AM625: A cyclic mode udma issue in SDK 10.01.10.04

Han Yusheng

Part Number: AM625

Tool/software:

Hello, I have a development board with am625 processor. I upgraded from sdk 09.01.00.08 to 10.01.10.04, and then the audio part had problems. I debugged and found that it seemed to be a problem with the cyclic dma part. I rolled back git patch e0a0ce8c2684c13fab0e65be767d036dfa592ee1 and then applied a patch f53378ff892b70c24ee6a0364ae94b690b70d11d, and then the problem was solved. I don't know whether it is a problem with my configuration or the dma part itself has a bug.
Before rolling back the patch, the bug in the audio part came from the long interval of cyclic mode dma interrupts, and each interrupt had to be separated by 1000 jiffies. I added a log to the udma_tx_status function in driver/dma/ti/k3-udma.c and found that while peer_bcnt keeps increasing, bcnt will remain unchanged for a while.
I noticed that the patch e0a0ce8c2684c13fab0e65be767d036dfa592ee1 adds the EOP attribute to the flag of cppi5_tr_type1_t, but I can't find out how the data of this cppi5_tr_type1_t structure is used.
This issue does not seem to be a codec problem. My board uses mcsap0 and mcasp2, and both have this problem.

3 months ago

0 Han Yusheng 3 months ago

Prodigy 10 points

@@ -3882,6 +3856,7 @@ static enum dma_status udma_tx_status(struct dma_chan *chan,
                        ret = DMA_COMPLETE;
                        delay = 0;
                }
+                printk("%s: bcnt=%u, peer_bcnt=%u\n", __func__, bcnt, peer_bcnt);
 
                dma_set_residue(txstate, residue);
                dma_set_in_flight_bytes(txstate, delay);

Comment the log print code. Thanks for anyone can give some suggestions.

0 Swargam Anil 3 months ago in reply to Han Yusheng

TI__Mastermind 46416 points

Hello

Regarding the DMA enablement, could you please clarify which core you are using to perform this operation? Is it the A53 core or a different core?

If you are using the A53 core, could you please confirm which operating system (OS) is running on the A53 core?

Regards,

Anil.

0 Han Yusheng 3 months ago in reply to Swargam Anil

Prodigy 10 points

Thanks for reply,

I used the A53 core on linux, in sdk 10.01.10.04, is linux 6.6.58.

0 Swargam Anil 3 months ago in reply to Han Yusheng

TI__Mastermind 46416 points

Hello Han Yusheng ,

I have routed your query to Linux expert .

You may get reply soon .

Regards,

Anil.

0 Han Yusheng 3 months ago in reply to Swargam Anil

Prodigy 10 points

Thanks very much Swargam Anil,

Sorry I do not know about the linux expert. How can I get to the linux expert?

Actually, I still have some questions about Linux on am625.

0 Suren Porwar 3 months ago in reply to Han Yusheng

TI__Mastermind 28385 points

Hi Han,

Can you test with the Linux Kernel 6.12 version or latest SDK 11.0 and provide us your observations. There have been a lot of fixes in DMA driver post 10.1 SDK as well.

Also, if you can attach the log for analysis.

Best Regards,

Suren

0 Han Yusheng 3 months ago in reply to Suren Porwar

Prodigy 10 points

Hi, Suren

Of course, no problem. I can upgrade the SDK again and see how it goes. But in this case, I need some time because I have other hardware modification work that has a higher priority (the new development board will still use am625 as the processor and Linux as the system).

It may take at least half a month to a month before I have time to upgrade. Do I need to close this issue temporarily or just leave it here?

Sincerely,

Han Yusheng

0 Suren Porwar 3 months ago in reply to Han Yusheng

TI__Mastermind 28385 points

Hi Han,

I will go ahead and close this one. Please open another one when you encounter the issue again with latest SDK release.

Best Regards,

Suren

0 Han Yusheng 3 months ago in reply to Suren Porwar

Prodigy 10 points

Ok, Thanks for your help

Processors

Processors forum

AM625: A cyclic mode udma issue in SDK 10.01.10.04