On Sat, Oct 03, 2020 at 05:54:38AM +0200, Eric Dumazet wrote: > > Sure, a WQ is already giving nice results on appliances, because there > you do not need strong isolation. > Would a kthread approach also work well on appliances ? Probably...
Right. I think we're on the same page. The only reason I'm bringing up multiple co-existing approaches now is to make sure they are designed in from the start instead of as afterthought. Two implementations already exist, but it doesn't look like that they can co-exist in the kernel. Like this NAPI_STATE_THREADED bit in this patch set. Should we burn that bit for kthread approach and another bit for workqueue based? I don't know. As the user of the feature I would be happy with any mechanism as long as I can flip between them in runtime :) Just like RPS and RFS.
