Dave Chinner wrote: > On Tue, Jun 21, 2022 at 04:48:06PM -0700, Dan Williams wrote: > > Dave Chinner wrote: [..] > > One experiment I want to see is turn off DAX and see what the unmount > > performance is when this is targeting page cache for mmap I/O. My > > suspicion is that at some point dirty-page pressure triggers writeback > > that just does not happen in the DAX case. > > I'd be more interesting in the result of a full bisect to find > the actual commit that caused/fixed the behaviour you are seeing.
Ok, the one-shot-in-the-dark revert guess indeed runs the risk of something else in that build modulating the failure. Sravani, lets attempt the full bisect for this to add more confidence.
