From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ilya Maximets <i.maximets@samsung.com>
Subject: Re: [PATCH v4] vhost: use SMP barriers instead of
	compiler ones.
Date: Mon, 21 Mar 2016 07:49:54 +0300
Message-ID: <56EF7D72.1050108@samsung.com>
References: <1456314438-4021-2-git-send-email-i.maximets@samsung.com>
 <1458303833-14815-1-git-send-email-i.maximets@samsung.com>
 <20160318124102.GV979@yliu-dev.sh.intel.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
Cc: dev@dpdk.org, Huawei Xie <huawei.xie@intel.com>,
 Dyasly Sergey <s.dyasly@samsung.com>
To: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Return-path: <dev-bounces@dpdk.org>
Received: from mailout1.w1.samsung.com (mailout1.w1.samsung.com
 [210.118.77.11]) by dpdk.org (Postfix) with ESMTP id 9176B28F2
 for <dev@dpdk.org>; Mon, 21 Mar 2016 05:49:57 +0100 (CET)
Received: from eucpsbgm1.samsung.com (unknown [203.254.199.244])
 by mailout1.w1.samsung.com
 (Oracle Communications Messaging Server 7.0.5.31.0 64bit (built May 5 2014))
 with ESMTP id <0O4D00I4VHF7G520@mailout1.w1.samsung.com> for dev@dpdk.org;
 Mon, 21 Mar 2016 04:49:55 +0000 (GMT)
In-reply-to: <20160318124102.GV979@yliu-dev.sh.intel.com>
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>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>



On 18.03.2016 15:41, Yuanhan Liu wrote:
> On Fri, Mar 18, 2016 at 03:23:53PM +0300, Ilya Maximets wrote:
>> Since commit 4c02e453cc62 ("eal: introduce SMP memory barriers") virtio
>> uses architecture dependent SMP barriers. vHost should use them too.
>>
>> Fixes: 4c02e453cc62 ("eal: introduce SMP memory barriers")
>>
>> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
>> ---
>>  lib/librte_vhost/vhost_rxtx.c | 7 ++++---
>>  1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/lib/librte_vhost/vhost_rxtx.c b/lib/librte_vhost/vhost_rxtx.c
>> index b4da665..859c669 100644
>> --- a/lib/librte_vhost/vhost_rxtx.c
>> +++ b/lib/librte_vhost/vhost_rxtx.c
>> @@ -315,7 +315,7 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
>>  			rte_prefetch0(&vq->desc[desc_indexes[i+1]]);
>>  	}
>>  
>> -	rte_compiler_barrier();
>> +	rte_smp_wmb();
>>  
>>  	/* Wait until it's our turn to add our buffer to the used ring. */
>>  	while (unlikely(vq->last_used_idx != res_start_idx))
>> @@ -565,7 +565,7 @@ virtio_dev_merge_rx(struct virtio_net *dev, uint16_t queue_id,
>>  
>>  		nr_used = copy_mbuf_to_desc_mergeable(dev, vq, start, end,
>>  						      pkts[pkt_idx]);
>> -		rte_compiler_barrier();
>> +		rte_smp_wmb();
>>  
>>  		/*
>>  		 * Wait until it's our turn to add our buffer
>> @@ -923,7 +923,8 @@ rte_vhost_dequeue_burst(struct virtio_net *dev, uint16_t queue_id,
>>  				sizeof(vq->used->ring[used_idx]));
>>  	}
>>  
>> -	rte_compiler_barrier();
>> +	rte_smp_wmb();
>> +	rte_smp_rmb();
> 
> rte_smp_mb?

rte_smp_mb() is a real mm_fence() on x86. And we don't need to synchronize reads with
writes here, only reads with reads and writes with writes. It is enough because next
increment uses read and write. Pair of barriers is better because it will not impact
on performance on x86.

Best regards, Ilya Maximets.