[PATCH 3/3] drm/vkms: fbdev emulation support

[Daniel Vetter]

On Tue, Oct 13, 2020 at 01:19:38PM +0300, Pekka Paalanen wrote:
> On Tue, 13 Oct 2020 09:53:44 +0200
> Daniel Vetter <daniel at ffwll.ch> wrote:
> 
> > On Tue, Oct 13, 2020 at 8:14 AM Pekka Paalanen <ppaalanen at gmail.com> wrote:
> > >
> > > On Mon, 12 Oct 2020 16:23:35 +0200
> > > Daniel Vetter <daniel at ffwll.ch> wrote:
> > >  
> > > > On Mon, Oct 12, 2020 at 02:40:58PM +0200, Neil Armstrong wrote:  
> > > > > Hi,
> > > > >  
> > >
> > > ...
> > >  
> > > > > It's weird because it the kernel is misconfigured and no console is specified on the cmdline
> > > > > this console could become the main console...
> > > > >
> > > > > It's a great feature, but couldn't this be a module parameter ?  
> > > >
> > > > If you have vkms enabled in a distro, you're doing it wrong.  
> > >
> > > That's really not a great position to take. I would prefer that
> > > if a random contributor writes a Weston patch and runs 'meson test', it
> > > will use VKMS to run Weston's DRM-backend tests on his machine
> > > automatically, maybe save for some seat and device node access
> > > permissions bits which distributions could be delivering as well.
> > >
> > > Just put the VKMS device node into a non-default seat, and Xorg etc.
> > > will happily ignore it.
> > >
> > > For the fbdev device node, I don't know. Maybe a module parameter
> > > really is a good choice there, defaulting to off. I have no interest in
> > > testing anything against fbdev, but other people might disagree of
> > > course.
> > >
> > > Why? Gitlab CI is still not running tests for every commit, just per
> > > MR, and it might even be infeasible too.
> > >
> > > I am also hoping for a future where I don't have to build my own kernel
> > > just to be able to run Weston DRM tests with VKMS. That means I want to
> > > be able to run my machine with VKMS loaded and active at all times,
> > > without affecting the normal desktop. I already have such a setup with
> > > an extra AMD card, but you can't run most KMS tests against real
> > > hardware drivers.  
> > 
> > I just realized that building vkms is no problem, since it doesn't
> > auto-load. And if our Grand Plans with configurability come true, then
> > your test-harness will want to do that loading and setup itself
> > anyway. With that there also shouldn't be any problems with fbcon,
> > since presumably you already have that bound to the real gpu.
> > 
> > So I think we're all fine here, for everyone.
> > 
> > Now if you built-in vkms, that's a different thing. And for that I
> > really think a "don't do that" is the right choice.
> 
> Very good.
> 
> My remaining wish is that VKMS would be fully configurable and usable
> by an ordinary user, but I suppose that should be solved with a
> privileged userspace daemon somewhat similar to logind that hands out
> VKMS "sessions" somehow.
> 
> Not sure configfs is the best choice for VKMS configuration, unless
> maybe unprivileged userspace could ask for a VKMS instance with its own
> configfs tree it can access without CAP_ADMIN...

My idea for testing would be to build a gitlab docker image and run that
under usermode linux, with vkms built-in. Given how few people work on
this, I don't think you get a fancy logind configuration session thing for
it anytime soon.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch