On Tue, Sep 29, 2020 at 11:30:24PM +0300, Yishai Hadas wrote: > > > flags. (see ODP_DMA_ADDR_MASK). Also, if we went through a > > > read->write access without invalidation why do we need to mask at > > > all ? the new access_mask should have the write access. > > Feels like a good idea to be safe here > It followed your note from V1 that the extra mask was really redundant, the > original code also didn't have it, but up-to-you. It wasn't hard to read from the diff that this was being done in all cases, not just as the result of ib_dma_map_page(). This is why I think it would be clearer with the control flow I suggested > > > > And all the pr_debugs around this code being touched should become > > > > mlx5_ib_dbg > > > We are in IB core, why mlx5_ib_debug ? > > oops, dev_dbg > Can it can done locally ? As long as the ib_device is avaiable the print function should be used Actually it is ibdev_dbg() these days Jason
