Apollo lake support for Centos 6.8

Issues related to hardware problems
matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/11 07:46:04

Hi, I just purchased Asrock's J3455-itx and was wondering how well it is supported under centos 6.8 x64.
Specifically I am worried about usb support as launching flash drive with install hangs due to usb handoff issues.
Also of importance support for storage controller(s) found on the board.

I already have a working system running on q1900-itx with 4 drives (one drive for boot and 3 more in linux software raid 5).
Would I be fine in simply changing the motherboard?

User avatar
TrevorH
Forum Moderator
Posts: 26076
Joined: 2009/09/24 10:40:56
Location: Brighton, UK

Re: Apollo lake support for Centos 6.8

Post by TrevorH » 2016/11/11 08:01:00

Any reason why not CentOS 7 given that it's newer and will have better hardware support? Also, bear in mind that CentOS 6 will go into production phase 3 next year and thus will start receiving only critical security updates for the final 3 years of its life.
CentOS 5 died in March 2017 - migrate NOW!
Full time Geek, part time moderator. Use the FAQ Luke

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/11 08:43:16

Mainly because its already setup and works fine :)
Reinstalling everything from scratch just doesn't sound like a good investment of time.
Other then that (sorry if that sounds fanboyish) I care deeply about initd and despise systemd.

EDIT: just wanted to add what I am doing with my current box (and intend to continue to do with the new motherboard).
1. Fileserver (Samba)
2. Smart home controller (using Pimatic)
3. VM host (pfsense fw vm and elastix vm for pbx) based on KVM
4. Torrent box.

As long as usb and storage works (and VM features of Apollo lake cpu are properly recognized for VM operation), I am good.
I don't care for graphics (I run my box in console mode), or audio (I have it disabled in bios).

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/12 16:52:25

Ok, I will try to replace the motherboard tomorrow and see how it goes.
Worst case, I will swap the q1900 back if it fails to boot or doesn't recognize something important.

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/13 11:33:19

Reporting back for the benefit of other users.
The server appears to be fine booted with the new motherboard with J3455-itx
The only issue is the message at boot saying that the cpu is unsupported.
Supressed the message by removing quiet from grub parameters.
Other then that, everything appears to be in order.
VMs start up and work fine.
All usb devices work.
Network card onboard works.
All hard drives recognized and linux raid5 works fine.

User avatar
TrevorH
Forum Moderator
Posts: 26076
Joined: 2009/09/24 10:40:56
Location: Brighton, UK

Re: Apollo lake support for Centos 6.8

Post by TrevorH » 2016/11/13 12:27:36

The only issue is the message at boot saying that the cpu is unsupported.
Supressed the message by removing quiet from grub parameters.
I'm pretty sure that removing quiet does not suppress that message - I suspect it just hides it by flooding you with more info than you can handle! Are you running the latest 2.6.32-642.6.2.el6 kernel? You need to be to be protected from CVE-2016-5195 which is a nasty local root exploit. I believe you can get shot of the unsupported hardware message by passing "unsupported_hardware" on the kernel command line if the latest kernel does not remove that message.
CentOS 5 died in March 2017 - migrate NOW!
Full time Geek, part time moderator. Use the FAQ Luke

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/13 16:05:06

You are right.
After having posted that I just thought the same.
The message is probably just hidden.
The thing is with quiet set to ON, the message stays on for the whole stage of kernel loading which is pretty long.
And this is unsettling.

I will try your suggestion about ""unsupported_hardware"" and write back if it helps.

To answer your question, yes I am running the latest kernel from Oct 26.

Here are some more impressions about this new board.
There were a couple hrtimer took x ns (they happened before, but I think never more then one in a very long time).
I will keep watching and see if they return.
I also had my usb network adapter (gigabit realtek 8169) getting usb disconnect once (again this happened before, was rather rare will keep watching).

I am not entirely sure whether the system is faster (boot takes in fact longer or so it feels).
I am booting off Samsung EVO 850 120 gig drive.
Webmin appears to be more responsive, but system load averages appear to be higher (with the pfsense vm taking about 100% cpu which is one core while it has been assigned two cores).
Then again system load averages could have risen due to yet another change in how kernel computes them (a few kernel updates ago I had system load averages ~ 200% sometimes going as high as 400%
By watching cpu clock rate ,it appears the system stays on lower clock rate then J1900 used to.
J1900 was usually at 1.3Ghz, this spends most of its time at 800Mhz).
AES-NI appears to be supported as evidenced by openssl benchmark.
Cpu temps appear to be roughly the same as before (this is a small Ml06B case and there is only one fan operating at 800 RPM).
The hw sensor chip is recognized as nct6793.
There are several bad temperature readings:
AUXTIN1: +110.0°C sensor = thermistor
AUXTIN2: +109.0°C sensor = thermistor
AUXTIN3: +110.0°C
SYSTIN: +117.0°C (high = +0.0°C, hyst = +0.0°C) sensor = thermistor

And some others which show 0C (possibly disconnected?).

The cpu capabilities appear to be correctly listed except VT-d which is supported according to Intel, but then again there is nowhere in bios to enabled it (after all its a consumer board) so possibly the support is turned off in bios.

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/14 16:55:52

OK after some more testing, I am pretty disappointed with the new cpu and motherboard.
It appears the performance of this new cpu is lower then J1900 under Centos 6.8.
Load averages and cpu usage of kvm processes are higher.
There is also an annoying issue with onboard network working with ssh and (possibly some other stuff).
This exhibits itself in very choppy input while working with putty.
It is also possible (though I haven't confirmed it yet) that samba is also affected by this.
Strangely enough copying files does appear to be faster.
I was able to reach over 100MB/s copying a large file to an ssd over the gigabit connection.
From what I remember, it used to be ~60-70 MB/s before.
Also VM running Elastix (based on centos 5.9) is eating significantly more cpu.
At idle with guest showing all zeroes as load average, the kvm process consumes over 100% cpu (more then one full core).
I also experienced (for the first time on with this installation) a CMCI storm

Nov 14 17:51:59 linuxserver kernel: CMCI storm detected: switching to poll mode
Nov 14 17:52:04 linuxserver kernel: br0: port 5(tap2) entering disabled state
Nov 14 17:52:04 linuxserver kernel: device tap2 left promiscuous mode
Nov 14 17:52:04 linuxserver kernel: br0: port 5(tap2) entering disabled state
Nov 14 17:53:26 linuxserver kernel: device tap2 entered promiscuous mode
Nov 14 17:53:26 linuxserver kernel: br0: port 5(tap2) entering forwarding state
Nov 14 17:53:41 linuxserver kernel: br0: port 5(tap2) entering forwarding state
Nov 14 17:56:59 linuxserver kernel: CMCI storm subsided: switching to interrupt mode

Apparently this is an issue with the network card?

matrix2020
Posts: 18
Joined: 2016/11/11 07:42:01

Re: Apollo lake support for Centos 6.8

Post by matrix2020 » 2016/11/15 07:54:04

So my saga continues :)
After having experienced another cmci storm during the night, I decided that desperate times call for desperate measures and disabled mce in kernel.
After reboot, I must say the system is significantly more responsive, load average has gone down and so did cpu usage for the VMs
The problem with ssh input lag also appears to be gone in the host (still happens with VMs).
I think I am suffering from the same problem as described here:
https://trick77.com/qemu-on-haswell-cau ... ce-events/
I also happen to run two VMs with qemu-kvm which are 32bit both.
I suppose the issues Intel had with Haswell could have been copied to the new Atom arch as well?
Anyway what are the dangers of disabling mce at boot?

Observation2017
Posts: 6
Joined: 2017/03/14 12:36:15

Re: Apollo lake support for Centos 6.8

Post by Observation2017 » 2017/03/14 14:16:23

What's the support status on this CPU. Been thinking about this CPU myself for a pfsense low power implementation.

Had some speculation on the reported poor performance which I wonder if anyone with real expertise can verify as yeah or nay or possibly.

That missing Vt-d might be critical since any I/O hardware gets replaced by inefficent emulation of the nearest generic device. Especially NICs with usable hardware acceleration.

Also wondering if the hardware AES instructions are active for this CPU in real mode or VMs. Unsupported status probably means "no" for both I would think. So VPN encryption maybe sucking up a huge emulation/unaccelerated mode charge.

BIOS settings? Not a *NIX family expert but I thought Linux and most the modern kernels throw out BIOS and load their own version of BIOS setup early in boot up process. Still "unsuppoerted" CPU is probably not correctly configured by default. But might be too hard to change a few critical settings yourself if someone expert has some advice.

Post Reply