* [ANN] netdev development stats for 7.4
@ 2026-10-22 20:57 Jakub Kicinski
0 siblings, 0 replies; only message in thread
From: Jakub Kicinski @ 2026-10-22 20:57 UTC (permalink / raw)
To: netdev
Preface
-------
As is tradition here are the development statistics based on mailing
list traffic on netdev@vger.
These stats are somewhat like LWN stats: https://lwn.net/Articles/1088776/
but more focused on mailing list participation. In particular "review
score" tries to capture the balance between reviewing other people's
code vs posting patches.
Previous 3 reports:
- for 7.1: https://lore.kernel.org/20260414182653.40d84ccc@kernel.org
- for 7.2: https://lore.kernel.org/20260617115319.43a5942d@kernel.org
- for 7.3: https://lore.kernel.org/20260819114659.271292db@kernel.org
Highlights
==========
Not sure how many people read these, but I expect more than any other
email on the list, so let me start with some highlights / announcements.
SUIE
----
SUIE, which maintainers use to order review queue, now supports
filtering by MAINTAINERS entries. If you find it difficult to sift
thru the list of patches please try it. Click the funnel icon, select
the subsystems, and enable the filter.
https://netdev.bots.linux.dev/suie.html?delegate=netdev
Clashiko
--------
Major change in this cycle is that we started emailing out the AI
reviews automatically. This obviously has negative implications
(extra work for submitters, risk of going down the wrong path,
higher ML volume as people respin for AI feedback).
With the growth of patch volume it was unavoidable. Not sure how
to explain this but I was basically 30min to an hour short of time
in a 24h day, and it was getting worse and worse. Not having to handle
the AI reports saves me at least an hour a day. We'll see what other
compromises we have to make if the volume continues to grow :(
From the technical side the AI reviews got reshuffled to support
multiple models. We test with GPT and Opus, and leech off main
sashiko to get Gemini reviews. This brought the false positive rate
(as measured by patches applied with Medium+ severity feedback) is
around 20%. The false negative (patches rejected without feedback)
is around 10%.
At least some of the 20% false positives are real problems, we don't
design locking and reset handling properly. Pure hallucinations are
infrequent.
Stats
=====
General
-------
The previous (7.3) release cycle spanned summer vacation season and
netdev.conf, so let's compare against last two releases.
We merged 43% more patches than in 7.3, 9% more than in 7.2.
We saw 48% more emails than during 7.3, 37% more than in 7.2.
There were 12% more participants on the list, 17% compared to 7.2.
Lines goes up.
Clashiko and series info bot were directly responsible for 13% of
the traffic, so traffic not counting the bots went up 35% (vs 7.3)
and 24% (vs 7.2). Hard to tell how much of the increase was indirectly
attributable to bots as people repost to fix AI reviews.
Almost half of the patches are fixing bugs. Of the patches posted
it's 39% and of the patches merged 46%. I think this reflects the
developer experience of development of new features being harder
and slower these days.
Number of necessary revisions went up for a single patch: +5.4%
and patch set: +1.6%. Cross company review recovered 4% to 58%
of commits carrying a review tag.
Developer rankings
------------------
Top reviewers (cs): Top reviewers (msg):
1 ( ) [49] Jakub Kicinski 1 ( ) [91] Jakub Kicinski
2 ( ) [28] Simon Horman 2 ( ) [70] Andrew Lunn
3 ( ) [25] Andrew Lunn 3 ( ) [42] Simon Horman
4 ( ) [19] Paolo Abeni 4 ( ) [27] Eric Dumazet
5 ( ) [17] Eric Dumazet 5 ( -1) [26] Paolo Abeni
6 ( -1) [11] Ido Schimmel 6 ( -1) [19] Ido Schimmel
7 ( -1) [ 7] Maxime Chevallier 7 ( +2) [17] Willem de Bruijn
8 ( +1) [ 7] Aleksandr Loktionov 8 ( -2) [15] Maxime Chevallier
9 (+34) [ 6] Hangbin Liu 9 (+18) [13] Nicolai Buchwitz
10 ( +2) [ 6] Willem de Bruijn 10 ( +1) [12] Aleksandr Loktionov
11 ( -4) [ 5] Kuniyuki Iwashima 11 ( -4) [11] Kuniyuki Iwashima
12 (+37) [ 5] Nicolai Buchwitz 12 (+44) [10] Hangbin Liu
13 ( ) [ 5] Krzysztof Kozlowski 13 ( +1) [ 9] Stanislav Fomichev
14 (+23) [ 4] David Laight 14 ( +6) [ 8] Lorenzo Bianconi
15 (***) [ 4] Daniel Machon 15 ( -7) [ 8] Krzysztof Kozlowski
Ido continues to handle the influx of fixes for L3,
Maxime is now officially a maintainer of stmmac (and helps with phylink).
Willem concentrates on virtio / tap / af_packet etc.
Hangbin reviews selftests, bonding, seg6, netlink etc.
Nicolai reviews embedded drivers (stmmac, macb, bcm* etc)
Thanks as always to everyone reviewing!
Top authors (cs): Top authors (msg):
1 (***) [9] Eric Dumazet 1 (***) [30] Eric Dumazet
2 (***) [5] Hui Peng 2 ( +6) [22] Kuniyuki Iwashima
3 (+22) [5] Wentao Liang 3 (+13) [20] Birger Koblitz
4 ( -1) [5] Jakub Kicinski 4 (***) [19] Allison Henderson
5 (+23) [4] Chengfeng Ye 5 (***) [19] Aleksei Sviridkin
6 (+24) [4] Jamal Hadi Salim 6 ( -4) [19] Tariq Toukan
7 ( -2) [4] Tariq Toukan 7 ( -4) [18] Pablo Neira Ayuso
8 ( -2) [4] Lorenzo Bianconi 8 (***) [18] Marc Kleine-Budde
9 ( +4) [4] Ratheesh Kannoth 9 (+12) [16] Jakub Kicinski
10 ( -6) [3] Ren Wei 10 (+12) [15] James Hilliard
11 (+12) [3] Zihan Xi 11 (-10) [15] Tony Nguyen
12 (***) [3] Aamir Ahmed 12 (+23) [15] Maxime Chevallier
13 (+23) [3] Myeonghun Pak 13 ( +1) [14] Lorenzo Bianconi
14 ( +7) [3] Rosen Penev 14 (-10) [14] Christian Marangi (Ansuel)
15 (***) [3] Daehyeon Ko 15 (***) [14] Zihan Xi
Folks who read the mailing list will note that roughly half
of the top authors by change set are LLM-aided newcomers.
The different between count of change set (left) and messages (right)
is striking, almost everyone on the right side is a long time
contributor. This suggests that the noobs send a ton of patches
and never follow up with new versions.
Top scores (positive): Top scores (negative):
1 ( ) [691] Jakub Kicinski 1 (+11) [77] Birger Koblitz
2 ( +1) [454] Andrew Lunn 2 (***) [74] Aleksei Sviridkin
3 ( -1) [382] Simon Horman 3 (+37) [64] Marc Kleine-Budde
4 ( ) [254] Paolo Abeni 4 (+12) [57] James Hilliard
5 ( ) [161] Ido Schimmel 5 (***) [57] Zihan Xi
6 (***) [124] Eric Dumazet 6 ( -2) [56] Tariq Toukan
7 ( +4) [ 90] Willem de Bruijn 7 (***) [54] Kyle Switch
8 ( ) [ 77] Aleksandr Loktionov 8 ( -7) [52] Tony Nguyen
9 ( ) [ 64] Krzysztof Kozlowski 9 ( -7) [52] Christian Marangi (Ansuel)
10 ( -4) [ 63] Maxime Chevallier 10 (***) [52] Allison Henderson
11 (+18) [ 52] David Laight 11 (+37) [49] David Yang
12 (+21) [ 51] Nicolai Buchwitz 12 (***) [47] Coia Prant
13 (+36) [ 50] Hangbin Liu 13 (***) [45] Ali Rouhi
14 ( +6) [ 46] Stanislav Fomichev 14 (***) [34] Christian Brauner
15 (+11) [ 40] Stefano Garzarella 15 ( -2) [33] Mingming Cao
Company rankings
----------------
Top reviewers (cs): Top reviewers (msg):
1 ( +1) [60] RedHat 1 ( ) [120] Meta
2 ( -1) [58] Meta 2 ( ) [106] RedHat
3 ( ) [25] Andrew Lunn 3 ( ) [ 70] Andrew Lunn
4 ( +2) [21] nVidia 4 ( +2) [ 40] nVidia
5 (***) [17] Eric Dumazet 5 ( ) [ 38] Google
6 ( -1) [15] Google 6 ( -2) [ 28] Intel
7 ( -3) [13] Intel 7 (***) [ 27] Eric Dumazet
8 ( -1) [11] Bootlin 8 ( -1) [ 22] Bootlin
9 (+21) [ 9] Kylin Software 9 (+26) [ 15] Kylin Software
10 ( -2) [ 7] Linaro 10 ( -2) [ 14] Linaro
Top authors (cs): Top authors (msg):
1 ( ) [13] Google 1 ( ) [68] Google
2 ( ) [11] Meta 2 ( ) [68] Meta
3 (***) [ 9] Eric Dumazet 3 ( ) [54] Intel
4 ( +1) [ 7] nVidia 4 ( ) [42] nVidia
5 ( -2) [ 7] Nebula Sec 5 ( +1) [38] Qualcomm
6 ( +5) [ 6] Qualcomm 6 ( -1) [32] RedHat
7 ( ) [ 6] RedHat 7 (***) [30] Eric Dumazet
8 ( +4) [ 6] Chinese Academy of Sciences 8 ( +1) [28] Microsoft
9 (+14) [ 6] Mojatatu 9 (+17) [28] Oracle
10 ( -6) [ 6] Kylin Software 10 ( +1) [26] Nebula Sec
Top scores (positive): Top scores (negative):
1 ( +1) [761] RedHat 1 ( ) [104] Nebula Sec
2 ( -1) [654] Meta 2 (+10) [ 77] Birger Koblitz
3 ( ) [454] Andrew Lunn 3 (***) [ 75] Pengutronix
4 ( +1) [162] nVidia 4 (***) [ 74] Ænix
5 (***) [124] Eric Dumazet 5 ( +4) [ 69] Microsoft
6 ( +4) [ 81] Bootlin 6 ( +5) [ 58] Qualcomm
7 (***) [ 56] Kylin Software 7 ( +6) [ 57] James Hilliard
8 ( +4) [ 52] Aculab 8 (+37) [ 54] Oracle
9 ( +5) [ 51] KUNBUS 9 (***) [ 54] Motorcomm
10 ( +1) [ 41] ARM 10 ( -6) [ 52] Christian Marangi (Ansuel)
Major changes in the company rankings is that Eric Dumazet
no longer works at Google, dropping Google to a "negative score".
Nebula Sec is one of the LLM fix companies, they used to do pretty
well but recently the patches are lower quality, and don't end up
getting merged.
Kylin OTOH moved to the positive side by cutting down on the bad
submissions and thanks to Hangbin's review effort.
Posting efficiency
------------------
Here are the stats for people who are best / worst at getting
code into the tree. I think at some point we may need to include
these in the "SUIE score".
Lowest postings/commit (min 10): Highest postings/commit (min 10):
1 [ 14@ 1.00] Ruizhe Zhou 1 [15@ 8.87] Ænix
2 [ 39@ 1.23] batman-adv 2 [26@ 8.54] Oracle
3 [ 34@ 1.26] TI 3 [22@ 8.18] Pengutronix
4 [ 31@ 1.71] NGI0 Core 4 [53@ 7.98] Intel
5 [ 84@ 1.76] Pablo Neira Ayuso 5 [32@ 7.06] Microsoft
6 [ 10@ 2.00] Andrew Halaney 6 [21@ 6.48] Microchip
7 [ 15@ 2.00] Karl Mehltretter 7 [48@ 6.31] Qualcomm
8 [ 69@ 2.10] Mojatatu 8 [20@ 6.25] Linaro
9 [ 17@ 2.12] OpenVPN 9 [11@ 5.82] Renesas
10 [112@ 2.14] Eric Dumazet 10 [19@ 5.74] Marvell
11 [ 21@ 2.19] Hemanth Selam 11 [31@ 5.68] Broadcom
12 [ 55@ 2.25] Anthropic 12 [11@ 5.55] Daniel Golle
13 [ 19@ 2.26] Isovalent 13 [17@ 5.35] AMD
14 [212@ 2.44] Meta 14 [52@ 5.02] RedHat
15 [ 23@ 2.48] Secunet 15 [26@ 4.88] Nebula Sec
^ ^
number of postings per
commits 1 commit
Nothing merged, people (min 30): Nothing merged, companies (min 30):
1 (***) [ 101] Kyle Switch 1 (***) [ 101] Motorcomm
2 (***) [ 61] Chih Kai Hsu 2 (***) [ 61] RealTek
3 (***) [ 54] Juha-Matti Tilli 3 (***) [ 54] Juha-Matti Tilli
4 (***) [ 42] Julius Bairaktaris 4 (***) [ 42] Julius Bairaktaris
5 (***) [ 41] Jacob Keller 5 (***) [ 35] Chelsio
6 (***) [ 41] Mohd Ayaan Anwar 6 (***) [ 32] Caleb James DeLisle
7 (***) [ 38] Wei Hu 7 (***) [ 32] Junhui Liu
8 (***) [ 36] Jens ES. Ostergaard
9 (***) [ 36] Pooventhiran G
10 (***) [ 35] Potnuri Bharat Teja
11 (***) [ 33] Sergey Temerkhanov
12 (***) [ 32] Caleb James DeLisle
13 (***) [ 32] Junhui Liu
14 (***) [ 30] Aleksandr Loktionov
Another potential metric to add to the mix is what percentage of
postings gets an AI review:
Replies from netdev-bot+sashiko@kernel.org:
Per posting, people (min 15): Per posting, companies (min 30):
1 [94% of 18] Arthur Crepin Leblond 1 [81% of 64] Nebula Matrix
2 [88% of 17] Jacky Chou 2 [67% of 46] Sagi Maimon
3 [85% of 26] Ali Firas 3 [65% of 48] Realsil
4 [81% of 64] Illusion Wang 4 [62% of 61] RealTek
5 [80% of 15] Parthiban Veerasooran 5 [62% of 53] Linutronix
6 [78% of 27] Ivan Vecera 6 [61% of 33] Aamir Ahmed
7 [68% of 19] Guangshuo Li 7 [58% of 125] Linaro
8 [67% of 46] Sagi Maimon 8 [58% of 36] Carlo Szelinsky
9 [67% of 24] Minxi Hou 9 [55% of 33] Corigine
10 [65% of 17] Satish Kharat 10 [54% of 101] Motorcomm
And a series info request:
Replies from netdev-bot+sinfo@kernel.org:
Per posting, people (min 15): Per posting, companies (min 30):
1 [39% of 18] Binbin Deng 1 [28% of 32] Jérémy Jean
2 [35% of 17] Qihang 2 [24% of 33] Minerva Networks
3 [33% of 15] Alexandre Cassen 3 [19% of 32] XPeng
4 [33% of 24] Henry Martin 4 [18% of 45] Daehyeon Ko
5 [33% of 15] Mahanta Jambigi 5 [14% of 49] Blockcast
6 [29% of 21] Ömer Mete Kaya 6 [10% of 39] Linkui Xiao
7 [28% of 32] Jérémy Jean 7 [ 9% of 43] TI
8 [24% of 33] Rosen Penev 8 [ 9% of 55] Chinese Academy of Sciences
9 [24% of 17] Bryam Vargas 9 [ 9% of 56] Chengfeng Ye
10 [20% of 15] Zxyan Zhu 10 [ 9% of 112] James Hilliard
The repeated series info requests are particularly frustrating.
For all fixes we ask:
- how was the issue discovered
- was the fix tested (esp. if its a driver fix and author may not have HW)
- was the bug validated/confirmed (IOW triggered to make sure it's not
an LLM hallucination)
https://www.kernel.org/doc/html/next/process/maintainer-netdev.html#bug-fixes
People (especially those blindly trusting LLMs) don't add this info,
even when they post a new version of patch where bot already asked
for it.
--
Code: https://github.com/kuba-moo/ml-stat
Raw output: https://netdev.bots.linux.dev/ml-stats/stats-7.4
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2026-10-22 20:57 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-22 20:57 [ANN] netdev development stats for 7.4 Jakub Kicinski
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox