As suggested by Comment #8 from Jan Kara I am forwarding this bug report: https://bugzilla.kernel.org/show_bug.cgi?id=201631 Begin forwarded message: Date: Tue, 11 Dec 2018 11:21:17 +0000 From: bugzilla-daemon@xxxxxxxxxxxxxxxxxxx To: erhard_f@xxxxxxxxxxx Subject: [Bug 201631] WARNING: CPU: 11 PID: 29593 at fs/ext4/inode.c:3927 .ext4_set_page_dirty+0x70/0xb0 https://bugzilla.kernel.org/show_bug.cgi?id=201631 Jan Kara (jack@xxxxxxx) changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |benh@xxxxxxxxxxxxxxxxxxx, | |jack@xxxxxxx, | |michael@xxxxxxxxxxxxxx --- Comment #8 from Jan Kara (jack@xxxxxxx) --- The lockups and ext4 warnings are likely a different thing. So please report the lockup to linux-fsdevel@xxxxxxxxxxxxxxx and also CC Al Viro <viro@xxxxxxxxxxxxxxxxxx> on it. The WARN_ON we hit in ext4 is likely: WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); because we don't expect page cache page to be marked dirty when unmapping it from page tables. It could be a powerpc specific race in handling of page table entry (PTE) flags as I don't see how PTE could be marked dirty without a page being already dirty. Seeing you hit this already twice on 4.20-rc's and never on 4.19.x, it could be some recent powerpc change that is triggering this. I've skimmed through powerpc changelogs and didn't find anything obviously matching but there were quite some mm changes pulled in this merge window. Powerpc guys, any idea how could _PAGE_DIRTY PTE bit get spuriously set? Normally, we should clear the _PAGE_DIRTY bit in page_mkclean() together with write-protecting the PTE and we are marking the page dirty whenever writeable PTE is created... So clean pagecache page with _PAGE_DIRTY in PTE is an invalid combination. -- You are receiving this mail because: You reported the bug. -- PGP-ID: 0x98891295 Fingerprint: 923B 911C 9366 E229 3149 9997 8922 516C 9889 1295 riot.im: @ernsteiswuerfel:matrix.org
