Need help with unexpected reboot in 10.1-RELEASE

[Natacha Porté]

Hello,

on Thursday 27 November 2014 at 03:08, Shane Ambler wrote:
> On 26/11/2014 18:59, Natacha Porté wrote:
> > Hello,
> >
> > last week, I updated my main personal computer from 9.2-RELEASE to
> > 10.1-RELEASE. Since then, I experienced four sudden and unexpected
> > reboots (is that what is called "crashes"?). They were immediate, so
> > it's not a kernel panic (which keeps the system unusable for 15s before
> > rebooting). Nothing appears in the logs, but who knows what could be in
> > the uncommitted buffers?
> 
> I haven't had reboots but my machine has hung, forcing me to reset
> nearly every day. When it doesn't hang the usb system fails to create
> new devices forcing me to restart to access a disk.

So for the record, in case anyone stumbles back here, I noticed that
nVidia proprietary drivers were upgraded in the ports (relatively) close
to the release of 10.1-RELEASE, so I happened to have simultaneously
upgraded FreeBSD base from 9.2 to 10.1 and nVidia drivers from 331 to
340.

Since I downgraded the drivers to 331 (the exact variant installed in my
9.2-RELEASE setup), I haven't experienced a single crash, despite a
heavy use of World of Warcraft. So I guess the culprit is very likely to
be nVidia proprietary drivers, which makes the problem out-of-topic for
this ML.

However I did experience one freeze (which I believe designates the same
reality as "my machine has hung), during a poudriere run. Unfortunately
I wasn't home when it happened, so I can only describe a sudden loss of
network connectivity (other hosts saying "host is down" when trying to
ping it), and when I got physical access, I couldn't make the screen
leave stand-by mode, and the keyboard LED didn't toggle. I'm afraid that
without network, screen or keyboard LED there is nothing left to judge
whether there is still any activity going on.

When I have time to babysit a poudriere run I will try again, that's a
completely different problem but I would love to see it solved too
(assuming it is indeed reproducible).


Considering all that, I'm not sure the questions below are still
relevant, but I will answer them in case it is somehow useful.

> > I run with a ZFS root, and the zpool is directly on the unsliced disk.
> > I have a nVidia graphics card, with the proprietary driver, on two
> > screens with two displays (":0" and ":0.1") and two window managers.
> > It's an amd64 platform.
> 
> How much ram? one disk in zpool?

8 GB for RAM, one disk (half a TB) for the system and one SSD (107GB)
for game installations, each of them alone in their dedicated zpool.

> > I doubt this is a purely hardware issue, since I generally choose my
> > hardware for its reliability, and I regularly reached three-digit days
> > of uptime with 9.2-RELEASE.
> 
> I used to install updates and restart monthly on 9.2.

I tend to keep an unhealthy amount of state in the various programs I
have opened, so I find seldom convenient to reboot. I might even be a
few CVE's late because of that.

> > I did take a snapshot of my 9.2-RELEASE, so I'm one zfs rollback away
> > from checking whether it sill happens with 9.2-RELEASE. However, if as
> > is likely it does work around the problem, I will probably have a hard
> > time motivating myself to come back to the problem, rather than just
> > waiting for the next release to see whether it has been magically solved
> > without me.
> 
> CAUTION - If you performed a zpool upgrade after upgrading to 10.1 then
> you can't read the zpool in 9.x so the rollback will fail. The way back
> will involve creating a new pool and transferring data.

I'm aware of the non-reversibility of zpool upgrades, which is why I
usually wait a few months before upgrading the zpool (when I don't
forget it altogether).

On top of that I can't seem to remember how to install the new
bootloader, which is necessary before upgrading the pool too because I
have a ZFS root. So I won't upgrade the pool before looking up (again)
how to upgrade the bootloader of a mounted disk.



Thanks for your help,
Natacha