2 * OS includes and handling of OS dependencies
4 * This header exists to pull in some common system headers that
5 * most code in QEMU will want, and to fix up some possible issues with
6 * it (missing defines, Windows weirdness, and so on).
8 * To avoid getting into possible circular include dependencies, this
9 * file should not include any other QEMU headers, with the exceptions
10 * of config-host.h, config-target.h, qemu/compiler.h,
11 * sysemu/os-posix.h, sysemu/os-win32.h, glib-compat.h and
12 * qemu/typedefs.h, all of which are doing a similar job to this file
13 * and are under similar constraints.
15 * This header also contains prototypes for functions defined in
16 * os-*.c and util/oslib-*.c; those would probably be better split
17 * out into separate header files.
19 * In an ideal world this header would contain only:
20 * (1) things which everybody needs
21 * (2) things without which code would work on most platforms but
22 * fail to compile or misbehave on a minority of host OSes
24 * This work is licensed under the terms of the GNU GPL, version 2 or later.
25 * See the COPYING file in the top-level directory.
30 #if !defined _FORTIFY_SOURCE && defined __OPTIMIZE__ && __OPTIMIZE__ && defined __linux__
31 # define _FORTIFY_SOURCE 2
34 #include "config-host.h"
36 #include CONFIG_TARGET
38 #include "exec/poison.h"
42 * HOST_WORDS_BIGENDIAN was replaced with HOST_BIG_ENDIAN. Prevent it from
45 #pragma GCC poison HOST_WORDS_BIGENDIAN
48 * TARGET_WORDS_BIGENDIAN was replaced with TARGET_BIG_ENDIAN. Prevent it from
51 #pragma GCC poison TARGET_WORDS_BIGENDIAN
53 #include "qemu/compiler.h"
55 /* Older versions of C++ don't get definitions of various macros from
56 * stdlib.h unless we define these macros before first inclusion of
59 #ifndef __STDC_CONSTANT_MACROS
60 #define __STDC_CONSTANT_MACROS
62 #ifndef __STDC_LIMIT_MACROS
63 #define __STDC_LIMIT_MACROS
65 #ifndef __STDC_FORMAT_MACROS
66 #define __STDC_FORMAT_MACROS
69 /* The following block of code temporarily renames the daemon() function so the
70 * compiler does not see the warning associated with it in stdlib.h on OSX
73 #define daemon qemu_fake_daemon_function
76 QEMU_EXTERN_C
int daemon(int, int);
80 /* as defined in sdkddkver.h */
82 #define _WIN32_WINNT 0x0602 /* Windows 8 API (should be >= the one from glib) */
84 /* reduces the number of implicitly included headers */
85 #ifndef WIN32_LEAN_AND_MEAN
86 #define WIN32_LEAN_AND_MEAN
90 /* enable C99/POSIX format strings (needs mingw32-runtime 3.15 or later) */
92 #define __USE_MINGW_ANSI_STDIO 1
96 * We need the FreeBSD "legacy" definitions. Rust needs the FreeBSD 11 system
97 * calls since it doesn't use libc at all, so we have to emulate that despite
98 * FreeBSD 11 being EOL'd.
101 #define _WANT_FREEBSD11_STAT
102 #define _WANT_FREEBSD11_STATFS
103 #define _WANT_FREEBSD11_DIRENT
104 #define _WANT_KERNEL_ERRNO
112 #include <sys/types.h>
118 #include <inttypes.h>
120 /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r
121 * function availability on recentish Mingw-w64 platforms. */
128 #include <sys/stat.h>
129 #include <sys/time.h>
131 /* setjmp must be declared before sysemu/os-win32.h
132 * because it is redefined there. */
140 #if defined(__linux__) && defined(__sparc__)
141 /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */
146 #include <sys/wait.h>
148 #define WIFEXITED(x) 1
149 #define WEXITSTATUS(x) (x)
153 #include <AvailabilityMacros.h>
157 * This is somewhat like a system header; it must be outside any extern "C"
158 * block because it includes system headers itself, including glib.h,
159 * which will not compile if inside an extern "C" block.
161 #include "glib-compat.h"
164 #include "sysemu/os-win32.h"
168 #include "sysemu/os-posix.h"
175 #include "qemu/typedefs.h"
178 * Mark a function that executes in coroutine context
180 * Functions that execute in coroutine context cannot be called directly from
181 * normal functions. In the future it would be nice to enable compiler or
182 * static checker support for catching such errors. This annotation might make
183 * it possible and in the meantime it serves as documentation.
187 * static void coroutine_fn foo(void) {
192 #define coroutine_fn QEMU_ANNOTATE("coroutine_fn")
198 * Mark a function that can suspend when executed in coroutine context,
199 * but can handle running in non-coroutine context too.
202 #define coroutine_mixed_fn QEMU_ANNOTATE("coroutine_mixed_fn")
204 #define coroutine_mixed_fn
208 * Mark a function that should not be called from a coroutine context.
209 * Usually there will be an analogous, coroutine_fn function that should
212 * When the function is also marked as coroutine_mixed_fn, the function should
213 * only be called if the caller does not know whether it is in coroutine
216 * Functions that are only no_coroutine_fn, on the other hand, should not
217 * be called from within coroutines at all. This for example includes
218 * functions that block.
220 * In the future it would be nice to enable compiler or static checker
221 * support for catching such errors. This annotation is the first step
222 * towards this, and in the meantime it serves as documentation.
226 * static void no_coroutine_fn foo(void) {
231 #define no_coroutine_fn QEMU_ANNOTATE("no_coroutine_fn")
233 #define no_coroutine_fn
238 * For mingw, as of v6.0.0, the function implementing the assert macro is
239 * not marked as noreturn, so the compiler cannot delete code following an
240 * assert(false) as unused. We rely on this within the code base to delete
241 * code that is unreachable when features are disabled.
242 * All supported versions of Glib's g_assert() satisfy this requirement.
246 #define assert(x) g_assert(x)
250 * qemu_build_not_reached()
252 * The compiler, during optimization, is expected to prove that a call
253 * to this function cannot be reached and remove it. If the compiler
254 * supports QEMU_ERROR, this will be reported at compile time; otherwise
255 * this will be reported at link time due to the missing symbol.
258 void QEMU_ERROR("code path is reachable")
259 qemu_build_not_reached_always(void);
260 #if defined(__OPTIMIZE__) && !defined(__NO_INLINE__)
261 #define qemu_build_not_reached() qemu_build_not_reached_always()
263 #define qemu_build_not_reached() g_assert_not_reached()
267 * qemu_build_assert()
269 * The compiler, during optimization, is expected to prove that the
272 #define qemu_build_assert(test) while (!(test)) qemu_build_not_reached()
275 * According to waitpid man page:
277 * This macro is not specified in POSIX.1-2001 and is not
278 * available on some UNIX implementations (e.g., AIX, SunOS).
279 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
282 #define WCOREDUMP(status) 0
285 * We have a lot of unaudited code that may fail in strange ways, or
286 * even be a security risk during migration, if you disable assertions
287 * at compile-time. You may comment out these safety checks if you
288 * absolutely want to disable assertion overhead, but it is not
289 * supported upstream so the risk is all yours. Meanwhile, please
290 * submit patches to remove any side-effects inside an assertion, or
291 * fixing error handling that should use Error instead of assert.
294 #error building with NDEBUG is not supported
296 #ifdef G_DISABLE_ASSERT
297 #error building with G_DISABLE_ASSERT is not supported
301 #define O_LARGEFILE 0
306 #ifndef MAP_ANONYMOUS
307 #define MAP_ANONYMOUS MAP_ANON
309 #ifndef MAP_NORESERVE
310 #define MAP_NORESERVE 0
313 #define ENOMEDIUM ENODEV
315 #if !defined(ENOTSUP)
318 #if !defined(ECANCELED)
319 #define ECANCELED 4097
321 #if !defined(EMEDIUMTYPE)
322 #define EMEDIUMTYPE 4098
324 #if !defined(ESHUTDOWN)
325 #define ESHUTDOWN 4099
328 #define RETRY_ON_EINTR(expr) \
330 ({ typeof(expr) __result; \
333 } while (__result == -1 && errno == EINTR); \
336 /* time_t may be either 32 or 64 bits depending on the host OS, and
337 * can be either signed or unsigned, so we can't just hardcode a
338 * specific maximum value. This is not a C preprocessor constant,
339 * so you can't use TIME_MAX in an #ifdef, but for our purposes
340 * this isn't a problem.
343 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
344 * Gnulib, and are under the LGPL v2.1 or (at your option) any
348 /* True if the real type T is signed. */
349 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
351 /* The width in bits of the integer type or expression T.
352 * Padding bits are not supported.
354 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
356 /* The maximum and minimum values for the integer type T. */
357 #define TYPE_MAXIMUM(t) \
358 ((t) (!TYPE_SIGNED(t) \
360 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
363 #define TIME_MAX TYPE_MAXIMUM(time_t)
366 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
367 * the wrong type. Our replacement isn't usable in preprocessor
368 * expressions, but it is sufficient for our needs. */
369 #ifdef HAVE_BROKEN_SIZE_MAX
371 #define SIZE_MAX ((size_t)-1)
375 * Two variations of MIN/MAX macros. The first is for runtime use, and
376 * evaluates arguments only once (so it is safe even with side
377 * effects), but will not work in constant contexts (such as array
378 * size declarations) because of the '{}'. The second is for constant
379 * expression use, where evaluating arguments twice is safe because
380 * the result is going to be constant anyway, but will not work in a
381 * runtime context because of a void expression where a value is
382 * expected. Thus, both gcc and clang will fail to compile if you use
383 * the wrong macro (even if the error may seem a bit cryptic).
385 * Note that neither form is usable as an #if condition; if you truly
386 * need to write conditional code that depends on a minimum or maximum
387 * determined by the pre-processor instead of the compiler, you'll
388 * have to open-code it. Sadly, Coverity is severely confused by the
389 * constant variants, so we have to dumb things down there.
391 * Preprocessor sorcery ahead: use different identifiers for the local
392 * variables in each expansion, so we can nest macro calls without
393 * shadowing variables.
395 #define MIN_INTERNAL(a, b, _a, _b) \
397 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
402 MIN_INTERNAL((a), (b), MAKE_IDENTFIER(_a), MAKE_IDENTFIER(_b))
404 #define MAX_INTERNAL(a, b, _a, _b) \
406 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
411 MAX_INTERNAL((a), (b), MAKE_IDENTFIER(_a), MAKE_IDENTFIER(_b))
414 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
415 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
417 # define MIN_CONST(a, b) \
418 __builtin_choose_expr( \
419 __builtin_constant_p(a) && __builtin_constant_p(b), \
420 (a) < (b) ? (a) : (b), \
422 # define MAX_CONST(a, b) \
423 __builtin_choose_expr( \
424 __builtin_constant_p(a) && __builtin_constant_p(b), \
425 (a) > (b) ? (a) : (b), \
430 * Minimum function that returns zero only if both values are zero.
431 * Intended for use with unsigned values only.
433 * Preprocessor sorcery ahead: use different identifiers for the local
434 * variables in each expansion, so we can nest macro calls without
435 * shadowing variables.
437 #define MIN_NON_ZERO_INTERNAL(a, b, _a, _b) \
439 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
440 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
442 #define MIN_NON_ZERO(a, b) \
443 MIN_NON_ZERO_INTERNAL((a), (b), MAKE_IDENTFIER(_a), MAKE_IDENTFIER(_b))
446 * Round number down to multiple. Safe when m is not a power of 2 (see
447 * ROUND_DOWN for a faster version when a power of 2 is guaranteed).
449 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
452 * Round number up to multiple. Safe when m is not a power of 2 (see
453 * ROUND_UP for a faster version when a power of 2 is guaranteed).
455 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
457 /* Check if n is a multiple of m */
458 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
460 /* n-byte align pointer down */
461 #define QEMU_ALIGN_PTR_DOWN(p, n) \
462 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
464 /* n-byte align pointer up */
465 #define QEMU_ALIGN_PTR_UP(p, n) \
466 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
468 /* Check if pointer p is n-bytes aligned */
469 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
472 * Round number down to multiple. Requires that d be a power of 2 (see
473 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
474 * numbers); works even if d is a smaller type than n.
477 #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d)))
481 * Round number up to multiple. Requires that d be a power of 2 (see
482 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
483 * numbers); works even if d is a smaller type than n.
486 #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d))
490 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
494 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
495 * pointer, not an array.
497 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
500 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
501 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
504 int qemu_daemon(int nochdir
, int noclose
);
505 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
,
507 void qemu_anon_ram_free(void *ptr
, size_t size
);
510 #define HAVE_CHARDEV_SERIAL 1
511 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
512 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
513 || defined(__GLIBC__) || defined(__APPLE__)
514 #define HAVE_CHARDEV_SERIAL 1
517 #if defined(__HAIKU__)
518 #define SIGIO SIGPOLL
521 #ifdef HAVE_MADVISE_WITHOUT_PROTOTYPE
523 * See MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156) for discussion
524 * about Solaris missing the madvise() prototype.
526 int madvise(char *, size_t, int);
529 #if defined(CONFIG_LINUX)
530 #ifndef BUS_MCEERR_AR
531 #define BUS_MCEERR_AR 4
533 #ifndef BUS_MCEERR_AO
534 #define BUS_MCEERR_AO 5
538 #if defined(__linux__) && \
539 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
540 || defined(__powerpc64__))
541 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
542 Valgrind does not support alignments larger than 1 MiB,
543 therefore we need special code which handles running on Valgrind. */
544 # define QEMU_VMALLOC_ALIGN (512 * 4096)
545 #elif defined(__linux__) && defined(__s390x__)
546 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
547 # define QEMU_VMALLOC_ALIGN (256 * 4096)
548 #elif defined(__linux__) && defined(__sparc__)
549 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size(), SHMLBA)
551 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size()
555 struct qemu_signalfd_siginfo
{
556 uint32_t ssi_signo
; /* Signal number */
557 int32_t ssi_errno
; /* Error number (unused) */
558 int32_t ssi_code
; /* Signal code */
559 uint32_t ssi_pid
; /* PID of sender */
560 uint32_t ssi_uid
; /* Real UID of sender */
561 int32_t ssi_fd
; /* File descriptor (SIGIO) */
562 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
563 uint32_t ssi_band
; /* Band event (SIGIO) */
564 uint32_t ssi_overrun
; /* POSIX timer overrun count */
565 uint32_t ssi_trapno
; /* Trap number that caused signal */
566 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
567 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
568 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
569 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
570 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
571 uint64_t ssi_addr
; /* Address that generated signal
572 (for hardware-generated signals) */
573 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
574 additional fields in the future) */
577 int qemu_signalfd(const sigset_t
*mask
);
578 void sigaction_invoke(struct sigaction
*action
,
579 struct qemu_signalfd_siginfo
*info
);
583 * Don't introduce new usage of this function, prefer the following
584 * qemu_open/qemu_create that take an "Error **errp"
586 int qemu_open_old(const char *name
, int flags
, ...);
587 int qemu_open(const char *name
, int flags
, Error
**errp
);
588 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
589 int qemu_close(int fd
);
590 int qemu_unlink(const char *name
);
592 int qemu_dup_flags(int fd
, int flags
);
593 int qemu_dup(int fd
);
594 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
595 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
596 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
597 bool qemu_has_ofd_lock(void);
600 #if defined(__HAIKU__) && defined(__i386__)
601 #define FMT_pid "%ld"
603 #define FMT_pid "%" PRId64
608 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
610 int qemu_get_thread_id(void);
618 * Use the same value as Linux for now.
622 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
623 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
627 static inline void qemu_timersub(const struct timeval
*val1
,
628 const struct timeval
*val2
,
631 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
632 if (val1
->tv_usec
< val2
->tv_usec
) {
634 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
636 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
640 #define qemu_timersub timersub
643 ssize_t
qemu_write_full(int fd
, const void *buf
, size_t count
)
644 G_GNUC_WARN_UNUSED_RESULT
;
646 void qemu_set_cloexec(int fd
);
648 /* Return a dynamically allocated directory path that is appropriate for storing
651 * The caller is responsible for releasing the value returned with g_free()
654 char *qemu_get_local_state_dir(void);
658 * @type: the auxiliary vector key to lookup
660 * Search the auxiliary vector for @type, returning the value
661 * or 0 if @type is not present.
663 unsigned long qemu_getauxval(unsigned long type
);
665 void qemu_set_tty_echo(int fd
, bool echo
);
667 typedef struct ThreadContext ThreadContext
;
671 * @fd: the fd mapped into the area, -1 for anonymous memory
672 * @area: start address of the are to preallocate
673 * @sz: the size of the area to preallocate
674 * @max_threads: maximum number of threads to use
675 * @errp: returns an error if this function fails
677 * Preallocate memory (populate/prefault page tables writable) for the virtual
678 * memory area starting at @area with the size of @sz. After a successful call,
679 * each page in the area was faulted in writable at least once, for example,
680 * after allocating file blocks for mapped files.
682 * Return: true on success, else false setting @errp with error.
684 bool qemu_prealloc_mem(int fd
, char *area
, size_t sz
, int max_threads
,
685 ThreadContext
*tc
, Error
**errp
);
689 * @pid: pid of a process
691 * For given @pid fetch its name. Caller is responsible for
692 * freeing the string when no longer needed.
693 * Returns allocated string on success, NULL on failure.
695 char *qemu_get_pid_name(pid_t pid
);
697 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
698 * when intptr_t is 32-bit and we are aligning a long long.
700 static inline uintptr_t qemu_real_host_page_size(void)
702 return getpagesize();
705 static inline intptr_t qemu_real_host_page_mask(void)
707 return -(intptr_t)qemu_real_host_page_size();
711 * After using getopt or getopt_long, if you need to parse another set
712 * of options, then you must reset optind. Unfortunately the way to
713 * do this varies between implementations of getopt.
715 static inline void qemu_reset_optind(void)
725 int qemu_fdatasync(int fd
);
728 * Sync changes made to the memory mapped file back to the backing
729 * storage. For POSIX compliant systems this will fallback
730 * to regular msync call. Otherwise it will trigger whole file sync
731 * (including the metadata case there is no support to skip that otherwise)
733 * @addr - start of the memory area to be synced
734 * @length - length of the are to be synced
735 * @fd - file descriptor for the file to be synced
736 * (mandatory only for POSIX non-compliant systems)
738 int qemu_msync(void *addr
, size_t length
, int fd
);
741 * qemu_get_host_physmem:
743 * Operating system agnostic way of querying host memory.
745 * Returns amount of physical memory on the system. This is purely
746 * advisery and may return 0 if we can't work it out. At the other
747 * end we saturate to SIZE_MAX if you are lucky enough to have that
750 size_t qemu_get_host_physmem(void);
753 * Toggle write/execute on the pages marked MAP_JIT
754 * for the current thread.
756 #if defined(MAC_OS_VERSION_11_0) && \
757 MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_VERSION_11_0
758 static inline void qemu_thread_jit_execute(void)
760 pthread_jit_write_protect_np(true);
763 static inline void qemu_thread_jit_write(void)
765 pthread_jit_write_protect_np(false);
768 static inline void qemu_thread_jit_write(void) {}
769 static inline void qemu_thread_jit_execute(void) {}
773 * Platforms which do not support system() return ENOSYS
775 #ifndef HAVE_SYSTEM_FUNCTION
776 #define system platform_does_not_support_system
777 static inline int platform_does_not_support_system(const char *command
)
782 #endif /* !HAVE_SYSTEM_FUNCTION */