Re: [PATCH 5.13 073/127] pipe: avoid unnecessary EPOLLET wakeups under normal loads

[Linus Torvalds]

On Tue, Aug 24, 2021 at 10:35 AM Sasha Levin <sashal@xxxxxxxxxx> wrote:
>
> On Tue, Aug 24, 2021 at 10:00:33AM -0700, Linus Torvalds wrote:
> >
> >Honestly, I don't understand the performance regression, but that's
> >likely on me, not on the kernel test robot.
>
> I'll drop it for now.
>
> Ideally we wouldn't take it at all, but I don't think any of us wants to
> field "my tests have regressed!" questions for the next 5 years or so.

I have a theory about what is going on, and it's not a new problem,
but it would explain how that test might be so bimodal in performance
if you happen to hit the exact right timing. And the test robot just
might not have hit the right timing previously.

See

    https://lore.kernel.org/lkml/CAHk-=wiKAg5QtrQOtvKNwkRUn0b2xufO54GPhUoTWxBgDzXWNA@xxxxxxxxxxxxxx/

with a test-patch in the next message in that thread.

So it looks like another case of an odd test, but it should be easy
enough to make that test happy too. Knock wood.

             Linus