On Thu, Nov 17, 2022 at 10:18 AM Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote: > > On Thu, Nov 17, 2022 at 5:49 PM Eric Dumazet <edumazet@xxxxxxxxxx> wrote: > > > > On Thu, Nov 17, 2022 at 9:42 AM Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote: > > > > > > > > > > > Yes, I agree. Your comments here have not been useful (or respectful) > > > so I am Ok with that. > > > > > > - Joel > > > > Well, I have discovered that some changes went in networking tree > > without network maintainers being involved nor CCed. > > > > What can I say ? > > > > It seems I have no say, right ? > > Sorry, I take responsibility for that. FWIW, the rxrpc change is not > yet in Linus's tree. > > Also FWIW, the rxrpc case came up because we detected that it does a > scheduler wakeup from the callback. We did both static and dynamic > testing to identify callbacks that do wakeups throughout the kernel > (kernel patch available on request), as the pattern observed is things > doing wakeups typically are for use cases that are not freeing memory > but something blocking, similar to synchronize_rcu(). So it was a > "trivial/obvious" change to make for rxrpc which I might have assumed > did not need much supervision because it just reverts that API to the > old behavior -- still probably no excuse. > > Again, we can talk this out no problem. But I would strongly recommend > not calling it "crazy thing", as we did all due diligence for almost a > year (talking about it at LPC, working through various code paths and > bugs, 4 different patch redesigns on the idea (including the opt-in > that you are bringing up), including a late night debugging session to > figure this out etc). Apologies. For me "crazy" does not have the same meaning, apparently. I will try to use more neutral words in the future. > > Just to clarify, I know you review/maintain a lot of the networking > code and I really appreciate that (not praising just for the sake). > And I care about the kernel too, just like you. I had no doubts about that, really.
