On Mon, Jul 01, 2019 at 07:02:06AM -0000, Michael van Elst wrote: > t...@giga.or.at (Thomas Klausner) writes: > > >So it looks like it could be a very extreme slowness instead of a > >complete deadlock. > > When it stops, try to reduce kern.maxvnodes to something low (like 100), > you can restore it, if the machine wakes up. > > If this is a memory shortage instead of a deadlock, this will easily > free kernel memory. You may need to wait 1-2 minutes to let the > page daemon collect free pages. > > If you cannot issue a sysctl command, you can patch the kernel variable > "desiredvnodes" with DDB instead.
>From some debugging so far, the cause for the hang seems to be that the nvme driver is waiting for an interrupt that doesn't come. At least once I got it to get unstuck by call "nvme_intr()" on the softc address from ddb. Thomas