NXP: tests/drivers/dma/loop_transfer fails on FRDM-K64F

[henrikbrixandersen]

Describe the bug

The tests/drivers/dma/loop_transfer test case fails on the NXP FRDM-K64F. Not all DMA RX buffers contain the DMA TX buffer string.

To Reproduce
Steps to reproduce the behavior:

1. cd tests/drivers/dma/loop_transfer; mkdir build; cd build
2. cmake -DBOARD=frdm_k64f
3. make flash
4. See the following output on the board console:

*** Booting Zephyr OS build zephyr-v2.3.0-898-g668e359d8558  ***
DMA memory to memory transfer started on DMA_0
Preparing DMA Controller
Starting the transfer and waiting for 1 second
TX data: The quick brown fox jumps over the lazy dog ....
block_size 48
I: channel 0 error status is 0x2
ERROR: unfinished transfer
Each RX buffer should contain the full TX buffer string.
TX data: The quick brown fox jumps over the lazy dog ....
RX data Loop 0: The quick brown fox jumps over the lazy dog ....
RX data Loop 1: The quick brown fox jups
RX data Loop 2: 
RX data Loop 3: 
RX data Loop 4: 
Finished: DMA

Expected behavior
All RX buffers contain the TX buffer content.

Impact
Not sure if the EDMA driver works correctly. The tests/drivers/dma/chan_blen_transfer test case passes, so something is working...

Environment (please complete the following information):

• OS: Linux
• Toolchain Zephyr SDK
• Commit SHA: 668e359

Additional context
I found the issue while enabling the eDMA driver on the NXP KE1xF SoC (see #26813).

[hakehuang]

@henrikbrixandersen this is a racing condtion which cause the dma access flash which will cause error on dma src, and it is not happened always, but with high possibilities. so the solution is to change the test cases dma.c as below:

#else
/* pad to times of 8*/
- static const char tx_data[] =
+ static char tx_data[] = 
	"The quick brown fox jumps over the lazy dog ....";
static __aligned(16) char rx_data[TRANSFER_LOOPS][RX_BUFF_SIZE] = { { 0 } };
#endif

or you can try my commit below:
hakehuang@f805872

let me know if this fixes your problem, I can create a pull request to for this fix

[henrikbrixandersen]

@hakehuang Thank you. I can confirm that your commit mentioned above fixes the race condition.