Re: [PATCH 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] mmc: mmci: DMA API violation due to map/unmap device mismatch
--

commit d094ad885249a508b0979254780f5dd030ce3bf9
Author: Frank Li <Frank.Li@nxp.com>

mmc: use dmaengine_get_dma_device() instead of chan->device->dev

This commit replaces direct access to the chan->device->dev field
with dmaengine_get_dma_device(chan) to retrieve the correct DMA device, 
especially for engines with per-channel IOMMU mappings. This also prepares
for making the DMA engine provider data structures private.

> diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
> index 416bdb184ed4c..5d81c96ca4f37 100644
> --- a/drivers/mmc/host/mmci.c
> +++ b/drivers/mmc/host/mmci.c
[ ... ]
> @@ -940,7 +940,7 @@ static void mmci_dma_unmap(struct mmci_host *host, struct mmc_data *data)
>  	else
>  		chan = dmae->tx_channel;
>  
> -	dma_unmap_sg(chan->device->dev, data->sg, data->sg_len,
> +	dma_unmap_sg(dmaengine_get_dma_device(chan), data->sg, data->sg_len,
>  		     mmc_get_dma_dir(data));
>  }

[Severity: High]
Does this introduce a device mismatch with the mapping paths in the same file?

The map path in _mmci_dmae_prep_data() still maps the scatterlist using the
underlying DMA controller device:

drivers/mmc/host/mmci.c:_mmci_dmae_prep_data() {
    ...
	nr_sg = dma_map_sg(device->dev, data->sg, data->sg_len,
			   mmc_get_dma_dir(data));
    ...
 unmap_exit:
	dma_unmap_sg(device->dev, data->sg, data->sg_len,
		     mmc_get_dma_dir(data));
}

Since dmaengine_get_dma_device(chan) can return a different device structure
than chan->device->dev, does passing mismatched device structures to
dma_map_sg() and dma_unmap_sg() violate the DMA API contract?

[Severity: Medium]
This is a pre-existing issue, but I noticed there are other files in the
MMC subsystem that still directly access chan->device->dev which appear to
have been missed by this patch's search strategy.

For example, in drivers/mmc/host/mxcmmc.c:mxcmci_setup_data():

	nents = dma_map_sg(host->dma->device->dev, data->sg,
				     data->sg_len,  host->dma_dir);

And in drivers/mmc/host/omap.c:mmc_omap_dma_cb():

		dev = c->device->dev;

Will leaving these in place cause build breakages when the DMA engine provider
structures are finally made private?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260917204240.1284796-1-Frank.Li@oss.nxp.com?part=1