XOrg weirdness

[Adam Jackson]

On Tue, 2007-02-27 at 21:59 +0000, Richard Hughes wrote:

> Well, I get this:
> 
> CPU: Core Solo / Duo, speed 1662.57 MHz (estimated)
> Counted CPU_CLK_UNHALTED events (Unhalted clock cycles) with a unit mask
> of 0x00 (Unhalted core cycles) count 860000
> CPU_CLK_UNHALT...|
>   samples|      %|
> ------------------
>     12704 82.6330 /usr/bin/Xorg
>         CPU_CLK_UNHALT...|
>           samples|      %|
>         ------------------
>             12066 94.9780 /usr/lib/xorg/modules/libfb.so

That's pretty well expected.  If anything's going to be burning CPU time
it's either fb or GLcore.  (Or the driver itself when it sits in
infinite loops waiting for the command queue to catch up, but that case
usually means wedged hardware.)  More interesting is knowing which
functions it's hitting, if you can get that.

Would also like to know whether the same slowdown happens when using
Option "ShadowFB".  If not then we can assume the slowdown is being
caused by a software fallback to something that happens to be in card
memory, and there might be a reasonable heuristic for evicting things
from card memory when that happens, or just a smarter way to write that
routine that hits the bus less.  But if it's still slow when shadowfb'd
then either it's some particularly naive code, or g-t is just asking us
to do a lot of really painful things.

> But also I get this (!!!):
> 
> oprofile: using NMI interrupt.
> 
> =================================
> [ INFO: inconsistent lock state ]
> 2.6.20-1.2949.fc7 #1

Yow!

- ajax