[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <56ae54cd-3ca5-4b31-4c9d-035fbf9d0474@grimberg.me>
Date: Thu, 9 Nov 2017 17:40:43 +0200
From: Sagi Grimberg <sagi@...mberg.me>
To: Saeed Mahameed <saeedm@...lanox.com>,
Thomas Gleixner <tglx@...utronix.de>
Cc: Jes Sorensen <jsorensen@...com>,
Tariq Toukan <tariqt@...lanox.com>,
Saeed Mahameed <saeedm@....mellanox.co.il>,
Networking <netdev@...r.kernel.org>,
Leon Romanovsky <leonro@...lanox.com>,
Kernel Team <kernel-team@...com>,
Christoph Hellwig <hch@....de>
Subject: Re: mlx5 broken affinity
>> Again, I think Jes or others can provide more information.
>
> Sagi, I believe Jes is not trying to argue about what initial affinity
> values you give to the driver, We have a very critical regression that
> is afflicting Live systems today and common tools that already exists
> in various distros, such as irqblanace which solely depends on
> smp_affinity sysfs entry which is now not write-able due to this
> regression. please see https://github.com/Irqbalance/irqbalance/blob/ma
> ster/activate.c
>
> Some users would like to have thier network traffic handled in some
> cores and free up other cores for other purposes, you just can't take
> that away from them.
>
> If revereting mlx5 patches would solve the issue, I am afraid that is
> the solution i am going to go with, until the regression is fixed.
Well, I completely agree with you Saeed, If this is causing regression
for mlx5 users and we can't find a way for managed interface to expose
this knob to userspace, I don't see any other choice as well.
Powered by blists - more mailing lists