Re: [PATCH] bus: fsl-mc: set dma_mask for the root DPRC device

[Ioana Ciornei]

On Fri, Jul 24, 2026 at 12:56:33AM +0200, Vincent Jardin via B4 Relay wrote:
> From: Vincent Jardin <vjardin@xxxxxxx>
> 
> On Layerscape platforms, when SMMU is disabled,
> fsl_mc_device_add() leads to the DPRC device with a NULL dma_mask.
> 
> It has been tested on LX2160A, no SMMU on the MC
> domain: the root DPRC can probe cleanly and fsl_mc children
> enumerate.
> 
> Fixes: a259ed1618d2 ("bus/fsl-mc: support dma configure for devices on fsl-mc bus")

Is this fixing an actual bug introduced by the blamed commit or are we
talking only about the warning message?

Commit 4d8bde883bfb ("OF: Don't set default coherent DMA mask") is the
one that added the dev_warn.

> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Vincent Jardin <vjardin@xxxxxxx>
> ---
> fsl_mc_device_add() already seeds a default dma_mask for non-DPRC
> objects, but never for the DPRC devices themselves. When the MC I/O
> domain is not behind an SMMU, the root DPRC reaches
> fsl_mc_dma_configure() -> of_dma_configure_id() with a NULL dma_mask
> and the probe warns "DMA mask not set".
> This mirrors the existing
> non-DPRC seeding into the DPRC branch.
> 
> Tested on LX2160A with no SMMU on the MC domain: the root DPRC probes
> cleanly and its fsl_mc children enumerate.
> ---
>  drivers/bus/fsl-mc/fsl-mc-bus.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
> index 64d75eed0d34..585345ac0179 100644
> --- a/drivers/bus/fsl-mc/fsl-mc-bus.c
> +++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
> @@ -832,6 +832,14 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
>  		error = get_dprc_icid(mc_io2, obj_desc->id, &mc_dev->icid);
>  		if (error < 0)
>  			goto error_cleanup_dev;
> +
> +		/*
> +		 * Seed dma_mask so platforms without an SMMU don't complain
> +		 * with "DMA mask not set" in of_dma_configure_id().
> +		 */
> +		mc_dev->dma_mask = FSL_MC_DEFAULT_DMA_MASK;
> +		mc_dev->dev.dma_mask = &mc_dev->dma_mask;
> +		mc_dev->dev.coherent_dma_mask = mc_dev->dma_mask;

Since as you said this is an exact mirror for the non-DPRC case, can you
move it to the common path before the if-else statement without
duplicating it?

Ioana