a disassembly even though explicit HUNK_SYMBOLs are not present. This type of disassembly is one of my main uses for Dis - to see what code my compiler is generating in test cases. There are many more special cases in Dis. Those interested in the gory details are invited to examine the source, which is included here. In particular, special handling is done for two types of Draco case statements, two types of Lattice C switch statements, one type of Aztec C switch statement, and a switch statement from some other compiler, which I assume is the GreenHills compiler used by Commodore. Dis can be stopped at any time by typing control-C. It may take a few seconds to stop if it is in the initial phase of processing symbolic and relocation information for a large text hunk. Those who get impatient with this phase are invited to rewrite the code to make it quicker. Perhaps a binary tree of relocation offsets would work. The culprit routines are 'insertRelocs' and 'mergeRelocs' in 'symbol.d'. Dis Options Several flags can be given to Dis to control how it works. A brief summary of them can be obtained by entering Dis -? In more detail, the flags are: -h - show all hunks. Some hunks, such as HUNK_NAME, HUNK_UNIT, HUNK_END, are of interest only to those who have to deal with object files directly, so Dis does not normally show them. Specifying '-h' will force it to do so. -x - raw dump of hunks. In this mode of operation, no disassembly or symbolic output is done. Instead, each hunk is dumped in a raw form. Data and text hunks are dumped in hexadecimal and ascii. Again, this form is probably only of use to those who need to know the details of AmigaDOS object files. -a - absolute. By default, Dis will use relative addresses on disassembly. This flag forces it to use absolute (hunk-relative) addresses. For text hunks with symbolic information, typically generated by a compiler, Dis will display position information and branch labels as relative to the most recent symbolic label, which means, for high-level languages, relative to the current function. With '-a' specified, all addresses and labels are expanded to 32 bits and are relative to the beginning of the text hunk. -s - show hex addresses. If Dis can find a symbolic value to show for an address, it will use that, rather than a hex value. '-s' forces it to show both. This can be useful if Dis is getting confused by some symbolic information, and is using it in inappropriate places. -f - force disassembly of all code. As explained previously, Dis by default only disassembles those parts of text hunks that it thinks are "reachable" from the initial code in the hunk. This method can miss parts of the code if those parts are not referenced directly. This is sometimes because a function is not being called, but can also be because parts of the code are only referenced indirectly, through a function table or jump table of some kind. Specifying flag '-f' forces Dis to treat the entire text hunk as code. This can lead to a mess, however, since most strings and constants don't disassemble very well. -n - Dis normally tries to generate labels for all branch and jump targets, and to use those labels at the branch points. Specifying '-n' tells it to not do that. Branches will then display both the hex form of the target and, as appropriate, the PC-relative form. -w - the previously mentioned technique of doing multiple scans of the code to find out what is code and what is data can be quite time consuming. Flag '-w' tells Dis to not work so hard. If '-n' is not specified, it will do a single pre-pass to identify labels, and then disassemble with what it has found by then. This will often work, but is confused by switch/case statements. If '-n' is specified along with '-w', then Dis will do no prescans, but will just disassemble in one pass. This can cut down on the time quite a bit, but can also miss a lot. Adding '-f' to the flag list gives a raw, non-labelled disassembly of the entire text hunk. -l - AmigaDOS libraries (in LIBS:) and devices (in DEVS:) have a very specific format, with a couple of header sections which point to the routines of the library/device. Flag '-l' tells Dis that it is looking at one of these. With this flag, it will try to find the special parts, and treat them separately. The code to do this works well for libraries generated by my 'fdCompile' program, but sometimes has trouble with other libraries/devices. (The difficulty seems to be that they put parts of the special stuff a long way down in the code, so that Dis doesn't find it early enough. Also, it seems that some of the Commodore-supplied ones don't follow the form layed out in the ROM Kernel Manuals.) -r - this flag tells Dis to show the relative form of branches and data references as well as a symbolic or label form. -p - by default, Dis shows a hexadecimal offset to the left of the disassembly it generates. This flag suppresses that part, which makes the output closer to something that an assembler will like. -c - by default, Dis will capitalize any non-MC68000 instructions or addressing forms that it finds. This is useful for people with compilers that can handle the MC68020/MC68881, etc. Specifying '-c' will suppress that capitalization. -q - this flag enables a quick summary mode. A single line only is printed for each hunk found in the object file. -d - a fully linked object file with no symbolic information has jumps and branches within it - to subroutines, run-time library routines, interface stubs, etc. One of the special cases in Dis is for a type of case statement in Draco which JMPs to a run-time routine to do a binary search through a sorted table. If no symbolic information is present, Dis has no way to distinguish this JMP from others. Flag '-d' tells Dis to treat all such JMPs as JMPs to those special run- time system routines. If you know that an object file is not fully linked, isn't compiled by Draco, or has symbolic information, this flag will not be needed. -k - problems similar to the above can occur in other object files with "unusual" code. Flag '-k' tells Dis to assume that the entire text hunk is code when doing prepasses to find labels. The normal checks for code v.s. data will be done when the actual disassembly is produced. The disadvantage of this flag is that spurious labels can be produced, and occasionally, data will be treated as code. Some Sample Disassemblies to Try dis dis dis -dl libs:disassemble.library dis -lkf libs:mathieeedoubtrans.library if you have enough patience, you can see the FPU instructions start at offset 0x385c dis c:format What? you don't have format copied to your C: directory so you can use it from CLI? Note the empty code and data hunks - must be a funny compiler Commodore uses. dis -q c:format dis -x c:format So does the trashcan icon look like a trashcan in hex? Expansion My original need was for a disassembler that would just nicely disassemble the object files my compiler produces for small test files. Dis has grown considerably from that initial requirement. As with any program like this, there is still much that could be done: - better heuristics for distinguishing code from data in text hunks - better handling of libraries and devices that don't follow the model that Dis likes - flags to allow some hunks to be skipped - flags to force certain regions to be treated as code or data - an interactive mode - intuition interface - flags to force output to be directly acceptable to assemblers I don't plan to do any of these things myself. Some minor changes may get done if something becomes a nuisance. Since the full source to Dis is included, however, others are invited to try their hand at some of them. I would especially be interested in improvements to the code/data distinguishing code.