Re: [PATCH v3] arm64: dts: nuvoton: ma35d1: add CAN nodes

[Jacky Huang <ychuang570808@gmail.com>]

Marc Kleine-Budde 於 2026/9/11 下午 01:56 寫道:
> On 11.09.2026 10:15:21, Jacky Huang wrote:
>>> [ ... ]
>>>> @@ -244,6 +244,58 @@ gpion: gpio@340 {
>>>>                         };
>>>>                 };
>>>>
>>>> +             can0: can@403c0000 {
>>>> +                     compatible = "bosch,m_can";
>>>> +                     reg = <0x0 0x403c0000 0x0 0x200>, <0x0 0x403c0200 0x0 0x2000>;
>>>> +                     reg-names = "m_can", "message_ram";
>>>> +                     interrupts = <GIC_SPI 71 IRQ_TYPE_LEVEL_HIGH>,
>>>> +                                  <GIC_SPI 131 IRQ_TYPE_LEVEL_HIGH>;
>>>> +                     interrupt-names = "int0", "int1";
>>>> +                     clocks = <&clk HCLK3>, <&clk CAN0_GATE>;
>>>> +                     clock-names = "hclk", "cclk";
>>>> +                     bosch,mram-cfg = <0x0 4 4 32 32 32 8 8>;
>>>> +                     status = "disabled";
>>>> +             };
>>> [Severity: Medium]
>>> Should this node include the hardware resets property (e.g.,
>>> resets = <&sys MA35D1_RESET_CANFD0>;)?
>>>
>>> Without the hardware reset line, the bosch,m_can driver cannot perform a
>>> hardware reset of the CAN controllers during probe or interface toggles.
>>> It must rely solely on software initialization, which can fail to fully
>>> recover the controller if the bootloader or a prior crashed state left
>>> the hardware in an unpredictable state.
>>>
>>> This omission pattern is identically repeated across the newly added can1,
>>> can2, and can3 nodes below.
>> The reset property is optional for the MA35D1 CAN nodes.
> yes, but....
>
>> The bosch,m_can binding does not require a reset property, and the
>> m_can driver obtains the reset control through
>> devm_reset_control_get_optional_shared().
>>
>> The MA35D1 CAN controllers also operate correctly without explicit
>> CAN reset properties in the existing platform DTS. Therefore, no
>> change is required for this patch.
> ...we've observed that the SW reset doesn't clear the CAN state and
> error counters, see commit 9271d0ea07c2 ("can: m_can: add support for
> optional reset") for details. As your SoC has a reset line connected to
> the M_CAN IP core, please add it to the dtsi.
>
> regards,
> Marc
>

Hi Marc,

Thanks for pointing this out. I agree that the hardware reset properties 
should
be described for the MA35D1 CAN controllers.

Zi-Yu, please add the corresponding CANFD reset properties to can0 
through can3 in v4.

I will re-review the updated version.

Best regards,
Jacky Huang