From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ferruh Yigit Subject: Re: [dpdk-stable] [PATCH] net/vmxnet3: keep link state consistent Date: Wed, 18 Apr 2018 17:23:19 +0100 Message-ID: References: <20180318014552.16703-1-3chas3@gmail.com> <7001889b-9a98-c353-ad9c-207bffc6869a@intel.com> <4064739.HSmrWIpKYf@xps> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Cc: Thomas Monjalon , Chas Williams <3chas3@gmail.com>, dev@dpdk.org, stable@dpdk.org To: Chas Williams Return-path: In-Reply-To: <4064739.HSmrWIpKYf@xps> Content-Language: en-US List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" On 4/17/2018 9:24 PM, Thomas Monjalon wrote: > 17/04/2018 21:25, Ferruh Yigit: >> On 4/5/2018 4:01 PM, Chas Williams wrote: >>> On Thu, Apr 5, 2018 at 6:03 AM, Thomas Monjalon wrote: >>>> 20/03/2018 15:12, Ferruh Yigit: >>>>> On 3/18/2018 1:45 AM, Chas Williams wrote: >>>>>> From: Chas Williams >>>>>> >>>>>> The vmxnet3 never attempts link speed negotiation. As a virtual device >>>>>> the link speed is vague at best. However, it is important for certain >>>>>> applications, like bonding, to see a consistent link_status. 802.3ad >>>>>> requires that only links of the same cost (link speed) be enslaved. >>>>>> Keeping the link status consistent in vmxnet3 avoids races with bonding >>>>>> enslavement. >>>> >>>> I don't understand the issue. >>>> Are you sure it is not an issue in bonding? >>> >>> 802.3ad "requires" you to bond together links of the same speed and duplex. The >>> primary reason for this (or so I gather) is to ensure that the >>> spanning-tree cost for >>> each port is the same. If you fail from one link to another, you >>> don't want a spanning >>> tree reconfiguration. >>> >>> The problem exists in general for most of the PMDs -- see >>> https://dpdk.org/ml/archives/dev/2018-April/094696.html >>> >>> The problem is more vexing for AUTONEG and bonding. I am still thinking about >>> that. You don't know until you go to activate the slave and bonding >>> only makes its >>> check during the setup phase. So for virtual adapters and bonding, not using >>> AUTONEG makes more sense because it is just easier to handle. >>> >>>> >>>> About the right value to set for virtual PMDs, I don't know, both are fakes. >>>> I thought that AUTONEG better convey the vague link speed you describe. >>> >>> It's not vague. There is no negotiation of any sort. The link speed >>> (and therefore cost) >>> of the link is fixed. While the particular rate you get from the >>> adapter depends >>> on a number of factors, the link speed isn't going to change. The >>> adapter is not >>> going to change the link speed from 10G to 1G or change from full duplex to half >>> duplex. >> >> Hi Chas, Thomas, >> >> What is the latest status of this patch? Is it agreed to convert link_autoneg to >> ETH_LINK_FIXED for following PMDs [1]? >> >> [1] >> pcap >> softnic >> vmxnet3 > > Yes, OK for ETH_LINK_FIXED. Thanks for clarification. Hi Chas, Do you prefer to re-make the patch to cover them all, or should I get this one as start, which one do you prefer? Thanks, ferruh > > >>>>>> Author: Thomas Monjalon >>>>>> Date: Fri Jan 5 18:38:55 2018 +0100 >>>>>> >>>>>> Fixes: 1e3a958f40b3 ("ethdev: fix link autonegotiation value") >>>>>> Cc: stable@dpdk.org >>>>> >>>>> There were a few more PMDs [1] they have been updated from FIXED to AUTONEG with >>>>> above commit, do you think should we update them back to FIXED as well? >>>>> >>>>> [1] >>>>> pcap >>>>> softnic >>>>> vmxnet3 >>>> >>>> Yes, they all can be fixed/LINK_FIXED :) I guess >>>> >>>> >>>> >> >> > > > > >