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