[Xen-devel] rochester and Debian buster

[Xen-devel] rochester and Debian buster

[Ian Jackson]

We discussed on irc the problems I have been having trying to get
buster's released kernel to run on the rochesters, which is wanted to
upgrade osstest to buster (which is currently Debian stable).

Unfortunately our previous conversations don't seem to have been
recorded anywhere.  Let's try at least to write things down now.

The symptom is that the machine thinks the network link is down, and
no network stuff happens, so the installer doesn't work.  (I don't
think I have checked at the switch end whether the link is actually
up.)

You suggested that maybe adding
  iommu.passthrough=1
to the kernel command line might help.  But it hasn't.

I have a memory of discussing the next steps and I think we discussed
upgrading the firmware.  If I remember rightly we agreed (with
Juergen) that upgrading the firmware on one of the two rochester
machines would be an acceptable risk.  Can we file a ticket to have
that done by our onsite technician ?

Other options would include trying a buster-backports kernel, if we
had some reason to think that a newer kernel would be better.
Certainly if this turns out to be a kernel bug, and a workaround is
awkward, the best fix would be to get the bugfix backported to the
Debian buster kernel (either the stable series, or -backports).

Currently I have rochester1 booked out and you are free to play with
it if you like.  We'll negotiate about that on irc.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [Xen-devel] rochester and Debian buster

[Julien Grall]

Hi Ian,

On 24/10/2019 16:47, Ian Jackson wrote:
> We discussed on irc the problems I have been having trying to get
> buster's released kernel to run on the rochesters, which is wanted to
> upgrade osstest to buster (which is currently Debian stable).
> 
> Unfortunately our previous conversations don't seem to have been
> recorded anywhere.  Let's try at least to write things down now.
> 
> The symptom is that the machine thinks the network link is down, and
> no network stuff happens, so the installer doesn't work.  (I don't
> think I have checked at the switch end whether the link is actually
> up.)

I have just remembered that we are not using the on-board network card but 
instead a USB dongle. Looking at the log, it looks like the kernel found the dongle:

Oct 24 10:57:31.499085 [   14.421064] asix 3-1.3:1.0 eth5: register 'asix' at 
usb-0000:00:11.0-1.3, ASIX AX88772B USB 2.0 Ethernet, 00:50:b6:22:00:67

But I am not sure which eth interface is linked to and how you found out the 
network is down.

> 
> You suggested that maybe adding
>    iommu.passthrough=1
> to the kernel command line might help.  But it hasn't.
> 
> I have a memory of discussing the next steps and I think we discussed
> upgrading the firmware.  If I remember rightly we agreed (with
> Juergen) that upgrading the firmware on one of the two rochester
> machines would be an acceptable risk.  Can we file a ticket to have
> that done by our onsite technician ?

I am not entirely sure whether this would help because I have had report that 
Debian Buster has been installed successfully on Thunder-X.

Anyway, updating the firmware to something more recent would be good. We are 
currently using T48, and the most recent seems to be F02 [1].

It is not entirely clear to me what's the difference between the Txx and Fxx 
release. I have asked some help and will let you know.

Cheers,

[1] https://www.gigabyte.com/ARM-Server/R150-T61-rev-110/support#support-dl-bios


-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [Xen-devel] rochester and Debian buster

[Ian Jackson]

Julien Grall writes ("Re: rochester and Debian buster"):
> I have just remembered that we are not using the on-board network card but 
> instead a USB dongle.

Yes.

> But I am not sure which eth interface is linked to and how you found
> out the network is down.

I didn't have clear notes about this so I decided to repro the
problem.  I reran my setup, and it *passed*, installing buster
successfully on rochester1.  This is very odd and I can't explain it.

The two jobs' (143098 vs 143287) runvars are identical apart from the
syslog server port (which is expected to vary) and the
`host_power_install' which is only set after a successful boot into
the installer.  The database records the previous one as `broken' and
the new one as `pass'.  The osstest revision was identical.  My
previous test run, that failed, was a really quite formal
mg-repro-setup run, so I don't think it had any kind of weirdness that
would explain the symptoms.

My recollection is that both `ip link show' and kernel log messages
agreed that there was no NIC with a working network link.  I don't
recall how many were detected but I think I looked at the MAC address
from the osstest configuration and saw it in the list from the
machine.

> I am not entirely sure whether this would help because I have had
> report that Debian Buster has been installed successfully on
> Thunder-X.

Well, now you have a second.

I have not done enough of these install attempts to know if there is
some kind of heisenbug here.  The situation seems concerning.  I guess
for now I will carry on and see if it does it again...

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel