On 2024/12/26 20:28, Shrikanth Hegde wrote: > > > On 12/26/24 17:20, Yicong Yang wrote: >> On 2024/12/26 17:23, Shrikanth Hegde wrote: >>> >>> >>> On 12/20/24 13:23, Yicong Yang wrote: >>>> From: Yicong Yang <yangyic...@hisilicon.com> >>>> >>>> The core CPU control framework supports runtime SMT control which >>>> is not yet supported on arm64. Besides the general vulnerabilities >>>> concerns we want this runtime control on our arm64 server for: >>>> >>>> - better single CPU performance in some cases >>>> - saving overall power consumption >>>> >>> >>> EAS is disabled when SMT is present. >>> I am curious to know how power saving happens here. >> >> EAS shouldn't work on non-asymmetic systems, so it's not the case here. > > Ok. so this is a symmetric system then?
yes, symmetric. > >> System wide power consumption comes down from the CPU offlining here. >> > > Ok. So SMT is enabled by default and then at runtime disable it to save power > by off-lining the sibling threads? > yes. > > Note: When enabling SMT, current behavior differs when a core is fully > offline on different archs. You may want to see which is behavior you need in > that case. i.e either online or skip. > > PowerPC change where we are skipping a fully offline core. > https://lore.kernel.org/all/20240731030126.956210-1-ny...@linux.ibm.com/ > Thanks for the information! Currently it's implemented as online and no special need for skip. We may need further support if skip is required in the future, currently for GENERIC_ARCH_TOPOLOGY an offline CPU's thread sibling only contains itself so there's no information for checking whether the whole core is offline or not. Thanks. >> Thanks. >> >>> >>>> This patchset implements it in the following aspects: >>>> >>>> - Provides a default topology_is_primary_thread() >>>> - support retrieve SMT thread number on OF based system >>>> - support retrieve SMT thread number on ACPI based system >>>> - select HOTPLUG_SMT for arm64 >>>> >>>> Tests has been done on our ACPI based arm64 server and on ACPI/OF\ >>>> based QEMU VMs. >>>> >>> . > > .
