Re: [RFC 00/50] staging: vchiq: Getting rid of the vchi/vchiq split

From: Greg KH
Date: Wed May 27 2020 - 08:09:13 EST

On Wed, May 27, 2020 at 01:53:05PM +0200, Nicolas Saenz Julienne wrote:
> vchi acts as a mid layer between vchiq and its kernel services, while
> arguably providing little to no benefit: half of the functions exposed
> are a 1:1 copy of vchiq's, and the rest provide some functionality which
> can be easly integrated into vchiq without all the churn. Moreover it
> has been found in the past as a blockage to further fixes in vchiq as
> every change needed its vchi counterpart, if even possible.
> Hence this series, which merges all vchi functionality into vchiq and
> provies a simpler and more concise API to services.
> I'm aware that kernel's vchi API tries to mimic its userspace
> counterpart (or vice versa). Obviously this breaks the parity, but I
> don't think it's a sane goal to have. There is little sense or gain from
> it, and adds impossible constraints to upstreaming the driver.
> Overall the series falls short of removing 1500 lines of code, which is
> pretty neat on itself.
> So far it has been tested trough bcm2835-camera, audio and vchiq-test. I
> can't do much about vc-sm-cma for now, but the changes are done in a way
> that shouldn't affect its behaviour.
> Note that the series builds up on RPi/Laurent's camera support series[1]
> and can't yet be merged. We'd have to coordinate here. We could either
> wait for the vc_sm_cma rework (if it's not going to take months and
> months to finish), or factor out all the vc-sm-cma stuff, merge that into
> the downstream kernel and take the rest of the series on top of
> Laurent's mmal-vchiq changes.

These look great, but due to the dependancy, I can't take them all.

But I can take the first 10, so I did that, they are now queued up,

greg k-h