Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[Frank Li]

On Wed, Sep 23, 2026 at 02:45:47PM -0400, Frank.Li@oss.nxp.com wrote:
> From: Frank Li <Frank.Li@nxp.com>
>
> Replace direct dma_chan::device::dev access with the proper
> dmaengine_get_dma_device() for consumer API
>
> chan->device->dev is not always the device used for DMA mapping.
> Some DMA engines support per-channel IOMMU mappings, so different
> channels may use different DMA devices.  dmaengine_get_dma_device()
> returns the correct device for each channel.
>
> This also prepares for making the DMA engine provider data structures
> private. DMA consumers should not access DMA engine internals directly.
>
> mxs-mmc.c wrong use mmc host device when do DMA map. Fix it by use
> dmaengine_get_dma_device().
>
> Assisted-by: LLM
> Reviewed-by: Linus Walleij <linusw@kernel.org>
> Reviewed-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
> Reviewed-by: Vinod Koul <vkoul@kernel.org>
> Signed-off-by: Frank Li <Frank.Li@nxp.com>
> ---
> Change in v4
> - Fix miss "R" at Reviewed-by tags.
> - Fix error path unmap dev at mxs-mmc.c (sashakio reported)
> - move dmaengine_get_dma_device(host->dma) after mxcmci_use_dma() to avoid
> defer host->dma if use pio mode (sashakio reported)
>
> Change in v3
> - add Linus Walleij and Vinod Koul's review by tags
> - Add missed omap and mxs-mmc.c (sashakio reported)
>
> Change in v2
> - find more place, which use device->dev for dma mapping. This time should
> clean all.
> - Reviewed-by: Linus Walleij <linusw@kernel.org>
>
> Cc: imx@lists.linux.dev
> Cc: vkoul@kernel.org
> ---
>  drivers/mmc/host/Kconfig                 |  2 +-

Please forget this version, I make misstake to include a debug code, sorry
for annoise.

Frank

Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[Arnd Bergmann]

On Wed, Sep 23, 2026, at 20:45, Frank.Li@oss.nxp.com wrote:
> @@ -214,7 +214,7 @@ config MMC_SDHCI_OF_AT91
>  config MMC_SDHCI_OF_ESDHC
>  	tristate "SDHCI OF support for the Freescale eSDHC controller"
>  	depends on MMC_SDHCI_PLTFM
> -	depends on PPC || ARCH_MXC || ARCH_LAYERSCAPE || COMPILE_TEST
> +	depends on PPC || ARCHMXC || ARCH_LAYERSCAPE || COMPILE_TEST
>  	select MMC_SDHCI_IO_ACCESSORS
>  	select FSL_GUTS

This looks like an accidental bug that should be removed from the patch.

> @@ -961,7 +961,7 @@ static void atmci_dma_cleanup(struct atmel_mci *host)
>  	struct mmc_data                 *data = host->data;
> 
>  	if (data)
> -		dma_unmap_sg(host->dma.chan->device->dev,
> +		dma_unmap_sg(dmaengine_get_dma_device(host->dma.chan),
>  				data->sg, data->sg_len,
>  				mmc_get_dma_dir(data));

Since you have the exact construct in a lot of drivers, would
it make sense to take this one step further and introduce
a new set of helpers in drivers/dma/dmaengine.h, such as

unsigned int dmaengine_map_sg(struct dma_chan *chan, struct scatterlist *sg, ...);

      Arnd

Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[Frank Li]

On Wed, Sep 23, 2026 at 08:57:04PM +0200, Arnd Bergmann wrote:
> On Wed, Sep 23, 2026, at 20:45, Frank.Li@oss.nxp.com wrote:
> > @@ -214,7 +214,7 @@ config MMC_SDHCI_OF_AT91
> >  config MMC_SDHCI_OF_ESDHC
> >  	tristate "SDHCI OF support for the Freescale eSDHC controller"
> >  	depends on MMC_SDHCI_PLTFM
> > -	depends on PPC || ARCH_MXC || ARCH_LAYERSCAPE || COMPILE_TEST
> > +	depends on PPC || ARCHMXC || ARCH_LAYERSCAPE || COMPILE_TEST
> >  	select MMC_SDHCI_IO_ACCESSORS
> >  	select FSL_GUTS
>
> This looks like an accidental bug that should be removed from the patch.

It is accidently including my local change to build test. I found it when
post it.

>
> > @@ -961,7 +961,7 @@ static void atmci_dma_cleanup(struct atmel_mci *host)
> >  	struct mmc_data                 *data = host->data;
> >
> >  	if (data)
> > -		dma_unmap_sg(host->dma.chan->device->dev,
> > +		dma_unmap_sg(dmaengine_get_dma_device(host->dma.chan),
> >  				data->sg, data->sg_len,
> >  				mmc_get_dma_dir(data));
>
> Since you have the exact construct in a lot of drivers, would
> it make sense to take this one step further and introduce
> a new set of helpers in drivers/dma/dmaengine.h, such as
>
> unsigned int dmaengine_map_sg(struct dma_chan *chan, struct scatterlist *sg, ...);

If add this, suppose should be set of map_*, like dmaengine_map_single(),

Vinod:
	Is it add new APIs like thats?

Frank

>
>       Arnd

Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[Frank Li]

On Wed, Sep 23, 2026 at 02:33:35PM -0500, Frank Li wrote:
> On Wed, Sep 23, 2026 at 08:57:04PM +0200, Arnd Bergmann wrote:
> > On Wed, Sep 23, 2026, at 20:45, Frank.Li@oss.nxp.com wrote:
> > > @@ -214,7 +214,7 @@ config MMC_SDHCI_OF_AT91
> > >  config MMC_SDHCI_OF_ESDHC
> > >  	tristate "SDHCI OF support for the Freescale eSDHC controller"
> > >  	depends on MMC_SDHCI_PLTFM
> > > -	depends on PPC || ARCH_MXC || ARCH_LAYERSCAPE || COMPILE_TEST
> > > +	depends on PPC || ARCHMXC || ARCH_LAYERSCAPE || COMPILE_TEST
> > >  	select MMC_SDHCI_IO_ACCESSORS
> > >  	select FSL_GUTS
> >
> > This looks like an accidental bug that should be removed from the patch.
>
> It is accidently including my local change to build test. I found it when
> post it.
>
> >
> > > @@ -961,7 +961,7 @@ static void atmci_dma_cleanup(struct atmel_mci *host)
> > >  	struct mmc_data                 *data = host->data;
> > >
> > >  	if (data)
> > > -		dma_unmap_sg(host->dma.chan->device->dev,
> > > +		dma_unmap_sg(dmaengine_get_dma_device(host->dma.chan),
> > >  				data->sg, data->sg_len,
> > >  				mmc_get_dma_dir(data));
> >
> > Since you have the exact construct in a lot of drivers, would
> > it make sense to take this one step further and introduce
> > a new set of helpers in drivers/dma/dmaengine.h, such as
> >
> > unsigned int dmaengine_map_sg(struct dma_chan *chan, struct scatterlist *sg, ...);
>
> If add this, suppose should be set of map_*, like dmaengine_map_single(),
>
> Vinod:
> 	Is it add new APIs like thats?

Maybe we can enhence dmaengine_prep_slave_sg(),

Most patten like

	sglen = dma_map_sg(chan->device->dev, data->sg,
			data->sg_len, mmc_get_dma_dir(data));

	dmaengine_slave_config(chan, &host->dma_conf);
	desc = dmaengine_prep_slave_sg(chan,
			data->sg, sglen, slave_dirn,
			DMA_PREP_INTERRUPT | DMA_CTRL_ACK);

let dmaengine_prep_slave_sg() include dma_map_sg(), which also reduce
error handle when return desc is NULL.

after call complete, auto ummap_sg(). We may add DMA_PREP_MAP flags to
keep back compatiblity.

after enhancement
	consumer just need call

	dmaengine_prep_slave_sg(chan, data->sg, data->sg_len, slave_dirn,
			        DMA_PREP_INTERRUPT | DMA_CTRL_ACK |
				DMA_PREP_MAP)

Frank
>
> Frank
>
> >
> >       Arnd

Re: [PATCH v4 1/1] mmc: use dmaengine_get_dma_device() instead of chan->device->dev

[Arnd Bergmann]

On Wed, Sep 23, 2026, at 23:40, Frank Li wrote:
> On Wed, Sep 23, 2026 at 02:33:35PM -0500, Frank Li wrote:
>> On Wed, Sep 23, 2026 at 08:57:04PM +0200, Arnd Bergmann wrote:

> after enhancement
> 	consumer just need call
>
> 	dmaengine_prep_slave_sg(chan, data->sg, data->sg_len, slave_dirn,
> 			        DMA_PREP_INTERRUPT | DMA_CTRL_ACK |
> 				DMA_PREP_MAP)

If that works, I think that would be perfect, good idea!

       Arnd