RE: [PATCH V1 5/6] watchdog: da9062: DA9062 watchdog driver

[Opensource [Steve Twiss]]

On 06 May 2015 17:02 Guenter Roeck wrote:

> On Wed, May 06, 2015 at 02:54:37PM +0000, Opensource [Steve Twiss] wrote:
> > On 18 April 2015 16:53 Guenter Roeck wrote:
> >
> > Hi Guenter,
> >
> > Thanks for your comments.
> >
> > > On 04/17/2015 07:23 AM, S Twiss wrote:
> > > > From: S Twiss <stwiss.opensource@xxxxxxxxxxx>
> > > >
> > > > Add watchdog driver support for DA9062
> > > >
> > > > Signed-off-by: Steve Twiss <stwiss.opensource@xxxxxxxxxxx>
> > > >
> > > Hi Steve,
> > >
> > > Key question here is if the da9062 is really so much different to the da9062
> > > that you can not use the same driver.
> >
> > The DA9062 watchdog driver does have some similarities with the DA9063 watchdog
> > base functionality -- however the watchdog component in the DA9062 chip has more
> > features yet to be added in software. I do intend to add these other features ...
> > however, if "not adding them here" is a problem I can drop the DA9062 watchdog
> > driver from this patch-set until I have time to write in the newer changes.
> >
> > > I am especially concerned about the added da9062_reset_watchdog_timer(),
> > > given the delay it introduces.
> >
> > After giving this some thought, I am going to remove this 300ms delay from the
> > reset_watchdog_timer() function for my next submission attempt. However
> > I am adding a 300ms delay into the stop() and update_timeout_register() functions
> > instead.
> >
> > The DA9062 watchdog ping (register CONTROL_F) is "windowed" for protection
> > against spurious writes -- i.e. the ping function cannot be called within a 250ms
> > time limit or the PMIC will reset. This windowing protection also extends to altering
> > the timeout scale in the CONTROL_D register -- in which case if the timeout
> > register is altered and the ping() function is called within the 250ms limit, the
> > PMIC will reset. The delay is there to stop that from happening.
> >
> > I realised my previous patch was over-sanitised: by putting the time delay into the
> > ping() function I was protecting CONTROL_D in stop() and update_timeout_register(),
> > but I was being too over-protective of the ping() function. Therefore if there was an
> > "incorrect trigger signal", the watchdog would not be allowed to fail because the
> > driver would have filtered out the errors.
> >
> Hi Steve,
> 
> From your description, it sounds like the protection is only necessary if there
> was a previous write to the same register(s). If so, it might make sense to
> record the time of such writes, and only add the delay if necessary, and only
> for the remainder of the time.
> 
> Would this be possible ?
> 

Hi Guenter,

I think so -- sounds like it should be possible.
Internally, there are several places where the two registers are written in succession.
Also, I'll have to re-write my tests in several places.

Probably the best solution would be to defer this watchdog driver for now, and I'll
re-submit it at a later date once the other parts of the DA9062 driver are [hopefully :)]
accepted. That way I can concentrate a solid block of time on the re-testing ... this
is the most time consuming.

Is that acceptable to you? -- I don't want to lose your existing comments from your
previous posts, so I will keep track of those changes you have already requested.

Regards,
Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/