On Wed, 2016-03-30 at 09:49 +0200, Jan Kara wrote: > On Tue 29-03-16 18:57:16, Verma, Vishal L wrote: > > > > On Mon, 2016-03-28 at 16:34 -0700, Dan Williams wrote: > > > > <> > > > > > > > > Seems kind of sad to fail the fault due to a bad block when we > > > were > > > going to zero it anyway, right? I'm not seeing a compelling > > > reason to > > > keep any zeroing in fs/dax.c. > > Agreed - but how do we do this? clear_pmem needs to be able to clear > > an > > arbitrary number of bytes, but to go through the driver, we'd need > > to > > send down a bio? If only the driver had an rw_bytes like interface > > that > > could be used by anyone... :) > Actually, my patches for page fault locking remove zeroing from > dax_insert_mapping() and __dax_pmd_fault() - the zeroing now happens > from > the filesystem only and the zeroing in those two functions is just a > dead > code... That should make things easier! Do you have a tree I could merge in to get this? (WIP is ok as we know that my series will depend on yours..) or, if you can distill out that patch on a 4.6-rc1 base, I could carry it in my series too (your v2's 3/10 doesn't apply on 4.6-rc1..) Thanks, -Vishal��.n������g��� �a��� �&ޖ)���)��h���&������梷���� �Ǟ�m������)������^�����������v���O��zf������