Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <689fcef44a12f33281dccfe35e56859bccd1c058.camel@sipsolutions.net>
Subject: Re: locking in wiphy_apply_custom_regulatory()
From:   Johannes Berg <johannes@sipsolutions.net>
To:     Arend van Spriel <arend.vanspriel@broadcom.com>
Cc:     linux-wireless <linux-wireless@vger.kernel.org>
Date:   Mon, 14 Feb 2022 10:17:38 +0100
In-Reply-To: <8c9879f5-fcfc-69e5-5803-e8b5b1fabfd0@broadcom.com>
References: <6562d8c3-27f6-490a-7732-6c300eb3aa64@broadcom.com>
         <a3d73398fa51b66e77d98cf1e883c72b66d6a3f4.camel@sipsolutions.net>
         <8c9879f5-fcfc-69e5-5803-e8b5b1fabfd0@broadcom.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.42.3 (3.42.3-1.fc35) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Mon, 2022-02-14 at 09:45 +0100, Arend van Spriel wrote:
> 
> Correct. Just not sure I understand why is has to be RTNL lock.

Well, does it? Now I forgot again from when I looked ;-)


Surely for nl80211_get_reg_do() for example it has to be, at least
today, though that is already protecting the regdom with RCU in the non-
wiphy case, so doesn't really need to use RTNL.

Similarly, nl80211_get_reg_dump() could use RCU.

So that's easy, but all the interaction e.g. with brcms_reg_notifier()
calling freq_reg_info() which uses it too, but then is called from
wiphy_update_regulatory() from e.g. update_all_wiphy_regulatory() only
with RTNL makes it all complicated ...


> So would wiphy mutex be sufficient. I guess my question is what is 
> protected by these lock in wiphy_apply_custom_regulatory() and is it 
> really necessary to have both.

See above.

Honestly, I wouldn't mind if wiphy mutex (or RCU) _was_ sufficient. But
we're not there, and the regulatory code is sufficiently complex and
called sufficiently infrequently that I just haven't bothered trying to
reduce its reliance on the RTNL.

> Just some experimental coding where I ended up calling 
> wiphy_apply_custom_regulatory() upon IFF_UP and hit deadlock because 
> RTNL was already taken. Anyway, that code already ended up in the 
> garbage bin, but wanted to ask anyway. Learning by asking (stupid) 
> questions ;-)
> 

Given the length and depth I had to go to to answer it, that really
couldn't be a stupid question :)

johannes