[Intel-gfx] Modesetting bug on 945GSE (Atom N270) with Chris Wilson drm-intel-next kernel

[Chris Wilson]

On Mon, 22 Nov 2010 10:58:48 +0000, Simon Farnsworth <simon.farnsworth at onelan.co.uk> wrote:
> On Monday 22 November 2010, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > The usual perf cpu profiling and latency charts would be interesting.
> > 
> The reason for moving to new bits is perf and sysprof telling me that I'm 
> probably out of CPU (one hyperthread fully utilised, the other always in use 
> to some extent).
> 
> What perf profiling and latency charts are interesting? I've been using sysprof 
> to try and get myself below 100% CPU, but I'd love to add more tools to my 
> toolbox.

In theory, we should be able to gather quite a bit of interesting
information for analysing deadlines from using perf timechart in
conjunction with i915:i915_gem_request_wait_begin,
i915:i915_flip_request, i915:i915_flip_complete. Just needs a lot more
effort for useful analysis.

> Argh. Noticed that I misspoke - I am using DVI, not VGA right now (enables me 
> to eliminate monitor effects when trying to spot whether missed deadlines are 
> visible to the user).
> 
> It's not cloning to 1280 pixels - it's compressing what looks like the full 
> 1680 pixels into a band about 500 pixels wide. I can try and take a photo if 
> it would help.

I think a clearer picture is forming in mind. Just a complete modesetting
screwup.

> xrandr --output VGA1 --off --output DVI1 --preferred doesn't change things.

Right looks like the other pipe is just residual values from probing for a
TV.

Mind also grabbing a drm.debug=0xe from a sane startup. Aside from a
silliness with attempting to use a HDMI command on a non-HDMI capable
connector, there was nothing outlandish in that dmesg that I could spot.
Hmm, I'll send a patch for that at least.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre