Re: Name hashing function causing a perf regression

From: Linus Torvalds <torvalds@linux-foundation.org>
To: Josef Bacik <jbacik@fb.com>, stable <stable@vger.kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Christoph Hellwig <hch@infradead.org>, Chris Mason <clm@fb.com>
Subject: Re: Name hashing function causing a perf regression
Date: Mon, 15 Sep 2014 09:22:57 -0700	[thread overview]
Message-ID: <CA+55aFyFYzXdPCR3nyjJBTYbEbTb1isrGJX1kDHe4EKjXTu2sA@mail.gmail.com> (raw)
In-Reply-To: <54170C08.5070603@fb.com>

On Mon, Sep 15, 2014 at 8:55 AM, Josef Bacik <jbacik@fb.com> wrote:
>
> I can't test on 3.17 proper since the Fusion IO driver doesn't build
> properly there and I'm not being paid to work on it anymore so I'm not
> fixing it ;).  Thanks for fixing this, I've pulled back 99d263d4c5b2 which
> will do us just fine.

Ok. I'm cc'ing stable to know that

    99d263d4c5b2 ("vfs: fix bad hashing of dentries")

should be added to the queues for 3.10+.

Greg&co - it's a simple fix for a performance regression. Not
end-of-the-world, but if it ends up being in the FB kernel trees,
might as well get it back-ported to the other stable trees too.

The other performance issues I found are actually potentially worse,
but they aren't regressions and they hit only when using namespaces.
Which mostly nobody does on old kernels anyway. More of a "systemd
uses namespaces for /tmp, and then it's quite noticeable as slowing
down pathname lookups there if you benchmark it".

So the single commit Josef mentions is likely sufficient for stable.

               Linus