From mboxrd@z Thu Jan 1 00:00:00 1970 From: John Fastabend Subject: Re: [patch net-next v4 2/7] switchdev: allow caller to explicitly request attr_set as deferred Date: Mon, 12 Oct 2015 22:48:11 -0700 Message-ID: <561C9B1B.4080408@gmail.com> References: <1444672467-20621-1-git-send-email-jiri@resnulli.us> <1444672986-20709-1-git-send-email-jiri@resnulli.us> <561C8B39.20604@gmail.com> <20151013054543.GB2242@nanopsycho.orion> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: Scott Feldman , Netdev , "David S. Miller" , Ido Schimmel , Elad Raz , Florian Fainelli , Guenter Roeck , Vivien Didelot , "andrew@lunn.ch" , David Laight , "stephen@networkplumber.org" To: Jiri Pirko Return-path: Received: from mail-pa0-f46.google.com ([209.85.220.46]:34461 "EHLO mail-pa0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751895AbbJMFs3 (ORCPT ); Tue, 13 Oct 2015 01:48:29 -0400 Received: by padhy16 with SMTP id hy16so10448336pad.1 for ; Mon, 12 Oct 2015 22:48:28 -0700 (PDT) In-Reply-To: <20151013054543.GB2242@nanopsycho.orion> Sender: netdev-owner@vger.kernel.org List-ID: On 15-10-12 10:45 PM, Jiri Pirko wrote: > Tue, Oct 13, 2015 at 06:40:25AM CEST, john.fastabend@gmail.com wrote: >> On 15-10-12 07:52 PM, Scott Feldman wrote: >>> On Mon, Oct 12, 2015 at 11:03 AM, Jiri Pirko wrote: >>>> From: Jiri Pirko >>>> >>>> Caller should know if he can call attr_set directly (when holding RTNL) >>>> or if he has to defer the att_set processing for later. >>>> >>>> This also allows drivers to sleep inside attr_set and report operation >>>> status back to switchdev core. Switchdev core then warns if status is >>>> not ok, instead of silent errors happening in drivers. >>>> >>>> Signed-off-by: Jiri Pirko >>>> --- >>>> include/net/switchdev.h | 1 + >>>> net/bridge/br_stp.c | 3 +- >>>> net/switchdev/switchdev.c | 107 ++++++++++++++++++++++++---------------------- >>>> 3 files changed, 59 insertions(+), 52 deletions(-) >>>> >>>> diff --git a/include/net/switchdev.h b/include/net/switchdev.h >>>> index d2879f2..6b109e4 100644 >>>> --- a/include/net/switchdev.h >>>> +++ b/include/net/switchdev.h >>>> @@ -17,6 +17,7 @@ >>>> >>>> #define SWITCHDEV_F_NO_RECURSE BIT(0) >>>> #define SWITCHDEV_F_SKIP_EOPNOTSUPP BIT(1) >>>> +#define SWITCHDEV_F_DEFER BIT(2) >>>> >>>> struct switchdev_trans_item { >>>> struct list_head list; >>>> diff --git a/net/bridge/br_stp.c b/net/bridge/br_stp.c >>>> index db6d243de..80c34d7 100644 >>>> --- a/net/bridge/br_stp.c >>>> +++ b/net/bridge/br_stp.c >>>> @@ -41,13 +41,14 @@ void br_set_state(struct net_bridge_port *p, unsigned int state) >>>> { >>>> struct switchdev_attr attr = { >>>> .id = SWITCHDEV_ATTR_ID_PORT_STP_STATE, >>>> + .flags = SWITCHDEV_F_DEFER, >>>> .u.stp_state = state, >>>> }; >>>> int err; >>>> >>>> p->state = state; >>>> err = switchdev_port_attr_set(p->dev, &attr); >>>> - if (err && err != -EOPNOTSUPP) >>>> + if (err) >>> >>> This looks like a problem as now all other non-switchdev ports will >>> get an WARN in the log when STP state changes. We should only WARN if >>> there was an err and the err is not -EOPNOTSUPP. >>> >>>> br_warn(p->br, "error setting offload STP state on port %u(%s)\n", >>>> (unsigned int) p->port_no, p->dev->name); >>>> } >>> >>> >>> >>>> struct switchdev_attr_set_work { >>>> struct work_struct work; >>>> struct net_device *dev; >>>> @@ -183,14 +226,17 @@ static void switchdev_port_attr_set_work(struct work_struct *work) >>>> { >>>> struct switchdev_attr_set_work *asw = >>>> container_of(work, struct switchdev_attr_set_work, work); >>>> + bool rtnl_locked = rtnl_is_locked(); >>>> int err; >>>> >>>> - rtnl_lock(); >>>> - err = switchdev_port_attr_set(asw->dev, &asw->attr); >>>> + if (!rtnl_locked) >>>> + rtnl_lock(); >>> >>> I'm not following this change. If someone else has rtnl_lock, we'll >>> not wait to grab it here ourselves, and proceed as if we have the >>> lock. But what if that someone else releases the lock in the middle >>> of us doing switchdev_port_attr_set_now? Seems we want to >>> unconditionally wait and grab the lock. We need to block anything >>> from moving while we do the attr set. >>> >> >> Also an additional race between setting rtnl_locked and the if stmt >> and then grabbing the lock. There seems to be a something of pattern >> around this where other subsystems use a rtnl_trylock and if it fails >> do a restart/re-queue operation to retry. Looks like how you handle >> it in the team driver at least. > > No, this is for different case. This is for case someone calls > switchdev_flush_defererd holding the rtnl_lock. > OK rather than funky if stmt could you just do a rtnl_trylock() and put a comment explaining the reasoning?