It’s also been reported that, if your motherboard supports it, setting the PCIe Link Speed to the promontory chipset to Gen2 from Gen3 will fix the issue; that may be a better fix that pci=nommconf but anything running through the chipset will be limited to at most 2 gigabytes/sec.
I have also been experimenting with a very slight FSB underclock (98mhz) on TR boards that support it, that may also “correct” the issue but I’m not real sure yet.
Following this with interest from home as I wait for my TR4 system to get built and arrive…
From reading about similar issues e.g. here and here I’m curious if you tried the “pcie_aspm=off” kernel setting mentioned at those links, and if so what the result was?
This stackexchange answer tries to ELI5 what pci=nommconf does - “disables Memory Mapped PCI Configuration Space” - which sounds a bit less performance-trashing than disabling memory mapped IO - is that correct, do you think?
Thanks for your output on this, btw, you seem to be the only easily findable resource on TR4/Linux M/B problems and solutions - good job!
this is what I meant, sorry, but it can still trash performance. Interesting. I wonder if disabling aspm in uefi will fix it. Perhaps the promontory chipset does not actually support pcie power saving modes? If so that’d be a trivial fix for board makers in uefi…
is there any verification that this is a serious error or needless system messages spam ??
i have applied the kernel option to ignore the aer errors and it seemed to work . but it still bugs me that these errors occurs, especially with the rate that they occur. my discussions with msi tech support about these were fruitless.
hopefully you will get further along with them .
just for refernce
the error occur on m y 1950x msi procarbon
kingston predator memory qvl
pny anarchy memory non qvl
team extreem 4133 non qvl
gtx1080
ati hd 6570
gt 610
gt 240
usb boot , sata boot. regular hdd boot .
any version of unix i have thrown at shows these errors (prox mox arch, solus, fedora, all they way upto kernel 4.13 ) … makes me wonder if they are occuring in windows and it is just hidden.
also makes me wonder if this is why windows 7 crashes on install at irql not equal to or less than regarding pci.sys.
i would love to have my mind put at ease as if this is a serious issue or not.
@wendell Have you tried finding speakers of the annual KVM Forum event ?
I mean looking up speakers and visit their sites/blogs and asking there ?
It could be that someone encountered, or maybe even resolved, the issues at hand. Or at least give more insight.
Just a quick update, I updated to 4.14.0, but the error still exists with ASPM enabled.
Hi, build a gentoo system on 1950X with Gigabyte Designare Ex MB, but the error is same as you mentioned.
After booting to system, I got the error flush my console. The only option for now is to remove the ASPM support from the kernel, but I know it is not the right solution. please do keep update the progress of the fix.
Thanks, adding pcie_aspm=off to grub mitigated the issue for me, no more kernel messages since.
So in theory if I understand it correctly, there is no negative impact to setting this, apart from “higher” power consumption at idle. So not a big deal on a GT 710.
Still hoping to see AMD and the Linux guys get together for fixing this flaw. Granted, the intersection of people buying Threadripper and people running Linux is quite small, but still, come on, implementing PCIe spec correctly cant be that hard, @AMD. smh