From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ferruh Yigit Subject: Re: [PATCH] net/sfc: enable TSO by default Date: Fri, 20 Jan 2017 14:47:27 +0000 Message-ID: References: <1484841140-24117-1-git-send-email-arybchenko@solarflare.com> <18339814.hMek8iyv21@xps13> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: dev@dpdk.org To: Andrew Rybchenko , Thomas Monjalon Return-path: Received: from mga01.intel.com (mga01.intel.com [192.55.52.88]) by dpdk.org (Postfix) with ESMTP id 0E6E137A8 for ; Fri, 20 Jan 2017 15:47:30 +0100 (CET) In-Reply-To: List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" On 1/20/2017 2:37 PM, Andrew Rybchenko wrote: > On 01/20/2017 05:29 PM, Thomas Monjalon wrote: >> 2017-01-19 15:52, Andrew Rybchenko: >>> Signed-off-by: Andrew Rybchenko >>> --- >>> It is a mistake that TSO support is compiled out by default. >>> We would be happy to enable it by deafult, but strictly speaking >>> it is not a bug fix. >>> >>> Arguments to enable are: >>> - be more feature-rich (and user-friendly) in default config >>> - the most of internal testing is done with TSO enabled >>> >>> config/common_base | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/config/common_base b/config/common_base >>> index b9fb8e2..134e9b9 100644 >>> --- a/config/common_base >>> +++ b/config/common_base >>> @@ -261,7 +261,7 @@ CONFIG_RTE_LIBRTE_BNXT_PMD=y >>> # >>> CONFIG_RTE_LIBRTE_SFC_EFX_PMD=y >>> CONFIG_RTE_LIBRTE_SFC_EFX_DEBUG=n >>> -CONFIG_RTE_LIBRTE_SFC_EFX_TSO=n >>> +CONFIG_RTE_LIBRTE_SFC_EFX_TSO=y >> There should not be such option in the build-time configuration. >> Why keeping it (even enabled by default)? > > Initially it was introduced since packet rates with TSO compiled out > were slightly but noticeable better. Finally we have changed approach > and performance improvements planned in 17.05 do not require it. > I'll be happy to remove it completely, if there is no problems to > include it in 17.02 (it will be a bit bigger but functionally equal to > this one). If it will be just removing ifdef, with keeping same functionality, I wouldn't mind bigger patch.