On Wed, Sep 03, 2025 at 05:54:32AM +0100, Al Viro wrote: > Branch force-pushed into > git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #work.mount > (also visible as #v3.mount, #v[12].mount being the previous versions) > Individual patches in followups. > > If nobody objects, this goes into #for-next. PS: survives LTP, xfstests and mount-related selftests. FWIW, I've spent the weekend trying to figure out what's going on with generic/475. Turns out that it was not a regression - it goes back at least to 6.12 and it's triggered by PREEMPT vs. PREEMPT_VOLUNTARY in config. The former gives several kinds of failures, with total frequency about 8%; the latter apparently works - if any similar failures happen, the frequency is at least an order of magnitude lower. One useful thing I've got out of that is a bunch of helpers for doing bisect for configs - semi-manual decomposing the difference between two configs into a series of small changes, allowing to do bisection on that. Unfortunately, the change it has converged to (and repeating it alone on the original config reproduces the effect) is not particulary useful - some race gets triggered by a config change that affects timings all over the place ;-/
