Hi Daniel,
thanks for your comments!
On 7/30/19 11:26 PM, Daniel Vetter wrote:
I have a couple of questions regarding this driver, thus the RFC: (1) By default, when loading the module a vt console binds to the fb. I think this is useful, but the cursor blink of the vt leads to an eternal refresh loop. Refresh on these displays is *extremely* slow (1 frame per 15s), so ideally I'd want the cursor to not blink. Is there some nice way to tell the vt/console driver to do this by default?
Hm maybe there is something, but this is the first epaper driver, so I guess even if it's exists already in fbcon, you'd need to wire through a flag from drm to drm fbdev emulation to fbdev core to fbcon that cursors should better not blink.
Thanks. I'll look into that.
(2) Is ioctl the correct interface for the driving waveform/refresh stuff?
For kms, no. In general kms is supposed to be standardized and generic, and uses properties. I think Emil already comment a bit with pointers what you should look into instead for each part. For partial updates we have the damage rectangle stuff now, but no idea whether that's good enough for what you need.
I suspect not. I looked into these properties, but as far as I understood they're tied to a mode, and modes are kind of static? If you want to display an animation on the epaper you might have to change these on a frame-per-frame basis. Can that be done with kms properties?
(3) I read that any drm driver has to be committed along with a libdrm implementation. I think the most likely interface for anyone to interact with this driver would be the fb dev. Do I have to make some userspace implementation as well anyway? If so, where would that go: libdrm or some sort of new libepaper?
Even more strict: we require full open source userspace, per https://dri.freedesktop.org/docs/drm/gpu/drm-uapi.html#open-source-userspace...
But since kms is standardized you should be able to create a driver for it with no new changes in userspace at all.
phew!
(4) The driver accepts both XRGB8888 and RGB565 (for compatibility with small LCDs). The driver's current approach to calculate a b/w/r "ternary" image from this data is to just take the MSBs of each color component, then make anything red (R>127, G,B<=127) red, anything black (R,G,B each <=127) black and anything else white. This is since the display's default state is white, and a pixel can turn either red or black from that. Note that it's the actual pixel changing color, i.e. there is no black and red sub-pixels. If you try to drive black and red at the same time, the chip just selects red for that pixel. What are your thoughts on this interface? I was thinking about using some indexed color mode, but that would limit possible future grayscale support[7].
Hm generally we're trying not to fake stuff in the kernel driver in drm. XRGB8888 is an exception because too much stuff blindly assumes it exists. I think what we want here is an epaper drm_fourcc.h code, at least long-term. But since that's new userspace api it also means we need some open source userspace to drive it. I think best approach would be to get the basic driver with XRGB8888 emulation merged first, and then figure out how to best add the specific epaper formats to fully expose the underlying capabilities.
I think I'll rip out the rgb565 part then.
The following patches all apply on v5.2. This is my first linux driver, so please be gentle but please do point out all my mistakes :) I'm aware the dt binding doc is still lacking.
So the great and also somewhat tricky bit is that support code for tiny drivers is evolving really quick, so would be good if you can rebase onto drm-misc-next from https://cgit.freedesktop.org/drm/drm-misc/ There's still tons in-flight, but that should at least help. One notable series that didn't land yet renames tindydrm to drm/tiny/, so maybe wait for that to land (it hopefully should land soon I think).
I'm aware of that and I'll rebase when that's landed.
- Jan