]> git.proxmox.com Git - mirror_qemu.git/blob - CODING_STYLE.rst
qdev: Move property code to qdev-properties.[ch]
[mirror_qemu.git] / CODING_STYLE.rst
1 =================
2 QEMU Coding Style
3 =================
4
5 .. contents:: Table of Contents
6
7 Please use the script checkpatch.pl in the scripts directory to check
8 patches before submitting.
9
10 Formatting and style
11 ********************
12
13 Whitespace
14 ==========
15
16 Of course, the most important aspect in any coding style is whitespace.
17 Crusty old coders who have trouble spotting the glasses on their noses
18 can tell the difference between a tab and eight spaces from a distance
19 of approximately fifteen parsecs. Many a flamewar has been fought and
20 lost on this issue.
21
22 QEMU indents are four spaces. Tabs are never used, except in Makefiles
23 where they have been irreversibly coded into the syntax.
24 Spaces of course are superior to tabs because:
25
26 * You have just one way to specify whitespace, not two. Ambiguity breeds
27 mistakes.
28 * The confusion surrounding 'use tabs to indent, spaces to justify' is gone.
29 * Tab indents push your code to the right, making your screen seriously
30 unbalanced.
31 * Tabs will be rendered incorrectly on editors who are misconfigured not
32 to use tab stops of eight positions.
33 * Tabs are rendered badly in patches, causing off-by-one errors in almost
34 every line.
35 * It is the QEMU coding style.
36
37 Do not leave whitespace dangling off the ends of lines.
38
39 Multiline Indent
40 ----------------
41
42 There are several places where indent is necessary:
43
44 * if/else
45 * while/for
46 * function definition & call
47
48 When breaking up a long line to fit within line width, we need a proper indent
49 for the following lines.
50
51 In case of if/else, while/for, align the secondary lines just after the
52 opening parenthesis of the first.
53
54 For example:
55
56 .. code-block:: c
57
58 if (a == 1 &&
59 b == 2) {
60
61 while (a == 1 &&
62 b == 2) {
63
64 In case of function, there are several variants:
65
66 * 4 spaces indent from the beginning
67 * align the secondary lines just after the opening parenthesis of the first
68
69 For example:
70
71 .. code-block:: c
72
73 do_something(x, y,
74 z);
75
76 do_something(x, y,
77 z);
78
79 do_something(x, do_another(y,
80 z));
81
82 Line width
83 ==========
84
85 Lines should be 80 characters; try not to make them longer.
86
87 Sometimes it is hard to do, especially when dealing with QEMU subsystems
88 that use long function or symbol names. Even in that case, do not make
89 lines much longer than 80 characters.
90
91 Rationale:
92
93 * Some people like to tile their 24" screens with a 6x4 matrix of 80x24
94 xterms and use vi in all of them. The best way to punish them is to
95 let them keep doing it.
96 * Code and especially patches is much more readable if limited to a sane
97 line length. Eighty is traditional.
98 * The four-space indentation makes the most common excuse ("But look
99 at all that white space on the left!") moot.
100 * It is the QEMU coding style.
101
102 Naming
103 ======
104
105 Variables are lower_case_with_underscores; easy to type and read. Structured
106 type names are in CamelCase; harder to type but standing out. Enum type
107 names and function type names should also be in CamelCase. Scalar type
108 names are lower_case_with_underscores_ending_with_a_t, like the POSIX
109 uint64_t and family. Note that this last convention contradicts POSIX
110 and is therefore likely to be changed.
111
112 Variable Naming Conventions
113 ---------------------------
114
115 A number of short naming conventions exist for variables that use
116 common QEMU types. For example, the architecture independent CPUState
117 is often held as a ``cs`` pointer variable, whereas the concrete
118 CPUArchState is usually held in a pointer called ``env``.
119
120 Likewise, in device emulation code the common DeviceState is usually
121 called ``dev``.
122
123 Function Naming Conventions
124 ---------------------------
125
126 Wrapped version of standard library or GLib functions use a ``qemu_``
127 prefix to alert readers that they are seeing a wrapped version, for
128 example ``qemu_strtol`` or ``qemu_mutex_lock``. Other utility functions
129 that are widely called from across the codebase should not have any
130 prefix, for example ``pstrcpy`` or bit manipulation functions such as
131 ``find_first_bit``.
132
133 The ``qemu_`` prefix is also used for functions that modify global
134 emulator state, for example ``qemu_add_vm_change_state_handler``.
135 However, if there is an obvious subsystem-specific prefix it should be
136 used instead.
137
138 Public functions from a file or subsystem (declared in headers) tend
139 to have a consistent prefix to show where they came from. For example,
140 ``tlb_`` for functions from ``cputlb.c`` or ``cpu_`` for functions
141 from cpus.c.
142
143 If there are two versions of a function to be called with or without a
144 lock held, the function that expects the lock to be already held
145 usually uses the suffix ``_locked``.
146
147
148 Block structure
149 ===============
150
151 Every indented statement is braced; even if the block contains just one
152 statement. The opening brace is on the line that contains the control
153 flow statement that introduces the new block; the closing brace is on the
154 same line as the else keyword, or on a line by itself if there is no else
155 keyword. Example:
156
157 .. code-block:: c
158
159 if (a == 5) {
160 printf("a was 5.\n");
161 } else if (a == 6) {
162 printf("a was 6.\n");
163 } else {
164 printf("a was something else entirely.\n");
165 }
166
167 Note that 'else if' is considered a single statement; otherwise a long if/
168 else if/else if/.../else sequence would need an indent for every else
169 statement.
170
171 An exception is the opening brace for a function; for reasons of tradition
172 and clarity it comes on a line by itself:
173
174 .. code-block:: c
175
176 void a_function(void)
177 {
178 do_something();
179 }
180
181 Rationale: a consistent (except for functions...) bracing style reduces
182 ambiguity and avoids needless churn when lines are added or removed.
183 Furthermore, it is the QEMU coding style.
184
185 Declarations
186 ============
187
188 Mixed declarations (interleaving statements and declarations within
189 blocks) are generally not allowed; declarations should be at the beginning
190 of blocks.
191
192 Every now and then, an exception is made for declarations inside a
193 #ifdef or #ifndef block: if the code looks nicer, such declarations can
194 be placed at the top of the block even if there are statements above.
195 On the other hand, however, it's often best to move that #ifdef/#ifndef
196 block to a separate function altogether.
197
198 Conditional statements
199 ======================
200
201 When comparing a variable for (in)equality with a constant, list the
202 constant on the right, as in:
203
204 .. code-block:: c
205
206 if (a == 1) {
207 /* Reads like: "If a equals 1" */
208 do_something();
209 }
210
211 Rationale: Yoda conditions (as in 'if (1 == a)') are awkward to read.
212 Besides, good compilers already warn users when '==' is mis-typed as '=',
213 even when the constant is on the right.
214
215 Comment style
216 =============
217
218 We use traditional C-style /``*`` ``*``/ comments and avoid // comments.
219
220 Rationale: The // form is valid in C99, so this is purely a matter of
221 consistency of style. The checkpatch script will warn you about this.
222
223 Multiline comment blocks should have a row of stars on the left,
224 and the initial /``*`` and terminating ``*``/ both on their own lines:
225
226 .. code-block:: c
227
228 /*
229 * like
230 * this
231 */
232
233 This is the same format required by the Linux kernel coding style.
234
235 (Some of the existing comments in the codebase use the GNU Coding
236 Standards form which does not have stars on the left, or other
237 variations; avoid these when writing new comments, but don't worry
238 about converting to the preferred form unless you're editing that
239 comment anyway.)
240
241 Rationale: Consistency, and ease of visually picking out a multiline
242 comment from the surrounding code.
243
244 Language usage
245 **************
246
247 Preprocessor
248 ============
249
250 Variadic macros
251 ---------------
252
253 For variadic macros, stick with this C99-like syntax:
254
255 .. code-block:: c
256
257 #define DPRINTF(fmt, ...) \
258 do { printf("IRQ: " fmt, ## __VA_ARGS__); } while (0)
259
260 Include directives
261 ------------------
262
263 Order include directives as follows:
264
265 .. code-block:: c
266
267 #include "qemu/osdep.h" /* Always first... */
268 #include <...> /* then system headers... */
269 #include "..." /* and finally QEMU headers. */
270
271 The "qemu/osdep.h" header contains preprocessor macros that affect the behavior
272 of core system headers like <stdint.h>. It must be the first include so that
273 core system headers included by external libraries get the preprocessor macros
274 that QEMU depends on.
275
276 Do not include "qemu/osdep.h" from header files since the .c file will have
277 already included it.
278
279 C types
280 =======
281
282 It should be common sense to use the right type, but we have collected
283 a few useful guidelines here.
284
285 Scalars
286 -------
287
288 If you're using "int" or "long", odds are good that there's a better type.
289 If a variable is counting something, it should be declared with an
290 unsigned type.
291
292 If it's host memory-size related, size_t should be a good choice (use
293 ssize_t only if required). Guest RAM memory offsets must use ram_addr_t,
294 but only for RAM, it may not cover whole guest address space.
295
296 If it's file-size related, use off_t.
297 If it's file-offset related (i.e., signed), use off_t.
298 If it's just counting small numbers use "unsigned int";
299 (on all but oddball embedded systems, you can assume that that
300 type is at least four bytes wide).
301
302 In the event that you require a specific width, use a standard type
303 like int32_t, uint32_t, uint64_t, etc. The specific types are
304 mandatory for VMState fields.
305
306 Don't use Linux kernel internal types like u32, __u32 or __le32.
307
308 Use hwaddr for guest physical addresses except pcibus_t
309 for PCI addresses. In addition, ram_addr_t is a QEMU internal address
310 space that maps guest RAM physical addresses into an intermediate
311 address space that can map to host virtual address spaces. Generally
312 speaking, the size of guest memory can always fit into ram_addr_t but
313 it would not be correct to store an actual guest physical address in a
314 ram_addr_t.
315
316 For CPU virtual addresses there are several possible types.
317 vaddr is the best type to use to hold a CPU virtual address in
318 target-independent code. It is guaranteed to be large enough to hold a
319 virtual address for any target, and it does not change size from target
320 to target. It is always unsigned.
321 target_ulong is a type the size of a virtual address on the CPU; this means
322 it may be 32 or 64 bits depending on which target is being built. It should
323 therefore be used only in target-specific code, and in some
324 performance-critical built-per-target core code such as the TLB code.
325 There is also a signed version, target_long.
326 abi_ulong is for the ``*``-user targets, and represents a type the size of
327 'void ``*``' in that target's ABI. (This may not be the same as the size of a
328 full CPU virtual address in the case of target ABIs which use 32 bit pointers
329 on 64 bit CPUs, like sparc32plus.) Definitions of structures that must match
330 the target's ABI must use this type for anything that on the target is defined
331 to be an 'unsigned long' or a pointer type.
332 There is also a signed version, abi_long.
333
334 Of course, take all of the above with a grain of salt. If you're about
335 to use some system interface that requires a type like size_t, pid_t or
336 off_t, use matching types for any corresponding variables.
337
338 Also, if you try to use e.g., "unsigned int" as a type, and that
339 conflicts with the signedness of a related variable, sometimes
340 it's best just to use the *wrong* type, if "pulling the thread"
341 and fixing all related variables would be too invasive.
342
343 Finally, while using descriptive types is important, be careful not to
344 go overboard. If whatever you're doing causes warnings, or requires
345 casts, then reconsider or ask for help.
346
347 Pointers
348 --------
349
350 Ensure that all of your pointers are "const-correct".
351 Unless a pointer is used to modify the pointed-to storage,
352 give it the "const" attribute. That way, the reader knows
353 up-front that this is a read-only pointer. Perhaps more
354 importantly, if we're diligent about this, when you see a non-const
355 pointer, you're guaranteed that it is used to modify the storage
356 it points to, or it is aliased to another pointer that is.
357
358 Typedefs
359 --------
360
361 Typedefs are used to eliminate the redundant 'struct' keyword, since type
362 names have a different style than other identifiers ("CamelCase" versus
363 "snake_case"). Each named struct type should have a CamelCase name and a
364 corresponding typedef.
365
366 Since certain C compilers choke on duplicated typedefs, you should avoid
367 them and declare a typedef only in one header file. For common types,
368 you can use "include/qemu/typedefs.h" for example. However, as a matter
369 of convenience it is also perfectly fine to use forward struct
370 definitions instead of typedefs in headers and function prototypes; this
371 avoids problems with duplicated typedefs and reduces the need to include
372 headers from other headers.
373
374 Reserved namespaces in C and POSIX
375 ----------------------------------
376
377 Underscore capital, double underscore, and underscore 't' suffixes should be
378 avoided.
379
380 Low level memory management
381 ===========================
382
383 Use of the malloc/free/realloc/calloc/valloc/memalign/posix_memalign
384 APIs is not allowed in the QEMU codebase. Instead of these routines,
385 use the GLib memory allocation routines g_malloc/g_malloc0/g_new/
386 g_new0/g_realloc/g_free or QEMU's qemu_memalign/qemu_blockalign/qemu_vfree
387 APIs.
388
389 Please note that g_malloc will exit on allocation failure, so there
390 is no need to test for failure (as you would have to with malloc).
391 Calling g_malloc with a zero size is valid and will return NULL.
392
393 Prefer g_new(T, n) instead of g_malloc(sizeof(T) ``*`` n) for the following
394 reasons:
395
396 * It catches multiplication overflowing size_t;
397 * It returns T ``*`` instead of void ``*``, letting compiler catch more type errors.
398
399 Declarations like
400
401 .. code-block:: c
402
403 T *v = g_malloc(sizeof(*v))
404
405 are acceptable, though.
406
407 Memory allocated by qemu_memalign or qemu_blockalign must be freed with
408 qemu_vfree, since breaking this will cause problems on Win32.
409
410 String manipulation
411 ===================
412
413 Do not use the strncpy function. As mentioned in the man page, it does *not*
414 guarantee a NULL-terminated buffer, which makes it extremely dangerous to use.
415 It also zeros trailing destination bytes out to the specified length. Instead,
416 use this similar function when possible, but note its different signature:
417
418 .. code-block:: c
419
420 void pstrcpy(char *dest, int dest_buf_size, const char *src)
421
422 Don't use strcat because it can't check for buffer overflows, but:
423
424 .. code-block:: c
425
426 char *pstrcat(char *buf, int buf_size, const char *s)
427
428 The same limitation exists with sprintf and vsprintf, so use snprintf and
429 vsnprintf.
430
431 QEMU provides other useful string functions:
432
433 .. code-block:: c
434
435 int strstart(const char *str, const char *val, const char **ptr)
436 int stristart(const char *str, const char *val, const char **ptr)
437 int qemu_strnlen(const char *s, int max_len)
438
439 There are also replacement character processing macros for isxyz and toxyz,
440 so instead of e.g. isalnum you should use qemu_isalnum.
441
442 Because of the memory management rules, you must use g_strdup/g_strndup
443 instead of plain strdup/strndup.
444
445 Printf-style functions
446 ======================
447
448 Whenever you add a new printf-style function, i.e., one with a format
449 string argument and following "..." in its prototype, be sure to use
450 gcc's printf attribute directive in the prototype.
451
452 This makes it so gcc's -Wformat and -Wformat-security options can do
453 their jobs and cross-check format strings with the number and types
454 of arguments.
455
456 C standard, implementation defined and undefined behaviors
457 ==========================================================
458
459 C code in QEMU should be written to the C99 language specification. A copy
460 of the final version of the C99 standard with corrigenda TC1, TC2, and TC3
461 included, formatted as a draft, can be downloaded from:
462
463 `<http://www.open-std.org/jtc1/sc22/WG14/www/docs/n1256.pdf>`_
464
465 The C language specification defines regions of undefined behavior and
466 implementation defined behavior (to give compiler authors enough leeway to
467 produce better code). In general, code in QEMU should follow the language
468 specification and avoid both undefined and implementation defined
469 constructs. ("It works fine on the gcc I tested it with" is not a valid
470 argument...) However there are a few areas where we allow ourselves to
471 assume certain behaviors because in practice all the platforms we care about
472 behave in the same way and writing strictly conformant code would be
473 painful. These are:
474
475 * you may assume that integers are 2s complement representation
476 * you may assume that right shift of a signed integer duplicates
477 the sign bit (ie it is an arithmetic shift, not a logical shift)
478
479 In addition, QEMU assumes that the compiler does not use the latitude
480 given in C99 and C11 to treat aspects of signed '<<' as undefined, as
481 documented in the GNU Compiler Collection manual starting at version 4.0.
482
483 Automatic memory deallocation
484 =============================
485
486 QEMU has a mandatory dependency either the GCC or CLang compiler. As
487 such it has the freedom to make use of a C language extension for
488 automatically running a cleanup function when a stack variable goes
489 out of scope. This can be used to simplify function cleanup paths,
490 often allowing many goto jumps to be eliminated, through automatic
491 free'ing of memory.
492
493 The GLib2 library provides a number of functions/macros for enabling
494 automatic cleanup:
495
496 `<https://developer.gnome.org/glib/stable/glib-Miscellaneous-Macros.html>`_
497
498 Most notably:
499
500 * g_autofree - will invoke g_free() on the variable going out of scope
501
502 * g_autoptr - for structs / objects, will invoke the cleanup func created
503 by a previous use of G_DEFINE_AUTOPTR_CLEANUP_FUNC. This is
504 supported for most GLib data types and GObjects
505
506 For example, instead of
507
508 .. code-block:: c
509
510 int somefunc(void) {
511 int ret = -1;
512 char *foo = g_strdup_printf("foo%", "wibble");
513 GList *bar = .....
514
515 if (eek) {
516 goto cleanup;
517 }
518
519 ret = 0;
520
521 cleanup:
522 g_free(foo);
523 g_list_free(bar);
524 return ret;
525 }
526
527 Using g_autofree/g_autoptr enables the code to be written as:
528
529 .. code-block:: c
530
531 int somefunc(void) {
532 g_autofree char *foo = g_strdup_printf("foo%", "wibble");
533 g_autoptr (GList) bar = .....
534
535 if (eek) {
536 return -1;
537 }
538
539 return 0;
540 }
541
542 While this generally results in simpler, less leak-prone code, there
543 are still some caveats to beware of
544
545 * Variables declared with g_auto* MUST always be initialized,
546 otherwise the cleanup function will use uninitialized stack memory
547
548 * If a variable declared with g_auto* holds a value which must
549 live beyond the life of the function, that value must be saved
550 and the original variable NULL'd out. This can be simpler using
551 g_steal_pointer
552
553
554 .. code-block:: c
555
556 char *somefunc(void) {
557 g_autofree char *foo = g_strdup_printf("foo%", "wibble");
558 g_autoptr (GList) bar = .....
559
560 if (eek) {
561 return NULL;
562 }
563
564 return g_steal_pointer(&foo);
565 }
566
567
568 QEMU Specific Idioms
569 ********************
570
571 Error handling and reporting
572 ============================
573
574 Reporting errors to the human user
575 ----------------------------------
576
577 Do not use printf(), fprintf() or monitor_printf(). Instead, use
578 error_report() or error_vreport() from error-report.h. This ensures the
579 error is reported in the right place (current monitor or stderr), and in
580 a uniform format.
581
582 Use error_printf() & friends to print additional information.
583
584 error_report() prints the current location. In certain common cases
585 like command line parsing, the current location is tracked
586 automatically. To manipulate it manually, use the loc_``*``() from
587 error-report.h.
588
589 Propagating errors
590 ------------------
591
592 An error can't always be reported to the user right where it's detected,
593 but often needs to be propagated up the call chain to a place that can
594 handle it. This can be done in various ways.
595
596 The most flexible one is Error objects. See error.h for usage
597 information.
598
599 Use the simplest suitable method to communicate success / failure to
600 callers. Stick to common methods: non-negative on success / -1 on
601 error, non-negative / -errno, non-null / null, or Error objects.
602
603 Example: when a function returns a non-null pointer on success, and it
604 can fail only in one way (as far as the caller is concerned), returning
605 null on failure is just fine, and certainly simpler and a lot easier on
606 the eyes than propagating an Error object through an Error ``*````*`` parameter.
607
608 Example: when a function's callers need to report details on failure
609 only the function really knows, use Error ``*````*``, and set suitable errors.
610
611 Do not report an error to the user when you're also returning an error
612 for somebody else to handle. Leave the reporting to the place that
613 consumes the error returned.
614
615 Handling errors
616 ---------------
617
618 Calling exit() is fine when handling configuration errors during
619 startup. It's problematic during normal operation. In particular,
620 monitor commands should never exit().
621
622 Do not call exit() or abort() to handle an error that can be triggered
623 by the guest (e.g., some unimplemented corner case in guest code
624 translation or device emulation). Guests should not be able to
625 terminate QEMU.
626
627 Note that &error_fatal is just another way to exit(1), and &error_abort
628 is just another way to abort().
629
630
631 trace-events style
632 ==================
633
634 0x prefix
635 ---------
636
637 In trace-events files, use a '0x' prefix to specify hex numbers, as in:
638
639 .. code-block::
640
641 some_trace(unsigned x, uint64_t y) "x 0x%x y 0x" PRIx64
642
643 An exception is made for groups of numbers that are hexadecimal by
644 convention and separated by the symbols '.', '/', ':', or ' ' (such as
645 PCI bus id):
646
647 .. code-block::
648
649 another_trace(int cssid, int ssid, int dev_num) "bus id: %x.%x.%04x"
650
651 However, you can use '0x' for such groups if you want. Anyway, be sure that
652 it is obvious that numbers are in hex, ex.:
653
654 .. code-block::
655
656 data_dump(uint8_t c1, uint8_t c2, uint8_t c3) "bytes (in hex): %02x %02x %02x"
657
658 Rationale: hex numbers are hard to read in logs when there is no 0x prefix,
659 especially when (occasionally) the representation doesn't contain any letters
660 and especially in one line with other decimal numbers. Number groups are allowed
661 to not use '0x' because for some things notations like %x.%x.%x are used not
662 only in Qemu. Also dumping raw data bytes with '0x' is less readable.
663
664 '#' printf flag
665 ---------------
666
667 Do not use printf flag '#', like '%#x'.
668
669 Rationale: there are two ways to add a '0x' prefix to printed number: '0x%...'
670 and '%#...'. For consistency the only one way should be used. Arguments for
671 '0x%' are:
672
673 * it is more popular
674 * '%#' omits the 0x for the value 0 which makes output inconsistent