Re: More LTTng patches

* Sergei Shtylyov ([email protected]) wrote:
> Hello.
> 
> Mathieu Desnoyers wrote:
> 
> >>Source: MontaVista Software, Inc.
> >>MR: 20036
> >>Type: Defect Fix
> >>Disposition: needs submitting to LTTng community
> >>Signed-off-by: Mark A. Greer <[email protected]>
> >>Description:
> >>   Apparently, gcc 4.2 has tighter constrainst for the 'i' asm operands
> >>   than previous gcc versions.  This patch changes a 'i' to a 'g' which
> >>   is a more relaxed constraint.
> 
> >>Index: linux-2.6.18/include/asm-powerpc/marker.h
> >>===================================================================
> >>--- linux-2.6.18.orig/include/asm-powerpc/marker.h
> >>+++ linux-2.6.18/include/asm-powerpc/marker.h
> >>@@ -40,7 +40,7 @@ struct __mark_marker {
> >>					"0:\n\t" \
> >>					"li %0,0;\n\t" \
> >>				: "=r" (condition) \
> >>-				: "i" (&__mark_c_##name)); \
> >>+				: "g" (&__mark_c_##name)); \
> >>		__mark_check_format(format, ## args); \
> >>		if (unlikely(condition)) { \
> >>			preempt_disable(); \
> 
> >I will also change it for asm-generic and asm-i386. I hope to find some
> >consistency in the way asm operands are treated across architectures.
> 
>    It looks like that issue got fixed in gcc 4.2:
> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27528
> 
>    The bad part is that when compiled with fixed gcc, the kernel receives 
> SIGSEGV and dies on bootup... "i" works again now. :-/
> 

Aargh...

I will commit a revert in the next release :

#       modified: include/asm-generic/marker.h
#       modified: include/asm-powerpc/marker.h

Will take a "i" operand.

asm-i386 takes a "m" operand (and uses it).

asm-ppc == asm-powerpc for this matter.

other architectures are currently using asm-generic, with the "i" operand. It is
never used inside the assembly.

Mathieu


OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68