TDA4VM: gpu question

lei tao

Part Number: TDA4VM

Tool/software:

Hi TI，

The GPU is stuck at the glFinish function. See dump.png for details.
Use 'cat /sys/kernel/debug/pvr/status' to view the status of the GPU. The Firmware Status shows NOT RESPONDING (KCCB stalled). See status.png for details.
Check the dmesg printout, which also contains abnormal information of PVR. See dmesg.log for details.

6644.dmesg.log

BR
Lei

over 1 year ago

0 lei tao over 1 year ago

Prodigy 170 points

SDK8.6

0 Erick Narvaez over 1 year ago in reply to lei tao

TI__Mastermind 36307 points

Hello Lei,

The GPU will throw a hardware recovery and firmware can stall for many reasons, so we need to try and narrow down what could have caused the issue. Please be aware of this application note to get a background on the debug process if it is driver related:

https://www.ti.com/lit/an/spradi3/spradi3.pdf

It does look like the GPU firmware has stalled. Can you please share some details about the failure:

1) Does it happen on every boot, or it reproduces randomly?

2) Does this happen on all boards, or only on one board?

3) Does this happen immediately in the application? Or does the application run normally and the reproduction of the issue can vary?

Thanks,

Erick

0 lei tao over 1 year ago in reply to Erick Narvaez

Prodigy 170 points

Hi Erick,

Erick Narvaez said:
1) Does it happen on every boot, or it reproduces randomly?

Randomly.

Erick Narvaez said:
2) Does this happen on all boards, or only on one board?

All boards.

Erick Narvaez said:
3) Does this happen immediately in the application? Or does the application run normally and the reproduction of the issue can vary?

Expand later.

BR，

Lei

0 lei tao over 1 year ago in reply to lei tao

Prodigy 170 points

Hi Erick,

Our problem looks similar to e2e, can you help confirm? Or what methods can be used to confirm?

0 Erick Narvaez over 1 year ago in reply to lei tao

TI__Mastermind 36307 points

Lei,

It may be related to the GPU cache coherency issue that is present in the SDK 8.x versions and older. There is an FAQ that explains the issue here:

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1218318/faq-tda4vm-are-there-any-known-bugs-and-patches-that-i-should-use-in-my-gpu-driver

However, the updated latest UM libraries and the latest KM patches are here: https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1316731/faq-tda4vl-q1-what-are-the-gpu-driver-bug-fixes-for-sdk-8-6-or-earlier

You can updated your libraries and drivers and see if you can still replicate the issue.

The first link contains a unit test you can run to see that your system is setup in a way that is resistant to the cache coherency issue:

rgx_kicksync_test -ver -nc 16 -loop 100 -n 10000 -r -seed 81576

If this test passes, your system has the patches applied correctly. If it fails, it is a good indication that you do not have the necessary patches or workarounds in place for the issue.

Regards,

Erick

Processors

Processors forum

TDA4VM: gpu question