[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v3 3/4] block: drop shared-tag fairness throttling
- To: Sumit Saxena <sumit.saxena@xxxxxxxxxxxx>
- From: Keith Busch <kbusch@xxxxxxxxxx>
- Date: Wed, 10 Jun 2026 10:35:59 -0600
- Authentication-results: eu.smtp.expurgate.cloud; dkim=pass header.s=k20260515 header.d=kernel.org header.i="@kernel.org" header.h="Date:From:To:Cc:Subject:References:In-Reply-To"
- Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, "Martin K . Petersen" <martin.petersen@xxxxxxxxxx>, Jens Axboe <axboe@xxxxxxxxx>, "James E . J . Bottomley" <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>, linux-scsi@xxxxxxxxxxxxxxx, linux-block@xxxxxxxxxxxxxxx, Adam Radford <aradford@xxxxxxxxx>, Khalid Aziz <khalid@xxxxxxxxxxxxxx>, Adaptec OEM Raid Solutions <aacraid@xxxxxxxxxxxxx>, Matthew Wilcox <willy@xxxxxxxxxxxxx>, Hannes Reinecke <hare@xxxxxxxx>, "Juergen E . Fischer" <fischer@xxxxxxxxx>, Russell King <linux@xxxxxxxxxxxxxxx>, linux-arm-kernel@xxxxxxxxxxxxxxxxxxx, Finn Thain <fthain@xxxxxxxxxxxxxx>, Michael Schmitz <schmitzmic@xxxxxxxxx>, Anil Gurumurthy <anil.gurumurthy@xxxxxxxxxx>, Sudarsana Kalluru <sudarsana.kalluru@xxxxxxxxxx>, Oliver Neukum <oliver@xxxxxxxxxx>, Ali Akcaagac <aliakc@xxxxxx>, Jamie Lenehan <lenehan@xxxxxxxxxxx>, Ram Vegesna <ram.vegesna@xxxxxxxxxxxx>, target-devel@xxxxxxxxxxxxxxx, Bradley Grove <linuxdrivers@xxxxxxxxxxxx>, Satish Kharat <satishkh@xxxxxxxxx>, Sesidhar Baddela <sebaddel@xxxxxxxxx>, Karan Tilak Kumar <kartilak@xxxxxxxxx>, Yihang Li <liyihang9@xxxxxxxxxxxxxx>, Don Brace <don.brace@xxxxxxxxxxxxx>, storagedev@xxxxxxxxxxxxx, HighPoint Linux Team <linux@xxxxxxxxxxxxxxxxxx>, Tyrel Datwyler <tyreld@xxxxxxxxxxxxx>, Madhavan Srinivasan <maddy@xxxxxxxxxxxxx>, Michael Ellerman <mpe@xxxxxxxxxxxxxx>, Nicholas Piggin <npiggin@xxxxxxxxx>, Christophe Leroy <chleroy@xxxxxxxxxx>, linuxppc-dev@xxxxxxxxxxxxxxxx, Brian King <brking@xxxxxxxxxx>, Lee Duncan <lduncan@xxxxxxxx>, Chris Leech <cleech@xxxxxxxxxx>, Mike Christie <michael.christie@xxxxxxxxxx>, open-iscsi@xxxxxxxxxxxxxxxx, Justin Tee <justin.tee@xxxxxxxxxxxx>, Paul Ely <paul.ely@xxxxxxxxxxxx>, Kashyap Desai <kashyap.desai@xxxxxxxxxxxx>, Shivasharan S <shivasharan.srikanteshwara@xxxxxxxxxxxx>, Chandrakanth Patil <chandrakanth.patil@xxxxxxxxxxxx>, megaraidlinux.pdl@xxxxxxxxxxxx, Sathya Prakash Veerichetty <sathya.prakash@xxxxxxxxxxxx>, Sreekanth Reddy <sreekanth.reddy@xxxxxxxxxxxx>, mpi3mr-linuxdrv.pdl@xxxxxxxxxxxx, Suganath Prabu Subramani <suganath-prabu.subramani@xxxxxxxxxxxx>, Ranjan Kumar <ranjan.kumar@xxxxxxxxxxxx>, MPT-FusionLinux.pdl@xxxxxxxxxxxx, Daniel Palmer <daniel@xxxxxxxxx>, GOTO Masanori <gotom@xxxxxxxxxxxx>, YOKOTA Hiroshi <yokota@xxxxxxxxxxxxxxxxxxxxxxx>, Jack Wang <jinpu.wang@xxxxxxxxxxxxxxx>, Geoff Levand <geoff@xxxxxxxxxxxxx>, Michael Reed <mdr@xxxxxxx>, Nilesh Javali <njavali@xxxxxxxxxxx>, GR-QLogic-Storage-Upstream@xxxxxxxxxxx, Narsimhulu Musini <nmusini@xxxxxxxxx>, "K . Y . Srinivasan" <kys@xxxxxxxxxxxxx>, Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>, Wei Liu <wei.liu@xxxxxxxxxx>, Dexuan Cui <decui@xxxxxxxxxxxxx>, Long Li <longli@xxxxxxxxxxxxx>, linux-hyperv@xxxxxxxxxxxxxxx, "Michael S . Tsirkin" <mst@xxxxxxxxxx>, Jason Wang <jasowang@xxxxxxxxxx>, Paolo Bonzini <pbonzini@xxxxxxxxxx>, Stefan Hajnoczi <stefanha@xxxxxxxxxx>, Eugenio Perez <eperezma@xxxxxxxxxx>, virtualization@xxxxxxxxxxxxxxx, Vishal Bhakta <vishal.bhakta@xxxxxxxxxxxx>, bcm-kernel-feedback-list@xxxxxxxxxxxx, Juergen Gross <jgross@xxxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxxx, Bart Van Assche <bvanassche@xxxxxxx>
- Delivery-date: Wed, 10 Jun 2026 16:36:15 +0000
- List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
On Wed, Jun 10, 2026 at 09:16:11PM +0530, Sumit Saxena wrote:
> The motivation for this change stems from performance issue we
> encountered due to false sharing of the 'nr_active_requests_shared_tags'
> counter
> on certain CPU architectures. I initially submitted a patch to move that
> counter to
> its own cache line to avoid conflicts with 'nr_requests' and other hot
> fields
> (see:
> https://patchwork.kernel.org/project/linux-scsi/patch/20260402074637.92417-3-sumit.saxena@xxxxxxxxxxxx/
> ).
>
> During the review, Bart shared his work, which eliminates the
> counter entirely by removing the fairness throttling. My testing confirmed
> that
> this approach resolved the performance issues and improved IOPS.
> This patch is part of a larger set, and I have reported the cumulative
> performance
> improvements in the cover letter.
So the problem is just the atomic operation accounting overhead? I
previously thought the device just really needed to consume all the tags
to hit performance.
|