Message ID | 20240108140118.1596-1-quic_sibis@quicinc.com |
---|---|
Headers | show |
Series | firmware: arm_scmi: Register and handle limits change notification | expand |
On Mon, Jan 08, 2024 at 07:31:15PM +0530, Sibi Sankar wrote: > This series registers for scmi limits change notifications and adds > perf_notify_support/perf_opp_xlate interfaces which are used by the > scmi cpufreq driver to determine the throttled frequency and apply HW > pressure. > Hi, a few initial remarks from the mere SCMI standpoint. Unlinke most SCMI protocols that expose domains info bits via an *info_get protocol operation, PERF does no do this since (till now) there wasn't a compelling reason (i.e. users) Ulf recently in his GenPD/SCMI series recently started exposing something and now you need to expose even more, adding also a new xlate ops. For the sake of simplicity, I think that we could now expose straight away the whole perf_domain_info and embedded structs via the usual *info_get. After having done that, you can just drop your patch 1 and 2 since you can access the needed info from the cpufreq_driver right away. Having said, I have already such patch ready (for my internal testing), I wll post it by the end of week after a minor cleanup, if you can bear with me. Thoughts ? Thanks, Cristian
On 1/17/24 15:11, Cristian Marussi wrote: > On Mon, Jan 08, 2024 at 07:31:15PM +0530, Sibi Sankar wrote: >> This series registers for scmi limits change notifications and adds >> perf_notify_support/perf_opp_xlate interfaces which are used by the >> scmi cpufreq driver to determine the throttled frequency and apply HW >> pressure. >> > > Hi, > > a few initial remarks from the mere SCMI standpoint. > > Unlinke most SCMI protocols that expose domains info bits via an > *info_get protocol operation, PERF does no do this since (till now) there > wasn't a compelling reason (i.e. users) > > Ulf recently in his GenPD/SCMI series recently started exposing something > and now you need to expose even more, adding also a new xlate ops. > > For the sake of simplicity, I think that we could now expose straight > away the whole perf_domain_info and embedded structs via the usual *info_get. > > After having done that, you can just drop your patch 1 and 2 since you > can access the needed info from the cpufreq_driver right away. > > Having said, I have already such patch ready (for my internal testing), I > wll post it by the end of week after a minor cleanup, if you can bear with me. > > Thoughts ? Ack, just from the naming I initially thought info_get would include everything but it just exposed minimal info. We certainly don't want to keep adding very similar ops just to expose more such info. I'll re-send the remainder of the series after you are done with your patches. Thanks. -Sibi > > Thanks, > Cristian