From mboxrd@z Thu Jan  1 00:00:00 1970
From: "Tan, Jianfeng" <jianfeng.tan@intel.com>
Subject: Re: [PATCH] eal: make resource initialization more robust
Date: Mon, 29 Feb 2016 09:50:27 +0800
Message-ID: <56D3A3E3.1090807@intel.com>
References: <1453661393-85704-1-git-send-email-jianfeng.tan@intel.com>
 <1454066522-80045-1-git-send-email-jianfeng.tan@intel.com>
 <2119610.dqEXMAUdDm@xps13>
Mime-Version: 1.0
Content-Type: text/plain; charset=windows-1252; format=flowed
Content-Transfer-Encoding: 7bit
Cc: dev@dpdk.org
To: Thomas Monjalon <thomas.monjalon@6wind.com>
Return-path: <dev-bounces@dpdk.org>
Received: from mga04.intel.com (mga04.intel.com [192.55.52.120])
 by dpdk.org (Postfix) with ESMTP id BF6422BCC
 for <dev@dpdk.org>; Mon, 29 Feb 2016 02:50:43 +0100 (CET)
In-Reply-To: <2119610.dqEXMAUdDm@xps13>
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>

Hi Thomas,

On 2/29/2016 5:12 AM, Thomas Monjalon wrote:
> Hi,
>
> 2016-01-29 19:22, Jianfeng Tan:
>> Current issue: DPDK is not that friendly to container environment, which
>> caused by that it pre-alloc resource like cores and hugepages. But there
>> are this or that resource limitations, for examples, cgroup, rlimit,
>> cpuset, etc.
>>
>> For cores, this patch makes use of pthread_getaffinity_np to further
>> narrow down detected cores before parsing coremask (-c), corelist (-l),
>> and coremap (--lcores).
>>
>> For hugepages, this patch adds a recover mechanism to the case that
>> there are no that many hugepages can be used. It relys on a mem access
>> to fault-in hugepages, and if fails with SIGBUS, recover to previously
>> saved stack environment with siglongjmp().
> They are some interesting ideas.
> However, I am not sure a library should try to be so smart silently.
> It needs more feedback to decide wether it can be the default behaviour
> or an option.
>
> Please send coremask and hugepage mapping as separate patches as they
> are totally different and may be integrated separately.

Good advise, thanks! I'll do it.

And one more thing FYI, coremask using pthread_getaffinity_np() may have 
issue on some Linux versions or distros: it excludes isolcpus. This is 
reported by Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>, 
and I'm still working it out.

Thanks,
Jianfeng

>
> Thanks