Re: Xen dom0 freeze after domU exits (was Re: Zombie kernel thread)

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Sun, Sep 02, 2018 at 09:04:58AM +0200, Emmanuel Dreyfus wrote:
> Hello
> 
> I follow up on my previous report: On XEN3_DOM0 kernel I very often freeze the
> kernel after a domU exits. 
> 
> It seems related to vnconfig destroying the domU vnd backend, with vnconfig
> process stuck in an I/O inside readdisklabel(). The backtrace inside
> readdisklabel() may be a bit different. Last time it was
> biowait/scan_mbr/readdisklabel, this time it is
> biowait/convertdisklabel/validate_label/readdisklabel
> 
> Any hint how to debug this? A system freeze means there is some thread entered a
> spl() and waits there before a splx(), is that correct?

No it's more complicated. a sleep will reset the spl level.
I suspect it's some kind of deadlock, and in your case it seems to be on
fstrans. A thread has suspended /, and then wants to read something from /.

How is your system partitioned ? All my test systems have a separate
partition for the file-backed virtual disks, this may be why I don't see
this.

BTW there has been recent fixes in the fstrans area and they have beeen pulled
up to netbsd-8:
http://releng.netbsd.org/cgi-bin/req-8.cgi?show=1000

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--