[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87ca050f-5643-4b90-8768-1d624e367cac@gmail.com>
Date: Mon, 1 Apr 2024 09:54:26 +0300
From: Tariq Toukan <ttoukan.linux@...il.com>
To: Jakub Kicinski <kuba@...nel.org>, Saeed Mahameed <saeed@...nel.org>
Cc: "David S. Miller" <davem@...emloft.net>, Paolo Abeni <pabeni@...hat.com>,
Eric Dumazet <edumazet@...gle.com>, Saeed Mahameed <saeedm@...dia.com>,
netdev@...r.kernel.org, Tariq Toukan <tariqt@...dia.com>,
Gal Pressman <gal@...dia.com>, Leon Romanovsky <leonro@...dia.com>,
Carolina Jubran <cjubran@...dia.com>
Subject: Re: [net 06/10] net/mlx5: RSS, Block changing channels number when
RXFH is configured
On 29/03/2024 8:31, Jakub Kicinski wrote:
> On Tue, 26 Mar 2024 07:46:42 -0700 Saeed Mahameed wrote:
>> Changing the channels number after configuring the receive
>> flow hash indirection table may alter the RSS table size.
>> The previous configuration may no longer be compatible with
>> the new receive flow hash indirection table.
>>
>> Block changing the channels number when RXFH is configured.
>
> Do I understand correctly that this will block all set_channels
> calls after indir table changes?
Right.
> This may be a little too risky
> for a fix. Perhaps okay for net-next, but not a fix.
>
This fixes an issue introduced only in v6.7, not before that.
> I'd think that setting indir table and then increasing the number
> of channels is a pretty legit maneuver, or even best practice
> to allocate a queue outside of RSS.
>
> Is it possible to make a narrower change, only rejecting the truly
> problematic transitions?
>
The rationale of having a "single flow" or "single "logic" is to make it
simple, and achieve a fine user experience.
Otherwise, users would, for example, question why increasing the number
of channels (after setting the indir table) from 24 channels to 120
works, but doesn't work when trying with 130 channels, although max num
channels is much higher.
The required order looks pretty natural: first set the desired num of
channels, and only then set your indirection table.
At the end, there are pros and cons for each solution.
If you still strongly prefer narrowing it down only to the truly
problematic transitions, then we'll have no big issue in changing this.
Powered by blists - more mailing lists