Re: regression with SG DMA buf allocations and IOMMU in low-mem