On Mon, 25 Aug 2025 02:34:30 -0700, Christoph Hellwig wrote: > On Tue, Aug 13, 2025 at 05:15:34PM +0800, alexjlzheng@xxxxxxxxx wrote: > > From: Jinliang Zheng <alexjlzheng@xxxxxxxxxxx> > > > > With iomap_folio_state, we can identify uptodate states at the block > > level, and a read_folio reading can correctly handle partially > > uptodate folios. > > > > Therefore, when a partial write occurs, accept the block-aligned > > partial write instead of rejecting the entire write. > > > > For example, suppose a folio is 2MB, blocksize is 4kB, and the copied > > bytes are 2MB-3kB. > > I'd still love to see some explanation of why you are doing this. > Do you have a workload that actually hits this regularly, and where > it makes a difference. Can you provide numbers to quantify them? Thank you for your reply. :) Actually, I discovered this while reading (and studying) the code for large folios. Given that short-writes are inherently unusual, I don't think this patchset will significantly improve performance in hot paths. It might help in scenarios with frequent memory hardware errors, but unfortunately, I haven't built a test scenario like that. I'm posting this patchset just because I think we can do better in exception handling: if we can reduce unnecessary copying, why not? Hahaha, just my personal opinion. :) thanks, Jinliang Zheng
