Hi Neeraj,
I agree.
There shouldn’t be any discrepancy.
But, there is.
What is wrong here?
Why do the GPIO measurement and GPIO I2C read give different results?
thanks,
Gil
Hi Neeraj,
Thank you.
Can you please read what I enclosed below and comment on that? (from your knowledge).
(1)
My guess it that SDMMC initialization procedure in the TF-A powers the SDMMC device off/on (maybe to bring it to some initial state) and by removing this line, the TF-A code will not know which supply to toggle, so it will not perform it.
So, the instability/oscillations in BUCK4 (due to the off/on) will not happen, thus will not affect the eMMC.
(2)
Then you wrote “In case of v3v3 supply of the sdmmc, you cannot do a power cycle of the sdmmc at tf-a level (because you will switch off almost all the platform). But this is not an issue because the romcode has already done it before.”
I assume this is another justification to remove this line from the TF-A dts.
(3)
Can you describe what this patch does?
(From you knowledge of the patch syntax, if you are familiar with it). I enclosed the patch.
thank you
Gil
Hi Neeraj,
I looked into the links you sent.
I compiled and sent few questions to the person who submitted the patch in ST.
That was five days ago and no reply from him.
Please take a look and see if you can assist.
************************************************************************
We encountered this issue just recently. Every few power-ons, the system is stuck and we get the messages as described at the beginning of the thread. Resetting the PMIC / STM32MP153C solves this.
At first you suggested to remove the line for vmmc-supply from the TF-A dts.
My guess it that SDMMC initialization procedure in the TF-A powers the SDMMC device off/on (maybe to bring it to some initial state) and by removing this line, the TF-A code will not know which supply to toggle, so it will not perform it.
So, the instability/oscillations in BUCK4 (due to the off/on) will not happen, thus will not affect the eMMC.
Then you wrote “In case of v3v3 supply of the sdmmc, you cannot do a power cycle of the sdmmc at tf-a level (because you will switch off almost all the platform). But this is not an issue because the romcode has already done it before.”
I assume this is another justification to remove this line from the TF-A dts.
In my case, our custom board connects v3v3 (BUCK4) to both vmmc and vqmmc supplies.
So, should I remove both lines (vmmc-supply = <&v3v3>; vqmmc-supply = <&v3v3>) from the TF-A dts?
Then, you wrote: “After internal discussions, we will deploy a correction that will have the same effect as the previous solution. So you can go on with it on your side.”
And, suggested a fix: a patch, to be applied to “regulator_core.c” (a file in the TF-A package).
Can you describe what this patch does?
What do you suggest we do? Apply the patch or modify the TF-A dts file?
*****************************************************************
thank you
Gil
hello Neeraj,
RANK 0 voltages, that we actively use, include BUCK2 (VDD_DDR) and the voltages that derive from it, LDO3/VTT_DDR and VREF_DDR.
We don’t use the other power rails that are RANK 0: LDO1, LDO2, LDO6.
However, I measure 0V on all of them.
This “erratic” behavior was observed on at least 4-5 boards (out of the 18 we assembled).
We use an eMMC Flash which has FSBL, SSBL, Linux, etc., on it.
I managed to capture the boot log when the issue happens.
Enclosed a text file with failing (problematic power-up) and successful boot logs.
I can understand why RANK 0 voltages are not there. It didn’t reach the place where they are intialized and enabled.
I am not able to accurately pin point where, during boot, the system fails.
Is it during the SoC PROM execution?
Was it able to load FSBL1 from the eMMC Flash?
What is CMD13?
Is everything we see in the boot log is a result of the SoC running its PROM code?
Meaning initial access to eMMC fails?
thanks,
Gil
Hi Michal,
I am thinking:
The FSBL initializes the clock and DDR, then loads the SSBL into the DDR and executes it.
Is there a known sequence that the FSBL follows when it initializes, before loading the SSBL?
Maybe we fail in some early stage of the FSBL init (clock initialization for example) and do not reach the stage where the DDR and DDR power (BUCK2) are initialized, and that is why BUCK2 is not defined as 1.35V and/or is not turned on?
thanks,
Gil
Hi Michal,
Thank you for the prompt response.
So, we have the ROM code in the STM32MP that loads the FSBL, from the eMMC flash, in the internal RAM and executes it.
the FSBL initializes the clock and DDR, then loads the SSBL into the DDR and executes it.
In our case, BUCK2=0V (VDD_DDR).
FSBL should change the default voltage to 1.35V and enable it. I assume it does that via I2C4.
The TF-A dts file you shared is very similar to what we use (I attached what we use).
Is there any difference between the TF-a dts files that might cause this issue?
It seems we have something marginal here since it only happens once every 5-10 power ups.
In addition, NRST solves this.
NRST resets the ST as well as the STPMIC1.
thanks,
Gil
Hi Neeraj,
The log level is at its maximum.
As soon as the kernel takes control, something goes wrong.
It doesn’t even reach a point where it can send a message to the UART so we can see it.
We saw a similar issue in some ST forum and it was suggested that there is a dependency between the kernel and UBOOT versions.
We will follow what you suggested and see how it goes.
thanks,
Gil
hello Neeraj,
Another question:
The U-Boot version we use is the one that comes with Dunfell: version: 2020.10
Should we use a more updated UBOOT version?
Are there any dependencies between the Linux kernel version and the UBOOT version?
thanks,
Gil
hello Neeraj,
Please see the attached files for details (“main” then “5_15_67_log”).
Do you have any suggestions for us?
thanks,
Gil
hello Neeraj,
We see an issue with how the device tree is handled in 5.15.67 versus 5.10 and 5.15.24.
Please see the attached file for details.
How do we proceed from here?
thank you
Gil
hello Neeraj,
Adding information to my above message:
We downloaded the kernel source from ST’s repository (where all the ST patches are already applied).
Some of the patches from the 5.15.24 do fail on the 5.15.67, but 14 patches fully pass.
So, some patches from the 5.15.24 are required for the 5.15.67. And some fail since they are already applied.
So, we can try and manually correct the failing patches.
Question is: Do we need also Octavo-created patches to be applied to the build of the 5.15.67 based Linux, so it can run on the Octavo SiP we use?
Will the 5.15.24 Octavo patches run successfully on the 5.15.67 sources? Are those enough, from Octavo, for the 5.15.67?
thank you
Gil
hello Neeraj,
Well, totally eight of the patches fail.
We need kernel version 5.15.67 (or later) for the SPI slave support. Do you know if the kernel code from ST, you directed me to, is what we need?
Are all the required patches from ST only? Doesn’t we need also Octavo-created patches to be applied to the build of the 5.15.67 based Linux, so it can run on the SiP we use?
thanks,
Gil
Hi Neeraj, Eshtaartha,
I input the information to three text files.
Please read them in this order: first.txt, second.txt, third.txt.
It is more updated and detailed than what I sent Martin.
Thank you
Gil
Hi Neeraj, Eshtaartha,
Thank you.
I am unable to reply to you with the full information I need to enclose.
I just sent an email to Martin Burgos, asking him to forward it to you.
thank you
Gil
Dear Eshtaartha,
Thank you for your reply.
We followed the guidelines.
The meta-layer Kirkstone support has Linux version 5.15.24.
It DOES NOT support SPI slave configurations.
Only from Linux version 5.15.67 SPI slave is supported.
Can we build the 5.15.67 Linux version and use the patches in your 5.15.24 version support?
thanks
Gil
Octavo Systems LLC all rights reserved
OCTAVO is registered in the U.S. Patent and Trademark Office. OSD, C-SiP, and the Octavo Logo are trademarks of Octavo Systems LLC.
"*" indicates required fields
"*" indicates required fields
"*" indicates required fields
"*" indicates required fields