util/oslib-posix.c

   1 /*
   2  * os-posix-lib.c
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2010 Red Hat, Inc.
   6  *
   7  * QEMU library functions on POSIX which are shared between QEMU and
   8  * the QEMU tools.
   9  *
  10  * Permission is hereby granted, free of charge, to any person obtaining a copy
  11  * of this software and associated documentation files (the "Software"), to deal
  12  * in the Software without restriction, including without limitation the rights
  13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  14  * copies of the Software, and to permit persons to whom the Software is
  15  * furnished to do so, subject to the following conditions:
  16  *
  17  * The above copyright notice and this permission notice shall be included in
  18  * all copies or substantial portions of the Software.
  19  *
  20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  26  * THE SOFTWARE.
  27  */
  28
  29 #include "qemu/osdep.h"
  30 #include <termios.h>
  31
  32 #include <glib/gprintf.h>
  33
  34 #include "qemu-common.h"
  35 #include "sysemu/sysemu.h"
  36 #include "trace.h"
  37 #include "qapi/error.h"
  38 #include "qemu/sockets.h"
  39 #include "qemu/thread.h"
  40 #include <libgen.h>
  41 #include <sys/signal.h>
  42 #include "qemu/cutils.h"
  43
  44 #ifdef CONFIG_LINUX
  45 #include <sys/syscall.h>
  46 #endif
  47
  48 #ifdef __FreeBSD__
  49 #include <sys/sysctl.h>
  50 #include <sys/user.h>
  51 #include <sys/thr.h>
  52 #include <libutil.h>
  53 #endif
  54
  55 #ifdef __NetBSD__
  56 #include <sys/sysctl.h>
  57 #include <lwp.h>
  58 #endif
  59
  60 #ifdef __APPLE__
  61 #include <mach-o/dyld.h>
  62 #endif
  63
  64 #include "qemu/mmap-alloc.h"
  65
  66 #ifdef CONFIG_DEBUG_STACK_USAGE
  67 #include "qemu/error-report.h"
  68 #endif
  69
  70 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  71
  72 struct MemsetThread {
  73     char *addr;
  74     size_t numpages;
  75     size_t hpagesize;
  76     QemuThread pgthread;
  77     sigjmp_buf env;
  78 };
  79 typedef struct MemsetThread MemsetThread;
  80
  81 static MemsetThread *memset_thread;
  82 static int memset_num_threads;
  83 static bool memset_thread_failed;
  84
  85 static QemuMutex page_mutex;
  86 static QemuCond page_cond;
  87 static bool threads_created_flag;
  88
  89 int qemu_get_thread_id(void)
  90 {
  91 #if defined(__linux__)
  92     return syscall(SYS_gettid);
  93 #elif defined(__FreeBSD__)
  94     /* thread id is up to INT_MAX */
  95     long tid;
  96     thr_self(&tid);
  97     return (int)tid;
  98 #elif defined(__NetBSD__)
  99     return _lwp_self();
 100 #else
 101     return getpid();
 102 #endif
 103 }
 104
 105 int qemu_daemon(int nochdir, int noclose)
 106 {
 107     return daemon(nochdir, noclose);
 108 }
 109
 110 bool qemu_write_pidfile(const char *path, Error **errp)
 111 {
 112     int fd;
 113     char pidstr[32];
 114
 115     while (1) {
 116         struct stat a, b;
 117         struct flock lock = {
 118             .l_type = F_WRLCK,
 119             .l_whence = SEEK_SET,
 120             .l_len = 0,
 121         };
 122
 123         fd = qemu_open(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
 124         if (fd == -1) {
 125             error_setg_errno(errp, errno, "Cannot open pid file");
 126             return false;
 127         }
 128
 129         if (fstat(fd, &b) < 0) {
 130             error_setg_errno(errp, errno, "Cannot stat file");
 131             goto fail_close;
 132         }
 133
 134         if (fcntl(fd, F_SETLK, &lock)) {
 135             error_setg_errno(errp, errno, "Cannot lock pid file");
 136             goto fail_close;
 137         }
 138
 139         /*
 140          * Now make sure the path we locked is the same one that now
 141          * exists on the filesystem.
 142          */
 143         if (stat(path, &a) < 0) {
 144             /*
 145              * PID file disappeared, someone else must be racing with
 146              * us, so try again.
 147              */
 148             close(fd);
 149             continue;
 150         }
 151
 152         if (a.st_ino == b.st_ino) {
 153             break;
 154         }
 155
 156         /*
 157          * PID file was recreated, someone else must be racing with
 158          * us, so try again.
 159          */
 160         close(fd);
 161     }
 162
 163     if (ftruncate(fd, 0) < 0) {
 164         error_setg_errno(errp, errno, "Failed to truncate pid file");
 165         goto fail_unlink;
 166     }
 167
 168     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 169     if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 170         error_setg(errp, "Failed to write pid file");
 171         goto fail_unlink;
 172     }
 173
 174     return true;
 175
 176 fail_unlink:
 177     unlink(path);
 178 fail_close:
 179     close(fd);
 180     return false;
 181 }
 182
 183 void *qemu_oom_check(void *ptr)
 184 {
 185     if (ptr == NULL) {
 186         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
 187         abort();
 188     }
 189     return ptr;
 190 }
 191
 192 void *qemu_try_memalign(size_t alignment, size_t size)
 193 {
 194     void *ptr;
 195
 196     if (alignment < sizeof(void*)) {
 197         alignment = sizeof(void*);
 198     }
 199
 200 #if defined(CONFIG_POSIX_MEMALIGN)
 201     int ret;
 202     ret = posix_memalign(&ptr, alignment, size);
 203     if (ret != 0) {
 204         errno = ret;
 205         ptr = NULL;
 206     }
 207 #elif defined(CONFIG_BSD)
 208     ptr = valloc(size);
 209 #else
 210     ptr = memalign(alignment, size);
 211 #endif
 212     trace_qemu_memalign(alignment, size, ptr);
 213     return ptr;
 214 }
 215
 216 void *qemu_memalign(size_t alignment, size_t size)
 217 {
 218     return qemu_oom_check(qemu_try_memalign(alignment, size));
 219 }
 220
 221 /* alloc shared memory pages */
 222 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
 223 {
 224     size_t align = QEMU_VMALLOC_ALIGN;
 225     void *ptr = qemu_ram_mmap(-1, size, align, shared, false);
 226
 227     if (ptr == MAP_FAILED) {
 228         return NULL;
 229     }
 230
 231     if (alignment) {
 232         *alignment = align;
 233     }
 234
 235     trace_qemu_anon_ram_alloc(size, ptr);
 236     return ptr;
 237 }
 238
 239 void qemu_vfree(void *ptr)
 240 {
 241     trace_qemu_vfree(ptr);
 242     free(ptr);
 243 }
 244
 245 void qemu_anon_ram_free(void *ptr, size_t size)
 246 {
 247     trace_qemu_anon_ram_free(ptr, size);
 248     qemu_ram_munmap(-1, ptr, size);
 249 }
 250
 251 void qemu_set_block(int fd)
 252 {
 253     int f;
 254     f = fcntl(fd, F_GETFL);
 255     assert(f != -1);
 256     f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 257     assert(f != -1);
 258 }
 259
 260 void qemu_set_nonblock(int fd)
 261 {
 262     int f;
 263     f = fcntl(fd, F_GETFL);
 264     assert(f != -1);
 265     f = fcntl(fd, F_SETFL, f | O_NONBLOCK);
 266 #ifdef __OpenBSD__
 267     if (f == -1) {
 268         /*
 269          * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on
 270          * memory devices and sets errno to ENODEV.
 271          * It's OK if we fail to set O_NONBLOCK on devices like /dev/null,
 272          * because they will never block anyway.
 273          */
 274         assert(errno == ENODEV);
 275     }
 276 #else
 277     assert(f != -1);
 278 #endif
 279 }
 280
 281 int socket_set_fast_reuse(int fd)
 282 {
 283     int val = 1, ret;
 284
 285     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 286                      (const char *)&val, sizeof(val));
 287
 288     assert(ret == 0);
 289
 290     return ret;
 291 }
 292
 293 void qemu_set_cloexec(int fd)
 294 {
 295     int f;
 296     f = fcntl(fd, F_GETFD);
 297     assert(f != -1);
 298     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 299     assert(f != -1);
 300 }
 301
 302 /*
 303  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 304  */
 305 int qemu_pipe(int pipefd[2])
 306 {
 307     int ret;
 308
 309 #ifdef CONFIG_PIPE2
 310     ret = pipe2(pipefd, O_CLOEXEC);
 311     if (ret != -1 || errno != ENOSYS) {
 312         return ret;
 313     }
 314 #endif
 315     ret = pipe(pipefd);
 316     if (ret == 0) {
 317         qemu_set_cloexec(pipefd[0]);
 318         qemu_set_cloexec(pipefd[1]);
 319     }
 320
 321     return ret;
 322 }
 323
 324 char *
 325 qemu_get_local_state_pathname(const char *relative_pathname)
 326 {
 327     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 328                            relative_pathname);
 329 }
 330
 331 void qemu_set_tty_echo(int fd, bool echo)
 332 {
 333     struct termios tty;
 334
 335     tcgetattr(fd, &tty);
 336
 337     if (echo) {
 338         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 339     } else {
 340         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 341     }
 342
 343     tcsetattr(fd, TCSANOW, &tty);
 344 }
 345
 346 static char exec_dir[PATH_MAX];
 347
 348 void qemu_init_exec_dir(const char *argv0)
 349 {
 350     char *dir;
 351     char *p = NULL;
 352     char buf[PATH_MAX];
 353
 354     assert(!exec_dir[0]);
 355
 356 #if defined(__linux__)
 357     {
 358         int len;
 359         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 360         if (len > 0) {
 361             buf[len] = 0;
 362             p = buf;
 363         }
 364     }
 365 #elif defined(__FreeBSD__) \
 366       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 367     {
 368 #if defined(__FreeBSD__)
 369         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 370 #else
 371         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 372 #endif
 373         size_t len = sizeof(buf) - 1;
 374
 375         *buf = '\0';
 376         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 377             *buf) {
 378             buf[sizeof(buf) - 1] = '\0';
 379             p = buf;
 380         }
 381     }
 382 #elif defined(__APPLE__)
 383     {
 384         char fpath[PATH_MAX];
 385         uint32_t len = sizeof(fpath);
 386         if (_NSGetExecutablePath(fpath, &len) == 0) {
 387             p = realpath(fpath, buf);
 388             if (!p) {
 389                 return;
 390             }
 391         }
 392     }
 393 #endif
 394     /* If we don't have any way of figuring out the actual executable
 395        location then try argv[0].  */
 396     if (!p) {
 397         if (!argv0) {
 398             return;
 399         }
 400         p = realpath(argv0, buf);
 401         if (!p) {
 402             return;
 403         }
 404     }
 405     dir = g_path_get_dirname(p);
 406
 407     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 408
 409     g_free(dir);
 410 }
 411
 412 char *qemu_get_exec_dir(void)
 413 {
 414     return g_strdup(exec_dir);
 415 }
 416
 417 static void sigbus_handler(int signal)
 418 {
 419     int i;
 420     if (memset_thread) {
 421         for (i = 0; i < memset_num_threads; i++) {
 422             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 423                 siglongjmp(memset_thread[i].env, 1);
 424             }
 425         }
 426     }
 427 }
 428
 429 static void *do_touch_pages(void *arg)
 430 {
 431     MemsetThread *memset_args = (MemsetThread *)arg;
 432     sigset_t set, oldset;
 433
 434     /*
 435      * On Linux, the page faults from the loop below can cause mmap_sem
 436      * contention with allocation of the thread stacks.  Do not start
 437      * clearing until all threads have been created.
 438      */
 439     qemu_mutex_lock(&page_mutex);
 440     while(!threads_created_flag){
 441         qemu_cond_wait(&page_cond, &page_mutex);
 442     }
 443     qemu_mutex_unlock(&page_mutex);
 444
 445     /* unblock SIGBUS */
 446     sigemptyset(&set);
 447     sigaddset(&set, SIGBUS);
 448     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 449
 450     if (sigsetjmp(memset_args->env, 1)) {
 451         memset_thread_failed = true;
 452     } else {
 453         char *addr = memset_args->addr;
 454         size_t numpages = memset_args->numpages;
 455         size_t hpagesize = memset_args->hpagesize;
 456         size_t i;
 457         for (i = 0; i < numpages; i++) {
 458             /*
 459              * Read & write back the same value, so we don't
 460              * corrupt existing user/app data that might be
 461              * stored.
 462              *
 463              * 'volatile' to stop compiler optimizing this away
 464              * to a no-op
 465              *
 466              * TODO: get a better solution from kernel so we
 467              * don't need to write at all so we don't cause
 468              * wear on the storage backing the region...
 469              */
 470             *(volatile char *)addr = *addr;
 471             addr += hpagesize;
 472         }
 473     }
 474     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 475     return NULL;
 476 }
 477
 478 static inline int get_memset_num_threads(int smp_cpus)
 479 {
 480     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 481     int ret = 1;
 482
 483     if (host_procs > 0) {
 484         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 485     }
 486     /* In case sysconf() fails, we fall back to single threaded */
 487     return ret;
 488 }
 489
 490 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 491                             int smp_cpus)
 492 {
 493     static gsize initialized = 0;
 494     size_t numpages_per_thread, leftover;
 495     char *addr = area;
 496     int i = 0;
 497
 498     if (g_once_init_enter(&initialized)) {
 499         qemu_mutex_init(&page_mutex);
 500         qemu_cond_init(&page_cond);
 501         g_once_init_leave(&initialized, 1);
 502     }
 503
 504     memset_thread_failed = false;
 505     threads_created_flag = false;
 506     memset_num_threads = get_memset_num_threads(smp_cpus);
 507     memset_thread = g_new0(MemsetThread, memset_num_threads);
 508     numpages_per_thread = numpages / memset_num_threads;
 509     leftover = numpages % memset_num_threads;
 510     for (i = 0; i < memset_num_threads; i++) {
 511         memset_thread[i].addr = addr;
 512         memset_thread[i].numpages = numpages_per_thread + (i < leftover);
 513         memset_thread[i].hpagesize = hpagesize;
 514         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 515                            do_touch_pages, &memset_thread[i],
 516                            QEMU_THREAD_JOINABLE);
 517         addr += memset_thread[i].numpages * hpagesize;
 518     }
 519
 520     qemu_mutex_lock(&page_mutex);
 521     threads_created_flag = true;
 522     qemu_cond_broadcast(&page_cond);
 523     qemu_mutex_unlock(&page_mutex);
 524
 525     for (i = 0; i < memset_num_threads; i++) {
 526         qemu_thread_join(&memset_thread[i].pgthread);
 527     }
 528     g_free(memset_thread);
 529     memset_thread = NULL;
 530
 531     return memset_thread_failed;
 532 }
 533
 534 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 535                      Error **errp)
 536 {
 537     int ret;
 538     struct sigaction act, oldact;
 539     size_t hpagesize = qemu_fd_getpagesize(fd);
 540     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 541
 542     memset(&act, 0, sizeof(act));
 543     act.sa_handler = &sigbus_handler;
 544     act.sa_flags = 0;
 545
 546     ret = sigaction(SIGBUS, &act, &oldact);
 547     if (ret) {
 548         error_setg_errno(errp, errno,
 549             "os_mem_prealloc: failed to install signal handler");
 550         return;
 551     }
 552
 553     /* touch pages simultaneously */
 554     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 555         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 556             "pages available to allocate guest RAM");
 557     }
 558
 559     ret = sigaction(SIGBUS, &oldact, NULL);
 560     if (ret) {
 561         /* Terminate QEMU since it can't recover from error */
 562         perror("os_mem_prealloc: failed to reinstall signal handler");
 563         exit(1);
 564     }
 565 }
 566
 567 char *qemu_get_pid_name(pid_t pid)
 568 {
 569     char *name = NULL;
 570
 571 #if defined(__FreeBSD__)
 572     /* BSDs don't have /proc, but they provide a nice substitute */
 573     struct kinfo_proc *proc = kinfo_getproc(pid);
 574
 575     if (proc) {
 576         name = g_strdup(proc->ki_comm);
 577         free(proc);
 578     }
 579 #else
 580     /* Assume a system with reasonable procfs */
 581     char *pid_path;
 582     size_t len;
 583
 584     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 585     g_file_get_contents(pid_path, &name, &len, NULL);
 586     g_free(pid_path);
 587 #endif
 588
 589     return name;
 590 }
 591
 592
 593 pid_t qemu_fork(Error **errp)
 594 {
 595     sigset_t oldmask, newmask;
 596     struct sigaction sig_action;
 597     int saved_errno;
 598     pid_t pid;
 599
 600     /*
 601      * Need to block signals now, so that child process can safely
 602      * kill off caller's signal handlers without a race.
 603      */
 604     sigfillset(&newmask);
 605     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 606         error_setg_errno(errp, errno,
 607                          "cannot block signals");
 608         return -1;
 609     }
 610
 611     pid = fork();
 612     saved_errno = errno;
 613
 614     if (pid < 0) {
 615         /* attempt to restore signal mask, but ignore failure, to
 616          * avoid obscuring the fork failure */
 617         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 618         error_setg_errno(errp, saved_errno,
 619                          "cannot fork child process");
 620         errno = saved_errno;
 621         return -1;
 622     } else if (pid) {
 623         /* parent process */
 624
 625         /* Restore our original signal mask now that the child is
 626          * safely running. Only documented failures are EFAULT (not
 627          * possible, since we are using just-grabbed mask) or EINVAL
 628          * (not possible, since we are using correct arguments).  */
 629         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 630     } else {
 631         /* child process */
 632         size_t i;
 633
 634         /* Clear out all signal handlers from parent so nothing
 635          * unexpected can happen in our child once we unblock
 636          * signals */
 637         sig_action.sa_handler = SIG_DFL;
 638         sig_action.sa_flags = 0;
 639         sigemptyset(&sig_action.sa_mask);
 640
 641         for (i = 1; i < NSIG; i++) {
 642             /* Only possible errors are EFAULT or EINVAL The former
 643              * won't happen, the latter we expect, so no need to check
 644              * return value */
 645             (void)sigaction(i, &sig_action, NULL);
 646         }
 647
 648         /* Unmask all signals in child, since we've no idea what the
 649          * caller's done with their signal mask and don't want to
 650          * propagate that to children */
 651         sigemptyset(&newmask);
 652         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 653             Error *local_err = NULL;
 654             error_setg_errno(&local_err, errno,
 655                              "cannot unblock signals");
 656             error_report_err(local_err);
 657             _exit(1);
 658         }
 659     }
 660     return pid;
 661 }
 662
 663 void *qemu_alloc_stack(size_t *sz)
 664 {
 665     void *ptr, *guardpage;
 666     int flags;
 667 #ifdef CONFIG_DEBUG_STACK_USAGE
 668     void *ptr2;
 669 #endif
 670     size_t pagesz = qemu_real_host_page_size;
 671 #ifdef _SC_THREAD_STACK_MIN
 672     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 673     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 674     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 675 #endif
 676     /* adjust stack size to a multiple of the page size */
 677     *sz = ROUND_UP(*sz, pagesz);
 678     /* allocate one extra page for the guard page */
 679     *sz += pagesz;
 680
 681     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 682 #if defined(MAP_STACK) && defined(__OpenBSD__)
 683     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 684      * Linux/FreeBSD/NetBSD have a flag with the same name
 685      * but have differing functionality. OpenBSD will SEGV
 686      * if it spots execution with a stack pointer pointing
 687      * at memory that was not allocated with MAP_STACK.
 688      */
 689     flags |= MAP_STACK;
 690 #endif
 691
 692     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 693     if (ptr == MAP_FAILED) {
 694         perror("failed to allocate memory for stack");
 695         abort();
 696     }
 697
 698 #if defined(HOST_IA64)
 699     /* separate register stack */
 700     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 701 #elif defined(HOST_HPPA)
 702     /* stack grows up */
 703     guardpage = ptr + *sz - pagesz;
 704 #else
 705     /* stack grows down */
 706     guardpage = ptr;
 707 #endif
 708     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 709         perror("failed to set up stack guard page");
 710         abort();
 711     }
 712
 713 #ifdef CONFIG_DEBUG_STACK_USAGE
 714     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 715         *(uint32_t *)ptr2 = 0xdeadbeaf;
 716     }
 717 #endif
 718
 719     return ptr;
 720 }
 721
 722 #ifdef CONFIG_DEBUG_STACK_USAGE
 723 static __thread unsigned int max_stack_usage;
 724 #endif
 725
 726 void qemu_free_stack(void *stack, size_t sz)
 727 {
 728 #ifdef CONFIG_DEBUG_STACK_USAGE
 729     unsigned int usage;
 730     void *ptr;
 731
 732     for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
 733          ptr += sizeof(uint32_t)) {
 734         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 735             break;
 736         }
 737     }
 738     usage = sz - (uintptr_t) (ptr - stack);
 739     if (usage > max_stack_usage) {
 740         error_report("thread %d max stack usage increased from %u to %u",
 741                      qemu_get_thread_id(), max_stack_usage, usage);
 742         max_stack_usage = usage;
 743     }
 744 #endif
 745
 746     munmap(stack, sz);
 747 }
 748
 749 void sigaction_invoke(struct sigaction *action,
 750                       struct qemu_signalfd_siginfo *info)
 751 {
 752     siginfo_t si = {};
 753     si.si_signo = info->ssi_signo;
 754     si.si_errno = info->ssi_errno;
 755     si.si_code = info->ssi_code;
 756
 757     /* Convert the minimal set of fields defined by POSIX.
 758      * Positive si_code values are reserved for kernel-generated
 759      * signals, where the valid siginfo fields are determined by
 760      * the signal number.  But according to POSIX, it is unspecified
 761      * whether SI_USER and SI_QUEUE have values less than or equal to
 762      * zero.
 763      */
 764     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 765         info->ssi_code <= 0) {
 766         /* SIGTERM, etc.  */
 767         si.si_pid = info->ssi_pid;
 768         si.si_uid = info->ssi_uid;
 769     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 770                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 771         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 772     } else if (info->ssi_signo == SIGCHLD) {
 773         si.si_pid = info->ssi_pid;
 774         si.si_status = info->ssi_status;
 775         si.si_uid = info->ssi_uid;
 776     }
 777     action->sa_sigaction(info->ssi_signo, &si, NULL);
 778 }