[elrepo] AMD EPYC Support

[Dave Chiluk]

Looking at your stack trace from presumably the guest it's stuck in

RIP: 0010:aesni_gcm_init+0x89/0x30f [aesni_intel]


Which implies that it might be some intel specific logic.  I'd suspect that
the cpu passthrough is not working as you'd expect and that the cpu model
is showing up as something intel specific which is causing your guest to go
down this code path.  You could double check by seeing what the Vendor and
Modle is showing up as.  So back to my original suggestion you should
likely try a modern version of qemu and try again.

Dave.

On Tue, Aug 21, 2018 at 11:34 AM Dave Chiluk <dchiluk at indeed.com> wrote:

> You should try Cent 7, or some other more modern distro.  It's
> unreasonable to expect ridiculously old software to work with the
> newest/latest and greatest toys.  Additionally you are unlikely to receive
> useful qemu/kvm specific support on this mailing list.  I suggest you try
> https://www.qemu.org/, where they will likely tell you to upgrade as
> well.  Alternatively you could try to find another repo that provides new
> qemu packages built on Centos 6.
>
> Dave.
>
> On Mon, Aug 20, 2018 at 2:54 PM <nick at ramnode.com> wrote:
>
>> So we’ve discovered two issues during testing so far:
>>
>>
>>
>>    1. With AES-NI enabled in BIOS and kernel-ml el6 installed (whether
>>    booted into or not), KVM guest template installs take exactly 10 minutes.
>>    libguestfs-test-tool takes 3 minutes after hanging at the section in the
>>    attached output.
>>
>>    With AES-NI disabled, installs take the regular time (<1 min) and
>>    libguestfs-test-tool completes in about 10 seconds.
>>
>>
>>
>>    1. With the host running kernel-ml el6, new KVM guests with CPU
>>    passthrough enabled display CPU cores incorrectly after installing the OS,
>>    running yum/apt-get update, and rebooting.
>>
>>
>>
>> # lscpu
>>
>> Architecture:          x86_64
>>
>> CPU op-mode(s):        32-bit, 64-bit
>>
>> Byte Order:            Little Endian
>>
>> CPU(s):                2
>>
>> On-line CPU(s) list:   0
>>
>> Off-line CPU(s) list:  1
>>
>>
>>
>> All cores show online after another reboot. This happens whether the OS
>> is installed via template or ISO. It happens regardless of AES being
>> enabled or disabled. It does not happen with CPU passthrough disabled.
>>
>>
>>
>>
>>
>> Has anyone seen anything like this or have any idea what additional
>> troubleshooting steps we should take?
>>
>> Regards,
>>
>> Nick
>>
>>
>>
>> *From:* Dave Chiluk <dchiluk at indeed.com>
>> *Sent:* Thursday, August 16, 2018 3:39 PM
>> *To:* nick at ramnode.com
>>
>> *Cc:* EL Repo General Mailing List <elrepo at lists.elrepo.org>
>>
>>
>> *Subject:* Re: [elrepo] AMD EPYC Support
>>
>>
>>
>> The packages for centos 6 exist, I'm curious to hear how it goes.
>> Considering you'll be running KVM on top of that, I'd guess there is
>> probably some performance loss by running old version of KVM.
>>
>>
>>
>> -- Change is hard --
>>
>>
>>
>> Dave.
>>
>>
>>
>>
>>
>>
>>
>> On Thu, Aug 16, 2018 at 2:02 PM <nick at ramnode.com> wrote:
>>
>> Thanks Dave. What do you think about trying kernel-ml on CentOS 6? We’re
>> trying to do a direct swap of Intel for AMD without disrupting the existing
>> systems (CentOS 6/kernel-lt el6 running KVM). Hoping to keep changes down
>> to just reconfiguring networking.
>>
>>
>>
>> Regards,
>>
>> Nick
>>
>>
>>
>> *From:* elrepo-bounces at lists.elrepo.org <elrepo-bounces at lists.elrepo.org>
>> *On Behalf Of *Dave Chiluk via elrepo
>> *Sent:* Thursday, August 16, 2018 10:24 AM
>> *To:* EL Repo General Mailing List <elrepo at lists.elrepo.org>
>> *Subject:* Re: [elrepo] AMD EPYC Support
>>
>>
>>
>> As hardware support is accomplished primarily through the kernel, and
>> you've likely replaced it with an elrepo kernel the Redhat documentation
>> would not apply.  Additionally since Epyc is such a new processor you might
>> have better luck running the kernel-ml instead of the kernel-lt as large
>> driver backports are typically not accepted according to the linux-stable
>> rules.
>>
>>
>>
>> If I were you and intent on running an Epyc machine I'd move up to cent
>> 7 + the kernel-ml, or move to Ubuntu Bionic which has a 4.15 kernel where
>> they may have attempted to backport some of the Epyc enablement.
>>
>>
>>
>> Good luck,
>>
>> Dave.
>>
>>
>>
>> On Wed, Aug 15, 2018 at 7:15 PM <nick at ramnode.com> wrote:
>>
>> Hello,
>>
>>
>>
>> Has anyone tried running kernel-lt 4.4 (el6) on AMD EPYC processors?
>> RedHat documentation seems to indicate that RHEL/CentOS 6 is not supported
>> on the EPYC lineup, but AMD’s documentation is unclear.
>>
>>
>>
>> Regards,
>>
>> Nick
>>
>> _______________________________________________
>> elrepo mailing list
>> elrepo at lists.elrepo.org
>> http://lists.elrepo.org/mailman/listinfo/elrepo
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.elrepo.org/pipermail/elrepo/attachments/20180821/b600d724/attachment-0001.html>