Pull Request For Linux To Remove Old Network Drivers, ISDN Subsystem Due To AI/LLM Noise
([Linux Networking] 6 Hours Ago
Removal)
- Reference: 0001629531
- News link: https://www.phoronix.com/news/Linux-7.1-PR-Remove-Old-Net
- Source link:
It was just days ago we reported on a proposal to [1]drop old network drivers due to AI-driven bug reports becoming a burden on upstream kernel developers. Last night that culminated with an initial pull request to clear out some old, unused networking drivers plus also clearing out the entire ISDN subsystem and more.
Due to AI / large language models (LLMs) generating more and more bug reports against all areas of the kernel -- including drivers that haven't been touched in years or even decades and likely not seeing any real use by users on vintage hardware with the leading-edge upstream kernels -- the networking subsystem is looking at gutting out that old code to eliminate the superfluous AI/LLM bug reports.
Jakub Kicinski sent out a pull request to begin clearing out those old drivers. So far Linux Torvalds has yet to honor that pull request but we'll see how it plays out over the next few days. The Linux 7.1 merge window is set to end on Sunday, 26 April. Kicinski argued in the [2]pull request :
"Hi Linus!
I've been trying to figure out how to survive the change in rate of submissions and we're doing 3 things in parallel:
- more/better LLM code review tools
- email bots guiding noobs and if needed throttling submissions
- deleting long-orphaned code (unless someone steps up immediately)
This PR is part of the 3rd point. I did move what could be built as a module to a GH repo, in case it helps someone. But I assume this is all basically unused code.
For the LLM reviewer part tweaking local prompts is showing its limits. They are missing 50%+ of bugs Sashiko finds. (Sashiko is using LLM APIs directly.) Sashiko/Gemini finds a lot of real issues but it reports side-issues and occasional false positives. Paolo says we spend ~40% of our time checking LLM outputs, sounds about right. I hacked up a Sashiko/Claude/semcode combo which is _much_ better in terms of false positives. But it also misses real things Sashiko/Gemini finds. Magic bullet I was hoping for yet to be found."
He further added in the pull request:
"Old code like amateur radio and NFC have long been a burden to core networking developers. syzbot loves to find bugs in BKL-era code, and noobs try to fix them.
If we want to have a fighting chance of surviving the LLM-pocalypse this code needs to find a dedicated owner or get deleted. We've talked about these deletions multiple times in the past and every time someone wanted the code to stay. It is never very clear to me how many of those people actually use the code vs are just nostalgic to see it go. Amateur radio did have occasional users (or so I think) but most users switched to user space implementations since its all super slow stuff. Nobody stepped up to maintain the kernel code.
We were lucky enough to find someone who wants to help with NFC so we're giving that a chance. Let's try to put the rest of this code behind us."
The pull request in its latest form gets rid of the 3com 3c509 / 3c515 / 3c574 / 3c589 drivers, the AMD Lance driver, AMD NMCLAN, SMSC SMC9194 / SMC91C92, Fujitsu FMVJ18X, and 8390 AX88190 / Ultra / WD80X3 drivers.
The pull additionally removes the CAIF network layer, the AX25 and amateur (Ham Radio) radio subsystem, unused ATM protocols and legacy ATM drivers, and also the ISDN subsystem and Bluetooth CMTP code. Yes, the entire ISDN subsystem with mISDN and CAPI are on the chopping block with it not being clear anyone is still using it with the latest upstream kernel. There is also now [3]this mod-orphan GitHub repository where the code is split off into an out-of-tree kernel module should anyone still be interested in using and maintaining it.
[4]
We'll see what Linus Torvalds decides to do for the Linux 7.1 merge window. This pull request removes 138.161 lines of old code from the networking area of the kernel.
[1] https://www.phoronix.com/news/Linux-Old-Network-AI
[2] https://lore.kernel.org/lkml/20260423235422.1541768-1-kuba@kernel.org/
[3] https://github.com/linux-netdev/mod-orphan
[4] https://www.phoronix.com/image-viewer.php?id=2026&image=old_net_remove_pr_lrg
Due to AI / large language models (LLMs) generating more and more bug reports against all areas of the kernel -- including drivers that haven't been touched in years or even decades and likely not seeing any real use by users on vintage hardware with the leading-edge upstream kernels -- the networking subsystem is looking at gutting out that old code to eliminate the superfluous AI/LLM bug reports.
Jakub Kicinski sent out a pull request to begin clearing out those old drivers. So far Linux Torvalds has yet to honor that pull request but we'll see how it plays out over the next few days. The Linux 7.1 merge window is set to end on Sunday, 26 April. Kicinski argued in the [2]pull request :
"Hi Linus!
I've been trying to figure out how to survive the change in rate of submissions and we're doing 3 things in parallel:
- more/better LLM code review tools
- email bots guiding noobs and if needed throttling submissions
- deleting long-orphaned code (unless someone steps up immediately)
This PR is part of the 3rd point. I did move what could be built as a module to a GH repo, in case it helps someone. But I assume this is all basically unused code.
For the LLM reviewer part tweaking local prompts is showing its limits. They are missing 50%+ of bugs Sashiko finds. (Sashiko is using LLM APIs directly.) Sashiko/Gemini finds a lot of real issues but it reports side-issues and occasional false positives. Paolo says we spend ~40% of our time checking LLM outputs, sounds about right. I hacked up a Sashiko/Claude/semcode combo which is _much_ better in terms of false positives. But it also misses real things Sashiko/Gemini finds. Magic bullet I was hoping for yet to be found."
He further added in the pull request:
"Old code like amateur radio and NFC have long been a burden to core networking developers. syzbot loves to find bugs in BKL-era code, and noobs try to fix them.
If we want to have a fighting chance of surviving the LLM-pocalypse this code needs to find a dedicated owner or get deleted. We've talked about these deletions multiple times in the past and every time someone wanted the code to stay. It is never very clear to me how many of those people actually use the code vs are just nostalgic to see it go. Amateur radio did have occasional users (or so I think) but most users switched to user space implementations since its all super slow stuff. Nobody stepped up to maintain the kernel code.
We were lucky enough to find someone who wants to help with NFC so we're giving that a chance. Let's try to put the rest of this code behind us."
The pull request in its latest form gets rid of the 3com 3c509 / 3c515 / 3c574 / 3c589 drivers, the AMD Lance driver, AMD NMCLAN, SMSC SMC9194 / SMC91C92, Fujitsu FMVJ18X, and 8390 AX88190 / Ultra / WD80X3 drivers.
The pull additionally removes the CAIF network layer, the AX25 and amateur (Ham Radio) radio subsystem, unused ATM protocols and legacy ATM drivers, and also the ISDN subsystem and Bluetooth CMTP code. Yes, the entire ISDN subsystem with mISDN and CAPI are on the chopping block with it not being clear anyone is still using it with the latest upstream kernel. There is also now [3]this mod-orphan GitHub repository where the code is split off into an out-of-tree kernel module should anyone still be interested in using and maintaining it.
[4]
We'll see what Linus Torvalds decides to do for the Linux 7.1 merge window. This pull request removes 138.161 lines of old code from the networking area of the kernel.
[1] https://www.phoronix.com/news/Linux-Old-Network-AI
[2] https://lore.kernel.org/lkml/20260423235422.1541768-1-kuba@kernel.org/
[3] https://github.com/linux-netdev/mod-orphan
[4] https://www.phoronix.com/image-viewer.php?id=2026&image=old_net_remove_pr_lrg