watchdog: watchdog_stop_on_reboot()

Commit Message

Paul Thomas Jan. 18, 2021, 5:39 p.m. UTC

Hello,

I'm going through our out-of-tree commits, and was wondering if there
is anything to be done with this one. Basically we remove both
watchdog_stop_on_reboot() calls (not sure why there are two to begin
with):

This seems to save us in certain conditions where the Software System
Reset fails[1]. I'm not sure if there's other use cases where you
really do want the watchdog to stop on reboot?

This is on an arm64 Zyny Ultrascale+ platform.

-Paul

[1] The one reproducible example of the Software System Reset failing
I have for our system is after a kernel oops from removing the
xilinx_dma module.
https://lore.kernel.org/lkml/CAD56B7d_9+0XuOBysxW=[e-mail]/

Comments

Guenter Roeck Jan. 18, 2021, 7:16 p.m. UTC | #1

On 1/18/21 9:39 AM, Paul Thomas wrote:
> Hello,
> 
> I'm going through our out-of-tree commits, and was wondering if there
> is anything to be done with this one. Basically we remove both
> watchdog_stop_on_reboot() calls (not sure why there are two to begin
> with):

Those were there from the beginning, only hand coded. One was implemented
using a shutdown notifier, one as shutdown callback. Over time both were
converted to use the watchdog API, without me or anyone else noticing
that the second call is redundant.

> --- a/drivers/watchdog/cadence_wdt.c
> +++ b/drivers/watchdog/cadence_wdt.c
> @@ -330,7 +330,6 @@ static int cdns_wdt_probe(struct platform_device *pdev)
> 
>         watchdog_init_timeout(cdns_wdt_device, wdt_timeout, dev);
>         watchdog_set_nowayout(cdns_wdt_device, nowayout);
> -       watchdog_stop_on_reboot(cdns_wdt_device);
>         watchdog_set_drvdata(cdns_wdt_device, wdt);
> 
>         wdt->clk = devm_clk_get(dev, NULL);
> @@ -359,7 +358,6 @@ static int cdns_wdt_probe(struct platform_device *pdev)
> 
>         spin_lock_init(&wdt->io_lock);
> 
> -       watchdog_stop_on_reboot(cdns_wdt_device);
>         watchdog_stop_on_unregister(cdns_wdt_device);
>         ret = devm_watchdog_register_device(dev, cdns_wdt_device);
>         if (ret)
> 
> This seems to save us in certain conditions where the Software System
> Reset fails[1]. I'm not sure if there's other use cases where you
> really do want the watchdog to stop on reboot?
> 

Good question. Unfortunately I don't have an answer. I am fine with removing both.

Guenter

> This is on an arm64 Zyny Ultrascale+ platform.
> 
> -Paul
> 
> [1] The one reproducible example of the Software System Reset failing
> I have for our system is after a kernel oops from removing the
> xilinx_dma module.
> https://lore.kernel.org/lkml/CAD56B7d_9+0XuOBysxW=[e-mail]/
>

Patch

--- a/drivers/watchdog/cadence_wdt.c
+++ b/drivers/watchdog/cadence_wdt.c
@@ -330,7 +330,6 @@  static int cdns_wdt_probe(struct platform_device *pdev)

        watchdog_init_timeout(cdns_wdt_device, wdt_timeout, dev);
        watchdog_set_nowayout(cdns_wdt_device, nowayout);
-       watchdog_stop_on_reboot(cdns_wdt_device);
        watchdog_set_drvdata(cdns_wdt_device, wdt);

        wdt->clk = devm_clk_get(dev, NULL);
@@ -359,7 +358,6 @@  static int cdns_wdt_probe(struct platform_device *pdev)

        spin_lock_init(&wdt->io_lock);

-       watchdog_stop_on_reboot(cdns_wdt_device);
        watchdog_stop_on_unregister(cdns_wdt_device);
        ret = devm_watchdog_register_device(dev, cdns_wdt_device);
        if (ret)