Network card stops working after a while. ELRepo?

[AlanBartlett]

I add this comment for completeness and posterity.

[quote]
cd builddir/build/SPECS/
edit file e1000e-kmod.spec and change kversion line to:
%{!?kversion: %define kversion 2.6.32-042stab031.1}
rpmbuild -ba e1000e-kmod.spec
[/quote]
There was no need to edit the spec file. The kernel version can be supplied on the rpmbuild command line.

[code]
cd builddir/build/SPECS/
[b]rpmbuild -ba --define 'kversion 2.6.32-042stab031.1' e1000e-kmod.spec[/b]
[/code]

[universe6]

Thanks for the trick. That basically means we can reduce the whole procedure to:

[code]
mock -r sl6-x86_64 --init
mock -r sl6-x86_64 --define='kversion `uname -r`' --installdeps -no-cleanup-after --rebuild e1000e-kmod-1.4.4-1.el6.elrepo.src.rpm[/code]

Of course still with the same problem

[code]rpm -ivh --test kmod-e1000e-1.4.4-1.el6.x86_64.rpm
error: Failed dependencies:
kernel(alloc_etherdev_mq) = 0x4070413e is needed by kmod-e1000e-1.4.4-1.el6.x86_64
...[/code]

[AlanBartlett]

[quote]
Of course still with the same problem

[code]rpm -ivh --test kmod-e1000e-1.4.4-1.el6.x86_64.rpm
error: Failed dependencies:
kernel(alloc_etherdev_mq) = 0x4070413e is needed by kmod-e1000e-1.4.4-1.el6.x86_64
...[/code]
[/quote]
There, I am unable to assist you other than to say that you will need to look at the configuration of your kernel and check what (symbols) it provides.

[universe6]

I understand, I just added this comment for posterity. So nobody thinks that would be the complete solution.

PS
Just to clarify. Downloading and installing the latest driver from e1000.sf.net works fine. Just creating an RPM for it is a bit of an issue, because I don't understand the kmodtool well, but I am getting there. Interesting topic!

[universe6]

One more update (and solution):

I built and installed all kind of drivers. Nothing helped. Then I installed plain vanilla CentOS and ElRepo drivers etc. Still the same problem. So it didn't seem to be driver related (or a bug in the non-standard CentOS kernel) ...

In the end I found the solution:

[code]echo performance > /sys/module/pcie_aspm/parameters/policy[/code]

Unfortunately this setting doesn't survive boots.

Alan, what would be the best way to make this setting permanent?

[AlanBartlett]

A quick, "off the top of my head" response would be to suggest putting that "echo" line into your [b]/etc/rc.d/rc.local[/b] file -- so that it is executed once, at the end, of every system boot. Would that be a satisfactory solution?

[dag]

You can provide module parameters for compiled-in modules on the kernel command line by doing this:

pcie_aspm.policy=performance

So you will have to edit your /etc/grub.conf and add that to (every) kernel command line.

[universe6]

Dag, I tested your suggestion and it worked. Thanks a lot.

Then TUV documentation doesn't seem to be correct, right?

See also the following topic (and link to TUV documentation):

https://www.centos.org/modules/newbb/viewtopic.php?topic_id=32863&forum=55

PS
Where did you learn about this setting?

[Brianmahler10]

I too have the same problem...My superMicro board with Intel 82574 nics drops connectivity after a while (30 min to several hours).
I noticed when it drops the nic changes to 1Gig (from 100M ) that I configured manually. Problem is that the switch that this server is connected to does not support Gig speeds (only supports Fast Ethernet 100Mb). Autoneg is not working, so I manually set the speed and duplex, and it works great ...for a while, then drops.

But I don't understand the solution listed in post #14;

echo performance > /sys/module/pcie_aspm/parameters/policy

How does one set this?

Thanks.
Brian