AM5748: Errata i940 Workaround

Stuart Baker

Part Number: AM5748
Other Parts Discussed in Thread: AM5728,

I'm working with a customer that is using the AM5728 and AM5748 in a high reliability system. They have encountered the MPU COUNTER_REALTIME Saturation issue (i940) after 388 days of continuous operation. They are using the Green Hills Integrity kernel on the Cortex-A15 cores, and will be engaging with Green Hills on a fix. It looks like there may be a patch out there for the Linux kernel to work around this issue. We have a few questions that can help us with this issue:

What was the first publicly published date of the i940 Errata for the AM572x and AM574x?
Is it possible to write a value to COUNTER_REALTIME in order to prime up the issue for reproducibility and validation of a fix without having to wait 388 days?
Where can we find the Linux patch file(s) related to this issue and applied to our standard TI Linux SDK?
1. These may be useful as reference for Green Hills to implement their own fix.
It looks like (based on simple Google search) the Linux patch is using a "DM Timer", and "DM Timer" is the same as "GP Timer". Are DM Timer and GP Timer the same thing?

Thanks,

Stuart

over 3 years ago

+1 Keerthy J over 3 years ago

TI__Guru*** 145310 points

Hi Stuart,

Stuart Baker said:
What was the first publicly published date of the i940 Errata for the AM572x and AM574x?

This document dates back to July 2017: https://www.ti.com/lit/er/sprz447d/sprz447d.pdf?ts=1631801510973&ref_url=https%253A%252F%252Fwww.google.com%252F

Has the description of errata.

Stuart Baker said:
Is it possible to write a value to COUNTER_REALTIME in order to prime up the issue for reproducibility and validation of a fix without having to wait 388 days?

I am not aware of this.

Stuart Baker said:
Where can we find the Linux patch file(s) related to this issue and applied to our standard TI Linux SDK?

These may be useful as reference for Green Hills to implement their own fix.

Check this: https://patchwork.kernel.org/project/linux-omap/list/?series=512953

This is already part of upstream Linux kernel.

Stuart Baker said:
It looks like (based on simple Google search) the Linux patch is using a "DM Timer", and "DM Timer" is the same as "GP Timer". Are DM Timer and GP Timer the same thing?

Yes they are the same.

If no other questions please click on verify answer.

- Keerthy

0 Stuart Baker over 3 years ago in reply to Keerthy J

TI__Genius 10390 points

Keerthy,

Before closing this thread, I would like to either know how to write a value to COUNTER_REALTIME or know for sure that it is not possible. Can you please help us with this?

Thanks,

Stuart

+1 Richard Woodruff over 3 years ago in reply to Stuart Baker

TI__Mastermind 19745 points

There is not a way to force the counter to a value to speed up testing or work around the up time limitation. The work arounds are to switch the time base or schedule periodic reboots.

The logic which makes up the real time counter has no programming interface other than the PRM_FRAC_INCREMENTER_NUMERATOR and PRM_FRAC_INCREMENTER_DENUMERATOR_RELOAD registers. These registers only impact the lower bits of the counter rate. The end 48 bit counter value is composed from a slow counting 40bit gray counter for high bits and fast counting alignment counter for low bits. Time freezes when the 40 bit counter reaches its max.

Regards,

Richard W.

Processors

Processors forum

AM5748: Errata i940 Workaround