Re: [RFC 00/21] drm: fb emulation: Step 3: Remove FB_KMS_HELPER config from drivers

[Daniel Vetter <daniel@ffwll.ch>]

On Tue, Jul 14, 2015 at 12:01:51PM +0530, Archit Taneja wrote:
> Hi,
> 
> On 07/13/2015 09:07 PM, Daniel Vetter wrote:
> >On Mon, Jul 13, 2015 at 11:30:34AM -0400, Alex Deucher wrote:
> >>On Mon, Jul 13, 2015 at 2:43 AM, Archit Taneja <architt@codeaurora.org> wrote:
> >>>With the top level DRM_FBDEV_EMULATION option, drivers don't need to
> >>>select DRM_FB_KMS_HELPER or other FB related config options in the
> >>>driver.
> >>>
> >>>Even if FBDEV_EMULATION isn't select, the stubbed out versions of
> >>>the drm_fb_helper functions will be called.
> >>>
> >>>There are some drivers which still need some FB stuff even after
> >>>these changes. They are qxl, udl and nouveau. These are handled a bit
> >>>differently compared to the other patches.
> >>
> >>Series is:
> >>Acked-by: Alex Deucher <alexander.deucher@amd.com>
> >
> >Archit for merging I think I'll just keep the current set of patches I
> >have for testing (well until there's a real bug discovered). Can you
> >please collect all the acks/r-bs/t-d when resending so I don't have to
> >digg them out when we merge the final versions of these?
> >
> 
> Sure. I'll keep a track of this.
> 
> Speaking of bugs, there was one thing I noticed in the series which I
> am not sure about.
> 
> One of the new helpers wraps around framebuffer_alloc(). Most drivers
> tend to pass the drm_device's dev pointer to this, and this is what I
> do in the helper too.
> 
> But some drivers(nouveau, radeon, qxl and some more) pass their
> pci_dev's dev pointer.

Hm, didn't even notice that one.

> Looking at the framebuffer code, it looks like the 'dev' we pass to
> framebuffer_alloc is what is used as the parent device when
> creating a fb device.
> 
> After this patch set, all drivers will use drm_device's dev when
> calling framebuffer_alloc. I was wondering if this could cause any
> fbcon issues?

I think it won't matter, this is just about discovering the fbdev instance
in sysfs. And all the legacy users that do care about fbdev device nodes
don't really use that anyway.

And I think conceptually it also makes more sense to have the emulated
fbdev device as a child of the kms device.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch