e1000e performance 30%+ slower on laptop Ubuntu vs Windows/Debian

[nate]

On 2021-07-23 20:34, Little Girl wrote:

> That's odd. Is that something that's commonly done? I've never heard
> of putting a driver into a kernel.

I think it's considered a best practice, so that the drivers are
developed inside the kernel along with the rest instead of being
a separate component. Note the driver was always in with
the kernel but the latest versions were often only available
through the Intel website, then at some points the people developing
the driver would have to submit patches to the kernel to merge the
changes. Their new approach gets things into the kernel more quickly,
that and in most cases downloading, compiling and installing a
3rd party module(and maintaining it across kernel upgrades) is beyond
the abilities of the user. So including it in the kernel is a good
thing generally.

The downside is that if you happen to need a newer driver it makes
it more difficult to get unless you are willing to adopt the kernel
it is distributed with.  I had a similar issue when I first got this
laptop. I used it for a year without ever using the Intel wireless
card on it. I went on a big 3 month trip and realized wireless was
not working. I went to download the driver to see that it required
a new major kernel revision to use(would not compile on the version
I had).  Fortunately there weren't too many issues upgrading to the
new major version but it was something I was not expecting having
used Linux for 25 years now.

Since my earliest days of Linux have always really disliked the
lack of a compatible binary driver interface to the kernel which
would make drivers/modules much more compatible across kernel
versions. I understand the reasons why the kernel team doesn't want
to do this and have long given up hope that it will ever happen,
but it would make user's life a lot easier (also help Android out
a lot as well as that is a huge sticking point in distributing
updates to Android devices). But it is what it is.

It's possible Intel will continue releasing standalone drivers on
their website going forward as new versions come out, though am not
optimistic they will.

> That's good, though, right?

Yes it just means the latest driver is in the kernel already so
that is generally good. Unless you have an edge case where the
newer driver breaks something but that is probably quite rare.

> I'm sorry. I know how frustrating that can be. Hopefully someone else
> will chime in here and offer something better.

No problem, not optimistic to find a solution at this point, I have
another somewhat unique case of a massive memory leak in Ubuntu 20
kernels vs Ubuntu 16 kernels(have 6 servers with 3 kernels and the
same workload running the trend is super clear) which I reported on
this list earlier in the year with no replies. Went back and am
gathering more info there, but not optimistic to find a real
solution(workaround is to reboot as the kernel is not freeing
the memory) in that case either.

thanks

nate