[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <91be26169ebbddf3c05cd19626478246cb72a72a.camel@codeconstruct.com.au>
Date: Fri, 22 Sep 2023 13:46:50 +0930
From: Andrew Jeffery <andrew@...econstruct.com.au>
To: Zev Weiss <zev@...ilderbeest.net>,
Wim Van Sebroeck <wim@...ux-watchdog.org>,
Guenter Roeck <linux@...ck-us.net>,
Joel Stanley <joel@....id.au>, Andrew Jeffery <andrew@...id.au>
Cc: linux-watchdog@...r.kernel.org, linux-aspeed@...ts.ozlabs.org,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
openbmc@...ts.ozlabs.org,
Thomas Weißschuh <linux@...ssschuh.net>,
Eddie James <eajames@...ux.ibm.com>,
Ivan Mikhaylov <i.mikhaylov@...ro.com>
Subject: Re: [PATCH] watchdog: aspeed: Add sysfs attributes for reset mask
bits
On Thu, 2023-09-21 at 18:35 -0700, Zev Weiss wrote:
> The AST2500 and AST2600 watchdog timers provide the ability to control
> which devices are reset by the watchdog timer via a reset mask
> resgister. Previously the driver ignored that register, leaving
> whatever configuration it found at boot and offering no way of
> altering its settings. Add a 'reset_ctrl' sysfs subdirectory with a
> file per bit so that userspace can determine which devices the reset
> is applied to.
>
> Note that not all bits in the hardware register are exposed -- in
> particular, the ARM CPU and SOC/misc reset bits are left hidden since
> clearing them can render the system unable to reboot.
>
> Signed-off-by: Zev Weiss <zev@...ilderbeest.net>
> ---
>
> I'm porting OpenBMC to a platform that requires that the LPC controller remain
> un-reset by a BMC reboot. With this patch userspace can control the reset
> mask of the Aspeed watchdog timer, with a few bits remaining unexposed so as
> to prevent some almost-certainly undesirable situations. If there are other
> bits that people feel shouldn't be exposed (or conversely if someone feels
> strongly that the "dangerous" bits _should_ be exposed) I can adjust
> accordingly.
Is there a reason this has to be managed by userspace? It sounds a lot
like a property of platform design, in which case exposing this feature
in the devicetree might be a better approach.
Andrew
Powered by blists - more mailing lists