netdev - RE: [PATCH/RFC net-next 2/3] devlink: Add new "max_vf

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 21 Sep 2022 01:47:37 +0000
From:   Yinjun Zhang <yinjun.zhang@...igine.com>
To:     Edward Cree <ecree.xilinx@...il.com>,
        Simon Horman <simon.horman@...igine.com>,
        David Miller <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>
CC:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        oss-drivers <oss-drivers@...igine.com>,
        Diana Wang <na.wang@...igine.com>,
        Nole Zhang <peng.zhang@...igine.com>
Subject: RE: [PATCH/RFC net-next 2/3] devlink: Add new "max_vf_queue" generic
 device param

On Tue, 20 Sep 2022 19:27:48 +0100, Edward Cree wrote:
> On 20/09/2022 16:14, Simon Horman wrote:
> > From: Peng Zhang <peng.zhang@...igine.com>
> >
> > VF max-queue-number is the MAX num of queues which the VF has.
> >
> > Add new device generic parameter to configure the max-queue-number
> > of the each VF to be generated dynamically.
> >
> > The string format is decided ad vendor specific. The suggested
> > format is ...-V-W-X-Y-Z, the V represents generating V VFs that
> > have 16 queues, the W represents generating W VFs that have 8
> > queues, and so on, the Z represents generating Z VFs that have
> > 1 queue.
> 
> I don't like this.
> If I'm correctly understanding, it hardcodes an assumption that
>  lower-numbered VFs will be the ones with more queues, and also

Usually all VFs have same max-q-num, so config like "...-0-N-0-.."
will be the most case. If you want some VFs have more queues
than the others, then yes, the lower-numbered VFs always have
more. We think it's OK, since the user can decide which VFs for
what use. If you need VFs with fewer queues, then use those
higher-numbered VFs.
And the format is just a recommendation, the parse process
is implemented in NIC driver, so it can be vendor specific.

>  makes it difficult to change a VF's max-queues on the fly.
> Why not instead have a per-VF operation to set that VF's max
>  queue count?  Ideally through the VF representor, perhaps as
>  an ethtool param/tunable, rather than devlink.  Then the
>  mechanism is flexible and makes no assumptions about policy.

The background of this configuration proposal is that we need a 
way to reallocate the queue resource that is shared by all VFs.
So it should be pre-configured before VFs are created, that's why
the configuration is per-PF or per-pcidev. We're not supposed to
change the max-q-number once the VF is created.

Thanks.

> 
> -ed