Re: [RFC v2 1/2] arm,smmu: switch to using iommu_fwspec functions

[Stefano Stabellini <sstabellini@kernel.org>]

On Tue, 21 Jul 2020, Brian Woods wrote:
> Modify the smmu driver so that it uses the iommu_fwspec helper
> functions.  This means both ARM IOMMU drivers will both use the
> iommu_fwspec helper functions.
> 
> Signed-off-by: Brian Woods <brian.woods@xilinx.com>

[...]

> @@ -1924,14 +1924,21 @@ static int arm_smmu_add_device(struct device *dev)
>  			ret = -ENOMEM;
>  			goto out_put_group;
>  		}
> +		cfg->fwspec = kzalloc(sizeof(struct iommu_fwspec), GFP_KERNEL);
> +		if (!cfg->fwspec) {
> +			kfree(cfg);
> +			ret = -ENOMEM;
> +			goto out_put_group;
> +		}
> +		iommu_fwspec_init(dev, smmu->dev);

Normally the fwspec structure is initialized in
xen/drivers/passthrough/device_tree.c:iommu_add_dt_device. However here
we are trying to use it with the legacy bindings, that of course don't
initialize in iommu_add_dt_device because they are different.

So I imagine this is the reason why we have to initialize iommu_fwspec here
indepdendently from iommu_add_dt_device.

However, why are we allocating the struct iommu_fwspec twice?

We are calling kzalloc, then iommu_fwspec_init is calling xzalloc.

Similarly, we are storing the pointer to struct iommu_fwspec in
cfg->fwspec but actually there is already a pointer to struct
iommu_fwspec in struct device (the one set by dev_iommu_fwspec_set.)

Do we actually need both?