This is the mail archive of the crossgcc@cygnus.com mailing list for the crossgcc project.
Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
---|---|---|
Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |
Glad to hear you are making progress! On Tue, 9 Sep 1997, Josef Wolf wrote: > > The initialised data appears to works for me, try adding -Wl,-Map,link.map > > and checking what section your foo variable is being put in. Then check > > the handling of that section in your startup code. > > I tried the same with coff and it worked well with my startup-code. It > looks like the ieee format does some strange things to pack and/or > compress initialized data, so just copying it from ROM to RAM won't work :-( > > Where could I find a description of the ieee-coding-format? Or (even > better ;-) a piece of code to unpack the ieee-format into the real > data-section? I have not come across this problem, when I use XRAY to download it creates a working image in memory. Are you actually storing the IEEE in flash or something? We just use objcopy to generate an S record file and use that for programing any flash/eproms (well actually we generally download to target RAM and then copy to FLASH, because we used to have overrun problems when trying to program via an un-flow controlled serial link:-). There are docs about the IEEE format on ftp.cygnus.com, I think they are under the embedded dir (but they might not be!). > > > If you need ieee695 files with debug info, you have probably already > > discovered it is not possible to directly produce them from the linker. > > You will have to compile and link with -g using a different object file > > format (m68k-coff with the default coff dbg format worked for me). Then > > run objcopy WITH the --debugging option (this appears to be > > equivalent to the "try and work flag" :-)). > > Ough! Strange, but it works this way! (at least "strings x.iee|grep '\.c$'" > gives me what I'd expect. Not tried the debugger yet) > > > I managed to get it mostly working with XRAY (2.3?), > > but I had to modify the compiler build to put constants in their own > > section (email me if you need to try this, I have a patch). > > Oups! I have no clue whether I need it because I don't understand this > sentence :-() Do you mean that XRAY need some magic values in some magic > sections to work properly? Hmm, I remember something like 'simulated_input', > 'simulated_output' and 'tags' in the MRI-Compiler with some coments about > XRAY around them. Maybe this is what you mean? My version of XRAY appeared to get upset because it saw something like:- int fred(void) { } const int initialised_var=42; int bill(void) { } It refused to load the resulting object file, unless I striped debug symbols.... The error message was something about overlapping sections. Not completely sure what was causing the error, but changing the compiler to put initialised data in a different section (instead of code) solved the problem. The problem might have been the type settings on the text section, but I didn't see the error unless text and initialised data was interleaved. I also had problems with other code (ie pSOS) being in sections with silly type settings (I can't remember if I fixed them with funny objcopy args or by changing the source). In case anyone is interested, I have attached my old COFF notes and patches. The patches do the following:- binutils - Allows gas to assemble something like "tst.l a5" on a cpu32 (-mcpu32). I needed this because I changed gcc to pass -mcpu32 to the assembler and -m68020 -mnobitfields to cc1, and gas didn't believe the chip could handle the resulting opcode. Cygnus patch - you know about this one already, in particular gcc v printf problem etc lib - Just reduces the number of libs which are built. In particular the resulting gcc will also reduce the number of libs built by newlib reducing its build time dramatically - of course you need to use this from day 1 or do a spring clean to make REAL SURE no one uses an out of date lib (you no longer build). coff - Change gcc to pass -mcpu32 to gas and -m68020 -mnobitfields to cc1 if you use -mcpu32. Also adds some cpp -Ds. coff2b - Change type info on .ctors and .dtors (to rd from x) and add .const (rd) section for initialised data. I am currently using an ELF/DWARF setup, because the SingleStep debugger from SDS would not read my IEEE files (though XRAY mainly worked, it did not understand the type info on some structures). I could post my patches for that, but they would probably not help you because DWARF appears to be only understood by gcc (not bfd/objcopy) so you can not convert it to ieee. > > > > - When using the standard-library (newlib) the linker always catches the > > > 68020-version of the library. This leads to problems with my 68332 :-() > > > This problem appears when I use m68k-coff-ld directly as well as when > > > I use it through m68k-coff-gcc. The --verbose output seems to be OK, so > > > I assume the libraries are not properly built? Here's the output: > > This problem disappeared somehow after a complete recompilation of the > tool-chain. > > Thank you very much! > Good luck Mark