AM6442: BCDMA CacheP_inv failed

Hello Shine,

Actually, cache_invalid is required after completion of DMA; otherwise, the processor can't read the data from destination memory.

About cache_invalid is why it is needed. I have covered that topic in the below thread. Please check the last response for cache_invalid.

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1301437/tmds64evm-data-transfer-timing-for-bcdma-samples/4968961#4968961

Regards,

S.Anil.

Hello Shine,

One more thing is that every time we don't need to call trpd_init for triggering dma,. trpd_init should be done only once.

Next, to start dma, use the que_raw function, which is based on the user's requirements for where they need to start dma.

Next, check the deque_raw function status as a completion of dma. After the completion of dma, do cache_inv. This step needs to be done based on how many times the user transfers DMA.

Regards,

S.Anil.

Hello S.Anil.

Thank you very much for your great support.

Customer understands why it needs to using CacheP_inv（）after DMA transfer now,  the further question is that why the code runs out of control after executing CacheP_inv（）.


Thanks
Regards
Shine

Hello Shina,

I verified customer code yesterday at a high level .

I need to check and debug further on my bench to see why the code goes out of control after calling cache_inv, and I will share the results once I have completed testing.

Regards,

S.Anil.

Hello Shina,

I have looked at the code, and the customer is trying to do a DMA transfer with the 4 byte?

Is this customer requirement to transfer only 4 bytes ? Can you please confirm how many bytes they need to transfer with the DMA ?

Actually, I have verified from my side, and the code seems to be working, and there is no issue at caheP_Inv.

and we don't need to call the App_udmaTrpdInit function every time, and this should be initiated only once since in the customer use case there is a change in

the source and destination buffer.

If you share above details I can give proper working code .

Is it application on NO RTOS or free RTOS ?

Regards,

S.Anil.

Hello S.Anil.,

Thank you very much.

Customer wants to use DMA to transfer large number of data to offload CPU. The 4 bytes is only for debug purpose.  So can the number of transfer be configured as preferred? 

He is using free RTOS. 

Thankds
Regards
Shine

Shine,

Yes, users can configure the large number of bytes transferred, and there is no problem.

My suggestion is that if the customer wants to create 5 UDMA channels on RTOS side ,. they can take the below example and add it to more channels in syscfg. One suggestion is that the App_udmaTrpdInit API should be called only once and not every time.

C:\ti\mcu_plus_sdk_am64x_09_00_00_31\examples\drivers\udma\udma_memcpy_interrupt\am64x-evm\r5fss0-0_freertos

Let me know if they need or face any problem after creating the example .

Regards,

S.Anil.

Hello S.Anil.,

Thank you very much.

The code customer sent is modified based on  mcu_plus_sdk_am64x_09_00_00_31\examples\drivers\udma\udma_memcpy_interrupt\am64x-evm\r5fss0-0_freertos. Because he didn't want to use interrupt, he did a cache_inv after the status checking in main() function was complete,  but the program would fly after cache_inv. He wants to solve this problem of flying, 

Thanks
Regards
Shine