Strange 'hangs' with RELENG_9

[László KÁROLYI]

Volodymyr Kostyrko wrote:

> Obtaining kernel dump is the way to go. You can occasionally find that
> obtaining dump on world built with clang is much easier than on
> gcc-compiled one. I remember at least one such case with broken zfs
> directory when trying to read such directory result in process with no
> state on gcc-compiled world and result in imminent panic in zfs code
> on clang-compiled world. Remember to set dumpdev to something
> appropriate.
>
I'm not near to the server for putting any dumpdevice in (if you meant
that way), and also, there's no kernel panic but just at the end of the
reboot process, and not all the time.

> This has impact on performance. You sure you really need that one? Can
> you try without it?
>
>> /etc/make.conf, the kernel was compiled with this settings:
>> CPUTYPE?=athlon64
> Compiling with clang you better turn that thing off.
Okay, I turned it off, recompiled the kernel, also turned the
vfs.zfs.vdev.max_pending option off in /boot/loader.conf, rebooted, but
the hangs are still there.

Moreover, I couldn't set SCHED_BSD in the kernel config, it said that
it's an illegal option. Maybe it does not exist in RELENG_9.

Any other ideas?