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 #include "config-host.h"
32 #include CONFIG_TARGET
34 #include "exec/poison.h"
38 * HOST_WORDS_BIGENDIAN was replaced with HOST_BIG_ENDIAN. Prevent it from
41 #pragma GCC poison HOST_WORDS_BIGENDIAN
44 * TARGET_WORDS_BIGENDIAN was replaced with TARGET_BIG_ENDIAN. Prevent it from
47 #pragma GCC poison TARGET_WORDS_BIGENDIAN
49 #include "qemu/compiler.h"
51 /* Older versions of C++ don't get definitions of various macros from
52 * stdlib.h unless we define these macros before first inclusion of
55 #ifndef __STDC_CONSTANT_MACROS
56 #define __STDC_CONSTANT_MACROS
58 #ifndef __STDC_LIMIT_MACROS
59 #define __STDC_LIMIT_MACROS
61 #ifndef __STDC_FORMAT_MACROS
62 #define __STDC_FORMAT_MACROS
65 /* The following block of code temporarily renames the daemon() function so the
66 * compiler does not see the warning associated with it in stdlib.h on OSX
69 #define daemon qemu_fake_daemon_function
72 QEMU_EXTERN_C
int daemon(int, int);
76 /* as defined in sdkddkver.h */
78 #define _WIN32_WINNT 0x0602 /* Windows 8 API (should be >= the one from glib) */
80 /* reduces the number of implicitly included headers */
81 #ifndef WIN32_LEAN_AND_MEAN
82 #define WIN32_LEAN_AND_MEAN
86 /* enable C99/POSIX format strings (needs mingw32-runtime 3.15 or later) */
88 #define __USE_MINGW_ANSI_STDIO 1
92 * We need the FreeBSD "legacy" definitions. Rust needs the FreeBSD 11 system
93 * calls since it doesn't use libc at all, so we have to emulate that despite
94 * FreeBSD 11 being EOL'd.
97 #define _WANT_FREEBSD11_STAT
98 #define _WANT_FREEBSD11_STATFS
99 #define _WANT_FREEBSD11_DIRENT
100 #define _WANT_KERNEL_ERRNO
108 #include <sys/types.h>
114 #include <inttypes.h>
116 /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r
117 * function availability on recentish Mingw-w64 platforms. */
124 #include <sys/stat.h>
125 #include <sys/time.h>
127 /* setjmp must be declared before sysemu/os-win32.h
128 * because it is redefined there. */
136 #if defined(__linux__) && defined(__sparc__)
137 /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */
142 #include <sys/wait.h>
144 #define WIFEXITED(x) 1
145 #define WEXITSTATUS(x) (x)
149 #include <AvailabilityMacros.h>
153 * This is somewhat like a system header; it must be outside any extern "C"
154 * block because it includes system headers itself, including glib.h,
155 * which will not compile if inside an extern "C" block.
157 #include "glib-compat.h"
160 #include "sysemu/os-win32.h"
164 #include "sysemu/os-posix.h"
171 #include "qemu/typedefs.h"
174 * Mark a function that executes in coroutine context
176 * Functions that execute in coroutine context cannot be called directly from
177 * normal functions. In the future it would be nice to enable compiler or
178 * static checker support for catching such errors. This annotation might make
179 * it possible and in the meantime it serves as documentation.
183 * static void coroutine_fn foo(void) {
188 #define coroutine_fn __attribute__((__annotate__("coroutine_fn")))
194 * Mark a function that can suspend when executed in coroutine context,
195 * but can handle running in non-coroutine context too.
198 #define coroutine_mixed_fn __attribute__((__annotate__("coroutine_mixed_fn")))
200 #define coroutine_mixed_fn
204 * Mark a function that should not be called from a coroutine context.
205 * Usually there will be an analogous, coroutine_fn function that should
208 * When the function is also marked as coroutine_mixed_fn, the function should
209 * only be called if the caller does not know whether it is in coroutine
212 * Functions that are only no_coroutine_fn, on the other hand, should not
213 * be called from within coroutines at all. This for example includes
214 * functions that block.
216 * In the future it would be nice to enable compiler or static checker
217 * support for catching such errors. This annotation is the first step
218 * towards this, and in the meantime it serves as documentation.
222 * static void no_coroutine_fn foo(void) {
227 #define no_coroutine_fn __attribute__((__annotate__("no_coroutine_fn")))
229 #define no_coroutine_fn
234 * For mingw, as of v6.0.0, the function implementing the assert macro is
235 * not marked as noreturn, so the compiler cannot delete code following an
236 * assert(false) as unused. We rely on this within the code base to delete
237 * code that is unreachable when features are disabled.
238 * All supported versions of Glib's g_assert() satisfy this requirement.
242 #define assert(x) g_assert(x)
246 * qemu_build_not_reached()
248 * The compiler, during optimization, is expected to prove that a call
249 * to this function cannot be reached and remove it. If the compiler
250 * supports QEMU_ERROR, this will be reported at compile time; otherwise
251 * this will be reported at link time due to the missing symbol.
254 void QEMU_ERROR("code path is reachable")
255 qemu_build_not_reached_always(void);
256 #if defined(__OPTIMIZE__) && !defined(__NO_INLINE__)
257 #define qemu_build_not_reached() qemu_build_not_reached_always()
259 #define qemu_build_not_reached() g_assert_not_reached()
263 * qemu_build_assert()
265 * The compiler, during optimization, is expected to prove that the
268 #define qemu_build_assert(test) while (!(test)) qemu_build_not_reached()
271 * According to waitpid man page:
273 * This macro is not specified in POSIX.1-2001 and is not
274 * available on some UNIX implementations (e.g., AIX, SunOS).
275 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
278 #define WCOREDUMP(status) 0
281 * We have a lot of unaudited code that may fail in strange ways, or
282 * even be a security risk during migration, if you disable assertions
283 * at compile-time. You may comment out these safety checks if you
284 * absolutely want to disable assertion overhead, but it is not
285 * supported upstream so the risk is all yours. Meanwhile, please
286 * submit patches to remove any side-effects inside an assertion, or
287 * fixing error handling that should use Error instead of assert.
290 #error building with NDEBUG is not supported
292 #ifdef G_DISABLE_ASSERT
293 #error building with G_DISABLE_ASSERT is not supported
297 #define O_LARGEFILE 0
302 #ifndef MAP_ANONYMOUS
303 #define MAP_ANONYMOUS MAP_ANON
305 #ifndef MAP_NORESERVE
306 #define MAP_NORESERVE 0
309 #define ENOMEDIUM ENODEV
311 #if !defined(ENOTSUP)
314 #if !defined(ECANCELED)
315 #define ECANCELED 4097
317 #if !defined(EMEDIUMTYPE)
318 #define EMEDIUMTYPE 4098
320 #if !defined(ESHUTDOWN)
321 #define ESHUTDOWN 4099
324 #define RETRY_ON_EINTR(expr) \
326 ({ typeof(expr) __result; \
329 } while (__result == -1 && errno == EINTR); \
332 /* time_t may be either 32 or 64 bits depending on the host OS, and
333 * can be either signed or unsigned, so we can't just hardcode a
334 * specific maximum value. This is not a C preprocessor constant,
335 * so you can't use TIME_MAX in an #ifdef, but for our purposes
336 * this isn't a problem.
339 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
340 * Gnulib, and are under the LGPL v2.1 or (at your option) any
344 /* True if the real type T is signed. */
345 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
347 /* The width in bits of the integer type or expression T.
348 * Padding bits are not supported.
350 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
352 /* The maximum and minimum values for the integer type T. */
353 #define TYPE_MAXIMUM(t) \
354 ((t) (!TYPE_SIGNED(t) \
356 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
359 #define TIME_MAX TYPE_MAXIMUM(time_t)
362 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
363 * the wrong type. Our replacement isn't usable in preprocessor
364 * expressions, but it is sufficient for our needs. */
365 #ifdef HAVE_BROKEN_SIZE_MAX
367 #define SIZE_MAX ((size_t)-1)
371 * Two variations of MIN/MAX macros. The first is for runtime use, and
372 * evaluates arguments only once (so it is safe even with side
373 * effects), but will not work in constant contexts (such as array
374 * size declarations) because of the '{}'. The second is for constant
375 * expression use, where evaluating arguments twice is safe because
376 * the result is going to be constant anyway, but will not work in a
377 * runtime context because of a void expression where a value is
378 * expected. Thus, both gcc and clang will fail to compile if you use
379 * the wrong macro (even if the error may seem a bit cryptic).
381 * Note that neither form is usable as an #if condition; if you truly
382 * need to write conditional code that depends on a minimum or maximum
383 * determined by the pre-processor instead of the compiler, you'll
384 * have to open-code it. Sadly, Coverity is severely confused by the
385 * constant variants, so we have to dumb things down there.
390 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
396 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
401 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
402 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
404 # define MIN_CONST(a, b) \
405 __builtin_choose_expr( \
406 __builtin_constant_p(a) && __builtin_constant_p(b), \
407 (a) < (b) ? (a) : (b), \
409 # define MAX_CONST(a, b) \
410 __builtin_choose_expr( \
411 __builtin_constant_p(a) && __builtin_constant_p(b), \
412 (a) > (b) ? (a) : (b), \
417 * Minimum function that returns zero only if both values are zero.
418 * Intended for use with unsigned values only.
421 #define MIN_NON_ZERO(a, b) \
423 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
424 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
429 * Round number down to multiple. Safe when m is not a power of 2 (see
430 * ROUND_DOWN for a faster version when a power of 2 is guaranteed).
432 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
435 * Round number up to multiple. Safe when m is not a power of 2 (see
436 * ROUND_UP for a faster version when a power of 2 is guaranteed).
438 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
440 /* Check if n is a multiple of m */
441 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
443 /* n-byte align pointer down */
444 #define QEMU_ALIGN_PTR_DOWN(p, n) \
445 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
447 /* n-byte align pointer up */
448 #define QEMU_ALIGN_PTR_UP(p, n) \
449 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
451 /* Check if pointer p is n-bytes aligned */
452 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
455 * Round number down to multiple. Requires that d be a power of 2 (see
456 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
457 * numbers); works even if d is a smaller type than n.
460 #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d)))
464 * Round number up to multiple. Requires that d be a power of 2 (see
465 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
466 * numbers); works even if d is a smaller type than n.
469 #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d))
473 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
477 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
478 * pointer, not an array.
480 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
483 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
484 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
487 int qemu_daemon(int nochdir
, int noclose
);
488 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
,
490 void qemu_anon_ram_free(void *ptr
, size_t size
);
493 #define HAVE_CHARDEV_SERIAL 1
494 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
495 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
496 || defined(__GLIBC__) || defined(__APPLE__)
497 #define HAVE_CHARDEV_SERIAL 1
500 #if defined(__HAIKU__)
501 #define SIGIO SIGPOLL
504 #ifdef HAVE_MADVISE_WITHOUT_PROTOTYPE
506 * See MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156) for discussion
507 * about Solaris missing the madvise() prototype.
509 int madvise(char *, size_t, int);
512 #if defined(CONFIG_LINUX)
513 #ifndef BUS_MCEERR_AR
514 #define BUS_MCEERR_AR 4
516 #ifndef BUS_MCEERR_AO
517 #define BUS_MCEERR_AO 5
521 #if defined(__linux__) && \
522 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
523 || defined(__powerpc64__))
524 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
525 Valgrind does not support alignments larger than 1 MiB,
526 therefore we need special code which handles running on Valgrind. */
527 # define QEMU_VMALLOC_ALIGN (512 * 4096)
528 #elif defined(__linux__) && defined(__s390x__)
529 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
530 # define QEMU_VMALLOC_ALIGN (256 * 4096)
531 #elif defined(__linux__) && defined(__sparc__)
532 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size(), SHMLBA)
534 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size()
538 struct qemu_signalfd_siginfo
{
539 uint32_t ssi_signo
; /* Signal number */
540 int32_t ssi_errno
; /* Error number (unused) */
541 int32_t ssi_code
; /* Signal code */
542 uint32_t ssi_pid
; /* PID of sender */
543 uint32_t ssi_uid
; /* Real UID of sender */
544 int32_t ssi_fd
; /* File descriptor (SIGIO) */
545 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
546 uint32_t ssi_band
; /* Band event (SIGIO) */
547 uint32_t ssi_overrun
; /* POSIX timer overrun count */
548 uint32_t ssi_trapno
; /* Trap number that caused signal */
549 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
550 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
551 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
552 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
553 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
554 uint64_t ssi_addr
; /* Address that generated signal
555 (for hardware-generated signals) */
556 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
557 additional fields in the future) */
560 int qemu_signalfd(const sigset_t
*mask
);
561 void sigaction_invoke(struct sigaction
*action
,
562 struct qemu_signalfd_siginfo
*info
);
566 * Don't introduce new usage of this function, prefer the following
567 * qemu_open/qemu_create that take an "Error **errp"
569 int qemu_open_old(const char *name
, int flags
, ...);
570 int qemu_open(const char *name
, int flags
, Error
**errp
);
571 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
572 int qemu_close(int fd
);
573 int qemu_unlink(const char *name
);
575 int qemu_dup_flags(int fd
, int flags
);
576 int qemu_dup(int fd
);
577 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
578 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
579 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
580 bool qemu_has_ofd_lock(void);
583 #if defined(__HAIKU__) && defined(__i386__)
584 #define FMT_pid "%ld"
586 #define FMT_pid "%" PRId64
591 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
593 int qemu_get_thread_id(void);
601 * Use the same value as Linux for now.
605 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
606 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
610 static inline void qemu_timersub(const struct timeval
*val1
,
611 const struct timeval
*val2
,
614 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
615 if (val1
->tv_usec
< val2
->tv_usec
) {
617 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
619 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
623 #define qemu_timersub timersub
626 ssize_t
qemu_write_full(int fd
, const void *buf
, size_t count
)
627 G_GNUC_WARN_UNUSED_RESULT
;
629 void qemu_set_cloexec(int fd
);
631 /* Return a dynamically allocated directory path that is appropriate for storing
634 * The caller is responsible for releasing the value returned with g_free()
637 char *qemu_get_local_state_dir(void);
641 * @type: the auxiliary vector key to lookup
643 * Search the auxiliary vector for @type, returning the value
644 * or 0 if @type is not present.
646 unsigned long qemu_getauxval(unsigned long type
);
648 void qemu_set_tty_echo(int fd
, bool echo
);
650 typedef struct ThreadContext ThreadContext
;
654 * @fd: the fd mapped into the area, -1 for anonymous memory
655 * @area: start address of the are to preallocate
656 * @sz: the size of the area to preallocate
657 * @max_threads: maximum number of threads to use
658 * @errp: returns an error if this function fails
660 * Preallocate memory (populate/prefault page tables writable) for the virtual
661 * memory area starting at @area with the size of @sz. After a successful call,
662 * each page in the area was faulted in writable at least once, for example,
663 * after allocating file blocks for mapped files.
665 void qemu_prealloc_mem(int fd
, char *area
, size_t sz
, int max_threads
,
666 ThreadContext
*tc
, Error
**errp
);
670 * @pid: pid of a process
672 * For given @pid fetch its name. Caller is responsible for
673 * freeing the string when no longer needed.
674 * Returns allocated string on success, NULL on failure.
676 char *qemu_get_pid_name(pid_t pid
);
678 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
679 * when intptr_t is 32-bit and we are aligning a long long.
681 static inline uintptr_t qemu_real_host_page_size(void)
683 return getpagesize();
686 static inline intptr_t qemu_real_host_page_mask(void)
688 return -(intptr_t)qemu_real_host_page_size();
692 * After using getopt or getopt_long, if you need to parse another set
693 * of options, then you must reset optind. Unfortunately the way to
694 * do this varies between implementations of getopt.
696 static inline void qemu_reset_optind(void)
706 int qemu_fdatasync(int fd
);
709 * Sync changes made to the memory mapped file back to the backing
710 * storage. For POSIX compliant systems this will fallback
711 * to regular msync call. Otherwise it will trigger whole file sync
712 * (including the metadata case there is no support to skip that otherwise)
714 * @addr - start of the memory area to be synced
715 * @length - length of the are to be synced
716 * @fd - file descriptor for the file to be synced
717 * (mandatory only for POSIX non-compliant systems)
719 int qemu_msync(void *addr
, size_t length
, int fd
);
722 * qemu_get_host_physmem:
724 * Operating system agnostic way of querying host memory.
726 * Returns amount of physical memory on the system. This is purely
727 * advisery and may return 0 if we can't work it out. At the other
728 * end we saturate to SIZE_MAX if you are lucky enough to have that
731 size_t qemu_get_host_physmem(void);
734 * Toggle write/execute on the pages marked MAP_JIT
735 * for the current thread.
737 #if defined(MAC_OS_VERSION_11_0) && \
738 MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_VERSION_11_0
739 static inline void qemu_thread_jit_execute(void)
741 pthread_jit_write_protect_np(true);
744 static inline void qemu_thread_jit_write(void)
746 pthread_jit_write_protect_np(false);
749 static inline void qemu_thread_jit_write(void) {}
750 static inline void qemu_thread_jit_execute(void) {}
754 * Platforms which do not support system() return ENOSYS
756 #ifndef HAVE_SYSTEM_FUNCTION
757 #define system platform_does_not_support_system
758 static inline int platform_does_not_support_system(const char *command
)
763 #endif /* !HAVE_SYSTEM_FUNCTION */