Project

General

Profile

Bug #9532

OpenIndiana panics when used as guest in Oracle VM 3.4.4

Added by Andreas Grüninger about 1 year ago. Updated about 1 year ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
-
Start date:
2018-05-11
Due date:
% Done:

0%

Estimated time:
Difficulty:
Medium
Tags:
needs-triage

Description

A virtual guest can be installed and started when OI-hipster-text-20171031.iso is used as installation media.
The installation is finished after about 5 minutes.

When OI-hipster-text-20180427.iso is used the installation is awfully slow and crashes after several hours.
When the VM installed with OI-hipster-text-20171031.is updated with pkg update the system can not be booted.
Both trials create the same panic:
https://gist.github.com/grueni/b3315b42d856ef766510b7fad21a8066#file-dilos_error1-jpg
The check with the latest version of DilOS creates the same error.

format started in the shell (OI-hipster-text-20180427.iso) looks like this:
https://gist.github.com/grueni/b3315b42d856ef766510b7fad21a8066#file-oi_format-jpg

OI-hipster-text-20180427.iso contains the changes to the xen drivers.

Oracle VM 3.4.4 contains Xen 4.4.4.
I checked it with the xen-modes "xen_hvm" and "xen_hvm_pv".

History

#1

Updated by Yuri Pankov about 1 year ago

Well, there are several issues filed here.

First off, does OI use the illumos-gate source as is? I strongly doubt it as it's missing the fix for #8079 -- the install ISO boot hanged for me, until I did a apix_enable/w0 from kmdb.

#2

Updated by Yuri Pankov about 1 year ago

(I'll be adding notes as I go through this)

With apix_enable set to 0, 20180427 installation was successful and fast.

I also don't see how "both trials create the same panic" is possible -- first screenshot shows failure to mount root because of changed driver (cmdk->xdf), and when installing 20180427, you simply can't choose the cmdk-backed one to get that kind of problem.

Note that I'm using Citrix XenServer 7 and not Oracle VM, which could be causing the difference, but it's unlikely.

#3

Updated by Michal Nowak about 1 year ago

Yuri Pankov wrote:

Well, there are several issues filed here.

First off, does OI use the illumos-gate source as is? I strongly doubt it as it's missing the fix for #8079 -- the install ISO boot hanged for me, until I did a apix_enable/w0 from kmdb.

I believe that OpenIndiana goes with vanilla illumos-gate apart from https://github.com/OpenIndiana/oi-userland/tree/oi/hipster/components/openindiana/illumos-gate/patches.

I test OpenIndiana on Xen 4.10 and this is where it gets https://openqa.oi.mnowak.cz/tests/3250. The HVM machine gets acpi, apic, and pae (in libvirt terminilogy).

#4

Updated by Yuri Pankov about 1 year ago

OK, the apix issue I'm seeing is solved (see #9533).

But I still need the clarification on the first issue in the description -- which exactly panic you are seeing when installing 20180427?

#5

Updated by Andreas Grüninger about 1 year ago

This panic is not available.

Trial 1 was DilOS and I got this screen.

Trial 2:
I let the installation running over night and found in the morning the system again in the installation screen.
When I disabled the ISO and start the installed system I get this screen.

Trial 3:
So I decided to install a system with the older OI which worked.
And in this running system I used "pkg update" which finishes successfully.
But when I start the new BE I get this screen again

AFAIK OI uses the plain villa code from illumos-gate.

Do you have any hint what can I try to do?
Before starting the installation of 20180427 I checked in the shell whether apix_enable is 0. And I choose in the boot options OFF for apix.
I tried too several OS types.
No chance, still awfully slow, 23% after 15 h.

#6

Updated by Andreas Grüninger about 1 year ago

I installed on identical hardware Xenserver 7.4 and created a vm with 20180427m without any problems.
Then I migrated the virtual disk to OVM 3.4.4 and replaced the disk of an existing vm.
This vm on OVM boots but it you have to wait for up to 2 hours.

After login I tried to install a debug system from my own repository.
"beadm create test" worked but took 3 or 4 hours. Also beadm mount.

The installation into this new boot environment crashes with:

panic[cpu0]/thread=ffffff0007ac5c40:
I/O to pool 'rpool' appears to be hung.

ffffff0007ac5a70 zfs:vdev_deadman+10e ()
ffffff0007ac5ac0 zfs:vdev_deadman+4b ()
ffffff0007ac5af0 zfs:spa_deadman+af ()
ffffff0007ac5b90 genunix:cyclic_softint+fd ()
ffffff0007ac5ba0 unix:cbe_low_level+14 ()
ffffff0007ac5bf0 unix:av_dispatch_softvect+78 ()
ffffff0007ac5c20 unix:dispatch_softint+39 ()
ffffff0007a05a30 unix:switch_sp_and_call+13 ()
ffffff0007a05a70 unix:dosoftint+44 ()
ffffff0007a05ad0 unix:do_interrupt+ba ()
ffffff0007a05ae0 unix:cmnint+ba ()
ffffff0007a05bd0 unix:mach_cpu_idle+6 ()
ffffff0007a05c00 unix:cpu_idle+11a ()
ffffff0007a05c20 unix:idle+41 ()
ffffff0007a05c30 unix:thread_start+8 ()

dumping to /dev/zvol/dsk/rpool/dump, offset 65536, content: kernel

During execution of the installation I observed with iostat that there waiting IOs and service time (asvc_t) up to 14000 ms and more.
The dump is available but it has a size of 93gb.

Also available in: Atom PDF