[v3,2/2] watchdog: mediatek: mt7988: add wdt support

Add support for watchdog and reset generator unit of the MediaTek
MT7988 SoC.

Signed-off-by: Daniel Golle <daniel@makrotopia.org>
---
v3: fix wrong function parameter name in kernel-doc comment
v2: call new toprgu_reset_sw_en_unlocked from toprgu_reset_update while
    holding lock.

 drivers/watchdog/mtk_wdt.c | 40 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

On 11/14/23 09:04, Daniel Golle wrote:
> Add support for watchdog and reset generator unit of the MediaTek
> MT7988 SoC.
> 
> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
> ---
> v3: fix wrong function parameter name in kernel-doc comment
> v2: call new toprgu_reset_sw_en_unlocked from toprgu_reset_update while
>      holding lock.
> 
>   drivers/watchdog/mtk_wdt.c | 40 ++++++++++++++++++++++++++++++++++++++
>   1 file changed, 40 insertions(+)
> 
> diff --git a/drivers/watchdog/mtk_wdt.c b/drivers/watchdog/mtk_wdt.c
> index b2330b16b497a..3b4ee7185feed 100644
> --- a/drivers/watchdog/mtk_wdt.c
> +++ b/drivers/watchdog/mtk_wdt.c
> @@ -58,6 +58,8 @@
>   #define WDT_SWSYSRST		0x18U
>   #define WDT_SWSYS_RST_KEY	0x88000000
>   
> +#define WDT_SWSYSRST_EN		0xfc
> +
>   #define DRV_NAME		"mtk-wdt"
>   #define DRV_VERSION		"1.0"
>   
> @@ -71,10 +73,12 @@ struct mtk_wdt_dev {
>   	struct reset_controller_dev rcdev;
>   	bool disable_wdt_extrst;
>   	bool reset_by_toprgu;
> +	bool has_swsysrst_en;
>   };
>   
>   struct mtk_wdt_data {
>   	int toprgu_sw_rst_num;
> +	bool has_swsysrst_en;
>   };
>   
>   static const struct mtk_wdt_data mt2712_data = {
> @@ -89,6 +93,11 @@ static const struct mtk_wdt_data mt7986_data = {
>   	.toprgu_sw_rst_num = MT7986_TOPRGU_SW_RST_NUM,
>   };
>   
> +static const struct mtk_wdt_data mt7988_data = {
> +	.toprgu_sw_rst_num = 24,

Kind of odd to have this defined locally, while the others are in include files,
but not worth arguing about. Please make it a define at the top of the file,
though.

Thanks,
Guenter

On Mon, Nov 20, 2023 at 09:19:46AM -0800, Guenter Roeck wrote:
> On 11/14/23 09:04, Daniel Golle wrote:
> > [...]
> > @@ -89,6 +93,11 @@ static const struct mtk_wdt_data mt7986_data = {
> >   	.toprgu_sw_rst_num = MT7986_TOPRGU_SW_RST_NUM,
> >   };
> > +static const struct mtk_wdt_data mt7988_data = {
> > +	.toprgu_sw_rst_num = 24,
> 
> Kind of odd to have this defined locally, while the others are in include files,
> but not worth arguing about.

On 11/20/23 09:33, Daniel Golle wrote:
> On Mon, Nov 20, 2023 at 09:19:46AM -0800, Guenter Roeck wrote:
>> On 11/14/23 09:04, Daniel Golle wrote:
>>> [...]
>>> @@ -89,6 +93,11 @@ static const struct mtk_wdt_data mt7986_data = {
>>>    	.toprgu_sw_rst_num = MT7986_TOPRGU_SW_RST_NUM,
>>>    };
>>> +static const struct mtk_wdt_data mt7988_data = {
>>> +	.toprgu_sw_rst_num = 24,
>>
>> Kind of odd to have this defined locally, while the others are in include files,
>> but not worth arguing about.
> 
>>From I have just learned from Krzysztof Kozlowski those headers shouldn't
> even exist in first place, as the listed IDs are not actually referenced
> anywhere in the driver, hence they aren't actually bindings [1].
> 
> Quote from that thread:
> | >>> Where is the driver change using these IDs?
> | >> It isn't needed as the driver doesn't list the IDs. [...]
> | > Then it is not a binding.
> ---
> 
> Now that they do exist it's too late to change that for everything
> already existing, I suppose. However, it also doesn't seem like adding
> such a header for MT7988 as well is going to be acknowledged, hence we
> will have to live with the inconsistency in the driver in which older
> SoCs will obtain the number of resets from a macro in their respective
> dt-bindings header while newer SoCs won't have such header and hence
> it will have to be defined in the driver itself (as that's also the
> only place where that number is being used).
> 

As I said, not worth arguing about. However, it seems to me that "too late
to change that for everything" isn't really correct. If MTxxxx_TOPRGU_RST_NUM
isn't supposed to be in devicetree include files, all those defines could be
removed from the from there and be added to the watchdog driver. I don't know
about the other defines in include/dt-bindings/reset/mediatek,mtXXXX-resets.h -
many of those _are_ used in dtsi files, but many others are not.

In summary, while I don't really know/understand what is supposed to be defined
in include/dt-bindings/, whatever is known to _not_ to be there (such as the
total number of reset pins on a SoC) could be moved into the driver(s) using it.

Of course, it might well be that there is a rule saying that anything in
include/dt-bindings/ must not be removed from it even if it is not supposed
to be there. In that case, my apologies for the noise.

Thanks,
Guenter