Seemingly RP235x hardware glitch setting "USE_LPOSC" in the AON timer

[mschnell1]

When the (POWMAN) AON timer is configured to use the XOSC, and then the USE_LPOSC is set in the TIMER register (with the timer stopped) , the state is not changed. The USING_XOSC bit still is set.
(Supposedly) Resetting the chip by starting a new debug session does not change the state and results in USING_XOSC still set and not being able to be reset by USE_LPOSC .
Only after power off-on, the AON timer can be set to use the LPOSC.
(As you normally will not change that state on the fly I suppose this is not really an issue...)
Did anybody experience the same ?

[mschnell1]

It's even worse !

With my initialization code I can configure the AON TIMER to show USING_LPOSC set and can confirm that indeed the low power oscillator is used (as with same the timer runs some 3 promille slower than when it is driven by XOSC and slightly varies with temperature).

But this only works when starting in the debugger by cargo run or cargo run --release.
When just powering up the device, the USING_LPOSC bit is not set.

Any help ?

[mschnell1]

More testing:
With the RP2350 chip (on a Pico 2 PCB) under test I find:
I did software (using the current Embassy distribution without nightly ) using TMER0 as system timer and simply configuring the POWMAN AON timer to run on LPOSC and checking it's count value.
Starting the system with power on, sometimes it finds the using_lposc bit in the TIMER register set, and sometimes cleared.
Nonetheless the counter seems to always use the LPOSC, hence the software should run anyway. But the bit stays clear until power down.

BTW, as said, when first configuring the counter to use XOSC and then setting it to use LPOSC, the bit never is set. re-starting the debug session does not end that situation, power on usually does. (I don't have access to the RUN pin.)

Should that be reported to the RASPI group ?

[mschnell1]

I might try to impose a delay before doing the configuration of the AON Timer ....

[mschnell1]

With the final version of my software this is not an issue any more either irrelevant or indeed not happening.