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 extern 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. */
107 #ifdef HAVE_SYS_SIGNAL_H
108 #include <sys/signal.h>
112 #include <sys/wait.h>
114 #define WIFEXITED(x) 1
115 #define WEXITSTATUS(x) (x)
119 #include "sysemu/os-win32.h"
123 #include "sysemu/os-posix.h"
126 #include "glib-compat.h"
127 #include "qemu/typedefs.h"
130 * For mingw, as of v6.0.0, the function implementing the assert macro is
131 * not marked as noreturn, so the compiler cannot delete code following an
132 * assert(false) as unused. We rely on this within the code base to delete
133 * code that is unreachable when features are disabled.
134 * All supported versions of Glib's g_assert() satisfy this requirement.
138 #define assert(x) g_assert(x)
142 * According to waitpid man page:
144 * This macro is not specified in POSIX.1-2001 and is not
145 * available on some UNIX implementations (e.g., AIX, SunOS).
146 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
149 #define WCOREDUMP(status) 0
152 * We have a lot of unaudited code that may fail in strange ways, or
153 * even be a security risk during migration, if you disable assertions
154 * at compile-time. You may comment out these safety checks if you
155 * absolutely want to disable assertion overhead, but it is not
156 * supported upstream so the risk is all yours. Meanwhile, please
157 * submit patches to remove any side-effects inside an assertion, or
158 * fixing error handling that should use Error instead of assert.
161 #error building with NDEBUG is not supported
163 #ifdef G_DISABLE_ASSERT
164 #error building with G_DISABLE_ASSERT is not supported
168 #define O_LARGEFILE 0
173 #ifndef MAP_ANONYMOUS
174 #define MAP_ANONYMOUS MAP_ANON
176 #ifndef MAP_FIXED_NOREPLACE
177 #define MAP_FIXED_NOREPLACE 0
180 #define ENOMEDIUM ENODEV
182 #if !defined(ENOTSUP)
185 #if !defined(ECANCELED)
186 #define ECANCELED 4097
188 #if !defined(EMEDIUMTYPE)
189 #define EMEDIUMTYPE 4098
191 #if !defined(ESHUTDOWN)
192 #define ESHUTDOWN 4099
195 /* time_t may be either 32 or 64 bits depending on the host OS, and
196 * can be either signed or unsigned, so we can't just hardcode a
197 * specific maximum value. This is not a C preprocessor constant,
198 * so you can't use TIME_MAX in an #ifdef, but for our purposes
199 * this isn't a problem.
202 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
203 * Gnulib, and are under the LGPL v2.1 or (at your option) any
207 /* True if the real type T is signed. */
208 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
210 /* The width in bits of the integer type or expression T.
211 * Padding bits are not supported.
213 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
215 /* The maximum and minimum values for the integer type T. */
216 #define TYPE_MAXIMUM(t) \
217 ((t) (!TYPE_SIGNED(t) \
219 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
222 #define TIME_MAX TYPE_MAXIMUM(time_t)
225 /* HOST_LONG_BITS is the size of a native pointer in bits. */
226 #if UINTPTR_MAX == UINT32_MAX
227 # define HOST_LONG_BITS 32
228 #elif UINTPTR_MAX == UINT64_MAX
229 # define HOST_LONG_BITS 64
231 # error Unknown pointer size
234 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
235 * the wrong type. Our replacement isn't usable in preprocessor
236 * expressions, but it is sufficient for our needs. */
237 #if defined(HAVE_BROKEN_SIZE_MAX) && HAVE_BROKEN_SIZE_MAX
239 #define SIZE_MAX ((size_t)-1)
243 * Two variations of MIN/MAX macros. The first is for runtime use, and
244 * evaluates arguments only once (so it is safe even with side
245 * effects), but will not work in constant contexts (such as array
246 * size declarations) because of the '{}'. The second is for constant
247 * expression use, where evaluating arguments twice is safe because
248 * the result is going to be constant anyway, but will not work in a
249 * runtime context because of a void expression where a value is
250 * expected. Thus, both gcc and clang will fail to compile if you use
251 * the wrong macro (even if the error may seem a bit cryptic).
253 * Note that neither form is usable as an #if condition; if you truly
254 * need to write conditional code that depends on a minimum or maximum
255 * determined by the pre-processor instead of the compiler, you'll
256 * have to open-code it. Sadly, Coverity is severely confused by the
257 * constant variants, so we have to dumb things down there.
262 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
268 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
273 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
274 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
276 # define MIN_CONST(a, b) \
277 __builtin_choose_expr( \
278 __builtin_constant_p(a) && __builtin_constant_p(b), \
279 (a) < (b) ? (a) : (b), \
281 # define MAX_CONST(a, b) \
282 __builtin_choose_expr( \
283 __builtin_constant_p(a) && __builtin_constant_p(b), \
284 (a) > (b) ? (a) : (b), \
289 * Minimum function that returns zero only if both values are zero.
290 * Intended for use with unsigned values only.
293 #define MIN_NON_ZERO(a, b) \
295 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
296 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
300 /* Round number down to multiple */
301 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
303 /* Round number up to multiple. Safe when m is not a power of 2 (see
304 * ROUND_UP for a faster version when a power of 2 is guaranteed) */
305 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
307 /* Check if n is a multiple of m */
308 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
310 /* n-byte align pointer down */
311 #define QEMU_ALIGN_PTR_DOWN(p, n) \
312 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
314 /* n-byte align pointer up */
315 #define QEMU_ALIGN_PTR_UP(p, n) \
316 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
318 /* Check if pointer p is n-bytes aligned */
319 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
321 /* Round number up to multiple. Requires that d be a power of 2 (see
322 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
323 * numbers); works even if d is a smaller type than n. */
325 #define ROUND_UP(n, d) (((n) + (d) - 1) & -(0 ? (n) : (d)))
329 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
333 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
334 * pointer, not an array.
336 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
339 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
340 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
343 int qemu_daemon(int nochdir
, int noclose
);
344 void *qemu_try_memalign(size_t alignment
, size_t size
);
345 void *qemu_memalign(size_t alignment
, size_t size
);
346 void *qemu_anon_ram_alloc(size_t size
, uint64_t *align
, bool shared
);
347 void qemu_vfree(void *ptr
);
348 void qemu_anon_ram_free(void *ptr
, size_t size
);
350 #define QEMU_MADV_INVALID -1
352 #if defined(CONFIG_MADVISE)
354 #define QEMU_MADV_WILLNEED MADV_WILLNEED
355 #define QEMU_MADV_DONTNEED MADV_DONTNEED
357 #define QEMU_MADV_DONTFORK MADV_DONTFORK
359 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
361 #ifdef MADV_MERGEABLE
362 #define QEMU_MADV_MERGEABLE MADV_MERGEABLE
364 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
366 #ifdef MADV_UNMERGEABLE
367 #define QEMU_MADV_UNMERGEABLE MADV_UNMERGEABLE
369 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
372 #define QEMU_MADV_DODUMP MADV_DODUMP
374 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
377 #define QEMU_MADV_DONTDUMP MADV_DONTDUMP
379 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
382 #define QEMU_MADV_HUGEPAGE MADV_HUGEPAGE
384 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
386 #ifdef MADV_NOHUGEPAGE
387 #define QEMU_MADV_NOHUGEPAGE MADV_NOHUGEPAGE
389 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
392 #define QEMU_MADV_REMOVE MADV_REMOVE
394 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
397 #elif defined(CONFIG_POSIX_MADVISE)
399 #define QEMU_MADV_WILLNEED POSIX_MADV_WILLNEED
400 #define QEMU_MADV_DONTNEED POSIX_MADV_DONTNEED
401 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
402 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
403 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
404 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
405 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
406 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
407 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
408 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
412 #define QEMU_MADV_WILLNEED QEMU_MADV_INVALID
413 #define QEMU_MADV_DONTNEED QEMU_MADV_INVALID
414 #define QEMU_MADV_DONTFORK QEMU_MADV_INVALID
415 #define QEMU_MADV_MERGEABLE QEMU_MADV_INVALID
416 #define QEMU_MADV_UNMERGEABLE QEMU_MADV_INVALID
417 #define QEMU_MADV_DODUMP QEMU_MADV_INVALID
418 #define QEMU_MADV_DONTDUMP QEMU_MADV_INVALID
419 #define QEMU_MADV_HUGEPAGE QEMU_MADV_INVALID
420 #define QEMU_MADV_NOHUGEPAGE QEMU_MADV_INVALID
421 #define QEMU_MADV_REMOVE QEMU_MADV_INVALID
426 #define HAVE_CHARDEV_SERIAL 1
427 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
428 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
429 || defined(__GLIBC__) || defined(__APPLE__)
430 #define HAVE_CHARDEV_SERIAL 1
433 #if defined(__linux__) || defined(__FreeBSD__) || \
434 defined(__FreeBSD_kernel__) || defined(__DragonFly__)
435 #define HAVE_CHARDEV_PARPORT 1
438 #if defined(__HAIKU__)
439 #define SIGIO SIGPOLL
442 #if defined(CONFIG_LINUX)
443 #ifndef BUS_MCEERR_AR
444 #define BUS_MCEERR_AR 4
446 #ifndef BUS_MCEERR_AO
447 #define BUS_MCEERR_AO 5
451 #if defined(__linux__) && \
452 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
453 || defined(__powerpc64__))
454 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
455 Valgrind does not support alignments larger than 1 MiB,
456 therefore we need special code which handles running on Valgrind. */
457 # define QEMU_VMALLOC_ALIGN (512 * 4096)
458 #elif defined(__linux__) && defined(__s390x__)
459 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
460 # define QEMU_VMALLOC_ALIGN (256 * 4096)
461 #elif defined(__linux__) && defined(__sparc__)
463 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size, SHMLBA)
465 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size
469 struct qemu_signalfd_siginfo
{
470 uint32_t ssi_signo
; /* Signal number */
471 int32_t ssi_errno
; /* Error number (unused) */
472 int32_t ssi_code
; /* Signal code */
473 uint32_t ssi_pid
; /* PID of sender */
474 uint32_t ssi_uid
; /* Real UID of sender */
475 int32_t ssi_fd
; /* File descriptor (SIGIO) */
476 uint32_t ssi_tid
; /* Kernel timer ID (POSIX timers) */
477 uint32_t ssi_band
; /* Band event (SIGIO) */
478 uint32_t ssi_overrun
; /* POSIX timer overrun count */
479 uint32_t ssi_trapno
; /* Trap number that caused signal */
480 int32_t ssi_status
; /* Exit status or signal (SIGCHLD) */
481 int32_t ssi_int
; /* Integer sent by sigqueue(2) */
482 uint64_t ssi_ptr
; /* Pointer sent by sigqueue(2) */
483 uint64_t ssi_utime
; /* User CPU time consumed (SIGCHLD) */
484 uint64_t ssi_stime
; /* System CPU time consumed (SIGCHLD) */
485 uint64_t ssi_addr
; /* Address that generated signal
486 (for hardware-generated signals) */
487 uint8_t pad
[48]; /* Pad size to 128 bytes (allow for
488 additional fields in the future) */
491 int qemu_signalfd(const sigset_t
*mask
);
492 void sigaction_invoke(struct sigaction
*action
,
493 struct qemu_signalfd_siginfo
*info
);
496 int qemu_madvise(void *addr
, size_t len
, int advice
);
497 int qemu_mprotect_rwx(void *addr
, size_t size
);
498 int qemu_mprotect_none(void *addr
, size_t size
);
500 int qemu_open(const char *name
, int flags
, ...);
501 int qemu_close(int fd
);
502 int qemu_unlink(const char *name
);
504 int qemu_dup(int fd
);
506 int qemu_lock_fd(int fd
, int64_t start
, int64_t len
, bool exclusive
);
507 int qemu_unlock_fd(int fd
, int64_t start
, int64_t len
);
508 int qemu_lock_fd_test(int fd
, int64_t start
, int64_t len
, bool exclusive
);
509 bool qemu_has_ofd_lock(void);
511 #if defined(__HAIKU__) && defined(__i386__)
512 #define FMT_pid "%ld"
514 #define FMT_pid "%" PRId64
519 bool qemu_write_pidfile(const char *pidfile
, Error
**errp
);
521 int qemu_get_thread_id(void);
529 * Use the same value as Linux for now.
533 ssize_t
readv(int fd
, const struct iovec
*iov
, int iov_cnt
);
534 ssize_t
writev(int fd
, const struct iovec
*iov
, int iov_cnt
);
540 static inline void qemu_timersub(const struct timeval
*val1
,
541 const struct timeval
*val2
,
544 res
->tv_sec
= val1
->tv_sec
- val2
->tv_sec
;
545 if (val1
->tv_usec
< val2
->tv_usec
) {
547 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
+ 1000 * 1000;
549 res
->tv_usec
= val1
->tv_usec
- val2
->tv_usec
;
553 #define qemu_timersub timersub
556 void qemu_set_cloexec(int fd
);
558 /* Starting on QEMU 2.5, qemu_hw_version() returns "2.5+" by default
559 * instead of QEMU_VERSION, so setting hw_version on MachineClass
560 * is no longer mandatory.
562 * Do NOT change this string, or it will break compatibility on all
563 * machine classes that don't set hw_version.
565 #define QEMU_HW_VERSION "2.5+"
567 /* QEMU "hardware version" setting. Used to replace code that exposed
568 * QEMU_VERSION to guests in the past and need to keep compatibility.
569 * Do not use qemu_hw_version() in new code.
571 void qemu_set_hw_version(const char *);
572 const char *qemu_hw_version(void);
574 void fips_set_state(bool requested
);
575 bool fips_get_state(void);
577 /* Return a dynamically allocated pathname denoting a file or directory that is
578 * appropriate for storing local state.
580 * @relative_pathname need not start with a directory separator; one will be
581 * added automatically.
583 * The caller is responsible for releasing the value returned with g_free()
586 char *qemu_get_local_state_pathname(const char *relative_pathname
);
588 /* Find program directory, and save it for later usage with
589 * qemu_get_exec_dir().
590 * Try OS specific API first, if not working, parse from argv0. */
591 void qemu_init_exec_dir(const char *argv0
);
593 /* Get the saved exec dir.
595 * The caller is responsible for releasing the value returned with g_free()
598 char *qemu_get_exec_dir(void);
602 * @type: the auxiliary vector key to lookup
604 * Search the auxiliary vector for @type, returning the value
605 * or 0 if @type is not present.
607 unsigned long qemu_getauxval(unsigned long type
);
609 void qemu_set_tty_echo(int fd
, bool echo
);
611 void os_mem_prealloc(int fd
, char *area
, size_t sz
, int smp_cpus
,
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 extern uintptr_t qemu_real_host_page_size
;
642 extern intptr_t qemu_real_host_page_mask
;
644 extern int qemu_icache_linesize
;
645 extern int qemu_icache_linesize_log
;
646 extern int qemu_dcache_linesize
;
647 extern int qemu_dcache_linesize_log
;
650 * After using getopt or getopt_long, if you need to parse another set
651 * of options, then you must reset optind. Unfortunately the way to
652 * do this varies between implementations of getopt.
654 static inline void qemu_reset_optind(void)
665 * qemu_get_host_name:
666 * @errp: Error object
668 * Operating system agnostic way of querying host name.
670 * Returns allocated hostname (caller should free), NULL on failure.
672 char *qemu_get_host_name(Error
**errp
);
675 * qemu_get_host_physmem:
677 * Operating system agnostic way of querying host memory.
679 * Returns amount of physical memory on the system. This is purely
680 * advisery and may return 0 if we can't work it out. At the other
681 * end we saturate to SIZE_MAX if you are lucky enough to have that
684 size_t qemu_get_host_physmem(void);