Hi Greg,

Thanks for the review.

> We should use a vendor prefix for this property "adi,sec" so that it
> will be the same as what the kernel will require. That way we will be
> able to switch to OF_UPSTREAM without a driver change once support
> lands in the kernel.

Makes sense. I'll rename it to "adi,secid" (the prefixed form of the
current name) in v2 so it matches what the kernel binding will use.

> I think you should just read it directly into the priv struct without
> a temporary, i.e. dev_read_u32(dev, "secid", &priv->secid).

Agreed, will do that in v2.

> Also, what do you think about using dev_read_u32_default() with a
> default of 3 (appropriately documented) so that any users with
> modified or custom board device trees will not be surprised by this if
> they update to the latest?

Fair point on not surprising existing users. My only hesitation is that
the SEC ID isn't really a value with a sensible universal default. It's
a fixed per-SoC hardware constant. It's 3 on the current SC5xx parts, but
on the upcoming ADSP SoC the watchdog maps to different SEC ID, for example.

Defaulting to 3 keeps existing SC5xx device trees working (they all want
3 anyway), so it does address the custom-DT case you raise. The trade-off
is that a future SoC whose DT omits the property would then probe
"successfully" while programming the wrong SEC source and silently
breaking the watchdog->reset path, rather than failing at probe.

I'd lean slightly towards keeping it required so a misconfiguration is caught
early. But I don't feel strongly, if you'd prefer the default for the
custom-DT case,
I'm okay to use dev_read_u32_default() with a comment documenting that 3 is
the SC5xx value. Which would you prefer?

Thanks,
Ozan


On Thu, Jul 30, 2026 at 7:44 PM Greg Malysa <[email protected]> wrote:
>
> Hi Ozan,
>
> On Thu, Jul 30, 2026 at 12:38 PM Ozan Durgut <[email protected]> wrote:
> >
> > From: UtsavAgarwalADI <[email protected]>
> >
> > The ADI watchdog drives a reset through a Security (SEC) controller
> > fault source. The source index was hardcoded to 3 in adi_wdt_start(),
> > which is only correct for the SC5xx parts that happen to use that
> > index.
> >
> > Read the index from a new "secid" device tree property so boards that
> > route the watchdog fault to a different SEC source can configure it.
> >
> > Set secid = <3> in the SC57x, SC58x, SC594 and SC598 device trees to
> > match the value they relied on before.
> >
> > Co-developed-by: Ozan Durgut <[email protected]>
> > Signed-off-by: Ozan Durgut <[email protected]>
> > Signed-off-by: UtsavAgarwalADI <[email protected]>
> > ---
> >  arch/arm/dts/sc57x.dtsi     | 1 +
> >  arch/arm/dts/sc58x.dtsi     | 1 +
> >  arch/arm/dts/sc594-som.dtsi | 1 +
> >  arch/arm/dts/sc598-som.dtsi | 1 +
> >  drivers/watchdog/adi_wdt.c  | 9 ++++++++-
> >  5 files changed, 12 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/arm/dts/sc57x.dtsi b/arch/arm/dts/sc57x.dtsi
> > index e4cc612959f..794e80c3cf1 100644
> > --- a/arch/arm/dts/sc57x.dtsi
> > +++ b/arch/arm/dts/sc57x.dtsi
> > @@ -113,6 +113,7 @@
> >
> >  &wdog {
> >         clocks = <&clk ADSP_SC57X_CLK_CGU0_SCLK0>;
> > +       secid = <3>;
> >  };
> >
> >  &eth0 {
> > diff --git a/arch/arm/dts/sc58x.dtsi b/arch/arm/dts/sc58x.dtsi
> > index 7b07589de47..d4008ba7988 100644
> > --- a/arch/arm/dts/sc58x.dtsi
> > +++ b/arch/arm/dts/sc58x.dtsi
> > @@ -158,6 +158,7 @@
> >
> >  &wdog {
> >         clocks = <&clk ADSP_SC58X_CLK_CGU0_SCLK0>;
> > +       secid = <3>;
> >  };
> >
> >  &eth0 {
> > diff --git a/arch/arm/dts/sc594-som.dtsi b/arch/arm/dts/sc594-som.dtsi
> > index 1c2adc601dd..7a7167fbbba 100644
> > --- a/arch/arm/dts/sc594-som.dtsi
> > +++ b/arch/arm/dts/sc594-som.dtsi
> > @@ -67,6 +67,7 @@
> >
> >  &wdog {
> >         clocks = <&clk ADSP_SC594_CLK_CGU0_SCLK0>;
> > +       secid = <3>;
> >  };
> >
> >  &i2c0 {
> > diff --git a/arch/arm/dts/sc598-som.dtsi b/arch/arm/dts/sc598-som.dtsi
> > index bc212ef25cb..bca3ff765d6 100644
> > --- a/arch/arm/dts/sc598-som.dtsi
> > +++ b/arch/arm/dts/sc598-som.dtsi
> > @@ -124,6 +124,7 @@
> >
> >  &wdog {
> >         clocks = <&clk ADSP_SC598_CLK_CGU0_SCLK0>;
> > +       secid = <3>;
> >  };
> >
>
> We should use a vendor prefix for this property "adi,sec" so that it
> will be the same as what the kernel will require. That way we will be
> able to switch to OF_UPSTREAM without a driver change once support
> lands in the kernel.
>
> >  &i2c0 {
> > diff --git a/drivers/watchdog/adi_wdt.c b/drivers/watchdog/adi_wdt.c
> > index 7d7cf98b55e..8e7ff958121 100644
> > --- a/drivers/watchdog/adi_wdt.c
> > +++ b/drivers/watchdog/adi_wdt.c
> > @@ -36,6 +36,7 @@ struct adi_wdt_priv {
> >         void __iomem *rcu_base;
> >         void __iomem *sec_base;
> >         void __iomem *wdt_base;
> > +       u32 secid;
> >         struct clk clock;
> >  };
> >
> > @@ -72,7 +73,7 @@ static int adi_wdt_start(struct udevice *dev, u64 
> > timeout_ms, ulong flags)
> >         iowrite32(0xc1, priv->sec_base + SEC_FCTL);
> >
> >         /* enable SEC fault source for watchdog0 */
> > -       setbits_32(priv->sec_base + SEC_SCTL0 + (3*8), 0x6);
> > +       setbits_32(priv->sec_base + SEC_SCTL0 + (priv->secid * 8), 0x6);
> >
> >         /* Enable SYSCD_RESETb input */
> >         iowrite32(0x100, priv->rcu_base + RCU_CTL);
> > @@ -95,6 +96,7 @@ static int adi_wdt_probe(struct udevice *dev)
> >         struct adi_wdt_priv *priv = dev_get_priv(dev);
> >         int ret;
> >         struct resource res;
> > +       u32 secid;
> >
> >         ret = dev_read_resource_byname(dev, "rcu", &res);
> >         if (ret)
> > @@ -111,6 +113,11 @@ static int adi_wdt_probe(struct udevice *dev)
> >                 return ret;
> >         priv->wdt_base = devm_ioremap(dev, res.start, resource_size(&res));
> >
> > +       ret = dev_read_u32(dev, "secid", &secid);
> > +       if (ret)
> > +               return ret;
> > +       priv->secid = secid;
> > +
>
> I think you should just read it directly into the priv struct without
> a temporary, i.e. dev_read_u32(dev, "secid", &priv->secid).
>
> Also, what do you think about using dev_read_u32_default() with a
> default of 3 (appropriately documented) so that any users with
> modified or custom board device trees will not be surprised by this if
> they update to the latest?
>
> >         ret = clk_get_by_name(dev, "sclk0", &priv->clock);
> >         if (ret < 0) {
> >                 printf("Can't get WDT clk: %d\n", ret);
> > --
> > 2.43.0
> >
>
> Thanks,
> Greg

Reply via email to