1 2 How to Get Your Change Into the Linux Kernel 3 or 4 Care And Operation Of Your Linus Torvalds 5 6 7 8For a person or company who wishes to submit a change to the Linux 9kernel, the process can sometimes be daunting if you're not familiar 10with "the system." This text is a collection of suggestions which 11can greatly increase the chances of your change being accepted. 12 13Read Documentation/SubmitChecklist for a list of items to check 14before submitting code. If you are submitting a driver, also read 15Documentation/SubmittingDrivers. 16 17 18 19-------------------------------------------- 20SECTION 1 - CREATING AND SENDING YOUR CHANGE 21-------------------------------------------- 22 23 24 251) "diff -up" 26------------ 27 28Use "diff -up" or "diff -uprN" to create patches. 29 30All changes to the Linux kernel occur in the form of patches, as 31generated by diff(1). When creating your patch, make sure to create it 32in "unified diff" format, as supplied by the '-u' argument to diff(1). 33Also, please use the '-p' argument which shows which C function each 34change is in - that makes the resultant diff a lot easier to read. 35Patches should be based in the root kernel source directory, 36not in any lower subdirectory. 37 38To create a patch for a single file, it is often sufficient to do: 39 40 SRCTREE= linux-2.6 41 MYFILE= drivers/net/mydriver.c 42 43 cd $SRCTREE 44 cp $MYFILE $MYFILE.orig 45 vi $MYFILE # make your change 46 cd .. 47 diff -up $SRCTREE/$MYFILE{.orig,} > /tmp/patch 48 49To create a patch for multiple files, you should unpack a "vanilla", 50or unmodified kernel source tree, and generate a diff against your 51own source tree. For example: 52 53 MYSRC= /devel/linux-2.6 54 55 tar xvfz linux-2.6.12.tar.gz 56 mv linux-2.6.12 linux-2.6.12-vanilla 57 diff -uprN -X linux-2.6.12-vanilla/Documentation/dontdiff \ 58 linux-2.6.12-vanilla $MYSRC > /tmp/patch 59 60"dontdiff" is a list of files which are generated by the kernel during 61the build process, and should be ignored in any diff(1)-generated 62patch. The "dontdiff" file is included in the kernel tree in 632.6.12 and later. For earlier kernel versions, you can get it 64from <http://www.xenotime.net/linux/doc/dontdiff>. 65 66Make sure your patch does not include any extra files which do not 67belong in a patch submission. Make sure to review your patch -after- 68generated it with diff(1), to ensure accuracy. 69 70If your changes produce a lot of deltas, you may want to look into 71splitting them into individual patches which modify things in 72logical stages. This will facilitate easier reviewing by other 73kernel developers, very important if you want your patch accepted. 74There are a number of scripts which can aid in this: 75 76Quilt: 77http://savannah.nongnu.org/projects/quilt 78 79Andrew Morton's patch scripts: 80http://www.zip.com.au/~akpm/linux/patches/ 81Instead of these scripts, quilt is the recommended patch management 82tool (see above). 83 84 85 862) Describe your changes. 87 88Describe the technical detail of the change(s) your patch includes. 89 90Be as specific as possible. The WORST descriptions possible include 91things like "update driver X", "bug fix for driver X", or "this patch 92includes updates for subsystem X. Please apply." 93 94If your description starts to get long, that's a sign that you probably 95need to split up your patch. See #3, next. 96 97 98 993) Separate your changes. 100 101Separate _logical changes_ into a single patch file. 102 103For example, if your changes include both bug fixes and performance 104enhancements for a single driver, separate those changes into two 105or more patches. If your changes include an API update, and a new 106driver which uses that new API, separate those into two patches. 107 108On the other hand, if you make a single change to numerous files, 109group those changes into a single patch. Thus a single logical change 110is contained within a single patch. 111 112If one patch depends on another patch in order for a change to be 113complete, that is OK. Simply note "this patch depends on patch X" 114in your patch description. 115 116If you cannot condense your patch set into a smaller set of patches, 117then only post say 15 or so at a time and wait for review and integration. 118 119 120 1214) Style check your changes. 122 123Check your patch for basic style violations, details of which can be 124found in Documentation/CodingStyle. Failure to do so simply wastes 125the reviewers time and will get your patch rejected, probably 126without even being read. 127 128At a minimum you should check your patches with the patch style 129checker prior to submission (scripts/checkpatch.pl). You should 130be able to justify all violations that remain in your patch. 131 132 133 1345) Select e-mail destination. 135 136Look through the MAINTAINERS file and the source code, and determine 137if your change applies to a specific subsystem of the kernel, with 138an assigned maintainer. If so, e-mail that person. 139 140If no maintainer is listed, or the maintainer does not respond, send 141your patch to the primary Linux kernel developer's mailing list, 142linux-kernel@vger.kernel.org. Most kernel developers monitor this 143e-mail list, and can comment on your changes. 144 145 146Do not send more than 15 patches at once to the vger mailing lists!!! 147 148 149Linus Torvalds is the final arbiter of all changes accepted into the 150Linux kernel. His e-mail address is <torvalds@linux-foundation.org>. 151He gets a lot of e-mail, so typically you should do your best to -avoid- 152sending him e-mail. 153 154Patches which are bug fixes, are "obvious" changes, or similarly 155require little discussion should be sent or CC'd to Linus. Patches 156which require discussion or do not have a clear advantage should 157usually be sent first to linux-kernel. Only after the patch is 158discussed should the patch then be submitted to Linus. 159 160 161 1626) Select your CC (e-mail carbon copy) list. 163 164Unless you have a reason NOT to do so, CC linux-kernel@vger.kernel.org. 165 166Other kernel developers besides Linus need to be aware of your change, 167so that they may comment on it and offer code review and suggestions. 168linux-kernel is the primary Linux kernel developer mailing list. 169Other mailing lists are available for specific subsystems, such as 170USB, framebuffer devices, the VFS, the SCSI subsystem, etc. See the 171MAINTAINERS file for a mailing list that relates specifically to 172your change. 173 174Majordomo lists of VGER.KERNEL.ORG at: 175 <http://vger.kernel.org/vger-lists.html> 176 177If changes affect userland-kernel interfaces, please send 178the MAN-PAGES maintainer (as listed in the MAINTAINERS file) 179a man-pages patch, or at least a notification of the change, 180so that some information makes its way into the manual pages. 181 182Even if the maintainer did not respond in step #4, make sure to ALWAYS 183copy the maintainer when you change their code. 184 185For small patches you may want to CC the Trivial Patch Monkey 186trivial@kernel.org managed by Adrian Bunk; which collects "trivial" 187patches. Trivial patches must qualify for one of the following rules: 188 Spelling fixes in documentation 189 Spelling fixes which could break grep(1) 190 Warning fixes (cluttering with useless warnings is bad) 191 Compilation fixes (only if they are actually correct) 192 Runtime fixes (only if they actually fix things) 193 Removing use of deprecated functions/macros (eg. check_region) 194 Contact detail and documentation fixes 195 Non-portable code replaced by portable code (even in arch-specific, 196 since people copy, as long as it's trivial) 197 Any fix by the author/maintainer of the file (ie. patch monkey 198 in re-transmission mode) 199URL: <http://www.kernel.org/pub/linux/kernel/people/bunk/trivial/> 200 201 202 2037) No MIME, no links, no compression, no attachments. Just plain text. 204 205Linus and other kernel developers need to be able to read and comment 206on the changes you are submitting. It is important for a kernel 207developer to be able to "quote" your changes, using standard e-mail 208tools, so that they may comment on specific portions of your code. 209 210For this reason, all patches should be submitting e-mail "inline". 211WARNING: Be wary of your editor's word-wrap corrupting your patch, 212if you choose to cut-n-paste your patch. 213 214Do not attach the patch as a MIME attachment, compressed or not. 215Many popular e-mail applications will not always transmit a MIME 216attachment as plain text, making it impossible to comment on your 217code. A MIME attachment also takes Linus a bit more time to process, 218decreasing the likelihood of your MIME-attached change being accepted. 219 220Exception: If your mailer is mangling patches then someone may ask 221you to re-send them using MIME. 222 223See Documentation/email-clients.txt for hints about configuring 224your e-mail client so that it sends your patches untouched. 225 2268) E-mail size. 227 228When sending patches to Linus, always follow step #7. 229 230Large changes are not appropriate for mailing lists, and some 231maintainers. If your patch, uncompressed, exceeds 40 kB in size, 232it is preferred that you store your patch on an Internet-accessible 233server, and provide instead a URL (link) pointing to your patch. 234 235 236 2379) Name your kernel version. 238 239It is important to note, either in the subject line or in the patch 240description, the kernel version to which this patch applies. 241 242If the patch does not apply cleanly to the latest kernel version, 243Linus will not apply it. 244 245 246 24710) Don't get discouraged. Re-submit. 248 249After you have submitted your change, be patient and wait. If Linus 250likes your change and applies it, it will appear in the next version 251of the kernel that he releases. 252 253However, if your change doesn't appear in the next version of the 254kernel, there could be any number of reasons. It's YOUR job to 255narrow down those reasons, correct what was wrong, and submit your 256updated change. 257 258It is quite common for Linus to "drop" your patch without comment. 259That's the nature of the system. If he drops your patch, it could be 260due to 261* Your patch did not apply cleanly to the latest kernel version. 262* Your patch was not sufficiently discussed on linux-kernel. 263* A style issue (see section 2). 264* An e-mail formatting issue (re-read this section). 265* A technical problem with your change. 266* He gets tons of e-mail, and yours got lost in the shuffle. 267* You are being annoying. 268 269When in doubt, solicit comments on linux-kernel mailing list. 270 271 272 27311) Include PATCH in the subject 274 275Due to high e-mail traffic to Linus, and to linux-kernel, it is common 276convention to prefix your subject line with [PATCH]. This lets Linus 277and other kernel developers more easily distinguish patches from other 278e-mail discussions. 279 280 281 28212) Sign your work 283 284To improve tracking of who did what, especially with patches that can 285percolate to their final resting place in the kernel through several 286layers of maintainers, we've introduced a "sign-off" procedure on 287patches that are being emailed around. 288 289The sign-off is a simple line at the end of the explanation for the 290patch, which certifies that you wrote it or otherwise have the right to 291pass it on as a open-source patch. The rules are pretty simple: if you 292can certify the below: 293 294 Developer's Certificate of Origin 1.1 295 296 By making a contribution to this project, I certify that: 297 298 (a) The contribution was created in whole or in part by me and I 299 have the right to submit it under the open source license 300 indicated in the file; or 301 302 (b) The contribution is based upon previous work that, to the best 303 of my knowledge, is covered under an appropriate open source 304 license and I have the right under that license to submit that 305 work with modifications, whether created in whole or in part 306 by me, under the same open source license (unless I am 307 permitted to submit under a different license), as indicated 308 in the file; or 309 310 (c) The contribution was provided directly to me by some other 311 person who certified (a), (b) or (c) and I have not modified 312 it. 313 314 (d) I understand and agree that this project and the contribution 315 are public and that a record of the contribution (including all 316 personal information I submit with it, including my sign-off) is 317 maintained indefinitely and may be redistributed consistent with 318 this project or the open source license(s) involved. 319 320then you just add a line saying 321 322 Signed-off-by: Random J Developer <random@developer.example.org> 323 324using your real name (sorry, no pseudonyms or anonymous contributions.) 325 326Some people also put extra tags at the end. They'll just be ignored for 327now, but you can do this to mark internal company procedures or just 328point out some special detail about the sign-off. 329 330 33113) When to use Acked-by: 332 333The Signed-off-by: tag indicates that the signer was involved in the 334development of the patch, or that he/she was in the patch's delivery path. 335 336If a person was not directly involved in the preparation or handling of a 337patch but wishes to signify and record their approval of it then they can 338arrange to have an Acked-by: line added to the patch's changelog. 339 340Acked-by: is often used by the maintainer of the affected code when that 341maintainer neither contributed to nor forwarded the patch. 342 343Acked-by: is not as formal as Signed-off-by:. It is a record that the acker 344has at least reviewed the patch and has indicated acceptance. Hence patch 345mergers will sometimes manually convert an acker's "yep, looks good to me" 346into an Acked-by:. 347 348Acked-by: does not necessarily indicate acknowledgement of the entire patch. 349For example, if a patch affects multiple subsystems and has an Acked-by: from 350one subsystem maintainer then this usually indicates acknowledgement of just 351the part which affects that maintainer's code. Judgement should be used here. 352 When in doubt people should refer to the original discussion in the mailing 353list archives. 354 355 35614) The canonical patch format 357 358The canonical patch subject line is: 359 360 Subject: [PATCH 001/123] subsystem: summary phrase 361 362The canonical patch message body contains the following: 363 364 - A "from" line specifying the patch author. 365 366 - An empty line. 367 368 - The body of the explanation, which will be copied to the 369 permanent changelog to describe this patch. 370 371 - The "Signed-off-by:" lines, described above, which will 372 also go in the changelog. 373 374 - A marker line containing simply "---". 375 376 - Any additional comments not suitable for the changelog. 377 378 - The actual patch (diff output). 379 380The Subject line format makes it very easy to sort the emails 381alphabetically by subject line - pretty much any email reader will 382support that - since because the sequence number is zero-padded, 383the numerical and alphabetic sort is the same. 384 385The "subsystem" in the email's Subject should identify which 386area or subsystem of the kernel is being patched. 387 388The "summary phrase" in the email's Subject should concisely 389describe the patch which that email contains. The "summary 390phrase" should not be a filename. Do not use the same "summary 391phrase" for every patch in a whole patch series (where a "patch 392series" is an ordered sequence of multiple, related patches). 393 394Bear in mind that the "summary phrase" of your email becomes 395a globally-unique identifier for that patch. It propagates 396all the way into the git changelog. The "summary phrase" may 397later be used in developer discussions which refer to the patch. 398People will want to google for the "summary phrase" to read 399discussion regarding that patch. 400 401A couple of example Subjects: 402 403 Subject: [patch 2/5] ext2: improve scalability of bitmap searching 404 Subject: [PATCHv2 001/207] x86: fix eflags tracking 405 406The "from" line must be the very first line in the message body, 407and has the form: 408 409 From: Original Author <author@example.com> 410 411The "from" line specifies who will be credited as the author of the 412patch in the permanent changelog. If the "from" line is missing, 413then the "From:" line from the email header will be used to determine 414the patch author in the changelog. 415 416The explanation body will be committed to the permanent source 417changelog, so should make sense to a competent reader who has long 418since forgotten the immediate details of the discussion that might 419have led to this patch. 420 421The "---" marker line serves the essential purpose of marking for patch 422handling tools where the changelog message ends. 423 424One good use for the additional comments after the "---" marker is for 425a diffstat, to show what files have changed, and the number of inserted 426and deleted lines per file. A diffstat is especially useful on bigger 427patches. Other comments relevant only to the moment or the maintainer, 428not suitable for the permanent changelog, should also go here. 429Use diffstat options "-p 1 -w 70" so that filenames are listed from the 430top of the kernel source tree and don't use too much horizontal space 431(easily fit in 80 columns, maybe with some indentation). 432 433See more details on the proper patch format in the following 434references. 435 436 437 438 439----------------------------------- 440SECTION 2 - HINTS, TIPS, AND TRICKS 441----------------------------------- 442 443This section lists many of the common "rules" associated with code 444submitted to the kernel. There are always exceptions... but you must 445have a really good reason for doing so. You could probably call this 446section Linus Computer Science 101. 447 448 449 4501) Read Documentation/CodingStyle 451 452Nuff said. If your code deviates too much from this, it is likely 453to be rejected without further review, and without comment. 454 455One significant exception is when moving code from one file to 456another -- in this case you should not modify the moved code at all in 457the same patch which moves it. This clearly delineates the act of 458moving the code and your changes. This greatly aids review of the 459actual differences and allows tools to better track the history of 460the code itself. 461 462Check your patches with the patch style checker prior to submission 463(scripts/checkpatch.pl). The style checker should be viewed as 464a guide not as the final word. If your code looks better with 465a violation then its probably best left alone. 466 467The checker reports at three levels: 468 - ERROR: things that are very likely to be wrong 469 - WARNING: things requiring careful review 470 - CHECK: things requiring thought 471 472You should be able to justify all violations that remain in your 473patch. 474 475 476 4772) #ifdefs are ugly 478 479Code cluttered with ifdefs is difficult to read and maintain. Don't do 480it. Instead, put your ifdefs in a header, and conditionally define 481'static inline' functions, or macros, which are used in the code. 482Let the compiler optimize away the "no-op" case. 483 484Simple example, of poor code: 485 486 dev = alloc_etherdev (sizeof(struct funky_private)); 487 if (!dev) 488 return -ENODEV; 489 #ifdef CONFIG_NET_FUNKINESS 490 init_funky_net(dev); 491 #endif 492 493Cleaned-up example: 494 495(in header) 496 #ifndef CONFIG_NET_FUNKINESS 497 static inline void init_funky_net (struct net_device *d) {} 498 #endif 499 500(in the code itself) 501 dev = alloc_etherdev (sizeof(struct funky_private)); 502 if (!dev) 503 return -ENODEV; 504 init_funky_net(dev); 505 506 507 5083) 'static inline' is better than a macro 509 510Static inline functions are greatly preferred over macros. 511They provide type safety, have no length limitations, no formatting 512limitations, and under gcc they are as cheap as macros. 513 514Macros should only be used for cases where a static inline is clearly 515suboptimal [there a few, isolated cases of this in fast paths], 516or where it is impossible to use a static inline function [such as 517string-izing]. 518 519'static inline' is preferred over 'static __inline__', 'extern inline', 520and 'extern __inline__'. 521 522 523 5244) Don't over-design. 525 526Don't try to anticipate nebulous future cases which may or may not 527be useful: "Make it as simple as you can, and no simpler." 528 529 530 531---------------------- 532SECTION 3 - REFERENCES 533---------------------- 534 535Andrew Morton, "The perfect patch" (tpp). 536 <http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt> 537 538Jeff Garzik, "Linux kernel patch submission format". 539 <http://linux.yyz.us/patch-format.html> 540 541Greg Kroah-Hartman, "How to piss off a kernel subsystem maintainer". 542 <http://www.kroah.com/log/2005/03/31/> 543 <http://www.kroah.com/log/2005/07/08/> 544 <http://www.kroah.com/log/2005/10/19/> 545 <http://www.kroah.com/log/2006/01/11/> 546 547NO!!!! No more huge patch bombs to linux-kernel@vger.kernel.org people! 548 <http://marc.theaimsgroup.com/?l=linux-kernel&m=112112749912944&w=2> 549 550Kernel Documentation/CodingStyle: 551 <http://users.sosdg.org/~qiyong/lxr/source/Documentation/CodingStyle> 552 553Linus Torvalds's mail on the canonical patch format: 554 <http://lkml.org/lkml/2005/4/7/183> 555-- 556