gprof's broken in 7-CURRENT

[Luoqi Chen]

> And to think that I mentioned this problem 5-6 months ago with gcc 4.2.x 
> and now it finally gets publicity...

Did I just hear someone volunteered to be a guinea pig?

Garrett, would you like to try out my fix? It's actually quite simple,

Index: profile.h
===================================================================
RCS file: /home/ncvs/src/sys/i386/include/profile.h,v
retrieving revision 1.42
diff -u -r1.42 profile.h
--- profile.h	28 Oct 2006 11:03:03 -0000	1.42
+++ profile.h	29 Nov 2007 20:07:15 -0000
@@ -115,7 +115,17 @@
 void									\
 mcount()								\
 {									\
-	uintfptr_t selfpc, frompc;					\
+	uintfptr_t selfpc, frompc, ecx;					\
+	/*								\
+	 * In gcc 4.2, ecx could be used in the caller as arg pointer	\
+	 * if stack realignment option is set (-mstackrealign) or	\
+	 * if the caller has the force_align_arg_pointer atrribute	\
+	 * (stack realignment is ALWAYS on for main).  Preserve ecx	\
+	 * here.  Use volatile to prevent this statement from being	\
+	 * rescheduled with the next two, just in case the compiler	\
+	 * decides to use ecx for either selfpc or frompc.		\
+	 */								\
+	__asm __volatile("movl %%ecx,%0" : "=m" (ecx));			\
 	/*								\
 	 * Find the return address for mcount,				\
 	 * and the return address for mcount's caller.			\
@@ -132,6 +142,7 @@
 	__asm("movl (%%ebp),%0" : "=r" (frompc));			\
 	frompc = ((uintfptr_t *)frompc)[1];				\
 	_mcount(frompc, selfpc);					\
+	__asm __volatile("movl %0,%%ecx" : : "m" (ecx));		\
 }
 #else /* !__GNUCLIKE_ASM */
 #define	MCOUNT

This file is in the kernel tree, but the change only affects libc build.
If you don't want to redo a world build, you may copy the patched profile.h
to /usr/include/machine and "make all install" inside /usr/src/lib/libc.
Then re-link -pg compiled code and try it out.

Thanks
-lq