Hi Gleb, > 2) The AMD CPPC switched on by default instead of Cool`n'Quiet caused > problems on > certain hardware. For laptops see bug 292615. > At Netflix A/B test we discovered that CPPC mode reports wrong frequency > levels > to powerd, which results in reduced performance. > This problem is still not resolved.
ShengYi will be, AFAIK, mostly unavailable for at least a month, so I'm handling the AMD CPPC problems (actually, I've done some part of the initial work as a reviewer and by proposing changes to it). I've been having some mail problems in the past days, so sorry if I missed something, but it's the first time I'm hearing Netflix experienced a problem caused by the new AMD CPPC. If powerd(8) tries to set frequencies to balance performance/efficiency (as it always has AFAIK), it is in effect incompatible with CPPC, both this new AMD implementation and the already existing Intel one. They do not allow to set frequency but just report the current real frequency which is *dynamic*, i.e., it can frequently change even for non-changing CPPC control values (that's the entire purpose of CPPC). In the future, we'll probably want to add support of CPPC to powerd(8). For now, it's either CPPC or powerd(8) (as for Intel), and a priori I'd recommend switching to CPPC entirely (once we have finished with the potential problems here). For pressing production issues, in the meantime you can set the 'machdep.hwpstate_amd_cppc_enable' tunable to 0 to force using the "manual" P-states that powerd(8) can manipulate. Please help us by intervening in bug 292615, adding comments about your problems and findings (or mail them to me if they must stay confidential), and please read the latest comments, apply the latest patch and play with the different knobs as explained there. Depending on your feedback, we may be able to provide more specific guidance. If CPPC mode reports a wrong frequency (how do you tell?), please comment there also, and depending on first analysis, we'll see if we want it to be a separate bug or not. Thanks and regards. -- Olivier Certner
signature.asc
Description: This is a digitally signed message part.
