Re: [PATCH] Speedup SMP kernel on UP box
Daniel Jacobowitz (dan@debian.org)
Sat, 16 Mar 2002 18:54:43 -0500
On Sat, Mar 16, 2002 at 03:37:26PM -0800, Paul Menage wrote:
> In article <0C01A29FBAE24448A792F5C68F5EA47D238DE0@nasdaq.ms.ensim.com>,
> you write:
> >@@ -9,9 +9,15 @@
> > */
> >
> > #ifdef CONFIG_SMP
> >-#define LOCK "lock ; "
> >+#define LOCK "\n1:\tlock ; "
> >+#define LOCK_ADDR "\n" \
> >+ ".section .lock.init,\"a\"\n\t" \
> >+ ".align 4\n\t" \
> >+ ".long 1b\n" \
> >+ ".previous\n"
>
>
> Why not do:
>
> #define LOCK "1: lock ; \n" \
> ".section .lock.init,\"a\"\n" \
> ".align 4\n"\
> ".long 1b\n"\
> ".previous\n"
>
> Then you don't need the LOCK_ADDR macro, so most of atomic.h can be
> left as is. The assembler doesn't seem to care that there's a section
> change between the lock prefix and the instruction that it's locking.
Local labels work forwards as well as backwards:
#define LOCK
".section .lock.init,\"a\"\n" \
".align 4\n"\
".long 1f\n"\
".previous\n"
"1: lock ; \n"
I recommend not using "1" for your label, though, because probably
other bits of code that use locks have local loops in them. It's to be
expected... just pick a less common number.
--
Daniel Jacobowitz Carnegie Mellon University
MontaVista Software Debian GNU/Linux Developer
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/