Hello, On Mon, Aug 25, 2025 at 12:13:53PM +0200, Jan Kara wrote: > True, but is that mechanism really needed? Given the approximate nature of > foreign flushing, couldn't we just always replace the oldest foreign entry > regardless of whether the writeback is running or not? I didn't give too > deep thought to this but from a quick look this should work just fine... Maybe, it's a bit difficult to predict. The thing was built to work around real cases of problem and FWIW we haven't heard of similar issues afterwards. Given the peculiarity and subtlety of the problem, I wonder keeping the behavior unchanged, if reasonably possible, is an easier path to follow here. Thanks. -- tejun
