Re: [PATCH v3 15/20] watchdog: s3c2410_wdt: Add support for Google tensor SoCs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Guenter,

Thanks for your review.

On Wed, 11 Oct 2023 at 22:20, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>
> On Wed, Oct 11, 2023 at 07:48:18PM +0100, Peter Griffin wrote:
> > This patch adds the compatibles and drvdata for the Google
> > gs101 & gs201 SoCs found in Pixel 6 and Pixel 7 phones. Similar
> > to Exynos850 it has two watchdog instances, one for each cluster
> > and has some control bits in PMU registers.
> >
> > The watchdog IP found in gs101 SoCs also supports a few
> > additional bits/features in the WTCON register which we add
> > support for and an additional register detailed below.
> >
> > dbgack-mask - Enables masking WDT interrupt and reset request
> > according to asserted DBGACK input
> >
> > windowed-mode - Enabled Windowed watchdog mode
> >
> > Windowed watchdog mode also has an additional register WTMINCNT.
> > If windowed watchdog is enabled and you reload WTCNT when the
> > value is greater than WTMINCNT, it prompts interrupt or reset
> > request as if the watchdog time has expired.
>
> Sorry, I don't understand what the code is doing here.

No need to be sorry, I appreciate the review feedback and questions :)

>
> It looks like it enables window mode unconditionally (?). If so,
> what is the impact ? Does it mean that any code requesting multiple
> keepalives in a row on the affected hardware will now cause an
> immediate reset ? If so, what is the rationale ?

Essentially yes, it stops continual keepalives being issued as a
keepalive is only considered valid at certain times in the windowed
cycle.

By way of example, if the watchdog interval is 30s and wtmincnt is set
to *half* of wtcnt, then we would have a "closed window" for the first
15s whereby issuing a keepalive will reset the system, and a "open
window" for the second half of the interval where issuing a keepalive
would reset wtcnt (and be considered a valid service of the watchdog).

The rationale is to stop the watchdog being re-armed "too much" and
it should enable detection of abnormally early as well as abnormally
late servicing of the watchdog.

> Alternatively, if it enables window mode and configures it such
> that WTMINCNT is always equal or larger than WTCNT, what is the
> point of enabling window mode in the first place ?

I looked again, and you are indeed correct that the code currently is
essentially turning windowed mode into a no-op. This appears to be a
bug in the downstream driver as well. Setting WTMINCNT to half of
WTCNT results in a 50% closed, 50% open window, and works on the
hardware as I would expect. I guess this is a good example of why
upstreaming your code, and public code review results in better quality
drivers.

Thanks,

Peter




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux