Hi Christoph, Still with my contrived memory layout where there is no physical memory the kernel can use below 4GB, it was possible to fail swiotlb_init(), but still not hit swiotlb_map_single() since all peripherals have a DMA_BIT_MASK() that is within the remaining addressable physical memory.
The second path could be backported to stable, but for the same reasons as the one we had just discussed before, this requires a very contrived test case that is not necessarily realistic or would warrant a stable backport IMHO. Thanks! Florian Fainelli (2): swiotlb: Group identical cleanup in swiotlb_cleanup() swiotlb: Return consistent SWIOTLB segments/nr_tbl kernel/dma/swiotlb.c | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-) -- 2.17.1 _______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu