Hi,

 

A few answers / thoughts, below (marked with RMo).

 

Thanks for all the help and suggestions!

 

... Russell


On Wed, Jan 5, 2011 02:23  PM, Khem Raj <[email protected]> wrote:


> 
On Wed, Jan 5, 2011 at 10:49 AM, Russell Morris
> <[email protected]> wrote:
> > Hi,
> >
> >
> >
> > Let me try to answer a few questions in one email ... :-). First of all, I 
> > tried the patch - unfortunately no joy. It does the same thing as earlier 
> > builds - let me try to explain, which will hopefully also answer the 
> > questions below.
> >
> >
> >
> > I applied the patch, and rebuilt from scratch with the minimal distro 
> > (deleted the TMPDIR completely before building). I built the 
> > helloworld-image, to get a statically linked executable, and also because 
> > it's a pretty small (=faster) build.
> 
> OK thats bad. Now can you recompile the kernel with user debugging
> enabled ? and reboot then it will dump lot more info on console on
> error you need to turn on CONFIG_DEBUG_USER in .config
 
[RMo] Sorry, a dumb question here - but how do I do this? I can see .config in 
the temp directory - is this where you want me to modify it?
> 
> >
> >
> >
> > I then looked at the helloworld executable, and a few interesting notes,
> >
> > - if I readelf -h helloworld, it reports "Version5 EABI" ... so I assume 
> > arm5te still?
> 
> thats EABI version it has nothing to do with ARM architecture versions
 
[RMo] OK, thanks!
> 
> >
> > - if I try to run helloworld using qemu-arm, it runs fine ... with no cpu 
> > selected (but I did some checking, and the default cpu for qemu-arm is the 
> > arm5te). If I try to run with a -cpu arm920t option I get the error message 
> > "qemu: uncaught target signal 4 (Illegal instruction) - core dumped"
> >
> 
> OK good so it seems there is still some intructions generated which
> are not supported in armv4t
 
[RMo] That's what it seems like. To confirm - what is the best way to test this 
... with qemu-arm, and/or on the target? Just trying to make sure I test it in 
a way that makes sense!
> 
> > - I was not able to run this on the target right now, as I'm not near it 
> > ... but when I did before I either got a core dump (illegal instruction), 
> > or it said basically that the file was not found (depending on the 
> > executable I tried to run).
> 
> Yes it wont change I think.
 
[RMo] Definitely agreed.
> 
> >
> >
> >
> > One more interesting fact - if I go inside TMPDIR, and then inside 
> > work/armv4t-oe-linux-gnueabi/gcc-cross-4.5-r28.0+svnr167948/gcc-4_5-branch/testsuite/gcc.target/arm,
> >  there is some sort of test file, with a filename of pr42235.c. Oddly 
> > enough the first line in this file says ... /* { dg-options "-mthumb -O2 
> > -march=armv5te" }  */
> 
> 
> thats just a gcc dejaGNU regression testcase it does not mean anything
> for compiling the root file system
 
[RMo] Ok, thanks!
> >
> >
> > Hopefully this all makes sense. I think this says that the executable is 
> > still targeting an armv5te ... but I could be wrong! Unfortunately it 
> > wouldn't be the first time I was off base, and certaintly it won't be the 
> > last ... :-(.
> >
> >
> >
> > Thanks for all your help!
> 
> as koen suggested try it with angstrom-2008 and see if that helps too.
 
[RMo] Absolutely - started already. I thought you were looking for the minimal 
distro, but I may have misunderstood. In any case, trying this now ... :-).
> 
> >
> >
> >
> > ... Russell
> >
> >
> >
> >
> >
> >
> > On Wed, Jan 5, 2011 11:45  AM, Khem Raj <[email protected]> wrote:
> >
> >
> >>
> > On Wed, Jan 5, 2011 at 7:11 AM, Phil Blundell <[email protected]> wrote:
> >> > On Wed, 2011-01-05 at 08:48 -0600, Russell Morris wrote:
> >> >> Just to confirm - have you run these on an armv4t target? Only asking 
> >> >> because my build completes fine, but the executables don't seem to run 
> >> >> on the target.
> >> >
> >> > What exactly happens when you try to run those executables?  Have you
> >> > inspected them to see if they look like the right kind of thing, and/or
> >> > compared them to working ones?
> >> >
> >> > p.
> >> >
> >> >
> >>
> >>
> >> yes as Phil asked you should try to localize the offending code in the
> >> faulty binary. So try to enable
> >> kernel debugging messages so it tells you where its faulting.
> >> Secondly if you can take a working system
> >> and see if the new binary faults in same way ? if not then link the
> >> binary statically and run it again on working
> >> system and see if it faults again. If it does then you can debug it
> >> >
> >> > _______________________________________________
> >> > Openembedded-devel mailing list
> >> > [email protected]
> >> > http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
> >> >
> >>
> >> _______________________________________________
> >> Openembedded-devel mailing list
> >> [email protected]
> >> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
> >>
> > _______________________________________________
> > Openembedded-devel mailing list
> > [email protected]
> > http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
> >
> 
> _______________________________________________
> Openembedded-devel mailing list
> [email protected]
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
> 
_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Reply via email to