[vfio-users] Framebuffer intercept: is it possible?

[Alex Williamson]

On Tue, Sep 15, 2015 at 2:15 PM, Wynter Woods <zerotri at gmail.com> wrote:

> I've been wondering about this a lot lately. Possibly by rewriting
> nouveau/any other gpu driver to support a framebuffer/read only state? I
> realize there would end up being a few differences between how Windows
> handles state and how we handle state on Linux but it shouldn't be so
> vastly different that it wouldn't be usable.
>

Just a thought, but by disabling mmap in QEMU (x-mmap=off, soon to be
x-no-mmap=true) and enabling vfio tracing in QEMU
(qemu.git:docs/tracing.txt), you can actually see exactly what gets written
to the device and it slows things down enough that you can actually see the
framebuffer being written so you can tell where it is.  You might then be
able to look back through the trace and see if the address of the
framebuffer is configurable or if it ever changes or maybe what it's size
is.  Referencing the nouveau or radeon drivers might help make sense of
things.  It would be interesting to see if a reliable screen scrape could
be done from the host.  It might open some interesting options.

Disabling mmap and making my eyes bleed on traces is how we got the quirks
for AMD and Nvidia.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20150915/17834e90/attachment.htm>