Re: [RFC PATCH 0/1] Suspending i.MX watchdog in WAIT mode

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

 





On 26. 09. 22 09:36, Andrej Picej wrote:


On 23. 09. 22 15:48, Guenter Roeck wrote:
On 9/23/22 00:27, Andrej Picej wrote:
On 22. 09. 22 16:56, Guenter Roeck wrote:
On 9/22/22 00:17, Andrej Picej wrote:
Hi Guenter,

On 21. 09. 22 16:18, Guenter Roeck wrote:
On 9/21/22 05:46, Andrej Picej wrote:
Hi all,

we are using i.MX6UL with its watchdog WDOG1 and kernel 5.15.62. It was discovered that the watchdog triggers reset when the device is put into
'Suspend-To-Idle' (WAIT) state.


Is that equivalent to "suspend" from Linux perspective, or some other
mode ? How does the device get into this state ?

I think WAIT mode maps to System idle mode in linux [1].


Sorry, I am not going to read that entire manual.


Perfectly understandable, you are busy guys, basically in chapter 21.1.1 it is written:
"System idle maps to WAIT mode."

Sorry don't quite understand your second question.
Do you mean how we trigger this state?
We trigger this state with:
 >     imx6ul-dev:~# echo freeze > /sys/power/state


So it is not "suspend" ? I am not sure if it is appropriate to stop
the watchdog timer in this situation. Normally it is only stopped
in suspend mode.


Well suspend/resume functions are still called when entering this "freeze" mode, but watchdog can not be disabled by software or pinged in this mode, that's why our idea is to set this bit.

Basically, that was my main question. Is it appropriate to stop watchdog in this situation? I guess not. Probably it is not meant to enter this mode for longer period of time.

Also, how does this interact or interfere with the suspend/resume code
in the driver, and does it behave the same for all chips supported by
the driver ? For example, in the current code, i.MX7D is handled
differently. What compatible entry are you using anyway ? There is none
for i.MX6UL. Did you make sure that the same bit doesn't mean something
else for other chips ?


I don't think it interact/interferes with suspend/resume code in any way since mx2+ watchdog is not stoppable during runtime. Not sure if watchdogs on other devices behave the same, sorry.

i.MX6UL devices use "fsl,imx6ul-wdt", "fsl,imx21-wdt" for compatible entry. I checked control registers with other supported devices by this driver:

- fsl,imx25-wdt (same behaviour -> WDW)
- fsl,imx27-wdt (bit is reserved)
- fsl,imx31-wdt (bit is reserved)
- fsl,imx35-wdt (same behaviour -> WDW)
- fsl,imx50-wdt (same behaviour -> WDW)
- fsl,imx51-wdt (same behaviour -> WDW)
- fsl,imx53-wdt (same behaviour -> WDW)
- fsl,imx6q-wdt (same behaviour -> WDW)
- fsl,imx6sl-wdt (same behaviour -> WDW)
- fsl,imx6sll-wdt (same behaviour -> WDW)
- fsl,imx6sx-wdt (same behaviour -> WDW)
- fsl,imx6ul-wdt (same behaviour -> WDW)
- fsl,imx7d-wdt (same behaviour -> WDW)
- fsl,imx8mm-wdt (same behaviour -> WDW)
- fsl,imx8mn-wdt (same behaviour -> WDW)
- fsl,imx8mp-wdt (same behaviour -> WDW)
- fsl,imx8mq-wdt (same behaviour -> WDW)
- fsl,ls1012a-wdt (bit is reserved)
- fsl,ls1043a-wdt (bit is reserved)
- fsl,vf610-wdt (same behaviour -> WDW)
- fsl,imx21-wdt (reserved)


And then there is fsl,imx7ulp-wdt which has its own driver.

Looking at this table WDW is not really i.MX6UL specific. It is strange that more people don't experience this problem. I guess using "freeze" mode is not very common, which is understandable, since other low power modes are more energy efficient.  > Anyway, if some people are using this "feature" of watchdog for WAIT mode supervision, setting this bit would break their use. I just wanted to get your opinion on this, which I got.


Oh, no, you are just giving up too early.

Well I'm not really sure if it is worth the struggle. But if you insist, I'll dig a bit deeper :).


Other watchdogs do get stopped in suspend. That is why the suspend function exists and is used in watchdog drivers in the first place. Yes, fsl,imx21-wdt
can not handle that, but that doesn't mean that others can't either. The
current workaround for fsl,imx21-wdt is to set the timeout to the maximum
and to stop the watchdog timer clock. You say that the suspend function is
called, so the watchdog clock _should_ be stopped. However, it looks like
that is not the case. Can you try to figure out the reason ?

Ok, so a quick look at clocks used in watchdog reveals an interesting find. The watchdog uses two clocks: - low-frequency reference clock (ipg_clk_32k) for its counter and control operation and
- peripheral bus clock for register read/write operations.

 From reference manual:
Low frequency (32.768 kHz) clock that continues to run in low-power
mode. It is assumed that the Clock Controller will provide this clock signal
synchronized to ipg_clk in the normal mode, and switch to a non-
synchronized signal in low-power mode when the ipg_clk is off.

In suspend function we are stopping wdog1_clk which is child of ipg_clk. This low-frequency reference clock is left running, which means that watchdog counter keeps running. The problem is that ipg_clk_32k is meant to be running in low-power mode, that's why this clock can not be gated off (no gating register).


Either case, I think it would be worthwhile exploring if the WDW bit can
be used on the CPUs supporting it to stop the watchdog in suspend mode.
How does your sysytem behave if put into suspend ? Does the watchdog still
reset it ? How is the watchdog clock configured ?

If the system is put into suspend mode, the watchdog is suspended, due to WDZST bit, which similar to WDW bit suspends the watchdog in low-power modes (STOP and DOZE). If this WDZST bit is NOT set, the watchdog triggers when system is put into "Suspend-to-RAM" and "Standby/Power-On Suspend".

I had a look at some of other devices watchdogs this driver supports and it looks like i.MX6, i.MX8, i.MX7 and i.MX5x have this low-frequency reference clock in common.


What if we used a new device-tree property "fsl,suspend-in-wait" to set WDW bit? And we leave WDW bit unset if this property is not defined, keeping the default watchdog behaviour as it is? Would that be acceptable?

Best regards,
Andrej



[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux