| Message ID | 20170216193053.5546-2-alexandre.belloni@free-electrons.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 02/16/2017 11:30 AM, Alexandre Belloni wrote: > The datasheet states: "When setting the WDDIS bit, and while it is set, the > fields WDV and WDD must not be modified." > > Ensure WDDIS is not set when changing the timeout and set it afterwards if > the watchdog was disabled. > > Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> > --- > Changes in v2: > - new patch > drivers/watchdog/sama5d4_wdt.c | 9 ++++++++- > 1 file changed, 8 insertions(+), 1 deletion(-) > > diff --git a/drivers/watchdog/sama5d4_wdt.c b/drivers/watchdog/sama5d4_wdt.c > index 2c6f5a70ae67..2a60251806d2 100644 > --- a/drivers/watchdog/sama5d4_wdt.c > +++ b/drivers/watchdog/sama5d4_wdt.c > @@ -90,11 +90,18 @@ static int sama5d4_wdt_set_timeout(struct watchdog_device *wdd, > u32 reg; > > reg = wdt_read(wdt, AT91_WDT_MR); > + > + if (reg & AT91_WDT_WDDIS) > + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); > + > reg &= ~AT91_WDT_WDV; > reg &= ~AT91_WDT_WDD; > reg |= AT91_WDT_SET_WDV(value); > reg |= AT91_WDT_SET_WDD(value); > - wdt_write(wdt, AT91_WDT_MR, reg); > + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); > + > + if (reg & AT91_WDT_WDDIS) > + wdt_write(wdt, AT91_WDT_MR, reg); > That means if the watchdog is running, the timeout would not be updated. It should be updated no matter if it is running or not. Guenter > wdd->timeout = timeout; > >
On 17/02/2017 at 06:40:33 -0800, Guenter Roeck wrote: > On 02/16/2017 11:30 AM, Alexandre Belloni wrote: > > The datasheet states: "When setting the WDDIS bit, and while it is set, the > > fields WDV and WDD must not be modified." > > > > Ensure WDDIS is not set when changing the timeout and set it afterwards if > > the watchdog was disabled. > > > > Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> > > --- > > Changes in v2: > > - new patch > > drivers/watchdog/sama5d4_wdt.c | 9 ++++++++- > > 1 file changed, 8 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/watchdog/sama5d4_wdt.c b/drivers/watchdog/sama5d4_wdt.c > > index 2c6f5a70ae67..2a60251806d2 100644 > > --- a/drivers/watchdog/sama5d4_wdt.c > > +++ b/drivers/watchdog/sama5d4_wdt.c > > @@ -90,11 +90,18 @@ static int sama5d4_wdt_set_timeout(struct watchdog_device *wdd, > > u32 reg; > > > > reg = wdt_read(wdt, AT91_WDT_MR); > > + > > + if (reg & AT91_WDT_WDDIS) > > + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); > > + > > reg &= ~AT91_WDT_WDV; > > reg &= ~AT91_WDT_WDD; > > reg |= AT91_WDT_SET_WDV(value); > > reg |= AT91_WDT_SET_WDD(value); > > - wdt_write(wdt, AT91_WDT_MR, reg); > > + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); > > + > > + if (reg & AT91_WDT_WDDIS) > > + wdt_write(wdt, AT91_WDT_MR, reg); > > > That means if the watchdog is running, the timeout would not be updated. > It should be updated no matter if it is running or not. > No, it is enabling the watchdog, then changing WDV and WDD and finally disabling the watchdog if necessary. So, WDV and WDD are always changed.
On 02/17/2017 07:16 AM, Alexandre Belloni wrote: > On 17/02/2017 at 06:40:33 -0800, Guenter Roeck wrote: >> On 02/16/2017 11:30 AM, Alexandre Belloni wrote: >>> The datasheet states: "When setting the WDDIS bit, and while it is set, the >>> fields WDV and WDD must not be modified." >>> >>> Ensure WDDIS is not set when changing the timeout and set it afterwards if >>> the watchdog was disabled. >>> >>> Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> >>> --- >>> Changes in v2: >>> - new patch >>> drivers/watchdog/sama5d4_wdt.c | 9 ++++++++- >>> 1 file changed, 8 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/watchdog/sama5d4_wdt.c b/drivers/watchdog/sama5d4_wdt.c >>> index 2c6f5a70ae67..2a60251806d2 100644 >>> --- a/drivers/watchdog/sama5d4_wdt.c >>> +++ b/drivers/watchdog/sama5d4_wdt.c >>> @@ -90,11 +90,18 @@ static int sama5d4_wdt_set_timeout(struct watchdog_device *wdd, >>> u32 reg; >>> >>> reg = wdt_read(wdt, AT91_WDT_MR); >>> + >>> + if (reg & AT91_WDT_WDDIS) >>> + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); >>> + >>> reg &= ~AT91_WDT_WDV; >>> reg &= ~AT91_WDT_WDD; >>> reg |= AT91_WDT_SET_WDV(value); >>> reg |= AT91_WDT_SET_WDD(value); >>> - wdt_write(wdt, AT91_WDT_MR, reg); >>> + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); >>> + >>> + if (reg & AT91_WDT_WDDIS) >>> + wdt_write(wdt, AT91_WDT_MR, reg); >>> >> That means if the watchdog is running, the timeout would not be updated. >> It should be updated no matter if it is running or not. >> > > No, it is enabling the watchdog, then changing WDV and WDD and finally > disabling the watchdog if necessary. So, WDV and WDD are always changed. > You are correct. Sorry for the noise. Seems odd that the watchdog must be _running_ to change the timeout. Usually, if there is a restriction, it is the opposite. I hope this doesn't cause race conditions, where the watchdog fires immediately after being enabled due to a low timeout. Guenter
On 19/02/2017 at 08:57:35 -0800, Guenter Roeck wrote: > > > That means if the watchdog is running, the timeout would not be updated. > > > It should be updated no matter if it is running or not. > > > > > > > No, it is enabling the watchdog, then changing WDV and WDD and finally > > disabling the watchdog if necessary. So, WDV and WDD are always changed. > > > You are correct. Sorry for the noise. > > Seems odd that the watchdog must be _running_ to change the timeout. > Usually, if there is a restriction, it is the opposite. I hope this > doesn't cause race conditions, where the watchdog fires immediately > after being enabled due to a low timeout. > While it is difficult to reproduce, I can confirm it races and sometimes reset the SoC without any good reason. It doesn't matter whether it is disabled or not I've raised the issue at Atmel last Thursday so I don't have any answer yet.
On 02/19/2017 03:52 PM, Alexandre Belloni wrote: > On 19/02/2017 at 08:57:35 -0800, Guenter Roeck wrote: >>>> That means if the watchdog is running, the timeout would not be updated. >>>> It should be updated no matter if it is running or not. >>>> >>> >>> No, it is enabling the watchdog, then changing WDV and WDD and finally >>> disabling the watchdog if necessary. So, WDV and WDD are always changed. >>> >> You are correct. Sorry for the noise. >> >> Seems odd that the watchdog must be _running_ to change the timeout. >> Usually, if there is a restriction, it is the opposite. I hope this >> doesn't cause race conditions, where the watchdog fires immediately >> after being enabled due to a low timeout. >> > > While it is difficult to reproduce, I can confirm it races and sometimes > reset the SoC without any good reason. It doesn't matter whether it is > disabled or not > Outch :-(. > I've raised the issue at Atmel last Thursday so I don't have any answer > yet. > Please keep us in the loop. Thanks, Guenter
Hi, On 19/02/2017 at 20:46:13 -0800, Guenter Roeck wrote: > Please keep us in the loop. > I've sent a new series that solves the issues we were seeing. I've tied to document as much as possible and to make the driver as simple as possible.
diff --git a/drivers/watchdog/sama5d4_wdt.c b/drivers/watchdog/sama5d4_wdt.c index 2c6f5a70ae67..2a60251806d2 100644 --- a/drivers/watchdog/sama5d4_wdt.c +++ b/drivers/watchdog/sama5d4_wdt.c @@ -90,11 +90,18 @@ static int sama5d4_wdt_set_timeout(struct watchdog_device *wdd, u32 reg; reg = wdt_read(wdt, AT91_WDT_MR); + + if (reg & AT91_WDT_WDDIS) + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); + reg &= ~AT91_WDT_WDV; reg &= ~AT91_WDT_WDD; reg |= AT91_WDT_SET_WDV(value); reg |= AT91_WDT_SET_WDD(value); - wdt_write(wdt, AT91_WDT_MR, reg); + wdt_write(wdt, AT91_WDT_MR, reg & ~AT91_WDT_WDDIS); + + if (reg & AT91_WDT_WDDIS) + wdt_write(wdt, AT91_WDT_MR, reg); wdd->timeout = timeout;
The datasheet states: "When setting the WDDIS bit, and while it is set, the fields WDV and WDD must not be modified." Ensure WDDIS is not set when changing the timeout and set it afterwards if the watchdog was disabled. Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> --- Changes in v2: - new patch drivers/watchdog/sama5d4_wdt.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)