mmcspi: fix STOP retry

[br]

Summary

Currently, "retry on crc error" code is not reached, but this is needed on CVA6 where the first attempt to READ/WRITE multiple blocks always return CRC error.

To fix that, do not alter "retries" variable of entire command (which is set to 0), instead retry static amount of times (default 5) the stop command only in case of CRC errors.

Test Plan

Tested on CVA6 synthesized on Genesys II board

[pkelsey]

Re: "the first attempt to READ/WRITE multiple blocks always return CRC error" -

Are you saying the first attempts to read/write multiple blocks return a CRC error for the stop command, or for a data block, or both? One reason I am asking is that the stop command is only used for write multiple if there is a block CRC error returned, so if retrying the stop command was relevant in that case it would mean both a block being written and the stop command itself experienced CRC errors.

Overall, I am wondering if there isn't some underlying issue with the behavior of this mmc device or this board that is being papered over with this change without being completely understood. It would be interesting to see logs captured with all tracing in this file enabled showing the affected commands that are returning CRC errors.

sys/dev/mmc/mmcspi.c

174

typo: reponse -> response

1439

I don't have any argument against separating the STOP_TRANSMISSION retry policy from the retry policy for the command it is being used to stop, and it seems that your empirical results support doing this.

1451

Agree with changing this to a check for MMC_ERR_BADCRC above. It looks like this is code that was not properly updated after checking for CRC error response was folded into mmcspi_send_cmd().

1622

typo: respone -> response

[br]

Thanks @pkelsey. I will look at this once again (I did not forget about this but need to restore my CVA6 setup).