Serial port less reliable after upgrade to 35.1, re-continuation

[albanderuaz]

Hello, this is another continuation of the topic https://forums.developer.nvidia.com/t/serial-port-less-reliable-after-upgrade-to-35-1-continuation/235316

I have performed the tests on the latest jetpack 5.1.2, we still have the same issue: serial errors go through the roof when cpu usage increases a bit.
@linuxdev Did you have a chance to perform the test?

Thanks for your help!

[linuxdev]

I’m not really set up for testing. I can say that there are driver differences I think other people have noticed when going from a JetPack 4 to a JetPack 5 change which validate the error rate. I don’t recall the exact thread, but there was possibly a patch in the latest JetPack 5 (L4T R35.4.1) which might help. I think there was also a case (not necessarily serial port, I am not sure, but maybe it was, but it could have been some other driver, e.g., ethernet) where someone used the 4.x kernel (L4T R32.x/JetPack 4.x has a 4.x version of the Linux kernel) driver in a 5.x kernel (L4T R35.x/JetPack 5.x uses a 5.x kernel version) and things improved dramatically. So a driver change is probably needed, but since JetPack 5.1.2 is the most recent, I’m betting any such change is in a future JetPack 5.1.3 or newer.

Maybe someone from NVIDIA can suggest if there is a driver change in store for either UART (hopefully) or maybe ethernet in the next release. I’m sorry, I just don’t recall the details close enough to say for certain which it was.

[albanderuaz]

Thank you for your answer!
Indeed, i have been monitoring kernel changes and I saw in the last release git log (Log - 8009bb70a484d47d6d0b7bfe336e9e6836897f07 - linux-5.10 - Gitiles) the commit “serial: tegra: fix uart error handler” which led me to trying again on R35.4.1.
I will keep monitoring for the next releases!