Re: efifb: patch to allow userspace to unbind efi framebuffer driver from VGA device

[Jonathan Campbell <jonathan@castus.tv>]

Perfect!

I removed the request_mem_region from my bochsfb driver and coded it to 
allocate an aperture struct, then register_framebuffer(). It kicks efifb 
off just as you said. Thanks!

Jonathan Campbell
> Hi Jonathan,
>
> On Sun, 18 Aug 2013 21:22:05 -0700 Jonathan Campbell wrote:
>> This patch allows userspace to direct efifb to let go of the VGA device
>> and unregister it's framebuffer. As far as I can tell, the Linux kernel
>> framebuffer console knows to let go when efifb unregisters it's
>> framebuffer. The problem I'm trying to solve is that I need efifb so the
>> kernel can show it's status on-screen during boot, but then I need efifb
>> to step aside and let a better driver load and take the VGA device later
>> during boot up.
>>
>> The custom Linux distribution I've made for myself uses a userspace
>> program to "boot" secondary VGA devices and load both fbdev and drm/kms
>> drivers to bring video online. When I wrote this, I needed the ability
>> for efifb to let go so that it could load bochsfb to better use the VGA
>> output of VirtualBox and bochs. Without it, my console is stuck at
>> whatever video mode the UEFI BIOS happened to leave it at and the more
>> specialized driver cannot acquire the resources it needs to do it's work.
>>
>> The patch adds a sysfs device file "bind_fb" to the
>> /sys/bus/platform/devices/efifb.0 filesystem tree. Writing "1" causes it
>> to re-register the framebuffer, "0" to unregister. Checks are in place
>> to not register the framebuffer if already registered, etc.
> That is the wrong approach.
>
> On fbdev subsystem there is infrastructure for replacing generic
> firmware drivers with specialized drivers (as is called by KMS
> drivers).
>
> Have a look at remove_conflicting_framebuffers() and its use by KMS
> drivers (i915, radeon, nouveau, mgag200, cirrus).
>
> Unloading efifb should be able to happen automatically when
> loading/probing new driver, without userspace help (so it can work when
> both are built-in).
>
> Bruno
>
>
>> I realize it's not perfect and I wanted to know what I could do to
>> improve it and eventually make it to mainline.
>>
>> On a related issue, when will efifb make use of the Graphics Output
>> Protocol through UEFI to offer modesetting? Is that possible? If no
>> specialized drivers are available it'd be nice to at least have basic
>> modesetting as needed.