freebsd stable - Oct 2014 - Heads-up: Possible regression between 10.0-RELEASE and 10.1-BETA1 with '/ on ZFS' setup

During testing of the 10.1-RC1, I set up a multi-disk VirtualBox VM to
test '/ on ZFS' in various scenarios.  FreeBSD 10.1-RC1 i386, when
installed on ZFS with more than one disk (i.e., mirror, raidz-1,
raidz-2, etc.) crashes when rebooting post-install.

This does not happen with a single-drive '/ on ZFS' setup under the same
configuration.

FreeBSD 10.0-RELEASE i386 does not exhibit this behavior, however
upgrading from 10.0-RELEASE to 10.0-BETA1 or later may exhibit
a double-fault panic on boot.

A possible solution to this is to set kern.kstack_pages=4 via loader(8),
however in my tests (solely in VirtualBox), this has been ineffective.

It is unclear to me right now if this is something specific to
VirtualBox or not, though this problem was reported recently through
Bugzilla ( https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=194015 ).

I am still bisecting the stable/10 tree to try to identify when this
possible regression was introduced, in addition to scrounging up the
hardware to be able to test this on a bare-metal machine to determine if
this is a VirtualBox (or hypervisor in general) specific problem, but
this is taking longer than initially expected.

To be perfectly clear, the panic does occur in my particular testing
environment as far back as 10.1-BETA1, so this is *not* something new to
the upcoming 10.1-RC1.

If you have a multi-drive '/ on ZFS' setup (mirror, raidz-1) *without*
PAE, and have upgraded to 10.1-BETA1 or later, please speak up in case
this is a problem specific to my testing environment, which will likely
be at least another day before I can verify is the case.

Glen

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL:
<http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20141003/cf0de8b5/attachment.sig>

----- Original Message ----- 
From: "Glen Barber" <gjb at FreeBSD.org>
> During testing of the 10.1-RC1, I set up a multi-disk VirtualBox VM to
> test '/ on ZFS' in various scenarios.  FreeBSD 10.1-RC1 i386, when
> installed on ZFS with more than one disk (i.e., mirror, raidz-1,
> raidz-2, etc.) crashes when rebooting post-install.
> 
> This does not happen with a single-drive '/ on ZFS' setup under the
same
> configuration.
> 
> FreeBSD 10.0-RELEASE i386 does not exhibit this behavior, however
> upgrading from 10.0-RELEASE to 10.0-BETA1 or later may exhibit
> a double-fault panic on boot.
> 
> A possible solution to this is to set kern.kstack_pages=4 via loader(8),
> however in my tests (solely in VirtualBox), this has been ineffective.
> 
> It is unclear to me right now if this is something specific to
> VirtualBox or not, though this problem was reported recently through
> Bugzilla ( https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=194015 ).
> 
> I am still bisecting the stable/10 tree to try to identify when this
> possible regression was introduced, in addition to scrounging up the
> hardware to be able to test this on a bare-metal machine to determine if
> this is a VirtualBox (or hypervisor in general) specific problem, but
> this is taking longer than initially expected.
> 
> To be perfectly clear, the panic does occur in my particular testing
> environment as far back as 10.1-BETA1, so this is *not* something new to
> the upcoming 10.1-RC1.
> 
> If you have a multi-drive '/ on ZFS' setup (mirror, raidz-1)
*without*
> PAE, and have upgraded to 10.1-BETA1 or later, please speak up in case
> this is a problem specific to my testing environment, which will likely
> be at least another day before I can verify is the case.
This has been a known issue on i386 since the switch to Clang see UPDATING:
20121223:
        After switching to Clang as the default compiler some users of ZFS
        on i386 systems started to experience stack overflow kernel panics.
        Please consider using 'options KSTACK_PAGES=4' in such
configurations.

In my experience your millage may vary but essentially without 4 stack pages
all bets are off in terms of stability.

    Regards
    Steve

On 04/10/2014 05:40, Glen Barber wrote:
> A possible solution to this is to set kern.kstack_pages=4 via loader(8),
Are you sure of this?
At least with the current CURRENT kern.kstack_pages is just a read-only sysctl
and not a tunable.
> however in my tests (solely in VirtualBox), this has been ineffective.

-- 
Andriy Gapon