<marex>
lynxeye: hey, are there any updates on your battle against TFA clock/power-domain control on MX8MM ?
<marex>
lynxeye: did GPCv2 win ?
<lynxeye>
marex: Not yet, but I'm actually currently working on finishing up a GPCv2 implementation. I guess having a working implementation might help to prove my point. ;)
<lynxeye>
BTW: It actually possible to independently gate 2D and 3D GPU on the 8MM. It's just the TF-A implementation mashing both cores together in a single domain.
<marex>
lynxeye: isnt there already an implementation for gpcv2 for mx8mq ?
<lynxeye>
marex: yes, I've upstreamed this as one of the first platform support things, because PCIe and GPU support needed it for correct oepration
<lynxeye>
However it is true that the 8MM is slightly more complex due to the internal structuring into subsystems and some of the control stuff being located inside the power domain, so it gets gated together with the peripherals
<marex>
lynxeye: can you push the mx8mm stuff somewhere, so I dont have to write it in parallel myself?
<marex>
lynxeye: lovely
<lynxeye>
but after seeing the actual implementation that's no reason to hide this stuff in TF-A
<marex>
lynxeye: I doubt there is ever a reason to hide it in TFA, it can only cause trouble
<lynxeye>
I need to give it some more testing and cleanups, but the plan is to send it out to the public ML this week.
<marex>
lynxeye: please CC me then
<lynxeye>
will do
<marex>
thanks
<marex>
lynxeye: do you need to rip anything out of the TFA ?
<marex>
lynxeye: I guess if you dont do the PSCI calls, its reasonably safe to keep the gpc code in there, although ...
<marex>
I dont particularly trust the TFA in general
<lynxeye>
yep, you can simply ignore the TF-A stuff
<marex>
lynxeye: well, without the GPC, what else is left in there ? DDR/clock init I guess
<lynxeye>
DDR reclocking and CPU power management stuff
<marex>
lynxeye: all of the useful stuff in realtime deployments :)
<lynxeye>
marex: at least it doesn't do things behind your back as with x86. If you don't want CPU power management or DDR reclocking in your realtime system, just don't call into TF-A services from the kernel.
<marex>
lynxeye: Perfection is achieved not when there is nothing left to add, but when there is nothing left to take away. - Antoine de Saint-Exupery.
<marex>
lynxeye: in other words, removing TFA altogether would be the better option in such case, and I have a feeling it won't even be such an undertaking anyway