hw/arm/virt: Add 5.0 HW compat props
[qemu/ar7.git] / util / oslib-posix.c
blob916f1be2243ac2b5ea53297fbfaaa365c3e5e0bf
1 /*
2 * os-posix-lib.c
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2010 Red Hat, Inc.
7 * QEMU library functions on POSIX which are shared between QEMU and
8 * the QEMU tools.
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:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
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.
29 #include "qemu/osdep.h"
30 #include <termios.h>
32 #include <glib/gprintf.h>
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"
44 #ifdef CONFIG_LINUX
45 #include <sys/syscall.h>
46 #endif
48 #ifdef __FreeBSD__
49 #include <sys/sysctl.h>
50 #include <sys/user.h>
51 #include <sys/thr.h>
52 #include <libutil.h>
53 #endif
55 #ifdef __NetBSD__
56 #include <sys/sysctl.h>
57 #include <lwp.h>
58 #endif
60 #include "qemu/mmap-alloc.h"
62 #ifdef CONFIG_DEBUG_STACK_USAGE
63 #include "qemu/error-report.h"
64 #endif
66 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
68 struct MemsetThread {
69 char *addr;
70 size_t numpages;
71 size_t hpagesize;
72 QemuThread pgthread;
73 sigjmp_buf env;
75 typedef struct MemsetThread MemsetThread;
77 static MemsetThread *memset_thread;
78 static int memset_num_threads;
79 static bool memset_thread_failed;
81 static QemuMutex page_mutex;
82 static QemuCond page_cond;
83 static bool threads_created_flag;
85 int qemu_get_thread_id(void)
87 #if defined(__linux__)
88 return syscall(SYS_gettid);
89 #elif defined(__FreeBSD__)
90 /* thread id is up to INT_MAX */
91 long tid;
92 thr_self(&tid);
93 return (int)tid;
94 #elif defined(__NetBSD__)
95 return _lwp_self();
96 #else
97 return getpid();
98 #endif
101 int qemu_daemon(int nochdir, int noclose)
103 return daemon(nochdir, noclose);
106 bool qemu_write_pidfile(const char *path, Error **errp)
108 int fd;
109 char pidstr[32];
111 while (1) {
112 struct stat a, b;
113 struct flock lock = {
114 .l_type = F_WRLCK,
115 .l_whence = SEEK_SET,
116 .l_len = 0,
119 fd = qemu_open(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
120 if (fd == -1) {
121 error_setg_errno(errp, errno, "Cannot open pid file");
122 return false;
125 if (fstat(fd, &b) < 0) {
126 error_setg_errno(errp, errno, "Cannot stat file");
127 goto fail_close;
130 if (fcntl(fd, F_SETLK, &lock)) {
131 error_setg_errno(errp, errno, "Cannot lock pid file");
132 goto fail_close;
136 * Now make sure the path we locked is the same one that now
137 * exists on the filesystem.
139 if (stat(path, &a) < 0) {
141 * PID file disappeared, someone else must be racing with
142 * us, so try again.
144 close(fd);
145 continue;
148 if (a.st_ino == b.st_ino) {
149 break;
153 * PID file was recreated, someone else must be racing with
154 * us, so try again.
156 close(fd);
159 if (ftruncate(fd, 0) < 0) {
160 error_setg_errno(errp, errno, "Failed to truncate pid file");
161 goto fail_unlink;
164 snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
165 if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
166 error_setg(errp, "Failed to write pid file");
167 goto fail_unlink;
170 return true;
172 fail_unlink:
173 unlink(path);
174 fail_close:
175 close(fd);
176 return false;
179 void *qemu_oom_check(void *ptr)
181 if (ptr == NULL) {
182 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
183 abort();
185 return ptr;
188 void *qemu_try_memalign(size_t alignment, size_t size)
190 void *ptr;
192 if (alignment < sizeof(void*)) {
193 alignment = sizeof(void*);
196 #if defined(CONFIG_POSIX_MEMALIGN)
197 int ret;
198 ret = posix_memalign(&ptr, alignment, size);
199 if (ret != 0) {
200 errno = ret;
201 ptr = NULL;
203 #elif defined(CONFIG_BSD)
204 ptr = valloc(size);
205 #else
206 ptr = memalign(alignment, size);
207 #endif
208 trace_qemu_memalign(alignment, size, ptr);
209 return ptr;
212 void *qemu_memalign(size_t alignment, size_t size)
214 return qemu_oom_check(qemu_try_memalign(alignment, size));
217 /* alloc shared memory pages */
218 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
220 size_t align = QEMU_VMALLOC_ALIGN;
221 void *ptr = qemu_ram_mmap(-1, size, align, shared, false);
223 if (ptr == MAP_FAILED) {
224 return NULL;
227 if (alignment) {
228 *alignment = align;
231 trace_qemu_anon_ram_alloc(size, ptr);
232 return ptr;
235 void qemu_vfree(void *ptr)
237 trace_qemu_vfree(ptr);
238 free(ptr);
241 void qemu_anon_ram_free(void *ptr, size_t size)
243 trace_qemu_anon_ram_free(ptr, size);
244 qemu_ram_munmap(-1, ptr, size);
247 void qemu_set_block(int fd)
249 int f;
250 f = fcntl(fd, F_GETFL);
251 assert(f != -1);
252 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
253 assert(f != -1);
256 void qemu_set_nonblock(int fd)
258 int f;
259 f = fcntl(fd, F_GETFL);
260 assert(f != -1);
261 f = fcntl(fd, F_SETFL, f | O_NONBLOCK);
262 #ifdef __OpenBSD__
263 if (f == -1) {
265 * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on
266 * memory devices and sets errno to ENODEV.
267 * It's OK if we fail to set O_NONBLOCK on devices like /dev/null,
268 * because they will never block anyway.
270 assert(errno == ENODEV);
272 #else
273 assert(f != -1);
274 #endif
277 int socket_set_fast_reuse(int fd)
279 int val = 1, ret;
281 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
282 (const char *)&val, sizeof(val));
284 assert(ret == 0);
286 return ret;
289 void qemu_set_cloexec(int fd)
291 int f;
292 f = fcntl(fd, F_GETFD);
293 assert(f != -1);
294 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
295 assert(f != -1);
299 * Creates a pipe with FD_CLOEXEC set on both file descriptors
301 int qemu_pipe(int pipefd[2])
303 int ret;
305 #ifdef CONFIG_PIPE2
306 ret = pipe2(pipefd, O_CLOEXEC);
307 if (ret != -1 || errno != ENOSYS) {
308 return ret;
310 #endif
311 ret = pipe(pipefd);
312 if (ret == 0) {
313 qemu_set_cloexec(pipefd[0]);
314 qemu_set_cloexec(pipefd[1]);
317 return ret;
320 char *
321 qemu_get_local_state_pathname(const char *relative_pathname)
323 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
324 relative_pathname);
327 void qemu_set_tty_echo(int fd, bool echo)
329 struct termios tty;
331 tcgetattr(fd, &tty);
333 if (echo) {
334 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
335 } else {
336 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
339 tcsetattr(fd, TCSANOW, &tty);
342 static char exec_dir[PATH_MAX];
344 void qemu_init_exec_dir(const char *argv0)
346 char *dir;
347 char *p = NULL;
348 char buf[PATH_MAX];
350 assert(!exec_dir[0]);
352 #if defined(__linux__)
354 int len;
355 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
356 if (len > 0) {
357 buf[len] = 0;
358 p = buf;
361 #elif defined(__FreeBSD__) \
362 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
364 #if defined(__FreeBSD__)
365 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
366 #else
367 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
368 #endif
369 size_t len = sizeof(buf) - 1;
371 *buf = '\0';
372 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
373 *buf) {
374 buf[sizeof(buf) - 1] = '\0';
375 p = buf;
378 #endif
379 /* If we don't have any way of figuring out the actual executable
380 location then try argv[0]. */
381 if (!p) {
382 if (!argv0) {
383 return;
385 p = realpath(argv0, buf);
386 if (!p) {
387 return;
390 dir = g_path_get_dirname(p);
392 pstrcpy(exec_dir, sizeof(exec_dir), dir);
394 g_free(dir);
397 char *qemu_get_exec_dir(void)
399 return g_strdup(exec_dir);
402 static void sigbus_handler(int signal)
404 int i;
405 if (memset_thread) {
406 for (i = 0; i < memset_num_threads; i++) {
407 if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
408 siglongjmp(memset_thread[i].env, 1);
414 static void *do_touch_pages(void *arg)
416 MemsetThread *memset_args = (MemsetThread *)arg;
417 sigset_t set, oldset;
420 * On Linux, the page faults from the loop below can cause mmap_sem
421 * contention with allocation of the thread stacks. Do not start
422 * clearing until all threads have been created.
424 qemu_mutex_lock(&page_mutex);
425 while(!threads_created_flag){
426 qemu_cond_wait(&page_cond, &page_mutex);
428 qemu_mutex_unlock(&page_mutex);
430 /* unblock SIGBUS */
431 sigemptyset(&set);
432 sigaddset(&set, SIGBUS);
433 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
435 if (sigsetjmp(memset_args->env, 1)) {
436 memset_thread_failed = true;
437 } else {
438 char *addr = memset_args->addr;
439 size_t numpages = memset_args->numpages;
440 size_t hpagesize = memset_args->hpagesize;
441 size_t i;
442 for (i = 0; i < numpages; i++) {
444 * Read & write back the same value, so we don't
445 * corrupt existing user/app data that might be
446 * stored.
448 * 'volatile' to stop compiler optimizing this away
449 * to a no-op
451 * TODO: get a better solution from kernel so we
452 * don't need to write at all so we don't cause
453 * wear on the storage backing the region...
455 *(volatile char *)addr = *addr;
456 addr += hpagesize;
459 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
460 return NULL;
463 static inline int get_memset_num_threads(int smp_cpus)
465 long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
466 int ret = 1;
468 if (host_procs > 0) {
469 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
471 /* In case sysconf() fails, we fall back to single threaded */
472 return ret;
475 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
476 int smp_cpus)
478 static gsize initialized = 0;
479 size_t numpages_per_thread, leftover;
480 char *addr = area;
481 int i = 0;
483 if (g_once_init_enter(&initialized)) {
484 qemu_mutex_init(&page_mutex);
485 qemu_cond_init(&page_cond);
486 g_once_init_leave(&initialized, 1);
489 memset_thread_failed = false;
490 threads_created_flag = false;
491 memset_num_threads = get_memset_num_threads(smp_cpus);
492 memset_thread = g_new0(MemsetThread, memset_num_threads);
493 numpages_per_thread = numpages / memset_num_threads;
494 leftover = numpages % memset_num_threads;
495 for (i = 0; i < memset_num_threads; i++) {
496 memset_thread[i].addr = addr;
497 memset_thread[i].numpages = numpages_per_thread + (i < leftover);
498 memset_thread[i].hpagesize = hpagesize;
499 qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
500 do_touch_pages, &memset_thread[i],
501 QEMU_THREAD_JOINABLE);
502 addr += memset_thread[i].numpages * hpagesize;
505 qemu_mutex_lock(&page_mutex);
506 threads_created_flag = true;
507 qemu_cond_broadcast(&page_cond);
508 qemu_mutex_unlock(&page_mutex);
510 for (i = 0; i < memset_num_threads; i++) {
511 qemu_thread_join(&memset_thread[i].pgthread);
513 g_free(memset_thread);
514 memset_thread = NULL;
516 return memset_thread_failed;
519 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
520 Error **errp)
522 int ret;
523 struct sigaction act, oldact;
524 size_t hpagesize = qemu_fd_getpagesize(fd);
525 size_t numpages = DIV_ROUND_UP(memory, hpagesize);
527 memset(&act, 0, sizeof(act));
528 act.sa_handler = &sigbus_handler;
529 act.sa_flags = 0;
531 ret = sigaction(SIGBUS, &act, &oldact);
532 if (ret) {
533 error_setg_errno(errp, errno,
534 "os_mem_prealloc: failed to install signal handler");
535 return;
538 /* touch pages simultaneously */
539 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
540 error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
541 "pages available to allocate guest RAM");
544 ret = sigaction(SIGBUS, &oldact, NULL);
545 if (ret) {
546 /* Terminate QEMU since it can't recover from error */
547 perror("os_mem_prealloc: failed to reinstall signal handler");
548 exit(1);
552 char *qemu_get_pid_name(pid_t pid)
554 char *name = NULL;
556 #if defined(__FreeBSD__)
557 /* BSDs don't have /proc, but they provide a nice substitute */
558 struct kinfo_proc *proc = kinfo_getproc(pid);
560 if (proc) {
561 name = g_strdup(proc->ki_comm);
562 free(proc);
564 #else
565 /* Assume a system with reasonable procfs */
566 char *pid_path;
567 size_t len;
569 pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
570 g_file_get_contents(pid_path, &name, &len, NULL);
571 g_free(pid_path);
572 #endif
574 return name;
578 pid_t qemu_fork(Error **errp)
580 sigset_t oldmask, newmask;
581 struct sigaction sig_action;
582 int saved_errno;
583 pid_t pid;
586 * Need to block signals now, so that child process can safely
587 * kill off caller's signal handlers without a race.
589 sigfillset(&newmask);
590 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
591 error_setg_errno(errp, errno,
592 "cannot block signals");
593 return -1;
596 pid = fork();
597 saved_errno = errno;
599 if (pid < 0) {
600 /* attempt to restore signal mask, but ignore failure, to
601 * avoid obscuring the fork failure */
602 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
603 error_setg_errno(errp, saved_errno,
604 "cannot fork child process");
605 errno = saved_errno;
606 return -1;
607 } else if (pid) {
608 /* parent process */
610 /* Restore our original signal mask now that the child is
611 * safely running. Only documented failures are EFAULT (not
612 * possible, since we are using just-grabbed mask) or EINVAL
613 * (not possible, since we are using correct arguments). */
614 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
615 } else {
616 /* child process */
617 size_t i;
619 /* Clear out all signal handlers from parent so nothing
620 * unexpected can happen in our child once we unblock
621 * signals */
622 sig_action.sa_handler = SIG_DFL;
623 sig_action.sa_flags = 0;
624 sigemptyset(&sig_action.sa_mask);
626 for (i = 1; i < NSIG; i++) {
627 /* Only possible errors are EFAULT or EINVAL The former
628 * won't happen, the latter we expect, so no need to check
629 * return value */
630 (void)sigaction(i, &sig_action, NULL);
633 /* Unmask all signals in child, since we've no idea what the
634 * caller's done with their signal mask and don't want to
635 * propagate that to children */
636 sigemptyset(&newmask);
637 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
638 Error *local_err = NULL;
639 error_setg_errno(&local_err, errno,
640 "cannot unblock signals");
641 error_report_err(local_err);
642 _exit(1);
645 return pid;
648 void *qemu_alloc_stack(size_t *sz)
650 void *ptr, *guardpage;
651 int flags;
652 #ifdef CONFIG_DEBUG_STACK_USAGE
653 void *ptr2;
654 #endif
655 size_t pagesz = qemu_real_host_page_size;
656 #ifdef _SC_THREAD_STACK_MIN
657 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
658 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
659 *sz = MAX(MAX(min_stack_sz, 0), *sz);
660 #endif
661 /* adjust stack size to a multiple of the page size */
662 *sz = ROUND_UP(*sz, pagesz);
663 /* allocate one extra page for the guard page */
664 *sz += pagesz;
666 flags = MAP_PRIVATE | MAP_ANONYMOUS;
667 #if defined(MAP_STACK) && defined(__OpenBSD__)
668 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
669 * Linux/FreeBSD/NetBSD have a flag with the same name
670 * but have differing functionality. OpenBSD will SEGV
671 * if it spots execution with a stack pointer pointing
672 * at memory that was not allocated with MAP_STACK.
674 flags |= MAP_STACK;
675 #endif
677 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
678 if (ptr == MAP_FAILED) {
679 perror("failed to allocate memory for stack");
680 abort();
683 #if defined(HOST_IA64)
684 /* separate register stack */
685 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
686 #elif defined(HOST_HPPA)
687 /* stack grows up */
688 guardpage = ptr + *sz - pagesz;
689 #else
690 /* stack grows down */
691 guardpage = ptr;
692 #endif
693 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
694 perror("failed to set up stack guard page");
695 abort();
698 #ifdef CONFIG_DEBUG_STACK_USAGE
699 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
700 *(uint32_t *)ptr2 = 0xdeadbeaf;
702 #endif
704 return ptr;
707 #ifdef CONFIG_DEBUG_STACK_USAGE
708 static __thread unsigned int max_stack_usage;
709 #endif
711 void qemu_free_stack(void *stack, size_t sz)
713 #ifdef CONFIG_DEBUG_STACK_USAGE
714 unsigned int usage;
715 void *ptr;
717 for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
718 ptr += sizeof(uint32_t)) {
719 if (*(uint32_t *)ptr != 0xdeadbeaf) {
720 break;
723 usage = sz - (uintptr_t) (ptr - stack);
724 if (usage > max_stack_usage) {
725 error_report("thread %d max stack usage increased from %u to %u",
726 qemu_get_thread_id(), max_stack_usage, usage);
727 max_stack_usage = usage;
729 #endif
731 munmap(stack, sz);
734 void sigaction_invoke(struct sigaction *action,
735 struct qemu_signalfd_siginfo *info)
737 siginfo_t si = {};
738 si.si_signo = info->ssi_signo;
739 si.si_errno = info->ssi_errno;
740 si.si_code = info->ssi_code;
742 /* Convert the minimal set of fields defined by POSIX.
743 * Positive si_code values are reserved for kernel-generated
744 * signals, where the valid siginfo fields are determined by
745 * the signal number. But according to POSIX, it is unspecified
746 * whether SI_USER and SI_QUEUE have values less than or equal to
747 * zero.
749 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
750 info->ssi_code <= 0) {
751 /* SIGTERM, etc. */
752 si.si_pid = info->ssi_pid;
753 si.si_uid = info->ssi_uid;
754 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
755 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
756 si.si_addr = (void *)(uintptr_t)info->ssi_addr;
757 } else if (info->ssi_signo == SIGCHLD) {
758 si.si_pid = info->ssi_pid;
759 si.si_status = info->ssi_status;
760 si.si_uid = info->ssi_uid;
762 action->sa_sigaction(info->ssi_signo, &si, NULL);