From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Message-ID: <555C1F88.90203@offcode.fi> Date: Wed, 20 May 2015 08:45:44 +0300 From: Timo Kokkonen MIME-Version: 1.0 To: Guenter Roeck , linux-arm-kernel@lists.infradead.org, linux-watchdog@vger.kernel.org, boris.brezillon@free-electrons.com, nicolas.ferre@atmel.com, alexandre.belloni@free-electrons.com CC: Wenyou.Yang@atmel.com Subject: Re: [PATCHv8 03/10] watchdog: core: Introduce default watchdog policy References: <1432023969-20736-1-git-send-email-timo.kokkonen@offcode.fi> <1432023969-20736-4-git-send-email-timo.kokkonen@offcode.fi> <555BDFD4.6050003@roeck-us.net> In-Reply-To: <555BDFD4.6050003@roeck-us.net> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 8bit List-ID: On 20.05.2015 04:13, Guenter Roeck wrote: > On 05/19/2015 01:26 AM, Timo Kokkonen wrote: >> Traditionally there have been no such thing as a policy for >> watchdogs. The watchdog timer was simply explicitly stopped at driver >> probe time and there was nothing else to do. Now that we have ability >> to work around HW disabilities in the watchdog core, it is time to >> introduce more policy options. >> >> The default option is to shut down the watchdog driver, as >> before. This ensures backward compatibility with current kernels and >> hardware that might have a watchdog, but user has not set up a >> watchdog daemon at all. >> >> A new option is introduced to keep the watchdog running or start it up >> at boot up. This is useful on many production systems that rely on the >> watchdog to reboot the device in case a crash. With this option even >> early kernel crashes or early user space crashes will lead to a reboot >> even though watchdog daemon has not opened and started the watchdog >> device yet. >> >> The third introduced option is to not touch the hardware at all in >> case bootloader have made an intelligent decision about the watchdog >> state and does not know how to tell kernel about it. The watchdog >> state is not touched until userspace takes over it. >> >> Signed-off-by: Timo Kokkonen > > Dislike. I don't think it is a good idea to have configuration options > like this. This means users will have to deal with those "default" > options for an entire distribution, which I don't think is desirable. > Maybe for you, but not for everyone, and definitely not for me. The configuration option was suggested by Uwe Kleine-König. I think it makes sense as there would not be otherwise any other way to enable the early-timeout-sec as device tree. I would think that for example in some server environment it would be good to enable watchdogs as early as possible and there probably are no device tree available. > I think you are trying to do too much in your patch set. > Can we possibly focus on getting early timeout to an acceptable form > instead of trying to do everything at the same time. I would be happy to do it the way that guarantees the inclusion, but I feel that I am getting conflicting feedback from different people giving review. I appreciate any concrete guidance you can give about the direction I should go with the patch set. Thanks, -Timo