mem-prealloc: reduce large guest start-up and migration time.
[qemu/ar7.git] / util / oslib-posix.c
blob956f66ab4a5caf6a7a59586283d341e48a964147
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 "sysemu/sysemu.h"
35 #include "trace.h"
36 #include "qapi/error.h"
37 #include "qemu/sockets.h"
38 #include <libgen.h>
39 #include <sys/signal.h>
40 #include "qemu/cutils.h"
42 #ifdef CONFIG_LINUX
43 #include <sys/syscall.h>
44 #endif
46 #ifdef __FreeBSD__
47 #include <sys/sysctl.h>
48 #include <sys/user.h>
49 #include <libutil.h>
50 #endif
52 #include "qemu/mmap-alloc.h"
54 #ifdef CONFIG_DEBUG_STACK_USAGE
55 #include "qemu/error-report.h"
56 #endif
58 #define MAX_MEM_PREALLOC_THREAD_COUNT (MIN(sysconf(_SC_NPROCESSORS_ONLN), 16))
60 struct MemsetThread {
61 char *addr;
62 uint64_t numpages;
63 uint64_t hpagesize;
64 QemuThread pgthread;
65 sigjmp_buf env;
67 typedef struct MemsetThread MemsetThread;
69 static MemsetThread *memset_thread;
70 static int memset_num_threads;
71 static bool memset_thread_failed;
73 int qemu_get_thread_id(void)
75 #if defined(__linux__)
76 return syscall(SYS_gettid);
77 #else
78 return getpid();
79 #endif
82 int qemu_daemon(int nochdir, int noclose)
84 return daemon(nochdir, noclose);
87 void *qemu_oom_check(void *ptr)
89 if (ptr == NULL) {
90 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
91 abort();
93 return ptr;
96 void *qemu_try_memalign(size_t alignment, size_t size)
98 void *ptr;
100 if (alignment < sizeof(void*)) {
101 alignment = sizeof(void*);
104 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
105 int ret;
106 ret = posix_memalign(&ptr, alignment, size);
107 if (ret != 0) {
108 errno = ret;
109 ptr = NULL;
111 #elif defined(CONFIG_BSD)
112 ptr = valloc(size);
113 #else
114 ptr = memalign(alignment, size);
115 #endif
116 trace_qemu_memalign(alignment, size, ptr);
117 return ptr;
120 void *qemu_memalign(size_t alignment, size_t size)
122 return qemu_oom_check(qemu_try_memalign(alignment, size));
125 /* alloc shared memory pages */
126 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
128 size_t align = QEMU_VMALLOC_ALIGN;
129 void *ptr = qemu_ram_mmap(-1, size, align, false);
131 if (ptr == MAP_FAILED) {
132 return NULL;
135 if (alignment) {
136 *alignment = align;
139 trace_qemu_anon_ram_alloc(size, ptr);
140 return ptr;
143 void qemu_vfree(void *ptr)
145 trace_qemu_vfree(ptr);
146 free(ptr);
149 void qemu_anon_ram_free(void *ptr, size_t size)
151 trace_qemu_anon_ram_free(ptr, size);
152 qemu_ram_munmap(ptr, size);
155 void qemu_set_block(int fd)
157 int f;
158 f = fcntl(fd, F_GETFL);
159 fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
162 void qemu_set_nonblock(int fd)
164 int f;
165 f = fcntl(fd, F_GETFL);
166 fcntl(fd, F_SETFL, f | O_NONBLOCK);
169 int socket_set_fast_reuse(int fd)
171 int val = 1, ret;
173 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
174 (const char *)&val, sizeof(val));
176 assert(ret == 0);
178 return ret;
181 void qemu_set_cloexec(int fd)
183 int f;
184 f = fcntl(fd, F_GETFD);
185 fcntl(fd, F_SETFD, f | FD_CLOEXEC);
189 * Creates a pipe with FD_CLOEXEC set on both file descriptors
191 int qemu_pipe(int pipefd[2])
193 int ret;
195 #ifdef CONFIG_PIPE2
196 ret = pipe2(pipefd, O_CLOEXEC);
197 if (ret != -1 || errno != ENOSYS) {
198 return ret;
200 #endif
201 ret = pipe(pipefd);
202 if (ret == 0) {
203 qemu_set_cloexec(pipefd[0]);
204 qemu_set_cloexec(pipefd[1]);
207 return ret;
210 int qemu_utimens(const char *path, const struct timespec *times)
212 struct timeval tv[2], tv_now;
213 struct stat st;
214 int i;
215 #ifdef CONFIG_UTIMENSAT
216 int ret;
218 ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
219 if (ret != -1 || errno != ENOSYS) {
220 return ret;
222 #endif
223 /* Fallback: use utimes() instead of utimensat() */
225 /* happy if special cases */
226 if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
227 return 0;
229 if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
230 return utimes(path, NULL);
233 /* prepare for hard cases */
234 if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
235 gettimeofday(&tv_now, NULL);
237 if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
238 stat(path, &st);
241 for (i = 0; i < 2; i++) {
242 if (times[i].tv_nsec == UTIME_NOW) {
243 tv[i].tv_sec = tv_now.tv_sec;
244 tv[i].tv_usec = tv_now.tv_usec;
245 } else if (times[i].tv_nsec == UTIME_OMIT) {
246 tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
247 tv[i].tv_usec = 0;
248 } else {
249 tv[i].tv_sec = times[i].tv_sec;
250 tv[i].tv_usec = times[i].tv_nsec / 1000;
254 return utimes(path, &tv[0]);
257 char *
258 qemu_get_local_state_pathname(const char *relative_pathname)
260 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
261 relative_pathname);
264 void qemu_set_tty_echo(int fd, bool echo)
266 struct termios tty;
268 tcgetattr(fd, &tty);
270 if (echo) {
271 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
272 } else {
273 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
276 tcsetattr(fd, TCSANOW, &tty);
279 static char exec_dir[PATH_MAX];
281 void qemu_init_exec_dir(const char *argv0)
283 char *dir;
284 char *p = NULL;
285 char buf[PATH_MAX];
287 assert(!exec_dir[0]);
289 #if defined(__linux__)
291 int len;
292 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
293 if (len > 0) {
294 buf[len] = 0;
295 p = buf;
298 #elif defined(__FreeBSD__)
300 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
301 size_t len = sizeof(buf) - 1;
303 *buf = '\0';
304 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
305 *buf) {
306 buf[sizeof(buf) - 1] = '\0';
307 p = buf;
310 #endif
311 /* If we don't have any way of figuring out the actual executable
312 location then try argv[0]. */
313 if (!p) {
314 if (!argv0) {
315 return;
317 p = realpath(argv0, buf);
318 if (!p) {
319 return;
322 dir = g_path_get_dirname(p);
324 pstrcpy(exec_dir, sizeof(exec_dir), dir);
326 g_free(dir);
329 char *qemu_get_exec_dir(void)
331 return g_strdup(exec_dir);
334 static void sigbus_handler(int signal)
336 int i;
337 if (memset_thread) {
338 for (i = 0; i < memset_num_threads; i++) {
339 if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
340 siglongjmp(memset_thread[i].env, 1);
346 static void *do_touch_pages(void *arg)
348 MemsetThread *memset_args = (MemsetThread *)arg;
349 char *addr = memset_args->addr;
350 uint64_t numpages = memset_args->numpages;
351 uint64_t hpagesize = memset_args->hpagesize;
352 sigset_t set, oldset;
353 int i = 0;
355 /* unblock SIGBUS */
356 sigemptyset(&set);
357 sigaddset(&set, SIGBUS);
358 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
360 if (sigsetjmp(memset_args->env, 1)) {
361 memset_thread_failed = true;
362 } else {
363 for (i = 0; i < numpages; i++) {
364 memset(addr, 0, 1);
365 addr += hpagesize;
368 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
369 return NULL;
372 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
373 int smp_cpus)
375 uint64_t numpages_per_thread, size_per_thread;
376 char *addr = area;
377 int i = 0;
379 memset_thread_failed = false;
380 memset_num_threads = MIN(smp_cpus, MAX_MEM_PREALLOC_THREAD_COUNT);
381 memset_thread = g_new0(MemsetThread, memset_num_threads);
382 numpages_per_thread = (numpages / memset_num_threads);
383 size_per_thread = (hpagesize * numpages_per_thread);
384 for (i = 0; i < memset_num_threads; i++) {
385 memset_thread[i].addr = addr;
386 memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
387 numpages : numpages_per_thread;
388 memset_thread[i].hpagesize = hpagesize;
389 qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
390 do_touch_pages, &memset_thread[i],
391 QEMU_THREAD_JOINABLE);
392 addr += size_per_thread;
393 numpages -= numpages_per_thread;
395 for (i = 0; i < memset_num_threads; i++) {
396 qemu_thread_join(&memset_thread[i].pgthread);
398 g_free(memset_thread);
399 memset_thread = NULL;
401 return memset_thread_failed;
404 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
405 Error **errp)
407 int ret;
408 struct sigaction act, oldact;
409 size_t hpagesize = qemu_fd_getpagesize(fd);
410 size_t numpages = DIV_ROUND_UP(memory, hpagesize);
412 memset(&act, 0, sizeof(act));
413 act.sa_handler = &sigbus_handler;
414 act.sa_flags = 0;
416 ret = sigaction(SIGBUS, &act, &oldact);
417 if (ret) {
418 error_setg_errno(errp, errno,
419 "os_mem_prealloc: failed to install signal handler");
420 return;
423 /* touch pages simultaneously */
424 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
425 error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
426 "pages available to allocate guest RAM\n");
429 ret = sigaction(SIGBUS, &oldact, NULL);
430 if (ret) {
431 /* Terminate QEMU since it can't recover from error */
432 perror("os_mem_prealloc: failed to reinstall signal handler");
433 exit(1);
438 static struct termios oldtty;
440 static void term_exit(void)
442 tcsetattr(0, TCSANOW, &oldtty);
445 static void term_init(void)
447 struct termios tty;
449 tcgetattr(0, &tty);
450 oldtty = tty;
452 tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
453 |INLCR|IGNCR|ICRNL|IXON);
454 tty.c_oflag |= OPOST;
455 tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
456 tty.c_cflag &= ~(CSIZE|PARENB);
457 tty.c_cflag |= CS8;
458 tty.c_cc[VMIN] = 1;
459 tty.c_cc[VTIME] = 0;
461 tcsetattr(0, TCSANOW, &tty);
463 atexit(term_exit);
466 int qemu_read_password(char *buf, int buf_size)
468 uint8_t ch;
469 int i, ret;
471 printf("password: ");
472 fflush(stdout);
473 term_init();
474 i = 0;
475 for (;;) {
476 ret = read(0, &ch, 1);
477 if (ret == -1) {
478 if (errno == EAGAIN || errno == EINTR) {
479 continue;
480 } else {
481 break;
483 } else if (ret == 0) {
484 ret = -1;
485 break;
486 } else {
487 if (ch == '\r' ||
488 ch == '\n') {
489 ret = 0;
490 break;
492 if (i < (buf_size - 1)) {
493 buf[i++] = ch;
497 term_exit();
498 buf[i] = '\0';
499 printf("\n");
500 return ret;
504 char *qemu_get_pid_name(pid_t pid)
506 char *name = NULL;
508 #if defined(__FreeBSD__)
509 /* BSDs don't have /proc, but they provide a nice substitute */
510 struct kinfo_proc *proc = kinfo_getproc(pid);
512 if (proc) {
513 name = g_strdup(proc->ki_comm);
514 free(proc);
516 #else
517 /* Assume a system with reasonable procfs */
518 char *pid_path;
519 size_t len;
521 pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
522 g_file_get_contents(pid_path, &name, &len, NULL);
523 g_free(pid_path);
524 #endif
526 return name;
530 pid_t qemu_fork(Error **errp)
532 sigset_t oldmask, newmask;
533 struct sigaction sig_action;
534 int saved_errno;
535 pid_t pid;
538 * Need to block signals now, so that child process can safely
539 * kill off caller's signal handlers without a race.
541 sigfillset(&newmask);
542 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
543 error_setg_errno(errp, errno,
544 "cannot block signals");
545 return -1;
548 pid = fork();
549 saved_errno = errno;
551 if (pid < 0) {
552 /* attempt to restore signal mask, but ignore failure, to
553 * avoid obscuring the fork failure */
554 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
555 error_setg_errno(errp, saved_errno,
556 "cannot fork child process");
557 errno = saved_errno;
558 return -1;
559 } else if (pid) {
560 /* parent process */
562 /* Restore our original signal mask now that the child is
563 * safely running. Only documented failures are EFAULT (not
564 * possible, since we are using just-grabbed mask) or EINVAL
565 * (not possible, since we are using correct arguments). */
566 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
567 } else {
568 /* child process */
569 size_t i;
571 /* Clear out all signal handlers from parent so nothing
572 * unexpected can happen in our child once we unblock
573 * signals */
574 sig_action.sa_handler = SIG_DFL;
575 sig_action.sa_flags = 0;
576 sigemptyset(&sig_action.sa_mask);
578 for (i = 1; i < NSIG; i++) {
579 /* Only possible errors are EFAULT or EINVAL The former
580 * won't happen, the latter we expect, so no need to check
581 * return value */
582 (void)sigaction(i, &sig_action, NULL);
585 /* Unmask all signals in child, since we've no idea what the
586 * caller's done with their signal mask and don't want to
587 * propagate that to children */
588 sigemptyset(&newmask);
589 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
590 Error *local_err = NULL;
591 error_setg_errno(&local_err, errno,
592 "cannot unblock signals");
593 error_report_err(local_err);
594 _exit(1);
597 return pid;
600 void *qemu_alloc_stack(size_t *sz)
602 void *ptr, *guardpage;
603 #ifdef CONFIG_DEBUG_STACK_USAGE
604 void *ptr2;
605 #endif
606 size_t pagesz = getpagesize();
607 #ifdef _SC_THREAD_STACK_MIN
608 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
609 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
610 *sz = MAX(MAX(min_stack_sz, 0), *sz);
611 #endif
612 /* adjust stack size to a multiple of the page size */
613 *sz = ROUND_UP(*sz, pagesz);
614 /* allocate one extra page for the guard page */
615 *sz += pagesz;
617 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
618 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
619 if (ptr == MAP_FAILED) {
620 abort();
623 #if defined(HOST_IA64)
624 /* separate register stack */
625 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
626 #elif defined(HOST_HPPA)
627 /* stack grows up */
628 guardpage = ptr + *sz - pagesz;
629 #else
630 /* stack grows down */
631 guardpage = ptr;
632 #endif
633 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
634 abort();
637 #ifdef CONFIG_DEBUG_STACK_USAGE
638 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
639 *(uint32_t *)ptr2 = 0xdeadbeaf;
641 #endif
643 return ptr;
646 #ifdef CONFIG_DEBUG_STACK_USAGE
647 static __thread unsigned int max_stack_usage;
648 #endif
650 void qemu_free_stack(void *stack, size_t sz)
652 #ifdef CONFIG_DEBUG_STACK_USAGE
653 unsigned int usage;
654 void *ptr;
656 for (ptr = stack + getpagesize(); ptr < stack + sz;
657 ptr += sizeof(uint32_t)) {
658 if (*(uint32_t *)ptr != 0xdeadbeaf) {
659 break;
662 usage = sz - (uintptr_t) (ptr - stack);
663 if (usage > max_stack_usage) {
664 error_report("thread %d max stack usage increased from %u to %u",
665 qemu_get_thread_id(), max_stack_usage, usage);
666 max_stack_usage = usage;
668 #endif
670 munmap(stack, sz);
673 void sigaction_invoke(struct sigaction *action,
674 struct qemu_signalfd_siginfo *info)
676 siginfo_t si = { 0 };
677 si.si_signo = info->ssi_signo;
678 si.si_errno = info->ssi_errno;
679 si.si_code = info->ssi_code;
681 /* Convert the minimal set of fields defined by POSIX.
682 * Positive si_code values are reserved for kernel-generated
683 * signals, where the valid siginfo fields are determined by
684 * the signal number. But according to POSIX, it is unspecified
685 * whether SI_USER and SI_QUEUE have values less than or equal to
686 * zero.
688 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
689 info->ssi_code <= 0) {
690 /* SIGTERM, etc. */
691 si.si_pid = info->ssi_pid;
692 si.si_uid = info->ssi_uid;
693 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
694 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
695 si.si_addr = (void *)(uintptr_t)info->ssi_addr;
696 } else if (info->ssi_signo == SIGCHLD) {
697 si.si_pid = info->ssi_pid;
698 si.si_status = info->ssi_status;
699 si.si_uid = info->ssi_uid;
700 } else if (info->ssi_signo == SIGIO) {
701 si.si_band = info->ssi_band;
703 action->sa_sigaction(info->ssi_signo, &si, NULL);