From: Tom Talpey <tom@talpey.com>
To: Haakon Bugge <haakon.bugge@oracle.com>
Cc: Chuck Lever III <chuck.lever@oracle.com>,
Jason Gunthorpe <jgg@nvidia.com>, Christoph Hellwig <hch@lst.de>,
Leon Romanovsky <leon@kernel.org>,
Doug Ledford <dledford@redhat.com>,
Leon Romanovsky <leonro@nvidia.com>,
Adit Ranadive <aditr@vmware.com>,
Anna Schumaker <anna.schumaker@netapp.com>,
Ariel Elior <aelior@marvell.com>,
Avihai Horon <avihaih@nvidia.com>,
Bart Van Assche <bvanassche@acm.org>,
Bernard Metzler <bmt@zurich.ibm.com>,
"David S. Miller" <davem@davemloft.net>,
Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>,
Devesh Sharma <devesh.sharma@broadcom.com>,
Faisal Latif <faisal.latif@intel.com>,
Jack Wang <jinpu.wang@ionos.com>,
Jakub Kicinski <kuba@kernel.org>,
Bruce Fields <bfields@fieldses.org>, Jens Axboe <axboe@fb.com>,
Karsten Graul <kgraul@linux.ibm.com>,
Keith Busch <kbusch@kernel.org>, Lijun Ou <oulijun@huawei.com>,
CIFS <linux-cifs@vger.kernel.org>,
LKML <linux-kernel@vger.kernel.org>,
Linux NFS Mailing List <linux-nfs@vger.kernel.org>,
"linux-nvme@lists.infradead.org" <linux-nvme@lists.infradead.org>,
OFED mailing list <linux-rdma@vger.kernel.org>,
"linux-s390@vger.kernel.org" <linux-s390@vger.kernel.org>,
Max Gurtovoy <maxg@mellanox.com>,
Max Gurtovoy <mgurtovoy@nvidia.com>,
"Md. Haris Iqbal" <haris.iqbal@ionos.com>,
Michael Guralnik <michaelgur@nvidia.com>,
Michal Kalderon <mkalderon@marvell.com>,
Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>,
Naresh Kumar PBS <nareshkumar.pbs@broadcom.com>,
Linux-Net <netdev@vger.kernel.org>,
Potnuri Bharat Teja <bharat@chelsio.com>,
"rds-devel@oss.oracle.com" <rds-devel@oss.oracle.com>,
Sagi Grimberg <sagi@grimberg.me>,
"samba-technical@lists.samba.org"
<samba-technical@lists.samba.org>,
Santosh Shilimkar <santosh.shilimkar@oracle.com>,
Selvin Xavier <selvin.xavier@broadcom.com>,
Shiraz Saleem <shiraz.saleem@intel.com>,
Somnath Kotur <somnath.kotur@broadcom.com>,
Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>,
Steve French <sfrench@samba.org>,
Trond Myklebust <trond.myklebust@hammerspace.com>,
VMware PV-Drivers <pv-drivers@vmware.com>,
Weihang Li <liweihang@huawei.com>,
Yishai Hadas <yishaih@nvidia.com>,
Zhu Yanjun <zyjzyj2000@gmail.com>
Subject: Re: [PATCH rdma-next 00/10] Enable relaxed ordering for ULPs
Date: Fri, 9 Apr 2021 13:49:15 -0400 [thread overview]
Message-ID: <7b9e7d9c-13d7-0d18-23b4-0d94409c7741@talpey.com> (raw)
In-Reply-To: <1FA38618-E245-4C53-BF49-6688CA93C660@oracle.com>
On 4/9/2021 12:27 PM, Haakon Bugge wrote:
>
>
>> On 9 Apr 2021, at 17:32, Tom Talpey <tom@talpey.com> wrote:
>>
>> On 4/9/2021 10:45 AM, Chuck Lever III wrote:
>>>> On Apr 9, 2021, at 10:26 AM, Tom Talpey <tom@talpey.com> wrote:
>>>>
>>>> On 4/6/2021 7:49 AM, Jason Gunthorpe wrote:
>>>>> On Mon, Apr 05, 2021 at 11:42:31PM +0000, Chuck Lever III wrote:
>>>>>
>>>>>> We need to get a better idea what correctness testing has been done,
>>>>>> and whether positive correctness testing results can be replicated
>>>>>> on a variety of platforms.
>>>>> RO has been rolling out slowly on mlx5 over a few years and storage
>>>>> ULPs are the last to change. eg the mlx5 ethernet driver has had RO
>>>>> turned on for a long time, userspace HPC applications have been using
>>>>> it for a while now too.
>>>>
>>>> I'd love to see RO be used more, it was always something the RDMA
>>>> specs supported and carefully architected for. My only concern is
>>>> that it's difficult to get right, especially when the platforms
>>>> have been running strictly-ordered for so long. The ULPs need
>>>> testing, and a lot of it.
>>>>
>>>>> We know there are platforms with broken RO implementations (like
>>>>> Haswell) but the kernel is supposed to globally turn off RO on all
>>>>> those cases. I'd be a bit surprised if we discover any more from this
>>>>> series.
>>>>> On the other hand there are platforms that get huge speed ups from
>>>>> turning this on, AMD is one example, there are a bunch in the ARM
>>>>> world too.
>>>>
>>>> My belief is that the biggest risk is from situations where completions
>>>> are batched, and therefore polling is used to detect them without
>>>> interrupts (which explicitly). The RO pipeline will completely reorder
>>>> DMA writes, and consumers which infer ordering from memory contents may
>>>> break. This can even apply within the provider code, which may attempt
>>>> to poll WR and CQ structures, and be tripped up.
>>> You are referring specifically to RPC/RDMA depending on Receive
>>> completions to guarantee that previous RDMA Writes have been
>>> retired? Or is there a particular implementation practice in
>>> the Linux RPC/RDMA code that worries you?
>>
>> Nothing in the RPC/RDMA code, which is IMO correct. The worry, which
>> is hopefully unfounded, is that the RO pipeline might not have flushed
>> when a completion is posted *after* posting an interrupt.
>>
>> Something like this...
>>
>> RDMA Write arrives
>> PCIe RO Write for data
>> PCIe RO Write for data
>> ...
>> RDMA Write arrives
>> PCIe RO Write for data
>> ...
>> RDMA Send arrives
>> PCIe RO Write for receive data
>> PCIe RO Write for receive descriptor
>
> Do you mean the Write of the CQE? It has to be Strongly Ordered for a correct implementation. Then it will shure prior written RO date has global visibility when the CQE can be observed.
I wasn't aware that a strongly-ordered PCIe Write will ensure that
prior relaxed-ordered writes went first. If that's the case, I'm
fine with it - as long as the providers are correctly coded!!
>> PCIe interrupt (flushes RO pipeline for all three ops above)
>
> Before the interrupt, the HCA will write the EQE (Event Queue Entry). This has to be a Strongly Ordered write to "push" prior written CQEs so that when the EQE is observed, the prior writes of CQEs have global visibility.
>
> And the MSI-X write likewise, to avoid spurious interrupts.
Ok, and yes agreed the same principle would apply.
Is there any implication if a PCIe switch were present on the
motherboard? The switch is allowed to do some creative routing
if the operation is relaxed, correct?
Tom.
> Thxs, Håkon
>
>>
>> RPC/RDMA polls CQ
>> Reaps receive completion
>>
>> RDMA Send arrives
>> PCIe RO Write for receive data
>> PCIe RO write for receive descriptor
>> Does *not* interrupt, since CQ not armed
>>
>> RPC/RDMA continues to poll CQ
>> Reaps receive completion
>> PCIe RO writes not yet flushed
>> Processes incomplete in-memory data
>> Bzzzt
>>
>> Hopefully, the adapter performs a PCIe flushing read, or something
>> to avoid this when an interrupt is not generated. Alternatively, I'm
>> overly paranoid.
>>
>> Tom.
>>
>>>> The Mellanox adapter, itself, historically has strict in-order DMA
>>>> semantics, and while it's great to relax that, changing it by default
>>>> for all consumers is something to consider very cautiously.
>>>>
>>>>> Still, obviously people should test on the platforms they have.
>>>>
>>>> Yes, and "test" be taken seriously with focus on ULP data integrity.
>>>> Speedups will mean nothing if the data is ever damaged.
>>> I agree that data integrity comes first.
>>> Since I currently don't have facilities to test RO in my lab, the
>>> community will have to agree on a set of tests and expected results
>>> that specifically exercise the corner cases you are concerned about.
>>> --
>>> Chuck Lever
>
next prev parent reply other threads:[~2021-04-09 17:49 UTC|newest]
Thread overview: 56+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-05 5:23 [PATCH rdma-next 00/10] Enable relaxed ordering for ULPs Leon Romanovsky
2021-04-05 5:23 ` [PATCH rdma-next 01/10] RDMA: Add access flags to ib_alloc_mr() and ib_mr_pool_init() Leon Romanovsky
2021-04-05 13:46 ` Christoph Hellwig
2021-04-06 5:24 ` Leon Romanovsky
2021-04-05 15:27 ` Bart Van Assche
2021-04-06 5:23 ` Leon Romanovsky
2021-04-06 5:27 ` Christoph Hellwig
2021-04-06 5:58 ` Leon Romanovsky
2021-04-06 12:13 ` Jason Gunthorpe
2021-04-06 12:30 ` Christoph Hellwig
2021-04-06 14:04 ` Jason Gunthorpe
2021-04-06 14:15 ` Christoph Hellwig
2021-04-06 14:40 ` Jason Gunthorpe
2021-04-06 14:54 ` Christoph Hellwig
2021-04-06 15:03 ` Christoph Hellwig
2021-04-07 18:28 ` Jason Gunthorpe
2021-04-05 5:23 ` [PATCH rdma-next 02/10] RDMA/core: Enable Relaxed Ordering in __ib_alloc_pd() Leon Romanovsky
2021-04-05 18:01 ` Tom Talpey
2021-04-05 20:40 ` Adit Ranadive
2021-04-06 6:28 ` Leon Romanovsky
2021-04-05 5:23 ` [PATCH rdma-next 03/10] RDMA/iser: Enable Relaxed Ordering Leon Romanovsky
2021-04-05 5:23 ` [PATCH rdma-next 04/10] RDMA/rtrs: " Leon Romanovsky
2021-04-05 5:23 ` [PATCH rdma-next 05/10] RDMA/srp: " Leon Romanovsky
2021-04-05 5:24 ` [PATCH rdma-next 06/10] nvme-rdma: " Leon Romanovsky
2021-04-05 5:24 ` [PATCH rdma-next 07/10] cifs: smbd: " Leon Romanovsky
2021-04-05 5:24 ` [PATCH rdma-next 08/10] net/rds: " Leon Romanovsky
2021-04-05 5:24 ` [PATCH rdma-next 09/10] net/smc: " Leon Romanovsky
2021-04-05 5:24 ` [PATCH rdma-next 10/10] xprtrdma: " Leon Romanovsky
2021-04-05 13:41 ` [PATCH rdma-next 00/10] Enable relaxed ordering for ULPs Christoph Hellwig
2021-04-05 14:08 ` Leon Romanovsky
2021-04-05 16:11 ` Santosh Shilimkar
2021-04-05 17:54 ` Tom Talpey
2021-04-05 20:07 ` Jason Gunthorpe
2021-04-05 23:42 ` Chuck Lever III
2021-04-05 23:50 ` Keith Busch
2021-04-06 5:12 ` Leon Romanovsky
2021-04-06 11:49 ` Jason Gunthorpe
2021-04-09 14:26 ` Tom Talpey
2021-04-09 14:45 ` Chuck Lever III
2021-04-09 15:32 ` Tom Talpey
2021-04-09 16:27 ` Haakon Bugge
2021-04-09 17:49 ` Tom Talpey [this message]
2021-04-10 13:30 ` David Laight
2021-04-12 18:32 ` Haakon Bugge
2021-04-12 20:20 ` Tom Talpey
2021-04-12 22:48 ` Jason Gunthorpe
2021-04-14 14:16 ` Tom Talpey
2021-04-14 14:41 ` David Laight
2021-04-14 14:49 ` Jason Gunthorpe
2021-04-14 14:44 ` Jason Gunthorpe
2021-04-09 16:40 ` Jason Gunthorpe
2021-04-09 17:44 ` Tom Talpey
2021-04-06 2:37 ` Honggang LI
2021-04-06 5:09 ` Leon Romanovsky
2021-04-06 11:53 ` Jason Gunthorpe
2021-04-11 10:09 ` Max Gurtovoy
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=7b9e7d9c-13d7-0d18-23b4-0d94409c7741@talpey.com \
--to=tom@talpey.com \
--cc=aditr@vmware.com \
--cc=aelior@marvell.com \
--cc=anna.schumaker@netapp.com \
--cc=avihaih@nvidia.com \
--cc=axboe@fb.com \
--cc=bfields@fieldses.org \
--cc=bharat@chelsio.com \
--cc=bmt@zurich.ibm.com \
--cc=bvanassche@acm.org \
--cc=chuck.lever@oracle.com \
--cc=davem@davemloft.net \
--cc=dennis.dalessandro@cornelisnetworks.com \
--cc=devesh.sharma@broadcom.com \
--cc=dledford@redhat.com \
--cc=faisal.latif@intel.com \
--cc=haakon.bugge@oracle.com \
--cc=haris.iqbal@ionos.com \
--cc=hch@lst.de \
--cc=jgg@nvidia.com \
--cc=jinpu.wang@ionos.com \
--cc=kbusch@kernel.org \
--cc=kgraul@linux.ibm.com \
--cc=kuba@kernel.org \
--cc=leon@kernel.org \
--cc=leonro@nvidia.com \
--cc=linux-cifs@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-nfs@vger.kernel.org \
--cc=linux-nvme@lists.infradead.org \
--cc=linux-rdma@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=liweihang@huawei.com \
--cc=maxg@mellanox.com \
--cc=mgurtovoy@nvidia.com \
--cc=michaelgur@nvidia.com \
--cc=mike.marciniszyn@cornelisnetworks.com \
--cc=mkalderon@marvell.com \
--cc=nareshkumar.pbs@broadcom.com \
--cc=netdev@vger.kernel.org \
--cc=oulijun@huawei.com \
--cc=pv-drivers@vmware.com \
--cc=rds-devel@oss.oracle.com \
--cc=sagi@grimberg.me \
--cc=samba-technical@lists.samba.org \
--cc=santosh.shilimkar@oracle.com \
--cc=selvin.xavier@broadcom.com \
--cc=sfrench@samba.org \
--cc=shiraz.saleem@intel.com \
--cc=somnath.kotur@broadcom.com \
--cc=sriharsha.basavapatna@broadcom.com \
--cc=trond.myklebust@hammerspace.com \
--cc=yishaih@nvidia.com \
--cc=zyjzyj2000@gmail.com \
/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).