09:24 <narmstrong> xdarklight: it booted on a s805x with no changes except the DT a pushed : [ 12.340274] soc soc0: Amlogic Meson GXL (Unknown) Revision 21:b (32:2) Detected

09:26 <xdarklight> narmstrong: nice - congrats :)

09:27 <xdarklight> narmstrong: does the lima driver also load (and does it render something for you)?

09:27 <narmstrong> xdarklight: did not test yet

09:29 <narmstrong> usb works

09:33 <xdarklight> at least something :)

09:34 <narmstrong> I'll need to check vdec and mali

09:35 <narmstrong> but there is no sdcard, so it's less easier to test :-)

09:39 <xdarklight> heh :)

09:39 <narmstrong> but with mainline u-boot, I can boot from USB ;-)

09:40 <xdarklight> :)

19:07 <xdarklight> narmstrong: I think it would make sense to use CCF for clkmsr now that jbrunet's patchset almost allows exposing GEN_CLK (except for wiring up the clkmsr bit in the mux)

19:08 <xdarklight> what do you think?

19:11 <jbrunet> xdarklight: we've talked about it but it going take some thinking. Not all clock in clock measure exist in ccf.

19:12 <jbrunet> Sure it is a good idea, maybe as a second step ?

19:14 <xdarklight> jbrunet: can you give me some more details what you discussed? Neil's current suggestion is basically a flat list of clock measurements in debugfs, can't we do the same with CCF (for example: we have "clk81" which is managed by CCF, and we couuld have a clkmsr_clk81 which is read-only and obtained from clkmsr)?

19:15 <xdarklight> obtained = rate and accuracy are obtained from clkmsr

19:16 <xdarklight> using debugfs would work too, I just want to understand the "why"

19:23 <jbrunet> xdarklight: there is not much more than that :) just that debugfs (as proposed - with clock that don't exist in ccf) could be useful, integration with ccf could also be nice to have. Perf counters is another idea

19:25 <jbrunet> Just thinking that ccf and perf could be done later on ... Debugfs is a good start IMHO

19:34 <jbrunet> Ohhh I understand (sorry I'm bit slow tonight) ... You are wondering why we need debugfs at all, and why not rely only on ccf ? Right ?

19:46 <xdarklight> jbrunet: yep, to me it looks like a duplication of what CCF provides - but then again, I might have missed something

19:46 <jbrunet> It is worth considering

19:48 <jbrunet> I'm bit concerned about how it will behave with ccf. Measuring a clock takes tens of uS

19:48 <jbrunet> Might be too much when ccf update the tree

19:51 <xdarklight> if that's really the case then I'm completely fine with debugfs (I don't know much about the perf framework, so I can't comment on that)

19:51 <xdarklight> that == being too slow when CCF updates it's tree

19:51 <jbrunet> Need to check it is a problem

19:51 <jbrunet> If