| 1 | /* |
| 2 | * OS includes and handling of OS dependencies |
| 3 | * |
| 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). |
| 7 | * |
| 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 | * system/os-posix.h, system/os-win32.h, system/os-wasm.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. |
| 14 | * |
| 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. |
| 18 | * |
| 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 |
| 23 | * |
| 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. |
| 26 | */ |
| 27 | #ifndef QEMU_OSDEP_H |
| 28 | #define QEMU_OSDEP_H |
| 29 | |
| 30 | #if !defined _FORTIFY_SOURCE && defined __OPTIMIZE__ && __OPTIMIZE__ && defined __linux__ |
| 31 | # define _FORTIFY_SOURCE 2 |
| 32 | #endif |
| 33 | |
| 34 | #include "config-host.h" |
| 35 | #ifdef COMPILING_PER_TARGET |
| 36 | #include CONFIG_TARGET |
| 37 | #else |
| 38 | #include "exec/poison.h" |
| 39 | #endif |
| 40 | |
| 41 | /* |
| 42 | * HOST_WORDS_BIGENDIAN was replaced with HOST_BIG_ENDIAN. Prevent it from |
| 43 | * creeping back in. |
| 44 | */ |
| 45 | #pragma GCC poison HOST_WORDS_BIGENDIAN |
| 46 | |
| 47 | /* |
| 48 | * TARGET_WORDS_BIGENDIAN was replaced with TARGET_BIG_ENDIAN. Prevent it from |
| 49 | * creeping back in. |
| 50 | */ |
| 51 | #pragma GCC poison TARGET_WORDS_BIGENDIAN |
| 52 | |
| 53 | #include "qemu/compiler.h" |
| 54 | |
| 55 | /* Older versions of C++ don't get definitions of various macros from |
| 56 | * stdlib.h unless we define these macros before first inclusion of |
| 57 | * that system header. |
| 58 | */ |
| 59 | #ifndef __STDC_CONSTANT_MACROS |
| 60 | #define __STDC_CONSTANT_MACROS |
| 61 | #endif |
| 62 | #ifndef __STDC_LIMIT_MACROS |
| 63 | #define __STDC_LIMIT_MACROS |
| 64 | #endif |
| 65 | #ifndef __STDC_FORMAT_MACROS |
| 66 | #define __STDC_FORMAT_MACROS |
| 67 | #endif |
| 68 | |
| 69 | /* The following block of code temporarily renames the daemon() function so the |
| 70 | * compiler does not see the warning associated with it in stdlib.h on OSX |
| 71 | */ |
| 72 | #ifdef __APPLE__ |
| 73 | #define daemon qemu_fake_daemon_function |
| 74 | #include <stdlib.h> |
| 75 | #undef daemon |
| 76 | QEMU_EXTERN_C int daemon(int, int); |
| 77 | #endif |
| 78 | |
| 79 | #ifdef _WIN32 |
| 80 | /* as defined in sdkddkver.h */ |
| 81 | #ifndef _WIN32_WINNT |
| 82 | #define _WIN32_WINNT 0x0602 /* Windows 8 API (should be >= the one from glib) */ |
| 83 | #endif |
| 84 | /* reduces the number of implicitly included headers */ |
| 85 | #ifndef WIN32_LEAN_AND_MEAN |
| 86 | #define WIN32_LEAN_AND_MEAN |
| 87 | #endif |
| 88 | #endif |
| 89 | |
| 90 | /* enable C99/POSIX format strings (needs mingw32-runtime 3.15 or later) */ |
| 91 | #ifdef __MINGW32__ |
| 92 | #define __USE_MINGW_ANSI_STDIO 1 |
| 93 | #endif |
| 94 | |
| 95 | /* |
| 96 | * We need the FreeBSD "legacy" definitions. Rust needs the FreeBSD 11 system |
| 97 | * calls since it doesn't use libc at all, so we have to emulate that despite |
| 98 | * FreeBSD 11 being EOL'd. |
| 99 | */ |
| 100 | #ifdef __FreeBSD__ |
| 101 | #define _WANT_FREEBSD11_STAT |
| 102 | #define _WANT_FREEBSD11_STATFS |
| 103 | #define _WANT_FREEBSD11_DIRENT |
| 104 | #define _WANT_KERNEL_ERRNO |
| 105 | #define _WANT_SEMUN |
| 106 | #endif |
| 107 | |
| 108 | #include <stdarg.h> |
| 109 | #include <stddef.h> |
| 110 | #include <stdbool.h> |
| 111 | #include <stdint.h> |
| 112 | #include <sys/types.h> |
| 113 | #include <stdlib.h> |
| 114 | #include <stdio.h> |
| 115 | |
| 116 | #include <string.h> |
| 117 | #include <inttypes.h> |
| 118 | #include <limits.h> |
| 119 | /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r |
| 120 | * function availability on recentish Mingw-w64 platforms. */ |
| 121 | #include <unistd.h> |
| 122 | #include <time.h> |
| 123 | #include <ctype.h> |
| 124 | #include <errno.h> |
| 125 | #include <fcntl.h> |
| 126 | #include <getopt.h> |
| 127 | #include <sys/stat.h> |
| 128 | #include <sys/time.h> |
| 129 | #include <assert.h> |
| 130 | /* setjmp must be declared before system/os-win32.h |
| 131 | * because it is redefined there. */ |
| 132 | #include <setjmp.h> |
| 133 | #include <signal.h> |
| 134 | |
| 135 | /* |
| 136 | * Avoid conflict with linux/arch/powerpc/include/uapi/asm/elf.h, included |
| 137 | * from <asm/sigcontext.h>, but we might as well do this unconditionally. |
| 138 | */ |
| 139 | #undef ELF_CLASS |
| 140 | #undef ELF_DATA |
| 141 | #undef ELF_ARCH |
| 142 | |
| 143 | /* |
| 144 | * Avoid conflict with Solaris FSCALE definition from <sys/param.h> header, |
| 145 | * but we might as well do this unconditionally. |
| 146 | */ |
| 147 | #undef FSCALE |
| 148 | |
| 149 | #ifdef CONFIG_IOVEC |
| 150 | #include <sys/uio.h> |
| 151 | #endif |
| 152 | |
| 153 | #if defined(__linux__) && defined(__sparc__) |
| 154 | /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */ |
| 155 | #include <sys/shm.h> |
| 156 | #endif |
| 157 | |
| 158 | #ifndef _WIN32 |
| 159 | #include <sys/wait.h> |
| 160 | #else |
| 161 | #define WIFEXITED(x) 1 |
| 162 | #define WEXITSTATUS(x) (x) |
| 163 | #endif |
| 164 | |
| 165 | #ifdef __APPLE__ |
| 166 | #include <AvailabilityMacros.h> |
| 167 | #endif |
| 168 | |
| 169 | /* |
| 170 | * This is somewhat like a system header; it must be outside any extern "C" |
| 171 | * block because it includes system headers itself, including glib.h, |
| 172 | * which will not compile if inside an extern "C" block. |
| 173 | */ |
| 174 | #include "glib-compat.h" |
| 175 | |
| 176 | #ifdef _WIN32 |
| 177 | #include "system/os-win32.h" |
| 178 | #endif |
| 179 | |
| 180 | #if defined(CONFIG_POSIX) && !defined(EMSCRIPTEN) |
| 181 | #include "system/os-posix.h" |
| 182 | #endif |
| 183 | |
| 184 | #if defined(EMSCRIPTEN) |
| 185 | #include "system/os-wasm.h" |
| 186 | #endif |
| 187 | |
| 188 | #ifdef __cplusplus |
| 189 | extern "C" { |
| 190 | #endif |
| 191 | |
| 192 | #include "qemu/typedefs.h" |
| 193 | |
| 194 | /** |
| 195 | * Mark a function that executes in coroutine context |
| 196 | * |
| 197 | * Functions that execute in coroutine context cannot be called directly from |
| 198 | * normal functions. In the future it would be nice to enable compiler or |
| 199 | * static checker support for catching such errors. This annotation might make |
| 200 | * it possible and in the meantime it serves as documentation. |
| 201 | * |
| 202 | * For example: |
| 203 | * |
| 204 | * static void coroutine_fn foo(void) { |
| 205 | * .... |
| 206 | * } |
| 207 | */ |
| 208 | #ifdef __clang__ |
| 209 | #define coroutine_fn QEMU_ANNOTATE("coroutine_fn") |
| 210 | #else |
| 211 | #define coroutine_fn |
| 212 | #endif |
| 213 | |
| 214 | /** |
| 215 | * Mark a function that can suspend when executed in coroutine context, |
| 216 | * but can handle running in non-coroutine context too. |
| 217 | */ |
| 218 | #ifdef __clang__ |
| 219 | #define coroutine_mixed_fn QEMU_ANNOTATE("coroutine_mixed_fn") |
| 220 | #else |
| 221 | #define coroutine_mixed_fn |
| 222 | #endif |
| 223 | |
| 224 | /** |
| 225 | * Mark a function that should not be called from a coroutine context. |
| 226 | * Usually there will be an analogous, coroutine_fn function that should |
| 227 | * be used instead. |
| 228 | * |
| 229 | * When the function is also marked as coroutine_mixed_fn, the function should |
| 230 | * only be called if the caller does not know whether it is in coroutine |
| 231 | * context. |
| 232 | * |
| 233 | * Functions that are only no_coroutine_fn, on the other hand, should not |
| 234 | * be called from within coroutines at all. This for example includes |
| 235 | * functions that block. |
| 236 | * |
| 237 | * In the future it would be nice to enable compiler or static checker |
| 238 | * support for catching such errors. This annotation is the first step |
| 239 | * towards this, and in the meantime it serves as documentation. |
| 240 | * |
| 241 | * For example: |
| 242 | * |
| 243 | * static void no_coroutine_fn foo(void) { |
| 244 | * .... |
| 245 | * } |
| 246 | */ |
| 247 | #ifdef __clang__ |
| 248 | #define no_coroutine_fn QEMU_ANNOTATE("no_coroutine_fn") |
| 249 | #else |
| 250 | #define no_coroutine_fn |
| 251 | #endif |
| 252 | |
| 253 | |
| 254 | /* |
| 255 | * For mingw, as of v6.0.0, the function implementing the assert macro is |
| 256 | * not marked as noreturn, so the compiler cannot delete code following an |
| 257 | * assert(false) as unused. We rely on this within the code base to delete |
| 258 | * code that is unreachable when features are disabled. |
| 259 | * All supported versions of Glib's g_assert() satisfy this requirement. |
| 260 | */ |
| 261 | #ifdef __MINGW32__ |
| 262 | #undef assert |
| 263 | #define assert(x) g_assert(x) |
| 264 | #endif |
| 265 | |
| 266 | /** |
| 267 | * qemu_build_not_reached() |
| 268 | * |
| 269 | * The compiler, during optimization, is expected to prove that a call |
| 270 | * to this function cannot be reached and remove it. If the compiler |
| 271 | * supports QEMU_ERROR, this will be reported at compile time; otherwise |
| 272 | * this will be reported at link time due to the missing symbol. |
| 273 | */ |
| 274 | G_NORETURN |
| 275 | void QEMU_ERROR("code path is reachable") |
| 276 | qemu_build_not_reached_always(void); |
| 277 | #if defined(__OPTIMIZE__) && !defined(__NO_INLINE__) |
| 278 | #define qemu_build_not_reached() qemu_build_not_reached_always() |
| 279 | #else |
| 280 | #define qemu_build_not_reached() g_assert_not_reached() |
| 281 | #endif |
| 282 | |
| 283 | /** |
| 284 | * qemu_build_assert() |
| 285 | * |
| 286 | * The compiler, during optimization, is expected to prove that the |
| 287 | * assertion is true. |
| 288 | */ |
| 289 | #define qemu_build_assert(test) while (!(test)) qemu_build_not_reached() |
| 290 | |
| 291 | /* |
| 292 | * According to waitpid man page: |
| 293 | * WCOREDUMP |
| 294 | * This macro is not specified in POSIX.1-2001 and is not |
| 295 | * available on some UNIX implementations (e.g., AIX, SunOS). |
| 296 | * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif. |
| 297 | */ |
| 298 | #ifndef WCOREDUMP |
| 299 | #define WCOREDUMP(status) 0 |
| 300 | #endif |
| 301 | /* |
| 302 | * We have a lot of unaudited code that may fail in strange ways, or |
| 303 | * even be a security risk during migration, if you disable assertions |
| 304 | * at compile-time. You may comment out these safety checks if you |
| 305 | * absolutely want to disable assertion overhead, but it is not |
| 306 | * supported upstream so the risk is all yours. Meanwhile, please |
| 307 | * submit patches to remove any side-effects inside an assertion, or |
| 308 | * fixing error handling that should use Error instead of assert. |
| 309 | */ |
| 310 | #ifdef NDEBUG |
| 311 | #error building with NDEBUG is not supported |
| 312 | #endif |
| 313 | #ifdef G_DISABLE_ASSERT |
| 314 | #error building with G_DISABLE_ASSERT is not supported |
| 315 | #endif |
| 316 | |
| 317 | #ifndef OFF_MAX |
| 318 | #define OFF_MAX (sizeof (off_t) == 8 ? INT64_MAX : INT32_MAX) |
| 319 | #endif |
| 320 | |
| 321 | #ifndef O_LARGEFILE |
| 322 | #define O_LARGEFILE 0 |
| 323 | #endif |
| 324 | #ifndef O_BINARY |
| 325 | #define O_BINARY 0 |
| 326 | #endif |
| 327 | #ifndef MAP_ANONYMOUS |
| 328 | #define MAP_ANONYMOUS MAP_ANON |
| 329 | #endif |
| 330 | #ifndef MAP_NORESERVE |
| 331 | #define MAP_NORESERVE 0 |
| 332 | #endif |
| 333 | #ifndef ENOMEDIUM |
| 334 | #define ENOMEDIUM ENODEV |
| 335 | #endif |
| 336 | #if !defined(ENOTSUP) |
| 337 | #define ENOTSUP 4096 |
| 338 | #endif |
| 339 | #if !defined(ECANCELED) |
| 340 | #define ECANCELED 4097 |
| 341 | #endif |
| 342 | #if !defined(EMEDIUMTYPE) |
| 343 | #define EMEDIUMTYPE 4098 |
| 344 | #endif |
| 345 | #if !defined(ESHUTDOWN) |
| 346 | #define ESHUTDOWN 4099 |
| 347 | #endif |
| 348 | |
| 349 | #define RETRY_ON_EINTR(expr) \ |
| 350 | (__extension__ \ |
| 351 | ({ typeof(expr) __result; \ |
| 352 | do { \ |
| 353 | __result = (expr); \ |
| 354 | } while (__result == -1 && errno == EINTR); \ |
| 355 | __result; })) |
| 356 | |
| 357 | /* time_t may be either 32 or 64 bits depending on the host OS, and |
| 358 | * can be either signed or unsigned, so we can't just hardcode a |
| 359 | * specific maximum value. This is not a C preprocessor constant, |
| 360 | * so you can't use TIME_MAX in an #ifdef, but for our purposes |
| 361 | * this isn't a problem. |
| 362 | */ |
| 363 | |
| 364 | /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from |
| 365 | * Gnulib, and are under the LGPL v2.1 or (at your option) any |
| 366 | * later version. |
| 367 | */ |
| 368 | |
| 369 | /* True if the real type T is signed. */ |
| 370 | #define TYPE_SIGNED(t) (!((t)0 < (t)-1)) |
| 371 | |
| 372 | /* The width in bits of the integer type or expression T. |
| 373 | * Padding bits are not supported. |
| 374 | */ |
| 375 | #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT) |
| 376 | |
| 377 | /* The maximum and minimum values for the integer type T. */ |
| 378 | #define TYPE_MAXIMUM(t) \ |
| 379 | ((t) (!TYPE_SIGNED(t) \ |
| 380 | ? (t)-1 \ |
| 381 | : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1))) |
| 382 | |
| 383 | #ifndef TIME_MAX |
| 384 | #define TIME_MAX TYPE_MAXIMUM(time_t) |
| 385 | #endif |
| 386 | |
| 387 | #ifndef PATH_MAX |
| 388 | #define PATH_MAX 1024 |
| 389 | #endif |
| 390 | |
| 391 | /* |
| 392 | * Use the same value as Linux for now. |
| 393 | */ |
| 394 | #ifndef IOV_MAX |
| 395 | #define IOV_MAX 1024 |
| 396 | #endif |
| 397 | |
| 398 | /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with |
| 399 | * the wrong type. Our replacement isn't usable in preprocessor |
| 400 | * expressions, but it is sufficient for our needs. */ |
| 401 | #ifdef HAVE_BROKEN_SIZE_MAX |
| 402 | #undef SIZE_MAX |
| 403 | #define SIZE_MAX ((size_t)-1) |
| 404 | #endif |
| 405 | |
| 406 | /* |
| 407 | * Two variations of MIN/MAX macros. The first is for runtime use, and |
| 408 | * evaluates arguments only once (so it is safe even with side |
| 409 | * effects), but will not work in constant contexts (such as array |
| 410 | * size declarations) because of the '{}'. The second is for constant |
| 411 | * expression use, where evaluating arguments twice is safe because |
| 412 | * the result is going to be constant anyway, but will not work in a |
| 413 | * runtime context because of a void expression where a value is |
| 414 | * expected. Thus, both gcc and clang will fail to compile if you use |
| 415 | * the wrong macro (even if the error may seem a bit cryptic). |
| 416 | * |
| 417 | * Note that neither form is usable as an #if condition; if you truly |
| 418 | * need to write conditional code that depends on a minimum or maximum |
| 419 | * determined by the pre-processor instead of the compiler, you'll |
| 420 | * have to open-code it. Sadly, Coverity is severely confused by the |
| 421 | * constant variants, so we have to dumb things down there. |
| 422 | * |
| 423 | * Preprocessor sorcery ahead: use different identifiers for the local |
| 424 | * variables in each expansion, so we can nest macro calls without |
| 425 | * shadowing variables. |
| 426 | */ |
| 427 | #define MIN_INTERNAL(a, b, _a, _b) \ |
| 428 | ({ \ |
| 429 | typeof(1 ? (a) : (b)) _a = (a), _b = (b); \ |
| 430 | _a < _b ? _a : _b; \ |
| 431 | }) |
| 432 | #undef MIN |
| 433 | #define MIN(a, b) \ |
| 434 | MIN_INTERNAL((a), (b), MAKE_IDENTIFIER(_a), MAKE_IDENTIFIER(_b)) |
| 435 | |
| 436 | #define MAX_INTERNAL(a, b, _a, _b) \ |
| 437 | ({ \ |
| 438 | typeof(1 ? (a) : (b)) _a = (a), _b = (b); \ |
| 439 | _a > _b ? _a : _b; \ |
| 440 | }) |
| 441 | #undef MAX |
| 442 | #define MAX(a, b) \ |
| 443 | MAX_INTERNAL((a), (b), MAKE_IDENTIFIER(_a), MAKE_IDENTIFIER(_b)) |
| 444 | |
| 445 | #ifdef __COVERITY__ |
| 446 | # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b)) |
| 447 | # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b)) |
| 448 | #else |
| 449 | # define MIN_CONST(a, b) \ |
| 450 | __builtin_choose_expr( \ |
| 451 | __builtin_constant_p(a) && __builtin_constant_p(b), \ |
| 452 | (a) < (b) ? (a) : (b), \ |
| 453 | ((void)0)) |
| 454 | # define MAX_CONST(a, b) \ |
| 455 | __builtin_choose_expr( \ |
| 456 | __builtin_constant_p(a) && __builtin_constant_p(b), \ |
| 457 | (a) > (b) ? (a) : (b), \ |
| 458 | ((void)0)) |
| 459 | #endif |
| 460 | |
| 461 | /* |
| 462 | * Minimum function that returns zero only if both values are zero. |
| 463 | * Intended for use with unsigned values only. |
| 464 | * |
| 465 | * Preprocessor sorcery ahead: use different identifiers for the local |
| 466 | * variables in each expansion, so we can nest macro calls without |
| 467 | * shadowing variables. |
| 468 | */ |
| 469 | #define MIN_NON_ZERO_INTERNAL(a, b, _a, _b) \ |
| 470 | ({ \ |
| 471 | typeof(1 ? (a) : (b)) _a = (a), _b = (b); \ |
| 472 | _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \ |
| 473 | }) |
| 474 | #define MIN_NON_ZERO(a, b) \ |
| 475 | MIN_NON_ZERO_INTERNAL((a), (b), MAKE_IDENTIFIER(_a), MAKE_IDENTIFIER(_b)) |
| 476 | |
| 477 | /* |
| 478 | * Round number down to multiple. Safe when m is not a power of 2 (see |
| 479 | * ROUND_DOWN for a faster version when a power of 2 is guaranteed). |
| 480 | */ |
| 481 | #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m)) |
| 482 | |
| 483 | /* |
| 484 | * Round number up to multiple. Safe when m is not a power of 2 (see |
| 485 | * ROUND_UP for a faster version when a power of 2 is guaranteed). |
| 486 | */ |
| 487 | #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m)) |
| 488 | |
| 489 | /* Check if n is a multiple of m */ |
| 490 | #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0) |
| 491 | |
| 492 | /* n-byte align pointer down */ |
| 493 | #define QEMU_ALIGN_PTR_DOWN(p, n) \ |
| 494 | ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n))) |
| 495 | |
| 496 | /* n-byte align pointer up */ |
| 497 | #define QEMU_ALIGN_PTR_UP(p, n) \ |
| 498 | ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n))) |
| 499 | |
| 500 | /* Check if pointer p is n-bytes aligned */ |
| 501 | #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n)) |
| 502 | |
| 503 | /* |
| 504 | * Round number down to multiple. Requires that d be a power of 2 (see |
| 505 | * QEMU_ALIGN_UP for a safer but slower version on arbitrary |
| 506 | * numbers); works even if d is a smaller type than n. |
| 507 | */ |
| 508 | #ifndef ROUND_DOWN |
| 509 | #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d))) |
| 510 | #endif |
| 511 | |
| 512 | /* |
| 513 | * Round number up to multiple. Requires that d be a power of 2 (see |
| 514 | * QEMU_ALIGN_UP for a safer but slower version on arbitrary |
| 515 | * numbers); works even if d is a smaller type than n. |
| 516 | */ |
| 517 | #ifndef ROUND_UP |
| 518 | #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d)) |
| 519 | #endif |
| 520 | |
| 521 | #ifndef DIV_ROUND_UP |
| 522 | #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d)) |
| 523 | #endif |
| 524 | |
| 525 | /* |
| 526 | * &(x)[0] is always a pointer - if it's same type as x then the argument is a |
| 527 | * pointer, not an array. |
| 528 | */ |
| 529 | #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \ |
| 530 | typeof(&(x)[0]))) |
| 531 | #ifndef ARRAY_SIZE |
| 532 | #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \ |
| 533 | QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x))) |
| 534 | #endif |
| 535 | |
| 536 | int qemu_daemon(int nochdir, int noclose); |
| 537 | void *qemu_anon_ram_alloc(size_t size, uint64_t *align, bool shared, |
| 538 | bool noreserve); |
| 539 | void qemu_anon_ram_free(void *ptr, size_t size); |
| 540 | int qemu_shm_alloc(size_t size, Error **errp); |
| 541 | |
| 542 | #ifdef _WIN32 |
| 543 | #define HAVE_CHARDEV_SERIAL 1 |
| 544 | #define HAVE_CHARDEV_PARALLEL 1 |
| 545 | #else |
| 546 | #if defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \ |
| 547 | || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \ |
| 548 | || defined(__GLIBC__) || defined(__APPLE__) |
| 549 | #define HAVE_CHARDEV_SERIAL 1 |
| 550 | #endif |
| 551 | #if defined(__linux__) || defined(__FreeBSD__) \ |
| 552 | || defined(__FreeBSD_kernel__) || defined(__DragonFly__) |
| 553 | #define HAVE_CHARDEV_PARALLEL 1 |
| 554 | #endif |
| 555 | #endif |
| 556 | |
| 557 | #if defined(__HAIKU__) |
| 558 | #define SIGIO SIGPOLL |
| 559 | #endif |
| 560 | |
| 561 | #ifdef HAVE_MADVISE_WITHOUT_PROTOTYPE |
| 562 | /* |
| 563 | * See MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156) for discussion |
| 564 | * about Solaris missing the madvise() prototype. |
| 565 | */ |
| 566 | int madvise(char *, size_t, int); |
| 567 | #endif |
| 568 | |
| 569 | #if defined(CONFIG_LINUX) |
| 570 | #ifndef BUS_MCEERR_AR |
| 571 | #define BUS_MCEERR_AR 4 |
| 572 | #endif |
| 573 | #ifndef BUS_MCEERR_AO |
| 574 | #define BUS_MCEERR_AO 5 |
| 575 | #endif |
| 576 | #endif |
| 577 | |
| 578 | #if defined(__linux__) && \ |
| 579 | (defined(__x86_64__) || defined(__aarch64__) \ |
| 580 | || defined(__powerpc64__) || defined(__riscv)) |
| 581 | /* Use 2 MiB alignment so transparent hugepages can be used by KVM. |
| 582 | Valgrind does not support alignments larger than 1 MiB, |
| 583 | therefore we need special code which handles running on Valgrind. */ |
| 584 | # define QEMU_VMALLOC_ALIGN (512 * 4096) |
| 585 | #elif defined(__linux__) && defined(__s390x__) |
| 586 | /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */ |
| 587 | # define QEMU_VMALLOC_ALIGN (256 * 4096) |
| 588 | #elif defined(__linux__) && defined(__sparc__) |
| 589 | # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size(), SHMLBA) |
| 590 | #elif defined(__linux__) && defined(__loongarch__) |
| 591 | /* |
| 592 | * For transparent hugepage optimization, it has better be huge page |
| 593 | * aligned. LoongArch host system supports two kinds of pagesize: 4K |
| 594 | * and 16K, here calculate huge page size from host page size |
| 595 | */ |
| 596 | # define QEMU_VMALLOC_ALIGN (qemu_real_host_page_size() * \ |
| 597 | qemu_real_host_page_size() / sizeof(long)) |
| 598 | #else |
| 599 | # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size() |
| 600 | #endif |
| 601 | |
| 602 | #ifdef CONFIG_POSIX |
| 603 | struct qemu_signalfd_siginfo { |
| 604 | uint32_t ssi_signo; /* Signal number */ |
| 605 | int32_t ssi_errno; /* Error number (unused) */ |
| 606 | int32_t ssi_code; /* Signal code */ |
| 607 | uint32_t ssi_pid; /* PID of sender */ |
| 608 | uint32_t ssi_uid; /* Real UID of sender */ |
| 609 | int32_t ssi_fd; /* File descriptor (SIGIO) */ |
| 610 | uint32_t ssi_tid; /* Kernel timer ID (POSIX timers) */ |
| 611 | uint32_t ssi_band; /* Band event (SIGIO) */ |
| 612 | uint32_t ssi_overrun; /* POSIX timer overrun count */ |
| 613 | uint32_t ssi_trapno; /* Trap number that caused signal */ |
| 614 | int32_t ssi_status; /* Exit status or signal (SIGCHLD) */ |
| 615 | int32_t ssi_int; /* Integer sent by sigqueue(2) */ |
| 616 | uint64_t ssi_ptr; /* Pointer sent by sigqueue(2) */ |
| 617 | uint64_t ssi_utime; /* User CPU time consumed (SIGCHLD) */ |
| 618 | uint64_t ssi_stime; /* System CPU time consumed (SIGCHLD) */ |
| 619 | uint64_t ssi_addr; /* Address that generated signal |
| 620 | (for hardware-generated signals) */ |
| 621 | uint8_t pad[48]; /* Pad size to 128 bytes (allow for |
| 622 | additional fields in the future) */ |
| 623 | }; |
| 624 | |
| 625 | int qemu_signalfd(const sigset_t *mask); |
| 626 | void sigaction_invoke(struct sigaction *action, |
| 627 | struct qemu_signalfd_siginfo *info); |
| 628 | #endif |
| 629 | |
| 630 | /* |
| 631 | * Don't introduce new usage of this function, prefer the following |
| 632 | * qemu_open/qemu_create that take an "Error **errp" |
| 633 | */ |
| 634 | int qemu_open_old(const char *name, int flags, ...); |
| 635 | int qemu_open(const char *name, int flags, Error **errp); |
| 636 | int qemu_create(const char *name, int flags, mode_t mode, Error **errp); |
| 637 | int qemu_close(int fd); |
| 638 | int qemu_unlink(const char *name); |
| 639 | #ifndef _WIN32 |
| 640 | int qemu_dup_flags(int fd, int flags); |
| 641 | int qemu_dup(int fd); |
| 642 | int qemu_lock_fd(int fd, int64_t start, int64_t len, bool exclusive); |
| 643 | int qemu_unlock_fd(int fd, int64_t start, int64_t len); |
| 644 | int qemu_lock_fd_test(int fd, int64_t start, int64_t len, bool exclusive); |
| 645 | bool qemu_has_ofd_lock(void); |
| 646 | int qemu_fcntl_addfl(int fd, int flag); |
| 647 | #endif |
| 648 | |
| 649 | bool qemu_has_direct_io(void); |
| 650 | |
| 651 | #ifdef WIN64 |
| 652 | #define FMT_pid "%" PRId64 |
| 653 | #else |
| 654 | #define FMT_pid "%d" |
| 655 | #endif |
| 656 | |
| 657 | bool qemu_write_pidfile(const char *pidfile, Error **errp); |
| 658 | |
| 659 | int qemu_get_thread_id(void); |
| 660 | |
| 661 | /** |
| 662 | * qemu_kill_thread: |
| 663 | * @tid: thread id. |
| 664 | * @sig: host signal. |
| 665 | * |
| 666 | * Send @sig to one of QEMU's own threads with identifier @tid. |
| 667 | */ |
| 668 | int qemu_kill_thread(int tid, int sig); |
| 669 | |
| 670 | #ifndef CONFIG_IOVEC |
| 671 | struct iovec { |
| 672 | void *iov_base; |
| 673 | size_t iov_len; |
| 674 | }; |
| 675 | |
| 676 | ssize_t readv(int fd, const struct iovec *iov, int iov_cnt); |
| 677 | ssize_t writev(int fd, const struct iovec *iov, int iov_cnt); |
| 678 | #endif |
| 679 | |
| 680 | #ifdef _WIN32 |
| 681 | static inline void qemu_timersub(const struct timeval *val1, |
| 682 | const struct timeval *val2, |
| 683 | struct timeval *res) |
| 684 | { |
| 685 | res->tv_sec = val1->tv_sec - val2->tv_sec; |
| 686 | if (val1->tv_usec < val2->tv_usec) { |
| 687 | res->tv_sec--; |
| 688 | res->tv_usec = val1->tv_usec - val2->tv_usec + 1000 * 1000; |
| 689 | } else { |
| 690 | res->tv_usec = val1->tv_usec - val2->tv_usec; |
| 691 | } |
| 692 | } |
| 693 | #else |
| 694 | #define qemu_timersub timersub |
| 695 | #endif |
| 696 | |
| 697 | ssize_t qemu_write_full(int fd, const void *buf, size_t count) |
| 698 | G_GNUC_WARN_UNUSED_RESULT; |
| 699 | |
| 700 | void qemu_set_cloexec(int fd); |
| 701 | bool qemu_set_blocking(int fd, bool block, Error **errp); |
| 702 | |
| 703 | /* |
| 704 | * Clear FD_CLOEXEC for a descriptor. |
| 705 | * |
| 706 | * The caller must guarantee that no other fork+exec's occur before the |
| 707 | * exec that is intended to inherit this descriptor, eg by suspending CPUs |
| 708 | * and blocking monitor commands. |
| 709 | */ |
| 710 | void qemu_clear_cloexec(int fd); |
| 711 | |
| 712 | /* Return a dynamically allocated directory path that is appropriate for storing |
| 713 | * local state. |
| 714 | * |
| 715 | * The caller is responsible for releasing the value returned with g_free() |
| 716 | * after use. |
| 717 | */ |
| 718 | char *qemu_get_local_state_dir(void); |
| 719 | |
| 720 | /** |
| 721 | * qemu_getauxval: |
| 722 | * @type: the auxiliary vector key to lookup |
| 723 | * |
| 724 | * Search the auxiliary vector for @type, returning the value |
| 725 | * or 0 if @type is not present. |
| 726 | */ |
| 727 | unsigned long qemu_getauxval(unsigned long type); |
| 728 | |
| 729 | void qemu_set_tty_echo(int fd, bool echo); |
| 730 | |
| 731 | typedef struct ThreadContext ThreadContext; |
| 732 | |
| 733 | /** |
| 734 | * qemu_prealloc_mem: |
| 735 | * @fd: the fd mapped into the area, -1 for anonymous memory |
| 736 | * @area: start address of the are to preallocate |
| 737 | * @sz: the size of the area to preallocate |
| 738 | * @max_threads: maximum number of threads to use |
| 739 | * @tc: prealloc context threads pointer, NULL if not in use |
| 740 | * @async: request asynchronous preallocation, requires @tc |
| 741 | * @errp: returns an error if this function fails |
| 742 | * |
| 743 | * Preallocate memory (populate/prefault page tables writable) for the virtual |
| 744 | * memory area starting at @area with the size of @sz. After a successful call, |
| 745 | * each page in the area was faulted in writable at least once, for example, |
| 746 | * after allocating file blocks for mapped files. |
| 747 | * |
| 748 | * When setting @async, allocation might be performed asynchronously. |
| 749 | * qemu_finish_async_prealloc_mem() must be called to finish any asynchronous |
| 750 | * preallocation. |
| 751 | * |
| 752 | * Return: true on success, else false setting @errp with error. |
| 753 | */ |
| 754 | bool qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads, |
| 755 | ThreadContext *tc, bool async, Error **errp); |
| 756 | |
| 757 | /** |
| 758 | * qemu_finish_async_prealloc_mem: |
| 759 | * @errp: returns an error if this function fails |
| 760 | * |
| 761 | * Finish all outstanding asynchronous memory preallocation. |
| 762 | * |
| 763 | * Return: true on success, else false setting @errp with error. |
| 764 | */ |
| 765 | bool qemu_finish_async_prealloc_mem(Error **errp); |
| 766 | |
| 767 | /** |
| 768 | * qemu_get_pid_name: |
| 769 | * @pid: pid of a process |
| 770 | * |
| 771 | * For given @pid fetch its name. Caller is responsible for |
| 772 | * freeing the string when no longer needed. |
| 773 | * Returns allocated string on success, NULL on failure. |
| 774 | */ |
| 775 | char *qemu_get_pid_name(pid_t pid); |
| 776 | |
| 777 | /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even |
| 778 | * when intptr_t is 32-bit and we are aligning a long long. |
| 779 | */ |
| 780 | static inline uintptr_t qemu_real_host_page_size(void) |
| 781 | { |
| 782 | return getpagesize(); |
| 783 | } |
| 784 | |
| 785 | static inline intptr_t qemu_real_host_page_mask(void) |
| 786 | { |
| 787 | return -(intptr_t)qemu_real_host_page_size(); |
| 788 | } |
| 789 | |
| 790 | /* |
| 791 | * After using getopt or getopt_long, if you need to parse another set |
| 792 | * of options, then you must reset optind. Unfortunately the way to |
| 793 | * do this varies between implementations of getopt. |
| 794 | */ |
| 795 | static inline void qemu_reset_optind(void) |
| 796 | { |
| 797 | #ifdef HAVE_OPTRESET |
| 798 | optind = 1; |
| 799 | optreset = 1; |
| 800 | #else |
| 801 | optind = 0; |
| 802 | #endif |
| 803 | } |
| 804 | |
| 805 | int qemu_fdatasync(int fd); |
| 806 | |
| 807 | /** |
| 808 | * qemu_close_all_open_fd: |
| 809 | * |
| 810 | * Close all open file descriptors except the ones supplied in the @skip array |
| 811 | * |
| 812 | * @skip: ordered array of distinct file descriptors that should not be closed |
| 813 | * if any, or NULL. |
| 814 | * @nskip: number of entries in the @skip array or 0 if @skip is NULL. |
| 815 | */ |
| 816 | void qemu_close_all_open_fd(const int *skip, unsigned int nskip); |
| 817 | |
| 818 | /** |
| 819 | * Sync changes made to the memory mapped file back to the backing |
| 820 | * storage. For POSIX compliant systems this will fallback |
| 821 | * to regular msync call. Otherwise it will trigger whole file sync |
| 822 | * (including the metadata case there is no support to skip that otherwise) |
| 823 | * |
| 824 | * @addr - start of the memory area to be synced |
| 825 | * @length - length of the are to be synced |
| 826 | * @fd - file descriptor for the file to be synced |
| 827 | * (mandatory only for POSIX non-compliant systems) |
| 828 | */ |
| 829 | int qemu_msync(void *addr, size_t length, int fd); |
| 830 | |
| 831 | /** |
| 832 | * qemu_get_host_physmem: |
| 833 | * |
| 834 | * Operating system agnostic way of querying host memory. |
| 835 | * |
| 836 | * Returns amount of physical memory on the system. This is purely |
| 837 | * advisery and may return 0 if we can't work it out. At the other |
| 838 | * end we saturate to SIZE_MAX if you are lucky enough to have that |
| 839 | * much memory. |
| 840 | */ |
| 841 | size_t qemu_get_host_physmem(void); |
| 842 | |
| 843 | /* |
| 844 | * Toggle write/execute on the pages marked MAP_JIT |
| 845 | * for the current thread. |
| 846 | */ |
| 847 | #ifdef __APPLE__ |
| 848 | static inline void qemu_thread_jit_execute(void) |
| 849 | { |
| 850 | pthread_jit_write_protect_np(true); |
| 851 | } |
| 852 | |
| 853 | static inline void qemu_thread_jit_write(void) |
| 854 | { |
| 855 | pthread_jit_write_protect_np(false); |
| 856 | } |
| 857 | #else |
| 858 | static inline void qemu_thread_jit_write(void) {} |
| 859 | static inline void qemu_thread_jit_execute(void) {} |
| 860 | #endif |
| 861 | |
| 862 | /** |
| 863 | * Platforms which do not support system() return ENOSYS |
| 864 | */ |
| 865 | #ifndef HAVE_SYSTEM_FUNCTION |
| 866 | #define system platform_does_not_support_system |
| 867 | static inline int platform_does_not_support_system(const char *command) |
| 868 | { |
| 869 | errno = ENOSYS; |
| 870 | return -1; |
| 871 | } |
| 872 | #endif /* !HAVE_SYSTEM_FUNCTION */ |
| 873 | |
| 874 | #ifdef __cplusplus |
| 875 | } |
| 876 | #endif |
| 877 | |
| 878 | #endif |