From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <bruce.richardson@intel.com>
Received: from mga03.intel.com (mga03.intel.com [134.134.136.65])
 by dpdk.org (Postfix) with ESMTP id 7A691595A
 for <dev@dpdk.org>; Mon,  2 Mar 2015 11:16:30 +0100 (CET)
Received: from fmsmga003.fm.intel.com ([10.253.24.29])
 by orsmga103.jf.intel.com with ESMTP; 02 Mar 2015 02:14:07 -0800
X-ExtLoop1: 1
X-IronPort-AV: E=Sophos;i="5.09,674,1418112000"; d="scan'208";a="461251009"
Received: from bricha3-mobl3.ger.corp.intel.com ([10.243.20.27])
 by FMSMGA003.fm.intel.com with SMTP; 02 Mar 2015 02:10:21 -0800
Received: by  (sSMTP sendmail emulation); Mon, 02 Mar 2015 10:16:25 +0025
Date: Mon, 2 Mar 2015 10:16:25 +0000
From: Bruce Richardson <bruce.richardson@intel.com>
To: "Wiles, Keith" <keith.wiles@intel.com>
Message-ID: <20150302101625.GA8520@bricha3-MOBL3>
References: <D117602F.1576B%keith.wiles@intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <D117602F.1576B%keith.wiles@intel.com>
Organization: Intel Shannon Ltd.
User-Agent: Mutt/1.5.23 (2014-03-12)
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] Do we need the refcnt set to zero again?
X-BeenThere: dev@dpdk.org
X-Mailman-Version: 2.1.15
Precedence: list
List-Id: patches and discussions about DPDK <dev.dpdk.org>
List-Unsubscribe: <http://dpdk.org/ml/options/dev>,
 <mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://dpdk.org/ml/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <http://dpdk.org/ml/listinfo/dev>,
 <mailto:dev-request@dpdk.org?subject=subscribe>
X-List-Received-Date: Mon, 02 Mar 2015 10:16:30 -0000

On Sat, Feb 28, 2015 at 06:08:16PM +0000, Wiles, Keith wrote:
> Looking that the code below does the rte_mbuf_refcnt_set(m,0) need to be present?
> 
> static inline struct rte_mbuf* __attribute__((always_inline))
> __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
> {
> __rte_mbuf_sanity_check(m, 0);
> 
> if (likely (rte_mbuf_refcnt_read(m) == 1) ||
> likely (rte_mbuf_refcnt_update(m, -1) == 0)) {
> 
> rte_mbuf_refcnt_set(m, 0);
> 
> /* if this is an indirect mbuf, then
> *  - detach mbuf
> *  - free attached mbuf segment
> */
> if (RTE_MBUF_INDIRECT(m)) {
> struct rte_mbuf *md = RTE_MBUF_FROM_BADDR(m->buf_addr);
> rte_pktmbuf_detach(m);
> if (rte_mbuf_refcnt_update(md, -1) == 0)
> __rte_mbuf_raw_free(md);
> }
> return(m);
> }
> return (NULL);
> }
> 
> It seems like the code could be this or did I miss a race-condition?

What you are really missing is the initial check for refcnt == 1. In the case
of the atomic refcnt, this allows us to skip the atomic decrement operation, 
which is very expensive, and instead just do a regular assignment of the refcnt
to zero, in the refcnt_set call.


/Bruce

> 
> static inline struct rte_mbuf* __attribute__((always_inline))
> __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
> {
> __rte_mbuf_sanity_check(m, 0);
> 
> /* The sanity check above should have checked for refcnt being zero */
> if ( likely (rte_mbuf_refcnt_update(m, -1) == 0 ) {
> 
> /* if this is an indirect mbuf, then
>  *  - detach mbuf
>  *  - free attached mbuf segment
>  */
> if (RTE_MBUF_INDIRECT(m)) {
> struct rte_mbuf *md = RTE_MBUF_FROM_BADDR(m->buf_addr);
> rte_pktmbuf_detach(m);
> if (rte_mbuf_refcnt_update(md, -1) == 0)
> __rte_mbuf_raw_free(md);
> }
> return(m);
> }
> return (NULL);
> }
>