[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <9ca1acdb-cc19-46d2-a0cf-1f2dc00678e6@linux.vnet.ibm.com>
Date: Fri, 26 May 2017 11:11:05 -0500
From: Thomas Falcon <tlfalcon@...ux.vnet.ibm.com>
To: David Miller <davem@...emloft.net>
Cc: netdev@...r.kernel.org, nfont@...ux.vnet.ibm.com,
jallen@...ux.vnet.ibm.com
Subject: Re: [PATCH net-next] ibmvnic: Enable TSO support
On 05/25/2017 01:49 PM, David Miller wrote:
> From: David Miller <davem@...emloft.net>
> Date: Thu, 25 May 2017 14:46:26 -0400 (EDT)
>
>> From: Thomas Falcon <tlfalcon@...ux.vnet.ibm.com>
>> Date: Wed, 24 May 2017 21:29:26 -0500
>>
>>> The feature is also enabled by a module parameter.
>>> This parameter is necessary because TSO can not easily be
>>> enabled or disabled in firmware without reinitializing the driver.
>> Sorry, this is unacceptable. When I say no module parameters,
>> I really really mean it.
>>
>> Users should not be burdoned with having to know a special knob for
>> every driver in order to adjust what is a generic feature.
>>
>> You'll have to find another way to accomodate this.
> Also, TSO helps without SG only because you haven't implemented
> support for xmit_more in this driver to decrease the number of
> doorball updates and VM enters.
>
> I bet if you added xmit_more support, TSO wouldn't give you much
> if any performance boost if you have to linearize.
>
Thanks for the feedback. I'm looking into providing that support, but for the time being we would like to continue with supporting TSO, without using a module parameter this time.
Powered by blists - more mailing lists