Re: [Regression] [PATCH] iommu: Avoid crash in init_no_remapping_devices if iommu is NULL

[Torsten Hilbrich <torsten.hilbrich@secunet.com>]

On 01.09.20 04:02, Lu Baolu wrote:
[...]
> This looks more like a generic issue, used-before-allocated, and should
> be fixed in iommu.c instead of VT-d driver. How about
> 
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index 8fd93a5b8764..a599da87eb60 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -190,6 +190,28 @@ static void dev_iommu_free(struct device *dev)
>         dev->iommu = NULL;
>  }
> 
> +void *dev_iommu_priv_get(struct device *dev)
> +{
> +       struct dev_iommu *param = dev_iommu_get(dev);
> +
> +       if (WARN_ON(!param))
> +               return ERR_PTR(-ENOMEM);
> +
> +        return param->priv;
> +}
> +EXPORT_SYMBOL_GPL(dev_iommu_priv_get);
> +
> +void dev_iommu_priv_set(struct device *dev, void *priv)
> +{
> +       struct dev_iommu *param = dev_iommu_get(dev);
> +
> +       if (WARN_ON(!param))
> +               return;
> +
> +        param->priv = priv;
> +}
> +EXPORT_SYMBOL_GPL(dev_iommu_priv_set);
> +

This fix would work in my case. I tested it with slight modification to
replace the inline routines in include/linux/iommu.h.

The WARN_ON was not triggered during my tests. However, looking at the
definition of dev_iommu_get this is to be expected.

Regards,

	Torsten
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu