On 7/25/21 4:09 PM, Mike Galbraith wrote: > On Sat, 2021-07-24 at 00:39 +0200, Vlastimil Babka wrote: >> >> If not, then I would expect this to work (I don't think they ever nest >> in the opposite order, also lockdep should tell us instead of >> -ENOBOOT?), but might be missing something... > > Yeah, like #ifndef CONFIG_PREMPT_RT at the bottom of the loop that our > useless damn eyeballs auto-correct instead of reporting :) Well doh, good catch. Hope fixing that helps then? I've so far convinced myself that the full patch will be needed - in my theory the ->partial cmpxchg in put_cpu_partial() itself can race with assignment in ___slab_alloc() in a way that will not cause crashes, but will leak slab pages. > -Mike >
