Hi Shimoda-san,

thanks for the patches and for providing a test case. I was not able to
reproduce the issue, though. I'll explain...

> Note that this patch series is tested by using additional debug code [1],
> because there is difficult to reproduce this issue. Before apply patch,
> When I enabled CONFIG_DMA_API_DEBUG and CONFIG_DMA_API_DEBUG_SG,
> I observed lacking dma unmapping on /sys/kernel/debug/dma-api/dump.
> And then I confirmed the patch can fix the issue.

So, I have this debug patch applied on top of mmc/next. I have the above
CONFIG_ symbols enabled. I have _not_ applied your three patches which
fix the issue. I mounted the eMMC and read a large file. I see the
injected timeouts happening:

[   94.079560] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for SD bus idle
[   94.088668] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for SD bus idle
[   94.097727] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for SD bus idle
[   94.106768] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for SD bus idle
[   94.115848] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for SD bus idle
[   99.300589] renesas_sdhi_internal_dmac ee140000.sd: timeout waiting for hardware interrupt (CMD13)

But I do not see any output from the DMA debug system about a missing
unmapping. I expected that, though, because your fixes are not applied.
The testfile could even be correctly checksummed after reading, just
awfully slow, of course.

Am I missing something?

All the best,

   Wolfram