2                Linux kernel coding style
   4This is a short document describing the preferred coding style for the
   5linux kernel.  Coding style is very personal, and I won't _force_ my
   6views on anybody, but this is what goes for anything that I have to be
   7able to maintain, and I'd prefer it for most other things too.  Please
   8at least consider the points made here.
  10First off, I'd suggest printing out a copy of the GNU coding standards,
  11and NOT read it.  Burn them, it's a great symbolic gesture.
  13Anyway, here goes:
  16                Chapter 1: Indentation
  18Tabs are 8 characters, and thus indentations are also 8 characters.
  19There are heretic movements that try to make indentations 4 (or even 2!)
  20characters deep, and that is akin to trying to define the value of PI to
  21be 3.
  23Rationale: The whole idea behind indentation is to clearly define where
  24a block of control starts and ends.  Especially when you've been looking
  25at your screen for 20 straight hours, you'll find it a lot easier to see
  26how the indentation works if you have large indentations.
  28Now, some people will claim that having 8-character indentations makes
  29the code move too far to the right, and makes it hard to read on a
  3080-character terminal screen.  The answer to that is that if you need
  31more than 3 levels of indentation, you're screwed anyway, and should fix
  32your program.
  34In short, 8-char indents make things easier to read, and have the added
  35benefit of warning you when you're nesting your functions too deep.
  36Heed that warning.
  38The preferred way to ease multiple indentation levels in a switch statement is
  39to align the "switch" and its subordinate "case" labels in the same column
  40instead of "double-indenting" the "case" labels.  E.g.:
  42        switch (suffix) {
  43        case 'G':
  44        case 'g':
  45                mem <<= 30;
  46                break;
  47        case 'M':
  48        case 'm':
  49                mem <<= 20;
  50                break;
  51        case 'K':
  52        case 'k':
  53                mem <<= 10;
  54                /* fall through */
  55        default:
  56                break;
  57        }
  59Don't put multiple statements on a single line unless you have
  60something to hide:
  62        if (condition) do_this;
  63          do_something_everytime;
  65Don't put multiple assignments on a single line either.  Kernel coding style
  66is super simple.  Avoid tricky expressions.
  68Outside of comments, documentation and except in Kconfig, spaces are never
  69used for indentation, and the above example is deliberately broken.
  71Get a decent editor and don't leave whitespace at the end of lines.
  74                Chapter 2: Breaking long lines and strings
  76Coding style is all about readability and maintainability using commonly
  77available tools.
  79The limit on the length of lines is 80 columns and this is a strongly
  80preferred limit.
  82Statements longer than 80 columns will be broken into sensible chunks, unless
  83exceeding 80 columns significantly increases readability and does not hide
  84information. Descendants are always substantially shorter than the parent and
  85are placed substantially to the right. The same applies to function headers
  86with a long argument list. However, never break user-visible strings such as
  87printk messages, because that breaks the ability to grep for them.
  90                Chapter 3: Placing Braces and Spaces
  92The other issue that always comes up in C styling is the placement of
  93braces.  Unlike the indent size, there are few technical reasons to
  94choose one placement strategy over the other, but the preferred way, as
  95shown to us by the prophets Kernighan and Ritchie, is to put the opening
  96brace last on the line, and put the closing brace first, thusly:
  98        if (x is true) {
  99                we do y
 100        }
 102This applies to all non-function statement blocks (if, switch, for,
 103while, do).  E.g.:
 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        }
 116However, there is one special case, namely functions: they have the
 117opening brace at the beginning of the next line, thus:
 119        int function(int x)
 120        {
 121                body of function
 122        }
 124Heretic people all over the world have claimed that this inconsistency
 125is ...  well ...  inconsistent, but all right-thinking people know that
 126(a) K&R are _right_ and (b) K&R are right.  Besides, functions are
 127special anyway (you can't nest them in C).
 129Note that the closing brace is empty on a line of its own, _except_ in
 130the cases where it is followed by a continuation of the same statement,
 131ie a "while" in a do-statement or an "else" in an if-statement, like
 134        do {
 135                body of do-loop
 136        } while (condition);
 140        if (x == y) {
 141                ..
 142        } else if (x > y) {
 143                ...
 144        } else {
 145                ....
 146        }
 148Rationale: K&R.
 150Also, note that this brace-placement also minimizes the number of empty
 151(or almost empty) lines, without any loss of readability.  Thus, as the
 152supply of new-lines on your screen is not a renewable resource (think
 15325-line terminal screens here), you have more empty lines to put
 154comments on.
 156Do not unnecessarily use braces where a single statement will do.
 158        if (condition)
 159                action();
 163        if (condition)
 164                do_this();
 165        else
 166                do_that();
 168This does not apply if only one branch of a conditional statement is a single
 169statement; in the latter case use braces in both branches:
 171        if (condition) {
 172                do_this();
 173                do_that();
 174        } else {
 175                otherwise();
 176        }
 178                3.1:  Spaces
 180Linux kernel style for use of spaces depends (mostly) on
 181function-versus-keyword usage.  Use a space after (most) keywords.  The
 182notable exceptions are sizeof, typeof, alignof, and __attribute__, which look
 183somewhat like functions (and are usually used with parentheses in Linux,
 184although they are not required in the language, as in: "sizeof info" after
 185"struct fileinfo info;" is declared).
 187So use a space after these keywords:
 189        if, switch, case, for, do, while
 191but not with sizeof, typeof, alignof, or __attribute__.  E.g.,
 193        s = sizeof(struct file);
 195Do not add spaces around (inside) parenthesized expressions.  This example is
 198        s = sizeof( struct file );
 200When declaring pointer data or a function that returns a pointer type, the
 201preferred use of '*' is adjacent to the data name or function name and not
 202adjacent to the type name.  Examples:
 204        char *linux_banner;
 205        unsigned long long memparse(char *ptr, char **retptr);
 206        char *match_strdup(substring_t *s);
 208Use one space around (on each side of) most binary and ternary operators,
 209such as any of these:
 211        =  +  -  <  >  *  /  %  |  &  ^  <=  >=  ==  !=  ?  :
 213but no space after unary operators:
 215        &  *  +  -  ~  !  sizeof  typeof  alignof  __attribute__  defined
 217no space before the postfix increment & decrement unary operators:
 219        ++  --
 221no space after the prefix increment & decrement unary operators:
 223        ++  --
 225and no space around the '.' and "->" structure member operators.
 227Do not leave trailing whitespace at the ends of lines.  Some editors with
 228"smart" indentation will insert whitespace at the beginning of new lines as
 229appropriate, so you can start typing the next line of code right away.
 230However, some such editors do not remove the whitespace if you end up not
 231putting a line of code there, such as if you leave a blank line.  As a result,
 232you end up with lines containing trailing whitespace.
 234Git will warn you about patches that introduce trailing whitespace, and can
 235optionally strip the trailing whitespace for you; however, if applying a series
 236of patches, this may make later patches in the series fail by changing their
 237context lines.
 240                Chapter 4: Naming
 242C is a Spartan language, and so should your naming be.  Unlike Modula-2
 243and Pascal programmers, C programmers do not use cute names like
 244ThisVariableIsATemporaryCounter.  A C programmer would call that
 245variable "tmp", which is much easier to write, and not the least more
 246difficult to understand.
 248HOWEVER, while mixed-case names are frowned upon, descriptive names for
 249global variables are a must.  To call a global function "foo" is a
 250shooting offense.
 252GLOBAL variables (to be used only if you _really_ need them) need to
 253have descriptive names, as do global functions.  If you have a function
 254that counts the number of active users, you should call that
 255"count_active_users()" or similar, you should _not_ call it "cntusr()".
 257Encoding the type of a function into the name (so-called Hungarian
 258notation) is brain damaged - the compiler knows the types anyway and can
 259check those, and it only confuses the programmer.  No wonder MicroSoft
 260makes buggy programs.
 262LOCAL variable names should be short, and to the point.  If you have
 263some random integer loop counter, it should probably be called "i".
 264Calling it "loop_counter" is non-productive, if there is no chance of it
 265being mis-understood.  Similarly, "tmp" can be just about any type of
 266variable that is used to hold a temporary value.
 268If you are afraid to mix up your local variable names, you have another
 269problem, which is called the function-growth-hormone-imbalance syndrome.
 270See chapter 6 (Functions).
 273                Chapter 5: Typedefs
 275Please don't use things like "vps_t".
 276It's a _mistake_ to use typedef for structures and pointers. When you see a
 278        vps_t a;
 280in the source, what does it mean?
 281In contrast, if it says
 283        struct virtual_container *a;
 285you can actually tell what "a" is.
 287Lots of people think that typedefs "help readability". Not so. They are
 288useful only for:
 290 (a) totally opaque objects (where the typedef is actively used to _hide_
 291     what the object is).
 293     Example: "pte_t" etc. opaque objects that you can only access using
 294     the proper accessor functions.
 296     NOTE! Opaqueness and "accessor functions" are not good in themselves.
 297     The reason we have them for things like pte_t etc. is that there
 298     really is absolutely _zero_ portably accessible information there.
 300 (b) Clear integer types, where the abstraction _helps_ avoid confusion
 301     whether it is "int" or "long".
 303     u8/u16/u32 are perfectly fine typedefs, although they fit into
 304     category (d) better than here.
 306     NOTE! Again - there needs to be a _reason_ for this. If something is
 307     "unsigned long", then there's no reason to do
 309        typedef unsigned long myflags_t;
 311     but if there is a clear reason for why it under certain circumstances
 312     might be an "unsigned int" and under other configurations might be
 313     "unsigned long", then by all means go ahead and use a typedef.
 315 (c) when you use sparse to literally create a _new_ type for
 316     type-checking.
 318 (d) New types which are identical to standard C99 types, in certain
 319     exceptional circumstances.
 321     Although it would only take a short amount of time for the eyes and
 322     brain to become accustomed to the standard types like 'uint32_t',
 323     some people object to their use anyway.
 325     Therefore, the Linux-specific 'u8/u16/u32/u64' types and their
 326     signed equivalents which are identical to standard types are
 327     permitted -- although they are not mandatory in new code of your
 328     own.
 330     When editing existing code which already uses one or the other set
 331     of types, you should conform to the existing choices in that code.
 333 (e) Types safe for use in userspace.
 335     In certain structures which are visible to userspace, we cannot
 336     require C99 types and cannot use the 'u32' form above. Thus, we
 337     use __u32 and similar types in all structures which are shared
 338     with userspace.
 340Maybe there are other cases too, but the rule should basically be to NEVER
 341EVER use a typedef unless you can clearly match one of those rules.
 343In general, a pointer, or a struct that has elements that can reasonably
 344be directly accessed should _never_ be a typedef.
 347                Chapter 6: Functions
 349Functions should be short and sweet, and do just one thing.  They should
 350fit on one or two screenfuls of text (the ISO/ANSI screen size is 80x24,
 351as we all know), and do one thing and do that well.
 353The maximum length of a function is inversely proportional to the
 354complexity and indentation level of that function.  So, if you have a
 355conceptually simple function that is just one long (but simple)
 356case-statement, where you have to do lots of small things for a lot of
 357different cases, it's OK to have a longer function.
 359However, if you have a complex function, and you suspect that a
 360less-than-gifted first-year high-school student might not even
 361understand what the function is all about, you should adhere to the
 362maximum limits all the more closely.  Use helper functions with
 363descriptive names (you can ask the compiler to in-line them if you think
 364it's performance-critical, and it will probably do a better job of it
 365than you would have done).
 367Another measure of the function is the number of local variables.  They
 368shouldn't exceed 5-10, or you're doing something wrong.  Re-think the
 369function, and split it into smaller pieces.  A human brain can
 370generally easily keep track of about 7 different things, anything more
 371and it gets confused.  You know you're brilliant, but maybe you'd like
 372to understand what you did 2 weeks from now.
 374In source files, separate functions with one blank line.  If the function is
 375exported, the EXPORT* macro for it should follow immediately after the closing
 376function brace line.  E.g.:
 378        int system_is_up(void)
 379        {
 380                return system_state == SYSTEM_RUNNING;
 381        }
 382        EXPORT_SYMBOL(system_is_up);
 384In function prototypes, include parameter names with their data types.
 385Although this is not required by the C language, it is preferred in Linux
 386because it is a simple way to add valuable information for the reader.
 389                Chapter 7: Centralized exiting of functions
 391Albeit deprecated by some people, the equivalent of the goto statement is
 392used frequently by compilers in form of the unconditional jump instruction.
 394The goto statement comes in handy when a function exits from multiple
 395locations and some common work such as cleanup has to be done.  If there is no
 396cleanup needed then just return directly.
 398Choose label names which say what the goto does or why the goto exists.  An
 399example of a good name could be "out_buffer:" if the goto frees "buffer".  Avoid
 400using GW-BASIC names like "err1:" and "err2:".  Also don't name them after the
 401goto location like "err_kmalloc_failed:"
 403The rationale for using gotos is:
 405- unconditional statements are easier to understand and follow
 406- nesting is reduced
 407- errors by not updating individual exit points when making
 408    modifications are prevented
 409- saves the compiler work to optimize redundant code away ;)
 411        int fun(int a)
 412        {
 413                int result = 0;
 414                char *buffer;
 416                buffer = kmalloc(SIZE, GFP_KERNEL);
 417                if (!buffer)
 418                        return -ENOMEM;
 420                if (condition1) {
 421                        while (loop1) {
 422                                ...
 423                        }
 424                        result = 1;
 425                        goto out_buffer;
 426                }
 427                ...
 428        out_buffer:
 429                kfree(buffer);
 430                return result;
 431        }
 433A common type of bug to be aware of is "one err bugs" which look like this:
 435        err:
 436                kfree(foo->bar);
 437                kfree(foo);
 438                return ret;
 440The bug in this code is that on some exit paths "foo" is NULL.  Normally the
 441fix for this is to split it up into two error labels "err_bar:" and "err_foo:".
 444                Chapter 8: Commenting
 446Comments are good, but there is also a danger of over-commenting.  NEVER
 447try to explain HOW your code works in a comment: it's much better to
 448write the code so that the _working_ is obvious, and it's a waste of
 449time to explain badly written code.
 451Generally, you want your comments to tell WHAT your code does, not HOW.
 452Also, try to avoid putting comments inside a function body: if the
 453function is so complex that you need to separately comment parts of it,
 454you should probably go back to chapter 6 for a while.  You can make
 455small comments to note or warn about something particularly clever (or
 456ugly), but try to avoid excess.  Instead, put the comments at the head
 457of the function, telling people what it does, and possibly WHY it does
 460When commenting the kernel API functions, please use the kernel-doc format.
 461See the files Documentation/kernel-doc-nano-HOWTO.txt and scripts/kernel-doc
 462for details.
 464Linux style for comments is the C89 "/* ... */" style.
 465Don't use C99-style "// ..." comments.
 467The preferred style for long (multi-line) comments is:
 469        /*
 470         * This is the preferred style for multi-line
 471         * comments in the Linux kernel source code.
 472         * Please use it consistently.
 473         *
 474         * Description:  A column of asterisks on the left side,
 475         * with beginning and ending almost-blank lines.
 476         */
 478For files in net/ and drivers/net/ the preferred style for long (multi-line)
 479comments is a little different.
 481        /* The preferred comment style for files in net/ and drivers/net
 482         * looks like this.
 483         *
 484         * It is nearly the same as the generally preferred comment style,
 485         * but there is no initial almost-blank line.
 486         */
 488It's also important to comment data, whether they are basic types or derived
 489types.  To this end, use just one data declaration per line (no commas for
 490multiple data declarations).  This leaves you room for a small comment on each
 491item, explaining its use.
 494                Chapter 9: You've made a mess of it
 496That's OK, we all do.  You've probably been told by your long-time Unix
 497user helper that "GNU emacs" automatically formats the C sources for
 498you, and you've noticed that yes, it does do that, but the defaults it
 499uses are less than desirable (in fact, they are worse than random
 500typing - an infinite number of monkeys typing into GNU emacs would never
 501make a good program).
 503So, you can either get rid of GNU emacs, or change it to use saner
 504values.  To do the latter, you can stick the following in your .emacs file:
 506(defun c-lineup-arglist-tabs-only (ignored)
 507  "Line up argument lists by tabs, not spaces"
 508  (let* ((anchor (c-langelem-pos c-syntactic-element))
 509         (column (c-langelem-2nd-pos c-syntactic-element))
 510         (offset (- (1+ column) anchor))
 511         (steps (floor offset c-basic-offset)))
 512    (* (max steps 1)
 513       c-basic-offset)))
 515(add-hook 'c-mode-common-hook
 516          (lambda ()
 517            ;; Add kernel style
 518            (c-add-style
 519             "linux-tabs-only"
 520             '("linux" (c-offsets-alist
 521                        (arglist-cont-nonempty
 522                         c-lineup-gcc-asm-reg
 523                         c-lineup-arglist-tabs-only))))))
 525(add-hook 'c-mode-hook
 526          (lambda ()
 527            (let ((filename (buffer-file-name)))
 528              ;; Enable kernel mode for the appropriate files
 529              (when (and filename
 530                         (string-match (expand-file-name "~/src/linux-trees")
 531                                       filename))
 532                (setq indent-tabs-mode t)
 533                (setq show-trailing-whitespace t)
 534                (c-set-style "linux-tabs-only")))))
 536This will make emacs go better with the kernel coding style for C
 537files below ~/src/linux-trees.
 539But even if you fail in getting emacs to do sane formatting, not
 540everything is lost: use "indent".
 542Now, again, GNU indent has the same brain-dead settings that GNU emacs
 543has, which is why you need to give it a few command line options.
 544However, that's not too bad, because even the makers of GNU indent
 545recognize the authority of K&R (the GNU people aren't evil, they are
 546just severely misguided in this matter), so you just give indent the
 547options "-kr -i8" (stands for "K&R, 8 character indents"), or use
 548"scripts/Lindent", which indents in the latest style.
 550"indent" has a lot of options, and especially when it comes to comment
 551re-formatting you may want to take a look at the man page.  But
 552remember: "indent" is not a fix for bad programming.
 555                Chapter 10: Kconfig configuration files
 557For all of the Kconfig* configuration files throughout the source tree,
 558the indentation is somewhat different.  Lines under a "config" definition
 559are indented with one tab, while help text is indented an additional two
 560spaces.  Example:
 562config AUDIT
 563        bool "Auditing support"
 564        depends on NET
 565        help
 566          Enable auditing infrastructure that can be used with another
 567          kernel subsystem, such as SELinux (which requires this for
 568          logging of avc messages output).  Does not do system-call
 569          auditing without CONFIG_AUDITSYSCALL.
 571Seriously dangerous features (such as write support for certain
 572filesystems) should advertise this prominently in their prompt string:
 574config ADFS_FS_RW
 575        bool "ADFS write support (DANGEROUS)"
 576        depends on ADFS_FS
 577        ...
 579For full documentation on the configuration files, see the file
 583                Chapter 11: Data structures
 585Data structures that have visibility outside the single-threaded
 586environment they are created and destroyed in should always have
 587reference counts.  In the kernel, garbage collection doesn't exist (and
 588outside the kernel garbage collection is slow and inefficient), which
 589means that you absolutely _have_ to reference count all your uses.
 591Reference counting means that you can avoid locking, and allows multiple
 592users to have access to the data structure in parallel - and not having
 593to worry about the structure suddenly going away from under them just
 594because they slept or did something else for a while.
 596Note that locking is _not_ a replacement for reference counting.
 597Locking is used to keep data structures coherent, while reference
 598counting is a memory management technique.  Usually both are needed, and
 599they are not to be confused with each other.
 601Many data structures can indeed have two levels of reference counting,
 602when there are users of different "classes".  The subclass count counts
 603the number of subclass users, and decrements the global count just once
 604when the subclass count goes to zero.
 606Examples of this kind of "multi-level-reference-counting" can be found in
 607memory management ("struct mm_struct": mm_users and mm_count), and in
 608filesystem code ("struct super_block": s_count and s_active).
 610Remember: if another thread can find your data structure, and you don't
 611have a reference count on it, you almost certainly have a bug.
 614                Chapter 12: Macros, Enums and RTL
 616Names of macros defining constants and labels in enums are capitalized.
 618        #define CONSTANT 0x12345
 620Enums are preferred when defining several related constants.
 622CAPITALIZED macro names are appreciated but macros resembling functions
 623may be named in lower case.
 625Generally, inline functions are preferable to macros resembling functions.
 627Macros with multiple statements should be enclosed in a do - while block:
 629        #define macrofun(a, b, c)                       \
 630                do {                                    \
 631                        if (a == 5)                     \
 632                                do_this(b, c);          \
 633                } while (0)
 635Things to avoid when using macros:
 6371) macros that affect control flow:
 639        #define FOO(x)                                  \
 640                do {                                    \
 641                        if (blah(x) < 0)                \
 642                                return -EBUGGERED;      \
 643                } while (0)
 645is a _very_ bad idea.  It looks like a function call but exits the "calling"
 646function; don't break the internal parsers of those who will read the code.
 6482) macros that depend on having a local variable with a magic name:
 650        #define FOO(val) bar(index, val)
 652might look like a good thing, but it's confusing as hell when one reads the
 653code and it's prone to breakage from seemingly innocent changes.
 6553) macros with arguments that are used as l-values: FOO(x) = y; will
 656bite you if somebody e.g. turns FOO into an inline function.
 6584) forgetting about precedence: macros defining constants using expressions
 659must enclose the expression in parentheses. Beware of similar issues with
 660macros using parameters.
 662        #define CONSTANT 0x4000
 663        #define CONSTEXP (CONSTANT | 3)
 6655) namespace collisions when defining local variables in macros resembling
 668#define FOO(x)                          \
 669({                                      \
 670        typeof(x) ret;                  \
 671        ret = calc_ret(x);              \
 672        (ret);                          \
 675ret is a common name for a local variable - __foo_ret is less likely
 676to collide with an existing variable.
 678The cpp manual deals with macros exhaustively. The gcc internals manual also
 679covers RTL which is used frequently with assembly language in the kernel.
 682                Chapter 13: Printing kernel messages
 684Kernel developers like to be seen as literate. Do mind the spelling
 685of kernel messages to make a good impression. Do not use crippled
 686words like "dont"; use "do not" or "don't" instead.  Make the messages
 687concise, clear, and unambiguous.
 689Kernel messages do not have to be terminated with a period.
 691Printing numbers in parentheses (%d) adds no value and should be avoided.
 693There are a number of driver model diagnostic macros in <linux/device.h>
 694which you should use to make sure messages are matched to the right device
 695and driver, and are tagged with the right level:  dev_err(), dev_warn(),
 696dev_info(), and so forth.  For messages that aren't associated with a
 697particular device, <linux/printk.h> defines pr_notice(), pr_info(),
 698pr_warn(), pr_err(), etc.
 700Coming up with good debugging messages can be quite a challenge; and once
 701you have them, they can be a huge help for remote troubleshooting.  However
 702debug message printing is handled differently than printing other non-debug
 703messages.  While the other pr_XXX() functions print unconditionally,
 704pr_debug() does not; it is compiled out by default, unless either DEBUG is
 705defined or CONFIG_DYNAMIC_DEBUG is set.  That is true for dev_dbg() also,
 706and a related convention uses VERBOSE_DEBUG to add dev_vdbg() messages to
 707the ones already enabled by DEBUG.
 709Many subsystems have Kconfig debug options to turn on -DDEBUG in the
 710corresponding Makefile; in other cases specific files #define DEBUG.  And
 711when a debug message should be unconditionally printed, such as if it is
 712already inside a debug-related #ifdef section, printk(KERN_DEBUG ...) can be
 716                Chapter 14: Allocating memory
 718The kernel provides the following general purpose memory allocators:
 719kmalloc(), kzalloc(), kmalloc_array(), kcalloc(), vmalloc(), and
 720vzalloc().  Please refer to the API documentation for further information
 721about them.
 723The preferred form for passing a size of a struct is the following:
 725        p = kmalloc(sizeof(*p), ...);
 727The alternative form where struct name is spelled out hurts readability and
 728introduces an opportunity for a bug when the pointer variable type is changed
 729but the corresponding sizeof that is passed to a memory allocator is not.
 731Casting the return value which is a void pointer is redundant. The conversion
 732from void pointer to any other pointer type is guaranteed by the C programming
 735The preferred form for allocating an array is the following:
 737        p = kmalloc_array(n, sizeof(...), ...);
 739The preferred form for allocating a zeroed array is the following:
 741        p = kcalloc(n, sizeof(...), ...);
 743Both forms check for overflow on the allocation size n * sizeof(...),
 744and return NULL if that occurred.
 747                Chapter 15: The inline disease
 749There appears to be a common misperception that gcc has a magic "make me
 750faster" speedup option called "inline". While the use of inlines can be
 751appropriate (for example as a means of replacing macros, see Chapter 12), it
 752very often is not. Abundant use of the inline keyword leads to a much bigger
 753kernel, which in turn slows the system as a whole down, due to a bigger
 754icache footprint for the CPU and simply because there is less memory
 755available for the pagecache. Just think about it; a pagecache miss causes a
 756disk seek, which easily takes 5 milliseconds. There are a LOT of cpu cycles
 757that can go into these 5 milliseconds.
 759A reasonable rule of thumb is to not put inline at functions that have more
 760than 3 lines of code in them. An exception to this rule are the cases where
 761a parameter is known to be a compiletime constant, and as a result of this
 762constantness you *know* the compiler will be able to optimize most of your
 763function away at compile time. For a good example of this later case, see
 764the kmalloc() inline function.
 766Often people argue that adding inline to functions that are static and used
 767only once is always a win since there is no space tradeoff. While this is
 768technically correct, gcc is capable of inlining these automatically without
 769help, and the maintenance issue of removing the inline when a second user
 770appears outweighs the potential value of the hint that tells gcc to do
 771something it would have done anyway.
 774                Chapter 16: Function return values and names
 776Functions can return values of many different kinds, and one of the
 777most common is a value indicating whether the function succeeded or
 778failed.  Such a value can be represented as an error-code integer
 779(-Exxx = failure, 0 = success) or a "succeeded" boolean (0 = failure,
 780non-zero = success).
 782Mixing up these two sorts of representations is a fertile source of
 783difficult-to-find bugs.  If the C language included a strong distinction
 784between integers and booleans then the compiler would find these mistakes
 785for us... but it doesn't.  To help prevent such bugs, always follow this
 788        If the name of a function is an action or an imperative command,
 789        the function should return an error-code integer.  If the name
 790        is a predicate, the function should return a "succeeded" boolean.
 792For example, "add work" is a command, and the add_work() function returns 0
 793for success or -EBUSY for failure.  In the same way, "PCI device present" is
 794a predicate, and the pci_dev_present() function returns 1 if it succeeds in
 795finding a matching device or 0 if it doesn't.
 797All EXPORTed functions must respect this convention, and so should all
 798public functions.  Private (static) functions need not, but it is
 799recommended that they do.
 801Functions whose return value is the actual result of a computation, rather
 802than an indication of whether the computation succeeded, are not subject to
 803this rule.  Generally they indicate failure by returning some out-of-range
 804result.  Typical examples would be functions that return pointers; they use
 805NULL or the ERR_PTR mechanism to report failure.
 808                Chapter 17:  Don't re-invent the kernel macros
 810The header file include/linux/kernel.h contains a number of macros that
 811you should use, rather than explicitly coding some variant of them yourself.
 812For example, if you need to calculate the length of an array, take advantage
 813of the macro
 815        #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 817Similarly, if you need to calculate the size of some structure member, use
 819        #define FIELD_SIZEOF(t, f) (sizeof(((t*)0)->f))
 821There are also min() and max() macros that do strict type checking if you
 822need them.  Feel free to peruse that header file to see what else is already
 823defined that you shouldn't reproduce in your code.
 826                Chapter 18:  Editor modelines and other cruft
 828Some editors can interpret configuration information embedded in source files,
 829indicated with special markers.  For example, emacs interprets lines marked
 830like this:
 832        -*- mode: c -*-
 834Or like this:
 836        /*
 837        Local Variables:
 838        compile-command: "gcc -DMAGIC_DEBUG_FLAG foo.c"
 839        End:
 840        */
 842Vim interprets markers that look like this:
 844        /* vim:set sw=8 noet */
 846Do not include any of these in source files.  People have their own personal
 847editor configurations, and your source files should not override them.  This
 848includes markers for indentation and mode configuration.  People may use their
 849own custom mode, or may have some other magic method for making indentation
 850work correctly.
 853                Chapter 19:  Inline assembly
 855In architecture-specific code, you may need to use inline assembly to interface
 856with CPU or platform functionality.  Don't hesitate to do so when necessary.
 857However, don't use inline assembly gratuitously when C can do the job.  You can
 858and should poke hardware from C when possible.
 860Consider writing simple helper functions that wrap common bits of inline
 861assembly, rather than repeatedly writing them with slight variations.  Remember
 862that inline assembly can use C parameters.
 864Large, non-trivial assembly functions should go in .S files, with corresponding
 865C prototypes defined in C header files.  The C prototypes for assembly
 866functions should use "asmlinkage".
 868You may need to mark your asm statement as volatile, to prevent GCC from
 869removing it if GCC doesn't notice any side effects.  You don't always need to
 870do so, though, and doing so unnecessarily can limit optimization.
 872When writing a single inline assembly statement containing multiple
 873instructions, put each instruction on a separate line in a separate quoted
 874string, and end each string except the last with \n\t to properly indent the
 875next instruction in the assembly output:
 877        asm ("magic %reg1, #42\n\t"
 878             "more_magic %reg2, %reg3"
 879             : /* outputs */ : /* inputs */ : /* clobbers */);
 882                Chapter 20: Conditional Compilation
 884Wherever possible, don't use preprocessor conditionals (#if, #ifdef) in .c
 885files; doing so makes code harder to read and logic harder to follow.  Instead,
 886use such conditionals in a header file defining functions for use in those .c
 887files, providing no-op stub versions in the #else case, and then call those
 888functions unconditionally from .c files.  The compiler will avoid generating
 889any code for the stub calls, producing identical results, but the logic will
 890remain easy to follow.
 892Prefer to compile out entire functions, rather than portions of functions or
 893portions of expressions.  Rather than putting an ifdef in an expression, factor
 894out part or all of the expression into a separate helper function and apply the
 895conditional to that function.
 897If you have a function or variable which may potentially go unused in a
 898particular configuration, and the compiler would warn about its definition
 899going unused, mark the definition as __maybe_unused rather than wrapping it in
 900a preprocessor conditional.  (However, if a function or variable *always* goes
 901unused, delete it.)
 903Within code, where possible, use the IS_ENABLED macro to convert a Kconfig
 904symbol into a C boolean expression, and use it in a normal C conditional:
 907                ...
 908        }
 910The compiler will constant-fold the conditional away, and include or exclude
 911the block of code just as with an #ifdef, so this will not add any runtime
 912overhead.  However, this approach still allows the C compiler to see the code
 913inside the block, and check it for correctness (syntax, types, symbol
 914references, etc).  Thus, you still have to use an #ifdef if the code inside the
 915block references symbols that will not exist if the condition is not met.
 917At the end of any non-trivial #if or #ifdef block (more than a few lines),
 918place a comment after the #endif on the same line, noting the conditional
 919expression used.  For instance:
 921        #ifdef CONFIG_SOMETHING
 922        ...
 923        #endif /* CONFIG_SOMETHING */
 926                Appendix I: References
 928The C Programming Language, Second Edition
 929by Brian W. Kernighan and Dennis M. Ritchie.
 930Prentice Hall, Inc., 1988.
 931ISBN 0-13-110362-8 (paperback), 0-13-110370-9 (hardback).
 933The Practice of Programming
 934by Brian W. Kernighan and Rob Pike.
 935Addison-Wesley, Inc., 1999.
 936ISBN 0-201-61586-X.
 938GNU manuals - where in compliance with K&R and this text - for cpp, gcc,
 939gcc internals and indent, all available from
 941WG14 is the international standardization working group for the programming
 942language C, URL:
 944Kernel CodingStyle, by at OLS 2002:
 947 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.