linux/Documentation/CodingStyle
<<
>>
Prefs
   1
   2                Linux kernel coding style
   3
   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.
   9
  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.
  12
  13Anyway, here goes:
  14
  15
  16                Chapter 1: Indentation
  17
  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.
  22
  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.
  27
  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.
  33
  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.
  37
  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.:
  41
  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        }
  58
  59
  60Don't put multiple statements on a single line unless you have
  61something to hide:
  62
  63        if (condition) do_this;
  64          do_something_everytime;
  65
  66Don't put multiple assignments on a single line either.  Kernel coding style
  67is super simple.  Avoid tricky expressions.
  68
  69Outside of comments, documentation and except in Kconfig, spaces are never
  70used for indentation, and the above example is deliberately broken.
  71
  72Get a decent editor and don't leave whitespace at the end of lines.
  73
  74
  75                Chapter 2: Breaking long lines and strings
  76
  77Coding style is all about readability and maintainability using commonly
  78available tools.
  79
  80The limit on the length of lines is 80 columns and this is a strongly
  81preferred limit.
  82
  83Statements longer than 80 columns will be broken into sensible chunks, unless
  84exceeding 80 columns significantly increases readability and does not hide
  85information. Descendants are always substantially shorter than the parent and
  86are placed substantially to the right. The same applies to function headers
  87with a long argument list. However, never break user-visible strings such as
  88printk messages, because that breaks the ability to grep for them.
  89
  90
  91                Chapter 3: Placing Braces and Spaces
  92
  93The other issue that always comes up in C styling is the placement of
  94braces.  Unlike the indent size, there are few technical reasons to
  95choose one placement strategy over the other, but the preferred way, as
  96shown to us by the prophets Kernighan and Ritchie, is to put the opening
  97brace last on the line, and put the closing brace first, thusly:
  98
  99        if (x is true) {
 100                we do y
 101        }
 102
 103This applies to all non-function statement blocks (if, switch, for,
 104while, do).  E.g.:
 105
 106        switch (action) {
 107        case KOBJ_ADD:
 108                return "add";
 109        case KOBJ_REMOVE:
 110                return "remove";
 111        case KOBJ_CHANGE:
 112                return "change";
 113        default:
 114                return NULL;
 115        }
 116
 117However, there is one special case, namely functions: they have the
 118opening brace at the beginning of the next line, thus:
 119
 120        int function(int x)
 121        {
 122                body of function
 123        }
 124
 125Heretic people all over the world have claimed that this inconsistency
 126is ...  well ...  inconsistent, but all right-thinking people know that
 127(a) K&R are _right_ and (b) K&R are right.  Besides, functions are
 128special anyway (you can't nest them in C).
 129
 130Note that the closing brace is empty on a line of its own, _except_ in
 131the cases where it is followed by a continuation of the same statement,
 132ie a "while" in a do-statement or an "else" in an if-statement, like
 133this:
 134
 135        do {
 136                body of do-loop
 137        } while (condition);
 138
 139and
 140
 141        if (x == y) {
 142                ..
 143        } else if (x > y) {
 144                ...
 145        } else {
 146                ....
 147        }
 148
 149Rationale: K&R.
 150
 151Also, note that this brace-placement also minimizes the number of empty
 152(or almost empty) lines, without any loss of readability.  Thus, as the
 153supply of new-lines on your screen is not a renewable resource (think
 15425-line terminal screens here), you have more empty lines to put
 155comments on.
 156
 157Do not unnecessarily use braces where a single statement will do.
 158
 159if (condition)
 160        action();
 161
 162and
 163
 164if (condition)
 165        do_this();
 166else
 167        do_that();
 168
 169This does not apply if only one branch of a conditional statement is a single
 170statement; in the latter case use braces in both branches:
 171
 172if (condition) {
 173        do_this();
 174        do_that();
 175} else {
 176        otherwise();
 177}
 178
 179                3.1:  Spaces
 180
 181Linux kernel style for use of spaces depends (mostly) on
 182function-versus-keyword usage.  Use a space after (most) keywords.  The
 183notable exceptions are sizeof, typeof, alignof, and __attribute__, which look
 184somewhat like functions (and are usually used with parentheses in Linux,
 185although they are not required in the language, as in: "sizeof info" after
 186"struct fileinfo info;" is declared).
 187
 188So use a space after these keywords:
 189        if, switch, case, for, do, while
 190but not with sizeof, typeof, alignof, or __attribute__.  E.g.,
 191        s = sizeof(struct file);
 192
 193Do not add spaces around (inside) parenthesized expressions.  This example is
 194*bad*:
 195
 196        s = sizeof( struct file );
 197
 198When declaring pointer data or a function that returns a pointer type, the
 199preferred use of '*' is adjacent to the data name or function name and not
 200adjacent to the type name.  Examples:
 201
 202        char *linux_banner;
 203        unsigned long long memparse(char *ptr, char **retptr);
 204        char *match_strdup(substring_t *s);
 205
 206Use one space around (on each side of) most binary and ternary operators,
 207such as any of these:
 208
 209        =  +  -  <  >  *  /  %  |  &  ^  <=  >=  ==  !=  ?  :
 210
 211but no space after unary operators:
 212        &  *  +  -  ~  !  sizeof  typeof  alignof  __attribute__  defined
 213
 214no space before the postfix increment & decrement unary operators:
 215        ++  --
 216
 217no space after the prefix increment & decrement unary operators:
 218        ++  --
 219
 220and no space around the '.' and "->" structure member operators.
 221
 222Do 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
 224appropriate, so you can start typing the next line of code right away.
 225However, some such editors do not remove the whitespace if you end up not
 226putting a line of code there, such as if you leave a blank line.  As a result,
 227you end up with lines containing trailing whitespace.
 228
 229Git will warn you about patches that introduce trailing whitespace, and can
 230optionally strip the trailing whitespace for you; however, if applying a series
 231of patches, this may make later patches in the series fail by changing their
 232context lines.
 233
 234
 235                Chapter 4: Naming
 236
 237C is a Spartan language, and so should your naming be.  Unlike Modula-2
 238and Pascal programmers, C programmers do not use cute names like
 239ThisVariableIsATemporaryCounter.  A C programmer would call that
 240variable "tmp", which is much easier to write, and not the least more
 241difficult to understand.
 242
 243HOWEVER, while mixed-case names are frowned upon, descriptive names for
 244global variables are a must.  To call a global function "foo" is a
 245shooting offense.
 246
 247GLOBAL variables (to be used only if you _really_ need them) need to
 248have descriptive names, as do global functions.  If you have a function
 249that counts the number of active users, you should call that
 250"count_active_users()" or similar, you should _not_ call it "cntusr()".
 251
 252Encoding the type of a function into the name (so-called Hungarian
 253notation) is brain damaged - the compiler knows the types anyway and can
 254check those, and it only confuses the programmer.  No wonder MicroSoft
 255makes buggy programs.
 256
 257LOCAL variable names should be short, and to the point.  If you have
 258some random integer loop counter, it should probably be called "i".
 259Calling it "loop_counter" is non-productive, if there is no chance of it
 260being mis-understood.  Similarly, "tmp" can be just about any type of
 261variable that is used to hold a temporary value.
 262
 263If you are afraid to mix up your local variable names, you have another
 264problem, which is called the function-growth-hormone-imbalance syndrome.
 265See chapter 6 (Functions).
 266
 267
 268                Chapter 5: Typedefs
 269
 270Please don't use things like "vps_t".
 271
 272It's a _mistake_ to use typedef for structures and pointers. When you see a
 273
 274        vps_t a;
 275
 276in the source, what does it mean?
 277
 278In contrast, if it says
 279
 280        struct virtual_container *a;
 281
 282you can actually tell what "a" is.
 283
 284Lots of people think that typedefs "help readability". Not so. They are
 285useful only for:
 286
 287 (a) totally opaque objects (where the typedef is actively used to _hide_
 288     what the object is).
 289
 290     Example: "pte_t" etc. opaque objects that you can only access using
 291     the proper accessor functions.
 292
 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.
 296
 297 (b) Clear integer types, where the abstraction _helps_ avoid confusion
 298     whether it is "int" or "long".
 299
 300     u8/u16/u32 are perfectly fine typedefs, although they fit into
 301     category (d) better than here.
 302
 303     NOTE! Again - there needs to be a _reason_ for this. If something is
 304     "unsigned long", then there's no reason to do
 305
 306        typedef unsigned long myflags_t;
 307
 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.
 311
 312 (c) when you use sparse to literally create a _new_ type for
 313     type-checking.
 314
 315 (d) New types which are identical to standard C99 types, in certain
 316     exceptional circumstances.
 317
 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.
 321
 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
 325     own.
 326
 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.
 329
 330 (e) Types safe for use in userspace.
 331
 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
 335     with userspace.
 336
 337Maybe there are other cases too, but the rule should basically be to NEVER
 338EVER use a typedef unless you can clearly match one of those rules.
 339
 340In general, a pointer, or a struct that has elements that can reasonably
 341be directly accessed should _never_ be a typedef.
 342
 343
 344                Chapter 6: Functions
 345
 346Functions should be short and sweet, and do just one thing.  They should
 347fit on one or two screenfuls of text (the ISO/ANSI screen size is 80x24,
 348as we all know), and do one thing and do that well.
 349
 350The maximum length of a function is inversely proportional to the
 351complexity and indentation level of that function.  So, if you have a
 352conceptually simple function that is just one long (but simple)
 353case-statement, where you have to do lots of small things for a lot of
 354different cases, it's OK to have a longer function.
 355
 356However, if you have a complex function, and you suspect that a
 357less-than-gifted first-year high-school student might not even
 358understand what the function is all about, you should adhere to the
 359maximum limits all the more closely.  Use helper functions with
 360descriptive names (you can ask the compiler to in-line them if you think
 361it's performance-critical, and it will probably do a better job of it
 362than you would have done).
 363
 364Another measure of the function is the number of local variables.  They
 365shouldn't exceed 5-10, or you're doing something wrong.  Re-think the
 366function, and split it into smaller pieces.  A human brain can
 367generally easily keep track of about 7 different things, anything more
 368and it gets confused.  You know you're brilliant, but maybe you'd like
 369to understand what you did 2 weeks from now.
 370
 371In source files, separate functions with one blank line.  If the function is
 372exported, the EXPORT* macro for it should follow immediately after the closing
 373function brace line.  E.g.:
 374
 375int system_is_up(void)
 376{
 377        return system_state == SYSTEM_RUNNING;
 378}
 379EXPORT_SYMBOL(system_is_up);
 380
 381In function prototypes, include parameter names with their data types.
 382Although this is not required by the C language, it is preferred in Linux
 383because it is a simple way to add valuable information for the reader.
 384
 385
 386                Chapter 7: Centralized exiting of functions
 387
 388Albeit deprecated by some people, the equivalent of the goto statement is
 389used frequently by compilers in form of the unconditional jump instruction.
 390
 391The goto statement comes in handy when a function exits from multiple
 392locations and some common work such as cleanup has to be done.
 393
 394The rationale is:
 395
 396- unconditional statements are easier to understand and follow
 397- nesting is reduced
 398- errors by not updating individual exit points when making
 399    modifications are prevented
 400- saves the compiler work to optimize redundant code away ;)
 401
 402int fun(int a)
 403{
 404        int result = 0;
 405        char *buffer = kmalloc(SIZE);
 406
 407        if (buffer == NULL)
 408                return -ENOMEM;
 409
 410        if (condition1) {
 411                while (loop1) {
 412                        ...
 413                }
 414                result = 1;
 415                goto out;
 416        }
 417        ...
 418out:
 419        kfree(buffer);
 420        return result;
 421}
 422
 423                Chapter 8: Commenting
 424
 425Comments are good, but there is also a danger of over-commenting.  NEVER
 426try to explain HOW your code works in a comment: it's much better to
 427write the code so that the _working_ is obvious, and it's a waste of
 428time to explain badly written code.
 429
 430Generally, you want your comments to tell WHAT your code does, not HOW.
 431Also, try to avoid putting comments inside a function body: if the
 432function is so complex that you need to separately comment parts of it,
 433you should probably go back to chapter 6 for a while.  You can make
 434small comments to note or warn about something particularly clever (or
 435ugly), but try to avoid excess.  Instead, put the comments at the head
 436of the function, telling people what it does, and possibly WHY it does
 437it.
 438
 439When commenting the kernel API functions, please use the kernel-doc format.
 440See the files Documentation/kernel-doc-nano-HOWTO.txt and scripts/kernel-doc
 441for details.
 442
 443Linux style for comments is the C89 "/* ... */" style.
 444Don't use C99-style "// ..." comments.
 445
 446The preferred style for long (multi-line) comments is:
 447
 448        /*
 449         * This is the preferred style for multi-line
 450         * comments in the Linux kernel source code.
 451         * Please use it consistently.
 452         *
 453         * Description:  A column of asterisks on the left side,
 454         * with beginning and ending almost-blank lines.
 455         */
 456
 457For files in net/ and drivers/net/ the preferred style for long (multi-line)
 458comments is a little different.
 459
 460        /* The preferred comment style for files in net/ and drivers/net
 461         * looks like this.
 462         *
 463         * It is nearly the same as the generally preferred comment style,
 464         * but there is no initial almost-blank line.
 465         */
 466
 467It's also important to comment data, whether they are basic types or derived
 468types.  To this end, use just one data declaration per line (no commas for
 469multiple data declarations).  This leaves you room for a small comment on each
 470item, explaining its use.
 471
 472
 473                Chapter 9: You've made a mess of it
 474
 475That's OK, we all do.  You've probably been told by your long-time Unix
 476user helper that "GNU emacs" automatically formats the C sources for
 477you, and you've noticed that yes, it does do that, but the defaults it
 478uses are less than desirable (in fact, they are worse than random
 479typing - an infinite number of monkeys typing into GNU emacs would never
 480make a good program).
 481
 482So, you can either get rid of GNU emacs, or change it to use saner
 483values.  To do the latter, you can stick the following in your .emacs file:
 484
 485(defun c-lineup-arglist-tabs-only (ignored)
 486  "Line up argument lists by tabs, not spaces"
 487  (let* ((anchor (c-langelem-pos c-syntactic-element))
 488         (column (c-langelem-2nd-pos c-syntactic-element))
 489         (offset (- (1+ column) anchor))
 490         (steps (floor offset c-basic-offset)))
 491    (* (max steps 1)
 492       c-basic-offset)))
 493
 494(add-hook 'c-mode-common-hook
 495          (lambda ()
 496            ;; Add kernel style
 497            (c-add-style
 498             "linux-tabs-only"
 499             '("linux" (c-offsets-alist
 500                        (arglist-cont-nonempty
 501                         c-lineup-gcc-asm-reg
 502                         c-lineup-arglist-tabs-only))))))
 503
 504(add-hook 'c-mode-hook
 505          (lambda ()
 506            (let ((filename (buffer-file-name)))
 507              ;; Enable kernel mode for the appropriate files
 508              (when (and filename
 509                         (string-match (expand-file-name "~/src/linux-trees")
 510                                       filename))
 511                (setq indent-tabs-mode t)
 512                (c-set-style "linux-tabs-only")))))
 513
 514This will make emacs go better with the kernel coding style for C
 515files below ~/src/linux-trees.
 516
 517But even if you fail in getting emacs to do sane formatting, not
 518everything is lost: use "indent".
 519
 520Now, again, GNU indent has the same brain-dead settings that GNU emacs
 521has, which is why you need to give it a few command line options.
 522However, that's not too bad, because even the makers of GNU indent
 523recognize the authority of K&R (the GNU people aren't evil, they are
 524just severely misguided in this matter), so you just give indent the
 525options "-kr -i8" (stands for "K&R, 8 character indents"), or use
 526"scripts/Lindent", which indents in the latest style.
 527
 528"indent" has a lot of options, and especially when it comes to comment
 529re-formatting you may want to take a look at the man page.  But
 530remember: "indent" is not a fix for bad programming.
 531
 532
 533                Chapter 10: Kconfig configuration files
 534
 535For all of the Kconfig* configuration files throughout the source tree,
 536the indentation is somewhat different.  Lines under a "config" definition
 537are indented with one tab, while help text is indented an additional two
 538spaces.  Example:
 539
 540config AUDIT
 541        bool "Auditing support"
 542        depends on NET
 543        help
 544          Enable auditing infrastructure that can be used with another
 545          kernel subsystem, such as SELinux (which requires this for
 546          logging of avc messages output).  Does not do system-call
 547          auditing without CONFIG_AUDITSYSCALL.
 548
 549Seriously dangerous features (such as write support for certain
 550filesystems) should advertise this prominently in their prompt string:
 551
 552config ADFS_FS_RW
 553        bool "ADFS write support (DANGEROUS)"
 554        depends on ADFS_FS
 555        ...
 556
 557For full documentation on the configuration files, see the file
 558Documentation/kbuild/kconfig-language.txt.
 559
 560
 561                Chapter 11: Data structures
 562
 563Data structures that have visibility outside the single-threaded
 564environment they are created and destroyed in should always have
 565reference counts.  In the kernel, garbage collection doesn't exist (and
 566outside the kernel garbage collection is slow and inefficient), which
 567means that you absolutely _have_ to reference count all your uses.
 568
 569Reference counting means that you can avoid locking, and allows multiple
 570users to have access to the data structure in parallel - and not having
 571to worry about the structure suddenly going away from under them just
 572because they slept or did something else for a while.
 573
 574Note that locking is _not_ a replacement for reference counting.
 575Locking is used to keep data structures coherent, while reference
 576counting is a memory management technique.  Usually both are needed, and
 577they are not to be confused with each other.
 578
 579Many data structures can indeed have two levels of reference counting,
 580when there are users of different "classes".  The subclass count counts
 581the number of subclass users, and decrements the global count just once
 582when the subclass count goes to zero.
 583
 584Examples of this kind of "multi-level-reference-counting" can be found in
 585memory management ("struct mm_struct": mm_users and mm_count), and in
 586filesystem code ("struct super_block": s_count and s_active).
 587
 588Remember: if another thread can find your data structure, and you don't
 589have a reference count on it, you almost certainly have a bug.
 590
 591
 592                Chapter 12: Macros, Enums and RTL
 593
 594Names of macros defining constants and labels in enums are capitalized.
 595
 596#define CONSTANT 0x12345
 597
 598Enums are preferred when defining several related constants.
 599
 600CAPITALIZED macro names are appreciated but macros resembling functions
 601may be named in lower case.
 602
 603Generally, inline functions are preferable to macros resembling functions.
 604
 605Macros with multiple statements should be enclosed in a do - while block:
 606
 607#define macrofun(a, b, c)                       \
 608        do {                                    \
 609                if (a == 5)                     \
 610                        do_this(b, c);          \
 611        } while (0)
 612
 613Things to avoid when using macros:
 614
 6151) macros that affect control flow:
 616
 617#define FOO(x)                                  \
 618        do {                                    \
 619                if (blah(x) < 0)                \
 620                        return -EBUGGERED;      \
 621        } while(0)
 622
 623is a _very_ bad idea.  It looks like a function call but exits the "calling"
 624function; don't break the internal parsers of those who will read the code.
 625
 6262) macros that depend on having a local variable with a magic name:
 627
 628#define FOO(val) bar(index, val)
 629
 630might look like a good thing, but it's confusing as hell when one reads the
 631code and it's prone to breakage from seemingly innocent changes.
 632
 6333) macros with arguments that are used as l-values: FOO(x) = y; will
 634bite you if somebody e.g. turns FOO into an inline function.
 635
 6364) forgetting about precedence: macros defining constants using expressions
 637must enclose the expression in parentheses. Beware of similar issues with
 638macros using parameters.
 639
 640#define CONSTANT 0x4000
 641#define CONSTEXP (CONSTANT | 3)
 642
 643The cpp manual deals with macros exhaustively. The gcc internals manual also
 644covers RTL which is used frequently with assembly language in the kernel.
 645
 646
 647                Chapter 13: Printing kernel messages
 648
 649Kernel developers like to be seen as literate. Do mind the spelling
 650of kernel messages to make a good impression. Do not use crippled
 651words like "dont"; use "do not" or "don't" instead.  Make the messages
 652concise, clear, and unambiguous.
 653
 654Kernel messages do not have to be terminated with a period.
 655
 656Printing numbers in parentheses (%d) adds no value and should be avoided.
 657
 658There are a number of driver model diagnostic macros in <linux/device.h>
 659which you should use to make sure messages are matched to the right device
 660and driver, and are tagged with the right level:  dev_err(), dev_warn(),
 661dev_info(), and so forth.  For messages that aren't associated with a
 662particular device, <linux/printk.h> defines pr_debug() and pr_info().
 663
 664Coming up with good debugging messages can be quite a challenge; and once
 665you have them, they can be a huge help for remote troubleshooting.  Such
 666messages should be compiled out when the DEBUG symbol is not defined (that
 667is, by default they are not included).  When you use dev_dbg() or pr_debug(),
 668that's automatic.  Many subsystems have Kconfig options to turn on -DDEBUG.
 669A related convention uses VERBOSE_DEBUG to add dev_vdbg() messages to the
 670ones already enabled by DEBUG.
 671
 672
 673                Chapter 14: Allocating memory
 674
 675The kernel provides the following general purpose memory allocators:
 676kmalloc(), kzalloc(), kmalloc_array(), kcalloc(), vmalloc(), and
 677vzalloc().  Please refer to the API documentation for further information
 678about them.
 679
 680The preferred form for passing a size of a struct is the following:
 681
 682        p = kmalloc(sizeof(*p), ...);
 683
 684The alternative form where struct name is spelled out hurts readability and
 685introduces an opportunity for a bug when the pointer variable type is changed
 686but the corresponding sizeof that is passed to a memory allocator is not.
 687
 688Casting the return value which is a void pointer is redundant. The conversion
 689from void pointer to any other pointer type is guaranteed by the C programming
 690language.
 691
 692The preferred form for allocating an array is the following:
 693
 694        p = kmalloc_array(n, sizeof(...), ...);
 695
 696The preferred form for allocating a zeroed array is the following:
 697
 698        p = kcalloc(n, sizeof(...), ...);
 699
 700Both forms check for overflow on the allocation size n * sizeof(...),
 701and return NULL if that occurred.
 702
 703
 704                Chapter 15: The inline disease
 705
 706There appears to be a common misperception that gcc has a magic "make me
 707faster" speedup option called "inline". While the use of inlines can be
 708appropriate (for example as a means of replacing macros, see Chapter 12), it
 709very often is not. Abundant use of the inline keyword leads to a much bigger
 710kernel, which in turn slows the system as a whole down, due to a bigger
 711icache footprint for the CPU and simply because there is less memory
 712available for the pagecache. Just think about it; a pagecache miss causes a
 713disk seek, which easily takes 5 milliseconds. There are a LOT of cpu cycles
 714that can go into these 5 milliseconds.
 715
 716A reasonable rule of thumb is to not put inline at functions that have more
 717than 3 lines of code in them. An exception to this rule are the cases where
 718a parameter is known to be a compiletime constant, and as a result of this
 719constantness you *know* the compiler will be able to optimize most of your
 720function away at compile time. For a good example of this later case, see
 721the kmalloc() inline function.
 722
 723Often people argue that adding inline to functions that are static and used
 724only once is always a win since there is no space tradeoff. While this is
 725technically correct, gcc is capable of inlining these automatically without
 726help, and the maintenance issue of removing the inline when a second user
 727appears outweighs the potential value of the hint that tells gcc to do
 728something it would have done anyway.
 729
 730
 731                Chapter 16: Function return values and names
 732
 733Functions can return values of many different kinds, and one of the
 734most common is a value indicating whether the function succeeded or
 735failed.  Such a value can be represented as an error-code integer
 736(-Exxx = failure, 0 = success) or a "succeeded" boolean (0 = failure,
 737non-zero = success).
 738
 739Mixing up these two sorts of representations is a fertile source of
 740difficult-to-find bugs.  If the C language included a strong distinction
 741between integers and booleans then the compiler would find these mistakes
 742for us... but it doesn't.  To help prevent such bugs, always follow this
 743convention:
 744
 745        If the name of a function is an action or an imperative command,
 746        the function should return an error-code integer.  If the name
 747        is a predicate, the function should return a "succeeded" boolean.
 748
 749For example, "add work" is a command, and the add_work() function returns 0
 750for success or -EBUSY for failure.  In the same way, "PCI device present" is
 751a predicate, and the pci_dev_present() function returns 1 if it succeeds in
 752finding a matching device or 0 if it doesn't.
 753
 754All EXPORTed functions must respect this convention, and so should all
 755public functions.  Private (static) functions need not, but it is
 756recommended that they do.
 757
 758Functions whose return value is the actual result of a computation, rather
 759than an indication of whether the computation succeeded, are not subject to
 760this rule.  Generally they indicate failure by returning some out-of-range
 761result.  Typical examples would be functions that return pointers; they use
 762NULL or the ERR_PTR mechanism to report failure.
 763
 764
 765                Chapter 17:  Don't re-invent the kernel macros
 766
 767The header file include/linux/kernel.h contains a number of macros that
 768you should use, rather than explicitly coding some variant of them yourself.
 769For example, if you need to calculate the length of an array, take advantage
 770of the macro
 771
 772  #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 773
 774Similarly, if you need to calculate the size of some structure member, use
 775
 776  #define FIELD_SIZEOF(t, f) (sizeof(((t*)0)->f))
 777
 778There are also min() and max() macros that do strict type checking if you
 779need them.  Feel free to peruse that header file to see what else is already
 780defined that you shouldn't reproduce in your code.
 781
 782
 783                Chapter 18:  Editor modelines and other cruft
 784
 785Some editors can interpret configuration information embedded in source files,
 786indicated with special markers.  For example, emacs interprets lines marked
 787like this:
 788
 789-*- mode: c -*-
 790
 791Or like this:
 792
 793/*
 794Local Variables:
 795compile-command: "gcc -DMAGIC_DEBUG_FLAG foo.c"
 796End:
 797*/
 798
 799Vim interprets markers that look like this:
 800
 801/* vim:set sw=8 noet */
 802
 803Do not include any of these in source files.  People have their own personal
 804editor configurations, and your source files should not override them.  This
 805includes markers for indentation and mode configuration.  People may use their
 806own custom mode, or may have some other magic method for making indentation
 807work correctly.
 808
 809
 810                Chapter 19:  Inline assembly
 811
 812In architecture-specific code, you may need to use inline assembly to interface
 813with CPU or platform functionality.  Don't hesitate to do so when necessary.
 814However, don't use inline assembly gratuitously when C can do the job.  You can
 815and should poke hardware from C when possible.
 816
 817Consider writing simple helper functions that wrap common bits of inline
 818assembly, rather than repeatedly writing them with slight variations.  Remember
 819that inline assembly can use C parameters.
 820
 821Large, non-trivial assembly functions should go in .S files, with corresponding
 822C prototypes defined in C header files.  The C prototypes for assembly
 823functions should use "asmlinkage".
 824
 825You may need to mark your asm statement as volatile, to prevent GCC from
 826removing it if GCC doesn't notice any side effects.  You don't always need to
 827do so, though, and doing so unnecessarily can limit optimization.
 828
 829When writing a single inline assembly statement containing multiple
 830instructions, put each instruction on a separate line in a separate quoted
 831string, and end each string except the last with \n\t to properly indent the
 832next instruction in the assembly output:
 833
 834        asm ("magic %reg1, #42\n\t"
 835             "more_magic %reg2, %reg3"
 836             : /* outputs */ : /* inputs */ : /* clobbers */);
 837
 838
 839
 840                Appendix I: References
 841
 842The C Programming Language, Second Edition
 843by Brian W. Kernighan and Dennis M. Ritchie.
 844Prentice Hall, Inc., 1988.
 845ISBN 0-13-110362-8 (paperback), 0-13-110370-9 (hardback).
 846URL: http://cm.bell-labs.com/cm/cs/cbook/
 847
 848The Practice of Programming
 849by Brian W. Kernighan and Rob Pike.
 850Addison-Wesley, Inc., 1999.
 851ISBN 0-201-61586-X.
 852URL: http://cm.bell-labs.com/cm/cs/tpop/
 853
 854GNU manuals - where in compliance with K&R and this text - for cpp, gcc,
 855gcc internals and indent, all available from http://www.gnu.org/manual/
 856
 857WG14 is the international standardization working group for the programming
 858language C, URL: http://www.open-std.org/JTC1/SC22/WG14/
 859
 860Kernel CodingStyle, by greg@kroah.com at OLS 2002:
 861http://www.kroah.com/linux/talks/ols_2002_kernel_codingstyle_talk/html/
 862
 863
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.