On Thu, May 02, 2024 at 10:23:41PM +0200, Andreas Gruenbacher wrote: > On Wed, Apr 3, 2024 at 7:24 PM Matthew Wilcox (Oracle) > <willy@xxxxxxxxxxxxx> wrote: > > For journalled data, folio migration currently works by writing the folio > > back, freeing the folio and faulting the new folio back in. We can > > bypass that by telling the migration code to migrate the buffer_heads > > attached to our folios. > > This part sounds reasonable, but I disagree with the following assertion: > > > That lets us delete gfs2_jdata_writepage() as it has no more callers. > > The reason is that the log flush code calls gfs2_jdata_writepage() > indirectly via mapping->a_ops->writepage. So with this patch, we end > up with a bunch of Oopses. > > Do you want to resend, or should I back out the gfs2_jdata_writepage > removal and add the remaining one-liner? Ugh, I see. If you could just add the one-liner for now, and I'll come back with a better proposal next for merge window?
