On Wed, Jun 19, 2024 at 7:34 PM Aaron Rainbolt <arainbolt@xxxxxxxxxx> wrote: > > On Wed, Jun 19, 2024 at 07:09:35PM +0200, Rafael J. Wysocki wrote: > > On Wed, Jun 19, 2024 at 6:33 AM Aaron Rainbolt <arainbolt@xxxxxxxxxx> wrote: > > > > > > acpi: Allow ignoring _OSC CPPC v2 bit via kernel parameter > > > > > > The _OSC is supposed to contain a bit indicating whether the hardware > > > supports CPPC v2 or not. This bit is not always set, causing CPPC v2 to > > > be considered absent. This results in severe single-core performance > > > issues with the EEVDF scheduler on heterogenous-core Intel processors. > > > > While some things can be affected by this, I don't immediately see a > > connection between CPPC v2, Intel hybrid processors and EEVDF. > > > > In particular, why would EEVDF alone be affected? > > > > Care to explain this? > > From what I understand, the EEVDF scheduler requires ITMT (which in turn > requires CPPC v2) in order to determine which cores are performance cores > and which cores are efficiency cores. When CPPC v2 is missing, ITMT is > also missing, and the scheduler no longer can figure out which cores are > which. Thus on a system with many efficiency cores and a few performance > cores, there's a pretty decent chance the scheduler will put an intensive > single-threaded load on an efficiency core rather than on a performance > core, which has obvious performance implications since efficiency cores > are slower than performance cores by design. So the above information should go into the changelog of your patch. > A good example of someone else hitting this issue can be seen here: > https://bugzilla.kernel.org/show_bug.cgi?id=218195 This issue was brought > onto the LKML here: > https://lore.kernel.org/lkml/a106fb4733d0a3f0d6d5792705cdb5cee13731f8.camel@xxxxxxxxxxxxxxx/T/ > Srinivas would have more info here, but I do not have his email so I can't > CC him on this. OK, CCed. Now, please check the patch I've just posted in this thread: https://lore.kernel.org/linux-acpi/12457165.O9o76ZdvQC@xxxxxxxxxxxxx/