[Libdlo] defio mmap issues

[Bernie Thompson]

On Mon, Jan 11, 2010 at 6:57 AM, <justin at dynam.ac> wrote:

> On a side note, and maybe for a latter email, I'm also getting horizontal
> artifacts on the screen after several hours of operations. Reading the
> archives, I saw someone else post on this, and there was a thought there
> was a race condition on the memmap. This is very reproducible for me
> (takes about 1 hour though) so if I can provide further info to help debug
> that, I'm willing to take on the challenge!
>

Here are some datapoints from trying to track down this issue:

* The problem happens both with the latest udlfb from git.plugable.com, and
with jaya's original defio driver patch posted a few months back. So it
could be a problem carried forward, but it doesn't look like a new problem
..
* It can be seen more frequently with faster processors and larger screens.
On my atom with 1920x1080, it's intermittent and small corruption.  On a
faster Celeron with 3 1680x1050 screens, it starts happening quickly with
any significant screen activity.
* And a fun one: on that multi-screen setup, once you get large areas of the
one screen (one DL device) to fail to render ... then moving the mouse and
triggering rendering on the *other* DL device/screen, for the corresponding
lines on the screen, refreshes the first device/screen properly (!)  Since
udlfb keeps no shared state between different DisplayLink devices,
apparently this confusion is happening at the defio/vma/page table level.

Defio is used by several other framebuffer drivers, but perhaps not at these
resolutions and not for multiple screens.  Has anyone seen other evidence
that would support/contradict this being a general defio problem?  Or
perhaps something simple/stupid that udlfb is doing to trigger it?

Thanks,
Bernie
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freedesktop.org/archives/libdlo/attachments/20100113/cb46bd5e/attachment.html