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. */
108 #include <sys/wait.h>
110 #define WIFEXITED(x) 1
111 #define WEXITSTATUS(x) (x)
115 * This is somewhat like a system header; it must be outside any extern "C"
116 * block because it includes system headers itself, including glib.h,
117 * which will not compile if inside an extern "C" block.
119 #include "glib-compat.h"
126 #include "sysemu/os-win32.h"
130 #include "sysemu/os-posix.h"
134 #include <AvailabilityMacros.h>
137 #include "qemu/typedefs.h"
140 * For mingw, as of v6.0.0, the function implementing the assert macro is
141 * not marked as noreturn, so the compiler cannot delete code following an
142 * assert(false) as unused. We rely on this within the code base to delete
143 * code that is unreachable when features are disabled.
144 * All supported versions of Glib's g_assert() satisfy this requirement.
148 #define assert(x) g_assert(x)
152 * According to waitpid man page:
154 * This macro is not specified in POSIX.1-2001 and is not
155 * available on some UNIX implementations (e.g., AIX, SunOS).
156 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
159 #define WCOREDUMP(status) 0
162 * We have a lot of unaudited code that may fail in strange ways, or
163 * even be a security risk during migration, if you disable assertions
164 * at compile-time. You may comment out these safety checks if you
165 * absolutely want to disable assertion overhead, but it is not
166 * supported upstream so the risk is all yours. Meanwhile, please
167 * submit patches to remove any side-effects inside an assertion, or
168 * fixing error handling that should use Error instead of assert.
171 #error building with NDEBUG is not supported
173 #ifdef G_DISABLE_ASSERT
174 #error building with G_DISABLE_ASSERT is not supported
178 #define O_LARGEFILE 0
183 #ifndef MAP_ANONYMOUS
184 #define MAP_ANONYMOUS MAP_ANON
186 #ifndef MAP_FIXED_NOREPLACE
187 #define MAP_FIXED_NOREPLACE 0
190 #define ENOMEDIUM ENODEV
192 #if !defined(ENOTSUP)
195 #if !defined(ECANCELED)
196 #define ECANCELED 4097
198 #if !defined(EMEDIUMTYPE)
199 #define EMEDIUMTYPE 4098
201 #if !defined(ESHUTDOWN)
202 #define ESHUTDOWN 4099
205 /* time_t may be either 32 or 64 bits depending on the host OS, and
206 * can be either signed or unsigned, so we can't just hardcode a
207 * specific maximum value. This is not a C preprocessor constant,
208 * so you can't use TIME_MAX in an #ifdef, but for our purposes
209 * this isn't a problem.
212 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
213 * Gnulib, and are under the LGPL v2.1 or (at your option) any
217 /* True if the real type T is signed. */
218 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
220 /* The width in bits of the integer type or expression T.
221 * Padding bits are not supported.
223 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
225 /* The maximum and minimum values for the integer type T. */
226 #define TYPE_MAXIMUM(t) \
227 ((t) (!TYPE_SIGNED(t) \
229 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
232 #define TIME_MAX TYPE_MAXIMUM(time_t)
235 /* HOST_LONG_BITS is the size of a native pointer in bits. */
236 #if UINTPTR_MAX == UINT32_MAX
237 # define HOST_LONG_BITS 32
238 #elif UINTPTR_MAX == UINT64_MAX
239 # define HOST_LONG_BITS 64
241 # error Unknown pointer size
244 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
245 * the wrong type. Our replacement isn't usable in preprocessor
246 * expressions, but it is sufficient for our needs. */
247 #if defined(HAVE_BROKEN_SIZE_MAX) && HAVE_BROKEN_SIZE_MAX
249 #define SIZE_MAX ((size_t)-1)
253 * Two variations of MIN/MAX macros. The first is for runtime use, and
254 * evaluates arguments only once (so it is safe even with side
255 * effects), but will not work in constant contexts (such as array
256 * size declarations) because of the '{}'. The second is for constant
257 * expression use, where evaluating arguments twice is safe because
258 * the result is going to be constant anyway, but will not work in a
259 * runtime context because of a void expression where a value is
260 * expected. Thus, both gcc and clang will fail to compile if you use
261 * the wrong macro (even if the error may seem a bit cryptic).
263 * Note that neither form is usable as an #if condition; if you truly
264 * need to write conditional code that depends on a minimum or maximum
265 * determined by the pre-processor instead of the compiler, you'll
266 * have to open-code it. Sadly, Coverity is severely confused by the
267 * constant variants, so we have to dumb things down there.
272 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
278 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
283 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
284 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
286 # define MIN_CONST(a, b) \
287 __builtin_choose_expr( \
288 __builtin_constant_p(a) && __builtin_constant_p(b), \
289 (a) < (b) ? (a) : (b), \
291 # define MAX_CONST(a, b) \
292 __builtin_choose_expr( \
293 __builtin_constant_p(a) && __builtin_constant_p(b), \
294 (a) > (b) ? (a) : (b), \
299 * Minimum function that returns zero only if both values are zero.
300 * Intended for use with unsigned values only.
303 #define MIN_NON_ZERO(a, b) \
305 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
306 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
310 /* Round number down to multiple */
311 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
313 /* Round number up to multiple. Safe when m is not a power of 2 (see
314 * ROUND_UP for a faster version when a power of 2 is guaranteed) */
315 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
317 /* Check if n is a multiple of m */
318 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
320 /* n-byte align pointer down */
321 #define QEMU_ALIGN_PTR_DOWN(p, n) \
322 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
324 /* n-byte align pointer up */
325 #define QEMU_ALIGN_PTR_UP(p, n) \
326 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
328 /* Check if pointer p is n-bytes aligned */
329 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
331 /* Round number up to multiple. Requires that d be a power of 2 (see
332 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
333 * numbers); works even if d is a smaller type than n. */
335 #define ROUND_UP(n, d) (((n) + (d) - 1) & -(0 ? (n) : (d)))
339 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
343 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
344 * pointer, not an array.
346 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
349 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
350 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
353 int qemu_daemon(int nochdir
, int noclose
);
354 void *qemu_try_memalign(size_t alignment
, size_t size
);
355 void *qemu_memalign(size_t alignment
, size_t size
);
356 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
);
357 void qemu_vfree(void *ptr
);
358 void qemu_anon_ram_free(void *ptr
, size_t size
);
360 #define QEMU_MADV_INVALID -1
362 #if defined(CONFIG_MADVISE)
364 #define QEMU_MADV_WILLNEED MADV_WILLNEED
365 #define QEMU_MADV_DONTNEED MADV_DONTNEED
367 #define QEMU_MADV_DONTFORK MADV_DONTFORK
369 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
371 #ifdef MADV_MERGEABLE
372 #define QEMU_MADV_MERGEABLE MADV_MERGEABLE
374 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
376 #ifdef MADV_UNMERGEABLE
377 #define QEMU_MADV_UNMERGEABLE MADV_UNMERGEABLE
379 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
382 #define QEMU_MADV_DODUMP MADV_DODUMP
384 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
387 #define QEMU_MADV_DONTDUMP MADV_DONTDUMP
389 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
392 #define QEMU_MADV_HUGEPAGE MADV_HUGEPAGE
394 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
396 #ifdef MADV_NOHUGEPAGE
397 #define QEMU_MADV_NOHUGEPAGE MADV_NOHUGEPAGE
399 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
402 #define QEMU_MADV_REMOVE MADV_REMOVE
404 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
407 #elif defined(CONFIG_POSIX_MADVISE)
409 #define QEMU_MADV_WILLNEED POSIX_MADV_WILLNEED
410 #define QEMU_MADV_DONTNEED POSIX_MADV_DONTNEED
411 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
412 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
413 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
414 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
415 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
416 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
417 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
418 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
422 #define QEMU_MADV_WILLNEED QEMU_MADV_INVALID
423 #define QEMU_MADV_DONTNEED QEMU_MADV_INVALID
424 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
425 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
426 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
427 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
428 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
429 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
430 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
431 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
436 #define HAVE_CHARDEV_SERIAL 1
437 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
438 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
439 || defined(__GLIBC__) || defined(__APPLE__)
440 #define HAVE_CHARDEV_SERIAL 1
443 #if defined(__linux__) || defined(__FreeBSD__) || \
444 defined(__FreeBSD_kernel__) || defined(__DragonFly__)
445 #define HAVE_CHARDEV_PARPORT 1
448 #if defined(__HAIKU__)
449 #define SIGIO SIGPOLL
452 #if defined(CONFIG_LINUX)
453 #ifndef BUS_MCEERR_AR
454 #define BUS_MCEERR_AR 4
456 #ifndef BUS_MCEERR_AO
457 #define BUS_MCEERR_AO 5
461 #if defined(__linux__) && \
462 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
463 || defined(__powerpc64__))
464 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
465 Valgrind does not support alignments larger than 1 MiB,
466 therefore we need special code which handles running on Valgrind. */
467 # define QEMU_VMALLOC_ALIGN (512 * 4096)
468 #elif defined(__linux__) && defined(__s390x__)
469 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
470 # define QEMU_VMALLOC_ALIGN (256 * 4096)
471 #elif defined(__linux__) && defined(__sparc__)
473 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size, SHMLBA)
475 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size
479 struct qemu_signalfd_siginfo
{
480 uint32_t ssi_signo
; /* Signal number */
481 int32_t ssi_errno
; /* Error number (unused) */
482 int32_t ssi_code
; /* Signal code */
483 uint32_t ssi_pid
; /* PID of sender */
484 uint32_t ssi_uid
; /* Real UID of sender */
485 int32_t ssi_fd
; /* File descriptor (SIGIO) */
486 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
487 uint32_t ssi_band
; /* Band event (SIGIO) */
488 uint32_t ssi_overrun
; /* POSIX timer overrun count */
489 uint32_t ssi_trapno
; /* Trap number that caused signal */
490 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
491 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
492 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
493 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
494 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
495 uint64_t ssi_addr
; /* Address that generated signal
496 (for hardware-generated signals) */
497 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
498 additional fields in the future) */
501 int qemu_signalfd(const sigset_t
*mask
);
502 void sigaction_invoke(struct sigaction
*action
,
503 struct qemu_signalfd_siginfo
*info
);
506 int qemu_madvise(void *addr
, size_t len
, int advice
);
507 int qemu_mprotect_rwx(void *addr
, size_t size
);
508 int qemu_mprotect_none(void *addr
, size_t size
);
511 * Don't introduce new usage of this function, prefer the following
512 * qemu_open/qemu_create that take an "Error **errp"
514 int qemu_open_old(const char *name
, int flags
, ...);
515 int qemu_open(const char *name
, int flags
, Error
**errp
);
516 int qemu_create(const char *name
, int flags
, mode_t mode
, Error
**errp
);
517 int qemu_close(int fd
);
518 int qemu_unlink(const char *name
);
520 int qemu_dup_flags(int fd
, int flags
);
521 int qemu_dup(int fd
);
522 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
523 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
524 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
525 bool qemu_has_ofd_lock(void);
528 #if defined(__HAIKU__) && defined(__i386__)
529 #define FMT_pid "%ld"
531 #define FMT_pid "%" PRId64
536 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
538 int qemu_get_thread_id(void);
546 * Use the same value as Linux for now.
550 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
551 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
557 static inline void qemu_timersub(const struct timeval
*val1
,
558 const struct timeval
*val2
,
561 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
562 if (val1
->tv_usec
< val2
->tv_usec
) {
564 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
566 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
570 #define qemu_timersub timersub
573 void qemu_set_cloexec(int fd
);
575 /* Starting on QEMU 2.5, qemu_hw_version() returns "2.5+" by default
576 * instead of QEMU_VERSION, so setting hw_version on MachineClass
577 * is no longer mandatory.
579 * Do NOT change this string, or it will break compatibility on all
580 * machine classes that don't set hw_version.
582 #define QEMU_HW_VERSION "2.5+"
584 /* QEMU "hardware version" setting. Used to replace code that exposed
585 * QEMU_VERSION to guests in the past and need to keep compatibility.
586 * Do not use qemu_hw_version() in new code.
588 void qemu_set_hw_version(const char *);
589 const char *qemu_hw_version(void);
591 void fips_set_state(bool requested
);
592 bool fips_get_state(void);
594 /* Return a dynamically allocated pathname denoting a file or directory that is
595 * appropriate for storing local state.
597 * @relative_pathname need not start with a directory separator; one will be
598 * added automatically.
600 * The caller is responsible for releasing the value returned with g_free()
603 char *qemu_get_local_state_pathname(const char *relative_pathname
);
605 /* Find program directory, and save it for later usage with
606 * qemu_get_exec_dir().
607 * Try OS specific API first, if not working, parse from argv0. */
608 void qemu_init_exec_dir(const char *argv0
);
610 /* Get the saved exec dir. */
611 const char *qemu_get_exec_dir(void);
615 * @type: the auxiliary vector key to lookup
617 * Search the auxiliary vector for @type, returning the value
618 * or 0 if @type is not present.
620 unsigned long qemu_getauxval(unsigned long type
);
622 void qemu_set_tty_echo(int fd
, bool echo
);
624 void os_mem_prealloc(int fd
, char *area
, size_t sz
, int smp_cpus
,
629 * @pid: pid of a process
631 * For given @pid fetch its name. Caller is responsible for
632 * freeing the string when no longer needed.
633 * Returns allocated string on success, NULL on failure.
635 char *qemu_get_pid_name(pid_t pid
);
640 * A version of fork that avoids signal handler race
641 * conditions that can lead to child process getting
642 * signals that are otherwise only expected by the
643 * parent. It also resets all signal handlers to the
646 * Returns 0 to child process, pid number to parent
649 pid_t
qemu_fork(Error
**errp
);
651 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
652 * when intptr_t is 32-bit and we are aligning a long long.
654 extern uintptr_t qemu_real_host_page_size
;
655 extern intptr_t qemu_real_host_page_mask
;
657 extern int qemu_icache_linesize
;
658 extern int qemu_icache_linesize_log
;
659 extern int qemu_dcache_linesize
;
660 extern int qemu_dcache_linesize_log
;
663 * After using getopt or getopt_long, if you need to parse another set
664 * of options, then you must reset optind. Unfortunately the way to
665 * do this varies between implementations of getopt.
667 static inline void qemu_reset_optind(void)
678 * qemu_get_host_name:
679 * @errp: Error object
681 * Operating system agnostic way of querying host name.
683 * Returns allocated hostname (caller should free), NULL on failure.
685 char *qemu_get_host_name(Error
**errp
);
688 * qemu_get_host_physmem:
690 * Operating system agnostic way of querying host memory.
692 * Returns amount of physical memory on the system. This is purely
693 * advisery and may return 0 if we can't work it out. At the other
694 * end we saturate to SIZE_MAX if you are lucky enough to have that
697 size_t qemu_get_host_physmem(void);
700 * Toggle write/execute on the pages marked MAP_JIT
701 * for the current thread.
703 #if defined(MAC_OS_VERSION_11_0) && \
704 MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_VERSION_11_0
705 static inline void qemu_thread_jit_execute(void)
707 if (__builtin_available(macOS
11.0, *)) {
708 pthread_jit_write_protect_np(true);
712 static inline void qemu_thread_jit_write(void)
714 if (__builtin_available(macOS
11.0, *)) {
715 pthread_jit_write_protect_np(false);
719 static inline void qemu_thread_jit_write(void) {}
720 static inline void qemu_thread_jit_execute(void) {}
724 * Platforms which do not support system() return ENOSYS
726 #ifndef HAVE_SYSTEM_FUNCTION
727 #define system platform_does_not_support_system
728 static inline int platform_does_not_support_system(const char *command
)
733 #endif /* !HAVE_SYSTEM_FUNCTION */