On Mon, Aug 10, 2009 at 01:30:44PM -0700, Roland Dreier wrote: > > > Looking at mlx4_write_mtt_chunk() I see that it calls > > mlx4_table_find() with a pointer to single dma_addr_t - dma_handle - > > while the dma addresses for the ICM memory is actually a list of > > different addresses covering possibly different sizes. I think > > mlx4_table_find() should be changed to support that, and then we can > > use calls to dma_sync_single_for_cpu()/dma_sync_single_for_device() > > with the correct dma addresses. > > No, I think we're careful that we write MTT ranges that don't cross a > page so there shouldn't be any problem.
But a contiguous ICM memory does not map, in general, to a contiguous DMA memory, so if dma_sync_single_for_*() does not harm anything than it does not do anything useful either. _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
