CentOS 7 Install fail - PCIe Bus Error

[donatell018]

I'm having trouble trying to install CentOS 7 from USB (and same problem with Red Hat 7).
The installers have worked on different hardware but this setup.

The install gets into an infinite loop with these errors (from what I could make out from my phones camera):

Code: Select all

PCIe bus error corrected
AER error Device Id=00ea 
Device=[8086:a11a]

I've tried various things, but could use some help.

This system has Intel Core i7 6820EQ (Skylake) CPU with 16Gb DDR4 RAM.
This link shows the basic system info (obtained from a working Red Hat 6.9 install on the hardware).
http://pastebin.centos.org/575231/

It appears there is a compatibility issue between kernel 3.10 and the Intel Skylake chipset.
I found this boot parameter for a Centos/Redhat 6.* work around, but it doesn't work for v7.

Code: Select all

pci=nommconf

The problem centers around the PCI architecture for the Intel Skylake generation of chips.
The above kernel option disables Memory-Mapped PCI Configuration Space. All devices are
accessed in the original way, which works around this problem for OS v6, but apparently not v7.

Can anyone please shed any light on how I can resolve this problem for CentOS 7?
Thanks in advance!
Ref:
https://unix.stackexchange.com/question ... er-bad-tlp
https://siweston.wordpress.com/2017/08/ ... 260-g2-pc/

[TrevorH]

Did you try pci=nomsi and/or pci=noaer ?

https://access.redhat.com/discussions/3360131

[donatell018]

Hi Trevor.
Other options I've tried as follows:

Code: Select all

pcie_aspm=off

This generates the same PCIe error as above.

In repeated attempts, I've tried each of the following options one after the other but get some sort of device controller error followed by system halt.

Code: Select all

pci=nommconf
pci=noaer
pci=nomsi

Also I can't access the boot logs:
Running the OS 7.4 install, I tried to access the console using Ctl-Alt-F1 and Ctl-Alt-F2,
but there's no response. All I'm getting is the infinity of errors.

Using the debug boot parameter also doesn't get around this.
It's like the failure happens too early in the install process.

Is there some way around to get at the log files?
Any quick solutions?
Thanks all.

[TrevorH]

The only mentions of that error I found where it had been solved were on Ubuntu and were solved by using nomsi. If I read /usr/share/doc/kernel-doc-3.10.0/Documentation/kernel-parameters.txt (part of kernel-doc) then I see you can have many many different pci= parameters so perhaps worth trying some of the others too. There's pci=bios, conf1, conf2, nommconf, nomsi, noaer, noearly and about 2 more pages of things to try. No idea if any of the ones you haven't already tried will help.

[donatell018]

Hi Trevor,
Thanks for the feedback.

Time constraints at the moment mean I'm going to attempt a *cough* Ubuntu install first.
This is simply because real Intel Skylake support seems to begin with linux kernel 4.3, not currently available
in the CentOS/Redhat ecosystem.

No doubt I'll come back to this at a later point...
Thanks

[donatell018]

To recap, CentOS/Redhat 6.9 installed OK, all be it with a work around.
Considering this, it's strange that CentOS/Redhat 7 installer fails completely.

I still might have to revisit this for a CentOS/Redhat 7 install solution (trying other boot params), but here's my current solution...

I’ve installed Ubuntu 16.04 LTS on this hardware (although it was slightly problematic).

The install completed successfully, but on reboot I had the same fault behaviour as was seen on Centos/Redhat 7*.
The work around was to use the “pci=noaer” boot parameter. Now boots successfully.

Ref:
https://askubuntu.com/questions/771899/ ... ect=1&lq=1