Message-ID: <6b0d4901-d543-4a06-a1e4-7f1558f5361f@hatter.bewilderbeest.net>
Date:   Mon, 25 Sep 2023 17:04:10 -0700
From:   Zev Weiss <zev@...ilderbeest.net>
To:     Andrew Jeffery <andrew@...id.au>, g@...ter.bewilderbeest.net
Cc:     Conor Dooley <conor+dt@...nel.org>,
        Guenter Roeck <linux@...ck-us.net>,
        Joel Stanley <joel@....id.au>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        "Milton D. Miller II" <mdmii@...look.com>,
        Rob Herring <robh+dt@...nel.org>,
        Wim Van Sebroeck <wim@...ux-watchdog.org>,
        devicetree@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        linux-aspeed@...ts.ozlabs.org, linux-kernel@...r.kernel.org,
        linux-watchdog@...r.kernel.org, openbmc@...ts.ozlabs.org,
        Eddie James <eajames@...ux.ibm.com>,
        Ivan Mikhaylov <i.mikhaylov@...ro.com>,
        Thomas Weißschuh <linux@...ssschuh.net>
Subject: Re: [PATCH 1/2] dt-bindings: watchdog: aspeed-wdt: Add
 aspeed,reset-mask property

On Sun, Sep 24, 2023 at 07:42:45PM PDT, Andrew Jeffery wrote:
>
>
>On Fri, 22 Sep 2023, at 20:12, Zev Weiss wrote:
>> This property configures the Aspeed watchdog timer's reset mask, which
>> controls which peripherals are reset when the watchdog timer expires.
>> Some platforms require that certain devices be left untouched across a
>> reboot; aspeed,reset-mask can now be used to express such constraints.
>>
>> Signed-off-by: Zev Weiss <zev@...ilderbeest.net>
>> ---
>>  .../bindings/watchdog/aspeed-wdt.txt          | 18 +++-
>>  include/dt-bindings/watchdog/aspeed-wdt.h     | 92 +++++++++++++++++++
>>  2 files changed, 109 insertions(+), 1 deletion(-)
>>  create mode 100644 include/dt-bindings/watchdog/aspeed-wdt.h
>>
>> diff --git a/Documentation/devicetree/bindings/watchdog/aspeed-wdt.txt
>> b/Documentation/devicetree/bindings/watchdog/aspeed-wdt.txt
>> index a8197632d6d2..3208adb3e52e 100644
>> --- a/Documentation/devicetree/bindings/watchdog/aspeed-wdt.txt
>> +++ b/Documentation/devicetree/bindings/watchdog/aspeed-wdt.txt
>> @@ -47,7 +47,15 @@ Optional properties for AST2500-compatible watchdogs:
>>  			   is configured as push-pull, then set the pulse
>>  			   polarity to active-high. The default is active-low.
>>
>> -Example:
>> +Optional properties for AST2500- and AST2600-compatible watchdogs:
>> + - aspeed,reset-mask: A bitmask indicating which peripherals will be reset if
>> +		      the watchdog timer expires.  On AST2500 this should be a
>> +		      single word defined using the AST2500_WDT_RESET_* macros;
>> +		      on AST2600 this should be a two-word array with the first
>> +		      word defined using the AST2600_WDT_RESET1_* macros and the
>> +		      second word defined using the AST2600_WDT_RESET2_* macros.
>> +
>> +Examples:
>>
>>  	wdt1: watchdog@...85000 {
>>  		compatible = "aspeed,ast2400-wdt";
>> @@ -55,3 +63,11 @@ Example:
>>  		aspeed,reset-type = "system";
>>  		aspeed,external-signal;
>>  	};
>> +
>> +	#include <dt-bindings/watchdog/aspeed-wdt.h>
>> +	wdt2: watchdog@...85040 {
>> +		compatible = "aspeed,ast2600-wdt";
>> +		reg = <0x1e785040 0x40>;
>> +		aspeed,reset-mask = <AST2600_WDT_RESET1_DEFAULT
>> +				     (AST2600_WDT_RESET2_DEFAULT & ~AST2600_WDT_RESET2_LPC)>;
>> +	};
>
>Rob has acked your current approach already, but I do wonder about an 
>alternative that aligns more with the clock/reset/interrupt properties. 
>Essentially, define a new generic watchdog property that is specified 
>on the controllers to be reset by the watchdog (or even on just the 
>watchdog node itself, emulating what you've proposed here):
>
>watchdog-resets = <phandle index>;
>
>The phandle links to the watchdog of interest, and the index specifies 
>the controller associated with the configuration. It might even be 
>useful to do:
>
>watchdog-resets = <phandle index enable>;
>
>"enable" could provide explicit control over whether somethings should 
>be reset or not (as a way to prevent reset if the controller targeted 
>by the provided index would otherwise be reset in accordance with the 
>default reset value in the watchdog controller).
>
>The macros from the dt-bindings header can then use macros to name the 
>indexes rather than define a mask tied to the register layout. The 
>index may still in some way represent the mask position. This has the 
>benefit of hiding the issue of one vs two configuration registers 
>between the AST2500 and AST2600 while also allowing other controllers 
>to exploit the binding (Nuvoton BMCs? Though maybe it's generalising 
>too early?).
>

Sorry, I'm having a bit of a hard time picturing exactly what you're 
suggesting here...to start with:

>property that is specified on the controllers to be reset by the
>watchdog

and

>or even on just the watchdog node itself

seem on the face of it like two fairly different approaches to me.  The 
former sounds more like existing clock/reset/etc. stuff, where the 
peripheral has a property describing its relationship to the "central" 
subsystem, and various peripheral drivers are all individually 
responsible for observing that property and calling in to the central 
subsystem to configure things for that peripheral appropriately; if I'm 
understanding you correctly, it might look something like:

   &spi1 {
     watchdog-resets = <&wdt1 WDT_INDEX_SPI1 0>;
   };

Or maybe something more like how pinctrl works, via phandles to subnodes 
of the central device?

   &wdt1 {
     wdt1_spi1_reset: spi1_reset {
       reg = <0x1c>;
       bit = <24>;
     };
   };

   &spi1 {
     watchdog-resets = <&wdt1_spi1_reset 0>;
   };

Either way, it seems like it'd be complicated by any insufficient 
granularity in the watchdog w.r.t. having independent control over the 
individual devices represented by separate DT nodes (such as how the 
AST2500 watchdog has a single SPI controller reset bit instead of one 
per SPI interface, or its "misc SOC controller" bit governing all sorts 
of odds and ends).

In the latter case (property on the wdt node), would it essentially just 
be kind of an indirection layer mapping hardware-independent device 
indices to specific registers/bits?  It's not obvious to me what purpose 
a phandle to the peripheral device node would serve (would the wdt 
driver have a good way of identifying what specific peripheral it's 
pointing to to know what bit to twiddle?), but maybe I'm 
misunderstanding what you're suggesting...


I guess my other uncertainty is the balance between generalization and 
applicability -- how many other watchdog devices have sufficient 
comparable configurability to make use of it?  I haven't pored over all 
of them, but from a random sampling of 20 so of the other existing wdt 
drivers I don't see any obvious candidates -- the closest I saw were 
cpwd.c, which apparently can distinguish between a CPU reset and a 
CPU/backplane/board reset, and realtek_otto_wdt.c, which can do a CPU or 
a SOC reset (though I don't have any of the hardware docs to know what 
capabilities other devices might provide that the drivers don't use).  
Do the Nuvoton BMCs have watchdogs with peripheral-granularity reset 
configuration?


Thanks,
Zev

Powered by blists - more mailing lists