<loki666>
I don't know what are the requried/recommended voltages, but with these changes it doesn't seems to crash (yet)
<apritzel>
interesting, so they bumped the lower OPPs by 50mV, which is also where you saw problems, right (lower frequencies)?
<loki666>
yes, until new I was only using the lowest and highest freq
<loki666>
are there any other h616 soc using the speed-5 bins, or is it only for H700 (I don't understand this part, just know I need to modify speed5 when 5bit is on in supported-hw)
<apritzel>
those bins do not seem to be measured or tested bins, but actually more the packages. So 0x6c00 means H700, and there is just one set of OPPs for all H700 chips
<loki666>
and Anbernics handhelds are then only H700 we know of
<apritzel>
and this particular OPP set came from some downstream BSP kernel, IIUC?
<apritzel>
yeah, so effectively bin bit 5 means "all Anbernic handhelds"
<loki666>
so could this be upstreamed?
<apritzel>
sure, it definitely should be!
<apritzel>
since it seems to fix a serious problem, at least for you
<loki666>
Ok, I'll do more tests, to see if it holds, and I'll probably come back to you for help on submitting that
<apritzel>
and increasing voltages should not affect other chips, it's just a tiny bit more wasteful, in the worst case
<apritzel>
we might want to keep the higher voltage for 1.4GHz, though, but please continue your testing with using just 1.1V there
<apritzel>
loki666: sure, happy to help with the logistics
<loki666>
for 1.4, I did bump it from 1.1v to 1.16v no ?
<loki666>
ah you mean keep 1.1v ?
<apritzel>
oh, right, sorry, misread that
<apritzel>
so keep whatever Anbernic uses, at lesat for now
<apritzel>
ah, I see now that the 1.5GHz OPP already uses 1.16V
<loki666>
yes this freq was holding, since I've been using performance gov for a while now
digetx has quit [Quit: No Ping reply in 180 seconds.]
digetx has joined #linux-sunxi
<loki666>
mmmh looks like 1.5Ghz is not *that* stable, or at least transition to/from it
<loki666>
if don't use that freq it seems stable enough (btw Anbernic is not using it)
<loki666>
should I disable it in opp-table, or bump to 1.2v ?
dsimic is now known as Guest1356
dsimic has joined #linux-sunxi
Guest1356 has quit [Ping timeout: 480 seconds]
<apritzel>
1.16V is already quite high, so I would just drop it. It's a bummer, but other packages don't support it either, and if in doubt we should follow Anbernic
<apritzel>
because they should have access to many more data points, and erring on the side of caution is what we aim for in mainline as well
<apritzel>
if people want to overclock their individual device, that's fine, but up to them
<loki666>
I think if it should have crashed, it would have already
<loki666>
so how should I make the commit message ?
<apritzel>
what to write, you mean? Mention that your device was very unstable, especially with some OPPs. And that the manufacturer uses more conservative values
<apritzel>
that should be a no-brainer, really: if we have credible reports of a device crashing, that should be fixed immediately
<apritzel>
roughly yes, but commit messages can be quite elaborate, so feel free to go into more detail
<apritzel>
for instance I would write that *your* device was very unstable, to make this more credible
<apritzel>
IIRC macromorgan and acmeplus never mentioned crashes?
<apritzel>
also please state how "unstable" manifested itself, and under what conditions
<apritzel>
IIUC this was quite easy to reproduce for you, which is an important detail
<apritzel>
and write a bit more about why you are dropping the 1.5GHz OPP, since this is the only downside of this patch, so would benefit from some rationale
<loki666>
ok
<apritzel>
and you said that 1.5 GHz worked fine before, when just running the perf governor, right?
<apritzel>
so can you try with the reparenting patch, if apparently transitioning to 1.5GHz is what's causing issues?