On Tue, 25 Mar 2014, Sasha Levin wrote: > I have a lead on this. Consider the following: > > kmem_cache_alloc > __slab_alloc > local_irq_save() > deactivate_slab > __cmpxchg_double_slab > slab_unlock > __bit_spin_unlock > preempt_enable > [ Page Fault ] > > With this trace, it manifests as a "BUG: sleeping function called from invalid > context at arch/x86/mm/fault.c" on a might_sleep() in the page fault handler > (which is an issue on it's own), but I suspect it's also the cause of the > trace > above - preemption enabled and a race that removed the page. > > Could someone confirm please? The preempt count is incremented earlier in bit_spin_lock so the preempt_enable() should not do anything. -- 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>