Tue, 12 May 2009 15:55:56 -0700
6840196: NUMA allocator: crash in fastdebug during startup on Linux
Summary: With libnuma >1.2 explicity use 1.1 symbols
Reviewed-by: ysr
1 /*
2 * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 // do not include precompiled header file
26 # include "incls/_os_linux.cpp.incl"
28 // put OS-includes here
29 # include <sys/types.h>
30 # include <sys/mman.h>
31 # include <pthread.h>
32 # include <signal.h>
33 # include <errno.h>
34 # include <dlfcn.h>
35 # include <stdio.h>
36 # include <unistd.h>
37 # include <sys/resource.h>
38 # include <pthread.h>
39 # include <sys/stat.h>
40 # include <sys/time.h>
41 # include <sys/times.h>
42 # include <sys/utsname.h>
43 # include <sys/socket.h>
44 # include <sys/wait.h>
45 # include <pwd.h>
46 # include <poll.h>
47 # include <semaphore.h>
48 # include <fcntl.h>
49 # include <string.h>
50 # include <syscall.h>
51 # include <sys/sysinfo.h>
52 # include <gnu/libc-version.h>
53 # include <sys/ipc.h>
54 # include <sys/shm.h>
55 # include <link.h>
57 #define MAX_PATH (2 * K)
59 // for timer info max values which include all bits
60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
61 #define SEC_IN_NANOSECS 1000000000LL
63 ////////////////////////////////////////////////////////////////////////////////
64 // global variables
65 julong os::Linux::_physical_memory = 0;
67 address os::Linux::_initial_thread_stack_bottom = NULL;
68 uintptr_t os::Linux::_initial_thread_stack_size = 0;
70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
72 Mutex* os::Linux::_createThread_lock = NULL;
73 pthread_t os::Linux::_main_thread;
74 int os::Linux::_page_size = -1;
75 bool os::Linux::_is_floating_stack = false;
76 bool os::Linux::_is_NPTL = false;
77 bool os::Linux::_supports_fast_thread_cpu_time = false;
78 const char * os::Linux::_glibc_version = NULL;
79 const char * os::Linux::_libpthread_version = NULL;
81 static jlong initial_time_count=0;
83 static int clock_tics_per_sec = 100;
85 // For diagnostics to print a message once. see run_periodic_checks
86 static sigset_t check_signal_done;
87 static bool check_signals = true;;
89 static pid_t _initial_pid = 0;
91 /* Signal number used to suspend/resume a thread */
93 /* do not use any signal number less than SIGSEGV, see 4355769 */
94 static int SR_signum = SIGUSR2;
95 sigset_t SR_sigset;
97 /* Used to protect dlsym() calls */
98 static pthread_mutex_t dl_mutex;
100 ////////////////////////////////////////////////////////////////////////////////
101 // utility functions
103 static int SR_initialize();
104 static int SR_finalize();
106 julong os::available_memory() {
107 return Linux::available_memory();
108 }
110 julong os::Linux::available_memory() {
111 // values in struct sysinfo are "unsigned long"
112 struct sysinfo si;
113 sysinfo(&si);
115 return (julong)si.freeram * si.mem_unit;
116 }
118 julong os::physical_memory() {
119 return Linux::physical_memory();
120 }
122 julong os::allocatable_physical_memory(julong size) {
123 #ifdef _LP64
124 return size;
125 #else
126 julong result = MIN2(size, (julong)3800*M);
127 if (!is_allocatable(result)) {
128 // See comments under solaris for alignment considerations
129 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
130 result = MIN2(size, reasonable_size);
131 }
132 return result;
133 #endif // _LP64
134 }
136 ////////////////////////////////////////////////////////////////////////////////
137 // environment support
139 bool os::getenv(const char* name, char* buf, int len) {
140 const char* val = ::getenv(name);
141 if (val != NULL && strlen(val) < (size_t)len) {
142 strcpy(buf, val);
143 return true;
144 }
145 if (len > 0) buf[0] = 0; // return a null string
146 return false;
147 }
150 // Return true if user is running as root.
152 bool os::have_special_privileges() {
153 static bool init = false;
154 static bool privileges = false;
155 if (!init) {
156 privileges = (getuid() != geteuid()) || (getgid() != getegid());
157 init = true;
158 }
159 return privileges;
160 }
163 #ifndef SYS_gettid
164 // i386: 224, ia64: 1105, amd64: 186, sparc 143
165 #ifdef __ia64__
166 #define SYS_gettid 1105
167 #elif __i386__
168 #define SYS_gettid 224
169 #elif __amd64__
170 #define SYS_gettid 186
171 #elif __sparc__
172 #define SYS_gettid 143
173 #else
174 #error define gettid for the arch
175 #endif
176 #endif
178 // Cpu architecture string
179 #if defined(IA64)
180 static char cpu_arch[] = "ia64";
181 #elif defined(IA32)
182 static char cpu_arch[] = "i386";
183 #elif defined(AMD64)
184 static char cpu_arch[] = "amd64";
185 #elif defined(SPARC)
186 # ifdef _LP64
187 static char cpu_arch[] = "sparcv9";
188 # else
189 static char cpu_arch[] = "sparc";
190 # endif
191 #else
192 #error Add appropriate cpu_arch setting
193 #endif
196 // pid_t gettid()
197 //
198 // Returns the kernel thread id of the currently running thread. Kernel
199 // thread id is used to access /proc.
200 //
201 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
202 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
203 //
204 pid_t os::Linux::gettid() {
205 int rslt = syscall(SYS_gettid);
206 if (rslt == -1) {
207 // old kernel, no NPTL support
208 return getpid();
209 } else {
210 return (pid_t)rslt;
211 }
212 }
214 // Most versions of linux have a bug where the number of processors are
215 // determined by looking at the /proc file system. In a chroot environment,
216 // the system call returns 1. This causes the VM to act as if it is
217 // a single processor and elide locking (see is_MP() call).
218 static bool unsafe_chroot_detected = false;
219 static const char *unstable_chroot_error = "/proc file system not found.\n"
220 "Java may be unstable running multithreaded in a chroot "
221 "environment on Linux when /proc filesystem is not mounted.";
223 void os::Linux::initialize_system_info() {
224 _processor_count = sysconf(_SC_NPROCESSORS_CONF);
225 if (_processor_count == 1) {
226 pid_t pid = os::Linux::gettid();
227 char fname[32];
228 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
229 FILE *fp = fopen(fname, "r");
230 if (fp == NULL) {
231 unsafe_chroot_detected = true;
232 } else {
233 fclose(fp);
234 }
235 }
236 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
237 assert(_processor_count > 0, "linux error");
238 }
240 void os::init_system_properties_values() {
241 // char arch[12];
242 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
244 // The next steps are taken in the product version:
245 //
246 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
247 // This library should be located at:
248 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
249 //
250 // If "/jre/lib/" appears at the right place in the path, then we
251 // assume libjvm[_g].so is installed in a JDK and we use this path.
252 //
253 // Otherwise exit with message: "Could not create the Java virtual machine."
254 //
255 // The following extra steps are taken in the debugging version:
256 //
257 // If "/jre/lib/" does NOT appear at the right place in the path
258 // instead of exit check for $JAVA_HOME environment variable.
259 //
260 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
261 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
262 // it looks like libjvm[_g].so is installed there
263 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
264 //
265 // Otherwise exit.
266 //
267 // Important note: if the location of libjvm.so changes this
268 // code needs to be changed accordingly.
270 // The next few definitions allow the code to be verbatim:
271 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
272 #define getenv(n) ::getenv(n)
274 /*
275 * See ld(1):
276 * The linker uses the following search paths to locate required
277 * shared libraries:
278 * 1: ...
279 * ...
280 * 7: The default directories, normally /lib and /usr/lib.
281 */
282 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
283 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
284 #else
285 #define DEFAULT_LIBPATH "/lib:/usr/lib"
286 #endif
288 #define EXTENSIONS_DIR "/lib/ext"
289 #define ENDORSED_DIR "/lib/endorsed"
290 #define REG_DIR "/usr/java/packages"
292 {
293 /* sysclasspath, java_home, dll_dir */
294 {
295 char *home_path;
296 char *dll_path;
297 char *pslash;
298 char buf[MAXPATHLEN];
299 os::jvm_path(buf, sizeof(buf));
301 // Found the full path to libjvm.so.
302 // Now cut the path to <java_home>/jre if we can.
303 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
304 pslash = strrchr(buf, '/');
305 if (pslash != NULL)
306 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
307 dll_path = malloc(strlen(buf) + 1);
308 if (dll_path == NULL)
309 return;
310 strcpy(dll_path, buf);
311 Arguments::set_dll_dir(dll_path);
313 if (pslash != NULL) {
314 pslash = strrchr(buf, '/');
315 if (pslash != NULL) {
316 *pslash = '\0'; /* get rid of /<arch> */
317 pslash = strrchr(buf, '/');
318 if (pslash != NULL)
319 *pslash = '\0'; /* get rid of /lib */
320 }
321 }
323 home_path = malloc(strlen(buf) + 1);
324 if (home_path == NULL)
325 return;
326 strcpy(home_path, buf);
327 Arguments::set_java_home(home_path);
329 if (!set_boot_path('/', ':'))
330 return;
331 }
333 /*
334 * Where to look for native libraries
335 *
336 * Note: Due to a legacy implementation, most of the library path
337 * is set in the launcher. This was to accomodate linking restrictions
338 * on legacy Linux implementations (which are no longer supported).
339 * Eventually, all the library path setting will be done here.
340 *
341 * However, to prevent the proliferation of improperly built native
342 * libraries, the new path component /usr/java/packages is added here.
343 * Eventually, all the library path setting will be done here.
344 */
345 {
346 char *ld_library_path;
348 /*
349 * Construct the invariant part of ld_library_path. Note that the
350 * space for the colon and the trailing null are provided by the
351 * nulls included by the sizeof operator (so actually we allocate
352 * a byte more than necessary).
353 */
354 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
355 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
356 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
358 /*
359 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
360 * should always exist (until the legacy problem cited above is
361 * addressed).
362 */
363 char *v = getenv("LD_LIBRARY_PATH");
364 if (v != NULL) {
365 char *t = ld_library_path;
366 /* That's +1 for the colon and +1 for the trailing '\0' */
367 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
368 sprintf(ld_library_path, "%s:%s", v, t);
369 }
370 Arguments::set_library_path(ld_library_path);
371 }
373 /*
374 * Extensions directories.
375 *
376 * Note that the space for the colon and the trailing null are provided
377 * by the nulls included by the sizeof operator (so actually one byte more
378 * than necessary is allocated).
379 */
380 {
381 char *buf = malloc(strlen(Arguments::get_java_home()) +
382 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
383 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
384 Arguments::get_java_home());
385 Arguments::set_ext_dirs(buf);
386 }
388 /* Endorsed standards default directory. */
389 {
390 char * buf;
391 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
392 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
393 Arguments::set_endorsed_dirs(buf);
394 }
395 }
397 #undef malloc
398 #undef getenv
399 #undef EXTENSIONS_DIR
400 #undef ENDORSED_DIR
402 // Done
403 return;
404 }
406 ////////////////////////////////////////////////////////////////////////////////
407 // breakpoint support
409 void os::breakpoint() {
410 BREAKPOINT;
411 }
413 extern "C" void breakpoint() {
414 // use debugger to set breakpoint here
415 }
417 ////////////////////////////////////////////////////////////////////////////////
418 // signal support
420 debug_only(static bool signal_sets_initialized = false);
421 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
423 bool os::Linux::is_sig_ignored(int sig) {
424 struct sigaction oact;
425 sigaction(sig, (struct sigaction*)NULL, &oact);
426 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
427 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
428 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
429 return true;
430 else
431 return false;
432 }
434 void os::Linux::signal_sets_init() {
435 // Should also have an assertion stating we are still single-threaded.
436 assert(!signal_sets_initialized, "Already initialized");
437 // Fill in signals that are necessarily unblocked for all threads in
438 // the VM. Currently, we unblock the following signals:
439 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
440 // by -Xrs (=ReduceSignalUsage));
441 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
442 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
443 // the dispositions or masks wrt these signals.
444 // Programs embedding the VM that want to use the above signals for their
445 // own purposes must, at this time, use the "-Xrs" option to prevent
446 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
447 // (See bug 4345157, and other related bugs).
448 // In reality, though, unblocking these signals is really a nop, since
449 // these signals are not blocked by default.
450 sigemptyset(&unblocked_sigs);
451 sigemptyset(&allowdebug_blocked_sigs);
452 sigaddset(&unblocked_sigs, SIGILL);
453 sigaddset(&unblocked_sigs, SIGSEGV);
454 sigaddset(&unblocked_sigs, SIGBUS);
455 sigaddset(&unblocked_sigs, SIGFPE);
456 sigaddset(&unblocked_sigs, SR_signum);
458 if (!ReduceSignalUsage) {
459 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
460 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
461 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
462 }
463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
465 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
466 }
467 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
468 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
469 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
470 }
471 }
472 // Fill in signals that are blocked by all but the VM thread.
473 sigemptyset(&vm_sigs);
474 if (!ReduceSignalUsage)
475 sigaddset(&vm_sigs, BREAK_SIGNAL);
476 debug_only(signal_sets_initialized = true);
478 }
480 // These are signals that are unblocked while a thread is running Java.
481 // (For some reason, they get blocked by default.)
482 sigset_t* os::Linux::unblocked_signals() {
483 assert(signal_sets_initialized, "Not initialized");
484 return &unblocked_sigs;
485 }
487 // These are the signals that are blocked while a (non-VM) thread is
488 // running Java. Only the VM thread handles these signals.
489 sigset_t* os::Linux::vm_signals() {
490 assert(signal_sets_initialized, "Not initialized");
491 return &vm_sigs;
492 }
494 // These are signals that are blocked during cond_wait to allow debugger in
495 sigset_t* os::Linux::allowdebug_blocked_signals() {
496 assert(signal_sets_initialized, "Not initialized");
497 return &allowdebug_blocked_sigs;
498 }
500 void os::Linux::hotspot_sigmask(Thread* thread) {
502 //Save caller's signal mask before setting VM signal mask
503 sigset_t caller_sigmask;
504 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
506 OSThread* osthread = thread->osthread();
507 osthread->set_caller_sigmask(caller_sigmask);
509 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
511 if (!ReduceSignalUsage) {
512 if (thread->is_VM_thread()) {
513 // Only the VM thread handles BREAK_SIGNAL ...
514 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
515 } else {
516 // ... all other threads block BREAK_SIGNAL
517 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
518 }
519 }
520 }
522 //////////////////////////////////////////////////////////////////////////////
523 // detecting pthread library
525 void os::Linux::libpthread_init() {
526 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
527 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
528 // generic name for earlier versions.
529 // Define macros here so we can build HotSpot on old systems.
530 # ifndef _CS_GNU_LIBC_VERSION
531 # define _CS_GNU_LIBC_VERSION 2
532 # endif
533 # ifndef _CS_GNU_LIBPTHREAD_VERSION
534 # define _CS_GNU_LIBPTHREAD_VERSION 3
535 # endif
537 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
538 if (n > 0) {
539 char *str = (char *)malloc(n);
540 confstr(_CS_GNU_LIBC_VERSION, str, n);
541 os::Linux::set_glibc_version(str);
542 } else {
543 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
544 static char _gnu_libc_version[32];
545 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
546 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
547 os::Linux::set_glibc_version(_gnu_libc_version);
548 }
550 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
551 if (n > 0) {
552 char *str = (char *)malloc(n);
553 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
554 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
555 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
556 // is the case. LinuxThreads has a hard limit on max number of threads.
557 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
558 // On the other hand, NPTL does not have such a limit, sysconf()
559 // will return -1 and errno is not changed. Check if it is really NPTL.
560 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
561 strstr(str, "NPTL") &&
562 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
563 free(str);
564 os::Linux::set_libpthread_version("linuxthreads");
565 } else {
566 os::Linux::set_libpthread_version(str);
567 }
568 } else {
569 // glibc before 2.3.2 only has LinuxThreads.
570 os::Linux::set_libpthread_version("linuxthreads");
571 }
573 if (strstr(libpthread_version(), "NPTL")) {
574 os::Linux::set_is_NPTL();
575 } else {
576 os::Linux::set_is_LinuxThreads();
577 }
579 // LinuxThreads have two flavors: floating-stack mode, which allows variable
580 // stack size; and fixed-stack mode. NPTL is always floating-stack.
581 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
582 os::Linux::set_is_floating_stack();
583 }
584 }
586 /////////////////////////////////////////////////////////////////////////////
587 // thread stack
589 // Force Linux kernel to expand current thread stack. If "bottom" is close
590 // to the stack guard, caller should block all signals.
591 //
592 // MAP_GROWSDOWN:
593 // A special mmap() flag that is used to implement thread stacks. It tells
594 // kernel that the memory region should extend downwards when needed. This
595 // allows early versions of LinuxThreads to only mmap the first few pages
596 // when creating a new thread. Linux kernel will automatically expand thread
597 // stack as needed (on page faults).
598 //
599 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
600 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
601 // region, it's hard to tell if the fault is due to a legitimate stack
602 // access or because of reading/writing non-exist memory (e.g. buffer
603 // overrun). As a rule, if the fault happens below current stack pointer,
604 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
605 // application (see Linux kernel fault.c).
606 //
607 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
608 // stack overflow detection.
609 //
610 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
611 // not use this flag. However, the stack of initial thread is not created
612 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
613 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
614 // and then attach the thread to JVM.
615 //
616 // To get around the problem and allow stack banging on Linux, we need to
617 // manually expand thread stack after receiving the SIGSEGV.
618 //
619 // There are two ways to expand thread stack to address "bottom", we used
620 // both of them in JVM before 1.5:
621 // 1. adjust stack pointer first so that it is below "bottom", and then
622 // touch "bottom"
623 // 2. mmap() the page in question
624 //
625 // Now alternate signal stack is gone, it's harder to use 2. For instance,
626 // if current sp is already near the lower end of page 101, and we need to
627 // call mmap() to map page 100, it is possible that part of the mmap() frame
628 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
629 // That will destroy the mmap() frame and cause VM to crash.
630 //
631 // The following code works by adjusting sp first, then accessing the "bottom"
632 // page to force a page fault. Linux kernel will then automatically expand the
633 // stack mapping.
634 //
635 // _expand_stack_to() assumes its frame size is less than page size, which
636 // should always be true if the function is not inlined.
638 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
639 #define NOINLINE
640 #else
641 #define NOINLINE __attribute__ ((noinline))
642 #endif
644 static void _expand_stack_to(address bottom) NOINLINE;
646 static void _expand_stack_to(address bottom) {
647 address sp;
648 size_t size;
649 volatile char *p;
651 // Adjust bottom to point to the largest address within the same page, it
652 // gives us a one-page buffer if alloca() allocates slightly more memory.
653 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
654 bottom += os::Linux::page_size() - 1;
656 // sp might be slightly above current stack pointer; if that's the case, we
657 // will alloca() a little more space than necessary, which is OK. Don't use
658 // os::current_stack_pointer(), as its result can be slightly below current
659 // stack pointer, causing us to not alloca enough to reach "bottom".
660 sp = (address)&sp;
662 if (sp > bottom) {
663 size = sp - bottom;
664 p = (volatile char *)alloca(size);
665 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
666 p[0] = '\0';
667 }
668 }
670 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
671 assert(t!=NULL, "just checking");
672 assert(t->osthread()->expanding_stack(), "expand should be set");
673 assert(t->stack_base() != NULL, "stack_base was not initialized");
675 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
676 sigset_t mask_all, old_sigset;
677 sigfillset(&mask_all);
678 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
679 _expand_stack_to(addr);
680 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
681 return true;
682 }
683 return false;
684 }
686 //////////////////////////////////////////////////////////////////////////////
687 // create new thread
689 static address highest_vm_reserved_address();
691 // check if it's safe to start a new thread
692 static bool _thread_safety_check(Thread* thread) {
693 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
694 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
695 // Heap is mmap'ed at lower end of memory space. Thread stacks are
696 // allocated (MAP_FIXED) from high address space. Every thread stack
697 // occupies a fixed size slot (usually 2Mbytes, but user can change
698 // it to other values if they rebuild LinuxThreads).
699 //
700 // Problem with MAP_FIXED is that mmap() can still succeed even part of
701 // the memory region has already been mmap'ed. That means if we have too
702 // many threads and/or very large heap, eventually thread stack will
703 // collide with heap.
704 //
705 // Here we try to prevent heap/stack collision by comparing current
706 // stack bottom with the highest address that has been mmap'ed by JVM
707 // plus a safety margin for memory maps created by native code.
708 //
709 // This feature can be disabled by setting ThreadSafetyMargin to 0
710 //
711 if (ThreadSafetyMargin > 0) {
712 address stack_bottom = os::current_stack_base() - os::current_stack_size();
714 // not safe if our stack extends below the safety margin
715 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
716 } else {
717 return true;
718 }
719 } else {
720 // Floating stack LinuxThreads or NPTL:
721 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
722 // there's not enough space left, pthread_create() will fail. If we come
723 // here, that means enough space has been reserved for stack.
724 return true;
725 }
726 }
728 // Thread start routine for all newly created threads
729 static void *java_start(Thread *thread) {
730 // Try to randomize the cache line index of hot stack frames.
731 // This helps when threads of the same stack traces evict each other's
732 // cache lines. The threads can be either from the same JVM instance, or
733 // from different JVM instances. The benefit is especially true for
734 // processors with hyperthreading technology.
735 static int counter = 0;
736 int pid = os::current_process_id();
737 alloca(((pid ^ counter++) & 7) * 128);
739 ThreadLocalStorage::set_thread(thread);
741 OSThread* osthread = thread->osthread();
742 Monitor* sync = osthread->startThread_lock();
744 // non floating stack LinuxThreads needs extra check, see above
745 if (!_thread_safety_check(thread)) {
746 // notify parent thread
747 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
748 osthread->set_state(ZOMBIE);
749 sync->notify_all();
750 return NULL;
751 }
753 // thread_id is kernel thread id (similar to Solaris LWP id)
754 osthread->set_thread_id(os::Linux::gettid());
756 if (UseNUMA) {
757 int lgrp_id = os::numa_get_group_id();
758 if (lgrp_id != -1) {
759 thread->set_lgrp_id(lgrp_id);
760 }
761 }
762 // initialize signal mask for this thread
763 os::Linux::hotspot_sigmask(thread);
765 // initialize floating point control register
766 os::Linux::init_thread_fpu_state();
768 // handshaking with parent thread
769 {
770 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
772 // notify parent thread
773 osthread->set_state(INITIALIZED);
774 sync->notify_all();
776 // wait until os::start_thread()
777 while (osthread->get_state() == INITIALIZED) {
778 sync->wait(Mutex::_no_safepoint_check_flag);
779 }
780 }
782 // call one more level start routine
783 thread->run();
785 return 0;
786 }
788 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
789 assert(thread->osthread() == NULL, "caller responsible");
791 // Allocate the OSThread object
792 OSThread* osthread = new OSThread(NULL, NULL);
793 if (osthread == NULL) {
794 return false;
795 }
797 // set the correct thread state
798 osthread->set_thread_type(thr_type);
800 // Initial state is ALLOCATED but not INITIALIZED
801 osthread->set_state(ALLOCATED);
803 thread->set_osthread(osthread);
805 // init thread attributes
806 pthread_attr_t attr;
807 pthread_attr_init(&attr);
808 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
810 // stack size
811 if (os::Linux::supports_variable_stack_size()) {
812 // calculate stack size if it's not specified by caller
813 if (stack_size == 0) {
814 stack_size = os::Linux::default_stack_size(thr_type);
816 switch (thr_type) {
817 case os::java_thread:
818 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
819 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
820 break;
821 case os::compiler_thread:
822 if (CompilerThreadStackSize > 0) {
823 stack_size = (size_t)(CompilerThreadStackSize * K);
824 break;
825 } // else fall through:
826 // use VMThreadStackSize if CompilerThreadStackSize is not defined
827 case os::vm_thread:
828 case os::pgc_thread:
829 case os::cgc_thread:
830 case os::watcher_thread:
831 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
832 break;
833 }
834 }
836 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
837 pthread_attr_setstacksize(&attr, stack_size);
838 } else {
839 // let pthread_create() pick the default value.
840 }
842 // glibc guard page
843 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
845 ThreadState state;
847 {
848 // Serialize thread creation if we are running with fixed stack LinuxThreads
849 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
850 if (lock) {
851 os::Linux::createThread_lock()->lock_without_safepoint_check();
852 }
854 pthread_t tid;
855 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
857 pthread_attr_destroy(&attr);
859 if (ret != 0) {
860 if (PrintMiscellaneous && (Verbose || WizardMode)) {
861 perror("pthread_create()");
862 }
863 // Need to clean up stuff we've allocated so far
864 thread->set_osthread(NULL);
865 delete osthread;
866 if (lock) os::Linux::createThread_lock()->unlock();
867 return false;
868 }
870 // Store pthread info into the OSThread
871 osthread->set_pthread_id(tid);
873 // Wait until child thread is either initialized or aborted
874 {
875 Monitor* sync_with_child = osthread->startThread_lock();
876 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
877 while ((state = osthread->get_state()) == ALLOCATED) {
878 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
879 }
880 }
882 if (lock) {
883 os::Linux::createThread_lock()->unlock();
884 }
885 }
887 // Aborted due to thread limit being reached
888 if (state == ZOMBIE) {
889 thread->set_osthread(NULL);
890 delete osthread;
891 return false;
892 }
894 // The thread is returned suspended (in state INITIALIZED),
895 // and is started higher up in the call chain
896 assert(state == INITIALIZED, "race condition");
897 return true;
898 }
900 /////////////////////////////////////////////////////////////////////////////
901 // attach existing thread
903 // bootstrap the main thread
904 bool os::create_main_thread(JavaThread* thread) {
905 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
906 return create_attached_thread(thread);
907 }
909 bool os::create_attached_thread(JavaThread* thread) {
910 #ifdef ASSERT
911 thread->verify_not_published();
912 #endif
914 // Allocate the OSThread object
915 OSThread* osthread = new OSThread(NULL, NULL);
917 if (osthread == NULL) {
918 return false;
919 }
921 // Store pthread info into the OSThread
922 osthread->set_thread_id(os::Linux::gettid());
923 osthread->set_pthread_id(::pthread_self());
925 // initialize floating point control register
926 os::Linux::init_thread_fpu_state();
928 // Initial thread state is RUNNABLE
929 osthread->set_state(RUNNABLE);
931 thread->set_osthread(osthread);
933 if (UseNUMA) {
934 int lgrp_id = os::numa_get_group_id();
935 if (lgrp_id != -1) {
936 thread->set_lgrp_id(lgrp_id);
937 }
938 }
940 if (os::Linux::is_initial_thread()) {
941 // If current thread is initial thread, its stack is mapped on demand,
942 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
943 // the entire stack region to avoid SEGV in stack banging.
944 // It is also useful to get around the heap-stack-gap problem on SuSE
945 // kernel (see 4821821 for details). We first expand stack to the top
946 // of yellow zone, then enable stack yellow zone (order is significant,
947 // enabling yellow zone first will crash JVM on SuSE Linux), so there
948 // is no gap between the last two virtual memory regions.
950 JavaThread *jt = (JavaThread *)thread;
951 address addr = jt->stack_yellow_zone_base();
952 assert(addr != NULL, "initialization problem?");
953 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
955 osthread->set_expanding_stack();
956 os::Linux::manually_expand_stack(jt, addr);
957 osthread->clear_expanding_stack();
958 }
960 // initialize signal mask for this thread
961 // and save the caller's signal mask
962 os::Linux::hotspot_sigmask(thread);
964 return true;
965 }
967 void os::pd_start_thread(Thread* thread) {
968 OSThread * osthread = thread->osthread();
969 assert(osthread->get_state() != INITIALIZED, "just checking");
970 Monitor* sync_with_child = osthread->startThread_lock();
971 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
972 sync_with_child->notify();
973 }
975 // Free Linux resources related to the OSThread
976 void os::free_thread(OSThread* osthread) {
977 assert(osthread != NULL, "osthread not set");
979 if (Thread::current()->osthread() == osthread) {
980 // Restore caller's signal mask
981 sigset_t sigmask = osthread->caller_sigmask();
982 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
983 }
985 delete osthread;
986 }
988 //////////////////////////////////////////////////////////////////////////////
989 // thread local storage
991 int os::allocate_thread_local_storage() {
992 pthread_key_t key;
993 int rslt = pthread_key_create(&key, NULL);
994 assert(rslt == 0, "cannot allocate thread local storage");
995 return (int)key;
996 }
998 // Note: This is currently not used by VM, as we don't destroy TLS key
999 // on VM exit.
1000 void os::free_thread_local_storage(int index) {
1001 int rslt = pthread_key_delete((pthread_key_t)index);
1002 assert(rslt == 0, "invalid index");
1003 }
1005 void os::thread_local_storage_at_put(int index, void* value) {
1006 int rslt = pthread_setspecific((pthread_key_t)index, value);
1007 assert(rslt == 0, "pthread_setspecific failed");
1008 }
1010 extern "C" Thread* get_thread() {
1011 return ThreadLocalStorage::thread();
1012 }
1014 //////////////////////////////////////////////////////////////////////////////
1015 // initial thread
1017 // Check if current thread is the initial thread, similar to Solaris thr_main.
1018 bool os::Linux::is_initial_thread(void) {
1019 char dummy;
1020 // If called before init complete, thread stack bottom will be null.
1021 // Can be called if fatal error occurs before initialization.
1022 if (initial_thread_stack_bottom() == NULL) return false;
1023 assert(initial_thread_stack_bottom() != NULL &&
1024 initial_thread_stack_size() != 0,
1025 "os::init did not locate initial thread's stack region");
1026 if ((address)&dummy >= initial_thread_stack_bottom() &&
1027 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1028 return true;
1029 else return false;
1030 }
1032 // Find the virtual memory area that contains addr
1033 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1034 FILE *fp = fopen("/proc/self/maps", "r");
1035 if (fp) {
1036 address low, high;
1037 while (!feof(fp)) {
1038 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1039 if (low <= addr && addr < high) {
1040 if (vma_low) *vma_low = low;
1041 if (vma_high) *vma_high = high;
1042 fclose (fp);
1043 return true;
1044 }
1045 }
1046 for (;;) {
1047 int ch = fgetc(fp);
1048 if (ch == EOF || ch == (int)'\n') break;
1049 }
1050 }
1051 fclose(fp);
1052 }
1053 return false;
1054 }
1056 // Locate initial thread stack. This special handling of initial thread stack
1057 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1058 // bogus value for initial thread.
1059 void os::Linux::capture_initial_stack(size_t max_size) {
1060 // stack size is the easy part, get it from RLIMIT_STACK
1061 size_t stack_size;
1062 struct rlimit rlim;
1063 getrlimit(RLIMIT_STACK, &rlim);
1064 stack_size = rlim.rlim_cur;
1066 // 6308388: a bug in ld.so will relocate its own .data section to the
1067 // lower end of primordial stack; reduce ulimit -s value a little bit
1068 // so we won't install guard page on ld.so's data section.
1069 stack_size -= 2 * page_size();
1071 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1072 // 7.1, in both cases we will get 2G in return value.
1073 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1074 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1075 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1076 // in case other parts in glibc still assumes 2M max stack size.
1077 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1078 #ifndef IA64
1079 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1080 #else
1081 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1082 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1083 #endif
1085 // Try to figure out where the stack base (top) is. This is harder.
1086 //
1087 // When an application is started, glibc saves the initial stack pointer in
1088 // a global variable "__libc_stack_end", which is then used by system
1089 // libraries. __libc_stack_end should be pretty close to stack top. The
1090 // variable is available since the very early days. However, because it is
1091 // a private interface, it could disappear in the future.
1092 //
1093 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1094 // to __libc_stack_end, it is very close to stack top, but isn't the real
1095 // stack top. Note that /proc may not exist if VM is running as a chroot
1096 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1097 // /proc/<pid>/stat could change in the future (though unlikely).
1098 //
1099 // We try __libc_stack_end first. If that doesn't work, look for
1100 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1101 // as a hint, which should work well in most cases.
1103 uintptr_t stack_start;
1105 // try __libc_stack_end first
1106 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1107 if (p && *p) {
1108 stack_start = *p;
1109 } else {
1110 // see if we can get the start_stack field from /proc/self/stat
1111 FILE *fp;
1112 int pid;
1113 char state;
1114 int ppid;
1115 int pgrp;
1116 int session;
1117 int nr;
1118 int tpgrp;
1119 unsigned long flags;
1120 unsigned long minflt;
1121 unsigned long cminflt;
1122 unsigned long majflt;
1123 unsigned long cmajflt;
1124 unsigned long utime;
1125 unsigned long stime;
1126 long cutime;
1127 long cstime;
1128 long prio;
1129 long nice;
1130 long junk;
1131 long it_real;
1132 uintptr_t start;
1133 uintptr_t vsize;
1134 uintptr_t rss;
1135 unsigned long rsslim;
1136 uintptr_t scodes;
1137 uintptr_t ecode;
1138 int i;
1140 // Figure what the primordial thread stack base is. Code is inspired
1141 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1142 // followed by command name surrounded by parentheses, state, etc.
1143 char stat[2048];
1144 int statlen;
1146 fp = fopen("/proc/self/stat", "r");
1147 if (fp) {
1148 statlen = fread(stat, 1, 2047, fp);
1149 stat[statlen] = '\0';
1150 fclose(fp);
1152 // Skip pid and the command string. Note that we could be dealing with
1153 // weird command names, e.g. user could decide to rename java launcher
1154 // to "java 1.4.2 :)", then the stat file would look like
1155 // 1234 (java 1.4.2 :)) R ... ...
1156 // We don't really need to know the command string, just find the last
1157 // occurrence of ")" and then start parsing from there. See bug 4726580.
1158 char * s = strrchr(stat, ')');
1160 i = 0;
1161 if (s) {
1162 // Skip blank chars
1163 do s++; while (isspace(*s));
1165 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1166 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */
1167 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
1168 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
1169 " %lu "
1170 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
1171 &state, /* 3 %c */
1172 &ppid, /* 4 %d */
1173 &pgrp, /* 5 %d */
1174 &session, /* 6 %d */
1175 &nr, /* 7 %d */
1176 &tpgrp, /* 8 %d */
1177 &flags, /* 9 %lu */
1178 &minflt, /* 10 %lu */
1179 &cminflt, /* 11 %lu */
1180 &majflt, /* 12 %lu */
1181 &cmajflt, /* 13 %lu */
1182 &utime, /* 14 %lu */
1183 &stime, /* 15 %lu */
1184 &cutime, /* 16 %ld */
1185 &cstime, /* 17 %ld */
1186 &prio, /* 18 %ld */
1187 &nice, /* 19 %ld */
1188 &junk, /* 20 %ld */
1189 &it_real, /* 21 %ld */
1190 &start, /* 22 UINTX_FORMAT */
1191 &vsize, /* 23 UINTX_FORMAT */
1192 &rss, /* 24 UINTX_FORMAT */
1193 &rsslim, /* 25 %lu */
1194 &scodes, /* 26 UINTX_FORMAT */
1195 &ecode, /* 27 UINTX_FORMAT */
1196 &stack_start); /* 28 UINTX_FORMAT */
1197 }
1199 if (i != 28 - 2) {
1200 assert(false, "Bad conversion from /proc/self/stat");
1201 // product mode - assume we are the initial thread, good luck in the
1202 // embedded case.
1203 warning("Can't detect initial thread stack location - bad conversion");
1204 stack_start = (uintptr_t) &rlim;
1205 }
1206 } else {
1207 // For some reason we can't open /proc/self/stat (for example, running on
1208 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1209 // most cases, so don't abort:
1210 warning("Can't detect initial thread stack location - no /proc/self/stat");
1211 stack_start = (uintptr_t) &rlim;
1212 }
1213 }
1215 // Now we have a pointer (stack_start) very close to the stack top, the
1216 // next thing to do is to figure out the exact location of stack top. We
1217 // can find out the virtual memory area that contains stack_start by
1218 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1219 // and its upper limit is the real stack top. (again, this would fail if
1220 // running inside chroot, because /proc may not exist.)
1222 uintptr_t stack_top;
1223 address low, high;
1224 if (find_vma((address)stack_start, &low, &high)) {
1225 // success, "high" is the true stack top. (ignore "low", because initial
1226 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1227 stack_top = (uintptr_t)high;
1228 } else {
1229 // failed, likely because /proc/self/maps does not exist
1230 warning("Can't detect initial thread stack location - find_vma failed");
1231 // best effort: stack_start is normally within a few pages below the real
1232 // stack top, use it as stack top, and reduce stack size so we won't put
1233 // guard page outside stack.
1234 stack_top = stack_start;
1235 stack_size -= 16 * page_size();
1236 }
1238 // stack_top could be partially down the page so align it
1239 stack_top = align_size_up(stack_top, page_size());
1241 if (max_size && stack_size > max_size) {
1242 _initial_thread_stack_size = max_size;
1243 } else {
1244 _initial_thread_stack_size = stack_size;
1245 }
1247 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1248 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1249 }
1251 ////////////////////////////////////////////////////////////////////////////////
1252 // time support
1254 // Time since start-up in seconds to a fine granularity.
1255 // Used by VMSelfDestructTimer and the MemProfiler.
1256 double os::elapsedTime() {
1258 return (double)(os::elapsed_counter()) * 0.000001;
1259 }
1261 jlong os::elapsed_counter() {
1262 timeval time;
1263 int status = gettimeofday(&time, NULL);
1264 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1265 }
1267 jlong os::elapsed_frequency() {
1268 return (1000 * 1000);
1269 }
1271 // For now, we say that linux does not support vtime. I have no idea
1272 // whether it can actually be made to (DLD, 9/13/05).
1274 bool os::supports_vtime() { return false; }
1275 bool os::enable_vtime() { return false; }
1276 bool os::vtime_enabled() { return false; }
1277 double os::elapsedVTime() {
1278 // better than nothing, but not much
1279 return elapsedTime();
1280 }
1282 jlong os::javaTimeMillis() {
1283 timeval time;
1284 int status = gettimeofday(&time, NULL);
1285 assert(status != -1, "linux error");
1286 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1287 }
1289 #ifndef CLOCK_MONOTONIC
1290 #define CLOCK_MONOTONIC (1)
1291 #endif
1293 void os::Linux::clock_init() {
1294 // we do dlopen's in this particular order due to bug in linux
1295 // dynamical loader (see 6348968) leading to crash on exit
1296 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1297 if (handle == NULL) {
1298 handle = dlopen("librt.so", RTLD_LAZY);
1299 }
1301 if (handle) {
1302 int (*clock_getres_func)(clockid_t, struct timespec*) =
1303 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1304 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1305 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1306 if (clock_getres_func && clock_gettime_func) {
1307 // See if monotonic clock is supported by the kernel. Note that some
1308 // early implementations simply return kernel jiffies (updated every
1309 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1310 // for nano time (though the monotonic property is still nice to have).
1311 // It's fixed in newer kernels, however clock_getres() still returns
1312 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1313 // resolution for now. Hopefully as people move to new kernels, this
1314 // won't be a problem.
1315 struct timespec res;
1316 struct timespec tp;
1317 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1318 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1319 // yes, monotonic clock is supported
1320 _clock_gettime = clock_gettime_func;
1321 } else {
1322 // close librt if there is no monotonic clock
1323 dlclose(handle);
1324 }
1325 }
1326 }
1327 }
1329 #ifndef SYS_clock_getres
1331 #if defined(IA32) || defined(AMD64)
1332 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1333 #else
1334 #error Value of SYS_clock_getres not known on this platform
1335 #endif
1337 #endif
1339 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1341 void os::Linux::fast_thread_clock_init() {
1342 if (!UseLinuxPosixThreadCPUClocks) {
1343 return;
1344 }
1345 clockid_t clockid;
1346 struct timespec tp;
1347 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1348 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1350 // Switch to using fast clocks for thread cpu time if
1351 // the sys_clock_getres() returns 0 error code.
1352 // Note, that some kernels may support the current thread
1353 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1354 // returned by the pthread_getcpuclockid().
1355 // If the fast Posix clocks are supported then the sys_clock_getres()
1356 // must return at least tp.tv_sec == 0 which means a resolution
1357 // better than 1 sec. This is extra check for reliability.
1359 if(pthread_getcpuclockid_func &&
1360 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1361 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1363 _supports_fast_thread_cpu_time = true;
1364 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1365 }
1366 }
1368 jlong os::javaTimeNanos() {
1369 if (Linux::supports_monotonic_clock()) {
1370 struct timespec tp;
1371 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1372 assert(status == 0, "gettime error");
1373 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1374 return result;
1375 } else {
1376 timeval time;
1377 int status = gettimeofday(&time, NULL);
1378 assert(status != -1, "linux error");
1379 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1380 return 1000 * usecs;
1381 }
1382 }
1384 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1385 if (Linux::supports_monotonic_clock()) {
1386 info_ptr->max_value = ALL_64_BITS;
1388 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1389 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1390 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1391 } else {
1392 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1393 info_ptr->max_value = ALL_64_BITS;
1395 // gettimeofday is a real time clock so it skips
1396 info_ptr->may_skip_backward = true;
1397 info_ptr->may_skip_forward = true;
1398 }
1400 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1401 }
1403 // Return the real, user, and system times in seconds from an
1404 // arbitrary fixed point in the past.
1405 bool os::getTimesSecs(double* process_real_time,
1406 double* process_user_time,
1407 double* process_system_time) {
1408 struct tms ticks;
1409 clock_t real_ticks = times(&ticks);
1411 if (real_ticks == (clock_t) (-1)) {
1412 return false;
1413 } else {
1414 double ticks_per_second = (double) clock_tics_per_sec;
1415 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1416 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1417 *process_real_time = ((double) real_ticks) / ticks_per_second;
1419 return true;
1420 }
1421 }
1424 char * os::local_time_string(char *buf, size_t buflen) {
1425 struct tm t;
1426 time_t long_time;
1427 time(&long_time);
1428 localtime_r(&long_time, &t);
1429 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1430 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1431 t.tm_hour, t.tm_min, t.tm_sec);
1432 return buf;
1433 }
1435 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1436 return localtime_r(clock, res);
1437 }
1439 ////////////////////////////////////////////////////////////////////////////////
1440 // runtime exit support
1442 // Note: os::shutdown() might be called very early during initialization, or
1443 // called from signal handler. Before adding something to os::shutdown(), make
1444 // sure it is async-safe and can handle partially initialized VM.
1445 void os::shutdown() {
1447 // allow PerfMemory to attempt cleanup of any persistent resources
1448 perfMemory_exit();
1450 // needs to remove object in file system
1451 AttachListener::abort();
1453 // flush buffered output, finish log files
1454 ostream_abort();
1456 // Check for abort hook
1457 abort_hook_t abort_hook = Arguments::abort_hook();
1458 if (abort_hook != NULL) {
1459 abort_hook();
1460 }
1462 }
1464 // Note: os::abort() might be called very early during initialization, or
1465 // called from signal handler. Before adding something to os::abort(), make
1466 // sure it is async-safe and can handle partially initialized VM.
1467 void os::abort(bool dump_core) {
1468 os::shutdown();
1469 if (dump_core) {
1470 #ifndef PRODUCT
1471 fdStream out(defaultStream::output_fd());
1472 out.print_raw("Current thread is ");
1473 char buf[16];
1474 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1475 out.print_raw_cr(buf);
1476 out.print_raw_cr("Dumping core ...");
1477 #endif
1478 ::abort(); // dump core
1479 }
1481 ::exit(1);
1482 }
1484 // Die immediately, no exit hook, no abort hook, no cleanup.
1485 void os::die() {
1486 // _exit() on LinuxThreads only kills current thread
1487 ::abort();
1488 }
1490 // unused on linux for now.
1491 void os::set_error_file(const char *logfile) {}
1493 intx os::current_thread_id() { return (intx)pthread_self(); }
1494 int os::current_process_id() {
1496 // Under the old linux thread library, linux gives each thread
1497 // its own process id. Because of this each thread will return
1498 // a different pid if this method were to return the result
1499 // of getpid(2). Linux provides no api that returns the pid
1500 // of the launcher thread for the vm. This implementation
1501 // returns a unique pid, the pid of the launcher thread
1502 // that starts the vm 'process'.
1504 // Under the NPTL, getpid() returns the same pid as the
1505 // launcher thread rather than a unique pid per thread.
1506 // Use gettid() if you want the old pre NPTL behaviour.
1508 // if you are looking for the result of a call to getpid() that
1509 // returns a unique pid for the calling thread, then look at the
1510 // OSThread::thread_id() method in osThread_linux.hpp file
1512 return (int)(_initial_pid ? _initial_pid : getpid());
1513 }
1515 // DLL functions
1517 const char* os::dll_file_extension() { return ".so"; }
1519 const char* os::get_temp_directory() { return "/tmp/"; }
1521 static bool file_exists(const char* filename) {
1522 struct stat statbuf;
1523 if (filename == NULL || strlen(filename) == 0) {
1524 return false;
1525 }
1526 return os::stat(filename, &statbuf) == 0;
1527 }
1529 void os::dll_build_name(char* buffer, size_t buflen,
1530 const char* pname, const char* fname) {
1531 // Copied from libhpi
1532 const size_t pnamelen = pname ? strlen(pname) : 0;
1534 // Quietly truncate on buffer overflow. Should be an error.
1535 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1536 *buffer = '\0';
1537 return;
1538 }
1540 if (pnamelen == 0) {
1541 snprintf(buffer, buflen, "lib%s.so", fname);
1542 } else if (strchr(pname, *os::path_separator()) != NULL) {
1543 int n;
1544 char** pelements = split_path(pname, &n);
1545 for (int i = 0 ; i < n ; i++) {
1546 // Really shouldn't be NULL, but check can't hurt
1547 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1548 continue; // skip the empty path values
1549 }
1550 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1551 if (file_exists(buffer)) {
1552 break;
1553 }
1554 }
1555 // release the storage
1556 for (int i = 0 ; i < n ; i++) {
1557 if (pelements[i] != NULL) {
1558 FREE_C_HEAP_ARRAY(char, pelements[i]);
1559 }
1560 }
1561 if (pelements != NULL) {
1562 FREE_C_HEAP_ARRAY(char*, pelements);
1563 }
1564 } else {
1565 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1566 }
1567 }
1569 const char* os::get_current_directory(char *buf, int buflen) {
1570 return getcwd(buf, buflen);
1571 }
1573 // check if addr is inside libjvm[_g].so
1574 bool os::address_is_in_vm(address addr) {
1575 static address libjvm_base_addr;
1576 Dl_info dlinfo;
1578 if (libjvm_base_addr == NULL) {
1579 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1580 libjvm_base_addr = (address)dlinfo.dli_fbase;
1581 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1582 }
1584 if (dladdr((void *)addr, &dlinfo)) {
1585 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1586 }
1588 return false;
1589 }
1591 bool os::dll_address_to_function_name(address addr, char *buf,
1592 int buflen, int *offset) {
1593 Dl_info dlinfo;
1595 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1596 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1597 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1598 return true;
1599 } else {
1600 if (buf) buf[0] = '\0';
1601 if (offset) *offset = -1;
1602 return false;
1603 }
1604 }
1606 struct _address_to_library_name {
1607 address addr; // input : memory address
1608 size_t buflen; // size of fname
1609 char* fname; // output: library name
1610 address base; // library base addr
1611 };
1613 static int address_to_library_name_callback(struct dl_phdr_info *info,
1614 size_t size, void *data) {
1615 int i;
1616 bool found = false;
1617 address libbase = NULL;
1618 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1620 // iterate through all loadable segments
1621 for (i = 0; i < info->dlpi_phnum; i++) {
1622 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1623 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1624 // base address of a library is the lowest address of its loaded
1625 // segments.
1626 if (libbase == NULL || libbase > segbase) {
1627 libbase = segbase;
1628 }
1629 // see if 'addr' is within current segment
1630 if (segbase <= d->addr &&
1631 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1632 found = true;
1633 }
1634 }
1635 }
1637 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1638 // so dll_address_to_library_name() can fall through to use dladdr() which
1639 // can figure out executable name from argv[0].
1640 if (found && info->dlpi_name && info->dlpi_name[0]) {
1641 d->base = libbase;
1642 if (d->fname) {
1643 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1644 }
1645 return 1;
1646 }
1647 return 0;
1648 }
1650 bool os::dll_address_to_library_name(address addr, char* buf,
1651 int buflen, int* offset) {
1652 Dl_info dlinfo;
1653 struct _address_to_library_name data;
1655 // There is a bug in old glibc dladdr() implementation that it could resolve
1656 // to wrong library name if the .so file has a base address != NULL. Here
1657 // we iterate through the program headers of all loaded libraries to find
1658 // out which library 'addr' really belongs to. This workaround can be
1659 // removed once the minimum requirement for glibc is moved to 2.3.x.
1660 data.addr = addr;
1661 data.fname = buf;
1662 data.buflen = buflen;
1663 data.base = NULL;
1664 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1666 if (rslt) {
1667 // buf already contains library name
1668 if (offset) *offset = addr - data.base;
1669 return true;
1670 } else if (dladdr((void*)addr, &dlinfo)){
1671 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1672 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1673 return true;
1674 } else {
1675 if (buf) buf[0] = '\0';
1676 if (offset) *offset = -1;
1677 return false;
1678 }
1679 }
1681 // Loads .dll/.so and
1682 // in case of error it checks if .dll/.so was built for the
1683 // same architecture as Hotspot is running on
1685 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1686 {
1687 void * result= ::dlopen(filename, RTLD_LAZY);
1688 if (result != NULL) {
1689 // Successful loading
1690 return result;
1691 }
1693 Elf32_Ehdr elf_head;
1695 // Read system error message into ebuf
1696 // It may or may not be overwritten below
1697 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1698 ebuf[ebuflen-1]='\0';
1699 int diag_msg_max_length=ebuflen-strlen(ebuf);
1700 char* diag_msg_buf=ebuf+strlen(ebuf);
1702 if (diag_msg_max_length==0) {
1703 // No more space in ebuf for additional diagnostics message
1704 return NULL;
1705 }
1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1710 if (file_descriptor < 0) {
1711 // Can't open library, report dlerror() message
1712 return NULL;
1713 }
1715 bool failed_to_read_elf_head=
1716 (sizeof(elf_head)!=
1717 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1719 ::close(file_descriptor);
1720 if (failed_to_read_elf_head) {
1721 // file i/o error - report dlerror() msg
1722 return NULL;
1723 }
1725 typedef struct {
1726 Elf32_Half code; // Actual value as defined in elf.h
1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1728 char elf_class; // 32 or 64 bit
1729 char endianess; // MSB or LSB
1730 char* name; // String representation
1731 } arch_t;
1733 #ifndef EM_486
1734 #define EM_486 6 /* Intel 80486 */
1735 #endif
1737 static const arch_t arch_array[]={
1738 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1739 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1740 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1741 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1742 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1743 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1744 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1745 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1747 };
1749 #if (defined IA32)
1750 static Elf32_Half running_arch_code=EM_386;
1751 #elif (defined AMD64)
1752 static Elf32_Half running_arch_code=EM_X86_64;
1753 #elif (defined IA64)
1754 static Elf32_Half running_arch_code=EM_IA_64;
1755 #elif (defined __sparc) && (defined _LP64)
1756 static Elf32_Half running_arch_code=EM_SPARCV9;
1757 #elif (defined __sparc) && (!defined _LP64)
1758 static Elf32_Half running_arch_code=EM_SPARC;
1759 #elif (defined __powerpc64__)
1760 static Elf32_Half running_arch_code=EM_PPC64;
1761 #elif (defined __powerpc__)
1762 static Elf32_Half running_arch_code=EM_PPC;
1763 #else
1764 #error Method os::dll_load requires that one of following is defined:\
1765 IA32, AMD64, IA64, __sparc, __powerpc__
1766 #endif
1768 // Identify compatability class for VM's architecture and library's architecture
1769 // Obtain string descriptions for architectures
1771 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1772 int running_arch_index=-1;
1774 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1775 if (running_arch_code == arch_array[i].code) {
1776 running_arch_index = i;
1777 }
1778 if (lib_arch.code == arch_array[i].code) {
1779 lib_arch.compat_class = arch_array[i].compat_class;
1780 lib_arch.name = arch_array[i].name;
1781 }
1782 }
1784 assert(running_arch_index != -1,
1785 "Didn't find running architecture code (running_arch_code) in arch_array");
1786 if (running_arch_index == -1) {
1787 // Even though running architecture detection failed
1788 // we may still continue with reporting dlerror() message
1789 return NULL;
1790 }
1792 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1793 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1794 return NULL;
1795 }
1797 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1798 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1799 return NULL;
1800 }
1802 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1803 if ( lib_arch.name!=NULL ) {
1804 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1805 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1806 lib_arch.name, arch_array[running_arch_index].name);
1807 } else {
1808 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1809 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1810 lib_arch.code,
1811 arch_array[running_arch_index].name);
1812 }
1813 }
1815 return NULL;
1816 }
1818 /*
1819 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1820 * chances are you might want to run the generated bits against glibc-2.0
1821 * libdl.so, so always use locking for any version of glibc.
1822 */
1823 void* os::dll_lookup(void* handle, const char* name) {
1824 pthread_mutex_lock(&dl_mutex);
1825 void* res = dlsym(handle, name);
1826 pthread_mutex_unlock(&dl_mutex);
1827 return res;
1828 }
1831 bool _print_ascii_file(const char* filename, outputStream* st) {
1832 int fd = open(filename, O_RDONLY);
1833 if (fd == -1) {
1834 return false;
1835 }
1837 char buf[32];
1838 int bytes;
1839 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1840 st->print_raw(buf, bytes);
1841 }
1843 close(fd);
1845 return true;
1846 }
1848 void os::print_dll_info(outputStream *st) {
1849 st->print_cr("Dynamic libraries:");
1851 char fname[32];
1852 pid_t pid = os::Linux::gettid();
1854 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1856 if (!_print_ascii_file(fname, st)) {
1857 st->print("Can not get library information for pid = %d\n", pid);
1858 }
1859 }
1862 void os::print_os_info(outputStream* st) {
1863 st->print("OS:");
1865 // Try to identify popular distros.
1866 // Most Linux distributions have /etc/XXX-release file, which contains
1867 // the OS version string. Some have more than one /etc/XXX-release file
1868 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1869 // so the order is important.
1870 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1871 !_print_ascii_file("/etc/sun-release", st) &&
1872 !_print_ascii_file("/etc/redhat-release", st) &&
1873 !_print_ascii_file("/etc/SuSE-release", st) &&
1874 !_print_ascii_file("/etc/turbolinux-release", st) &&
1875 !_print_ascii_file("/etc/gentoo-release", st) &&
1876 !_print_ascii_file("/etc/debian_version", st)) {
1877 st->print("Linux");
1878 }
1879 st->cr();
1881 // kernel
1882 st->print("uname:");
1883 struct utsname name;
1884 uname(&name);
1885 st->print(name.sysname); st->print(" ");
1886 st->print(name.release); st->print(" ");
1887 st->print(name.version); st->print(" ");
1888 st->print(name.machine);
1889 st->cr();
1891 // Print warning if unsafe chroot environment detected
1892 if (unsafe_chroot_detected) {
1893 st->print("WARNING!! ");
1894 st->print_cr(unstable_chroot_error);
1895 }
1897 // libc, pthread
1898 st->print("libc:");
1899 st->print(os::Linux::glibc_version()); st->print(" ");
1900 st->print(os::Linux::libpthread_version()); st->print(" ");
1901 if (os::Linux::is_LinuxThreads()) {
1902 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1903 }
1904 st->cr();
1906 // rlimit
1907 st->print("rlimit:");
1908 struct rlimit rlim;
1910 st->print(" STACK ");
1911 getrlimit(RLIMIT_STACK, &rlim);
1912 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1913 else st->print("%uk", rlim.rlim_cur >> 10);
1915 st->print(", CORE ");
1916 getrlimit(RLIMIT_CORE, &rlim);
1917 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1918 else st->print("%uk", rlim.rlim_cur >> 10);
1920 st->print(", NPROC ");
1921 getrlimit(RLIMIT_NPROC, &rlim);
1922 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1923 else st->print("%d", rlim.rlim_cur);
1925 st->print(", NOFILE ");
1926 getrlimit(RLIMIT_NOFILE, &rlim);
1927 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1928 else st->print("%d", rlim.rlim_cur);
1930 st->print(", AS ");
1931 getrlimit(RLIMIT_AS, &rlim);
1932 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1933 else st->print("%uk", rlim.rlim_cur >> 10);
1934 st->cr();
1936 // load average
1937 st->print("load average:");
1938 double loadavg[3];
1939 os::loadavg(loadavg, 3);
1940 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1941 st->cr();
1942 }
1944 void os::print_memory_info(outputStream* st) {
1946 st->print("Memory:");
1947 st->print(" %dk page", os::vm_page_size()>>10);
1949 // values in struct sysinfo are "unsigned long"
1950 struct sysinfo si;
1951 sysinfo(&si);
1953 st->print(", physical " UINT64_FORMAT "k",
1954 os::physical_memory() >> 10);
1955 st->print("(" UINT64_FORMAT "k free)",
1956 os::available_memory() >> 10);
1957 st->print(", swap " UINT64_FORMAT "k",
1958 ((jlong)si.totalswap * si.mem_unit) >> 10);
1959 st->print("(" UINT64_FORMAT "k free)",
1960 ((jlong)si.freeswap * si.mem_unit) >> 10);
1961 st->cr();
1962 }
1964 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
1965 // but they're the same for all the linux arch that we support
1966 // and they're the same for solaris but there's no common place to put this.
1967 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1968 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1969 "ILL_COPROC", "ILL_BADSTK" };
1971 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1972 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1973 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1975 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1977 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1979 void os::print_siginfo(outputStream* st, void* siginfo) {
1980 st->print("siginfo:");
1982 const int buflen = 100;
1983 char buf[buflen];
1984 siginfo_t *si = (siginfo_t*)siginfo;
1985 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1986 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1987 st->print("si_errno=%s", buf);
1988 } else {
1989 st->print("si_errno=%d", si->si_errno);
1990 }
1991 const int c = si->si_code;
1992 assert(c > 0, "unexpected si_code");
1993 switch (si->si_signo) {
1994 case SIGILL:
1995 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1996 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1997 break;
1998 case SIGFPE:
1999 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2000 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2001 break;
2002 case SIGSEGV:
2003 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2004 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2005 break;
2006 case SIGBUS:
2007 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2008 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2009 break;
2010 default:
2011 st->print(", si_code=%d", si->si_code);
2012 // no si_addr
2013 }
2015 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2016 UseSharedSpaces) {
2017 FileMapInfo* mapinfo = FileMapInfo::current_info();
2018 if (mapinfo->is_in_shared_space(si->si_addr)) {
2019 st->print("\n\nError accessing class data sharing archive." \
2020 " Mapped file inaccessible during execution, " \
2021 " possible disk/network problem.");
2022 }
2023 }
2024 st->cr();
2025 }
2028 static void print_signal_handler(outputStream* st, int sig,
2029 char* buf, size_t buflen);
2031 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2032 st->print_cr("Signal Handlers:");
2033 print_signal_handler(st, SIGSEGV, buf, buflen);
2034 print_signal_handler(st, SIGBUS , buf, buflen);
2035 print_signal_handler(st, SIGFPE , buf, buflen);
2036 print_signal_handler(st, SIGPIPE, buf, buflen);
2037 print_signal_handler(st, SIGXFSZ, buf, buflen);
2038 print_signal_handler(st, SIGILL , buf, buflen);
2039 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2040 print_signal_handler(st, SR_signum, buf, buflen);
2041 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2042 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2043 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2044 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2045 }
2047 static char saved_jvm_path[MAXPATHLEN] = {0};
2049 // Find the full path to the current module, libjvm.so or libjvm_g.so
2050 void os::jvm_path(char *buf, jint len) {
2051 // Error checking.
2052 if (len < MAXPATHLEN) {
2053 assert(false, "must use a large-enough buffer");
2054 buf[0] = '\0';
2055 return;
2056 }
2057 // Lazy resolve the path to current module.
2058 if (saved_jvm_path[0] != 0) {
2059 strcpy(buf, saved_jvm_path);
2060 return;
2061 }
2063 char dli_fname[MAXPATHLEN];
2064 bool ret = dll_address_to_library_name(
2065 CAST_FROM_FN_PTR(address, os::jvm_path),
2066 dli_fname, sizeof(dli_fname), NULL);
2067 assert(ret != 0, "cannot locate libjvm");
2068 if (realpath(dli_fname, buf) == NULL)
2069 return;
2071 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2072 // Support for the gamma launcher. Typical value for buf is
2073 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2074 // the right place in the string, then assume we are installed in a JDK and
2075 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2076 // up the path so it looks like libjvm.so is installed there (append a
2077 // fake suffix hotspot/libjvm.so).
2078 const char *p = buf + strlen(buf) - 1;
2079 for (int count = 0; p > buf && count < 5; ++count) {
2080 for (--p; p > buf && *p != '/'; --p)
2081 /* empty */ ;
2082 }
2084 if (strncmp(p, "/jre/lib/", 9) != 0) {
2085 // Look for JAVA_HOME in the environment.
2086 char* java_home_var = ::getenv("JAVA_HOME");
2087 if (java_home_var != NULL && java_home_var[0] != 0) {
2088 // Check the current module name "libjvm.so" or "libjvm_g.so".
2089 p = strrchr(buf, '/');
2090 assert(strstr(p, "/libjvm") == p, "invalid library name");
2091 p = strstr(p, "_g") ? "_g" : "";
2093 if (realpath(java_home_var, buf) == NULL)
2094 return;
2095 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2096 if (0 == access(buf, F_OK)) {
2097 // Use current module name "libjvm[_g].so" instead of
2098 // "libjvm"debug_only("_g")".so" since for fastdebug version
2099 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2100 // It is used when we are choosing the HPI library's name
2101 // "libhpi[_g].so" in hpi::initialize_get_interface().
2102 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2103 } else {
2104 // Go back to path of .so
2105 if (realpath(dli_fname, buf) == NULL)
2106 return;
2107 }
2108 }
2109 }
2110 }
2112 strcpy(saved_jvm_path, buf);
2113 }
2115 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2116 // no prefix required, not even "_"
2117 }
2119 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2120 // no suffix required
2121 }
2123 ////////////////////////////////////////////////////////////////////////////////
2124 // sun.misc.Signal support
2126 static volatile jint sigint_count = 0;
2128 static void
2129 UserHandler(int sig, void *siginfo, void *context) {
2130 // 4511530 - sem_post is serialized and handled by the manager thread. When
2131 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2132 // don't want to flood the manager thread with sem_post requests.
2133 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2134 return;
2136 // Ctrl-C is pressed during error reporting, likely because the error
2137 // handler fails to abort. Let VM die immediately.
2138 if (sig == SIGINT && is_error_reported()) {
2139 os::die();
2140 }
2142 os::signal_notify(sig);
2143 }
2145 void* os::user_handler() {
2146 return CAST_FROM_FN_PTR(void*, UserHandler);
2147 }
2149 extern "C" {
2150 typedef void (*sa_handler_t)(int);
2151 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2152 }
2154 void* os::signal(int signal_number, void* handler) {
2155 struct sigaction sigAct, oldSigAct;
2157 sigfillset(&(sigAct.sa_mask));
2158 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2159 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2161 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2162 // -1 means registration failed
2163 return (void *)-1;
2164 }
2166 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2167 }
2169 void os::signal_raise(int signal_number) {
2170 ::raise(signal_number);
2171 }
2173 /*
2174 * The following code is moved from os.cpp for making this
2175 * code platform specific, which it is by its very nature.
2176 */
2178 // Will be modified when max signal is changed to be dynamic
2179 int os::sigexitnum_pd() {
2180 return NSIG;
2181 }
2183 // a counter for each possible signal value
2184 static volatile jint pending_signals[NSIG+1] = { 0 };
2186 // Linux(POSIX) specific hand shaking semaphore.
2187 static sem_t sig_sem;
2189 void os::signal_init_pd() {
2190 // Initialize signal structures
2191 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2193 // Initialize signal semaphore
2194 ::sem_init(&sig_sem, 0, 0);
2195 }
2197 void os::signal_notify(int sig) {
2198 Atomic::inc(&pending_signals[sig]);
2199 ::sem_post(&sig_sem);
2200 }
2202 static int check_pending_signals(bool wait) {
2203 Atomic::store(0, &sigint_count);
2204 for (;;) {
2205 for (int i = 0; i < NSIG + 1; i++) {
2206 jint n = pending_signals[i];
2207 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2208 return i;
2209 }
2210 }
2211 if (!wait) {
2212 return -1;
2213 }
2214 JavaThread *thread = JavaThread::current();
2215 ThreadBlockInVM tbivm(thread);
2217 bool threadIsSuspended;
2218 do {
2219 thread->set_suspend_equivalent();
2220 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2221 ::sem_wait(&sig_sem);
2223 // were we externally suspended while we were waiting?
2224 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2225 if (threadIsSuspended) {
2226 //
2227 // The semaphore has been incremented, but while we were waiting
2228 // another thread suspended us. We don't want to continue running
2229 // while suspended because that would surprise the thread that
2230 // suspended us.
2231 //
2232 ::sem_post(&sig_sem);
2234 thread->java_suspend_self();
2235 }
2236 } while (threadIsSuspended);
2237 }
2238 }
2240 int os::signal_lookup() {
2241 return check_pending_signals(false);
2242 }
2244 int os::signal_wait() {
2245 return check_pending_signals(true);
2246 }
2248 ////////////////////////////////////////////////////////////////////////////////
2249 // Virtual Memory
2251 int os::vm_page_size() {
2252 // Seems redundant as all get out
2253 assert(os::Linux::page_size() != -1, "must call os::init");
2254 return os::Linux::page_size();
2255 }
2257 // Solaris allocates memory by pages.
2258 int os::vm_allocation_granularity() {
2259 assert(os::Linux::page_size() != -1, "must call os::init");
2260 return os::Linux::page_size();
2261 }
2263 // Rationale behind this function:
2264 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2265 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2266 // samples for JITted code. Here we create private executable mapping over the code cache
2267 // and then we can use standard (well, almost, as mapping can change) way to provide
2268 // info for the reporting script by storing timestamp and location of symbol
2269 void linux_wrap_code(char* base, size_t size) {
2270 static volatile jint cnt = 0;
2272 if (!UseOprofile) {
2273 return;
2274 }
2276 char buf[40];
2277 int num = Atomic::add(1, &cnt);
2279 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2280 unlink(buf);
2282 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2284 if (fd != -1) {
2285 off_t rv = lseek(fd, size-2, SEEK_SET);
2286 if (rv != (off_t)-1) {
2287 if (write(fd, "", 1) == 1) {
2288 mmap(base, size,
2289 PROT_READ|PROT_WRITE|PROT_EXEC,
2290 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2291 }
2292 }
2293 close(fd);
2294 unlink(buf);
2295 }
2296 }
2298 // NOTE: Linux kernel does not really reserve the pages for us.
2299 // All it does is to check if there are enough free pages
2300 // left at the time of mmap(). This could be a potential
2301 // problem.
2302 bool os::commit_memory(char* addr, size_t size, bool exec) {
2303 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2304 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2305 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2306 return res != (uintptr_t) MAP_FAILED;
2307 }
2309 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2310 bool exec) {
2311 return commit_memory(addr, size, exec);
2312 }
2314 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2316 void os::free_memory(char *addr, size_t bytes) {
2317 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2318 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2319 }
2321 void os::numa_make_global(char *addr, size_t bytes) {
2322 Linux::numa_interleave_memory(addr, bytes);
2323 }
2325 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2326 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2327 }
2329 bool os::numa_topology_changed() { return false; }
2331 size_t os::numa_get_groups_num() {
2332 int max_node = Linux::numa_max_node();
2333 return max_node > 0 ? max_node + 1 : 1;
2334 }
2336 int os::numa_get_group_id() {
2337 int cpu_id = Linux::sched_getcpu();
2338 if (cpu_id != -1) {
2339 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2340 if (lgrp_id != -1) {
2341 return lgrp_id;
2342 }
2343 }
2344 return 0;
2345 }
2347 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2348 for (size_t i = 0; i < size; i++) {
2349 ids[i] = i;
2350 }
2351 return size;
2352 }
2354 bool os::get_page_info(char *start, page_info* info) {
2355 return false;
2356 }
2358 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2359 return end;
2360 }
2362 extern "C" void numa_warn(int number, char *where, ...) { }
2363 extern "C" void numa_error(char *where) { }
2366 // If we are running with libnuma version > 2, then we should
2367 // be trying to use symbols with versions 1.1
2368 // If we are running with earlier version, which did not have symbol versions,
2369 // we should use the base version.
2370 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2371 void *f = dlvsym(handle, name, "libnuma_1.1");
2372 if (f == NULL) {
2373 f = dlsym(handle, name);
2374 }
2375 return f;
2376 }
2378 bool os::Linux::libnuma_init() {
2379 // sched_getcpu() should be in libc.
2380 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2381 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2383 if (sched_getcpu() != -1) { // Does it work?
2384 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2385 if (handle != NULL) {
2386 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2387 libnuma_dlsym(handle, "numa_node_to_cpus")));
2388 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2389 libnuma_dlsym(handle, "numa_max_node")));
2390 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2391 libnuma_dlsym(handle, "numa_available")));
2392 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2393 libnuma_dlsym(handle, "numa_tonode_memory")));
2394 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2395 libnuma_dlsym(handle, "numa_interleave_memory")));
2398 if (numa_available() != -1) {
2399 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2400 // Create a cpu -> node mapping
2401 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2402 rebuild_cpu_to_node_map();
2403 return true;
2404 }
2405 }
2406 }
2407 return false;
2408 }
2410 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2411 // The table is later used in get_node_by_cpu().
2412 void os::Linux::rebuild_cpu_to_node_map() {
2413 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2414 // in libnuma (possible values are starting from 16,
2415 // and continuing up with every other power of 2, but less
2416 // than the maximum number of CPUs supported by kernel), and
2417 // is a subject to change (in libnuma version 2 the requirements
2418 // are more reasonable) we'll just hardcode the number they use
2419 // in the library.
2420 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2422 size_t cpu_num = os::active_processor_count();
2423 size_t cpu_map_size = NCPUS / BitsPerCLong;
2424 size_t cpu_map_valid_size =
2425 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2427 cpu_to_node()->clear();
2428 cpu_to_node()->at_grow(cpu_num - 1);
2429 size_t node_num = numa_get_groups_num();
2431 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2432 for (size_t i = 0; i < node_num; i++) {
2433 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2434 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2435 if (cpu_map[j] != 0) {
2436 for (size_t k = 0; k < BitsPerCLong; k++) {
2437 if (cpu_map[j] & (1UL << k)) {
2438 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2439 }
2440 }
2441 }
2442 }
2443 }
2444 }
2445 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2446 }
2448 int os::Linux::get_node_by_cpu(int cpu_id) {
2449 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2450 return cpu_to_node()->at(cpu_id);
2451 }
2452 return -1;
2453 }
2455 GrowableArray<int>* os::Linux::_cpu_to_node;
2456 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2457 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2458 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2459 os::Linux::numa_available_func_t os::Linux::_numa_available;
2460 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2461 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2462 unsigned long* os::Linux::_numa_all_nodes;
2464 bool os::uncommit_memory(char* addr, size_t size) {
2465 return ::mmap(addr, size, PROT_NONE,
2466 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2467 != MAP_FAILED;
2468 }
2470 static address _highest_vm_reserved_address = NULL;
2472 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2473 // at 'requested_addr'. If there are existing memory mappings at the same
2474 // location, however, they will be overwritten. If 'fixed' is false,
2475 // 'requested_addr' is only treated as a hint, the return value may or
2476 // may not start from the requested address. Unlike Linux mmap(), this
2477 // function returns NULL to indicate failure.
2478 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2479 char * addr;
2480 int flags;
2482 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2483 if (fixed) {
2484 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2485 flags |= MAP_FIXED;
2486 }
2488 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2489 // to PROT_EXEC if executable when we commit the page.
2490 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2491 flags, -1, 0);
2493 if (addr != MAP_FAILED) {
2494 // anon_mmap() should only get called during VM initialization,
2495 // don't need lock (actually we can skip locking even it can be called
2496 // from multiple threads, because _highest_vm_reserved_address is just a
2497 // hint about the upper limit of non-stack memory regions.)
2498 if ((address)addr + bytes > _highest_vm_reserved_address) {
2499 _highest_vm_reserved_address = (address)addr + bytes;
2500 }
2501 }
2503 return addr == MAP_FAILED ? NULL : addr;
2504 }
2506 // Don't update _highest_vm_reserved_address, because there might be memory
2507 // regions above addr + size. If so, releasing a memory region only creates
2508 // a hole in the address space, it doesn't help prevent heap-stack collision.
2509 //
2510 static int anon_munmap(char * addr, size_t size) {
2511 return ::munmap(addr, size) == 0;
2512 }
2514 char* os::reserve_memory(size_t bytes, char* requested_addr,
2515 size_t alignment_hint) {
2516 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2517 }
2519 bool os::release_memory(char* addr, size_t size) {
2520 return anon_munmap(addr, size);
2521 }
2523 static address highest_vm_reserved_address() {
2524 return _highest_vm_reserved_address;
2525 }
2527 static bool linux_mprotect(char* addr, size_t size, int prot) {
2528 // Linux wants the mprotect address argument to be page aligned.
2529 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2531 // According to SUSv3, mprotect() should only be used with mappings
2532 // established by mmap(), and mmap() always maps whole pages. Unaligned
2533 // 'addr' likely indicates problem in the VM (e.g. trying to change
2534 // protection of malloc'ed or statically allocated memory). Check the
2535 // caller if you hit this assert.
2536 assert(addr == bottom, "sanity check");
2538 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2539 return ::mprotect(bottom, size, prot) == 0;
2540 }
2542 // Set protections specified
2543 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2544 bool is_committed) {
2545 unsigned int p = 0;
2546 switch (prot) {
2547 case MEM_PROT_NONE: p = PROT_NONE; break;
2548 case MEM_PROT_READ: p = PROT_READ; break;
2549 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2550 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2551 default:
2552 ShouldNotReachHere();
2553 }
2554 // is_committed is unused.
2555 return linux_mprotect(addr, bytes, p);
2556 }
2558 bool os::guard_memory(char* addr, size_t size) {
2559 return linux_mprotect(addr, size, PROT_NONE);
2560 }
2562 bool os::unguard_memory(char* addr, size_t size) {
2563 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2564 }
2566 // Large page support
2568 static size_t _large_page_size = 0;
2570 bool os::large_page_init() {
2571 if (!UseLargePages) return false;
2573 if (LargePageSizeInBytes) {
2574 _large_page_size = LargePageSizeInBytes;
2575 } else {
2576 // large_page_size on Linux is used to round up heap size. x86 uses either
2577 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2578 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2579 // page as large as 256M.
2580 //
2581 // Here we try to figure out page size by parsing /proc/meminfo and looking
2582 // for a line with the following format:
2583 // Hugepagesize: 2048 kB
2584 //
2585 // If we can't determine the value (e.g. /proc is not mounted, or the text
2586 // format has been changed), we'll use the largest page size supported by
2587 // the processor.
2589 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2591 FILE *fp = fopen("/proc/meminfo", "r");
2592 if (fp) {
2593 while (!feof(fp)) {
2594 int x = 0;
2595 char buf[16];
2596 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2597 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2598 _large_page_size = x * K;
2599 break;
2600 }
2601 } else {
2602 // skip to next line
2603 for (;;) {
2604 int ch = fgetc(fp);
2605 if (ch == EOF || ch == (int)'\n') break;
2606 }
2607 }
2608 }
2609 fclose(fp);
2610 }
2611 }
2613 const size_t default_page_size = (size_t)Linux::page_size();
2614 if (_large_page_size > default_page_size) {
2615 _page_sizes[0] = _large_page_size;
2616 _page_sizes[1] = default_page_size;
2617 _page_sizes[2] = 0;
2618 }
2620 // Large page support is available on 2.6 or newer kernel, some vendors
2621 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2622 // We optimistically assume the support is available. If later it turns out
2623 // not true, VM will automatically switch to use regular page size.
2624 return true;
2625 }
2627 #ifndef SHM_HUGETLB
2628 #define SHM_HUGETLB 04000
2629 #endif
2631 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2632 // "exec" is passed in but not used. Creating the shared image for
2633 // the code cache doesn't have an SHM_X executable permission to check.
2634 assert(UseLargePages, "only for large pages");
2636 key_t key = IPC_PRIVATE;
2637 char *addr;
2639 bool warn_on_failure = UseLargePages &&
2640 (!FLAG_IS_DEFAULT(UseLargePages) ||
2641 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2642 );
2643 char msg[128];
2645 // Create a large shared memory region to attach to based on size.
2646 // Currently, size is the total size of the heap
2647 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2648 if (shmid == -1) {
2649 // Possible reasons for shmget failure:
2650 // 1. shmmax is too small for Java heap.
2651 // > check shmmax value: cat /proc/sys/kernel/shmmax
2652 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2653 // 2. not enough large page memory.
2654 // > check available large pages: cat /proc/meminfo
2655 // > increase amount of large pages:
2656 // echo new_value > /proc/sys/vm/nr_hugepages
2657 // Note 1: different Linux may use different name for this property,
2658 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2659 // Note 2: it's possible there's enough physical memory available but
2660 // they are so fragmented after a long run that they can't
2661 // coalesce into large pages. Try to reserve large pages when
2662 // the system is still "fresh".
2663 if (warn_on_failure) {
2664 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2665 warning(msg);
2666 }
2667 return NULL;
2668 }
2670 // attach to the region
2671 addr = (char*)shmat(shmid, NULL, 0);
2672 int err = errno;
2674 // Remove shmid. If shmat() is successful, the actual shared memory segment
2675 // will be deleted when it's detached by shmdt() or when the process
2676 // terminates. If shmat() is not successful this will remove the shared
2677 // segment immediately.
2678 shmctl(shmid, IPC_RMID, NULL);
2680 if ((intptr_t)addr == -1) {
2681 if (warn_on_failure) {
2682 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2683 warning(msg);
2684 }
2685 return NULL;
2686 }
2688 return addr;
2689 }
2691 bool os::release_memory_special(char* base, size_t bytes) {
2692 // detaching the SHM segment will also delete it, see reserve_memory_special()
2693 int rslt = shmdt(base);
2694 return rslt == 0;
2695 }
2697 size_t os::large_page_size() {
2698 return _large_page_size;
2699 }
2701 // Linux does not support anonymous mmap with large page memory. The only way
2702 // to reserve large page memory without file backing is through SysV shared
2703 // memory API. The entire memory region is committed and pinned upfront.
2704 // Hopefully this will change in the future...
2705 bool os::can_commit_large_page_memory() {
2706 return false;
2707 }
2709 bool os::can_execute_large_page_memory() {
2710 return false;
2711 }
2713 // Reserve memory at an arbitrary address, only if that area is
2714 // available (and not reserved for something else).
2716 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2717 const int max_tries = 10;
2718 char* base[max_tries];
2719 size_t size[max_tries];
2720 const size_t gap = 0x000000;
2722 // Assert only that the size is a multiple of the page size, since
2723 // that's all that mmap requires, and since that's all we really know
2724 // about at this low abstraction level. If we need higher alignment,
2725 // we can either pass an alignment to this method or verify alignment
2726 // in one of the methods further up the call chain. See bug 5044738.
2727 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2729 // Repeatedly allocate blocks until the block is allocated at the
2730 // right spot. Give up after max_tries. Note that reserve_memory() will
2731 // automatically update _highest_vm_reserved_address if the call is
2732 // successful. The variable tracks the highest memory address every reserved
2733 // by JVM. It is used to detect heap-stack collision if running with
2734 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2735 // space than needed, it could confuse the collision detecting code. To
2736 // solve the problem, save current _highest_vm_reserved_address and
2737 // calculate the correct value before return.
2738 address old_highest = _highest_vm_reserved_address;
2740 // Linux mmap allows caller to pass an address as hint; give it a try first,
2741 // if kernel honors the hint then we can return immediately.
2742 char * addr = anon_mmap(requested_addr, bytes, false);
2743 if (addr == requested_addr) {
2744 return requested_addr;
2745 }
2747 if (addr != NULL) {
2748 // mmap() is successful but it fails to reserve at the requested address
2749 anon_munmap(addr, bytes);
2750 }
2752 int i;
2753 for (i = 0; i < max_tries; ++i) {
2754 base[i] = reserve_memory(bytes);
2756 if (base[i] != NULL) {
2757 // Is this the block we wanted?
2758 if (base[i] == requested_addr) {
2759 size[i] = bytes;
2760 break;
2761 }
2763 // Does this overlap the block we wanted? Give back the overlapped
2764 // parts and try again.
2766 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2767 if (top_overlap >= 0 && top_overlap < bytes) {
2768 unmap_memory(base[i], top_overlap);
2769 base[i] += top_overlap;
2770 size[i] = bytes - top_overlap;
2771 } else {
2772 size_t bottom_overlap = base[i] + bytes - requested_addr;
2773 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2774 unmap_memory(requested_addr, bottom_overlap);
2775 size[i] = bytes - bottom_overlap;
2776 } else {
2777 size[i] = bytes;
2778 }
2779 }
2780 }
2781 }
2783 // Give back the unused reserved pieces.
2785 for (int j = 0; j < i; ++j) {
2786 if (base[j] != NULL) {
2787 unmap_memory(base[j], size[j]);
2788 }
2789 }
2791 if (i < max_tries) {
2792 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2793 return requested_addr;
2794 } else {
2795 _highest_vm_reserved_address = old_highest;
2796 return NULL;
2797 }
2798 }
2800 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2801 return ::read(fd, buf, nBytes);
2802 }
2804 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2805 // Solaris uses poll(), linux uses park().
2806 // Poll() is likely a better choice, assuming that Thread.interrupt()
2807 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2808 // SIGSEGV, see 4355769.
2810 const int NANOSECS_PER_MILLISECS = 1000000;
2812 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2813 assert(thread == Thread::current(), "thread consistency check");
2815 ParkEvent * const slp = thread->_SleepEvent ;
2816 slp->reset() ;
2817 OrderAccess::fence() ;
2819 if (interruptible) {
2820 jlong prevtime = javaTimeNanos();
2822 for (;;) {
2823 if (os::is_interrupted(thread, true)) {
2824 return OS_INTRPT;
2825 }
2827 jlong newtime = javaTimeNanos();
2829 if (newtime - prevtime < 0) {
2830 // time moving backwards, should only happen if no monotonic clock
2831 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2832 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2833 } else {
2834 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2835 }
2837 if(millis <= 0) {
2838 return OS_OK;
2839 }
2841 prevtime = newtime;
2843 {
2844 assert(thread->is_Java_thread(), "sanity check");
2845 JavaThread *jt = (JavaThread *) thread;
2846 ThreadBlockInVM tbivm(jt);
2847 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2849 jt->set_suspend_equivalent();
2850 // cleared by handle_special_suspend_equivalent_condition() or
2851 // java_suspend_self() via check_and_wait_while_suspended()
2853 slp->park(millis);
2855 // were we externally suspended while we were waiting?
2856 jt->check_and_wait_while_suspended();
2857 }
2858 }
2859 } else {
2860 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2861 jlong prevtime = javaTimeNanos();
2863 for (;;) {
2864 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2865 // the 1st iteration ...
2866 jlong newtime = javaTimeNanos();
2868 if (newtime - prevtime < 0) {
2869 // time moving backwards, should only happen if no monotonic clock
2870 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2871 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2872 } else {
2873 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2874 }
2876 if(millis <= 0) break ;
2878 prevtime = newtime;
2879 slp->park(millis);
2880 }
2881 return OS_OK ;
2882 }
2883 }
2885 int os::naked_sleep() {
2886 // %% make the sleep time an integer flag. for now use 1 millisec.
2887 return os::sleep(Thread::current(), 1, false);
2888 }
2890 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2891 void os::infinite_sleep() {
2892 while (true) { // sleep forever ...
2893 ::sleep(100); // ... 100 seconds at a time
2894 }
2895 }
2897 // Used to convert frequent JVM_Yield() to nops
2898 bool os::dont_yield() {
2899 return DontYieldALot;
2900 }
2902 void os::yield() {
2903 sched_yield();
2904 }
2906 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2908 void os::yield_all(int attempts) {
2909 // Yields to all threads, including threads with lower priorities
2910 // Threads on Linux are all with same priority. The Solaris style
2911 // os::yield_all() with nanosleep(1ms) is not necessary.
2912 sched_yield();
2913 }
2915 // Called from the tight loops to possibly influence time-sharing heuristics
2916 void os::loop_breaker(int attempts) {
2917 os::yield_all(attempts);
2918 }
2920 ////////////////////////////////////////////////////////////////////////////////
2921 // thread priority support
2923 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2924 // only supports dynamic priority, static priority must be zero. For real-time
2925 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2926 // However, for large multi-threaded applications, SCHED_RR is not only slower
2927 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2928 // of 5 runs - Sep 2005).
2929 //
2930 // The following code actually changes the niceness of kernel-thread/LWP. It
2931 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2932 // not the entire user process, and user level threads are 1:1 mapped to kernel
2933 // threads. It has always been the case, but could change in the future. For
2934 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2935 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2937 int os::java_to_os_priority[MaxPriority + 1] = {
2938 19, // 0 Entry should never be used
2940 4, // 1 MinPriority
2941 3, // 2
2942 2, // 3
2944 1, // 4
2945 0, // 5 NormPriority
2946 -1, // 6
2948 -2, // 7
2949 -3, // 8
2950 -4, // 9 NearMaxPriority
2952 -5 // 10 MaxPriority
2953 };
2955 static int prio_init() {
2956 if (ThreadPriorityPolicy == 1) {
2957 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2958 // if effective uid is not root. Perhaps, a more elegant way of doing
2959 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2960 if (geteuid() != 0) {
2961 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2962 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2963 }
2964 ThreadPriorityPolicy = 0;
2965 }
2966 }
2967 return 0;
2968 }
2970 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2971 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2973 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2974 return (ret == 0) ? OS_OK : OS_ERR;
2975 }
2977 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2978 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2979 *priority_ptr = java_to_os_priority[NormPriority];
2980 return OS_OK;
2981 }
2983 errno = 0;
2984 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2985 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2986 }
2988 // Hint to the underlying OS that a task switch would not be good.
2989 // Void return because it's a hint and can fail.
2990 void os::hint_no_preempt() {}
2992 ////////////////////////////////////////////////////////////////////////////////
2993 // suspend/resume support
2995 // the low-level signal-based suspend/resume support is a remnant from the
2996 // old VM-suspension that used to be for java-suspension, safepoints etc,
2997 // within hotspot. Now there is a single use-case for this:
2998 // - calling get_thread_pc() on the VMThread by the flat-profiler task
2999 // that runs in the watcher thread.
3000 // The remaining code is greatly simplified from the more general suspension
3001 // code that used to be used.
3002 //
3003 // The protocol is quite simple:
3004 // - suspend:
3005 // - sends a signal to the target thread
3006 // - polls the suspend state of the osthread using a yield loop
3007 // - target thread signal handler (SR_handler) sets suspend state
3008 // and blocks in sigsuspend until continued
3009 // - resume:
3010 // - sets target osthread state to continue
3011 // - sends signal to end the sigsuspend loop in the SR_handler
3012 //
3013 // Note that the SR_lock plays no role in this suspend/resume protocol.
3014 //
3016 static void resume_clear_context(OSThread *osthread) {
3017 osthread->set_ucontext(NULL);
3018 osthread->set_siginfo(NULL);
3020 // notify the suspend action is completed, we have now resumed
3021 osthread->sr.clear_suspended();
3022 }
3024 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3025 osthread->set_ucontext(context);
3026 osthread->set_siginfo(siginfo);
3027 }
3029 //
3030 // Handler function invoked when a thread's execution is suspended or
3031 // resumed. We have to be careful that only async-safe functions are
3032 // called here (Note: most pthread functions are not async safe and
3033 // should be avoided.)
3034 //
3035 // Note: sigwait() is a more natural fit than sigsuspend() from an
3036 // interface point of view, but sigwait() prevents the signal hander
3037 // from being run. libpthread would get very confused by not having
3038 // its signal handlers run and prevents sigwait()'s use with the
3039 // mutex granting granting signal.
3040 //
3041 // Currently only ever called on the VMThread
3042 //
3043 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3044 // Save and restore errno to avoid confusing native code with EINTR
3045 // after sigsuspend.
3046 int old_errno = errno;
3048 Thread* thread = Thread::current();
3049 OSThread* osthread = thread->osthread();
3050 assert(thread->is_VM_thread(), "Must be VMThread");
3051 // read current suspend action
3052 int action = osthread->sr.suspend_action();
3053 if (action == SR_SUSPEND) {
3054 suspend_save_context(osthread, siginfo, context);
3056 // Notify the suspend action is about to be completed. do_suspend()
3057 // waits until SR_SUSPENDED is set and then returns. We will wait
3058 // here for a resume signal and that completes the suspend-other
3059 // action. do_suspend/do_resume is always called as a pair from
3060 // the same thread - so there are no races
3062 // notify the caller
3063 osthread->sr.set_suspended();
3065 sigset_t suspend_set; // signals for sigsuspend()
3067 // get current set of blocked signals and unblock resume signal
3068 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3069 sigdelset(&suspend_set, SR_signum);
3071 // wait here until we are resumed
3072 do {
3073 sigsuspend(&suspend_set);
3074 // ignore all returns until we get a resume signal
3075 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3077 resume_clear_context(osthread);
3079 } else {
3080 assert(action == SR_CONTINUE, "unexpected sr action");
3081 // nothing special to do - just leave the handler
3082 }
3084 errno = old_errno;
3085 }
3088 static int SR_initialize() {
3089 struct sigaction act;
3090 char *s;
3091 /* Get signal number to use for suspend/resume */
3092 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3093 int sig = ::strtol(s, 0, 10);
3094 if (sig > 0 || sig < _NSIG) {
3095 SR_signum = sig;
3096 }
3097 }
3099 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3100 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3102 sigemptyset(&SR_sigset);
3103 sigaddset(&SR_sigset, SR_signum);
3105 /* Set up signal handler for suspend/resume */
3106 act.sa_flags = SA_RESTART|SA_SIGINFO;
3107 act.sa_handler = (void (*)(int)) SR_handler;
3109 // SR_signum is blocked by default.
3110 // 4528190 - We also need to block pthread restart signal (32 on all
3111 // supported Linux platforms). Note that LinuxThreads need to block
3112 // this signal for all threads to work properly. So we don't have
3113 // to use hard-coded signal number when setting up the mask.
3114 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3116 if (sigaction(SR_signum, &act, 0) == -1) {
3117 return -1;
3118 }
3120 // Save signal flag
3121 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3122 return 0;
3123 }
3125 static int SR_finalize() {
3126 return 0;
3127 }
3130 // returns true on success and false on error - really an error is fatal
3131 // but this seems the normal response to library errors
3132 static bool do_suspend(OSThread* osthread) {
3133 // mark as suspended and send signal
3134 osthread->sr.set_suspend_action(SR_SUSPEND);
3135 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3136 assert_status(status == 0, status, "pthread_kill");
3138 // check status and wait until notified of suspension
3139 if (status == 0) {
3140 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3141 os::yield_all(i);
3142 }
3143 osthread->sr.set_suspend_action(SR_NONE);
3144 return true;
3145 }
3146 else {
3147 osthread->sr.set_suspend_action(SR_NONE);
3148 return false;
3149 }
3150 }
3152 static void do_resume(OSThread* osthread) {
3153 assert(osthread->sr.is_suspended(), "thread should be suspended");
3154 osthread->sr.set_suspend_action(SR_CONTINUE);
3156 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3157 assert_status(status == 0, status, "pthread_kill");
3158 // check status and wait unit notified of resumption
3159 if (status == 0) {
3160 for (int i = 0; osthread->sr.is_suspended(); i++) {
3161 os::yield_all(i);
3162 }
3163 }
3164 osthread->sr.set_suspend_action(SR_NONE);
3165 }
3167 ////////////////////////////////////////////////////////////////////////////////
3168 // interrupt support
3170 void os::interrupt(Thread* thread) {
3171 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3172 "possibility of dangling Thread pointer");
3174 OSThread* osthread = thread->osthread();
3176 if (!osthread->interrupted()) {
3177 osthread->set_interrupted(true);
3178 // More than one thread can get here with the same value of osthread,
3179 // resulting in multiple notifications. We do, however, want the store
3180 // to interrupted() to be visible to other threads before we execute unpark().
3181 OrderAccess::fence();
3182 ParkEvent * const slp = thread->_SleepEvent ;
3183 if (slp != NULL) slp->unpark() ;
3184 }
3186 // For JSR166. Unpark even if interrupt status already was set
3187 if (thread->is_Java_thread())
3188 ((JavaThread*)thread)->parker()->unpark();
3190 ParkEvent * ev = thread->_ParkEvent ;
3191 if (ev != NULL) ev->unpark() ;
3193 }
3195 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3196 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3197 "possibility of dangling Thread pointer");
3199 OSThread* osthread = thread->osthread();
3201 bool interrupted = osthread->interrupted();
3203 if (interrupted && clear_interrupted) {
3204 osthread->set_interrupted(false);
3205 // consider thread->_SleepEvent->reset() ... optional optimization
3206 }
3208 return interrupted;
3209 }
3211 ///////////////////////////////////////////////////////////////////////////////////
3212 // signal handling (except suspend/resume)
3214 // This routine may be used by user applications as a "hook" to catch signals.
3215 // The user-defined signal handler must pass unrecognized signals to this
3216 // routine, and if it returns true (non-zero), then the signal handler must
3217 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3218 // routine will never retun false (zero), but instead will execute a VM panic
3219 // routine kill the process.
3220 //
3221 // If this routine returns false, it is OK to call it again. This allows
3222 // the user-defined signal handler to perform checks either before or after
3223 // the VM performs its own checks. Naturally, the user code would be making
3224 // a serious error if it tried to handle an exception (such as a null check
3225 // or breakpoint) that the VM was generating for its own correct operation.
3226 //
3227 // This routine may recognize any of the following kinds of signals:
3228 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3229 // It should be consulted by handlers for any of those signals.
3230 //
3231 // The caller of this routine must pass in the three arguments supplied
3232 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3233 // field of the structure passed to sigaction(). This routine assumes that
3234 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3235 //
3236 // Note that the VM will print warnings if it detects conflicting signal
3237 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3238 //
3239 extern "C" int
3240 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3241 void* ucontext, int abort_if_unrecognized);
3243 void signalHandler(int sig, siginfo_t* info, void* uc) {
3244 assert(info != NULL && uc != NULL, "it must be old kernel");
3245 JVM_handle_linux_signal(sig, info, uc, true);
3246 }
3249 // This boolean allows users to forward their own non-matching signals
3250 // to JVM_handle_linux_signal, harmlessly.
3251 bool os::Linux::signal_handlers_are_installed = false;
3253 // For signal-chaining
3254 struct sigaction os::Linux::sigact[MAXSIGNUM];
3255 unsigned int os::Linux::sigs = 0;
3256 bool os::Linux::libjsig_is_loaded = false;
3257 typedef struct sigaction *(*get_signal_t)(int);
3258 get_signal_t os::Linux::get_signal_action = NULL;
3260 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3261 struct sigaction *actp = NULL;
3263 if (libjsig_is_loaded) {
3264 // Retrieve the old signal handler from libjsig
3265 actp = (*get_signal_action)(sig);
3266 }
3267 if (actp == NULL) {
3268 // Retrieve the preinstalled signal handler from jvm
3269 actp = get_preinstalled_handler(sig);
3270 }
3272 return actp;
3273 }
3275 static bool call_chained_handler(struct sigaction *actp, int sig,
3276 siginfo_t *siginfo, void *context) {
3277 // Call the old signal handler
3278 if (actp->sa_handler == SIG_DFL) {
3279 // It's more reasonable to let jvm treat it as an unexpected exception
3280 // instead of taking the default action.
3281 return false;
3282 } else if (actp->sa_handler != SIG_IGN) {
3283 if ((actp->sa_flags & SA_NODEFER) == 0) {
3284 // automaticlly block the signal
3285 sigaddset(&(actp->sa_mask), sig);
3286 }
3288 sa_handler_t hand;
3289 sa_sigaction_t sa;
3290 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3291 // retrieve the chained handler
3292 if (siginfo_flag_set) {
3293 sa = actp->sa_sigaction;
3294 } else {
3295 hand = actp->sa_handler;
3296 }
3298 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3299 actp->sa_handler = SIG_DFL;
3300 }
3302 // try to honor the signal mask
3303 sigset_t oset;
3304 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3306 // call into the chained handler
3307 if (siginfo_flag_set) {
3308 (*sa)(sig, siginfo, context);
3309 } else {
3310 (*hand)(sig);
3311 }
3313 // restore the signal mask
3314 pthread_sigmask(SIG_SETMASK, &oset, 0);
3315 }
3316 // Tell jvm's signal handler the signal is taken care of.
3317 return true;
3318 }
3320 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3321 bool chained = false;
3322 // signal-chaining
3323 if (UseSignalChaining) {
3324 struct sigaction *actp = get_chained_signal_action(sig);
3325 if (actp != NULL) {
3326 chained = call_chained_handler(actp, sig, siginfo, context);
3327 }
3328 }
3329 return chained;
3330 }
3332 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3333 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3334 return &sigact[sig];
3335 }
3336 return NULL;
3337 }
3339 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3340 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3341 sigact[sig] = oldAct;
3342 sigs |= (unsigned int)1 << sig;
3343 }
3345 // for diagnostic
3346 int os::Linux::sigflags[MAXSIGNUM];
3348 int os::Linux::get_our_sigflags(int sig) {
3349 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3350 return sigflags[sig];
3351 }
3353 void os::Linux::set_our_sigflags(int sig, int flags) {
3354 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3355 sigflags[sig] = flags;
3356 }
3358 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3359 // Check for overwrite.
3360 struct sigaction oldAct;
3361 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3363 void* oldhand = oldAct.sa_sigaction
3364 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3365 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3366 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3367 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3368 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3369 if (AllowUserSignalHandlers || !set_installed) {
3370 // Do not overwrite; user takes responsibility to forward to us.
3371 return;
3372 } else if (UseSignalChaining) {
3373 // save the old handler in jvm
3374 save_preinstalled_handler(sig, oldAct);
3375 // libjsig also interposes the sigaction() call below and saves the
3376 // old sigaction on it own.
3377 } else {
3378 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3379 }
3380 }
3382 struct sigaction sigAct;
3383 sigfillset(&(sigAct.sa_mask));
3384 sigAct.sa_handler = SIG_DFL;
3385 if (!set_installed) {
3386 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3387 } else {
3388 sigAct.sa_sigaction = signalHandler;
3389 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3390 }
3391 // Save flags, which are set by ours
3392 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3393 sigflags[sig] = sigAct.sa_flags;
3395 int ret = sigaction(sig, &sigAct, &oldAct);
3396 assert(ret == 0, "check");
3398 void* oldhand2 = oldAct.sa_sigaction
3399 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3400 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3401 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3402 }
3404 // install signal handlers for signals that HotSpot needs to
3405 // handle in order to support Java-level exception handling.
3407 void os::Linux::install_signal_handlers() {
3408 if (!signal_handlers_are_installed) {
3409 signal_handlers_are_installed = true;
3411 // signal-chaining
3412 typedef void (*signal_setting_t)();
3413 signal_setting_t begin_signal_setting = NULL;
3414 signal_setting_t end_signal_setting = NULL;
3415 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3416 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3417 if (begin_signal_setting != NULL) {
3418 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3419 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3420 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3421 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3422 libjsig_is_loaded = true;
3423 assert(UseSignalChaining, "should enable signal-chaining");
3424 }
3425 if (libjsig_is_loaded) {
3426 // Tell libjsig jvm is setting signal handlers
3427 (*begin_signal_setting)();
3428 }
3430 set_signal_handler(SIGSEGV, true);
3431 set_signal_handler(SIGPIPE, true);
3432 set_signal_handler(SIGBUS, true);
3433 set_signal_handler(SIGILL, true);
3434 set_signal_handler(SIGFPE, true);
3435 set_signal_handler(SIGXFSZ, true);
3437 if (libjsig_is_loaded) {
3438 // Tell libjsig jvm finishes setting signal handlers
3439 (*end_signal_setting)();
3440 }
3442 // We don't activate signal checker if libjsig is in place, we trust ourselves
3443 // and if UserSignalHandler is installed all bets are off
3444 if (CheckJNICalls) {
3445 if (libjsig_is_loaded) {
3446 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3447 check_signals = false;
3448 }
3449 if (AllowUserSignalHandlers) {
3450 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3451 check_signals = false;
3452 }
3453 }
3454 }
3455 }
3457 // This is the fastest way to get thread cpu time on Linux.
3458 // Returns cpu time (user+sys) for any thread, not only for current.
3459 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3460 // It might work on 2.6.10+ with a special kernel/glibc patch.
3461 // For reference, please, see IEEE Std 1003.1-2004:
3462 // http://www.unix.org/single_unix_specification
3464 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3465 struct timespec tp;
3466 int rc = os::Linux::clock_gettime(clockid, &tp);
3467 assert(rc == 0, "clock_gettime is expected to return 0 code");
3469 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3470 }
3472 /////
3473 // glibc on Linux platform uses non-documented flag
3474 // to indicate, that some special sort of signal
3475 // trampoline is used.
3476 // We will never set this flag, and we should
3477 // ignore this flag in our diagnostic
3478 #ifdef SIGNIFICANT_SIGNAL_MASK
3479 #undef SIGNIFICANT_SIGNAL_MASK
3480 #endif
3481 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3483 static const char* get_signal_handler_name(address handler,
3484 char* buf, int buflen) {
3485 int offset;
3486 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3487 if (found) {
3488 // skip directory names
3489 const char *p1, *p2;
3490 p1 = buf;
3491 size_t len = strlen(os::file_separator());
3492 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3493 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3494 } else {
3495 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3496 }
3497 return buf;
3498 }
3500 static void print_signal_handler(outputStream* st, int sig,
3501 char* buf, size_t buflen) {
3502 struct sigaction sa;
3504 sigaction(sig, NULL, &sa);
3506 // See comment for SIGNIFICANT_SIGNAL_MASK define
3507 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3509 st->print("%s: ", os::exception_name(sig, buf, buflen));
3511 address handler = (sa.sa_flags & SA_SIGINFO)
3512 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3513 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3515 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3516 st->print("SIG_DFL");
3517 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3518 st->print("SIG_IGN");
3519 } else {
3520 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3521 }
3523 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3525 address rh = VMError::get_resetted_sighandler(sig);
3526 // May be, handler was resetted by VMError?
3527 if(rh != NULL) {
3528 handler = rh;
3529 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3530 }
3532 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3534 // Check: is it our handler?
3535 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3536 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3537 // It is our signal handler
3538 // check for flags, reset system-used one!
3539 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3540 st->print(
3541 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3542 os::Linux::get_our_sigflags(sig));
3543 }
3544 }
3545 st->cr();
3546 }
3549 #define DO_SIGNAL_CHECK(sig) \
3550 if (!sigismember(&check_signal_done, sig)) \
3551 os::Linux::check_signal_handler(sig)
3553 // This method is a periodic task to check for misbehaving JNI applications
3554 // under CheckJNI, we can add any periodic checks here
3556 void os::run_periodic_checks() {
3558 if (check_signals == false) return;
3560 // SEGV and BUS if overridden could potentially prevent
3561 // generation of hs*.log in the event of a crash, debugging
3562 // such a case can be very challenging, so we absolutely
3563 // check the following for a good measure:
3564 DO_SIGNAL_CHECK(SIGSEGV);
3565 DO_SIGNAL_CHECK(SIGILL);
3566 DO_SIGNAL_CHECK(SIGFPE);
3567 DO_SIGNAL_CHECK(SIGBUS);
3568 DO_SIGNAL_CHECK(SIGPIPE);
3569 DO_SIGNAL_CHECK(SIGXFSZ);
3572 // ReduceSignalUsage allows the user to override these handlers
3573 // see comments at the very top and jvm_solaris.h
3574 if (!ReduceSignalUsage) {
3575 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3576 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3577 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3578 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3579 }
3581 DO_SIGNAL_CHECK(SR_signum);
3582 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3583 }
3585 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3587 static os_sigaction_t os_sigaction = NULL;
3589 void os::Linux::check_signal_handler(int sig) {
3590 char buf[O_BUFLEN];
3591 address jvmHandler = NULL;
3594 struct sigaction act;
3595 if (os_sigaction == NULL) {
3596 // only trust the default sigaction, in case it has been interposed
3597 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3598 if (os_sigaction == NULL) return;
3599 }
3601 os_sigaction(sig, (struct sigaction*)NULL, &act);
3604 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3606 address thisHandler = (act.sa_flags & SA_SIGINFO)
3607 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3608 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3611 switch(sig) {
3612 case SIGSEGV:
3613 case SIGBUS:
3614 case SIGFPE:
3615 case SIGPIPE:
3616 case SIGILL:
3617 case SIGXFSZ:
3618 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3619 break;
3621 case SHUTDOWN1_SIGNAL:
3622 case SHUTDOWN2_SIGNAL:
3623 case SHUTDOWN3_SIGNAL:
3624 case BREAK_SIGNAL:
3625 jvmHandler = (address)user_handler();
3626 break;
3628 case INTERRUPT_SIGNAL:
3629 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3630 break;
3632 default:
3633 if (sig == SR_signum) {
3634 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3635 } else {
3636 return;
3637 }
3638 break;
3639 }
3641 if (thisHandler != jvmHandler) {
3642 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3643 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3644 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3645 // No need to check this sig any longer
3646 sigaddset(&check_signal_done, sig);
3647 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3648 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3649 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3650 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3651 // No need to check this sig any longer
3652 sigaddset(&check_signal_done, sig);
3653 }
3655 // Dump all the signal
3656 if (sigismember(&check_signal_done, sig)) {
3657 print_signal_handlers(tty, buf, O_BUFLEN);
3658 }
3659 }
3661 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3663 extern bool signal_name(int signo, char* buf, size_t len);
3665 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3666 if (0 < exception_code && exception_code <= SIGRTMAX) {
3667 // signal
3668 if (!signal_name(exception_code, buf, size)) {
3669 jio_snprintf(buf, size, "SIG%d", exception_code);
3670 }
3671 return buf;
3672 } else {
3673 return NULL;
3674 }
3675 }
3677 // this is called _before_ the most of global arguments have been parsed
3678 void os::init(void) {
3679 char dummy; /* used to get a guess on initial stack address */
3680 // first_hrtime = gethrtime();
3682 // With LinuxThreads the JavaMain thread pid (primordial thread)
3683 // is different than the pid of the java launcher thread.
3684 // So, on Linux, the launcher thread pid is passed to the VM
3685 // via the sun.java.launcher.pid property.
3686 // Use this property instead of getpid() if it was correctly passed.
3687 // See bug 6351349.
3688 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3690 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3692 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3694 init_random(1234567);
3696 ThreadCritical::initialize();
3698 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3699 if (Linux::page_size() == -1) {
3700 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3701 }
3702 init_page_sizes((size_t) Linux::page_size());
3704 Linux::initialize_system_info();
3706 // main_thread points to the aboriginal thread
3707 Linux::_main_thread = pthread_self();
3709 Linux::clock_init();
3710 initial_time_count = os::elapsed_counter();
3711 pthread_mutex_init(&dl_mutex, NULL);
3712 }
3714 // To install functions for atexit system call
3715 extern "C" {
3716 static void perfMemory_exit_helper() {
3717 perfMemory_exit();
3718 }
3719 }
3721 // this is called _after_ the global arguments have been parsed
3722 jint os::init_2(void)
3723 {
3724 Linux::fast_thread_clock_init();
3726 // Allocate a single page and mark it as readable for safepoint polling
3727 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3728 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3730 os::set_polling_page( polling_page );
3732 #ifndef PRODUCT
3733 if(Verbose && PrintMiscellaneous)
3734 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3735 #endif
3737 if (!UseMembar) {
3738 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3739 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3740 os::set_memory_serialize_page( mem_serialize_page );
3742 #ifndef PRODUCT
3743 if(Verbose && PrintMiscellaneous)
3744 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3745 #endif
3746 }
3748 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3750 // initialize suspend/resume support - must do this before signal_sets_init()
3751 if (SR_initialize() != 0) {
3752 perror("SR_initialize failed");
3753 return JNI_ERR;
3754 }
3756 Linux::signal_sets_init();
3757 Linux::install_signal_handlers();
3759 size_t threadStackSizeInBytes = ThreadStackSize * K;
3760 if (threadStackSizeInBytes != 0 &&
3761 threadStackSizeInBytes < Linux::min_stack_allowed) {
3762 tty->print_cr("\nThe stack size specified is too small, "
3763 "Specify at least %dk",
3764 Linux::min_stack_allowed / K);
3765 return JNI_ERR;
3766 }
3768 // Make the stack size a multiple of the page size so that
3769 // the yellow/red zones can be guarded.
3770 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3771 vm_page_size()));
3773 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3775 Linux::libpthread_init();
3776 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3777 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3778 Linux::glibc_version(), Linux::libpthread_version(),
3779 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3780 }
3782 if (UseNUMA) {
3783 if (!Linux::libnuma_init()) {
3784 UseNUMA = false;
3785 } else {
3786 if ((Linux::numa_max_node() < 1)) {
3787 // There's only one node(they start from 0), disable NUMA.
3788 UseNUMA = false;
3789 }
3790 }
3791 if (!UseNUMA && ForceNUMA) {
3792 UseNUMA = true;
3793 }
3794 }
3796 if (MaxFDLimit) {
3797 // set the number of file descriptors to max. print out error
3798 // if getrlimit/setrlimit fails but continue regardless.
3799 struct rlimit nbr_files;
3800 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3801 if (status != 0) {
3802 if (PrintMiscellaneous && (Verbose || WizardMode))
3803 perror("os::init_2 getrlimit failed");
3804 } else {
3805 nbr_files.rlim_cur = nbr_files.rlim_max;
3806 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3807 if (status != 0) {
3808 if (PrintMiscellaneous && (Verbose || WizardMode))
3809 perror("os::init_2 setrlimit failed");
3810 }
3811 }
3812 }
3814 // Initialize lock used to serialize thread creation (see os::create_thread)
3815 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3817 // Initialize HPI.
3818 jint hpi_result = hpi::initialize();
3819 if (hpi_result != JNI_OK) {
3820 tty->print_cr("There was an error trying to initialize the HPI library.");
3821 return hpi_result;
3822 }
3824 // at-exit methods are called in the reverse order of their registration.
3825 // atexit functions are called on return from main or as a result of a
3826 // call to exit(3C). There can be only 32 of these functions registered
3827 // and atexit() does not set errno.
3829 if (PerfAllowAtExitRegistration) {
3830 // only register atexit functions if PerfAllowAtExitRegistration is set.
3831 // atexit functions can be delayed until process exit time, which
3832 // can be problematic for embedded VM situations. Embedded VMs should
3833 // call DestroyJavaVM() to assure that VM resources are released.
3835 // note: perfMemory_exit_helper atexit function may be removed in
3836 // the future if the appropriate cleanup code can be added to the
3837 // VM_Exit VMOperation's doit method.
3838 if (atexit(perfMemory_exit_helper) != 0) {
3839 warning("os::init2 atexit(perfMemory_exit_helper) failed");
3840 }
3841 }
3843 // initialize thread priority policy
3844 prio_init();
3846 return JNI_OK;
3847 }
3849 // Mark the polling page as unreadable
3850 void os::make_polling_page_unreadable(void) {
3851 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3852 fatal("Could not disable polling page");
3853 };
3855 // Mark the polling page as readable
3856 void os::make_polling_page_readable(void) {
3857 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3858 fatal("Could not enable polling page");
3859 }
3860 };
3862 int os::active_processor_count() {
3863 // Linux doesn't yet have a (official) notion of processor sets,
3864 // so just return the number of online processors.
3865 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3866 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3867 return online_cpus;
3868 }
3870 bool os::distribute_processes(uint length, uint* distribution) {
3871 // Not yet implemented.
3872 return false;
3873 }
3875 bool os::bind_to_processor(uint processor_id) {
3876 // Not yet implemented.
3877 return false;
3878 }
3880 ///
3882 // Suspends the target using the signal mechanism and then grabs the PC before
3883 // resuming the target. Used by the flat-profiler only
3884 ExtendedPC os::get_thread_pc(Thread* thread) {
3885 // Make sure that it is called by the watcher for the VMThread
3886 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3887 assert(thread->is_VM_thread(), "Can only be called for VMThread");
3889 ExtendedPC epc;
3891 OSThread* osthread = thread->osthread();
3892 if (do_suspend(osthread)) {
3893 if (osthread->ucontext() != NULL) {
3894 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3895 } else {
3896 // NULL context is unexpected, double-check this is the VMThread
3897 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3898 }
3899 do_resume(osthread);
3900 }
3901 // failure means pthread_kill failed for some reason - arguably this is
3902 // a fatal problem, but such problems are ignored elsewhere
3904 return epc;
3905 }
3907 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3908 {
3909 if (is_NPTL()) {
3910 return pthread_cond_timedwait(_cond, _mutex, _abstime);
3911 } else {
3912 #ifndef IA64
3913 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3914 // word back to default 64bit precision if condvar is signaled. Java
3915 // wants 53bit precision. Save and restore current value.
3916 int fpu = get_fpu_control_word();
3917 #endif // IA64
3918 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3919 #ifndef IA64
3920 set_fpu_control_word(fpu);
3921 #endif // IA64
3922 return status;
3923 }
3924 }
3926 ////////////////////////////////////////////////////////////////////////////////
3927 // debug support
3929 #ifndef PRODUCT
3930 static address same_page(address x, address y) {
3931 int page_bits = -os::vm_page_size();
3932 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3933 return x;
3934 else if (x > y)
3935 return (address)(intptr_t(y) | ~page_bits) + 1;
3936 else
3937 return (address)(intptr_t(y) & page_bits);
3938 }
3940 bool os::find(address addr) {
3941 Dl_info dlinfo;
3942 memset(&dlinfo, 0, sizeof(dlinfo));
3943 if (dladdr(addr, &dlinfo)) {
3944 tty->print(PTR_FORMAT ": ", addr);
3945 if (dlinfo.dli_sname != NULL) {
3946 tty->print("%s+%#x", dlinfo.dli_sname,
3947 addr - (intptr_t)dlinfo.dli_saddr);
3948 } else if (dlinfo.dli_fname) {
3949 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3950 } else {
3951 tty->print("<absolute address>");
3952 }
3953 if (dlinfo.dli_fname) {
3954 tty->print(" in %s", dlinfo.dli_fname);
3955 }
3956 if (dlinfo.dli_fbase) {
3957 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3958 }
3959 tty->cr();
3961 if (Verbose) {
3962 // decode some bytes around the PC
3963 address begin = same_page(addr-40, addr);
3964 address end = same_page(addr+40, addr);
3965 address lowest = (address) dlinfo.dli_sname;
3966 if (!lowest) lowest = (address) dlinfo.dli_fbase;
3967 if (begin < lowest) begin = lowest;
3968 Dl_info dlinfo2;
3969 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3970 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3971 end = (address) dlinfo2.dli_saddr;
3972 Disassembler::decode(begin, end);
3973 }
3974 return true;
3975 }
3976 return false;
3977 }
3979 #endif
3981 ////////////////////////////////////////////////////////////////////////////////
3982 // misc
3984 // This does not do anything on Linux. This is basically a hook for being
3985 // able to use structured exception handling (thread-local exception filters)
3986 // on, e.g., Win32.
3987 void
3988 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3989 JavaCallArguments* args, Thread* thread) {
3990 f(value, method, args, thread);
3991 }
3993 void os::print_statistics() {
3994 }
3996 int os::message_box(const char* title, const char* message) {
3997 int i;
3998 fdStream err(defaultStream::error_fd());
3999 for (i = 0; i < 78; i++) err.print_raw("=");
4000 err.cr();
4001 err.print_raw_cr(title);
4002 for (i = 0; i < 78; i++) err.print_raw("-");
4003 err.cr();
4004 err.print_raw_cr(message);
4005 for (i = 0; i < 78; i++) err.print_raw("=");
4006 err.cr();
4008 char buf[16];
4009 // Prevent process from exiting upon "read error" without consuming all CPU
4010 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4012 return buf[0] == 'y' || buf[0] == 'Y';
4013 }
4015 int os::stat(const char *path, struct stat *sbuf) {
4016 char pathbuf[MAX_PATH];
4017 if (strlen(path) > MAX_PATH - 1) {
4018 errno = ENAMETOOLONG;
4019 return -1;
4020 }
4021 hpi::native_path(strcpy(pathbuf, path));
4022 return ::stat(pathbuf, sbuf);
4023 }
4025 bool os::check_heap(bool force) {
4026 return true;
4027 }
4029 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4030 return ::vsnprintf(buf, count, format, args);
4031 }
4033 // Is a (classpath) directory empty?
4034 bool os::dir_is_empty(const char* path) {
4035 DIR *dir = NULL;
4036 struct dirent *ptr;
4038 dir = opendir(path);
4039 if (dir == NULL) return true;
4041 /* Scan the directory */
4042 bool result = true;
4043 char buf[sizeof(struct dirent) + MAX_PATH];
4044 while (result && (ptr = ::readdir(dir)) != NULL) {
4045 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4046 result = false;
4047 }
4048 }
4049 closedir(dir);
4050 return result;
4051 }
4053 // create binary file, rewriting existing file if required
4054 int os::create_binary_file(const char* path, bool rewrite_existing) {
4055 int oflags = O_WRONLY | O_CREAT;
4056 if (!rewrite_existing) {
4057 oflags |= O_EXCL;
4058 }
4059 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4060 }
4062 // return current position of file pointer
4063 jlong os::current_file_offset(int fd) {
4064 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4065 }
4067 // move file pointer to the specified offset
4068 jlong os::seek_to_file_offset(int fd, jlong offset) {
4069 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4070 }
4072 // Map a block of memory.
4073 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4074 char *addr, size_t bytes, bool read_only,
4075 bool allow_exec) {
4076 int prot;
4077 int flags;
4079 if (read_only) {
4080 prot = PROT_READ;
4081 flags = MAP_SHARED;
4082 } else {
4083 prot = PROT_READ | PROT_WRITE;
4084 flags = MAP_PRIVATE;
4085 }
4087 if (allow_exec) {
4088 prot |= PROT_EXEC;
4089 }
4091 if (addr != NULL) {
4092 flags |= MAP_FIXED;
4093 }
4095 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4096 fd, file_offset);
4097 if (mapped_address == MAP_FAILED) {
4098 return NULL;
4099 }
4100 return mapped_address;
4101 }
4104 // Remap a block of memory.
4105 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4106 char *addr, size_t bytes, bool read_only,
4107 bool allow_exec) {
4108 // same as map_memory() on this OS
4109 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4110 allow_exec);
4111 }
4114 // Unmap a block of memory.
4115 bool os::unmap_memory(char* addr, size_t bytes) {
4116 return munmap(addr, bytes) == 0;
4117 }
4119 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4121 static clockid_t thread_cpu_clockid(Thread* thread) {
4122 pthread_t tid = thread->osthread()->pthread_id();
4123 clockid_t clockid;
4125 // Get thread clockid
4126 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4127 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4128 return clockid;
4129 }
4131 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4132 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4133 // of a thread.
4134 //
4135 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4136 // the fast estimate available on the platform.
4138 jlong os::current_thread_cpu_time() {
4139 if (os::Linux::supports_fast_thread_cpu_time()) {
4140 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4141 } else {
4142 // return user + sys since the cost is the same
4143 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4144 }
4145 }
4147 jlong os::thread_cpu_time(Thread* thread) {
4148 // consistent with what current_thread_cpu_time() returns
4149 if (os::Linux::supports_fast_thread_cpu_time()) {
4150 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4151 } else {
4152 return slow_thread_cpu_time(thread, true /* user + sys */);
4153 }
4154 }
4156 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4157 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4158 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4159 } else {
4160 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4161 }
4162 }
4164 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4165 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4166 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4167 } else {
4168 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4169 }
4170 }
4172 //
4173 // -1 on error.
4174 //
4176 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4177 static bool proc_pid_cpu_avail = true;
4178 static bool proc_task_unchecked = true;
4179 static const char *proc_stat_path = "/proc/%d/stat";
4180 pid_t tid = thread->osthread()->thread_id();
4181 int i;
4182 char *s;
4183 char stat[2048];
4184 int statlen;
4185 char proc_name[64];
4186 int count;
4187 long sys_time, user_time;
4188 char string[64];
4189 int idummy;
4190 long ldummy;
4191 FILE *fp;
4193 // We first try accessing /proc/<pid>/cpu since this is faster to
4194 // process. If this file is not present (linux kernels 2.5 and above)
4195 // then we open /proc/<pid>/stat.
4196 if ( proc_pid_cpu_avail ) {
4197 sprintf(proc_name, "/proc/%d/cpu", tid);
4198 fp = fopen(proc_name, "r");
4199 if ( fp != NULL ) {
4200 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4201 fclose(fp);
4202 if ( count != 3 ) return -1;
4204 if (user_sys_cpu_time) {
4205 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4206 } else {
4207 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4208 }
4209 }
4210 else proc_pid_cpu_avail = false;
4211 }
4213 // The /proc/<tid>/stat aggregates per-process usage on
4214 // new Linux kernels 2.6+ where NPTL is supported.
4215 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4216 // See bug 6328462.
4217 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4218 // and possibly in some other cases, so we check its availability.
4219 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4220 // This is executed only once
4221 proc_task_unchecked = false;
4222 fp = fopen("/proc/self/task", "r");
4223 if (fp != NULL) {
4224 proc_stat_path = "/proc/self/task/%d/stat";
4225 fclose(fp);
4226 }
4227 }
4229 sprintf(proc_name, proc_stat_path, tid);
4230 fp = fopen(proc_name, "r");
4231 if ( fp == NULL ) return -1;
4232 statlen = fread(stat, 1, 2047, fp);
4233 stat[statlen] = '\0';
4234 fclose(fp);
4236 // Skip pid and the command string. Note that we could be dealing with
4237 // weird command names, e.g. user could decide to rename java launcher
4238 // to "java 1.4.2 :)", then the stat file would look like
4239 // 1234 (java 1.4.2 :)) R ... ...
4240 // We don't really need to know the command string, just find the last
4241 // occurrence of ")" and then start parsing from there. See bug 4726580.
4242 s = strrchr(stat, ')');
4243 i = 0;
4244 if (s == NULL ) return -1;
4246 // Skip blank chars
4247 do s++; while (isspace(*s));
4249 count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4250 &idummy, &idummy, &idummy, &idummy, &idummy,
4251 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4252 &user_time, &sys_time);
4253 if ( count != 12 ) return -1;
4254 if (user_sys_cpu_time) {
4255 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4256 } else {
4257 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4258 }
4259 }
4261 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4262 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4263 info_ptr->may_skip_backward = false; // elapsed time not wall time
4264 info_ptr->may_skip_forward = false; // elapsed time not wall time
4265 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4266 }
4268 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4269 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4270 info_ptr->may_skip_backward = false; // elapsed time not wall time
4271 info_ptr->may_skip_forward = false; // elapsed time not wall time
4272 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4273 }
4275 bool os::is_thread_cpu_time_supported() {
4276 return true;
4277 }
4279 // System loadavg support. Returns -1 if load average cannot be obtained.
4280 // Linux doesn't yet have a (official) notion of processor sets,
4281 // so just return the system wide load average.
4282 int os::loadavg(double loadavg[], int nelem) {
4283 return ::getloadavg(loadavg, nelem);
4284 }
4286 void os::pause() {
4287 char filename[MAX_PATH];
4288 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4289 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4290 } else {
4291 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4292 }
4294 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4295 if (fd != -1) {
4296 struct stat buf;
4297 close(fd);
4298 while (::stat(filename, &buf) == 0) {
4299 (void)::poll(NULL, 0, 100);
4300 }
4301 } else {
4302 jio_fprintf(stderr,
4303 "Could not open pause file '%s', continuing immediately.\n", filename);
4304 }
4305 }
4307 extern "C" {
4309 /**
4310 * NOTE: the following code is to keep the green threads code
4311 * in the libjava.so happy. Once the green threads is removed,
4312 * these code will no longer be needed.
4313 */
4314 int
4315 jdk_waitpid(pid_t pid, int* status, int options) {
4316 return waitpid(pid, status, options);
4317 }
4319 int
4320 fork1() {
4321 return fork();
4322 }
4324 int
4325 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4326 return sem_init(sem, pshared, value);
4327 }
4329 int
4330 jdk_sem_post(sem_t *sem) {
4331 return sem_post(sem);
4332 }
4334 int
4335 jdk_sem_wait(sem_t *sem) {
4336 return sem_wait(sem);
4337 }
4339 int
4340 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4341 return pthread_sigmask(how , newmask, oldmask);
4342 }
4344 }
4346 // Refer to the comments in os_solaris.cpp park-unpark.
4347 //
4348 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4349 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4350 // For specifics regarding the bug see GLIBC BUGID 261237 :
4351 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4352 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4353 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4354 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4355 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4356 // and monitorenter when we're using 1-0 locking. All those operations may result in
4357 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4358 // of libpthread avoids the problem, but isn't practical.
4359 //
4360 // Possible remedies:
4361 //
4362 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4363 // This is palliative and probabilistic, however. If the thread is preempted
4364 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4365 // than the minimum period may have passed, and the abstime may be stale (in the
4366 // past) resultin in a hang. Using this technique reduces the odds of a hang
4367 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4368 //
4369 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4370 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4371 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4372 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4373 // thread.
4374 //
4375 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4376 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4377 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4378 // This also works well. In fact it avoids kernel-level scalability impediments
4379 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4380 // timers in a graceful fashion.
4381 //
4382 // 4. When the abstime value is in the past it appears that control returns
4383 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4384 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4385 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4386 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4387 // It may be possible to avoid reinitialization by checking the return
4388 // value from pthread_cond_timedwait(). In addition to reinitializing the
4389 // condvar we must establish the invariant that cond_signal() is only called
4390 // within critical sections protected by the adjunct mutex. This prevents
4391 // cond_signal() from "seeing" a condvar that's in the midst of being
4392 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4393 // desirable signal-after-unlock optimization that avoids futile context switching.
4394 //
4395 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4396 // structure when a condvar is used or initialized. cond_destroy() would
4397 // release the helper structure. Our reinitialize-after-timedwait fix
4398 // put excessive stress on malloc/free and locks protecting the c-heap.
4399 //
4400 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4401 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4402 // and only enabling the work-around for vulnerable environments.
4404 // utility to compute the abstime argument to timedwait:
4405 // millis is the relative timeout time
4406 // abstime will be the absolute timeout time
4407 // TODO: replace compute_abstime() with unpackTime()
4409 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4410 if (millis < 0) millis = 0;
4411 struct timeval now;
4412 int status = gettimeofday(&now, NULL);
4413 assert(status == 0, "gettimeofday");
4414 jlong seconds = millis / 1000;
4415 millis %= 1000;
4416 if (seconds > 50000000) { // see man cond_timedwait(3T)
4417 seconds = 50000000;
4418 }
4419 abstime->tv_sec = now.tv_sec + seconds;
4420 long usec = now.tv_usec + millis * 1000;
4421 if (usec >= 1000000) {
4422 abstime->tv_sec += 1;
4423 usec -= 1000000;
4424 }
4425 abstime->tv_nsec = usec * 1000;
4426 return abstime;
4427 }
4430 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4431 // Conceptually TryPark() should be equivalent to park(0).
4433 int os::PlatformEvent::TryPark() {
4434 for (;;) {
4435 const int v = _Event ;
4436 guarantee ((v == 0) || (v == 1), "invariant") ;
4437 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4438 }
4439 }
4441 void os::PlatformEvent::park() { // AKA "down()"
4442 // Invariant: Only the thread associated with the Event/PlatformEvent
4443 // may call park().
4444 // TODO: assert that _Assoc != NULL or _Assoc == Self
4445 int v ;
4446 for (;;) {
4447 v = _Event ;
4448 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4449 }
4450 guarantee (v >= 0, "invariant") ;
4451 if (v == 0) {
4452 // Do this the hard way by blocking ...
4453 int status = pthread_mutex_lock(_mutex);
4454 assert_status(status == 0, status, "mutex_lock");
4455 guarantee (_nParked == 0, "invariant") ;
4456 ++ _nParked ;
4457 while (_Event < 0) {
4458 status = pthread_cond_wait(_cond, _mutex);
4459 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4460 // Treat this the same as if the wait was interrupted
4461 if (status == ETIME) { status = EINTR; }
4462 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4463 }
4464 -- _nParked ;
4466 // In theory we could move the ST of 0 into _Event past the unlock(),
4467 // but then we'd need a MEMBAR after the ST.
4468 _Event = 0 ;
4469 status = pthread_mutex_unlock(_mutex);
4470 assert_status(status == 0, status, "mutex_unlock");
4471 }
4472 guarantee (_Event >= 0, "invariant") ;
4473 }
4475 int os::PlatformEvent::park(jlong millis) {
4476 guarantee (_nParked == 0, "invariant") ;
4478 int v ;
4479 for (;;) {
4480 v = _Event ;
4481 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4482 }
4483 guarantee (v >= 0, "invariant") ;
4484 if (v != 0) return OS_OK ;
4486 // We do this the hard way, by blocking the thread.
4487 // Consider enforcing a minimum timeout value.
4488 struct timespec abst;
4489 compute_abstime(&abst, millis);
4491 int ret = OS_TIMEOUT;
4492 int status = pthread_mutex_lock(_mutex);
4493 assert_status(status == 0, status, "mutex_lock");
4494 guarantee (_nParked == 0, "invariant") ;
4495 ++_nParked ;
4497 // Object.wait(timo) will return because of
4498 // (a) notification
4499 // (b) timeout
4500 // (c) thread.interrupt
4501 //
4502 // Thread.interrupt and object.notify{All} both call Event::set.
4503 // That is, we treat thread.interrupt as a special case of notification.
4504 // The underlying Solaris implementation, cond_timedwait, admits
4505 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4506 // JVM from making those visible to Java code. As such, we must
4507 // filter out spurious wakeups. We assume all ETIME returns are valid.
4508 //
4509 // TODO: properly differentiate simultaneous notify+interrupt.
4510 // In that case, we should propagate the notify to another waiter.
4512 while (_Event < 0) {
4513 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4514 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4515 pthread_cond_destroy (_cond);
4516 pthread_cond_init (_cond, NULL) ;
4517 }
4518 assert_status(status == 0 || status == EINTR ||
4519 status == ETIME || status == ETIMEDOUT,
4520 status, "cond_timedwait");
4521 if (!FilterSpuriousWakeups) break ; // previous semantics
4522 if (status == ETIME || status == ETIMEDOUT) break ;
4523 // We consume and ignore EINTR and spurious wakeups.
4524 }
4525 --_nParked ;
4526 if (_Event >= 0) {
4527 ret = OS_OK;
4528 }
4529 _Event = 0 ;
4530 status = pthread_mutex_unlock(_mutex);
4531 assert_status(status == 0, status, "mutex_unlock");
4532 assert (_nParked == 0, "invariant") ;
4533 return ret;
4534 }
4536 void os::PlatformEvent::unpark() {
4537 int v, AnyWaiters ;
4538 for (;;) {
4539 v = _Event ;
4540 if (v > 0) {
4541 // The LD of _Event could have reordered or be satisfied
4542 // by a read-aside from this processor's write buffer.
4543 // To avoid problems execute a barrier and then
4544 // ratify the value.
4545 OrderAccess::fence() ;
4546 if (_Event == v) return ;
4547 continue ;
4548 }
4549 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4550 }
4551 if (v < 0) {
4552 // Wait for the thread associated with the event to vacate
4553 int status = pthread_mutex_lock(_mutex);
4554 assert_status(status == 0, status, "mutex_lock");
4555 AnyWaiters = _nParked ;
4556 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4557 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4558 AnyWaiters = 0 ;
4559 pthread_cond_signal (_cond);
4560 }
4561 status = pthread_mutex_unlock(_mutex);
4562 assert_status(status == 0, status, "mutex_unlock");
4563 if (AnyWaiters != 0) {
4564 status = pthread_cond_signal(_cond);
4565 assert_status(status == 0, status, "cond_signal");
4566 }
4567 }
4569 // Note that we signal() _after dropping the lock for "immortal" Events.
4570 // This is safe and avoids a common class of futile wakeups. In rare
4571 // circumstances this can cause a thread to return prematurely from
4572 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4573 // simply re-test the condition and re-park itself.
4574 }
4577 // JSR166
4578 // -------------------------------------------------------
4580 /*
4581 * The solaris and linux implementations of park/unpark are fairly
4582 * conservative for now, but can be improved. They currently use a
4583 * mutex/condvar pair, plus a a count.
4584 * Park decrements count if > 0, else does a condvar wait. Unpark
4585 * sets count to 1 and signals condvar. Only one thread ever waits
4586 * on the condvar. Contention seen when trying to park implies that someone
4587 * is unparking you, so don't wait. And spurious returns are fine, so there
4588 * is no need to track notifications.
4589 */
4592 #define NANOSECS_PER_SEC 1000000000
4593 #define NANOSECS_PER_MILLISEC 1000000
4594 #define MAX_SECS 100000000
4595 /*
4596 * This code is common to linux and solaris and will be moved to a
4597 * common place in dolphin.
4598 *
4599 * The passed in time value is either a relative time in nanoseconds
4600 * or an absolute time in milliseconds. Either way it has to be unpacked
4601 * into suitable seconds and nanoseconds components and stored in the
4602 * given timespec structure.
4603 * Given time is a 64-bit value and the time_t used in the timespec is only
4604 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4605 * overflow if times way in the future are given. Further on Solaris versions
4606 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4607 * number of seconds, in abstime, is less than current_time + 100,000,000.
4608 * As it will be 28 years before "now + 100000000" will overflow we can
4609 * ignore overflow and just impose a hard-limit on seconds using the value
4610 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4611 * years from "now".
4612 */
4614 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4615 assert (time > 0, "convertTime");
4617 struct timeval now;
4618 int status = gettimeofday(&now, NULL);
4619 assert(status == 0, "gettimeofday");
4621 time_t max_secs = now.tv_sec + MAX_SECS;
4623 if (isAbsolute) {
4624 jlong secs = time / 1000;
4625 if (secs > max_secs) {
4626 absTime->tv_sec = max_secs;
4627 }
4628 else {
4629 absTime->tv_sec = secs;
4630 }
4631 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4632 }
4633 else {
4634 jlong secs = time / NANOSECS_PER_SEC;
4635 if (secs >= MAX_SECS) {
4636 absTime->tv_sec = max_secs;
4637 absTime->tv_nsec = 0;
4638 }
4639 else {
4640 absTime->tv_sec = now.tv_sec + secs;
4641 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4642 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4643 absTime->tv_nsec -= NANOSECS_PER_SEC;
4644 ++absTime->tv_sec; // note: this must be <= max_secs
4645 }
4646 }
4647 }
4648 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4649 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4650 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4651 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4652 }
4654 void Parker::park(bool isAbsolute, jlong time) {
4655 // Optional fast-path check:
4656 // Return immediately if a permit is available.
4657 if (_counter > 0) {
4658 _counter = 0 ;
4659 return ;
4660 }
4662 Thread* thread = Thread::current();
4663 assert(thread->is_Java_thread(), "Must be JavaThread");
4664 JavaThread *jt = (JavaThread *)thread;
4666 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4667 // Check interrupt before trying to wait
4668 if (Thread::is_interrupted(thread, false)) {
4669 return;
4670 }
4672 // Next, demultiplex/decode time arguments
4673 timespec absTime;
4674 if (time < 0) { // don't wait at all
4675 return;
4676 }
4677 if (time > 0) {
4678 unpackTime(&absTime, isAbsolute, time);
4679 }
4682 // Enter safepoint region
4683 // Beware of deadlocks such as 6317397.
4684 // The per-thread Parker:: mutex is a classic leaf-lock.
4685 // In particular a thread must never block on the Threads_lock while
4686 // holding the Parker:: mutex. If safepoints are pending both the
4687 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4688 ThreadBlockInVM tbivm(jt);
4690 // Don't wait if cannot get lock since interference arises from
4691 // unblocking. Also. check interrupt before trying wait
4692 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4693 return;
4694 }
4696 int status ;
4697 if (_counter > 0) { // no wait needed
4698 _counter = 0;
4699 status = pthread_mutex_unlock(_mutex);
4700 assert (status == 0, "invariant") ;
4701 return;
4702 }
4704 #ifdef ASSERT
4705 // Don't catch signals while blocked; let the running threads have the signals.
4706 // (This allows a debugger to break into the running thread.)
4707 sigset_t oldsigs;
4708 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4709 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4710 #endif
4712 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4713 jt->set_suspend_equivalent();
4714 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4716 if (time == 0) {
4717 status = pthread_cond_wait (_cond, _mutex) ;
4718 } else {
4719 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4720 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4721 pthread_cond_destroy (_cond) ;
4722 pthread_cond_init (_cond, NULL);
4723 }
4724 }
4725 assert_status(status == 0 || status == EINTR ||
4726 status == ETIME || status == ETIMEDOUT,
4727 status, "cond_timedwait");
4729 #ifdef ASSERT
4730 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4731 #endif
4733 _counter = 0 ;
4734 status = pthread_mutex_unlock(_mutex) ;
4735 assert_status(status == 0, status, "invariant") ;
4736 // If externally suspended while waiting, re-suspend
4737 if (jt->handle_special_suspend_equivalent_condition()) {
4738 jt->java_suspend_self();
4739 }
4741 }
4743 void Parker::unpark() {
4744 int s, status ;
4745 status = pthread_mutex_lock(_mutex);
4746 assert (status == 0, "invariant") ;
4747 s = _counter;
4748 _counter = 1;
4749 if (s < 1) {
4750 if (WorkAroundNPTLTimedWaitHang) {
4751 status = pthread_cond_signal (_cond) ;
4752 assert (status == 0, "invariant") ;
4753 status = pthread_mutex_unlock(_mutex);
4754 assert (status == 0, "invariant") ;
4755 } else {
4756 status = pthread_mutex_unlock(_mutex);
4757 assert (status == 0, "invariant") ;
4758 status = pthread_cond_signal (_cond) ;
4759 assert (status == 0, "invariant") ;
4760 }
4761 } else {
4762 pthread_mutex_unlock(_mutex);
4763 assert (status == 0, "invariant") ;
4764 }
4765 }
4768 extern char** environ;
4770 #ifndef __NR_fork
4771 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4772 #endif
4774 #ifndef __NR_execve
4775 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4776 #endif
4778 // Run the specified command in a separate process. Return its exit value,
4779 // or -1 on failure (e.g. can't fork a new process).
4780 // Unlike system(), this function can be called from signal handler. It
4781 // doesn't block SIGINT et al.
4782 int os::fork_and_exec(char* cmd) {
4783 const char * argv[4] = {"sh", "-c", cmd, NULL};
4785 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4786 // pthread_atfork handlers and reset pthread library. All we need is a
4787 // separate process to execve. Make a direct syscall to fork process.
4788 // On IA64 there's no fork syscall, we have to use fork() and hope for
4789 // the best...
4790 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4791 IA64_ONLY(fork();)
4793 if (pid < 0) {
4794 // fork failed
4795 return -1;
4797 } else if (pid == 0) {
4798 // child process
4800 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4801 // first to kill every thread on the thread list. Because this list is
4802 // not reset by fork() (see notes above), execve() will instead kill
4803 // every thread in the parent process. We know this is the only thread
4804 // in the new process, so make a system call directly.
4805 // IA64 should use normal execve() from glibc to match the glibc fork()
4806 // above.
4807 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4808 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4810 // execve failed
4811 _exit(-1);
4813 } else {
4814 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4815 // care about the actual exit code, for now.
4817 int status;
4819 // Wait for the child process to exit. This returns immediately if
4820 // the child has already exited. */
4821 while (waitpid(pid, &status, 0) < 0) {
4822 switch (errno) {
4823 case ECHILD: return 0;
4824 case EINTR: break;
4825 default: return -1;
4826 }
4827 }
4829 if (WIFEXITED(status)) {
4830 // The child exited normally; get its exit code.
4831 return WEXITSTATUS(status);
4832 } else if (WIFSIGNALED(status)) {
4833 // The child exited because of a signal
4834 // The best value to return is 0x80 + signal number,
4835 // because that is what all Unix shells do, and because
4836 // it allows callers to distinguish between process exit and
4837 // process death by signal.
4838 return 0x80 + WTERMSIG(status);
4839 } else {
4840 // Unknown exit code; pass it through
4841 return status;
4842 }
4843 }
4844 }