Development statistics for 6.3

The 6.3 kernel was released on April 24 after a nine-week development cycle. As is the case with all mainline releases, this is a major kernel release with a lot of changes and a big pile of new features. The time has come, yet again, for a look at where that work came from and who supported it.

The 6.3 development cycle saw the merging of 14,424 non-merge changesets from 1,971 developers, which is a bit of a slowdown from 6.2. Of those developers, 250 made their first kernel contribution for this release. The work merged for 6.2 deleted over 513,000 lines of code — far more than the usual — but the kernel still grew by over 131,000 lines.

The most active developers in this cycle were:

Most active 6.3 developers

By changesets Krzysztof Kozlowski 387 2.7% Dmitry Baryshkov 317 2.2% Arnd Bergmann 185 1.3% Andy Shevchenko 175 1.2% Christoph Hellwig 167 1.2% Uwe Kleine-König 163 1.1% Konrad Dybcio 118 0.8% Sean Christopherson 113 0.8% Martin Kaiser 113 0.8% Chuck Lever 109 0.8% Hans de Goede 104 0.7% Johan Hovold 99 0.7% Thomas Zimmermann 99 0.7% Ville Syrjälä 98 0.7% Mark Brown 97 0.7% Vladimir Oltean 96 0.7% Greg Kroah-Hartman 96 0.7% Randy Dunlap 95 0.7% Jakub Kicinski 93 0.6% Jonathan Cameron 92 0.6%

By changed lines Arnd Bergmann 160437 16.4% Kalle Valo 53435 5.5% Greg Kroah‑Hartman 52609 5.4% Hans Verkuil 28249 2.9% Cai Huoqing 19975 2.0% Wenjing Liu 18159 1.9% Thierry Reding 13698 1.4% Dmitry Baryshkov 12724 1.3% Trevor Wu 12633 1.3% Abel Vesa 11843 1.2% Jakub Kicinski 11591 1.2% Krzysztof Kozlowski 9418 1.0% Steen Hegelund 9124 0.9% Jacek Lawrynowicz 8802 0.9% Herbert Xu 7601 0.8% Ondrej Zary 7584 0.8% Shazad Hussain 7438 0.8% Herve Codina 7032 0.7% Bjorn Andersson 6943 0.7% Neil Armstrong 6769 0.7%

This is the fourth release in a row where Krzysztof Kozlowski appears in the top two changeset contributors; he continues his work with devicetree files. Dmitry Baryshkov worked extensively on a number of Qualcomm device drivers. Among other things, Arnd Bergmann removed a lot of old architecture and device-support code. Andy Shevchenko contributed cleanups across large parts of the driver tree, and Christoph Hellwig continues to refactor code in the block and filesystem areas.

In the changed-lines column, Bergmann's removal work got rid of just over 158,000 lines of code. Kalle Valo added a new Qualcomm WiFi driver. Greg Kroah-Hartman worked throughout the device-driver tree and removed the unneeded r8188eu driver from the staging tree. Hans Verkuil removed a number of old media drivers, and Cai Huoqing removed a set of obsolete graphics drivers.

The top testers and reviewers this time around were:

Test and review credits in 6.3
Tested-by Daniel Wheeler 134 8.2% Philipp Hortmann 112 6.9% Ulf Hansson 44 2.7% Tony Lindgren 44 2.7% Scott Mayhew 41 2.5% Niklas Schnelle 34 2.1% Gurucharan G 34 2.1% Andrew Halaney 33 2.0% Florian Fainelli 23 1.4% Mingming Su 23 1.4%	Reviewed-by Konrad Dybcio 352 4.0% Krzysztof Kozlowski 225 2.5% Rob Herring 146 1.6% Simon Horman 142 1.6% Christoph Hellwig 133 1.5% Laurent Pinchart 126 1.4% AngeloGioacchino Del Regno 124 1.4% Linus Walleij 118 1.3% Dmitry Baryshkov 108 1.2% Hans de Goede 103 1.2%

Daniel Wheeler and Philipp Hortmann are reliably the top testers, regularly adding their tags to Realtek and AMD graphics driver patches, respectively. Ulf Hansson and Tony Lindgren, instead, both tested many of the same patches to the cpuidle subsystem. On the review side, Konrad Dybcio reviewed 352 patches to Qualcomm drivers — at a rate of nearly six patches for every day of the development cycle, weekends and holidays included. Kozlowski and Rob Herring both focused mainly on devicetree patches.

This time around, 1,358 patches (9.4% of the total) had Tested-by tags, while 6,902 (47.9%) had Reviewed-by tags. The increase in the number of patches with Reviewed-by tags noted in the 6.2 development-statistics article continues with 6.3.

A total of 220 employers (that could be identified) supported work on 6.3, a slight drop from 6.2. The most active employers were:

Most active 6.3 employers

By changesets Linaro 1752 12.1% Intel 1416 9.8% Red Hat 1013 7.0% (Unknown) 957 6.6% Google 840 5.8% (None) 686 4.8% AMD 601 4.2% IBM 460 3.2% NVIDIA 455 3.2% Huawei Technologies 413 2.9% Oracle 393 2.7% Meta 363 2.5% SUSE 320 2.2% (Consultant) 300 2.1% Pengutronix 265 1.8% Renesas Electronics 224 1.6% Qualcomm 210 1.5% NXP Semiconductors 201 1.4% Microchip Technology Inc. 166 1.2% Linux Foundation 165 1.1%

By lines changed Linaro 236941 24.2% Qualcomm 80099 8.2% (Unknown) 61511 6.3% Intel 57448 5.9% Linux Foundation 53935 5.5% Red Hat 50334 5.1% AMD 38130 3.9% NVIDIA 35199 3.6% Cisco 28249 2.9% Google 24424 2.5% IBM 21713 2.2% Meta 21334 2.2% (None) 18667 1.9% Microchip Technology Inc. 17778 1.8% MediaTek 17113 1.8% Oracle 12501 1.3% (Consultant) 11013 1.1% Bootlin 8681 0.9% SUSE 7865 0.8% Renesas Electronics 6893 0.7%

Linaro continues its longstanding trend of increasing its contributions over time. In general, though, this table looks about the same as it always does.

Of course, not all companies contribute to the kernel in the same way; each has its own reasons for contributing, and those reasons will drive the work that is done. Some insight can perhaps be gained by looking at which companies dominate in which parts of the kernel. For the following analysis, contributions merged after the 5.17 release were considered, giving just over one year of history.

During that period, 89,392 non-merge changesets landed in the mainline. Of those, 12,579 (14%) touched files in arch/, while 48,132 (54%) touched files in either drivers/ or sound/ — together reflecting work to support specific hardware. The top employers working in those areas were:

Most active employers, 5.18 to 6.3
Architecture subsystems Linaro 1941 15.4% Google 1359 10.8% IBM 1050 8.3% (Unknown) 789 6.3% Intel 638 5.1% (None) 569 4.5% Red Hat 529 4.2% Arm 430 3.4% Renesas Electronics 324 2.6% CS Group 240 1.9%	Driver subsystems Intel 7189 14.9% AMD 4147 8.6% (Unknown) 3292 6.8% Linaro 2667 5.5% (None) 2437 5.1% Huawei Technologies 2154 4.5% Red Hat 2122 4.4% NVIDIA 1831 3.8% Google 1738 3.6% Pengutronix 1430 3.0%

The list of companies working on architecture-specific support is mostly unsurprising. Linaro exists to support the Arm architecture, as does Arm itself. IBM works on the Power architecture, while Intel is focused on x86. Google might seem to a bit of an outlier, but remember that the company is active in both cloud computing and mobile devices. Google's most active contributor under arch/ (Sean Christopherson) has seemingly been rewriting the KVM subsystem on his own, while many other Google developers work on Arm support.

Intel and AMD naturally dominate on the drivers side; supporting their GPUs alone brings a lot of changes into the kernel.

The filesystem and block layers are another area of interest; 6,037 changesets (7% of the total) touched these areas. The core kernel (somewhat arbitrarily defined as the kernel/ and mm/ directories), instead, saw only 4,682 changes — 5% of the total — during this time.

Most active employers, 5.18 to 6.3
Filesystem and block layer Red Hat 877 14.5% SUSE 859 14.2% Oracle 706 11.7% Meta 614 10.2% Huawei Technologies 551 9.1% (Consultant) 456 7.6% (Unknown) 279 4.6% Google 235 3.9% Microsoft 218 3.6% Alibaba 150 2.5%	Core kernel Google 575 12.3% Oracle 537 11.5% Huawei Technologies 468 10.0% Red Hat 456 9.7% Meta 421 9.0% Intel 293 6.3% (Unknown) 206 4.4% (None) 183 3.9% ByteDance 143 3.1% Amazon.com 136 2.9%

The filesystem and block patches came primarily from distributors and companies that run massive data centers of their own. The core-kernel list is similar, but the distributors are less active in that part of the kernel.

Another significant part of the kernel is the networking subsystem. A huge amount of work enters the kernel through the networking tree during each merge window, but only 4,168 changesets (just under 5% of the total) touched core networking; most of the rest applied to the network-interface drivers. Finally, there is the all-important Documentation directory, with the devicetree (Documentation/devicetree) files excluded.

Most active employers, 5.18 to 6.3
Networking Intel 509 12.2% Red Hat 498 11.9% Google 437 10.5% Meta 322 7.7% (Unknown) 317 7.6% NVIDIA 257 6.2% Huawei Technologies 175 4.2% NXP Semiconductors 175 4.2% Oracle 161 3.9% Amazon.com 154 3.7%	Documentation (Unknown) 263 10.8% (None) 251 10.3% Google 233 9.6% Intel 214 8.8% Red Hat 142 5.8% Meta 126 5.2% Loongson 123 5.1% Huawei Technologies 88 3.6% AMD 64 2.6% Amazon.com 55 2.3%

The presence of companies like Red Hat, Google, and Meta in the networking list is not particularly surprising, but one might wonder about a couple of the others. Fully half of Intel's contribution to the networking subsystem comes in the form of Johannes Berg's WiFi work. NVIDIA, instead, found its way into this subsystem by way of its acquisition of Mellanox in 2020.

The Documentation numbers, instead, show a high proportion of developers who are not affiliated with any employer at all. This might be interpreted to mean that companies are relatively reluctant to pay developers to work on documentation; it also reflects the fact that documentation is a common starting place for new developers.

There is exactly one company — Google — are exactly two companies — Google and Red Hat — that spread their contributions widely enough to appear on all of the above lists.

While the reasons driving contributions to the kernel vary; that work all adds up to an impressive body of work, with regular releases every nine or ten weeks. This work looks set to continue in the near future; as of this writing there are just over 12,000 changesets waiting in linux-next for the 6.4 development cycle. Look here for an update on that work once the 6.4 cycle completes.

Index entries for this article
Kernel	Releases/6.3

---

to post comments