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
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) {
175 #define coroutine_fn __attribute__((__annotate__("coroutine_fn")))
181 * Mark a function that can suspend when executed in coroutine context,
182 * but can handle running in non-coroutine context too.
185 #define coroutine_mixed_fn __attribute__((__annotate__("coroutine_mixed_fn")))
187 #define coroutine_mixed_fn
191 * Mark a function that should not be called from a coroutine context.
192 * Usually there will be an analogous, coroutine_fn function that should
195 * When the function is also marked as coroutine_mixed_fn, the function should
196 * only be called if the caller does not know whether it is in coroutine
199 * Functions that are only no_coroutine_fn, on the other hand, should not
200 * be called from within coroutines at all. This for example includes
201 * functions that block.
203 * In the future it would be nice to enable compiler or static checker
204 * support for catching such errors. This annotation is the first step
205 * towards this, and in the meantime it serves as documentation.
209 * static void no_coroutine_fn foo(void) {
214 #define no_coroutine_fn __attribute__((__annotate__("no_coroutine_fn")))
216 #define no_coroutine_fn
221 * For mingw, as of v6.0.0, the function implementing the assert macro is
222 * not marked as noreturn, so the compiler cannot delete code following an
223 * assert(false) as unused. We rely on this within the code base to delete
224 * code that is unreachable when features are disabled.
225 * All supported versions of Glib's g_assert() satisfy this requirement.
229 #define assert(x) g_assert(x)
233 * qemu_build_not_reached()
235 * The compiler, during optimization, is expected to prove that a call
236 * to this function cannot be reached and remove it. If the compiler
237 * supports QEMU_ERROR, this will be reported at compile time; otherwise
238 * this will be reported at link time due to the missing symbol.
241 void QEMU_ERROR("code path is reachable")
242 qemu_build_not_reached_always(void);
243 #if defined(__OPTIMIZE__) && !defined(__NO_INLINE__)
244 #define qemu_build_not_reached() qemu_build_not_reached_always()
246 #define qemu_build_not_reached() g_assert_not_reached()
250 * qemu_build_assert()
252 * The compiler, during optimization, is expected to prove that the
255 #define qemu_build_assert(test) while (!(test)) qemu_build_not_reached()
258 * According to waitpid man page:
260 * This macro is not specified in POSIX.1-2001 and is not
261 * available on some UNIX implementations (e.g., AIX, SunOS).
262 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
265 #define WCOREDUMP(status) 0
268 * We have a lot of unaudited code that may fail in strange ways, or
269 * even be a security risk during migration, if you disable assertions
270 * at compile-time. You may comment out these safety checks if you
271 * absolutely want to disable assertion overhead, but it is not
272 * supported upstream so the risk is all yours. Meanwhile, please
273 * submit patches to remove any side-effects inside an assertion, or
274 * fixing error handling that should use Error instead of assert.
277 #error building with NDEBUG is not supported
279 #ifdef G_DISABLE_ASSERT
280 #error building with G_DISABLE_ASSERT is not supported
284 #define O_LARGEFILE 0
289 #ifndef MAP_ANONYMOUS
290 #define MAP_ANONYMOUS MAP_ANON
292 #ifndef MAP_FIXED_NOREPLACE
293 #define MAP_FIXED_NOREPLACE 0
295 #ifndef MAP_NORESERVE
296 #define MAP_NORESERVE 0
299 #define ENOMEDIUM ENODEV
301 #if !defined(ENOTSUP)
304 #if !defined(ECANCELED)
305 #define ECANCELED 4097
307 #if !defined(EMEDIUMTYPE)
308 #define EMEDIUMTYPE 4098
310 #if !defined(ESHUTDOWN)
311 #define ESHUTDOWN 4099
314 #define RETRY_ON_EINTR(expr) \
316 ({ typeof(expr) __result; \
319 } while (__result == -1 && errno == EINTR); \
322 /* time_t may be either 32 or 64 bits depending on the host OS, and
323 * can be either signed or unsigned, so we can't just hardcode a
324 * specific maximum value. This is not a C preprocessor constant,
325 * so you can't use TIME_MAX in an #ifdef, but for our purposes
326 * this isn't a problem.
329 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
330 * Gnulib, and are under the LGPL v2.1 or (at your option) any
334 /* True if the real type T is signed. */
335 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
337 /* The width in bits of the integer type or expression T.
338 * Padding bits are not supported.
340 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
342 /* The maximum and minimum values for the integer type T. */
343 #define TYPE_MAXIMUM(t) \
344 ((t) (!TYPE_SIGNED(t) \
346 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
349 #define TIME_MAX TYPE_MAXIMUM(time_t)
352 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
353 * the wrong type. Our replacement isn't usable in preprocessor
354 * expressions, but it is sufficient for our needs. */
355 #ifdef HAVE_BROKEN_SIZE_MAX
357 #define SIZE_MAX ((size_t)-1)
361 * Two variations of MIN/MAX macros. The first is for runtime use, and
362 * evaluates arguments only once (so it is safe even with side
363 * effects), but will not work in constant contexts (such as array
364 * size declarations) because of the '{}'. The second is for constant
365 * expression use, where evaluating arguments twice is safe because
366 * the result is going to be constant anyway, but will not work in a
367 * runtime context because of a void expression where a value is
368 * expected. Thus, both gcc and clang will fail to compile if you use
369 * the wrong macro (even if the error may seem a bit cryptic).
371 * Note that neither form is usable as an #if condition; if you truly
372 * need to write conditional code that depends on a minimum or maximum
373 * determined by the pre-processor instead of the compiler, you'll
374 * have to open-code it. Sadly, Coverity is severely confused by the
375 * constant variants, so we have to dumb things down there.
380 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
386 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
391 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
392 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
394 # define MIN_CONST(a, b) \
395 __builtin_choose_expr( \
396 __builtin_constant_p(a) && __builtin_constant_p(b), \
397 (a) < (b) ? (a) : (b), \
399 # define MAX_CONST(a, b) \
400 __builtin_choose_expr( \
401 __builtin_constant_p(a) && __builtin_constant_p(b), \
402 (a) > (b) ? (a) : (b), \
407 * Minimum function that returns zero only if both values are zero.
408 * Intended for use with unsigned values only.
411 #define MIN_NON_ZERO(a, b) \
413 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
414 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
419 * Round number down to multiple. Safe when m is not a power of 2 (see
420 * ROUND_DOWN for a faster version when a power of 2 is guaranteed).
422 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
425 * Round number up to multiple. Safe when m is not a power of 2 (see
426 * ROUND_UP for a faster version when a power of 2 is guaranteed).
428 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
430 /* Check if n is a multiple of m */
431 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
433 /* n-byte align pointer down */
434 #define QEMU_ALIGN_PTR_DOWN(p, n) \
435 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
437 /* n-byte align pointer up */
438 #define QEMU_ALIGN_PTR_UP(p, n) \
439 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
441 /* Check if pointer p is n-bytes aligned */
442 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
445 * Round number down to multiple. Requires that d be a power of 2 (see
446 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
447 * numbers); works even if d is a smaller type than n.
450 #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d)))
454 * Round number up to multiple. Requires that d be a power of 2 (see
455 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
456 * numbers); works even if d is a smaller type than n.
459 #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d))
463 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
467 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
468 * pointer, not an array.
470 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
473 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
474 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
477 int qemu_daemon(int nochdir
, int noclose
);
478 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
,
480 void qemu_anon_ram_free(void *ptr
, size_t size
);
483 #define HAVE_CHARDEV_SERIAL 1
484 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
485 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
486 || defined(__GLIBC__) || defined(__APPLE__)
487 #define HAVE_CHARDEV_SERIAL 1
490 #if defined(__HAIKU__)
491 #define SIGIO SIGPOLL
494 #ifdef HAVE_MADVISE_WITHOUT_PROTOTYPE
496 * See MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156) for discussion
497 * about Solaris missing the madvise() prototype.
499 extern int madvise(char *, size_t, int);
502 #if defined(CONFIG_LINUX)
503 #ifndef BUS_MCEERR_AR
504 #define BUS_MCEERR_AR 4
506 #ifndef BUS_MCEERR_AO
507 #define BUS_MCEERR_AO 5
511 #if defined(__linux__) && \
512 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
513 || defined(__powerpc64__))
514 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
515 Valgrind does not support alignments larger than 1 MiB,
516 therefore we need special code which handles running on Valgrind. */
517 # define QEMU_VMALLOC_ALIGN (512 * 4096)
518 #elif defined(__linux__) && defined(__s390x__)
519 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
520 # define QEMU_VMALLOC_ALIGN (256 * 4096)
521 #elif defined(__linux__) && defined(__sparc__)
522 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size(), SHMLBA)
524 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size()
528 struct qemu_signalfd_siginfo
{
529 uint32_t ssi_signo
; /* Signal number */
530 int32_t ssi_errno
; /* Error number (unused) */
531 int32_t ssi_code
; /* Signal code */
532 uint32_t ssi_pid
; /* PID of sender */
533 uint32_t ssi_uid
; /* Real UID of sender */
534 int32_t ssi_fd
; /* File descriptor (SIGIO) */
535 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
536 uint32_t ssi_band
; /* Band event (SIGIO) */
537 uint32_t ssi_overrun
; /* POSIX timer overrun count */
538 uint32_t ssi_trapno
; /* Trap number that caused signal */
539 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
540 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
541 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
542 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
543 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
544 uint64_t ssi_addr
; /* Address that generated signal
545 (for hardware-generated signals) */
546 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
547 additional fields in the future) */
550 int qemu_signalfd(const sigset_t
*mask
);
551 void sigaction_invoke(struct sigaction
*action
,
552 struct qemu_signalfd_siginfo
*info
);
556 * Don't introduce new usage of this function, prefer the following
557 * qemu_open/qemu_create that take an "Error **errp"
559 int qemu_open_old(const char *name
, int flags
, ...);
560 int qemu_open(const char *name
, int flags
, Error
**errp
);
561 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
562 int qemu_close(int fd
);
563 int qemu_unlink(const char *name
);
565 int qemu_dup_flags(int fd
, int flags
);
566 int qemu_dup(int fd
);
567 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
568 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
569 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
570 bool qemu_has_ofd_lock(void);
573 #if defined(__HAIKU__) && defined(__i386__)
574 #define FMT_pid "%ld"
576 #define FMT_pid "%" PRId64
581 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
583 int qemu_get_thread_id(void);
591 * Use the same value as Linux for now.
595 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
596 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
600 static inline void qemu_timersub(const struct timeval
*val1
,
601 const struct timeval
*val2
,
604 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
605 if (val1
->tv_usec
< val2
->tv_usec
) {
607 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
609 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
613 #define qemu_timersub timersub
616 ssize_t
qemu_write_full(int fd
, const void *buf
, size_t count
)
617 G_GNUC_WARN_UNUSED_RESULT
;
619 void qemu_set_cloexec(int fd
);
621 /* Return a dynamically allocated directory path that is appropriate for storing
624 * The caller is responsible for releasing the value returned with g_free()
627 char *qemu_get_local_state_dir(void);
631 * @type: the auxiliary vector key to lookup
633 * Search the auxiliary vector for @type, returning the value
634 * or 0 if @type is not present.
636 unsigned long qemu_getauxval(unsigned long type
);
638 void qemu_set_tty_echo(int fd
, bool echo
);
640 typedef struct ThreadContext ThreadContext
;
644 * @fd: the fd mapped into the area, -1 for anonymous memory
645 * @area: start address of the are to preallocate
646 * @sz: the size of the area to preallocate
647 * @max_threads: maximum number of threads to use
648 * @errp: returns an error if this function fails
650 * Preallocate memory (populate/prefault page tables writable) for the virtual
651 * memory area starting at @area with the size of @sz. After a successful call,
652 * each page in the area was faulted in writable at least once, for example,
653 * after allocating file blocks for mapped files.
655 void qemu_prealloc_mem(int fd
, char *area
, size_t sz
, int max_threads
,
656 ThreadContext
*tc
, Error
**errp
);
660 * @pid: pid of a process
662 * For given @pid fetch its name. Caller is responsible for
663 * freeing the string when no longer needed.
664 * Returns allocated string on success, NULL on failure.
666 char *qemu_get_pid_name(pid_t pid
);
668 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
669 * when intptr_t is 32-bit and we are aligning a long long.
671 static inline uintptr_t qemu_real_host_page_size(void)
673 return getpagesize();
676 static inline intptr_t qemu_real_host_page_mask(void)
678 return -(intptr_t)qemu_real_host_page_size();
682 * After using getopt or getopt_long, if you need to parse another set
683 * of options, then you must reset optind. Unfortunately the way to
684 * do this varies between implementations of getopt.
686 static inline void qemu_reset_optind(void)
696 int qemu_fdatasync(int fd
);
699 * Sync changes made to the memory mapped file back to the backing
700 * storage. For POSIX compliant systems this will fallback
701 * to regular msync call. Otherwise it will trigger whole file sync
702 * (including the metadata case there is no support to skip that otherwise)
704 * @addr - start of the memory area to be synced
705 * @length - length of the are to be synced
706 * @fd - file descriptor for the file to be synced
707 * (mandatory only for POSIX non-compliant systems)
709 int qemu_msync(void *addr
, size_t length
, int fd
);
712 * qemu_get_host_physmem:
714 * Operating system agnostic way of querying host memory.
716 * Returns amount of physical memory on the system. This is purely
717 * advisery and may return 0 if we can't work it out. At the other
718 * end we saturate to SIZE_MAX if you are lucky enough to have that
721 size_t qemu_get_host_physmem(void);
724 * Toggle write/execute on the pages marked MAP_JIT
725 * for the current thread.
727 #if defined(MAC_OS_VERSION_11_0) && \
728 MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_VERSION_11_0
729 static inline void qemu_thread_jit_execute(void)
731 pthread_jit_write_protect_np(true);
734 static inline void qemu_thread_jit_write(void)
736 pthread_jit_write_protect_np(false);
739 static inline void qemu_thread_jit_write(void) {}
740 static inline void qemu_thread_jit_execute(void) {}
744 * Platforms which do not support system() return ENOSYS
746 #ifndef HAVE_SYSTEM_FUNCTION
747 #define system platform_does_not_support_system
748 static inline int platform_does_not_support_system(const char *command
)
753 #endif /* !HAVE_SYSTEM_FUNCTION */