On Mon, May 22, 2023 at 10:57:03PM +0900, Tetsuo Handa wrote: > On 2023/05/16 10:44, Huang, Ying wrote: > > Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> writes: > > > > > > > The above is based on analysis in > > > > https://lore.kernel.org/all/20190107204627.GA25526@xxxxxxxxxxx/ > > https://lore.kernel.org/linux-mm/d642e597-cf7d-b410-16ce-22dff483fd8e@xxxxxxxxxxxxxxxxxxx/ > > > > Tetsuo's patch avoids to take pgdat->kswapd_wait.lock when timer base > > lock is held via adding check for __GFP_KSWAPD_RECLAIM, so breaks the > > circular dependency chain. > > Yes. Mel, any questions on this patch? > No, I do not, the deadlock issue is more clear now. > Thomas Gleixner described this lock as kswapd_wait::lock at > https://lkml.kernel.org/r/168476016890.404.6911447269153588182.tip-bot2@tip-bot2 . > Should I resubmit this patch with s/pgdat->kswapd_wait/pgdat->kswapd_wait.lock/ or > s/pgdat->kswapd_wait/kswapd_wait::lock/ ? > I don't think a revision of what's in Andrew's tree is necessary. It could be improved by outlining the exact deadlock similar based on this thread but it's somewhat overkill as the fix has other obvious justifications on its own. Acked-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> Thanks. -- Mel Gorman SUSE Labs
