From mboxrd@z Thu Jan  1 00:00:00 1970
From: Eric Dumazet <eric.dumazet@gmail.com>
Subject: Re: [PATCH net 2/2] bnx2x: fix napi poll return value for repoll
Date: Tue, 20 Jan 2015 07:44:37 -0800
Message-ID: <1421768677.4832.7.camel@edumazet-glaptop2.roam.corp.google.com>
References: <1421759776-376-1-git-send-email-_govind@gmx.com>
	 <1421759776-376-3-git-send-email-_govind@gmx.com>
	 <1421765480.4832.2.camel@edumazet-glaptop2.roam.corp.google.com>
	 <alpine.LNX.2.03.1501202027430.31736@ws.cisco>
Mime-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Cc: Willem de Bruijn <willemb@google.com>, davem@davemloft.net,
	netdev@vger.kernel.org, ariel.elior@qlogic.com, ssujith@cisco.com,
	benve@cisco.com
To: Govindarajulu Varadarajan <_govind@gmx.com>
Return-path: <netdev-owner@vger.kernel.org>
Received: from mail-ie0-f171.google.com ([209.85.223.171]:48017 "EHLO
	mail-ie0-f171.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1752130AbbATPoj (ORCPT
	<rfc822;netdev@vger.kernel.org>); Tue, 20 Jan 2015 10:44:39 -0500
Received: by mail-ie0-f171.google.com with SMTP id tr6so7329779ieb.2
        for <netdev@vger.kernel.org>; Tue, 20 Jan 2015 07:44:39 -0800 (PST)
In-Reply-To: <alpine.LNX.2.03.1501202027430.31736@ws.cisco>
Sender: netdev-owner@vger.kernel.org
List-ID: <netdev.vger.kernel.org>

On Tue, 2015-01-20 at 20:54 +0530, Govindarajulu Varadarajan wrote:
> On Tue, 20 Jan 2015, Eric Dumazet wrote:
> 
> > On Tue, 2015-01-20 at 18:46 +0530, Govindarajulu Varadarajan wrote:
> >> With the commit d75b1ade567ffab ("net: less interrupt masking in NAPI") napi repoll
> >> is done only when work_done == budget. When in busy_poll is we return 0 in
> >> napi_poll. We should return budget. Also do not return workdone > budget.
> >>
> >
> > I am not sure.
> >
> 
> This is based on f41281d02f8b94e136f78cb1b6a5d78182c222bd &
> 9dfa9a27b620640322588df399eb8f624b48d877
> 
> I do not know about bnx2x, but in enic driver, when busy_poll is enables
> rq clean is not happening. This is because, in napi_poll() when
> work_done < budget, we do not repoll. At this point, enic has disables rq intr
> and has not called napi_complete. Driver assumes that napi will repoll.
> Which does not happen.
> 
> Lot of drivers I have checked return full budget if they want to repoll.
> eg. mlx4_en_poll_rx_cq()

I was referring to the "workdone > budget" condition.

But yes, the fix about busy poll seems needed.