From mboxrd@z Thu Jan 1 00:00:00 1970 From: Stephen Hemminger Subject: Re: [PATCH v5 00/15] common ethdev linkstatus functions Date: Wed, 17 Jan 2018 08:05:55 -0800 Message-ID: <20180117080555.5afd8c78@xeon-e3> References: <20180116183755.24542-1-stephen@networkplumber.org> <42f4f5a2-008d-c3e8-4c00-ed9ef59065c8@solarflare.com> <27f327e5-5632-69cc-feaa-10cf9384a701@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: Andrew Rybchenko , dev@dpdk.org To: Ferruh Yigit Return-path: Received: from mail-pl0-f67.google.com (mail-pl0-f67.google.com [209.85.160.67]) by dpdk.org (Postfix) with ESMTP id 6381C3257 for ; Wed, 17 Jan 2018 17:05:59 +0100 (CET) Received: by mail-pl0-f67.google.com with SMTP id bi12so9202768plb.6 for ; Wed, 17 Jan 2018 08:05:59 -0800 (PST) In-Reply-To: <27f327e5-5632-69cc-feaa-10cf9384a701@intel.com> 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 Wed, 17 Jan 2018 14:32:17 +0000 Ferruh Yigit wrote: > On 1/17/2018 7:56 AM, Andrew Rybchenko wrote: > > On 01/16/2018 09:37 PM, Stephen Hemminger wrote: > >> While reviewing drivers, noticed a lot of unnecessary > >> duplication of code in drivers for handling the eth_dev link status > >> information. While consolidating this, it also became obvious that > >> some drivers behave differently for no good reason. > >> > >> It also was a good chance to introduce atomic exchange primitives > >> in EAL because there are other places using cmpset where not > >> necessary (such as bonding). > >> > >> Mostly only compile tested only, don't have all of the hardware > >> available (except ixgbe and virtio) to test. > >> > >> Note: the eth_dev_link_update function return value is inconsistent > >> across drivers. Should be changed to be void. > > > > I would say "link_update" callback return value is inconsistent across > > drivers. I'm not sure which direction is right here: make it consistent > > or make it void. Also any changes in link information could be > > important. As I understand it should not happen without up/down, > > but bugs with loss of intermediate transitions are definitely possible. > > So, notifying about any changes in link information is definitely safer. > > May be not now. > > Again, why not return previous link status, it is simple enough to prevent > inconsistent usage. > > rte_eth_link_get() already discards the return value, so won't be a problem there. > > For the cases PMD would like know about link changes, they will need to > implement almost same link_update function with a return value, so why not use > existing link_update function? > > Like been in virtio, link_update() used in interrupt handler, and calls a > callback process if status changes. When link_update() return status changed to > void, I guess they will need to implement another version of the link_update > with return and use it. The interrupt and non-interrupt model are different. Also the driver internally may want to do something different, this is about the return value for dev_ops->link_update. The code in rte_eth_dev never used the return value. The bonding driver was expecting it to work but it doesn't. Anyway drivers shouldn't in general be directly calling other devices eth_dev_ops