On Tue, Dec 11, 2012 at 08:58:47PM -0500, Rick Macklem wrote: > Ok, I'll test r243598 and then r243599 and r243835, just to > see if it really is this. > > I'll email when I have done this. If you test only r243598, I am sure that you would experience corruption. The r243599 should cause the deadlocks. > > > > > > > Also, do you use the post-r244095 kernel ? > > > > > > Before and after. The most recent tests were post-r244095. > > > (If anything the more recent kernels hang more easily.) > > > > > > > > > > > > > > Is your machine SMP ? > > > > > > Old, slow single core i386. > > > > Try this. Please note that this is mostly a debugging facility. > > > It seemed to help, but didn't stop the hangs completely. > r244125 without the patch would hang somewhere in a kernel > build. r244125 plus this patch ran almost 2 kernel builds > before it got hung. Can you try to look into the system state on the hang (on the kernel with the if (1 || patch applied) ? Using the ddb and recipe from the web page. Basically watch for a thread looping in the mnt_active iterator and threads owning vnode interlocks.
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:33 UTC