Re: [PATCH v4] watchdog: qcom: support pre-timeout when the bark irq is available

[Guenter Roeck <linux@roeck-us.net>]

On Thu, Sep 05, 2019 at 11:34:03PM +0200, Jorge Ramirez wrote:
> On 9/5/19 23:19, Guenter Roeck wrote:
> > On Thu, Sep 05, 2019 at 11:00:35PM +0200, Jorge Ramirez-Ortiz wrote:
> >> Use the bark interrupt as the pre-timeout notifier whenever this
> >> interrupt is available.
> >>
> >> By default, the pretimeout notification shall occur one second earlier
> >> than the timeout.
> >>
> >> Signed-off-by: Jorge Ramirez-Ortiz <jorge.ramirez-ortiz@linaro.org>
> >> ---
> >>  v4:
> >>      address Guenter Roeck comments as follows:
> >>        remove unnecessary include and private variable
> >>        provide macro for WDT EN register values
> >>        use pretimeout as per its API intent
> >>        handle EPROBE_DEFER on get_irq
> >>      also:
> >>        handle the irq registration as done in pm8916_wdt.c
> >>  v3:
> >>     remove unnecesary variable added to private.
> >>
> >>  v2:
> >>     register the pre-timeout notifier instead.
> >>
> >>  v1:
> >>  drivers/watchdog/qcom-wdt.c | 64 ++++++++++++++++++++++++++++++++++---
> >>  1 file changed, 59 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/drivers/watchdog/qcom-wdt.c b/drivers/watchdog/qcom-wdt.c
> >> index 7be7f87be28f..0f1d29eeb81d 100644
> >> --- a/drivers/watchdog/qcom-wdt.c
> >> +++ b/drivers/watchdog/qcom-wdt.c
> >> @@ -10,6 +10,7 @@
> >>  #include <linux/platform_device.h>
> >>  #include <linux/watchdog.h>
> >>  #include <linux/of_device.h>
> >> +#include <linux/interrupt.h>
> >>  
> >>  enum wdt_reg {
> >>  	WDT_RST,
> >> @@ -19,6 +20,9 @@ enum wdt_reg {
> >>  	WDT_BITE_TIME,
> >>  };
> >>  
> >> +#define QCOM_WDT_ENABLE		BIT(0)
> >> +#define QCOM_WDT_ENABLE_IRQ	BIT(1)
> >> +
> > 
> > Using BIT() requires "#include <linux/bits.h>".
> 
> do you want it explicitly in the file even if it builds?
> 

May I kindly suggest to read Documentation/process/submit-checklist.rst ?

> > 
> >>  static const u32 reg_offset_data_apcs_tmr[] = {
> >>  	[WDT_RST] = 0x38,
> >>  	[WDT_EN] = 0x40,
> >> @@ -54,15 +58,38 @@ struct qcom_wdt *to_qcom_wdt(struct watchdog_device *wdd)
> >>  	return container_of(wdd, struct qcom_wdt, wdd);
> >>  }
> >>  
> >> +static inline int qcom_get_enable(struct watchdog_device *wdd)
> >> +{
> >> +	int enable = QCOM_WDT_ENABLE;
> >> +
> >> +	if (wdd->info->options & WDIOF_PRETIMEOUT)
> >> +		enable |= QCOM_WDT_ENABLE_IRQ;
> >> +
> > 
> > Again, the condition needs to be that pretimeout != 0,
> > not that it is supported.
> 
> no I dont think so. doing that would propagate a possible error in some
> pretimeout setup code which would end up enabling an interrupt when it
> shouldnt. so I dont think that doing that would be correct.
> 
If the pretimeout setup code is buggy, it needs to be fixed.

> The interrupt should only be enabled if WDIOF_PRETIMEOUT is configured
> (independently of the pretimeout value); as a matter of fact, if
> pretimeout is 0, the interrupt will trigger at the same time than bark
> (which is what the original code used to do).
> 
The original code did not set bit 1 of the WDT_EN register,
and it did not set the bark time.

> so I'd rather keep this condition unless you strongly oppose to it.
> 

Please feel free to petition  to Wim.

> > 
> >> +	return enable;
> >> +}
> >> +
> >> +static irqreturn_t qcom_wdt_isr(int irq, void *arg)
> >> +{
> >> +	struct watchdog_device *wdd = arg;
> >> +
> >> +	watchdog_notify_pretimeout(wdd);
> >> +
> >> +	return IRQ_HANDLED;
> >> +}
> >> +
> >>  static int qcom_wdt_start(struct watchdog_device *wdd)
> >>  {
> >>  	struct qcom_wdt *wdt = to_qcom_wdt(wdd);
> >> +	unsigned int bark = wdd->timeout;
> >> +
> >> +	if (wdd->pretimeout)
> >> +		bark = bark - wdd->pretimeout;
> > 
> > The if() just adds code and doesn't otherwise do any good.
> 
> yeah, was just for clarity and it is surely removed by the compiler. but
> sure will remove
> 
> > 
> >>  
> >>  	writel(0, wdt_addr(wdt, WDT_EN));
> >>  	writel(1, wdt_addr(wdt, WDT_RST));
> >> -	writel(wdd->timeout * wdt->rate, wdt_addr(wdt, WDT_BARK_TIME));
> >> +	writel(bark * wdt->rate, wdt_addr(wdt, WDT_BARK_TIME));
> >>  	writel(wdd->timeout * wdt->rate, wdt_addr(wdt, WDT_BITE_TIME));
> >> -	writel(1, wdt_addr(wdt, WDT_EN));
> >> +	writel(qcom_get_enable(wdd), wdt_addr(wdt, WDT_EN));
> >>  	return 0;
> >>  }
> >>  
> >> @@ -89,6 +116,13 @@ static int qcom_wdt_set_timeout(struct watchdog_device *wdd,
> >>  	return qcom_wdt_start(wdd);
> >>  }
> >>  
> >> +static int qcom_wdt_set_pretimeout(struct watchdog_device *wdd,
> >> +				   unsigned int timeout)
> >> +{
> >> +	wdd->pretimeout = timeout;
> >> +	return qcom_wdt_start(wdd);
> >> +}
> >> +
> >>  static int qcom_wdt_restart(struct watchdog_device *wdd, unsigned long action,
> >>  			    void *data)
> >>  {
> >> @@ -105,7 +139,7 @@ static int qcom_wdt_restart(struct watchdog_device *wdd, unsigned long action,
> >>  	writel(1, wdt_addr(wdt, WDT_RST));
> >>  	writel(timeout, wdt_addr(wdt, WDT_BARK_TIME));
> >>  	writel(timeout, wdt_addr(wdt, WDT_BITE_TIME));
> >> -	writel(1, wdt_addr(wdt, WDT_EN));
> >> +	writel(qcom_get_enable(wdd), wdt_addr(wdt, WDT_EN));
> >>  
> >>  	/*
> >>  	 * Actually make sure the above sequence hits hardware before sleeping.
> >> @@ -121,6 +155,7 @@ static const struct watchdog_ops qcom_wdt_ops = {
> >>  	.stop		= qcom_wdt_stop,
> >>  	.ping		= qcom_wdt_ping,
> >>  	.set_timeout	= qcom_wdt_set_timeout,
> >> +	.set_pretimeout	= qcom_wdt_set_pretimeout,
> >>  	.restart        = qcom_wdt_restart,
> >>  	.owner		= THIS_MODULE,
> >>  };
> >> @@ -133,6 +168,15 @@ static const struct watchdog_info qcom_wdt_info = {
> >>  	.identity	= KBUILD_MODNAME,
> >>  };
> >>  
> >> +static const struct watchdog_info qcom_wdt_pt_info = {
> >> +	.options	= WDIOF_KEEPALIVEPING
> >> +			| WDIOF_MAGICCLOSE
> >> +			| WDIOF_SETTIMEOUT
> >> +			| WDIOF_PRETIMEOUT
> >> +			| WDIOF_CARDRESET,
> >> +	.identity	= KBUILD_MODNAME,
> >> +};
> >> +
> >>  static void qcom_clk_disable_unprepare(void *data)
> >>  {
> >>  	clk_disable_unprepare(data);
> >> @@ -146,7 +190,7 @@ static int qcom_wdt_probe(struct platform_device *pdev)
> >>  	struct device_node *np = dev->of_node;
> >>  	const u32 *regs;
> >>  	u32 percpu_offset;
> >> -	int ret;
> >> +	int irq, ret;
> >>  
> >>  	regs = of_device_get_match_data(dev);
> >>  	if (!regs) {
> >> @@ -204,7 +248,17 @@ static int qcom_wdt_probe(struct platform_device *pdev)
> >>  		return -EINVAL;
> >>  	}
> >>  
> >> -	wdt->wdd.info = &qcom_wdt_info;
> >> +	irq = platform_get_irq(pdev, 0);
> >> +	if (irq > 0) {
> >> +		if (devm_request_irq(dev, irq, qcom_wdt_isr,
> >> +				     IRQF_TRIGGER_RISING, "wdt_bark",
> >> +				     &wdt->wdd))
> >> +			irq = 0;
> >> +	} else if (irq == -EPROBE_DEFER)
> >> +		return -EPROBE_DEFER;
> >> +
> >> +	wdt->wdd.info = irq > 0 ? &qcom_wdt_pt_info : &qcom_wdt_info;
> >> +	wdt->wdd.pretimeout = irq > 0 ? 1 : 0;
> > 
> > Why repeat the conditional ? It seems to me that something like
> > 
> > 	wdt->wdd.info = &qcom_wdt_info;
> > 	...
> > 	if (irq > 0) {
> > 		wdt->wdd.info = &qcom_wdt_pt_info;
> > 		wdt->wdd.pretimeout = 1;
> > 		...
> > 	}
> > 
> > would be much easier and avoid the repeated conditionals.
> 
> I agree. will change.
> 
> > 
> >>  	wdt->wdd.ops = &qcom_wdt_ops;
> >>  	wdt->wdd.min_timeout = 1;
> >>  	wdt->wdd.max_timeout = 0x10000000U / wdt->rate;
> >> -- 
> >> 2.23.0
> >>
> > 
>