1 .. _development_process:
3 How the development process works
4 =================================
6 Linux kernel development in the early 1990's was a pretty loose affair,
7 with relatively small numbers of users and developers involved. With a
8 user base in the millions and with some 2,000 developers involved over the
9 course of one year, the kernel has since had to evolve a number of
10 processes to keep development happening smoothly. A solid understanding of
11 how the process works is required in order to be an effective part of it.
16 The kernel developers use a loosely time-based release process, with a new
17 major kernel release happening every two or three months. The recent
18 release history looks like this:
20 ====== =================
22 2.6.37 January 4, 2011
23 2.6.36 October 20, 2010
26 2.6.33 February 24, 2010
27 ====== =================
29 Every 2.6.x release is a major kernel release with new features, internal
30 API changes, and more. A typical 2.6 release can contain nearly 10,000
31 changesets with changes to several hundred thousand lines of code. 2.6 is
32 thus the leading edge of Linux kernel development; the kernel uses a
33 rolling development model which is continually integrating major changes.
35 A relatively straightforward discipline is followed with regard to the
36 merging of patches for each release. At the beginning of each development
37 cycle, the "merge window" is said to be open. At that time, code which is
38 deemed to be sufficiently stable (and which is accepted by the development
39 community) is merged into the mainline kernel. The bulk of changes for a
40 new development cycle (and all of the major changes) will be merged during
41 this time, at a rate approaching 1,000 changes ("patches," or "changesets")
44 (As an aside, it is worth noting that the changes integrated during the
45 merge window do not come out of thin air; they have been collected, tested,
46 and staged ahead of time. How that process works will be described in
49 The merge window lasts for approximately two weeks. At the end of this
50 time, Linus Torvalds will declare that the window is closed and release the
51 first of the "rc" kernels. For the kernel which is destined to be 2.6.40,
52 for example, the release which happens at the end of the merge window will
53 be called 2.6.40-rc1. The -rc1 release is the signal that the time to
54 merge new features has passed, and that the time to stabilize the next
57 Over the next six to ten weeks, only patches which fix problems should be
58 submitted to the mainline. On occasion a more significant change will be
59 allowed, but such occasions are rare; developers who try to merge new
60 features outside of the merge window tend to get an unfriendly reception.
61 As a general rule, if you miss the merge window for a given feature, the
62 best thing to do is to wait for the next development cycle. (An occasional
63 exception is made for drivers for previously-unsupported hardware; if they
64 touch no in-tree code, they cannot cause regressions and should be safe to
67 As fixes make their way into the mainline, the patch rate will slow over
68 time. Linus releases new -rc kernels about once a week; a normal series
69 will get up to somewhere between -rc6 and -rc9 before the kernel is
70 considered to be sufficiently stable and the final 2.6.x release is made.
71 At that point the whole process starts over again.
73 As an example, here is how the 2.6.38 development cycle went (all dates in
76 ============== ===============================
77 January 4 2.6.37 stable release
78 January 18 2.6.38-rc1, merge window closes
82 February 15 2.6.38-rc5
83 February 21 2.6.38-rc6
86 March 14 2.6.38 stable release
87 ============== ===============================
89 How do the developers decide when to close the development cycle and create
90 the stable release? The most significant metric used is the list of
91 regressions from previous releases. No bugs are welcome, but those which
92 break systems which worked in the past are considered to be especially
93 serious. For this reason, patches which cause regressions are looked upon
94 unfavorably and are quite likely to be reverted during the stabilization
97 The developers' goal is to fix all known regressions before the stable
98 release is made. In the real world, this kind of perfection is hard to
99 achieve; there are just too many variables in a project of this size.
100 There comes a point where delaying the final release just makes the problem
101 worse; the pile of changes waiting for the next merge window will grow
102 larger, creating even more regressions the next time around. So most 2.6.x
103 kernels go out with a handful of known regressions though, hopefully, none
106 Once a stable release is made, its ongoing maintenance is passed off to the
107 "stable team," currently consisting of Greg Kroah-Hartman. The stable team
108 will release occasional updates to the stable release using the 2.6.x.y
109 numbering scheme. To be considered for an update release, a patch must (1)
110 fix a significant bug, and (2) already be merged into the mainline for the
111 next development kernel. Kernels will typically receive stable updates for
112 a little more than one development cycle past their initial release. So,
113 for example, the 2.6.36 kernel's history looked like:
115 ============== ===============================
116 October 10 2.6.36 stable release
121 ============== ===============================
123 2.6.36.4 was the final stable update for the 2.6.36 release.
125 Some kernels are designated "long term" kernels; they will receive support
126 for a longer period. As of this writing, the current long term kernels
127 and their maintainers are:
129 ====== ====================== ===========================
130 2.6.27 Willy Tarreau (Deep-frozen stable kernel)
131 2.6.32 Greg Kroah-Hartman
132 2.6.35 Andi Kleen (Embedded flag kernel)
133 ====== ====================== ===========================
135 The selection of a kernel for long-term support is purely a matter of a
136 maintainer having the need and the time to maintain that release. There
137 are no known plans for long-term support for any specific upcoming
141 The lifecycle of a patch
142 ------------------------
144 Patches do not go directly from the developer's keyboard into the mainline
145 kernel. There is, instead, a somewhat involved (if somewhat informal)
146 process designed to ensure that each patch is reviewed for quality and that
147 each patch implements a change which is desirable to have in the mainline.
148 This process can happen quickly for minor fixes, or, in the case of large
149 and controversial changes, go on for years. Much developer frustration
150 comes from a lack of understanding of this process or from attempts to
153 In the hopes of reducing that frustration, this document will describe how
154 a patch gets into the kernel. What follows below is an introduction which
155 describes the process in a somewhat idealized way. A much more detailed
156 treatment will come in later sections.
158 The stages that a patch goes through are, generally:
160 - Design. This is where the real requirements for the patch - and the way
161 those requirements will be met - are laid out. Design work is often
162 done without involving the community, but it is better to do this work
163 in the open if at all possible; it can save a lot of time redesigning
166 - Early review. Patches are posted to the relevant mailing list, and
167 developers on that list reply with any comments they may have. This
168 process should turn up any major problems with a patch if all goes
171 - Wider review. When the patch is getting close to ready for mainline
172 inclusion, it should be accepted by a relevant subsystem maintainer -
173 though this acceptance is not a guarantee that the patch will make it
174 all the way to the mainline. The patch will show up in the maintainer's
175 subsystem tree and into the -next trees (described below). When the
176 process works, this step leads to more extensive review of the patch and
177 the discovery of any problems resulting from the integration of this
178 patch with work being done by others.
180 - Please note that most maintainers also have day jobs, so merging
181 your patch may not be their highest priority. If your patch is
182 getting feedback about changes that are needed, you should either
183 make those changes or justify why they should not be made. If your
184 patch has no review complaints but is not being merged by its
185 appropriate subsystem or driver maintainer, you should be persistent
186 in updating the patch to the current kernel so that it applies cleanly
187 and keep sending it for review and merging.
189 - Merging into the mainline. Eventually, a successful patch will be
190 merged into the mainline repository managed by Linus Torvalds. More
191 comments and/or problems may surface at this time; it is important that
192 the developer be responsive to these and fix any issues which arise.
194 - Stable release. The number of users potentially affected by the patch
195 is now large, so, once again, new problems may arise.
197 - Long-term maintenance. While it is certainly possible for a developer
198 to forget about code after merging it, that sort of behavior tends to
199 leave a poor impression in the development community. Merging code
200 eliminates some of the maintenance burden, in that others will fix
201 problems caused by API changes. But the original developer should
202 continue to take responsibility for the code if it is to remain useful
205 One of the largest mistakes made by kernel developers (or their employers)
206 is to try to cut the process down to a single "merging into the mainline"
207 step. This approach invariably leads to frustration for everybody
210 How patches get into the Kernel
211 -------------------------------
213 There is exactly one person who can merge patches into the mainline kernel
214 repository: Linus Torvalds. But, of the over 9,500 patches which went
215 into the 2.6.38 kernel, only 112 (around 1.3%) were directly chosen by Linus
216 himself. The kernel project has long since grown to a size where no single
217 developer could possibly inspect and select every patch unassisted. The
218 way the kernel developers have addressed this growth is through the use of
219 a lieutenant system built around a chain of trust.
221 The kernel code base is logically broken down into a set of subsystems:
222 networking, specific architecture support, memory management, video
223 devices, etc. Most subsystems have a designated maintainer, a developer
224 who has overall responsibility for the code within that subsystem. These
225 subsystem maintainers are the gatekeepers (in a loose way) for the portion
226 of the kernel they manage; they are the ones who will (usually) accept a
227 patch for inclusion into the mainline kernel.
229 Subsystem maintainers each manage their own version of the kernel source
230 tree, usually (but certainly not always) using the git source management
231 tool. Tools like git (and related tools like quilt or mercurial) allow
232 maintainers to track a list of patches, including authorship information
233 and other metadata. At any given time, the maintainer can identify which
234 patches in his or her repository are not found in the mainline.
236 When the merge window opens, top-level maintainers will ask Linus to "pull"
237 the patches they have selected for merging from their repositories. If
238 Linus agrees, the stream of patches will flow up into his repository,
239 becoming part of the mainline kernel. The amount of attention that Linus
240 pays to specific patches received in a pull operation varies. It is clear
241 that, sometimes, he looks quite closely. But, as a general rule, Linus
242 trusts the subsystem maintainers to not send bad patches upstream.
244 Subsystem maintainers, in turn, can pull patches from other maintainers.
245 For example, the networking tree is built from patches which accumulated
246 first in trees dedicated to network device drivers, wireless networking,
247 etc. This chain of repositories can be arbitrarily long, though it rarely
248 exceeds two or three links. Since each maintainer in the chain trusts
249 those managing lower-level trees, this process is known as the "chain of
252 Clearly, in a system like this, getting patches into the kernel depends on
253 finding the right maintainer. Sending patches directly to Linus is not
254 normally the right way to go.
260 The chain of subsystem trees guides the flow of patches into the kernel,
261 but it also raises an interesting question: what if somebody wants to look
262 at all of the patches which are being prepared for the next merge window?
263 Developers will be interested in what other changes are pending to see
264 whether there are any conflicts to worry about; a patch which changes a
265 core kernel function prototype, for example, will conflict with any other
266 patches which use the older form of that function. Reviewers and testers
267 want access to the changes in their integrated form before all of those
268 changes land in the mainline kernel. One could pull changes from all of
269 the interesting subsystem trees, but that would be a big and error-prone
272 The answer comes in the form of -next trees, where subsystem trees are
273 collected for testing and review. The older of these trees, maintained by
274 Andrew Morton, is called "-mm" (for memory management, which is how it got
275 started). The -mm tree integrates patches from a long list of subsystem
276 trees; it also has some patches aimed at helping with debugging.
278 Beyond that, -mm contains a significant collection of patches which have
279 been selected by Andrew directly. These patches may have been posted on a
280 mailing list, or they may apply to a part of the kernel for which there is
281 no designated subsystem tree. As a result, -mm operates as a sort of
282 subsystem tree of last resort; if there is no other obvious path for a
283 patch into the mainline, it is likely to end up in -mm. Miscellaneous
284 patches which accumulate in -mm will eventually either be forwarded on to
285 an appropriate subsystem tree or be sent directly to Linus. In a typical
286 development cycle, approximately 5-10% of the patches going into the
287 mainline get there via -mm.
289 The current -mm patch is available in the "mmotm" (-mm of the moment)
292 http://www.ozlabs.org/~akpm/mmotm/
294 Use of the MMOTM tree is likely to be a frustrating experience, though;
295 there is a definite chance that it will not even compile.
297 The primary tree for next-cycle patch merging is linux-next, maintained by
298 Stephen Rothwell. The linux-next tree is, by design, a snapshot of what
299 the mainline is expected to look like after the next merge window closes.
300 Linux-next trees are announced on the linux-kernel and linux-next mailing
301 lists when they are assembled; they can be downloaded from:
303 http://www.kernel.org/pub/linux/kernel/next/
305 Linux-next has become an integral part of the kernel development process;
306 all patches merged during a given merge window should really have found
307 their way into linux-next some time before the merge window opens.
313 The kernel source tree contains the drivers/staging/ directory, where
314 many sub-directories for drivers or filesystems that are on their way to
315 being added to the kernel tree live. They remain in drivers/staging while
316 they still need more work; once complete, they can be moved into the
317 kernel proper. This is a way to keep track of drivers that aren't
318 up to Linux kernel coding or quality standards, but people may want to use
319 them and track development.
321 Greg Kroah-Hartman currently maintains the staging tree. Drivers that
322 still need work are sent to him, with each driver having its own
323 subdirectory in drivers/staging/. Along with the driver source files, a
324 TODO file should be present in the directory as well. The TODO file lists
325 the pending work that the driver needs for acceptance into the kernel
326 proper, as well as a list of people that should be Cc'd for any patches to
327 the driver. Current rules require that drivers contributed to staging
328 must, at a minimum, compile properly.
330 Staging can be a relatively easy way to get new drivers into the mainline
331 where, with luck, they will come to the attention of other developers and
332 improve quickly. Entry into staging is not the end of the story, though;
333 code in staging which is not seeing regular progress will eventually be
334 removed. Distributors also tend to be relatively reluctant to enable
335 staging drivers. So staging is, at best, a stop on the way toward becoming
336 a proper mainline driver.
342 As can be seen from the above text, the kernel development process depends
343 heavily on the ability to herd collections of patches in various
344 directions. The whole thing would not work anywhere near as well as it
345 does without suitably powerful tools. Tutorials on how to use these tools
346 are well beyond the scope of this document, but there is space for a few
349 By far the dominant source code management system used by the kernel
350 community is git. Git is one of a number of distributed version control
351 systems being developed in the free software community. It is well tuned
352 for kernel development, in that it performs quite well when dealing with
353 large repositories and large numbers of patches. It also has a reputation
354 for being difficult to learn and use, though it has gotten better over
355 time. Some sort of familiarity with git is almost a requirement for kernel
356 developers; even if they do not use it for their own work, they'll need git
357 to keep up with what other developers (and the mainline) are doing.
359 Git is now packaged by almost all Linux distributions. There is a home
364 That page has pointers to documentation and tutorials.
366 Among the kernel developers who do not use git, the most popular choice is
367 almost certainly Mercurial:
369 http://www.selenic.com/mercurial/
371 Mercurial shares many features with git, but it provides an interface which
372 many find easier to use.
374 The other tool worth knowing about is Quilt:
376 http://savannah.nongnu.org/projects/quilt/
378 Quilt is a patch management system, rather than a source code management
379 system. It does not track history over time; it is, instead, oriented
380 toward tracking a specific set of changes against an evolving code base.
381 Some major subsystem maintainers use quilt to manage patches intended to go
382 upstream. For the management of certain kinds of trees (-mm, for example),
383 quilt is the best tool for the job.
389 A great deal of Linux kernel development work is done by way of mailing
390 lists. It is hard to be a fully-functioning member of the community
391 without joining at least one list somewhere. But Linux mailing lists also
392 represent a potential hazard to developers, who risk getting buried under a
393 load of electronic mail, running afoul of the conventions used on the Linux
396 Most kernel mailing lists are run on vger.kernel.org; the master list can
399 http://vger.kernel.org/vger-lists.html
401 There are lists hosted elsewhere, though; a number of them are at
404 The core mailing list for kernel development is, of course, linux-kernel.
405 This list is an intimidating place to be; volume can reach 500 messages per
406 day, the amount of noise is high, the conversation can be severely
407 technical, and participants are not always concerned with showing a high
408 degree of politeness. But there is no other place where the kernel
409 development community comes together as a whole; developers who avoid this
410 list will miss important information.
412 There are a few hints which can help with linux-kernel survival:
414 - Have the list delivered to a separate folder, rather than your main
415 mailbox. One must be able to ignore the stream for sustained periods of
418 - Do not try to follow every conversation - nobody else does. It is
419 important to filter on both the topic of interest (though note that
420 long-running conversations can drift away from the original subject
421 without changing the email subject line) and the people who are
424 - Do not feed the trolls. If somebody is trying to stir up an angry
425 response, ignore them.
427 - When responding to linux-kernel email (or that on other lists) preserve
428 the Cc: header for all involved. In the absence of a strong reason (such
429 as an explicit request), you should never remove recipients. Always make
430 sure that the person you are responding to is in the Cc: list. This
431 convention also makes it unnecessary to explicitly ask to be copied on
432 replies to your postings.
434 - Search the list archives (and the net as a whole) before asking
435 questions. Some developers can get impatient with people who clearly
436 have not done their homework.
438 - Avoid top-posting (the practice of putting your answer above the quoted
439 text you are responding to). It makes your response harder to read and
440 makes a poor impression.
442 - Ask on the correct mailing list. Linux-kernel may be the general meeting
443 point, but it is not the best place to find developers from all
446 The last point - finding the correct mailing list - is a common place for
447 beginning developers to go wrong. Somebody who asks a networking-related
448 question on linux-kernel will almost certainly receive a polite suggestion
449 to ask on the netdev list instead, as that is the list frequented by most
450 networking developers. Other lists exist for the SCSI, video4linux, IDE,
451 filesystem, etc. subsystems. The best place to look for mailing lists is
452 in the MAINTAINERS file packaged with the kernel source.
455 Getting started with Kernel development
456 ---------------------------------------
458 Questions about how to get started with the kernel development process are
459 common - from both individuals and companies. Equally common are missteps
460 which make the beginning of the relationship harder than it has to be.
462 Companies often look to hire well-known developers to get a development
463 group started. This can, in fact, be an effective technique. But it also
464 tends to be expensive and does not do much to grow the pool of experienced
465 kernel developers. It is possible to bring in-house developers up to speed
466 on Linux kernel development, given the investment of a bit of time. Taking
467 this time can endow an employer with a group of developers who understand
468 the kernel and the company both, and who can help to train others as well.
469 Over the medium term, this is often the more profitable approach.
471 Individual developers are often, understandably, at a loss for a place to
472 start. Beginning with a large project can be intimidating; one often wants
473 to test the waters with something smaller first. This is the point where
474 some developers jump into the creation of patches fixing spelling errors or
475 minor coding style issues. Unfortunately, such patches create a level of
476 noise which is distracting for the development community as a whole, so,
477 increasingly, they are looked down upon. New developers wishing to
478 introduce themselves to the community will not get the sort of reception
479 they wish for by these means.
481 Andrew Morton gives this advice for aspiring kernel developers
485 The #1 project for all kernel beginners should surely be "make sure
486 that the kernel runs perfectly at all times on all machines which
487 you can lay your hands on". Usually the way to do this is to work
488 with others on getting things fixed up (this can require
489 persistence!) but that's fine - it's a part of kernel development.
491 (http://lwn.net/Articles/283982/).
493 In the absence of obvious problems to fix, developers are advised to look
494 at the current lists of regressions and open bugs in general. There is
495 never any shortage of issues in need of fixing; by addressing these issues,
496 developers will gain experience with the process while, at the same time,
497 building respect with the rest of the development community.
git.samba.org / sfrench / cifs-2.6.git / blob