FYI, we noticed a +668 bytes kernel size regression due to commit: commit: 65c908b7bd08144e5298258920f61255aa7628ee (dma-mapping: add a dma_ops_bypass flag to struct device) git://git.infradead.org/users/hch/misc.git dma-bypass.4 Details as below (size data is obtained by `nm --size-sort vmlinux`): 4861c37e: dma-mapping: inline the fast path dma-direct calls 65c908b7: dma-mapping: add a dma_ops_bypass flag to struct device +---------------------------------+----------+----------+-------+ | symbol | 4861c37e | 65c908b7 | delta | +---------------------------------+----------+----------+-------+ | bzImage | 439680 | 439840 | 160 | | nm.t.dma_alloc_direct | 0 | 103 | 103 | | nm.t.dma_map_direct | 0 | 102 | 102 | | nm.T.dma_get_sgtable_attrs | 84 | 135 | 51 | | nm.T.dma_mmap_attrs | 84 | 135 | 51 | | nm.T.dma_free_attrs | 73 | 113 | 40 | | nm.T.dma_alloc_attrs | 76 | 113 | 37 | | nm.T.dma_map_resource | 100 | 132 | 32 | | nm.T.dma_max_mapping_size | 38 | 65 | 27 | | nm.T.dma_map_sg_attrs | 69 | 95 | 26 | | nm.T.dma_get_required_mask | 50 | 73 | 23 | | nm.T.dma_can_mmap | 33 | 52 | 19 | | nm.T.dma_cache_sync | 56 | 75 | 19 | | nm.T.dma_sync_sg_for_cpu | 56 | 75 | 19 | | nm.T.dma_sync_sg_for_device | 56 | 75 | 19 | | nm.T.dma_sync_single_for_cpu | 56 | 75 | 19 | | nm.T.dma_sync_single_for_device | 56 | 75 | 19 | | nm.T.dma_unmap_page_attrs | 61 | 80 | 19 | | nm.T.dma_unmap_resource | 61 | 80 | 19 | | nm.T.dma_unmap_sg_attrs | 61 | 80 | 19 | | nm.T.dma_map_page_attrs | 226 | 231 | 5 | +---------------------------------+----------+----------+-------+ Thanks, Kbuild test robot