[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <791dc5e2-4de3-1eb9-8de2-4f2904335756@roeck-us.net>
Date: Thu, 22 Sep 2022 10:43:41 -0700
From: Guenter Roeck <linux@...ck-us.net>
To: rwright@....com, Vladimir Panteleev <git@...dimir.panteleev.md>
Cc: Wim Van Sebroeck <wim@...ux-watchdog.org>,
linux-watchdog@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] watchdog: sp5100_tco: Add "action" module parameter
On 9/22/22 10:36, rwright@....com wrote:
> On Tue, Sep 20, 2022 at 09:27:21AM +0000, Vladimir Panteleev wrote:
>> Allow configuring the "action" bit, as documented in [1].
>>
>> Previously, the only action supported by this module was to reset the
>> system (0). It can now be configured to power off (1) instead.
>>
>> [1]: https://www.amd.com/system/files/TechDocs/44413.pdf
>
> The referenced AMD document 44413 is over 10 years old, and I'm
> concerned when I try to line that document up against the newer versions
> that are implemented in AMD's EPYC processors, that the bit being
> manipulated as SP5100_WDT_ACTION_RESET is effectively reserved in the
> newer references, for example:
>
> https://www.amd.com/system/files/TechDocs/55772-A1-PUB.zip
>
> Is Core::X86::Msr::CpuWdtCfg in the newer document is the same as
> WatchDogControl in the cited 44413.pdf? If so, then I would point out
> that bit 2 is now included in what is called, CpuWdtTimeBase where
> values 2-3H are reserved, meaning bit 2 effectively must be zero.
>
I think those may be different watchdogs. Chapter 9.2.6 ("Watchdog
Timer (WDT) Registers") in 55772-A1-PUB still lists bit 2 of the
control register as:
WatchdogAction. Read-write. Reset: 0. 0=System reset. 1=System power off. This bit determines the action to
be taken when the watchdog timer expires. The bit is only valid when the watchdog is enabled.
Guenter
Powered by blists - more mailing lists