Message-ID: <52D21226.4070306@lwfinger.net> (sfid-20140112_045551_987547_E28940F2)
Date: Sat, 11 Jan 2014 21:55:18 -0600
From: Larry Finger <Larry.Finger@lwfinger.net>
MIME-Version: 1.0
To: Ben Hutchings <ben@decadent.org.uk>
CC: linville@tuxdriver.com, linux-wireless@vger.kernel.org,
	netdev@vger.kernel.org, Stable <stable@vger.kernel.org>
Subject: Re: [PATCH 2/3] b43: Fix oops if firmware is not available
References: <1389469714-13040-1-git-send-email-Larry.Finger@lwfinger.net>	 <1389469714-13040-3-git-send-email-Larry.Finger@lwfinger.net> <1389497251.3720.40.camel@deadeye.wl.decadent.org.uk>
In-Reply-To: <1389497251.3720.40.camel@deadeye.wl.decadent.org.uk>
Content-Type: text/plain; charset=UTF-8; format=flowed
Sender: linux-wireless-owner@vger.kernel.org

On 01/11/2014 09:27 PM, Ben Hutchings wrote:
> On Sat, 2014-01-11 at 13:48 -0600, Larry Finger wrote:
>> On openSUSE systems, the script that installs the firmware for b43 also
>> unloads and reloads the driver. When the firmware was not previously
>> available, the driver has stalled at a wait_for_completion(). When the
>> unload routine releases that hold, the driver encounters structures
>> that have already been deleted and generates a fatal condition. When
>> the user does a manual restart, the file system cleanup frequently
>> results in the firmware files being deleted and the user is never able
>> to install the firmware. The fix is to change the wait_for_completion()
>> with a wait_for_completion_timeout() with a 60 second wait period.
>>
>> There is a potential race condition; however, the chances that less
>> than a minute has elapsed between the initial driver load and a
>> subsequent unload is very unlikely.
>
> A minute-long race is 'unlikely' to be hit?  Seriously?!

Ben,

If you force a reboot before the minute expires, nothing weird happens. The only 
race condition happens when the user has to log in, open a terminal, run a 
script that downloads 13.5 MiB of files from the Internet, and then executes the 
firmware extraction program. On my 10 Mbps external line and a 2 GHz CPU, that 
takes 32 s, plus any time to enter the password for a sudo operation. That was 
the basis for my conclusion that a race is unlikely.

What is the minimum time that should be allowed for a request_firmware_nowait() 
to respond? I know we had to go to asynchronous fw loading because the 
synchronous  version would timeout at 30 s.

Larry