[RFC] AST2500 video engine driver design

[Andrew Jeffery]

On Sat, 19 May 2018, at 01:12, Eddie James wrote:
> Hi,
> 
> I'm working on a driver for the AST2500 video engine; just wanted to 
> post up the plan in case anyone has any ideas. The purpose behind this 
> development is to be able to use IKVM with video from the host and i/o 
> from a remote user.
> 
> The video engine is a piece of hardware in the AST24/2500 that captures 
> and/or compresses video data. Typically the input is the host graphics, 
> writing it's RGB output to the VGA memory on the BMC. The video engine 
> will capture and compress the data and store it in memory. The driver 
> should provide an interface for an application to read that compressed 
> video in order to get it on the network.
> 
> The driver interface will use the Video4Linux2 API. This involves 
> standard file operations like open, read, and close, as well as a video 
> ioctl interface for configuring the device. Configuration options will 
> probably be limited, as I think there will only be one mode that will 
> provide enough performance for IKVM. I do want to be able to configure 
> the video source; either direct from host at the BMC VGA memory or from 
> the BMC DRM driver.
> 
> The most substantial part of the driver will just be the hardware setup 
> and manipulation. Intel has done some work here already (thanks Ed) 
> which has been useful for configuring the hardware and doing the 
> capture/compress loop. I'll be following what I think is fairly common 
> for video drivers and only initializing the engine once a file 
> descriptor is opened, and then allow more file descriptors to be opened 
> as necessary without further touching the engine, though changing 
> configs will obviously do the necessary engine operations. Once the last 
> file descriptor is released, the engine and clocks will be shut down. 
> The read operation will trigger a single frame capture/compress for now.
> 
> I still have details to hash out regarding streaming, user-space 
> buffers, etc, which will probably be necessary to get the performance we 
> need.
> 
> Any comments/questions/concerns?
> 

Sounds reasonable to me, but I don't have much experience in V4L land. Keep us posted on your progress, and pipe up if you are looking for opinions on the driver design.

Cheers,

Andrew