1         Linux kernel coding style
  2 
  3 This is a short document describing the preferred coding style for the
  4 linux kernel.  Coding style is very personal, and I won't _force_ my
  5 views on anybody, but this is what goes for anything that I have to be
  6 able to maintain, and I'd prefer it for most other things too.  Please
  7 at least consider the points made here.
  8 
  9 First off, I'd suggest printing out a copy of the GNU coding standards,
 10 and NOT read it.  Burn them, it's a great symbolic gesture.
 11 
 12 Anyway, here goes:
 13 
 14 
 15          Chapter 1: Indentation
 16 
 17 Tabs are 8 characters, and thus indentations are also 8 characters.
 18 There are heretic movements that try to make indentations 4 (or even 2!)
 19 characters deep, and that is akin to trying to define the value of PI to
 20 be 3.
 21 
 22 Rationale: The whole idea behind indentation is to clearly define where
 23 a block of control starts and ends.  Especially when you've been looking
 24 at your screen for 20 straight hours, you'll find it a lot easier to see
 25 how the indentation works if you have large indentations.
 26 
 27 Now, some people will claim that having 8-character indentations makes
 28 the code move too far to the right, and makes it hard to read on a
 29 80-character terminal screen.  The answer to that is that if you need
 30 more than 3 levels of indentation, you're screwed anyway, and should fix
 31 your program.
 32 
 33 In short, 8-char indents make things easier to read, and have the added
 34 benefit of warning you when you're nesting your functions too deep.
 35 Heed that warning.
 36 
 37 The preferred way to ease multiple indentation levels in a switch statement is
 38 to align the "switch" and its subordinate "case" labels in the same column
 39 instead of "double-indenting" the "case" labels.  E.g.:
 40 
 41     switch (suffix) {
 42     case 'G':
 43     case 'g':
 44         mem <<= 30;
 45         break;
 46     case 'M':
 47     case 'm':
 48         mem <<= 20;
 49         break;
 50     case 'K':
 51     case 'k':
 52         mem <<= 10;
 53         /* fall through */
 54     default:
 55         break;
 56     }
 57 
 58 
 59 Don't put multiple statements on a single line unless you have
 60 something to hide:
 61 
 62     if (condition) do_this;
 63       do_something_everytime;
 64 
 65 Don't put multiple assignments on a single line either.  Kernel coding style
 66 is super simple.  Avoid tricky expressions.
 67 
 68 Outside of comments, documentation and except in Kconfig, spaces are never
 69 used for indentation, and the above example is deliberately broken.
 70 
 71 Get a decent editor and don't leave whitespace at the end of lines.
 72 
 73 
 74         Chapter 2: Breaking long lines and strings
 75 
 76 Coding style is all about readability and maintainability using commonly
 77 available tools.
 78 
 79 The limit on the length of lines is 80 columns and this is a strongly
 80 preferred limit.
 81 
 82 Statements longer than 80 columns will be broken into sensible chunks, unless
 83 exceeding 80 columns significantly increases readability and does not hide
 84 information. Descendants are always substantially shorter than the parent and
 85 are placed substantially to the right. The same applies to function headers
 86 with a long argument list. However, never break user-visible strings such as
 87 printk messages, because that breaks the ability to grep for them.
 88 
 89 
 90         Chapter 3: Placing Braces and Spaces
 91 
 92 The other issue that always comes up in C styling is the placement of
 93 braces.  Unlike the indent size, there are few technical reasons to
 94 choose one placement strategy over the other, but the preferred way, as
 95 shown to us by the prophets Kernighan and Ritchie, is to put the opening
 96 brace last on the line, and put the closing brace first, thusly:
 97 
 98     if (x is true) {
 99         we do y
100     }
101 
102 This applies to all non-function statement blocks (if, switch, for,
103 while, do).  E.g.:
104 
105     switch (action) {
106     case KOBJ_ADD:
107         return "add";
108     case KOBJ_REMOVE:
109         return "remove";
110     case KOBJ_CHANGE:
111         return "change";
112     default:
113         return NULL;
114     }
115 
116 However, there is one special case, namely functions: they have the
117 opening brace at the beginning of the next line, thus:
118 
119     int function(int x)
120     {
121         body of function
122     }
123 
124 Heretic people all over the world have claimed that this inconsistency
125 is ...  well ...  inconsistent, but all right-thinking people know that
126 (a) K&R are _right_ and (b) K&R are right.  Besides, functions are
127 special anyway (you can't nest them in C).
128 
129 Note that the closing brace is empty on a line of its own, _except_ in
130 the cases where it is followed by a continuation of the same statement,
131 ie a "while" in a do-statement or an "else" in an if-statement, like
132 this:
133 
134     do {
135         body of do-loop
136     } while (condition);
137 
138 and
139 
140     if (x == y) {
141         ..
142     } else if (x > y) {
143         ...
144     } else {
145         ....
146     }
147 
148 Rationale: K&R.
149 
150 Also, note that this brace-placement also minimizes the number of empty
151 (or almost empty) lines, without any loss of readability.  Thus, as the
152 supply of new-lines on your screen is not a renewable resource (think
153 25-line terminal screens here), you have more empty lines to put
154 comments on.
155 
156 Do not unnecessarily use braces where a single statement will do.
157 
158 if (condition)
159     action();
160 
161 and
162 
163 if (condition)
164     do_this();
165 else
166     do_that();
167 
168 This does not apply if only one branch of a conditional statement is a single
169 statement; in the latter case use braces in both branches:
170 
171 if (condition) {
172     do_this();
173     do_that();
174 } else {
175     otherwise();
176 }
177 
178         3.1:  Spaces
179 
180 Linux kernel style for use of spaces depends (mostly) on
181 function-versus-keyword usage.  Use a space after (most) keywords.  The
182 notable exceptions are sizeof, typeof, alignof, and __attribute__, which look
183 somewhat like functions (and are usually used with parentheses in Linux,
184 although they are not required in the language, as in: "sizeof info" after
185 "struct fileinfo info;" is declared).
186 
187 So use a space after these keywords:
188     if, switch, case, for, do, while
189 but not with sizeof, typeof, alignof, or __attribute__.  E.g.,
190     s = sizeof(struct file);
191 
192 Do not add spaces around (inside) parenthesized expressions.  This example is
193 *bad*:
194 
195     s = sizeof( struct file );
196 
197 When declaring pointer data or a function that returns a pointer type, the
198 preferred use of '*' is adjacent to the data name or function name and not
199 adjacent to the type name.  Examples:
200 
201     char *linux_banner;
202     unsigned long long memparse(char *ptr, char **retptr);
203     char *match_strdup(substring_t *s);
204 
205 Use one space around (on each side of) most binary and ternary operators,
206 such as any of these:
207 
208     =  +  -  <  >  *  /  %  |  &  ^  <=  >=  ==  !=  ?  :
209 
210 but no space after unary operators:
211     &  *  +  -  ~  !  sizeof  typeof  alignof  __attribute__  defined
212 
213 no space before the postfix increment & decrement unary operators:
214     ++  --
215 
216 no space after the prefix increment & decrement unary operators:
217     ++  --
218 
219 and no space around the '.' and "->" structure member operators.
220 
221 Do not leave trailing whitespace at the ends of lines.  Some editors with
222 "smart" indentation will insert whitespace at the beginning of new lines as
223 appropriate, so you can start typing the next line of code right away.
224 However, some such editors do not remove the whitespace if you end up not
225 putting a line of code there, such as if you leave a blank line.  As a result,
226 you end up with lines containing trailing whitespace.
227 
228 Git will warn you about patches that introduce trailing whitespace, and can
229 optionally strip the trailing whitespace for you; however, if applying a series
230 of patches, this may make later patches in the series fail by changing their
231 context lines.
232 
233 
234         Chapter 4: Naming
235 
236 C is a Spartan language, and so should your naming be.  Unlike Modula-2
237 and Pascal programmers, C programmers do not use cute names like
238 ThisVariableIsATemporaryCounter.  A C programmer would call that
239 variable "tmp", which is much easier to write, and not the least more
240 difficult to understand.
241 
242 HOWEVER, while mixed-case names are frowned upon, descriptive names for
243 global variables are a must.  To call a global function "foo" is a
244 shooting offense.
245 
246 GLOBAL variables (to be used only if you _really_ need them) need to
247 have descriptive names, as do global functions.  If you have a function
248 that counts the number of active users, you should call that
249 "count_active_users()" or similar, you should _not_ call it "cntusr()".
250 
251 Encoding the type of a function into the name (so-called Hungarian
252 notation) is brain damaged - the compiler knows the types anyway and can
253 check those, and it only confuses the programmer.  No wonder MicroSoft
254 makes buggy programs.
255 
256 LOCAL variable names should be short, and to the point.  If you have
257 some random integer loop counter, it should probably be called "i".
258 Calling it "loop_counter" is non-productive, if there is no chance of it
259 being mis-understood.  Similarly, "tmp" can be just about any type of
260 variable that is used to hold a temporary value.
261 
262 If you are afraid to mix up your local variable names, you have another
263 problem, which is called the function-growth-hormone-imbalance syndrome.
264 See chapter 6 (Functions).
265 
266 
267         Chapter 5: Typedefs
268 
269 Please don't use things like "vps_t".
270 
271 It's a _mistake_ to use typedef for structures and pointers. When you see a
272 
273     vps_t a;
274 
275 in the source, what does it mean?
276 
277 In contrast, if it says
278 
279     struct virtual_container *a;
280 
281 you can actually tell what "a" is.
282 
283 Lots of people think that typedefs "help readability". Not so. They are
284 useful only for:
285 
286  (a) totally opaque objects (where the typedef is actively used to _hide_
287      what the object is).
288 
289      Example: "pte_t" etc. opaque objects that you can only access using
290      the proper accessor functions.
291 
292      NOTE! Opaqueness and "accessor functions" are not good in themselves.
293      The reason we have them for things like pte_t etc. is that there
294      really is absolutely _zero_ portably accessible information there.
295 
296  (b) Clear integer types, where the abstraction _helps_ avoid confusion
297      whether it is "int" or "long".
298 
299      u8/u16/u32 are perfectly fine typedefs, although they fit into
300      category (d) better than here.
301 
302      NOTE! Again - there needs to be a _reason_ for this. If something is
303      "unsigned long", then there's no reason to do
304 
305     typedef unsigned long myflags_t;
306 
307      but if there is a clear reason for why it under certain circumstances
308      might be an "unsigned int" and under other configurations might be
309      "unsigned long", then by all means go ahead and use a typedef.
310 
311  (c) when you use sparse to literally create a _new_ type for
312      type-checking.
313 
314  (d) New types which are identical to standard C99 types, in certain
315      exceptional circumstances.
316 
317      Although it would only take a short amount of time for the eyes and
318      brain to become accustomed to the standard types like 'uint32_t',
319      some people object to their use anyway.
320 
321      Therefore, the Linux-specific 'u8/u16/u32/u64' types and their
322      signed equivalents which are identical to standard types are
323      permitted -- although they are not mandatory in new code of your
324      own.
325 
326      When editing existing code which already uses one or the other set
327      of types, you should conform to the existing choices in that code.
328 
329  (e) Types safe for use in userspace.
330 
331      In certain structures which are visible to userspace, we cannot
332      require C99 types and cannot use the 'u32' form above. Thus, we
333      use __u32 and similar types in all structures which are shared
334      with userspace.
335 
336 Maybe there are other cases too, but the rule should basically be to NEVER
337 EVER use a typedef unless you can clearly match one of those rules.
338 
339 In general, a pointer, or a struct that has elements that can reasonably
340 be directly accessed should _never_ be a typedef.
341 
342 
343         Chapter 6: Functions
344 
345 Functions should be short and sweet, and do just one thing.  They should
346 fit on one or two screenfuls of text (the ISO/ANSI screen size is 80x24,
347 as we all know), and do one thing and do that well.
348 
349 The maximum length of a function is inversely proportional to the
350 complexity and indentation level of that function.  So, if you have a
351 conceptually simple function that is just one long (but simple)
352 case-statement, where you have to do lots of small things for a lot of
353 different cases, it's OK to have a longer function.
354 
355 However, if you have a complex function, and you suspect that a
356 less-than-gifted first-year high-school student might not even
357 understand what the function is all about, you should adhere to the
358 maximum limits all the more closely.  Use helper functions with
359 descriptive names (you can ask the compiler to in-line them if you think
360 it's performance-critical, and it will probably do a better job of it
361 than you would have done).
362 
363 Another measure of the function is the number of local variables.  They
364 shouldn't exceed 5-10, or you're doing something wrong.  Re-think the
365 function, and split it into smaller pieces.  A human brain can
366 generally easily keep track of about 7 different things, anything more
367 and it gets confused.  You know you're brilliant, but maybe you'd like
368 to understand what you did 2 weeks from now.
369 
370 In source files, separate functions with one blank line.  If the function is
371 exported, the EXPORT* macro for it should follow immediately after the closing
372 function brace line.  E.g.:
373 
374 int system_is_up(void)
375 {
376     return system_state == SYSTEM_RUNNING;
377 }
378 EXPORT_SYMBOL(system_is_up);
379 
380 In function prototypes, include parameter names with their data types.
381 Although this is not required by the C language, it is preferred in Linux
382 because it is a simple way to add valuable information for the reader.
383 
384 
385         Chapter 7: Centralized exiting of functions
386 
387 Albeit deprecated by some people, the equivalent of the goto statement is
388 used frequently by compilers in form of the unconditional jump instruction.
389 
390 The goto statement comes in handy when a function exits from multiple
391 locations and some common work such as cleanup has to be done.
392 
393 The rationale is:
394 
395 - unconditional statements are easier to understand and follow
396 - nesting is reduced
397 - errors by not updating individual exit points when making
398     modifications are prevented
399 - saves the compiler work to optimize redundant code away ;)
400 
401 int fun(int a)
402 {
403     int result = 0;
404     char *buffer = kmalloc(SIZE);
405 
406     if (buffer == NULL)
407         return -ENOMEM;
408 
409     if (condition1) {
410         while (loop1) {
411             ...
412         }
413         result = 1;
414         goto out;
415     }
416     ...
417 out:
418     kfree(buffer);
419     return result;
420 }
421 
422         Chapter 8: Commenting
423 
424 Comments are good, but there is also a danger of over-commenting.  NEVER
425 try to explain HOW your code works in a comment: it's much better to
426 write the code so that the _working_ is obvious, and it's a waste of
427 time to explain badly written code.
428 
429 Generally, you want your comments to tell WHAT your code does, not HOW.
430 Also, try to avoid putting comments inside a function body: if the
431 function is so complex that you need to separately comment parts of it,
432 you should probably go back to chapter 6 for a while.  You can make
433 small comments to note or warn about something particularly clever (or
434 ugly), but try to avoid excess.  Instead, put the comments at the head
435 of the function, telling people what it does, and possibly WHY it does
436 it.
437 
438 When commenting the kernel API functions, please use the kernel-doc format.
439 See the files Documentation/kernel-doc-nano-HOWTO.txt and scripts/kernel-doc
440 for details.
441 
442 Linux style for comments is the C89 "/* ... */" style.
443 Don't use C99-style "http:// ..." comments.
444 
445 The preferred style for long (multi-line) comments is:
446 
447     /*
448      * This is the preferred style for multi-line
449      * comments in the Linux kernel source code.
450      * Please use it consistently.
451      *
452      * Description:  A column of asterisks on the left side,
453      * with beginning and ending almost-blank lines.
454      */
455 
456 For files in net/ and drivers/net/ the preferred style for long (multi-line)
457 comments is a little different.
458 
459     /* The preferred comment style for files in net/ and drivers/net
460      * looks like this.
461      *
462      * It is nearly the same as the generally preferred comment style,
463      * but there is no initial almost-blank line.
464      */
465 
466 It's also important to comment data, whether they are basic types or derived
467 types.  To this end, use just one data declaration per line (no commas for
468 multiple data declarations).  This leaves you room for a small comment on each
469 item, explaining its use.
470 
471 
472         Chapter 9: You've made a mess of it
473 
474 That's OK, we all do.  You've probably been told by your long-time Unix
475 user helper that "GNU emacs" automatically formats the C sources for
476 you, and you've noticed that yes, it does do that, but the defaults it
477 uses are less than desirable (in fact, they are worse than random
478 typing - an infinite number of monkeys typing into GNU emacs would never
479 make a good program).
480 
481 So, you can either get rid of GNU emacs, or change it to use saner
482 values.  To do the latter, you can stick the following in your .emacs file:
483 
484 (defun c-lineup-arglist-tabs-only (ignored)
485   "Line up argument lists by tabs, not spaces"
486   (let* ((anchor (c-langelem-pos c-syntactic-element))
487      (column (c-langelem-2nd-pos c-syntactic-element))
488      (offset (- (1+ column) anchor))
489      (steps (floor offset c-basic-offset)))
490     (* (max steps 1)
491        c-basic-offset)))
492 
493 (add-hook 'c-mode-common-hook
494           (lambda ()
495             ;; Add kernel style
496             (c-add-style
497              "linux-tabs-only"
498              '("linux" (c-offsets-alist
499                         (arglist-cont-nonempty
500                          c-lineup-gcc-asm-reg
501                          c-lineup-arglist-tabs-only))))))
502 
503 (add-hook 'c-mode-hook
504           (lambda ()
505             (let ((filename (buffer-file-name)))
506               ;; Enable kernel mode for the appropriate files
507               (when (and filename
508                          (string-match (expand-file-name "~/src/linux-trees")
509                                        filename))
510                 (setq indent-tabs-mode t)
511                 (c-set-style "linux-tabs-only")))))
512 
513 This will make emacs go better with the kernel coding style for C
514 files below ~/src/linux-trees.
515 
516 But even if you fail in getting emacs to do sane formatting, not
517 everything is lost: use "indent".
518 
519 Now, again, GNU indent has the same brain-dead settings that GNU emacs
520 has, which is why you need to give it a few command line options.
521 However, that's not too bad, because even the makers of GNU indent
522 recognize the authority of K&R (the GNU people aren't evil, they are
523 just severely misguided in this matter), so you just give indent the
524 options "-kr -i8" (stands for "K&R, 8 character indents"), or use
525 "scripts/Lindent", which indents in the latest style.
526 
527 "indent" has a lot of options, and especially when it comes to comment
528 re-formatting you may want to take a look at the man page.  But
529 remember: "indent" is not a fix for bad programming.
530 
531 
532         Chapter 10: Kconfig configuration files
533 
534 For all of the Kconfig* configuration files throughout the source tree,
535 the indentation is somewhat different.  Lines under a "config" definition
536 are indented with one tab, while help text is indented an additional two
537 spaces.  Example:
538 
539 config AUDIT
540     bool "Auditing support"
541     depends on NET
542     help
543       Enable auditing infrastructure that can be used with another
544       kernel subsystem, such as SELinux (which requires this for
545       logging of avc messages output).  Does not do system-call
546       auditing without CONFIG_AUDITSYSCALL.
547 
548 Features that might still be considered unstable should be defined as
549 dependent on "EXPERIMENTAL":
550 
551 config SLUB
552     depends on EXPERIMENTAL && !ARCH_USES_SLAB_PAGE_STRUCT
553     bool "SLUB (Unqueued Allocator)"
554     ...
555 
556 while seriously dangerous features (such as write support for certain
557 filesystems) should advertise this prominently in their prompt string:
558 
559 config ADFS_FS_RW
560     bool "ADFS write support (DANGEROUS)"
561     depends on ADFS_FS
562     ...
563 
564 For full documentation on the configuration files, see the file
565 Documentation/kbuild/kconfig-language.txt.
566 
567 
568         Chapter 11: Data structures
569 
570 Data structures that have visibility outside the single-threaded
571 environment they are created and destroyed in should always have
572 reference counts.  In the kernel, garbage collection doesn't exist (and
573 outside the kernel garbage collection is slow and inefficient), which
574 means that you absolutely _have_ to reference count all your uses.
575 
576 Reference counting means that you can avoid locking, and allows multiple
577 users to have access to the data structure in parallel - and not having
578 to worry about the structure suddenly going away from under them just
579 because they slept or did something else for a while.
580 
581 Note that locking is _not_ a replacement for reference counting.
582 Locking is used to keep data structures coherent, while reference
583 counting is a memory management technique.  Usually both are needed, and
584 they are not to be confused with each other.
585 
586 Many data structures can indeed have two levels of reference counting,
587 when there are users of different "classes".  The subclass count counts
588 the number of subclass users, and decrements the global count just once
589 when the subclass count goes to zero.
590 
591 Examples of this kind of "multi-level-reference-counting" can be found in
592 memory management ("struct mm_struct": mm_users and mm_count), and in
593 filesystem code ("struct super_block": s_count and s_active).
594 
595 Remember: if another thread can find your data structure, and you don't
596 have a reference count on it, you almost certainly have a bug.
597 
598 
599         Chapter 12: Macros, Enums and RTL
600 
601 Names of macros defining constants and labels in enums are capitalized.
602 
603 #define CONSTANT 0x12345
604 
605 Enums are preferred when defining several related constants.
606 
607 CAPITALIZED macro names are appreciated but macros resembling functions
608 may be named in lower case.
609 
610 Generally, inline functions are preferable to macros resembling functions.
611 
612 Macros with multiple statements should be enclosed in a do - while block:
613 
614 #define macrofun(a, b, c)             \
615     do {                    \
616         if (a == 5)            \
617             do_this(b, c);        \
618     } while (0)
619 
620 Things to avoid when using macros:
621 
622 1) macros that affect control flow:
623 
624 #define FOO(x)                    \
625     do {                    \
626         if (blah(x) < 0)        \
627             return -EBUGGERED;    \
628     } while(0)
629 
630 is a _very_ bad idea.  It looks like a function call but exits the "calling"
631 function; don't break the internal parsers of those who will read the code.
632 
633 2) macros that depend on having a local variable with a magic name:
634 
635 #define FOO(val) bar(index, val)
636 
637 might look like a good thing, but it's confusing as hell when one reads the
638 code and it's prone to breakage from seemingly innocent changes.
639 
640 3) macros with arguments that are used as l-values: FOO(x) = y; will
641 bite you if somebody e.g. turns FOO into an inline function.
642 
643 4) forgetting about precedence: macros defining constants using expressions
644 must enclose the expression in parentheses. Beware of similar issues with
645 macros using parameters.
646 
647 #define CONSTANT 0x4000
648 #define CONSTEXP (CONSTANT | 3)
649 
650 The cpp manual deals with macros exhaustively. The gcc internals manual also
651 covers RTL which is used frequently with assembly language in the kernel.
652 
653 
654         Chapter 13: Printing kernel messages
655 
656 Kernel developers like to be seen as literate. Do mind the spelling
657 of kernel messages to make a good impression. Do not use crippled
658 words like "dont"; use "do not" or "don't" instead.  Make the messages
659 concise, clear, and unambiguous.
660 
661 Kernel messages do not have to be terminated with a period.
662 
663 Printing numbers in parentheses (%d) adds no value and should be avoided.
664 
665 There are a number of driver model diagnostic macros in <linux/device.h>
666 which you should use to make sure messages are matched to the right device
667 and driver, and are tagged with the right level:  dev_err(), dev_warn(),
668 dev_info(), and so forth.  For messages that aren't associated with a
669 particular device, <linux/printk.h> defines pr_debug() and pr_info().
670 
671 Coming up with good debugging messages can be quite a challenge; and once
672 you have them, they can be a huge help for remote troubleshooting.  Such
673 messages should be compiled out when the DEBUG symbol is not defined (that
674 is, by default they are not included).  When you use dev_dbg() or pr_debug(),
675 that's automatic.  Many subsystems have Kconfig options to turn on -DDEBUG.
676 A related convention uses VERBOSE_DEBUG to add dev_vdbg() messages to the
677 ones already enabled by DEBUG.
678 
679 
680         Chapter 14: Allocating memory
681 
682 The kernel provides the following general purpose memory allocators:
683 kmalloc(), kzalloc(), kmalloc_array(), kcalloc(), vmalloc(), and
684 vzalloc().  Please refer to the API documentation for further information
685 about them.
686 
687 The preferred form for passing a size of a struct is the following:
688 
689     p = kmalloc(sizeof(*p), ...);
690 
691 The alternative form where struct name is spelled out hurts readability and
692 introduces an opportunity for a bug when the pointer variable type is changed
693 but the corresponding sizeof that is passed to a memory allocator is not.
694 
695 Casting the return value which is a void pointer is redundant. The conversion
696 from void pointer to any other pointer type is guaranteed by the C programming
697 language.
698 
699 The preferred form for allocating an array is the following:
700 
701     p = kmalloc_array(n, sizeof(...), ...);
702 
703 The preferred form for allocating a zeroed array is the following:
704 
705     p = kcalloc(n, sizeof(...), ...);
706 
707 Both forms check for overflow on the allocation size n * sizeof(...),
708 and return NULL if that occurred.
709 
710 
711         Chapter 15: The inline disease
712 
713 There appears to be a common misperception that gcc has a magic "make me
714 faster" speedup option called "inline". While the use of inlines can be
715 appropriate (for example as a means of replacing macros, see Chapter 12), it
716 very often is not. Abundant use of the inline keyword leads to a much bigger
717 kernel, which in turn slows the system as a whole down, due to a bigger
718 icache footprint for the CPU and simply because there is less memory
719 available for the pagecache. Just think about it; a pagecache miss causes a
720 disk seek, which easily takes 5 milliseconds. There are a LOT of cpu cycles
721 that can go into these 5 milliseconds.
722 
723 A reasonable rule of thumb is to not put inline at functions that have more
724 than 3 lines of code in them. An exception to this rule are the cases where
725 a parameter is known to be a compiletime constant, and as a result of this
726 constantness you *know* the compiler will be able to optimize most of your
727 function away at compile time. For a good example of this later case, see
728 the kmalloc() inline function.
729 
730 Often people argue that adding inline to functions that are static and used
731 only once is always a win since there is no space tradeoff. While this is
732 technically correct, gcc is capable of inlining these automatically without
733 help, and the maintenance issue of removing the inline when a second user
734 appears outweighs the potential value of the hint that tells gcc to do
735 something it would have done anyway.
736 
737 
738         Chapter 16: Function return values and names
739 
740 Functions can return values of many different kinds, and one of the
741 most common is a value indicating whether the function succeeded or
742 failed.  Such a value can be represented as an error-code integer
743 (-Exxx = failure, 0 = success) or a "succeeded" boolean (0 = failure,
744 non-zero = success).
745 
746 Mixing up these two sorts of representations is a fertile source of
747 difficult-to-find bugs.  If the C language included a strong distinction
748 between integers and booleans then the compiler would find these mistakes
749 for us... but it doesn't.  To help prevent such bugs, always follow this
750 convention:
751 
752     If the name of a function is an action or an imperative command,
753     the function should return an error-code integer.  If the name
754     is a predicate, the function should return a "succeeded" boolean.
755 
756 For example, "add work" is a command, and the add_work() function returns 0
757 for success or -EBUSY for failure.  In the same way, "PCI device present" is
758 a predicate, and the pci_dev_present() function returns 1 if it succeeds in
759 finding a matching device or 0 if it doesn't.
760 
761 All EXPORTed functions must respect this convention, and so should all
762 public functions.  Private (static) functions need not, but it is
763 recommended that they do.
764 
765 Functions whose return value is the actual result of a computation, rather
766 than an indication of whether the computation succeeded, are not subject to
767 this rule.  Generally they indicate failure by returning some out-of-range
768 result.  Typical examples would be functions that return pointers; they use
769 NULL or the ERR_PTR mechanism to report failure.
770 
771 
772         Chapter 17:  Don't re-invent the kernel macros
773 
774 The header file include/linux/kernel.h contains a number of macros that
775 you should use, rather than explicitly coding some variant of them yourself.
776 For example, if you need to calculate the length of an array, take advantage
777 of the macro
778 
779   #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
780 
781 Similarly, if you need to calculate the size of some structure member, use
782 
783   #define FIELD_SIZEOF(t, f) (sizeof(((t*)0)->f))
784 
785 There are also min() and max() macros that do strict type checking if you
786 need them.  Feel free to peruse that header file to see what else is already
787 defined that you shouldn't reproduce in your code.
788 
789 
790         Chapter 18:  Editor modelines and other cruft
791 
792 Some editors can interpret configuration information embedded in source files,
793 indicated with special markers.  For example, emacs interprets lines marked
794 like this:
795 
796 -*- mode: c -*-
797 
798 Or like this:
799 
800 /*
801 Local Variables:
802 compile-command: "gcc -DMAGIC_DEBUG_FLAG foo.c"
803 End:
804 */
805 
806 Vim interprets markers that look like this:
807 
808 /* vim:set sw=8 noet */
809 
810 Do not include any of these in source files.  People have their own personal
811 editor configurations, and your source files should not override them.  This
812 includes markers for indentation and mode configuration.  People may use their
813 own custom mode, or may have some other magic method for making indentation
814 work correctly.
815 
816 
817         Chapter 19:  Inline assembly
818 
819 In architecture-specific code, you may need to use inline assembly to interface
820 with CPU or platform functionality.  Don't hesitate to do so when necessary.
821 However, don't use inline assembly gratuitously when C can do the job.  You can
822 and should poke hardware from C when possible.
823 
824 Consider writing simple helper functions that wrap common bits of inline
825 assembly, rather than repeatedly writing them with slight variations.  Remember
826 that inline assembly can use C parameters.
827 
828 Large, non-trivial assembly functions should go in .S files, with corresponding
829 C prototypes defined in C header files.  The C prototypes for assembly
830 functions should use "asmlinkage".
831 
832 You may need to mark your asm statement as volatile, to prevent GCC from
833 removing it if GCC doesn't notice any side effects.  You don't always need to
834 do so, though, and doing so unnecessarily can limit optimization.
835 
836 When writing a single inline assembly statement containing multiple
837 instructions, put each instruction on a separate line in a separate quoted
838 string, and end each string except the last with \n\t to properly indent the
839 next instruction in the assembly output:
840 
841     asm ("magic %reg1, #42\n\t"
842          "more_magic %reg2, %reg3"
843          : /* outputs */ : /* inputs */ : /* clobbers */);
844 
845 
846 
847         Appendix I: References
848 
849 The C Programming Language, Second Edition
850 by Brian W. Kernighan and Dennis M. Ritchie.
851 Prentice Hall, Inc., 1988.
852 ISBN 0-13-110362-8 (paperback), 0-13-110370-9 (hardback).
853 URL: http://cm.bell-labs.com/cm/cs/cbook/
854 
855 The Practice of Programming
856 by Brian W. Kernighan and Rob Pike.
857 Addison-Wesley, Inc., 1999.
858 ISBN 0-201-61586-X.
859 URL: http://cm.bell-labs.com/cm/cs/tpop/
860 
861 GNU manuals - where in compliance with K&R and this text - for cpp, gcc,
862 gcc internals and indent, all available from http://www.gnu.org/manual/
863 
864 WG14 is the international standardization working group for the programming
865 language C, URL: http://www.open-std.org/JTC1/SC22/WG14/
866 
867 Kernel CodingStyle, by greg@kroah.com at OLS 2002:
868 http://www.kroah.com/linux/talks/ols_2002_kernel_codingstyle_talk/html/