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 0x0601 /* Windows 7 API (should be in sync with 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
95 #include <sys/types.h>
101 #include <inttypes.h>
103 /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r
104 * function availability on recentish Mingw-w64 platforms. */
111 #include <sys/stat.h>
112 #include <sys/time.h>
114 /* setjmp must be declared before sysemu/os-win32.h
115 * because it is redefined there. */
123 #if defined(__linux__) && defined(__sparc__)
124 /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */
129 #include <sys/wait.h>
131 #define WIFEXITED(x) 1
132 #define WEXITSTATUS(x) (x)
136 #include <AvailabilityMacros.h>
140 * This is somewhat like a system header; it must be outside any extern "C"
141 * block because it includes system headers itself, including glib.h,
142 * which will not compile if inside an extern "C" block.
144 #include "glib-compat.h"
147 #include "sysemu/os-win32.h"
151 #include "sysemu/os-posix.h"
158 #include "qemu/typedefs.h"
161 * Mark a function that executes in coroutine context
163 * Functions that execute in coroutine context cannot be called directly from
164 * normal functions. In the future it would be nice to enable compiler or
165 * static checker support for catching such errors. This annotation might make
166 * it possible and in the meantime it serves as documentation.
170 * static void coroutine_fn foo(void) {
177 * For mingw, as of v6.0.0, the function implementing the assert macro is
178 * not marked as noreturn, so the compiler cannot delete code following an
179 * assert(false) as unused. We rely on this within the code base to delete
180 * code that is unreachable when features are disabled.
181 * All supported versions of Glib's g_assert() satisfy this requirement.
185 #define assert(x) g_assert(x)
189 * qemu_build_not_reached()
191 * The compiler, during optimization, is expected to prove that a call
192 * to this function cannot be reached and remove it. If the compiler
193 * supports QEMU_ERROR, this will be reported at compile time; otherwise
194 * this will be reported at link time due to the missing symbol.
197 void QEMU_ERROR("code path is reachable")
198 qemu_build_not_reached_always(void);
199 #if defined(__OPTIMIZE__) && !defined(__NO_INLINE__)
200 #define qemu_build_not_reached() qemu_build_not_reached_always()
202 #define qemu_build_not_reached() g_assert_not_reached()
206 * qemu_build_assert()
208 * The compiler, during optimization, is expected to prove that the
211 #define qemu_build_assert(test) while (!(test)) qemu_build_not_reached()
214 * According to waitpid man page:
216 * This macro is not specified in POSIX.1-2001 and is not
217 * available on some UNIX implementations (e.g., AIX, SunOS).
218 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
221 #define WCOREDUMP(status) 0
224 * We have a lot of unaudited code that may fail in strange ways, or
225 * even be a security risk during migration, if you disable assertions
226 * at compile-time. You may comment out these safety checks if you
227 * absolutely want to disable assertion overhead, but it is not
228 * supported upstream so the risk is all yours. Meanwhile, please
229 * submit patches to remove any side-effects inside an assertion, or
230 * fixing error handling that should use Error instead of assert.
233 #error building with NDEBUG is not supported
235 #ifdef G_DISABLE_ASSERT
236 #error building with G_DISABLE_ASSERT is not supported
240 #define O_LARGEFILE 0
245 #ifndef MAP_ANONYMOUS
246 #define MAP_ANONYMOUS MAP_ANON
248 #ifndef MAP_FIXED_NOREPLACE
249 #define MAP_FIXED_NOREPLACE 0
251 #ifndef MAP_NORESERVE
252 #define MAP_NORESERVE 0
255 #define ENOMEDIUM ENODEV
257 #if !defined(ENOTSUP)
260 #if !defined(ECANCELED)
261 #define ECANCELED 4097
263 #if !defined(EMEDIUMTYPE)
264 #define EMEDIUMTYPE 4098
266 #if !defined(ESHUTDOWN)
267 #define ESHUTDOWN 4099
270 #define RETRY_ON_EINTR(expr) \
272 ({ typeof(expr) __result; \
275 } while (__result == -1 && errno == EINTR); \
278 /* time_t may be either 32 or 64 bits depending on the host OS, and
279 * can be either signed or unsigned, so we can't just hardcode a
280 * specific maximum value. This is not a C preprocessor constant,
281 * so you can't use TIME_MAX in an #ifdef, but for our purposes
282 * this isn't a problem.
285 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
286 * Gnulib, and are under the LGPL v2.1 or (at your option) any
290 /* True if the real type T is signed. */
291 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
293 /* The width in bits of the integer type or expression T.
294 * Padding bits are not supported.
296 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
298 /* The maximum and minimum values for the integer type T. */
299 #define TYPE_MAXIMUM(t) \
300 ((t) (!TYPE_SIGNED(t) \
302 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
305 #define TIME_MAX TYPE_MAXIMUM(time_t)
308 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
309 * the wrong type. Our replacement isn't usable in preprocessor
310 * expressions, but it is sufficient for our needs. */
311 #ifdef HAVE_BROKEN_SIZE_MAX
313 #define SIZE_MAX ((size_t)-1)
317 * Two variations of MIN/MAX macros. The first is for runtime use, and
318 * evaluates arguments only once (so it is safe even with side
319 * effects), but will not work in constant contexts (such as array
320 * size declarations) because of the '{}'. The second is for constant
321 * expression use, where evaluating arguments twice is safe because
322 * the result is going to be constant anyway, but will not work in a
323 * runtime context because of a void expression where a value is
324 * expected. Thus, both gcc and clang will fail to compile if you use
325 * the wrong macro (even if the error may seem a bit cryptic).
327 * Note that neither form is usable as an #if condition; if you truly
328 * need to write conditional code that depends on a minimum or maximum
329 * determined by the pre-processor instead of the compiler, you'll
330 * have to open-code it. Sadly, Coverity is severely confused by the
331 * constant variants, so we have to dumb things down there.
336 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
342 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
347 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
348 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
350 # define MIN_CONST(a, b) \
351 __builtin_choose_expr( \
352 __builtin_constant_p(a) && __builtin_constant_p(b), \
353 (a) < (b) ? (a) : (b), \
355 # define MAX_CONST(a, b) \
356 __builtin_choose_expr( \
357 __builtin_constant_p(a) && __builtin_constant_p(b), \
358 (a) > (b) ? (a) : (b), \
363 * Minimum function that returns zero only if both values are zero.
364 * Intended for use with unsigned values only.
367 #define MIN_NON_ZERO(a, b) \
369 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
370 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
375 * Round number down to multiple. Safe when m is not a power of 2 (see
376 * ROUND_DOWN for a faster version when a power of 2 is guaranteed).
378 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
381 * Round number up to multiple. Safe when m is not a power of 2 (see
382 * ROUND_UP for a faster version when a power of 2 is guaranteed).
384 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
386 /* Check if n is a multiple of m */
387 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
389 /* n-byte align pointer down */
390 #define QEMU_ALIGN_PTR_DOWN(p, n) \
391 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
393 /* n-byte align pointer up */
394 #define QEMU_ALIGN_PTR_UP(p, n) \
395 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
397 /* Check if pointer p is n-bytes aligned */
398 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
401 * Round number down to multiple. Requires that d be a power of 2 (see
402 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
403 * numbers); works even if d is a smaller type than n.
406 #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d)))
410 * Round number up to multiple. Requires that d be a power of 2 (see
411 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
412 * numbers); works even if d is a smaller type than n.
415 #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d))
419 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
423 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
424 * pointer, not an array.
426 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
429 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
430 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
433 int qemu_daemon(int nochdir
, int noclose
);
434 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
,
436 void qemu_anon_ram_free(void *ptr
, size_t size
);
439 #define HAVE_CHARDEV_SERIAL 1
440 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
441 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
442 || defined(__GLIBC__) || defined(__APPLE__)
443 #define HAVE_CHARDEV_SERIAL 1
446 #if defined(__HAIKU__)
447 #define SIGIO SIGPOLL
450 #ifdef HAVE_MADVISE_WITHOUT_PROTOTYPE
452 * See MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156) for discussion
453 * about Solaris missing the madvise() prototype.
455 extern int madvise(char *, size_t, int);
458 #if defined(CONFIG_LINUX)
459 #ifndef BUS_MCEERR_AR
460 #define BUS_MCEERR_AR 4
462 #ifndef BUS_MCEERR_AO
463 #define BUS_MCEERR_AO 5
467 #if defined(__linux__) && \
468 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
469 || defined(__powerpc64__))
470 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
471 Valgrind does not support alignments larger than 1 MiB,
472 therefore we need special code which handles running on Valgrind. */
473 # define QEMU_VMALLOC_ALIGN (512 * 4096)
474 #elif defined(__linux__) && defined(__s390x__)
475 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
476 # define QEMU_VMALLOC_ALIGN (256 * 4096)
477 #elif defined(__linux__) && defined(__sparc__)
478 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size(), SHMLBA)
480 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size()
484 struct qemu_signalfd_siginfo
{
485 uint32_t ssi_signo
; /* Signal number */
486 int32_t ssi_errno
; /* Error number (unused) */
487 int32_t ssi_code
; /* Signal code */
488 uint32_t ssi_pid
; /* PID of sender */
489 uint32_t ssi_uid
; /* Real UID of sender */
490 int32_t ssi_fd
; /* File descriptor (SIGIO) */
491 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
492 uint32_t ssi_band
; /* Band event (SIGIO) */
493 uint32_t ssi_overrun
; /* POSIX timer overrun count */
494 uint32_t ssi_trapno
; /* Trap number that caused signal */
495 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
496 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
497 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
498 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
499 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
500 uint64_t ssi_addr
; /* Address that generated signal
501 (for hardware-generated signals) */
502 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
503 additional fields in the future) */
506 int qemu_signalfd(const sigset_t
*mask
);
507 void sigaction_invoke(struct sigaction
*action
,
508 struct qemu_signalfd_siginfo
*info
);
512 * Don't introduce new usage of this function, prefer the following
513 * qemu_open/qemu_create that take an "Error **errp"
515 int qemu_open_old(const char *name
, int flags
, ...);
516 int qemu_open(const char *name
, int flags
, Error
**errp
);
517 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
518 int qemu_close(int fd
);
519 int qemu_unlink(const char *name
);
521 int qemu_dup_flags(int fd
, int flags
);
522 int qemu_dup(int fd
);
523 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
524 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
525 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
526 bool qemu_has_ofd_lock(void);
529 #if defined(__HAIKU__) && defined(__i386__)
530 #define FMT_pid "%ld"
532 #define FMT_pid "%" PRId64
537 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
539 int qemu_get_thread_id(void);
547 * Use the same value as Linux for now.
551 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
552 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
556 static inline void qemu_timersub(const struct timeval
*val1
,
557 const struct timeval
*val2
,
560 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
561 if (val1
->tv_usec
< val2
->tv_usec
) {
563 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
565 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
569 #define qemu_timersub timersub
572 ssize_t
qemu_write_full(int fd
, const void *buf
, size_t count
)
573 G_GNUC_WARN_UNUSED_RESULT
;
575 void qemu_set_cloexec(int fd
);
577 /* Return a dynamically allocated directory path that is appropriate for storing
580 * The caller is responsible for releasing the value returned with g_free()
583 char *qemu_get_local_state_dir(void);
587 * @type: the auxiliary vector key to lookup
589 * Search the auxiliary vector for @type, returning the value
590 * or 0 if @type is not present.
592 unsigned long qemu_getauxval(unsigned long type
);
594 void qemu_set_tty_echo(int fd
, bool echo
);
596 typedef struct ThreadContext ThreadContext
;
600 * @fd: the fd mapped into the area, -1 for anonymous memory
601 * @area: start address of the are to preallocate
602 * @sz: the size of the area to preallocate
603 * @max_threads: maximum number of threads to use
604 * @errp: returns an error if this function fails
606 * Preallocate memory (populate/prefault page tables writable) for the virtual
607 * memory area starting at @area with the size of @sz. After a successful call,
608 * each page in the area was faulted in writable at least once, for example,
609 * after allocating file blocks for mapped files.
611 void qemu_prealloc_mem(int fd
, char *area
, size_t sz
, int max_threads
,
612 ThreadContext
*tc
, Error
**errp
);
616 * @pid: pid of a process
618 * For given @pid fetch its name. Caller is responsible for
619 * freeing the string when no longer needed.
620 * Returns allocated string on success, NULL on failure.
622 char *qemu_get_pid_name(pid_t pid
);
627 * A version of fork that avoids signal handler race
628 * conditions that can lead to child process getting
629 * signals that are otherwise only expected by the
630 * parent. It also resets all signal handlers to the
633 * Returns 0 to child process, pid number to parent
636 pid_t
qemu_fork(Error
**errp
);
638 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
639 * when intptr_t is 32-bit and we are aligning a long long.
641 static inline uintptr_t qemu_real_host_page_size(void)
643 return getpagesize();
646 static inline intptr_t qemu_real_host_page_mask(void)
648 return -(intptr_t)qemu_real_host_page_size();
652 * After using getopt or getopt_long, if you need to parse another set
653 * of options, then you must reset optind. Unfortunately the way to
654 * do this varies between implementations of getopt.
656 static inline void qemu_reset_optind(void)
666 int qemu_fdatasync(int fd
);
669 * Sync changes made to the memory mapped file back to the backing
670 * storage. For POSIX compliant systems this will fallback
671 * to regular msync call. Otherwise it will trigger whole file sync
672 * (including the metadata case there is no support to skip that otherwise)
674 * @addr - start of the memory area to be synced
675 * @length - length of the are to be synced
676 * @fd - file descriptor for the file to be synced
677 * (mandatory only for POSIX non-compliant systems)
679 int qemu_msync(void *addr
, size_t length
, int fd
);
682 * qemu_get_host_physmem:
684 * Operating system agnostic way of querying host memory.
686 * Returns amount of physical memory on the system. This is purely
687 * advisery and may return 0 if we can't work it out. At the other
688 * end we saturate to SIZE_MAX if you are lucky enough to have that
691 size_t qemu_get_host_physmem(void);
694 * Toggle write/execute on the pages marked MAP_JIT
695 * for the current thread.
697 #if defined(MAC_OS_VERSION_11_0) && \
698 MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_VERSION_11_0
699 static inline void qemu_thread_jit_execute(void)
701 pthread_jit_write_protect_np(true);
704 static inline void qemu_thread_jit_write(void)
706 pthread_jit_write_protect_np(false);
709 static inline void qemu_thread_jit_write(void) {}
710 static inline void qemu_thread_jit_execute(void) {}
714 * Platforms which do not support system() return ENOSYS
716 #ifndef HAVE_SYSTEM_FUNCTION
717 #define system platform_does_not_support_system
718 static inline int platform_does_not_support_system(const char *command
)
723 #endif /* !HAVE_SYSTEM_FUNCTION */