> On Wed, 19 May 1999, Martin Lucina wrote:
>
> > Yes, I've seen it too. If you look at the comments in the front of
> > tgafb.c, it's listed as a problem ("Some redraws can stall the
> > kernel...") but I don't know what causes it.
A bit more info on this: magic sysrq'ing the PC when it's happening shows
that it's memmove, for which there is no arch/alpha/lib routine, so it
does it byte at a time. Not the most efficient on an architecture that
can't do byte addressing.. :-P
It's being called from fbcon_cfb8_bmove. (Incidentally, can you shed any
light on the reason that fbcon uses 'mymemmove' instead of just 'memmove'?
There's a comment in <video/fbcon.h> about it.)
I'm recompiling now with an 'optimised' (read 'hacked')
lib/string.c:memmove to see if it makes things any better. For the
perfect-alignment case, we're doing 12 times as many memory accesses as we
need to. :-(
*reboots*
Yup, it's about 12 times faster now. ;-)
So someone just needs to write lib/arch/alpha/memmove.S. Cc'd to
linux-kernel in case someone has time on their hands.
Tim.
*/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/