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

[Alice Guo (OSS)]

> -----Original Message-----
> From: Marco Felsch <m.felsch at pengutronix.de>
> Sent: Thursday, August 25, 2022 5:02 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
> 
> On 22-08-25, Alice Guo (OSS) wrote:
> >
> > > -----Original Message-----
> > > From: Marco Felsch <m.felsch at pengutronix.de>
> > > Sent: Thursday, August 25, 2022 3:50 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
> > >
> > > 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
> >
> > Hi Marco,
> >
> > The ROM code of i.MX7ULP configures the WDOG to support 16-bit unlock
> > command. I plan to add a comment to explain it in code, and keep
> > "mb(); writel_relaxed; writel_relaxed; mb()" unchanged.
> 
> As said, the writel_relaxed() is also 32bit writes. Please just use the correct APIs
> for the stuff you wanna do. In your case, if you want 16bit access functions you
> need to use writew() function and you can drop the mb(). But after reading the
> RM again, I think you can go with writel() since 32bit writes are okay, but you
> need to pass the two 16bit unlock commands. For that you just need to ensure
> that the timing is correct, as noted in the RM.
> 
> Also I read that the WDOG1 is only programmed in serial downloader mode to
> reset the device if no activity was found on the serial/usb bus. Also this only
> happen if the fuse for it was burned, but the RM says also:
> "The watchdog is enabled by default after reset.". Can you please read the
> Watchdog status register and post it here?
> 
> Also after after reading the RM I found the ULK bit "This read-only bit indicates
> whether WDOG is unlocked or not." If this bit is 0 the device is locked and we
> need a unlock command else we can just write the config to the device. The
> unlock command can than be: two writel() commands which are adding the
> barriers for free.
> 
> Regards,
>   Marco

Hi Marco,

1. Two 16-bit writes(0xC520, 0xD928) to the CNT register must be finished within 16 bus clocks.
The CNT register is still a 32-bit register, so it is right to use writel_relaxed().

2. After test, we found that memory barriers cannot be added between two 16-bit writes.
writel() cannot be used.

3. Software must make updates within 128 bus clocks after unlocking and before WDOG closing unlock window.
The ULK bit cannot be used to indicate that WDOG can be configured.

Best Regards,
Alice Guo