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"
37 #include "qemu/compiler.h"
39 /* Older versions of C++ don't get definitions of various macros from
40 * stdlib.h unless we define these macros before first inclusion of
43 #ifndef __STDC_CONSTANT_MACROS
44 #define __STDC_CONSTANT_MACROS
46 #ifndef __STDC_LIMIT_MACROS
47 #define __STDC_LIMIT_MACROS
49 #ifndef __STDC_FORMAT_MACROS
50 #define __STDC_FORMAT_MACROS
53 /* The following block of code temporarily renames the daemon() function so the
54 * compiler does not see the warning associated with it in stdlib.h on OSX
57 #define daemon qemu_fake_daemon_function
60 QEMU_EXTERN_C
int daemon(int, int);
64 /* as defined in sdkddkver.h */
66 #define _WIN32_WINNT 0x0600 /* Vista */
68 /* reduces the number of implicitly included headers */
69 #ifndef WIN32_LEAN_AND_MEAN
70 #define WIN32_LEAN_AND_MEAN
74 /* enable C99/POSIX format strings (needs mingw32-runtime 3.15 or later) */
76 #define __USE_MINGW_ANSI_STDIO 1
83 #include <sys/types.h>
91 /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r
92 * function availability on recentish Mingw-w64 platforms. */
100 #include <sys/time.h>
102 /* setjmp must be declared before sysemu/os-win32.h
103 * because it is redefined there. */
111 #if defined(__linux__) && defined(__sparc__)
112 /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */
117 #include <sys/wait.h>
119 #define WIFEXITED(x) 1
120 #define WEXITSTATUS(x) (x)
124 #include <AvailabilityMacros.h>
128 * This is somewhat like a system header; it must be outside any extern "C"
129 * block because it includes system headers itself, including glib.h,
130 * which will not compile if inside an extern "C" block.
132 #include "glib-compat.h"
135 #include "sysemu/os-win32.h"
139 #include "sysemu/os-posix.h"
146 #include "qemu/typedefs.h"
149 * For mingw, as of v6.0.0, the function implementing the assert macro is
150 * not marked as noreturn, so the compiler cannot delete code following an
151 * assert(false) as unused. We rely on this within the code base to delete
152 * code that is unreachable when features are disabled.
153 * All supported versions of Glib's g_assert() satisfy this requirement.
157 #define assert(x) g_assert(x)
161 * According to waitpid man page:
163 * This macro is not specified in POSIX.1-2001 and is not
164 * available on some UNIX implementations (e.g., AIX, SunOS).
165 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
168 #define WCOREDUMP(status) 0
171 * We have a lot of unaudited code that may fail in strange ways, or
172 * even be a security risk during migration, if you disable assertions
173 * at compile-time. You may comment out these safety checks if you
174 * absolutely want to disable assertion overhead, but it is not
175 * supported upstream so the risk is all yours. Meanwhile, please
176 * submit patches to remove any side-effects inside an assertion, or
177 * fixing error handling that should use Error instead of assert.
180 #error building with NDEBUG is not supported
182 #ifdef G_DISABLE_ASSERT
183 #error building with G_DISABLE_ASSERT is not supported
187 #define O_LARGEFILE 0
192 #ifndef MAP_ANONYMOUS
193 #define MAP_ANONYMOUS MAP_ANON
195 #ifndef MAP_FIXED_NOREPLACE
196 #define MAP_FIXED_NOREPLACE 0
199 #define ENOMEDIUM ENODEV
201 #if !defined(ENOTSUP)
204 #if !defined(ECANCELED)
205 #define ECANCELED 4097
207 #if !defined(EMEDIUMTYPE)
208 #define EMEDIUMTYPE 4098
210 #if !defined(ESHUTDOWN)
211 #define ESHUTDOWN 4099
214 /* time_t may be either 32 or 64 bits depending on the host OS, and
215 * can be either signed or unsigned, so we can't just hardcode a
216 * specific maximum value. This is not a C preprocessor constant,
217 * so you can't use TIME_MAX in an #ifdef, but for our purposes
218 * this isn't a problem.
221 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
222 * Gnulib, and are under the LGPL v2.1 or (at your option) any
226 /* True if the real type T is signed. */
227 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
229 /* The width in bits of the integer type or expression T.
230 * Padding bits are not supported.
232 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
234 /* The maximum and minimum values for the integer type T. */
235 #define TYPE_MAXIMUM(t) \
236 ((t) (!TYPE_SIGNED(t) \
238 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
241 #define TIME_MAX TYPE_MAXIMUM(time_t)
244 /* HOST_LONG_BITS is the size of a native pointer in bits. */
245 #if UINTPTR_MAX == UINT32_MAX
246 # define HOST_LONG_BITS 32
247 #elif UINTPTR_MAX == UINT64_MAX
248 # define HOST_LONG_BITS 64
250 # error Unknown pointer size
253 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
254 * the wrong type. Our replacement isn't usable in preprocessor
255 * expressions, but it is sufficient for our needs. */
256 #if defined(HAVE_BROKEN_SIZE_MAX) && HAVE_BROKEN_SIZE_MAX
258 #define SIZE_MAX ((size_t)-1)
262 * Two variations of MIN/MAX macros. The first is for runtime use, and
263 * evaluates arguments only once (so it is safe even with side
264 * effects), but will not work in constant contexts (such as array
265 * size declarations) because of the '{}'. The second is for constant
266 * expression use, where evaluating arguments twice is safe because
267 * the result is going to be constant anyway, but will not work in a
268 * runtime context because of a void expression where a value is
269 * expected. Thus, both gcc and clang will fail to compile if you use
270 * the wrong macro (even if the error may seem a bit cryptic).
272 * Note that neither form is usable as an #if condition; if you truly
273 * need to write conditional code that depends on a minimum or maximum
274 * determined by the pre-processor instead of the compiler, you'll
275 * have to open-code it. Sadly, Coverity is severely confused by the
276 * constant variants, so we have to dumb things down there.
281 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
287 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
292 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
293 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
295 # define MIN_CONST(a, b) \
296 __builtin_choose_expr( \
297 __builtin_constant_p(a) && __builtin_constant_p(b), \
298 (a) < (b) ? (a) : (b), \
300 # define MAX_CONST(a, b) \
301 __builtin_choose_expr( \
302 __builtin_constant_p(a) && __builtin_constant_p(b), \
303 (a) > (b) ? (a) : (b), \
308 * Minimum function that returns zero only if both values are zero.
309 * Intended for use with unsigned values only.
312 #define MIN_NON_ZERO(a, b) \
314 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
315 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
319 /* Round number down to multiple */
320 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
322 /* Round number up to multiple. Safe when m is not a power of 2 (see
323 * ROUND_UP for a faster version when a power of 2 is guaranteed) */
324 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
326 /* Check if n is a multiple of m */
327 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
329 /* n-byte align pointer down */
330 #define QEMU_ALIGN_PTR_DOWN(p, n) \
331 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
333 /* n-byte align pointer up */
334 #define QEMU_ALIGN_PTR_UP(p, n) \
335 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
337 /* Check if pointer p is n-bytes aligned */
338 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
340 /* Round number up to multiple. Requires that d be a power of 2 (see
341 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
342 * numbers); works even if d is a smaller type than n. */
344 #define ROUND_UP(n, d) (((n) + (d) - 1) & -(0 ? (n) : (d)))
348 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
352 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
353 * pointer, not an array.
355 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
358 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
359 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
362 int qemu_daemon(int nochdir
, int noclose
);
363 void *qemu_try_memalign(size_t alignment
, size_t size
);
364 void *qemu_memalign(size_t alignment
, size_t size
);
365 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
);
366 void qemu_vfree(void *ptr
);
367 void qemu_anon_ram_free(void *ptr
, size_t size
);
369 #define QEMU_MADV_INVALID -1
371 #if defined(CONFIG_MADVISE)
373 #define QEMU_MADV_WILLNEED MADV_WILLNEED
374 #define QEMU_MADV_DONTNEED MADV_DONTNEED
376 #define QEMU_MADV_DONTFORK MADV_DONTFORK
378 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
380 #ifdef MADV_MERGEABLE
381 #define QEMU_MADV_MERGEABLE MADV_MERGEABLE
383 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
385 #ifdef MADV_UNMERGEABLE
386 #define QEMU_MADV_UNMERGEABLE MADV_UNMERGEABLE
388 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
391 #define QEMU_MADV_DODUMP MADV_DODUMP
393 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
396 #define QEMU_MADV_DONTDUMP MADV_DONTDUMP
398 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
401 #define QEMU_MADV_HUGEPAGE MADV_HUGEPAGE
403 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
405 #ifdef MADV_NOHUGEPAGE
406 #define QEMU_MADV_NOHUGEPAGE MADV_NOHUGEPAGE
408 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
411 #define QEMU_MADV_REMOVE MADV_REMOVE
413 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
416 #elif defined(CONFIG_POSIX_MADVISE)
418 #define QEMU_MADV_WILLNEED POSIX_MADV_WILLNEED
419 #define QEMU_MADV_DONTNEED POSIX_MADV_DONTNEED
420 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
421 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
422 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
423 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
424 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
425 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
426 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
427 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
431 #define QEMU_MADV_WILLNEED QEMU_MADV_INVALID
432 #define QEMU_MADV_DONTNEED QEMU_MADV_INVALID
433 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
434 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
435 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
436 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
437 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
438 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
439 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
440 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
445 #define HAVE_CHARDEV_SERIAL 1
446 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
447 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
448 || defined(__GLIBC__) || defined(__APPLE__)
449 #define HAVE_CHARDEV_SERIAL 1
452 #if defined(__linux__) || defined(__FreeBSD__) || \
453 defined(__FreeBSD_kernel__) || defined(__DragonFly__)
454 #define HAVE_CHARDEV_PARPORT 1
457 #if defined(__HAIKU__)
458 #define SIGIO SIGPOLL
461 #if defined(CONFIG_LINUX)
462 #ifndef BUS_MCEERR_AR
463 #define BUS_MCEERR_AR 4
465 #ifndef BUS_MCEERR_AO
466 #define BUS_MCEERR_AO 5
470 #if defined(__linux__) && \
471 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
472 || defined(__powerpc64__))
473 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
474 Valgrind does not support alignments larger than 1 MiB,
475 therefore we need special code which handles running on Valgrind. */
476 # define QEMU_VMALLOC_ALIGN (512 * 4096)
477 #elif defined(__linux__) && defined(__s390x__)
478 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
479 # define QEMU_VMALLOC_ALIGN (256 * 4096)
480 #elif defined(__linux__) && defined(__sparc__)
481 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size, SHMLBA)
483 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size
487 struct qemu_signalfd_siginfo
{
488 uint32_t ssi_signo
; /* Signal number */
489 int32_t ssi_errno
; /* Error number (unused) */
490 int32_t ssi_code
; /* Signal code */
491 uint32_t ssi_pid
; /* PID of sender */
492 uint32_t ssi_uid
; /* Real UID of sender */
493 int32_t ssi_fd
; /* File descriptor (SIGIO) */
494 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
495 uint32_t ssi_band
; /* Band event (SIGIO) */
496 uint32_t ssi_overrun
; /* POSIX timer overrun count */
497 uint32_t ssi_trapno
; /* Trap number that caused signal */
498 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
499 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
500 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
501 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
502 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
503 uint64_t ssi_addr
; /* Address that generated signal
504 (for hardware-generated signals) */
505 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
506 additional fields in the future) */
509 int qemu_signalfd(const sigset_t
*mask
);
510 void sigaction_invoke(struct sigaction
*action
,
511 struct qemu_signalfd_siginfo
*info
);
514 int qemu_madvise(void *addr
, size_t len
, int advice
);
515 int qemu_mprotect_rwx(void *addr
, size_t size
);
516 int qemu_mprotect_none(void *addr
, size_t size
);
519 * Don't introduce new usage of this function, prefer the following
520 * qemu_open/qemu_create that take an "Error **errp"
522 int qemu_open_old(const char *name
, int flags
, ...);
523 int qemu_open(const char *name
, int flags
, Error
**errp
);
524 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
525 int qemu_close(int fd
);
526 int qemu_unlink(const char *name
);
528 int qemu_dup_flags(int fd
, int flags
);
529 int qemu_dup(int fd
);
530 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
531 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
532 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
533 bool qemu_has_ofd_lock(void);
536 #if defined(__HAIKU__) && defined(__i386__)
537 #define FMT_pid "%ld"
539 #define FMT_pid "%" PRId64
544 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
546 int qemu_get_thread_id(void);
554 * Use the same value as Linux for now.
558 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
559 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
563 static inline void qemu_timersub(const struct timeval
*val1
,
564 const struct timeval
*val2
,
567 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
568 if (val1
->tv_usec
< val2
->tv_usec
) {
570 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
572 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
576 #define qemu_timersub timersub
579 void qemu_set_cloexec(int fd
);
581 /* Starting on QEMU 2.5, qemu_hw_version() returns "2.5+" by default
582 * instead of QEMU_VERSION, so setting hw_version on MachineClass
583 * is no longer mandatory.
585 * Do NOT change this string, or it will break compatibility on all
586 * machine classes that don't set hw_version.
588 #define QEMU_HW_VERSION "2.5+"
590 /* QEMU "hardware version" setting. Used to replace code that exposed
591 * QEMU_VERSION to guests in the past and need to keep compatibility.
592 * Do not use qemu_hw_version() in new code.
594 void qemu_set_hw_version(const char *);
595 const char *qemu_hw_version(void);
597 void fips_set_state(bool requested
);
598 bool fips_get_state(void);
600 /* Return a dynamically allocated pathname denoting a file or directory that is
601 * appropriate for storing local state.
603 * @relative_pathname need not start with a directory separator; one will be
604 * added automatically.
606 * The caller is responsible for releasing the value returned with g_free()
609 char *qemu_get_local_state_pathname(const char *relative_pathname
);
611 /* Find program directory, and save it for later usage with
612 * qemu_get_exec_dir().
613 * Try OS specific API first, if not working, parse from argv0. */
614 void qemu_init_exec_dir(const char *argv0
);
616 /* Get the saved exec dir. */
617 const char *qemu_get_exec_dir(void);
621 * @type: the auxiliary vector key to lookup
623 * Search the auxiliary vector for @type, returning the value
624 * or 0 if @type is not present.
626 unsigned long qemu_getauxval(unsigned long type
);
628 void qemu_set_tty_echo(int fd
, bool echo
);
630 void os_mem_prealloc(int fd
, char *area
, size_t sz
, int smp_cpus
,
635 * @pid: pid of a process
637 * For given @pid fetch its name. Caller is responsible for
638 * freeing the string when no longer needed.
639 * Returns allocated string on success, NULL on failure.
641 char *qemu_get_pid_name(pid_t pid
);
646 * A version of fork that avoids signal handler race
647 * conditions that can lead to child process getting
648 * signals that are otherwise only expected by the
649 * parent. It also resets all signal handlers to the
652 * Returns 0 to child process, pid number to parent
655 pid_t
qemu_fork(Error
**errp
);
657 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
658 * when intptr_t is 32-bit and we are aligning a long long.
660 extern uintptr_t qemu_real_host_page_size
;
661 extern intptr_t qemu_real_host_page_mask
;
663 extern int qemu_icache_linesize
;
664 extern int qemu_icache_linesize_log
;
665 extern int qemu_dcache_linesize
;
666 extern int qemu_dcache_linesize_log
;
669 * After using getopt or getopt_long, if you need to parse another set
670 * of options, then you must reset optind. Unfortunately the way to
671 * do this varies between implementations of getopt.
673 static inline void qemu_reset_optind(void)
684 * qemu_get_host_name:
685 * @errp: Error object
687 * Operating system agnostic way of querying host name.
689 * Returns allocated hostname (caller should free), NULL on failure.
691 char *qemu_get_host_name(Error
**errp
);
694 * qemu_get_host_physmem:
696 * Operating system agnostic way of querying host memory.
698 * Returns amount of physical memory on the system. This is purely
699 * advisery and may return 0 if we can't work it out. At the other
700 * end we saturate to SIZE_MAX if you are lucky enough to have that
703 size_t qemu_get_host_physmem(void);
706 * Toggle write/execute on the pages marked MAP_JIT
707 * for the current thread.
709 #if defined(MAC_OS_VERSION_11_0) && \
710 MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_VERSION_11_0
711 static inline void qemu_thread_jit_execute(void)
713 if (__builtin_available(macOS
11.0, *)) {
714 pthread_jit_write_protect_np(true);
718 static inline void qemu_thread_jit_write(void)
720 if (__builtin_available(macOS
11.0, *)) {
721 pthread_jit_write_protect_np(false);
725 static inline void qemu_thread_jit_write(void) {}
726 static inline void qemu_thread_jit_execute(void) {}
730 * Platforms which do not support system() return ENOSYS
732 #ifndef HAVE_SYSTEM_FUNCTION
733 #define system platform_does_not_support_system
734 static inline int platform_does_not_support_system(const char *command
)
739 #endif /* !HAVE_SYSTEM_FUNCTION */