From mboxrd@z Thu Jan  1 00:00:00 1970
From: Olivier MATZ <olivier.matz@6wind.com>
Subject: Re: [PATCH] kni: use bulk functions to allocate and free
 mbufs
Date: Wed, 11 Jan 2017 17:38:25 +0100
Message-ID: <20170111173825.7caecf62@glumotte.dev.6wind.com>
References: <1483048216-2936-1-git-send-email-s.vyazmitinov@brain4net.com>
 <20170111081759.7b1ee146@xeon-e3>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Cc: Sergey Vyazmitinov <s.vyazmitinov@brain4net.com>,
 olivier.matz@6wind.com, ferruh.yigit@intel.com, dev@dpdk.org
To: Stephen Hemminger <stephen@networkplumber.org>
Return-path: <dev-bounces@dpdk.org>
Received: from mail-wj0-f176.google.com (mail-wj0-f176.google.com
 [209.85.210.176]) by dpdk.org (Postfix) with ESMTP id E60E5F60E
 for <dev@dpdk.org>; Wed, 11 Jan 2017 17:38:33 +0100 (CET)
Received: by mail-wj0-f176.google.com with SMTP id ew7so93109537wjc.3
 for <dev@dpdk.org>; Wed, 11 Jan 2017 08:38:33 -0800 (PST)
In-Reply-To: <20170111081759.7b1ee146@xeon-e3>
List-Id: DPDK patches and discussions <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>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>

On Wed, 11 Jan 2017 08:17:59 -0800, Stephen Hemminger
<stephen@networkplumber.org> wrote:
> On Fri, 30 Dec 2016 04:50:16 +0700
> Sergey Vyazmitinov <s.vyazmitinov@brain4net.com> wrote:
> 
> >  /**
> > + * Free n packets mbuf back into its original mempool.
> > + *
> > + * Free each mbuf, and all its segments in case of chained
> > buffers. Each
> > + * segment is added back into its original mempool.
> > + *
> > + * @param mp
> > + *   The packets mempool.
> > + * @param mbufs
> > + *   The packets mbufs array to be freed.
> > + * @param n
> > + *   Number of packets.
> > + */
> > +static inline void rte_pktmbuf_free_bulk(struct rte_mempool *mp,
> > +		struct rte_mbuf **mbufs, unsigned n)
> > +{
> > +	struct rte_mbuf *mbuf, *m_next;
> > +	unsigned i;
> > +	for (i = 0; i < n; ++i) {
> > +		mbuf = mbufs[i];
> > +		__rte_mbuf_sanity_check(mbuf, 1);
> > +
> > +		mbuf = mbuf->next;
> > +		while (mbuf != NULL) {
> > +			m_next = mbuf->next;
> > +			rte_pktmbuf_free_seg(mbuf);
> > +			mbuf = m_next;
> > +		}
> > +	}
> > +	rte_mempool_put_bulk(mp, (void * const *)mbufs, n);
> > +}  
> 
> The mbufs may come from different pools. You need to handle that.

I have an implementation for that in an endless-work-in-progress
patchset:

 /**
+ * Free several mbufs segments.
+ *
+ * This function frees a table of mbufs, ensuring that each mbuf is
+ * returned into its original pool. It is the equivalent of calling
+ * rte_pktmbuf_free_seg() on all mbuf of the table.
+ *
+ * @param mbufs
+ *    Array of mbuf pointers.
+ * @param n
+ *    Array size.
+ */
+static inline void
+rte_pktmbuf_free_seg_bulk(struct rte_mbuf * const *m_tab, unsigned n)
+{
+       struct rte_mbuf *m;
+       struct rte_mbuf * const *start = NULL;
+       unsigned n_free = 0, i;
+       struct rte_mempool *free_pool = NULL;
+
+       for (i = 0; i < n; i++) {
+               m = m_tab[i];
+
+               if (__rte_pktmbuf_prefree_seg(m) == NULL) {
+                       if (n_free != 0)
+                               rte_mempool_put_bulk(free_pool,
+                                       (void * const *)start, n_free);
+
+                       free_pool = NULL;
+                       n_free = 0;
+                       continue;
+               }
+
+               if (unlikely(m->pool != free_pool)) {
+                       if (n_free != 0)
+                               rte_mempool_put_bulk(free_pool,
+                                       (void * const *)start, n_free);
+
+                       free_pool = m->pool;
+                       start = &m_tab[i];
+                       n_free = 1;
+               } else {
+                       n_free++;
+               }
+       }
+
+       if (n_free != 0)
+               rte_mempool_put_bulk(free_pool,
+                       (void * const *)start, n_free);
+}


In the same patch, I also remove the tx_free_bulk_mbuf() functions
that does almost the same job in specific drivers. Unfortunately,
this patch needs to be rebased and better tested, so it's not ready yet.


Regards,
Olivier