2015-12-14 13:57 GMT+03:00 Michal Hocko <mhocko@xxxxxxx>: > On Mon 14-12-15 13:13:22, Andrey Ryabinin wrote: >> 2015-12-14 13:05 GMT+03:00 Michal Hocko <mhocko@xxxxxxx>: >> > On Mon 14-12-15 11:24:33, Kirill A. Shutemov wrote: >> >> On Thu, Dec 10, 2015 at 05:48:01PM +0200, Mika Westerberg wrote: >> >> > Hi Kirill, >> >> > >> >> > I got following crash on my desktop machine while building swift. It >> >> > reproduces pretty easily on 4.4-rc4. >> >> > >> >> > Before it happens the ld process is killed by OOM killer. I attached the >> >> > whole dmesg. >> >> > >> >> > [ 254.740603] page:ffffea00111c31c0 count:2 mapcount:0 mapping: (null) index:0x0 >> >> > [ 254.740636] flags: 0x5fff8000048028(uptodate|lru|swapcache|swapbacked) >> >> > [ 254.740655] page dumped because: VM_BUG_ON_PAGE(!PageLocked(page)) >> >> > [ 254.740679] ------------[ cut here ]------------ >> >> > [ 254.740690] kernel BUG at mm/memcontrol.c:5270! >> >> >> >> >> >> Hm. I don't see how this can happen. >> > >> > What a coincidence. I have just posted a similar report: >> > http://lkml.kernel.org/r/20151214100156.GA4540@xxxxxxxxxxxxxx except I >> > have hit the VM_BUG_ON from a different path. My suspicion is that >> > somebody unlocks the page while we are waiting on the writeback. >> > I am trying to reproduce this now. >> >> Guys, this is fixed in rc5 - dfd01f026058a ("sched/wait: Fix the >> signal handling fix"). >> http://lkml.kernel.org/r/<20151212162342.GF11257@xxxxxxxxxxxxxxxxxxx> > > Hmm, so you think that some callpath was doing wait_on_page_locked and > the above bug would allow a race and then unlock the page under our > feet? It rather more simple, read report carefully from the link I gave. __wait_on_bit_lock() in __lock_page() could just return -EINTR and leave the page unlocked. So in rc4 lock_page() simply didn't work (sometimes). > That would make some sense to me but I haven't checked the code to > see which path that would be. I am also not able to reproduce this > again... > -- > Michal Hocko > SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>