Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Content-Type: text/plain;
        charset=utf-8
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH] x86: only use ERMS for user copies for larger sizes
From:   Andy Lutomirski <luto@amacapital.net>
In-Reply-To: <658cdb28-e3e5-c0af-368f-c26daf9986ac@kernel.dk>
Date:   Wed, 21 Nov 2018 11:26:09 -0700
Cc:     Linus Torvalds <torvalds@linux-foundation.org>, pabeni@redhat.com,
        Ingo Molnar <mingo@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, bp@alien8.de,
        Peter Anvin <hpa@zytor.com>,
        the arch/x86 maintainers <x86@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Andrew Lutomirski <luto@kernel.org>,
        Peter Zijlstra <a.p.zijlstra@chello.nl>, dvlasenk@redhat.com,
        brgerst@gmail.com,
        Linux List Kernel Mailing <linux-kernel@vger.kernel.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <9E7DFB44-8A2A-48CF-972E-6CB5122CCA20@amacapital.net>
References: <02bfc577-32a5-66be-64bf-d476b7d447d2@kernel.dk> <20181121063609.GA109082@gmail.com> <48e27a3a-2bb2-ff41-3512-8aeb3fd59e57@kernel.dk> <1c22125bb5d22c2dcd686d0d3b390f115894f746.camel@redhat.com> <CAHk-=wgnjB+=o0771=J_YkQzabU5aadh6pN3x9Vk4HPs3MHL3g@mail.gmail.com> <658cdb28-e3e5-c0af-368f-c26daf9986ac@kernel.dk>
To:     Jens Axboe <axboe@kernel.dk>, dave.hansen@intel.com
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On Nov 21, 2018, at 11:04 AM, Jens Axboe <axboe@kernel.dk> wrote:
>=20
>> On 11/21/18 10:27 AM, Linus Torvalds wrote:
>>> On Wed, Nov 21, 2018 at 5:45 AM Paolo Abeni <pabeni@redhat.com> wrote:
>>>=20
>>> In my experiments 64 bytes was the break even point for all the CPUs I
>>> had handy, but I guess that may change with other models.
>>=20
>> Note that experiments with memcpy speed are almost invariably broken.
>> microbenchmarks don't show the impact of I$, but they also don't show
>> the impact of _behavior_.
>>=20
>> For example, there might be things like "repeat strings do cacheline
>> optimizations" that end up meaning that cachelines stay in L2, for
>> example, and are never brought into L1. That can be a really good
>> thing, but it can also mean that now the result isn't as close to the
>> CPU, and the subsequent use of the cacheline can be costlier.
>=20
> Totally agree, which is why all my testing was NOT microbenchmarking.
>=20
>> I say "go for upping the limit to 128 bytes".
>=20
> See below...
>=20
>> That said, if the aio user copy is _so_ critical that it's this
>> noticeable, there may be other issues. Sometimes _real_ cost of small
>> user copies is often the STAC/CLAC, more so than the "rep movs".
>>=20
>> It would be interesting to know exactly which copy it is that matters
>> so much...  *inlining* the erms case might show that nicely in
>> profiles.
>=20
> Oh I totally agree, which is why I since went a different route. The
> copy that matters is the copy_from_user() of the iocb, which is 64
> bytes. Even for 4k IOs, copying 64b per IO is somewhat counter
> productive for O_DIRECT.

Can we maybe use this as an excuse to ask for some reasonable instructions t=
o access user memory?  Intel already did all the dirty work of giving someth=
ing resembling sane semantics for the kernel doing a user-privileged access w=
ith WRUSS.  How about WRUSER, RDUSER, and maybe even the REP variants?  And I=
 suppose LOCK CMPXCHGUSER.

Or Intel could try to make STAC and CLAC be genuinely fast (0 or 1 cycles an=
d no stalls *ought* to be possible if it were handled in the front end, as l=
ong as there aren=E2=80=99t any PUSHF or POPF instructions in the pipeline).=
  As it stands, I assume that both instructions prevent any following memory=
 accesses from starting until they retire, and they might even be nastily mi=
crocoded to handle the overloading of AC.