On Tue, Sep 03, 2019 at 10:06:27PM -0700, Christoph Hellwig wrote: > On Tue, Sep 03, 2019 at 04:40:24PM -0700, Darrick J. Wong wrote: > > /me shrugs -- it's been broken for years, apparently, and we've been > > arguing with almost no action for months. Developers who are building > > things off of 5.3 should probably just add the patch (or turn off slub > > debugging).... > > Well, this is a new breakage with the same old root cause. The root > cause is that slub with debugging enabled gives us unaligned memory. > > But with my log recovery changes we may now use slub for allocations > larger than a single page (before we only used it for smaller than > page sizes allocations in the buffer cache), and that newly trips > up in the pmem driver (and a few others). Another ping. It would be really sad to release 5.3 with a known regression we've fixed two weeks before the release.
