Re: [Gems-users] Using Magic Instruction on sparc64-linux

[Gene Kishinevsky <genek@xxxxxxxxxxxx>]

Thanks a lot to everyone for the responses. Replacing __sparc with 
__sparc__ in magic-instruction.h file and inserting MAGIC(4 << 16) into 
the benchmark code did the trick!
-Gene

Kevin Moore wrote:
> Just a quick followup to Dan's comment.  Simics will only break 
> execution on one specific value: (4 << 16).  Other arguments can be 
> detected by simics callbacks (see ruby/commands.C), but will not stop 
> the simulation.
> --Kevin
>
>
> On Mar 19, 2006, at 9:54 AM, Dan Gibson wrote:
>
>   
> > Gene,
> >
> > Your problem seems to be related to Simics. To my knowledge, we have
> > only used Simics with the sarek (sparc/solaris) target, and you
> > mentioned you are using bagel. Since bagel is a sparc machine, I'm
> > surprised that the __sparc symbol was undefined. You might try passing
> > -D__sparc to your compiler to force the symbol to be defined, but I am
> > not clear as to whether the magic breakpoint interface is the same for
> > bagel as for sarek.
> >
> > As I look the the magic-instruction.h file provided with Simics 2.2.19,
> > I notice that they intend to use "sethi 'n' %g0" as a magic breakpoint,
> > where n is an integer greater than zero. You might try using '1' rather
> > than '0' in your example program below and see if that works.
> >
> > In the past, I know there have been some reordering concerns with magic
> > breakpoints as well. Optimizing compilers may reorder or remove
> > instructions--try compiling without optimiziations if you have them 
> > enabled.
> > I also would suggest you have a look at the Simics documentation, 
> > and/or
> > the Simics website (www.simics.net).
> >
> > Regards,
> > Dan
> >
> > Gene Michael Kishinevsky wrote:
> >
> >     
> > > Hi all,
> > >
> > > We've tried using the magic instruction today on our simulated system 
> > > with
> > > no success. We're using the bagle system provided by Simics, which is 
> > > a sun
> > > sparc64 machine with linux on it. I used the magic-breakpoint-enable 
> > > command
> > > in Simics and then ran my simple program on the simulated machine. 
> > > However,
> > > we never got the simulation to stop. We tried changing the 
> > > instructions
> > > around and even using the magic-instruction.h library definition 
> > > instead of
> > > our own. When using the library we got an "unsupported architecture" 
> > > error
> > > from the following section of code in the magic library.
> > >
> > > #else  /* !__sparc && !__i386 && !__powerpc__ */
> > > #error "Unsupported architecture"
> > > #endif /* !__sparc && !__i386 && !__powerpc__ */
> > >
> > > I've included our test program below. If anyone has any suggestions 
> > > we'd be
> > > very grateful.
> > >
> > > Thanks a lot.
> > >
> > > -Gene
> > >
> > >
> > > #include <stdio.h>
> > > //#include "magic-instruction.h"
> > >
> > > #define MAGIC_INSTRUCTION do {    \
> > >   __asm__ __volatile__ ("sethi 0, %g0");  \
> > > } while (0)
> > >
> > > main() {
> > >  printf("Hello world.");
> > >  MAGIC_INSTRUCTION;
> > >  printf("After magic I should have stoppped!");
> > > }
> > >
> > > _______________________________________________
> > > Gems-users mailing list
> > > Gems-users@xxxxxxxxxxx
> > > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> > >
> > >
> > >
> > >       
> > _______________________________________________
> > Gems-users mailing list
> > Gems-users@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> >     
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>