On Thu, May 09, 2024 at 08:08:28AM -0700, Darrick J. Wong wrote: > Oh. Right, this is for bs>ps. For that case it makes sense to fail the > mount. I was only thinking about bs<=ps with large folios, where it > doesn't. > > (Would we use the zero-hugepage for large folios on a 4k fsblock fs?) The direct I/O zeroing code always deals with sub-blocksize amounts. The buffered zeroing path can deal with larger amounts in a few cases, but that goes through the page cache anyway.
