Re: [PATCH v10 5/5] Watchdog: ARM SBSA Generic Watchdog half timeout panic support

[Guenter Roeck]

On 02/06/2016 10:02 AM, Fu Wei wrote:
> Hi Guenter,
>
> On 6 February 2016 at 07:54, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > On 02/05/2016 10:21 AM, Fu Wei wrote:
> > > On 5 February 2016 at 22:42, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > > > On 02/05/2016 01:51 AM, Fu Wei wrote:
> > > > > Hi Guenter,
> > > > >
> > > > > On 4 February 2016 at 13:17, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > > > > > On 02/03/2016 03:00 PM, Fu Wei wrote:
> > > > > > > On 4 February 2016 at 02:45, Timur Tabi <timur@xxxxxxxxxxxxxx> wrote:
> > > > > > > > Fu Wei wrote:
> > > > > > > > > As you know I have made the pre-timeout support patch, If people
> > > > > > > > > like
> > > > > > > > > it, i am happy to go on upstream it separately.
> > > > > > > > >
> > > > > > > > > If we want to use pre-timeout here, user only can use get_pretimeout
> > > > > > > > > and disable panic by setting pretimeout to 0
> > > > > > > > > but user can not really set pretimeout, because "pre-timeout  ==
> > > > > > > > > timeout / 2 (always)".
> > > > > > > > > if user want to change pretimeout, he/she has to set_time instead.
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > Ok, I think patches 4 and 5 should be combined, and I think the
> > > > > > > > Kconfig
> > > > > > > > entry should be removed and just use panic_enabled.
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > Agreed.
> > > > >
> > > > >
> > > > >
> > > > > np, will do
> > > > >
> > > > > > > NP, will update this patchset like that ,  thanks :-)
> > > > > >
> > > > > > Also, if panic is enabled, the timeout needs to be adjusted accordingly
> > > > > > (to only panic after the entire timeout period has expired, not after
> > > > > > half of it). We can not panic the system after timeout / 2.
> > > > >
> > > > >
> > > > >
> > > > > OK, my thought is
> > > > >
> > > > > if panic is enabled :
> > > > > |--------WOR-------WS0--------WOR-------WS1
> > > > > |------timeout------(panic)------timeout-----reset
> > > > >
> > > > > if panic is disabled .
> > > > > |--------WOR-------WS0--------WOR-------WS1
> > > > > |---------------------timeout---------------------reset
> > > > >
> > > > >     panic_enabled only can be configured when module is loaded by module
> > > > > parameter
> > > > >
> > > > > But user should know that max_timeout(panic_enable) =
> > > > > max_timeout(panic_disable) / 2
> > > >
> > > > That means you'll have to update max_timeout accordingly.
> > >
> > >
> > > panic_enabled only can be configured when module is loaded, so we
> > > don't need to update it.
> > >
> > > max_timeout will only be set up in the init stage.
> > >
> > > Does it make sense ? :-)
> > Not sure I understand your problem or question.
> >
> > max_timeout will have to reflect the correct maximum timeout, under
> > all circumstances. It will have to be set to the correct value before
> > the watchdog driver is registered.
>
> yes, understood,  my thought is :
>
> in  static int sbsa_gwdt_probe(struct platform_device *pdev)
>
> if (action) {
>          wdd->min_timeout = 1;
>          wdd->max_timeout = U32_MAX / gwdt->clk;
> } else {
>          wdd->min_timeout = 2;
>          wdd->max_timeout = U32_MAX / gwdt->clk * 2;

Pretty much, though you would also have to adjust all calculations
using gwdt->clk, in both set_timeout() and get_timeout(). Wonder if
you could adjust gwdt->clk instead.

Does min_timeout really have to be 2 if panic is disabled ?
The only reason seems to be the calculation in sbsa_gwdt_set_timeout().

	writel(timeout / 2 * gwdt->clk, gwdt->control_base + SBSA_GWDT_WOR);

Maybe you could use something like

	writel(timeout * (gwdt->clk / 2), ...);

instead. Or, as mentioned above, adjust the value of gwdt->clk
to include the factor.

Thanks,
Guenter