Ping? I'm attaching a slightly different patch after having fixed
a small style issue.

I can now run the gdb-testsuite without crashing my sparc-solaris
machine; not sure what have changed in the meantime. In any case, I
had no new regressions and one new PASS (stack-checking.exp).

OK to apply?


Jerome Guitton (guitton@adacore.com):

> This patch should fix gdb.base/stack-checking.exp on sparc-solaris.
> GCC generates some new sequences for stack checking, we just need to
> recognize them. Compare:
> 
>          sethi <some immediate>,%g1
>          sub %sp, %g1, %g1
>          sethi  <some immediate>, %g4
>          sub  %g1, %g4, %g4
>          cmp  %g1, %g4
>          be  <disp>
>          add  %g1, -<some immediate>, %g1
>          ba  <disp>
>          st  [%g0, [%g1]]
>          clr [%g4 - some immediate]
> 
> with this new sequence:
> 
>          sethi <some immediate>,%g1
>          sethi  <some immediate>, %g4
>          sub %sp, %g1, %g1
>          sub  %g1, %g4, %g4
>          cmp  %g1, %g4
>          be  <disp>
>          add  %g1, -<some immediate>, %g1
>          ba  <disp>
>          st  [%g0, [%g1 - 0]]
>          [clr [%g4 - some immediate]]
> 
> Two differences:
> * first, the probing loop has been slightly reordered;
> * second, to clear a memory location, GCC may now generate
> "st %g0, [%g1+0]" instead of "st %g0, [%g1]". The same effect,
> but a different instruction format.
> 
> Also, the last probe may not be generated if the probe loop leaves no
> remainder.
> 
> I cannot run the testsuite on sparc solaris, as it crashes the machine
> pretty badly. We ought to identify the test that causes this crash.
> I could run stack-checking.exp and check that it fixes one regression
> though. And I tested against AdaCore's testsuite, which
> is already a fairly good insurance that this patch is safe.
> OK to apply?