Hi Julein > On 16 Apr 2021, at 4:23 pm, Julien Grall <jul...@xen.org> wrote: > > > > On 16/04/2021 16:01, Rahul Singh wrote: >> Hi Julien, > > Hi Rahul, > >>> On 16 Apr 2021, at 3:35 pm, Julien Grall <jul...@xen.org> wrote: >>> >>> Hi, >>> >>> On 16/04/2021 12:25, Rahul Singh wrote: >>>> Revert the code that associates the group pointer with the S2CR as this >>>> code causing an issue when the SMMU device has more than one master >>>> device. >>> >>> It is not clear to me why this change was first added. Are we missing any >>> feature when reverting it? >> This feature was added when we backported the code from Linux to fix the >> stream match conflict issue >> as part of commit "xen/arm: smmuv1: Intelligent SMR allocation”. >> This is an extra feature added to allocate IOMMU group based on stream-id. >> If two device has the >> same stream-id then we assign those devices to the same group. > > If we revert the patch, then it would not be possible to use the SMMU if two > devices use the same stream-id. Is that correct?
No. If we revert the patch we can use the SMMU if two devices use the same stream-id without any issue but each device will be in a separate group.This is same behaviour before the code is merged. > >> This code was removed from Linux >> later point in time when IOMMU group handling is done by Linux code not by a >> specific IOMMU driver. > > Right.... But Linux still support that option. Is that correct? Yes Linux support IOMMU groups handling at Linux code level and not by every IOMMU driver. That why I though when we add the support for IOMMU group at XEN level we can add this functionality. > >> Therefore I think it is ok revert the code. > > I am ok with the principle of (partially) reverting patch to unblock the > situation. But I have to admit, I don't quite understand why this is reverted > rather than fixed. As I mention earlier I reverted this patch until we have proper support for IOMMU groups for ARM. Once we have IOMMU group support for ARM we can add this functionality. Regards, Rahul > > Cheers, > > -- > Julien Grall
