Re: [PATCH v3 2/4] iommu/dma-iommu: Handle deferred devices

[Christoph Hellwig]

On Mon, May 06, 2019 at 07:52:04PM +0100, Tom Murphy wrote:
> +static int handle_deferred_device(struct device *dev)
> +{
> +	struct iommu_domain *domain;
> +	const struct iommu_ops *ops;
> +
> +	if (!is_kdump_kernel())
> +		return 0;
> +
> +	domain = iommu_get_domain_for_dev(dev);

> -	dma_handle =__iommu_dma_map(dev, phys, size,
> +	if (unlikely(handle_deferred_device(dev)))
> +		return DMA_MAPPING_ERROR;
> +
> +	dma_handle = __iommu_dma_map(dev, phys, size,

__iommu_dma_map already looks up the domain, and as far as I can
tell all callers need the handle_deferred_device call.  Should we
just move it to there and pass the domain from the caller?

Also shouldn't the iommu_attach_device call inside
handle_deferred_device also get an unlikely marker?