[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210310202327.GA237124@roeck-us.net>
Date: Wed, 10 Mar 2021 12:23:27 -0800
From: Guenter Roeck <linux@...ck-us.net>
To: Sai Prakash Ranjan <saiprakash.ranjan@...eaurora.org>
Cc: Andy Gross <agross@...nel.org>,
Bjorn Andersson <bjorn.andersson@...aro.org>,
Douglas Anderson <dianders@...omium.org>,
Wim Van Sebroeck <wim@...ux-watchdog.org>,
linux-watchdog@...r.kernel.org, Stephen Boyd <swboyd@...omium.org>,
Matthias Kaehlcke <mka@...omium.org>,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
linux-arm-msm@...r.kernel.org
Subject: Re: [PATCH] watchdog: qcom: Move suspend/resume to
suspend_late/resume_early
On Thu, Mar 11, 2021 at 01:50:04AM +0530, Sai Prakash Ranjan wrote:
> During suspend/resume usecases and tests, it is common to see issues
> such as lockups either in suspend path or resume path because of the
> bugs in the corresponding device driver pm handling code. In such cases,
> it is important that watchdog is active to make sure that we either
> receive a watchdog pretimeout notification or a bite causing reset
> instead of a hang causing us to hard reset the machine.
>
> There are good reasons as to why we need this because:
>
> * We can have a watchdog pretimeout governor set to panic in which
> case we can have a backtrace which would help identify the issue
> with the particular driver and cause a normal reboot.
>
> * Even in case where there is no pretimeout support, a watchdog
> bite is still useful because some firmware has debug support to dump
> CPU core context on watchdog bite for post-mortem analysis.
>
> * One more usecase which comes to mind is of warm reboot. In case we
> hard reset the target, a cold reboot could be induced resulting in
> lose of ddr contents thereby losing all the debug info.
>
> Currently, the watchdog pm callback just invokes the usual suspend
> and resume callback which do not have any special ordering in the
> sense that a watchdog can be suspended before the buggy device driver
> suspend callback and watchdog resume can happen after the buggy device
> driver resume callback. This would mean that the watchdog will not be
> active when the buggy driver cause the lockups thereby hanging the
> system. So to make sure this doesn't happen, move the watchdog pm to
> use late/early system pm callbacks which will ensure that the watchdog
> is suspended late and resumed early so that it can catch such issues.
>
> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@...eaurora.org>
Reviewed-by: Guenter Roeck <linux@...ck-us.net>
> ---
> drivers/watchdog/qcom-wdt.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/watchdog/qcom-wdt.c b/drivers/watchdog/qcom-wdt.c
> index e38a87ffe5f5..0d2209c5eaca 100644
> --- a/drivers/watchdog/qcom-wdt.c
> +++ b/drivers/watchdog/qcom-wdt.c
> @@ -329,7 +329,9 @@ static int __maybe_unused qcom_wdt_resume(struct device *dev)
> return 0;
> }
>
> -static SIMPLE_DEV_PM_OPS(qcom_wdt_pm_ops, qcom_wdt_suspend, qcom_wdt_resume);
> +static const struct dev_pm_ops qcom_wdt_pm_ops = {
> + SET_LATE_SYSTEM_SLEEP_PM_OPS(qcom_wdt_suspend, qcom_wdt_resume)
> +};
>
> static const struct of_device_id qcom_wdt_of_table[] = {
> { .compatible = "qcom,kpss-timer", .data = &match_data_apcs_tmr },
> --
> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
> of Code Aurora Forum, hosted by The Linux Foundation
>
Powered by blists - more mailing lists