On Sat, Sep 20, 2025 at 10:16:21AM -0000, Stuart Henderson wrote: > Not 100% sure where pppoe will end up, but I think it's in softnet which > can move between cores. > > The scheduler doesn't distinguish between P and E cores (or between > "pseudo cores" if SMT is enabled) so if you want consistent performance > and have a way to disable E cores, that may be a good idea.
On machines which don't have a way to disable E cores in the BIOS, it's also possible to disable use of them, (or any other specific cores), at the OS level with a fairly tivial kernel patch. Not ideal if you're otherwise happy running the stock kernel, but it's a possible work-around if you need to solve a specific problem. With the increasing number of machines with these sorts of low-power cores that it's quite reasonable for people to want to disable, I wonder whether a sysctl similar to hw.smt would be useful? I've hard coded patches for the few machines I manage that are affected, but a more general mechanism might be useful to others. But remember that there are CPUs on the market with more than two types of core. So if it was a binary on/off switch, then a decision would need to be made over what to enable and disable. And something more complicated than a simple on/off twiddle might not be welcome.
