Public service announcement: Changes that went in for the v4.5 kernel seems to be disrupting page pinning for vfio. This potentially means that VM pages could change out from under the IOMMU mapping resulting in incorrect DMA operations. Since v4.5 kernels are now starting to trickle into stable distribution releases, I would encourage transparent huge pages be disabled in those kernels for the time being. This can easily be done by adding the kernel boot parameter 'transparent_hugepage=never'. Use of hugetlbfs for VM memory would likely also avoid this issue. I expect we'll have a fix soon, which will be backported to v4.5 via stable. Thanks,
Alex
_______________________________________________ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
