RE: [PATCH] watchdog: da9062: add power management ops

From: Adam Thomson <Adam.Thomson.Opensource@diasemi.com>
To: Guenter Roeck <linux@roeck-us.net>,
	Adam Thomson <Adam.Thomson.Opensource@diasemi.com>,
	Marco Felsch <m.felsch@pengutronix.de>
Cc: Support Opensource <Support.Opensource@diasemi.com>,
	"wim@linux-watchdog.org" <wim@linux-watchdog.org>,
	"linux-watchdog@vger.kernel.org" <linux-watchdog@vger.kernel.org>,
	"kernel@pengutronix.de" <kernel@pengutronix.de>
Subject: RE: [PATCH] watchdog: da9062: add power management ops
Date: Mon, 2 Dec 2019 14:44:43 +0000	[thread overview]
Message-ID: <AM5PR1001MB0994B8FDCC3C1B991433344780430@AM5PR1001MB0994.EURPRD10.PROD.OUTLOOK.COM> (raw)
In-Reply-To: <1feb2b8e-af46-ffa9-873a-fd9b624294b9@roeck-us.net>

On 02 December 2019 13:38, Adam Thomson wrote:

> On 12/2/19 5:27 AM, Adam Thomson wrote:
> > Hi Marco,
> >
> >>> Hmmm. Wouldn't that then be a board specific fix rather than this change?
> >>> Am concerned relying on software to reenable the watchdog on resume
> could
> >> allow
> >>> for a hang scenario potentially if that code never gets to execute. Other
> >>> systems shouldn't need this fix, assuming they don't have issues at the HW
> >>> level, so this this seems like it could be making those systems less robust. If
> >>> we are to do this at the driver level, maybe this should be an option through
> DT
> >>> to help faulty systems, but I'm thinking this shouldn't be default behaviour.
> >>
> >> I don't think that we should rely on the OTP values. Those values are
> >> set by Dialog, the SoM manufacturers or by the customer itself and the
> >> time shows that this is error prone too. At least if the customer or the
> >> SoM manufacturer don't ask the Dialog Support..
> >>
> >> You're right with the (re-)enabling but there are other drivers using
> >> such an approach.. IMHO it is safer to go this way rather than trust
> >> the OTP and the PCB layout. I would rather add a dt-compatible to tell
> >> the driver to do nothing during suspend/resume e.g.:
> >>
> >>      - dlg,use-hw-pm
> >>
> >> or something. Because the user needs to validate the PCB and the OTP
> >> values.
> >
> > The thing is the DT FW is supposed to be fairly static so I would rather not
> > enforce this change on users if they adopt a kernel version with this update in.
> > I also still think it's safer if the HW does this for us. I would have hoped for
> > most designs this would be caught in early bring up where it can be rectified
> > with minimal impact, although I'm guessing that's not the case in your scenario.
> >
> 
> dla,use-sw-pm ?
> dla,hw-pm-broken ?

Yes, I'd probably opt for 'dlg,use-sw-pm' or maybe 'dlg,manual-pm' to cover it.

> 
> Guenter