Hello,
I'm a migrating from openSTLinux ecosystem 5 to 6 on an existing product of ours (with a stm32mp157) and we are really focused on power consumption efficiency.
The main difference between our first version and the second is that we first had TF-A + SPMIN and now we are using the OP-TEE OS.
Previous :
- TF-A 2.4 +SPMIN
- U-Boot 2020.10
- Linux 5.10
Bump to :
- TF-A 2.10
- OP-TEE 4.0.0
- U-Boot 2023.10
- Linux 6.6
We achieved to reach almost the same application state as we were with our version 5 except about the power consumption.
We set the MPU on standby with suspend to RAM mode and we measured the consumption of the overall board and we found that our system used around 10mW more with version 6. After a lot of research, we found out the culprit which is the OP-TEE DDR driver that doesn't latch the CKOE and never disables the DDRPHYCEN bit.
Indeed, in ST version of arm-trusted-firmware 2.4-stm32mp, the CKOE pin is disabled when entering the DDR Self Refresh mode and enabled again when leaving that mode.
Also, this time in version 2.10 of the TF-A, the DDRPHYCEN is disabled when entering in Self Refresh mode and enabled again at leaving that mode.
But it is not done in OP-TEE OS (here and here) which is now in charge of making the DDR enter in Self Refresh. It is enabled at start but never disabled when entering the Self Refresh.
So to test my theory, I added back those lines in the OP-TEE OS driver, run a test and my board booted, entered in suspend and woke up without any problem and the consumption was back to what I had in version 5.
I didn't find any information about why those lines are missing. Is that wanted by the developers or is it an oversight ? If this is wanted by the developers, is there a way to workaround and have the TF-A take care of that part again ?
Don't hesitate to ask for more information if needed or if I wasn't clear.
Thanks in advance for your answer
