On Sat, Dec 21, 2019 at 03:03:53PM +0000, Tom Murphy wrote:
> In the intel iommu driver devices which only support 32bit DMA can't be
> direct mapped. The implementation of this is weird. Currently we assign
> it a direct mapped domain and then remove the domain later and replace
> it with a domain of type IOMMU_DOMAIN_IDENTITY. We should just assign it
> a domain of type IOMMU_DOMAIN_IDENTITY from the begging rather than
> needlessly swapping domains.
>
> Signed-off-by: Tom Murphy <murphyt7@xxxxxx>
> ---
> drivers/iommu/intel-iommu.c | 88 +++++++++++++------------------------
> 1 file changed, 31 insertions(+), 57 deletions(-)
>
> diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
> index 0c8d81f56a30..c1ea66467918 100644
> --- a/drivers/iommu/intel-iommu.c
> +++ b/drivers/iommu/intel-iommu.c
> @@ -5640,7 +5609,12 @@ static int intel_iommu_add_device(struct device *dev)
> domain = iommu_get_domain_for_dev(dev);
> dmar_domain = to_dmar_domain(domain);
> if (domain->type == IOMMU_DOMAIN_DMA) {
> - if (device_def_domain_type(dev) == IOMMU_DOMAIN_IDENTITY) {
> + /*
> + * We check dma_mask >= dma_get_required_mask(dev) because
> + * 32 bit DMA falls back to non-identity mapping.
> + */
> + if (device_def_domain_type(dev) == IOMMU_DOMAIN_IDENTITY &&
> + dma_mask >= dma_get_required_mask(dev)) {
> ret = iommu_request_dm_for_dev(dev);
> if (ret) {
> dmar_remove_one_dev_info(dev);
> --
> 2.20.1
>
Should this be dma_direct_get_required_mask? dma_get_required_mask may
return DMA_BIT_MASK(32) -- it callbacks into intel_get_required_mask,
but I'm not sure what iommu_no_mapping(dev) will do at this point?
