On Thu, Mar 18, 2021 at 11:55:43AM +0000, Matthew Wilcox wrote: > On Thu, Mar 18, 2021 at 11:29:28AM +0000, Luis Henriques wrote: > > On Thu, Mar 18, 2021 at 02:03:02PM +0300, Kirill A. Shutemov wrote: > > > On Thu, Mar 18, 2021 at 11:59:59AM +0100, Miklos Szeredi wrote: > > > > > [16247.536348] page:00000000dfe36ab1 refcount:673 mapcount:0 mapping:00000000f982a7f8 index:0x1400 pfn:0x4c65e00 > > > > > [16247.536359] head:00000000dfe36ab1 order:9 compound_mapcount:0 compound_pincount:0 > > > > > > > > This is a compound page alright. Have no idea how it got into fuse's > > > > pagecache. > > > > > > > > > Luis, do you have CONFIG_READ_ONLY_THP_FOR_FS enabled? > > > > Yes, it looks like Tumbleweed kernels have that config option enabled by > > default. And it this feature was introduced in 5.4 (the bug doesn't seem > > to be reproducible in 5.3). > > Can you try adding this patch? > > https://git.infradead.org/users/willy/pagecache.git/commitdiff/369a4fcd78369b7a026bdef465af9669bde98ef4 Yep, sure. Unfortunately, the testing round-trip can be a bit high. I'll push a new kernel build and ask the reporter to give it a try. [ I'll add this patch on top of the s/BUG_ON/VM_BUG_ON_PAGE change. ] Cheers, -- Luís
