Re: [PATCH v4 2/2] dma: add IOMMU static calls with clear default ops

[Leon Romanovsky]

On Wed, Sep 11, 2024 at 10:04:45AM +0200, Christoph Hellwig wrote:
> On Wed, Sep 11, 2024 at 09:43:05AM +0300, Leon Romanovsky wrote:
> > Thanks for the report, I'm looking into it. However, it is unclear to me
> > why my patch is causing this issue. The change in dma_supported() should
> > produce WARN_ON [1] if new path is taken, otherwise, we return to
> > previous behavior.
> 
> dma-iommu never implemented .dma_supported and thus claims to support
> all dma masks.  To restore that behavior we'd need something like the
> patch below:
> 
> diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
> index 7550b5dc5e55df..d23a4d5a6b31a1 100644
> --- a/kernel/dma/mapping.c
> +++ b/kernel/dma/mapping.c
> @@ -841,17 +841,19 @@ static int dma_supported(struct device *dev, u64 mask)
>  {
>  	const struct dma_map_ops *ops = get_dma_ops(dev);
>  
> -	if (WARN_ON(ops && use_dma_iommu(dev)))
> -		return false;

The below code still has merit. It is an error to have ops and take
dma-iommu path.

>  	/*
>  	 * ->dma_supported sets the bypass flag, so we must always call
>  	 * into the method here unless the device is truly direct mapped.
>  	 */
> -	if (!ops)
> -		return dma_direct_supported(dev, mask);
> -	if (!ops->dma_supported)
> -		return 1;
> -	return ops->dma_supported(dev, mask);
> +	if (ops) {
> +		if (!ops->dma_supported)
> +			return 1;
> +		return ops->dma_supported(dev, mask);
> +	}
> +
> +	if (use_dma_iommu(dev))
> +		return true;

I would simply put this hunk below if (WARN_ON ...) without any other
changes. Should I send a patch?

Thanks

> +	return dma_direct_supported(dev, mask);
>  }
>  
>  bool dma_pci_p2pdma_supported(struct device *dev)