[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170220173216.GD31264@wotan.suse.de>
Date: Mon, 20 Feb 2017 18:32:16 +0100
From: "Luis R. Rodriguez" <mcgrof@...nel.org>
To: "Grumbach, Emmanuel" <emmanuel.grumbach@...el.com>
Cc: "Luis R. Rodriguez" <mcgrof@...nel.org>,
"Berg, Johannes" <johannes.berg@...el.com>,
"Coelho, Luciano" <luciano.coelho@...el.com>,
"tj@...nel.org" <tj@...nel.org>,
"arjan@...ux.intel.com" <arjan@...ux.intel.com>,
"ming.lei@...onical.com" <ming.lei@...onical.com>,
"zajec5@...il.com" <zajec5@...il.com>,
"jeyu@...hat.com" <jeyu@...hat.com>,
"rusty@...tcorp.com.au" <rusty@...tcorp.com.au>,
"pmladek@...e.com" <pmladek@...e.com>,
"gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
linuxwifi <linuxwifi@...el.com>,
"linux-wireless@...r.kernel.org" <linux-wireless@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [RFC 1/5] iwlwifi: fix drv cleanup on opmode registration failure
On Sun, Feb 19, 2017 at 09:16:01AM +0000, Grumbach, Emmanuel wrote:
> > This should in theory fix a detangled drv from the drv list should either of the
> > opmode modules loaded and handled registration for the drv.
> >
> > The path of having the opmode registration deal with the drv opmode start is
> > actually the more common path. The other path, from the async callback is
> > rathe rare (1/8 or so times for me) -- it happens when the the opmode
> > driver's init routine completed prior to the driver's async callback opmode
> > start call.
>
> I'd claim it should never happen unless you have several devices on the system using the same
> opmode, or unless you do:
> modprobe iwlwifi #which will load iwl{d,m}vm
> rmmod iwl{d,m}vm #and do _not_ remove iwlwifi
> modprobe iwlwifi
That is indeed one way one can easily reproduce this. There are however other
ways too. Try a loop of
modprobe -r iwlmvm (which removes iwlwifi) followed by modprobe iwlmvm;
while this check for which path is taken, or better yet check if the
list of drvs is empty on opmode registration. Every now and then I see
the list is empty.
I have a feeling this is then also a rare rarely observed by your QA team
as well, so this code then is also stitching together a set of sequence
calls for both paths.
> > Signed-off-by: Luis R. Rodriguez <mcgrof@...nel.org>
> > ---
>
> Luca is OOO, but this looks fine to me.
Reviewed-by ?
Luis
Powered by blists - more mailing lists