Unbreak high resolution profiling a little: use dummy asms to prevent

timing loops being optimized away.

Once apon a time, gcc promised not to optimize away timing loops, but
gcc started optimizing away the call to a null function in the timing
loop here some time between gcc-3.3.3 and gcc-3.4.6, and it started
optimizing away the timing loop itself some time between gcc-3.4.6
and gcc-4.2.

1 files changed, 6 insertions, 1 deletions

diff --git a/sys/libkern/mcount.c b/sys/libkern/mcount.c
index d967fe8..e26f6fa 100644
--- a/sys/libkern/mcount.c
+++ b/sys/libkern/mcount.c
@@ -268,18 +268,23 @@ mexitcount(selfpc)
 	}
 }
 
+#ifndef __GNUCLIKE_ASM
+#error "This file uses null asms to prevent timing loops being optimized away."
+#endif
+
 void
 empty_loop()
 {
 	int i;
 
 	for (i = 0; i < CALIB_SCALE; i++)
-		;
+		__asm __volatile("");
 }
 
 void
 nullfunc()
 {
+	__asm __volatile("");
 }
 
 void