On Thu, Dec 19, 2013 at 03:24:16PM -0500, Dave Jones wrote: > Yes. Note the original trace in this thread was a VM_BUG_ON(atomic_read(&page->_count) <= 0); > > Right after these crashes btw, the box locks up solid. So bad that traces don't > always make it over usb-serial. Annoying. I think I finally have an idea what's going on now. Kent's changes in e34ecee2ae791df674dfb466ce40692ca6218e43 are broken and result in a memory leak of the aio kioctx. This eventually leads to the system running out of memory, which ends up triggering the otherwise hard to hit error paths in aio_setup_ring(). Linus' suggested changes should fix the badness in the aio_setup_ring(), but more work has to be done to fix up the percpu reference counting tie in with the aio code. I'll fix this up in the morning if nobody beats me to it over night, as I'm just heading out right now. -ben > Dave > -- "Thought is the essence of where you are now." -- 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>