[PATCH 2/7] watchdog: imx7ulp: Add explict memory barrier for unlock sequence

[Marco Felsch]

On 22-08-24, Alice Guo (OSS) wrote:
> 
> > -----Original Message-----
> > From: Marco Felsch <m.felsch at pengutronix.de>
> > Sent: Wednesday, August 24, 2022 5:06 PM
> > To: Alice Guo (OSS) <alice.guo at oss.nxp.com>
> > Cc: Guenter Roeck <linux at roeck-us.net>; wim at linux-watchdog.org;
> > shawnguo at kernel.org; s.hauer at pengutronix.de; festevam at gmail.com;
> > linux-arm-kernel at lists.infradead.org; linux-kernel at vger.kernel.org;
> > dl-linux-imx <linux-imx at nxp.com>; kernel at pengutronix.de;
> > linux-watchdog at vger.kernel.org
> > Subject: Re: [PATCH 2/7] watchdog: imx7ulp: Add explict memory barrier for
> > unlock sequence
> > 
> > Hi Alice,
> > 
> > On 22-08-24, Alice Guo (OSS) wrote:
> > > > -----Original Message-----
> > > > From: Marco Felsch <m.felsch at pengutronix.de>
> > > > Sent: Wednesday, August 24, 2022 4:04 PM
> > > > To: Alice Guo (OSS) <alice.guo at oss.nxp.com>
> > > > Cc: Guenter Roeck <linux at roeck-us.net>; wim at linux-watchdog.org;
> > > > shawnguo at kernel.org; s.hauer at pengutronix.de; festevam at gmail.com;
> > > > linux-arm-kernel at lists.infradead.org; linux-kernel at vger.kernel.org;
> > > > dl-linux-imx <linux-imx at nxp.com>; kernel at pengutronix.de;
> > > > linux-watchdog at vger.kernel.org
> > > > Subject: Re: [PATCH 2/7] watchdog: imx7ulp: Add explict memory
> > > > barrier for unlock sequence
> > > >
> > > > Hi Alice,
> > > >
> > > > On 22-08-24, Alice Guo (OSS) wrote:
> > > >
> > > > ...
> > > >
> > > > > > > > Hi Guenter and Marco,
> > > > > > > >
> > > > > > > > 1. did you see any issues?
> > > > > > > > This WDOG Timer first appeared in i.MX7ULP, no one report
> > > > > > > > issues probably because few people use i.MX7ULP. This issue
> > > > > > > > was found when we did a stress test on it. When we
> > > > > > > > reconfigure the WDOG Timer, there is a certain probability
> > > > > > > > that it reset. The reason for the error is that when
> > > > > > > > WDOG_CS[CMD32EN] is 0, the unlock sequence is two 16-bit
> > > > > > > > writes (0xC520, 0xD928) to the CNT register within 16 bus
> > > > > > > > clocks, and improper unlock sequence causes the
> > > > WDOG to reset.
> > > > > > > > Adding mb() is to guarantee that two 16-bit writes are
> > > > > > > > finished within 16
> > > > > > bus clocks.
> > > > > > >
> > > > > > > After this explanation the whole imx7ulp_wdt_init() seems a
> > > > > > > bit buggy because writel_relaxed() as well as writel() are
> > > > > > > 32bit access
> > > > functions.
> > > > > > > So the very first thing to do is to enable the 32-bit mode.
> > > > > > >
> > > > > > Agreed. This is much better than having extra code to deal with
> > > > > > both 16-bit and 32-bit access.
> > > > > >
> > > > > > > Also this is a explanation worth to be added to the commit
> > > > > > > message
> > > > > > > ;)
> > > > > > >
> > > > > >
> > > > > > Definitely. Also, the use of mb(), if it should indeed be
> > > > > > needed, would have to be explained in a code comment.
> > > > > >
> > > > > > Thanks,
> > > > > > Guenter
> > > > >
> > > > > Hi Marco and Guenter,
> > > > >
> > > > > Thank you for your comments. I plan to enable support for 32-bit
> > > > > unlock command write words in bootloader. In this way, there is no
> > > > > need to distinguish whether the unlock command is a 32-bit command
> > > > > or a 16-bit command in driver.
> > > >
> > > > Please don't move this into the bootloader, enabling it within the
> > > > init seq. is just fine. If you move it into the bootloader then you
> > > > can't ensure that the bit is set since there are plenty of bootloaders out
> > there.
> > > >
> > > > As I said, just drop the "16bit" unlock sequence from the init
> > > > function because the unlock is handled just fine in all the watchdog_ops.
> > > >
> > > > Regards,
> > > >   Marco
> > >
> > > Hi Marco,
> > >
> > > Sorry, I did not tell you that all watchdog control bits, timeout
> > > value, and window value cannot be set until the watchdog is unlocked.
> > 
> > You don't have to according the RM:
> > 8<----------------------------------------------------------------------
> > 59.5.2 Disable Watchdog after Reset
> > 
> > All of watchdog registers are unlocked by reset. Therefore, unlock sequence is
> > unnecessary, but it needs to write all of watchdog registers to make the new
> > configuration take effect. The code snippet below shows an example of
> > disabling watchdog after reset.
> > 8<----------------------------------------------------------------------
> > 
> > also the RM tells us:
> > 8<----------------------------------------------------------------------
> > 59.4.3.1 Configuring the Watchdog Once
> > 
> > The new configuration takes effect only after all registers except CNT are
> > written after reset. Otherwise, the WDOG uses the reset values by default. If
> > window mode is not used (CS[WIN] is 0), writing to WIN is not required to
> > make the new configuration take effect.
> > 8<----------------------------------------------------------------------
> > 
> > > Support for 32-bit unlock command write words in enabled in
> > > imx7ulp_wdt_init now.
> > 
> > So.. after reading the IMX7ULP RM, which was not my intention, I found out
> > that most of the WDOG_CS regiter bits are write-once bits. This means if you
> > didn't set it within the bootloader you still in case "59.4.3.1".
> > 
> > So the imx7ulp_wdt_init() function just needs to check if the
> > WDOG_CS_UPDATE bit was set. If it is not the case, then you need to write the
> > WDOG_CS register as currently done. If the bit is set, than you need know that
> > the bootloader did the job for you and you can exit
> > imx7ulp_wdt_init() early. In both cases the unlock is not required.
> > 
> > Can you please check/test if this is working for you?
> > 
> > Regards,
> >   Marco
> > 
> 
> Hi Marco,
> 
> Rom code has already configured the WDOG once, so we cannot use "
> Configuring the Watchdog Once".

What? How does the ROM code configure the WDOG? Also this would be worth a
comment within the code. Also still assume that this "16bit unlock" seq.
is useless since you writing 32bit anyway.

Regards,
  Marco