On Sat, Jun 11, 2022 at 07:34:01PM -0400, Chris Mason wrote: > On 6/9/22 5:15 PM, Chris Mason wrote: > > On 6/8/22 8:53 PM, Dave Chinner wrote: > > > > Reviewed-by: Dave Chinner <dchinner@xxxxxxxxxx> > > > > Thanks! Johannes and I are both going on vacation, but I'll get an > > experiment rolled to enough hosts to see if the long tails get shorter. > > We're unlikely to come back with results before July. > > > > Of course, the easiest way to test my theory is live patching against v5.6, > but there's a wrinkle because v5.6 still has xfs_vm_writepage() > > Looks like the iomap conversion deleted the warning that Jan was originally > fixing, and I went through some hoops to trigger skipping the pages from > inside writepage(). As you noted in the commit to delete writepage, this is > pretty hard to trigger but it does happen once I get down to a few hundred > MB free. It doesn't seem to impact fsx runs or other load, and we > unmount/xfs_repair cleanly. > > I don't like to ask people to think about ancient kernels, but am I missing > any huge problems? I've got this patch backported on v5.6, and writepage is > still in place. Just backport the patch to remove ->writepage as well? I've run test kernels since ~2017 with the writepage delete in it and never seen any problems as a result of it. That would seem to me to be the simplest way of avoiding unexpected issues with reclaim writeback and truncate races... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
