1 #ifdef USE_PRAGMA_IDENT_SRC
2 #pragma ident "@(#)os_linux.cpp 1.259 08/11/24 12:20:22 JVM"
3 #endif
4 /*
5 * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved.
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7 *
8 * This code is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License version 2 only, as
10 * published by the Free Software Foundation.
11 *
12 * This code is distributed in the hope that it will be useful, but WITHOUT
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 * version 2 for more details (a copy is included in the LICENSE file that
16 * accompanied this code).
17 *
18 * You should have received a copy of the GNU General Public License version
19 * 2 along with this work; if not, write to the Free Software Foundation,
20 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
21 *
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
24 * have any questions.
25 *
26 */
27
28 // do not include precompiled header file
29 # include "incls/_os_linux.cpp.incl"
30
31 // put OS-includes here
32 # include <sys/types.h>
33 # include <sys/mman.h>
34 # include <pthread.h>
35 # include <signal.h>
36 # include <errno.h>
37 # include <dlfcn.h>
38 # include <stdio.h>
39 # include <unistd.h>
40 # include <sys/resource.h>
41 # include <pthread.h>
42 # include <sys/stat.h>
43 # include <sys/time.h>
44 # include <sys/times.h>
45 # include <sys/utsname.h>
46 # include <sys/socket.h>
47 # include <sys/wait.h>
48 # include <pwd.h>
49 # include <poll.h>
50 # include <semaphore.h>
51 # include <fcntl.h>
52 # include <string.h>
53 # include <syscall.h>
54 # include <sys/sysinfo.h>
55 # include <gnu/libc-version.h>
56 # include <sys/ipc.h>
57 # include <sys/shm.h>
58 # include <link.h>
59
60 #define MAX_PATH (2 * K)
61
62 // for timer info max values which include all bits
63 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
64 #define SEC_IN_NANOSECS 1000000000LL
65
66 ////////////////////////////////////////////////////////////////////////////////
67 // global variables
68 julong os::Linux::_physical_memory = 0;
69
70 address os::Linux::_initial_thread_stack_bottom = NULL;
71 uintptr_t os::Linux::_initial_thread_stack_size = 0;
72
73 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
74 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
75 Mutex* os::Linux::_createThread_lock = NULL;
76 pthread_t os::Linux::_main_thread;
77 int os::Linux::_page_size = -1;
78 bool os::Linux::_is_floating_stack = false;
79 bool os::Linux::_is_NPTL = false;
80 bool os::Linux::_supports_fast_thread_cpu_time = false;
81 const char * os::Linux::_glibc_version = NULL;
82 const char * os::Linux::_libpthread_version = NULL;
83
84 static jlong initial_time_count=0;
85
86 static int clock_tics_per_sec = 100;
87
88 // For diagnostics to print a message once. see run_periodic_checks
89 static sigset_t check_signal_done;
90 static bool check_signals = true;;
91
92 static pid_t _initial_pid = 0;
93
94 /* Signal number used to suspend/resume a thread */
95
96 /* do not use any signal number less than SIGSEGV, see 4355769 */
97 static int SR_signum = SIGUSR2;
98 sigset_t SR_sigset;
99
100 /* Used to protect dlsym() calls */
101 static pthread_mutex_t dl_mutex;
102
103 ////////////////////////////////////////////////////////////////////////////////
104 // utility functions
105
106 static int SR_initialize();
107 static int SR_finalize();
108
109 julong os::available_memory() {
110 return Linux::available_memory();
111 }
112
113 julong os::Linux::available_memory() {
114 // values in struct sysinfo are "unsigned long"
115 struct sysinfo si;
116 sysinfo(&si);
117
118 return (julong)si.freeram * si.mem_unit;
119 }
120
121 julong os::physical_memory() {
122 return Linux::physical_memory();
123 }
124
125 julong os::allocatable_physical_memory(julong size) {
126 #ifdef _LP64
127 return size;
128 #else
129 julong result = MIN2(size, (julong)3800*M);
130 if (!is_allocatable(result)) {
131 // See comments under solaris for alignment considerations
132 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
133 result = MIN2(size, reasonable_size);
134 }
135 return result;
136 #endif // _LP64
137 }
138
139 ////////////////////////////////////////////////////////////////////////////////
140 // environment support
141
142 bool os::getenv(const char* name, char* buf, int len) {
143 const char* val = ::getenv(name);
144 if (val != NULL && strlen(val) < (size_t)len) {
145 strcpy(buf, val);
146 return true;
147 }
148 if (len > 0) buf[0] = 0; // return a null string
149 return false;
150 }
151
152
153 // Return true if user is running as root.
154
155 bool os::have_special_privileges() {
156 static bool init = false;
157 static bool privileges = false;
158 if (!init) {
159 privileges = (getuid() != geteuid()) || (getgid() != getegid());
160 init = true;
161 }
162 return privileges;
163 }
164
165
166 #ifndef SYS_gettid
167 // i386: 224, ia64: 1105, amd64: 186, sparc 143
168 #ifdef __ia64__
169 #define SYS_gettid 1105
170 #elif __i386__
171 #define SYS_gettid 224
172 #elif __amd64__
173 #define SYS_gettid 186
174 #elif __sparc__
175 #define SYS_gettid 143
176 #else
177 #error define gettid for the arch
178 #endif
179 #endif
180
181 // Cpu architecture string
182 #if defined(IA64)
183 static char cpu_arch[] = "ia64";
184 #elif defined(IA32)
185 static char cpu_arch[] = "i386";
186 #elif defined(AMD64)
187 static char cpu_arch[] = "amd64";
188 #elif defined(SPARC)
189 # ifdef _LP64
190 static char cpu_arch[] = "sparcv9";
191 # else
192 static char cpu_arch[] = "sparc";
193 # endif
194 #else
195 #error Add appropriate cpu_arch setting
196 #endif
197
198
199 // pid_t gettid()
200 //
201 // Returns the kernel thread id of the currently running thread. Kernel
202 // thread id is used to access /proc.
203 //
204 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
205 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
206 //
207 pid_t os::Linux::gettid() {
208 int rslt = syscall(SYS_gettid);
209 if (rslt == -1) {
210 // old kernel, no NPTL support
211 return getpid();
212 } else {
213 return (pid_t)rslt;
214 }
215 }
216
217 // Most versions of linux have a bug where the number of processors are
218 // determined by looking at the /proc file system. In a chroot environment,
219 // the system call returns 1. This causes the VM to act as if it is
220 // a single processor and elide locking (see is_MP() call).
221 static bool unsafe_chroot_detected = false;
222 static const char *unstable_chroot_error = "/proc file system not found.\n"
223 "Java may be unstable running multithreaded in a chroot "
224 "environment on Linux when /proc filesystem is not mounted.";
225
226 void os::Linux::initialize_system_info() {
227 _processor_count = sysconf(_SC_NPROCESSORS_CONF);
228 if (_processor_count == 1) {
229 pid_t pid = os::Linux::gettid();
230 char fname[32];
231 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
232 FILE *fp = fopen(fname, "r");
233 if (fp == NULL) {
234 unsafe_chroot_detected = true;
235 } else {
236 fclose(fp);
237 }
238 }
239 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
240 assert(_processor_count > 0, "linux error");
241 }
242
243 void os::init_system_properties_values() {
244 // char arch[12];
245 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
246
247 // The next steps are taken in the product version:
248 //
249 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
250 // This library should be located at:
251 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
252 //
253 // If "/jre/lib/" appears at the right place in the path, then we
254 // assume libjvm[_g].so is installed in a JDK and we use this path.
255 //
256 // Otherwise exit with message: "Could not create the Java virtual machine."
257 //
258 // The following extra steps are taken in the debugging version:
259 //
260 // If "/jre/lib/" does NOT appear at the right place in the path
261 // instead of exit check for $JAVA_HOME environment variable.
262 //
263 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
264 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
265 // it looks like libjvm[_g].so is installed there
266 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
267 //
268 // Otherwise exit.
269 //
270 // Important note: if the location of libjvm.so changes this
271 // code needs to be changed accordingly.
272
273 // The next few definitions allow the code to be verbatim:
274 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
275 #define getenv(n) ::getenv(n)
276
277 /*
278 * See ld(1):
279 * The linker uses the following search paths to locate required
280 * shared libraries:
281 * 1: ...
282 * ...
283 * 7: The default directories, normally /lib and /usr/lib.
284 */
285 #define DEFAULT_LIBPATH "/lib:/usr/lib"
286
287 #define EXTENSIONS_DIR "/lib/ext"
288 #define ENDORSED_DIR "/lib/endorsed"
289 #define REG_DIR "/usr/java/packages"
290
291 {
292 /* sysclasspath, java_home, dll_dir */
293 {
294 char *home_path;
295 char *dll_path;
296 char *pslash;
297 char buf[MAXPATHLEN];
298 os::jvm_path(buf, sizeof(buf));
299
300 // Found the full path to libjvm.so.
301 // Now cut the path to <java_home>/jre if we can.
302 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
303 pslash = strrchr(buf, '/');
304 if (pslash != NULL)
305 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
306 dll_path = malloc(strlen(buf) + 1);
307 if (dll_path == NULL)
308 return;
309 strcpy(dll_path, buf);
310 Arguments::set_dll_dir(dll_path);
311
312 if (pslash != NULL) {
313 pslash = strrchr(buf, '/');
314 if (pslash != NULL) {
315 *pslash = '\0'; /* get rid of /<arch> */
316 pslash = strrchr(buf, '/');
317 if (pslash != NULL)
318 *pslash = '\0'; /* get rid of /lib */
319 }
320 }
321
322 home_path = malloc(strlen(buf) + 1);
323 if (home_path == NULL)
324 return;
325 strcpy(home_path, buf);
326 Arguments::set_java_home(home_path);
327
328 if (!set_boot_path('/', ':'))
329 return;
330 }
331
332 /*
333 * Where to look for native libraries
334 *
335 * Note: Due to a legacy implementation, most of the library path
336 * is set in the launcher. This was to accomodate linking restrictions
337 * on legacy Linux implementations (which are no longer supported).
338 * Eventually, all the library path setting will be done here.
339 *
340 * However, to prevent the proliferation of improperly built native
341 * libraries, the new path component /usr/java/packages is added here.
342 * Eventually, all the library path setting will be done here.
343 */
344 {
345 char *ld_library_path;
346
347 /*
348 * Construct the invariant part of ld_library_path. Note that the
349 * space for the colon and the trailing null are provided by the
350 * nulls included by the sizeof operator (so actually we allocate
351 * a byte more than necessary).
352 */
353 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
354 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
355 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
356
357 /*
358 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
359 * should always exist (until the legacy problem cited above is
360 * addressed).
361 */
362 char *v = getenv("LD_LIBRARY_PATH");
363 if (v != NULL) {
364 char *t = ld_library_path;
365 /* That's +1 for the colon and +1 for the trailing '\0' */
366 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
367 sprintf(ld_library_path, "%s:%s", v, t);
368 }
369 Arguments::set_library_path(ld_library_path);
370 }
371
372 /*
373 * Extensions directories.
374 *
375 * Note that the space for the colon and the trailing null are provided
376 * by the nulls included by the sizeof operator (so actually one byte more
377 * than necessary is allocated).
378 */
379 {
380 char *buf = malloc(strlen(Arguments::get_java_home()) +
381 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
382 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
383 Arguments::get_java_home());
384 Arguments::set_ext_dirs(buf);
385 }
386
387 /* Endorsed standards default directory. */
388 {
389 char * buf;
390 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
391 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
392 Arguments::set_endorsed_dirs(buf);
393 }
394 }
395
396 #undef malloc
397 #undef getenv
398 #undef EXTENSIONS_DIR
399 #undef ENDORSED_DIR
400
401 // Done
402 return;
403 }
404
405 ////////////////////////////////////////////////////////////////////////////////
406 // breakpoint support
407
408 void os::breakpoint() {
409 BREAKPOINT;
410 }
411
412 extern "C" void breakpoint() {
413 // use debugger to set breakpoint here
414 }
415
416 ////////////////////////////////////////////////////////////////////////////////
417 // signal support
418
419 debug_only(static bool signal_sets_initialized = false);
420 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
421
422 bool os::Linux::is_sig_ignored(int sig) {
423 struct sigaction oact;
424 sigaction(sig, (struct sigaction*)NULL, &oact);
425 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
426 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
427 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
428 return true;
429 else
430 return false;
431 }
432
433 void os::Linux::signal_sets_init() {
434 // Should also have an assertion stating we are still single-threaded.
435 assert(!signal_sets_initialized, "Already initialized");
436 // Fill in signals that are necessarily unblocked for all threads in
437 // the VM. Currently, we unblock the following signals:
438 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
439 // by -Xrs (=ReduceSignalUsage));
440 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
441 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
442 // the dispositions or masks wrt these signals.
443 // Programs embedding the VM that want to use the above signals for their
444 // own purposes must, at this time, use the "-Xrs" option to prevent
445 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
446 // (See bug 4345157, and other related bugs).
447 // In reality, though, unblocking these signals is really a nop, since
448 // these signals are not blocked by default.
449 sigemptyset(&unblocked_sigs);
450 sigemptyset(&allowdebug_blocked_sigs);
451 sigaddset(&unblocked_sigs, SIGILL);
452 sigaddset(&unblocked_sigs, SIGSEGV);
453 sigaddset(&unblocked_sigs, SIGBUS);
454 sigaddset(&unblocked_sigs, SIGFPE);
455 sigaddset(&unblocked_sigs, SR_signum);
456
457 if (!ReduceSignalUsage) {
458 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
459 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
460 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
461 }
462 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
463 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
464 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
465 }
466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
468 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
469 }
470 }
471 // Fill in signals that are blocked by all but the VM thread.
472 sigemptyset(&vm_sigs);
473 if (!ReduceSignalUsage)
474 sigaddset(&vm_sigs, BREAK_SIGNAL);
475 debug_only(signal_sets_initialized = true);
476
477 }
478
479 // These are signals that are unblocked while a thread is running Java.
480 // (For some reason, they get blocked by default.)
481 sigset_t* os::Linux::unblocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &unblocked_sigs;
484 }
485
486 // These are the signals that are blocked while a (non-VM) thread is
487 // running Java. Only the VM thread handles these signals.
488 sigset_t* os::Linux::vm_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &vm_sigs;
491 }
492
493 // These are signals that are blocked during cond_wait to allow debugger in
494 sigset_t* os::Linux::allowdebug_blocked_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &allowdebug_blocked_sigs;
497 }
498
499 void os::Linux::hotspot_sigmask(Thread* thread) {
500
501 //Save caller's signal mask before setting VM signal mask
502 sigset_t caller_sigmask;
503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
504
505 OSThread* osthread = thread->osthread();
506 osthread->set_caller_sigmask(caller_sigmask);
507
508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
509
510 if (!ReduceSignalUsage) {
511 if (thread->is_VM_thread()) {
512 // Only the VM thread handles BREAK_SIGNAL ...
513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
514 } else {
515 // ... all other threads block BREAK_SIGNAL
516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
517 }
518 }
519 }
520
521 //////////////////////////////////////////////////////////////////////////////
522 // detecting pthread library
523
524 void os::Linux::libpthread_init() {
525 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
526 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
527 // generic name for earlier versions.
528 // Define macros here so we can build HotSpot on old systems.
529 # ifndef _CS_GNU_LIBC_VERSION
530 # define _CS_GNU_LIBC_VERSION 2
531 # endif
532 # ifndef _CS_GNU_LIBPTHREAD_VERSION
533 # define _CS_GNU_LIBPTHREAD_VERSION 3
534 # endif
535
536 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
537 if (n > 0) {
538 char *str = (char *)malloc(n);
539 confstr(_CS_GNU_LIBC_VERSION, str, n);
540 os::Linux::set_glibc_version(str);
541 } else {
542 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
543 static char _gnu_libc_version[32];
544 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
545 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
546 os::Linux::set_glibc_version(_gnu_libc_version);
547 }
548
549 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
550 if (n > 0) {
551 char *str = (char *)malloc(n);
552 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
553 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
554 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
555 // is the case. LinuxThreads has a hard limit on max number of threads.
556 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
557 // On the other hand, NPTL does not have such a limit, sysconf()
558 // will return -1 and errno is not changed. Check if it is really NPTL.
559 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
560 strstr(str, "NPTL") &&
561 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
562 free(str);
563 os::Linux::set_libpthread_version("linuxthreads");
564 } else {
565 os::Linux::set_libpthread_version(str);
566 }
567 } else {
568 // glibc before 2.3.2 only has LinuxThreads.
569 os::Linux::set_libpthread_version("linuxthreads");
570 }
571
572 if (strstr(libpthread_version(), "NPTL")) {
573 os::Linux::set_is_NPTL();
574 } else {
575 os::Linux::set_is_LinuxThreads();
576 }
577
578 // LinuxThreads have two flavors: floating-stack mode, which allows variable
579 // stack size; and fixed-stack mode. NPTL is always floating-stack.
580 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
581 os::Linux::set_is_floating_stack();
582 }
583 }
584
585 /////////////////////////////////////////////////////////////////////////////
586 // thread stack
587
588 // Force Linux kernel to expand current thread stack. If "bottom" is close
589 // to the stack guard, caller should block all signals.
590 //
591 // MAP_GROWSDOWN:
592 // A special mmap() flag that is used to implement thread stacks. It tells
593 // kernel that the memory region should extend downwards when needed. This
594 // allows early versions of LinuxThreads to only mmap the first few pages
595 // when creating a new thread. Linux kernel will automatically expand thread
596 // stack as needed (on page faults).
597 //
598 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
599 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
600 // region, it's hard to tell if the fault is due to a legitimate stack
601 // access or because of reading/writing non-exist memory (e.g. buffer
602 // overrun). As a rule, if the fault happens below current stack pointer,
603 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
604 // application (see Linux kernel fault.c).
605 //
606 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
607 // stack overflow detection.
608 //
609 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
610 // not use this flag. However, the stack of initial thread is not created
611 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
612 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
613 // and then attach the thread to JVM.
614 //
615 // To get around the problem and allow stack banging on Linux, we need to
616 // manually expand thread stack after receiving the SIGSEGV.
617 //
618 // There are two ways to expand thread stack to address "bottom", we used
619 // both of them in JVM before 1.5:
620 // 1. adjust stack pointer first so that it is below "bottom", and then
621 // touch "bottom"
622 // 2. mmap() the page in question
623 //
624 // Now alternate signal stack is gone, it's harder to use 2. For instance,
625 // if current sp is already near the lower end of page 101, and we need to
626 // call mmap() to map page 100, it is possible that part of the mmap() frame
627 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
628 // That will destroy the mmap() frame and cause VM to crash.
629 //
630 // The following code works by adjusting sp first, then accessing the "bottom"
631 // page to force a page fault. Linux kernel will then automatically expand the
632 // stack mapping.
633 //
634 // _expand_stack_to() assumes its frame size is less than page size, which
635 // should always be true if the function is not inlined.
636
637 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
638 #define NOINLINE
639 #else
640 #define NOINLINE __attribute__ ((noinline))
641 #endif
642
643 static void _expand_stack_to(address bottom) NOINLINE;
644
645 static void _expand_stack_to(address bottom) {
646 address sp;
647 size_t size;
648 volatile char *p;
649
650 // Adjust bottom to point to the largest address within the same page, it
651 // gives us a one-page buffer if alloca() allocates slightly more memory.
652 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
653 bottom += os::Linux::page_size() - 1;
654
655 // sp might be slightly above current stack pointer; if that's the case, we
656 // will alloca() a little more space than necessary, which is OK. Don't use
657 // os::current_stack_pointer(), as its result can be slightly below current
658 // stack pointer, causing us to not alloca enough to reach "bottom".
659 sp = (address)&sp;
660
661 if (sp > bottom) {
662 size = sp - bottom;
663 p = (volatile char *)alloca(size);
664 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
665 p[0] = '\0';
666 }
667 }
668
669 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
670 assert(t!=NULL, "just checking");
671 assert(t->osthread()->expanding_stack(), "expand should be set");
672 assert(t->stack_base() != NULL, "stack_base was not initialized");
673
674 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
675 sigset_t mask_all, old_sigset;
676 sigfillset(&mask_all);
677 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
678 _expand_stack_to(addr);
679 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
680 return true;
681 }
682 return false;
683 }
684
685 //////////////////////////////////////////////////////////////////////////////
686 // create new thread
687
688 static address highest_vm_reserved_address();
689
690 // check if it's safe to start a new thread
691 static bool _thread_safety_check(Thread* thread) {
692 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
693 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
694 // Heap is mmap'ed at lower end of memory space. Thread stacks are
695 // allocated (MAP_FIXED) from high address space. Every thread stack
696 // occupies a fixed size slot (usually 2Mbytes, but user can change
697 // it to other values if they rebuild LinuxThreads).
698 //
699 // Problem with MAP_FIXED is that mmap() can still succeed even part of
700 // the memory region has already been mmap'ed. That means if we have too
701 // many threads and/or very large heap, eventually thread stack will
702 // collide with heap.
703 //
704 // Here we try to prevent heap/stack collision by comparing current
705 // stack bottom with the highest address that has been mmap'ed by JVM
706 // plus a safety margin for memory maps created by native code.
707 //
708 // This feature can be disabled by setting ThreadSafetyMargin to 0
709 //
710 if (ThreadSafetyMargin > 0) {
711 address stack_bottom = os::current_stack_base() - os::current_stack_size();
712
713 // not safe if our stack extends below the safety margin
714 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
715 } else {
716 return true;
717 }
718 } else {
719 // Floating stack LinuxThreads or NPTL:
720 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
721 // there's not enough space left, pthread_create() will fail. If we come
722 // here, that means enough space has been reserved for stack.
723 return true;
724 }
725 }
726
727 // Thread start routine for all newly created threads
728 static void *java_start(Thread *thread) {
729 // Try to randomize the cache line index of hot stack frames.
730 // This helps when threads of the same stack traces evict each other's
731 // cache lines. The threads can be either from the same JVM instance, or
732 // from different JVM instances. The benefit is especially true for
733 // processors with hyperthreading technology.
734 static int counter = 0;
735 int pid = os::current_process_id();
736 alloca(((pid ^ counter++) & 7) * 128);
737
738 ThreadLocalStorage::set_thread(thread);
739
740 OSThread* osthread = thread->osthread();
741 Monitor* sync = osthread->startThread_lock();
742
743 // non floating stack LinuxThreads needs extra check, see above
744 if (!_thread_safety_check(thread)) {
745 // notify parent thread
746 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
747 osthread->set_state(ZOMBIE);
748 sync->notify_all();
749 return NULL;
750 }
751
752 // thread_id is kernel thread id (similar to Solaris LWP id)
753 osthread->set_thread_id(os::Linux::gettid());
754
755 if (UseNUMA) {
756 int lgrp_id = os::numa_get_group_id();
757 if (lgrp_id != -1) {
758 thread->set_lgrp_id(lgrp_id);
759 }
760 }
761 // initialize signal mask for this thread
762 os::Linux::hotspot_sigmask(thread);
763
764 // initialize floating point control register
765 os::Linux::init_thread_fpu_state();
766
767 // handshaking with parent thread
768 {
769 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
770
771 // notify parent thread
772 osthread->set_state(INITIALIZED);
773 sync->notify_all();
774
775 // wait until os::start_thread()
776 while (osthread->get_state() == INITIALIZED) {
777 sync->wait(Mutex::_no_safepoint_check_flag);
778 }
779 }
780
781 // call one more level start routine
782 thread->run();
783
784 return 0;
785 }
786
787 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
788 assert(thread->osthread() == NULL, "caller responsible");
789
790 // Allocate the OSThread object
791 OSThread* osthread = new OSThread(NULL, NULL);
792 if (osthread == NULL) {
793 return false;
794 }
795
796 // set the correct thread state
797 osthread->set_thread_type(thr_type);
798
799 // Initial state is ALLOCATED but not INITIALIZED
800 osthread->set_state(ALLOCATED);
801
802 thread->set_osthread(osthread);
803
804 // init thread attributes
805 pthread_attr_t attr;
806 pthread_attr_init(&attr);
807 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
808
809 // stack size
810 if (os::Linux::supports_variable_stack_size()) {
811 // calculate stack size if it's not specified by caller
812 if (stack_size == 0) {
813 stack_size = os::Linux::default_stack_size(thr_type);
814
815 switch (thr_type) {
816 case os::java_thread:
817 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
818 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
819 break;
820 case os::compiler_thread:
821 if (CompilerThreadStackSize > 0) {
822 stack_size = (size_t)(CompilerThreadStackSize * K);
823 break;
824 } // else fall through:
825 // use VMThreadStackSize if CompilerThreadStackSize is not defined
826 case os::vm_thread:
827 case os::pgc_thread:
828 case os::cgc_thread:
829 case os::watcher_thread:
830 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
831 break;
832 }
833 }
834
835 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
836 pthread_attr_setstacksize(&attr, stack_size);
837 } else {
838 // let pthread_create() pick the default value.
839 }
840
841 // glibc guard page
842 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
843
844 ThreadState state;
845
846 {
847 // Serialize thread creation if we are running with fixed stack LinuxThreads
848 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
849 if (lock) {
850 os::Linux::createThread_lock()->lock_without_safepoint_check();
851 }
852
853 pthread_t tid;
854 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
855
856 pthread_attr_destroy(&attr);
857
858 if (ret != 0) {
859 if (PrintMiscellaneous && (Verbose || WizardMode)) {
860 perror("pthread_create()");
861 }
862 // Need to clean up stuff we've allocated so far
863 thread->set_osthread(NULL);
864 delete osthread;
865 if (lock) os::Linux::createThread_lock()->unlock();
866 return false;
867 }
868
869 // Store pthread info into the OSThread
870 osthread->set_pthread_id(tid);
871
872 // Wait until child thread is either initialized or aborted
873 {
874 Monitor* sync_with_child = osthread->startThread_lock();
875 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
876 while ((state = osthread->get_state()) == ALLOCATED) {
877 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
878 }
879 }
880
881 if (lock) {
882 os::Linux::createThread_lock()->unlock();
883 }
884 }
885
886 // Aborted due to thread limit being reached
887 if (state == ZOMBIE) {
888 thread->set_osthread(NULL);
889 delete osthread;
890 return false;
891 }
892
893 // The thread is returned suspended (in state INITIALIZED),
894 // and is started higher up in the call chain
895 assert(state == INITIALIZED, "race condition");
896 return true;
897 }
898
899 /////////////////////////////////////////////////////////////////////////////
900 // attach existing thread
901
902 // bootstrap the main thread
903 bool os::create_main_thread(JavaThread* thread) {
904 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
905 return create_attached_thread(thread);
906 }
907
908 bool os::create_attached_thread(JavaThread* thread) {
909 #ifdef ASSERT
910 thread->verify_not_published();
911 #endif
912
913 // Allocate the OSThread object
914 OSThread* osthread = new OSThread(NULL, NULL);
915
916 if (osthread == NULL) {
917 return false;
918 }
919
920 // Store pthread info into the OSThread
921 osthread->set_thread_id(os::Linux::gettid());
922 osthread->set_pthread_id(::pthread_self());
923
924 // initialize floating point control register
925 os::Linux::init_thread_fpu_state();
926
927 // Initial thread state is RUNNABLE
928 osthread->set_state(RUNNABLE);
929
930 thread->set_osthread(osthread);
931
932 if (UseNUMA) {
933 int lgrp_id = os::numa_get_group_id();
934 if (lgrp_id != -1) {
935 thread->set_lgrp_id(lgrp_id);
936 }
937 }
938
939 if (os::Linux::is_initial_thread()) {
940 // If current thread is initial thread, its stack is mapped on demand,
941 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
942 // the entire stack region to avoid SEGV in stack banging.
943 // It is also useful to get around the heap-stack-gap problem on SuSE
944 // kernel (see 4821821 for details). We first expand stack to the top
945 // of yellow zone, then enable stack yellow zone (order is significant,
946 // enabling yellow zone first will crash JVM on SuSE Linux), so there
947 // is no gap between the last two virtual memory regions.
948
949 JavaThread *jt = (JavaThread *)thread;
950 address addr = jt->stack_yellow_zone_base();
951 assert(addr != NULL, "initialization problem?");
952 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
953
954 osthread->set_expanding_stack();
955 os::Linux::manually_expand_stack(jt, addr);
956 osthread->clear_expanding_stack();
957 }
958
959 // initialize signal mask for this thread
960 // and save the caller's signal mask
961 os::Linux::hotspot_sigmask(thread);
962
963 return true;
964 }
965
966 void os::pd_start_thread(Thread* thread) {
967 OSThread * osthread = thread->osthread();
968 assert(osthread->get_state() != INITIALIZED, "just checking");
969 Monitor* sync_with_child = osthread->startThread_lock();
970 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
971 sync_with_child->notify();
972 }
973
974 // Free Linux resources related to the OSThread
975 void os::free_thread(OSThread* osthread) {
976 assert(osthread != NULL, "osthread not set");
977
978 if (Thread::current()->osthread() == osthread) {
979 // Restore caller's signal mask
980 sigset_t sigmask = osthread->caller_sigmask();
981 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
982 }
983
984 delete osthread;
985 }
986
987 //////////////////////////////////////////////////////////////////////////////
988 // thread local storage
989
990 int os::allocate_thread_local_storage() {
991 pthread_key_t key;
992 int rslt = pthread_key_create(&key, NULL);
993 assert(rslt == 0, "cannot allocate thread local storage");
994 return (int)key;
995 }
996
997 // Note: This is currently not used by VM, as we don't destroy TLS key
998 // on VM exit.
999 void os::free_thread_local_storage(int index) {
1000 int rslt = pthread_key_delete((pthread_key_t)index);
1001 assert(rslt == 0, "invalid index");
1002 }
1003
1004 void os::thread_local_storage_at_put(int index, void* value) {
1005 int rslt = pthread_setspecific((pthread_key_t)index, value);
1006 assert(rslt == 0, "pthread_setspecific failed");
1007 }
1008
1009 extern "C" Thread* get_thread() {
1010 return ThreadLocalStorage::thread();
1011 }
1012
1013 //////////////////////////////////////////////////////////////////////////////
1014 // initial thread
1015
1016 // Check if current thread is the initial thread, similar to Solaris thr_main.
1017 bool os::Linux::is_initial_thread(void) {
1018 char dummy;
1019 // If called before init complete, thread stack bottom will be null.
1020 // Can be called if fatal error occurs before initialization.
1021 if (initial_thread_stack_bottom() == NULL) return false;
1022 assert(initial_thread_stack_bottom() != NULL &&
1023 initial_thread_stack_size() != 0,
1024 "os::init did not locate initial thread's stack region");
1025 if ((address)&dummy >= initial_thread_stack_bottom() &&
1026 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1027 return true;
1028 else return false;
1029 }
1030
1031 // Find the virtual memory area that contains addr
1032 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1033 FILE *fp = fopen("/proc/self/maps", "r");
1034 if (fp) {
1035 address low, high;
1036 while (!feof(fp)) {
1037 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1038 if (low <= addr && addr < high) {
1039 if (vma_low) *vma_low = low;
1040 if (vma_high) *vma_high = high;
1041 fclose (fp);
1042 return true;
1043 }
1044 }
1045 for (;;) {
1046 int ch = fgetc(fp);
1047 if (ch == EOF || ch == (int)'\n') break;
1048 }
1049 }
1050 fclose(fp);
1051 }
1052 return false;
1053 }
1054
1055 // Locate initial thread stack. This special handling of initial thread stack
1056 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1057 // bogus value for initial thread.
1058 void os::Linux::capture_initial_stack(size_t max_size) {
1059 // stack size is the easy part, get it from RLIMIT_STACK
1060 size_t stack_size;
1061 struct rlimit rlim;
1062 getrlimit(RLIMIT_STACK, &rlim);
1063 stack_size = rlim.rlim_cur;
1064
1065 // 6308388: a bug in ld.so will relocate its own .data section to the
1066 // lower end of primordial stack; reduce ulimit -s value a little bit
1067 // so we won't install guard page on ld.so's data section.
1068 stack_size -= 2 * page_size();
1069
1070 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1071 // 7.1, in both cases we will get 2G in return value.
1072 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1073 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1074 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1075 // in case other parts in glibc still assumes 2M max stack size.
1076 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1077 #ifndef IA64
1078 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1079 #else
1080 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1081 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1082 #endif
1083
1084 // Try to figure out where the stack base (top) is. This is harder.
1085 //
1086 // When an application is started, glibc saves the initial stack pointer in
1087 // a global variable "__libc_stack_end", which is then used by system
1088 // libraries. __libc_stack_end should be pretty close to stack top. The
1089 // variable is available since the very early days. However, because it is
1090 // a private interface, it could disappear in the future.
1091 //
1092 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1093 // to __libc_stack_end, it is very close to stack top, but isn't the real
1094 // stack top. Note that /proc may not exist if VM is running as a chroot
1095 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1096 // /proc/<pid>/stat could change in the future (though unlikely).
1097 //
1098 // We try __libc_stack_end first. If that doesn't work, look for
1099 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1100 // as a hint, which should work well in most cases.
1101
1102 uintptr_t stack_start;
1103
1104 // try __libc_stack_end first
1105 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1106 if (p && *p) {
1107 stack_start = *p;
1108 } else {
1109 // see if we can get the start_stack field from /proc/self/stat
1110 FILE *fp;
1111 int pid;
1112 char state;
1113 int ppid;
1114 int pgrp;
1115 int session;
1116 int nr;
1117 int tpgrp;
1118 unsigned long flags;
1119 unsigned long minflt;
1120 unsigned long cminflt;
1121 unsigned long majflt;
1122 unsigned long cmajflt;
1123 unsigned long utime;
1124 unsigned long stime;
1125 long cutime;
1126 long cstime;
1127 long prio;
1128 long nice;
1129 long junk;
1130 long it_real;
1131 uintptr_t start;
1132 uintptr_t vsize;
1133 uintptr_t rss;
1134 unsigned long rsslim;
1135 uintptr_t scodes;
1136 uintptr_t ecode;
1137 int i;
1138
1139 // Figure what the primordial thread stack base is. Code is inspired
1140 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1141 // followed by command name surrounded by parentheses, state, etc.
1142 char stat[2048];
1143 int statlen;
1144
1145 fp = fopen("/proc/self/stat", "r");
1146 if (fp) {
1147 statlen = fread(stat, 1, 2047, fp);
1148 stat[statlen] = '\0';
1149 fclose(fp);
1150
1151 // Skip pid and the command string. Note that we could be dealing with
1152 // weird command names, e.g. user could decide to rename java launcher
1153 // to "java 1.4.2 :)", then the stat file would look like
1154 // 1234 (java 1.4.2 :)) R ... ...
1155 // We don't really need to know the command string, just find the last
1156 // occurrence of ")" and then start parsing from there. See bug 4726580.
1157 char * s = strrchr(stat, ')');
1158
1159 i = 0;
1160 if (s) {
1161 // Skip blank chars
1162 do s++; while (isspace(*s));
1163
1164 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1165 /* 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 */
1166 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu",
1167 &state, /* 3 %c */
1168 &ppid, /* 4 %d */
1169 &pgrp, /* 5 %d */
1170 &session, /* 6 %d */
1171 &nr, /* 7 %d */
1172 &tpgrp, /* 8 %d */
1173 &flags, /* 9 %lu */
1174 &minflt, /* 10 %lu */
1175 &cminflt, /* 11 %lu */
1176 &majflt, /* 12 %lu */
1177 &cmajflt, /* 13 %lu */
1178 &utime, /* 14 %lu */
1179 &stime, /* 15 %lu */
1180 &cutime, /* 16 %ld */
1181 &cstime, /* 17 %ld */
1182 &prio, /* 18 %ld */
1183 &nice, /* 19 %ld */
1184 &junk, /* 20 %ld */
1185 &it_real, /* 21 %ld */
1186 &start, /* 22 %lu */
1187 &vsize, /* 23 %lu */
1188 &rss, /* 24 %ld */
1189 &rsslim, /* 25 %lu */
1190 &scodes, /* 26 %lu */
1191 &ecode, /* 27 %lu */
1192 &stack_start); /* 28 %lu */
1193 }
1194
1195 if (i != 28 - 2) {
1196 assert(false, "Bad conversion from /proc/self/stat");
1197 // product mode - assume we are the initial thread, good luck in the
1198 // embedded case.
1199 warning("Can't detect initial thread stack location - bad conversion");
1200 stack_start = (uintptr_t) &rlim;
1201 }
1202 } else {
1203 // For some reason we can't open /proc/self/stat (for example, running on
1204 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1205 // most cases, so don't abort:
1206 warning("Can't detect initial thread stack location - no /proc/self/stat");
1207 stack_start = (uintptr_t) &rlim;
1208 }
1209 }
1210
1211 // Now we have a pointer (stack_start) very close to the stack top, the
1212 // next thing to do is to figure out the exact location of stack top. We
1213 // can find out the virtual memory area that contains stack_start by
1214 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1215 // and its upper limit is the real stack top. (again, this would fail if
1216 // running inside chroot, because /proc may not exist.)
1217
1218 uintptr_t stack_top;
1219 address low, high;
1220 if (find_vma((address)stack_start, &low, &high)) {
1221 // success, "high" is the true stack top. (ignore "low", because initial
1222 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1223 stack_top = (uintptr_t)high;
1224 } else {
1225 // failed, likely because /proc/self/maps does not exist
1226 warning("Can't detect initial thread stack location - find_vma failed");
1227 // best effort: stack_start is normally within a few pages below the real
1228 // stack top, use it as stack top, and reduce stack size so we won't put
1229 // guard page outside stack.
1230 stack_top = stack_start;
1231 stack_size -= 16 * page_size();
1232 }
1233
1234 // stack_top could be partially down the page so align it
1235 stack_top = align_size_up(stack_top, page_size());
1236
1237 if (max_size && stack_size > max_size) {
1238 _initial_thread_stack_size = max_size;
1239 } else {
1240 _initial_thread_stack_size = stack_size;
1241 }
1242
1243 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1244 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1245 }
1246
1247 ////////////////////////////////////////////////////////////////////////////////
1248 // time support
1249
1250 // Time since start-up in seconds to a fine granularity.
1251 // Used by VMSelfDestructTimer and the MemProfiler.
1252 double os::elapsedTime() {
1253
1254 return (double)(os::elapsed_counter()) * 0.000001;
1255 }
1256
1257 jlong os::elapsed_counter() {
1258 timeval time;
1259 int status = gettimeofday(&time, NULL);
1260 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1261 }
1262
1263 jlong os::elapsed_frequency() {
1264 return (1000 * 1000);
1265 }
1266
1267 // For now, we say that linux does not support vtime. I have no idea
1268 // whether it can actually be made to (DLD, 9/13/05).
1269
1270 bool os::supports_vtime() { return false; }
1271 bool os::enable_vtime() { return false; }
1272 bool os::vtime_enabled() { return false; }
1273 double os::elapsedVTime() {
1274 // better than nothing, but not much
1275 return elapsedTime();
1276 }
1277
1278 jlong os::javaTimeMillis() {
1279 timeval time;
1280 int status = gettimeofday(&time, NULL);
1281 assert(status != -1, "linux error");
1282 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1283 }
1284
1285 #ifndef CLOCK_MONOTONIC
1286 #define CLOCK_MONOTONIC (1)
1287 #endif
1288
1289 void os::Linux::clock_init() {
1290 // we do dlopen's in this particular order due to bug in linux
1291 // dynamical loader (see 6348968) leading to crash on exit
1292 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1293 if (handle == NULL) {
1294 handle = dlopen("librt.so", RTLD_LAZY);
1295 }
1296
1297 if (handle) {
1298 int (*clock_getres_func)(clockid_t, struct timespec*) =
1299 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1300 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1301 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1302 if (clock_getres_func && clock_gettime_func) {
1303 // See if monotonic clock is supported by the kernel. Note that some
1304 // early implementations simply return kernel jiffies (updated every
1305 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1306 // for nano time (though the monotonic property is still nice to have).
1307 // It's fixed in newer kernels, however clock_getres() still returns
1308 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1309 // resolution for now. Hopefully as people move to new kernels, this
1310 // won't be a problem.
1311 struct timespec res;
1312 struct timespec tp;
1313 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1314 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1315 // yes, monotonic clock is supported
1316 _clock_gettime = clock_gettime_func;
1317 } else {
1318 // close librt if there is no monotonic clock
1319 dlclose(handle);
1320 }
1321 }
1322 }
1323 }
1324
1325 #ifndef SYS_clock_getres
1326
1327 #if defined(IA32) || defined(AMD64)
1328 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1329 #else
1330 #error Value of SYS_clock_getres not known on this platform
1331 #endif
1332
1333 #endif
1334
1335 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1336
1337 void os::Linux::fast_thread_clock_init() {
1338 if (!UseLinuxPosixThreadCPUClocks) {
1339 return;
1340 }
1341 clockid_t clockid;
1342 struct timespec tp;
1343 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1344 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1345
1346 // Switch to using fast clocks for thread cpu time if
1347 // the sys_clock_getres() returns 0 error code.
1348 // Note, that some kernels may support the current thread
1349 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1350 // returned by the pthread_getcpuclockid().
1351 // If the fast Posix clocks are supported then the sys_clock_getres()
1352 // must return at least tp.tv_sec == 0 which means a resolution
1353 // better than 1 sec. This is extra check for reliability.
1354
1355 if(pthread_getcpuclockid_func &&
1356 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1357 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1358
1359 _supports_fast_thread_cpu_time = true;
1360 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1361 }
1362 }
1363
1364 jlong os::javaTimeNanos() {
1365 if (Linux::supports_monotonic_clock()) {
1366 struct timespec tp;
1367 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1368 assert(status == 0, "gettime error");
1369 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1370 return result;
1371 } else {
1372 timeval time;
1373 int status = gettimeofday(&time, NULL);
1374 assert(status != -1, "linux error");
1375 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1376 return 1000 * usecs;
1377 }
1378 }
1379
1380 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1381 if (Linux::supports_monotonic_clock()) {
1382 info_ptr->max_value = ALL_64_BITS;
1383
1384 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1385 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1386 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1387 } else {
1388 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1389 info_ptr->max_value = ALL_64_BITS;
1390
1391 // gettimeofday is a real time clock so it skips
1392 info_ptr->may_skip_backward = true;
1393 info_ptr->may_skip_forward = true;
1394 }
1395
1396 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1397 }
1398
1399 // Return the real, user, and system times in seconds from an
1400 // arbitrary fixed point in the past.
1401 bool os::getTimesSecs(double* process_real_time,
1402 double* process_user_time,
1403 double* process_system_time) {
1404 struct tms ticks;
1405 clock_t real_ticks = times(&ticks);
1406
1407 if (real_ticks == (clock_t) (-1)) {
1408 return false;
1409 } else {
1410 double ticks_per_second = (double) clock_tics_per_sec;
1411 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1412 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1413 *process_real_time = ((double) real_ticks) / ticks_per_second;
1414
1415 return true;
1416 }
1417 }
1418
1419
1420 char * os::local_time_string(char *buf, size_t buflen) {
1421 struct tm t;
1422 time_t long_time;
1423 time(&long_time);
1424 localtime_r(&long_time, &t);
1425 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1426 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1427 t.tm_hour, t.tm_min, t.tm_sec);
1428 return buf;
1429 }
1430
1431 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1432 return localtime_r(clock, res);
1433 }
1434
1435 ////////////////////////////////////////////////////////////////////////////////
1436 // runtime exit support
1437
1438 // Note: os::shutdown() might be called very early during initialization, or
1439 // called from signal handler. Before adding something to os::shutdown(), make
1440 // sure it is async-safe and can handle partially initialized VM.
1441 void os::shutdown() {
1442
1443 // allow PerfMemory to attempt cleanup of any persistent resources
1444 perfMemory_exit();
1445
1446 // needs to remove object in file system
1447 AttachListener::abort();
1448
1449 // flush buffered output, finish log files
1450 ostream_abort();
1451
1452 // Check for abort hook
1453 abort_hook_t abort_hook = Arguments::abort_hook();
1454 if (abort_hook != NULL) {
1455 abort_hook();
1456 }
1457
1458 }
1459
1460 // Note: os::abort() might be called very early during initialization, or
1461 // called from signal handler. Before adding something to os::abort(), make
1462 // sure it is async-safe and can handle partially initialized VM.
1463 void os::abort(bool dump_core) {
1464 os::shutdown();
1465 if (dump_core) {
1466 #ifndef PRODUCT
1467 fdStream out(defaultStream::output_fd());
1468 out.print_raw("Current thread is ");
1469 char buf[16];
1470 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1471 out.print_raw_cr(buf);
1472 out.print_raw_cr("Dumping core ...");
1473 #endif
1474 ::abort(); // dump core
1475 }
1476
1477 ::exit(1);
1478 }
1479
1480 // Die immediately, no exit hook, no abort hook, no cleanup.
1481 void os::die() {
1482 // _exit() on LinuxThreads only kills current thread
1483 ::abort();
1484 }
1485
1486 // unused on linux for now.
1487 void os::set_error_file(const char *logfile) {}
1488
1489 intx os::current_thread_id() { return (intx)pthread_self(); }
1490 int os::current_process_id() {
1491
1492 // Under the old linux thread library, linux gives each thread
1493 // its own process id. Because of this each thread will return
1494 // a different pid if this method were to return the result
1495 // of getpid(2). Linux provides no api that returns the pid
1496 // of the launcher thread for the vm. This implementation
1497 // returns a unique pid, the pid of the launcher thread
1498 // that starts the vm 'process'.
1499
1500 // Under the NPTL, getpid() returns the same pid as the
1501 // launcher thread rather than a unique pid per thread.
1502 // Use gettid() if you want the old pre NPTL behaviour.
1503
1504 // if you are looking for the result of a call to getpid() that
1505 // returns a unique pid for the calling thread, then look at the
1506 // OSThread::thread_id() method in osThread_linux.hpp file
1507
1508 return (int)(_initial_pid ? _initial_pid : getpid());
1509 }
1510
1511 // DLL functions
1512
1513 const char* os::dll_file_extension() { return ".so"; }
1514
1515 const char* os::get_temp_directory() { return "/tmp/"; }
1516
1517 static bool file_exists(const char* filename) {
1518 struct stat statbuf;
1519 if (filename == NULL || strlen(filename) == 0) {
1520 return false;
1521 }
1522 return os::stat(filename, &statbuf) == 0;
1523 }
1524
1525 void os::dll_build_name(char* buffer, size_t buflen,
1526 const char* pname, const char* fname) {
1527 // Copied from libhpi
1528 const size_t pnamelen = pname ? strlen(pname) : 0;
1529
1530 // Quietly truncate on buffer overflow. Should be an error.
1531 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1532 *buffer = '\0';
1533 return;
1534 }
1535
1536 if (pnamelen == 0) {
1537 snprintf(buffer, buflen, "lib%s.so", fname);
1538 } else if (strchr(pname, *os::path_separator()) != NULL) {
1539 int n;
1540 char** pelements = split_path(pname, &n);
1541 for (int i = 0 ; i < n ; i++) {
1542 // Really shouldn't be NULL, but check can't hurt
1543 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1544 continue; // skip the empty path values
1545 }
1546 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1547 if (file_exists(buffer)) {
1548 break;
1549 }
1550 }
1551 // release the storage
1552 for (int i = 0 ; i < n ; i++) {
1553 if (pelements[i] != NULL) {
1554 FREE_C_HEAP_ARRAY(char, pelements[i]);
1555 }
1556 }
1557 if (pelements != NULL) {
1558 FREE_C_HEAP_ARRAY(char*, pelements);
1559 }
1560 } else {
1561 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1562 }
1563 }
1564
1565 const char* os::get_current_directory(char *buf, int buflen) {
1566 return getcwd(buf, buflen);
1567 }
1568
1569 // check if addr is inside libjvm[_g].so
1570 bool os::address_is_in_vm(address addr) {
1571 static address libjvm_base_addr;
1572 Dl_info dlinfo;
1573
1574 if (libjvm_base_addr == NULL) {
1575 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1576 libjvm_base_addr = (address)dlinfo.dli_fbase;
1577 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1578 }
1579
1580 if (dladdr((void *)addr, &dlinfo)) {
1581 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1582 }
1583
1584 return false;
1585 }
1586
1587 bool os::dll_address_to_function_name(address addr, char *buf,
1588 int buflen, int *offset) {
1589 Dl_info dlinfo;
1590
1591 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1592 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1593 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1594 return true;
1595 } else {
1596 if (buf) buf[0] = '\0';
1597 if (offset) *offset = -1;
1598 return false;
1599 }
1600 }
1601
1602 struct _address_to_library_name {
1603 address addr; // input : memory address
1604 size_t buflen; // size of fname
1605 char* fname; // output: library name
1606 address base; // library base addr
1607 };
1608
1609 static int address_to_library_name_callback(struct dl_phdr_info *info,
1610 size_t size, void *data) {
1611 int i;
1612 bool found = false;
1613 address libbase = NULL;
1614 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1615
1616 // iterate through all loadable segments
1617 for (i = 0; i < info->dlpi_phnum; i++) {
1618 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1619 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1620 // base address of a library is the lowest address of its loaded
1621 // segments.
1622 if (libbase == NULL || libbase > segbase) {
1623 libbase = segbase;
1624 }
1625 // see if 'addr' is within current segment
1626 if (segbase <= d->addr &&
1627 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1628 found = true;
1629 }
1630 }
1631 }
1632
1633 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1634 // so dll_address_to_library_name() can fall through to use dladdr() which
1635 // can figure out executable name from argv[0].
1636 if (found && info->dlpi_name && info->dlpi_name[0]) {
1637 d->base = libbase;
1638 if (d->fname) {
1639 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1640 }
1641 return 1;
1642 }
1643 return 0;
1644 }
1645
1646 bool os::dll_address_to_library_name(address addr, char* buf,
1647 int buflen, int* offset) {
1648 Dl_info dlinfo;
1649 struct _address_to_library_name data;
1650
1651 // There is a bug in old glibc dladdr() implementation that it could resolve
1652 // to wrong library name if the .so file has a base address != NULL. Here
1653 // we iterate through the program headers of all loaded libraries to find
1654 // out which library 'addr' really belongs to. This workaround can be
1655 // removed once the minimum requirement for glibc is moved to 2.3.x.
1656 data.addr = addr;
1657 data.fname = buf;
1658 data.buflen = buflen;
1659 data.base = NULL;
1660 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1661
1662 if (rslt) {
1663 // buf already contains library name
1664 if (offset) *offset = addr - data.base;
1665 return true;
1666 } else if (dladdr((void*)addr, &dlinfo)){
1667 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1668 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1669 return true;
1670 } else {
1671 if (buf) buf[0] = '\0';
1672 if (offset) *offset = -1;
1673 return false;
1674 }
1675 }
1676
1677 // Loads .dll/.so and
1678 // in case of error it checks if .dll/.so was built for the
1679 // same architecture as Hotspot is running on
1680
1681 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1682 {
1683 void * result= ::dlopen(filename, RTLD_LAZY);
1684 if (result != NULL) {
1685 // Successful loading
1686 return result;
1687 }
1688
1689 Elf32_Ehdr elf_head;
1690
1691 // Read system error message into ebuf
1692 // It may or may not be overwritten below
1693 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1694 ebuf[ebuflen-1]='\0';
1695 int diag_msg_max_length=ebuflen-strlen(ebuf);
1696 char* diag_msg_buf=ebuf+strlen(ebuf);
1697
1698 if (diag_msg_max_length==0) {
1699 // No more space in ebuf for additional diagnostics message
1700 return NULL;
1701 }
1702
1703
1704 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1705
1706 if (file_descriptor < 0) {
1707 // Can't open library, report dlerror() message
1708 return NULL;
1709 }
1710
1711 bool failed_to_read_elf_head=
1712 (sizeof(elf_head)!=
1713 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1714
1715 ::close(file_descriptor);
1716 if (failed_to_read_elf_head) {
1717 // file i/o error - report dlerror() msg
1718 return NULL;
1719 }
1720
1721 typedef struct {
1722 Elf32_Half code; // Actual value as defined in elf.h
1723 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1724 char elf_class; // 32 or 64 bit
1725 char endianess; // MSB or LSB
1726 char* name; // String representation
1727 } arch_t;
1728
1729 #ifndef EM_486
1730 #define EM_486 6 /* Intel 80486 */
1731 #endif
1732
1733 static const arch_t arch_array[]={
1734 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1737 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1738 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1741 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1742 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1743 };
1744
1745 #if (defined IA32)
1746 static Elf32_Half running_arch_code=EM_386;
1747 #elif (defined AMD64)
1748 static Elf32_Half running_arch_code=EM_X86_64;
1749 #elif (defined IA64)
1750 static Elf32_Half running_arch_code=EM_IA_64;
1751 #elif (defined __sparc) && (defined _LP64)
1752 static Elf32_Half running_arch_code=EM_SPARCV9;
1753 #elif (defined __sparc) && (!defined _LP64)
1754 static Elf32_Half running_arch_code=EM_SPARC;
1755 #elif (defined __powerpc64__)
1756 static Elf32_Half running_arch_code=EM_PPC64;
1757 #elif (defined __powerpc__)
1758 static Elf32_Half running_arch_code=EM_PPC;
1759 #else
1760 #error Method os::dll_load requires that one of following is defined:\
1761 IA32, AMD64, IA64, __sparc, __powerpc__
1762 #endif
1763
1764 // Identify compatability class for VM's architecture and library's architecture
1765 // Obtain string descriptions for architectures
1766
1767 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1768 int running_arch_index=-1;
1769
1770 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1771 if (running_arch_code == arch_array[i].code) {
1772 running_arch_index = i;
1773 }
1774 if (lib_arch.code == arch_array[i].code) {
1775 lib_arch.compat_class = arch_array[i].compat_class;
1776 lib_arch.name = arch_array[i].name;
1777 }
1778 }
1779
1780 assert(running_arch_index != -1,
1781 "Didn't find running architecture code (running_arch_code) in arch_array");
1782 if (running_arch_index == -1) {
1783 // Even though running architecture detection failed
1784 // we may still continue with reporting dlerror() message
1785 return NULL;
1786 }
1787
1788 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1789 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1790 return NULL;
1791 }
1792
1793 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1794 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1795 return NULL;
1796 }
1797
1798 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1799 if ( lib_arch.name!=NULL ) {
1800 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1801 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1802 lib_arch.name, arch_array[running_arch_index].name);
1803 } else {
1804 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1805 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1806 lib_arch.code,
1807 arch_array[running_arch_index].name);
1808 }
1809 }
1810
1811 return NULL;
1812 }
1813
1814 /*
1815 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1816 * chances are you might want to run the generated bits against glibc-2.0
1817 * libdl.so, so always use locking for any version of glibc.
1818 */
1819 void* os::dll_lookup(void* handle, const char* name) {
1820 pthread_mutex_lock(&dl_mutex);
1821 void* res = dlsym(handle, name);
1822 pthread_mutex_unlock(&dl_mutex);
1823 return res;
1824 }
1825
1826
1827 bool _print_ascii_file(const char* filename, outputStream* st) {
1828 int fd = open(filename, O_RDONLY);
1829 if (fd == -1) {
1830 return false;
1831 }
1832
1833 char buf[32];
1834 int bytes;
1835 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1836 st->print_raw(buf, bytes);
1837 }
1838
1839 close(fd);
1840
1841 return true;
1842 }
1843
1844 void os::print_dll_info(outputStream *st) {
1845 st->print_cr("Dynamic libraries:");
1846
1847 char fname[32];
1848 pid_t pid = os::Linux::gettid();
1849
1850 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1851
1852 if (!_print_ascii_file(fname, st)) {
1853 st->print("Can not get library information for pid = %d\n", pid);
1854 }
1855 }
1856
1857
1858 void os::print_os_info(outputStream* st) {
1859 st->print("OS:");
1860
1861 // Try to identify popular distros.
1862 // Most Linux distributions have /etc/XXX-release file, which contains
1863 // the OS version string. Some have more than one /etc/XXX-release file
1864 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1865 // so the order is important.
1866 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1867 !_print_ascii_file("/etc/sun-release", st) &&
1868 !_print_ascii_file("/etc/redhat-release", st) &&
1869 !_print_ascii_file("/etc/SuSE-release", st) &&
1870 !_print_ascii_file("/etc/turbolinux-release", st) &&
1871 !_print_ascii_file("/etc/gentoo-release", st) &&
1872 !_print_ascii_file("/etc/debian_version", st)) {
1873 st->print("Linux");
1874 }
1875 st->cr();
1876
1877 // kernel
1878 st->print("uname:");
1879 struct utsname name;
1880 uname(&name);
1881 st->print(name.sysname); st->print(" ");
1882 st->print(name.release); st->print(" ");
1883 st->print(name.version); st->print(" ");
1884 st->print(name.machine);
1885 st->cr();
1886
1887 // Print warning if unsafe chroot environment detected
1888 if (unsafe_chroot_detected) {
1889 st->print("WARNING!! ");
1890 st->print_cr(unstable_chroot_error);
1891 }
1892
1893 // libc, pthread
1894 st->print("libc:");
1895 st->print(os::Linux::glibc_version()); st->print(" ");
1896 st->print(os::Linux::libpthread_version()); st->print(" ");
1897 if (os::Linux::is_LinuxThreads()) {
1898 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1899 }
1900 st->cr();
1901
1902 // rlimit
1903 st->print("rlimit:");
1904 struct rlimit rlim;
1905
1906 st->print(" STACK ");
1907 getrlimit(RLIMIT_STACK, &rlim);
1908 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1909 else st->print("%uk", rlim.rlim_cur >> 10);
1910
1911 st->print(", CORE ");
1912 getrlimit(RLIMIT_CORE, &rlim);
1913 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1914 else st->print("%uk", rlim.rlim_cur >> 10);
1915
1916 st->print(", NPROC ");
1917 getrlimit(RLIMIT_NPROC, &rlim);
1918 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1919 else st->print("%d", rlim.rlim_cur);
1920
1921 st->print(", NOFILE ");
1922 getrlimit(RLIMIT_NOFILE, &rlim);
1923 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1924 else st->print("%d", rlim.rlim_cur);
1925
1926 st->print(", AS ");
1927 getrlimit(RLIMIT_AS, &rlim);
1928 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1929 else st->print("%uk", rlim.rlim_cur >> 10);
1930 st->cr();
1931
1932 // load average
1933 st->print("load average:");
1934 double loadavg[3];
1935 os::loadavg(loadavg, 3);
1936 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1937 st->cr();
1938 }
1939
1940 void os::print_memory_info(outputStream* st) {
1941
1942 st->print("Memory:");
1943 st->print(" %dk page", os::vm_page_size()>>10);
1944
1945 // values in struct sysinfo are "unsigned long"
1946 struct sysinfo si;
1947 sysinfo(&si);
1948
1949 st->print(", physical " UINT64_FORMAT "k",
1950 os::physical_memory() >> 10);
1951 st->print("(" UINT64_FORMAT "k free)",
1952 os::available_memory() >> 10);
1953 st->print(", swap " UINT64_FORMAT "k",
1954 ((jlong)si.totalswap * si.mem_unit) >> 10);
1955 st->print("(" UINT64_FORMAT "k free)",
1956 ((jlong)si.freeswap * si.mem_unit) >> 10);
1957 st->cr();
1958 }
1959
1960 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
1961 // but they're the same for all the linux arch that we support
1962 // and they're the same for solaris but there's no common place to put this.
1963 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1964 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1965 "ILL_COPROC", "ILL_BADSTK" };
1966
1967 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1968 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1969 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1970
1971 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1972
1973 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1974
1975 void os::print_siginfo(outputStream* st, void* siginfo) {
1976 st->print("siginfo:");
1977
1978 const int buflen = 100;
1979 char buf[buflen];
1980 siginfo_t *si = (siginfo_t*)siginfo;
1981 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1982 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1983 st->print("si_errno=%s", buf);
1984 } else {
1985 st->print("si_errno=%d", si->si_errno);
1986 }
1987 const int c = si->si_code;
1988 assert(c > 0, "unexpected si_code");
1989 switch (si->si_signo) {
1990 case SIGILL:
1991 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1992 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1993 break;
1994 case SIGFPE:
1995 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1996 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1997 break;
1998 case SIGSEGV:
1999 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2000 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2001 break;
2002 case SIGBUS:
2003 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2004 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2005 break;
2006 default:
2007 st->print(", si_code=%d", si->si_code);
2008 // no si_addr
2009 }
2010
2011 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2012 UseSharedSpaces) {
2013 FileMapInfo* mapinfo = FileMapInfo::current_info();
2014 if (mapinfo->is_in_shared_space(si->si_addr)) {
2015 st->print("\n\nError accessing class data sharing archive." \
2016 " Mapped file inaccessible during execution, " \
2017 " possible disk/network problem.");
2018 }
2019 }
2020 st->cr();
2021 }
2022
2023
2024 static void print_signal_handler(outputStream* st, int sig,
2025 char* buf, size_t buflen);
2026
2027 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2028 st->print_cr("Signal Handlers:");
2029 print_signal_handler(st, SIGSEGV, buf, buflen);
2030 print_signal_handler(st, SIGBUS , buf, buflen);
2031 print_signal_handler(st, SIGFPE , buf, buflen);
2032 print_signal_handler(st, SIGPIPE, buf, buflen);
2033 print_signal_handler(st, SIGXFSZ, buf, buflen);
2034 print_signal_handler(st, SIGILL , buf, buflen);
2035 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2036 print_signal_handler(st, SR_signum, buf, buflen);
2037 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2038 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2039 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2040 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2041 }
2042
2043 static char saved_jvm_path[MAXPATHLEN] = {0};
2044
2045 // Find the full path to the current module, libjvm.so or libjvm_g.so
2046 void os::jvm_path(char *buf, jint len) {
2047 // Error checking.
2048 if (len < MAXPATHLEN) {
2049 assert(false, "must use a large-enough buffer");
2050 buf[0] = '\0';
2051 return;
2052 }
2053 // Lazy resolve the path to current module.
2054 if (saved_jvm_path[0] != 0) {
2055 strcpy(buf, saved_jvm_path);
2056 return;
2057 }
2058
2059 char dli_fname[MAXPATHLEN];
2060 bool ret = dll_address_to_library_name(
2061 CAST_FROM_FN_PTR(address, os::jvm_path),
2062 dli_fname, sizeof(dli_fname), NULL);
2063 assert(ret != 0, "cannot locate libjvm");
2064 realpath(dli_fname, buf);
2065
2066 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2067 // Support for the gamma launcher. Typical value for buf is
2068 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2069 // the right place in the string, then assume we are installed in a JDK and
2070 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2071 // up the path so it looks like libjvm.so is installed there (append a
2072 // fake suffix hotspot/libjvm.so).
2073 const char *p = buf + strlen(buf) - 1;
2074 for (int count = 0; p > buf && count < 5; ++count) {
2075 for (--p; p > buf && *p != '/'; --p)
2076 /* empty */ ;
2077 }
2078
2079 if (strncmp(p, "/jre/lib/", 9) != 0) {
2080 // Look for JAVA_HOME in the environment.
2081 char* java_home_var = ::getenv("JAVA_HOME");
2082 if (java_home_var != NULL && java_home_var[0] != 0) {
2083 // Check the current module name "libjvm.so" or "libjvm_g.so".
2084 p = strrchr(buf, '/');
2085 assert(strstr(p, "/libjvm") == p, "invalid library name");
2086 p = strstr(p, "_g") ? "_g" : "";
2087
2088 realpath(java_home_var, buf);
2089 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2090 if (0 == access(buf, F_OK)) {
2091 // Use current module name "libjvm[_g].so" instead of
2092 // "libjvm"debug_only("_g")".so" since for fastdebug version
2093 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2094 // It is used when we are choosing the HPI library's name
2095 // "libhpi[_g].so" in hpi::initialize_get_interface().
2096 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2097 } else {
2098 // Go back to path of .so
2099 realpath(dli_fname, buf);
2100 }
2101 }
2102 }
2103 }
2104
2105 strcpy(saved_jvm_path, buf);
2106 }
2107
2108 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2109 // no prefix required, not even "_"
2110 }
2111
2112 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2113 // no suffix required
2114 }
2115
2116 ////////////////////////////////////////////////////////////////////////////////
2117 // sun.misc.Signal support
2118
2119 static volatile jint sigint_count = 0;
2120
2121 static void
2122 UserHandler(int sig, void *siginfo, void *context) {
2123 // 4511530 - sem_post is serialized and handled by the manager thread. When
2124 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2125 // don't want to flood the manager thread with sem_post requests.
2126 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2127 return;
2128
2129 // Ctrl-C is pressed during error reporting, likely because the error
2130 // handler fails to abort. Let VM die immediately.
2131 if (sig == SIGINT && is_error_reported()) {
2132 os::die();
2133 }
2134
2135 os::signal_notify(sig);
2136 }
2137
2138 void* os::user_handler() {
2139 return CAST_FROM_FN_PTR(void*, UserHandler);
2140 }
2141
2142 extern "C" {
2143 typedef void (*sa_handler_t)(int);
2144 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2145 }
2146
2147 void* os::signal(int signal_number, void* handler) {
2148 struct sigaction sigAct, oldSigAct;
2149
2150 sigfillset(&(sigAct.sa_mask));
2151 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2152 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2153
2154 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2155 // -1 means registration failed
2156 return (void *)-1;
2157 }
2158
2159 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2160 }
2161
2162 void os::signal_raise(int signal_number) {
2163 ::raise(signal_number);
2164 }
2165
2166 /*
2167 * The following code is moved from os.cpp for making this
2168 * code platform specific, which it is by its very nature.
2169 */
2170
2171 // Will be modified when max signal is changed to be dynamic
2172 int os::sigexitnum_pd() {
2173 return NSIG;
2174 }
2175
2176 // a counter for each possible signal value
2177 static volatile jint pending_signals[NSIG+1] = { 0 };
2178
2179 // Linux(POSIX) specific hand shaking semaphore.
2180 static sem_t sig_sem;
2181
2182 void os::signal_init_pd() {
2183 // Initialize signal structures
2184 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2185
2186 // Initialize signal semaphore
2187 ::sem_init(&sig_sem, 0, 0);
2188 }
2189
2190 void os::signal_notify(int sig) {
2191 Atomic::inc(&pending_signals[sig]);
2192 ::sem_post(&sig_sem);
2193 }
2194
2195 static int check_pending_signals(bool wait) {
2196 Atomic::store(0, &sigint_count);
2197 for (;;) {
2198 for (int i = 0; i < NSIG + 1; i++) {
2199 jint n = pending_signals[i];
2200 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2201 return i;
2202 }
2203 }
2204 if (!wait) {
2205 return -1;
2206 }
2207 JavaThread *thread = JavaThread::current();
2208 ThreadBlockInVM tbivm(thread);
2209
2210 bool threadIsSuspended;
2211 do {
2212 thread->set_suspend_equivalent();
2213 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2214 ::sem_wait(&sig_sem);
2215
2216 // were we externally suspended while we were waiting?
2217 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2218 if (threadIsSuspended) {
2219 //
2220 // The semaphore has been incremented, but while we were waiting
2221 // another thread suspended us. We don't want to continue running
2222 // while suspended because that would surprise the thread that
2223 // suspended us.
2224 //
2225 ::sem_post(&sig_sem);
2226
2227 thread->java_suspend_self();
2228 }
2229 } while (threadIsSuspended);
2230 }
2231 }
2232
2233 int os::signal_lookup() {
2234 return check_pending_signals(false);
2235 }
2236
2237 int os::signal_wait() {
2238 return check_pending_signals(true);
2239 }
2240
2241 ////////////////////////////////////////////////////////////////////////////////
2242 // Virtual Memory
2243
2244 int os::vm_page_size() {
2245 // Seems redundant as all get out
2246 assert(os::Linux::page_size() != -1, "must call os::init");
2247 return os::Linux::page_size();
2248 }
2249
2250 // Solaris allocates memory by pages.
2251 int os::vm_allocation_granularity() {
2252 assert(os::Linux::page_size() != -1, "must call os::init");
2253 return os::Linux::page_size();
2254 }
2255
2256 // Rationale behind this function:
2257 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2258 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2259 // samples for JITted code. Here we create private executable mapping over the code cache
2260 // and then we can use standard (well, almost, as mapping can change) way to provide
2261 // info for the reporting script by storing timestamp and location of symbol
2262 void linux_wrap_code(char* base, size_t size) {
2263 static volatile jint cnt = 0;
2264
2265 if (!UseOprofile) {
2266 return;
2267 }
2268
2269 char buf[40];
2270 int num = Atomic::add(1, &cnt);
2271
2272 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2273 unlink(buf);
2274
2275 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2276
2277 if (fd != -1) {
2278 off_t rv = lseek(fd, size-2, SEEK_SET);
2279 if (rv != (off_t)-1) {
2280 if (write(fd, "", 1) == 1) {
2281 mmap(base, size,
2282 PROT_READ|PROT_WRITE|PROT_EXEC,
2283 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2284 }
2285 }
2286 close(fd);
2287 unlink(buf);
2288 }
2289 }
2290
2291 // NOTE: Linux kernel does not really reserve the pages for us.
2292 // All it does is to check if there are enough free pages
2293 // left at the time of mmap(). This could be a potential
2294 // problem.
2295 bool os::commit_memory(char* addr, size_t size) {
2296 uintptr_t res = (uintptr_t) ::mmap(addr, size,
2297 PROT_READ|PROT_WRITE|PROT_EXEC,
2298 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2299 return res != (uintptr_t) MAP_FAILED;
2300 }
2301
2302 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) {
2303 return commit_memory(addr, size);
2304 }
2305
2306 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2307
2308 void os::free_memory(char *addr, size_t bytes) {
2309 uncommit_memory(addr, bytes);
2310 }
2311
2312 void os::numa_make_global(char *addr, size_t bytes) {
2313 Linux::numa_interleave_memory(addr, bytes);
2314 }
2315
2316 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2317 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2318 }
2319
2320 bool os::numa_topology_changed() { return false; }
2321
2322 size_t os::numa_get_groups_num() {
2323 int max_node = Linux::numa_max_node();
2324 return max_node > 0 ? max_node + 1 : 1;
2325 }
2326
2327 int os::numa_get_group_id() {
2328 int cpu_id = Linux::sched_getcpu();
2329 if (cpu_id != -1) {
2330 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2331 if (lgrp_id != -1) {
2332 return lgrp_id;
2333 }
2334 }
2335 return 0;
2336 }
2337
2338 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2339 for (size_t i = 0; i < size; i++) {
2340 ids[i] = i;
2341 }
2342 return size;
2343 }
2344
2345 bool os::get_page_info(char *start, page_info* info) {
2346 return false;
2347 }
2348
2349 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2350 return end;
2351 }
2352
2353 extern "C" void numa_warn(int number, char *where, ...) { }
2354 extern "C" void numa_error(char *where) { }
2355
2356 bool os::Linux::libnuma_init() {
2357 // sched_getcpu() should be in libc.
2358 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2359 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2360
2361 if (sched_getcpu() != -1) { // Does it work?
2362 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2363 if (handle != NULL) {
2364 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2365 dlsym(handle, "numa_node_to_cpus")));
2366 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2367 dlsym(handle, "numa_max_node")));
2368 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2369 dlsym(handle, "numa_available")));
2370 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2371 dlsym(handle, "numa_tonode_memory")));
2372 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2373 dlsym(handle, "numa_interleave_memory")));
2374
2375
2376 if (numa_available() != -1) {
2377 set_numa_all_nodes((unsigned long*)dlsym(handle, "numa_all_nodes"));
2378 // Create a cpu -> node mapping
2379 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2380 rebuild_cpu_to_node_map();
2381 return true;
2382 }
2383 }
2384 }
2385 return false;
2386 }
2387
2388 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2389 // The table is later used in get_node_by_cpu().
2390 void os::Linux::rebuild_cpu_to_node_map() {
2391 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2392 // in libnuma (possible values are starting from 16,
2393 // and continuing up with every other power of 2, but less
2394 // than the maximum number of CPUs supported by kernel), and
2395 // is a subject to change (in libnuma version 2 the requirements
2396 // are more reasonable) we'll just hardcode the number they use
2397 // in the library.
2398 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2399
2400 size_t cpu_num = os::active_processor_count();
2401 size_t cpu_map_size = NCPUS / BitsPerCLong;
2402 size_t cpu_map_valid_size =
2403 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2404
2405 cpu_to_node()->clear();
2406 cpu_to_node()->at_grow(cpu_num - 1);
2407 size_t node_num = numa_get_groups_num();
2408
2409 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2410 for (size_t i = 0; i < node_num; i++) {
2411 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2412 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2413 if (cpu_map[j] != 0) {
2414 for (size_t k = 0; k < BitsPerCLong; k++) {
2415 if (cpu_map[j] & (1UL << k)) {
2416 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2417 }
2418 }
2419 }
2420 }
2421 }
2422 }
2423 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2424 }
2425
2426 int os::Linux::get_node_by_cpu(int cpu_id) {
2427 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2428 return cpu_to_node()->at(cpu_id);
2429 }
2430 return -1;
2431 }
2432
2433 GrowableArray<int>* os::Linux::_cpu_to_node;
2434 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2435 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2436 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2437 os::Linux::numa_available_func_t os::Linux::_numa_available;
2438 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2439 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2440 unsigned long* os::Linux::_numa_all_nodes;
2441
2442 bool os::uncommit_memory(char* addr, size_t size) {
2443 return ::mmap(addr, size,
2444 PROT_READ|PROT_WRITE|PROT_EXEC,
2445 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2446 != MAP_FAILED;
2447 }
2448
2449 static address _highest_vm_reserved_address = NULL;
2450
2451 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2452 // at 'requested_addr'. If there are existing memory mappings at the same
2453 // location, however, they will be overwritten. If 'fixed' is false,
2454 // 'requested_addr' is only treated as a hint, the return value may or
2455 // may not start from the requested address. Unlike Linux mmap(), this
2456 // function returns NULL to indicate failure.
2457 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2458 char * addr;
2459 int flags;
2460
2461 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2462 if (fixed) {
2463 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2464 flags |= MAP_FIXED;
2465 }
2466
2467 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
2468 flags, -1, 0);
2469
2470 if (addr != MAP_FAILED) {
2471 // anon_mmap() should only get called during VM initialization,
2472 // don't need lock (actually we can skip locking even it can be called
2473 // from multiple threads, because _highest_vm_reserved_address is just a
2474 // hint about the upper limit of non-stack memory regions.)
2475 if ((address)addr + bytes > _highest_vm_reserved_address) {
2476 _highest_vm_reserved_address = (address)addr + bytes;
2477 }
2478 }
2479
2480 return addr == MAP_FAILED ? NULL : addr;
2481 }
2482
2483 // Don't update _highest_vm_reserved_address, because there might be memory
2484 // regions above addr + size. If so, releasing a memory region only creates
2485 // a hole in the address space, it doesn't help prevent heap-stack collision.
2486 //
2487 static int anon_munmap(char * addr, size_t size) {
2488 return ::munmap(addr, size) == 0;
2489 }
2490
2491 char* os::reserve_memory(size_t bytes, char* requested_addr,
2492 size_t alignment_hint) {
2493 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2494 }
2495
2496 bool os::release_memory(char* addr, size_t size) {
2497 return anon_munmap(addr, size);
2498 }
2499
2500 static address highest_vm_reserved_address() {
2501 return _highest_vm_reserved_address;
2502 }
2503
2504 static bool linux_mprotect(char* addr, size_t size, int prot) {
2505 // Linux wants the mprotect address argument to be page aligned.
2506 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2507
2508 // According to SUSv3, mprotect() should only be used with mappings
2509 // established by mmap(), and mmap() always maps whole pages. Unaligned
2510 // 'addr' likely indicates problem in the VM (e.g. trying to change
2511 // protection of malloc'ed or statically allocated memory). Check the
2512 // caller if you hit this assert.
2513 assert(addr == bottom, "sanity check");
2514
2515 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2516 return ::mprotect(bottom, size, prot) == 0;
2517 }
2518
2519 // Set protections specified
2520 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2521 bool is_committed) {
2522 unsigned int p = 0;
2523 switch (prot) {
2524 case MEM_PROT_NONE: p = PROT_NONE; break;
2525 case MEM_PROT_READ: p = PROT_READ; break;
2526 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2527 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2528 default:
2529 ShouldNotReachHere();
2530 }
2531 // is_committed is unused.
2532 return linux_mprotect(addr, bytes, p);
2533 }
2534
2535 bool os::guard_memory(char* addr, size_t size) {
2536 return linux_mprotect(addr, size, PROT_NONE);
2537 }
2538
2539 bool os::unguard_memory(char* addr, size_t size) {
2540 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2541 }
2542
2543 // Large page support
2544
2545 static size_t _large_page_size = 0;
2546
2547 bool os::large_page_init() {
2548 if (!UseLargePages) return false;
2549
2550 if (LargePageSizeInBytes) {
2551 _large_page_size = LargePageSizeInBytes;
2552 } else {
2553 // large_page_size on Linux is used to round up heap size. x86 uses either
2554 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2555 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2556 // page as large as 256M.
2557 //
2558 // Here we try to figure out page size by parsing /proc/meminfo and looking
2559 // for a line with the following format:
2560 // Hugepagesize: 2048 kB
2561 //
2562 // If we can't determine the value (e.g. /proc is not mounted, or the text
2563 // format has been changed), we'll use the largest page size supported by
2564 // the processor.
2565
2566 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2567
2568 FILE *fp = fopen("/proc/meminfo", "r");
2569 if (fp) {
2570 while (!feof(fp)) {
2571 int x = 0;
2572 char buf[16];
2573 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2574 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2575 _large_page_size = x * K;
2576 break;
2577 }
2578 } else {
2579 // skip to next line
2580 for (;;) {
2581 int ch = fgetc(fp);
2582 if (ch == EOF || ch == (int)'\n') break;
2583 }
2584 }
2585 }
2586 fclose(fp);
2587 }
2588 }
2589
2590 const size_t default_page_size = (size_t)Linux::page_size();
2591 if (_large_page_size > default_page_size) {
2592 _page_sizes[0] = _large_page_size;
2593 _page_sizes[1] = default_page_size;
2594 _page_sizes[2] = 0;
2595 }
2596
2597 // Large page support is available on 2.6 or newer kernel, some vendors
2598 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2599 // We optimistically assume the support is available. If later it turns out
2600 // not true, VM will automatically switch to use regular page size.
2601 return true;
2602 }
2603
2604 #ifndef SHM_HUGETLB
2605 #define SHM_HUGETLB 04000
2606 #endif
2607
2608 char* os::reserve_memory_special(size_t bytes) {
2609 assert(UseLargePages, "only for large pages");
2610
2611 key_t key = IPC_PRIVATE;
2612 char *addr;
2613
2614 bool warn_on_failure = UseLargePages &&
2615 (!FLAG_IS_DEFAULT(UseLargePages) ||
2616 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2617 );
2618 char msg[128];
2619
2620 // Create a large shared memory region to attach to based on size.
2621 // Currently, size is the total size of the heap
2622 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2623 if (shmid == -1) {
2624 // Possible reasons for shmget failure:
2625 // 1. shmmax is too small for Java heap.
2626 // > check shmmax value: cat /proc/sys/kernel/shmmax
2627 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2628 // 2. not enough large page memory.
2629 // > check available large pages: cat /proc/meminfo
2630 // > increase amount of large pages:
2631 // echo new_value > /proc/sys/vm/nr_hugepages
2632 // Note 1: different Linux may use different name for this property,
2633 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2634 // Note 2: it's possible there's enough physical memory available but
2635 // they are so fragmented after a long run that they can't
2636 // coalesce into large pages. Try to reserve large pages when
2637 // the system is still "fresh".
2638 if (warn_on_failure) {
2639 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2640 warning(msg);
2641 }
2642 return NULL;
2643 }
2644
2645 // attach to the region
2646 addr = (char*)shmat(shmid, NULL, 0);
2647 int err = errno;
2648
2649 // Remove shmid. If shmat() is successful, the actual shared memory segment
2650 // will be deleted when it's detached by shmdt() or when the process
2651 // terminates. If shmat() is not successful this will remove the shared
2652 // segment immediately.
2653 shmctl(shmid, IPC_RMID, NULL);
2654
2655 if ((intptr_t)addr == -1) {
2656 if (warn_on_failure) {
2657 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2658 warning(msg);
2659 }
2660 return NULL;
2661 }
2662
2663 return addr;
2664 }
2665
2666 bool os::release_memory_special(char* base, size_t bytes) {
2667 // detaching the SHM segment will also delete it, see reserve_memory_special()
2668 int rslt = shmdt(base);
2669 return rslt == 0;
2670 }
2671
2672 size_t os::large_page_size() {
2673 return _large_page_size;
2674 }
2675
2676 // Linux does not support anonymous mmap with large page memory. The only way
2677 // to reserve large page memory without file backing is through SysV shared
2678 // memory API. The entire memory region is committed and pinned upfront.
2679 // Hopefully this will change in the future...
2680 bool os::can_commit_large_page_memory() {
2681 return false;
2682 }
2683
2684 bool os::can_execute_large_page_memory() {
2685 return false;
2686 }
2687
2688 // Reserve memory at an arbitrary address, only if that area is
2689 // available (and not reserved for something else).
2690
2691 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2692 const int max_tries = 10;
2693 char* base[max_tries];
2694 size_t size[max_tries];
2695 const size_t gap = 0x000000;
2696
2697 // Assert only that the size is a multiple of the page size, since
2698 // that's all that mmap requires, and since that's all we really know
2699 // about at this low abstraction level. If we need higher alignment,
2700 // we can either pass an alignment to this method or verify alignment
2701 // in one of the methods further up the call chain. See bug 5044738.
2702 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2703
2704 // Repeatedly allocate blocks until the block is allocated at the
2705 // right spot. Give up after max_tries. Note that reserve_memory() will
2706 // automatically update _highest_vm_reserved_address if the call is
2707 // successful. The variable tracks the highest memory address every reserved
2708 // by JVM. It is used to detect heap-stack collision if running with
2709 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2710 // space than needed, it could confuse the collision detecting code. To
2711 // solve the problem, save current _highest_vm_reserved_address and
2712 // calculate the correct value before return.
2713 address old_highest = _highest_vm_reserved_address;
2714
2715 // Linux mmap allows caller to pass an address as hint; give it a try first,
2716 // if kernel honors the hint then we can return immediately.
2717 char * addr = anon_mmap(requested_addr, bytes, false);
2718 if (addr == requested_addr) {
2719 return requested_addr;
2720 }
2721
2722 if (addr != NULL) {
2723 // mmap() is successful but it fails to reserve at the requested address
2724 anon_munmap(addr, bytes);
2725 }
2726
2727 int i;
2728 for (i = 0; i < max_tries; ++i) {
2729 base[i] = reserve_memory(bytes);
2730
2731 if (base[i] != NULL) {
2732 // Is this the block we wanted?
2733 if (base[i] == requested_addr) {
2734 size[i] = bytes;
2735 break;
2736 }
2737
2738 // Does this overlap the block we wanted? Give back the overlapped
2739 // parts and try again.
2740
2741 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2742 if (top_overlap >= 0 && top_overlap < bytes) {
2743 unmap_memory(base[i], top_overlap);
2744 base[i] += top_overlap;
2745 size[i] = bytes - top_overlap;
2746 } else {
2747 size_t bottom_overlap = base[i] + bytes - requested_addr;
2748 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2749 unmap_memory(requested_addr, bottom_overlap);
2750 size[i] = bytes - bottom_overlap;
2751 } else {
2752 size[i] = bytes;
2753 }
2754 }
2755 }
2756 }
2757
2758 // Give back the unused reserved pieces.
2759
2760 for (int j = 0; j < i; ++j) {
2761 if (base[j] != NULL) {
2762 unmap_memory(base[j], size[j]);
2763 }
2764 }
2765
2766 if (i < max_tries) {
2767 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2768 return requested_addr;
2769 } else {
2770 _highest_vm_reserved_address = old_highest;
2771 return NULL;
2772 }
2773 }
2774
2775 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2776 return ::read(fd, buf, nBytes);
2777 }
2778
2779 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2780 // Solaris uses poll(), linux uses park().
2781 // Poll() is likely a better choice, assuming that Thread.interrupt()
2782 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2783 // SIGSEGV, see 4355769.
2784
2785 const int NANOSECS_PER_MILLISECS = 1000000;
2786
2787 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2788 assert(thread == Thread::current(), "thread consistency check");
2789
2790 ParkEvent * const slp = thread->_SleepEvent ;
2791 slp->reset() ;
2792 OrderAccess::fence() ;
2793
2794 if (interruptible) {
2795 jlong prevtime = javaTimeNanos();
2796
2797 for (;;) {
2798 if (os::is_interrupted(thread, true)) {
2799 return OS_INTRPT;
2800 }
2801
2802 jlong newtime = javaTimeNanos();
2803
2804 if (newtime - prevtime < 0) {
2805 // time moving backwards, should only happen if no monotonic clock
2806 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2807 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2808 } else {
2809 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2810 }
2811
2812 if(millis <= 0) {
2813 return OS_OK;
2814 }
2815
2816 prevtime = newtime;
2817
2818 {
2819 assert(thread->is_Java_thread(), "sanity check");
2820 JavaThread *jt = (JavaThread *) thread;
2821 ThreadBlockInVM tbivm(jt);
2822 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2823
2824 jt->set_suspend_equivalent();
2825 // cleared by handle_special_suspend_equivalent_condition() or
2826 // java_suspend_self() via check_and_wait_while_suspended()
2827
2828 slp->park(millis);
2829
2830 // were we externally suspended while we were waiting?
2831 jt->check_and_wait_while_suspended();
2832 }
2833 }
2834 } else {
2835 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2836 jlong prevtime = javaTimeNanos();
2837
2838 for (;;) {
2839 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2840 // the 1st iteration ...
2841 jlong newtime = javaTimeNanos();
2842
2843 if (newtime - prevtime < 0) {
2844 // time moving backwards, should only happen if no monotonic clock
2845 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2846 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2847 } else {
2848 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2849 }
2850
2851 if(millis <= 0) break ;
2852
2853 prevtime = newtime;
2854 slp->park(millis);
2855 }
2856 return OS_OK ;
2857 }
2858 }
2859
2860 int os::naked_sleep() {
2861 // %% make the sleep time an integer flag. for now use 1 millisec.
2862 return os::sleep(Thread::current(), 1, false);
2863 }
2864
2865 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2866 void os::infinite_sleep() {
2867 while (true) { // sleep forever ...
2868 ::sleep(100); // ... 100 seconds at a time
2869 }
2870 }
2871
2872 // Used to convert frequent JVM_Yield() to nops
2873 bool os::dont_yield() {
2874 return DontYieldALot;
2875 }
2876
2877 void os::yield() {
2878 sched_yield();
2879 }
2880
2881 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2882
2883 void os::yield_all(int attempts) {
2884 // Yields to all threads, including threads with lower priorities
2885 // Threads on Linux are all with same priority. The Solaris style
2886 // os::yield_all() with nanosleep(1ms) is not necessary.
2887 sched_yield();
2888 }
2889
2890 // Called from the tight loops to possibly influence time-sharing heuristics
2891 void os::loop_breaker(int attempts) {
2892 os::yield_all(attempts);
2893 }
2894
2895 ////////////////////////////////////////////////////////////////////////////////
2896 // thread priority support
2897
2898 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2899 // only supports dynamic priority, static priority must be zero. For real-time
2900 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2901 // However, for large multi-threaded applications, SCHED_RR is not only slower
2902 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2903 // of 5 runs - Sep 2005).
2904 //
2905 // The following code actually changes the niceness of kernel-thread/LWP. It
2906 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2907 // not the entire user process, and user level threads are 1:1 mapped to kernel
2908 // threads. It has always been the case, but could change in the future. For
2909 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2910 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2911
2912 int os::java_to_os_priority[MaxPriority + 1] = {
2913 19, // 0 Entry should never be used
2914
2915 4, // 1 MinPriority
2916 3, // 2
2917 2, // 3
2918
2919 1, // 4
2920 0, // 5 NormPriority
2921 -1, // 6
2922
2923 -2, // 7
2924 -3, // 8
2925 -4, // 9 NearMaxPriority
2926
2927 -5 // 10 MaxPriority
2928 };
2929
2930 static int prio_init() {
2931 if (ThreadPriorityPolicy == 1) {
2932 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2933 // if effective uid is not root. Perhaps, a more elegant way of doing
2934 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2935 if (geteuid() != 0) {
2936 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2937 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2938 }
2939 ThreadPriorityPolicy = 0;
2940 }
2941 }
2942 return 0;
2943 }
2944
2945 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2946 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2947
2948 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2949 return (ret == 0) ? OS_OK : OS_ERR;
2950 }
2951
2952 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2953 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2954 *priority_ptr = java_to_os_priority[NormPriority];
2955 return OS_OK;
2956 }
2957
2958 errno = 0;
2959 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2960 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2961 }
2962
2963 // Hint to the underlying OS that a task switch would not be good.
2964 // Void return because it's a hint and can fail.
2965 void os::hint_no_preempt() {}
2966
2967 ////////////////////////////////////////////////////////////////////////////////
2968 // suspend/resume support
2969
2970 // the low-level signal-based suspend/resume support is a remnant from the
2971 // old VM-suspension that used to be for java-suspension, safepoints etc,
2972 // within hotspot. Now there is a single use-case for this:
2973 // - calling get_thread_pc() on the VMThread by the flat-profiler task
2974 // that runs in the watcher thread.
2975 // The remaining code is greatly simplified from the more general suspension
2976 // code that used to be used.
2977 //
2978 // The protocol is quite simple:
2979 // - suspend:
2980 // - sends a signal to the target thread
2981 // - polls the suspend state of the osthread using a yield loop
2982 // - target thread signal handler (SR_handler) sets suspend state
2983 // and blocks in sigsuspend until continued
2984 // - resume:
2985 // - sets target osthread state to continue
2986 // - sends signal to end the sigsuspend loop in the SR_handler
2987 //
2988 // Note that the SR_lock plays no role in this suspend/resume protocol.
2989 //
2990
2991 static void resume_clear_context(OSThread *osthread) {
2992 osthread->set_ucontext(NULL);
2993 osthread->set_siginfo(NULL);
2994
2995 // notify the suspend action is completed, we have now resumed
2996 osthread->sr.clear_suspended();
2997 }
2998
2999 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3000 osthread->set_ucontext(context);
3001 osthread->set_siginfo(siginfo);
3002 }
3003
3004 //
3005 // Handler function invoked when a thread's execution is suspended or
3006 // resumed. We have to be careful that only async-safe functions are
3007 // called here (Note: most pthread functions are not async safe and
3008 // should be avoided.)
3009 //
3010 // Note: sigwait() is a more natural fit than sigsuspend() from an
3011 // interface point of view, but sigwait() prevents the signal hander
3012 // from being run. libpthread would get very confused by not having
3013 // its signal handlers run and prevents sigwait()'s use with the
3014 // mutex granting granting signal.
3015 //
3016 // Currently only ever called on the VMThread
3017 //
3018 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3019 // Save and restore errno to avoid confusing native code with EINTR
3020 // after sigsuspend.
3021 int old_errno = errno;
3022
3023 Thread* thread = Thread::current();
3024 OSThread* osthread = thread->osthread();
3025 assert(thread->is_VM_thread(), "Must be VMThread");
3026 // read current suspend action
3027 int action = osthread->sr.suspend_action();
3028 if (action == SR_SUSPEND) {
3029 suspend_save_context(osthread, siginfo, context);
3030
3031 // Notify the suspend action is about to be completed. do_suspend()
3032 // waits until SR_SUSPENDED is set and then returns. We will wait
3033 // here for a resume signal and that completes the suspend-other
3034 // action. do_suspend/do_resume is always called as a pair from
3035 // the same thread - so there are no races
3036
3037 // notify the caller
3038 osthread->sr.set_suspended();
3039
3040 sigset_t suspend_set; // signals for sigsuspend()
3041
3042 // get current set of blocked signals and unblock resume signal
3043 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3044 sigdelset(&suspend_set, SR_signum);
3045
3046 // wait here until we are resumed
3047 do {
3048 sigsuspend(&suspend_set);
3049 // ignore all returns until we get a resume signal
3050 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3051
3052 resume_clear_context(osthread);
3053
3054 } else {
3055 assert(action == SR_CONTINUE, "unexpected sr action");
3056 // nothing special to do - just leave the handler
3057 }
3058
3059 errno = old_errno;
3060 }
3061
3062
3063 static int SR_initialize() {
3064 struct sigaction act;
3065 char *s;
3066 /* Get signal number to use for suspend/resume */
3067 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3068 int sig = ::strtol(s, 0, 10);
3069 if (sig > 0 || sig < _NSIG) {
3070 SR_signum = sig;
3071 }
3072 }
3073
3074 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3075 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3076
3077 sigemptyset(&SR_sigset);
3078 sigaddset(&SR_sigset, SR_signum);
3079
3080 /* Set up signal handler for suspend/resume */
3081 act.sa_flags = SA_RESTART|SA_SIGINFO;
3082 act.sa_handler = (void (*)(int)) SR_handler;
3083
3084 // SR_signum is blocked by default.
3085 // 4528190 - We also need to block pthread restart signal (32 on all
3086 // supported Linux platforms). Note that LinuxThreads need to block
3087 // this signal for all threads to work properly. So we don't have
3088 // to use hard-coded signal number when setting up the mask.
3089 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3090
3091 if (sigaction(SR_signum, &act, 0) == -1) {
3092 return -1;
3093 }
3094
3095 // Save signal flag
3096 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3097 return 0;
3098 }
3099
3100 static int SR_finalize() {
3101 return 0;
3102 }
3103
3104
3105 // returns true on success and false on error - really an error is fatal
3106 // but this seems the normal response to library errors
3107 static bool do_suspend(OSThread* osthread) {
3108 // mark as suspended and send signal
3109 osthread->sr.set_suspend_action(SR_SUSPEND);
3110 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3111 assert_status(status == 0, status, "pthread_kill");
3112
3113 // check status and wait until notified of suspension
3114 if (status == 0) {
3115 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3116 os::yield_all(i);
3117 }
3118 osthread->sr.set_suspend_action(SR_NONE);
3119 return true;
3120 }
3121 else {
3122 osthread->sr.set_suspend_action(SR_NONE);
3123 return false;
3124 }
3125 }
3126
3127 static void do_resume(OSThread* osthread) {
3128 assert(osthread->sr.is_suspended(), "thread should be suspended");
3129 osthread->sr.set_suspend_action(SR_CONTINUE);
3130
3131 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3132 assert_status(status == 0, status, "pthread_kill");
3133 // check status and wait unit notified of resumption
3134 if (status == 0) {
3135 for (int i = 0; osthread->sr.is_suspended(); i++) {
3136 os::yield_all(i);
3137 }
3138 }
3139 osthread->sr.set_suspend_action(SR_NONE);
3140 }
3141
3142 ////////////////////////////////////////////////////////////////////////////////
3143 // interrupt support
3144
3145 void os::interrupt(Thread* thread) {
3146 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3147 "possibility of dangling Thread pointer");
3148
3149 OSThread* osthread = thread->osthread();
3150
3151 if (!osthread->interrupted()) {
3152 osthread->set_interrupted(true);
3153 // More than one thread can get here with the same value of osthread,
3154 // resulting in multiple notifications. We do, however, want the store
3155 // to interrupted() to be visible to other threads before we execute unpark().
3156 OrderAccess::fence();
3157 ParkEvent * const slp = thread->_SleepEvent ;
3158 if (slp != NULL) slp->unpark() ;
3159 }
3160
3161 // For JSR166. Unpark even if interrupt status already was set
3162 if (thread->is_Java_thread())
3163 ((JavaThread*)thread)->parker()->unpark();
3164
3165 ParkEvent * ev = thread->_ParkEvent ;
3166 if (ev != NULL) ev->unpark() ;
3167
3168 }
3169
3170 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3171 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3172 "possibility of dangling Thread pointer");
3173
3174 OSThread* osthread = thread->osthread();
3175
3176 bool interrupted = osthread->interrupted();
3177
3178 if (interrupted && clear_interrupted) {
3179 osthread->set_interrupted(false);
3180 // consider thread->_SleepEvent->reset() ... optional optimization
3181 }
3182
3183 return interrupted;
3184 }
3185
3186 ///////////////////////////////////////////////////////////////////////////////////
3187 // signal handling (except suspend/resume)
3188
3189 // This routine may be used by user applications as a "hook" to catch signals.
3190 // The user-defined signal handler must pass unrecognized signals to this
3191 // routine, and if it returns true (non-zero), then the signal handler must
3192 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3193 // routine will never retun false (zero), but instead will execute a VM panic
3194 // routine kill the process.
3195 //
3196 // If this routine returns false, it is OK to call it again. This allows
3197 // the user-defined signal handler to perform checks either before or after
3198 // the VM performs its own checks. Naturally, the user code would be making
3199 // a serious error if it tried to handle an exception (such as a null check
3200 // or breakpoint) that the VM was generating for its own correct operation.
3201 //
3202 // This routine may recognize any of the following kinds of signals:
3203 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3204 // It should be consulted by handlers for any of those signals.
3205 //
3206 // The caller of this routine must pass in the three arguments supplied
3207 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3208 // field of the structure passed to sigaction(). This routine assumes that
3209 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3210 //
3211 // Note that the VM will print warnings if it detects conflicting signal
3212 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3213 //
3214 extern "C" int
3215 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3216 void* ucontext, int abort_if_unrecognized);
3217
3218 void signalHandler(int sig, siginfo_t* info, void* uc) {
3219 assert(info != NULL && uc != NULL, "it must be old kernel");
3220 JVM_handle_linux_signal(sig, info, uc, true);
3221 }
3222
3223
3224 // This boolean allows users to forward their own non-matching signals
3225 // to JVM_handle_linux_signal, harmlessly.
3226 bool os::Linux::signal_handlers_are_installed = false;
3227
3228 // For signal-chaining
3229 struct sigaction os::Linux::sigact[MAXSIGNUM];
3230 unsigned int os::Linux::sigs = 0;
3231 bool os::Linux::libjsig_is_loaded = false;
3232 typedef struct sigaction *(*get_signal_t)(int);
3233 get_signal_t os::Linux::get_signal_action = NULL;
3234
3235 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3236 struct sigaction *actp = NULL;
3237
3238 if (libjsig_is_loaded) {
3239 // Retrieve the old signal handler from libjsig
3240 actp = (*get_signal_action)(sig);
3241 }
3242 if (actp == NULL) {
3243 // Retrieve the preinstalled signal handler from jvm
3244 actp = get_preinstalled_handler(sig);
3245 }
3246
3247 return actp;
3248 }
3249
3250 static bool call_chained_handler(struct sigaction *actp, int sig,
3251 siginfo_t *siginfo, void *context) {
3252 // Call the old signal handler
3253 if (actp->sa_handler == SIG_DFL) {
3254 // It's more reasonable to let jvm treat it as an unexpected exception
3255 // instead of taking the default action.
3256 return false;
3257 } else if (actp->sa_handler != SIG_IGN) {
3258 if ((actp->sa_flags & SA_NODEFER) == 0) {
3259 // automaticlly block the signal
3260 sigaddset(&(actp->sa_mask), sig);
3261 }
3262
3263 sa_handler_t hand;
3264 sa_sigaction_t sa;
3265 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3266 // retrieve the chained handler
3267 if (siginfo_flag_set) {
3268 sa = actp->sa_sigaction;
3269 } else {
3270 hand = actp->sa_handler;
3271 }
3272
3273 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3274 actp->sa_handler = SIG_DFL;
3275 }
3276
3277 // try to honor the signal mask
3278 sigset_t oset;
3279 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3280
3281 // call into the chained handler
3282 if (siginfo_flag_set) {
3283 (*sa)(sig, siginfo, context);
3284 } else {
3285 (*hand)(sig);
3286 }
3287
3288 // restore the signal mask
3289 pthread_sigmask(SIG_SETMASK, &oset, 0);
3290 }
3291 // Tell jvm's signal handler the signal is taken care of.
3292 return true;
3293 }
3294
3295 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3296 bool chained = false;
3297 // signal-chaining
3298 if (UseSignalChaining) {
3299 struct sigaction *actp = get_chained_signal_action(sig);
3300 if (actp != NULL) {
3301 chained = call_chained_handler(actp, sig, siginfo, context);
3302 }
3303 }
3304 return chained;
3305 }
3306
3307 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3308 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3309 return &sigact[sig];
3310 }
3311 return NULL;
3312 }
3313
3314 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3315 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3316 sigact[sig] = oldAct;
3317 sigs |= (unsigned int)1 << sig;
3318 }
3319
3320 // for diagnostic
3321 int os::Linux::sigflags[MAXSIGNUM];
3322
3323 int os::Linux::get_our_sigflags(int sig) {
3324 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3325 return sigflags[sig];
3326 }
3327
3328 void os::Linux::set_our_sigflags(int sig, int flags) {
3329 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3330 sigflags[sig] = flags;
3331 }
3332
3333 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3334 // Check for overwrite.
3335 struct sigaction oldAct;
3336 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3337
3338 void* oldhand = oldAct.sa_sigaction
3339 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3340 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3341 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3342 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3343 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3344 if (AllowUserSignalHandlers || !set_installed) {
3345 // Do not overwrite; user takes responsibility to forward to us.
3346 return;
3347 } else if (UseSignalChaining) {
3348 // save the old handler in jvm
3349 save_preinstalled_handler(sig, oldAct);
3350 // libjsig also interposes the sigaction() call below and saves the
3351 // old sigaction on it own.
3352 } else {
3353 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3354 }
3355 }
3356
3357 struct sigaction sigAct;
3358 sigfillset(&(sigAct.sa_mask));
3359 sigAct.sa_handler = SIG_DFL;
3360 if (!set_installed) {
3361 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3362 } else {
3363 sigAct.sa_sigaction = signalHandler;
3364 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3365 }
3366 // Save flags, which are set by ours
3367 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3368 sigflags[sig] = sigAct.sa_flags;
3369
3370 int ret = sigaction(sig, &sigAct, &oldAct);
3371 assert(ret == 0, "check");
3372
3373 void* oldhand2 = oldAct.sa_sigaction
3374 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3375 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3376 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3377 }
3378
3379 // install signal handlers for signals that HotSpot needs to
3380 // handle in order to support Java-level exception handling.
3381
3382 void os::Linux::install_signal_handlers() {
3383 if (!signal_handlers_are_installed) {
3384 signal_handlers_are_installed = true;
3385
3386 // signal-chaining
3387 typedef void (*signal_setting_t)();
3388 signal_setting_t begin_signal_setting = NULL;
3389 signal_setting_t end_signal_setting = NULL;
3390 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3391 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3392 if (begin_signal_setting != NULL) {
3393 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3394 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3395 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3396 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3397 libjsig_is_loaded = true;
3398 assert(UseSignalChaining, "should enable signal-chaining");
3399 }
3400 if (libjsig_is_loaded) {
3401 // Tell libjsig jvm is setting signal handlers
3402 (*begin_signal_setting)();
3403 }
3404
3405 set_signal_handler(SIGSEGV, true);
3406 set_signal_handler(SIGPIPE, true);
3407 set_signal_handler(SIGBUS, true);
3408 set_signal_handler(SIGILL, true);
3409 set_signal_handler(SIGFPE, true);
3410 set_signal_handler(SIGXFSZ, true);
3411
3412 if (libjsig_is_loaded) {
3413 // Tell libjsig jvm finishes setting signal handlers
3414 (*end_signal_setting)();
3415 }
3416
3417 // We don't activate signal checker if libjsig is in place, we trust ourselves
3418 // and if UserSignalHandler is installed all bets are off
3419 if (CheckJNICalls) {
3420 if (libjsig_is_loaded) {
3421 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3422 check_signals = false;
3423 }
3424 if (AllowUserSignalHandlers) {
3425 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3426 check_signals = false;
3427 }
3428 }
3429 }
3430 }
3431
3432 // This is the fastest way to get thread cpu time on Linux.
3433 // Returns cpu time (user+sys) for any thread, not only for current.
3434 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3435 // It might work on 2.6.10+ with a special kernel/glibc patch.
3436 // For reference, please, see IEEE Std 1003.1-2004:
3437 // http://www.unix.org/single_unix_specification
3438
3439 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3440 struct timespec tp;
3441 int rc = os::Linux::clock_gettime(clockid, &tp);
3442 assert(rc == 0, "clock_gettime is expected to return 0 code");
3443
3444 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3445 }
3446
3447 /////
3448 // glibc on Linux platform uses non-documented flag
3449 // to indicate, that some special sort of signal
3450 // trampoline is used.
3451 // We will never set this flag, and we should
3452 // ignore this flag in our diagnostic
3453 #ifdef SIGNIFICANT_SIGNAL_MASK
3454 #undef SIGNIFICANT_SIGNAL_MASK
3455 #endif
3456 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3457
3458 static const char* get_signal_handler_name(address handler,
3459 char* buf, int buflen) {
3460 int offset;
3461 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3462 if (found) {
3463 // skip directory names
3464 const char *p1, *p2;
3465 p1 = buf;
3466 size_t len = strlen(os::file_separator());
3467 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3468 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3469 } else {
3470 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3471 }
3472 return buf;
3473 }
3474
3475 static void print_signal_handler(outputStream* st, int sig,
3476 char* buf, size_t buflen) {
3477 struct sigaction sa;
3478
3479 sigaction(sig, NULL, &sa);
3480
3481 // See comment for SIGNIFICANT_SIGNAL_MASK define
3482 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3483
3484 st->print("%s: ", os::exception_name(sig, buf, buflen));
3485
3486 address handler = (sa.sa_flags & SA_SIGINFO)
3487 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3488 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3489
3490 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3491 st->print("SIG_DFL");
3492 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3493 st->print("SIG_IGN");
3494 } else {
3495 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3496 }
3497
3498 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3499
3500 address rh = VMError::get_resetted_sighandler(sig);
3501 // May be, handler was resetted by VMError?
3502 if(rh != NULL) {
3503 handler = rh;
3504 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3505 }
3506
3507 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3508
3509 // Check: is it our handler?
3510 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3511 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3512 // It is our signal handler
3513 // check for flags, reset system-used one!
3514 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3515 st->print(
3516 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3517 os::Linux::get_our_sigflags(sig));
3518 }
3519 }
3520 st->cr();
3521 }
3522
3523
3524 #define DO_SIGNAL_CHECK(sig) \
3525 if (!sigismember(&check_signal_done, sig)) \
3526 os::Linux::check_signal_handler(sig)
3527
3528 // This method is a periodic task to check for misbehaving JNI applications
3529 // under CheckJNI, we can add any periodic checks here
3530
3531 void os::run_periodic_checks() {
3532
3533 if (check_signals == false) return;
3534
3535 // SEGV and BUS if overridden could potentially prevent
3536 // generation of hs*.log in the event of a crash, debugging
3537 // such a case can be very challenging, so we absolutely
3538 // check the following for a good measure:
3539 DO_SIGNAL_CHECK(SIGSEGV);
3540 DO_SIGNAL_CHECK(SIGILL);
3541 DO_SIGNAL_CHECK(SIGFPE);
3542 DO_SIGNAL_CHECK(SIGBUS);
3543 DO_SIGNAL_CHECK(SIGPIPE);
3544 DO_SIGNAL_CHECK(SIGXFSZ);
3545
3546
3547 // ReduceSignalUsage allows the user to override these handlers
3548 // see comments at the very top and jvm_solaris.h
3549 if (!ReduceSignalUsage) {
3550 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3551 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3552 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3553 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3554 }
3555
3556 DO_SIGNAL_CHECK(SR_signum);
3557 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3558 }
3559
3560 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3561
3562 static os_sigaction_t os_sigaction = NULL;
3563
3564 void os::Linux::check_signal_handler(int sig) {
3565 char buf[O_BUFLEN];
3566 address jvmHandler = NULL;
3567
3568
3569 struct sigaction act;
3570 if (os_sigaction == NULL) {
3571 // only trust the default sigaction, in case it has been interposed
3572 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3573 if (os_sigaction == NULL) return;
3574 }
3575
3576 os_sigaction(sig, (struct sigaction*)NULL, &act);
3577
3578
3579 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3580
3581 address thisHandler = (act.sa_flags & SA_SIGINFO)
3582 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3583 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3584
3585
3586 switch(sig) {
3587 case SIGSEGV:
3588 case SIGBUS:
3589 case SIGFPE:
3590 case SIGPIPE:
3591 case SIGILL:
3592 case SIGXFSZ:
3593 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3594 break;
3595
3596 case SHUTDOWN1_SIGNAL:
3597 case SHUTDOWN2_SIGNAL:
3598 case SHUTDOWN3_SIGNAL:
3599 case BREAK_SIGNAL:
3600 jvmHandler = (address)user_handler();
3601 break;
3602
3603 case INTERRUPT_SIGNAL:
3604 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3605 break;
3606
3607 default:
3608 if (sig == SR_signum) {
3609 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3610 } else {
3611 return;
3612 }
3613 break;
3614 }
3615
3616 if (thisHandler != jvmHandler) {
3617 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3618 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3619 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3620 // No need to check this sig any longer
3621 sigaddset(&check_signal_done, sig);
3622 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3623 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3624 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3625 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3626 // No need to check this sig any longer
3627 sigaddset(&check_signal_done, sig);
3628 }
3629
3630 // Dump all the signal
3631 if (sigismember(&check_signal_done, sig)) {
3632 print_signal_handlers(tty, buf, O_BUFLEN);
3633 }
3634 }
3635
3636 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3637
3638 extern bool signal_name(int signo, char* buf, size_t len);
3639
3640 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3641 if (0 < exception_code && exception_code <= SIGRTMAX) {
3642 // signal
3643 if (!signal_name(exception_code, buf, size)) {
3644 jio_snprintf(buf, size, "SIG%d", exception_code);
3645 }
3646 return buf;
3647 } else {
3648 return NULL;
3649 }
3650 }
3651
3652 // this is called _before_ the most of global arguments have been parsed
3653 void os::init(void) {
3654 char dummy; /* used to get a guess on initial stack address */
3655 // first_hrtime = gethrtime();
3656
3657 // With LinuxThreads the JavaMain thread pid (primordial thread)
3658 // is different than the pid of the java launcher thread.
3659 // So, on Linux, the launcher thread pid is passed to the VM
3660 // via the sun.java.launcher.pid property.
3661 // Use this property instead of getpid() if it was correctly passed.
3662 // See bug 6351349.
3663 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3664
3665 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3666
3667 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3668
3669 init_random(1234567);
3670
3671 ThreadCritical::initialize();
3672
3673 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3674 if (Linux::page_size() == -1) {
3675 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3676 }
3677 init_page_sizes((size_t) Linux::page_size());
3678
3679 Linux::initialize_system_info();
3680
3681 // main_thread points to the aboriginal thread
3682 Linux::_main_thread = pthread_self();
3683
3684 Linux::clock_init();
3685 initial_time_count = os::elapsed_counter();
3686 pthread_mutex_init(&dl_mutex, NULL);
3687 }
3688
3689 // To install functions for atexit system call
3690 extern "C" {
3691 static void perfMemory_exit_helper() {
3692 perfMemory_exit();
3693 }
3694 }
3695
3696 // this is called _after_ the global arguments have been parsed
3697 jint os::init_2(void)
3698 {
3699 Linux::fast_thread_clock_init();
3700
3701 // Allocate a single page and mark it as readable for safepoint polling
3702 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3703 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3704
3705 os::set_polling_page( polling_page );
3706
3707 #ifndef PRODUCT
3708 if(Verbose && PrintMiscellaneous)
3709 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3710 #endif
3711
3712 if (!UseMembar) {
3713 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3714 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3715 os::set_memory_serialize_page( mem_serialize_page );
3716
3717 #ifndef PRODUCT
3718 if(Verbose && PrintMiscellaneous)
3719 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3720 #endif
3721 }
3722
3723 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3724
3725 // initialize suspend/resume support - must do this before signal_sets_init()
3726 if (SR_initialize() != 0) {
3727 perror("SR_initialize failed");
3728 return JNI_ERR;
3729 }
3730
3731 Linux::signal_sets_init();
3732 Linux::install_signal_handlers();
3733
3734 size_t threadStackSizeInBytes = ThreadStackSize * K;
3735 if (threadStackSizeInBytes != 0 &&
3736 threadStackSizeInBytes < Linux::min_stack_allowed) {
3737 tty->print_cr("\nThe stack size specified is too small, "
3738 "Specify at least %dk",
3739 Linux::min_stack_allowed / K);
3740 return JNI_ERR;
3741 }
3742
3743 // Make the stack size a multiple of the page size so that
3744 // the yellow/red zones can be guarded.
3745 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3746 vm_page_size()));
3747
3748 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3749
3750 Linux::libpthread_init();
3751 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3752 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3753 Linux::glibc_version(), Linux::libpthread_version(),
3754 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3755 }
3756
3757 if (UseNUMA) {
3758 if (!Linux::libnuma_init()) {
3759 UseNUMA = false;
3760 } else {
3761 if ((Linux::numa_max_node() < 1)) {
3762 // There's only one node(they start from 0), disable NUMA.
3763 UseNUMA = false;
3764 }
3765 }
3766 if (!UseNUMA && ForceNUMA) {
3767 UseNUMA = true;
3768 }
3769 }
3770
3771 if (MaxFDLimit) {
3772 // set the number of file descriptors to max. print out error
3773 // if getrlimit/setrlimit fails but continue regardless.
3774 struct rlimit nbr_files;
3775 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3776 if (status != 0) {
3777 if (PrintMiscellaneous && (Verbose || WizardMode))
3778 perror("os::init_2 getrlimit failed");
3779 } else {
3780 nbr_files.rlim_cur = nbr_files.rlim_max;
3781 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3782 if (status != 0) {
3783 if (PrintMiscellaneous && (Verbose || WizardMode))
3784 perror("os::init_2 setrlimit failed");
3785 }
3786 }
3787 }
3788
3789 // Initialize lock used to serialize thread creation (see os::create_thread)
3790 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3791
3792 // Initialize HPI.
3793 jint hpi_result = hpi::initialize();
3794 if (hpi_result != JNI_OK) {
3795 tty->print_cr("There was an error trying to initialize the HPI library.");
3796 return hpi_result;
3797 }
3798
3799 // at-exit methods are called in the reverse order of their registration.
3800 // atexit functions are called on return from main or as a result of a
3801 // call to exit(3C). There can be only 32 of these functions registered
3802 // and atexit() does not set errno.
3803
3804 if (PerfAllowAtExitRegistration) {
3805 // only register atexit functions if PerfAllowAtExitRegistration is set.
3806 // atexit functions can be delayed until process exit time, which
3807 // can be problematic for embedded VM situations. Embedded VMs should
3808 // call DestroyJavaVM() to assure that VM resources are released.
3809
3810 // note: perfMemory_exit_helper atexit function may be removed in
3811 // the future if the appropriate cleanup code can be added to the
3812 // VM_Exit VMOperation's doit method.
3813 if (atexit(perfMemory_exit_helper) != 0) {
3814 warning("os::init2 atexit(perfMemory_exit_helper) failed");
3815 }
3816 }
3817
3818 // initialize thread priority policy
3819 prio_init();
3820
3821 return JNI_OK;
3822 }
3823
3824 // Mark the polling page as unreadable
3825 void os::make_polling_page_unreadable(void) {
3826 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3827 fatal("Could not disable polling page");
3828 };
3829
3830 // Mark the polling page as readable
3831 void os::make_polling_page_readable(void) {
3832 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3833 fatal("Could not enable polling page");
3834 }
3835 };
3836
3837 int os::active_processor_count() {
3838 // Linux doesn't yet have a (official) notion of processor sets,
3839 // so just return the number of online processors.
3840 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3841 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3842 return online_cpus;
3843 }
3844
3845 bool os::distribute_processes(uint length, uint* distribution) {
3846 // Not yet implemented.
3847 return false;
3848 }
3849
3850 bool os::bind_to_processor(uint processor_id) {
3851 // Not yet implemented.
3852 return false;
3853 }
3854
3855 ///
3856
3857 // Suspends the target using the signal mechanism and then grabs the PC before
3858 // resuming the target. Used by the flat-profiler only
3859 ExtendedPC os::get_thread_pc(Thread* thread) {
3860 // Make sure that it is called by the watcher for the VMThread
3861 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3862 assert(thread->is_VM_thread(), "Can only be called for VMThread");
3863
3864 ExtendedPC epc;
3865
3866 OSThread* osthread = thread->osthread();
3867 if (do_suspend(osthread)) {
3868 if (osthread->ucontext() != NULL) {
3869 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3870 } else {
3871 // NULL context is unexpected, double-check this is the VMThread
3872 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3873 }
3874 do_resume(osthread);
3875 }
3876 // failure means pthread_kill failed for some reason - arguably this is
3877 // a fatal problem, but such problems are ignored elsewhere
3878
3879 return epc;
3880 }
3881
3882 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3883 {
3884 if (is_NPTL()) {
3885 return pthread_cond_timedwait(_cond, _mutex, _abstime);
3886 } else {
3887 #ifndef IA64
3888 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3889 // word back to default 64bit precision if condvar is signaled. Java
3890 // wants 53bit precision. Save and restore current value.
3891 int fpu = get_fpu_control_word();
3892 #endif // IA64
3893 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3894 #ifndef IA64
3895 set_fpu_control_word(fpu);
3896 #endif // IA64
3897 return status;
3898 }
3899 }
3900
3901 ////////////////////////////////////////////////////////////////////////////////
3902 // debug support
3903
3904 #ifndef PRODUCT
3905 static address same_page(address x, address y) {
3906 int page_bits = -os::vm_page_size();
3907 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3908 return x;
3909 else if (x > y)
3910 return (address)(intptr_t(y) | ~page_bits) + 1;
3911 else
3912 return (address)(intptr_t(y) & page_bits);
3913 }
3914
3915 bool os::find(address addr) {
3916 Dl_info dlinfo;
3917 memset(&dlinfo, 0, sizeof(dlinfo));
3918 if (dladdr(addr, &dlinfo)) {
3919 tty->print(PTR_FORMAT ": ", addr);
3920 if (dlinfo.dli_sname != NULL) {
3921 tty->print("%s+%#x", dlinfo.dli_sname,
3922 addr - (intptr_t)dlinfo.dli_saddr);
3923 } else if (dlinfo.dli_fname) {
3924 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3925 } else {
3926 tty->print("<absolute address>");
3927 }
3928 if (dlinfo.dli_fname) {
3929 tty->print(" in %s", dlinfo.dli_fname);
3930 }
3931 if (dlinfo.dli_fbase) {
3932 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3933 }
3934 tty->cr();
3935
3936 if (Verbose) {
3937 // decode some bytes around the PC
3938 address begin = same_page(addr-40, addr);
3939 address end = same_page(addr+40, addr);
3940 address lowest = (address) dlinfo.dli_sname;
3941 if (!lowest) lowest = (address) dlinfo.dli_fbase;
3942 if (begin < lowest) begin = lowest;
3943 Dl_info dlinfo2;
3944 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3945 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3946 end = (address) dlinfo2.dli_saddr;
3947 Disassembler::decode(begin, end);
3948 }
3949 return true;
3950 }
3951 return false;
3952 }
3953
3954 #endif
3955
3956 ////////////////////////////////////////////////////////////////////////////////
3957 // misc
3958
3959 // This does not do anything on Linux. This is basically a hook for being
3960 // able to use structured exception handling (thread-local exception filters)
3961 // on, e.g., Win32.
3962 void
3963 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3964 JavaCallArguments* args, Thread* thread) {
3965 f(value, method, args, thread);
3966 }
3967
3968 void os::print_statistics() {
3969 }
3970
3971 int os::message_box(const char* title, const char* message) {
3972 int i;
3973 fdStream err(defaultStream::error_fd());
3974 for (i = 0; i < 78; i++) err.print_raw("=");
3975 err.cr();
3976 err.print_raw_cr(title);
3977 for (i = 0; i < 78; i++) err.print_raw("-");
3978 err.cr();
3979 err.print_raw_cr(message);
3980 for (i = 0; i < 78; i++) err.print_raw("=");
3981 err.cr();
3982
3983 char buf[16];
3984 // Prevent process from exiting upon "read error" without consuming all CPU
3985 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3986
3987 return buf[0] == 'y' || buf[0] == 'Y';
3988 }
3989
3990 int os::stat(const char *path, struct stat *sbuf) {
3991 char pathbuf[MAX_PATH];
3992 if (strlen(path) > MAX_PATH - 1) {
3993 errno = ENAMETOOLONG;
3994 return -1;
3995 }
3996 hpi::native_path(strcpy(pathbuf, path));
3997 return ::stat(pathbuf, sbuf);
3998 }
3999
4000 bool os::check_heap(bool force) {
4001 return true;
4002 }
4003
4004 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4005 return ::vsnprintf(buf, count, format, args);
4006 }
4007
4008 // Is a (classpath) directory empty?
4009 bool os::dir_is_empty(const char* path) {
4010 DIR *dir = NULL;
4011 struct dirent *ptr;
4012
4013 dir = opendir(path);
4014 if (dir == NULL) return true;
4015
4016 /* Scan the directory */
4017 bool result = true;
4018 char buf[sizeof(struct dirent) + MAX_PATH];
4019 while (result && (ptr = ::readdir(dir)) != NULL) {
4020 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4021 result = false;
4022 }
4023 }
4024 closedir(dir);
4025 return result;
4026 }
4027
4028 // create binary file, rewriting existing file if required
4029 int os::create_binary_file(const char* path, bool rewrite_existing) {
4030 int oflags = O_WRONLY | O_CREAT;
4031 if (!rewrite_existing) {
4032 oflags |= O_EXCL;
4033 }
4034 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4035 }
4036
4037 // return current position of file pointer
4038 jlong os::current_file_offset(int fd) {
4039 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4040 }
4041
4042 // move file pointer to the specified offset
4043 jlong os::seek_to_file_offset(int fd, jlong offset) {
4044 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4045 }
4046
4047 // Map a block of memory.
4048 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4049 char *addr, size_t bytes, bool read_only,
4050 bool allow_exec) {
4051 int prot;
4052 int flags;
4053
4054 if (read_only) {
4055 prot = PROT_READ;
4056 flags = MAP_SHARED;
4057 } else {
4058 prot = PROT_READ | PROT_WRITE;
4059 flags = MAP_PRIVATE;
4060 }
4061
4062 if (allow_exec) {
4063 prot |= PROT_EXEC;
4064 }
4065
4066 if (addr != NULL) {
4067 flags |= MAP_FIXED;
4068 }
4069
4070 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4071 fd, file_offset);
4072 if (mapped_address == MAP_FAILED) {
4073 return NULL;
4074 }
4075 return mapped_address;
4076 }
4077
4078
4079 // Remap a block of memory.
4080 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4081 char *addr, size_t bytes, bool read_only,
4082 bool allow_exec) {
4083 // same as map_memory() on this OS
4084 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4085 allow_exec);
4086 }
4087
4088
4089 // Unmap a block of memory.
4090 bool os::unmap_memory(char* addr, size_t bytes) {
4091 return munmap(addr, bytes) == 0;
4092 }
4093
4094 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4095
4096 static clockid_t thread_cpu_clockid(Thread* thread) {
4097 pthread_t tid = thread->osthread()->pthread_id();
4098 clockid_t clockid;
4099
4100 // Get thread clockid
4101 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4102 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4103 return clockid;
4104 }
4105
4106 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4107 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4108 // of a thread.
4109 //
4110 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4111 // the fast estimate available on the platform.
4112
4113 jlong os::current_thread_cpu_time() {
4114 if (os::Linux::supports_fast_thread_cpu_time()) {
4115 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4116 } else {
4117 // return user + sys since the cost is the same
4118 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4119 }
4120 }
4121
4122 jlong os::thread_cpu_time(Thread* thread) {
4123 // consistent with what current_thread_cpu_time() returns
4124 if (os::Linux::supports_fast_thread_cpu_time()) {
4125 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4126 } else {
4127 return slow_thread_cpu_time(thread, true /* user + sys */);
4128 }
4129 }
4130
4131 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4132 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4133 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4134 } else {
4135 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4136 }
4137 }
4138
4139 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4140 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4141 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4142 } else {
4143 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4144 }
4145 }
4146
4147 //
4148 // -1 on error.
4149 //
4150
4151 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4152 static bool proc_pid_cpu_avail = true;
4153 static bool proc_task_unchecked = true;
4154 static const char *proc_stat_path = "/proc/%d/stat";
4155 pid_t tid = thread->osthread()->thread_id();
4156 int i;
4157 char *s;
4158 char stat[2048];
4159 int statlen;
4160 char proc_name[64];
4161 int count;
4162 long sys_time, user_time;
4163 char string[64];
4164 int idummy;
4165 long ldummy;
4166 FILE *fp;
4167
4168 // We first try accessing /proc/<pid>/cpu since this is faster to
4169 // process. If this file is not present (linux kernels 2.5 and above)
4170 // then we open /proc/<pid>/stat.
4171 if ( proc_pid_cpu_avail ) {
4172 sprintf(proc_name, "/proc/%d/cpu", tid);
4173 fp = fopen(proc_name, "r");
4174 if ( fp != NULL ) {
4175 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4176 fclose(fp);
4177 if ( count != 3 ) return -1;
4178
4179 if (user_sys_cpu_time) {
4180 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4181 } else {
4182 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4183 }
4184 }
4185 else proc_pid_cpu_avail = false;
4186 }
4187
4188 // The /proc/<tid>/stat aggregates per-process usage on
4189 // new Linux kernels 2.6+ where NPTL is supported.
4190 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4191 // See bug 6328462.
4192 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4193 // and possibly in some other cases, so we check its availability.
4194 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4195 // This is executed only once
4196 proc_task_unchecked = false;
4197 fp = fopen("/proc/self/task", "r");
4198 if (fp != NULL) {
4199 proc_stat_path = "/proc/self/task/%d/stat";
4200 fclose(fp);
4201 }
4202 }
4203
4204 sprintf(proc_name, proc_stat_path, tid);
4205 fp = fopen(proc_name, "r");
4206 if ( fp == NULL ) return -1;
4207 statlen = fread(stat, 1, 2047, fp);
4208 stat[statlen] = '\0';
4209 fclose(fp);
4210
4211 // Skip pid and the command string. Note that we could be dealing with
4212 // weird command names, e.g. user could decide to rename java launcher
4213 // to "java 1.4.2 :)", then the stat file would look like
4214 // 1234 (java 1.4.2 :)) R ... ...
4215 // We don't really need to know the command string, just find the last
4216 // occurrence of ")" and then start parsing from there. See bug 4726580.
4217 s = strrchr(stat, ')');
4218 i = 0;
4219 if (s == NULL ) return -1;
4220
4221 // Skip blank chars
4222 do s++; while (isspace(*s));
4223
4224 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4225 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4226 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4227 &user_time, &sys_time);
4228 if ( count != 13 ) return -1;
4229 if (user_sys_cpu_time) {
4230 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4231 } else {
4232 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4233 }
4234 }
4235
4236 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4237 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4238 info_ptr->may_skip_backward = false; // elapsed time not wall time
4239 info_ptr->may_skip_forward = false; // elapsed time not wall time
4240 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4241 }
4242
4243 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4244 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4245 info_ptr->may_skip_backward = false; // elapsed time not wall time
4246 info_ptr->may_skip_forward = false; // elapsed time not wall time
4247 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4248 }
4249
4250 bool os::is_thread_cpu_time_supported() {
4251 return true;
4252 }
4253
4254 // System loadavg support. Returns -1 if load average cannot be obtained.
4255 // Linux doesn't yet have a (official) notion of processor sets,
4256 // so just return the system wide load average.
4257 int os::loadavg(double loadavg[], int nelem) {
4258 return ::getloadavg(loadavg, nelem);
4259 }
4260
4261 void os::pause() {
4262 char filename[MAX_PATH];
4263 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4264 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4265 } else {
4266 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4267 }
4268
4269 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4270 if (fd != -1) {
4271 struct stat buf;
4272 close(fd);
4273 while (::stat(filename, &buf) == 0) {
4274 (void)::poll(NULL, 0, 100);
4275 }
4276 } else {
4277 jio_fprintf(stderr,
4278 "Could not open pause file '%s', continuing immediately.\n", filename);
4279 }
4280 }
4281
4282 extern "C" {
4283
4284 /**
4285 * NOTE: the following code is to keep the green threads code
4286 * in the libjava.so happy. Once the green threads is removed,
4287 * these code will no longer be needed.
4288 */
4289 int
4290 jdk_waitpid(pid_t pid, int* status, int options) {
4291 return waitpid(pid, status, options);
4292 }
4293
4294 int
4295 fork1() {
4296 return fork();
4297 }
4298
4299 int
4300 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4301 return sem_init(sem, pshared, value);
4302 }
4303
4304 int
4305 jdk_sem_post(sem_t *sem) {
4306 return sem_post(sem);
4307 }
4308
4309 int
4310 jdk_sem_wait(sem_t *sem) {
4311 return sem_wait(sem);
4312 }
4313
4314 int
4315 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4316 return pthread_sigmask(how , newmask, oldmask);
4317 }
4318
4319 }
4320
4321 // Refer to the comments in os_solaris.cpp park-unpark.
4322 //
4323 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4324 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4325 // For specifics regarding the bug see GLIBC BUGID 261237 :
4326 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4327 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4328 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4329 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4330 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4331 // and monitorenter when we're using 1-0 locking. All those operations may result in
4332 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4333 // of libpthread avoids the problem, but isn't practical.
4334 //
4335 // Possible remedies:
4336 //
4337 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4338 // This is palliative and probabilistic, however. If the thread is preempted
4339 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4340 // than the minimum period may have passed, and the abstime may be stale (in the
4341 // past) resultin in a hang. Using this technique reduces the odds of a hang
4342 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4343 //
4344 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4345 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4346 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4347 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4348 // thread.
4349 //
4350 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4351 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4352 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4353 // This also works well. In fact it avoids kernel-level scalability impediments
4354 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4355 // timers in a graceful fashion.
4356 //
4357 // 4. When the abstime value is in the past it appears that control returns
4358 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4359 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4360 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4361 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4362 // It may be possible to avoid reinitialization by checking the return
4363 // value from pthread_cond_timedwait(). In addition to reinitializing the
4364 // condvar we must establish the invariant that cond_signal() is only called
4365 // within critical sections protected by the adjunct mutex. This prevents
4366 // cond_signal() from "seeing" a condvar that's in the midst of being
4367 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4368 // desirable signal-after-unlock optimization that avoids futile context switching.
4369 //
4370 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4371 // structure when a condvar is used or initialized. cond_destroy() would
4372 // release the helper structure. Our reinitialize-after-timedwait fix
4373 // put excessive stress on malloc/free and locks protecting the c-heap.
4374 //
4375 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4376 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4377 // and only enabling the work-around for vulnerable environments.
4378
4379 // utility to compute the abstime argument to timedwait:
4380 // millis is the relative timeout time
4381 // abstime will be the absolute timeout time
4382 // TODO: replace compute_abstime() with unpackTime()
4383
4384 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4385 if (millis < 0) millis = 0;
4386 struct timeval now;
4387 int status = gettimeofday(&now, NULL);
4388 assert(status == 0, "gettimeofday");
4389 jlong seconds = millis / 1000;
4390 millis %= 1000;
4391 if (seconds > 50000000) { // see man cond_timedwait(3T)
4392 seconds = 50000000;
4393 }
4394 abstime->tv_sec = now.tv_sec + seconds;
4395 long usec = now.tv_usec + millis * 1000;
4396 if (usec >= 1000000) {
4397 abstime->tv_sec += 1;
4398 usec -= 1000000;
4399 }
4400 abstime->tv_nsec = usec * 1000;
4401 return abstime;
4402 }
4403
4404
4405 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4406 // Conceptually TryPark() should be equivalent to park(0).
4407
4408 int os::PlatformEvent::TryPark() {
4409 for (;;) {
4410 const int v = _Event ;
4411 guarantee ((v == 0) || (v == 1), "invariant") ;
4412 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4413 }
4414 }
4415
4416 void os::PlatformEvent::park() { // AKA "down()"
4417 // Invariant: Only the thread associated with the Event/PlatformEvent
4418 // may call park().
4419 // TODO: assert that _Assoc != NULL or _Assoc == Self
4420 int v ;
4421 for (;;) {
4422 v = _Event ;
4423 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4424 }
4425 guarantee (v >= 0, "invariant") ;
4426 if (v == 0) {
4427 // Do this the hard way by blocking ...
4428 int status = pthread_mutex_lock(_mutex);
4429 assert_status(status == 0, status, "mutex_lock");
4430 guarantee (_nParked == 0, "invariant") ;
4431 ++ _nParked ;
4432 while (_Event < 0) {
4433 status = pthread_cond_wait(_cond, _mutex);
4434 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4435 // Treat this the same as if the wait was interrupted
4436 if (status == ETIME) { status = EINTR; }
4437 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4438 }
4439 -- _nParked ;
4440
4441 // In theory we could move the ST of 0 into _Event past the unlock(),
4442 // but then we'd need a MEMBAR after the ST.
4443 _Event = 0 ;
4444 status = pthread_mutex_unlock(_mutex);
4445 assert_status(status == 0, status, "mutex_unlock");
4446 }
4447 guarantee (_Event >= 0, "invariant") ;
4448 }
4449
4450 int os::PlatformEvent::park(jlong millis) {
4451 guarantee (_nParked == 0, "invariant") ;
4452
4453 int v ;
4454 for (;;) {
4455 v = _Event ;
4456 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4457 }
4458 guarantee (v >= 0, "invariant") ;
4459 if (v != 0) return OS_OK ;
4460
4461 // We do this the hard way, by blocking the thread.
4462 // Consider enforcing a minimum timeout value.
4463 struct timespec abst;
4464 compute_abstime(&abst, millis);
4465
4466 int ret = OS_TIMEOUT;
4467 int status = pthread_mutex_lock(_mutex);
4468 assert_status(status == 0, status, "mutex_lock");
4469 guarantee (_nParked == 0, "invariant") ;
4470 ++_nParked ;
4471
4472 // Object.wait(timo) will return because of
4473 // (a) notification
4474 // (b) timeout
4475 // (c) thread.interrupt
4476 //
4477 // Thread.interrupt and object.notify{All} both call Event::set.
4478 // That is, we treat thread.interrupt as a special case of notification.
4479 // The underlying Solaris implementation, cond_timedwait, admits
4480 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4481 // JVM from making those visible to Java code. As such, we must
4482 // filter out spurious wakeups. We assume all ETIME returns are valid.
4483 //
4484 // TODO: properly differentiate simultaneous notify+interrupt.
4485 // In that case, we should propagate the notify to another waiter.
4486
4487 while (_Event < 0) {
4488 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4489 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4490 pthread_cond_destroy (_cond);
4491 pthread_cond_init (_cond, NULL) ;
4492 }
4493 assert_status(status == 0 || status == EINTR ||
4494 status == ETIME || status == ETIMEDOUT,
4495 status, "cond_timedwait");
4496 if (!FilterSpuriousWakeups) break ; // previous semantics
4497 if (status == ETIME || status == ETIMEDOUT) break ;
4498 // We consume and ignore EINTR and spurious wakeups.
4499 }
4500 --_nParked ;
4501 if (_Event >= 0) {
4502 ret = OS_OK;
4503 }
4504 _Event = 0 ;
4505 status = pthread_mutex_unlock(_mutex);
4506 assert_status(status == 0, status, "mutex_unlock");
4507 assert (_nParked == 0, "invariant") ;
4508 return ret;
4509 }
4510
4511 void os::PlatformEvent::unpark() {
4512 int v, AnyWaiters ;
4513 for (;;) {
4514 v = _Event ;
4515 if (v > 0) {
4516 // The LD of _Event could have reordered or be satisfied
4517 // by a read-aside from this processor's write buffer.
4518 // To avoid problems execute a barrier and then
4519 // ratify the value.
4520 OrderAccess::fence() ;
4521 if (_Event == v) return ;
4522 continue ;
4523 }
4524 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4525 }
4526 if (v < 0) {
4527 // Wait for the thread associated with the event to vacate
4528 int status = pthread_mutex_lock(_mutex);
4529 assert_status(status == 0, status, "mutex_lock");
4530 AnyWaiters = _nParked ;
4531 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4532 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4533 AnyWaiters = 0 ;
4534 pthread_cond_signal (_cond);
4535 }
4536 status = pthread_mutex_unlock(_mutex);
4537 assert_status(status == 0, status, "mutex_unlock");
4538 if (AnyWaiters != 0) {
4539 status = pthread_cond_signal(_cond);
4540 assert_status(status == 0, status, "cond_signal");
4541 }
4542 }
4543
4544 // Note that we signal() _after dropping the lock for "immortal" Events.
4545 // This is safe and avoids a common class of futile wakeups. In rare
4546 // circumstances this can cause a thread to return prematurely from
4547 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4548 // simply re-test the condition and re-park itself.
4549 }
4550
4551
4552 // JSR166
4553 // -------------------------------------------------------
4554
4555 /*
4556 * The solaris and linux implementations of park/unpark are fairly
4557 * conservative for now, but can be improved. They currently use a
4558 * mutex/condvar pair, plus a a count.
4559 * Park decrements count if > 0, else does a condvar wait. Unpark
4560 * sets count to 1 and signals condvar. Only one thread ever waits
4561 * on the condvar. Contention seen when trying to park implies that someone
4562 * is unparking you, so don't wait. And spurious returns are fine, so there
4563 * is no need to track notifications.
4564 */
4565
4566
4567 #define NANOSECS_PER_SEC 1000000000
4568 #define NANOSECS_PER_MILLISEC 1000000
4569 #define MAX_SECS 100000000
4570 /*
4571 * This code is common to linux and solaris and will be moved to a
4572 * common place in dolphin.
4573 *
4574 * The passed in time value is either a relative time in nanoseconds
4575 * or an absolute time in milliseconds. Either way it has to be unpacked
4576 * into suitable seconds and nanoseconds components and stored in the
4577 * given timespec structure.
4578 * Given time is a 64-bit value and the time_t used in the timespec is only
4579 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4580 * overflow if times way in the future are given. Further on Solaris versions
4581 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4582 * number of seconds, in abstime, is less than current_time + 100,000,000.
4583 * As it will be 28 years before "now + 100000000" will overflow we can
4584 * ignore overflow and just impose a hard-limit on seconds using the value
4585 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4586 * years from "now".
4587 */
4588
4589 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4590 assert (time > 0, "convertTime");
4591
4592 struct timeval now;
4593 int status = gettimeofday(&now, NULL);
4594 assert(status == 0, "gettimeofday");
4595
4596 time_t max_secs = now.tv_sec + MAX_SECS;
4597
4598 if (isAbsolute) {
4599 jlong secs = time / 1000;
4600 if (secs > max_secs) {
4601 absTime->tv_sec = max_secs;
4602 }
4603 else {
4604 absTime->tv_sec = secs;
4605 }
4606 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4607 }
4608 else {
4609 jlong secs = time / NANOSECS_PER_SEC;
4610 if (secs >= MAX_SECS) {
4611 absTime->tv_sec = max_secs;
4612 absTime->tv_nsec = 0;
4613 }
4614 else {
4615 absTime->tv_sec = now.tv_sec + secs;
4616 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4617 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4618 absTime->tv_nsec -= NANOSECS_PER_SEC;
4619 ++absTime->tv_sec; // note: this must be <= max_secs
4620 }
4621 }
4622 }
4623 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4624 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4625 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4626 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4627 }
4628
4629 void Parker::park(bool isAbsolute, jlong time) {
4630 // Optional fast-path check:
4631 // Return immediately if a permit is available.
4632 if (_counter > 0) {
4633 _counter = 0 ;
4634 return ;
4635 }
4636
4637 Thread* thread = Thread::current();
4638 assert(thread->is_Java_thread(), "Must be JavaThread");
4639 JavaThread *jt = (JavaThread *)thread;
4640
4641 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4642 // Check interrupt before trying to wait
4643 if (Thread::is_interrupted(thread, false)) {
4644 return;
4645 }
4646
4647 // Next, demultiplex/decode time arguments
4648 timespec absTime;
4649 if (time < 0) { // don't wait at all
4650 return;
4651 }
4652 if (time > 0) {
4653 unpackTime(&absTime, isAbsolute, time);
4654 }
4655
4656
4657 // Enter safepoint region
4658 // Beware of deadlocks such as 6317397.
4659 // The per-thread Parker:: mutex is a classic leaf-lock.
4660 // In particular a thread must never block on the Threads_lock while
4661 // holding the Parker:: mutex. If safepoints are pending both the
4662 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4663 ThreadBlockInVM tbivm(jt);
4664
4665 // Don't wait if cannot get lock since interference arises from
4666 // unblocking. Also. check interrupt before trying wait
4667 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4668 return;
4669 }
4670
4671 int status ;
4672 if (_counter > 0) { // no wait needed
4673 _counter = 0;
4674 status = pthread_mutex_unlock(_mutex);
4675 assert (status == 0, "invariant") ;
4676 return;
4677 }
4678
4679 #ifdef ASSERT
4680 // Don't catch signals while blocked; let the running threads have the signals.
4681 // (This allows a debugger to break into the running thread.)
4682 sigset_t oldsigs;
4683 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4684 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4685 #endif
4686
4687 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4688 jt->set_suspend_equivalent();
4689 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4690
4691 if (time == 0) {
4692 status = pthread_cond_wait (_cond, _mutex) ;
4693 } else {
4694 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4695 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4696 pthread_cond_destroy (_cond) ;
4697 pthread_cond_init (_cond, NULL);
4698 }
4699 }
4700 assert_status(status == 0 || status == EINTR ||
4701 status == ETIME || status == ETIMEDOUT,
4702 status, "cond_timedwait");
4703
4704 #ifdef ASSERT
4705 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4706 #endif
4707
4708 _counter = 0 ;
4709 status = pthread_mutex_unlock(_mutex) ;
4710 assert_status(status == 0, status, "invariant") ;
4711 // If externally suspended while waiting, re-suspend
4712 if (jt->handle_special_suspend_equivalent_condition()) {
4713 jt->java_suspend_self();
4714 }
4715
4716 }
4717
4718 void Parker::unpark() {
4719 int s, status ;
4720 status = pthread_mutex_lock(_mutex);
4721 assert (status == 0, "invariant") ;
4722 s = _counter;
4723 _counter = 1;
4724 if (s < 1) {
4725 if (WorkAroundNPTLTimedWaitHang) {
4726 status = pthread_cond_signal (_cond) ;
4727 assert (status == 0, "invariant") ;
4728 status = pthread_mutex_unlock(_mutex);
4729 assert (status == 0, "invariant") ;
4730 } else {
4731 status = pthread_mutex_unlock(_mutex);
4732 assert (status == 0, "invariant") ;
4733 status = pthread_cond_signal (_cond) ;
4734 assert (status == 0, "invariant") ;
4735 }
4736 } else {
4737 pthread_mutex_unlock(_mutex);
4738 assert (status == 0, "invariant") ;
4739 }
4740 }
4741
4742
4743 extern char** environ;
4744
4745 #ifndef __NR_fork
4746 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4747 #endif
4748
4749 #ifndef __NR_execve
4750 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4751 #endif
4752
4753 // Run the specified command in a separate process. Return its exit value,
4754 // or -1 on failure (e.g. can't fork a new process).
4755 // Unlike system(), this function can be called from signal handler. It
4756 // doesn't block SIGINT et al.
4757 int os::fork_and_exec(char* cmd) {
4758 const char * argv[4] = {"sh", "-c", cmd, NULL};
4759
4760 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4761 // pthread_atfork handlers and reset pthread library. All we need is a
4762 // separate process to execve. Make a direct syscall to fork process.
4763 // On IA64 there's no fork syscall, we have to use fork() and hope for
4764 // the best...
4765 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4766 IA64_ONLY(fork();)
4767
4768 if (pid < 0) {
4769 // fork failed
4770 return -1;
4771
4772 } else if (pid == 0) {
4773 // child process
4774
4775 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4776 // first to kill every thread on the thread list. Because this list is
4777 // not reset by fork() (see notes above), execve() will instead kill
4778 // every thread in the parent process. We know this is the only thread
4779 // in the new process, so make a system call directly.
4780 // IA64 should use normal execve() from glibc to match the glibc fork()
4781 // above.
4782 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4783 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4784
4785 // execve failed
4786 _exit(-1);
4787
4788 } else {
4789 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4790 // care about the actual exit code, for now.
4791
4792 int status;
4793
4794 // Wait for the child process to exit. This returns immediately if
4795 // the child has already exited. */
4796 while (waitpid(pid, &status, 0) < 0) {
4797 switch (errno) {
4798 case ECHILD: return 0;
4799 case EINTR: break;
4800 default: return -1;
4801 }
4802 }
4803
4804 if (WIFEXITED(status)) {
4805 // The child exited normally; get its exit code.
4806 return WEXITSTATUS(status);
4807 } else if (WIFSIGNALED(status)) {
4808 // The child exited because of a signal
4809 // The best value to return is 0x80 + signal number,
4810 // because that is what all Unix shells do, and because
4811 // it allows callers to distinguish between process exit and
4812 // process death by signal.
4813 return 0x80 + WTERMSIG(status);
4814 } else {
4815 // Unknown exit code; pass it through
4816 return status;
4817 }
4818 }
4819 }