[U-Boot] [linux-sunxi] Re: [PATCH] video: add cfb console driver for sunxi

["Henrik Nordström" <henrik@henriknordstrom.net>]

m?n 2014-08-04 klockan 18:53 +0200 skrev Michal Suchanek:

> Then you have to define what is compatible.

Yes, and the part I responded to was odd unexpected combinations of
u-boot only partially providing the right information (missing in DT) or
kernel not having enabled support.

Likewise there is the opposite situation which is also in the
unsupported area, simplefb nodes in the DT but framebuffer not actually
enabled by u-boot.

Exact details on what needs to be provided in DT and how for glitch free
handover from u-boot to KMS is a bit premature to dive into now.

> eg. you can define that if you do have simplefb in u-boot you should
> have it in kernel.

Maybe. I am only saying that there is no need to try to cover all corner
cases and odd/partial combinations. It's sufficient to cover two cases:
all enabled, all disabled. In both u-boot, kernel and DT. (well, enabled
in kernel but not in DT should be the same as disabled in kernel, so
three cases).

> If you do not have simplefb in kernel the fb memory may get reused and
> the clocks should get disabled as unused, eventually. Unless you set
> the debug option not to disable them. And KMS might eventually take
> over reprogramming or re-enabling the clocks again if you have it.

Memory would be lost today if the region is reserved in DT, but may
eventually be reclaimed when support is available in kernel. And user
should expect strange display effects as clock sources and eventually
memory gets reprogrammed and/or repurposed without anything shutting off
the disaplay engine. I do not see a problem here.

My preferred solution to this problem is to have u-boot shut off (or not
enable) the display engine before starting the kernel, and to provide a
DT without simplefb nodes. Don't worry too much about this odd case in
the kernel.

> Or as pointed out earlier you might find this unacceptable and even
> with simplefb compiled out the KMS driver would need enough simplefb
> logic to analyze the simplefb DT node and reuse the clocks and memory
> without reprogramming.

Maybe, but glich free handover is not a primary or even high priority
goal, or even what I really commented on.

It's something that can be looked into when KMS is functional. And
missing support should cause no more than a temporary display glitch
until KMS have fully claimed and programed the display components.
Likely on the same scale as KMS selecting another display mode.

Regards
Henrik