Intel's I219-LM Gigabit Ethernet Adapter Ran at 60% Maximum Speed on Linux Since 2020

Phoronix reports: If you rely on an Intel I219-LM Gigabit Ethernet adapter, you will want to look forward to upgrading your Linux kernel build soon... A fix was committed Thursday after Intel engineers discovered this particular Ethernet chipset had only been running at around 60% of its maximum speed due to a regression introduced back in 2020...

Since the release of Linux 5.8 in mid-2020, this Ethernet adapter had been running at around 60% of its advertised potential due to an e1000e driver regression.

Well

[jonathantn]

MOAR speed!

Re:

[Z00L00K]

Makes me wonder if they really did test the drivers before they released them.

Re:

[arglebargle_xiv]

Makes me wonder why no-one noticed it until now. That is, did anyone actually need the extra speed?

Re: Well

[partymola]

I don't think you understood it right. TSO should only be used on 10/100 and wasn't being properly disabled in other cases. The issue was affecting high speed links.

Re:

[martinfb]

Makes me wonder HOW they tested the drivers before they released them.

Re:

[arglebargle_xiv]

Their fault for still running TSO when everything's moved on to TSO/E or CMS.or z/OS Unix.

Funny thing to be running on a network adapter though...

The description reads a little different

[Arnonyrnous Covvard]

It seems there is a hardware bug in that chip which causes issues if a particular feature isn't disabled. The "regression" was a patch that was supposed to turn off that broken feature, except it didn't really turn it off. That sounds more like a failed workaround than a regression. Also, what are Intel engineers doing if it takes this long to notice that their hardware is running at 60%?

Re: The description reads a little different

[Z00L00K]

Obviously they didn't test it enough.

Re: The description reads a little different

[rhardy]

It has been challenging to get Intel to take Linux support seriously.
I had a case open for driver issues with this chip i.e. including huge performance differences between their Linux and Windows drivers.
They kept not doing anything about it so I gave up and changed hardware to something that worked. Glad they finally fixed that.

Re:

[CompuGeek]

I concur with @rhardy. And it's not just for the I219-LM. The situation with the igc driver for the I-225V chipset is a real debacle that makes it unusable for most users. Windows as well as Linux users have problems. People expect better from Intel since it's a top name in the industry.

Re:

[Firethorn]

Yeah, I know that for a while when I was shopping around for motherboard/laptop type systems, that "Intel" network connection was a positive thing for me. Might mean my picking one system over another if they're of similar price.

Yes, I would check reviews semi-regularly, but generally the intel stuff was noted for being stable and dependable. You can't really increase speed over the standard, of course. But an assurance of "it just works" for something peripheral is something you want.

Re:

[NoWayNoShapeNoForm]

The reviewers aren't really doing their job either. They're happy to drool over all the latest benchmarks for CPU and GPU performance, but everything else in a system/mobo just gets a checkbox.

Most Linux code reviewers, the people that actually sign off the code before it's sent to Linus as a "git pull", compile the proposed changes.

If the compile works, then they look for unexpected interactions with other parts of Linux, and if none are found THEY MIGHT DO ACTUAL TESTS

Much of the Linux codebase lacks rigorous and documented performance & regression testing. Yes, more and more programmers are including module tests and better code building checks in their code packages, but I have rarely se

Re:

[Anonymous Coward]

Also, what are Intel engineers doing if it takes this long to notice

Running Windows?

Re: The description reads a little different

[jrronimo]

Not unless the router has this Intel chip in it. Let's use an example: A Linux box with this Intel I219-LM in it and a Mac mini, both hardwired to a router rated for 1,000 Mbps speeds (gigabit). Both devices are rated for 1,000 Mbps operation. The Mac mini would be able to push data to the router at 1,000 Mbps no problem. The Linux box with the faulty driver would only see speeds in the range of 600 Mbps because of this bug. Of course if this is your house and your ISP only offers you 100 Mbps operation y

Re:

[omnichad]

Except the bug only affects 10/100 connections, so you would see gigabit.

Re:The description reads a little different

[omnichad]

This bug only hits in 10/100 mode. If you are in Gigabit mode, you are unaffected. I looked at the commit to try to figure out if it affects I219-V (still don't know).

The reason the feature wasn't disabled? For some reason, they set the flags in the watchdog code instead of the hardware probe code.

Source: https://git.kernel.org/pub/scm... [kernel.org]

Re:

[F.Ultra]

According to a post down below this only affected the 10/100 mode and not the 1000 mode so that is probably why few people noticed and no one filed a bug report until now.

Re:The description reads a little different

[thegarbz]

That sounds more like a failed workaround than a regression.

It doesn't matter what something is. Fix, workaround, legit feature ... if it worked on one release and not the next it is a regression.

on which boards and cards?

[Anonymouse Cowtard]

File under They always fix it eventually.

In other words ...

[fahrbot-bot]

Intel's I219-LM Gigabit Ethernet Adapter Ran at 60% Maximum Speed

They bit off more than they could through.

Are VMs effected?

[pcjunky]

Does this apply to the e1000 drivers in KVM?

Windows

[echostorm]

That chip seems to have some sort of issue on Windows 11 also- it cant seem to handle more than some arbitrary number of simultaneous connections but I haven't had time to look into it, it at least loads everything reasonably quickly on KDE Neon. Just disabled the chip about a year ago and have been using wifi.