From: Catalin Marinas <catalin.marinas@arm.com>
To: Jason Gunthorpe <jgg@nvidia.com>
Cc: Mark Rutland <mark.rutland@arm.com>,
Leon Romanovsky <leon@kernel.org>, Arnd Bergmann <arnd@arndb.de>,
linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
linux-rdma@vger.kernel.org, llvm@lists.linux.dev,
Michael Guralnik <michaelgur@mellanox.com>,
Nathan Chancellor <nathan@kernel.org>,
Nick Desaulniers <ndesaulniers@google.com>,
Will Deacon <will@kernel.org>
Subject: Re: [PATCH rdma-next 1/2] arm64/io: add memcpy_toio_64
Date: Mon, 4 Dec 2023 17:31:47 +0000 [thread overview]
Message-ID: <ZW4NAzI_jvwoq8dL@arm.com> (raw)
In-Reply-To: <20231127134505.GI436702@nvidia.com>
On Mon, Nov 27, 2023 at 09:45:05AM -0400, Jason Gunthorpe wrote:
> On Mon, Nov 27, 2023 at 12:42:41PM +0000, Catalin Marinas wrote:
> > > > What's the actual requirement here? Is this just for performance?
> > >
> > > Yes, just performance.
> >
> > Do you have any rough numbers (percentage)? It's highly
> > microarchitecture-dependent until we get the ST64B instruction.
>
> The current C code is an open coded store loop. The kernel does 250
> tries and measures if any one of them succeeds to combine.
>
> On x86, and older ARM cores we see that 100% of the time at least 1 in
> 250 tries succeeds.
>
> With the new CPU cores we see more like 9 out of 10 time there are 0
> in 250 tries that succeed. Ie we can go thousands of times without
> seeing any successful WC combine.
>
> The STP block brings it back to 100% of the time 1 in 250 succeed.
That's a bit confusing to me: 1 in 250 succeeding is still pretty rare.
But I guess what your benchmark says is that at least 1 succeeded to
write-combine and it might as well be all 250 tries. It's more
interesting to see if there's actual performance gain in real-world
traffic, not just some artificial benchmark (I may have misunderstood
your numbers above).
> However, in userspace we have long been using ST4 to create a
> single-instruction 64 byte store on ARM64. As far as I know this is
> highly reliable. I don't have direct data on the STP configuration.
Personally I'd optimise the mempcy_toio() arm64 implementation to do
STPs if the alignment is right (like we do for classic memcpy()).
There's a slight overhead for alignment checking but I suspect it would
be lost as long as you can get the write-combining. Not sure whether the
interspersed reads in memcpy_toio() would somehow prevent the
write-combining.
A memcpy_toio_64() can use the new ST64B instruction if available or
fall back to memcpy_toio() on arm64. It should also have the DGH
instruction (io_stop_wc()) but only if falling back to classic
memcpy_toio(). We don't need DGH with ST64B.
> > More of a bike-shedding, I wonder whether the __iowrite*_copy()
> > semantics are better suited for what you need in terms of ordering (not
> > that mempcy_toio() to Normal NC memory gives us any ordering).
>
> I have the same remark I gave to Niklas, this does not require
> alignment or an exact 64 byte size. It was clearly made to support WC
> stores since Pathscale did it, but I don't see this mapping nicely to
> the future 64 byte store instructions are we getting.
As above, I'd suggest just using memcpy_toio() as a fallback if ST64B is
not available.
> We could name it __iowrite512_copy() if that makes more sense?
I've been thinking at the __iowrite*_copy() and these also take a
'count' argument. I assume in this instance we don't really need one, so
it's just additional overhead (more like API clutter, I doubt it makes
much difference for performance). I'd say just stick to the
mempcy_toio_64() but have the io_stop_wc() inside this function as we
won't need it with ST64B.
Well, unless someone has a better name for this function.
--
Catalin
next prev parent reply other threads:[~2023-12-04 17:31 UTC|newest]
Thread overview: 68+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-11-23 19:04 [PATCH rdma-next 0/2] Add and use memcpy_toio_64() Leon Romanovsky
2023-11-23 19:04 ` [PATCH rdma-next 1/2] arm64/io: add memcpy_toio_64 Leon Romanovsky
2023-11-24 10:16 ` Mark Rutland
2023-11-24 12:23 ` Jason Gunthorpe
2023-11-27 12:42 ` Catalin Marinas
2023-11-27 13:45 ` Jason Gunthorpe
2023-12-04 17:31 ` Catalin Marinas [this message]
2023-12-04 18:23 ` Jason Gunthorpe
2023-12-05 17:21 ` Catalin Marinas
2023-12-05 17:51 ` Jason Gunthorpe
2023-12-05 19:34 ` Catalin Marinas
2023-12-05 19:51 ` Jason Gunthorpe
2023-12-06 11:09 ` Catalin Marinas
2023-12-06 12:59 ` Jason Gunthorpe
2024-01-16 18:51 ` Jason Gunthorpe
2024-01-17 12:30 ` Mark Rutland
2024-01-17 12:36 ` Jason Gunthorpe
2024-01-17 12:41 ` Jason Gunthorpe
2024-01-17 13:29 ` Mark Rutland
2024-01-23 20:38 ` Catalin Marinas
2024-01-24 1:27 ` Jason Gunthorpe
2024-01-24 8:26 ` Marc Zyngier
2024-01-24 13:06 ` Jason Gunthorpe
2024-01-24 13:32 ` Marc Zyngier
2024-01-24 15:52 ` Jason Gunthorpe
2024-01-24 17:54 ` Catalin Marinas
2024-01-25 1:29 ` Jason Gunthorpe
2024-01-26 16:15 ` Catalin Marinas
2024-01-26 17:09 ` Jason Gunthorpe
2024-01-24 11:38 ` Mark Rutland
2024-01-24 12:40 ` Catalin Marinas
2024-01-24 13:27 ` Jason Gunthorpe
2024-01-24 17:22 ` Catalin Marinas
2024-01-24 19:26 ` Jason Gunthorpe
2024-01-25 17:43 ` Jason Gunthorpe
2024-01-26 14:56 ` Catalin Marinas
2024-01-26 15:24 ` Jason Gunthorpe
2024-01-17 14:07 ` Mark Rutland
2024-01-17 15:28 ` Jason Gunthorpe
2024-01-17 16:05 ` Will Deacon
2024-01-18 16:18 ` Jason Gunthorpe
2024-01-24 11:31 ` Mark Rutland
2023-11-24 12:58 ` Robin Murphy
2023-11-24 13:45 ` Jason Gunthorpe
2023-11-24 15:32 ` Robin Murphy
2023-11-24 14:10 ` Niklas Schnelle
2023-11-24 14:20 ` Jason Gunthorpe
2023-11-24 14:48 ` Niklas Schnelle
2023-11-24 14:53 ` Niklas Schnelle
2023-11-24 14:55 ` Jason Gunthorpe
2023-11-24 15:59 ` Niklas Schnelle
2023-11-24 16:06 ` Jason Gunthorpe
2023-11-27 17:43 ` Niklas Schnelle
2023-11-27 17:51 ` Jason Gunthorpe
2023-11-28 16:28 ` Niklas Schnelle
2024-01-16 17:33 ` Jason Gunthorpe
2024-01-17 13:20 ` Niklas Schnelle
2024-01-17 13:26 ` Jason Gunthorpe
2024-01-17 17:55 ` Jason Gunthorpe
2024-01-18 13:46 ` Niklas Schnelle
2024-01-18 14:00 ` Jason Gunthorpe
2024-01-18 15:59 ` Niklas Schnelle
2024-01-18 16:21 ` Jason Gunthorpe
2024-01-18 16:25 ` Niklas Schnelle
2024-01-19 11:52 ` Niklas Schnelle
2024-02-16 12:09 ` Niklas Schnelle
2024-02-16 12:39 ` Jason Gunthorpe
2023-11-23 19:04 ` [PATCH rdma-next 2/2] IB/mlx5: Use memcpy_toio_64() for write combining stores Leon Romanovsky
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZW4NAzI_jvwoq8dL@arm.com \
--to=catalin.marinas@arm.com \
--cc=arnd@arndb.de \
--cc=jgg@nvidia.com \
--cc=leon@kernel.org \
--cc=linux-arch@vger.kernel.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-rdma@vger.kernel.org \
--cc=llvm@lists.linux.dev \
--cc=mark.rutland@arm.com \
--cc=michaelgur@mellanox.com \
--cc=nathan@kernel.org \
--cc=ndesaulniers@google.com \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).