Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <f53b1ea0c733b58a242fd18a3e6c97a7b00ed01e.camel@kernel.org>
Subject: Re: [PATCH] Revert "net: linkwatch: add check for netdevice being
 present to linkwatch_do_dev"
From:   Saeed Mahameed <saeed@kernel.org>
To:     Jakub Kicinski <kuba@kernel.org>
Cc:     David Miller <davem@davemloft.net>, hkallweit1@gmail.com,
        geert+renesas@glider.be, f.fainelli@gmail.com, andrew@lunn.ch,
        gaku.inami.xh@renesas.com, yoshihiro.shimoda.uh@renesas.com,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Date:   Thu, 24 Sep 2020 11:16:57 -0700
In-Reply-To: <20200924090311.745cac3b@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
References: <e6f50a85-aa25-5fb7-7fd2-158668d55378@gmail.com>
         <a7ff1afd2e1fc2232103ceb9aa763064daf90212.camel@kernel.org>
         <20200923.172125.1341776337290371000.davem@davemloft.net>
         <20200923.172349.872678515629678579.davem@davemloft.net>
         <2cf4178e970d2737e7ba866ebc83a7ec30ca8ad1.camel@kernel.org>
         <20200924090311.745cac3b@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.36.5 (3.36.5-1.fc32) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Thu, 2020-09-24 at 09:03 -0700, Jakub Kicinski wrote:
> On Wed, 23 Sep 2020 22:49:37 -0700 Saeed Mahameed wrote:
> > 2) Another problematic scenario which i see is repeated in many
> > drivers:
> > 
> > shutdown/suspend()
> >     rtnl_lock()
> >     netif_device_detach()//Mark !present;
> >     stop()->carrier_off()->linkwatch_event()
> >     // at this point device is still IFF_UP and !present
> >     // due to the early detach above..  
> >     rtnl_unlock();
> 
> Maybe we can solve this by providing drivers with a better helper for
> the suspend use case?
> 
> AFAIU netif_device_detach() is used by both IO errors and drivers
> willingly detaching the device during normal operation (e.g. for
> suspend).
> 
> Since the suspend path can sleep if we have a separate helper perhaps
> we could fire off the appropriate events synchronously, and
> quiescence
> the stack properly?

I was thinking something similar, a more heavy
weight netif_device_detach(), which will be used in all drivers suspend
flows.
 
1) clear IFF_UP
2) ndo_stop()
3) fire events
4) mark !present
...

5) suspend device


but went and sampled some drivers and found there are many variations
for using netif_device_detach it is not going to be a simple task.